X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=ofproto%2Fofproto-dpif.c;h=d32f6d16374f13a298d30605b9504fa07dac30d5;hb=72e8bf28bb38e8816435c64859fb350215b6a9e6;hp=2afd94a9cdc69cca85c9c66d6be149b8537c1090;hpb=805e4d07ecd8ce86c6f36bdb83d2e7fee56891c7;p=sliver-openvswitch.git diff --git a/ofproto/ofproto-dpif.c b/ofproto/ofproto-dpif.c index 2afd94a9c..d32f6d163 100644 --- a/ofproto/ofproto-dpif.c +++ b/ofproto/ofproto-dpif.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks. + * Copyright (c) 2009, 2010, 2011, 2012 Nicira, Inc. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -42,9 +42,14 @@ #include "odp-util.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-sflow.h" #include "poll-loop.h" +#include "simap.h" +#include "smap.h" #include "timer.h" #include "unaligned.h" #include "unixctl.h" @@ -53,25 +58,25 @@ VLOG_DEFINE_THIS_MODULE(ofproto_dpif); -COVERAGE_DEFINE(ofproto_dpif_ctlr_action); COVERAGE_DEFINE(ofproto_dpif_expired); -COVERAGE_DEFINE(ofproto_dpif_no_packet_in); COVERAGE_DEFINE(ofproto_dpif_xlate); COVERAGE_DEFINE(facet_changed_rule); -COVERAGE_DEFINE(facet_invalidated); 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 32 +#define MAX_RESUBMIT_RECURSION 64 /* Number of implemented OpenFlow tables. */ enum { N_TABLES = 255 }; -BUILD_ASSERT_DECL(N_TABLES >= 1 && 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; @@ -103,15 +108,19 @@ static struct rule_dpif *rule_dpif_cast(const struct rule *rule) } static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *, - const struct flow *, uint8_t table); - + 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 *, - uint64_t packets, uint64_t bytes, - long long int used); - + const struct dpif_flow_stats *); static tag_type rule_calculate_tag(const struct flow *, - const struct flow_wildcards *, - uint32_t basis); + const struct minimask *, uint32_t basis); static void rule_invalidate(const struct rule_dpif *); #define MAX_MIRRORS 32 @@ -145,8 +154,8 @@ static void update_mirror_stats(struct ofproto_dpif *ofproto, uint64_t packets, uint64_t bytes); struct ofbundle { - struct ofproto_dpif *ofproto; /* Owning ofproto. */ struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */ + struct ofproto_dpif *ofproto; /* Owning ofproto. */ void *aux; /* Key supplied by ofproto's client. */ char *name; /* Identifier for log messages. */ @@ -175,8 +184,9 @@ 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(struct ofproto_dpif *, - uint16_t in_port, bool warn); +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 @@ -209,17 +219,13 @@ struct action_xlate_ctx { * revalidating without a packet to refer to. */ const struct ofpbuf *packet; - /* Should OFPP_NORMAL update the MAC learning table? We want to update it - * 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_macs; - - /* Should "learn" actions update the flow table? We want to update it if - * we are actually processing a packet, or in most cases if we are - * accounting for packets that the datapath has processed, but not if we - * are just revalidating. */ - bool may_flow_mod; + /* 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; @@ -229,21 +235,35 @@ struct action_xlate_ctx { * timeouts.) */ uint8_t tcp_flags; - /* If nonnull, called just before executing a resubmit action. In - * addition, disables logging of traces when the recursion depth is - * exceeded. + /* 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 *); + 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. */ - bool may_set_up_flow; /* True ordinarily; false if the actions must - * be reassessed for every packet. */ + 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? */ @@ -259,17 +279,131 @@ struct action_xlate_ctx { 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. */ - uint16_t sflow_odp_port; /* Output port for composing sFlow action. */ + 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 struct ofpbuf *xlate_actions(struct action_xlate_ctx *, - const union ofp_action *in, size_t n_in); +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: + * + * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the + * case just after the subfacet is created, just before the subfacet is + * destroyed, or if the datapath returns an error when we try to install a + * subfacet. + * + * - SF_FAST_PATH: The subfacet's actions are installed in the datapath. + * + * - SF_SLOW_PATH: An action that sends every packet for the subfacet through + * ofproto_dpif is installed in the datapath. + */ +enum subfacet_path { + SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */ + SF_FAST_PATH, /* Full actions are installed. */ + 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. */ +struct subfacet { + /* Owners. */ + 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. */ + + /* 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. */ + + 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 *, + const struct nlattr *key, size_t key_len, + uint32_t key_hash, + const struct flow *flow); +static void subfacet_destroy(struct subfacet *); +static void subfacet_destroy__(struct subfacet *); +static void subfacet_destroy_batch(struct ofproto_dpif *, + struct subfacet **, int n); +static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *, + struct ofpbuf *key); +static void subfacet_reset_dp_stats(struct subfacet *, + struct dpif_flow_stats *); +static void subfacet_update_time(struct subfacet *, long long int used); +static void subfacet_update_stats(struct subfacet *, + const struct dpif_flow_stats *); +static void subfacet_make_actions(struct subfacet *, + const struct ofpbuf *packet, + struct ofpbuf *odp_actions); +static int subfacet_install(struct subfacet *, + const struct nlattr *actions, size_t actions_len, + struct dpif_flow_stats *, enum slow_path_reason); +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. * @@ -329,22 +463,28 @@ struct facet { * 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 may_install; /* Reassess actions for every packet? */ 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. */ + + /* 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.) */ + struct subfacet one_subfacet; }; -static struct facet *facet_create(struct rule_dpif *, const struct flow *); +static struct facet *facet_create(struct rule_dpif *, + const struct flow *, uint32_t hash); static void facet_remove(struct facet *); static void facet_free(struct facet *); -static struct facet *facet_find(struct ofproto_dpif *, const struct flow *); +static struct facet *facet_find(struct ofproto_dpif *, + const struct flow *, uint32_t hash); static struct facet *facet_lookup_valid(struct ofproto_dpif *, - const struct flow *); -static bool facet_revalidate(struct facet *); + const struct flow *, uint32_t hash); +static void facet_revalidate(struct facet *); static bool facet_check_consistency(struct facet *); static void facet_flush_stats(struct facet *); @@ -352,70 +492,13 @@ 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_account(struct facet *, bool may_flow_mod); +static void facet_learn(struct facet *); +static void facet_account(struct facet *); static bool facet_is_controller_flow(struct facet *); -/* A dpif flow and actions associated with a facet. - * - * See also the large comment on struct facet. */ -struct subfacet { - /* Owners. */ - 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. */ - - /* 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. */ - - 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. */ - - bool installed; /* 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. */ -}; - -static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness, - const struct nlattr *key, - size_t key_len, ovs_be16 initial_tci); -static struct subfacet *subfacet_find(struct ofproto_dpif *, - const struct nlattr *key, size_t key_len); -static void subfacet_destroy(struct subfacet *); -static void subfacet_destroy__(struct subfacet *); -static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *, - struct ofpbuf *key); -static void subfacet_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); -static int subfacet_install(struct subfacet *, - const struct nlattr *actions, size_t actions_len, - struct dpif_flow_stats *); -static void subfacet_uninstall(struct subfacet *); - struct ofport_dpif { + struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */ struct ofport up; uint32_t odp_port; @@ -470,11 +553,15 @@ struct vlan_splinter { static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *, uint32_t realdev, ovs_be16 vlan_tci); -static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *, - uint16_t vlandev, int *vid); +static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *); static void vsp_remove(struct ofport_dpif *); static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid); +static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *, + uint16_t ofp_port); +static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *, + uint32_t odp_port); + static struct ofport_dpif * ofport_dpif_cast(const struct ofport *ofport) { @@ -483,6 +570,7 @@ ofport_dpif_cast(const struct ofport *ofport) } static void port_run(struct ofport_dpif *); +static void port_run_fast(struct ofport_dpif *); static void port_wait(struct ofport_dpif *); static int set_cfm(struct ofport *, const struct cfm_settings *); static void ofport_clear_priorities(struct ofport_dpif *); @@ -503,11 +591,50 @@ struct table_dpif { uint32_t basis; /* Keeps each table's tags separate. */ }; +/* Reasons that we might need to revalidate every facet, and corresponding + * coverage counters. + * + * A value of 0 means that there is no need to revalidate. + * + * It would be nice to have some cleaner way to integrate with coverage + * counters, but with only a few reasons I guess this is good enough for + * now. */ +enum revalidate_reason { + REV_RECONFIGURE = 1, /* Switch configuration changed. */ + REV_STP, /* Spanning tree protocol port status change. */ + REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/ + REV_FLOW_TABLE, /* Flow table changed. */ + REV_INCONSISTENCY /* Facet self-check failed. */ +}; +COVERAGE_DEFINE(rev_reconfigure); +COVERAGE_DEFINE(rev_stp); +COVERAGE_DEFINE(rev_port_toggled); +COVERAGE_DEFINE(rev_flow_table); +COVERAGE_DEFINE(rev_inconsistency); + +/* All datapaths of a given type share a single dpif backer instance. */ +struct dpif_backer { + char *type; + int refcount; + struct dpif *dpif; + struct timer next_expiration; + struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */ +}; + +/* All existing ofproto_backer instances, indexed by ofproto->up.type. */ +static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers); + +static struct ofport_dpif * +odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port); + struct ofproto_dpif { struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */ struct ofproto up; - struct dpif *dpif; - int max_ports; + struct dpif_backer *backer; + + /* 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; @@ -518,18 +645,17 @@ struct ofproto_dpif { struct hmap bundles; /* Contains "struct ofbundle"s. */ struct mac_learning *ml; struct ofmirror *mirrors[MAX_MIRRORS]; + bool has_mirrors; bool has_bonded_bundles; - /* Expiration. */ - struct timer next_expiration; - /* Facets. */ struct hmap facets; struct hmap subfacets; + struct governor *governor; /* Revalidation. */ struct table_dpif tables[N_TABLES]; - bool need_revalidate; + enum revalidate_reason need_revalidate; struct tag_set revalidate_set; /* Support for debugging async flow mods. */ @@ -546,6 +672,11 @@ struct ofproto_dpif { /* VLAN splinters. */ struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */ struct hmap vlandev_map; /* vlandev -> (realdev,vid). */ + + /* Ports. */ + struct sset ports; /* Set of port names. */ + struct sset port_poll_set; /* Queued names for port_poll() reply. */ + int port_poll_errno; /* Last errno for port_poll() reply. */ }; /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only @@ -564,9 +695,9 @@ ofproto_dpif_cast(const struct ofproto *ofproto) return CONTAINER_OF(ofproto, struct ofproto_dpif, up); } -static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *, +static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *, uint16_t ofp_port); -static struct ofport_dpif *get_odp_port(struct ofproto_dpif *, +static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *, uint32_t odp_port); static void ofproto_trace(struct ofproto_dpif *, const struct flow *, const struct ofpbuf *, ovs_be16 initial_tci, @@ -578,26 +709,47 @@ static void update_learning_table(struct ofproto_dpif *, struct ofbundle *); /* Upcalls. */ #define FLOW_MISS_MAX_BATCH 50 -static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch); +static int handle_upcalls(struct dpif_backer *, unsigned int max_batch); /* Flow expiration. */ -static int expire(struct ofproto_dpif *); +static int expire(struct dpif_backer *); /* NetFlow. */ static void send_netflow_active_timeouts(struct ofproto_dpif *); /* 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 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); /* Factory functions. */ +static void +init(const struct shash *iface_hints) +{ + struct shash_node *node; + + /* Make a local copy, since we don't own 'iface_hints' elements. */ + SHASH_FOR_EACH(node, iface_hints) { + const struct iface_hint *orig_hint = node->data; + struct iface_hint *new_hint = xmalloc(sizeof *new_hint); + + new_hint->br_name = xstrdup(orig_hint->br_name); + new_hint->br_type = xstrdup(orig_hint->br_type); + new_hint->ofp_port = orig_hint->ofp_port; + + shash_add(&init_ofp_ports, node->name, new_hint); + } +} + static void enumerate_types(struct sset *types) { @@ -607,7 +759,17 @@ enumerate_types(struct sset *types) static int enumerate_names(const char *type, struct sset *names) { - return dp_enumerate_names(type, names); + struct ofproto_dpif *ofproto; + + sset_clear(names); + HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { + if (strcmp(type, ofproto->up.type)) { + continue; + } + sset_add(names, ofproto->up.name); + } + + return 0; } static int @@ -624,8 +786,142 @@ del(const char *type, const char *name) return error; } +static const char * +port_open_type(const char *datapath_type, const char *port_type) +{ + return dpif_port_open_type(datapath_type, port_type); +} + +/* Type functions. */ + +static int +type_run(const char *type) +{ + struct dpif_backer *backer; + char *devname; + int error; + + backer = shash_find_data(&all_dpif_backers, type); + if (!backer) { + /* This is not necessarily a problem, since backers are only + * created on demand. */ + return 0; + } + + dpif_run(backer->dpif); + + if (timer_expired(&backer->next_expiration)) { + int delay = expire(backer); + timer_set_duration(&backer->next_expiration, delay); + } + + /* Check for port changes in the dpif. */ + while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) { + struct ofproto_dpif *ofproto = NULL; + struct dpif_port port; + + /* Don't report on the datapath's device. */ + if (!strcmp(devname, dpif_base_name(backer->dpif))) { + continue; + } + + HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, + &all_ofproto_dpifs) { + if (sset_contains(&ofproto->ports, devname)) { + break; + } + } + + 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; + } + 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 (error != EAGAIN) { + struct ofproto_dpif *ofproto; + + /* There was some sort of error, so propagate it to all + * ofprotos that use this backer. */ + HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, + &all_ofproto_dpifs) { + if (ofproto->backer == backer) { + sset_clear(&ofproto->port_poll_set); + ofproto->port_poll_errno = error; + } + } + } + + return 0; +} + +static int +type_run_fast(const char *type) +{ + struct dpif_backer *backer; + unsigned int work; + + backer = shash_find_data(&all_dpif_backers, type); + if (!backer) { + /* This is not necessarily a problem, since backers are only + * created on demand. */ + return 0; + } + + /* Handle one or more batches of upcalls, until there's nothing left to do + * or until we do a fixed total amount of work. + * + * We do work in batches because it can be much cheaper to set up a number + * of flows and fire off their patches all at once. We do multiple batches + * because in some cases handling a packet can cause another packet to be + * queued almost immediately as part of the return flow. Both + * optimizations can make major improvements on some benchmarks and + * presumably for real traffic as well. */ + work = 0; + while (work < FLOW_MISS_MAX_BATCH) { + int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work); + if (retval <= 0) { + return -retval; + } + work += retval; + } + + return 0; +} + +static void +type_wait(const char *type) +{ + struct dpif_backer *backer; + + backer = shash_find_data(&all_dpif_backers, type); + if (!backer) { + /* This is not necessarily a problem, since backers are only + * created on demand. */ + return; + } + + timer_wait(&backer->next_expiration); +} + /* Basic life-cycle. */ +static int add_internal_flows(struct ofproto_dpif *); + static struct ofproto * alloc(void) { @@ -640,33 +936,146 @@ dealloc(struct ofproto *ofproto_) free(ofproto); } -static int -construct(struct ofproto *ofproto_) +static void +close_dpif_backer(struct dpif_backer *backer) { - struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - const char *name = ofproto->up.name; + struct shash_node *node; + + assert(backer->refcount > 0); + + if (--backer->refcount) { + return; + } + + hmap_destroy(&backer->odp_to_ofport_map); + node = shash_find(&all_dpif_backers, backer->type); + free(backer->type); + shash_delete(&all_dpif_backers, node); + dpif_close(backer->dpif); + + free(backer); +} + +/* Datapath port slated for removal from datapath. */ +struct odp_garbage { + struct list list_node; + uint32_t odp_port; +}; + +static int +open_dpif_backer(const char *type, struct dpif_backer **backerp) +{ + struct dpif_backer *backer; + struct dpif_port_dump port_dump; + struct dpif_port port; + struct shash_node *node; + struct list garbage_list; + struct odp_garbage *garbage, *next; + struct sset names; + char *backer_name; + const char *name; int error; - int i; - error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif); + backer = shash_find_data(&all_dpif_backers, type); + if (backer) { + backer->refcount++; + *backerp = backer; + return 0; + } + + backer_name = xasprintf("ovs-%s", type); + + /* Remove any existing datapaths, since we assume we're the only + * userspace controlling the datapath. */ + sset_init(&names); + dp_enumerate_names(type, &names); + SSET_FOR_EACH(name, &names) { + struct dpif *old_dpif; + + /* Don't remove our backer if it exists. */ + if (!strcmp(name, backer_name)) { + continue; + } + + if (dpif_open(name, type, &old_dpif)) { + VLOG_WARN("couldn't open old datapath %s to remove it", name); + } else { + dpif_delete(old_dpif); + dpif_close(old_dpif); + } + } + sset_destroy(&names); + + backer = xmalloc(sizeof *backer); + + error = dpif_create_and_open(backer_name, type, &backer->dpif); + free(backer_name); if (error) { - VLOG_ERR("failed to open datapath %s: %s", name, strerror(error)); + VLOG_ERR("failed to open datapath of type %s: %s", type, + strerror(error)); return error; } - ofproto->max_ports = dpif_get_max_ports(ofproto->dpif); - ofproto->n_matches = 0; + backer->type = xstrdup(type); + backer->refcount = 1; + hmap_init(&backer->odp_to_ofport_map); + timer_set_duration(&backer->next_expiration, 1000); + *backerp = backer; + + dpif_flow_flush(backer->dpif); + + /* Loop through the ports already on the datapath and remove any + * that we don't need anymore. */ + list_init(&garbage_list); + dpif_port_dump_start(&port_dump, backer->dpif); + while (dpif_port_dump_next(&port_dump, &port)) { + node = shash_find(&init_ofp_ports, port.name); + if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) { + garbage = xmalloc(sizeof *garbage); + garbage->odp_port = port.port_no; + list_push_front(&garbage_list, &garbage->list_node); + } + } + dpif_port_dump_done(&port_dump); - dpif_flow_flush(ofproto->dpif); - dpif_recv_purge(ofproto->dpif); + LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) { + dpif_port_del(backer->dpif, garbage->odp_port); + list_remove(&garbage->list_node); + free(garbage); + } + + shash_add(&all_dpif_backers, type, backer); + + error = dpif_recv_set(backer->dpif, true); + if (error) { + VLOG_ERR("failed to listen on datapath of type %s: %s", + type, strerror(error)); + close_dpif_backer(backer); + return error; + } + + return error; +} + +static int +construct(struct ofproto *ofproto_) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + struct shash_node *node, *next; + int max_ports; + int error; + int i; - error = dpif_recv_set(ofproto->dpif, true); + error = open_dpif_backer(ofproto->up.type, &ofproto->backer); if (error) { - VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error)); - dpif_close(ofproto->dpif); return error; } + max_ports = dpif_get_max_ports(ofproto->backer->dpif); + ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX)); + + ofproto->n_matches = 0; + ofproto->netflow = NULL; ofproto->sflow = NULL; ofproto->stp = NULL; @@ -677,10 +1086,9 @@ construct(struct ofproto *ofproto_) } ofproto->has_bonded_bundles = false; - timer_set_duration(&ofproto->next_expiration, 1000); - hmap_init(&ofproto->facets); hmap_init(&ofproto->subfacets); + ofproto->governor = NULL; for (i = 0; i < N_TABLES; i++) { struct table_dpif *table = &ofproto->tables[i]; @@ -689,27 +1097,114 @@ construct(struct ofproto *ofproto_) table->other_table = NULL; table->basis = random_uint32(); } - ofproto->need_revalidate = false; + 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->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; + + if (!strcmp(iface_hint->br_name, ofproto->up.name)) { + /* Check if the datapath already has this port. */ + if (dpif_port_exists(ofproto->backer->dpif, node->name)) { + sset_add(&ofproto->ports, node->name); + } + + free(iface_hint->br_name); + free(iface_hint->br_type); + shash_delete(&init_ofp_ports, node); + } + } + hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node, hash_string(ofproto->up.name, 0)); memset(&ofproto->stats, 0, sizeof ofproto->stats); ofproto_init_tables(ofproto_, N_TABLES); + error = add_internal_flows(ofproto); + ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY; + + return error; +} + +static int +add_internal_flow(struct ofproto_dpif *ofproto, int id, + const struct ofpbuf *ofpacts, struct rule_dpif **rulep) +{ + struct ofputil_flow_mod fm; + int error; + + match_init_catchall(&fm.match); + fm.priority = 0; + match_set_reg(&fm.match, 0, id); + fm.new_cookie = htonll(0); + fm.cookie = htonll(0); + fm.cookie_mask = htonll(0); + fm.table_id = TBL_INTERNAL; + fm.command = OFPFC_ADD; + fm.idle_timeout = 0; + fm.hard_timeout = 0; + fm.buffer_id = 0; + fm.out_port = 0; + fm.flags = 0; + fm.ofpacts = ofpacts->data; + fm.ofpacts_len = ofpacts->size; + + error = ofproto_flow_mod(&ofproto->up, &fm); + if (error) { + VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)", + id, ofperr_to_string(error)); + return error; + } + + *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL); + assert(*rulep != NULL); return 0; } +static int +add_internal_flows(struct ofproto_dpif *ofproto) +{ + struct ofpact_controller *controller; + uint64_t ofpacts_stub[128 / 8]; + struct ofpbuf ofpacts; + int error; + int id; + + ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); + id = 1; + + controller = ofpact_put_CONTROLLER(&ofpacts); + controller->max_len = UINT16_MAX; + controller->controller_id = 0; + controller->reason = OFPR_NO_MATCH; + ofpact_pad(&ofpacts); + + error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule); + if (error) { + return error; + } + + ofpbuf_clear(&ofpacts); + error = add_internal_flow(ofproto, id++, &ofpacts, + &ofproto->no_packet_in_rule); + return error; +} + static void complete_operations(struct ofproto_dpif *ofproto) { @@ -753,36 +1248,27 @@ destruct(struct ofproto *ofproto_) hmap_destroy(&ofproto->facets); hmap_destroy(&ofproto->subfacets); + governor_destroy(ofproto->governor); hmap_destroy(&ofproto->vlandev_map); hmap_destroy(&ofproto->realdev_vid_map); - dpif_close(ofproto->dpif); + sset_destroy(&ofproto->ports); + sset_destroy(&ofproto->port_poll_set); + + close_dpif_backer(ofproto->backer); } static int run_fast(struct ofproto *ofproto_) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - unsigned int work; + struct ofport_dpif *ofport; - /* Handle one or more batches of upcalls, until there's nothing left to do - * or until we do a fixed total amount of work. - * - * We do work in batches because it can be much cheaper to set up a number - * of flows and fire off their patches all at once. We do multiple batches - * because in some cases handling a packet can cause another packet to be - * queued almost immediately as part of the return flow. Both - * optimizations can make major improvements on some benchmarks and - * presumably for real traffic as well. */ - work = 0; - while (work < FLOW_MISS_MAX_BATCH) { - int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work); - if (retval <= 0) { - return -retval; - } - work += retval; + HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { + port_run_fast(ofport); } + return 0; } @@ -797,18 +1283,12 @@ run(struct ofproto *ofproto_) if (!clogged) { complete_operations(ofproto); } - dpif_run(ofproto->dpif); error = run_fast(ofproto_); if (error) { return error; } - if (timer_expired(&ofproto->next_expiration)) { - int delay = expire(ofproto); - timer_set_duration(&ofproto->next_expiration, delay); - } - if (ofproto->netflow) { if (netflow_run(ofproto->netflow)) { send_netflow_active_timeouts(ofproto); @@ -833,13 +1313,21 @@ run(struct ofproto *ofproto_) || !tag_set_is_empty(&ofproto->revalidate_set)) { struct tag_set revalidate_set = ofproto->revalidate_set; bool revalidate_all = ofproto->need_revalidate; - struct facet *facet, *next; + 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 = false; + ofproto->need_revalidate = 0; - HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) { + HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { if (revalidate_all || tag_set_intersects(&revalidate_set, facet->tags)) { facet_revalidate(facet); @@ -855,11 +1343,29 @@ run(struct ofproto *ofproto_) struct facet, hmap_node); if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) { if (!facet_check_consistency(facet)) { - ofproto->need_revalidate = true; + ofproto->need_revalidate = REV_INCONSISTENCY; } } } + if (ofproto->governor) { + size_t n_subfacets; + + governor_run(ofproto->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(&ofproto->subfacets); + if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold + && governor_is_idle(ofproto->governor)) { + governor_destroy(ofproto->governor); + ofproto->governor = NULL; + } + } + return 0; } @@ -874,8 +1380,8 @@ wait(struct ofproto *ofproto_) poll_immediate_wake(); } - dpif_wait(ofproto->dpif); - dpif_recv_wait(ofproto->dpif); + dpif_wait(ofproto->backer->dpif); + dpif_recv_wait(ofproto->backer->dpif); if (ofproto->sflow) { dpif_sflow_wait(ofproto->sflow); } @@ -897,32 +1403,46 @@ wait(struct ofproto *ofproto_) /* Shouldn't happen, but if it does just go around again. */ VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); poll_immediate_wake(); - } else { - timer_wait(&ofproto->next_expiration); + } + if (ofproto->governor) { + governor_wait(ofproto->governor); } } +static void +get_memory_usage(const struct ofproto *ofproto_, struct simap *usage) +{ + const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + + simap_increase(usage, "facets", hmap_count(&ofproto->facets)); + simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets)); +} + static void flush(struct ofproto *ofproto_) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - struct facet *facet, *next_facet; - - HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { - /* Mark the facet as not installed so that facet_remove() doesn't - * bother trying to uninstall it. There is no point in uninstalling it - * individually since we are about to blow away all the facets with - * dpif_flow_flush(). */ - struct subfacet *subfacet; + struct subfacet *subfacet, *next_subfacet; + struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH]; + int n_batch; - LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { - subfacet->installed = false; - subfacet->dp_packet_count = 0; - subfacet->dp_byte_count = 0; + n_batch = 0; + HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, + &ofproto->subfacets) { + if (subfacet->path != SF_NOT_INSTALLED) { + batch[n_batch++] = subfacet; + if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { + subfacet_destroy_batch(ofproto, batch, n_batch); + n_batch = 0; + } + } else { + subfacet_destroy(subfacet); } - facet_remove(facet); } - dpif_flow_flush(ofproto->dpif); + + if (n_batch > 0) { + subfacet_destroy_batch(ofproto, batch, n_batch); + } } static void @@ -945,17 +1465,17 @@ get_features(struct ofproto *ofproto_ OVS_UNUSED, } static void -get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots) +get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); struct dpif_dp_stats s; strcpy(ots->name, "classifier"); - dpif_get_dp_stats(ofproto->dpif, &s); - put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed)); - put_32aligned_be64(&ots->matched_count, - htonll(s.n_hit + ofproto->n_matches)); + dpif_get_dp_stats(ofproto->backer->dpif, &s); + + ots->lookup_count = htonll(s.n_hit + s.n_missed); + ots->matched_count = htonll(s.n_hit + ofproto->n_matches); } static struct ofport * @@ -977,9 +1497,10 @@ port_construct(struct ofport *port_) { struct ofport_dpif *port = ofport_dpif_cast(port_); struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); + struct dpif_port dpif_port; + int error; - ofproto->need_revalidate = true; - port->odp_port = ofp_port_to_odp_port(port->up.ofp_port); + ofproto->need_revalidate = REV_RECONFIGURE; port->bundle = NULL; port->cfm = NULL; port->tag = tag_create_random(); @@ -991,8 +1512,28 @@ port_construct(struct ofport *port_) port->vlandev_vid = 0; port->carrier_seq = netdev_get_carrier_resets(port->up.netdev); + error = dpif_port_query_by_name(ofproto->backer->dpif, + netdev_get_name(port->up.netdev), + &dpif_port); + if (error) { + return error; + } + + 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; + } + + hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node, + hash_int(port->odp_port, 0)); + if (ofproto->sflow) { - dpif_sflow_add_port(ofproto->sflow, port_); + dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port); } return 0; @@ -1003,8 +1544,19 @@ 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); + + if (dpif_port_exists(ofproto->backer->dpif, devname)) { + /* 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); + } - ofproto->need_revalidate = true; + sset_find_and_delete(&ofproto->ports, devname); + hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node); + ofproto->need_revalidate = REV_RECONFIGURE; bundle_remove(port_); set_cfm(port_, NULL); if (ofproto->sflow) { @@ -1033,8 +1585,9 @@ port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config) enum ofputil_port_config changed = old_config ^ port->up.pp.config; if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP | - OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD)) { - ofproto->need_revalidate = true; + OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD | + OFPUTIL_PC_NO_PACKET_IN)) { + ofproto->need_revalidate = REV_RECONFIGURE; if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) { bundle_update(port->bundle); @@ -1053,17 +1606,17 @@ set_sflow(struct ofproto *ofproto_, if (!ds) { struct ofport_dpif *ofport; - ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif); + ds = ofproto->sflow = dpif_sflow_create(); HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { - dpif_sflow_add_port(ds, &ofport->up); + dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port); } - ofproto->need_revalidate = true; + ofproto->need_revalidate = REV_RECONFIGURE; } dpif_sflow_set_options(ds, sflow_options); } else { if (ds) { dpif_sflow_destroy(ds); - ofproto->need_revalidate = true; + ofproto->need_revalidate = REV_RECONFIGURE; ofproto->sflow = NULL; } } @@ -1083,7 +1636,7 @@ set_cfm(struct ofport *ofport_, const struct cfm_settings *s) struct ofproto_dpif *ofproto; ofproto = ofproto_dpif_cast(ofport->up.ofproto); - ofproto->need_revalidate = true; + ofproto->need_revalidate = REV_RECONFIGURE; ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev)); } @@ -1106,6 +1659,14 @@ get_cfm_fault(const struct ofport *ofport_) return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1; } +static int +get_cfm_opup(const struct ofport *ofport_) +{ + struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); + + return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1; +} + static int get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps, size_t *n_rmps) @@ -1119,6 +1680,14 @@ get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps, return -1; } } + +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. */ @@ -1155,7 +1724,7 @@ 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 = true; + ofproto->need_revalidate = REV_RECONFIGURE; } if (s) { @@ -1228,7 +1797,7 @@ update_stp_port_state(struct ofport_dpif *ofport) fwd_change = stp_forward_in_state(ofport->stp_state) != stp_forward_in_state(state); - ofproto->need_revalidate = true; + ofproto->need_revalidate = REV_STP; ofport->stp_state = state; ofport->stp_state_entered = time_msec(); @@ -1414,7 +1983,7 @@ set_queues(struct ofport *ofport_, uint8_t dscp; dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK; - if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue, + if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue, &priority)) { continue; } @@ -1426,12 +1995,12 @@ set_queues(struct ofport *ofport_, pdscp = xmalloc(sizeof *pdscp); pdscp->priority = priority; pdscp->dscp = dscp; - ofproto->need_revalidate = true; + ofproto->need_revalidate = REV_RECONFIGURE; } if (pdscp->dscp != dscp) { pdscp->dscp = dscp; - ofproto->need_revalidate = true; + ofproto->need_revalidate = REV_RECONFIGURE; } hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0)); @@ -1439,7 +2008,7 @@ set_queues(struct ofport *ofport_, if (!hmap_is_empty(&ofport->priorities)) { ofport_clear_priorities(ofport); - ofproto->need_revalidate = true; + ofproto->need_revalidate = REV_RECONFIGURE; } hmap_swap(&new, &ofport->priorities); @@ -1466,7 +2035,7 @@ bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos) struct mac_learning *ml = ofproto->ml; struct mac_entry *mac, *next_mac; - ofproto->need_revalidate = true; + ofproto->need_revalidate = REV_RECONFIGURE; LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { if (mac->port.p == bundle) { if (all_ofprotos) { @@ -1543,7 +2112,7 @@ bundle_del_port(struct ofport_dpif *port) { struct ofbundle *bundle = port->bundle; - bundle->ofproto->need_revalidate = true; + bundle->ofproto->need_revalidate = REV_RECONFIGURE; list_remove(&port->bundle_node); port->bundle = NULL; @@ -1571,7 +2140,7 @@ bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port, } if (port->bundle != bundle) { - bundle->ofproto->need_revalidate = true; + bundle->ofproto->need_revalidate = REV_RECONFIGURE; if (port->bundle) { bundle_del_port(port); } @@ -1584,7 +2153,7 @@ bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port, } } if (lacp) { - port->bundle->ofproto->need_revalidate = true; + port->bundle->ofproto->need_revalidate = REV_RECONFIGURE; lacp_slave_register(bundle->lacp, port, lacp); } @@ -1612,7 +2181,7 @@ bundle_destroy(struct ofbundle *bundle) mirror_destroy(m); } else if (hmapx_find_and_delete(&m->srcs, bundle) || hmapx_find_and_delete(&m->dsts, bundle)) { - ofproto->need_revalidate = true; + ofproto->need_revalidate = REV_RECONFIGURE; } } } @@ -1684,7 +2253,7 @@ bundle_set(struct ofproto *ofproto_, void *aux, /* LACP. */ if (s->lacp) { if (!bundle->lacp) { - ofproto->need_revalidate = true; + ofproto->need_revalidate = REV_RECONFIGURE; bundle->lacp = lacp_create(); } lacp_configure(bundle->lacp, s->lacp); @@ -1747,7 +2316,7 @@ bundle_set(struct ofproto *ofproto_, void *aux, break; case PORT_VLAN_TRUNK: - trunks = (unsigned long *) s->trunks; + trunks = CONST_CAST(unsigned long *, s->trunks); break; case PORT_VLAN_NATIVE_UNTAGGED: @@ -1764,7 +2333,7 @@ bundle_set(struct ofproto *ofproto_, void *aux, bitmap_set1(trunks, vlan); bitmap_set0(trunks, 0); } else { - trunks = (unsigned long *) s->trunks; + trunks = CONST_CAST(unsigned long *, s->trunks); } break; @@ -1790,11 +2359,11 @@ 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 = true; + ofproto->need_revalidate = REV_RECONFIGURE; } } else { bundle->bond = bond_create(s->bond); - ofproto->need_revalidate = true; + ofproto->need_revalidate = REV_RECONFIGURE; } LIST_FOR_EACH (port, bundle_node, &bundle->ports) { @@ -2099,7 +2668,8 @@ mirror_set(struct ofproto *ofproto_, void *aux, } } - ofproto->need_revalidate = true; + ofproto->need_revalidate = REV_RECONFIGURE; + ofproto->has_mirrors = true; mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); mirror_update_dups(ofproto); @@ -2112,13 +2682,14 @@ mirror_destroy(struct ofmirror *mirror) struct ofproto_dpif *ofproto; mirror_mask_t mirror_bit; struct ofbundle *bundle; + int i; if (!mirror) { return; } ofproto = mirror->ofproto; - ofproto->need_revalidate = true; + ofproto->need_revalidate = REV_RECONFIGURE; mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); mirror_bit = MIRROR_MASK_C(1) << mirror->idx; @@ -2137,6 +2708,14 @@ mirror_destroy(struct ofmirror *mirror) 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; + } + } } static int @@ -2179,8 +2758,7 @@ static void forward_bpdu_changed(struct ofproto *ofproto_) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - /* Revalidate cached flows whenever forward_bpdu option changes. */ - ofproto->need_revalidate = true; + ofproto->need_revalidate = REV_RECONFIGURE; } static void @@ -2193,25 +2771,39 @@ set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time) /* Ports. */ static struct ofport_dpif * -get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port) +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(struct ofproto_dpif *ofproto, uint32_t odp_port) +get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port) { - return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port)); + return get_ofp_port(ofproto, odp_port_to_ofp_port(ofproto, odp_port)); } static void -ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port, +ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto, + struct ofproto_port *ofproto_port, struct dpif_port *dpif_port) { ofproto_port->name = dpif_port->name; ofproto_port->type = dpif_port->type; - ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no); + ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no); +} + +static void +port_run_fast(struct ofport_dpif *ofport) +{ + if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) { + struct ofpbuf packet; + + ofpbuf_init(&packet, 0); + cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr); + send_packet(ofport, &packet); + ofpbuf_uninit(&packet); + } } static void @@ -2223,20 +2815,16 @@ port_run(struct ofport_dpif *ofport) ofport->carrier_seq = carrier_seq; + port_run_fast(ofport); if (ofport->cfm) { - cfm_run(ofport->cfm); + int cfm_opup = cfm_get_opup(ofport->cfm); - if (cfm_should_send_ccm(ofport->cfm)) { - struct ofpbuf packet; + cfm_run(ofport->cfm); + enable = enable && !cfm_get_fault(ofport->cfm); - ofpbuf_init(&packet, 0); - cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr); - send_packet(ofport, &packet); - ofpbuf_uninit(&packet); + if (cfm_opup >= 0) { + enable = enable && cfm_opup; } - - enable = enable && !cfm_get_fault(ofport->cfm) - && cfm_get_opup(ofport->cfm); } if (ofport->bundle) { @@ -2250,7 +2838,7 @@ port_run(struct ofport_dpif *ofport) struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); if (ofproto->has_bundle_action) { - ofproto->need_revalidate = true; + ofproto->need_revalidate = REV_PORT_TOGGLED; } } @@ -2273,23 +2861,27 @@ port_query_by_name(const struct ofproto *ofproto_, const char *devname, struct dpif_port dpif_port; int error; - error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port); + if (!sset_contains(&ofproto->ports, devname)) { + return ENODEV; + } + error = dpif_port_query_by_name(ofproto->backer->dpif, + devname, &dpif_port); if (!error) { - ofproto_port_from_dpif_port(ofproto_port, &dpif_port); + ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port); } return error; } static int -port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp) +port_add(struct ofproto *ofproto_, struct netdev *netdev) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - uint16_t odp_port; + uint32_t odp_port = UINT32_MAX; int error; - error = dpif_port_add(ofproto->dpif, netdev, &odp_port); + error = dpif_port_add(ofproto->backer->dpif, netdev, &odp_port); if (!error) { - *ofp_portp = odp_port_to_ofp_port(odp_port); + sset_add(&ofproto->ports, netdev_get_name(netdev)); } return error; } @@ -2298,9 +2890,12 @@ static int port_del(struct ofproto *ofproto_, uint16_t ofp_port) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - int error; + uint32_t odp_port = ofp_port_to_odp_port(ofproto, ofp_port); + int error = 0; - error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port)); + if (odp_port != OFPP_NONE) { + error = dpif_port_del(ofproto->backer->dpif, odp_port); + } if (!error) { struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); if (ofport) { @@ -2340,7 +2935,7 @@ port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats) /* ofproto->stats.rx_packets represents packets that were received on * some port and we processed internally and dropped (e.g. STP). - * Account fro them as if they had been forwarded to OFPP_LOCAL. */ + * Account for them as if they had been forwarded to OFPP_LOCAL. */ if (stats->tx_packets != UINT64_MAX) { stats->tx_packets += ofproto->stats.rx_packets; @@ -2372,19 +2967,18 @@ ofproto_update_local_port_stats(const struct ofproto *ofproto_, } struct port_dump_state { - struct dpif_port_dump dump; - bool done; + uint32_t bucket; + uint32_t offset; }; static int -port_dump_start(const struct ofproto *ofproto_, void **statep) +port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep) { - struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); struct port_dump_state *state; *statep = state = xmalloc(sizeof *state); - dpif_port_dump_start(&state->dump, ofproto->dpif); - state->done = false; + state->bucket = 0; + state->offset = 0; return 0; } @@ -2392,17 +2986,21 @@ static int port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_, struct ofproto_port *port) { + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); struct port_dump_state *state = state_; - struct dpif_port dpif_port; + struct sset_node *node; - if (dpif_port_dump_next(&state->dump, &dpif_port)) { - ofproto_port_from_dpif_port(port, &dpif_port); - return 0; - } else { - int error = dpif_port_dump_done(&state->dump); - state->done = true; - return error ? error : EOF; + while ((node = sset_at_position(&ofproto->ports, &state->bucket, + &state->offset))) { + int error; + + error = port_query_by_name(ofproto_, node->name, port); + if (error != ENODEV) { + return error; + } } + + return EOF; } static int @@ -2410,9 +3008,6 @@ port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) { struct port_dump_state *state = state_; - if (!state->done) { - dpif_port_dump_done(&state->dump); - } free(state); return 0; } @@ -2421,14 +3016,26 @@ static int port_poll(const struct ofproto *ofproto_, char **devnamep) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - return dpif_port_poll(ofproto->dpif, devnamep); + + if (ofproto->port_poll_errno) { + int error = ofproto->port_poll_errno; + ofproto->port_poll_errno = 0; + return error; + } + + if (sset_is_empty(&ofproto->port_poll_set)) { + return EAGAIN; + } + + *devnamep = sset_pop(&ofproto->port_poll_set); + return 0; } static void port_poll_wait(const struct ofproto *ofproto_) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - dpif_port_poll_wait(ofproto->dpif); + dpif_port_poll_wait(ofproto->backer->dpif); } static int @@ -2453,17 +3060,22 @@ port_is_lacp_current(const struct ofport *ofport_) * It's possible to batch more than that, but the benefit might be minimal. */ struct flow_miss { struct hmap_node hmap_node; + struct ofproto_dpif *ofproto; struct flow flow; enum odp_key_fitness key_fitness; const struct nlattr *key; size_t key_len; ovs_be16 initial_tci; struct 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; + struct subfacet *subfacet; /* Subfacet */ + void *garbage; /* Pointer to pass to free(), NULL if none. */ + uint64_t stub[1024 / 8]; /* Temporary buffer. */ }; /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each @@ -2487,49 +3099,42 @@ send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet, flow_get_metadata(flow, &pin.fmd); - /* Registers aren't meaningful on a miss. */ - memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks); - connmgr_send_packet_in(ofproto->up.connmgr, &pin); } -static bool +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 false; + return 0; } if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) { if (packet) { cfm_process_heartbeat(ofport->cfm, packet); } - return true; + 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 true; + return SLOW_LACP; } else if (ofproto->stp && stp_should_process_flow(flow)) { if (packet) { stp_process_packet(ofport, packet); } - return true; + return SLOW_STP; } - return false; + return 0; } static struct flow_miss * -flow_miss_create(struct hmap *todo, const struct flow *flow, - enum odp_key_fitness key_fitness, - const struct nlattr *key, size_t key_len, - ovs_be16 initial_tci) +flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash) { - uint32_t hash = flow_hash(flow, 0); struct flow_miss *miss; HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) { @@ -2538,168 +3143,270 @@ flow_miss_create(struct hmap *todo, const struct flow *flow, } } - miss = xmalloc(sizeof *miss); - hmap_insert(todo, &miss->hmap_node, hash); - miss->flow = *flow; - miss->key_fitness = key_fitness; - miss->key = key; - miss->key_len = key_len; - miss->initial_tci = initial_tci; - list_init(&miss->packets); - return miss; + return NULL; } +/* Partially Initializes 'op' as an "execute" operation for 'miss' and + * 'packet'. The caller must initialize op->actions and op->actions_len. If + * 'miss' is associated with a subfacet the caller must also initialize the + * returned op->subfacet, and if anything needs to be freed after processing + * the op, the caller must initialize op->garbage also. */ static void -handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss, - struct flow_miss_op *ops, size_t *n_ops) -{ - const struct flow *flow = &miss->flow; - struct ofpbuf *packet, *next_packet; - struct subfacet *subfacet; - struct facet *facet; +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) { + /* This packet was received on a VLAN splinter port. We + * added a VLAN to the packet to make the packet resemble + * the flow, but the actions were composed assuming that + * the packet contained no VLAN. So, we must remove the + * VLAN header from the packet before trying to execute the + * actions. */ + eth_pop_vlan(packet); + } - facet = facet_lookup_valid(ofproto, flow); - if (!facet) { - struct rule_dpif *rule; + op->subfacet = NULL; + op->garbage = NULL; + 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; +} - rule = rule_dpif_lookup(ofproto, flow, 0); - if (!rule) { - /* Don't send a packet-in if OFPUTIL_PC_NO_PACKET_IN asserted. */ - struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port); - if (port) { - if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) { - COVERAGE_INC(ofproto_dpif_no_packet_in); - /* XXX install 'drop' flow entry */ - return; - } - } else { - VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, - flow->in_port); - } +/* 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) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); - LIST_FOR_EACH (packet, list_node, &miss->packets) { - send_packet_in_miss(ofproto, packet, flow); - } + ofproto->n_matches++; + + if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) { + /* + * Extra-special case for fail-open mode. + * + * We are in fail-open mode and the packet matched the fail-open + * rule, but we are connected to a controller too. We should send + * the packet up to the controller in the hope that it will try to + * set up a flow and thereby allow us to exit fail-open. + * + * See the top-level comment in fail-open.c for more information. + */ + send_packet_in_miss(ofproto, packet, flow); + } +} + +/* Figures out whether a flow that missed in 'ofproto', whose details are in + * 'miss', 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) +{ + if (!ofproto->governor) { + size_t n_subfacets; - return; + n_subfacets = hmap_count(&ofproto->subfacets); + if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) { + return true; } - facet = facet_create(rule, flow); + ofproto->governor = governor_create(ofproto->up.name); } - subfacet = subfacet_create(facet, - miss->key_fitness, miss->key, miss->key_len, - miss->initial_tci); + return governor_should_install_flow(ofproto->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'. */ +static void +handle_flow_miss_without_facet(struct flow_miss *miss, + struct rule_dpif *rule, + 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_SAFE (packet, next_packet, list_node, &miss->packets) { + LIST_FOR_EACH (packet, list_node, &miss->packets) { + struct flow_miss_op *op = &ops[*n_ops]; struct dpif_flow_stats stats; - struct flow_miss_op *op; - struct dpif_execute *execute; + struct ofpbuf odp_actions; - ofproto->n_matches++; + COVERAGE_INC(facet_suppress); - if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) { - /* - * Extra-special case for fail-open mode. - * - * We are in fail-open mode and the packet matched the fail-open - * rule, but we are connected to a controller too. We should send - * the packet up to the controller in the hope that it will try to - * set up a flow and thereby allow us to exit fail-open. - * - * See the top-level comment in fail-open.c for more information. - */ - send_packet_in_miss(ofproto, packet, flow); + ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub); + + dpif_flow_stats_extract(&miss->flow, packet, now, &stats); + rule_credit_stats(rule, &stats); + + action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci, + rule, 0, packet); + ctx.resubmit_stats = &stats; + xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, + &odp_actions); + + if (odp_actions.size) { + 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); + + (*n_ops)++; + } else { + ofpbuf_uninit(&odp_actions); } + } +} + +/* Handles 'miss', which matches 'facet'. May add any required datapath + * operations to 'ops', incrementing '*n_ops' for each new op. + * + * All of the packets in 'miss' are considered to have arrived at time 'now'. + * This is really important only for new facets: if we just called time_msec() + * here, then the new subfacet or its packets could look (occasionally) as + * though it was used some time after the facet was used. That can make a + * one-packet flow look like it has a nonzero duration, which looks odd in + * e.g. NetFlow statistics. */ +static void +handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet, + long long int now, + 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); + + 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; - if (!facet->may_install || !subfacet->actions) { - subfacet_make_actions(subfacet, packet); + handle_flow_miss_common(facet->rule, packet, &miss->flow); + + ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub); + if (!subfacet->actions || subfacet->slow) { + subfacet_make_actions(subfacet, packet, &odp_actions); } - dpif_flow_stats_extract(&facet->flow, packet, &stats); + dpif_flow_stats_extract(&facet->flow, packet, now, &stats); subfacet_update_stats(subfacet, &stats); - if (!subfacet->actions_len) { - /* No actions to execute, so skip talking to the dpif. */ - continue; - } + if (subfacet->actions_len) { + 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); + } - if (flow->vlan_tci != subfacet->initial_tci) { - /* This packet was received on a VLAN splinter port. We added - * a VLAN to the packet to make the packet resemble the flow, - * but the actions were composed assuming that the packet - * contained no VLAN. So, we must remove the VLAN header from - * the packet before trying to execute the actions. */ - eth_pop_vlan(packet); + (*n_ops)++; + } else { + ofpbuf_uninit(&odp_actions); } + } - op = &ops[(*n_ops)++]; - execute = &op->dpif_op.u.execute; - op->subfacet = subfacet; - op->dpif_op.type = DPIF_OP_EXECUTE; - execute->key = miss->key; - execute->key_len = miss->key_len; - execute->actions = (facet->may_install - ? subfacet->actions - : xmemdup(subfacet->actions, - subfacet->actions_len)); - execute->actions_len = subfacet->actions_len; - execute->packet = packet; - } - - if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) { + want_path = subfacet_want_path(subfacet->slow); + 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; op->dpif_op.type = DPIF_OP_FLOW_PUT; put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; put->key = miss->key; put->key_len = miss->key_len; - put->actions = subfacet->actions; - put->actions_len = subfacet->actions_len; + if (want_path == SF_FAST_PATH) { + put->actions = subfacet->actions; + put->actions_len = subfacet->actions_len; + } else { + compose_slow_path(ofproto, &facet->flow, subfacet->slow, + op->stub, sizeof op->stub, + &put->actions, &put->actions_len); + } put->stats = NULL; } } -/* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of - * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns - * an ODP_FIT_* value that indicates how well 'key' fits our expectations for - * what a flow key should contain. - * - * This function also includes some logic to help make VLAN splinters - * transparent to the rest of the upcall processing logic. In particular, if - * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by - * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN - * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull). +/* Handles flow miss 'miss'. May add any required datapath operations + * to 'ops', incrementing '*n_ops' for each new op. */ +static void +handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops, + size_t *n_ops) +{ + struct ofproto_dpif *ofproto = miss->ofproto; + struct facet *facet; + long long int now; + uint32_t hash; + + /* The caller must ensure that miss->hmap_node.hash contains + * flow_hash(miss->flow, 0). */ + hash = miss->hmap_node.hash; + + facet = facet_lookup_valid(ofproto, &miss->flow, hash); + 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); + return; + } + + facet = facet_create(rule, &miss->flow, hash); + now = facet->used; + } else { + now = time_msec(); + } + handle_flow_miss_with_facet(miss, facet, now, 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.) - */ + * received on VLAN splinters.) */ static enum odp_key_fitness -ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto, - const struct nlattr *key, size_t key_len, - struct flow *flow, ovs_be16 *initial_tci, - struct ofpbuf *packet) +ofproto_dpif_vsp_adjust(const struct ofproto_dpif *ofproto, + enum odp_key_fitness fitness, + struct flow *flow, ovs_be16 *initial_tci, + struct ofpbuf *packet) { - enum odp_key_fitness fitness; - uint16_t realdev; - int vid; - - fitness = odp_flow_key_to_flow(key, key_len, flow); if (fitness == ODP_FIT_ERROR) { return fitness; } *initial_tci = flow->vlan_tci; - realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid); - if (realdev) { - /* 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->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI); + 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 @@ -2728,14 +3435,16 @@ ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto, } static void -handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls, +handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls, size_t n_upcalls) { struct dpif_upcall *upcall; - struct flow_miss *miss, *next_miss; + 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; @@ -2749,37 +3458,58 @@ handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls, * the packets that have the same flow in the same "flow_miss" structure so * that we can process them together. */ hmap_init(&todo); + n_misses = 0; for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) { + struct flow_miss *miss = &misses[n_misses]; + struct flow_miss *existing_miss; enum odp_key_fitness fitness; - struct flow_miss *miss; - ovs_be16 initial_tci; + struct ofproto_dpif *ofproto; + struct ofport_dpif *port; + uint32_t odp_in_port; struct flow flow; + uint32_t hash; - /* Obtain metadata and check userspace/kernel agreement on flow match, - * then set 'flow''s header pointers. */ - fitness = ofproto_dpif_extract_flow_key(ofproto, - upcall->key, upcall->key_len, - &flow, &initial_tci, - upcall->packet); - if (fitness == ODP_FIT_ERROR) { - ofpbuf_delete(upcall->packet); + 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); continue; } - flow_extract(upcall->packet, flow.skb_priority, flow.tun_id, - flow.in_port, &flow); + ofproto = ofproto_dpif_cast(port->up.ofproto); + odp_in_port = flow.in_port; + flow.in_port = port->up.ofp_port; - /* Handle 802.1ag, LACP, and STP specially. */ - if (process_special(ofproto, &flow, upcall->packet)) { - ofproto_update_local_port_stats(&ofproto->up, - 0, upcall->packet->size); - ofpbuf_delete(upcall->packet); - ofproto->n_matches++; + /* 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) { continue; } + 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. */ - miss = flow_miss_create(&todo, &flow, fitness, - upcall->key, upcall->key_len, initial_tci); + hash = flow_hash(&miss->flow, 0); + existing_miss = flow_miss_find(&todo, &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++; + } else { + miss = existing_miss; + } list_push_back(&miss->packets, &upcall->packet->list_node); } @@ -2787,7 +3517,7 @@ handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls, * operations to batch. */ n_ops = 0; HMAP_FOR_EACH (miss, hmap_node, &todo) { - handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops); + handle_flow_miss(miss, flow_miss_ops, &n_ops); } assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); @@ -2795,112 +3525,159 @@ handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls, for (i = 0; i < n_ops; i++) { dpif_ops[i] = &flow_miss_ops[i].dpif_op; } - dpif_operate(ofproto->dpif, dpif_ops, n_ops); + dpif_operate(backer->dpif, dpif_ops, n_ops); /* Free memory and update facets. */ for (i = 0; i < n_ops; i++) { struct flow_miss_op *op = &flow_miss_ops[i]; - struct dpif_execute *execute; switch (op->dpif_op.type) { case DPIF_OP_EXECUTE: - execute = &op->dpif_op.u.execute; - if (op->subfacet->actions != execute->actions) { - free((struct nlattr *) execute->actions); - } break; case DPIF_OP_FLOW_PUT: if (!op->dpif_op.error) { - op->subfacet->installed = true; + op->subfacet->path = subfacet_want_path(op->subfacet->slow); } break; + + case DPIF_OP_FLOW_DEL: + NOT_REACHED(); } - } - HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) { - ofpbuf_list_delete(&miss->packets); - hmap_remove(&todo, &miss->hmap_node); - free(miss); + + free(op->garbage); } hmap_destroy(&todo); } +static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL } +classify_upcall(const struct dpif_upcall *upcall) +{ + union user_action_cookie cookie; + + /* First look at the upcall type. */ + switch (upcall->type) { + case DPIF_UC_ACTION: + break; + + case DPIF_UC_MISS: + return MISS_UPCALL; + + case DPIF_N_UC_TYPES: + default: + VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type); + return BAD_UPCALL; + } + + /* "action" upcalls need a closer look. */ + memcpy(&cookie, &upcall->userdata, sizeof(cookie)); + switch (cookie.type) { + case USER_ACTION_COOKIE_SFLOW: + return SFLOW_UPCALL; + + case 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); + return BAD_UPCALL; + } +} + static void -handle_userspace_upcall(struct ofproto_dpif *ofproto, - struct dpif_upcall *upcall) +handle_sflow_upcall(struct dpif_backer *backer, + const struct dpif_upcall *upcall) { - struct user_action_cookie cookie; + 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; - memcpy(&cookie, &upcall->userdata, sizeof(cookie)); + fitness = odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); - fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key, - upcall->key_len, &flow, - &initial_tci, upcall->packet); - if (fitness == ODP_FIT_ERROR) { - ofpbuf_delete(upcall->packet); + port = odp_port_to_ofport(backer, flow.in_port); + if (!port) { return; } - if (cookie.type == USER_ACTION_COOKIE_SFLOW) { - if (ofproto->sflow) { - dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, - &cookie); - } - } else { - VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata); + ofproto = ofproto_dpif_cast(port->up.ofproto); + if (!ofproto->sflow) { + 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) { + return; } - ofpbuf_delete(upcall->packet); + + memcpy(&cookie, &upcall->userdata, sizeof(cookie)); + dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, + odp_in_port, &cookie); } static int -handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch) +handle_upcalls(struct dpif_backer *backer, unsigned int max_batch) { struct dpif_upcall misses[FLOW_MISS_MAX_BATCH]; + struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH]; + uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8]; + int n_processed; int n_misses; int i; - assert (max_batch <= FLOW_MISS_MAX_BATCH); + assert(max_batch <= FLOW_MISS_MAX_BATCH); n_misses = 0; - for (i = 0; i < max_batch; i++) { + for (n_processed = 0; n_processed < max_batch; n_processed++) { struct dpif_upcall *upcall = &misses[n_misses]; + struct ofpbuf *buf = &miss_bufs[n_misses]; int error; - error = dpif_recv(ofproto->dpif, upcall); + ofpbuf_use_stub(buf, miss_buf_stubs[n_misses], + sizeof miss_buf_stubs[n_misses]); + error = dpif_recv(backer->dpif, upcall, buf); if (error) { + ofpbuf_uninit(buf); break; } - switch (upcall->type) { - case DPIF_UC_ACTION: - handle_userspace_upcall(ofproto, upcall); - break; - - case DPIF_UC_MISS: + switch (classify_upcall(upcall)) { + case MISS_UPCALL: /* Handle it later. */ n_misses++; break; - case DPIF_N_UC_TYPES: - default: - VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, - upcall->type); + case SFLOW_UPCALL: + handle_sflow_upcall(backer, upcall); + ofpbuf_uninit(buf); + break; + + case BAD_UPCALL: + ofpbuf_uninit(buf); break; } } - handle_miss_upcalls(ofproto, misses, n_misses); + /* Handle deferred MISS_UPCALL processing. */ + handle_miss_upcalls(backer, misses, n_misses); + for (i = 0; i < n_misses; i++) { + ofpbuf_uninit(&miss_bufs[i]); + } - return i; + return n_processed; } /* Flow expiration. */ static int subfacet_max_idle(const struct ofproto_dpif *); -static void update_stats(struct ofproto_dpif *); +static void update_stats(struct dpif_backer *); static void rule_expire(struct rule_dpif *); static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle); @@ -2911,42 +3688,108 @@ static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle); * * Returns the number of milliseconds after which it should be called again. */ static int -expire(struct ofproto_dpif *ofproto) +expire(struct dpif_backer *backer) { - struct rule_dpif *rule, *next_rule; - struct oftable *table; - int dp_max_idle; + struct ofproto_dpif *ofproto; + int max_idle = INT32_MAX; - /* Update stats for each flow in the datapath. */ - update_stats(ofproto); + /* Update stats for each flow in the backer. */ + update_stats(backer); - /* Expire subfacets that have been idle too long. */ - dp_max_idle = subfacet_max_idle(ofproto); - expire_subfacets(ofproto, dp_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; - /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */ - OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { - struct cls_cursor cursor; + if (ofproto->backer != backer) { + continue; + } - cls_cursor_init(&cursor, &table->cls, NULL); - CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { - rule_expire(rule); + /* 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); + } } - } - /* All outstanding data in existing flows has been accounted, so it's a - * good time to do bond rebalancing. */ - if (ofproto->has_bonded_bundles) { - struct ofbundle *bundle; + /* All outstanding data in existing flows has been accounted, so it's a + * good time to do bond rebalancing. */ + if (ofproto->has_bonded_bundles) { + struct ofbundle *bundle; - HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { - if (bundle->bond) { - bond_rebalance(bundle->bond, &ofproto->revalidate_set); + HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { + if (bundle->bond) { + bond_rebalance(bundle->bond, &ofproto->revalidate_set); + } } } } - return MIN(dp_max_idle, 1000); + return MIN(max_idle, 1000); +} + +/* Updates flow table statistics given that the datapath just reported 'stats' + * as 'subfacet''s statistics. */ +static void +update_subfacet_stats(struct subfacet *subfacet, + const struct dpif_flow_stats *stats) +{ + struct facet *facet = subfacet->facet; + + if (stats->n_packets >= subfacet->dp_packet_count) { + uint64_t extra = stats->n_packets - subfacet->dp_packet_count; + facet->packet_count += extra; + } else { + VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); + } + + if (stats->n_bytes >= subfacet->dp_byte_count) { + facet->byte_count += stats->n_bytes - subfacet->dp_byte_count; + } else { + VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); + } + + subfacet->dp_packet_count = stats->n_packets; + subfacet->dp_byte_count = stats->n_bytes; + + 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, + const struct nlattr *key, size_t key_len) +{ + if (!VLOG_DROP_WARN(&rl)) { + struct ds s; + + ds_init(&s); + odp_flow_key_format(key, key_len, &s); + VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s)); + ds_destroy(&s); + } + + COVERAGE_INC(facet_unexpected); + dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL); } /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. @@ -2961,56 +3804,54 @@ expire(struct ofproto_dpif *ofproto) * datapath do not justify the benefit of having perfectly accurate statistics. */ static void -update_stats(struct ofproto_dpif *p) +update_stats(struct dpif_backer *backer) { const struct dpif_flow_stats *stats; struct dpif_flow_dump dump; const struct nlattr *key; size_t key_len; - dpif_flow_dump_start(&dump, p->dpif); + dpif_flow_dump_start(&dump, backer->dpif); while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) { + struct flow flow; struct subfacet *subfacet; + enum odp_key_fitness fitness; + struct ofproto_dpif *ofproto; + struct ofport_dpif *port; + uint32_t key_hash; - subfacet = subfacet_find(p, key, key_len); - if (subfacet && subfacet->installed) { - struct facet *facet = subfacet->facet; - - if (stats->n_packets >= subfacet->dp_packet_count) { - uint64_t extra = stats->n_packets - subfacet->dp_packet_count; - facet->packet_count += extra; - } else { - VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); - } - - if (stats->n_bytes >= subfacet->dp_byte_count) { - facet->byte_count += stats->n_bytes - subfacet->dp_byte_count; - } else { - VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); - } + fitness = odp_flow_key_to_flow(key, key_len, &flow); + if (fitness == ODP_FIT_ERROR) { + continue; + } - subfacet->dp_packet_count = stats->n_packets; - subfacet->dp_byte_count = stats->n_bytes; + 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; + } - facet->tcp_flags |= stats->tcp_flags; + ofproto = ofproto_dpif_cast(port->up.ofproto); + flow.in_port = port->up.ofp_port; + key_hash = odp_flow_key_hash(key, key_len); - subfacet_update_time(subfacet, stats->used); - facet_account(facet, true); - facet_push_stats(facet); - } else { - if (!VLOG_DROP_WARN(&rl)) { - struct ds s; + subfacet = subfacet_find(ofproto, key, key_len, key_hash, &flow); + switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) { + case SF_FAST_PATH: + update_subfacet_stats(subfacet, stats); + break; - ds_init(&s); - odp_flow_key_format(key, key_len, &s); - VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s)); - ds_destroy(&s); - } + case SF_SLOW_PATH: + /* Stats are updated per-packet. */ + break; - COVERAGE_INC(facet_unexpected); - /* There's a flow in the datapath that we know nothing about, or a - * flow that shouldn't be installed but was anyway. Delete it. */ - dpif_flow_del(p->dpif, key, key_len, NULL); + case SF_NOT_INSTALLED: + default: + delete_unexpected_flow(ofproto, key, key_len); + break; } } dpif_flow_dump_done(&dump); @@ -3107,15 +3948,41 @@ subfacet_max_idle(const struct ofproto_dpif *ofproto) static void expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle) { - long long int cutoff = time_msec() - dp_max_idle; + /* Cutoff time for most flows. */ + long long int normal_cutoff = time_msec() - dp_max_idle; + + /* We really want to keep flows for special protocols around, so use a more + * conservative cutoff. */ + long long int special_cutoff = time_msec() - 10000; + struct subfacet *subfacet, *next_subfacet; + struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH]; + int n_batch; + n_batch = 0; HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, &ofproto->subfacets) { + long long int cutoff; + + cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP) + ? special_cutoff + : normal_cutoff); if (subfacet->used < cutoff) { - subfacet_destroy(subfacet); + if (subfacet->path != SF_NOT_INSTALLED) { + batch[n_batch++] = subfacet; + if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { + subfacet_destroy_batch(ofproto, batch, n_batch); + n_batch = 0; + } + } else { + subfacet_destroy(subfacet); + } } } + + if (n_batch > 0) { + subfacet_destroy_batch(ofproto, batch, n_batch); + } } /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, @@ -3127,6 +3994,11 @@ rule_expire(struct rule_dpif *rule) long long int now; uint8_t reason; + if (rule->up.pending) { + /* We'll have to expire it later. */ + return; + } + /* Has 'rule' expired? */ now = time_msec(); if (rule->up.hard_timeout @@ -3159,17 +4031,19 @@ rule_expire(struct rule_dpif *rule) * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in * the ofproto's classifier table. * + * 'hash' must be the return value of flow_hash(flow, 0). + * * 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) +facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); struct facet *facet; facet = xzalloc(sizeof *facet); facet->used = time_msec(); - hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0)); + hmap_insert(&ofproto->facets, &facet->hmap_node, hash); list_push_back(&rule->facets, &facet->list_node); facet->rule = rule; facet->flow = *flow; @@ -3200,9 +4074,10 @@ execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, int error; ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); - odp_flow_key_from_flow(&key, flow); + odp_flow_key_from_flow(&key, flow, + ofp_port_to_odp_port(ofproto, flow->in_port)); - error = dpif_execute(ofproto->dpif, key.data, key.size, + error = dpif_execute(ofproto->backer->dpif, key.data, key.size, odp_actions, actions_len, packet); ofpbuf_delete(packet); @@ -3246,42 +4121,43 @@ facet_remove(struct facet *facet) facet_free(facet); } +/* Feed information from 'facet' back into the learning table to keep it in + * sync with what is actually flowing through the datapath. */ static void -facet_account(struct facet *facet, bool may_flow_mod) +facet_learn(struct facet *facet) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); - uint64_t n_bytes; - struct subfacet *subfacet; - const struct nlattr *a; - unsigned int left; - ovs_be16 vlan_tci; + struct action_xlate_ctx ctx; - if (facet->byte_count <= facet->accounted_bytes) { + if (!facet->has_learn + && !facet->has_normal + && (!facet->has_fin_timeout + || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) { return; } - n_bytes = facet->byte_count - facet->accounted_bytes; - facet->accounted_bytes = facet->byte_count; - - /* Feed information from the active flows back into the learning table to - * ensure that table is always in sync with what is actually flowing - * through the datapath. */ - if (facet->has_learn || facet->has_normal - || (facet->has_fin_timeout - && facet->tcp_flags & (TCP_FIN | TCP_RST))) { - struct action_xlate_ctx ctx; - action_xlate_ctx_init(&ctx, ofproto, &facet->flow, - facet->flow.vlan_tci, - facet->rule, facet->tcp_flags, NULL); - ctx.may_learn_macs = true; - ctx.may_flow_mod = may_flow_mod; - ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions, - facet->rule->up.n_actions)); - } + 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); +} + +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) { return; } + n_bytes = facet->byte_count - facet->accounted_bytes; /* This loop feeds byte counters to bond_account() for rebalancing to use * as a basis. We also need to track the actual VLAN on which the packet @@ -3326,10 +4202,18 @@ facet_account(struct facet *facet, bool may_flow_mod) static bool facet_is_controller_flow(struct facet *facet) { - return (facet - && facet->rule->up.n_actions == 1 - && action_outputs_to_port(&facet->rule->up.actions[0], - htons(OFPP_CONTROLLER))); + if (facet) { + const struct rule *rule = &facet->rule->up; + const struct ofpact *ofpacts = rule->ofpacts; + size_t ofpacts_len = rule->ofpacts_len; + + if (ofpacts_len > 0 && + ofpacts->type == OFPACT_CONTROLLER && + ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) { + return true; + } + } + return false; } /* Folds all of 'facet''s statistics into its rule. Also updates the @@ -3348,7 +4232,10 @@ facet_flush_stats(struct facet *facet) } facet_push_stats(facet); - facet_account(facet, false); + if (facet->accounted_bytes < facet->byte_count) { + facet_account(facet); + facet->accounted_bytes = facet->byte_count; + } if (ofproto->netflow && !facet_is_controller_flow(facet)) { struct ofexpired expired; @@ -3373,15 +4260,17 @@ facet_flush_stats(struct facet *facet) /* 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). + * * 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) +facet_find(struct ofproto_dpif *ofproto, + const struct flow *flow, uint32_t hash) { struct facet *facet; - HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0), - &ofproto->facets) { + HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) { if (flow_equal(flow, &facet->flow)) { return facet; } @@ -3393,25 +4282,63 @@ facet_find(struct ofproto_dpif *ofproto, const struct flow *flow) /* 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). + * * The returned facet is guaranteed to be valid. */ static struct facet * -facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow) +facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow, + uint32_t hash) { - struct facet *facet = facet_find(ofproto, flow); + struct facet *facet; - /* The facet we found might not be valid, since we could be in need of - * revalidation. If it is not valid, don't return it. */ + facet = facet_find(ofproto, flow, hash); if (facet && (ofproto->need_revalidate - || tag_set_intersects(&ofproto->revalidate_set, facet->tags)) - && !facet_revalidate(facet)) { - COVERAGE_INC(facet_invalidated); - return NULL; + || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) { + facet_revalidate(facet); } 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) { @@ -3419,23 +4346,19 @@ facet_check_consistency(struct facet *facet) struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); + uint64_t odp_actions_stub[1024 / 8]; + struct ofpbuf odp_actions; + 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, 0); - if (!rule) { - if (!VLOG_DROP_WARN(&rl)) { - char *s = flow_to_string(&facet->flow); - VLOG_WARN("%s: facet should not exist", s); - free(s); - } - return false; - } else if (rule != facet->rule) { + rule = rule_dpif_lookup(ofproto, &facet->flow); + ok = rule == facet->rule; + if (!ok) { may_log = !VLOG_DROP_WARN(&rl); - ok = false; if (may_log) { struct ds s; @@ -3452,80 +4375,79 @@ facet_check_consistency(struct facet *facet) VLOG_WARN("%s", ds_cstr(&s)); ds_destroy(&s); } - } else { - ok = true; } /* 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 *odp_actions; - bool actions_changed; - bool should_install; + struct ofpbuf key; + struct ds s; action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci, rule, 0, NULL); - odp_actions = xlate_actions(&ctx, rule->up.actions, - rule->up.n_actions); - - should_install = (ctx.may_set_up_flow - && subfacet->key_fitness != ODP_FIT_TOO_LITTLE); - if (!should_install && !subfacet->installed) { - /* The actions for uninstallable flows may vary from one packet to - * the next, so don't compare the actions. */ - goto next; - } - - actions_changed = (subfacet->actions_len != odp_actions->size - || memcmp(subfacet->actions, odp_actions->data, - subfacet->actions_len)); - if (should_install != subfacet->installed || actions_changed) { - if (ok) { - may_log = !VLOG_DROP_WARN(&rl); - ok = false; - } + xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, + &odp_actions); - if (may_log) { - struct odputil_keybuf keybuf; - struct ofpbuf key; - struct ds s; + 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; + } - ds_init(&s); - subfacet_get_key(subfacet, &keybuf, &key); - odp_flow_key_format(key.data, key.size, &s); + 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; + } - ds_put_cstr(&s, ": inconsistency in subfacet"); - if (should_install != subfacet->installed) { - enum odp_key_fitness fitness = subfacet->key_fitness; + if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) { + continue; + } - ds_put_format(&s, " (should%s have been installed)", - should_install ? "" : " not"); - ds_put_format(&s, " (may_set_up_flow=%s, fitness=%s)", - ctx.may_set_up_flow ? "true" : "false", - odp_key_fitness_to_string(fitness)); - } - if (actions_changed) { - 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, ')'); - } - VLOG_WARN("%s", ds_cstr(&s)); - ds_destroy(&s); - } + /* Inconsistency! */ + if (ok) { + may_log = !VLOG_DROP_WARN(&rl); + ok = false; + } + if (!may_log) { + /* Rate-limited, skip reporting. */ + continue; } - next: - ofpbuf_delete(odp_actions); + 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, ')'); + } + VLOG_WARN("%s", ds_cstr(&s)); + ds_destroy(&s); } + ofpbuf_uninit(&odp_actions); return ok; } @@ -3536,12 +4458,8 @@ 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. - * - * - If there is none, destroys 'facet'. - * - * Returns true if 'facet' still exists, false if it has been destroyed. */ -static bool + * where it is and recompiles its actions anyway. */ +static void facet_revalidate(struct facet *facet) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); @@ -3552,20 +4470,16 @@ facet_revalidate(struct facet *facet) struct actions *new_actions; struct action_xlate_ctx ctx; + uint64_t odp_actions_stub[1024 / 8]; + struct ofpbuf odp_actions; + struct rule_dpif *new_rule; struct subfacet *subfacet; - bool actions_changed; int i; COVERAGE_INC(facet_revalidate); - /* Determine the new rule. */ - new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0); - if (!new_rule) { - /* No new rule, so delete the facet. */ - facet_remove(facet); - return false; - } + new_rule = rule_dpif_lookup(ofproto, &facet->flow); /* Calculate new datapath actions. * @@ -3578,43 +4492,36 @@ facet_revalidate(struct facet *facet) 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) { - struct ofpbuf *odp_actions; - bool should_install; + enum slow_path_reason slow; action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci, new_rule, 0, NULL); - odp_actions = xlate_actions(&ctx, new_rule->up.actions, - new_rule->up.n_actions); - actions_changed = (subfacet->actions_len != odp_actions->size - || memcmp(subfacet->actions, odp_actions->data, - subfacet->actions_len)); - - should_install = (ctx.may_set_up_flow - && subfacet->key_fitness != ODP_FIT_TOO_LITTLE); - if (actions_changed || should_install != subfacet->installed) { - if (should_install) { - struct dpif_flow_stats stats; + xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len, + &odp_actions); - subfacet_install(subfacet, - odp_actions->data, odp_actions->size, &stats); - subfacet_update_stats(subfacet, &stats); - } else { - subfacet_uninstall(subfacet); - } + slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; + if (subfacet_should_install(subfacet, slow, &odp_actions)) { + struct dpif_flow_stats stats; + + subfacet_install(subfacet, + odp_actions.data, odp_actions.size, &stats, slow); + subfacet_update_stats(subfacet, &stats); if (!new_actions) { new_actions = xcalloc(list_size(&facet->subfacets), sizeof *new_actions); } - new_actions[i].odp_actions = xmemdup(odp_actions->data, - odp_actions->size); - new_actions[i].actions_len = odp_actions->size; + new_actions[i].odp_actions = xmemdup(odp_actions.data, + odp_actions.size); + new_actions[i].actions_len = odp_actions.size; } - ofpbuf_delete(odp_actions); i++; } + ofpbuf_uninit(&odp_actions); + if (new_actions) { facet_flush_stats(facet); } @@ -3622,23 +4529,24 @@ facet_revalidate(struct facet *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->may_install = ctx.may_set_up_flow; facet->has_learn = ctx.has_learn; facet->has_normal = ctx.has_normal; facet->has_fin_timeout = ctx.has_fin_timeout; facet->mirrors = ctx.mirrors; - if (new_actions) { - i = 0; - LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { - if (new_actions[i].odp_actions) { - free(subfacet->actions); - subfacet->actions = new_actions[i].odp_actions; - subfacet->actions_len = new_actions[i].actions_len; - } - i++; + + 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; } - free(new_actions); + i++; } + free(new_actions); + if (facet->rule != new_rule) { COVERAGE_INC(facet_changed_rule); list_remove(&facet->list_node); @@ -3647,8 +4555,6 @@ facet_revalidate(struct facet *facet) facet->used = new_rule->up.created; facet->prev_used = facet->used; } - - return true; } /* Updates 'facet''s used time. Caller is responsible for calling @@ -3677,76 +4583,61 @@ facet_reset_counters(struct facet *facet) static void facet_push_stats(struct facet *facet) { - uint64_t new_packets, new_bytes; + 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); - new_packets = facet->packet_count - facet->prev_packet_count; - new_bytes = facet->byte_count - facet->prev_byte_count; + 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; - if (new_packets || new_bytes || facet->used > facet->prev_used) { + 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, - new_packets, new_bytes, facet->used); + flow_push_stats(facet->rule, &facet->flow, &stats); update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto), - facet->mirrors, new_packets, new_bytes); + facet->mirrors, stats.n_packets, stats.n_bytes); } } -struct ofproto_push { - struct action_xlate_ctx ctx; - uint64_t packets; - uint64_t bytes; - long long int used; -}; - static void -push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) +rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats) { - struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx); - - if (rule) { - rule->packet_count += push->packets; - rule->byte_count += push->bytes; - ofproto_rule_update_used(&rule->up, push->used); - } + rule->packet_count += stats->n_packets; + rule->byte_count += stats->n_bytes; + ofproto_rule_update_used(&rule->up, stats->used); } /* 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, uint64_t packets, uint64_t bytes, - long long int used) + const struct flow *flow, const struct dpif_flow_stats *stats) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); - struct ofproto_push push; - - push.packets = packets; - push.bytes = bytes; - push.used = used; + struct action_xlate_ctx ctx; - ofproto_rule_update_used(&rule->up, used); + ofproto_rule_update_used(&rule->up, stats->used); - action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, rule, + action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule, 0, NULL); - push.ctx.resubmit_hook = push_resubmit; - ofpbuf_delete(xlate_actions(&push.ctx, - rule->up.actions, rule->up.n_actions)); + 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 ofproto_dpif *ofproto, + const struct nlattr *key, size_t key_len, uint32_t key_hash, + const struct flow *flow) { struct subfacet *subfacet; @@ -3764,65 +4655,70 @@ subfacet_find__(struct ofproto_dpif *ofproto, } /* Searches 'facet' (within 'ofproto') for a subfacet with the specified - * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if - * there is one, otherwise creates and returns a new subfacet. + * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the + * existing subfacet if there is one, otherwise creates and returns a + * new subfacet. * * If the returned subfacet is new, then subfacet->actions will be NULL, in * which case the caller must populate the actions with * subfacet_make_actions(). */ static struct subfacet * -subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness, - const struct nlattr *key, size_t key_len, ovs_be16 initial_tci) +subfacet_create(struct facet *facet, struct flow_miss *miss, + long long int now) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); - uint32_t key_hash = odp_flow_key_hash(key, key_len); + enum odp_key_fitness key_fitness = miss->key_fitness; + const struct nlattr *key = miss->key; + size_t key_len = miss->key_len; + uint32_t key_hash; struct subfacet *subfacet; - subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow); - if (subfacet) { - if (subfacet->facet == facet) { - return subfacet; + key_hash = odp_flow_key_hash(key, key_len); + + if (list_is_empty(&facet->subfacets)) { + subfacet = &facet->one_subfacet; + } else { + subfacet = subfacet_find(ofproto, key, key_len, key_hash, + &facet->flow); + if (subfacet) { + if (subfacet->facet == facet) { + return subfacet; + } + + /* This shouldn't happen. */ + VLOG_ERR_RL(&rl, "subfacet with wrong facet"); + subfacet_destroy(subfacet); } - /* This shouldn't happen. */ - VLOG_ERR_RL(&rl, "subfacet with wrong facet"); - subfacet_destroy(subfacet); + subfacet = xmalloc(sizeof *subfacet); } - subfacet = xzalloc(sizeof *subfacet); hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash); list_push_back(&facet->subfacets, &subfacet->list_node); subfacet->facet = facet; - subfacet->used = time_msec(); 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->installed = false; - subfacet->initial_tci = initial_tci; + subfacet->used = 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; return subfacet; } -/* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and - * 'flow'. Returns the subfacet if one exists, otherwise NULL. */ -static struct subfacet * -subfacet_find(struct ofproto_dpif *ofproto, - const struct nlattr *key, size_t key_len) -{ - uint32_t key_hash = odp_flow_key_hash(key, key_len); - enum odp_key_fitness fitness; - struct flow flow; - - fitness = odp_flow_key_to_flow(key, key_len, &flow); - if (fitness == ODP_FIT_ERROR) { - return NULL; - } - - return subfacet_find__(ofproto, key, key_len, key_hash, &flow); -} - /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from * its facet within 'ofproto', and frees it. */ static void @@ -3836,7 +4732,9 @@ subfacet_destroy__(struct subfacet *subfacet) list_remove(&subfacet->list_node); free(subfacet->key); free(subfacet->actions); - free(subfacet); + if (subfacet != &facet->one_subfacet) { + free(subfacet); + } } /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the @@ -3854,6 +4752,34 @@ subfacet_destroy(struct subfacet *subfacet) } } +static void +subfacet_destroy_batch(struct ofproto_dpif *ofproto, + struct subfacet **subfacets, int n) +{ + struct odputil_keybuf keybufs[SUBFACET_DESTROY_MAX_BATCH]; + struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH]; + struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH]; + struct ofpbuf keys[SUBFACET_DESTROY_MAX_BATCH]; + struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH]; + int i; + + for (i = 0; i < n; i++) { + ops[i].type = DPIF_OP_FLOW_DEL; + subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]); + ops[i].u.flow_del.key = keys[i].data; + ops[i].u.flow_del.key_len = keys[i].size; + ops[i].u.flow_del.stats = &stats[i]; + opsp[i] = &ops[i]; + } + + dpif_operate(ofproto->backer->dpif, opsp, n); + for (i = 0; i < n; i++) { + subfacet_reset_dp_stats(subfacets[i], &stats[i]); + subfacets[i]->path = SF_NOT_INSTALLED; + subfacet_destroy(subfacets[i]); + } +} + /* 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. */ @@ -3861,43 +4787,47 @@ 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, &subfacet->facet->flow); + 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. */ +/* 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) +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 ofpbuf *odp_actions; + struct action_xlate_ctx ctx; action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci, rule, 0, packet); - odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions); + xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions); facet->tags = ctx.tags; - facet->may_install = ctx.may_set_up_flow; 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); } - - ofpbuf_delete(odp_actions); } /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len' @@ -3909,10 +4839,13 @@ subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet) static int subfacet_install(struct subfacet *subfacet, const struct nlattr *actions, size_t actions_len, - struct dpif_flow_stats *stats) + struct dpif_flow_stats *stats, + enum slow_path_reason slow) { struct facet *facet = subfacet->facet; struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); + enum subfacet_path path = subfacet_want_path(slow); + uint64_t slow_path_stub[128 / 8]; struct odputil_keybuf keybuf; enum dpif_flow_put_flags flags; struct ofpbuf key; @@ -3923,22 +4856,38 @@ subfacet_install(struct subfacet *subfacet, flags |= DPIF_FP_ZERO_STATS; } + if (path == SF_SLOW_PATH) { + compose_slow_path(ofproto, &facet->flow, slow, + slow_path_stub, sizeof slow_path_stub, + &actions, &actions_len); + } + subfacet_get_key(subfacet, &keybuf, &key); - ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size, + ret = dpif_flow_put(ofproto->backer->dpif, flags, key.data, key.size, actions, actions_len, stats); if (stats) { subfacet_reset_dp_stats(subfacet, stats); } + if (!ret) { + 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->installed) { + 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; @@ -3947,12 +4896,13 @@ subfacet_uninstall(struct subfacet *subfacet) int error; subfacet_get_key(subfacet, &keybuf, &key); - error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats); + error = dpif_flow_del(ofproto->backer->dpif, + key.data, key.size, &stats); subfacet_reset_dp_stats(subfacet, &stats); if (!error) { subfacet_update_stats(subfacet, &stats); } - subfacet->installed = false; + subfacet->path = SF_NOT_INSTALLED; } else { assert(subfacet->dp_packet_count == 0); assert(subfacet->dp_byte_count == 0); @@ -4015,8 +4965,21 @@ subfacet_update_stats(struct subfacet *subfacet, /* Rules. */ static struct rule_dpif * -rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow, - uint8_t table_id) +rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow) +{ + struct rule_dpif *rule; + + rule = rule_dpif_lookup__(ofproto, flow, 0); + if (rule) { + return rule; + } + + 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 cls_rule *cls_rule; struct classifier *cls; @@ -4040,6 +5003,23 @@ rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow, return rule_dpif_cast(rule_from_cls_rule(cls_rule)); } +static 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); + if (!port) { + VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port); + return ofproto->miss_rule; + } + + if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) { + return ofproto->no_packet_in_rule; + } + return ofproto->miss_rule; +} + static void complete_operation(struct rule_dpif *rule) { @@ -4076,13 +5056,6 @@ rule_construct(struct rule *rule_) struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); struct rule_dpif *victim; uint8_t table_id; - enum ofperr error; - - error = validate_actions(rule->up.actions, rule->up.n_actions, - &rule->up.cr.flow, ofproto->max_ports); - if (error) { - return error; - } rule->packet_count = 0; rule->byte_count = 0; @@ -4109,10 +5082,17 @@ rule_construct(struct rule *rule_) } table_id = rule->up.table_id; - rule->tag = (victim ? victim->tag - : table_id == 0 ? 0 - : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc, - ofproto->tables[table_id].basis)); + if (victim) { + rule->tag = victim->tag; + } else if (table_id == 0) { + rule->tag = 0; + } else { + struct flow flow; + + miniflow_expand(&rule->up.cr.match.flow, &flow); + rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask, + ofproto->tables[table_id].basis); + } complete_operation(rule); return 0; @@ -4157,21 +5137,26 @@ rule_execute(struct rule *rule_, const struct flow *flow, { 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; - struct ofpbuf *odp_actions; - size_t size; + uint64_t odp_actions_stub[1024 / 8]; + struct ofpbuf odp_actions; + 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, packet_get_tcp_flags(packet, flow), packet); - odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions); - size = packet->size; - if (execute_odp_actions(ofproto, flow, odp_actions->data, - odp_actions->size, packet)) { - rule->packet_count++; - rule->byte_count += size; - flow_push_stats(rule, flow, 1, size, time_msec()); - } - ofpbuf_delete(odp_actions); + rule, stats.tcp_flags, packet); + ctx.resubmit_stats = &stats; + xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions); + + execute_odp_actions(ofproto, flow, odp_actions.data, + odp_actions.size, packet); + + ofpbuf_uninit(&odp_actions); return 0; } @@ -4180,15 +5165,6 @@ static void rule_modify_actions(struct rule *rule_) { struct rule_dpif *rule = rule_dpif_cast(rule_); - struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); - enum ofperr error; - - error = validate_actions(rule->up.actions, rule->up.n_actions, - &rule->up.cr.flow, ofproto->max_ports); - if (error) { - ofoperation_complete(rule->up.pending, error); - return; - } complete_operation(rule); } @@ -4202,11 +5178,11 @@ 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; - uint16_t odp_port; + uint32_t odp_port; struct flow flow; int error; - flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow); + flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow); odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port, flow.vlan_tci); if (odp_port != ofport->odp_port) { @@ -4215,13 +5191,14 @@ send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) } ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); - odp_flow_key_from_flow(&key, &flow); + 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->dpif, + error = dpif_execute(ofproto->backer->dpif, key.data, key.size, odp_actions.data, odp_actions.size, packet); @@ -4237,24 +5214,90 @@ send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) /* OpenFlow to datapath action translation. */ -static void do_xlate_actions(const union ofp_action *in, size_t n_in, - struct action_xlate_ctx *ctx); +static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len, + struct action_xlate_ctx *); static void xlate_normal(struct action_xlate_ctx *); +/* 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 struct user_action_cookie *cookie) + const union user_action_cookie *cookie) { uint32_t pid; - pid = dpif_port_get_pid(ofproto->dpif, - ofp_port_to_odp_port(flow->in_port)); + pid = dpif_port_get_pid(ofproto->backer->dpif, + ofp_port_to_odp_port(ofproto, flow->in_port)); return odp_put_userspace_action(pid, cookie, 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, @@ -4262,24 +5305,15 @@ compose_sflow_action(const struct ofproto_dpif *ofproto, const struct flow *flow, uint32_t odp_port) { - uint32_t port_ifindex; uint32_t probability; - struct user_action_cookie cookie; + union user_action_cookie cookie; size_t sample_offset, actions_offset; - int cookie_offset, n_output; + int cookie_offset; if (!ofproto->sflow || flow->in_port == OFPP_NONE) { return 0; } - if (odp_port == OVSP_NONE) { - port_ifindex = 0; - n_output = 0; - } else { - port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port); - n_output = 1; - } - sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE); /* Number of packets out of UINT_MAX to sample. */ @@ -4287,11 +5321,8 @@ compose_sflow_action(const struct ofproto_dpif *ofproto, nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability); actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS); - - cookie.type = USER_ACTION_COOKIE_SFLOW; - cookie.data = port_ifindex; - cookie.n_output = n_output; - cookie.vlan_tci = 0; + 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); @@ -4319,27 +5350,18 @@ static void fix_sflow_action(struct action_xlate_ctx *ctx) { const struct flow *base = &ctx->base_flow; - struct user_action_cookie *cookie; + 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 != NULL); + sizeof(*cookie)); assert(cookie->type == USER_ACTION_COOKIE_SFLOW); - if (ctx->sflow_n_outputs) { - cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow, - ctx->sflow_odp_port); - } - if (ctx->sflow_n_outputs >= 255) { - cookie->n_output = 255; - } else { - cookie->n_output = ctx->sflow_n_outputs; - } - cookie->vlan_tci = base->vlan_tci; + compose_sflow_cookie(ctx->ofproto, base->vlan_tci, + ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie); } static void @@ -4347,7 +5369,7 @@ 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); - uint16_t odp_port = ofp_port_to_odp_port(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; @@ -4355,8 +5377,11 @@ compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port, if (ofport) { struct priority_to_dscp *pdscp; - if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD - || (check_stp && !stp_forward_in_state(ofport->stp_state))) { + 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; } @@ -4394,7 +5419,7 @@ compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port) static void xlate_table_action(struct action_xlate_ctx *ctx, - uint16_t in_port, uint8_t table_id) + uint16_t in_port, uint8_t table_id, bool may_packet_in) { if (ctx->recurse < MAX_RESUBMIT_RECURSION) { struct ofproto_dpif *ofproto = ctx->ofproto; @@ -4408,16 +5433,16 @@ xlate_table_action(struct action_xlate_ctx *ctx, /* 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); + 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 + ctx->tags |= (rule && rule->tag ? rule->tag : rule_calculate_tag(&ctx->flow, - &table->other_table->wc, + &table->other_table->mask, table->basis)); } } @@ -4430,12 +5455,27 @@ xlate_table_action(struct action_xlate_ctx *ctx, 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.actions, rule->up.n_actions, ctx); + do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx); ctx->rule = old_rule; ctx->recurse--; } @@ -4451,18 +5491,23 @@ xlate_table_action(struct action_xlate_ctx *ctx, } static void -xlate_resubmit_table(struct action_xlate_ctx *ctx, - const struct nx_action_resubmit *nar) +xlate_ofpact_resubmit(struct action_xlate_ctx *ctx, + const struct ofpact_resubmit *resubmit) { uint16_t in_port; uint8_t table_id; - in_port = (nar->in_port == htons(OFPP_IN_PORT) - ? ctx->flow.in_port - : ntohs(nar->in_port)); - table_id = nar->table == 255 ? ctx->table_id : nar->table; + 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); + xlate_table_action(ctx, in_port, table_id, false); } static void @@ -4495,7 +5540,7 @@ execute_controller_action(struct action_xlate_ctx *ctx, int len, struct ofputil_packet_in pin; struct ofpbuf *packet; - ctx->may_set_up_flow = false; + ctx->slow |= SLOW_CONTROLLER; if (!ctx->packet) { return; } @@ -4507,7 +5552,13 @@ execute_controller_action(struct action_xlate_ctx *ctx, int len, eth_pop_vlan(packet); eh = packet->l2; - assert(eh->eth_type == ctx->flow.dl_type); + + /* 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); @@ -4548,7 +5599,7 @@ execute_controller_action(struct action_xlate_ctx *ctx, int len, } static bool -compose_dec_ttl(struct action_xlate_ctx *ctx) +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)) { @@ -4559,7 +5610,12 @@ compose_dec_ttl(struct action_xlate_ctx *ctx) ctx->flow.nw_ttl--; return false; } else { - execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0); + 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; @@ -4567,8 +5623,8 @@ compose_dec_ttl(struct action_xlate_ctx *ctx) } static void -xlate_output_action__(struct action_xlate_ctx *ctx, - uint16_t port, uint16_t max_len) +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; @@ -4579,7 +5635,7 @@ xlate_output_action__(struct action_xlate_ctx *ctx, compose_output_action(ctx, ctx->flow.in_port); break; case OFPP_TABLE: - xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id); + xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in); break; case OFPP_NORMAL: xlate_normal(ctx); @@ -4599,6 +5655,8 @@ xlate_output_action__(struct action_xlate_ctx *ctx, default: if (port != ctx->flow.in_port) { compose_output_action(ctx, port); + } else { + xlate_report(ctx, "skipping output to input port"); } break; } @@ -4615,44 +5673,33 @@ xlate_output_action__(struct action_xlate_ctx *ctx, static void xlate_output_reg_action(struct action_xlate_ctx *ctx, - const struct nx_action_output_reg *naor) + const struct ofpact_output_reg *or) { - struct mf_subfield src; - uint64_t ofp_port; - - nxm_decode(&src, naor->src, naor->ofs_nbits); - ofp_port = mf_get_subfield(&src, &ctx->flow); - - if (ofp_port <= UINT16_MAX) { - xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len)); + 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_output_action(struct action_xlate_ctx *ctx, - const struct ofp_action_output *oao) -{ - xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len)); -} - static void xlate_enqueue_action(struct action_xlate_ctx *ctx, - const struct ofp_action_enqueue *oae) + const struct ofpact_enqueue *enqueue) { - uint16_t ofp_port; + uint16_t ofp_port = enqueue->port; + uint32_t queue_id = enqueue->queue; uint32_t flow_priority, priority; int error; - error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id), - &priority); + /* 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, ntohs(oae->port), 0); + xlate_output_action(ctx, enqueue->port, 0, false); return; } - /* Figure out datapath output port. */ - ofp_port = ntohs(oae->port); + /* Check output port. */ if (ofp_port == OFPP_IN_PORT) { ofp_port = ctx->flow.in_port; } else if (ofp_port == ctx->flow.in_port) { @@ -4674,21 +5721,17 @@ xlate_enqueue_action(struct action_xlate_ctx *ctx, } static void -xlate_set_queue_action(struct action_xlate_ctx *ctx, - const struct nx_action_set_queue *nasq) +xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id) { - uint32_t priority; - int error; + uint32_t skb_priority; - error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id), - &priority); - if (error) { - /* Couldn't translate queue to a priority, so ignore. A warning + 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. */ - return; } - - ctx->flow.skb_priority = priority; } struct xlate_reg_state { @@ -4698,9 +5741,9 @@ struct xlate_reg_state { static void xlate_autopath(struct action_xlate_ctx *ctx, - const struct nx_action_autopath *naa) + const struct ofpact_autopath *ap) { - uint16_t ofp_port = ntohl(naa->id); + uint16_t ofp_port = ap->port; struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port); if (!port || !port->bundle) { @@ -4713,7 +5756,7 @@ xlate_autopath(struct action_xlate_ctx *ctx, ofp_port = slave->up.ofp_port; } } - autopath_execute(naa, &ctx->flow, ofp_port); + nxm_reg_load(&ap->dst, ofp_port, &ctx->flow); } static bool @@ -4738,15 +5781,32 @@ slave_enabled_cb(uint16_t ofp_port, void *ofproto_) } } +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 nx_action_learn *learn) + 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; - learn_execute(learn, &ctx->flow, &fm); + 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)) { @@ -4754,7 +5814,7 @@ xlate_learn_action(struct action_xlate_ctx *ctx, ofperr_get_name(error)); } - free(fm.actions); + ofpbuf_uninit(&ofpacts); } /* Reduces '*timeout' to no more than 'max'. A value of zero in either case @@ -4769,13 +5829,13 @@ reduce_timeout(uint16_t max, uint16_t *timeout) static void xlate_fin_timeout(struct action_xlate_ctx *ctx, - const struct nx_action_fin_timeout *naft) + const struct ofpact_fin_timeout *oft) { if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) { struct rule_dpif *rule = ctx->rule; - reduce_timeout(ntohs(naft->fin_idle_timeout), &rule->up.idle_timeout); - reduce_timeout(ntohs(naft->fin_hard_timeout), &rule->up.hard_timeout); + reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout); + reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout); } } @@ -4801,13 +5861,12 @@ may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx) } static void -do_xlate_actions(const union ofp_action *in, size_t n_in, +do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len, struct action_xlate_ctx *ctx) { const struct ofport_dpif *port; - const union ofp_action *ia; bool was_evictable = true; - size_t left; + const struct ofpact *a; port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); if (port && !may_receive(port, ctx)) { @@ -4820,183 +5879,177 @@ do_xlate_actions(const union ofp_action *in, size_t n_in, was_evictable = ctx->rule->up.evictable; ctx->rule->up.evictable = false; } - OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) { - const struct ofp_action_dl_addr *oada; - const struct nx_action_resubmit *nar; - const struct nx_action_set_tunnel *nast; - const struct nx_action_set_queue *nasq; - const struct nx_action_multipath *nam; - const struct nx_action_autopath *naa; - const struct nx_action_bundle *nab; - const struct nx_action_output_reg *naor; - const struct nx_action_controller *nac; - enum ofputil_action_code code; - ovs_be64 tun_id; + OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) { + struct ofpact_controller *controller; + const struct ofpact_metadata *metadata; if (ctx->exit) { break; } - code = ofputil_decode_action_unsafe(ia); - switch (code) { - case OFPUTIL_OFPAT10_OUTPUT: - xlate_output_action(ctx, &ia->output); + 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 OFPUTIL_OFPAT10_SET_VLAN_VID: + case OFPACT_SET_VLAN_VID: ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK); - ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI); + ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid) + | htons(VLAN_CFI)); break; - case OFPUTIL_OFPAT10_SET_VLAN_PCP: + case OFPACT_SET_VLAN_PCP: ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); - ctx->flow.vlan_tci |= htons( - (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI); + ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp + << VLAN_PCP_SHIFT) + | VLAN_CFI); break; - case OFPUTIL_OFPAT10_STRIP_VLAN: + case OFPACT_STRIP_VLAN: ctx->flow.vlan_tci = htons(0); break; - case OFPUTIL_OFPAT10_SET_DL_SRC: - oada = ((struct ofp_action_dl_addr *) ia); - memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN); + 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 OFPUTIL_OFPAT10_SET_DL_DST: - oada = ((struct ofp_action_dl_addr *) ia); - memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN); + case OFPACT_SET_ETH_DST: + memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac, + ETH_ADDR_LEN); break; - case OFPUTIL_OFPAT10_SET_NW_SRC: - ctx->flow.nw_src = ia->nw_addr.nw_addr; + case OFPACT_SET_IPV4_SRC: + ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4; break; - case OFPUTIL_OFPAT10_SET_NW_DST: - ctx->flow.nw_dst = ia->nw_addr.nw_addr; + case OFPACT_SET_IPV4_DST: + ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4; break; - case OFPUTIL_OFPAT10_SET_NW_TOS: + 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 |= ia->nw_tos.nw_tos & IP_DSCP_MASK; + ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp; } break; - case OFPUTIL_OFPAT10_SET_TP_SRC: - ctx->flow.tp_src = ia->tp_port.tp_port; - break; - - case OFPUTIL_OFPAT10_SET_TP_DST: - ctx->flow.tp_dst = ia->tp_port.tp_port; - break; - - case OFPUTIL_OFPAT10_ENQUEUE: - xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia); + case OFPACT_SET_L4_SRC_PORT: + ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port); break; - case OFPUTIL_NXAST_RESUBMIT: - nar = (const struct nx_action_resubmit *) ia; - xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id); + case OFPACT_SET_L4_DST_PORT: + ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port); break; - case OFPUTIL_NXAST_RESUBMIT_TABLE: - xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia); + case OFPACT_RESUBMIT: + xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a)); break; - case OFPUTIL_NXAST_SET_TUNNEL: - nast = (const struct nx_action_set_tunnel *) ia; - tun_id = htonll(ntohl(nast->tun_id)); - ctx->flow.tun_id = tun_id; + case OFPACT_SET_TUNNEL: + ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id); break; - case OFPUTIL_NXAST_SET_QUEUE: - nasq = (const struct nx_action_set_queue *) ia; - xlate_set_queue_action(ctx, nasq); + case OFPACT_SET_QUEUE: + xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id); break; - case OFPUTIL_NXAST_POP_QUEUE: + case OFPACT_POP_QUEUE: ctx->flow.skb_priority = ctx->orig_skb_priority; break; - case OFPUTIL_NXAST_REG_MOVE: - nxm_execute_reg_move((const struct nx_action_reg_move *) ia, - &ctx->flow); - break; - - case OFPUTIL_NXAST_REG_LOAD: - nxm_execute_reg_load((const struct nx_action_reg_load *) ia, - &ctx->flow); + case OFPACT_REG_MOVE: + nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow); break; - case OFPUTIL_NXAST_NOTE: - /* Nothing to do. */ + case OFPACT_REG_LOAD: + nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow); break; - case OFPUTIL_NXAST_SET_TUNNEL64: - tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id; - ctx->flow.tun_id = tun_id; + case OFPACT_DEC_TTL: + if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) { + goto out; + } break; - case OFPUTIL_NXAST_MULTIPATH: - nam = (const struct nx_action_multipath *) ia; - multipath_execute(nam, &ctx->flow); + case OFPACT_NOTE: + /* Nothing to do. */ break; - case OFPUTIL_NXAST_AUTOPATH: - naa = (const struct nx_action_autopath *) ia; - xlate_autopath(ctx, naa); + case OFPACT_MULTIPATH: + multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow); break; - case OFPUTIL_NXAST_BUNDLE: - ctx->ofproto->has_bundle_action = true; - nab = (const struct nx_action_bundle *) ia; - xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow, - slave_enabled_cb, - ctx->ofproto), 0); + case OFPACT_AUTOPATH: + xlate_autopath(ctx, ofpact_get_AUTOPATH(a)); break; - case OFPUTIL_NXAST_BUNDLE_LOAD: + case OFPACT_BUNDLE: ctx->ofproto->has_bundle_action = true; - nab = (const struct nx_action_bundle *) ia; - bundle_execute_load(nab, &ctx->flow, slave_enabled_cb, - ctx->ofproto); + xlate_bundle_action(ctx, ofpact_get_BUNDLE(a)); break; - case OFPUTIL_NXAST_OUTPUT_REG: - naor = (const struct nx_action_output_reg *) ia; - xlate_output_reg_action(ctx, naor); + case OFPACT_OUTPUT_REG: + xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a)); break; - case OFPUTIL_NXAST_LEARN: + case OFPACT_LEARN: ctx->has_learn = true; - if (ctx->may_flow_mod) { - xlate_learn_action(ctx, (const struct nx_action_learn *) ia); - } - break; - - case OFPUTIL_NXAST_DEC_TTL: - if (compose_dec_ttl(ctx)) { - goto out; + if (ctx->may_learn) { + xlate_learn_action(ctx, ofpact_get_LEARN(a)); } break; - case OFPUTIL_NXAST_EXIT: + case OFPACT_EXIT: ctx->exit = true; break; - case OFPUTIL_NXAST_FIN_TIMEOUT: + case OFPACT_FIN_TIMEOUT: ctx->has_fin_timeout = true; - xlate_fin_timeout(ctx, (const struct nx_action_fin_timeout *) ia); + 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 OFPUTIL_NXAST_CONTROLLER: - nac = (const struct nx_action_controller *) ia; - execute_controller_action(ctx, ntohs(nac->max_len), nac->reason, - ntohs(nac->controller_id)); + 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: @@ -5020,28 +6073,39 @@ action_xlate_ctx_init(struct action_xlate_ctx *ctx, ctx->ofproto = ofproto; ctx->flow = *flow; ctx->base_flow = ctx->flow; - ctx->base_flow.tun_id = 0; + 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_macs = packet != NULL; - ctx->may_flow_mod = packet != NULL; + ctx->may_learn = packet != NULL; ctx->tcp_flags = tcp_flags; ctx->resubmit_hook = NULL; + ctx->report_hook = NULL; + ctx->resubmit_stats = NULL; } -static struct ofpbuf * +/* 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 union ofp_action *in, size_t n_in) + const struct ofpact *ofpacts, size_t ofpacts_len, + struct ofpbuf *odp_actions) { - struct flow orig_flow = ctx->flow; + /* 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); - ctx->odp_actions = ofpbuf_new(512); - ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE); + ofpbuf_clear(odp_actions); + ofpbuf_reserve(odp_actions, NL_A_U32_SIZE); + + ctx->odp_actions = odp_actions; ctx->tags = 0; - ctx->may_set_up_flow = true; + ctx->slow = 0; ctx->has_learn = false; ctx->has_normal = false; ctx->has_fin_timeout = false; @@ -5053,6 +6117,16 @@ xlate_actions(struct action_xlate_ctx *ctx, 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: @@ -5062,7 +6136,7 @@ xlate_actions(struct action_xlate_ctx *ctx, break; case OFPC_FRAG_DROP: - return ctx->odp_actions; + return; case OFPC_FRAG_REASM: NOT_REACHED(); @@ -5076,43 +6150,70 @@ xlate_actions(struct action_xlate_ctx *ctx, } } - if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) { - ctx->may_set_up_flow = false; - return ctx->odp_actions; + 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); - struct flow original_flow = ctx->flow; ovs_be16 initial_tci = ctx->base_flow.vlan_tci; add_sflow_action(ctx); - do_xlate_actions(in, n_in, ctx); - - if (ctx->max_resubmit_trigger && !ctx->resubmit_hook - && !VLOG_DROP_ERR(&trace_rl)) { - struct ds ds = DS_EMPTY_INITIALIZER; - - ofproto_trace(ctx->ofproto, &original_flow, ctx->packet, - initial_tci, &ds); - VLOG_ERR("Trace triggered by excessive resubmit recursion:\n%s", - ds_cstr(&ds)); - ds_destroy(&ds); + 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->may_set_up_flow = false; + 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); } } - add_mirror_actions(ctx, &orig_flow); + 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); +} - return ctx->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. */ @@ -5296,47 +6397,6 @@ vlan_is_mirrored(const struct ofmirror *m, int vlan) return !m->vlans || bitmap_is_set(m->vlans, vlan); } -/* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored - * to a VLAN. In general most packets may be mirrored but we want to drop - * protocols that may confuse switches. */ -static bool -eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN]) -{ - /* If you change this function's behavior, please update corresponding - * documentation in vswitch.xml at the same time. */ - if (dst[0] != 0x01) { - /* All the currently banned MACs happen to start with 01 currently, so - * this is a quick way to eliminate most of the good ones. */ - } else { - if (eth_addr_is_reserved(dst)) { - /* Drop STP, IEEE pause frames, and other reserved protocols - * (01-80-c2-00-00-0x). */ - return false; - } - - if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) { - /* Cisco OUI. */ - if ((dst[3] & 0xfe) == 0xcc && - (dst[4] & 0xfe) == 0xcc && - (dst[5] & 0xfe) == 0xcc) { - /* Drop the following protocols plus others following the same - pattern: - - CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc) - Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd) - STP Uplink Fast (01-00-0c-cd-cd-cd) */ - return false; - } - - if (!(dst[3] | dst[4] | dst[5])) { - /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */ - return false; - } - } - } - return true; -} - static void add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow) { @@ -5349,7 +6409,7 @@ add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow) size_t left; in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port, - ctx->packet != NULL); + ctx->packet != NULL, NULL); if (!in_bundle) { return; } @@ -5403,7 +6463,7 @@ add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow) m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; if (!vlan_is_mirrored(m, vlan)) { - mirrors &= mirrors - 1; + mirrors = zero_rightmost_1bit(mirrors); continue; } @@ -5411,8 +6471,8 @@ add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow) ctx->mirrors |= m->dup_mirrors; if (m->out) { output_normal(ctx, m->out, vlan); - } else if (eth_dst_may_rspan(orig_flow->dl_dst) - && vlan != 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) { @@ -5433,7 +6493,7 @@ update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors, return; } - for (; mirrors; mirrors &= mirrors - 1) { + for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) { struct ofmirror *m; m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; @@ -5509,22 +6569,26 @@ update_learning_table(struct ofproto_dpif *ofproto, } static struct ofbundle * -lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn) +lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port, + bool warn, struct ofport_dpif **in_ofportp) { struct ofport_dpif *ofport; - /* 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; - } - /* 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 @@ -5534,6 +6598,9 @@ lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn) * 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); @@ -5559,14 +6626,17 @@ lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn) * so in one special case. */ static bool -is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow, - struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags) +is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port, + uint16_t vlan) { + struct ofproto_dpif *ofproto = ctx->ofproto; + struct flow *flow = &ctx->flow; struct ofbundle *in_bundle = in_port->bundle; /* Drop frames for reserved multicast addresses * only if forward_bpdu option is absent. */ - if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) { + if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) { + xlate_report(ctx, "packet has reserved destination MAC, dropping"); return false; } @@ -5574,11 +6644,12 @@ is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow, struct mac_entry *mac; switch (bond_check_admissibility(in_bundle->bond, in_port, - flow->dl_dst, tags)) { + flow->dl_dst, &ctx->tags)) { case BV_ACCEPT: break; case BV_DROP: + xlate_report(ctx, "bonding refused admissibility, dropping"); return false; case BV_DROP_IF_MOVED: @@ -5586,6 +6657,8 @@ is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow, 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; @@ -5607,15 +6680,12 @@ xlate_normal(struct action_xlate_ctx *ctx) ctx->has_normal = true; in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port, - ctx->packet != NULL); + ctx->packet != NULL, &in_port); if (!in_bundle) { + xlate_report(ctx, "no input bundle, dropping"); return; } - /* We know 'in_port' exists unless it is "ofpp_none_bundle", - * since lookup_input_bundle() succeeded. */ - in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); - /* Drop malformed frames. */ if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) && !(ctx->flow.vlan_tci & htons(VLAN_CFI))) { @@ -5625,6 +6695,7 @@ xlate_normal(struct action_xlate_ctx *ctx) "VLAN tag received on port %s", ctx->ofproto->up.name, in_bundle->name); } + xlate_report(ctx, "partial VLAN tag, dropping"); return; } @@ -5636,24 +6707,25 @@ xlate_normal(struct action_xlate_ctx *ctx) "%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; } /* 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); /* Check other admissibility requirements. */ - if (in_port && - !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) { + if (in_port && !is_admissible(ctx, in_port, vlan)) { return; } /* Learn source MAC. */ - if (ctx->may_learn_macs) { + if (ctx->may_learn) { update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle); } @@ -5662,11 +6734,15 @@ xlate_normal(struct action_xlate_ctx *ctx) &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) @@ -5700,18 +6776,17 @@ xlate_normal(struct action_xlate_ctx *ctx) * a few more, but not all of the facets or even all of the facets that * resubmit to the table modified by MAC learning). */ -/* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted +/* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted * into an OpenFlow table with the given 'basis'. */ static tag_type -rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc, +rule_calculate_tag(const struct flow *flow, const struct minimask *mask, uint32_t secret) { - if (flow_wildcards_is_catchall(wc)) { + if (minimask_is_catchall(mask)) { return 0; } else { - struct flow tag_flow = *flow; - flow_zero_wildcards(&tag_flow, wc); - return tag_create_deterministic(flow_hash(&tag_flow, secret)); + uint32_t hash = flow_hash_in_minimask(flow, mask, secret); + return tag_create_deterministic(hash); } } @@ -5764,7 +6839,7 @@ 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 = true; + ofproto->need_revalidate = REV_FLOW_TABLE; } } @@ -5788,7 +6863,7 @@ rule_invalidate(const struct rule_dpif *rule) if (table->other_table && rule->tag) { tag_set_add(&ofproto->revalidate_set, rule->tag); } else { - ofproto->need_revalidate = true; + ofproto->need_revalidate = REV_FLOW_TABLE; } } } @@ -5800,7 +6875,7 @@ set_frag_handling(struct ofproto *ofproto_, struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); if (frag_handling != OFPC_FRAG_REASM) { - ofproto->need_revalidate = true; + ofproto->need_revalidate = REV_RECONFIGURE; return true; } else { return false; @@ -5810,42 +6885,36 @@ set_frag_handling(struct ofproto *ofproto_, static enum ofperr packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, const struct flow *flow, - const union ofp_action *ofp_actions, size_t n_ofp_actions) + const struct ofpact *ofpacts, size_t ofpacts_len) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - enum ofperr error; + struct odputil_keybuf keybuf; + struct dpif_flow_stats stats; - if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) { - return OFPERR_NXBRC_BAD_IN_PORT; - } + struct ofpbuf key; - error = validate_actions(ofp_actions, n_ofp_actions, flow, - ofproto->max_ports); - if (!error) { - struct odputil_keybuf keybuf; - struct ofpbuf *odp_actions; - struct ofproto_push push; - 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); + ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); + odp_flow_key_from_flow(&key, flow, + ofp_port_to_odp_port(ofproto, flow->in_port)); - action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, NULL, - packet_get_tcp_flags(packet, flow), packet); + dpif_flow_stats_extract(flow, packet, time_msec(), &stats); - /* Ensure that resubmits in 'ofp_actions' get accounted to their - * matching rules. */ - push.packets = 1; - push.bytes = packet->size; - push.used = time_msec(); - push.ctx.resubmit_hook = push_resubmit; + action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL, + packet_get_tcp_flags(packet, flow), packet); + ctx.resubmit_stats = &stats; - odp_actions = xlate_actions(&push.ctx, ofp_actions, n_ofp_actions); - dpif_execute(ofproto->dpif, key.data, key.size, - odp_actions->data, odp_actions->size, packet); - ofpbuf_delete(odp_actions); - } - return error; + ofpbuf_use_stub(&odp_actions, + odp_actions_stub, sizeof odp_actions_stub); + xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions); + dpif_execute(ofproto->backer->dpif, key.data, key.size, + odp_actions.data, odp_actions.size, packet); + ofpbuf_uninit(&odp_actions); + + return 0; } /* NetFlow. */ @@ -5874,7 +6943,7 @@ get_netflow_ids(const struct ofproto *ofproto_, { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id); + dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id); } static void @@ -5886,11 +6955,10 @@ send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) struct ofexpired expired; LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { - if (subfacet->installed) { + if (subfacet->path == SF_FAST_PATH) { struct dpif_flow_stats stats; - subfacet_install(subfacet, subfacet->actions, - subfacet->actions_len, &stats); + subfacet_reinstall(subfacet, &stats); subfacet_update_stats(subfacet, &stats); } } @@ -5998,7 +7066,7 @@ trace_format_rule(struct ds *result, uint8_t table_id, int level, ds_put_char_multiple(result, '\t', level); ds_put_cstr(result, "OpenFlow "); - ofp_print_actions(result, rule->up.actions, rule->up.n_actions); + ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result); ds_put_char(result, '\n'); } @@ -6056,6 +7124,17 @@ trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule); } +static void +trace_report(struct action_xlate_ctx *ctx, const char *s) +{ + struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx); + struct ds *result = trace->result; + + ds_put_char_multiple(result, '\t', ctx->recurse); + ds_put_cstr(result, s); + ds_put_char(result, '\n'); +} + static void ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], void *aux OVS_UNUSED) @@ -6083,23 +7162,51 @@ ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], /* ofproto/trace dpname flow [-generate] */ const char *flow_s = argv[2]; const char *generate_s = argv[3]; - 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; - } + /* 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; + } - /* Convert odp_key to flow. */ - error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data, - odp_key.size, &flow, - &initial_tci, NULL); - if (error == ODP_FIT_ERROR) { - unixctl_command_reply_error(conn, "Invalid flow"); - goto exit; + 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"); + 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); + goto exit; + } + + initial_tci = flow.vlan_tci; + vsp_adjust_flow(ofproto, &flow); } /* Generate a packet, if requested. */ @@ -6107,15 +7214,17 @@ ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], packet = ofpbuf_new(0); flow_compose(packet, &flow); } - } else if (argc == 6) { - /* ofproto/trace dpname priority tun_id in_port packet */ + } else if (argc == 7) { + /* ofproto/trace dpname priority tun_id in_port mark packet */ const char *priority_s = argv[2]; const char *tun_id_s = argv[3]; const char *in_port_s = argv[4]; - const char *packet_s = argv[5]; - uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s)); + const char *mark_s = argv[5]; + const char *packet_s = argv[6]; + uint32_t in_port = atoi(in_port_s); ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0)); uint32_t priority = atoi(priority_s); + uint32_t mark = atoi(mark_s); const char *msg; msg = eth_from_hex(packet_s, &packet); @@ -6129,7 +7238,8 @@ ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], ds_put_cstr(&result, s); free(s); - flow_extract(packet, priority, tun_id, in_port, &flow); + flow_extract(packet, priority, mark, NULL, in_port, &flow); + flow.tunnel.tun_id = tun_id; initial_tci = flow.vlan_tci; } else { unixctl_command_reply_error(conn, "Bad command syntax"); @@ -6156,34 +7266,84 @@ 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, 0); + rule = rule_dpif_lookup(ofproto, flow); + trace_format_rule(ds, 0, 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"); + } + if (rule) { + uint64_t odp_actions_stub[1024 / 8]; + struct ofpbuf odp_actions; + struct trace_ctx trace; - struct ofpbuf *odp_actions; uint8_t tcp_flags; tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0; trace.result = ds; 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; - odp_actions = xlate_actions(&trace.ctx, - rule->up.actions, rule->up.n_actions); + trace.ctx.report_hook = trace_report; + xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len, + &odp_actions); 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_delete(odp_actions); + format_odp_actions(ds, odp_actions.data, odp_actions.size); + ofpbuf_uninit(&odp_actions); + + if (trace.ctx.slow) { + enum slow_path_reason 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; + } - if (!trace.ctx.may_set_up_flow) { - if (packet) { - ds_put_cstr(ds, "\nThis flow is not cachable."); - } else { - ds_put_cstr(ds, "\nThe datapath actions are incomplete--" - "for complete actions, please supply a packet."); + slow &= ~bit; + } + + if (slow & ~SLOW_MATCH) { + ds_put_cstr(ds, "\nThe datapath actions above do not reflect " + "the special slow-path processing."); } } } @@ -6220,7 +7380,7 @@ ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply) } } if (errors) { - ofproto->need_revalidate = true; + ofproto->need_revalidate = REV_INCONSISTENCY; } if (errors) { @@ -6256,6 +7416,212 @@ ofproto_dpif_self_check(struct unixctl_conn *conn, ds_destroy(&reply); } +/* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list + * of the 'ofproto_shash' nodes. It is the responsibility of the caller + * to destroy 'ofproto_shash' and free the returned value. */ +static const struct shash_node ** +get_ofprotos(struct shash *ofproto_shash) +{ + const struct ofproto_dpif *ofproto; + + HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { + char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name); + shash_add_nocopy(ofproto_shash, name, ofproto); + } + + return shash_sort(ofproto_shash); +} + +static void +ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED, + const char *argv[] OVS_UNUSED, + void *aux OVS_UNUSED) +{ + struct ds ds = DS_EMPTY_INITIALIZER; + struct shash ofproto_shash; + const struct shash_node **sorted_ofprotos; + int i; + + shash_init(&ofproto_shash); + sorted_ofprotos = get_ofprotos(&ofproto_shash); + for (i = 0; i < shash_count(&ofproto_shash); i++) { + const struct shash_node *node = sorted_ofprotos[i]; + ds_put_format(&ds, "%s\n", node->name); + } + + shash_destroy(&ofproto_shash); + free(sorted_ofprotos); + + unixctl_command_reply(conn, ds_cstr(&ds)); + ds_destroy(&ds); +} + +static void +show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds) +{ + struct dpif_dp_stats s; + const struct shash_node **ports; + int i; + + dpif_get_dp_stats(ofproto->backer->dpif, &s); + + 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); + + netdev_close(netdev); + } + ds_put_char(ds, ')'); + } + ds_put_char(ds, '\n'); + } + free(ports); +} + +static void +ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc, + const char *argv[], void *aux OVS_UNUSED) +{ + struct ds ds = DS_EMPTY_INITIALIZER; + const struct ofproto_dpif *ofproto; + + if (argc > 1) { + int i; + for (i = 1; i < argc; i++) { + ofproto = ofproto_dpif_lookup(argv[i]); + if (!ofproto) { + ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps " + "for help)", argv[i]); + unixctl_command_reply_error(conn, ds_cstr(&ds)); + return; + } + show_dp_format(ofproto, &ds); + } + } else { + struct shash ofproto_shash; + const struct shash_node **sorted_ofprotos; + int i; + + shash_init(&ofproto_shash); + sorted_ofprotos = get_ofprotos(&ofproto_shash); + for (i = 0; i < shash_count(&ofproto_shash); i++) { + const struct shash_node *node = sorted_ofprotos[i]; + show_dp_format(node->data, &ds); + } + + shash_destroy(&ofproto_shash); + free(sorted_ofprotos); + } + + unixctl_command_reply(conn, ds_cstr(&ds)); + ds_destroy(&ds); +} + +static void +ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn, + int argc OVS_UNUSED, const char *argv[], + void *aux OVS_UNUSED) +{ + struct ds ds = DS_EMPTY_INITIALIZER; + const struct ofproto_dpif *ofproto; + struct subfacet *subfacet; + + ofproto = ofproto_dpif_lookup(argv[1]); + if (!ofproto) { + unixctl_command_reply_error(conn, "no such bridge"); + return; + } + + HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) { + struct odputil_keybuf keybuf; + struct ofpbuf key; + + subfacet_get_key(subfacet, &keybuf, &key); + odp_flow_key_format(key.data, key.size, &ds); + + ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:", + subfacet->dp_packet_count, subfacet->dp_byte_count); + if (subfacet->used) { + ds_put_format(&ds, "%.3fs", + (time_msec() - subfacet->used) / 1000.0); + } else { + ds_put_format(&ds, "never"); + } + if (subfacet->facet->tcp_flags) { + ds_put_cstr(&ds, ", flags:"); + packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags); + } + + ds_put_cstr(&ds, ", actions:"); + format_odp_actions(&ds, subfacet->actions, subfacet->actions_len); + ds_put_char(&ds, '\n'); + } + + unixctl_command_reply(conn, ds_cstr(&ds)); + ds_destroy(&ds); +} + +static void +ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn, + int argc OVS_UNUSED, const char *argv[], + void *aux OVS_UNUSED) +{ + struct ds ds = DS_EMPTY_INITIALIZER; + struct ofproto_dpif *ofproto; + + ofproto = ofproto_dpif_lookup(argv[1]); + if (!ofproto) { + unixctl_command_reply_error(conn, "no such bridge"); + return; + } + + flush(&ofproto->up); + + unixctl_command_reply(conn, ds_cstr(&ds)); + ds_destroy(&ds); +} + static void ofproto_dpif_unixctl_init(void) { @@ -6267,8 +7633,8 @@ ofproto_dpif_unixctl_init(void) unixctl_command_register( "ofproto/trace", - "bridge {tun_id in_port packet | odp_flow [-generate]}", - 2, 5, ofproto_unixctl_trace, NULL); + "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}", + 2, 6, ofproto_unixctl_trace, NULL); unixctl_command_register("fdb/flush", "[bridge]", 0, 1, ofproto_unixctl_fdb_flush, NULL); unixctl_command_register("fdb/show", "bridge", 1, 1, @@ -6279,6 +7645,14 @@ ofproto_dpif_unixctl_init(void) ofproto_dpif_unclog, NULL); unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1, ofproto_dpif_self_check, NULL); + unixctl_command_register("dpif/dump-dps", "", 0, 0, + ofproto_unixctl_dpif_dump_dps, NULL); + unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX, + ofproto_unixctl_dpif_show, NULL); + unixctl_command_register("dpif/dump-flows", "bridge", 1, 1, + ofproto_unixctl_dpif_dump_flows, NULL); + unixctl_command_register("dpif/del-flows", "bridge", 1, 1, + ofproto_unixctl_dpif_del_flows, NULL); } /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) @@ -6299,7 +7673,7 @@ set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid) return 0; } - ofproto->need_revalidate = true; + ofproto->need_revalidate = REV_RECONFIGURE; if (ofport->realdev_ofp_port) { vsp_remove(ofport); @@ -6338,16 +7712,17 @@ vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto, uint32_t realdev_odp_port, ovs_be16 vlan_tci) { if (!hmap_is_empty(&ofproto->realdev_vid_map)) { - uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port); + uint16_t realdev_ofp_port; int vid = vlan_tci_to_vid(vlan_tci); const struct vlan_splinter *vsp; + realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port); HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node, hash_realdev_vid(realdev_ofp_port, vid), &ofproto->realdev_vid_map) { if (vsp->realdev_ofp_port == realdev_ofp_port && vsp->vid == vid) { - return ofp_port_to_odp_port(vsp->vlandev_ofp_port); + return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port); } } } @@ -6396,6 +7771,30 @@ vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto, return 0; } +/* Given 'flow', a flow representing a packet received on 'ofproto', checks + * whether 'flow->in_port' represents a Linux VLAN device. If so, changes + * 'flow->in_port' to the "real" device backing the VLAN device, sets + * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is + * always the case unless VLAN splinters are enabled), returns false without + * making any changes. */ +static bool +vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow) +{ + uint16_t realdev; + int vid; + + realdev = vsp_vlandev_to_realdev(ofproto, flow->in_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->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI); + return true; +} + static void vsp_remove(struct ofport_dpif *port) { @@ -6438,11 +7837,52 @@ vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid) VLOG_ERR("duplicate vlan device record"); } } - + +static uint32_t +ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port) +{ + const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); + return ofport ? ofport->odp_port : OVSP_NONE; +} + +static struct ofport_dpif * +odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port) +{ + struct ofport_dpif *port; + + HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, + hash_int(odp_port, 0), + &backer->odp_to_ofport_map) { + if (port->odp_port == odp_port) { + return port; + } + } + + return NULL; +} + +static uint16_t +odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port) +{ + struct ofport_dpif *port; + + port = odp_port_to_ofport(ofproto->backer, odp_port); + if (port && ofproto == ofproto_dpif_cast(port->up.ofproto)) { + return port->up.ofp_port; + } else { + return OFPP_NONE; + } +} + const struct ofproto_class ofproto_dpif_class = { + init, enumerate_types, enumerate_names, del, + port_open_type, + type_run, + type_run_fast, + type_wait, alloc, construct, destruct, @@ -6450,6 +7890,7 @@ const struct ofproto_class ofproto_dpif_class = { run, run_fast, wait, + get_memory_usage, flush, get_features, get_tables, @@ -6484,7 +7925,9 @@ const struct ofproto_class ofproto_dpif_class = { set_sflow, set_cfm, get_cfm_fault, + get_cfm_opup, get_cfm_remote_mpids, + get_cfm_health, set_stp, get_stp_status, set_stp_port,