X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=ofproto%2Fofproto-dpif.c;h=c9d4a04969fa3965eeb3cd76b034a50049cac47f;hb=4a1b8f30e5956655b59200d673964b9f3f9ad711;hp=211ed428fb7083138dc45feada002c92419d8110;hpb=c423da59a8e0d605dfdea7b9bd4c42eb2eba8885;p=sliver-openvswitch.git diff --git a/ofproto/ofproto-dpif.c b/ofproto/ofproto-dpif.c index 211ed428f..4fc90c67b 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. @@ -16,11 +16,12 @@ #include +#include "ofproto/ofproto-dpif.h" #include "ofproto/ofproto-provider.h" #include -#include "autopath.h" +#include "bfd.h" #include "bond.h" #include "bundle.h" #include "byte-order.h" @@ -36,21 +37,27 @@ #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" #include "odp-util.h" +#include "odp-execute.h" #include "ofp-util.h" #include "ofpbuf.h" #include "ofp-actions.h" #include "ofp-parse.h" #include "ofp-print.h" #include "ofproto-dpif-governor.h" +#include "ofproto-dpif-ipfix.h" +#include "ofproto-dpif-mirror.h" #include "ofproto-dpif-sflow.h" +#include "ofproto-dpif-xlate.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" @@ -59,99 +66,28 @@ VLOG_DEFINE_THIS_MODULE(ofproto_dpif); COVERAGE_DEFINE(ofproto_dpif_expired); -COVERAGE_DEFINE(ofproto_dpif_xlate); COVERAGE_DEFINE(facet_changed_rule); COVERAGE_DEFINE(facet_revalidate); COVERAGE_DEFINE(facet_unexpected); COVERAGE_DEFINE(facet_suppress); - -/* Maximum depth of flow table recursion (due to resubmit actions) in a - * flow translation. */ -#define MAX_RESUBMIT_RECURSION 64 +COVERAGE_DEFINE(subfacet_install_fail); /* Number of implemented OpenFlow tables. */ enum { N_TABLES = 255 }; enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */ BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255); -struct ofport_dpif; -struct ofproto_dpif; struct flow_miss; - -struct rule_dpif { - struct rule up; - - /* These statistics: - * - * - Do include packets and bytes from facets that have been deleted or - * whose own statistics have been folded into the rule. - * - * - Do include packets and bytes sent "by hand" that were accounted to - * the rule without any facet being involved (this is a rare corner - * case in rule_execute()). - * - * - Do not include packet or bytes that can be obtained from any facet's - * packet_count or byte_count member or that can be obtained from the - * datapath by, e.g., dpif_flow_get() for any subfacet. - */ - uint64_t packet_count; /* Number of packets received. */ - uint64_t byte_count; /* Number of bytes received. */ - - tag_type tag; /* Caches rule_calculate_tag() result. */ - - struct list facets; /* List of "struct facet"s. */ -}; - -static struct rule_dpif *rule_dpif_cast(const struct rule *rule) -{ - return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL; -} +struct facet; static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *, - const struct flow *); -static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *, - const struct flow *, - uint8_t table); -static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto, - const struct flow *flow); - -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 tag_type rule_calculate_tag(const struct flow *, - const struct minimask *, uint32_t basis); -static void rule_invalidate(const struct rule_dpif *); - -#define MAX_MIRRORS 32 -typedef uint32_t mirror_mask_t; -#define MIRROR_MASK_C(X) UINT32_C(X) -BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS); -struct ofmirror { - struct ofproto_dpif *ofproto; /* Owning ofproto. */ - size_t idx; /* In ofproto's "mirrors" array. */ - void *aux; /* Key supplied by ofproto's client. */ - char *name; /* Identifier for log messages. */ - - /* Selection criteria. */ - struct hmapx srcs; /* Contains "struct ofbundle *"s. */ - struct hmapx dsts; /* Contains "struct ofbundle *"s. */ - unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */ - - /* Output (exactly one of out == NULL and out_vlan == -1 is true). */ - struct ofbundle *out; /* Output port or NULL. */ - int out_vlan; /* Output VLAN or -1. */ - mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */ + const struct flow *, + struct flow_wildcards *wc); - /* Counters. */ - int64_t packet_count; /* Number of packets sent. */ - int64_t byte_count; /* Number of bytes sent. */ -}; - -static void mirror_destroy(struct ofmirror *); -static void update_mirror_stats(struct ofproto_dpif *ofproto, - mirror_mask_t mirrors, - uint64_t packets, uint64_t bytes); +static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes); +static void rule_invalidate(const struct rule_dpif *); +static tag_type rule_calculate_tag(const struct flow *, + const struct minimask *, uint32_t secret); struct ofbundle { struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */ @@ -171,11 +107,6 @@ struct ofbundle { /* Status. */ bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */ - - /* Port mirroring info. */ - mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ - mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ - mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */ }; static void bundle_remove(struct ofport *); @@ -184,131 +115,18 @@ static void bundle_destroy(struct ofbundle *); static void bundle_del_port(struct ofport_dpif *); static void bundle_run(struct ofbundle *); static void bundle_wait(struct ofbundle *); -static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *, - uint16_t in_port, bool warn, - struct ofport_dpif **in_ofportp); - -/* A controller may use OFPP_NONE as the ingress port to indicate that - * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for - * when an input bundle is needed for validation (e.g., mirroring or - * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have - * any 'port' structs, so care must be taken when dealing with it. */ -static struct ofbundle ofpp_none_bundle = { - .name = "OFPP_NONE", - .vlan_mode = PORT_VLAN_TRUNK -}; static void stp_run(struct ofproto_dpif *ofproto); static void stp_wait(struct ofproto_dpif *ofproto); static int set_stp_port(struct ofport *, const struct ofproto_port_stp_settings *); -static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan); - -struct action_xlate_ctx { -/* action_xlate_ctx_init() initializes these members. */ - - /* The ofproto. */ - struct ofproto_dpif *ofproto; - - /* Flow to which the OpenFlow actions apply. xlate_actions() will modify - * this flow when actions change header fields. */ - struct flow flow; - - /* The packet corresponding to 'flow', or a null pointer if we are - * revalidating without a packet to refer to. */ - const struct ofpbuf *packet; - - /* Should OFPP_NORMAL update the MAC learning table? Should "learn" - * actions update the flow table? - * - * We want to update these tables if we are actually processing a packet, - * or if we are accounting for packets that the datapath has processed, but - * not if we are just revalidating. */ - bool may_learn; - - /* The rule that we are currently translating, or NULL. */ - struct rule_dpif *rule; - - /* Union of the set of TCP flags seen so far in this flow. (Used only by - * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules' - * timeouts.) */ - uint8_t tcp_flags; - - /* If nonnull, flow translation calls this function just before executing a - * resubmit or OFPP_TABLE action. In addition, disables logging of traces - * when the recursion depth is exceeded. - * - * 'rule' is the rule being submitted into. It will be null if the - * resubmit or OFPP_TABLE action didn't find a matching rule. - * - * This is normally null so the client has to set it manually after - * calling action_xlate_ctx_init(). */ - void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule); - - /* If nonnull, flow translation calls this function to report some - * significant decision, e.g. to explain why OFPP_NORMAL translation - * dropped a packet. */ - void (*report_hook)(struct action_xlate_ctx *, const char *s); - - /* If nonnull, flow translation credits the specified statistics to each - * rule reached through a resubmit or OFPP_TABLE action. - * - * This is normally null so the client has to set it manually after - * calling action_xlate_ctx_init(). */ - const struct dpif_flow_stats *resubmit_stats; - -/* xlate_actions() initializes and uses these members. The client might want - * to look at them after it returns. */ - - struct ofpbuf *odp_actions; /* Datapath actions. */ - tag_type tags; /* Tags associated with actions. */ - enum slow_path_reason slow; /* 0 if fast path may be used. */ - bool has_learn; /* Actions include NXAST_LEARN? */ - bool has_normal; /* Actions output to OFPP_NORMAL? */ - bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */ - uint16_t nf_output_iface; /* Output interface index for NetFlow. */ - mirror_mask_t mirrors; /* Bitmap of associated mirrors. */ - -/* xlate_actions() initializes and uses these members, but the client has no - * reason to look at them. */ - - int recurse; /* Recursion level, via xlate_table_action. */ - bool max_resubmit_trigger; /* Recursed too deeply during translation. */ - struct flow base_flow; /* Flow at the last commit. */ - uint32_t orig_skb_priority; /* Priority when packet arrived. */ - uint8_t table_id; /* OpenFlow table ID where flow was found. */ - uint32_t sflow_n_outputs; /* Number of output ports. */ - uint32_t sflow_odp_port; /* Output port for composing sFlow action. */ - uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */ - bool exit; /* No further actions should be processed. */ - struct flow orig_flow; /* Copy of original flow. */ -}; - -static void action_xlate_ctx_init(struct action_xlate_ctx *, - struct ofproto_dpif *, const struct flow *, - ovs_be16 initial_tci, 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, - struct ofpbuf *odp_actions); -static void xlate_actions_for_side_effects(struct action_xlate_ctx *, - const struct ofpact *ofpacts, - size_t ofpacts_len); - -static size_t put_userspace_action(const struct ofproto_dpif *, - struct ofpbuf *odp_actions, - const struct flow *, - const union user_action_cookie *); - static void compose_slow_path(const struct ofproto_dpif *, const struct flow *, enum slow_path_reason, uint64_t *stub, size_t stub_size, const struct nlattr **actionsp, size_t *actions_lenp); -static void xlate_report(struct action_xlate_ctx *ctx, const char *s); - /* A subfacet (see "struct subfacet" below) has three possible installation * states: * @@ -328,8 +146,6 @@ enum subfacet_path { SF_SLOW_PATH, /* Send-to-userspace action is installed. */ }; -static const char *subfacet_path_to_string(enum subfacet_path); - /* A dpif flow and actions associated with a facet. * * See also the large comment on struct facet. */ @@ -338,99 +154,73 @@ struct subfacet { struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */ struct list list_node; /* In struct facet's 'facets' list. */ struct facet *facet; /* Owning facet. */ + struct dpif_backer *backer; /* Owning backer. */ - /* 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. */ - /* Datapath actions. - * - * These should be essentially identical for every subfacet in a facet, but - * may differ in trivial ways due to VLAN splinters. */ - size_t actions_len; /* Number of bytes in actions[]. */ - struct nlattr *actions; /* Datapath actions. */ - - 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. */ - - /* 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; }; #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 *, +static struct subfacet *subfacet_find(struct dpif_backer *, const struct nlattr *key, size_t key_len, - uint32_t key_hash, - const struct flow *flow); + uint32_t key_hash); static void subfacet_destroy(struct subfacet *); static void subfacet_destroy__(struct subfacet *); -static void subfacet_destroy_batch(struct ofproto_dpif *, +static void subfacet_destroy_batch(struct dpif_backer *, struct subfacet **, int n); -static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *, - struct ofpbuf *key); static void subfacet_reset_dp_stats(struct subfacet *, struct dpif_flow_stats *); -static void subfacet_update_time(struct subfacet *, long long int used); static void subfacet_update_stats(struct subfacet *, const struct dpif_flow_stats *); -static void subfacet_make_actions(struct subfacet *, - const struct ofpbuf *packet, - struct ofpbuf *odp_actions); static int subfacet_install(struct subfacet *, - const struct nlattr *actions, size_t actions_len, - struct dpif_flow_stats *, enum slow_path_reason); + const struct ofpbuf *odp_actions, + struct dpif_flow_stats *); static void subfacet_uninstall(struct subfacet *); -static enum subfacet_path subfacet_want_path(enum slow_path_reason); - -/* An exact-match instantiation of an OpenFlow flow. +/* A unique, non-overlapping instantiation of an OpenFlow flow. * * A facet associates a "struct flow", which represents the Open vSwitch - * userspace idea of an exact-match flow, with one or more subfacets. Each - * subfacet tracks the datapath's idea of the exact-match flow equivalent to - * the facet. When the kernel module (or other dpif implementation) and Open - * vSwitch userspace agree on the definition of a flow key, there is exactly - * one subfacet per facet. If the dpif implementation supports more-specific - * flow matching than userspace, however, a facet can have more than one - * subfacet, each of which corresponds to some distinction in flow that - * userspace simply doesn't understand. + * userspace idea of an exact-match flow, with one or more subfacets. + * While the facet is created based on an exact-match flow, it is stored + * within the ofproto based on the wildcards that could be expressed + * based on the flow table and other configuration. (See the 'wc' + * description in "struct xlate_out" for more details.) + * + * Each subfacet tracks the datapath's idea of the flow equivalent to + * the facet. When the kernel module (or other dpif implementation) and + * Open vSwitch userspace agree on the definition of a flow key, there + * is exactly one subfacet per facet. If the dpif implementation + * supports more-specific flow matching than userspace, however, a facet + * can have more than one subfacet. Examples include the dpif + * implementation not supporting the same wildcards as userspace or some + * distinction in flow that userspace simply doesn't understand. * - * Flow expiration works in terms of subfacets, so a facet must have at least - * one subfacet or it will never expire, leaking memory. */ + * Flow expiration works in terms of subfacets, so a facet must have at + * least one subfacet or it will never expire, leaking memory. */ struct facet { /* Owners. */ struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ - struct list list_node; /* In owning rule's 'facets' list. */ - struct rule_dpif *rule; /* Owning rule. */ + struct ofproto_dpif *ofproto; /* Owned data. */ struct list subfacets; long long int used; /* Time last used; time created if not used. */ /* Key. */ - struct flow flow; + struct flow flow; /* Flow of the creating subfacet. */ + struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */ /* These statistics: * @@ -457,43 +247,38 @@ struct facet { struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */ - /* Properties of datapath actions. - * - * Every subfacet has its own actions because actions can differ slightly - * between splintered and non-splintered subfacets due to the VLAN tag - * being initially different (present vs. absent). All of them have these - * properties in common so we just store one copy of them here. */ - bool has_learn; /* Actions include NXAST_LEARN? */ - bool has_normal; /* Actions output to OFPP_NORMAL? */ - bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */ - tag_type tags; /* Tags that would require revalidation. */ - mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */ + struct xlate_out xout; + bool fail_open; /* Facet matched the fail open rule. */ /* Storage for a single subfacet, to reduce malloc() time and space * overhead. (A facet always has at least one subfacet and in the common - * 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 *, - const struct flow *, uint32_t hash); +static struct facet *facet_create(const struct flow_miss *, struct rule_dpif *, + struct xlate_out *, + struct dpif_flow_stats *); static void facet_remove(struct facet *); static void facet_free(struct facet *); -static struct facet *facet_find(struct ofproto_dpif *, - const struct flow *, uint32_t hash); +static struct facet *facet_find(struct ofproto_dpif *, const struct flow *); static struct facet *facet_lookup_valid(struct ofproto_dpif *, - const struct flow *, uint32_t hash); -static void facet_revalidate(struct facet *); + const struct flow *); +static bool facet_revalidate(struct facet *); static bool facet_check_consistency(struct facet *); static void facet_flush_stats(struct facet *); -static void facet_update_time(struct facet *, long long int used); static void facet_reset_counters(struct facet *); -static void facet_push_stats(struct facet *); +static void facet_push_stats(struct facet *, bool may_learn); static void facet_learn(struct facet *); static void facet_account(struct facet *); +static void push_all_stats(void); static bool facet_is_controller_flow(struct facet *); @@ -501,14 +286,16 @@ struct ofport_dpif { struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */ struct ofport up; - uint32_t odp_port; + odp_port_t odp_port; struct ofbundle *bundle; /* Bundle that contains this port, if any. */ struct list bundle_node; /* In struct ofbundle's "ports" list. */ struct cfm *cfm; /* Connectivity Fault Management, if any. */ + struct bfd *bfd; /* BFD, if any. */ tag_type tag; /* Tag associated with this port. */ - uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */ bool may_enable; /* May be enabled in bonds. */ + bool is_tunnel; /* This port is a tunnel. */ long long int carrier_seq; /* Carrier status changes. */ + struct ofport_dpif *peer; /* Peer if patch port. */ /* Spanning tree. */ struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */ @@ -523,7 +310,7 @@ struct ofport_dpif { * drivers in old versions of Linux that do not properly support VLANs when * VLAN devices are not used. When broken device drivers are no longer in * widespread use, we will delete these interfaces. */ - uint16_t realdev_ofp_port; + ofp_port_t realdev_ofp_port; int vlandev_vid; }; @@ -546,34 +333,35 @@ struct priority_to_dscp { struct vlan_splinter { struct hmap_node realdev_vid_node; struct hmap_node vlandev_node; - uint16_t realdev_ofp_port; - uint16_t vlandev_ofp_port; + ofp_port_t realdev_ofp_port; + ofp_port_t vlandev_ofp_port; int vid; }; -static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *, - uint32_t realdev, ovs_be16 vlan_tci); 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 void vsp_add(struct ofport_dpif *, ofp_port_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 odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *, + ofp_port_t); + +static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *, + odp_port_t); static struct ofport_dpif * ofport_dpif_cast(const struct ofport *ofport) { - assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class); return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL; } static void port_run(struct ofport_dpif *); static void port_run_fast(struct ofport_dpif *); static void port_wait(struct ofport_dpif *); +static int set_bfd(struct ofport *, const struct smap *); static int set_cfm(struct ofport *, const struct cfm_settings *); static void ofport_clear_priorities(struct ofport_dpif *); +static void ofport_update_peer(struct ofport_dpif *); +static void run_fast_rl(void); struct dpif_completion { struct list list_node; @@ -602,16 +390,34 @@ struct table_dpif { enum revalidate_reason { REV_RECONFIGURE = 1, /* Switch configuration changed. */ REV_STP, /* Spanning tree protocol port status change. */ + REV_BOND, /* Bonding changed. */ REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/ REV_FLOW_TABLE, /* Flow table changed. */ + REV_MAC_LEARNING, /* Mac learning changed. */ REV_INCONSISTENCY /* Facet self-check failed. */ }; COVERAGE_DEFINE(rev_reconfigure); COVERAGE_DEFINE(rev_stp); +COVERAGE_DEFINE(rev_bond); COVERAGE_DEFINE(rev_port_toggled); COVERAGE_DEFINE(rev_flow_table); +COVERAGE_DEFINE(rev_mac_learning); COVERAGE_DEFINE(rev_inconsistency); +/* Drop keys are odp flow keys which have drop flows installed in the kernel. + * These are datapath flows which have no associated ofproto, if they did we + * would use facets. */ +struct drop_key { + struct hmap_node hmap_node; + struct nlattr *key; + size_t key_len; +}; + +struct avg_subfacet_rates { + double add_rate; /* Moving average of new flows created per minute. */ + double del_rate; /* Moving average of flows deleted per minute. */ +}; + /* All datapaths of a given type share a single dpif backer instance. */ struct dpif_backer { char *type; @@ -619,13 +425,53 @@ struct dpif_backer { struct dpif *dpif; struct timer next_expiration; struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */ + + struct simap tnl_backers; /* Set of dpif ports backing tunnels. */ + + /* Facet revalidation flags applying to facets which use this backer. */ + enum revalidate_reason need_revalidate; /* Revalidate every facet. */ + struct tag_set revalidate_set; /* Revalidate only matching facets. */ + + struct hmap drop_keys; /* Set of dropped odp keys. */ + bool recv_set_enable; /* Enables or disables receiving packets. */ + + struct hmap subfacets; + struct governor *governor; + + /* 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 max_n_subfacet; /* Maximum number of flows */ + unsigned avg_n_subfacet; /* Average number of flows. */ + long long int avg_subfacet_life; /* Average life span of subfacets. */ + + /* The average number of subfacets... */ + struct avg_subfacet_rates hourly; /* ...over the last hour. */ + struct avg_subfacet_rates daily; /* ...over the last day. */ + struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */ + long long int last_minute; /* Last time 'hourly' was updated. */ + + /* Number of subfacets added or deleted since 'last_minute'. */ + unsigned subfacet_add_count; + unsigned subfacet_del_count; + + /* Number of subfacets added or deleted from 'created' to 'last_minute.' */ + unsigned long long int total_subfacet_add_count; + unsigned long long int total_subfacet_del_count; }; /* 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); +odp_port_to_ofport(const struct dpif_backer *, odp_port_t odp_port); +static void update_moving_averages(struct dpif_backer *backer); struct ofproto_dpif { struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */ @@ -635,33 +481,27 @@ struct ofproto_dpif { /* Special OpenFlow rules. */ struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */ struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */ - - /* Statistics. */ - uint64_t n_matches; + struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */ /* 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; + struct mbridge *mbridge; /* Facets. */ - struct hmap facets; - struct hmap subfacets; - struct governor *governor; + struct classifier facets; /* Contains 'struct facet's. */ + 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; - bool has_bundle_action; /* True when the first bundle action appears. */ struct netdev_stats stats; /* To account packets generated and consumed in * userspace. */ @@ -674,39 +514,39 @@ struct ofproto_dpif { struct hmap vlandev_map; /* vlandev -> (realdev,vid). */ /* Ports. */ - struct sset ports; /* Set of port names. */ + struct sset ports; /* Set of standard port names. */ + struct sset ghost_ports; /* Ports with no datapath port. */ struct sset port_poll_set; /* Queued names for port_poll() reply. */ int port_poll_errno; /* Last errno for port_poll() reply. */ + + /* Per ofproto's dpif stats. */ + uint64_t n_hit; + uint64_t n_missed; }; /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only * for debugging the asynchronous flow_mod implementation.) */ static bool clogged; +/* By default, flows in the datapath are wildcarded (megaflows). They + * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */ +static bool enable_megaflows = true; + /* All existing ofproto_dpif instances, indexed by ->up.name. */ static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs); static void ofproto_dpif_unixctl_init(void); -static struct ofproto_dpif * +static inline 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); } -static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *, - uint16_t ofp_port); -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 *); +static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto, + ofp_port_t ofp_port); -/* Packet processing. */ -static void update_learning_table(struct ofproto_dpif *, - const struct flow *, int vlan, - struct ofbundle *); /* Upcalls. */ #define FLOW_MISS_MAX_BATCH 50 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch); @@ -719,16 +559,26 @@ static void send_netflow_active_timeouts(struct ofproto_dpif *); /* Utilities. */ 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 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); + +int +ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto, + struct ofputil_flow_mod *fm) +{ + return ofproto_flow_mod(&ofproto->up, fm); +} + +void +ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto, + struct ofputil_packet_in *pin) +{ + connmgr_send_packet_in(ofproto->up.connmgr, pin); +} /* Factory functions. */ @@ -794,12 +644,31 @@ port_open_type(const char *datapath_type, const char *port_type) /* Type functions. */ +static void process_dpif_port_changes(struct dpif_backer *); +static void process_dpif_all_ports_changed(struct dpif_backer *); +static void process_dpif_port_change(struct dpif_backer *, + const char *devname); +static void process_dpif_port_error(struct dpif_backer *, int error); + +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) { @@ -810,75 +679,328 @@ type_run(const char *type) dpif_run(backer->dpif); - if (timer_expired(&backer->next_expiration)) { - int delay = expire(backer); - timer_set_duration(&backer->next_expiration, delay); + /* 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(); } - /* Check for port changes in the dpif. */ - while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) { - struct ofproto_dpif *ofproto = NULL; - struct dpif_port port; + /* If vswitchd started with other_config:flow_restore_wait set as "true", + * and the configuration has now changed to "false", enable receiving + * packets from the datapath. */ + if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) { + int error; + + backer->recv_set_enable = true; - /* Don't report on the datapath's device. */ - if (!strcmp(devname, dpif_base_name(backer->dpif))) { - continue; + error = dpif_recv_set(backer->dpif, backer->recv_set_enable); + if (error) { + VLOG_ERR("Failed to enable receiving packets in dpif."); + return error; } + dpif_flow_flush(backer->dpif); + backer->need_revalidate = REV_RECONFIGURE; + } - HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, - &all_ofproto_dpifs) { - if (sset_contains(&ofproto->ports, devname)) { - break; + 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) { + char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; + const char *dp_port; + + if (!iter->is_tunnel) { + continue; + } + + dp_port = netdev_vport_get_dpif_port(iter->up.netdev, + namebuf, sizeof namebuf); + 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) { + odp_port_t odp_port = ODPP_NONE; + + if (!dpif_port_add(backer->dpif, iter->up.netdev, + &odp_port)) { + simap_put(&backer->tnl_backers, dp_port, + odp_to_u32(odp_port)); + node = simap_find(&backer->tnl_backers, dp_port); + } + } + } + + iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE; + if (tnl_port_reconfigure(iter, iter->up.netdev, + iter->odp_port)) { + backer->need_revalidate = REV_RECONFIGURE; + } } } - 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; + SIMAP_FOR_EACH (node, &tmp_backers) { + dpif_port_del(backer->dpif, u32_to_odp(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_BOND: COVERAGE_INC(rev_bond); break; + case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break; + case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break; + case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); 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; + struct cls_cursor cursor; + + if (ofproto->backer != backer) { + continue; + } + + if (need_revalidate) { + struct ofport_dpif *ofport; + struct ofbundle *bundle; + + xlate_ofproto_set(ofproto, ofproto->up.name, ofproto->ml, + ofproto->mbridge, ofproto->sflow, + ofproto->ipfix, ofproto->up.frag_handling, + ofproto->up.forward_bpdu, + connmgr_has_in_band(ofproto->up.connmgr), + ofproto->netflow != NULL, + ofproto->stp != NULL); + + HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { + xlate_bundle_set(ofproto, bundle, bundle->name, + bundle->vlan_mode, bundle->vlan, + bundle->trunks, bundle->use_priority_tags, + bundle->bond, bundle->lacp, + bundle->floodable); + } + + HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { + xlate_ofport_set(ofproto, ofport->bundle, ofport, + ofport->up.ofp_port, ofport->odp_port, + ofport->up.netdev, ofport->cfm, + ofport->bfd, ofport->peer, + ofport->up.pp.config, ofport->stp_state, + ofport->is_tunnel, ofport->may_enable); + } + } + + cls_cursor_init(&cursor, &ofproto->facets, NULL); + CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) { + if (need_revalidate + || tag_set_intersects(&revalidate_set, facet->xout.tags)) { + facet_revalidate(facet); + run_fast_rl(); + } } - dpif_port_destroy(&port); - } 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); } + } - free(devname); + if (!backer->recv_set_enable) { + /* Wake up before a max of 1000ms. */ + timer_set_duration(&backer->next_expiration, 1000); + } else if (timer_expired(&backer->next_expiration)) { + int delay = expire(backer); + timer_set_duration(&backer->next_expiration, delay); } - if (error != EAGAIN) { - struct ofproto_dpif *ofproto; + process_dpif_port_changes(backer); - /* 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; - } + if (backer->governor) { + size_t n_subfacets; + + governor_run(backer->governor); + + /* If the governor has shrunk to its minimum size and the number of + * subfacets has dwindled, then drop the governor entirely. + * + * For hysteresis, the number of subfacets to drop the governor is + * smaller than the number needed to trigger its creation. */ + n_subfacets = hmap_count(&backer->subfacets); + if (n_subfacets * 4 < flow_eviction_threshold + && governor_is_idle(backer->governor)) { + governor_destroy(backer->governor); + backer->governor = NULL; } } return 0; } +/* Check for and handle port changes in 'backer''s dpif. */ +static void +process_dpif_port_changes(struct dpif_backer *backer) +{ + for (;;) { + char *devname; + int error; + + error = dpif_port_poll(backer->dpif, &devname); + switch (error) { + case EAGAIN: + return; + + case ENOBUFS: + process_dpif_all_ports_changed(backer); + break; + + case 0: + process_dpif_port_change(backer, devname); + free(devname); + break; + + default: + process_dpif_port_error(backer, error); + break; + } + } +} + +static void +process_dpif_all_ports_changed(struct dpif_backer *backer) +{ + struct ofproto_dpif *ofproto; + struct dpif_port dpif_port; + struct dpif_port_dump dump; + struct sset devnames; + const char *devname; + + sset_init(&devnames); + HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { + if (ofproto->backer == backer) { + struct ofport *ofport; + + HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) { + sset_add(&devnames, netdev_get_name(ofport->netdev)); + } + } + } + DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) { + sset_add(&devnames, dpif_port.name); + } + + SSET_FOR_EACH (devname, &devnames) { + process_dpif_port_change(backer, devname); + } + sset_destroy(&devnames); +} + +static void +process_dpif_port_change(struct dpif_backer *backer, const char *devname) +{ + struct ofproto_dpif *ofproto; + struct dpif_port port; + + /* Don't report on the datapath's device. */ + if (!strcmp(devname, dpif_base_name(backer->dpif))) { + return; + } + + HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, + &all_ofproto_dpifs) { + if (simap_contains(&ofproto->backer->tnl_backers, devname)) { + return; + } + } + + 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); + } else { + struct ofport_dpif *ofport; + + ofport = ofport_dpif_cast(shash_find_data( + &ofproto->up.port_by_name, devname)); + if (ofport + && ofport->odp_port != port.port_no + && !odp_port_to_ofport(backer, port.port_no)) + { + /* 'ofport''s datapath port number has changed from + * 'ofport->odp_port' to 'port.port_no'. Update our internal data + * structures to match. */ + hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node); + ofport->odp_port = port.port_no; + hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node, + hash_odp_port(port.port_no)); + backer->need_revalidate = REV_RECONFIGURE; + } + } + dpif_port_destroy(&port); +} + +/* Propagate 'error' to all ofprotos based on 'backer'. */ +static void +process_dpif_port_error(struct dpif_backer *backer, int error) +{ + struct ofproto_dpif *ofproto; + + 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; + } + } +} + static int -type_run_fast(const char *type) +dpif_backer_run_fast(struct dpif_backer *backer, int max_batch) { - struct dpif_backer *backer; unsigned int work; - backer = shash_find_data(&all_dpif_backers, type); - if (!backer) { - /* This is not necessarily a problem, since backers are only - * created on demand. */ + /* If recv_set_enable is false, we should not handle upcalls. */ + if (!backer->recv_set_enable) { return 0; } @@ -892,8 +1014,8 @@ type_run_fast(const char *type) * optimizations can make major improvements on some benchmarks and * presumably for real traffic as well. */ work = 0; - while (work < FLOW_MISS_MAX_BATCH) { - int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work); + while (work < max_batch) { + int retval = handle_upcalls(backer, max_batch - work); if (retval <= 0) { return -retval; } @@ -903,8 +1025,8 @@ type_run_fast(const char *type) return 0; } -static void -type_wait(const char *type) +static int +type_run_fast(const char *type) { struct dpif_backer *backer; @@ -912,54 +1034,118 @@ type_wait(const char *type) if (!backer) { /* This is not necessarily a problem, since backers are only * created on demand. */ - return; + return 0; } - timer_wait(&backer->next_expiration); -} - -/* Basic life-cycle. */ - -static int add_internal_flows(struct ofproto_dpif *); - -static struct ofproto * -alloc(void) -{ - struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto); - return &ofproto->up; + return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH); } static void -dealloc(struct ofproto *ofproto_) +run_fast_rl(void) { - struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - free(ofproto); -} + static long long int port_rl = LLONG_MIN; + static unsigned int backer_rl = 0; -static void -close_dpif_backer(struct dpif_backer *backer) + 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; + } + + if (backer->governor) { + governor_wait(backer->governor); + } + + timer_wait(&backer->next_expiration); +} + +/* Basic life-cycle. */ + +static int add_internal_flows(struct ofproto_dpif *); + +static struct ofproto * +alloc(void) +{ + struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto); + return &ofproto->up; +} + +static void +dealloc(struct ofproto *ofproto_) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + free(ofproto); +} + +static void +close_dpif_backer(struct dpif_backer *backer) { struct shash_node *node; - assert(backer->refcount > 0); + ovs_assert(backer->refcount > 0); if (--backer->refcount) { return; } + drop_key_clear(backer); + hmap_destroy(&backer->drop_keys); + + simap_destroy(&backer->tnl_backers); hmap_destroy(&backer->odp_to_ofport_map); node = shash_find(&all_dpif_backers, backer->type); free(backer->type); shash_delete(&all_dpif_backers, node); dpif_close(backer->dpif); + ovs_assert(hmap_is_empty(&backer->subfacets)); + hmap_destroy(&backer->subfacets); + governor_destroy(backer->governor); + free(backer); } /* Datapath port slated for removal from datapath. */ struct odp_garbage { struct list list_node; - uint32_t odp_port; + odp_port_t odp_port; }; static int @@ -1012,17 +1198,27 @@ open_dpif_backer(const char *type, struct dpif_backer **backerp) free(backer_name); if (error) { VLOG_ERR("failed to open datapath of type %s: %s", type, - strerror(error)); + ovs_strerror(error)); + free(backer); return error; } backer->type = xstrdup(type); + backer->governor = NULL; backer->refcount = 1; hmap_init(&backer->odp_to_ofport_map); + hmap_init(&backer->drop_keys); + hmap_init(&backer->subfacets); timer_set_duration(&backer->next_expiration, 1000); + backer->need_revalidate = 0; + simap_init(&backer->tnl_backers); + tag_set_init(&backer->revalidate_set); + backer->recv_set_enable = !ofproto_get_flow_restore_wait(); *backerp = backer; - dpif_flow_flush(backer->dpif); + if (backer->recv_set_enable) { + dpif_flow_flush(backer->dpif); + } /* Loop through the ports already on the datapath and remove any * that we don't need anymore. */ @@ -1046,14 +1242,27 @@ open_dpif_backer(const char *type, struct dpif_backer **backerp) shash_add(&all_dpif_backers, type, backer); - error = dpif_recv_set(backer->dpif, true); + error = dpif_recv_set(backer->dpif, backer->recv_set_enable); if (error) { VLOG_ERR("failed to listen on datapath of type %s: %s", - type, strerror(error)); + type, ovs_strerror(error)); close_dpif_backer(backer); return error; } + backer->max_n_subfacet = 0; + backer->created = time_msec(); + backer->last_minute = backer->created; + memset(&backer->hourly, 0, sizeof backer->hourly); + memset(&backer->daily, 0, sizeof backer->daily); + memset(&backer->lifetime, 0, sizeof backer->lifetime); + backer->subfacet_add_count = 0; + backer->subfacet_del_count = 0; + backer->total_subfacet_add_count = 0; + backer->total_subfacet_del_count = 0; + backer->avg_n_subfacet = 0; + backer->avg_subfacet_life = 0; + return error; } @@ -1062,7 +1271,7 @@ construct(struct ofproto *ofproto_) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); struct shash_node *node, *next; - int max_ports; + odp_port_t max_ports; int error; int i; @@ -1072,23 +1281,20 @@ construct(struct ofproto *ofproto_) } max_ports = dpif_get_max_ports(ofproto->backer->dpif); - ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX)); - - ofproto->n_matches = 0; + ofproto_init_max_ports(ofproto_, u16_to_ofp(MIN(odp_to_u32(max_ports), + ofp_to_u16(OFPP_MAX)))); ofproto->netflow = NULL; ofproto->sflow = NULL; + ofproto->ipfix = NULL; ofproto->stp = NULL; hmap_init(&ofproto->bundles); ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME); - for (i = 0; i < MAX_MIRRORS; i++) { - ofproto->mirrors[i] = NULL; - } + ofproto->mbridge = mbridge_create(); ofproto->has_bonded_bundles = false; - hmap_init(&ofproto->facets); - hmap_init(&ofproto->subfacets); - ofproto->governor = NULL; + classifier_init(&ofproto->facets); + ofproto->consistency_rl = LLONG_MIN; for (i = 0; i < N_TABLES; i++) { struct table_dpif *table = &ofproto->tables[i]; @@ -1097,25 +1303,21 @@ 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); ofproto_dpif_unixctl_init(); - ofproto->has_mirrors = false; - ofproto->has_bundle_action = false; - 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) { - const struct iface_hint *iface_hint = node->data; + struct iface_hint *iface_hint = node->data; if (!strcmp(iface_hint->br_name, ofproto->up.name)) { /* Check if the datapath already has this port. */ @@ -1125,6 +1327,7 @@ construct(struct ofproto *ofproto_) free(iface_hint->br_name); free(iface_hint->br_type); + free(iface_hint); shash_delete(&init_ofp_ports, node); } } @@ -1137,6 +1340,9 @@ 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; + return error; } @@ -1153,6 +1359,7 @@ add_internal_flow(struct ofproto_dpif *ofproto, int id, fm.new_cookie = htonll(0); fm.cookie = htonll(0); fm.cookie_mask = htonll(0); + fm.modify_cookie = false; fm.table_id = TBL_INTERNAL; fm.command = OFPFC_ADD; fm.idle_timeout = 0; @@ -1170,8 +1377,9 @@ add_internal_flow(struct ofproto_dpif *ofproto, int id, return error; } - *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL); - assert(*rulep != NULL); + *rulep = rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, + TBL_INTERNAL); + ovs_assert(*rulep != NULL); return 0; } @@ -1202,6 +1410,12 @@ add_internal_flows(struct ofproto_dpif *ofproto) ofpbuf_clear(&ofpacts); error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->no_packet_in_rule); + if (error) { + return error; + } + + error = add_internal_flow(ofproto, id++, &ofpacts, + &ofproto->drop_frags_rule); return error; } @@ -1223,7 +1437,9 @@ destruct(struct ofproto *ofproto_) struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); struct rule_dpif *rule, *next_rule; struct oftable *table; - int i; + + ofproto->backer->need_revalidate = REV_RECONFIGURE; + xlate_remove_ofproto(ofproto); hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node); complete_operations(ofproto); @@ -1237,23 +1453,20 @@ destruct(struct ofproto *ofproto_) } } - for (i = 0; i < MAX_MIRRORS; i++) { - mirror_destroy(ofproto->mirrors[i]); - } + mbridge_unref(ofproto->mbridge); netflow_destroy(ofproto->netflow); - dpif_sflow_destroy(ofproto->sflow); + dpif_sflow_unref(ofproto->sflow); hmap_destroy(&ofproto->bundles); - mac_learning_destroy(ofproto->ml); + mac_learning_unref(ofproto->ml); - hmap_destroy(&ofproto->facets); - hmap_destroy(&ofproto->subfacets); - governor_destroy(ofproto->governor); + classifier_destroy(&ofproto->facets); hmap_destroy(&ofproto->vlandev_map); hmap_destroy(&ofproto->realdev_vid_map); sset_destroy(&ofproto->ports); + sset_destroy(&ofproto->ghost_ports); sset_destroy(&ofproto->port_poll_set); close_dpif_backer(ofproto->backer); @@ -1265,6 +1478,12 @@ run_fast(struct ofproto *ofproto_) struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); struct ofport_dpif *ofport; + /* Do not perform any periodic activity required by 'ofproto' while + * waiting for flow restore to complete. */ + if (ofproto_get_flow_restore_wait()) { + return 0; + } + HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { port_run_fast(ofport); } @@ -1284,6 +1503,19 @@ run(struct ofproto *ofproto_) complete_operations(ofproto); } + if (mbridge_need_revalidate(ofproto->mbridge)) { + ofproto->backer->need_revalidate = REV_RECONFIGURE; + ovs_rwlock_wrlock(&ofproto->ml->rwlock); + mac_learning_flush(ofproto->ml); + ovs_rwlock_unlock(&ofproto->ml->rwlock); + } + + /* Do not perform any periodic activity below required by 'ofproto' while + * waiting for flow restore to complete. */ + if (ofproto_get_flow_restore_wait()) { + return 0; + } + error = run_fast(ofproto_); if (error) { return error; @@ -1306,63 +1538,33 @@ 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); - } - } + ovs_rwlock_wrlock(&ofproto->ml->rwlock); + if (mac_learning_run(ofproto->ml)) { + ofproto->backer->need_revalidate = REV_MAC_LEARNING; } + ovs_rwlock_unlock(&ofproto->ml->rwlock); /* 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 + && !classifier_is_empty(&ofproto->facets) + && !ofproto->backer->need_revalidate) { + struct cls_table *table; + struct cls_rule *cr; struct facet *facet; - facet = CONTAINER_OF(hmap_random_node(&ofproto->facets), - struct facet, hmap_node); - if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) { - if (!facet_check_consistency(facet)) { - ofproto->need_revalidate = REV_INCONSISTENCY; - } - } - } - - if (ofproto->governor) { - size_t n_subfacets; + ofproto->consistency_rl = time_msec() + 250; - governor_run(ofproto->governor); + table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables), + struct cls_table, hmap_node); + cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule, + hmap_node); + facet = CONTAINER_OF(cr, struct facet, cr); - /* If the governor has shrunk to its minimum size and the number of - * subfacets has dwindled, then drop the governor entirely. - * - * For hysteresis, the number of subfacets to drop the governor is - * smaller than the number needed to trigger its creation. */ - n_subfacets = hmap_count(&ofproto->subfacets); - if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold - && governor_is_idle(ofproto->governor)) { - governor_destroy(ofproto->governor); - ofproto->governor = NULL; + if (!tag_set_intersects(&ofproto->backer->revalidate_set, + facet->xout.tags)) { + if (!facet_check_consistency(facet)) { + ofproto->backer->need_revalidate = REV_INCONSISTENCY; + } } } @@ -1380,12 +1582,16 @@ wait(struct ofproto *ofproto_) poll_immediate_wake(); } + if (ofproto_get_flow_restore_wait()) { + return; + } + dpif_wait(ofproto->backer->dpif); dpif_recv_wait(ofproto->backer->dpif); if (ofproto->sflow) { dpif_sflow_wait(ofproto->sflow); } - 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) { @@ -1397,25 +1603,32 @@ wait(struct ofproto *ofproto_) if (ofproto->netflow) { netflow_wait(ofproto->netflow); } + ovs_rwlock_rdlock(&ofproto->ml->rwlock); mac_learning_wait(ofproto->ml); + ovs_rwlock_unlock(&ofproto->ml->rwlock); 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(); } - if (ofproto->governor) { - governor_wait(ofproto->governor); - } } static void get_memory_usage(const struct ofproto *ofproto_, struct simap *usage) { const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + struct cls_cursor cursor; + size_t n_subfacets = 0; + struct facet *facet; - simap_increase(usage, "facets", hmap_count(&ofproto->facets)); - simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets)); + simap_increase(usage, "facets", classifier_count(&ofproto->facets)); + + cls_cursor_init(&cursor, &ofproto->facets, NULL); + CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { + n_subfacets += list_size(&facet->subfacets); + } + simap_increase(usage, "subfacets", n_subfacets); } static void @@ -1428,11 +1641,15 @@ flush(struct ofproto *ofproto_) n_batch = 0; HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, - &ofproto->subfacets) { + &ofproto->backer->subfacets) { + if (subfacet->facet->ofproto != ofproto) { + continue; + } + if (subfacet->path != SF_NOT_INSTALLED) { batch[n_batch++] = subfacet; if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { - subfacet_destroy_batch(ofproto, batch, n_batch); + subfacet_destroy_batch(ofproto->backer, batch, n_batch); n_batch = 0; } } else { @@ -1441,7 +1658,7 @@ flush(struct ofproto *ofproto_) } if (n_batch > 0) { - subfacet_destroy_batch(ofproto, batch, n_batch); + subfacet_destroy_batch(ofproto->backer, batch, n_batch); } } @@ -1469,13 +1686,19 @@ get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); struct dpif_dp_stats s; + uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags; + uint64_t n_lookup; strcpy(ots->name, "classifier"); dpif_get_dp_stats(ofproto->backer->dpif, &s); + rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes); + rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes); + rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes); - ots->lookup_count = htonll(s.n_hit + s.n_missed); - ots->matched_count = htonll(s.n_hit + ofproto->n_matches); + n_lookup = s.n_hit + s.n_missed - n_dropped_frags; + ots->lookup_count = htonll(n_lookup); + ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in); } static struct ofport * @@ -1497,23 +1720,40 @@ 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; + char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; struct dpif_port dpif_port; int error; - ofproto->need_revalidate = REV_RECONFIGURE; + ofproto->backer->need_revalidate = REV_RECONFIGURE; port->bundle = NULL; port->cfm = NULL; + port->bfd = NULL; port->tag = tag_create_random(); port->may_enable = true; port->stp_port = NULL; port->stp_state = STP_DISABLED; + port->is_tunnel = false; + port->peer = NULL; hmap_init(&port->priorities); port->realdev_ofp_port = 0; port->vlandev_vid = 0; - port->carrier_seq = netdev_get_carrier_resets(port->up.netdev); + port->carrier_seq = netdev_get_carrier_resets(netdev); + + if (netdev_vport_is_patch(netdev)) { + /* By bailing out here, we don't submit the port to the sFlow module + * to be considered for counter polling export. This is correct + * because the patch port represents an interface that sFlow considers + * to be "internal" to the switch as a whole, and therefore not an + * candidate for counter polling. */ + port->odp_port = ODPP_NONE; + ofport_update_peer(port); + return 0; + } error = dpif_port_query_by_name(ofproto->backer->dpif, - netdev_get_name(port->up.netdev), + netdev_vport_get_dpif_port(netdev, namebuf, + sizeof namebuf), &dpif_port); if (error) { return error; @@ -1521,16 +1761,23 @@ port_construct(struct ofport *port_) port->odp_port = dpif_port.port_no; - /* Sanity-check that a mapping doesn't already exist. This - * shouldn't happen. */ - if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) { - VLOG_ERR("port %s already has an OpenFlow port number\n", - dpif_port.name); - return EBUSY; - } + if (netdev_get_tunnel_config(netdev)) { + tnl_port_add(port, port->up.netdev, port->odp_port); + port->is_tunnel = true; + } else { + /* Sanity-check that a mapping doesn't already exist. This + * shouldn't happen for non-tunnel ports. */ + if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) { + VLOG_ERR("port %s already has an OpenFlow port number", + dpif_port.name); + dpif_port_destroy(&dpif_port); + return EBUSY; + } - hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node, - hash_int(port->odp_port, 0)); + hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node, + hash_odp_port(port->odp_port)); + } + dpif_port_destroy(&dpif_port); if (ofproto->sflow) { dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port); @@ -1545,20 +1792,39 @@ 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 *devname = netdev_get_name(port->up.netdev); + char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; + const char *dp_port_name; + + ofproto->backer->need_revalidate = REV_RECONFIGURE; + xlate_ofport_remove(port); - if (dpif_port_exists(ofproto->backer->dpif, devname)) { + dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf, + sizeof namebuf); + if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) { /* The underlying device is still there, so delete it. This * happens when the ofproto is being destroyed, since the caller * assumes that removal of attached ports will happen as part of * destruction. */ - dpif_port_del(ofproto->backer->dpif, port->odp_port); + if (!port->is_tunnel) { + dpif_port_del(ofproto->backer->dpif, port->odp_port); + } + } + + if (port->peer) { + port->peer->peer = NULL; + port->peer = NULL; + } + + if (port->odp_port != ODPP_NONE && !port->is_tunnel) { + hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node); } + tnl_port_del(port); sset_find_and_delete(&ofproto->ports, devname); - hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node); - ofproto->need_revalidate = REV_RECONFIGURE; + sset_find_and_delete(&ofproto->ghost_ports, devname); bundle_remove(port_); set_cfm(port_, NULL); + set_bfd(port_, NULL); if (ofproto->sflow) { dpif_sflow_del_port(ofproto->sflow, port->odp_port); } @@ -1575,6 +1841,18 @@ port_modified(struct ofport *port_) if (port->bundle && port->bundle->bond) { bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev); } + + if (port->cfm) { + cfm_set_netdev(port->cfm, port->up.netdev); + } + + if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev, + port->odp_port)) { + ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate = + REV_RECONFIGURE; + } + + ofport_update_peer(port); } static void @@ -1587,7 +1865,7 @@ 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); @@ -1610,19 +1888,45 @@ set_sflow(struct ofproto *ofproto_, HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port); } - ofproto->need_revalidate = REV_RECONFIGURE; + ofproto->backer->need_revalidate = REV_RECONFIGURE; } dpif_sflow_set_options(ds, sflow_options); } else { if (ds) { - dpif_sflow_destroy(ds); - ofproto->need_revalidate = REV_RECONFIGURE; + dpif_sflow_unref(ds); + 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_unref(di); + ofproto->ipfix = NULL; + } + } + return 0; +} + static int set_cfm(struct ofport *ofport_, const struct cfm_settings *s) { @@ -1636,8 +1940,8 @@ set_cfm(struct ofport *ofport_, const struct cfm_settings *s) struct ofproto_dpif *ofproto; ofproto = ofproto_dpif_cast(ofport->up.ofproto); - ofproto->need_revalidate = REV_RECONFIGURE; - ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev)); + ofproto->backer->need_revalidate = REV_RECONFIGURE; + ofport->cfm = cfm_create(ofport->up.netdev); } if (cfm_configure(ofport->cfm, s)) { @@ -1646,48 +1950,56 @@ set_cfm(struct ofport *ofport_, const struct cfm_settings *s) error = EINVAL; } - cfm_destroy(ofport->cfm); + cfm_unref(ofport->cfm); ofport->cfm = NULL; return error; } -static int -get_cfm_fault(const struct ofport *ofport_) +static bool +get_cfm_status(const struct ofport *ofport_, + struct ofproto_cfm_status *status) { struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); - return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1; + if (ofport->cfm) { + status->faults = cfm_get_fault(ofport->cfm); + status->remote_opstate = cfm_get_opup(ofport->cfm); + status->health = cfm_get_health(ofport->cfm); + cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps); + return true; + } else { + return false; + } } static int -get_cfm_opup(const struct ofport *ofport_) +set_bfd(struct ofport *ofport_, const struct smap *cfg) { + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); + struct bfd *old; + + old = ofport->bfd; + ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg); + if (ofport->bfd != old) { + ofproto->backer->need_revalidate = REV_RECONFIGURE; + } - return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1; + return 0; } static int -get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps, - size_t *n_rmps) +get_bfd_status(struct ofport *ofport_, struct smap *smap) { struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); - if (ofport->cfm) { - cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps); + if (ofport->bfd) { + bfd_get_status(ofport->bfd, smap); return 0; } else { - return -1; + return ENOENT; } } - -static int -get_cfm_health(const struct ofport *ofport_) -{ - struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); - - return ofport->cfm ? cfm_get_health(ofport->cfm) : -1; -} /* Spanning Tree. */ @@ -1724,7 +2036,7 @@ 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) { @@ -1746,7 +2058,7 @@ set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s) set_stp_port(ofport, NULL); } - stp_destroy(ofproto->stp); + stp_unref(ofproto->stp); ofproto->stp = NULL; } @@ -1792,12 +2104,14 @@ 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); + ovs_rwlock_wrlock(&ofproto->ml->rwlock); + mac_learning_flush(ofproto->ml); + ovs_rwlock_unlock(&ofproto->ml->rwlock); } 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(); @@ -1897,7 +2211,9 @@ stp_run(struct ofproto_dpif *ofproto) } if (stp_check_and_reset_fdb_flush(ofproto->stp)) { - mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); + ovs_rwlock_wrlock(&ofproto->ml->rwlock); + mac_learning_flush(ofproto->ml); + ovs_rwlock_unlock(&ofproto->ml->rwlock); } } } @@ -1910,14 +2226,16 @@ stp_wait(struct ofproto_dpif *ofproto) } } -/* Returns true if STP should process 'flow'. */ -static bool -stp_should_process_flow(const struct flow *flow) +/* Returns true if STP should process 'flow'. Sets fields in 'wc' that + * were used to make the determination.*/ +bool +stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc) { + memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst); return eth_addr_equals(flow->dl_dst, eth_addr_stp); } -static void +void stp_process_packet(const struct ofport_dpif *ofport, const struct ofpbuf *packet) { @@ -1941,6 +2259,13 @@ stp_process_packet(const struct ofport_dpif *ofport, } } +int +ofproto_dpif_queue_to_priority(const struct ofproto_dpif *ofproto, + uint32_t queue_id, uint32_t *priority) +{ + return dpif_queue_to_priority(ofproto->backer->dpif, queue_id, priority); +} + static struct priority_to_dscp * get_priority(const struct ofport_dpif *ofport, uint32_t priority) { @@ -1956,6 +2281,15 @@ get_priority(const struct ofport_dpif *ofport, uint32_t priority) return NULL; } +bool +ofproto_dpif_dscp_from_priority(const struct ofport_dpif *ofport, + uint32_t priority, uint8_t *dscp) +{ + struct priority_to_dscp *pdscp = get_priority(ofport, priority); + *dscp = pdscp ? pdscp->dscp : 0; + return pdscp != NULL; +} + static void ofport_clear_priorities(struct ofport_dpif *ofport) { @@ -1995,12 +2329,12 @@ 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)); @@ -2008,7 +2342,7 @@ 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); @@ -2035,7 +2369,8 @@ 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; + ovs_rwlock_wrlock(&ml->rwlock); LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { if (mac->port.p == bundle) { if (all_ofprotos) { @@ -2045,12 +2380,12 @@ bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos) if (o != ofproto) { struct mac_entry *e; - e = mac_learning_lookup(o->ml, mac->mac, mac->vlan, - NULL); + ovs_rwlock_wrlock(&o->ml->rwlock); + e = mac_learning_lookup(o->ml, mac->mac, mac->vlan); if (e) { - tag_set_add(&o->revalidate_set, e->tag); mac_learning_expire(o->ml, e); } + ovs_rwlock_unlock(&o->ml->rwlock); } } } @@ -2058,6 +2393,7 @@ bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos) mac_learning_expire(ml, mac); } } + ovs_rwlock_unlock(&ml->rwlock); } static struct ofbundle * @@ -2074,24 +2410,6 @@ bundle_lookup(const struct ofproto_dpif *ofproto, void *aux) return NULL; } -/* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the - * ones that are found to 'bundles'. */ -static void -bundle_lookup_multiple(struct ofproto_dpif *ofproto, - void **auxes, size_t n_auxes, - struct hmapx *bundles) -{ - size_t i; - - hmapx_init(bundles); - for (i = 0; i < n_auxes; i++) { - struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]); - if (bundle) { - hmapx_add(bundles, bundle); - } - } -} - static void bundle_update(struct ofbundle *bundle) { @@ -2112,7 +2430,7 @@ 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; @@ -2128,9 +2446,8 @@ bundle_del_port(struct ofport_dpif *port) } static bool -bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port, - struct lacp_slave_settings *lacp, - uint32_t bond_stable_id) +bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port, + struct lacp_slave_settings *lacp) { struct ofport_dpif *port; @@ -2140,7 +2457,7 @@ 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); } @@ -2153,12 +2470,10 @@ 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; } @@ -2167,24 +2482,15 @@ bundle_destroy(struct ofbundle *bundle) { struct ofproto_dpif *ofproto; struct ofport_dpif *port, *next_port; - int i; if (!bundle) { return; } ofproto = bundle->ofproto; - for (i = 0; i < MAX_MIRRORS; i++) { - struct ofmirror *m = ofproto->mirrors[i]; - if (m) { - if (m->out == 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; - } - } - } + mbridge_unregister_bundle(ofproto->mbridge, bundle->aux); + + xlate_bundle_remove(bundle); LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { bundle_del_port(port); @@ -2194,8 +2500,8 @@ bundle_destroy(struct ofbundle *bundle) hmap_remove(&ofproto->bundles, &bundle->hmap_node); free(bundle->name); free(bundle->trunks); - lacp_destroy(bundle->lacp); - bond_destroy(bundle->bond); + lacp_unref(bundle->lacp); + bond_unref(bundle->bond); free(bundle); } @@ -2217,8 +2523,8 @@ 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) { @@ -2239,10 +2545,7 @@ bundle_set(struct ofproto *ofproto_, void *aux, bundle->bond = NULL; bundle->floodable = true; - - bundle->src_mirrors = 0; - bundle->dst_mirrors = 0; - bundle->mirror_out = 0; + mbridge_register_bundle(ofproto->mbridge, bundle); } if (!bundle->name || strcmp(s->name, bundle->name)) { @@ -2253,12 +2556,12 @@ 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); } else { - lacp_destroy(bundle->lacp); + lacp_unref(bundle->lacp); bundle->lacp = NULL; } @@ -2266,8 +2569,7 @@ 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; } } @@ -2285,7 +2587,7 @@ 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); @@ -2359,19 +2661,18 @@ 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); + bond_unref(bundle->bond); bundle->bond = NULL; } @@ -2395,7 +2696,7 @@ bundle_remove(struct ofport *port_) if (list_is_empty(&bundle->ports)) { bundle_destroy(bundle); } else if (list_is_short(&bundle->ports)) { - bond_destroy(bundle->bond); + bond_unref(bundle->bond); bundle->bond = NULL; } } @@ -2424,7 +2725,7 @@ send_pdu_cb(void *port_, const void *pdu, size_t pdu_size) } else { VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " "%s (%s)", port->bundle->name, - netdev_get_name(port->up.netdev), strerror(error)); + netdev_get_name(port->up.netdev), ovs_strerror(error)); } } @@ -2436,6 +2737,7 @@ bundle_send_learning_packets(struct ofbundle *bundle) struct mac_entry *e; error = n_packets = n_errors = 0; + ovs_rwlock_rdlock(&ofproto->ml->rwlock); LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { if (e->port.p != bundle) { struct ofpbuf *learning_packet; @@ -2458,12 +2760,13 @@ bundle_send_learning_packets(struct ofbundle *bundle) n_packets++; } } + ovs_rwlock_unlock(&ofproto->ml->rwlock); if (n_errors) { static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " "packets, last error was: %s", - bundle->name, n_errors, n_packets, strerror(error)); + bundle->name, n_errors, n_packets, ovs_strerror(error)); } else { VLOG_DBG("bond %s: sent %d gratuitous learning packets", bundle->name, n_packets); @@ -2483,8 +2786,10 @@ bundle_run(struct ofbundle *bundle) bond_slave_set_may_enable(bundle->bond, port, port->may_enable); } - bond_run(bundle->bond, &bundle->ofproto->revalidate_set, - lacp_status(bundle->lacp)); + if (bond_run(bundle->bond, lacp_status(bundle->lacp))) { + bundle->ofproto->backer->need_revalidate = REV_BOND; + } + if (bond_should_send_learning_packets(bundle->bond)) { bundle_send_learning_packets(bundle); } @@ -2505,292 +2810,159 @@ bundle_wait(struct ofbundle *bundle) /* Mirrors. */ static int -mirror_scan(struct ofproto_dpif *ofproto) +mirror_set__(struct ofproto *ofproto_, void *aux, + const struct ofproto_mirror_settings *s) { - int idx; + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + struct ofbundle **srcs, **dsts; + int error; + size_t i; - for (idx = 0; idx < MAX_MIRRORS; idx++) { - if (!ofproto->mirrors[idx]) { - return idx; - } + if (!s) { + mirror_destroy(ofproto->mbridge, aux); + return 0; } - return -1; -} -static struct ofmirror * -mirror_lookup(struct ofproto_dpif *ofproto, void *aux) -{ - int i; + srcs = xmalloc(s->n_srcs * sizeof *srcs); + dsts = xmalloc(s->n_dsts * sizeof *dsts); - for (i = 0; i < MAX_MIRRORS; i++) { - struct ofmirror *mirror = ofproto->mirrors[i]; - if (mirror && mirror->aux == aux) { - return mirror; - } + for (i = 0; i < s->n_srcs; i++) { + srcs[i] = bundle_lookup(ofproto, s->srcs[i]); } - return NULL; -} - -/* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */ -static void -mirror_update_dups(struct ofproto_dpif *ofproto) -{ - int i; - - for (i = 0; i < MAX_MIRRORS; i++) { - struct ofmirror *m = ofproto->mirrors[i]; - - if (m) { - m->dup_mirrors = MIRROR_MASK_C(1) << i; - } + for (i = 0; i < s->n_dsts; i++) { + dsts[i] = bundle_lookup(ofproto, s->dsts[i]); } - for (i = 0; i < MAX_MIRRORS; i++) { - struct ofmirror *m1 = ofproto->mirrors[i]; - int j; - - if (!m1) { - continue; - } - - for (j = i + 1; j < MAX_MIRRORS; j++) { - struct ofmirror *m2 = ofproto->mirrors[j]; + error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts, + s->n_dsts, s->src_vlans, + bundle_lookup(ofproto, s->out_bundle), s->out_vlan); + free(srcs); + free(dsts); + return error; +} - if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) { - m1->dup_mirrors |= MIRROR_MASK_C(1) << j; - m2->dup_mirrors |= m1->dup_mirrors; - } - } - } +static int +mirror_get_stats__(struct ofproto *ofproto, void *aux, + uint64_t *packets, uint64_t *bytes) +{ + push_all_stats(); + return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets, + bytes); } static int -mirror_set(struct ofproto *ofproto_, void *aux, - const struct ofproto_mirror_settings *s) +set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - mirror_mask_t mirror_bit; - struct ofbundle *bundle; - struct ofmirror *mirror; - struct ofbundle *out; - struct hmapx srcs; /* Contains "struct ofbundle *"s. */ - struct hmapx dsts; /* Contains "struct ofbundle *"s. */ - int out_vlan; - - mirror = mirror_lookup(ofproto, aux); - if (!s) { - mirror_destroy(mirror); - return 0; + ovs_rwlock_wrlock(&ofproto->ml->rwlock); + if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { + mac_learning_flush(ofproto->ml); } - if (!mirror) { - int idx; - - idx = mirror_scan(ofproto); - if (idx < 0) { - VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " - "cannot create %s", - ofproto->up.name, MAX_MIRRORS, s->name); - return EFBIG; - } + ovs_rwlock_unlock(&ofproto->ml->rwlock); + return 0; +} - mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror); - mirror->ofproto = ofproto; - mirror->idx = idx; - mirror->aux = aux; - mirror->out_vlan = -1; - mirror->name = NULL; - } +static bool +is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + struct ofbundle *bundle = bundle_lookup(ofproto, aux); + return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0; +} - if (!mirror->name || strcmp(s->name, mirror->name)) { - free(mirror->name); - mirror->name = xstrdup(s->name); - } +static void +forward_bpdu_changed(struct ofproto *ofproto_) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + ofproto->backer->need_revalidate = REV_RECONFIGURE; +} - /* Get the new configuration. */ - if (s->out_bundle) { - out = bundle_lookup(ofproto, s->out_bundle); - if (!out) { - mirror_destroy(mirror); - return EINVAL; - } - out_vlan = -1; - } else { - out = NULL; - out_vlan = s->out_vlan; - } - bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs); - bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts); - - /* If the configuration has not changed, do nothing. */ - if (hmapx_equals(&srcs, &mirror->srcs) - && hmapx_equals(&dsts, &mirror->dsts) - && vlan_bitmap_equal(mirror->vlans, s->src_vlans) - && mirror->out == out - && mirror->out_vlan == out_vlan) - { - hmapx_destroy(&srcs); - hmapx_destroy(&dsts); - return 0; - } +static void +set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time, + size_t max_entries) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + ovs_rwlock_wrlock(&ofproto->ml->rwlock); + mac_learning_set_idle_time(ofproto->ml, idle_time); + mac_learning_set_max_entries(ofproto->ml, max_entries); + ovs_rwlock_unlock(&ofproto->ml->rwlock); +} + +/* Ports. */ - hmapx_swap(&srcs, &mirror->srcs); - hmapx_destroy(&srcs); +static struct ofport_dpif * +get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port) +{ + struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port); + return ofport ? ofport_dpif_cast(ofport) : NULL; +} - hmapx_swap(&dsts, &mirror->dsts); - hmapx_destroy(&dsts); +static struct ofport_dpif * +get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port) +{ + struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port); + return port && &ofproto->up == port->up.ofproto ? port : NULL; +} - free(mirror->vlans); - mirror->vlans = vlan_bitmap_clone(s->src_vlans); - - mirror->out = out; - mirror->out_vlan = out_vlan; - - /* Update bundles. */ - mirror_bit = MIRROR_MASK_C(1) << mirror->idx; - HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) { - if (hmapx_contains(&mirror->srcs, bundle)) { - bundle->src_mirrors |= mirror_bit; - } else { - bundle->src_mirrors &= ~mirror_bit; - } - - if (hmapx_contains(&mirror->dsts, bundle)) { - bundle->dst_mirrors |= mirror_bit; - } else { - bundle->dst_mirrors &= ~mirror_bit; - } - - if (mirror->out == bundle) { - bundle->mirror_out |= mirror_bit; - } else { - bundle->mirror_out &= ~mirror_bit; - } - } - - ofproto->need_revalidate = REV_RECONFIGURE; - ofproto->has_mirrors = true; - mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); - mirror_update_dups(ofproto); - - return 0; +static void +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(ofproto, dpif_port->port_no); } static void -mirror_destroy(struct ofmirror *mirror) +ofport_update_peer(struct ofport_dpif *ofport) { - struct ofproto_dpif *ofproto; - mirror_mask_t mirror_bit; - struct ofbundle *bundle; - int i; + const struct ofproto_dpif *ofproto; + struct dpif_backer *backer; + const char *peer_name; - if (!mirror) { + if (!netdev_vport_is_patch(ofport->up.netdev)) { return; } - ofproto = mirror->ofproto; - ofproto->need_revalidate = REV_RECONFIGURE; - mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); + backer = ofproto_dpif_cast(ofport->up.ofproto)->backer; + backer->need_revalidate = REV_RECONFIGURE; - mirror_bit = MIRROR_MASK_C(1) << mirror->idx; - HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { - bundle->src_mirrors &= ~mirror_bit; - bundle->dst_mirrors &= ~mirror_bit; - bundle->mirror_out &= ~mirror_bit; + if (ofport->peer) { + ofport->peer->peer = NULL; + ofport->peer = NULL; } - hmapx_destroy(&mirror->srcs); - hmapx_destroy(&mirror->dsts); - free(mirror->vlans); - - ofproto->mirrors[mirror->idx] = NULL; - free(mirror->name); - free(mirror); - - mirror_update_dups(ofproto); - - ofproto->has_mirrors = false; - for (i = 0; i < MAX_MIRRORS; i++) { - if (ofproto->mirrors[i]) { - ofproto->has_mirrors = true; - break; - } + peer_name = netdev_vport_patch_peer(ofport->up.netdev); + if (!peer_name) { + return; } -} -static int -mirror_get_stats(struct ofproto *ofproto_, void *aux, - uint64_t *packets, uint64_t *bytes) -{ - struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - struct ofmirror *mirror = mirror_lookup(ofproto, aux); + HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { + struct ofport *peer_ofport; + struct ofport_dpif *peer; + const char *peer_peer; - if (!mirror) { - *packets = *bytes = UINT64_MAX; - return 0; - } + if (ofproto->backer != backer) { + continue; + } - *packets = mirror->packet_count; - *bytes = mirror->byte_count; + peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name); + if (!peer_ofport) { + continue; + } - return 0; -} + peer = ofport_dpif_cast(peer_ofport); + peer_peer = netdev_vport_patch_peer(peer->up.netdev); + if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev), + peer_peer)) { + ofport->peer = peer; + ofport->peer->peer = ofport; + } -static int -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); + return; } - return 0; -} - -static bool -is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux) -{ - struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - struct ofbundle *bundle = bundle_lookup(ofproto, aux); - return bundle && bundle->mirror_out != 0; -} - -static void -forward_bpdu_changed(struct ofproto *ofproto_) -{ - struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - ofproto->need_revalidate = REV_RECONFIGURE; -} - -static void -set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time) -{ - struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - mac_learning_set_idle_time(ofproto->ml, idle_time); -} - -/* Ports. */ - -static struct ofport_dpif * -get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port) -{ - struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port); - return ofport ? ofport_dpif_cast(ofport) : NULL; -} - -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(ofproto, odp_port)); -} - -static void -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(ofproto, dpif_port->port_no); } static void @@ -2804,6 +2976,15 @@ port_run_fast(struct ofport_dpif *ofport) send_packet(ofport, &packet); ofpbuf_uninit(&packet); } + + if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) { + struct ofpbuf packet; + + ofpbuf_init(&packet, 0); + bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr); + send_packet(ofport, &packet); + ofpbuf_uninit(&packet); + } } static void @@ -2816,6 +2997,7 @@ port_run(struct ofport_dpif *ofport) ofport->carrier_seq = carrier_seq; port_run_fast(ofport); + if (ofport->cfm) { int cfm_opup = cfm_get_opup(ofport->cfm); @@ -2827,6 +3009,11 @@ port_run(struct ofport_dpif *ofport) } } + if (ofport->bfd) { + bfd_run(ofport->bfd); + enable = enable && bfd_forwarding(ofport->bfd); + } + if (ofport->bundle) { enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport); if (carrier_changed) { @@ -2836,10 +3023,7 @@ port_run(struct ofport_dpif *ofport) if (ofport->may_enable != enable) { 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; } ofport->may_enable = enable; @@ -2851,6 +3035,10 @@ port_wait(struct ofport_dpif *ofport) if (ofport->cfm) { cfm_wait(ofport->cfm); } + + if (ofport->bfd) { + bfd_wait(ofport->bfd); + } } static int @@ -2861,6 +3049,24 @@ port_query_by_name(const struct ofproto *ofproto_, const char *devname, struct dpif_port dpif_port; int error; + if (sset_contains(&ofproto->ghost_ports, devname)) { + const char *type = netdev_get_type_from_name(devname); + + /* We may be called before ofproto->up.port_by_name is populated with + * the appropriate ofport. For this reason, we must get the name and + * type from the netdev layer directly. */ + if (type) { + const struct ofport *ofport; + + ofport = shash_find_data(&ofproto->up.port_by_name, devname); + ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE; + ofproto_port->name = xstrdup(devname); + ofproto_port->type = xstrdup(type); + return 0; + } + return ENODEV; + } + if (!sset_contains(&ofproto->ports, devname)) { return ENODEV; } @@ -2876,29 +3082,55 @@ static int port_add(struct ofproto *ofproto_, struct netdev *netdev) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - uint32_t odp_port = UINT32_MAX; - int error; + const char *devname = netdev_get_name(netdev); + char namebuf[NETDEV_VPORT_NAME_BUFSIZE]; + const char *dp_port_name; - error = dpif_port_add(ofproto->backer->dpif, netdev, &odp_port); - if (!error) { - sset_add(&ofproto->ports, netdev_get_name(netdev)); + if (netdev_vport_is_patch(netdev)) { + sset_add(&ofproto->ghost_ports, netdev_get_name(netdev)); + return 0; } - return error; + + dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf); + if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) { + odp_port_t port_no = ODPP_NONE; + 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, odp_to_u32(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) +port_del(struct ofproto *ofproto_, ofp_port_t ofp_port) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - uint32_t odp_port = ofp_port_to_odp_port(ofproto, ofp_port); + struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); int error = 0; - if (odp_port != OFPP_NONE) { - error = dpif_port_del(ofproto->backer->dpif, odp_port); + if (!ofport) { + return 0; } - if (!error) { - struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); - if (ofport) { + + sset_find_and_delete(&ofproto->ghost_ports, + netdev_get_name(ofport->up.netdev)); + ofproto->backer->need_revalidate = REV_RECONFIGURE; + if (!ofport->is_tunnel) { + 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 @@ -2915,9 +3147,11 @@ 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 @@ -2949,57 +3183,56 @@ port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats) return error; } -/* Account packets for LOCAL port. */ -static void -ofproto_update_local_port_stats(const struct ofproto *ofproto_, - size_t tx_size, size_t rx_size) -{ - struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - - if (rx_size) { - ofproto->stats.rx_packets++; - ofproto->stats.rx_bytes += rx_size; - } - if (tx_size) { - ofproto->stats.tx_packets++; - ofproto->stats.tx_bytes += tx_size; - } -} - struct port_dump_state { uint32_t bucket; uint32_t offset; + bool ghost; + + struct ofproto_port port; + bool has_port; }; static int port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep) { - struct port_dump_state *state; - - *statep = state = xmalloc(sizeof *state); - state->bucket = 0; - state->offset = 0; + *statep = xzalloc(sizeof(struct port_dump_state)); return 0; } static int -port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_, +port_dump_next(const struct ofproto *ofproto_, void *state_, struct ofproto_port *port) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); struct port_dump_state *state = state_; + const struct sset *sset; struct sset_node *node; - while ((node = sset_at_position(&ofproto->ports, &state->bucket, - &state->offset))) { + if (state->has_port) { + ofproto_port_destroy(&state->port); + state->has_port = false; + } + sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports; + while ((node = sset_at_position(sset, &state->bucket, &state->offset))) { int error; - error = port_query_by_name(ofproto_, node->name, port); - if (error != ENODEV) { + error = port_query_by_name(ofproto_, node->name, &state->port); + if (!error) { + *port = state->port; + state->has_port = true; + return 0; + } else if (error != ENODEV) { return error; } } + if (!state->ghost) { + state->ghost = true; + state->bucket = 0; + state->offset = 0; + return port_dump_next(ofproto_, state_, port); + } + return EOF; } @@ -3008,6 +3241,9 @@ port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) { struct port_dump_state *state = state_; + if (state->has_port) { + ofproto_port_destroy(&state->port); + } free(state); return 0; } @@ -3065,17 +3301,23 @@ struct flow_miss { enum odp_key_fitness key_fitness; const struct nlattr *key; size_t key_len; - ovs_be16 initial_tci; 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. */ + + uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */ + struct xlate_out xout; + bool xout_garbage; /* 'xout' needs to be uninitialized? */ + + struct ofpbuf mask; /* Flow mask for "put" ops. */ + struct odputil_keybuf maskbuf; + + /* If this is a "put" op, then a pointer to the subfacet that should + * be marked as uninstalled if the operation fails. */ + struct subfacet *subfacet; }; /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each @@ -3102,43 +3344,14 @@ send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet, connmgr_send_packet_in(ofproto->up.connmgr, &pin); } -static enum slow_path_reason -process_special(struct ofproto_dpif *ofproto, const struct flow *flow, - 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)) { - if (packet) { - cfm_process_heartbeat(ofport->cfm, packet); - } - return SLOW_CFM; - } else if (ofport->bundle && ofport->bundle->lacp - && flow->dl_type == htons(ETH_TYPE_LACP)) { - if (packet) { - lacp_process_packet(ofport->bundle->lacp, ofport, packet); - } - return SLOW_LACP; - } else if (ofproto->stp && stp_should_process_flow(flow)) { - if (packet) { - stp_process_packet(ofport, packet); - } - return SLOW_STP; - } - 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; } } @@ -3155,7 +3368,9 @@ 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.in_port.ofp_port + != vsp_realdev_to_vlandev(miss->ofproto, miss->flow.in_port.ofp_port, + miss->flow.vlan_tci)) { /* This packet was received on a VLAN splinter port. We * added a VLAN to the packet to make the packet resemble * the flow, but the actions were composed assuming that @@ -3166,24 +3381,21 @@ init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet, } op->subfacet = NULL; - op->garbage = NULL; + op->xout_garbage = false; op->dpif_op.type = DPIF_OP_EXECUTE; op->dpif_op.u.execute.key = miss->key; op->dpif_op.u.execute.key_len = miss->key_len; op->dpif_op.u.execute.packet = packet; + ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf); } /* Helper for handle_flow_miss_without_facet() and * handle_flow_miss_with_facet(). */ static void -handle_flow_miss_common(struct rule_dpif *rule, - struct ofpbuf *packet, const struct flow *flow) +handle_flow_miss_common(struct ofproto_dpif *ofproto, struct ofpbuf *packet, + const struct flow *flow, bool fail_open) { - struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); - - ofproto->n_matches++; - - if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) { + if (fail_open) { /* * Extra-special case for fail-open mode. * @@ -3199,72 +3411,78 @@ handle_flow_miss_common(struct rule_dpif *rule, } /* Figures out whether a flow that missed in 'ofproto', whose details are in - * 'miss', is likely to be worth tracking in detail in userspace and (usually) - * installing a datapath flow. The answer is usually "yes" (a return value of - * true). However, for short flows the cost of bookkeeping is much higher than - * the benefits, so when the datapath holds a large number of flows we impose - * some heuristics to decide which flows are likely to be worth tracking. */ + * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace + * and (usually) installing a datapath flow. The answer is usually "yes" (a + * return value of true). However, for short flows the cost of bookkeeping is + * much higher than the benefits, so when the datapath holds a large number of + * flows we impose some heuristics to decide which flows are likely to be worth + * tracking. */ static bool -flow_miss_should_make_facet(struct ofproto_dpif *ofproto, - struct flow_miss *miss, uint32_t hash) +flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc) { - if (!ofproto->governor) { + struct dpif_backer *backer = miss->ofproto->backer; + uint32_t hash; + + switch (flow_miss_model) { + case OFPROTO_HANDLE_MISS_AUTO: + break; + case OFPROTO_HANDLE_MISS_WITH_FACETS: + return true; + case OFPROTO_HANDLE_MISS_WITHOUT_FACETS: + return false; + } + + if (!backer->governor) { size_t n_subfacets; - n_subfacets = hmap_count(&ofproto->subfacets); - if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) { + n_subfacets = hmap_count(&backer->subfacets); + if (n_subfacets * 2 <= flow_eviction_threshold) { return true; } - ofproto->governor = governor_create(ofproto->up.name); + backer->governor = governor_create(); } - return governor_should_install_flow(ofproto->governor, hash, + hash = flow_hash_in_wildcards(&miss->flow, wc, 0); + return governor_should_install_flow(backer->governor, hash, list_size(&miss->packets)); } -/* Handles 'miss', which matches 'rule', without creating a facet or subfacet - * or creating any datapath flow. May add an "execute" operation to 'ops' and - * increment '*n_ops'. */ +/* Handles 'miss' without creating a facet or subfacet or creating any datapath + * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'. + * May add an "execute" operation to 'ops' and increment '*n_ops'. */ static void -handle_flow_miss_without_facet(struct flow_miss *miss, - struct rule_dpif *rule, +handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout, + struct flow_miss *miss, struct flow_miss_op *ops, size_t *n_ops) { - struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); - long long int now = time_msec(); - struct action_xlate_ctx ctx; struct ofpbuf *packet; LIST_FOR_EACH (packet, list_node, &miss->packets) { - struct flow_miss_op *op = &ops[*n_ops]; - struct dpif_flow_stats stats; - struct ofpbuf odp_actions; COVERAGE_INC(facet_suppress); - ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub); + handle_flow_miss_common(miss->ofproto, packet, &miss->flow, + rule->up.cr.priority == FAIL_OPEN_PRIORITY); - dpif_flow_stats_extract(&miss->flow, packet, now, &stats); - rule_credit_stats(rule, &stats); + if (xout->slow) { + struct xlate_in xin; - action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci, - rule, 0, packet); - ctx.resubmit_stats = &stats; - xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, - &odp_actions); + xlate_in_init(&xin, miss->ofproto, &miss->flow, rule, 0, packet); + xlate_actions_for_side_effects(&xin); + } - if (odp_actions.size) { + if (xout->odp_actions.size) { + struct flow_miss_op *op = &ops[*n_ops]; struct dpif_execute *execute = &op->dpif_op.u.execute; init_flow_miss_execute_op(miss, packet, op); - execute->actions = odp_actions.data; - execute->actions_len = odp_actions.size; - op->garbage = ofpbuf_get_uninit_pointer(&odp_actions); + xlate_out_copy(&op->xout, xout); + execute->actions = op->xout.odp_actions.data; + execute->actions_len = op->xout.odp_actions.size; + op->xout_garbage = true; (*n_ops)++; - } else { - ofpbuf_uninit(&odp_actions); } } } @@ -3277,72 +3495,93 @@ handle_flow_miss_without_facet(struct flow_miss *miss, * here, then the new subfacet or its packets could look (occasionally) as * though it was used some time after the facet was used. That can make a * one-packet flow look like it has a nonzero duration, which looks odd in - * e.g. NetFlow statistics. */ + * e.g. NetFlow statistics. + * + * If non-null, 'stats' will be folded into 'facet'. */ static void handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet, - long long int now, + long long int now, struct dpif_flow_stats *stats, struct flow_miss_op *ops, size_t *n_ops) { - struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); enum subfacet_path want_path; struct subfacet *subfacet; struct ofpbuf *packet; - subfacet = subfacet_create(facet, miss, now); + want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH; LIST_FOR_EACH (packet, list_node, &miss->packets) { struct flow_miss_op *op = &ops[*n_ops]; - struct dpif_flow_stats stats; - struct ofpbuf odp_actions; - handle_flow_miss_common(facet->rule, packet, &miss->flow); + handle_flow_miss_common(miss->ofproto, packet, &miss->flow, + facet->fail_open); - ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub); - if (!subfacet->actions || subfacet->slow) { - subfacet_make_actions(subfacet, packet, &odp_actions); - } + if (want_path != SF_FAST_PATH) { + struct rule_dpif *rule; + struct xlate_in xin; - dpif_flow_stats_extract(&facet->flow, packet, now, &stats); - subfacet_update_stats(subfacet, &stats); + rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL); + xlate_in_init(&xin, facet->ofproto, &miss->flow, rule, 0, packet); + xlate_actions_for_side_effects(&xin); + } - if (subfacet->actions_len) { + if (facet->xout.odp_actions.size) { struct dpif_execute *execute = &op->dpif_op.u.execute; init_flow_miss_execute_op(miss, packet, op); - op->subfacet = subfacet; - if (!subfacet->slow) { - execute->actions = subfacet->actions; - execute->actions_len = subfacet->actions_len; - ofpbuf_uninit(&odp_actions); - } else { - execute->actions = odp_actions.data; - execute->actions_len = odp_actions.size; - op->garbage = ofpbuf_get_uninit_pointer(&odp_actions); - } - + execute->actions = facet->xout.odp_actions.data, + execute->actions_len = facet->xout.odp_actions.size; (*n_ops)++; - } else { - ofpbuf_uninit(&odp_actions); } } - want_path = subfacet_want_path(subfacet->slow); + /* Don't install the flow if it's the result of the "userspace" + * action for an already installed facet. This can occur when a + * datapath flow with wildcards has a "userspace" action and flows + * sent to userspace result in a different subfacet, which will then + * be rejected as overlapping by the datapath. */ + if (miss->upcall_type == DPIF_UC_ACTION + && !list_is_empty(&facet->subfacets)) { + if (stats) { + facet->used = MAX(facet->used, stats->used); + facet->packet_count += stats->n_packets; + facet->byte_count += stats->n_bytes; + facet->tcp_flags |= stats->tcp_flags; + } + return; + } + + subfacet = subfacet_create(facet, miss, now); + if (stats) { + subfacet_update_stats(subfacet, stats); + } + if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) { struct flow_miss_op *op = &ops[(*n_ops)++]; struct dpif_flow_put *put = &op->dpif_op.u.flow_put; - op->subfacet = subfacet; - op->garbage = NULL; + subfacet->path = want_path; + + ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf); + if (enable_megaflows) { + odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks, + &miss->flow, UINT32_MAX); + } + + op->xout_garbage = false; op->dpif_op.type = DPIF_OP_FLOW_PUT; - put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; + op->subfacet = subfacet; + put->flags = DPIF_FP_CREATE; put->key = miss->key; put->key_len = miss->key_len; + put->mask = op->mask.data; + put->mask_len = op->mask.size; + if (want_path == SF_FAST_PATH) { - put->actions = subfacet->actions; - put->actions_len = subfacet->actions_len; + put->actions = facet->xout.odp_actions.data; + put->actions_len = facet->xout.odp_actions.size; } else { - compose_slow_path(ofproto, &facet->flow, subfacet->slow, - op->stub, sizeof op->stub, + compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow, + op->slow_stub, sizeof op->slow_stub, &put->actions, &put->actions_len); } put->stats = NULL; @@ -3356,104 +3595,209 @@ handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops, size_t *n_ops) { struct ofproto_dpif *ofproto = miss->ofproto; + struct dpif_flow_stats stats__; + struct dpif_flow_stats *stats = &stats__; + struct ofpbuf *packet; struct facet *facet; long long int now; - uint32_t hash; - /* The caller must ensure that miss->hmap_node.hash contains - * flow_hash(miss->flow, 0). */ - hash = miss->hmap_node.hash; + now = time_msec(); + memset(stats, 0, sizeof *stats); + stats->used = now; + LIST_FOR_EACH (packet, list_node, &miss->packets) { + stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow); + stats->n_bytes += packet->size; + stats->n_packets++; + } - facet = facet_lookup_valid(ofproto, &miss->flow, hash); + facet = facet_lookup_valid(ofproto, &miss->flow); if (!facet) { - struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow); - - if (!flow_miss_should_make_facet(ofproto, miss, hash)) { - handle_flow_miss_without_facet(miss, rule, ops, n_ops); + struct flow_wildcards wc; + struct rule_dpif *rule; + struct xlate_out xout; + struct xlate_in xin; + + flow_wildcards_init_catchall(&wc); + rule = rule_dpif_lookup(ofproto, &miss->flow, &wc); + rule_credit_stats(rule, stats); + + xlate_in_init(&xin, ofproto, &miss->flow, rule, stats->tcp_flags, + NULL); + xin.resubmit_stats = stats; + xin.may_learn = true; + xlate_actions(&xin, &xout); + flow_wildcards_or(&xout.wc, &xout.wc, &wc); + + /* There does not exist a bijection between 'struct flow' and datapath + * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental + * assumption used throughout the facet and subfacet handling code. + * Since we have to handle these misses in userspace anyway, we simply + * skip facet creation, avoiding the problem altogether. */ + if (miss->key_fitness == ODP_FIT_TOO_LITTLE + || !flow_miss_should_make_facet(miss, &xout.wc)) { + handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops); return; } - facet = facet_create(rule, &miss->flow, hash); - now = facet->used; - } else { - now = time_msec(); + facet = facet_create(miss, rule, &xout, stats); + stats = NULL; } - handle_flow_miss_with_facet(miss, facet, now, ops, n_ops); + handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops); } -/* 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). The - * caller must have called odp_flow_key_to_flow() and supply 'fitness' and - * 'flow' from its output. The 'flow' argument must have had the "in_port" - * member converted to the OpenFlow number. - * - * 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_vsp_adjust(const struct ofproto_dpif *ofproto, - enum odp_key_fitness fitness, - struct flow *flow, ovs_be16 *initial_tci, - struct ofpbuf *packet) +static struct drop_key * +drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key, + size_t key_len) { - if (fitness == ODP_FIT_ERROR) { - return fitness; - } - *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); - } + struct drop_key *drop_key; - /* Let the caller know that we can't reproduce 'key' from 'flow'. */ - if (fitness == ODP_FIT_PERFECT) { - fitness = ODP_FIT_TOO_MUCH; + 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 fitness; + return NULL; } static void -handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls, - size_t n_upcalls) +drop_key_clear(struct dpif_backer *backer) { - struct dpif_upcall *upcall; - struct flow_miss *miss; - struct flow_miss misses[FLOW_MISS_MAX_BATCH]; - struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2]; - struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2]; - struct hmap todo; - int n_misses; - size_t n_ops; - size_t i; + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); + struct drop_key *drop_key, *next; - if (!n_upcalls) { - return; - } + HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) { + int error; - /* Construct the to-do list. - * + 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)", + ovs_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. + * + * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets + * 'flow''s in_port to OFPP_NONE. + * + * 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). + * + * 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, odp_port_t *odp_in_port) +{ + const struct ofport_dpif *port; + enum odp_key_fitness fitness; + int error = ENODEV; + + fitness = odp_flow_key_to_flow(key, key_len, flow); + if (fitness == ODP_FIT_ERROR) { + error = EINVAL; + goto exit; + } + + if (odp_in_port) { + *odp_in_port = flow->in_port.odp_port; + } + + port = (tnl_port_should_receive(flow) + ? tnl_port_receive(flow) + : odp_port_to_ofport(backer, flow->in_port.odp_port)); + flow->in_port.ofp_port = port ? port->up.ofp_port : OFPP_NONE; + if (!port) { + goto exit; + } + + /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port + * it's theoretically possible that we'll receive an ofport belonging to an + * entirely different datapath. In practice, this can't happen because no + * platforms has two separate datapaths which each support tunneling. */ + ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer); + + 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); + } + +exit: + if (fitnessp) { + *fitnessp = fitness; + } + return error; +} + +static void +handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls, + size_t n_upcalls) +{ + struct dpif_upcall *upcall; + struct flow_miss *miss; + struct flow_miss misses[FLOW_MISS_MAX_BATCH]; + struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2]; + struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2]; + struct hmap todo; + int n_misses; + size_t n_ops; + size_t i; + + if (!n_upcalls) { + return; + } + + /* Construct the to-do list. + * * This just amounts to extracting the flow from each packet and sticking * the packets that have the same flow in the same "flow_miss" structure so * that we can process them together. */ @@ -3462,48 +3806,63 @@ handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls, for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) { struct flow_miss *miss = &misses[n_misses]; struct flow_miss *existing_miss; - enum odp_key_fitness fitness; struct ofproto_dpif *ofproto; - struct ofport_dpif *port; - uint32_t odp_in_port; + odp_port_t odp_in_port; struct flow flow; uint32_t hash; + int error; - fitness = odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); - port = odp_port_to_ofport(backer, flow.in_port); - if (!port) { - /* 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. */ - VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32, - flow.in_port); + error = ofproto_receive(backer, upcall->packet, upcall->key, + upcall->key_len, &flow, &miss->key_fitness, + &ofproto, &odp_in_port); + if (error == ENODEV) { + struct drop_key *drop_key; + + /* Received packet on datapath port for which we couldn't + * associate an ofproto. This can happen if a port is removed + * while traffic is being received. Print a rate-limited message + * in case it happens frequently. Install a drop flow so + * that future packets of the flow are inexpensively dropped + * in the kernel. */ + VLOG_INFO_RL(&rl, "received packet on unassociated datapath port " + "%"PRIu32, odp_in_port); + + drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len); + if (!drop_key) { + int ret; + ret = dpif_flow_put(backer->dpif, + DPIF_FP_CREATE | DPIF_FP_MODIFY, + upcall->key, upcall->key_len, + NULL, 0, NULL, 0, NULL); + + if (!ret) { + drop_key = xmalloc(sizeof *drop_key); + drop_key->key = xmemdup(upcall->key, upcall->key_len); + drop_key->key_len = upcall->key_len; + + hmap_insert(&backer->drop_keys, &drop_key->hmap_node, + hash_bytes(drop_key->key, drop_key->key_len, 0)); + } + } continue; } - ofproto = ofproto_dpif_cast(port->up.ofproto); - odp_in_port = flow.in_port; - flow.in_port = port->up.ofp_port; - - /* Obtain metadata and check userspace/kernel agreement on flow match, - * then set 'flow''s header pointers. */ - miss->key_fitness = ofproto_dpif_vsp_adjust(ofproto, fitness, - &flow, &miss->initial_tci, upcall->packet); - if (miss->key_fitness == ODP_FIT_ERROR) { + if (error) { continue; } - flow_extract(upcall->packet, flow.skb_priority, - &flow.tunnel, flow.in_port, &miss->flow); + + 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++; @@ -3519,7 +3878,7 @@ handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls, HMAP_FOR_EACH (miss, hmap_node, &todo) { handle_flow_miss(miss, flow_miss_ops, &n_ops); } - assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); + ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); /* Execute batch. */ for (i = 0; i < n_ops; i++) { @@ -3527,32 +3886,42 @@ handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls, } dpif_operate(backer->dpif, dpif_ops, n_ops); - /* Free memory and update facets. */ 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); + if (dpif_ops[i]->error != 0 + && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT + && flow_miss_ops[i].subfacet) { + struct subfacet *subfacet = flow_miss_ops[i].subfacet; + + COVERAGE_INC(subfacet_install_fail); + + /* Zero-out subfacet counters when installation failed, but + * datapath reported hits. This should not happen and + * indicates a bug, since if the datapath flow exists, we + * should not be attempting to create a new subfacet. A + * buggy datapath could trigger this, so just zero out the + * counters and log an error. */ + if (subfacet->dp_packet_count || subfacet->dp_byte_count) { + VLOG_ERR_RL(&rl, "failed to install subfacet for which " + "datapath reported hits"); + subfacet->dp_packet_count = subfacet->dp_byte_count = 0; } - break; - case DPIF_OP_FLOW_DEL: - NOT_REACHED(); + subfacet->path = SF_NOT_INSTALLED; } - free(op->garbage); + /* Free memory. */ + if (flow_miss_ops[i].xout_garbage) { + xlate_out_uninit(&flow_miss_ops[i].xout); + } } hmap_destroy(&todo); } -static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL } +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. */ @@ -3570,17 +3939,34 @@ classify_upcall(const struct dpif_upcall *upcall) } /* "action" upcalls need a closer look. */ - memcpy(&cookie, &upcall->userdata, sizeof(cookie)); - switch (cookie.type) { - case USER_ACTION_COOKIE_SFLOW: + if (!upcall->userdata) { + VLOG_WARN_RL(&rl, "action upcall missing cookie"); + return BAD_UPCALL; + } + userdata_len = nl_attr_get_size(upcall->userdata); + if (userdata_len < sizeof cookie.type + || userdata_len > sizeof cookie) { + VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu", + userdata_len); + return BAD_UPCALL; + } + memset(&cookie, 0, sizeof cookie); + memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len); + if (userdata_len == sizeof cookie.sflow + && cookie.type == USER_ACTION_COOKIE_SFLOW) { return SFLOW_UPCALL; - - case USER_ACTION_COOKIE_SLOW_PATH: + } else if (userdata_len == sizeof cookie.slow_path + && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) { return MISS_UPCALL; - - case USER_ACTION_COOKIE_UNSPEC: - default: - VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata); + } else if (userdata_len == sizeof cookie.flow_sample + && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) { + return FLOW_SAMPLE_UPCALL; + } else if (userdata_len == sizeof cookie.ipfix + && cookie.type == USER_ACTION_COOKIE_IPFIX) { + return IPFIX_UPCALL; + } else { + VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16 + " and size %zu", cookie.type, userdata_len); return BAD_UPCALL; } } @@ -3591,35 +3977,63 @@ handle_sflow_upcall(struct dpif_backer *backer, { struct ofproto_dpif *ofproto; union user_action_cookie cookie; - enum odp_key_fitness fitness; - struct ofport_dpif *port; - ovs_be16 initial_tci; struct flow flow; - uint32_t odp_in_port; + odp_port_t odp_in_port; - fitness = odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); - - port = odp_port_to_ofport(backer, flow.in_port); - if (!port) { + if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len, + &flow, NULL, &ofproto, &odp_in_port) + || !ofproto->sflow) { return; } - ofproto = ofproto_dpif_cast(port->up.ofproto); - if (!ofproto->sflow) { + 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) + || !ofproto->ipfix) { return; } - odp_in_port = flow.in_port; - flow.in_port = port->up.ofp_port; - fitness = ofproto_dpif_vsp_adjust(ofproto, fitness, &flow, - &initial_tci, upcall->packet); - if (fitness == ODP_FIT_ERROR) { + 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) + || !ofproto->ipfix) { return; } - memcpy(&cookie, &upcall->userdata, sizeof(cookie)); - dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, - odp_in_port, &cookie); + /* 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 @@ -3632,7 +4046,7 @@ handle_upcalls(struct dpif_backer *backer, unsigned int max_batch) int n_misses; int i; - assert(max_batch <= FLOW_MISS_MAX_BATCH); + ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH); n_misses = 0; for (n_processed = 0; n_processed < max_batch; n_processed++) { @@ -3659,6 +4073,16 @@ handle_upcalls(struct dpif_backer *backer, unsigned int max_batch) ofpbuf_uninit(buf); break; + case FLOW_SAMPLE_UPCALL: + handle_flow_sample_upcall(backer, upcall); + ofpbuf_uninit(buf); + break; + + case IPFIX_UPCALL: + handle_ipfix_upcall(backer, upcall); + ofpbuf_uninit(buf); + break; + case BAD_UPCALL: ofpbuf_uninit(buf); break; @@ -3676,10 +4100,10 @@ handle_upcalls(struct dpif_backer *backer, unsigned int max_batch) /* Flow expiration. */ -static int subfacet_max_idle(const struct ofproto_dpif *); +static int subfacet_max_idle(const struct dpif_backer *); static void update_stats(struct dpif_backer *); static void rule_expire(struct rule_dpif *); -static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle); +static void expire_subfacets(struct dpif_backer *, int dp_max_idle); /* This function is called periodically by run(). Its job is to collect * updates for the flows that have been installed into the datapath, most @@ -3691,35 +4115,50 @@ static int expire(struct dpif_backer *backer) { struct ofproto_dpif *ofproto; - int max_idle = INT32_MAX; + size_t n_subfacets; + int max_idle; + + /* Periodically clear out the drop keys in an effort to keep them + * relatively few. */ + drop_key_clear(backer); /* Update stats for each flow in the backer. */ update_stats(backer); + n_subfacets = hmap_count(&backer->subfacets); + if (n_subfacets) { + struct subfacet *subfacet; + long long int total, now; + + total = 0; + now = time_msec(); + HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) { + total += now - subfacet->created; + } + backer->avg_subfacet_life += total / n_subfacets; + } + backer->avg_subfacet_life /= 2; + + backer->avg_n_subfacet += n_subfacets; + backer->avg_n_subfacet /= 2; + + backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets); + + max_idle = subfacet_max_idle(backer); + expire_subfacets(backer, max_idle); + HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { - struct rule_dpif *rule, *next_rule; - struct oftable *table; - int dp_max_idle; + struct rule *rule, *next_rule; if (ofproto->backer != backer) { continue; } - /* 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. */ - OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { - struct cls_cursor cursor; - - cls_cursor_init(&cursor, &table->cls, NULL); - CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { - rule_expire(rule); - } + LIST_FOR_EACH_SAFE (rule, next_rule, expirable, + &ofproto->up.expirable) { + rule_expire(rule_dpif_cast(rule)); } /* All outstanding data in existing flows has been accounted, so it's a @@ -3729,7 +4168,7 @@ expire(struct dpif_backer *backer) HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { if (bundle->bond) { - bond_rebalance(bundle->bond, &ofproto->revalidate_set); + bond_rebalance(bundle->bond); } } } @@ -3745,38 +4184,41 @@ update_subfacet_stats(struct subfacet *subfacet, const struct dpif_flow_stats *stats) { struct facet *facet = subfacet->facet; + struct dpif_flow_stats diff; + + diff.tcp_flags = stats->tcp_flags; + diff.used = stats->used; if (stats->n_packets >= subfacet->dp_packet_count) { - uint64_t extra = stats->n_packets - subfacet->dp_packet_count; - facet->packet_count += extra; + diff.n_packets = stats->n_packets - subfacet->dp_packet_count; } else { VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); + diff.n_packets = 0; } if (stats->n_bytes >= subfacet->dp_byte_count) { - facet->byte_count += stats->n_bytes - subfacet->dp_byte_count; + diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count; } else { VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); + diff.n_bytes = 0; } + facet->ofproto->n_hit += diff.n_packets; subfacet->dp_packet_count = stats->n_packets; subfacet->dp_byte_count = stats->n_bytes; + subfacet_update_stats(subfacet, &diff); - facet->tcp_flags |= stats->tcp_flags; - - subfacet_update_time(subfacet, stats->used); if (facet->accounted_bytes < facet->byte_count) { facet_learn(facet); 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 ofproto_dpif *ofproto, +delete_unexpected_flow(struct dpif_backer *backer, const struct nlattr *key, size_t key_len) { if (!VLOG_DROP_WARN(&rl)) { @@ -3784,12 +4226,12 @@ delete_unexpected_flow(struct ofproto_dpif *ofproto, ds_init(&s); odp_flow_key_format(key, key_len, &s); - VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s)); + VLOG_WARN("unexpected flow: %s", ds_cstr(&s)); ds_destroy(&s); } COVERAGE_INC(facet_unexpected); - dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL); + dpif_flow_del(backer->dpif, key, key_len, NULL); } /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. @@ -3802,43 +4244,28 @@ delete_unexpected_flow(struct ofproto_dpif *ofproto, * avoided by calling update_stats() whenever rules are created or * deleted. However, the performance impact of making so many calls to the * datapath do not justify the benefit of having perfectly accurate statistics. + * + * In addition, this function maintains per ofproto flow hit counts. The patch + * port is not treated specially. e.g. A packet ingress from br0 patched into + * br1 will increase the hit count of br0 by 1, however, does not affect + * the hit or miss counts of br1. */ static void update_stats(struct dpif_backer *backer) { const struct dpif_flow_stats *stats; struct dpif_flow_dump dump; - const struct nlattr *key; - size_t key_len; + const struct nlattr *key, *mask; + size_t key_len, mask_len; dpif_flow_dump_start(&dump, backer->dpif); - while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) { - struct flow flow; + while (dpif_flow_dump_next(&dump, &key, &key_len, + &mask, &mask_len, NULL, NULL, &stats)) { struct subfacet *subfacet; - enum odp_key_fitness fitness; - struct ofproto_dpif *ofproto; - struct ofport_dpif *port; uint32_t key_hash; - fitness = odp_flow_key_to_flow(key, key_len, &flow); - if (fitness == ODP_FIT_ERROR) { - continue; - } - - port = odp_port_to_ofport(backer, flow.in_port); - if (!port) { - /* This flow is for a port for which we couldn't associate an - * ofproto. This can happen if a port is removed while - * traffic is being received. Ignore this flow, since it - * will get timed out. */ - continue; - } - - ofproto = ofproto_dpif_cast(port->up.ofproto); - flow.in_port = port->up.ofp_port; key_hash = odp_flow_key_hash(key, key_len); - - subfacet = subfacet_find(ofproto, key, key_len, key_hash, &flow); + subfacet = subfacet_find(backer, key, key_len, key_hash); switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) { case SF_FAST_PATH: update_subfacet_stats(subfacet, stats); @@ -3850,11 +4277,14 @@ update_stats(struct dpif_backer *backer) case SF_NOT_INSTALLED: default: - delete_unexpected_flow(ofproto, key, key_len); + delete_unexpected_flow(backer, key, key_len); break; } + run_fast_rl(); } dpif_flow_dump_done(&dump); + + update_moving_averages(backer); } /* Calculates and returns the number of milliseconds of idle time after which @@ -3862,7 +4292,7 @@ update_stats(struct dpif_backer *backer) * its statistics into its facet, and when a facet's last subfacet expires, we * fold its statistic into its rule. */ static int -subfacet_max_idle(const struct ofproto_dpif *ofproto) +subfacet_max_idle(const struct dpif_backer *backer) { /* * Idle time histogram. @@ -3885,7 +4315,7 @@ subfacet_max_idle(const struct ofproto_dpif *ofproto) * that is installed in the kernel gets dropped in the appropriate bucket. * After the histogram has been built, we compute the cutoff so that only * the most-recently-used 1% of subfacets (but at least - * ofproto->up.flow_eviction_threshold flows) are kept cached. At least + * flow_eviction_threshold flows) are kept cached. At least * the most-recently-used bucket of subfacets is kept, so actually an * arbitrary number of subfacets can be kept in any given expiration run * (though the next run will delete most of those unless they receive @@ -3903,14 +4333,14 @@ subfacet_max_idle(const struct ofproto_dpif *ofproto) long long int now; int i; - total = hmap_count(&ofproto->subfacets); - if (total <= ofproto->up.flow_eviction_threshold) { + total = hmap_count(&backer->subfacets); + if (total <= flow_eviction_threshold) { return N_BUCKETS * BUCKET_WIDTH; } /* Build histogram. */ now = time_msec(); - HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) { + HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) { long long int idle = now - subfacet->used; int bucket = (idle <= 0 ? 0 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 @@ -3923,7 +4353,7 @@ subfacet_max_idle(const struct ofproto_dpif *ofproto) do { subtotal += buckets[bucket++]; } while (bucket < N_BUCKETS && - subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100)); + subtotal < MAX(flow_eviction_threshold, total / 100)); if (VLOG_IS_DBG_ENABLED()) { struct ds s; @@ -3938,7 +4368,7 @@ subfacet_max_idle(const struct ofproto_dpif *ofproto) ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); } } - VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s)); + VLOG_INFO("%s (msec:count)", ds_cstr(&s)); ds_destroy(&s); } @@ -3946,7 +4376,7 @@ subfacet_max_idle(const struct ofproto_dpif *ofproto) } static void -expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle) +expire_subfacets(struct dpif_backer *backer, int dp_max_idle) { /* Cutoff time for most flows. */ long long int normal_cutoff = time_msec() - dp_max_idle; @@ -3961,17 +4391,18 @@ expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle) n_batch = 0; HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, - &ofproto->subfacets) { + &backer->subfacets) { long long int cutoff; - cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP) + cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP + | SLOW_STP) ? special_cutoff : normal_cutoff); if (subfacet->used < cutoff) { if (subfacet->path != SF_NOT_INSTALLED) { batch[n_batch++] = subfacet; if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { - subfacet_destroy_batch(ofproto, batch, n_batch); + subfacet_destroy_batch(backer, batch, n_batch); n_batch = 0; } } else { @@ -3981,7 +4412,7 @@ expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle) } if (n_batch > 0) { - subfacet_destroy_batch(ofproto, batch, n_batch); + subfacet_destroy_batch(backer, batch, n_batch); } } @@ -3990,7 +4421,6 @@ expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle) static void rule_expire(struct rule_dpif *rule) { - struct facet *facet, *next_facet; long long int now; uint8_t reason; @@ -4013,57 +4443,67 @@ rule_expire(struct rule_dpif *rule) COVERAGE_INC(ofproto_dpif_expired); - /* Update stats. (This is a no-op if the rule expired due to an idle - * timeout, because that only happens when the rule has no facets left.) */ - LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { - facet_remove(facet); - } - /* Get rid of the rule. */ ofproto_rule_expire(&rule->up, reason); } /* Facets. */ -/* Creates and returns a new facet owned by 'rule', given a 'flow'. +/* Creates and returns a new facet based on 'miss'. * * The caller must already have determined that no facet with an identical - * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in - * the ofproto's classifier table. + * 'miss->flow' exists in 'miss->ofproto'. * - * 'hash' must be the return value of flow_hash(flow, 0). + * 'rule' and 'xout' must have been created based on 'miss'. + * + * 'facet'' statistics are initialized based on 'stats'. * * The facet will initially have no subfacets. The caller should create (at * least) one subfacet with subfacet_create(). */ static struct facet * -facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash) +facet_create(const struct flow_miss *miss, struct rule_dpif *rule, + struct xlate_out *xout, struct dpif_flow_stats *stats) { - struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); + struct ofproto_dpif *ofproto = miss->ofproto; struct facet *facet; + struct match match; facet = xzalloc(sizeof *facet); - facet->used = time_msec(); - hmap_insert(&ofproto->facets, &facet->hmap_node, hash); - list_push_back(&rule->facets, &facet->list_node); - facet->rule = rule; - facet->flow = *flow; + facet->ofproto = miss->ofproto; + facet->packet_count = facet->prev_packet_count = stats->n_packets; + facet->byte_count = facet->prev_byte_count = stats->n_bytes; + facet->tcp_flags = stats->tcp_flags; + facet->used = stats->used; + facet->flow = miss->flow; + facet->learn_rl = time_msec() + 500; + list_init(&facet->subfacets); netflow_flow_init(&facet->nf_flow); netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); + xlate_out_copy(&facet->xout, xout); + + match_init(&match, &facet->flow, &facet->xout.wc); + cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY); + classifier_insert(&ofproto->facets, &facet->cr); + + facet->nf_flow.output_iface = facet->xout.nf_output_iface; + facet->fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY; + return facet; } static void facet_free(struct facet *facet) { - free(facet); + if (facet) { + xlate_out_uninit(&facet->xout); + free(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, @@ -4075,16 +4515,14 @@ execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); odp_flow_key_from_flow(&key, flow, - ofp_port_to_odp_port(ofproto, flow->in_port)); + ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port)); error = dpif_execute(ofproto->backer->dpif, key.data, key.size, odp_actions, actions_len, packet); - - ofpbuf_delete(packet); return !error; } -/* Remove 'facet' from 'ofproto' and free up the associated memory: +/* Remove 'facet' from its ofproto and free up the associated memory: * * - If 'facet' was installed in the datapath, uninstalls it and updates its * rule's statistics, via subfacet_uninstall(). @@ -4094,10 +4532,9 @@ execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, static void 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) { @@ -4116,8 +4553,8 @@ facet_remove(struct facet *facet) &facet->subfacets) { subfacet_destroy__(subfacet); } - hmap_remove(&ofproto->facets, &facet->hmap_node); - list_remove(&facet->list_node); + classifier_remove(&facet->ofproto->facets, &facet->cr); + cls_rule_destroy(&facet->cr); facet_free(facet); } @@ -4126,35 +4563,33 @@ facet_remove(struct facet *facet) static void facet_learn(struct facet *facet) { - struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); - struct action_xlate_ctx ctx; + long long int now = time_msec(); + + if (!facet->xout.has_fin_timeout && now < facet->learn_rl) { + return; + } - if (!facet->has_learn - && !facet->has_normal - && (!facet->has_fin_timeout + facet->learn_rl = now + 500; + + if (!facet->xout.has_learn + && !facet->xout.has_normal + && (!facet->xout.has_fin_timeout || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) { return; } - action_xlate_ctx_init(&ctx, ofproto, &facet->flow, - facet->flow.vlan_tci, - facet->rule, facet->tcp_flags, NULL); - ctx.may_learn = true; - xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts, - facet->rule->up.ofpacts_len); + facet_push_stats(facet, true); } static void facet_account(struct facet *facet) { - struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); - struct subfacet *subfacet; const struct nlattr *a; unsigned int left; ovs_be16 vlan_tci; uint64_t n_bytes; - if (!facet->has_normal || !ofproto->has_bonded_bundles) { + if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) { return; } n_bytes = facet->byte_count - facet->accounted_bytes; @@ -4167,17 +4602,15 @@ 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) { + NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data, + facet->xout.odp_actions.size) { const struct ovs_action_push_vlan *vlan; struct ofport_dpif *port; switch (nl_attr_type(a)) { case OVS_ACTION_ATTR_OUTPUT: - port = get_odp_port(ofproto, nl_attr_get_u32(a)); + port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a)); if (port && port->bundle && port->bundle->bond) { bond_account(port->bundle->bond, &facet->flow, vlan_tci_to_vid(vlan_tci), n_bytes); @@ -4203,9 +4636,11 @@ static bool facet_is_controller_flow(struct facet *facet) { if (facet) { - const struct rule *rule = &facet->rule->up; - const struct ofpact *ofpacts = rule->ofpacts; - size_t ofpacts_len = rule->ofpacts_len; + struct ofproto_dpif *ofproto = facet->ofproto; + const struct rule_dpif *rule = rule_dpif_lookup(ofproto, &facet->flow, + NULL); + const struct ofpact *ofpacts = rule->up.ofpacts; + size_t ofpacts_len = rule->up.ofpacts_len; if (ofpacts_len > 0 && ofpacts->type == OFPACT_CONTROLLER && @@ -4223,15 +4658,15 @@ facet_is_controller_flow(struct facet *facet) static void facet_flush_stats(struct facet *facet) { - struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); + struct ofproto_dpif *ofproto = facet->ofproto; 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); + facet_push_stats(facet, false); if (facet->accounted_bytes < facet->byte_count) { facet_account(facet); facet->accounted_bytes = facet->byte_count; @@ -4246,9 +4681,6 @@ facet_flush_stats(struct facet *facet) netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); } - facet->rule->packet_count += facet->packet_count; - facet->rule->byte_count += facet->byte_count; - /* Reset counters to prevent double counting if 'facet' ever gets * reinstalled. */ facet_reset_counters(facet); @@ -4257,197 +4689,86 @@ facet_flush_stats(struct facet *facet) facet->tcp_flags = 0; } -/* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. - * Returns it if found, otherwise a null pointer. - * - * 'hash' must be the return value of flow_hash(flow, 0). +/* Searches 'ofproto''s table of facets for one which would be responsible for + * 'flow'. Returns it if found, otherwise a null pointer. * * The returned facet might need revalidation; use facet_lookup_valid() * instead if that is important. */ static struct facet * -facet_find(struct ofproto_dpif *ofproto, - const struct flow *flow, uint32_t hash) +facet_find(struct ofproto_dpif *ofproto, const struct flow *flow) { - struct facet *facet; - - HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) { - if (flow_equal(flow, &facet->flow)) { - return facet; - } - } - - return NULL; + struct cls_rule *cr = classifier_lookup(&ofproto->facets, flow, NULL); + return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL; } -/* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. - * Returns it if found, otherwise a null pointer. - * - * 'hash' must be the return value of flow_hash(flow, 0). +/* Searches 'ofproto''s table of facets for one capable that covers + * 'flow'. Returns it if found, otherwise a null pointer. * * The returned facet is guaranteed to be valid. */ static struct facet * -facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow, - uint32_t hash) +facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow) { struct facet *facet; - facet = facet_find(ofproto, flow, hash); + facet = facet_find(ofproto, flow); if (facet - && (ofproto->need_revalidate - || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) { - facet_revalidate(facet); + && (ofproto->backer->need_revalidate + || tag_set_intersects(&ofproto->backer->revalidate_set, + facet->xout.tags)) + && !facet_revalidate(facet)) { + return NULL; } return facet; } -static const char * -subfacet_path_to_string(enum subfacet_path path) -{ - switch (path) { - case SF_NOT_INSTALLED: - return "not installed"; - case SF_FAST_PATH: - return "in fast path"; - case SF_SLOW_PATH: - return "in slow path"; - default: - return ""; - } -} - -/* Returns the path in which a subfacet should be installed if its 'slow' - * member has the specified value. */ -static enum subfacet_path -subfacet_want_path(enum slow_path_reason slow) -{ - return slow ? SF_SLOW_PATH : SF_FAST_PATH; -} - -/* Returns true if 'subfacet' needs to have its datapath flow updated, - * supposing that its actions have been recalculated as 'want_actions' and that - * 'slow' is nonzero iff 'subfacet' should be in the slow path. */ -static bool -subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow, - const struct ofpbuf *want_actions) -{ - enum subfacet_path want_path = subfacet_want_path(slow); - return (want_path != subfacet->path - || (want_path == SF_FAST_PATH - && (subfacet->actions_len != want_actions->size - || memcmp(subfacet->actions, want_actions->data, - subfacet->actions_len)))); -} - static bool facet_check_consistency(struct facet *facet) { static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); - struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); - - uint64_t odp_actions_stub[1024 / 8]; - struct ofpbuf odp_actions; + struct xlate_out xout; + struct xlate_in xin; struct rule_dpif *rule; - struct subfacet *subfacet; - bool may_log = false; - bool ok; - - /* Check the rule for consistency. */ - rule = rule_dpif_lookup(ofproto, &facet->flow); - ok = rule == facet->rule; - if (!ok) { - may_log = !VLOG_DROP_WARN(&rl); - if (may_log) { - struct ds s; - - ds_init(&s); - flow_format(&s, &facet->flow); - ds_put_format(&s, ": facet associated with wrong rule (was " - "table=%"PRIu8",", facet->rule->up.table_id); - cls_rule_format(&facet->rule->up.cr, &s); - ds_put_format(&s, ") (should have been table=%"PRIu8",", - rule->up.table_id); - cls_rule_format(&rule->up.cr, &s); - ds_put_char(&s, ')'); - - VLOG_WARN("%s", ds_cstr(&s)); - ds_destroy(&s); - } - } + bool ok, fail_open; /* Check the datapath actions for consistency. */ - 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; + rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL); + xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL); + xlate_actions(&xin, &xout); - action_xlate_ctx_init(&ctx, ofproto, &facet->flow, - subfacet->initial_tci, rule, 0, NULL); - xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, - &odp_actions); - - if (subfacet->path == SF_NOT_INSTALLED) { - /* This only happens if the datapath reported an error when we - * tried to install the flow. Don't flag another error here. */ - continue; - } + fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY; + ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions) + && facet->xout.slow == xout.slow + && facet->fail_open == fail_open; + if (!ok && !VLOG_DROP_WARN(&rl)) { + struct ds s = DS_EMPTY_INITIALIZER; - want_path = subfacet_want_path(subfacet->slow); - if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) { - /* The actions for slow-path flows may legitimately vary from one - * packet to the next. We're done. */ - continue; - } + flow_format(&s, &facet->flow); + ds_put_cstr(&s, ": inconsistency in facet"); - if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) { - continue; + if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) { + ds_put_cstr(&s, " (actions were: "); + format_odp_actions(&s, facet->xout.odp_actions.data, + facet->xout.odp_actions.size); + ds_put_cstr(&s, ") (correct actions: "); + format_odp_actions(&s, xout.odp_actions.data, + xout.odp_actions.size); + ds_put_char(&s, ')'); } - /* Inconsistency! */ - if (ok) { - may_log = !VLOG_DROP_WARN(&rl); - ok = false; - } - if (!may_log) { - /* Rate-limited, skip reporting. */ - continue; + if (facet->xout.slow != xout.slow) { + ds_put_format(&s, " slow path incorrect. should be %d", xout.slow); } - ds_init(&s); - subfacet_get_key(subfacet, &keybuf, &key); - odp_flow_key_format(key.data, key.size, &s); - - ds_put_cstr(&s, ": inconsistency in subfacet"); - if (want_path != subfacet->path) { - enum odp_key_fitness fitness = subfacet->key_fitness; - - ds_put_format(&s, " (%s, fitness=%s)", - subfacet_path_to_string(subfacet->path), - odp_key_fitness_to_string(fitness)); - ds_put_format(&s, " (should have been %s)", - subfacet_path_to_string(want_path)); - } else if (want_path == SF_FAST_PATH) { - ds_put_cstr(&s, " (actions were: "); - format_odp_actions(&s, subfacet->actions, - subfacet->actions_len); - ds_put_cstr(&s, ") (correct actions: "); - format_odp_actions(&s, odp_actions.data, odp_actions.size); - ds_put_char(&s, ')'); - } else { - ds_put_cstr(&s, " (actions: "); - format_odp_actions(&s, subfacet->actions, - subfacet->actions_len); - ds_put_char(&s, ')'); + if (facet->fail_open != fail_open) { + ds_put_format(&s, " fail open incorrect. should be %s", + fail_open ? "true" : "false"); } - VLOG_WARN("%s", ds_cstr(&s)); ds_destroy(&s); } - ofpbuf_uninit(&odp_actions); + xlate_out_uninit(&xout); return ok; } @@ -4458,120 +4779,105 @@ facet_check_consistency(struct facet *facet) * 'facet' to the new rule and recompiles its actions. * * - If the rule found is the same as 'facet''s current rule, leaves 'facet' - * where it is and recompiles its actions anyway. */ -static void + * where it is and recompiles its actions anyway. + * + * - If any of 'facet''s subfacets correspond to a new flow according to + * ofproto_receive(), 'facet' is removed. + * + * Returns true if 'facet' is still valid. False if 'facet' was removed. */ +static bool facet_revalidate(struct facet *facet) { - struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); - struct actions { - struct nlattr *odp_actions; - size_t actions_len; - }; - struct actions *new_actions; - - struct action_xlate_ctx ctx; - uint64_t odp_actions_stub[1024 / 8]; - struct ofpbuf odp_actions; - + struct ofproto_dpif *ofproto = facet->ofproto; struct rule_dpif *new_rule; struct subfacet *subfacet; - int i; + struct flow_wildcards wc; + struct xlate_out xout; + struct xlate_in xin; COVERAGE_INC(facet_revalidate); - new_rule = rule_dpif_lookup(ofproto, &facet->flow); + /* 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); + if (error + || recv_ofproto != ofproto + || facet != facet_find(ofproto, &recv_flow)) { + facet_remove(facet); + return false; + } + } + + flow_wildcards_init_catchall(&wc); + new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc); /* Calculate new datapath actions. * * We do not modify any 'facet' state yet, because we might need to, e.g., * emit a NetFlow expiration and, if so, we need to have the old state * around to properly compose it. */ + xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL); + xlate_actions(&xin, &xout); + flow_wildcards_or(&xout.wc, &xout.wc, &wc); - /* If the datapath actions changed or the installability changed, - * then we need to talk to the datapath. */ - i = 0; - new_actions = NULL; - memset(&ctx, 0, sizeof ctx); - ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); - LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { - enum slow_path_reason slow; - - action_xlate_ctx_init(&ctx, ofproto, &facet->flow, - subfacet->initial_tci, new_rule, 0, NULL); - xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len, - &odp_actions); - - slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; - if (subfacet_should_install(subfacet, slow, &odp_actions)) { - struct dpif_flow_stats stats; + /* A facet's slow path reason should only change under dramatic + * circumstances. Rather than try to update everything, it's simpler to + * remove the facet and start over. + * + * More importantly, if a facet's wildcards change, it will be relatively + * difficult to figure out if its subfacets still belong to it, and if not + * which facet they may belong to. Again, to avoid the complexity, we + * simply give up instead. */ + if (facet->xout.slow != xout.slow + || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) { + facet_remove(facet); + xlate_out_uninit(&xout); + return false; + } - subfacet_install(subfacet, - odp_actions.data, odp_actions.size, &stats, slow); - subfacet_update_stats(subfacet, &stats); + if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) { + LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) { + if (subfacet->path == SF_FAST_PATH) { + struct dpif_flow_stats stats; - if (!new_actions) { - new_actions = xcalloc(list_size(&facet->subfacets), - sizeof *new_actions); + subfacet_install(subfacet, &xout.odp_actions, &stats); + subfacet_update_stats(subfacet, &stats); } - new_actions[i].odp_actions = xmemdup(odp_actions.data, - odp_actions.size); - new_actions[i].actions_len = odp_actions.size; } - i++; - } - ofpbuf_uninit(&odp_actions); - - if (new_actions) { facet_flush_stats(facet); - } - - /* Update 'facet' now that we've taken care of all the old state. */ - facet->tags = ctx.tags; - facet->nf_flow.output_iface = ctx.nf_output_iface; - facet->has_learn = ctx.has_learn; - facet->has_normal = ctx.has_normal; - facet->has_fin_timeout = ctx.has_fin_timeout; - facet->mirrors = ctx.mirrors; - - i = 0; - LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { - subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; - if (new_actions && new_actions[i].odp_actions) { - free(subfacet->actions); - subfacet->actions = new_actions[i].odp_actions; - subfacet->actions_len = new_actions[i].actions_len; - } - i++; + ofpbuf_clear(&facet->xout.odp_actions); + ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data, + xout.odp_actions.size); } - free(new_actions); - if (facet->rule != new_rule) { - COVERAGE_INC(facet_changed_rule); - list_remove(&facet->list_node); - list_push_back(&new_rule->facets, &facet->list_node); - facet->rule = new_rule; - facet->used = new_rule->up.created; - facet->prev_used = facet->used; - } + /* Update 'facet' now that we've taken care of all the old state. */ + facet->xout.tags = xout.tags; + facet->xout.slow = xout.slow; + facet->xout.has_learn = xout.has_learn; + facet->xout.has_normal = xout.has_normal; + facet->xout.has_fin_timeout = xout.has_fin_timeout; + facet->xout.nf_output_iface = xout.nf_output_iface; + facet->xout.mirrors = xout.mirrors; + facet->nf_flow.output_iface = facet->xout.nf_output_iface; + facet->used = MAX(facet->used, new_rule->up.created); + facet->fail_open = new_rule->up.cr.priority == FAIL_OPEN_PRIORITY; + + xlate_out_uninit(&xout); + return true; } -/* Updates 'facet''s used time. Caller is responsible for calling - * facet_push_stats() to update the flows which 'facet' resubmits into. */ static void -facet_update_time(struct facet *facet, long long int used) -{ - struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); - if (used > facet->used) { - facet->used = used; - ofproto_rule_update_used(&facet->rule->up, used); - netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used); - } -} - -static void -facet_reset_counters(struct facet *facet) +facet_reset_counters(struct facet *facet) { facet->packet_count = 0; facet->byte_count = 0; @@ -4581,72 +4887,102 @@ facet_reset_counters(struct facet *facet) } static void -facet_push_stats(struct facet *facet) +facet_push_stats(struct facet *facet, bool may_learn) { 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; stats.used = facet->used; - stats.tcp_flags = 0; + stats.tcp_flags = facet->tcp_flags; + + if (may_learn || stats.n_packets || facet->used > facet->prev_used) { + struct ofproto_dpif *ofproto = facet->ofproto; + struct ofport_dpif *in_port; + struct rule_dpif *rule; + struct xlate_in xin; - if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) { facet->prev_packet_count = facet->packet_count; facet->prev_byte_count = facet->byte_count; facet->prev_used = facet->used; - flow_push_stats(facet->rule, &facet->flow, &stats); + in_port = get_ofp_port(ofproto, facet->flow.in_port.ofp_port); + if (in_port && in_port->is_tunnel) { + netdev_vport_inc_rx(in_port->up.netdev, &stats); + } + + rule = rule_dpif_lookup(ofproto, &facet->flow, NULL); + rule_credit_stats(rule, &stats); + netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, + facet->used); + netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags); + mirror_update_stats(ofproto->mbridge, facet->xout.mirrors, + stats.n_packets, stats.n_bytes); - update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto), - facet->mirrors, stats.n_packets, stats.n_bytes); + xlate_in_init(&xin, ofproto, &facet->flow, rule, stats.tcp_flags, + NULL); + xin.resubmit_stats = &stats; + xin.may_learn = may_learn; + xlate_actions_for_side_effects(&xin); } } static void -rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats) +push_all_stats__(bool run_fast) { - rule->packet_count += stats->n_packets; - rule->byte_count += stats->n_bytes; - ofproto_rule_update_used(&rule->up, stats->used); + 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 cls_cursor cursor; + struct facet *facet; + + cls_cursor_init(&cursor, &ofproto->facets, NULL); + CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { + facet_push_stats(facet, false); + if (run_fast) { + run_fast_rl(); + } + } + } + + rl = time_msec() + 100; } -/* Pushes flow statistics to the rules which 'flow' resubmits into given - * 'rule''s actions and mirrors. */ static void -flow_push_stats(struct rule_dpif *rule, - const struct flow *flow, const struct dpif_flow_stats *stats) +push_all_stats(void) { - struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); - struct action_xlate_ctx ctx; + push_all_stats__(true); +} +void +rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats) +{ + rule->packet_count += stats->n_packets; + rule->byte_count += stats->n_bytes; ofproto_rule_update_used(&rule->up, stats->used); - - action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule, - 0, NULL); - ctx.resubmit_stats = stats; - xlate_actions_for_side_effects(&ctx, rule->up.ofpacts, - rule->up.ofpacts_len); } /* 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 dpif_backer *backer, 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)) { + &backer->subfacets) { + if (subfacet->key_len == key_len + && !memcmp(key, subfacet->key, key_len)) { return subfacet; } } @@ -4657,16 +4993,12 @@ subfacet_find(struct ofproto_dpif *ofproto, /* Searches 'facet' (within 'ofproto') for a subfacet with the specified * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the * existing subfacet if there is one, otherwise creates and returns a - * new subfacet. - * - * 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(). */ + * new subfacet. */ static struct subfacet * subfacet_create(struct facet *facet, struct flow_miss *miss, long long int now) { - struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); + struct dpif_backer *backer = miss->ofproto->backer; enum odp_key_fitness key_fitness = miss->key_fitness; const struct nlattr *key = miss->key; size_t key_len = miss->key_len; @@ -4678,8 +5010,7 @@ subfacet_create(struct facet *facet, struct flow_miss *miss, if (list_is_empty(&facet->subfacets)) { subfacet = &facet->one_subfacet; } else { - subfacet = subfacet_find(ofproto, key, key_len, key_hash, - &facet->flow); + subfacet = subfacet_find(backer, key, key_len, key_hash); if (subfacet) { if (subfacet->facet == facet) { return subfacet; @@ -4693,29 +5024,20 @@ subfacet_create(struct facet *facet, struct flow_miss *miss, subfacet = xmalloc(sizeof *subfacet); } - hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash); + hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash); list_push_back(&facet->subfacets, &subfacet->list_node); subfacet->facet = facet; subfacet->key_fitness = key_fitness; - if (key_fitness != ODP_FIT_PERFECT) { - subfacet->key = xmemdup(key, key_len); - subfacet->key_len = key_len; - } else { - subfacet->key = NULL; - subfacet->key_len = 0; - } + subfacet->key = xmemdup(key, key_len); + subfacet->key_len = key_len; subfacet->used = now; + subfacet->created = now; subfacet->dp_packet_count = 0; subfacet->dp_byte_count = 0; - subfacet->actions_len = 0; - subfacet->actions = NULL; - subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE - ? SLOW_MATCH - : 0); subfacet->path = SF_NOT_INSTALLED; - subfacet->initial_tci = miss->initial_tci; - subfacet->odp_in_port = miss->odp_in_port; + subfacet->backer = backer; + backer->subfacet_add_count++; return subfacet; } @@ -4725,13 +5047,15 @@ static void subfacet_destroy__(struct subfacet *subfacet) { struct facet *facet = subfacet->facet; - struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); + struct ofproto_dpif *ofproto = facet->ofproto; + + /* Update ofproto stats before uninstall the subfacet. */ + ofproto->backer->subfacet_del_count++; subfacet_uninstall(subfacet); - hmap_remove(&ofproto->subfacets, &subfacet->hmap_node); + hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node); list_remove(&subfacet->list_node); free(subfacet->key); - free(subfacet->actions); if (subfacet != &facet->one_subfacet) { free(subfacet); } @@ -4753,80 +5077,28 @@ subfacet_destroy(struct subfacet *subfacet) } static void -subfacet_destroy_batch(struct ofproto_dpif *ofproto, +subfacet_destroy_batch(struct dpif_backer *backer, struct subfacet **subfacets, int n) { - struct odputil_keybuf keybufs[SUBFACET_DESTROY_MAX_BATCH]; struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH]; struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH]; - struct ofpbuf keys[SUBFACET_DESTROY_MAX_BATCH]; struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH]; int i; for (i = 0; i < n; i++) { ops[i].type = DPIF_OP_FLOW_DEL; - subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]); - ops[i].u.flow_del.key = keys[i].data; - ops[i].u.flow_del.key_len = keys[i].size; + ops[i].u.flow_del.key = subfacets[i]->key; + ops[i].u.flow_del.key_len = subfacets[i]->key_len; ops[i].u.flow_del.stats = &stats[i]; opsp[i] = &ops[i]; } - dpif_operate(ofproto->backer->dpif, opsp, n); + dpif_operate(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]); - } -} - -/* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes - * that can be used to refer to 'subfacet'. The caller must provide 'keybuf' - * for use as temporary storage. */ -static void -subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf, - struct ofpbuf *key) -{ - - if (!subfacet->key) { - struct flow *flow = &subfacet->facet->flow; - - ofpbuf_use_stack(key, keybuf, sizeof *keybuf); - odp_flow_key_from_flow(key, flow, subfacet->odp_in_port); - } else { - ofpbuf_use_const(key, subfacet->key, subfacet->key_len); - } -} - -/* Composes the datapath actions for 'subfacet' based on its rule's actions. - * Translates the actions into 'odp_actions', which the caller must have - * initialized and is responsible for uninitializing. */ -static void -subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet, - struct ofpbuf *odp_actions) -{ - struct facet *facet = subfacet->facet; - struct rule_dpif *rule = facet->rule; - struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); - - struct action_xlate_ctx ctx; - - action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci, - rule, 0, packet); - xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions); - facet->tags = ctx.tags; - facet->has_learn = ctx.has_learn; - facet->has_normal = ctx.has_normal; - facet->has_fin_timeout = ctx.has_fin_timeout; - facet->nf_flow.output_iface = ctx.nf_output_iface; - facet->mirrors = ctx.mirrors; - - subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; - if (subfacet->actions_len != odp_actions->size - || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) { - free(subfacet->actions); - subfacet->actions_len = odp_actions->size; - subfacet->actions = xmemdup(odp_actions->data, odp_actions->size); + run_fast_rl(); } } @@ -4837,75 +5109,73 @@ subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet, * * Returns 0 if successful, otherwise a positive errno value. */ static int -subfacet_install(struct subfacet *subfacet, - const struct nlattr *actions, size_t actions_len, - struct dpif_flow_stats *stats, - enum slow_path_reason slow) +subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions, + struct dpif_flow_stats *stats) { struct facet *facet = subfacet->facet; - struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); - enum subfacet_path path = subfacet_want_path(slow); + enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH; + const struct nlattr *actions = odp_actions->data; + size_t actions_len = odp_actions->size; + struct odputil_keybuf maskbuf; + struct ofpbuf mask; + 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; + flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE + : DPIF_FP_MODIFY; if (stats) { flags |= DPIF_FP_ZERO_STATS; } if (path == SF_SLOW_PATH) { - compose_slow_path(ofproto, &facet->flow, slow, + compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow, slow_path_stub, sizeof slow_path_stub, &actions, &actions_len); } - subfacet_get_key(subfacet, &keybuf, &key); - ret = dpif_flow_put(ofproto->backer->dpif, flags, key.data, key.size, + ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf); + if (enable_megaflows) { + odp_flow_key_from_mask(&mask, &facet->xout.wc.masks, + &facet->flow, UINT32_MAX); + } + + ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key, + subfacet->key_len, mask.data, mask.size, actions, actions_len, stats); if (stats) { subfacet_reset_dp_stats(subfacet, stats); } - if (!ret) { + if (ret) { + COVERAGE_INC(subfacet_install_fail); + } else { subfacet->path = path; } return ret; } -static int -subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats) -{ - return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len, - stats, subfacet->slow); -} - /* If 'subfacet' is installed in the datapath, uninstalls it. */ static void 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 ofproto_dpif *ofproto = subfacet->facet->ofproto; struct dpif_flow_stats stats; - struct ofpbuf key; int error; - subfacet_get_key(subfacet, &keybuf, &key); - error = dpif_flow_del(ofproto->backer->dpif, - key.data, key.size, &stats); + error = dpif_flow_del(ofproto->backer->dpif, subfacet->key, + subfacet->key_len, &stats); subfacet_reset_dp_stats(subfacet, &stats); if (!error) { subfacet_update_stats(subfacet, &stats); } subfacet->path = SF_NOT_INSTALLED; } else { - assert(subfacet->dp_packet_count == 0); - assert(subfacet->dp_byte_count == 0); + ovs_assert(subfacet->dp_packet_count == 0); + ovs_assert(subfacet->dp_byte_count == 0); } } @@ -4929,17 +5199,6 @@ subfacet_reset_dp_stats(struct subfacet *subfacet, subfacet->dp_byte_count = 0; } -/* Updates 'subfacet''s used time. The caller is responsible for calling - * facet_push_stats() to update the flows which 'subfacet' resubmits into. */ -static void -subfacet_update_time(struct subfacet *subfacet, long long int used) -{ - if (used > subfacet->used) { - subfacet->used = used; - facet_update_time(subfacet->facet, used); - } -} - /* Folds the statistics from 'stats' into the counters in 'subfacet'. * * Because of the meaning of a subfacet's counters, it only makes sense to do @@ -4953,23 +5212,25 @@ subfacet_update_stats(struct subfacet *subfacet, if (stats->n_packets || stats->used > subfacet->used) { struct facet *facet = subfacet->facet; - subfacet_update_time(subfacet, stats->used); + subfacet->used = MAX(subfacet->used, stats->used); + facet->used = MAX(facet->used, stats->used); facet->packet_count += stats->n_packets; facet->byte_count += stats->n_bytes; facet->tcp_flags |= stats->tcp_flags; - facet_push_stats(facet); - netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); } } /* Rules. */ +/* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets + * the fields that were relevant as part of the lookup. */ static struct rule_dpif * -rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow) +rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow, + struct flow_wildcards *wc) { struct rule_dpif *rule; - rule = rule_dpif_lookup__(ofproto, flow, 0); + rule = rule_dpif_lookup_in_table(ofproto, flow, wc, 0); if (rule) { return rule; } @@ -4977,40 +5238,52 @@ rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow) return rule_dpif_miss_rule(ofproto, flow); } -static struct rule_dpif * -rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow, - uint8_t table_id) +struct rule_dpif * +rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, + const struct flow *flow, struct flow_wildcards *wc, + uint8_t table_id) { struct cls_rule *cls_rule; struct classifier *cls; + bool frag; if (table_id >= N_TABLES) { return NULL; } + if (wc) { + memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type); + wc->masks.nw_frag |= FLOW_NW_FRAG_MASK; + } + cls = &ofproto->up.tables[table_id].cls; - if (flow->nw_frag & FLOW_NW_FRAG_ANY - && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) { - /* For OFPC_NORMAL frag_handling, we must pretend that transport ports - * are unavailable. */ + frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0; + if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) { + /* We must pretend that transport ports are unavailable. */ struct flow ofpc_normal_flow = *flow; ofpc_normal_flow.tp_src = htons(0); ofpc_normal_flow.tp_dst = htons(0); - cls_rule = classifier_lookup(cls, &ofpc_normal_flow); + cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc); + } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) { + cls_rule = &ofproto->drop_frags_rule->up.cr; + if (wc) { + flow_wildcards_init_exact(wc); + } } else { - cls_rule = classifier_lookup(cls, flow); + cls_rule = classifier_lookup(cls, flow, wc); } return rule_dpif_cast(rule_from_cls_rule(cls_rule)); } -static struct rule_dpif * +struct rule_dpif * rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow) { struct ofport_dpif *port; - port = get_ofp_port(ofproto, flow->in_port); + port = get_ofp_port(ofproto, flow->in_port.ofp_port); if (!port) { - VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port); + VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16, + flow->in_port.ofp_port); return ofproto->miss_rule; } @@ -5060,28 +5333,8 @@ rule_construct(struct rule *rule_) rule->packet_count = 0; rule->byte_count = 0; - victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending)); - if (victim && !list_is_empty(&victim->facets)) { - struct facet *facet; - - rule->facets = victim->facets; - list_moved(&rule->facets); - LIST_FOR_EACH (facet, list_node, &rule->facets) { - /* XXX: We're only clearing our local counters here. It's possible - * that quite a few packets are unaccounted for in the datapath - * statistics. These will be accounted to the new rule instead of - * cleared as required. This could be fixed by clearing out the - * datapath statistics for this facet, but currently it doesn't - * seem worth it. */ - facet_reset_counters(facet); - facet->rule = rule; - } - } else { - /* Must avoid list_moved() in this case. */ - list_init(&rule->facets); - } - table_id = rule->up.table_id; + victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending)); if (victim) { rule->tag = victim->tag; } else if (table_id == 0) { @@ -5099,65 +5352,56 @@ rule_construct(struct rule *rule_) } static void -rule_destruct(struct rule *rule_) +rule_destruct(struct rule *rule) { - struct rule_dpif *rule = rule_dpif_cast(rule_); - struct facet *facet, *next_facet; - - LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { - facet_revalidate(facet); - } - - complete_operation(rule); + complete_operation(rule_dpif_cast(rule)); } static void rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) { struct rule_dpif *rule = rule_dpif_cast(rule_); - struct facet *facet; + + /* push_all_stats() can handle flow misses which, when using the learn + * action, can cause rules to be added and deleted. This can corrupt our + * caller's datastructures which assume that rule_get_stats() doesn't have + * an impact on the flow table. To be safe, we disable miss handling. */ + push_all_stats__(false); /* Start from historical data for 'rule' itself that are no longer tracked * in facets. This counts, for example, facets that have expired. */ *packets = rule->packet_count; *bytes = rule->byte_count; - - /* Add any statistics that are tracked by facets. This includes - * statistical data recently updated by ofproto_update_stats() as well as - * stats for packets that were executed "by hand" via dpif_execute(). */ - LIST_FOR_EACH (facet, list_node, &rule->facets) { - *packets += facet->packet_count; - *bytes += facet->byte_count; - } } -static enum ofperr -rule_execute(struct rule *rule_, const struct flow *flow, - struct ofpbuf *packet) +static void +rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow, + struct ofpbuf *packet) { - struct rule_dpif *rule = rule_dpif_cast(rule_); struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); - struct dpif_flow_stats stats; - - struct action_xlate_ctx ctx; - uint64_t odp_actions_stub[1024 / 8]; - struct ofpbuf odp_actions; + struct xlate_out xout; + struct xlate_in xin; dpif_flow_stats_extract(flow, packet, time_msec(), &stats); rule_credit_stats(rule, &stats); - ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); - action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, - rule, stats.tcp_flags, packet); - ctx.resubmit_stats = &stats; - xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions); + xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet); + xin.resubmit_stats = &stats; + xlate_actions(&xin, &xout); - execute_odp_actions(ofproto, flow, odp_actions.data, - odp_actions.size, packet); + execute_odp_actions(ofproto, flow, xout.odp_actions.data, + xout.odp_actions.size, packet); - ofpbuf_uninit(&odp_actions); + xlate_out_uninit(&xout); +} +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; } @@ -5171,1588 +5415,126 @@ rule_modify_actions(struct rule *rule_) /* Sends 'packet' out 'ofport'. * May modify 'packet'. - * Returns 0 if successful, otherwise a positive errno value. */ -static int -send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) -{ - const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); - struct ofpbuf key, odp_actions; - struct odputil_keybuf keybuf; - uint32_t odp_port; - struct flow flow; - int error; - - flow_extract(packet, 0, NULL, OFPP_LOCAL, &flow); - odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port, - flow.vlan_tci); - if (odp_port != ofport->odp_port) { - eth_pop_vlan(packet); - flow.vlan_tci = htons(0); - } - - ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); - odp_flow_key_from_flow(&key, &flow, - ofp_port_to_odp_port(ofproto, flow.in_port)); - - ofpbuf_init(&odp_actions, 32); - compose_sflow_action(ofproto, &odp_actions, &flow, odp_port); - - nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port); - error = dpif_execute(ofproto->backer->dpif, - key.data, key.size, - odp_actions.data, odp_actions.size, - packet); - ofpbuf_uninit(&odp_actions); - - if (error) { - VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)", - ofproto->up.name, odp_port, strerror(error)); - } - ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0); - return error; -} - -/* OpenFlow to datapath action translation. */ - -static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len, - struct action_xlate_ctx *); -static void xlate_normal(struct action_xlate_ctx *); - -/* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'. - * The action will state 'slow' as the reason that the action is in the slow - * path. (This is purely informational: it allows a human viewing "ovs-dpctl - * dump-flows" output to see why a flow is in the slow path.) - * - * The 'stub_size' bytes in 'stub' will be used to store the action. - * 'stub_size' must be large enough for the action. - * - * The action and its size will be stored in '*actionsp' and '*actions_lenp', - * respectively. */ -static void -compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow, - enum slow_path_reason slow, - uint64_t *stub, size_t stub_size, - const struct nlattr **actionsp, size_t *actions_lenp) -{ - union user_action_cookie cookie; - struct ofpbuf buf; - - cookie.type = USER_ACTION_COOKIE_SLOW_PATH; - cookie.slow_path.unused = 0; - cookie.slow_path.reason = slow; - - ofpbuf_use_stack(&buf, stub, stub_size); - if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) { - uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT16_MAX); - odp_put_userspace_action(pid, &cookie, &buf); - } else { - put_userspace_action(ofproto, &buf, flow, &cookie); - } - *actionsp = buf.data; - *actions_lenp = buf.size; -} - -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) -{ - uint32_t pid; - - pid = dpif_port_get_pid(ofproto->backer->dpif, - ofp_port_to_odp_port(ofproto, flow->in_port)); - - return odp_put_userspace_action(pid, cookie, odp_actions); -} - -static void -compose_sflow_cookie(const struct ofproto_dpif *ofproto, - ovs_be16 vlan_tci, uint32_t odp_port, - unsigned int n_outputs, union user_action_cookie *cookie) -{ - int ifindex; - - cookie->type = USER_ACTION_COOKIE_SFLOW; - cookie->sflow.vlan_tci = vlan_tci; - - /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output - * port information") for the interpretation of cookie->output. */ - switch (n_outputs) { - case 0: - /* 0x40000000 | 256 means "packet dropped for unknown reason". */ - cookie->sflow.output = 0x40000000 | 256; - break; - - case 1: - ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port); - if (ifindex) { - cookie->sflow.output = ifindex; - break; - } - /* Fall through. */ - default: - /* 0x80000000 means "multiple output ports. */ - cookie->sflow.output = 0x80000000 | n_outputs; - break; - } -} - -/* Compose SAMPLE action for sFlow. */ -static size_t -compose_sflow_action(const struct ofproto_dpif *ofproto, - struct ofpbuf *odp_actions, - const struct flow *flow, - uint32_t odp_port) -{ - 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; -} - -/* SAMPLE action must be first action in any given list of actions. - * At this point we do not have all information required to build it. So try to - * build sample action as complete as possible. */ -static void -add_sflow_action(struct action_xlate_ctx *ctx) -{ - ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto, - ctx->odp_actions, - &ctx->flow, OVSP_NONE); - ctx->sflow_odp_port = 0; - ctx->sflow_n_outputs = 0; -} - -/* 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. */ -static void -fix_sflow_action(struct action_xlate_ctx *ctx) -{ - const struct flow *base = &ctx->base_flow; - union user_action_cookie *cookie; - - if (!ctx->user_cookie_offset) { - return; - } - - cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset, - sizeof(*cookie)); - assert(cookie->type == USER_ACTION_COOKIE_SFLOW); - - compose_sflow_cookie(ctx->ofproto, base->vlan_tci, - ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie); -} - -static void -compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port, - bool check_stp) -{ - const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port); - uint32_t odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port); - ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci; - uint8_t flow_nw_tos = ctx->flow.nw_tos; - uint16_t out_port; - - if (ofport) { - struct priority_to_dscp *pdscp; - - 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; - } - - pdscp = get_priority(ofport, ctx->flow.skb_priority); - if (pdscp) { - ctx->flow.nw_tos &= ~IP_DSCP_MASK; - ctx->flow.nw_tos |= pdscp->dscp; - } - } 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. */ - } - - 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); - } - commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions); - nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port); - - ctx->sflow_odp_port = odp_port; - ctx->sflow_n_outputs++; - ctx->nf_output_iface = ofp_port; - ctx->flow.vlan_tci = flow_vlan_tci; - ctx->flow.nw_tos = flow_nw_tos; -} - -static void -compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port) -{ - compose_output_action__(ctx, ofp_port, true); -} - -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; - - 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)); - } - } - - /* 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); - } - - 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); - ctx->rule = old_rule; - ctx->recurse--; - } - - ctx->table_id = old_table_id; - } else { - static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); - - VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times", - MAX_RESUBMIT_RECURSION); - ctx->max_resubmit_trigger = true; - } -} - -static void -xlate_ofpact_resubmit(struct action_xlate_ctx *ctx, - const struct ofpact_resubmit *resubmit) -{ - uint16_t in_port; - uint8_t table_id; - - in_port = resubmit->in_port; - if (in_port == OFPP_IN_PORT) { - in_port = ctx->flow.in_port; - } - - table_id = resubmit->table_id; - if (table_id == 255) { - table_id = ctx->table_id; - } - - xlate_table_action(ctx, in_port, table_id, false); -} - -static void -flood_packets(struct action_xlate_ctx *ctx, bool all) -{ - struct ofport_dpif *ofport; - - HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) { - uint16_t ofp_port = ofport->up.ofp_port; - - if (ofp_port == ctx->flow.in_port) { - continue; - } - - if (all) { - compose_output_action__(ctx, ofp_port, false); - } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) { - compose_output_action(ctx, ofp_port); - } - } - - ctx->nf_output_iface = NF_OUT_FLOOD; -} - -static void -execute_controller_action(struct action_xlate_ctx *ctx, int len, - enum ofp_packet_in_reason reason, - uint16_t controller_id) -{ - struct ofputil_packet_in pin; - struct ofpbuf *packet; - - ctx->slow |= SLOW_CONTROLLER; - if (!ctx->packet) { - return; - } - - packet = ofpbuf_clone(ctx->packet); - - if (packet->l2 && packet->l3) { - struct eth_header *eh; - - 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); - - if (ctx->flow.vlan_tci & htons(VLAN_CFI)) { - eth_push_vlan(packet, ctx->flow.vlan_tci); - } - - if (packet->l4) { - if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { - packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst, - ctx->flow.nw_tos, ctx->flow.nw_ttl); - } - - if (packet->l7) { - if (ctx->flow.nw_proto == IPPROTO_TCP) { - packet_set_tcp_port(packet, ctx->flow.tp_src, - ctx->flow.tp_dst); - } else if (ctx->flow.nw_proto == IPPROTO_UDP) { - packet_set_udp_port(packet, ctx->flow.tp_src, - ctx->flow.tp_dst); - } - } - } - } - - pin.packet = packet->data; - pin.packet_len = packet->size; - pin.reason = reason; - pin.controller_id = controller_id; - pin.table_id = ctx->table_id; - pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0; - - pin.send_len = len; - flow_get_metadata(&ctx->flow, &pin.fmd); - - connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin); - ofpbuf_delete(packet); -} - -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 void -xlate_output_action(struct action_xlate_ctx *ctx, - uint16_t port, uint16_t max_len, bool may_packet_in) -{ - uint16_t prev_nf_output_iface = ctx->nf_output_iface; - - ctx->nf_output_iface = NF_OUT_DROP; - - switch (port) { - case OFPP_IN_PORT: - compose_output_action(ctx, ctx->flow.in_port); - break; - case OFPP_TABLE: - xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in); - break; - case OFPP_NORMAL: - xlate_normal(ctx); - break; - case OFPP_FLOOD: - flood_packets(ctx, false); - break; - case OFPP_ALL: - flood_packets(ctx, true); - break; - case OFPP_CONTROLLER: - execute_controller_action(ctx, max_len, OFPR_ACTION, 0); - break; - case OFPP_NONE: - break; - case OFPP_LOCAL: - default: - if (port != ctx->flow.in_port) { - compose_output_action(ctx, port); - } else { - xlate_report(ctx, "skipping output to input port"); - } - break; - } - - if (prev_nf_output_iface == NF_OUT_FLOOD) { - ctx->nf_output_iface = NF_OUT_FLOOD; - } else if (ctx->nf_output_iface == NF_OUT_DROP) { - ctx->nf_output_iface = prev_nf_output_iface; - } else if (prev_nf_output_iface != NF_OUT_DROP && - ctx->nf_output_iface != NF_OUT_FLOOD) { - ctx->nf_output_iface = NF_OUT_MULTI; - } -} - -static void -xlate_output_reg_action(struct action_xlate_ctx *ctx, - const struct ofpact_output_reg *or) -{ - uint64_t port = mf_get_subfield(&or->src, &ctx->flow); - if (port <= UINT16_MAX) { - xlate_output_action(ctx, port, or->max_len, false); - } -} - -static void -xlate_enqueue_action(struct action_xlate_ctx *ctx, - const struct ofpact_enqueue *enqueue) -{ - uint16_t ofp_port = enqueue->port; - uint32_t queue_id = enqueue->queue; - uint32_t flow_priority, priority; - int error; - - /* Translate queue to 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); - return; - } - - /* Check output port. */ - if (ofp_port == OFPP_IN_PORT) { - ofp_port = ctx->flow.in_port; - } else if (ofp_port == ctx->flow.in_port) { - return; - } - - /* Add datapath actions. */ - flow_priority = ctx->flow.skb_priority; - ctx->flow.skb_priority = priority; - compose_output_action(ctx, ofp_port); - ctx->flow.skb_priority = flow_priority; - - /* Update NetFlow output port. */ - if (ctx->nf_output_iface == NF_OUT_DROP) { - ctx->nf_output_iface = ofp_port; - } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { - ctx->nf_output_iface = NF_OUT_MULTI; - } -} - -static void -xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id) -{ - uint32_t 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 - * has already been logged. */ - } -} - -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_) -{ - struct ofproto_dpif *ofproto = ofproto_; - struct ofport_dpif *port; - - switch (ofp_port) { - case OFPP_IN_PORT: - case OFPP_TABLE: - case OFPP_NORMAL: - case OFPP_FLOOD: - case OFPP_ALL: - case OFPP_NONE: - return true; - case OFPP_CONTROLLER: /* Not supported by the bundle action. */ - return false; - default: - port = get_ofp_port(ofproto, ofp_port); - return port ? port->may_enable : false; - } -} - -static void -xlate_bundle_action(struct action_xlate_ctx *ctx, - const struct ofpact_bundle *bundle) -{ - uint16_t port; - - port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto); - if (bundle->dst.field) { - nxm_reg_load(&bundle->dst, port, &ctx->flow); - } else { - xlate_output_action(ctx, port, 0, false); - } -} - -static void -xlate_learn_action(struct action_xlate_ctx *ctx, - const struct ofpact_learn *learn) -{ - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1); - struct ofputil_flow_mod fm; - uint64_t ofpacts_stub[1024 / 8]; - struct ofpbuf ofpacts; - int error; - - ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); - learn_execute(learn, &ctx->flow, &fm, &ofpacts); - - error = ofproto_flow_mod(&ctx->ofproto->up, &fm); - if (error && !VLOG_DROP_WARN(&rl)) { - VLOG_WARN("learning action failed to modify flow table (%s)", - ofperr_get_name(error)); - } - - ofpbuf_uninit(&ofpacts); -} - -/* Reduces '*timeout' to no more than 'max'. A value of zero in either case - * means "infinite". */ -static void -reduce_timeout(uint16_t max, uint16_t *timeout) -{ - if (max && (!*timeout || *timeout > max)) { - *timeout = max; - } -} - -static void -xlate_fin_timeout(struct action_xlate_ctx *ctx, - const struct ofpact_fin_timeout *oft) -{ - if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) { - struct rule_dpif *rule = ctx->rule; - - reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout); - reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout); - } -} - -static bool -may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx) -{ - if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) - ? OFPUTIL_PC_NO_RECV_STP - : OFPUTIL_PC_NO_RECV)) { - return false; - } - - /* Only drop packets here if both forwarding and learning are - * disabled. If just learning is enabled, we need to have - * OFPP_NORMAL and the learning action have a look at the packet - * before we can drop it. */ - if (!stp_forward_in_state(port->stp_state) - && !stp_learn_in_state(port->stp_state)) { - return false; - } - - 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; - } - OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) { - struct ofpact_controller *controller; - const struct ofpact_metadata *metadata; - - if (ctx->exit) { - break; - } - - switch (a->type) { - case OFPACT_OUTPUT: - xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port, - ofpact_get_OUTPUT(a)->max_len, true); - break; - - case OFPACT_CONTROLLER: - controller = ofpact_get_CONTROLLER(a); - execute_controller_action(ctx, controller->max_len, - controller->reason, - controller->controller_id); - break; - - case OFPACT_ENQUEUE: - xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a)); - break; - - case OFPACT_SET_VLAN_VID: - ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK); - ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid) - | htons(VLAN_CFI)); - break; - - case OFPACT_SET_VLAN_PCP: - ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); - ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp - << VLAN_PCP_SHIFT) - | VLAN_CFI); - break; - - case OFPACT_STRIP_VLAN: - ctx->flow.vlan_tci = htons(0); - break; - - case OFPACT_PUSH_VLAN: - /* TODO:XXX 802.1AD(QinQ) */ - ctx->flow.vlan_tci = htons(VLAN_CFI); - break; - - case OFPACT_SET_ETH_SRC: - memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac, - ETH_ADDR_LEN); - break; - - case OFPACT_SET_ETH_DST: - memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac, - ETH_ADDR_LEN); - break; - - case OFPACT_SET_IPV4_SRC: - 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; - break; - - case OFPACT_SET_IPV4_DSCP: - /* OpenFlow 1.0 only supports IPv4. */ - if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { - ctx->flow.nw_tos &= ~IP_DSCP_MASK; - ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp; - } - break; - - case OFPACT_SET_L4_SRC_PORT: - 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); - break; - - case OFPACT_RESUBMIT: - xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a)); - break; - - case OFPACT_SET_TUNNEL: - ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id); - break; - - case OFPACT_SET_QUEUE: - xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id); - break; - - case OFPACT_POP_QUEUE: - ctx->flow.skb_priority = ctx->orig_skb_priority; - break; - - case OFPACT_REG_MOVE: - nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow); - break; - - case OFPACT_REG_LOAD: - nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow); - break; - - case OFPACT_DEC_TTL: - if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) { - goto out; - } - break; - - case OFPACT_NOTE: - /* Nothing to do. */ - break; - - case OFPACT_MULTIPATH: - 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)); - break; - - case OFPACT_OUTPUT_REG: - xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a)); - break; - - case OFPACT_LEARN: - ctx->has_learn = true; - if (ctx->may_learn) { - xlate_learn_action(ctx, ofpact_get_LEARN(a)); - } - break; - - case OFPACT_EXIT: - ctx->exit = true; - break; - - case OFPACT_FIN_TIMEOUT: - ctx->has_fin_timeout = true; - xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a)); - break; - - case OFPACT_CLEAR_ACTIONS: - /* TODO: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. - */ - break; - - case OFPACT_WRITE_METADATA: - metadata = ofpact_get_WRITE_METADATA(a); - ctx->flow.metadata &= ~metadata->mask; - ctx->flow.metadata |= metadata->metadata & metadata->mask; - break; - - case OFPACT_GOTO_TABLE: { - /* TODO:XXX remove recursion */ - /* It is assumed that goto-table is 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); - 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; - } -} - -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, - uint8_t tcp_flags, const struct ofpbuf *packet) -{ - ctx->ofproto = ofproto; - ctx->flow = *flow; - ctx->base_flow = ctx->flow; - memset(&ctx->base_flow.tunnel, 0, sizeof ctx->base_flow.tunnel); - ctx->base_flow.vlan_tci = initial_tci; - ctx->rule = rule; - ctx->packet = packet; - ctx->may_learn = packet != NULL; - ctx->tcp_flags = tcp_flags; - ctx->resubmit_hook = NULL; - ctx->report_hook = NULL; - ctx->resubmit_stats = NULL; -} - -/* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts' - * into datapath actions in 'odp_actions', using 'ctx'. */ -static void -xlate_actions(struct action_xlate_ctx *ctx, - const struct ofpact *ofpacts, size_t ofpacts_len, - struct ofpbuf *odp_actions) -{ - /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so - * that in the future we always keep a copy of the original flow for - * tracing purposes. */ - static bool hit_resubmit_limit; - - enum slow_path_reason special; - - COVERAGE_INC(ofproto_dpif_xlate); - - ofpbuf_clear(odp_actions); - ofpbuf_reserve(odp_actions, NL_A_U32_SIZE); - - ctx->odp_actions = odp_actions; - ctx->tags = 0; - ctx->slow = 0; - ctx->has_learn = false; - ctx->has_normal = false; - ctx->has_fin_timeout = false; - ctx->nf_output_iface = NF_OUT_DROP; - ctx->mirrors = 0; - ctx->recurse = 0; - ctx->max_resubmit_trigger = false; - ctx->orig_skb_priority = ctx->flow.skb_priority; - ctx->table_id = 0; - ctx->exit = false; - - 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; - } - - if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) { - switch (ctx->ofproto->up.frag_handling) { - case OFPC_FRAG_NORMAL: - /* We must pretend that transport ports are unavailable. */ - ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0); - ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0); - break; - - case OFPC_FRAG_DROP: - return; - - case OFPC_FRAG_REASM: - NOT_REACHED(); - - case OFPC_FRAG_NX_MATCH: - /* Nothing to do. */ - break; - - case OFPC_INVALID_TTL_TO_CONTROLLER: - NOT_REACHED(); - } - } - - special = process_special(ctx->ofproto, &ctx->flow, 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; - - add_sflow_action(ctx); - do_xlate_actions(ofpacts, ofpacts_len, ctx); - - if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) { - if (!hit_resubmit_limit) { - /* We didn't record the original flow. Make sure we do from - * now on. */ - hit_resubmit_limit = true; - } 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); - VLOG_ERR("Trace triggered by excessive resubmit " - "recursion:\n%s", ds_cstr(&ds)); - ds_destroy(&ds); - } - } - - if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow, - ctx->odp_actions->data, - ctx->odp_actions->size)) { - ctx->slow |= SLOW_IN_BAND; - if (ctx->packet - && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow, - ctx->packet)) { - compose_output_action(ctx, OFPP_LOCAL); - } - } - if (ctx->ofproto->has_mirrors) { - add_mirror_actions(ctx, &ctx->orig_flow); - } - fix_sflow_action(ctx); - } -} - -/* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts' - * into datapath actions, using 'ctx', and discards the datapath actions. */ -static void -xlate_actions_for_side_effects(struct action_xlate_ctx *ctx, - const struct ofpact *ofpacts, - size_t ofpacts_len) -{ - uint64_t odp_actions_stub[1024 / 8]; - struct ofpbuf odp_actions; - - ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); - xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions); - ofpbuf_uninit(&odp_actions); -} - -static void -xlate_report(struct action_xlate_ctx *ctx, const char *s) -{ - if (ctx->report_hook) { - ctx->report_hook(ctx, s); - } -} - -/* OFPP_NORMAL implementation. */ - -static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *); - -/* Given 'vid', the VID obtained from the 802.1Q header that was received as - * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle', - * the bundle on which the packet was received, returns the VLAN to which the - * packet belongs. - * - * Both 'vid' and the return value are in the range 0...4095. */ -static uint16_t -input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid) -{ - switch (in_bundle->vlan_mode) { - case PORT_VLAN_ACCESS: - return in_bundle->vlan; - break; - - case PORT_VLAN_TRUNK: - return vid; - - case PORT_VLAN_NATIVE_UNTAGGED: - case PORT_VLAN_NATIVE_TAGGED: - return vid ? vid : in_bundle->vlan; - - default: - NOT_REACHED(); - } -} - -/* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'. - * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs - * a warning. - * - * 'vid' should be the VID obtained from the 802.1Q header that was received as - * part of a packet (specify 0 if there was no 802.1Q header), in the range - * 0...4095. */ -static bool -input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn) -{ - /* Allow any VID on the OFPP_NONE port. */ - if (in_bundle == &ofpp_none_bundle) { - return true; - } - - switch (in_bundle->vlan_mode) { - case PORT_VLAN_ACCESS: - if (vid) { - if (warn) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged " - "packet received on port %s configured as VLAN " - "%"PRIu16" access port", - in_bundle->ofproto->up.name, vid, - in_bundle->name, in_bundle->vlan); - } - return false; - } - return true; - - case PORT_VLAN_NATIVE_UNTAGGED: - case PORT_VLAN_NATIVE_TAGGED: - if (!vid) { - /* Port must always carry its native VLAN. */ - return true; - } - /* Fall through. */ - case PORT_VLAN_TRUNK: - if (!ofbundle_includes_vlan(in_bundle, vid)) { - if (warn) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet " - "received on port %s not configured for trunking " - "VLAN %"PRIu16, - in_bundle->ofproto->up.name, vid, - in_bundle->name, vid); - } - return false; - } - return true; - - default: - NOT_REACHED(); - } - -} - -/* Given 'vlan', the VLAN that a packet belongs to, and - * 'out_bundle', a bundle on which the packet is to be output, returns the VID - * that should be included in the 802.1Q header. (If the return value is 0, - * then the 802.1Q header should only be included in the packet if there is a - * nonzero PCP.) - * - * Both 'vlan' and the return value are in the range 0...4095. */ -static uint16_t -output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan) -{ - switch (out_bundle->vlan_mode) { - case PORT_VLAN_ACCESS: - return 0; - - case PORT_VLAN_TRUNK: - case PORT_VLAN_NATIVE_TAGGED: - return vlan; - - case PORT_VLAN_NATIVE_UNTAGGED: - return vlan == out_bundle->vlan ? 0 : vlan; - - default: - NOT_REACHED(); - } -} - -static void -output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle, - uint16_t vlan) -{ - struct ofport_dpif *port; - uint16_t vid; - ovs_be16 tci, old_tci; - - vid = output_vlan_to_vid(out_bundle, vlan); - if (!out_bundle->bond) { - port = ofbundle_get_a_port(out_bundle); - } else { - port = bond_choose_output_slave(out_bundle->bond, &ctx->flow, - vid, &ctx->tags); - if (!port) { - /* No slaves enabled, so drop packet. */ - return; - } - } - - old_tci = ctx->flow.vlan_tci; - tci = htons(vid); - if (tci || out_bundle->use_priority_tags) { - tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK); - if (tci) { - tci |= htons(VLAN_CFI); - } - } - ctx->flow.vlan_tci = tci; - - compose_output_action(ctx, port->up.ofp_port); - ctx->flow.vlan_tci = old_tci; -} - -static int -mirror_mask_ffs(mirror_mask_t mask) -{ - BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); - return ffs(mask); -} - -static bool -ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan) -{ - return (bundle->vlan_mode != PORT_VLAN_ACCESS - && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan))); -} - -static bool -ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan) -{ - return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan); -} - -/* Returns an arbitrary interface within 'bundle'. */ -static struct ofport_dpif * -ofbundle_get_a_port(const struct ofbundle *bundle) -{ - return CONTAINER_OF(list_front(&bundle->ports), - struct ofport_dpif, bundle_node); -} - -static bool -vlan_is_mirrored(const struct ofmirror *m, int vlan) -{ - return !m->vlans || bitmap_is_set(m->vlans, vlan); -} - -static void -add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow) -{ - struct ofproto_dpif *ofproto = ctx->ofproto; - mirror_mask_t mirrors; - struct ofbundle *in_bundle; - uint16_t vlan; - uint16_t vid; - const struct nlattr *a; - size_t left; - - in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port, - ctx->packet != NULL, NULL); - if (!in_bundle) { - return; - } - mirrors = in_bundle->src_mirrors; - - /* Drop frames on bundles reserved for mirroring. */ - if (in_bundle->mirror_out) { - if (ctx->packet != NULL) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " - "%s, which is reserved exclusively for mirroring", - ctx->ofproto->up.name, in_bundle->name); - } - return; - } - - /* Check VLAN. */ - vid = vlan_tci_to_vid(orig_flow->vlan_tci); - if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) { - return; - } - vlan = input_vid_to_vlan(in_bundle, vid); - - /* Look at the output ports to check for destination selections. */ - - NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data, - ctx->odp_actions->size) { - enum ovs_action_attr type = nl_attr_type(a); - struct ofport_dpif *ofport; - - if (type != OVS_ACTION_ATTR_OUTPUT) { - continue; - } - - ofport = get_odp_port(ofproto, nl_attr_get_u32(a)); - if (ofport && ofport->bundle) { - mirrors |= ofport->bundle->dst_mirrors; - } - } - - if (!mirrors) { - return; - } - - /* Restore the original packet before adding the mirror actions. */ - ctx->flow = *orig_flow; - - while (mirrors) { - struct ofmirror *m; - - m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; - - if (!vlan_is_mirrored(m, vlan)) { - mirrors = zero_rightmost_1bit(mirrors); - continue; - } - - mirrors &= ~m->dup_mirrors; - ctx->mirrors |= m->dup_mirrors; - if (m->out) { - output_normal(ctx, m->out, vlan); - } else if (vlan != m->out_vlan - && !eth_addr_is_reserved(orig_flow->dl_dst)) { - struct ofbundle *bundle; - - HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { - if (ofbundle_includes_vlan(bundle, m->out_vlan) - && !bundle->mirror_out) { - output_normal(ctx, bundle, m->out_vlan); - } - } - } - } -} - -static void -update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors, - uint64_t packets, uint64_t bytes) -{ - if (!mirrors) { - return; - } - - for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) { - struct ofmirror *m; - - m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; - - if (!m) { - /* In normal circumstances 'm' will not be NULL. However, - * if mirrors are reconfigured, we can temporarily get out - * of sync in facet_revalidate(). We could "correct" the - * mirror list before reaching here, but doing that would - * not properly account the traffic stats we've currently - * accumulated for previous mirror configuration. */ - continue; - } - - m->packet_count += packets; - m->byte_count += bytes; - } -} - -/* A VM broadcasts a gratuitous ARP to indicate that it has resumed after - * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to - * indicate this; newer upstream kernels use gratuitous ARP requests. */ -static bool -is_gratuitous_arp(const struct flow *flow) -{ - return (flow->dl_type == htons(ETH_TYPE_ARP) - && eth_addr_is_broadcast(flow->dl_dst) - && (flow->nw_proto == ARP_OP_REPLY - || (flow->nw_proto == ARP_OP_REQUEST - && flow->nw_src == flow->nw_dst))); -} - -static void -update_learning_table(struct ofproto_dpif *ofproto, - const struct flow *flow, int vlan, - struct ofbundle *in_bundle) -{ - struct mac_entry *mac; - - /* Don't learn the OFPP_NONE port. */ - if (in_bundle == &ofpp_none_bundle) { - return; - } - - if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) { - return; - } - - mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan); - if (is_gratuitous_arp(flow)) { - /* We don't want to learn from gratuitous ARP packets that are - * reflected back over bond slaves so we lock the learning table. */ - if (!in_bundle->bond) { - mac_entry_set_grat_arp_lock(mac); - } else if (mac_entry_is_grat_arp_locked(mac)) { - return; - } - } - - if (mac_entry_is_new(mac) || mac->port.p != in_bundle) { - /* The log messages here could actually be useful in debugging, - * so keep the rate limit relatively high. */ - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); - VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " - "on port %s in VLAN %d", - ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src), - in_bundle->name, vlan); - - mac->port.p = in_bundle; - tag_set_add(&ofproto->revalidate_set, - mac_learning_changed(ofproto->ml, mac)); - } -} - -static struct ofbundle * -lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port, - bool warn, struct ofport_dpif **in_ofportp) -{ - struct ofport_dpif *ofport; - - /* Find the port and bundle for the received packet. */ - ofport = get_ofp_port(ofproto, in_port); - if (in_ofportp) { - *in_ofportp = ofport; - } - if (ofport && ofport->bundle) { - return ofport->bundle; - } - - /* Special-case OFPP_NONE, which a controller may use as the ingress - * port for traffic that it is sourcing. */ - if (in_port == OFPP_NONE) { - return &ofpp_none_bundle; - } - - /* Odd. A few possible reasons here: - * - * - We deleted a port but there are still a few packets queued up - * from it. - * - * - Someone externally added a port (e.g. "ovs-dpctl add-if") that - * we don't know about. - * - * - The ofproto client didn't configure the port as part of a bundle. - * This is particularly likely to happen if a packet was received on the - * port after it was created, but before the client had a chance to - * configure its bundle. - */ - if (warn) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - - VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " - "port %"PRIu16, ofproto->up.name, in_port); - } - return NULL; -} - -/* Determines whether packets in 'flow' within 'ofproto' should be forwarded or - * dropped. Returns true if they may be forwarded, false if they should be - * dropped. - * - * 'in_port' must be the ofport_dpif that corresponds to flow->in_port. - * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull). - * - * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as - * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as - * checked by input_vid_is_valid(). - * - * May also add tags to '*tags', although the current implementation only does - * so in one special case. - */ -static bool -is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port, - uint16_t vlan) + * Returns 0 if successful, otherwise a positive errno value. */ +static int +send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) { - struct ofproto_dpif *ofproto = ctx->ofproto; - struct flow *flow = &ctx->flow; - struct ofbundle *in_bundle = in_port->bundle; + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); + uint64_t odp_actions_stub[1024 / 8]; + struct ofpbuf key, odp_actions; + struct dpif_flow_stats stats; + struct odputil_keybuf keybuf; + struct ofpact_output output; + struct xlate_out xout; + struct xlate_in xin; + struct flow flow; + union flow_in_port in_port_; + int error; - /* Drop frames for reserved multicast addresses - * only if forward_bpdu option is absent. */ - if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) { - xlate_report(ctx, "packet has reserved destination MAC, dropping"); - return false; - } + ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); + ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); - if (in_bundle->bond) { - struct mac_entry *mac; + /* Use OFPP_NONE as the in_port to avoid special packet processing. */ + in_port_.ofp_port = OFPP_NONE; + flow_extract(packet, 0, 0, NULL, &in_port_, &flow); + odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto, + OFPP_LOCAL)); + dpif_flow_stats_extract(&flow, packet, time_msec(), &stats); - switch (bond_check_admissibility(in_bundle->bond, in_port, - flow->dl_dst, &ctx->tags)) { - case BV_ACCEPT: - break; + ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output); + output.port = ofport->up.ofp_port; + output.max_len = 0; - case BV_DROP: - xlate_report(ctx, "bonding refused admissibility, dropping"); - return false; + xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet); + xin.ofpacts_len = sizeof output; + xin.ofpacts = &output.ofpact; + xin.resubmit_stats = &stats; + xlate_actions(&xin, &xout); - case BV_DROP_IF_MOVED: - mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL); - if (mac && mac->port.p != in_bundle && - (!is_gratuitous_arp(flow) - || mac_entry_is_grat_arp_locked(mac))) { - xlate_report(ctx, "SLB bond thinks this packet looped back, " - "dropping"); - return false; - } - break; - } + error = dpif_execute(ofproto->backer->dpif, + key.data, key.size, + xout.odp_actions.data, xout.odp_actions.size, + packet); + xlate_out_uninit(&xout); + + if (error) { + VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)", + ofproto->up.name, netdev_get_name(ofport->up.netdev), + ovs_strerror(error)); } - return true; + ofproto->stats.tx_packets++; + ofproto->stats.tx_bytes += packet->size; + return error; } +/* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'. + * The action will state 'slow' as the reason that the action is in the slow + * path. (This is purely informational: it allows a human viewing "ovs-dpctl + * dump-flows" output to see why a flow is in the slow path.) + * + * The 'stub_size' bytes in 'stub' will be used to store the action. + * 'stub_size' must be large enough for the action. + * + * The action and its size will be stored in '*actionsp' and '*actions_lenp', + * respectively. */ static void -xlate_normal(struct action_xlate_ctx *ctx) +compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow, + enum slow_path_reason slow, + uint64_t *stub, size_t stub_size, + const struct nlattr **actionsp, size_t *actions_lenp) { - struct ofport_dpif *in_port; - struct ofbundle *in_bundle; - struct mac_entry *mac; - uint16_t vlan; - uint16_t vid; - - ctx->has_normal = true; - - in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port, - ctx->packet != NULL, &in_port); - if (!in_bundle) { - xlate_report(ctx, "no input bundle, dropping"); - return; - } + union user_action_cookie cookie; + struct ofpbuf buf; - /* Drop malformed frames. */ - if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) && - !(ctx->flow.vlan_tci & htons(VLAN_CFI))) { - if (ctx->packet != NULL) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial " - "VLAN tag received on port %s", - ctx->ofproto->up.name, in_bundle->name); - } - xlate_report(ctx, "partial VLAN tag, dropping"); - return; - } + cookie.type = USER_ACTION_COOKIE_SLOW_PATH; + cookie.slow_path.unused = 0; + cookie.slow_path.reason = slow; - /* Drop frames on bundles reserved for mirroring. */ - if (in_bundle->mirror_out) { - if (ctx->packet != NULL) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " - "%s, which is reserved exclusively for mirroring", - ctx->ofproto->up.name, in_bundle->name); - } - xlate_report(ctx, "input port is mirror output port, dropping"); - return; + ofpbuf_use_stack(&buf, stub, stub_size); + if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) { + uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, + ODPP_NONE); + odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf); + } else { + put_userspace_action(ofproto, &buf, flow, &cookie, + sizeof cookie.slow_path); } + *actionsp = buf.data; + *actions_lenp = buf.size; +} - /* Check VLAN. */ - vid = vlan_tci_to_vid(ctx->flow.vlan_tci); - if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) { - xlate_report(ctx, "disallowed VLAN VID for this input port, dropping"); - return; - } - vlan = input_vid_to_vlan(in_bundle, vid); +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 size_t cookie_size) +{ + uint32_t pid; - /* Check other admissibility requirements. */ - if (in_port && !is_admissible(ctx, in_port, vlan)) { - return; - } + pid = dpif_port_get_pid(ofproto->backer->dpif, + ofp_port_to_odp_port(ofproto, + flow->in_port.ofp_port)); - /* Learn source MAC. */ - if (ctx->may_learn) { - update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle); - } + return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions); +} - /* Determine output bundle. */ - mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan, - &ctx->tags); - if (mac) { - if (mac->port.p != in_bundle) { - xlate_report(ctx, "forwarding to learned port"); - output_normal(ctx, mac->port.p, vlan); - } else { - xlate_report(ctx, "learned port is input port, dropping"); - } - } else { - struct ofbundle *bundle; - - xlate_report(ctx, "no learned MAC for destination, flooding"); - HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) { - if (bundle != in_bundle - && ofbundle_includes_vlan(bundle, vlan) - && bundle->floodable - && !bundle->mirror_out) { - output_normal(ctx, bundle, vlan); - } +tag_type +calculate_flow_tag(struct ofproto_dpif *ofproto, const struct flow *flow, + uint8_t table_id, struct rule_dpif *rule) +{ + if (table_id > 0 && table_id < N_TABLES) { + struct table_dpif *table = &ofproto->tables[table_id]; + if (table->other_table) { + return (rule && rule->tag + ? rule->tag + : rule_calculate_tag(flow, &table->other_table->mask, + table->basis)); } - ctx->nf_output_iface = NF_OUT_FLOOD; } + + return 0; } /* Optimized flow revalidation. @@ -6839,7 +5621,7 @@ 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; } } @@ -6857,13 +5639,13 @@ 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; } } } @@ -6873,9 +5655,8 @@ 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; @@ -6890,29 +5671,27 @@ packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); struct odputil_keybuf keybuf; struct dpif_flow_stats stats; - + struct xlate_out xout; + struct xlate_in xin; struct ofpbuf key; - struct action_xlate_ctx ctx; - uint64_t odp_actions_stub[1024 / 8]; - struct ofpbuf odp_actions; ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); odp_flow_key_from_flow(&key, flow, - ofp_port_to_odp_port(ofproto, flow->in_port)); + ofp_port_to_odp_port(ofproto, + flow->in_port.ofp_port)); dpif_flow_stats_extract(flow, packet, time_msec(), &stats); - action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL, - packet_get_tcp_flags(packet, flow), packet); - ctx.resubmit_stats = &stats; + xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet); + xin.resubmit_stats = &stats; + xin.ofpacts_len = ofpacts_len; + xin.ofpacts = ofpacts; - ofpbuf_use_stub(&odp_actions, - odp_actions_stub, sizeof odp_actions_stub); - xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions); + xlate_actions(&xin, &xout); dpif_execute(ofproto->backer->dpif, key.data, key.size, - odp_actions.data, odp_actions.size, packet); - ofpbuf_uninit(&odp_actions); + xout.odp_actions.data, xout.odp_actions.size, packet); + xlate_out_uninit(&xout); return 0; } @@ -6928,13 +5707,16 @@ set_netflow(struct ofproto *ofproto_, if (netflow_options) { if (!ofproto->netflow) { ofproto->netflow = netflow_create(); + ofproto->backer->need_revalidate = REV_RECONFIGURE; } return netflow_set_options(ofproto->netflow, netflow_options); - } else { + } else if (ofproto->netflow) { + ofproto->backer->need_revalidate = REV_RECONFIGURE; netflow_destroy(ofproto->netflow); ofproto->netflow = NULL; - return 0; } + + return 0; } static void @@ -6958,7 +5740,8 @@ send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) if (subfacet->path == SF_FAST_PATH) { struct dpif_flow_stats stats; - subfacet_reinstall(subfacet, &stats); + subfacet_install(subfacet, &facet->xout.odp_actions, + &stats); subfacet_update_stats(subfacet, &stats); } } @@ -6974,9 +5757,11 @@ send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) static void send_netflow_active_timeouts(struct ofproto_dpif *ofproto) { + struct cls_cursor cursor; struct facet *facet; - HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { + cls_cursor_init(&cursor, &ofproto->facets, NULL); + CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { send_active_timeout(ofproto, facet); } } @@ -7007,16 +5792,27 @@ 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); + ovs_rwlock_wrlock(&ofproto->ml->rwlock); + mac_learning_flush(ofproto->ml); + ovs_rwlock_unlock(&ofproto->ml->rwlock); } else { HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { - mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); + ovs_rwlock_wrlock(&ofproto->ml->rwlock); + mac_learning_flush(ofproto->ml); + ovs_rwlock_unlock(&ofproto->ml->rwlock); } } unixctl_command_reply(conn, "table successfully flushed"); } +static struct ofport_dpif * +ofbundle_get_a_port(const struct ofbundle *bundle) +{ + return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif, + bundle_node); +} + static void ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED, const char *argv[], void *aux OVS_UNUSED) @@ -7032,26 +5828,31 @@ ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED, } ds_put_cstr(&ds, " port VLAN MAC Age\n"); + ovs_rwlock_rdlock(&ofproto->ml->rwlock); LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { struct ofbundle *bundle = e->port.p; - ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", - ofbundle_get_a_port(bundle)->odp_port, - e->vlan, ETH_ADDR_ARGS(e->mac), + char name[OFP_MAX_PORT_NAME_LEN]; + + ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port, + name, sizeof name); + ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n", + name, e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(ofproto->ml, e)); } + ovs_rwlock_unlock(&ofproto->ml->rwlock); unixctl_command_reply(conn, ds_cstr(&ds)); ds_destroy(&ds); } struct trace_ctx { - struct action_xlate_ctx ctx; + struct xlate_out xout; + struct xlate_in xin; struct flow flow; struct ds *result; }; static void -trace_format_rule(struct ds *result, uint8_t table_id, int level, - const struct rule_dpif *rule) +trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule) { ds_put_char_multiple(result, '\t', level); if (!rule) { @@ -7060,7 +5861,7 @@ trace_format_rule(struct ds *result, uint8_t table_id, int level, } ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ", - table_id, ntohll(rule->up.flow_cookie)); + rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie)); cls_rule_format(&rule->up.cr, result); ds_put_char(result, '\n'); @@ -7072,15 +5873,15 @@ trace_format_rule(struct ds *result, uint8_t table_id, int level, static void trace_format_flow(struct ds *result, int level, const char *title, - struct trace_ctx *trace) + struct trace_ctx *trace) { ds_put_char_multiple(result, '\t', level); ds_put_format(result, "%s: ", title); - if (flow_equal(&trace->ctx.flow, &trace->flow)) { + if (flow_equal(&trace->xin.flow, &trace->flow)) { ds_put_cstr(result, "unchanged"); } else { - flow_format(result, &trace->ctx.flow); - trace->flow = trace->ctx.flow; + flow_format(result, &trace->xin.flow); + trace->flow = trace->xin.flow; } ds_put_char(result, '\n'); } @@ -7103,7 +5904,7 @@ static void trace_format_odp(struct ds *result, int level, const char *title, struct trace_ctx *trace) { - struct ofpbuf *odp_actions = trace->ctx.odp_actions; + struct ofpbuf *odp_actions = &trace->xout.odp_actions; ds_put_char_multiple(result, '\t', level); ds_put_format(result, "%s: ", title); @@ -7112,25 +5913,25 @@ trace_format_odp(struct ds *result, int level, const char *title, } static void -trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) +trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse) { - struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx); + struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin); struct ds *result = trace->result; ds_put_char(result, '\n'); - trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); - trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace); - trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace); - trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule); + trace_format_flow(result, recurse + 1, "Resubmitted flow", trace); + trace_format_regs(result, recurse + 1, "Resubmitted regs", trace); + trace_format_odp(result, recurse + 1, "Resubmitted odp", trace); + trace_format_rule(result, recurse + 1, rule); } static void -trace_report(struct action_xlate_ctx *ctx, const char *s) +trace_report(struct xlate_in *xin, const char *s, int recurse) { - struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx); + struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin); struct ds *result = trace->result; - ds_put_char_multiple(result, '\t', ctx->recurse); + ds_put_char_multiple(result, '\t', recurse); ds_put_cstr(result, s); ds_put_char(result, '\n'); } @@ -7139,124 +5940,125 @@ static void ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], void *aux OVS_UNUSED) { - const char *dpname = argv[1]; + const struct dpif_backer *backer; struct ofproto_dpif *ofproto; - struct ofpbuf odp_key; + struct ofpbuf odp_key, odp_mask; struct ofpbuf *packet; - ovs_be16 initial_tci; struct ds result; struct flow flow; char *s; packet = NULL; - ofpbuf_init(&odp_key, 0); + backer = NULL; ds_init(&result); + ofpbuf_init(&odp_key, 0); + ofpbuf_init(&odp_mask, 0); - ofproto = ofproto_dpif_lookup(dpname); - if (!ofproto) { - unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list " - "for help)"); - goto exit; + /* Handle "-generate" or a hex string as the last argument. */ + if (!strcmp(argv[argc - 1], "-generate")) { + packet = ofpbuf_new(0); + argc--; + } else { + const char *error = eth_from_hex(argv[argc - 1], &packet); + if (!error) { + argc--; + } else if (argc == 4) { + /* The 3-argument form must end in "-generate' or a hex string. */ + unixctl_command_reply_error(conn, error); + goto exit; + } } - if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) { - /* ofproto/trace dpname flow [-generate] */ - const char *flow_s = argv[2]; - const char *generate_s = argv[3]; - /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like - * flow. We guess which type it is based on whether 'flow_s' contains - * an '(', since a datapath flow always contains '(') but an - * OpenFlow-like flow should not (in fact it's allowed but I believe - * that's not documented anywhere). - * - * An alternative would be to try to parse 'flow_s' both ways, but then - * it would be tricky giving a sensible error message. After all, do - * you just say "syntax error" or do you present both error messages? - * Both choices seem lousy. */ - if (strchr(flow_s, '(')) { - enum odp_key_fitness fitness; - int error; - - /* Convert string to datapath key. */ - ofpbuf_init(&odp_key, 0); - error = odp_flow_key_from_string(flow_s, NULL, &odp_key); - if (error) { - unixctl_command_reply_error(conn, "Bad flow syntax"); - goto exit; + /* Parse the flow and determine whether a datapath or + * bridge is specified. If function odp_flow_key_from_string() + * returns 0, the flow is a odp_flow. If function + * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */ + if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) { + /* If the odp_flow is the second argument, + * the datapath name is the first argument. */ + if (argc == 3) { + const char *dp_type; + if (!strncmp(argv[1], "ovs-", 4)) { + dp_type = argv[1] + 4; + } else { + dp_type = argv[1]; } - - fitness = odp_flow_key_to_flow(odp_key.data, odp_key.size, &flow); - flow.in_port = odp_port_to_ofp_port(ofproto, flow.in_port); - - /* Convert odp_key to flow. */ - error = ofproto_dpif_vsp_adjust(ofproto, fitness, &flow, - &initial_tci, NULL); - if (error == ODP_FIT_ERROR) { - unixctl_command_reply_error(conn, "Invalid flow"); + backer = shash_find_data(&all_dpif_backers, dp_type); + if (!backer) { + unixctl_command_reply_error(conn, "Cannot find datapath " + "of this name"); goto exit; } } else { - char *error_s; - - error_s = parse_ofp_exact_flow(&flow, argv[2]); - if (error_s) { - unixctl_command_reply_error(conn, error_s); - free(error_s); + /* No datapath name specified, so there should be only one + * datapath. */ + struct shash_node *node; + if (shash_count(&all_dpif_backers) != 1) { + unixctl_command_reply_error(conn, "Must specify datapath " + "name, there is more than one type of datapath"); goto exit; } - - initial_tci = flow.vlan_tci; - vsp_adjust_flow(ofproto, &flow); + node = shash_first(&all_dpif_backers); + backer = node->data; } - /* Generate a packet, if requested. */ - if (generate_s) { - packet = ofpbuf_new(0); - flow_compose(packet, &flow); + /* Extract the ofproto_dpif object from the ofproto_receive() + * function. */ + if (ofproto_receive(backer, NULL, odp_key.data, + odp_key.size, &flow, NULL, &ofproto, NULL)) { + unixctl_command_reply_error(conn, "Invalid datapath flow"); + goto exit; } - } else if (argc == 6) { - /* ofproto/trace dpname priority tun_id in_port 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]; - 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); - const char *msg; - - msg = eth_from_hex(packet_s, &packet); - if (msg) { - unixctl_command_reply_error(conn, msg); + ds_put_format(&result, "Bridge: %s\n", ofproto->up.name); + } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) { + if (argc != 3) { + unixctl_command_reply_error(conn, "Must specify bridge name"); goto exit; } - ds_put_cstr(&result, "Packet: "); - s = ofp_packet_to_string(packet->data, packet->size); - ds_put_cstr(&result, s); - free(s); - - flow_extract(packet, priority, NULL, in_port, &flow); - flow.tunnel.tun_id = tun_id; - initial_tci = flow.vlan_tci; + ofproto = ofproto_dpif_lookup(argv[1]); + if (!ofproto) { + unixctl_command_reply_error(conn, "Unknown bridge name"); + goto exit; + } } else { - unixctl_command_reply_error(conn, "Bad command syntax"); + unixctl_command_reply_error(conn, "Bad flow syntax"); goto exit; } - ofproto_trace(ofproto, &flow, packet, initial_tci, &result); + /* Generate a packet, if requested. */ + if (packet) { + if (!packet->size) { + flow_compose(packet, &flow); + } else { + union flow_in_port in_port_; + + in_port_ = flow.in_port; + ds_put_cstr(&result, "Packet: "); + s = ofp_packet_to_string(packet->data, packet->size); + ds_put_cstr(&result, s); + free(s); + + /* Use the metadata from the flow and the packet argument + * to reconstruct the flow. */ + flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL, + &in_port_, &flow); + } + } + + ofproto_trace(ofproto, &flow, packet, &result); unixctl_command_reply(conn, ds_cstr(&result)); exit: ds_destroy(&result); ofpbuf_delete(packet); ofpbuf_uninit(&odp_key); + ofpbuf_uninit(&odp_mask); } -static void +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, struct ds *ds) { struct rule_dpif *rule; @@ -7264,21 +6066,24 @@ ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow, flow_format(ds, flow); ds_put_char(ds, '\n'); - rule = rule_dpif_lookup(ofproto, flow); + rule = rule_dpif_lookup(ofproto, flow, NULL); - trace_format_rule(ds, 0, 0, rule); + trace_format_rule(ds, 0, rule); if (rule == ofproto->miss_rule) { ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n"); } else if (rule == ofproto->no_packet_in_rule) { ds_put_cstr(ds, "\nNo match, packets dropped because " "OFPPC_NO_PACKET_IN is set on in_port.\n"); + } else if (rule == ofproto->drop_frags_rule) { + ds_put_cstr(ds, "\nPackets dropped because they are IP fragments " + "and the fragment handling mode is \"drop\".\n"); } if (rule) { uint64_t odp_actions_stub[1024 / 8]; struct ofpbuf odp_actions; - struct trace_ctx trace; + struct match match; uint8_t tcp_flags; tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0; @@ -7286,64 +6091,50 @@ 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, - rule, tcp_flags, packet); - trace.ctx.resubmit_hook = trace_resubmit; - trace.ctx.report_hook = trace_report; - xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len, - &odp_actions); + xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet); + trace.xin.resubmit_hook = trace_resubmit; + trace.xin.report_hook = trace_report; + + xlate_actions(&trace.xin, &trace.xout); ds_put_char(ds, '\n'); trace_format_flow(ds, 0, "Final flow", &trace); - ds_put_cstr(ds, "Datapath actions: "); - format_odp_actions(ds, odp_actions.data, odp_actions.size); - ofpbuf_uninit(&odp_actions); - if (trace.ctx.slow) { - enum slow_path_reason slow; + match_init(&match, flow, &trace.xout.wc); + ds_put_cstr(ds, "Relevant fields: "); + match_format(&match, ds, OFP_DEFAULT_PRIORITY); + ds_put_char(ds, '\n'); + + ds_put_cstr(ds, "Datapath actions: "); + format_odp_actions(ds, trace.xout.odp_actions.data, + trace.xout.odp_actions.size); + if (trace.xout.slow) { ds_put_cstr(ds, "\nThis flow is handled by the userspace " "slow path because it:"); - for (slow = trace.ctx.slow; slow; ) { - enum slow_path_reason bit = rightmost_1bit(slow); - - switch (bit) { - case SLOW_CFM: - ds_put_cstr(ds, "\n\t- Consists of CFM packets."); - break; - case SLOW_LACP: - ds_put_cstr(ds, "\n\t- Consists of LACP packets."); - break; - case SLOW_STP: - ds_put_cstr(ds, "\n\t- Consists of STP packets."); - break; - case SLOW_IN_BAND: - ds_put_cstr(ds, "\n\t- Needs in-band special case " - "processing."); - if (!packet) { - ds_put_cstr(ds, "\n\t (The datapath actions are " - "incomplete--for complete actions, " - "please supply a packet.)"); - } - break; - case SLOW_CONTROLLER: - ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages " - "to the OpenFlow controller."); - break; - case SLOW_MATCH: - ds_put_cstr(ds, "\n\t- Needs more specific matching " - "than the datapath supports."); - break; - } - - slow &= ~bit; - } - - if (slow & ~SLOW_MATCH) { - ds_put_cstr(ds, "\nThe datapath actions above do not reflect " - "the special slow-path processing."); + switch (trace.xout.slow) { + case SLOW_CFM: + ds_put_cstr(ds, "\n\t- Consists of CFM packets."); + break; + case SLOW_LACP: + ds_put_cstr(ds, "\n\t- Consists of LACP packets."); + break; + case SLOW_STP: + ds_put_cstr(ds, "\n\t- Consists of STP packets."); + break; + case SLOW_BFD: + ds_put_cstr(ds, "\n\t- Consists of BFD packets."); + break; + case SLOW_CONTROLLER: + ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages " + "to the OpenFlow controller."); + break; + case __SLOW_MAX: + NOT_REACHED(); } } + + xlate_out_uninit(&trace.xout); } } @@ -7368,17 +6159,19 @@ ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, static void ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply) { + struct cls_cursor cursor; struct facet *facet; int errors; errors = 0; - HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { + cls_cursor_init(&cursor, &ofproto->facets, NULL); + CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { if (!facet_check_consistency(facet)) { errors++; } } if (errors) { - ofproto->need_revalidate = REV_INCONSISTENCY; + ofproto->backer->need_revalidate = REV_INCONSISTENCY; } if (errors) { @@ -7455,107 +6248,218 @@ ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED, } static void -show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds) +show_dp_rates(struct ds *ds, const char *heading, + const struct avg_subfacet_rates *rates) { - struct dpif_dp_stats s; - const struct shash_node **ports; - int i; + ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n", + heading, rates->add_rate, rates->del_rate); +} - dpif_get_dp_stats(ofproto->backer->dpif, &s); +static void +dpif_show_backer(const struct dpif_backer *backer, struct ds *ds) +{ + const struct shash_node **ofprotos; + struct ofproto_dpif *ofproto; + struct shash ofproto_shash; + uint64_t n_hit, n_missed; + long long int minutes; + size_t i; - ds_put_format(ds, "%s (%s):\n", ofproto->up.name, - dpif_name(ofproto->backer->dpif)); - /* xxx It would be better to show bridge-specific stats instead - * xxx of dp ones. */ - ds_put_format(ds, - "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n", - s.n_hit, s.n_missed, s.n_lost); - ds_put_format(ds, "\tflows: %zu\n", - hmap_count(&ofproto->subfacets)); - - 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); - - ds_put_format(ds, "\t%s %u/%u:", name, ofport->ofp_port, - ofp_port_to_odp_port(ofproto, ofport->ofp_port)); - if (strcmp(type, "system")) { - struct netdev *netdev; - int error; - - 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); + n_hit = n_missed = 0; + HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { + if (ofproto->backer == backer) { + n_missed += ofproto->n_missed; + n_hit += ofproto->n_hit; + } + } + + ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n", + dpif_name(backer->dpif), n_hit, n_missed); + ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u," + " life span: %lldms\n", hmap_count(&backer->subfacets), + backer->avg_n_subfacet, backer->max_n_subfacet, + backer->avg_subfacet_life); + + minutes = (time_msec() - backer->created) / (1000 * 60); + if (minutes >= 60) { + show_dp_rates(ds, "\thourly avg:", &backer->hourly); + } + if (minutes >= 60 * 24) { + show_dp_rates(ds, "\tdaily avg:", &backer->daily); + } + show_dp_rates(ds, "\toverall avg:", &backer->lifetime); + + shash_init(&ofproto_shash); + ofprotos = get_ofprotos(&ofproto_shash); + for (i = 0; i < shash_count(&ofproto_shash); i++) { + struct ofproto_dpif *ofproto = ofprotos[i]->data; + const struct shash_node **ports; + size_t j; + + if (ofproto->backer != backer) { + continue; + } + + ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n", + ofproto->up.name, ofproto->n_hit, ofproto->n_missed); + + ports = shash_sort(&ofproto->up.port_by_name); + for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) { + const struct shash_node *node = ports[j]; + struct ofport *ofport = node->data; + struct smap config; + odp_port_t odp_port; + + ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev), + ofport->ofp_port); - netdev_close(netdev); + odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port); + if (odp_port != ODPP_NONE) { + ds_put_format(ds, "%"PRIu32":", odp_port); + } else { + ds_put_cstr(ds, "none:"); + } + + ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev)); + + smap_init(&config); + if (!netdev_get_config(ofport->netdev, &config)) { + 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); + ds_put_char(ds, ')'); + ds_put_char(ds, '\n'); } - ds_put_char(ds, '\n'); + free(ports); } - free(ports); + shash_destroy(&ofproto_shash); + free(ofprotos); } static void -ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc, - const char *argv[], void *aux OVS_UNUSED) +ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED, + const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) +{ + struct ds ds = DS_EMPTY_INITIALIZER; + const struct shash_node **backers; + int i; + + backers = shash_sort(&all_dpif_backers); + for (i = 0; i < shash_count(&all_dpif_backers); i++) { + dpif_show_backer(backers[i]->data, &ds); + } + free(backers); + + unixctl_command_reply(conn, ds_cstr(&ds)); + ds_destroy(&ds); +} + +/* Dump the megaflow (facet) cache. This is useful to check the + * correctness of flow wildcarding, since the same mechanism is used for + * both xlate caching and kernel wildcarding. + * + * It's important to note that in the output the flow description uses + * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports. + * + * This command is only needed for advanced debugging, so it's not + * documented in the man page. */ +static void +ofproto_unixctl_dpif_dump_megaflows(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; + long long int now = time_msec(); + struct cls_cursor cursor; + struct facet *facet; - 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; + ofproto = ofproto_dpif_lookup(argv[1]); + if (!ofproto) { + unixctl_command_reply_error(conn, "no such bridge"); + return; + } - 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); + cls_cursor_init(&cursor, &ofproto->facets, NULL); + CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { + cls_rule_format(&facet->cr, &ds); + ds_put_cstr(&ds, ", "); + ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets)); + ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0); + ds_put_cstr(&ds, "Datapath actions: "); + if (facet->xout.slow) { + uint64_t slow_path_stub[128 / 8]; + const struct nlattr *actions; + size_t actions_len; + + compose_slow_path(ofproto, &facet->flow, facet->xout.slow, + slow_path_stub, sizeof slow_path_stub, + &actions, &actions_len); + format_odp_actions(&ds, actions, actions_len); + } else { + format_odp_actions(&ds, facet->xout.odp_actions.data, + facet->xout.odp_actions.size); } - - shash_destroy(&ofproto_shash); - free(sorted_ofprotos); + ds_put_cstr(&ds, "\n"); } + ds_chomp(&ds, '\n'); unixctl_command_reply(conn, ds_cstr(&ds)); ds_destroy(&ds); } +/* Disable using the megaflows. + * + * This command is only needed for advanced debugging, so it's not + * documented in the man page. */ +static void +ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn, + int argc OVS_UNUSED, + const char *argv[] OVS_UNUSED, + void *aux OVS_UNUSED) +{ + struct ofproto_dpif *ofproto; + + enable_megaflows = false; + + HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { + flush(&ofproto->up); + } + + unixctl_command_reply(conn, "megaflows disabled"); +} + +/* Re-enable using megaflows. + * + * This command is only needed for advanced debugging, so it's not + * documented in the man page. */ +static void +ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn, + int argc OVS_UNUSED, + const char *argv[] OVS_UNUSED, + void *aux OVS_UNUSED) +{ + struct ofproto_dpif *ofproto; + + enable_megaflows = true; + + HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { + flush(&ofproto->up); + } + + unixctl_command_reply(conn, "megaflows enabled"); +} + static void ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn, int argc OVS_UNUSED, const char *argv[], @@ -7571,12 +6475,25 @@ ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn, return; } - HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) { - struct odputil_keybuf keybuf; - struct ofpbuf key; + update_stats(ofproto->backer); + + HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) { + struct facet *facet = subfacet->facet; + struct odputil_keybuf maskbuf; + struct ofpbuf mask; + + if (facet->ofproto != ofproto) { + continue; + } + + ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf); + if (enable_megaflows) { + odp_flow_key_from_mask(&mask, &facet->xout.wc.masks, + &facet->flow, UINT32_MAX); + } - subfacet_get_key(subfacet, &keybuf, &key); - odp_flow_key_format(key.data, key.size, &ds); + odp_flow_format(subfacet->key, subfacet->key_len, + mask.data, mask.size, &ds); ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:", subfacet->dp_packet_count, subfacet->dp_byte_count); @@ -7592,7 +6509,19 @@ ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn, } ds_put_cstr(&ds, ", actions:"); - format_odp_actions(&ds, subfacet->actions, subfacet->actions_len); + if (facet->xout.slow) { + uint64_t slow_path_stub[128 / 8]; + const struct nlattr *actions; + size_t actions_len; + + compose_slow_path(ofproto, &facet->flow, facet->xout.slow, + slow_path_stub, sizeof slow_path_stub, + &actions, &actions_len); + format_odp_actions(&ds, actions, actions_len); + } else { + format_odp_actions(&ds, facet->xout.odp_actions.data, + facet->xout.odp_actions.size); + } ds_put_char(&ds, '\n'); } @@ -7631,8 +6560,8 @@ ofproto_dpif_unixctl_init(void) unixctl_command_register( "ofproto/trace", - "bridge {priority tun_id in_port packet | odp_flow [-generate]}", - 2, 5, ofproto_unixctl_trace, NULL); + "[dp_name]|bridge odp_flow|br_flow [-generate|packet]", + 1, 3, ofproto_unixctl_trace, NULL); unixctl_command_register("fdb/flush", "[bridge]", 0, 1, ofproto_unixctl_fdb_flush, NULL); unixctl_command_register("fdb/show", "bridge", 1, 1, @@ -7645,12 +6574,18 @@ ofproto_dpif_unixctl_init(void) 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/show", "", 0, 0, 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); + unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1, + ofproto_unixctl_dpif_dump_megaflows, NULL); + unixctl_command_register("dpif/disable-megaflows", "", 0, 0, + ofproto_unixctl_dpif_disable_megaflows, NULL); + unixctl_command_register("dpif/enable-megaflows", "", 0, 0, + ofproto_unixctl_dpif_enable_megaflows, NULL); } /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) @@ -7661,7 +6596,7 @@ ofproto_dpif_unixctl_init(void) * widespread use, we will delete these interfaces. */ static int -set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid) +set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); @@ -7671,7 +6606,7 @@ 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); @@ -7693,46 +6628,51 @@ set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid) } static uint32_t -hash_realdev_vid(uint16_t realdev_ofp_port, int vid) +hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid) +{ + return hash_2words(ofp_to_u16(realdev_ofp_port), vid); +} + +bool +ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto) { - return hash_2words(realdev_ofp_port, vid); + return !hmap_is_empty(&ofproto->realdev_vid_map); } -/* Returns the ODP port number of the Linux VLAN device that corresponds to - * 'vlan_tci' on the network device with port number 'realdev_odp_port' in - * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9, - * it would return the port number of eth0.9. +/* Returns the OFP port number of the Linux VLAN device that corresponds to + * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in + * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and + * 'vlan_tci' 9, it would return the port number of eth0.9. * - * Unless VLAN splinters are enabled for port 'realdev_odp_port', this - * function just returns its 'realdev_odp_port' argument. */ -static uint32_t + * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this + * function just returns its 'realdev_ofp_port' argument. */ +ofp_port_t vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto, - uint32_t realdev_odp_port, ovs_be16 vlan_tci) + ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci) { if (!hmap_is_empty(&ofproto->realdev_vid_map)) { - 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(ofproto, vsp->vlandev_ofp_port); + return vsp->vlandev_ofp_port; } } } - return realdev_odp_port; + return realdev_ofp_port; } static struct vlan_splinter * -vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port) +vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port) { struct vlan_splinter *vsp; - HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0), + HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, + hash_ofp_port(vlandev_ofp_port), &ofproto->vlandev_map) { if (vsp->vlandev_ofp_port == vlandev_ofp_port) { return vsp; @@ -7751,9 +6691,9 @@ vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port) * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux * VLAN device. Unless VLAN splinters are enabled, this is what this function * always does.*/ -static uint16_t +static ofp_port_t vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto, - uint16_t vlandev_ofp_port, int *vid) + ofp_port_t vlandev_ofp_port, int *vid) { if (!hmap_is_empty(&ofproto->vlandev_map)) { const struct vlan_splinter *vsp; @@ -7778,17 +6718,17 @@ vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto, static bool vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow) { - uint16_t realdev; + ofp_port_t realdev; int vid; - realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid); + realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid); if (!realdev) { return false; } /* Cause the flow to be processed as if it came in on the real device with * the VLAN device's VLAN ID. */ - flow->in_port = realdev; + flow->in_port.ofp_port = realdev; flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI); return true; } @@ -7812,7 +6752,7 @@ vsp_remove(struct ofport_dpif *port) } static void -vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid) +vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); @@ -7823,7 +6763,7 @@ vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid) vsp = xmalloc(sizeof *vsp); hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node, - hash_int(port->up.ofp_port, 0)); + hash_ofp_port(port->up.ofp_port)); hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node, hash_realdev_vid(realdev_ofp_port, vid)); vsp->realdev_ofp_port = realdev_ofp_port; @@ -7836,20 +6776,19 @@ vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid) } } -static uint32_t -ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port) +static odp_port_t +ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port) { const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); - return ofport ? ofport->odp_port : OVSP_NONE; + return ofport ? ofport->odp_port : ODPP_NONE; } static struct ofport_dpif * -odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port) +odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port) { struct ofport_dpif *port; - HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, - hash_int(odp_port, 0), + HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port), &backer->odp_to_ofport_map) { if (port->odp_port == odp_port) { return port; @@ -7859,19 +6798,64 @@ odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port) return NULL; } -static uint16_t -odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port) +static ofp_port_t +odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port) { struct ofport_dpif *port; port = odp_port_to_ofport(ofproto->backer, odp_port); - if (port && ofproto == ofproto_dpif_cast(port->up.ofproto)) { + if (port && &ofproto->up == port->up.ofproto) { return port->up.ofp_port; } else { return OFPP_NONE; } } +/* Compute exponentially weighted moving average, adding 'new' as the newest, + * most heavily weighted element. 'base' designates the rate of decay: after + * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e + * (about .37). */ +static void +exp_mavg(double *avg, int base, double new) +{ + *avg = (*avg * (base - 1) + new) / base; +} + +static void +update_moving_averages(struct dpif_backer *backer) +{ + const int min_ms = 60 * 1000; /* milliseconds in one minute. */ + long long int minutes = (time_msec() - backer->created) / min_ms; + + if (minutes > 0) { + backer->lifetime.add_rate = (double) backer->total_subfacet_add_count + / minutes; + backer->lifetime.del_rate = (double) backer->total_subfacet_del_count + / minutes; + } else { + backer->lifetime.add_rate = 0.0; + backer->lifetime.del_rate = 0.0; + } + + /* Update hourly averages on the minute boundaries. */ + if (time_msec() - backer->last_minute >= min_ms) { + exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count); + exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count); + + /* Update daily averages on the hour boundaries. */ + if ((backer->last_minute - backer->created) / min_ms % 60 == 59) { + exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate); + exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate); + } + + backer->total_subfacet_add_count += backer->subfacet_add_count; + backer->total_subfacet_del_count += backer->subfacet_del_count; + backer->subfacet_add_count = 0; + backer->subfacet_del_count = 0; + backer->last_minute += min_ms; + } +} + const struct ofproto_class ofproto_dpif_class = { init, enumerate_types, @@ -7921,11 +6905,11 @@ const struct ofproto_class ofproto_dpif_class = { set_netflow, get_netflow_ids, set_sflow, + set_ipfix, set_cfm, - get_cfm_fault, - get_cfm_opup, - get_cfm_remote_mpids, - get_cfm_health, + get_cfm_status, + set_bfd, + get_bfd_status, set_stp, get_stp_status, set_stp_port, @@ -7933,11 +6917,15 @@ const struct ofproto_class ofproto_dpif_class = { set_queues, bundle_set, bundle_remove, - mirror_set, - mirror_get_stats, + mirror_set__, + mirror_get_stats__, set_flood_vlans, is_mirror_output_bundle, forward_bpdu_changed, - set_mac_idle_time, + set_mac_table_config, set_realdev, + NULL, /* meter_get_features */ + NULL, /* meter_set */ + NULL, /* meter_get */ + NULL, /* meter_del */ };