/*
- * Copyright (c) 2009, 2010, 2011 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.
#include "lacp.h"
#include "learn.h"
#include "mac-learning.h"
+#include "meta-flow.h"
#include "multipath.h"
#include "netdev.h"
#include "netlink.h"
#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"
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;
- long long int used; /* Time last used; time created if not used. */
-
/* These statistics:
*
* - Do include packets and bytes from facets that have been deleted or
*
* - Do not include packet or bytes that can be obtained from any facet's
* packet_count or byte_count member or that can be obtained from the
- * datapath by, e.g., dpif_flow_get() for any facet.
+ * datapath by, e.g., dpif_flow_get() for any subfacet.
*/
uint64_t packet_count; /* Number of packets received. */
uint64_t byte_count; /* Number of bytes received. */
}
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 *,
+ const struct dpif_flow_stats *);
+static tag_type rule_calculate_tag(const struct flow *,
+ const struct minimask *, uint32_t basis);
+static void rule_invalidate(const struct rule_dpif *);
#define MAX_MIRRORS 32
typedef uint32_t mirror_mask_t;
struct hmapx dsts; /* Contains "struct ofbundle *"s. */
unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
- /* Output (mutually exclusive). */
+ /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
struct ofbundle *out; /* Output port or NULL. */
int out_vlan; /* Output VLAN or -1. */
+ mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
+
+ /* Counters. */
+ int64_t packet_count; /* Number of packets sent. */
+ int64_t byte_count; /* Number of bytes sent. */
};
static void mirror_destroy(struct ofmirror *);
+static void update_mirror_stats(struct ofproto_dpif *ofproto,
+ mirror_mask_t mirrors,
+ uint64_t packets, uint64_t bytes);
-/* A group of one or more OpenFlow ports. */
-#define OFBUNDLE_FLOOD ((struct ofbundle *) 1)
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. */
* NULL if all VLANs are trunked. */
struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
struct bond *bond; /* Nonnull iff more than one port. */
+ bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
/* Status. */
- bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */
+ bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
/* Port mirroring info. */
mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
static void bundle_del_port(struct ofport_dpif *);
static void bundle_run(struct ofbundle *);
static void bundle_wait(struct ofbundle *);
+static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
+ uint16_t in_port, bool warn,
+ struct ofport_dpif **in_ofportp);
+
+/* A controller may use OFPP_NONE as the ingress port to indicate that
+ * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
+ * when an input bundle is needed for validation (e.g., mirroring or
+ * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
+ * any 'port' structs, so care must be taken when dealing with it. */
+static struct ofbundle ofpp_none_bundle = {
+ .name = "OFPP_NONE",
+ .vlan_mode = PORT_VLAN_TRUNK
+};
static void stp_run(struct ofproto_dpif *ofproto);
static void stp_wait(struct ofproto_dpif *ofproto);
+static int set_stp_port(struct ofport *,
+ const struct ofproto_port_stp_settings *);
+
+static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
struct action_xlate_ctx {
/* action_xlate_ctx_init() initializes these members. */
* revalidating without a packet to refer to. */
const struct ofpbuf *packet;
- /* Should OFPP_NORMAL MAC learning and NXAST_LEARN actions execute? We
- * want to execute them 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. */
+ /* 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;
- /* If nonnull, called just before executing a resubmit action.
+ /* The rule that we are currently translating, or NULL. */
+ struct rule_dpif *rule;
+
+ /* Union of the set of TCP flags seen so far in this flow. (Used only by
+ * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
+ * timeouts.) */
+ uint8_t tcp_flags;
+
+ /* If nonnull, flow translation calls this function just before executing a
+ * resubmit or OFPP_TABLE action. In addition, disables logging of traces
+ * when the recursion depth is exceeded.
+ *
+ * 'rule' is the rule being submitted into. It will be null if the
+ * resubmit or OFPP_TABLE action didn't find a matching rule.
+ *
+ * This is normally null so the client has to set it manually after
+ * calling action_xlate_ctx_init(). */
+ void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
+
+ /* If nonnull, flow translation calls this function to report some
+ * significant decision, e.g. to explain why OFPP_NORMAL translation
+ * dropped a packet. */
+ void (*report_hook)(struct action_xlate_ctx *, const char *s);
+
+ /* If nonnull, flow translation credits the specified statistics to each
+ * rule reached through a resubmit or OFPP_TABLE action.
*
* This is normally null so the client has to set it manually after
* calling action_xlate_ctx_init(). */
- void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
+ 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? */
uint16_t nf_output_iface; /* Output interface index for NetFlow. */
+ mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
/* xlate_actions() initializes and uses these members, but the client has no
* reason to look at them. */
int recurse; /* Recursion level, via xlate_table_action. */
+ bool max_resubmit_trigger; /* Recursed too deeply during translation. */
struct flow base_flow; /* Flow at the last commit. */
- uint32_t original_priority; /* Priority when packet arrived. */
+ 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 *,
- const struct ofpbuf *);
-static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
- const union ofp_action *in, size_t n_in);
+ ovs_be16 initial_tci, struct rule_dpif *,
+ uint8_t tcp_flags, const struct ofpbuf *);
+static void xlate_actions(struct action_xlate_ctx *,
+ const struct ofpact *ofpacts, size_t ofpacts_len,
+ struct ofpbuf *odp_actions);
+static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
+ const struct ofpact *ofpacts,
+ size_t ofpacts_len);
+
+static size_t put_userspace_action(const struct ofproto_dpif *,
+ struct ofpbuf *odp_actions,
+ const struct flow *,
+ const union user_action_cookie *);
+
+static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
+ enum slow_path_reason,
+ uint64_t *stub, size_t stub_size,
+ const struct nlattr **actionsp,
+ size_t *actions_lenp);
+
+static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
+
+/* A subfacet (see "struct subfacet" below) has three possible installation
+ * states:
+ *
+ * - 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;
+};
-/* An exact-match instantiation of an OpenFlow flow. */
+#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.
+ *
+ * A facet associates a "struct flow", which represents the Open vSwitch
+ * userspace idea of an exact-match flow, with one or more subfacets. Each
+ * subfacet tracks the datapath's idea of the exact-match flow equivalent to
+ * the facet. When the kernel module (or other dpif implementation) and Open
+ * vSwitch userspace agree on the definition of a flow key, there is exactly
+ * one subfacet per facet. If the dpif implementation supports more-specific
+ * flow matching than userspace, however, a facet can have more than one
+ * subfacet, each of which corresponds to some distinction in flow that
+ * userspace simply doesn't understand.
+ *
+ * Flow expiration works in terms of subfacets, so a facet must have at least
+ * one subfacet or it will never expire, leaking memory. */
struct facet {
+ /* Owners. */
+ struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
+ struct list list_node; /* In owning rule's 'facets' list. */
+ struct rule_dpif *rule; /* Owning rule. */
+
+ /* Owned data. */
+ struct list subfacets;
long long int used; /* Time last used; time created if not used. */
+ /* Key. */
+ struct flow flow;
+
/* These statistics:
*
* - Do include packets and bytes sent "by hand", e.g. with
* dpif_execute().
*
* - Do include packets and bytes that were obtained from the datapath
- * when its statistics were reset (e.g. dpif_flow_put() with
+ * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
* DPIF_FP_ZERO_STATS).
+ *
+ * - Do not include packets or bytes that can be obtained from the
+ * datapath for any existing subfacet.
*/
uint64_t packet_count; /* Number of packets received. */
uint64_t byte_count; /* Number of bytes received. */
- uint64_t dp_packet_count; /* Last known packet count in the datapath. */
- uint64_t dp_byte_count; /* Last known byte count in the datapath. */
-
- uint64_t rs_packet_count; /* Packets pushed to resubmit children. */
- uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */
- long long int rs_used; /* Used time pushed to resubmit children. */
+ /* Resubmit statistics. */
+ uint64_t prev_packet_count; /* Number of packets from last stats push. */
+ uint64_t prev_byte_count; /* Number of bytes from last stats push. */
+ long long int prev_used; /* Used time from last stats push. */
+ /* Accounting. */
uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
+ struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
+ uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
- struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
- struct list list_node; /* In owning rule's 'facets' list. */
- struct rule_dpif *rule; /* Owning rule. */
- struct flow flow; /* Exact-match flow. */
- bool installed; /* Installed in datapath? */
- bool may_install; /* True ordinarily; false if actions must
- * be reassessed for every packet. */
+ /* Properties of datapath actions.
+ *
+ * Every subfacet has its own actions because actions can differ slightly
+ * between splintered and non-splintered subfacets due to the VLAN tag
+ * being initially different (present vs. absent). All of them have these
+ * properties in common so we just store one copy of them here. */
bool has_learn; /* Actions include NXAST_LEARN? */
bool has_normal; /* Actions output to OFPP_NORMAL? */
- size_t actions_len; /* Number of bytes in actions[]. */
- struct nlattr *actions; /* Datapath actions. */
- tag_type tags; /* Tags. */
- struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
+ 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 void facet_remove(struct ofproto_dpif *, struct facet *);
+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 ofproto_dpif *, struct facet *);
-
-static bool execute_controller_action(struct ofproto_dpif *,
- const struct flow *,
- const struct nlattr *odp_actions,
- size_t actions_len,
- struct ofpbuf *packet);
-static void facet_execute(struct ofproto_dpif *, struct facet *,
- struct ofpbuf *packet);
-
-static int facet_put__(struct ofproto_dpif *, struct facet *,
- const struct nlattr *actions, size_t actions_len,
- struct dpif_flow_stats *);
-static void facet_install(struct ofproto_dpif *, struct facet *,
- bool zero_stats);
-static void facet_uninstall(struct ofproto_dpif *, struct facet *);
-static void facet_flush_stats(struct ofproto_dpif *, struct facet *);
-
-static void facet_make_actions(struct ofproto_dpif *, struct facet *,
- const struct ofpbuf *packet);
-static void facet_update_time(struct ofproto_dpif *, struct facet *,
- long long int used);
-static void facet_update_stats(struct ofproto_dpif *, struct facet *,
- const struct dpif_flow_stats *);
+ 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 *);
+
+static void facet_update_time(struct facet *, long long int used);
static void facet_reset_counters(struct facet *);
-static void facet_reset_dp_stats(struct facet *, struct dpif_flow_stats *);
static void facet_push_stats(struct facet *);
-static void facet_account(struct ofproto_dpif *, struct facet *);
+static void facet_learn(struct facet *);
+static void facet_account(struct facet *);
static bool facet_is_controller_flow(struct facet *);
-static void flow_push_stats(const struct rule_dpif *,
- struct flow *, uint64_t packets, uint64_t bytes,
- long long int used);
-
-static uint32_t rule_calculate_tag(const struct flow *,
- const struct flow_wildcards *,
- uint32_t basis);
-static void rule_invalidate(const struct rule_dpif *);
-
struct ofport_dpif {
+ struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
struct ofport up;
uint32_t odp_port;
tag_type tag; /* Tag associated with this port. */
uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
bool may_enable; /* May be enabled in bonds. */
+ long long int carrier_seq; /* Carrier status changes. */
+ /* Spanning tree. */
struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
long long int stp_state_entered;
+
+ struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
+
+ /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
+ *
+ * This is deprecated. It is only for compatibility with broken device
+ * drivers in old versions of Linux that do not properly support VLANs when
+ * VLAN devices are not used. When broken device drivers are no longer in
+ * widespread use, we will delete these interfaces. */
+ uint16_t realdev_ofp_port;
+ int vlandev_vid;
+};
+
+/* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
+ * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
+ * traffic egressing the 'ofport' with that priority should be marked with. */
+struct priority_to_dscp {
+ struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
+ uint32_t priority; /* Priority of this queue (see struct flow). */
+
+ uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
+};
+
+/* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
+ *
+ * This is deprecated. It is only for compatibility with broken device drivers
+ * in old versions of Linux that do not properly support VLANs when VLAN
+ * devices are not used. When broken device drivers are no longer in
+ * widespread use, we will delete these interfaces. */
+struct vlan_splinter {
+ struct hmap_node realdev_vid_node;
+ struct hmap_node vlandev_node;
+ uint16_t realdev_ofp_port;
+ uint16_t vlandev_ofp_port;
+ int vid;
};
+static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
+ uint32_t realdev, ovs_be16 vlan_tci);
+static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
+static void vsp_remove(struct ofport_dpif *);
+static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
+
+static 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)
{
}
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 *);
struct dpif_completion {
struct list list_node;
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;
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. */
struct list completions;
bool has_bundle_action; /* True when the first bundle action appears. */
+ struct netdev_stats stats; /* To account packets generated and consumed in
+ * userspace. */
/* Spanning tree. */
struct stp *stp;
long long int stp_last_tick;
+
+ /* 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
* for debugging the asynchronous flow_mod implementation.) */
static bool clogged;
+/* All existing ofproto_dpif instances, indexed by ->up.name. */
+static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
+
static void ofproto_dpif_unixctl_init(void);
static struct ofproto_dpif *
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,
+ struct ds *);
/* Packet processing. */
static void update_learning_table(struct ofproto_dpif *,
const struct flow *, int vlan,
struct ofbundle *);
-static bool is_admissible(struct ofproto_dpif *, const struct flow *,
- bool have_packet, tag_type *, int *vlanp,
- struct ofbundle **in_bundlep);
-
/* Upcalls. */
#define FLOW_MISS_MAX_BATCH 50
-static void handle_upcall(struct ofproto_dpif *, struct dpif_upcall *);
-static void handle_miss_upcalls(struct ofproto_dpif *,
- struct dpif_upcall *, size_t n);
+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(struct ofproto_dpif *, uint32_t odp_port,
- const 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 int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
+static size_t compose_sflow_action(const struct ofproto_dpif *,
+ struct ofpbuf *odp_actions,
+ const struct flow *, uint32_t odp_port);
+static void add_mirror_actions(struct action_xlate_ctx *ctx,
+ const struct flow *flow);
/* Global variables. */
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+
+/* Initial mappings of port to bridge mappings. */
+static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
\f
/* 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)
{
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
return error;
}
\f
+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);
+}
+\f
/* Basic life-cycle. */
+static int add_internal_flows(struct ofproto_dpif *);
+
static struct ofproto *
alloc(void)
{
free(ofproto);
}
-static int
-construct(struct ofproto *ofproto_, int *n_tablesp)
+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_mask(ofproto->dpif,
- ((1u << DPIF_UC_MISS) |
- (1u << DPIF_UC_ACTION)));
+ 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;
hmap_init(&ofproto->bundles);
- ofproto->ml = mac_learning_create();
+ ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
for (i = 0; i < MAX_MIRRORS; i++) {
ofproto->mirrors[i] = NULL;
}
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];
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;
- *n_tablesp = N_TABLES;
+ 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)
{
{
struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
struct rule_dpif *rule, *next_rule;
- struct classifier *table;
+ struct oftable *table;
int i;
+ hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
complete_operations(ofproto);
OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
struct cls_cursor cursor;
- cls_cursor_init(&cursor, table, NULL);
+ cls_cursor_init(&cursor, &table->cls, NULL);
CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
ofproto_rule_destroy(&rule->up);
}
mac_learning_destroy(ofproto->ml);
hmap_destroy(&ofproto->facets);
+ hmap_destroy(&ofproto->subfacets);
+ governor_destroy(ofproto->governor);
- dpif_close(ofproto->dpif);
+ hmap_destroy(&ofproto->vlandev_map);
+ hmap_destroy(&ofproto->realdev_vid_map);
+
+ sset_destroy(&ofproto->ports);
+ sset_destroy(&ofproto->port_poll_set);
+
+ close_dpif_backer(ofproto->backer);
}
static int
-run(struct ofproto *ofproto_)
+run_fast(struct ofproto *ofproto_)
{
struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
- struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
struct ofport_dpif *ofport;
- struct ofbundle *bundle;
- size_t n_misses;
- int i;
- if (!clogged) {
- complete_operations(ofproto);
+ HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
+ port_run_fast(ofport);
}
- dpif_run(ofproto->dpif);
- n_misses = 0;
- for (i = 0; i < FLOW_MISS_MAX_BATCH; i++) {
- struct dpif_upcall *upcall = &misses[n_misses];
- int error;
+ return 0;
+}
- error = dpif_recv(ofproto->dpif, upcall);
- if (error) {
- if (error == ENODEV && n_misses == 0) {
- return error;
- }
- break;
- }
+static int
+run(struct ofproto *ofproto_)
+{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+ struct ofport_dpif *ofport;
+ struct ofbundle *bundle;
+ int error;
- if (upcall->type == DPIF_UC_MISS) {
- /* Handle it later. */
- n_misses++;
- } else {
- handle_upcall(ofproto, upcall);
- }
+ if (!clogged) {
+ complete_operations(ofproto);
}
- handle_miss_upcalls(ofproto, misses, n_misses);
-
- if (timer_expired(&ofproto->next_expiration)) {
- int delay = expire(ofproto);
- timer_set_duration(&ofproto->next_expiration, delay);
+ error = run_fast(ofproto_);
+ if (error) {
+ return error;
}
if (ofproto->netflow) {
- netflow_run(ofproto->netflow);
+ if (netflow_run(ofproto->netflow)) {
+ send_netflow_active_timeouts(ofproto);
+ }
}
if (ofproto->sflow) {
dpif_sflow_run(ofproto->sflow);
|| !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(ofproto, facet);
+ facet_revalidate(facet);
+ }
+ }
+ }
+
+ /* Check the consistency of a random facet, to aid debugging. */
+ if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
+ struct facet *facet;
+
+ facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
+ struct facet, hmap_node);
+ if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
+ if (!facet_check_consistency(facet)) {
+ ofproto->need_revalidate = REV_INCONSISTENCY;
}
}
}
+ if (ofproto->governor) {
+ size_t n_subfacets;
+
+ 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;
}
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);
}
HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
bundle_wait(bundle);
}
+ if (ofproto->netflow) {
+ netflow_wait(ofproto->netflow);
+ }
mac_learning_wait(ofproto->ml);
stp_wait(ofproto);
if (ofproto->need_revalidate) {
/* Shouldn't happen, but if it does just go around again. */
VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
poll_immediate_wake();
- } else {
- timer_wait(&ofproto->next_expiration);
}
+ if (ofproto->governor) {
+ governor_wait(ofproto->governor);
+ }
+}
+
+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;
+ 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) {
+ 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);
+ }
+ }
- 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(). */
- facet->installed = false;
- facet->dp_packet_count = 0;
- facet->dp_byte_count = 0;
- facet_remove(ofproto, facet);
+ if (n_batch > 0) {
+ subfacet_destroy_batch(ofproto, batch, n_batch);
}
- dpif_flow_flush(ofproto->dpif);
}
static void
get_features(struct ofproto *ofproto_ OVS_UNUSED,
- bool *arp_match_ip, uint32_t *actions)
+ bool *arp_match_ip, enum ofputil_action_bitmap *actions)
{
*arp_match_ip = true;
- *actions = ((1u << OFPAT_OUTPUT) |
- (1u << OFPAT_SET_VLAN_VID) |
- (1u << OFPAT_SET_VLAN_PCP) |
- (1u << OFPAT_STRIP_VLAN) |
- (1u << OFPAT_SET_DL_SRC) |
- (1u << OFPAT_SET_DL_DST) |
- (1u << OFPAT_SET_NW_SRC) |
- (1u << OFPAT_SET_NW_DST) |
- (1u << OFPAT_SET_NW_TOS) |
- (1u << OFPAT_SET_TP_SRC) |
- (1u << OFPAT_SET_TP_DST) |
- (1u << OFPAT_ENQUEUE));
+ *actions = (OFPUTIL_A_OUTPUT |
+ OFPUTIL_A_SET_VLAN_VID |
+ OFPUTIL_A_SET_VLAN_PCP |
+ OFPUTIL_A_STRIP_VLAN |
+ OFPUTIL_A_SET_DL_SRC |
+ OFPUTIL_A_SET_DL_DST |
+ OFPUTIL_A_SET_NW_SRC |
+ OFPUTIL_A_SET_NW_DST |
+ OFPUTIL_A_SET_NW_TOS |
+ OFPUTIL_A_SET_TP_SRC |
+ OFPUTIL_A_SET_TP_DST |
+ OFPUTIL_A_ENQUEUE);
}
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));
-}
-
-static int
-set_netflow(struct ofproto *ofproto_,
- const struct netflow_options *netflow_options)
-{
- struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+ dpif_get_dp_stats(ofproto->backer->dpif, &s);
- if (netflow_options) {
- if (!ofproto->netflow) {
- ofproto->netflow = netflow_create();
- }
- return netflow_set_options(ofproto->netflow, netflow_options);
- } else {
- netflow_destroy(ofproto->netflow);
- ofproto->netflow = NULL;
- return 0;
- }
+ ots->lookup_count = htonll(s.n_hit + s.n_missed);
+ ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
}
static struct ofport *
{
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();
port->may_enable = true;
port->stp_port = NULL;
port->stp_state = STP_DISABLED;
+ hmap_init(&port->priorities);
+ port->realdev_ofp_port = 0;
+ 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->odp_port,
- netdev_get_name(port->up.netdev));
+ dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
}
return 0;
{
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) {
dpif_sflow_del_port(ofproto->sflow, port->odp_port);
}
+
+ ofport_clear_priorities(port);
+ hmap_destroy(&port->priorities);
}
static void
}
static void
-port_reconfigured(struct ofport *port_, ovs_be32 old_config)
+port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
{
struct ofport_dpif *port = ofport_dpif_cast(port_);
struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
- ovs_be32 changed = old_config ^ port->up.opp.config;
+ enum ofputil_port_config changed = old_config ^ port->up.pp.config;
- if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
- OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
- ofproto->need_revalidate = true;
+ if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
+ OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
+ OFPUTIL_PC_NO_PACKET_IN)) {
+ ofproto->need_revalidate = REV_RECONFIGURE;
- if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) {
+ if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
bundle_update(port->bundle);
}
}
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->odp_port,
- netdev_get_name(ofport->up.netdev));
+ 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;
}
}
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));
}
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)
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;
+}
\f
/* Spanning Tree. */
VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
"with unknown MAC", ofproto->up.name, port_num);
} else {
- send_packet(ofproto_dpif_cast(ofport->up.ofproto),
- ofport->odp_port, pkt);
+ send_packet(ofport, pkt);
}
}
ofpbuf_delete(pkt);
/* Only revalidate flows if the configuration changed. */
if (!s != !ofproto->stp) {
- ofproto->need_revalidate = true;
+ ofproto->need_revalidate = REV_RECONFIGURE;
}
if (s) {
stp_set_max_age(ofproto->stp, s->max_age);
stp_set_forward_delay(ofproto->stp, s->fwd_delay);
} else {
+ struct ofport *ofport;
+
+ HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
+ set_stp_port(ofport, NULL);
+ }
+
stp_destroy(ofproto->stp);
ofproto->stp = NULL;
}
/* Update state. */
if (ofport->stp_state != state) {
- ovs_be32 of_state;
+ enum ofputil_port_state of_state;
bool fwd_change;
VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
if (stp_learn_in_state(ofport->stp_state)
!= stp_learn_in_state(state)) {
/* xxx Learning action flows should also be flushed. */
- mac_learning_flush(ofproto->ml);
+ mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
}
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();
- if (fwd_change) {
+ if (fwd_change && ofport->bundle) {
bundle_update(ofport->bundle);
}
/* Update the STP state bits in the OpenFlow port description. */
- of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK))
- | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN
- : state == STP_LEARNING ? OFPPS_STP_LEARN
- : state == STP_FORWARDING ? OFPPS_STP_FORWARD
- : state == STP_BLOCKING ? OFPPS_STP_BLOCK
- : 0);
+ of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
+ of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
+ : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
+ : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
+ : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
+ : 0);
ofproto_port_set_state(&ofport->up, of_state);
}
}
s->state = stp_port_get_state(sp);
s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
s->role = stp_port_get_role(sp);
+ stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
return 0;
}
update_stp_port_state(ofport);
}
}
+
+ if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
+ mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
+ }
}
}
}
}
\f
+static struct priority_to_dscp *
+get_priority(const struct ofport_dpif *ofport, uint32_t priority)
+{
+ struct priority_to_dscp *pdscp;
+ uint32_t hash;
+
+ hash = hash_int(priority, 0);
+ HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
+ if (pdscp->priority == priority) {
+ return pdscp;
+ }
+ }
+ return NULL;
+}
+
+static void
+ofport_clear_priorities(struct ofport_dpif *ofport)
+{
+ struct priority_to_dscp *pdscp, *next;
+
+ HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
+ hmap_remove(&ofport->priorities, &pdscp->hmap_node);
+ free(pdscp);
+ }
+}
+
+static int
+set_queues(struct ofport *ofport_,
+ const struct ofproto_port_queue *qdscp_list,
+ size_t n_qdscp)
+{
+ struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
+ struct hmap new = HMAP_INITIALIZER(&new);
+ size_t i;
+
+ for (i = 0; i < n_qdscp; i++) {
+ struct priority_to_dscp *pdscp;
+ uint32_t priority;
+ uint8_t dscp;
+
+ dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
+ if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
+ &priority)) {
+ continue;
+ }
+
+ pdscp = get_priority(ofport, priority);
+ if (pdscp) {
+ hmap_remove(&ofport->priorities, &pdscp->hmap_node);
+ } else {
+ pdscp = xmalloc(sizeof *pdscp);
+ pdscp->priority = priority;
+ pdscp->dscp = dscp;
+ ofproto->need_revalidate = REV_RECONFIGURE;
+ }
+
+ if (pdscp->dscp != dscp) {
+ pdscp->dscp = dscp;
+ ofproto->need_revalidate = REV_RECONFIGURE;
+ }
+
+ hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
+ }
+
+ if (!hmap_is_empty(&ofport->priorities)) {
+ ofport_clear_priorities(ofport);
+ ofproto->need_revalidate = REV_RECONFIGURE;
+ }
+
+ hmap_swap(&new, &ofport->priorities);
+ hmap_destroy(&new);
+
+ return 0;
+}
+\f
/* Bundles. */
-/* Expires all MAC learning entries associated with 'port' and forces ofproto
- * to revalidate every flow. */
+/* Expires all MAC learning entries associated with 'bundle' and forces its
+ * ofproto to revalidate every flow.
+ *
+ * Normally MAC learning entries are removed only from the ofproto associated
+ * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
+ * are removed from every ofproto. When patch ports and SLB bonds are in use
+ * and a VM migration happens and the gratuitous ARPs are somehow lost, this
+ * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
+ * with the host from which it migrated. */
static void
-bundle_flush_macs(struct ofbundle *bundle)
+bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
{
struct ofproto_dpif *ofproto = bundle->ofproto;
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) {
+ struct ofproto_dpif *o;
+
+ HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
+ if (o != ofproto) {
+ struct mac_entry *e;
+
+ e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
+ NULL);
+ if (e) {
+ tag_set_add(&o->revalidate_set, e->tag);
+ mac_learning_expire(o->ml, e);
+ }
+ }
+ }
+ }
+
mac_learning_expire(ml, mac);
}
}
bundle->floodable = true;
LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
- if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)
- || !stp_forward_in_state(port->stp_state)) {
+ if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
+ || !stp_forward_in_state(port->stp_state)) {
bundle->floodable = false;
break;
}
{
struct ofbundle *bundle = port->bundle;
- bundle->ofproto->need_revalidate = true;
+ bundle->ofproto->need_revalidate = REV_RECONFIGURE;
list_remove(&port->bundle_node);
port->bundle = NULL;
}
if (port->bundle != bundle) {
- bundle->ofproto->need_revalidate = true;
+ bundle->ofproto->need_revalidate = REV_RECONFIGURE;
if (port->bundle) {
bundle_del_port(port);
}
port->bundle = bundle;
list_push_back(&bundle->ports, &port->bundle_node);
- if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)
- || !stp_forward_in_state(port->stp_state)) {
+ if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
+ || !stp_forward_in_state(port->stp_state)) {
bundle->floodable = false;
}
}
if (lacp) {
- port->bundle->ofproto->need_revalidate = true;
+ port->bundle->ofproto->need_revalidate = REV_RECONFIGURE;
lacp_slave_register(bundle->lacp, port, lacp);
}
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;
}
}
}
bundle_del_port(port);
}
- bundle_flush_macs(bundle);
+ bundle_flush_macs(bundle, true);
hmap_remove(&ofproto->bundles, &bundle->hmap_node);
free(bundle->name);
free(bundle->trunks);
bundle->vlan_mode = PORT_VLAN_TRUNK;
bundle->vlan = -1;
bundle->trunks = NULL;
+ bundle->use_priority_tags = s->use_priority_tags;
bundle->lacp = NULL;
bundle->bond = NULL;
/* 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);
}
/* Set VLAN tagging mode */
- if (s->vlan_mode != bundle->vlan_mode) {
+ if (s->vlan_mode != bundle->vlan_mode
+ || s->use_priority_tags != bundle->use_priority_tags) {
bundle->vlan_mode = s->vlan_mode;
+ bundle->use_priority_tags = s->use_priority_tags;
need_flush = true;
}
break;
case PORT_VLAN_TRUNK:
- trunks = (unsigned long *) s->trunks;
+ trunks = CONST_CAST(unsigned long *, s->trunks);
break;
case PORT_VLAN_NATIVE_UNTAGGED:
bitmap_set1(trunks, vlan);
bitmap_set0(trunks, 0);
} else {
- trunks = (unsigned long *) s->trunks;
+ trunks = CONST_CAST(unsigned long *, s->trunks);
}
break;
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) {
/* If we changed something that would affect MAC learning, un-learn
* everything on this port and force flow revalidation. */
if (need_flush) {
- bundle_flush_macs(bundle);
+ bundle_flush_macs(bundle, false);
}
return 0;
pdu_size);
memcpy(packet_pdu, pdu, pdu_size);
- send_packet(ofproto_dpif_cast(port->up.ofproto), port->odp_port,
- &packet);
+ send_packet(port, &packet);
ofpbuf_uninit(&packet);
} else {
VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
if (e->port.p != bundle) {
struct ofpbuf *learning_packet;
struct ofport_dpif *port;
+ void *port_void;
int ret;
- learning_packet = bond_compose_learning_packet(bundle->bond, e->mac,
- e->vlan,
- (void **)&port);
- ret = send_packet(ofproto_dpif_cast(port->up.ofproto),
- port->odp_port, learning_packet);
+ /* The assignment to "port" is unnecessary but makes "grep"ing for
+ * struct ofport_dpif more effective. */
+ learning_packet = bond_compose_learning_packet(bundle->bond,
+ e->mac, e->vlan,
+ &port_void);
+ port = port_void;
+ ret = send_packet(port, learning_packet);
ofpbuf_delete(learning_packet);
if (ret) {
error = ret;
}
bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
- lacp_negotiated(bundle->lacp));
+ lacp_status(bundle->lacp));
if (bond_should_send_learning_packets(bundle->bond)) {
bundle_send_learning_packets(bundle);
}
return NULL;
}
+/* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
+static void
+mirror_update_dups(struct ofproto_dpif *ofproto)
+{
+ int i;
+
+ for (i = 0; i < MAX_MIRRORS; i++) {
+ struct ofmirror *m = ofproto->mirrors[i];
+
+ if (m) {
+ m->dup_mirrors = MIRROR_MASK_C(1) << i;
+ }
+ }
+
+ for (i = 0; i < MAX_MIRRORS; i++) {
+ struct ofmirror *m1 = ofproto->mirrors[i];
+ int j;
+
+ if (!m1) {
+ continue;
+ }
+
+ for (j = i + 1; j < MAX_MIRRORS; j++) {
+ struct ofmirror *m2 = ofproto->mirrors[j];
+
+ if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
+ m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
+ m2->dup_mirrors |= m1->dup_mirrors;
+ }
+ }
+ }
+}
+
static int
mirror_set(struct ofproto *ofproto_, void *aux,
const struct ofproto_mirror_settings *s)
}
}
- ofproto->need_revalidate = true;
- mac_learning_flush(ofproto->ml);
+ ofproto->need_revalidate = REV_RECONFIGURE;
+ ofproto->has_mirrors = true;
+ mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
+ mirror_update_dups(ofproto);
return 0;
}
struct ofproto_dpif *ofproto;
mirror_mask_t mirror_bit;
struct ofbundle *bundle;
+ int i;
if (!mirror) {
return;
}
ofproto = mirror->ofproto;
- ofproto->need_revalidate = true;
- mac_learning_flush(ofproto->ml);
+ ofproto->need_revalidate = REV_RECONFIGURE;
+ mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
ofproto->mirrors[mirror->idx] = NULL;
free(mirror->name);
free(mirror);
+
+ mirror_update_dups(ofproto);
+
+ ofproto->has_mirrors = false;
+ for (i = 0; i < MAX_MIRRORS; i++) {
+ if (ofproto->mirrors[i]) {
+ ofproto->has_mirrors = true;
+ break;
+ }
+ }
+}
+
+static int
+mirror_get_stats(struct ofproto *ofproto_, void *aux,
+ uint64_t *packets, uint64_t *bytes)
+{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+ struct ofmirror *mirror = mirror_lookup(ofproto, aux);
+
+ if (!mirror) {
+ *packets = *bytes = UINT64_MAX;
+ return 0;
+ }
+
+ *packets = mirror->packet_count;
+ *bytes = mirror->byte_count;
+
+ return 0;
}
static int
{
struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
- ofproto->need_revalidate = true;
- mac_learning_flush(ofproto->ml);
+ mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
}
return 0;
}
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
+set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
+{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+ mac_learning_set_idle_time(ofproto->ml, idle_time);
}
\f
/* 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
port_run(struct ofport_dpif *ofport)
{
+ long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
+ bool carrier_changed = carrier_seq != ofport->carrier_seq;
bool enable = netdev_get_carrier(ofport->up.netdev);
+ 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.opp.hw_addr);
- send_packet(ofproto_dpif_cast(ofport->up.ofproto),
- ofport->odp_port, &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) {
enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
+ if (carrier_changed) {
+ lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
+ }
}
if (ofport->may_enable != enable) {
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;
}
}
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;
}
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) {
return error;
}
+static int
+port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
+{
+ struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
+ int error;
+
+ error = netdev_get_stats(ofport->up.netdev, stats);
+
+ if (!error && ofport->odp_port == OVSP_LOCAL) {
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
+
+ /* ofproto->stats.tx_packets represents packets that we created
+ * internally and sent to some port (e.g. packets sent with
+ * send_packet()). Account for them as if they had come from
+ * OFPP_LOCAL and got forwarded. */
+
+ if (stats->rx_packets != UINT64_MAX) {
+ stats->rx_packets += ofproto->stats.tx_packets;
+ }
+
+ if (stats->rx_bytes != UINT64_MAX) {
+ stats->rx_bytes += ofproto->stats.tx_bytes;
+ }
+
+ /* ofproto->stats.rx_packets represents packets that were received on
+ * some port and we processed internally and dropped (e.g. STP).
+ * 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;
+ }
+
+ if (stats->tx_bytes != UINT64_MAX) {
+ stats->tx_bytes += ofproto->stats.rx_bytes;
+ }
+ }
+
+ return error;
+}
+
+/* Account packets for LOCAL port. */
+static void
+ofproto_update_local_port_stats(const struct ofproto *ofproto_,
+ size_t tx_size, size_t rx_size)
+{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+
+ if (rx_size) {
+ ofproto->stats.rx_packets++;
+ ofproto->stats.rx_bytes += rx_size;
+ }
+ if (tx_size) {
+ ofproto->stats.tx_packets++;
+ ofproto->stats.tx_bytes += tx_size;
+ }
+}
+
struct port_dump_state {
- 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;
}
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
{
struct port_dump_state *state = state_;
- if (!state->done) {
- dpif_port_dump_done(&state->dump);
- }
free(state);
return 0;
}
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
* 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 {
- union dpif_op dpif_op;
- struct facet *facet;
+ struct dpif_op dpif_op;
+ struct subfacet *subfacet; /* Subfacet */
+ void *garbage; /* Pointer to pass to free(), NULL if none. */
+ uint64_t stub[1024 / 8]; /* Temporary buffer. */
};
/* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
* OpenFlow controller as necessary according to their individual
- * configurations.
- *
- * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
- * ownership is transferred to this function. */
+ * configurations. */
static void
-send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
- const struct flow *flow, bool clone)
+send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
+ const struct flow *flow)
{
struct ofputil_packet_in pin;
- pin.packet = packet;
- pin.in_port = flow->in_port;
+ pin.packet = packet->data;
+ pin.packet_len = packet->size;
pin.reason = OFPR_NO_MATCH;
- pin.buffer_id = 0; /* not yet known */
- pin.send_len = 0; /* not used for flow table misses */
- connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
- clone ? NULL : packet);
-}
+ pin.controller_id = 0;
-/* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_ACTION to each
- * OpenFlow controller as necessary according to their individual
- * configurations.
- *
- * 'send_len' should be the number of bytes of 'packet' to send to the
- * controller, as specified in the action that caused the packet to be sent.
- *
- * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
- * Otherwise, ownership is transferred to this function. */
-static void
-send_packet_in_action(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
- uint64_t userdata, const struct flow *flow, bool clone)
-{
- struct ofputil_packet_in pin;
- struct user_action_cookie cookie;
+ pin.table_id = 0;
+ pin.cookie = 0;
+
+ pin.send_len = 0; /* not used for flow table misses */
- memcpy(&cookie, &userdata, sizeof(cookie));
+ flow_get_metadata(flow, &pin.fmd);
- pin.packet = packet;
- pin.in_port = flow->in_port;
- pin.reason = OFPR_ACTION;
- pin.buffer_id = 0; /* not yet known */
- pin.send_len = cookie.data;
- connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
- clone ? NULL : packet);
+ 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,
- const struct nlattr *key, size_t key_len)
+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) {
}
}
- miss = xmalloc(sizeof *miss);
- hmap_insert(todo, &miss->hmap_node, hash);
- miss->flow = *flow;
- miss->key = key;
- miss->key_len = key_len;
- 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)
+init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
+ struct flow_miss_op *op)
{
- const struct flow *flow = &miss->flow;
- struct ofpbuf *packet, *next_packet;
- struct facet *facet;
+ 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 OFPPC_NO_PACKET_IN asserted. */
- struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
- if (port) {
- if (port->up.opp.config & htonl(OFPPC_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_SAFE (packet, next_packet, list_node,
- &miss->packets) {
- list_remove(&packet->list_node);
- send_packet_in_miss(ofproto, packet, flow, false);
- }
+ ofproto->n_matches++;
- return;
+ 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;
+
+ 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);
}
- LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
- list_remove(&packet->list_node);
- ofproto->n_matches++;
+ return governor_should_install_flow(ofproto->governor, hash,
+ list_size(&miss->packets));
+}
- 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, true);
+/* 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 (packet, list_node, &miss->packets) {
+ struct flow_miss_op *op = &ops[*n_ops];
+ struct dpif_flow_stats stats;
+ struct ofpbuf odp_actions;
+
+ COVERAGE_INC(facet_suppress);
+
+ ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
+
+ 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);
- if (!facet->may_install) {
- facet_make_actions(ofproto, facet, packet);
+ LIST_FOR_EACH (packet, list_node, &miss->packets) {
+ struct flow_miss_op *op = &ops[*n_ops];
+ struct dpif_flow_stats stats;
+ struct ofpbuf odp_actions;
+
+ handle_flow_miss_common(facet->rule, packet, &miss->flow);
+
+ ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
+ if (!subfacet->actions || subfacet->slow) {
+ subfacet_make_actions(subfacet, packet, &odp_actions);
}
- if (!execute_controller_action(ofproto, &facet->flow,
- facet->actions, facet->actions_len,
- packet)) {
- struct flow_miss_op *op = &ops[(*n_ops)++];
- struct dpif_execute *execute = &op->dpif_op.execute;
- op->facet = facet;
- execute->type = DPIF_OP_EXECUTE;
- execute->key = miss->key;
- execute->key_len = miss->key_len;
- execute->actions
- = (facet->may_install
- ? facet->actions
- : xmemdup(facet->actions, facet->actions_len));
- execute->actions_len = facet->actions_len;
- execute->packet = packet;
+ dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
+ subfacet_update_stats(subfacet, &stats);
+
+ 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);
+ }
+
+ (*n_ops)++;
+ } else {
+ ofpbuf_uninit(&odp_actions);
}
}
- if (facet->may_install) {
+ 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.flow_put;
+ struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
- op->facet = facet;
- put->type = DPIF_OP_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 = facet->actions;
- put->actions_len = facet->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;
}
}
+/* 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.) */
+static enum odp_key_fitness
+ofproto_dpif_vsp_adjust(const struct ofproto_dpif *ofproto,
+ enum odp_key_fitness fitness,
+ struct flow *flow, ovs_be16 *initial_tci,
+ struct ofpbuf *packet)
+{
+ if (fitness == ODP_FIT_ERROR) {
+ return fitness;
+ }
+ *initial_tci = flow->vlan_tci;
+
+ if (vsp_adjust_flow(ofproto, flow)) {
+ if (packet) {
+ /* Make the packet resemble the flow, so that it gets sent to an
+ * OpenFlow controller properly, so that it looks correct for
+ * sFlow, and so that flow_extract() will get the correct vlan_tci
+ * if it is called on 'packet'.
+ *
+ * The allocated space inside 'packet' probably also contains
+ * 'key', that is, both 'packet' and 'key' are probably part of a
+ * struct dpif_upcall (see the large comment on that structure
+ * definition), so pushing data on 'packet' is in general not a
+ * good idea since it could overwrite 'key' or free it as a side
+ * effect. However, it's OK in this special case because we know
+ * that 'packet' is inside a Netlink attribute: pushing 4 bytes
+ * will just overwrite the 4-byte "struct nlattr", which is fine
+ * since we don't need that header anymore. */
+ eth_push_vlan(packet, flow->vlan_tci);
+ }
+
+ /* Let the caller know that we can't reproduce 'key' from 'flow'. */
+ if (fitness == ODP_FIT_PERFECT) {
+ fitness = ODP_FIT_TOO_MUCH;
+ }
+ }
+
+ return fitness;
+}
+
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];
- union dpif_op *dpif_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;
* 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;
+ struct flow_miss *miss = &misses[n_misses];
+ struct flow_miss *existing_miss;
+ enum odp_key_fitness fitness;
+ struct ofproto_dpif *ofproto;
+ struct ofport_dpif *port;
+ uint32_t odp_in_port;
struct flow flow;
-
- /* Obtain in_port and tun_id, at least, then set 'flow''s header
- * pointers. */
- odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
- flow_extract(upcall->packet, flow.priority, flow.tun_id,
- flow.in_port, &flow);
-
- /* Handle 802.1ag, LACP, and STP specially. */
- if (process_special(ofproto, &flow, upcall->packet)) {
- ofpbuf_delete(upcall->packet);
- ofproto->n_matches++;
+ uint32_t hash;
+
+ 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;
+ }
+ ofproto = ofproto_dpif_cast(port->up.ofproto);
+ odp_in_port = flow.in_port;
+ flow.in_port = port->up.ofp_port;
+
+ /* Obtain metadata and check userspace/kernel agreement on flow match,
+ * then set 'flow''s header pointers. */
+ miss->key_fitness = ofproto_dpif_vsp_adjust(ofproto, fitness,
+ &flow, &miss->initial_tci, upcall->packet);
+ if (miss->key_fitness == ODP_FIT_ERROR) {
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, upcall->key, upcall->key_len);
+ 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);
}
/* Process each element in the to-do list, constructing the set of
* operations to batch. */
n_ops = 0;
- HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
- handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
- ofpbuf_list_delete(&miss->packets);
- hmap_remove(&todo, &miss->hmap_node);
- free(miss);
+ HMAP_FOR_EACH (miss, hmap_node, &todo) {
+ handle_flow_miss(miss, flow_miss_ops, &n_ops);
}
assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
- hmap_destroy(&todo);
/* Execute batch. */
for (i = 0; i < n_ops; i++) {
dpif_ops[i] = &flow_miss_ops[i].dpif_op;
}
- dpif_operate(ofproto->dpif, dpif_ops, n_ops);
+ dpif_operate(backer->dpif, dpif_ops, n_ops);
/* Free memory and update facets. */
for (i = 0; i < n_ops; i++) {
struct flow_miss_op *op = &flow_miss_ops[i];
- struct dpif_execute *execute;
- struct dpif_flow_put *put;
switch (op->dpif_op.type) {
case DPIF_OP_EXECUTE:
- execute = &op->dpif_op.execute;
- if (op->facet->actions != execute->actions) {
- free((struct nlattr *) execute->actions);
- }
- ofpbuf_delete((struct ofpbuf *) execute->packet);
break;
case DPIF_OP_FLOW_PUT:
- put = &op->dpif_op.flow_put;
- if (!put->error) {
- op->facet->installed = true;
+ if (!op->dpif_op.error) {
+ op->subfacet->path = subfacet_want_path(op->subfacet->slow);
}
break;
+
+ case DPIF_OP_FLOW_DEL:
+ NOT_REACHED();
}
+
+ free(op->garbage);
}
+ hmap_destroy(&todo);
}
-static void
-handle_userspace_upcall(struct ofproto_dpif *ofproto,
- struct dpif_upcall *upcall)
+static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
+classify_upcall(const struct dpif_upcall *upcall)
{
- struct flow flow;
- struct user_action_cookie cookie;
+ 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;
- if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
- if (ofproto->sflow) {
- odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
- dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie);
- }
- ofpbuf_delete(upcall->packet);
+ case USER_ACTION_COOKIE_SLOW_PATH:
+ return MISS_UPCALL;
- } else if (cookie.type == USER_ACTION_COOKIE_CONTROLLER) {
- COVERAGE_INC(ofproto_dpif_ctlr_action);
- odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
- send_packet_in_action(ofproto, upcall->packet, upcall->userdata,
- &flow, false);
- } else {
+ case USER_ACTION_COOKIE_UNSPEC:
+ default:
VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
+ return BAD_UPCALL;
}
}
static void
-handle_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
+handle_sflow_upcall(struct dpif_backer *backer,
+ const struct dpif_upcall *upcall)
{
- switch (upcall->type) {
- case DPIF_UC_ACTION:
- handle_userspace_upcall(ofproto, upcall);
- break;
+ 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;
- case DPIF_UC_MISS:
- /* The caller handles these. */
- NOT_REACHED();
+ fitness = odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
- case DPIF_N_UC_TYPES:
- default:
- VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
- break;
+ port = odp_port_to_ofport(backer, flow.in_port);
+ if (!port) {
+ return;
+ }
+
+ 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;
+ }
+
+ memcpy(&cookie, &upcall->userdata, sizeof(cookie));
+ dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
+ odp_in_port, &cookie);
+}
+
+static int
+handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
+{
+ struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
+ struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
+ uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
+ int n_processed;
+ int n_misses;
+ int i;
+
+ assert(max_batch <= FLOW_MISS_MAX_BATCH);
+
+ n_misses = 0;
+ for (n_processed = 0; n_processed < max_batch; n_processed++) {
+ struct dpif_upcall *upcall = &misses[n_misses];
+ struct ofpbuf *buf = &miss_bufs[n_misses];
+ int error;
+
+ ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
+ sizeof miss_buf_stubs[n_misses]);
+ error = dpif_recv(backer->dpif, upcall, buf);
+ if (error) {
+ ofpbuf_uninit(buf);
+ break;
+ }
+
+ switch (classify_upcall(upcall)) {
+ case MISS_UPCALL:
+ /* Handle it later. */
+ n_misses++;
+ break;
+
+ case SFLOW_UPCALL:
+ handle_sflow_upcall(backer, upcall);
+ ofpbuf_uninit(buf);
+ break;
+
+ case BAD_UPCALL:
+ ofpbuf_uninit(buf);
+ break;
+ }
}
+
+ /* 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 n_processed;
}
\f
/* Flow expiration. */
-static int facet_max_idle(const struct ofproto_dpif *);
-static void update_stats(struct ofproto_dpif *);
+static int subfacet_max_idle(const struct ofproto_dpif *);
+static void update_stats(struct dpif_backer *);
static void rule_expire(struct rule_dpif *);
-static void expire_facets(struct ofproto_dpif *, int dp_max_idle);
+static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
/* This function is called periodically by run(). Its job is to collect
* updates for the flows that have been installed into the datapath, most
*
* 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 classifier *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 facets that have been idle too long. */
- dp_max_idle = facet_max_idle(ofproto);
- expire_facets(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, 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.
* 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 facet *facet;
struct flow flow;
+ struct subfacet *subfacet;
+ enum odp_key_fitness fitness;
+ struct ofproto_dpif *ofproto;
+ struct ofport_dpif *port;
+ uint32_t key_hash;
- if (odp_flow_key_to_flow(key, key_len, &flow)) {
- struct ds s;
-
- ds_init(&s);
- odp_flow_key_format(key, key_len, &s);
- VLOG_WARN_RL(&rl, "failed to convert datapath flow key to flow: %s",
- ds_cstr(&s));
- ds_destroy(&s);
-
+ fitness = odp_flow_key_to_flow(key, key_len, &flow);
+ if (fitness == ODP_FIT_ERROR) {
continue;
}
- facet = facet_find(p, &flow);
- if (facet && facet->installed) {
+ 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;
+ }
- if (stats->n_packets >= facet->dp_packet_count) {
- uint64_t extra = stats->n_packets - facet->dp_packet_count;
- facet->packet_count += extra;
- } else {
- VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
- }
+ ofproto = ofproto_dpif_cast(port->up.ofproto);
+ flow.in_port = port->up.ofp_port;
+ key_hash = odp_flow_key_hash(key, key_len);
- if (stats->n_bytes >= facet->dp_byte_count) {
- facet->byte_count += stats->n_bytes - facet->dp_byte_count;
- } else {
- VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
- }
+ 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;
- facet->dp_packet_count = stats->n_packets;
- facet->dp_byte_count = stats->n_bytes;
+ case SF_SLOW_PATH:
+ /* Stats are updated per-packet. */
+ break;
- facet_update_time(p, facet, stats->used);
- facet_account(p, facet);
- facet_push_stats(facet);
- } else {
- /* There's a flow in the datapath that we know nothing about.
- * Delete it. */
- COVERAGE_INC(facet_unexpected);
- 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);
}
/* Calculates and returns the number of milliseconds of idle time after which
- * facets should expire from the datapath and we should fold their statistics
- * into their parent rules in userspace. */
+ * subfacets should expire from the datapath. When a subfacet expires, we fold
+ * its statistics into its facet, and when a facet's last subfacet expires, we
+ * fold its statistic into its rule. */
static int
-facet_max_idle(const struct ofproto_dpif *ofproto)
+subfacet_max_idle(const struct ofproto_dpif *ofproto)
{
/*
* Idle time histogram.
*
- * Most of the time a switch has a relatively small number of facets. When
- * this is the case we might as well keep statistics for all of them in
- * userspace and to cache them in the kernel datapath for performance as
+ * Most of the time a switch has a relatively small number of subfacets.
+ * When this is the case we might as well keep statistics for all of them
+ * in userspace and to cache them in the kernel datapath for performance as
* well.
*
- * As the number of facets increases, the memory required to maintain
+ * As the number of subfacets increases, the memory required to maintain
* statistics about them in userspace and in the kernel becomes
- * significant. However, with a large number of facets it is likely that
- * only a few of them are "heavy hitters" that consume a large amount of
- * bandwidth. At this point, only heavy hitters are worth caching in the
- * kernel and maintaining in userspaces; other facets we can discard.
+ * significant. However, with a large number of subfacets it is likely
+ * that only a few of them are "heavy hitters" that consume a large amount
+ * of bandwidth. At this point, only heavy hitters are worth caching in
+ * the kernel and maintaining in userspaces; other subfacets we can
+ * discard.
*
* The technique used to compute the idle time is to build a histogram with
- * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
+ * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
* that is installed in the kernel gets dropped in the appropriate bucket.
* After the histogram has been built, we compute the cutoff so that only
- * the most-recently-used 1% of facets (but at least
+ * the most-recently-used 1% of subfacets (but at least
* ofproto->up.flow_eviction_threshold flows) are kept cached. At least
- * the most-recently-used bucket of facets is kept, so actually an
- * arbitrary number of facets can be kept in any given expiration run
+ * the most-recently-used bucket of subfacets is kept, so actually an
+ * arbitrary number of subfacets can be kept in any given expiration run
* (though the next run will delete most of those unless they receive
* additional data).
*
- * This requires a second pass through the facets, in addition to the pass
- * made by update_stats(), because the former function never looks
- * at uninstallable facets.
+ * This requires a second pass through the subfacets, in addition to the
+ * pass made by update_stats(), because the former function never looks at
+ * uninstallable subfacets.
*/
enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
int buckets[N_BUCKETS] = { 0 };
int total, subtotal, bucket;
- struct facet *facet;
+ struct subfacet *subfacet;
long long int now;
int i;
- total = hmap_count(&ofproto->facets);
+ total = hmap_count(&ofproto->subfacets);
if (total <= ofproto->up.flow_eviction_threshold) {
return N_BUCKETS * BUCKET_WIDTH;
}
/* Build histogram. */
now = time_msec();
- HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
- long long int idle = now - facet->used;
+ HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
+ long long int idle = now - subfacet->used;
int bucket = (idle <= 0 ? 0
: idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
: (unsigned int) idle / BUCKET_WIDTH);
}
static void
-facet_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
-{
- if (ofproto->netflow && !facet_is_controller_flow(facet) &&
- netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
- struct ofexpired expired;
-
- if (facet->installed) {
- struct dpif_flow_stats stats;
-
- facet_put__(ofproto, facet, facet->actions, facet->actions_len,
- &stats);
- facet_update_stats(ofproto, facet, &stats);
- }
-
- expired.flow = facet->flow;
- expired.packet_count = facet->packet_count;
- expired.byte_count = facet->byte_count;
- expired.used = facet->used;
- netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
+expire_subfacets(struct ofproto_dpif *ofproto, int 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) {
+ 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);
+ }
+ }
}
-}
-
-static void
-expire_facets(struct ofproto_dpif *ofproto, int dp_max_idle)
-{
- long long int cutoff = time_msec() - dp_max_idle;
- struct facet *facet, *next_facet;
- HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
- facet_active_timeout(ofproto, facet);
- if (facet->used < cutoff) {
- facet_remove(ofproto, facet);
- }
+ if (n_batch > 0) {
+ subfacet_destroy_batch(ofproto, batch, n_batch);
}
}
static void
rule_expire(struct rule_dpif *rule)
{
- struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
struct facet *facet, *next_facet;
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
&& now > rule->up.modified + rule->up.hard_timeout * 1000) {
reason = OFPRR_HARD_TIMEOUT;
- } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
- && now > rule->used + rule->up.idle_timeout * 1000) {
+ } else if (rule->up.idle_timeout
+ && now > rule->up.used + rule->up.idle_timeout * 1000) {
reason = OFPRR_IDLE_TIMEOUT;
} else {
return;
/* Update stats. (This is a no-op if the rule expired due to an idle
* timeout, because that only happens when the rule has no facets left.) */
LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
- facet_remove(ofproto, facet);
+ facet_remove(facet);
}
/* Get rid of the rule. */
* 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
* the ofproto's classifier table.
*
- * The facet will initially have no ODP actions. The caller should fix that
- * by calling facet_make_actions(). */
+ * '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;
+ list_init(&facet->subfacets);
netflow_flow_init(&facet->nf_flow);
netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
static void
facet_free(struct facet *facet)
{
- free(facet->actions);
free(facet);
}
-static bool
-execute_controller_action(struct ofproto_dpif *ofproto,
- const struct flow *flow,
- const struct nlattr *odp_actions, size_t actions_len,
- struct ofpbuf *packet)
-{
- if (actions_len
- && odp_actions->nla_type == OVS_ACTION_ATTR_USERSPACE
- && NLA_ALIGN(odp_actions->nla_len) == actions_len) {
- /* As an optimization, avoid a round-trip from userspace to kernel to
- * userspace. This also avoids possibly filling up kernel packet
- * buffers along the way.
- *
- * This optimization will not accidentally catch sFlow
- * OVS_ACTION_ATTR_USERSPACE actions, since those are encapsulated
- * inside OVS_ACTION_ATTR_SAMPLE. */
- const struct nlattr *nla;
-
- nla = nl_attr_find_nested(odp_actions, OVS_USERSPACE_ATTR_USERDATA);
- send_packet_in_action(ofproto, packet, nl_attr_get_u64(nla), flow,
- false);
- return true;
- } else {
- return false;
- }
-}
-
/* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
* 'packet', which arrived on 'in_port'.
*
struct ofpbuf key;
int error;
- if (execute_controller_action(ofproto, flow, odp_actions, actions_len,
- packet)) {
- return true;
- }
-
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);
return !error;
}
-/* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
- * statistics appropriately. 'packet' must have at least sizeof(struct
- * ofp_packet_in) bytes of headroom.
- *
- * For correct results, 'packet' must actually be in 'facet''s flow; that is,
- * applying flow_extract() to 'packet' would yield the same flow as
- * 'facet->flow'.
- *
- * 'facet' must have accurately composed datapath actions; that is, it must
- * not be in need of revalidation.
- *
- * Takes ownership of 'packet'. */
-static void
-facet_execute(struct ofproto_dpif *ofproto, struct facet *facet,
- struct ofpbuf *packet)
-{
- struct dpif_flow_stats stats;
-
- assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
-
- dpif_flow_stats_extract(&facet->flow, packet, &stats);
- stats.used = time_msec();
- if (execute_odp_actions(ofproto, &facet->flow,
- facet->actions, facet->actions_len, packet)) {
- facet_update_stats(ofproto, facet, &stats);
- }
-}
-
/* Remove 'facet' from 'ofproto' and free up the associated memory:
*
* - If 'facet' was installed in the datapath, uninstalls it and updates its
- * rule's statistics, via facet_uninstall().
+ * rule's statistics, via subfacet_uninstall().
*
* - Removes 'facet' from its rule and from ofproto->facets.
*/
static void
-facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
-{
- facet_uninstall(ofproto, facet);
- facet_flush_stats(ofproto, facet);
- hmap_remove(&ofproto->facets, &facet->hmap_node);
- list_remove(&facet->list_node);
- facet_free(facet);
-}
-
-/* Composes the datapath actions for 'facet' based on its rule's actions. */
-static void
-facet_make_actions(struct ofproto_dpif *p, struct facet *facet,
- const struct ofpbuf *packet)
+facet_remove(struct facet *facet)
{
- const struct rule_dpif *rule = facet->rule;
- struct ofpbuf *odp_actions;
- struct action_xlate_ctx ctx;
-
- action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
- odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_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->nf_flow.output_iface = ctx.nf_output_iface;
-
- if (facet->actions_len != odp_actions->size
- || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
- free(facet->actions);
- facet->actions_len = odp_actions->size;
- facet->actions = xmemdup(odp_actions->data, odp_actions->size);
- }
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
+ struct subfacet *subfacet, *next_subfacet;
- ofpbuf_delete(odp_actions);
-}
-
-/* Updates 'facet''s flow in the datapath setting its actions to 'actions_len'
- * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
- * in the datapath will be zeroed and 'stats' will be updated with traffic new
- * since 'facet' was last updated.
- *
- * Returns 0 if successful, otherwise a positive errno value.*/
-static int
-facet_put__(struct ofproto_dpif *ofproto, struct facet *facet,
- const struct nlattr *actions, size_t actions_len,
- struct dpif_flow_stats *stats)
-{
- struct odputil_keybuf keybuf;
- enum dpif_flow_put_flags flags;
- struct ofpbuf key;
- int ret;
+ assert(!list_is_empty(&facet->subfacets));
- flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
- if (stats) {
- flags |= DPIF_FP_ZERO_STATS;
+ /* First uninstall all of the subfacets to get final statistics. */
+ LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
+ subfacet_uninstall(subfacet);
}
- ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
- odp_flow_key_from_flow(&key, &facet->flow);
-
- ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
- actions, actions_len, stats);
+ /* Flush the final stats to the rule.
+ *
+ * This might require us to have at least one subfacet around so that we
+ * can use its actions for accounting in facet_account(), which is why we
+ * have uninstalled but not yet destroyed the subfacets. */
+ facet_flush_stats(facet);
- if (stats) {
- facet_reset_dp_stats(facet, stats);
+ /* Now we're really all done so destroy everything. */
+ LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
+ &facet->subfacets) {
+ subfacet_destroy__(subfacet);
}
-
- return ret;
+ hmap_remove(&ofproto->facets, &facet->hmap_node);
+ list_remove(&facet->list_node);
+ facet_free(facet);
}
-/* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
- * 'zero_stats' is true, clears any existing statistics from the datapath for
- * '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_install(struct ofproto_dpif *p, struct facet *facet, bool zero_stats)
+facet_learn(struct facet *facet)
{
- struct dpif_flow_stats stats;
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
+ struct action_xlate_ctx ctx;
- if (facet->may_install
- && !facet_put__(p, facet, facet->actions, facet->actions_len,
- zero_stats ? &stats : NULL)) {
- facet->installed = true;
+ if (!facet->has_learn
+ && !facet->has_normal
+ && (!facet->has_fin_timeout
+ || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
+ return;
}
+
+ action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
+ facet->flow.vlan_tci,
+ facet->rule, facet->tcp_flags, NULL);
+ ctx.may_learn = true;
+ xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
+ facet->rule->up.ofpacts_len);
}
static void
-facet_account(struct ofproto_dpif *ofproto, struct facet *facet)
+facet_account(struct facet *facet)
{
- uint64_t n_bytes;
+ 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;
-
- if (facet->byte_count <= facet->accounted_bytes) {
- 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) {
- struct action_xlate_ctx ctx;
-
- action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
- ctx.may_learn = true;
- ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
- facet->rule->up.n_actions));
- }
+ 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
* is going to be sent to ensure that it matches the one passed to
* bond_choose_output_slave(). (Otherwise, we will account to the wrong
- * hash bucket.) */
+ * hash bucket.)
+ *
+ * We use the actions from an arbitrary subfacet because they should all
+ * be equally valid for our purpose. */
+ subfacet = CONTAINER_OF(list_front(&facet->subfacets),
+ struct subfacet, list_node);
vlan_tci = facet->flow.vlan_tci;
- NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->actions, facet->actions_len) {
+ NL_ATTR_FOR_EACH_UNSAFE (a, left,
+ subfacet->actions, subfacet->actions_len) {
+ const struct ovs_action_push_vlan *vlan;
struct ofport_dpif *port;
switch (nl_attr_type(a)) {
- const struct nlattr *nested;
case OVS_ACTION_ATTR_OUTPUT:
port = get_odp_port(ofproto, nl_attr_get_u32(a));
if (port && port->bundle && port->bundle->bond) {
}
break;
- case OVS_ACTION_ATTR_POP:
- if (nl_attr_get_u16(a) == OVS_KEY_ATTR_8021Q) {
- vlan_tci = htons(0);
- }
+ case OVS_ACTION_ATTR_POP_VLAN:
+ vlan_tci = htons(0);
break;
- case OVS_ACTION_ATTR_PUSH:
- nested = nl_attr_get(a);
- if (nl_attr_type(nested) == OVS_KEY_ATTR_8021Q) {
- const struct ovs_key_8021q *q_key;
-
- q_key = nl_attr_get_unspec(nested, sizeof(*q_key));
- vlan_tci = q_key->q_tci;
- }
+ case OVS_ACTION_ATTR_PUSH_VLAN:
+ vlan = nl_attr_get(a);
+ vlan_tci = vlan->vlan_tci;
break;
}
}
}
-/* If 'rule' is installed in the datapath, uninstalls it. */
-static void
-facet_uninstall(struct ofproto_dpif *p, struct facet *facet)
-{
- if (facet->installed) {
- struct odputil_keybuf keybuf;
- struct dpif_flow_stats stats;
- struct ofpbuf key;
- int error;
-
- ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
- odp_flow_key_from_flow(&key, &facet->flow);
-
- error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
- facet_reset_dp_stats(facet, &stats);
- if (!error) {
- facet_update_stats(p, facet, &stats);
- }
- facet->installed = false;
- } else {
- assert(facet->dp_packet_count == 0);
- assert(facet->dp_byte_count == 0);
- }
-}
-
/* Returns true if the only action for 'facet' is to send to the controller.
* (We don't report NetFlow expiration messages for such facets because they
* are just part of the control logic for the network, not real traffic). */
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;
-/* Resets 'facet''s datapath statistics counters. This should be called when
- * 'facet''s statistics are cleared in the datapath. If 'stats' is non-null,
- * it should contain the statistics returned by dpif when 'facet' was reset in
- * the datapath. 'stats' will be modified to only included statistics new
- * since 'facet' was last updated. */
-static void
-facet_reset_dp_stats(struct facet *facet, struct dpif_flow_stats *stats)
-{
- if (stats && facet->dp_packet_count <= stats->n_packets
- && facet->dp_byte_count <= stats->n_bytes) {
- stats->n_packets -= facet->dp_packet_count;
- stats->n_bytes -= facet->dp_byte_count;
+ if (ofpacts_len > 0 &&
+ ofpacts->type == OFPACT_CONTROLLER &&
+ ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
+ return true;
+ }
}
-
- facet->dp_packet_count = 0;
- facet->dp_byte_count = 0;
+ return false;
}
/* Folds all of 'facet''s statistics into its rule. Also updates the
* 'facet''s statistics in the datapath should have been zeroed and folded into
* its packet and byte counts before this function is called. */
static void
-facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
+facet_flush_stats(struct facet *facet)
{
- assert(!facet->dp_byte_count);
- assert(!facet->dp_packet_count);
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
+ struct subfacet *subfacet;
+
+ LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
+ assert(!subfacet->dp_byte_count);
+ assert(!subfacet->dp_packet_count);
+ }
facet_push_stats(facet);
- facet_account(ofproto, facet);
+ 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;
facet_reset_counters(facet);
netflow_flow_clear(&facet->nf_flow);
+ facet->tcp_flags = 0;
}
/* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
* Returns it if found, otherwise a null pointer.
*
+ * 'hash' must be the return value of flow_hash(flow, 0).
+ *
* 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;
}
/* 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(ofproto, facet)) {
- COVERAGE_INC(facet_invalidated);
- return NULL;
+ || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) {
+ facet_revalidate(facet);
}
return facet;
}
-/* Re-searches 'ofproto''s classifier for a rule matching '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 "<error>";
+ }
+}
+
+/* Returns the path in which a subfacet should be installed if its 'slow'
+ * member has the specified value. */
+static enum subfacet_path
+subfacet_want_path(enum slow_path_reason slow)
+{
+ return slow ? SF_SLOW_PATH : SF_FAST_PATH;
+}
+
+/* Returns true if 'subfacet' needs to have its datapath flow updated,
+ * supposing that its actions have been recalculated as 'want_actions' and that
+ * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
+static bool
+subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
+ const struct ofpbuf *want_actions)
+{
+ enum subfacet_path want_path = subfacet_want_path(slow);
+ return (want_path != subfacet->path
+ || (want_path == SF_FAST_PATH
+ && (subfacet->actions_len != want_actions->size
+ || memcmp(subfacet->actions, want_actions->data,
+ subfacet->actions_len))));
+}
+
+static bool
+facet_check_consistency(struct facet *facet)
+{
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
+
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
+
+ uint64_t odp_actions_stub[1024 / 8];
+ struct ofpbuf odp_actions;
+
+ struct rule_dpif *rule;
+ struct subfacet *subfacet;
+ bool may_log = false;
+ bool ok;
+
+ /* Check the rule for consistency. */
+ rule = rule_dpif_lookup(ofproto, &facet->flow);
+ ok = rule == facet->rule;
+ if (!ok) {
+ may_log = !VLOG_DROP_WARN(&rl);
+ if (may_log) {
+ struct ds s;
+
+ ds_init(&s);
+ flow_format(&s, &facet->flow);
+ ds_put_format(&s, ": facet associated with wrong rule (was "
+ "table=%"PRIu8",", facet->rule->up.table_id);
+ cls_rule_format(&facet->rule->up.cr, &s);
+ ds_put_format(&s, ") (should have been table=%"PRIu8",",
+ rule->up.table_id);
+ cls_rule_format(&rule->up.cr, &s);
+ ds_put_char(&s, ')');
+
+ VLOG_WARN("%s", ds_cstr(&s));
+ ds_destroy(&s);
+ }
+ }
+
+ /* Check the datapath actions for consistency. */
+ ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
+ LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
+ enum subfacet_path want_path;
+ struct odputil_keybuf keybuf;
+ struct action_xlate_ctx ctx;
+ struct ofpbuf key;
+ struct ds s;
+
+ action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
+ subfacet->initial_tci, rule, 0, NULL);
+ xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
+ &odp_actions);
+
+ if (subfacet->path == SF_NOT_INSTALLED) {
+ /* This only happens if the datapath reported an error when we
+ * tried to install the flow. Don't flag another error here. */
+ continue;
+ }
+
+ want_path = subfacet_want_path(subfacet->slow);
+ if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
+ /* The actions for slow-path flows may legitimately vary from one
+ * packet to the next. We're done. */
+ continue;
+ }
+
+ if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
+ continue;
+ }
+
+ /* Inconsistency! */
+ if (ok) {
+ may_log = !VLOG_DROP_WARN(&rl);
+ ok = false;
+ }
+ if (!may_log) {
+ /* Rate-limited, skip reporting. */
+ continue;
+ }
+
+ 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;
+}
+
+/* Re-searches the classifier for 'facet':
*
* - If the rule found is different from 'facet''s current rule, moves
* '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
-facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
+ * 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);
+ struct actions {
+ struct nlattr *odp_actions;
+ size_t actions_len;
+ };
+ struct actions *new_actions;
+
struct action_xlate_ctx ctx;
- struct ofpbuf *odp_actions;
+ uint64_t odp_actions_stub[1024 / 8];
+ struct ofpbuf odp_actions;
+
struct rule_dpif *new_rule;
- bool actions_changed;
+ struct subfacet *subfacet;
+ 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(ofproto, facet);
- return false;
- }
+ new_rule = rule_dpif_lookup(ofproto, &facet->flow);
/* Calculate new datapath actions.
*
* We do not modify any 'facet' state yet, because we might need to, e.g.,
* emit a NetFlow expiration and, if so, we need to have the old state
* around to properly compose it. */
- action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
- odp_actions = xlate_actions(&ctx,
- new_rule->up.actions, new_rule->up.n_actions);
- actions_changed = (facet->actions_len != odp_actions->size
- || memcmp(facet->actions, odp_actions->data,
- facet->actions_len));
/* If the datapath actions changed or the installability changed,
* then we need to talk to the datapath. */
- if (actions_changed || ctx.may_set_up_flow != facet->installed) {
- if (ctx.may_set_up_flow) {
+ i = 0;
+ new_actions = NULL;
+ memset(&ctx, 0, sizeof ctx);
+ ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
+ LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
+ enum slow_path_reason slow;
+
+ action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
+ subfacet->initial_tci, new_rule, 0, NULL);
+ xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
+ &odp_actions);
+
+ slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
+ if (subfacet_should_install(subfacet, slow, &odp_actions)) {
struct dpif_flow_stats stats;
- facet_put__(ofproto, facet,
- odp_actions->data, odp_actions->size, &stats);
- facet_update_stats(ofproto, facet, &stats);
- } else {
- facet_uninstall(ofproto, facet);
+ 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;
}
- /* The datapath flow is gone or has zeroed stats, so push stats out of
- * 'facet' into 'rule'. */
- facet_flush_stats(ofproto, facet);
+ i++;
+ }
+ ofpbuf_uninit(&odp_actions);
+
+ if (new_actions) {
+ facet_flush_stats(facet);
}
/* Update 'facet' now that we've taken care of all the old state. */
facet->tags = ctx.tags;
facet->nf_flow.output_iface = ctx.nf_output_iface;
- facet->may_install = ctx.may_set_up_flow;
facet->has_learn = ctx.has_learn;
facet->has_normal = ctx.has_normal;
- if (actions_changed) {
- free(facet->actions);
- facet->actions_len = odp_actions->size;
- facet->actions = xmemdup(odp_actions->data, odp_actions->size);
+ facet->has_fin_timeout = ctx.has_fin_timeout;
+ facet->mirrors = ctx.mirrors;
+
+ i = 0;
+ LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
+ subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
+
+ if (new_actions && new_actions[i].odp_actions) {
+ free(subfacet->actions);
+ subfacet->actions = new_actions[i].odp_actions;
+ subfacet->actions_len = new_actions[i].actions_len;
+ }
+ i++;
}
+ free(new_actions);
+
if (facet->rule != new_rule) {
COVERAGE_INC(facet_changed_rule);
list_remove(&facet->list_node);
list_push_back(&new_rule->facets, &facet->list_node);
facet->rule = new_rule;
facet->used = new_rule->up.created;
- facet->rs_used = facet->used;
+ facet->prev_used = facet->used;
}
-
- ofpbuf_delete(odp_actions);
-
- return true;
}
/* Updates 'facet''s used time. Caller is responsible for calling
* facet_push_stats() to update the flows which 'facet' resubmits into. */
static void
-facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
- long long int used)
+facet_update_time(struct facet *facet, long long int used)
{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
if (used > facet->used) {
facet->used = used;
- if (used > facet->rule->used) {
- facet->rule->used = used;
- }
+ ofproto_rule_update_used(&facet->rule->up, used);
netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
}
}
-/* Folds the statistics from 'stats' into the counters in 'facet'.
- *
- * Because of the meaning of a facet's counters, it only makes sense to do this
- * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
- * packet that was sent by hand or if it represents statistics that have been
- * cleared out of the datapath. */
-static void
-facet_update_stats(struct ofproto_dpif *ofproto, struct facet *facet,
- const struct dpif_flow_stats *stats)
-{
- if (stats->n_packets || stats->used > facet->used) {
- facet_update_time(ofproto, facet, stats->used);
- facet->packet_count += stats->n_packets;
- facet->byte_count += stats->n_bytes;
- facet_push_stats(facet);
- netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
- }
-}
-
static void
facet_reset_counters(struct facet *facet)
{
facet->packet_count = 0;
facet->byte_count = 0;
- facet->rs_packet_count = 0;
- facet->rs_byte_count = 0;
+ facet->prev_packet_count = 0;
+ facet->prev_byte_count = 0;
facet->accounted_bytes = 0;
}
static void
facet_push_stats(struct facet *facet)
{
- uint64_t rs_packets, rs_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);
- assert(facet->packet_count >= facet->rs_packet_count);
- assert(facet->byte_count >= facet->rs_byte_count);
- assert(facet->used >= facet->rs_used);
+ stats.n_packets = facet->packet_count - facet->prev_packet_count;
+ stats.n_bytes = facet->byte_count - facet->prev_byte_count;
+ stats.used = facet->used;
+ stats.tcp_flags = 0;
- rs_packets = facet->packet_count - facet->rs_packet_count;
- rs_bytes = facet->byte_count - facet->rs_byte_count;
+ 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;
- if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
- facet->rs_packet_count = facet->packet_count;
- facet->rs_byte_count = facet->byte_count;
- facet->rs_used = facet->used;
+ flow_push_stats(facet->rule, &facet->flow, &stats);
- flow_push_stats(facet->rule, &facet->flow,
- rs_packets, rs_bytes, facet->used);
+ update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
+ 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;
- rule->used = MAX(push->used, rule->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. */
+ * 'rule''s actions and mirrors. */
static void
-flow_push_stats(const struct rule_dpif *rule,
- struct flow *flow, uint64_t packets, uint64_t bytes,
- long long int used)
+flow_push_stats(struct rule_dpif *rule,
+ const struct flow *flow, const struct dpif_flow_stats *stats)
{
struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
- struct ofproto_push push;
+ struct action_xlate_ctx ctx;
- push.packets = packets;
- push.bytes = bytes;
- push.used = used;
+ ofproto_rule_update_used(&rule->up, stats->used);
- action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
- push.ctx.resubmit_hook = push_resubmit;
- ofpbuf_delete(xlate_actions(&push.ctx,
- rule->up.actions, rule->up.n_actions));
+ action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
+ 0, NULL);
+ ctx.resubmit_stats = stats;
+ xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
+ rule->up.ofpacts_len);
}
\f
-/* Rules. */
+/* 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)
+{
+ struct subfacet *subfacet;
+
+ HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
+ &ofproto->subfacets) {
+ if (subfacet->key
+ ? (subfacet->key_len == key_len
+ && !memcmp(key, subfacet->key, key_len))
+ : flow_equal(flow, &subfacet->facet->flow)) {
+ return subfacet;
+ }
+ }
-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;
+ return NULL;
+}
- if (table_id >= N_TABLES) {
- return NULL;
+/* Searches 'facet' (within 'ofproto') for a subfacet with the specified
+ * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
+ * existing subfacet if there is one, otherwise creates and returns a
+ * new subfacet.
+ *
+ * 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, struct flow_miss *miss,
+ long long int now)
+{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
+ enum odp_key_fitness key_fitness = miss->key_fitness;
+ const struct nlattr *key = miss->key;
+ size_t key_len = miss->key_len;
+ uint32_t key_hash;
+ struct subfacet *subfacet;
+
+ key_hash = odp_flow_key_hash(key, key_len);
+
+ if (list_is_empty(&facet->subfacets)) {
+ subfacet = &facet->one_subfacet;
+ } else {
+ subfacet = subfacet_find(ofproto, key, key_len, key_hash,
+ &facet->flow);
+ if (subfacet) {
+ if (subfacet->facet == facet) {
+ return subfacet;
+ }
+
+ /* This shouldn't happen. */
+ VLOG_ERR_RL(&rl, "subfacet with wrong facet");
+ subfacet_destroy(subfacet);
+ }
+
+ subfacet = xmalloc(sizeof *subfacet);
}
- cls = &ofproto->up.tables[table_id];
- if (flow->nw_frag & FLOW_NW_FRAG_ANY
- && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
- /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
- * are unavailable. */
- struct flow ofpc_normal_flow = *flow;
- ofpc_normal_flow.tp_src = htons(0);
- ofpc_normal_flow.tp_dst = htons(0);
- cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
+ hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
+ list_push_back(&facet->subfacets, &subfacet->list_node);
+ subfacet->facet = facet;
+ subfacet->key_fitness = key_fitness;
+ if (key_fitness != ODP_FIT_PERFECT) {
+ subfacet->key = xmemdup(key, key_len);
+ subfacet->key_len = key_len;
} else {
- cls_rule = classifier_lookup(cls, flow);
- }
+ subfacet->key = NULL;
+ subfacet->key_len = 0;
+ }
+ 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;
+}
+
+/* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
+ * its facet within 'ofproto', and frees it. */
+static void
+subfacet_destroy__(struct subfacet *subfacet)
+{
+ struct facet *facet = subfacet->facet;
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
+
+ subfacet_uninstall(subfacet);
+ hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
+ list_remove(&subfacet->list_node);
+ free(subfacet->key);
+ free(subfacet->actions);
+ if (subfacet != &facet->one_subfacet) {
+ free(subfacet);
+ }
+}
+
+/* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
+ * last remaining subfacet in its facet destroys the facet too. */
+static void
+subfacet_destroy(struct subfacet *subfacet)
+{
+ struct facet *facet = subfacet->facet;
+
+ if (list_is_singleton(&facet->subfacets)) {
+ /* facet_remove() needs at least one subfacet (it will remove it). */
+ facet_remove(facet);
+ } else {
+ subfacet_destroy__(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. */
+static void
+subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
+ struct ofpbuf *key)
+{
+
+ if (!subfacet->key) {
+ struct flow *flow = &subfacet->facet->flow;
+
+ ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
+ odp_flow_key_from_flow(key, flow, subfacet->odp_in_port);
+ } else {
+ ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
+ }
+}
+
+/* Composes the datapath actions for 'subfacet' based on its rule's actions.
+ * Translates the actions into 'odp_actions', which the caller must have
+ * initialized and is responsible for uninitializing. */
+static void
+subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
+ struct ofpbuf *odp_actions)
+{
+ struct facet *facet = subfacet->facet;
+ struct rule_dpif *rule = facet->rule;
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
+
+ struct action_xlate_ctx ctx;
+
+ action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
+ rule, 0, packet);
+ xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
+ facet->tags = ctx.tags;
+ facet->has_learn = ctx.has_learn;
+ facet->has_normal = ctx.has_normal;
+ facet->has_fin_timeout = ctx.has_fin_timeout;
+ facet->nf_flow.output_iface = ctx.nf_output_iface;
+ facet->mirrors = ctx.mirrors;
+
+ subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
+ if (subfacet->actions_len != odp_actions->size
+ || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
+ free(subfacet->actions);
+ subfacet->actions_len = odp_actions->size;
+ subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
+ }
+}
+
+/* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
+ * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
+ * in the datapath will be zeroed and 'stats' will be updated with traffic new
+ * since 'subfacet' was last updated.
+ *
+ * Returns 0 if successful, otherwise a positive errno value. */
+static int
+subfacet_install(struct subfacet *subfacet,
+ const struct nlattr *actions, size_t actions_len,
+ struct dpif_flow_stats *stats,
+ enum slow_path_reason slow)
+{
+ 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;
+ int ret;
+
+ flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
+ if (stats) {
+ flags |= DPIF_FP_ZERO_STATS;
+ }
+
+ if (path == SF_SLOW_PATH) {
+ compose_slow_path(ofproto, &facet->flow, slow,
+ slow_path_stub, sizeof slow_path_stub,
+ &actions, &actions_len);
+ }
+
+ subfacet_get_key(subfacet, &keybuf, &key);
+ ret = dpif_flow_put(ofproto->backer->dpif, flags, key.data, key.size,
+ 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->path != SF_NOT_INSTALLED) {
+ struct rule_dpif *rule = subfacet->facet->rule;
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
+ struct odputil_keybuf keybuf;
+ struct dpif_flow_stats stats;
+ struct ofpbuf key;
+ int error;
+
+ subfacet_get_key(subfacet, &keybuf, &key);
+ error = dpif_flow_del(ofproto->backer->dpif,
+ key.data, key.size, &stats);
+ subfacet_reset_dp_stats(subfacet, &stats);
+ if (!error) {
+ subfacet_update_stats(subfacet, &stats);
+ }
+ subfacet->path = SF_NOT_INSTALLED;
+ } else {
+ assert(subfacet->dp_packet_count == 0);
+ assert(subfacet->dp_byte_count == 0);
+ }
+}
+
+/* Resets 'subfacet''s datapath statistics counters. This should be called
+ * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
+ * non-null, it should contain the statistics returned by dpif when 'subfacet'
+ * was reset in the datapath. 'stats' will be modified to include only
+ * statistics new since 'subfacet' was last updated. */
+static void
+subfacet_reset_dp_stats(struct subfacet *subfacet,
+ struct dpif_flow_stats *stats)
+{
+ if (stats
+ && subfacet->dp_packet_count <= stats->n_packets
+ && subfacet->dp_byte_count <= stats->n_bytes) {
+ stats->n_packets -= subfacet->dp_packet_count;
+ stats->n_bytes -= subfacet->dp_byte_count;
+ }
+
+ subfacet->dp_packet_count = 0;
+ subfacet->dp_byte_count = 0;
+}
+
+/* Updates 'subfacet''s used time. The caller is responsible for calling
+ * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
+static void
+subfacet_update_time(struct subfacet *subfacet, long long int used)
+{
+ if (used > subfacet->used) {
+ subfacet->used = used;
+ facet_update_time(subfacet->facet, used);
+ }
+}
+
+/* Folds the statistics from 'stats' into the counters in 'subfacet'.
+ *
+ * Because of the meaning of a subfacet's counters, it only makes sense to do
+ * this if 'stats' are not tracked in the datapath, that is, if 'stats'
+ * represents a packet that was sent by hand or if it represents statistics
+ * that have been cleared out of the datapath. */
+static void
+subfacet_update_stats(struct subfacet *subfacet,
+ const struct dpif_flow_stats *stats)
+{
+ if (stats->n_packets || stats->used > subfacet->used) {
+ struct facet *facet = subfacet->facet;
+
+ subfacet_update_time(subfacet, stats->used);
+ facet->packet_count += stats->n_packets;
+ facet->byte_count += stats->n_bytes;
+ facet->tcp_flags |= stats->tcp_flags;
+ facet_push_stats(facet);
+ netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
+ }
+}
+\f
+/* Rules. */
+
+static struct rule_dpif *
+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;
+
+ if (table_id >= N_TABLES) {
+ return NULL;
+ }
+
+ cls = &ofproto->up.tables[table_id].cls;
+ if (flow->nw_frag & FLOW_NW_FRAG_ANY
+ && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
+ /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
+ * are unavailable. */
+ struct flow ofpc_normal_flow = *flow;
+ ofpc_normal_flow.tp_src = htons(0);
+ ofpc_normal_flow.tp_dst = htons(0);
+ cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
+ } else {
+ cls_rule = classifier_lookup(cls, 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)
{
free(rule);
}
-static int
+static enum ofperr
rule_construct(struct rule *rule_)
{
struct rule_dpif *rule = rule_dpif_cast(rule_);
struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
struct rule_dpif *victim;
uint8_t table_id;
- int error;
-
- error = validate_actions(rule->up.actions, rule->up.n_actions,
- &rule->up.cr.flow, ofproto->max_ports);
- if (error) {
- return error;
- }
- rule->used = rule->up.created;
rule->packet_count = 0;
rule->byte_count = 0;
}
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;
rule_destruct(struct rule *rule_)
{
struct rule_dpif *rule = rule_dpif_cast(rule_);
- struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
struct facet *facet, *next_facet;
LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
- facet_revalidate(ofproto, facet);
+ facet_revalidate(facet);
}
complete_operation(rule);
}
}
-static int
-rule_execute(struct rule *rule_, struct flow *flow, struct ofpbuf *packet)
+static enum ofperr
+rule_execute(struct rule *rule_, const struct flow *flow,
+ struct ofpbuf *packet)
{
struct rule_dpif *rule = rule_dpif_cast(rule_);
struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
+
+ struct dpif_flow_stats stats;
+
struct action_xlate_ctx ctx;
- struct ofpbuf *odp_actions;
- struct facet *facet;
- size_t size;
+ uint64_t odp_actions_stub[1024 / 8];
+ struct ofpbuf odp_actions;
- /* First look for a related facet. If we find one, account it to that. */
- facet = facet_lookup_valid(ofproto, flow);
- if (facet && facet->rule == rule) {
- if (!facet->may_install) {
- facet_make_actions(ofproto, facet, packet);
- }
- facet_execute(ofproto, facet, packet);
- return 0;
- }
+ dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
+ rule_credit_stats(rule, &stats);
- /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
- * create a new facet for it and use that. */
- if (rule_dpif_lookup(ofproto, flow, 0) == rule) {
- facet = facet_create(rule, flow);
- facet_make_actions(ofproto, facet, packet);
- facet_execute(ofproto, facet, packet);
- facet_install(ofproto, facet, true);
- return 0;
- }
+ ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
+ action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
+ rule, stats.tcp_flags, packet);
+ ctx.resubmit_stats = &stats;
+ xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
- /* We can't account anything to a facet. If we were to try, then that
- * facet would have a non-matching rule, busting our invariants. */
- action_xlate_ctx_init(&ctx, ofproto, 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->used = time_msec();
- rule->packet_count++;
- rule->byte_count += size;
- flow_push_stats(rule, flow, 1, size, rule->used);
- }
- ofpbuf_delete(odp_actions);
+ execute_odp_actions(ofproto, flow, odp_actions.data,
+ odp_actions.size, packet);
+
+ ofpbuf_uninit(&odp_actions);
return 0;
}
rule_modify_actions(struct rule *rule_)
{
struct rule_dpif *rule = rule_dpif_cast(rule_);
- struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
- int 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);
}
\f
-/* Sends 'packet' out of port 'odp_port' within 'ofproto'.
+/* Sends 'packet' out 'ofport'.
+ * May modify 'packet'.
* Returns 0 if successful, otherwise a positive errno value. */
static int
-send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port,
- const struct ofpbuf *packet)
+send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
{
+ const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
struct ofpbuf key, odp_actions;
struct odputil_keybuf keybuf;
+ uint32_t odp_port;
struct flow flow;
int error;
- flow_extract((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) {
+ eth_pop_vlan(packet);
+ flow.vlan_tci = htons(0);
+ }
+
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);
VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
ofproto->up.name, odp_port, strerror(error));
}
+ ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
return error;
}
\f
/* 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)
{
- size_t offset;
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;
- offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_USERSPACE);
- nl_msg_put_u32(odp_actions, OVS_USERSPACE_ATTR_PID, pid);
- nl_msg_put_unspec(odp_actions, OVS_USERSPACE_ATTR_USERDATA,
- cookie, sizeof *cookie);
- nl_msg_end_nested(odp_actions, offset);
+ 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;
- return odp_actions->size - NLA_ALIGN(sizeof *cookie);
+ 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. */
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. */
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);
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;
-}
-
-static void
-commit_action__(struct ofpbuf *odp_actions,
- enum ovs_action_attr act_type,
- enum ovs_key_attr key_type,
- const void *key, size_t key_size)
-{
- size_t offset = nl_msg_start_nested(odp_actions, act_type);
-
- nl_msg_put_unspec(odp_actions, key_type, key, key_size);
- nl_msg_end_nested(odp_actions, offset);
+ compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
+ ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
}
static void
-commit_set_tun_id_action(const struct flow *flow, struct flow *base,
- struct ofpbuf *odp_actions)
+compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
+ bool check_stp)
{
- if (base->tun_id == flow->tun_id) {
- return;
- }
- base->tun_id = flow->tun_id;
+ const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
+ uint32_t odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port);
+ ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
+ uint8_t flow_nw_tos = ctx->flow.nw_tos;
+ uint16_t out_port;
- commit_action__(odp_actions, OVS_ACTION_ATTR_SET,
- OVS_KEY_ATTR_TUN_ID, &base->tun_id, sizeof(base->tun_id));
-}
+ if (ofport) {
+ struct priority_to_dscp *pdscp;
-static void
-commit_set_ether_addr_action(const struct flow *flow, struct flow *base,
- struct ofpbuf *odp_actions)
-{
- struct ovs_key_ethernet eth_key;
+ if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
+ xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
+ return;
+ } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
+ xlate_report(ctx, "STP not in forwarding state, skipping output");
+ return;
+ }
- if (eth_addr_equals(base->dl_src, flow->dl_src) &&
- eth_addr_equals(base->dl_dst, flow->dl_dst)) {
- return;
+ pdscp = get_priority(ofport, ctx->flow.skb_priority);
+ if (pdscp) {
+ ctx->flow.nw_tos &= ~IP_DSCP_MASK;
+ ctx->flow.nw_tos |= pdscp->dscp;
+ }
+ } else {
+ /* We may not have an ofport record for this port, but it doesn't hurt
+ * to allow forwarding to it anyhow. Maybe such a port will appear
+ * later and we're pre-populating the flow table. */
}
- memcpy(base->dl_src, flow->dl_src, ETH_ADDR_LEN);
- memcpy(base->dl_dst, flow->dl_dst, ETH_ADDR_LEN);
-
- memcpy(eth_key.eth_src, base->dl_src, ETH_ADDR_LEN);
- memcpy(eth_key.eth_dst, base->dl_dst, ETH_ADDR_LEN);
+ out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
+ ctx->flow.vlan_tci);
+ if (out_port != odp_port) {
+ ctx->flow.vlan_tci = htons(0);
+ }
+ commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
+ nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
- commit_action__(odp_actions, OVS_ACTION_ATTR_SET,
- OVS_KEY_ATTR_ETHERNET, ð_key, sizeof(eth_key));
+ ctx->sflow_odp_port = odp_port;
+ ctx->sflow_n_outputs++;
+ ctx->nf_output_iface = ofp_port;
+ ctx->flow.vlan_tci = flow_vlan_tci;
+ ctx->flow.nw_tos = flow_nw_tos;
}
static void
-commit_vlan_action(struct action_xlate_ctx *ctx, ovs_be16 new_tci)
+compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
{
- struct flow *base = &ctx->base_flow;
-
- if (base->vlan_tci == new_tci) {
- return;
- }
-
- if (base->vlan_tci & htons(VLAN_CFI)) {
- nl_msg_put_u16(ctx->odp_actions, OVS_ACTION_ATTR_POP,
- OVS_KEY_ATTR_8021Q);
- }
-
- if (new_tci & htons(VLAN_CFI)) {
- struct ovs_key_8021q q_key;
-
- q_key.q_tpid = htons(ETH_TYPE_VLAN);
- q_key.q_tci = new_tci & ~htons(VLAN_CFI);
-
- commit_action__(ctx->odp_actions, OVS_ACTION_ATTR_PUSH,
- OVS_KEY_ATTR_8021Q, &q_key, sizeof(q_key));
- }
- base->vlan_tci = new_tci;
+ compose_output_action__(ctx, ofp_port, true);
}
static void
-commit_set_nw_action(const struct flow *flow, struct flow *base,
- struct ofpbuf *odp_actions)
+xlate_table_action(struct action_xlate_ctx *ctx,
+ uint16_t in_port, uint8_t table_id, bool may_packet_in)
{
- struct ovs_key_ipv4 ipv4_key;
-
- if (base->dl_type != htons(ETH_TYPE_IP) ||
- !base->nw_src || !base->nw_dst) {
- return;
- }
+ if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
+ struct ofproto_dpif *ofproto = ctx->ofproto;
+ struct rule_dpif *rule;
+ uint16_t old_in_port;
+ uint8_t old_table_id;
- if (base->nw_src == flow->nw_src &&
- base->nw_dst == flow->nw_dst &&
- base->nw_tos == flow->nw_tos &&
- base->nw_ttl == flow->nw_ttl &&
- base->nw_frag == flow->nw_frag) {
- return;
- }
-
- ipv4_key.ipv4_src = base->nw_src = flow->nw_src;
- ipv4_key.ipv4_dst = base->nw_dst = flow->nw_dst;
- ipv4_key.ipv4_proto = base->nw_proto;
- ipv4_key.ipv4_tos = flow->nw_tos;
- ipv4_key.ipv4_ttl = flow->nw_ttl;
- ipv4_key.ipv4_frag = (base->nw_frag == 0 ? OVS_FRAG_TYPE_NONE
- : base->nw_frag == FLOW_NW_FRAG_ANY
- ? OVS_FRAG_TYPE_FIRST : OVS_FRAG_TYPE_LATER);
-
- commit_action__(odp_actions, OVS_ACTION_ATTR_SET,
- OVS_KEY_ATTR_IPV4, &ipv4_key, sizeof(ipv4_key));
-}
-
-static void
-commit_set_port_action(const struct flow *flow, struct flow *base,
- struct ofpbuf *odp_actions)
-{
- if (!base->tp_src || !base->tp_dst) {
- return;
- }
-
- if (base->tp_src == flow->tp_src &&
- base->tp_dst == flow->tp_dst) {
- return;
- }
-
- if (flow->nw_proto == IPPROTO_TCP) {
- struct ovs_key_tcp port_key;
-
- port_key.tcp_src = base->tp_src = flow->tp_src;
- port_key.tcp_dst = base->tp_dst = flow->tp_dst;
-
- commit_action__(odp_actions, OVS_ACTION_ATTR_SET,
- OVS_KEY_ATTR_TCP, &port_key, sizeof(port_key));
-
- } else if (flow->nw_proto == IPPROTO_UDP) {
- struct ovs_key_udp port_key;
-
- port_key.udp_src = base->tp_src = flow->tp_src;
- port_key.udp_dst = base->tp_dst = flow->tp_dst;
-
- commit_action__(odp_actions, OVS_ACTION_ATTR_SET,
- OVS_KEY_ATTR_UDP, &port_key, sizeof(port_key));
- }
-}
-
-static void
-commit_set_priority_action(const struct flow *flow, struct flow *base,
- struct ofpbuf *odp_actions)
-{
- if (base->priority == flow->priority) {
- return;
- }
- base->priority = flow->priority;
-
- commit_action__(odp_actions, OVS_ACTION_ATTR_SET,
- OVS_KEY_ATTR_PRIORITY, &base->priority,
- sizeof(base->priority));
-}
-
-static void
-commit_odp_actions(struct action_xlate_ctx *ctx)
-{
- const struct flow *flow = &ctx->flow;
- struct flow *base = &ctx->base_flow;
- struct ofpbuf *odp_actions = ctx->odp_actions;
-
- commit_set_tun_id_action(flow, base, odp_actions);
- commit_set_ether_addr_action(flow, base, odp_actions);
- commit_vlan_action(ctx, flow->vlan_tci);
- commit_set_nw_action(flow, base, odp_actions);
- commit_set_port_action(flow, base, odp_actions);
- commit_set_priority_action(flow, base, odp_actions);
-}
-
-static void
-compose_output_action(struct action_xlate_ctx *ctx, uint16_t odp_port)
-{
- nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
- ctx->sflow_odp_port = odp_port;
- ctx->sflow_n_outputs++;
-}
-
-static void
-add_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
-{
- const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
- uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
-
- if (ofport) {
- if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)
- || !stp_forward_in_state(ofport->stp_state)) {
- /* Forwarding disabled on port. */
- return;
- }
- } else {
- /*
- * We don't have an ofport record for this port, but it doesn't hurt to
- * allow forwarding to it anyhow. Maybe such a port will appear later
- * and we're pre-populating the flow table.
- */
- }
-
- commit_odp_actions(ctx);
- compose_output_action(ctx, odp_port);
- ctx->nf_output_iface = ofp_port;
-}
-
-static void
-xlate_table_action(struct action_xlate_ctx *ctx,
- uint16_t in_port, uint8_t table_id)
-{
- if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
- struct ofproto_dpif *ofproto = ctx->ofproto;
- struct rule_dpif *rule;
- uint16_t old_in_port;
- uint8_t old_table_id;
-
- old_table_id = ctx->table_id;
- ctx->table_id = table_id;
+ old_table_id = ctx->table_id;
+ ctx->table_id = table_id;
/* Look up a flow with 'in_port' as the input port. */
old_in_port = ctx->flow.in_port;
ctx->flow.in_port = in_port;
- rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
+ 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));
}
}
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++;
- do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
+ ctx->rule = rule;
+ do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
+ ctx->rule = old_rule;
ctx->recurse--;
}
VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
MAX_RESUBMIT_RECURSION);
+ ctx->max_resubmit_trigger = true;
}
}
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
-flood_packets(struct action_xlate_ctx *ctx, ovs_be32 mask)
+flood_packets(struct action_xlate_ctx *ctx, bool all)
{
struct ofport_dpif *ofport;
- commit_odp_actions(ctx);
HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
uint16_t ofp_port = ofport->up.ofp_port;
- if (ofp_port != ctx->flow.in_port
- && !(ofport->up.opp.config & mask)
- && stp_forward_in_state(ofport->stp_state)) {
- compose_output_action(ctx, ofport->odp_port);
+
+ if (ofp_port == ctx->flow.in_port) {
+ continue;
+ }
+
+ if (all) {
+ compose_output_action__(ctx, ofp_port, false);
+ } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
+ compose_output_action(ctx, ofp_port);
}
}
}
static void
-compose_controller_action(struct action_xlate_ctx *ctx, int len)
+execute_controller_action(struct action_xlate_ctx *ctx, int len,
+ enum ofp_packet_in_reason reason,
+ uint16_t controller_id)
+{
+ struct ofputil_packet_in pin;
+ struct ofpbuf *packet;
+
+ ctx->slow |= SLOW_CONTROLLER;
+ if (!ctx->packet) {
+ return;
+ }
+
+ packet = ofpbuf_clone(ctx->packet);
+
+ if (packet->l2 && packet->l3) {
+ struct eth_header *eh;
+
+ eth_pop_vlan(packet);
+ eh = packet->l2;
+
+ /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
+ * LLC frame. Calculating the Ethernet type of these frames is more
+ * trouble than seems appropriate for a simple assertion. */
+ assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
+ || eh->eth_type == ctx->flow.dl_type);
+
+ memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
+ memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
+
+ if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
+ eth_push_vlan(packet, ctx->flow.vlan_tci);
+ }
+
+ if (packet->l4) {
+ if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
+ packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
+ ctx->flow.nw_tos, ctx->flow.nw_ttl);
+ }
+
+ if (packet->l7) {
+ if (ctx->flow.nw_proto == IPPROTO_TCP) {
+ packet_set_tcp_port(packet, ctx->flow.tp_src,
+ ctx->flow.tp_dst);
+ } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
+ packet_set_udp_port(packet, ctx->flow.tp_src,
+ ctx->flow.tp_dst);
+ }
+ }
+ }
+ }
+
+ pin.packet = packet->data;
+ pin.packet_len = packet->size;
+ pin.reason = reason;
+ pin.controller_id = controller_id;
+ pin.table_id = ctx->table_id;
+ pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
+
+ pin.send_len = len;
+ flow_get_metadata(&ctx->flow, &pin.fmd);
+
+ connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
+ ofpbuf_delete(packet);
+}
+
+static bool
+compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
{
- struct user_action_cookie cookie;
+ if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
+ ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
+ return false;
+ }
+
+ if (ctx->flow.nw_ttl > 1) {
+ ctx->flow.nw_ttl--;
+ return false;
+ } else {
+ size_t i;
+
+ for (i = 0; i < ids->n_controllers; i++) {
+ execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
+ ids->cnt_ids[i]);
+ }
- cookie.type = USER_ACTION_COOKIE_CONTROLLER;
- cookie.data = len;
- cookie.n_output = 0;
- cookie.vlan_tci = 0;
- put_userspace_action(ctx->ofproto, ctx->odp_actions, &ctx->flow, &cookie);
+ /* Stop processing for current table. */
+ return true;
+ }
}
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;
switch (port) {
case OFPP_IN_PORT:
- add_output_action(ctx, ctx->flow.in_port);
+ 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);
break;
case OFPP_FLOOD:
- flood_packets(ctx, htonl(OFPPC_NO_FLOOD));
+ flood_packets(ctx, false);
break;
case OFPP_ALL:
- flood_packets(ctx, htonl(0));
+ flood_packets(ctx, true);
break;
case OFPP_CONTROLLER:
- commit_odp_actions(ctx);
- compose_controller_action(ctx, max_len);
- break;
- case OFPP_LOCAL:
- add_output_action(ctx, OFPP_LOCAL);
+ execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
break;
case OFPP_NONE:
break;
+ case OFPP_LOCAL:
default:
if (port != ctx->flow.in_port) {
- add_output_action(ctx, port);
+ compose_output_action(ctx, port);
+ } else {
+ xlate_report(ctx, "skipping output to input port");
}
break;
}
static void
xlate_output_reg_action(struct action_xlate_ctx *ctx,
- const struct nx_action_output_reg *naor)
+ const struct ofpact_output_reg *or)
{
- uint64_t ofp_port;
-
- ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &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, odp_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) {
return;
}
- odp_port = ofp_port_to_odp_port(ofp_port);
/* Add datapath actions. */
- flow_priority = ctx->flow.priority;
- ctx->flow.priority = priority;
- add_output_action(ctx, odp_port);
- ctx->flow.priority = flow_priority;
+ flow_priority = ctx->flow.skb_priority;
+ ctx->flow.skb_priority = priority;
+ compose_output_action(ctx, ofp_port);
+ ctx->flow.skb_priority = flow_priority;
/* Update NetFlow output port. */
if (ctx->nf_output_iface == NF_OUT_DROP) {
- ctx->nf_output_iface = odp_port;
+ ctx->nf_output_iface = ofp_port;
} else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
ctx->nf_output_iface = NF_OUT_MULTI;
}
}
static void
-xlate_set_queue_action(struct action_xlate_ctx *ctx,
- 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.priority = priority;
}
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) {
ofp_port = slave->up.ofp_port;
}
}
- autopath_execute(naa, &ctx->flow, ofp_port);
+ nxm_reg_load(&ap->dst, ofp_port, &ctx->flow);
}
static bool
}
}
+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)) {
- char *msg = ofputil_error_to_string(error);
- VLOG_WARN("learning action failed to modify flow table (%s)", msg);
- free(msg);
+ VLOG_WARN("learning action failed to modify flow table (%s)",
+ ofperr_get_name(error));
+ }
+
+ ofpbuf_uninit(&ofpacts);
+}
+
+/* Reduces '*timeout' to no more than 'max'. A value of zero in either case
+ * means "infinite". */
+static void
+reduce_timeout(uint16_t max, uint16_t *timeout)
+{
+ if (max && (!*timeout || *timeout > max)) {
+ *timeout = max;
}
+}
- free(fm.actions);
+static void
+xlate_fin_timeout(struct action_xlate_ctx *ctx,
+ const struct ofpact_fin_timeout *oft)
+{
+ if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
+ struct rule_dpif *rule = ctx->rule;
+
+ reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
+ reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
+ }
}
static bool
may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
{
- if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
- ? htonl(OFPPC_NO_RECV_STP)
- : htonl(OFPPC_NO_RECV))) {
+ if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
+ ? OFPUTIL_PC_NO_RECV_STP
+ : OFPUTIL_PC_NO_RECV)) {
return false;
}
}
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;
- size_t left;
+ bool was_evictable = true;
+ const struct ofpact *a;
port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
if (port && !may_receive(port, ctx)) {
return;
}
- 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;
- enum ofputil_action_code code;
- ovs_be64 tun_id;
+ if (ctx->rule) {
+ /* Don't let the rule we're working on get evicted underneath us. */
+ was_evictable = ctx->rule->up.evictable;
+ ctx->rule->up.evictable = false;
+ }
+ OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
+ struct ofpact_controller *controller;
+ const struct ofpact_metadata *metadata;
if (ctx->exit) {
break;
}
- code = ofputil_decode_action_unsafe(ia);
- switch (code) {
- case OFPUTIL_OFPAT_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 OFPUTIL_OFPAT_SET_VLAN_VID:
- ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
- ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
+ case OFPACT_CONTROLLER:
+ controller = ofpact_get_CONTROLLER(a);
+ execute_controller_action(ctx, controller->max_len,
+ controller->reason,
+ controller->controller_id);
break;
- case OFPUTIL_OFPAT_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);
+ case OFPACT_ENQUEUE:
+ xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
break;
- case OFPUTIL_OFPAT_STRIP_VLAN:
- ctx->flow.vlan_tci = htons(0);
+ case OFPACT_SET_VLAN_VID:
+ ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
+ ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
+ | htons(VLAN_CFI));
break;
- case OFPUTIL_OFPAT_SET_DL_SRC:
- oada = ((struct ofp_action_dl_addr *) ia);
- memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
+ case OFPACT_SET_VLAN_PCP:
+ ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
+ ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
+ << VLAN_PCP_SHIFT)
+ | VLAN_CFI);
break;
- case OFPUTIL_OFPAT_SET_DL_DST:
- oada = ((struct ofp_action_dl_addr *) ia);
- memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
+ case OFPACT_STRIP_VLAN:
+ ctx->flow.vlan_tci = htons(0);
break;
- case OFPUTIL_OFPAT_SET_NW_SRC:
- ctx->flow.nw_src = ia->nw_addr.nw_addr;
+ case OFPACT_PUSH_VLAN:
+ /* TODO:XXX 802.1AD(QinQ) */
+ ctx->flow.vlan_tci = htons(VLAN_CFI);
break;
- case OFPUTIL_OFPAT_SET_NW_DST:
- ctx->flow.nw_dst = ia->nw_addr.nw_addr;
+ case OFPACT_SET_ETH_SRC:
+ memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
+ ETH_ADDR_LEN);
break;
- case OFPUTIL_OFPAT_SET_NW_TOS:
- ctx->flow.nw_tos &= ~IP_DSCP_MASK;
- ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
+ case OFPACT_SET_ETH_DST:
+ memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
+ ETH_ADDR_LEN);
break;
- case OFPUTIL_OFPAT_SET_TP_SRC:
- ctx->flow.tp_src = ia->tp_port.tp_port;
+ case OFPACT_SET_IPV4_SRC:
+ ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
break;
- case OFPUTIL_OFPAT_SET_TP_DST:
- ctx->flow.tp_dst = ia->tp_port.tp_port;
+ case OFPACT_SET_IPV4_DST:
+ ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
break;
- case OFPUTIL_OFPAT_ENQUEUE:
- xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
+ case OFPACT_SET_IPV4_DSCP:
+ /* OpenFlow 1.0 only supports IPv4. */
+ if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
+ ctx->flow.nw_tos &= ~IP_DSCP_MASK;
+ ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
+ }
break;
- case 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_SRC_PORT:
+ ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
break;
- case OFPUTIL_NXAST_RESUBMIT_TABLE:
- xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
+ case OFPACT_SET_L4_DST_PORT:
+ ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
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_RESUBMIT:
+ xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
break;
- case OFPUTIL_NXAST_SET_QUEUE:
- nasq = (const struct nx_action_set_queue *) ia;
- xlate_set_queue_action(ctx, nasq);
+ case OFPACT_SET_TUNNEL:
+ ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
break;
- case OFPUTIL_NXAST_POP_QUEUE:
- ctx->flow.priority = ctx->original_priority;
+ case OFPACT_SET_QUEUE:
+ xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
break;
- case OFPUTIL_NXAST_REG_MOVE:
- nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
- &ctx->flow);
+ case OFPACT_POP_QUEUE:
+ ctx->flow.skb_priority = ctx->orig_skb_priority;
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_learn) {
- xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
+ xlate_learn_action(ctx, ofpact_get_LEARN(a));
}
break;
- case OFPUTIL_NXAST_EXIT:
+ case OFPACT_EXIT:
ctx->exit = true;
break;
+
+ case OFPACT_FIN_TIMEOUT:
+ ctx->has_fin_timeout = true;
+ xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
+ break;
+
+ case OFPACT_CLEAR_ACTIONS:
+ /* TODO:XXX
+ * Nothing to do because writa-actions is not supported for now.
+ * When writa-actions is supported, clear-actions also must
+ * be supported at the same time.
+ */
+ break;
+
+ case OFPACT_WRITE_METADATA:
+ metadata = ofpact_get_WRITE_METADATA(a);
+ ctx->flow.metadata &= ~metadata->mask;
+ ctx->flow.metadata |= metadata->metadata & metadata->mask;
+ break;
+
+ case OFPACT_GOTO_TABLE: {
+ /* TODO:XXX remove recursion */
+ /* It is assumed that goto-table is last action */
+ struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
+ assert(ctx->table_id < ogt->table_id);
+ xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
+ break;
+ }
}
}
+out:
/* We've let OFPP_NORMAL and the learning action look at the packet,
* so drop it now if forwarding is disabled. */
if (port && !stp_forward_in_state(port->stp_state)) {
ofpbuf_clear(ctx->odp_actions);
add_sflow_action(ctx);
}
+ if (ctx->rule) {
+ ctx->rule->up.evictable = was_evictable;
+ }
}
static void
action_xlate_ctx_init(struct action_xlate_ctx *ctx,
struct ofproto_dpif *ofproto, const struct flow *flow,
- const struct ofpbuf *packet)
+ ovs_be16 initial_tci, struct rule_dpif *rule,
+ uint8_t tcp_flags, const struct ofpbuf *packet)
{
ctx->ofproto = ofproto;
ctx->flow = *flow;
+ ctx->base_flow = ctx->flow;
+ memset(&ctx->base_flow.tunnel, 0, sizeof ctx->base_flow.tunnel);
+ ctx->base_flow.vlan_tci = initial_tci;
+ ctx->rule = rule;
ctx->packet = packet;
ctx->may_learn = packet != NULL;
+ ctx->tcp_flags = tcp_flags;
ctx->resubmit_hook = NULL;
+ ctx->report_hook = NULL;
+ ctx->resubmit_stats = NULL;
}
-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)
{
+ /* 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;
ctx->nf_output_iface = NF_OUT_DROP;
+ ctx->mirrors = 0;
ctx->recurse = 0;
- ctx->original_priority = ctx->flow.priority;
- ctx->base_flow = ctx->flow;
- ctx->base_flow.tun_id = 0;
+ ctx->max_resubmit_trigger = false;
+ ctx->orig_skb_priority = ctx->flow.skb_priority;
ctx->table_id = 0;
ctx->exit = false;
+ if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
+ /* Do this conditionally because the copy is expensive enough that it
+ * shows up in profiles.
+ *
+ * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
+ * believe that I wasn't using it without initializing it if I kept it
+ * in a local variable. */
+ ctx->orig_flow = ctx->flow;
+ }
+
if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
switch (ctx->ofproto->up.frag_handling) {
case OFPC_FRAG_NORMAL:
break;
case OFPC_FRAG_DROP:
- return ctx->odp_actions;
+ return;
case OFPC_FRAG_REASM:
NOT_REACHED();
case OFPC_FRAG_NX_MATCH:
/* Nothing to do. */
break;
+
+ case OFPC_INVALID_TTL_TO_CONTROLLER:
+ NOT_REACHED();
}
}
- 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);
+ ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
+
add_sflow_action(ctx);
- do_xlate_actions(in, n_in, ctx);
+ do_xlate_actions(ofpacts, ofpacts_len, ctx);
+
+ if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
+ if (!hit_resubmit_limit) {
+ /* We didn't record the original flow. Make sure we do from
+ * now on. */
+ hit_resubmit_limit = true;
+ } else if (!VLOG_DROP_ERR(&trace_rl)) {
+ struct ds ds = DS_EMPTY_INITIALIZER;
+
+ ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
+ initial_tci, &ds);
+ VLOG_ERR("Trace triggered by excessive resubmit "
+ "recursion:\n%s", ds_cstr(&ds));
+ ds_destroy(&ds);
+ }
+ }
if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
ctx->odp_actions->data,
ctx->odp_actions->size)) {
- ctx->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, OVSP_LOCAL);
+ compose_output_action(ctx, OFPP_LOCAL);
}
}
+ if (ctx->ofproto->has_mirrors) {
+ add_mirror_actions(ctx, &ctx->orig_flow);
+ }
fix_sflow_action(ctx);
}
-
- return ctx->odp_actions;
}
-\f
-/* OFPP_NORMAL implementation. */
-struct dst {
- struct ofport_dpif *port;
- uint16_t vid;
-};
+/* 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;
-struct dst_set {
- struct dst builtin[32];
- struct dst *dsts;
- size_t n, allocated;
-};
+ ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
+ xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
+ ofpbuf_uninit(&odp_actions);
+}
-static void dst_set_init(struct dst_set *);
-static void dst_set_add(struct dst_set *, const struct dst *);
-static void dst_set_free(struct dst_set *);
+static void
+xlate_report(struct action_xlate_ctx *ctx, const char *s)
+{
+ if (ctx->report_hook) {
+ ctx->report_hook(ctx, s);
+ }
+}
+\f
+/* OFPP_NORMAL implementation. */
static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
}
}
+/* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
+ * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
+ * a warning.
+ *
+ * 'vid' should be the VID obtained from the 802.1Q header that was received as
+ * part of a packet (specify 0 if there was no 802.1Q header), in the range
+ * 0...4095. */
+static bool
+input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
+{
+ /* Allow any VID on the OFPP_NONE port. */
+ if (in_bundle == &ofpp_none_bundle) {
+ return true;
+ }
+
+ switch (in_bundle->vlan_mode) {
+ case PORT_VLAN_ACCESS:
+ if (vid) {
+ if (warn) {
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+ VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
+ "packet received on port %s configured as VLAN "
+ "%"PRIu16" access port",
+ in_bundle->ofproto->up.name, vid,
+ in_bundle->name, in_bundle->vlan);
+ }
+ return false;
+ }
+ return true;
+
+ case PORT_VLAN_NATIVE_UNTAGGED:
+ case PORT_VLAN_NATIVE_TAGGED:
+ if (!vid) {
+ /* Port must always carry its native VLAN. */
+ return true;
+ }
+ /* Fall through. */
+ case PORT_VLAN_TRUNK:
+ if (!ofbundle_includes_vlan(in_bundle, vid)) {
+ if (warn) {
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+ VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
+ "received on port %s not configured for trunking "
+ "VLAN %"PRIu16,
+ in_bundle->ofproto->up.name, vid,
+ in_bundle->name, vid);
+ }
+ return false;
+ }
+ return true;
+
+ default:
+ NOT_REACHED();
+ }
+
+}
+
/* Given 'vlan', the VLAN that a packet belongs to, and
* 'out_bundle', a bundle on which the packet is to be output, returns the VID
* that should be included in the 802.1Q header. (If the return value is 0,
}
}
-static bool
-set_dst(struct action_xlate_ctx *ctx, struct dst *dst,
- const struct ofbundle *in_bundle, const struct ofbundle *out_bundle)
+static void
+output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
+ uint16_t vlan)
{
- uint16_t vlan;
+ struct ofport_dpif *port;
+ uint16_t vid;
+ ovs_be16 tci, old_tci;
+
+ vid = output_vlan_to_vid(out_bundle, vlan);
+ if (!out_bundle->bond) {
+ port = ofbundle_get_a_port(out_bundle);
+ } else {
+ port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
+ vid, &ctx->tags);
+ if (!port) {
+ /* No slaves enabled, so drop packet. */
+ return;
+ }
+ }
- vlan = input_vid_to_vlan(in_bundle, vlan_tci_to_vid(ctx->flow.vlan_tci));
- dst->vid = output_vlan_to_vid(out_bundle, vlan);
+ old_tci = ctx->flow.vlan_tci;
+ tci = htons(vid);
+ if (tci || out_bundle->use_priority_tags) {
+ tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
+ if (tci) {
+ tci |= htons(VLAN_CFI);
+ }
+ }
+ ctx->flow.vlan_tci = tci;
- dst->port = (!out_bundle->bond
- ? ofbundle_get_a_port(out_bundle)
- : bond_choose_output_slave(out_bundle->bond, &ctx->flow,
- dst->vid, &ctx->tags));
- return dst->port != NULL;
+ compose_output_action(ctx, port->up.ofp_port);
+ ctx->flow.vlan_tci = old_tci;
}
static int
return ffs(mask);
}
-static void
-dst_set_init(struct dst_set *set)
+static bool
+ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
{
- set->dsts = set->builtin;
- set->n = 0;
- set->allocated = ARRAY_SIZE(set->builtin);
-}
-
-static void
-dst_set_add(struct dst_set *set, const struct dst *dst)
-{
- if (set->n >= set->allocated) {
- size_t new_allocated;
- struct dst *new_dsts;
-
- new_allocated = set->allocated * 2;
- new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
- memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
-
- dst_set_free(set);
-
- set->dsts = new_dsts;
- set->allocated = new_allocated;
- }
- set->dsts[set->n++] = *dst;
-}
-
-static void
-dst_set_free(struct dst_set *set)
-{
- if (set->dsts != set->builtin) {
- free(set->dsts);
- }
-}
-
-static bool
-dst_is_duplicate(const struct dst_set *set, const struct dst *test)
-{
- size_t i;
- for (i = 0; i < set->n; i++) {
- if (set->dsts[i].vid == test->vid
- && set->dsts[i].port == test->port) {
- return true;
- }
- }
- return false;
-}
-
-static bool
-ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
-{
- return (bundle->vlan_mode != PORT_VLAN_ACCESS
- && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
+ return (bundle->vlan_mode != PORT_VLAN_ACCESS
+ && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
}
static bool
struct ofport_dpif, bundle_node);
}
-static void
-compose_dsts(struct action_xlate_ctx *ctx, uint16_t vlan,
- const struct ofbundle *in_bundle,
- const struct ofbundle *out_bundle, struct dst_set *set)
-{
- struct dst dst;
-
- if (out_bundle == OFBUNDLE_FLOOD) {
- struct ofbundle *bundle;
-
- HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
- if (bundle != in_bundle
- && ofbundle_includes_vlan(bundle, vlan)
- && bundle->floodable
- && !bundle->mirror_out
- && set_dst(ctx, &dst, in_bundle, bundle)) {
- dst_set_add(set, &dst);
- }
- }
- ctx->nf_output_iface = NF_OUT_FLOOD;
- } else if (out_bundle && set_dst(ctx, &dst, in_bundle, out_bundle)) {
- dst_set_add(set, &dst);
- ctx->nf_output_iface = dst.port->odp_port;
- }
-}
-
static bool
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])
+static void
+add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
{
- /* 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:
+ struct ofproto_dpif *ofproto = ctx->ofproto;
+ mirror_mask_t mirrors;
+ struct ofbundle *in_bundle;
+ uint16_t vlan;
+ uint16_t vid;
+ const struct nlattr *a;
+ size_t left;
- 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;
- }
+ in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
+ ctx->packet != NULL, NULL);
+ if (!in_bundle) {
+ return;
+ }
+ mirrors = in_bundle->src_mirrors;
- if (!(dst[3] | dst[4] | dst[5])) {
- /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
- return false;
- }
+ /* Drop frames on bundles reserved for mirroring. */
+ if (in_bundle->mirror_out) {
+ if (ctx->packet != NULL) {
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+ VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
+ "%s, which is reserved exclusively for mirroring",
+ ctx->ofproto->up.name, in_bundle->name);
}
+ return;
}
- return true;
-}
-static void
-compose_mirror_dsts(struct action_xlate_ctx *ctx,
- uint16_t vlan, const struct ofbundle *in_bundle,
- struct dst_set *set)
-{
- struct ofproto_dpif *ofproto = ctx->ofproto;
- mirror_mask_t mirrors;
- uint16_t flow_vid;
- size_t i;
+ /* Check VLAN. */
+ vid = vlan_tci_to_vid(orig_flow->vlan_tci);
+ if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
+ return;
+ }
+ vlan = input_vid_to_vlan(in_bundle, vid);
- mirrors = in_bundle->src_mirrors;
- for (i = 0; i < set->n; i++) {
- mirrors |= set->dsts[i].port->bundle->dst_mirrors;
+ /* Look at the output ports to check for destination selections. */
+
+ NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
+ ctx->odp_actions->size) {
+ enum ovs_action_attr type = nl_attr_type(a);
+ struct ofport_dpif *ofport;
+
+ if (type != OVS_ACTION_ATTR_OUTPUT) {
+ continue;
+ }
+
+ ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
+ if (ofport && ofport->bundle) {
+ mirrors |= ofport->bundle->dst_mirrors;
+ }
}
if (!mirrors) {
return;
}
- flow_vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
+ /* Restore the original packet before adding the mirror actions. */
+ ctx->flow = *orig_flow;
+
while (mirrors) {
- struct ofmirror *m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
- if (vlan_is_mirrored(m, vlan)) {
- struct dst dst;
-
- if (m->out) {
- if (set_dst(ctx, &dst, in_bundle, m->out)
- && !dst_is_duplicate(set, &dst)) {
- dst_set_add(set, &dst);
- }
- } else if (eth_dst_may_rspan(ctx->flow.dl_dst)) {
- struct ofbundle *bundle;
-
- HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
- if (ofbundle_includes_vlan(bundle, m->out_vlan)
- && set_dst(ctx, &dst, in_bundle, bundle))
- {
- /* set_dst() got dst->vid from the input packet's VLAN,
- * not from m->out_vlan, so recompute it. */
- dst.vid = output_vlan_to_vid(bundle, m->out_vlan);
-
- if (dst_is_duplicate(set, &dst)) {
- continue;
- }
+ struct ofmirror *m;
- if (bundle == in_bundle && dst.vid == flow_vid) {
- /* Don't send out input port on same VLAN. */
- continue;
- }
- dst_set_add(set, &dst);
- }
+ m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
+
+ if (!vlan_is_mirrored(m, vlan)) {
+ mirrors = zero_rightmost_1bit(mirrors);
+ continue;
+ }
+
+ mirrors &= ~m->dup_mirrors;
+ ctx->mirrors |= m->dup_mirrors;
+ if (m->out) {
+ output_normal(ctx, m->out, vlan);
+ } else if (vlan != m->out_vlan
+ && !eth_addr_is_reserved(orig_flow->dl_dst)) {
+ struct ofbundle *bundle;
+
+ HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
+ if (ofbundle_includes_vlan(bundle, m->out_vlan)
+ && !bundle->mirror_out) {
+ output_normal(ctx, bundle, m->out_vlan);
}
}
}
- mirrors &= mirrors - 1;
}
}
static void
-compose_actions(struct action_xlate_ctx *ctx, uint16_t vlan,
- const struct ofbundle *in_bundle,
- const struct ofbundle *out_bundle)
+update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
+ uint64_t packets, uint64_t bytes)
{
- uint16_t initial_vid, cur_vid;
- const struct dst *dst;
- struct dst_set set;
-
- dst_set_init(&set);
- compose_dsts(ctx, vlan, in_bundle, out_bundle, &set);
- compose_mirror_dsts(ctx, vlan, in_bundle, &set);
- if (!set.n) {
- dst_set_free(&set);
+ if (!mirrors) {
return;
}
- /* Output all the packets we can without having to change the VLAN. */
- commit_odp_actions(ctx);
- initial_vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
- for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
- if (dst->vid != initial_vid) {
- continue;
- }
- compose_output_action(ctx, dst->port->odp_port);
- }
-
- /* Then output the rest. */
- cur_vid = initial_vid;
- for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
- if (dst->vid == initial_vid) {
- continue;
- }
- if (dst->vid != cur_vid) {
- ovs_be16 tci;
+ for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
+ struct ofmirror *m;
- tci = htons(dst->vid);
- tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
- if (tci) {
- tci |= htons(VLAN_CFI);
- }
- commit_vlan_action(ctx, tci);
+ m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
- cur_vid = dst->vid;
+ if (!m) {
+ /* In normal circumstances 'm' will not be NULL. However,
+ * if mirrors are reconfigured, we can temporarily get out
+ * of sync in facet_revalidate(). We could "correct" the
+ * mirror list before reaching here, but doing that would
+ * not properly account the traffic stats we've currently
+ * accumulated for previous mirror configuration. */
+ continue;
}
- compose_output_action(ctx, dst->port->odp_port);
- }
-
- dst_set_free(&set);
-}
-/* Returns the effective vlan of a packet, taking into account both the
- * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
- * the packet is untagged and -1 indicates it has an invalid header and
- * should be dropped. */
-static int
-flow_get_vlan(struct ofproto_dpif *ofproto, const struct flow *flow,
- struct ofbundle *in_bundle, bool have_packet)
-{
- int vlan = vlan_tci_to_vid(flow->vlan_tci);
- if (vlan) {
- if (in_bundle->vlan_mode == PORT_VLAN_ACCESS) {
- /* Drop tagged packet on access port */
- if (have_packet) {
- static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
- VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
- "packet received on port %s configured with "
- "implicit VLAN %"PRIu16,
- ofproto->up.name, vlan,
- in_bundle->name, in_bundle->vlan);
- }
- return -1;
- } else if (ofbundle_includes_vlan(in_bundle, vlan)) {
- return vlan;
- } else {
- /* Drop packets from a VLAN not member of the trunk */
- if (have_packet) {
- static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
- VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
- "packet received on port %s not configured for "
- "trunking VLAN %d",
- ofproto->up.name, vlan, in_bundle->name, vlan);
- }
- return -1;
- }
- } else {
- if (in_bundle->vlan_mode != PORT_VLAN_TRUNK) {
- return in_bundle->vlan;
- } else {
- return ofbundle_includes_vlan(in_bundle, 0) ? 0 : -1;
- }
+ m->packet_count += packets;
+ m->byte_count += bytes;
}
}
{
struct mac_entry *mac;
+ /* Don't learn the OFPP_NONE port. */
+ if (in_bundle == &ofpp_none_bundle) {
+ return;
+ }
+
if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
return;
}
}
}
-/* Determines whether packets in 'flow' within 'br' should be forwarded or
+static struct ofbundle *
+lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
+ bool warn, struct ofport_dpif **in_ofportp)
+{
+ struct ofport_dpif *ofport;
+
+ /* Find the port and bundle for the received packet. */
+ ofport = get_ofp_port(ofproto, in_port);
+ if (in_ofportp) {
+ *in_ofportp = ofport;
+ }
+ if (ofport && ofport->bundle) {
+ return ofport->bundle;
+ }
+
+ /* Special-case OFPP_NONE, which a controller may use as the ingress
+ * port for traffic that it is sourcing. */
+ if (in_port == OFPP_NONE) {
+ return &ofpp_none_bundle;
+ }
+
+ /* Odd. A few possible reasons here:
+ *
+ * - We deleted a port but there are still a few packets queued up
+ * from it.
+ *
+ * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
+ * we don't know about.
+ *
+ * - The ofproto client didn't configure the port as part of a bundle.
+ * This is particularly likely to happen if a packet was received on the
+ * port after it was created, but before the client had a chance to
+ * configure its bundle.
+ */
+ if (warn) {
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+
+ VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
+ "port %"PRIu16, ofproto->up.name, in_port);
+ }
+ return NULL;
+}
+
+/* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
* dropped. Returns true if they may be forwarded, false if they should be
* dropped.
*
- * If 'have_packet' is true, it indicates that the caller is processing a
- * received packet. If 'have_packet' is false, then the caller is just
- * revalidating an existing flow because configuration has changed. Either
- * way, 'have_packet' only affects logging (there is no point in logging errors
- * during revalidation).
- *
- * Sets '*in_portp' to the input port. This will be a null pointer if
- * flow->in_port does not designate a known input port (in which case
- * is_admissible() returns false).
+ * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
+ * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
*
- * When returning true, sets '*vlanp' to the effective VLAN of the input
- * packet, as returned by flow_get_vlan().
+ * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
+ * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
+ * checked by input_vid_is_valid().
*
* May also add tags to '*tags', although the current implementation only does
* so in one special case.
*/
static bool
-is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
- bool have_packet,
- tag_type *tags, int *vlanp, struct ofbundle **in_bundlep)
+is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
+ uint16_t vlan)
{
- struct ofport_dpif *in_port;
- struct ofbundle *in_bundle;
- int vlan;
-
- /* Find the port and bundle for the received packet. */
- in_port = get_ofp_port(ofproto, flow->in_port);
- *in_bundlep = in_bundle = in_port ? in_port->bundle : NULL;
- if (!in_port || !in_bundle) {
- /* No interface? Something fishy... */
- if (have_packet) {
- /* Odd. A few possible reasons here:
- *
- * - We deleted a port but there are still a few packets queued up
- * from it.
- *
- * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
- * we don't know about.
- *
- * - Packet arrived on the local port but the local port is not
- * part of a bundle.
- */
- static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
-
- VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
- "port %"PRIu16,
- ofproto->up.name, flow->in_port);
- }
- *vlanp = -1;
- return false;
- }
- *vlanp = vlan = flow_get_vlan(ofproto, flow, in_bundle, have_packet);
- if (vlan < 0) {
- return false;
- }
-
- /* 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) {
- return false;
- }
+ struct ofproto_dpif *ofproto = ctx->ofproto;
+ struct flow *flow = &ctx->flow;
+ struct ofbundle *in_bundle = in_port->bundle;
- /* Drop frames on bundles reserved for mirroring. */
- if (in_bundle->mirror_out) {
- if (have_packet) {
- static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
- VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
- "%s, which is reserved exclusively for mirroring",
- ofproto->up.name, in_bundle->name);
- }
+ /* Drop frames for reserved multicast addresses
+ * only if forward_bpdu option is absent. */
+ if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
+ xlate_report(ctx, "packet has reserved destination MAC, dropping");
return false;
}
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:
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;
static void
xlate_normal(struct action_xlate_ctx *ctx)
{
+ struct ofport_dpif *in_port;
struct ofbundle *in_bundle;
- struct ofbundle *out_bundle;
struct mac_entry *mac;
- int vlan;
+ uint16_t vlan;
+ uint16_t vid;
ctx->has_normal = true;
- /* Check whether we should drop packets in this flow. */
- if (!is_admissible(ctx->ofproto, &ctx->flow, ctx->packet != NULL,
- &ctx->tags, &vlan, &in_bundle)) {
- out_bundle = NULL;
- goto done;
+ in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
+ ctx->packet != NULL, &in_port);
+ if (!in_bundle) {
+ xlate_report(ctx, "no input bundle, dropping");
+ return;
+ }
+
+ /* Drop malformed frames. */
+ if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
+ !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
+ if (ctx->packet != NULL) {
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+ VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
+ "VLAN tag received on port %s",
+ ctx->ofproto->up.name, in_bundle->name);
+ }
+ xlate_report(ctx, "partial VLAN tag, dropping");
+ return;
+ }
+
+ /* Drop frames on bundles reserved for mirroring. */
+ if (in_bundle->mirror_out) {
+ if (ctx->packet != NULL) {
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+ VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
+ "%s, which is reserved exclusively for mirroring",
+ ctx->ofproto->up.name, in_bundle->name);
+ }
+ xlate_report(ctx, "input port is mirror output port, dropping");
+ return;
+ }
+
+ /* 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, in_port, vlan)) {
+ return;
}
/* Learn source MAC. */
mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
&ctx->tags);
if (mac) {
- out_bundle = mac->port.p;
- } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) {
- /* If we are revalidating but don't have a learning entry then eject
- * the flow. Installing a flow that floods packets opens up a window
- * of time where we could learn from a packet reflected on a bond and
- * blackhole packets before the learning table is updated to reflect
- * the correct port. */
- ctx->may_set_up_flow = false;
- return;
+ 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 {
- out_bundle = OFBUNDLE_FLOOD;
- }
-
- /* Don't send packets out their input bundles. */
- if (in_bundle == out_bundle) {
- out_bundle = NULL;
- }
+ struct ofbundle *bundle;
-done:
- if (in_bundle) {
- compose_actions(ctx, vlan, in_bundle, out_bundle);
+ xlate_report(ctx, "no learned MAC for destination, flooding");
+ HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
+ if (bundle != in_bundle
+ && ofbundle_includes_vlan(bundle, vlan)
+ && bundle->floodable
+ && !bundle->mirror_out) {
+ output_normal(ctx, bundle, vlan);
+ }
+ }
+ ctx->nf_output_iface = NF_OUT_FLOOD;
}
}
\f
* 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 uint32_t
-rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
+static tag_type
+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);
}
}
table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
{
struct table_dpif *table = &ofproto->tables[table_id];
- const struct classifier *cls = &ofproto->up.tables[table_id];
+ const struct oftable *oftable = &ofproto->up.tables[table_id];
struct cls_table *catchall, *other;
struct cls_table *t;
catchall = other = NULL;
- switch (hmap_count(&cls->tables)) {
+ switch (hmap_count(&oftable->cls.tables)) {
case 0:
/* We could tag this OpenFlow table but it would make the logic a
* little harder and it's a corner case that doesn't seem worth it
case 1:
case 2:
- HMAP_FOR_EACH (t, hmap_node, &cls->tables) {
+ HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
if (cls_table_is_catchall(t)) {
catchall = t;
} else if (!other) {
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;
}
}
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;
}
}
}
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;
}
}
-static int
+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_);
- int error;
+ struct odputil_keybuf keybuf;
+ struct dpif_flow_stats stats;
- if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
- return ofp_mkerr_nicira(OFPET_BAD_REQUEST, 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 action_xlate_ctx ctx;
- struct ofpbuf *odp_actions;
- struct ofpbuf key;
+ struct action_xlate_ctx ctx;
+ uint64_t odp_actions_stub[1024 / 8];
+ struct ofpbuf odp_actions;
+
+ ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
+ odp_flow_key_from_flow(&key, flow,
+ ofp_port_to_odp_port(ofproto, flow->in_port));
+
+ dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
+
+ action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
+ packet_get_tcp_flags(packet, flow), packet);
+ ctx.resubmit_stats = &stats;
+
+ 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;
+}
+\f
+/* NetFlow. */
- ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
- odp_flow_key_from_flow(&key, flow);
+static int
+set_netflow(struct ofproto *ofproto_,
+ const struct netflow_options *netflow_options)
+{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
- action_xlate_ctx_init(&ctx, ofproto, flow, packet);
- odp_actions = xlate_actions(&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);
+ if (netflow_options) {
+ if (!ofproto->netflow) {
+ ofproto->netflow = netflow_create();
+ }
+ return netflow_set_options(ofproto->netflow, netflow_options);
+ } else {
+ netflow_destroy(ofproto->netflow);
+ ofproto->netflow = NULL;
+ return 0;
}
- return error;
}
static void
{
struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
- dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
-}
-\f
-static struct ofproto_dpif *
-ofproto_dpif_lookup(const char *name)
-{
- struct ofproto *ofproto = ofproto_lookup(name);
- return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class
- ? ofproto_dpif_cast(ofproto)
- : NULL);
+ dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
}
static void
-ofproto_unixctl_fdb_flush(struct unixctl_conn *conn,
- const char *args, void *aux OVS_UNUSED)
+send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
{
- const struct ofproto_dpif *ofproto;
+ if (!facet_is_controller_flow(facet) &&
+ netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
+ struct subfacet *subfacet;
+ struct ofexpired expired;
- ofproto = ofproto_dpif_lookup(args);
- if (!ofproto) {
- unixctl_command_reply(conn, 501, "no such bridge");
- return;
- }
- mac_learning_flush(ofproto->ml);
+ LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
+ if (subfacet->path == SF_FAST_PATH) {
+ struct dpif_flow_stats stats;
- unixctl_command_reply(conn, 200, "table successfully flushed");
-}
+ subfacet_reinstall(subfacet, &stats);
+ subfacet_update_stats(subfacet, &stats);
+ }
+ }
+
+ expired.flow = facet->flow;
+ expired.packet_count = facet->packet_count;
+ expired.byte_count = facet->byte_count;
+ expired.used = facet->used;
+ netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
+ }
+}
+
+static void
+send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
+{
+ struct facet *facet;
+
+ HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
+ send_active_timeout(ofproto, facet);
+ }
+}
+\f
+static struct ofproto_dpif *
+ofproto_dpif_lookup(const char *name)
+{
+ struct ofproto_dpif *ofproto;
+
+ HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
+ hash_string(name, 0), &all_ofproto_dpifs) {
+ if (!strcmp(ofproto->up.name, name)) {
+ return ofproto;
+ }
+ }
+ return NULL;
+}
+
+static void
+ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
+ const char *argv[], void *aux OVS_UNUSED)
+{
+ struct ofproto_dpif *ofproto;
+
+ if (argc > 1) {
+ ofproto = ofproto_dpif_lookup(argv[1]);
+ if (!ofproto) {
+ unixctl_command_reply_error(conn, "no such bridge");
+ return;
+ }
+ mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
+ } else {
+ HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
+ mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
+ }
+ }
+
+ unixctl_command_reply(conn, "table successfully flushed");
+}
static void
-ofproto_unixctl_fdb_show(struct unixctl_conn *conn,
- const char *args, void *aux OVS_UNUSED)
+ofproto_unixctl_fdb_show(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;
const struct mac_entry *e;
- ofproto = ofproto_dpif_lookup(args);
+ ofproto = ofproto_dpif_lookup(argv[1]);
if (!ofproto) {
- unixctl_command_reply(conn, 501, "no such bridge");
+ unixctl_command_reply_error(conn, "no such bridge");
return;
}
struct ofbundle *bundle = e->port.p;
ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
ofbundle_get_a_port(bundle)->odp_port,
- e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
+ e->vlan, ETH_ADDR_ARGS(e->mac),
+ mac_entry_age(ofproto->ml, e));
}
- unixctl_command_reply(conn, 200, ds_cstr(&ds));
+ unixctl_command_reply(conn, ds_cstr(&ds));
ds_destroy(&ds);
}
-struct ofproto_trace {
+struct trace_ctx {
struct action_xlate_ctx ctx;
struct flow flow;
struct ds *result;
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');
}
static void
trace_format_flow(struct ds *result, int level, const char *title,
- struct ofproto_trace *trace)
+ struct trace_ctx *trace)
{
ds_put_char_multiple(result, '\t', level);
ds_put_format(result, "%s: ", title);
static void
trace_format_regs(struct ds *result, int level, const char *title,
- struct ofproto_trace *trace)
+ struct trace_ctx *trace)
{
size_t i;
ds_put_char(result, '\n');
}
+static void
+trace_format_odp(struct ds *result, int level, const char *title,
+ struct trace_ctx *trace)
+{
+ struct ofpbuf *odp_actions = trace->ctx.odp_actions;
+
+ ds_put_char_multiple(result, '\t', level);
+ ds_put_format(result, "%s: ", title);
+ format_odp_actions(result, odp_actions->data, odp_actions->size);
+ ds_put_char(result, '\n');
+}
+
static void
trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
{
- struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
+ struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
struct ds *result = trace->result;
ds_put_char(result, '\n');
trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
+ trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
}
static void
-ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
+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)
{
- char *dpname, *arg1, *arg2, *arg3, *arg4;
- char *args = xstrdup(args_);
- char *save_ptr = NULL;
+ const char *dpname = argv[1];
struct ofproto_dpif *ofproto;
struct ofpbuf odp_key;
struct ofpbuf *packet;
- struct rule_dpif *rule;
+ ovs_be16 initial_tci;
struct ds result;
struct flow flow;
char *s;
ofpbuf_init(&odp_key, 0);
ds_init(&result);
- dpname = strtok_r(args, " ", &save_ptr);
- arg1 = strtok_r(NULL, " ", &save_ptr);
- arg2 = strtok_r(NULL, " ", &save_ptr);
- arg3 = strtok_r(NULL, " ", &save_ptr);
- arg4 = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
- if (dpname && arg1 && (!arg2 || !strcmp(arg2, "-generate")) && !arg3) {
+ ofproto = ofproto_dpif_lookup(dpname);
+ if (!ofproto) {
+ unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
+ "for help)");
+ goto exit;
+ }
+ if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
/* ofproto/trace dpname flow [-generate] */
- int error;
+ const char *flow_s = argv[2];
+ const char *generate_s = argv[3];
+
+ /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
+ * flow. We guess which type it is based on whether 'flow_s' contains
+ * an '(', since a datapath flow always contains '(') but an
+ * OpenFlow-like flow should not (in fact it's allowed but I believe
+ * that's not documented anywhere).
+ *
+ * An alternative would be to try to parse 'flow_s' both ways, but then
+ * it would be tricky giving a sensible error message. After all, do
+ * you just say "syntax error" or do you present both error messages?
+ * Both choices seem lousy. */
+ if (strchr(flow_s, '(')) {
+ enum odp_key_fitness fitness;
+ int error;
+
+ /* Convert string to datapath key. */
+ ofpbuf_init(&odp_key, 0);
+ error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
+ if (error) {
+ unixctl_command_reply_error(conn, "Bad flow syntax");
+ goto exit;
+ }
- /* Convert string to datapath key. */
- ofpbuf_init(&odp_key, 0);
- error = odp_flow_key_from_string(arg1, &odp_key);
- if (error) {
- unixctl_command_reply(conn, 501, "Bad flow syntax");
- 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 = odp_flow_key_to_flow(odp_key.data, odp_key.size, &flow);
- if (error) {
- unixctl_command_reply(conn, 501, "Invalid flow");
- goto exit;
+ /* 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. */
- if (arg2) {
+ if (generate_s) {
packet = ofpbuf_new(0);
flow_compose(packet, &flow);
}
- } else if (dpname && arg1 && arg2 && arg3 && arg4) {
- /* ofproto/trace dpname priority tun_id in_port packet */
- uint16_t in_port;
- ovs_be64 tun_id;
- uint32_t priority;
-
- priority = atoi(arg1);
- tun_id = htonll(strtoull(arg2, NULL, 0));
- in_port = ofp_port_to_odp_port(atoi(arg3));
-
- packet = ofpbuf_new(strlen(args) / 2);
- arg4 = ofpbuf_put_hex(packet, arg4, NULL);
- arg4 += strspn(arg4, " ");
- if (*arg4 != '\0') {
- unixctl_command_reply(conn, 501, "Trailing garbage in command");
- goto exit;
- }
- if (packet->size < ETH_HEADER_LEN) {
- unixctl_command_reply(conn, 501,
- "Packet data too short for Ethernet");
+ } else if (argc == 7) {
+ /* ofproto/trace dpname priority tun_id in_port mark packet */
+ const char *priority_s = argv[2];
+ const char *tun_id_s = argv[3];
+ const char *in_port_s = argv[4];
+ const char *mark_s = argv[5];
+ const char *packet_s = argv[6];
+ uint32_t in_port = atoi(in_port_s);
+ ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
+ uint32_t priority = atoi(priority_s);
+ uint32_t mark = atoi(mark_s);
+ const char *msg;
+
+ msg = eth_from_hex(packet_s, &packet);
+ if (msg) {
+ unixctl_command_reply_error(conn, msg);
goto exit;
}
ds_put_cstr(&result, "Packet: ");
- s = ofp_packet_to_string(packet->data, packet->size, packet->size);
+ s = ofp_packet_to_string(packet->data, packet->size);
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(conn, 501, "Bad command syntax");
+ unixctl_command_reply_error(conn, "Bad command syntax");
goto exit;
}
- ofproto = ofproto_dpif_lookup(dpname);
- if (!ofproto) {
- unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
- "for help)");
- goto exit;
- }
+ ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
+ unixctl_command_reply(conn, ds_cstr(&result));
- ds_put_cstr(&result, "Flow: ");
- flow_format(&result, &flow);
- ds_put_char(&result, '\n');
+exit:
+ ds_destroy(&result);
+ ofpbuf_delete(packet);
+ ofpbuf_uninit(&odp_key);
+}
- rule = rule_dpif_lookup(ofproto, &flow, 0);
- trace_format_rule(&result, 0, 0, rule);
- if (rule) {
- struct ofproto_trace trace;
- struct ofpbuf *odp_actions;
+static void
+ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
+ const struct ofpbuf *packet, ovs_be16 initial_tci,
+ struct ds *ds)
+{
+ struct rule_dpif *rule;
+
+ ds_put_cstr(ds, "Flow: ");
+ flow_format(ds, flow);
+ ds_put_char(ds, '\n');
+
+ rule = rule_dpif_lookup(ofproto, flow);
- trace.result = &result;
- trace.flow = flow;
- action_xlate_ctx_init(&trace.ctx, ofproto, &flow, packet);
+ 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;
+ 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);
-
- ds_put_char(&result, '\n');
- trace_format_flow(&result, 0, "Final flow", &trace);
- ds_put_cstr(&result, "Datapath actions: ");
- format_odp_actions(&result, odp_actions->data, odp_actions->size);
- ofpbuf_delete(odp_actions);
-
- if (!trace.ctx.may_set_up_flow) {
- if (packet) {
- ds_put_cstr(&result, "\nThis flow is not cachable.");
- } else {
- ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
- "for complete actions, please supply a packet.");
+ 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_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;
+ }
+
+ slow &= ~bit;
+ }
+
+ if (slow & ~SLOW_MATCH) {
+ ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
+ "the special slow-path processing.");
}
}
}
-
- unixctl_command_reply(conn, 200, ds_cstr(&result));
-
-exit:
- ds_destroy(&result);
- ofpbuf_delete(packet);
- ofpbuf_uninit(&odp_key);
- free(args);
}
static void
-ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED,
- const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
+ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
+ const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
{
clogged = true;
- unixctl_command_reply(conn, 200, NULL);
+ unixctl_command_reply(conn, NULL);
}
static void
-ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED,
- const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
+ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
+ const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
{
clogged = false;
- unixctl_command_reply(conn, 200, NULL);
+ unixctl_command_reply(conn, NULL);
+}
+
+/* Runs a self-check of flow translations in 'ofproto'. Appends a message to
+ * 'reply' describing the results. */
+static void
+ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
+{
+ struct facet *facet;
+ int errors;
+
+ errors = 0;
+ HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
+ if (!facet_check_consistency(facet)) {
+ errors++;
+ }
+ }
+ if (errors) {
+ ofproto->need_revalidate = REV_INCONSISTENCY;
+ }
+
+ if (errors) {
+ ds_put_format(reply, "%s: self-check failed (%d errors)\n",
+ ofproto->up.name, errors);
+ } else {
+ ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
+ }
+}
+
+static void
+ofproto_dpif_self_check(struct unixctl_conn *conn,
+ int argc, const char *argv[], void *aux OVS_UNUSED)
+{
+ struct ds reply = DS_EMPTY_INITIALIZER;
+ struct ofproto_dpif *ofproto;
+
+ if (argc > 1) {
+ ofproto = ofproto_dpif_lookup(argv[1]);
+ if (!ofproto) {
+ unixctl_command_reply_error(conn, "Unknown ofproto (use "
+ "ofproto/list for help)");
+ return;
+ }
+ ofproto_dpif_self_check__(ofproto, &reply);
+ } else {
+ HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
+ ofproto_dpif_self_check__(ofproto, &reply);
+ }
+ }
+
+ unixctl_command_reply(conn, ds_cstr(&reply));
+ 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
}
registered = true;
- unixctl_command_register("ofproto/trace",
- "bridge {tun_id in_port packet | odp_flow [-generate]}",
- ofproto_unixctl_trace, NULL);
- unixctl_command_register("fdb/flush", "bridge", ofproto_unixctl_fdb_flush,
- NULL);
- unixctl_command_register("fdb/show", "bridge", ofproto_unixctl_fdb_show,
- NULL);
- unixctl_command_register("ofproto/clog", "", ofproto_dpif_clog, NULL);
- unixctl_command_register("ofproto/unclog", "", ofproto_dpif_unclog, NULL);
+ unixctl_command_register(
+ "ofproto/trace",
+ "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,
+ ofproto_unixctl_fdb_show, NULL);
+ unixctl_command_register("ofproto/clog", "", 0, 0,
+ ofproto_dpif_clog, NULL);
+ unixctl_command_register("ofproto/unclog", "", 0, 0,
+ ofproto_dpif_unclog, NULL);
+ unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
+ ofproto_dpif_self_check, NULL);
+ unixctl_command_register("dpif/dump-dps", "", 0, 0,
+ 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);
}
\f
+/* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
+ *
+ * This is deprecated. It is only for compatibility with broken device drivers
+ * in old versions of Linux that do not properly support VLANs when VLAN
+ * devices are not used. When broken device drivers are no longer in
+ * widespread use, we will delete these interfaces. */
+
+static int
+set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
+{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
+ struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
+
+ if (realdev_ofp_port == ofport->realdev_ofp_port
+ && vid == ofport->vlandev_vid) {
+ return 0;
+ }
+
+ ofproto->need_revalidate = REV_RECONFIGURE;
+
+ if (ofport->realdev_ofp_port) {
+ vsp_remove(ofport);
+ }
+ if (realdev_ofp_port && ofport->bundle) {
+ /* vlandevs are enslaved to their realdevs, so they are not allowed to
+ * themselves be part of a bundle. */
+ bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
+ }
+
+ ofport->realdev_ofp_port = realdev_ofp_port;
+ ofport->vlandev_vid = vid;
+
+ if (realdev_ofp_port) {
+ vsp_add(ofport, realdev_ofp_port, vid);
+ }
+
+ return 0;
+}
+
+static uint32_t
+hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
+{
+ return hash_2words(realdev_ofp_port, vid);
+}
+
+/* Returns the ODP port number of the Linux VLAN device that corresponds to
+ * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
+ * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
+ * it would return the port number of eth0.9.
+ *
+ * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
+ * function just returns its 'realdev_odp_port' argument. */
+static uint32_t
+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;
+ int vid = vlan_tci_to_vid(vlan_tci);
+ const struct vlan_splinter *vsp;
+
+ realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
+ HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
+ hash_realdev_vid(realdev_ofp_port, vid),
+ &ofproto->realdev_vid_map) {
+ if (vsp->realdev_ofp_port == realdev_ofp_port
+ && vsp->vid == vid) {
+ return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
+ }
+ }
+ }
+ return realdev_odp_port;
+}
+
+static struct vlan_splinter *
+vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
+{
+ struct vlan_splinter *vsp;
+
+ HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
+ &ofproto->vlandev_map) {
+ if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
+ return vsp;
+ }
+ }
+
+ return NULL;
+}
+
+/* Returns the OpenFlow port number of the "real" device underlying the Linux
+ * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
+ * VLAN VID of the Linux VLAN device in '*vid'. For example, given
+ * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
+ * eth0 and store 9 in '*vid'.
+ *
+ * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
+ * VLAN device. Unless VLAN splinters are enabled, this is what this function
+ * always does.*/
+static uint16_t
+vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
+ uint16_t vlandev_ofp_port, int *vid)
+{
+ if (!hmap_is_empty(&ofproto->vlandev_map)) {
+ const struct vlan_splinter *vsp;
+
+ vsp = vlandev_find(ofproto, vlandev_ofp_port);
+ if (vsp) {
+ if (vid) {
+ *vid = vsp->vid;
+ }
+ return vsp->realdev_ofp_port;
+ }
+ }
+ 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)
+{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
+ struct vlan_splinter *vsp;
+
+ vsp = vlandev_find(ofproto, port->up.ofp_port);
+ if (vsp) {
+ hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
+ hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
+ free(vsp);
+
+ port->realdev_ofp_port = 0;
+ } else {
+ VLOG_ERR("missing vlan device record");
+ }
+}
+
+static void
+vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
+{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
+
+ if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
+ && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
+ == realdev_ofp_port)) {
+ struct vlan_splinter *vsp;
+
+ vsp = xmalloc(sizeof *vsp);
+ hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
+ hash_int(port->up.ofp_port, 0));
+ hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
+ hash_realdev_vid(realdev_ofp_port, vid));
+ vsp->realdev_ofp_port = realdev_ofp_port;
+ vsp->vlandev_ofp_port = port->up.ofp_port;
+ vsp->vid = vid;
+
+ port->realdev_ofp_port = realdev_ofp_port;
+ } else {
+ 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,
dealloc,
run,
+ run_fast,
wait,
+ get_memory_usage,
flush,
get_features,
get_tables,
port_query_by_name,
port_add,
port_del,
+ port_get_stats,
port_dump_start,
port_dump_next,
port_dump_done,
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,
get_stp_port_status,
+ set_queues,
bundle_set,
bundle_remove,
mirror_set,
+ mirror_get_stats,
set_flood_vlans,
is_mirror_output_bundle,
forward_bpdu_changed,
+ set_mac_idle_time,
+ set_realdev,
};
projects / openvswitch / blobdiff