/*
- * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
+ * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 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 "bond.h"
#include "bundle.h"
#include "byte-order.h"
+#include "connectivity.h"
#include "connmgr.h"
#include "coverage.h"
#include "cfm.h"
#include "dpif.h"
#include "dynamic-string.h"
#include "fail-open.h"
+#include "guarded-list.h"
#include "hmapx.h"
#include "lacp.h"
#include "learn.h"
#include "ofp-actions.h"
#include "ofp-parse.h"
#include "ofp-print.h"
-#include "ofproto-dpif-governor.h"
#include "ofproto-dpif-ipfix.h"
+#include "ofproto-dpif-mirror.h"
+#include "ofproto-dpif-monitor.h"
+#include "ofproto-dpif-rid.h"
#include "ofproto-dpif-sflow.h"
+#include "ofproto-dpif-upcall.h"
#include "ofproto-dpif-xlate.h"
#include "poll-loop.h"
+#include "seq.h"
#include "simap.h"
#include "smap.h"
#include "timer.h"
VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
COVERAGE_DEFINE(ofproto_dpif_expired);
-COVERAGE_DEFINE(facet_changed_rule);
-COVERAGE_DEFINE(facet_revalidate);
-COVERAGE_DEFINE(facet_unexpected);
-COVERAGE_DEFINE(facet_suppress);
-COVERAGE_DEFINE(subfacet_install_fail);
+COVERAGE_DEFINE(packet_in_overflow);
+
+/* Number of implemented OpenFlow tables. */
+enum { N_TABLES = 255 };
+enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
+BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
+
+/* No bfd/cfm status change. */
+#define NO_STATUS_CHANGE -1
struct flow_miss;
-struct facet;
-static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
- const struct flow *,
- struct flow_wildcards *wc);
+struct rule_dpif {
+ struct rule up;
+
+ /* These statistics:
+ *
+ * - Do include packets and bytes from datapath flows which have not
+ * recently been processed by a revalidator. */
+ struct ovs_mutex stats_mutex;
+ struct dpif_flow_stats stats OVS_GUARDED;
+};
+
+static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes,
+ long long int *used);
+static struct rule_dpif *rule_dpif_cast(const struct rule *);
+static void rule_expire(struct rule_dpif *);
+
+struct group_dpif {
+ struct ofgroup up;
-static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
-static void rule_invalidate(const struct rule_dpif *);
+ /* These statistics:
+ *
+ * - Do include packets and bytes from datapath flows which have not
+ * recently been processed by a revalidator. */
+ struct ovs_mutex stats_mutex;
+ uint64_t packet_count OVS_GUARDED; /* Number of packets received. */
+ uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */
+ struct bucket_counter *bucket_stats OVS_GUARDED; /* Bucket statistics. */
+};
-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);
+struct ofbundle {
+ 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. */
+
+ /* Configuration. */
+ struct list ports; /* Contains "struct ofport"s. */
+ enum port_vlan_mode vlan_mode; /* VLAN mode */
+ int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
+ unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
+ * 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 OFPUTIL_PC_NO_FLOOD set. */
+};
static void bundle_remove(struct ofport *);
static void bundle_update(struct ofbundle *);
static int set_stp_port(struct ofport *,
const struct ofproto_port_stp_settings *);
-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);
-
-/* 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. */
-};
-
-/* 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. */
- struct dpif_backer *backer; /* Owning backer. */
-
- enum odp_key_fitness key_fitness;
- struct nlattr *key;
- int key_len;
-
- long long int used; /* Time last used; time created if not used. */
- long long int created; /* Time created. */
-
- uint64_t dp_packet_count; /* Last known packet count in the datapath. */
- uint64_t dp_byte_count; /* Last known byte count in the datapath. */
+struct ofport_dpif {
+ struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
+ struct ofport up;
- enum subfacet_path path; /* Installed in datapath? */
-};
-
-#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 dpif_backer *,
- const struct nlattr *key, size_t key_len,
- uint32_t key_hash);
-static void subfacet_destroy(struct subfacet *);
-static void subfacet_destroy__(struct subfacet *);
-static void subfacet_destroy_batch(struct dpif_backer *,
- struct subfacet **, int n);
-static void subfacet_reset_dp_stats(struct subfacet *,
- struct dpif_flow_stats *);
-static void subfacet_update_stats(struct subfacet *,
- const struct dpif_flow_stats *);
-static int subfacet_install(struct subfacet *,
- const struct ofpbuf *odp_actions,
- struct dpif_flow_stats *);
-static void subfacet_uninstall(struct subfacet *);
-
-/* A unique, non-overlapping instantiation of an OpenFlow flow.
- *
- * A facet associates a "struct flow", which represents the Open vSwitch
- * userspace idea of an exact-match flow, with one or more subfacets.
- * While the facet is created based on an exact-match flow, it is stored
- * within the ofproto based on the wildcards that could be expressed
- * based on the flow table and other configuration. (See the 'wc'
- * description in "struct xlate_out" for more details.)
- *
- * Each subfacet tracks the datapath's idea of the flow equivalent to
- * the facet. When the kernel module (or other dpif implementation) and
- * Open vSwitch userspace agree on the definition of a flow key, there
- * is exactly one subfacet per facet. If the dpif implementation
- * supports more-specific flow matching than userspace, however, a facet
- * can have more than one subfacet. Examples include the dpif
- * implementation not supporting the same wildcards as userspace or some
- * distinction in flow that userspace simply doesn't understand.
- *
- * Flow expiration works in terms of subfacets, so a facet must have at
- * least one subfacet or it will never expire, leaking memory. */
-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; /* Flow of the creating subfacet. */
- struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
-
- /* 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 a subfacet's statistics were reset (e.g. dpif_flow_put() with
- * DPIF_FP_ZERO_STATS).
+ odp_port_t odp_port;
+ struct ofbundle *bundle; /* Bundle that contains this port, if any. */
+ struct list bundle_node; /* In struct ofbundle's "ports" list. */
+ struct cfm *cfm; /* Connectivity Fault Management, if any. */
+ struct bfd *bfd; /* BFD, if any. */
+ bool may_enable; /* May be enabled in bonds. */
+ bool is_tunnel; /* This port is a tunnel. */
+ bool is_layer3; /* This is a layer 3 port. */
+ long long int carrier_seq; /* Carrier status changes. */
+ struct ofport_dpif *peer; /* Peer if patch port. */
+
+ /* Spanning tree. */
+ struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
+ enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
+ long long int stp_state_entered;
+
+ /* Queue to DSCP mapping. */
+ struct ofproto_port_queue *qdscp;
+ size_t n_qdscp;
+
+ /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
*
- * - 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. */
-
- /* 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 xlate_out xout;
-
- /* 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. However, 'one_subfacet' may not
- * always be valid, since it could have been removed after newer
- * subfacets were pushed onto the 'subfacets' list.) */
- struct subfacet one_subfacet;
-
- long long int learn_rl; /* Rate limiter for facet_learn(). */
-};
-
-static struct facet *facet_create(const struct flow_miss *, struct rule_dpif *,
- struct xlate_out *,
- struct dpif_flow_stats *);
-static void facet_remove(struct facet *);
-static void facet_free(struct facet *);
-
-static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
-static struct facet *facet_lookup_valid(struct ofproto_dpif *,
- const struct flow *);
-static bool facet_revalidate(struct facet *);
-static bool facet_check_consistency(struct facet *);
-
-static void facet_flush_stats(struct facet *);
-
-static void facet_reset_counters(struct facet *);
-static void facet_push_stats(struct facet *, bool may_learn);
-static void facet_learn(struct facet *);
-static void facet_account(struct facet *);
-static void push_all_stats(void);
-
-static bool facet_is_controller_flow(struct facet *);
-
-/* 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. */
+ * 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. */
+ ofp_port_t realdev_ofp_port;
+ int vlandev_vid;
};
/* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
int vid;
};
-static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
static void vsp_remove(struct ofport_dpif *);
static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
+static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
+ ofp_port_t);
+
static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
- odp_port_t odp_port);
+ odp_port_t);
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_bfd(struct ofport *, const struct smap *);
static int set_cfm(struct ofport *, const struct cfm_settings *);
-static void ofport_clear_priorities(struct ofport_dpif *);
static void ofport_update_peer(struct ofport_dpif *);
-static void run_fast_rl(void);
struct dpif_completion {
struct list list_node;
struct ofoperation *op;
};
-/* Reasons that we might need to revalidate every facet, and corresponding
- * coverage counters.
+/* Reasons that we might need to revalidate every datapath flow, and
+ * corresponding coverage counters.
*
* A value of 0 means that there is no need to revalidate.
*
enum revalidate_reason {
REV_RECONFIGURE = 1, /* Switch configuration changed. */
REV_STP, /* Spanning tree protocol port status change. */
+ REV_BOND, /* Bonding changed. */
REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
REV_FLOW_TABLE, /* Flow table changed. */
- REV_INCONSISTENCY /* Facet self-check failed. */
+ REV_MAC_LEARNING, /* Mac learning changed. */
};
COVERAGE_DEFINE(rev_reconfigure);
COVERAGE_DEFINE(rev_stp);
+COVERAGE_DEFINE(rev_bond);
COVERAGE_DEFINE(rev_port_toggled);
COVERAGE_DEFINE(rev_flow_table);
-COVERAGE_DEFINE(rev_inconsistency);
-
-/* Drop keys are odp flow keys which have drop flows installed in the kernel.
- * These are datapath flows which have no associated ofproto, if they did we
- * would use facets. */
-struct drop_key {
- struct hmap_node hmap_node;
- struct nlattr *key;
- size_t key_len;
-};
-
-struct avg_subfacet_rates {
- double add_rate; /* Moving average of new flows created per minute. */
- double del_rate; /* Moving average of flows deleted per minute. */
-};
+COVERAGE_DEFINE(rev_mac_learning);
/* All datapaths of a given type share a single dpif backer instance. */
struct dpif_backer {
char *type;
int refcount;
struct dpif *dpif;
- struct timer next_expiration;
- struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
+ struct udpif *udpif;
+
+ struct ovs_rwlock odp_to_ofport_lock;
+ struct hmap odp_to_ofport_map OVS_GUARDED; /* Contains "struct ofport"s. */
struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
- /* Facet revalidation flags applying to facets which use this backer. */
- enum revalidate_reason need_revalidate; /* Revalidate every facet. */
- struct tag_set revalidate_set; /* Revalidate only matching facets. */
+ enum revalidate_reason need_revalidate; /* Revalidate all flows. */
- struct hmap drop_keys; /* Set of dropped odp keys. */
bool recv_set_enable; /* Enables or disables receiving packets. */
- struct hmap subfacets;
- struct governor *governor;
+ /* Recirculation. */
+ struct recirc_id_pool *rid_pool; /* Recirculation ID pool. */
+ bool enable_recirc; /* True if the datapath supports recirculation */
- /* Subfacet statistics.
- *
- * These keep track of the total number of subfacets added and deleted and
- * flow life span. They are useful for computing the flow rates stats
- * exposed via "ovs-appctl dpif/show". The goal is to learn about
- * traffic patterns in ways that we can use later to improve Open vSwitch
- * performance in new situations. */
- long long int created; /* Time when it is created. */
- unsigned max_n_subfacet; /* Maximum number of flows */
- unsigned avg_n_subfacet; /* Average number of flows. */
- long long int avg_subfacet_life; /* Average life span of subfacets. */
-
- /* The average number of subfacets... */
- struct avg_subfacet_rates hourly; /* ...over the last hour. */
- struct avg_subfacet_rates daily; /* ...over the last day. */
- struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
- long long int last_minute; /* Last time 'hourly' was updated. */
-
- /* Number of subfacets added or deleted since 'last_minute'. */
- unsigned subfacet_add_count;
- unsigned subfacet_del_count;
-
- /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
- unsigned long long int total_subfacet_add_count;
- unsigned long long int total_subfacet_del_count;
+ /* True if the datapath supports variable-length
+ * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.
+ * False if the datapath supports only 8-byte (or shorter) userdata. */
+ bool variable_length_userdata;
+
+ /* Maximum number of MPLS label stack entries that the datapath supports
+ * in a match */
+ size_t max_mpls_depth;
};
/* All existing ofproto_backer instances, indexed by ofproto->up.type. */
static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
-static void drop_key_clear(struct dpif_backer *);
-static struct ofport_dpif *
-odp_port_to_ofport(const struct dpif_backer *, odp_port_t odp_port);
-static void update_moving_averages(struct dpif_backer *backer);
-
-/* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
- * for debugging the asynchronous flow_mod implementation.) */
-static bool clogged;
+struct ofproto_dpif {
+ struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
+ struct ofproto up;
+ struct dpif_backer *backer;
-/* By default, flows in the datapath are wildcarded (megaflows). They
- * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
-static bool enable_megaflows = true;
+ uint64_t dump_seq; /* Last read of udpif_dump_seq(). */
+
+ /* 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. */
+ struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
+
+ /* Bridging. */
+ struct netflow *netflow;
+ struct dpif_sflow *sflow;
+ struct dpif_ipfix *ipfix;
+ struct hmap bundles; /* Contains "struct ofbundle"s. */
+ struct mac_learning *ml;
+ bool has_bonded_bundles;
+ bool lacp_enabled;
+ struct mbridge *mbridge;
+
+ struct ovs_mutex stats_mutex;
+ struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
+ * consumed in userspace. */
+
+ /* Spanning tree. */
+ struct stp *stp;
+ long long int stp_last_tick;
+
+ /* VLAN splinters. */
+ struct ovs_mutex vsp_mutex;
+ struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
+ struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */
+
+ /* Ports. */
+ struct sset ports; /* Set of standard port names. */
+ struct sset ghost_ports; /* Ports with no datapath port. */
+ struct sset port_poll_set; /* Queued names for port_poll() reply. */
+ int port_poll_errno; /* Last errno for port_poll() reply. */
+ uint64_t change_seq; /* Connectivity status changes. */
+
+ /* Work queues. */
+ struct guarded_list pins; /* Contains "struct ofputil_packet_in"s. */
+ struct seq *pins_seq; /* For notifying 'pins' reception. */
+ uint64_t pins_seqno;
+};
/* 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);
-/* Upcalls. */
-#define FLOW_MISS_MAX_BATCH 50
-static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
+static inline struct ofproto_dpif *
+ofproto_dpif_cast(const struct ofproto *ofproto)
+{
+ ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
+ return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
+}
-/* Flow expiration. */
-static int expire(struct dpif_backer *);
+size_t
+ofproto_dpif_get_max_mpls_depth(const struct ofproto_dpif *ofproto)
+{
+ return ofproto->backer->max_mpls_depth;
+}
-/* NetFlow. */
-static void send_netflow_active_timeouts(struct ofproto_dpif *);
+bool
+ofproto_dpif_get_enable_recirc(const struct ofproto_dpif *ofproto)
+{
+ return ofproto->backer->enable_recirc;
+}
-/* Utilities. */
-static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
+static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
+ ofp_port_t ofp_port);
+static void ofproto_trace(struct ofproto_dpif *, struct flow *,
+ const struct ofpbuf *packet,
+ const struct ofpact[], size_t ofpacts_len,
+ struct ds *);
/* 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);
+
+/* Executes 'fm'. The caller retains ownership of 'fm' and everything in
+ * it. */
+void
+ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
+ struct ofputil_flow_mod *fm)
+{
+ ofproto_flow_mod(&ofproto->up, fm);
+}
+
+/* Resets the modified time for 'rule' or an equivalent rule. If 'rule' is not
+ * in the classifier, but an equivalent rule is, unref 'rule' and ref the new
+ * rule. Otherwise if 'rule' is no longer installed in the classifier,
+ * reinstall it.
+ *
+ * Returns the rule whose modified time has been reset. */
+struct rule_dpif *
+ofproto_dpif_refresh_rule(struct rule_dpif *rule)
+{
+ return rule_dpif_cast(ofproto_refresh_rule(&rule->up));
+}
+
+/* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
+ * Takes ownership of 'pin' and pin->packet. */
+void
+ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
+ struct ofproto_packet_in *pin)
+{
+ if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
+ COVERAGE_INC(packet_in_overflow);
+ free(CONST_CAST(void *, pin->up.packet));
+ free(pin);
+ }
+
+ /* Wakes up main thread for packet-in I/O. */
+ seq_change(ofproto->pins_seq);
+}
+
+/* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
+ * packet rather than to send the packet to the controller.
+ *
+ * This function returns false to indicate that a packet_in message
+ * for a "table-miss" should be sent to at least one controller.
+ * False otherwise. */
+bool
+ofproto_dpif_wants_packet_in_on_miss(struct ofproto_dpif *ofproto)
+{
+ return connmgr_wants_packet_in_on_miss(ofproto->up.connmgr);
+}
\f
/* Factory functions. */
/* Type functions. */
+static void process_dpif_port_changes(struct dpif_backer *);
+static void process_dpif_all_ports_changed(struct dpif_backer *);
+static void process_dpif_port_change(struct dpif_backer *,
+ const char *devname);
+static void process_dpif_port_error(struct dpif_backer *, int error);
+
static struct ofproto_dpif *
lookup_ofproto_dpif_by_port_name(const char *name)
{
static int
type_run(const char *type)
{
- static long long int push_timer = LLONG_MIN;
struct dpif_backer *backer;
- char *devname;
- int error;
backer = shash_find_data(&all_dpif_backers, type);
if (!backer) {
dpif_run(backer->dpif);
- /* The most natural place to push facet statistics is when they're pulled
- * from the datapath. However, when there are many flows in the datapath,
- * this expensive operation can occur so frequently, that it reduces our
- * ability to quickly set up flows. To reduce the cost, we push statistics
- * here instead. */
- if (time_msec() > push_timer) {
- push_timer = time_msec() + 2000;
- push_all_stats();
- }
-
/* If vswitchd started with other_config:flow_restore_wait set as "true",
* and the configuration has now changed to "false", enable receiving
* packets from the datapath. */
if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
+ int error;
+
backer->recv_set_enable = true;
error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
backer->need_revalidate = REV_RECONFIGURE;
}
- if (backer->need_revalidate
- || !tag_set_is_empty(&backer->revalidate_set)) {
- struct tag_set revalidate_set = backer->revalidate_set;
- bool need_revalidate = backer->need_revalidate;
+ if (backer->recv_set_enable) {
+ udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
+ }
+
+ if (backer->need_revalidate) {
struct ofproto_dpif *ofproto;
struct simap_node *node;
struct simap tmp_backers;
switch (backer->need_revalidate) {
case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
case REV_STP: COVERAGE_INC(rev_stp); break;
+ case REV_BOND: COVERAGE_INC(rev_bond); break;
case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
- case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
+ case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
}
-
- if (backer->need_revalidate) {
- /* Clear the drop_keys in case we should now be accepting some
- * formerly dropped flows. */
- drop_key_clear(backer);
- }
-
- /* Clear the revalidation flags. */
- tag_set_init(&backer->revalidate_set);
backer->need_revalidate = 0;
HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
- struct facet *facet, *next;
- struct cls_cursor cursor;
+ struct ofport_dpif *ofport;
+ struct ofbundle *bundle;
if (ofproto->backer != backer) {
continue;
}
- cls_cursor_init(&cursor, &ofproto->facets, NULL);
- CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
- if (need_revalidate
- || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
- facet_revalidate(facet);
- run_fast_rl();
- }
+ ovs_rwlock_wrlock(&xlate_rwlock);
+ xlate_ofproto_set(ofproto, ofproto->up.name,
+ ofproto->backer->dpif, ofproto->miss_rule,
+ ofproto->no_packet_in_rule, ofproto->ml,
+ ofproto->stp, ofproto->mbridge,
+ ofproto->sflow, ofproto->ipfix,
+ ofproto->netflow, ofproto->up.frag_handling,
+ ofproto->up.forward_bpdu,
+ connmgr_has_in_band(ofproto->up.connmgr),
+ ofproto->backer->enable_recirc,
+ ofproto->backer->variable_length_userdata,
+ ofproto->backer->max_mpls_depth);
+
+ HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
+ xlate_bundle_set(ofproto, bundle, bundle->name,
+ bundle->vlan_mode, bundle->vlan,
+ bundle->trunks, bundle->use_priority_tags,
+ bundle->bond, bundle->lacp,
+ bundle->floodable);
}
+
+ HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
+ int stp_port = ofport->stp_port
+ ? stp_port_no(ofport->stp_port)
+ : -1;
+ xlate_ofport_set(ofproto, ofport->bundle, ofport,
+ ofport->up.ofp_port, ofport->odp_port,
+ ofport->up.netdev, ofport->cfm,
+ ofport->bfd, ofport->peer, stp_port,
+ ofport->qdscp, ofport->n_qdscp,
+ ofport->up.pp.config, ofport->up.pp.state,
+ ofport->is_tunnel, ofport->may_enable);
+ }
+ ovs_rwlock_unlock(&xlate_rwlock);
}
- }
- if (!backer->recv_set_enable) {
- /* Wake up before a max of 1000ms. */
- timer_set_duration(&backer->next_expiration, 1000);
- } else if (timer_expired(&backer->next_expiration)) {
- int delay = expire(backer);
- timer_set_duration(&backer->next_expiration, delay);
+ udpif_revalidate(backer->udpif);
}
- /* Check for port changes in the dpif. */
- while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
- struct ofproto_dpif *ofproto;
- struct dpif_port port;
+ process_dpif_port_changes(backer);
- /* Don't report on the datapath's device. */
- if (!strcmp(devname, dpif_base_name(backer->dpif))) {
- goto next;
- }
+ return 0;
+}
- HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
- &all_ofproto_dpifs) {
- if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
- goto next;
- }
- }
+/* Check for and handle port changes in 'backer''s dpif. */
+static void
+process_dpif_port_changes(struct dpif_backer *backer)
+{
+ for (;;) {
+ char *devname;
+ int error;
- ofproto = lookup_ofproto_dpif_by_port_name(devname);
- if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
- /* The port was removed. If we know the datapath,
- * report it through poll_set(). If we don't, it may be
- * notifying us of a removal we initiated, so ignore it.
- * If there's a pending ENOBUFS, let it stand, since
- * everything will be reevaluated. */
- if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
- sset_add(&ofproto->port_poll_set, devname);
- ofproto->port_poll_errno = 0;
- }
- } else if (!ofproto) {
- /* The port was added, but we don't know with which
- * ofproto we should associate it. Delete it. */
- dpif_port_del(backer->dpif, port.port_no);
- }
- dpif_port_destroy(&port);
+ error = dpif_port_poll(backer->dpif, &devname);
+ switch (error) {
+ case EAGAIN:
+ return;
- next:
- free(devname);
- }
+ case ENOBUFS:
+ process_dpif_all_ports_changed(backer);
+ break;
- if (error != EAGAIN) {
- struct ofproto_dpif *ofproto;
+ case 0:
+ process_dpif_port_change(backer, devname);
+ free(devname);
+ break;
- /* 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;
- }
+ default:
+ process_dpif_port_error(backer, error);
+ break;
}
}
+}
- if (backer->governor) {
- size_t n_subfacets;
+static void
+process_dpif_all_ports_changed(struct dpif_backer *backer)
+{
+ struct ofproto_dpif *ofproto;
+ struct dpif_port dpif_port;
+ struct dpif_port_dump dump;
+ struct sset devnames;
+ const char *devname;
- governor_run(backer->governor);
+ sset_init(&devnames);
+ HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
+ if (ofproto->backer == backer) {
+ struct ofport *ofport;
- /* If the governor has shrunk to its minimum size and the number of
- * subfacets has dwindled, then drop the governor entirely.
- *
- * For hysteresis, the number of subfacets to drop the governor is
- * smaller than the number needed to trigger its creation. */
- n_subfacets = hmap_count(&backer->subfacets);
- if (n_subfacets * 4 < flow_eviction_threshold
- && governor_is_idle(backer->governor)) {
- governor_destroy(backer->governor);
- backer->governor = NULL;
+ HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
+ sset_add(&devnames, netdev_get_name(ofport->netdev));
+ }
}
}
+ DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
+ sset_add(&devnames, dpif_port.name);
+ }
- return 0;
+ SSET_FOR_EACH (devname, &devnames) {
+ process_dpif_port_change(backer, devname);
+ }
+ sset_destroy(&devnames);
}
-static int
-dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
+static void
+process_dpif_port_change(struct dpif_backer *backer, const char *devname)
{
- unsigned int work;
+ struct ofproto_dpif *ofproto;
+ struct dpif_port port;
- /* If recv_set_enable is false, we should not handle upcalls. */
- if (!backer->recv_set_enable) {
- return 0;
+ /* Don't report on the datapath's device. */
+ if (!strcmp(devname, dpif_base_name(backer->dpif))) {
+ return;
}
- /* Handle one or more batches of upcalls, until there's nothing left to do
- * or until we do a fixed total amount of work.
- *
- * We do work in batches because it can be much cheaper to set up a number
- * of flows and fire off their patches all at once. We do multiple batches
- * because in some cases handling a packet can cause another packet to be
- * queued almost immediately as part of the return flow. Both
- * optimizations can make major improvements on some benchmarks and
- * presumably for real traffic as well. */
- work = 0;
- while (work < max_batch) {
- int retval = handle_upcalls(backer, max_batch - work);
- if (retval <= 0) {
- return -retval;
+ HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
+ &all_ofproto_dpifs) {
+ if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
+ return;
}
- work += retval;
}
- return 0;
-}
-
-static int
-type_run_fast(const char *type)
-{
- struct dpif_backer *backer;
+ ofproto = lookup_ofproto_dpif_by_port_name(devname);
+ if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
+ /* The port was removed. If we know the datapath,
+ * report it through poll_set(). If we don't, it may be
+ * notifying us of a removal we initiated, so ignore it.
+ * If there's a pending ENOBUFS, let it stand, since
+ * everything will be reevaluated. */
+ if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
+ sset_add(&ofproto->port_poll_set, devname);
+ ofproto->port_poll_errno = 0;
+ }
+ } else if (!ofproto) {
+ /* The port was added, but we don't know with which
+ * ofproto we should associate it. Delete it. */
+ dpif_port_del(backer->dpif, port.port_no);
+ } else {
+ struct ofport_dpif *ofport;
- 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;
+ ofport = ofport_dpif_cast(shash_find_data(
+ &ofproto->up.port_by_name, devname));
+ if (ofport
+ && ofport->odp_port != port.port_no
+ && !odp_port_to_ofport(backer, port.port_no))
+ {
+ /* 'ofport''s datapath port number has changed from
+ * 'ofport->odp_port' to 'port.port_no'. Update our internal data
+ * structures to match. */
+ ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
+ hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
+ ofport->odp_port = port.port_no;
+ hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
+ hash_odp_port(port.port_no));
+ ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
+ backer->need_revalidate = REV_RECONFIGURE;
+ }
}
-
- return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
+ dpif_port_destroy(&port);
}
+/* Propagate 'error' to all ofprotos based on 'backer'. */
static void
-run_fast_rl(void)
+process_dpif_port_error(struct dpif_backer *backer, int error)
{
- static long long int port_rl = LLONG_MIN;
- static unsigned int backer_rl = 0;
-
- if (time_msec() >= port_rl) {
- struct ofproto_dpif *ofproto;
- struct ofport_dpif *ofport;
-
- HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
-
- HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
- port_run_fast(ofport);
- }
- }
- port_rl = time_msec() + 200;
- }
-
- /* XXX: We have to be careful not to do too much work in this function. If
- * we call dpif_backer_run_fast() too often, or with too large a batch,
- * performance improves signifcantly, but at a cost. It's possible for the
- * number of flows in the datapath to increase without bound, and for poll
- * loops to take 10s of seconds. The correct solution to this problem,
- * long term, is to separate flow miss handling into it's own thread so it
- * isn't affected by revalidations, and expirations. Until then, this is
- * the best we can do. */
- if (++backer_rl >= 10) {
- struct shash_node *node;
+ struct ofproto_dpif *ofproto;
- backer_rl = 0;
- SHASH_FOR_EACH (node, &all_dpif_backers) {
- dpif_backer_run_fast(node->data, 1);
+ 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;
}
- if (backer->governor) {
- governor_wait(backer->governor);
- }
-
- timer_wait(&backer->next_expiration);
+ dpif_wait(backer->dpif);
}
\f
/* Basic life-cycle. */
static void
close_dpif_backer(struct dpif_backer *backer)
{
- struct shash_node *node;
-
ovs_assert(backer->refcount > 0);
if (--backer->refcount) {
return;
}
- drop_key_clear(backer);
- hmap_destroy(&backer->drop_keys);
+ udpif_destroy(backer->udpif);
simap_destroy(&backer->tnl_backers);
+ ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
hmap_destroy(&backer->odp_to_ofport_map);
- node = shash_find(&all_dpif_backers, backer->type);
+ shash_find_and_delete(&all_dpif_backers, backer->type);
+ recirc_id_pool_destroy(backer->rid_pool);
free(backer->type);
- shash_delete(&all_dpif_backers, node);
dpif_close(backer->dpif);
-
- ovs_assert(hmap_is_empty(&backer->subfacets));
- hmap_destroy(&backer->subfacets);
- governor_destroy(backer->governor);
-
free(backer);
}
odp_port_t odp_port;
};
+static bool check_variable_length_userdata(struct dpif_backer *backer);
+static size_t check_max_mpls_depth(struct dpif_backer *backer);
+static bool check_recirc(struct dpif_backer *backer);
+
static int
open_dpif_backer(const char *type, struct dpif_backer **backerp)
{
struct shash_node *node;
struct list garbage_list;
struct odp_garbage *garbage, *next;
+
struct sset names;
char *backer_name;
const char *name;
free(backer);
return error;
}
+ backer->udpif = udpif_create(backer, backer->dpif);
backer->type = xstrdup(type);
- backer->governor = NULL;
backer->refcount = 1;
hmap_init(&backer->odp_to_ofport_map);
- hmap_init(&backer->drop_keys);
- hmap_init(&backer->subfacets);
- timer_set_duration(&backer->next_expiration, 1000);
+ ovs_rwlock_init(&backer->odp_to_ofport_lock);
backer->need_revalidate = 0;
simap_init(&backer->tnl_backers);
- tag_set_init(&backer->revalidate_set);
backer->recv_set_enable = !ofproto_get_flow_restore_wait();
*backerp = backer;
close_dpif_backer(backer);
return error;
}
+ backer->enable_recirc = check_recirc(backer);
+ backer->variable_length_userdata = check_variable_length_userdata(backer);
+ backer->max_mpls_depth = check_max_mpls_depth(backer);
+ backer->rid_pool = recirc_id_pool_create();
- backer->max_n_subfacet = 0;
- backer->created = time_msec();
- backer->last_minute = backer->created;
- memset(&backer->hourly, 0, sizeof backer->hourly);
- memset(&backer->daily, 0, sizeof backer->daily);
- memset(&backer->lifetime, 0, sizeof backer->lifetime);
- backer->subfacet_add_count = 0;
- backer->subfacet_del_count = 0;
- backer->total_subfacet_add_count = 0;
- backer->total_subfacet_del_count = 0;
- backer->avg_n_subfacet = 0;
- backer->avg_subfacet_life = 0;
+ if (backer->recv_set_enable) {
+ udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
+ }
return error;
}
+/* Tests whether 'backer''s datapath supports recirculation Only newer datapath
+ * supports OVS_KEY_ATTR in OVS_ACTION_ATTR_USERSPACE actions. We need to
+ * disable some features on older datapaths that don't support this feature.
+ *
+ * Returns false if 'backer' definitely does not support recirculation, true if
+ * it seems to support recirculation or if at least the error we get is
+ * ambiguous. */
+static bool
+check_recirc(struct dpif_backer *backer)
+{
+ struct flow flow;
+ struct odputil_keybuf keybuf;
+ struct ofpbuf key;
+ int error;
+ bool enable_recirc = false;
+
+ memset(&flow, 0, sizeof flow);
+ flow.recirc_id = 1;
+ flow.dp_hash = 1;
+
+ ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
+ odp_flow_key_from_flow(&key, &flow, NULL, 0);
+
+ error = dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
+ ofpbuf_data(&key), ofpbuf_size(&key), NULL, 0, NULL,
+ 0, NULL);
+ if (error && error != EEXIST) {
+ if (error != EINVAL) {
+ VLOG_WARN("%s: Reciculation flow probe failed (%s)",
+ dpif_name(backer->dpif), ovs_strerror(error));
+ }
+ goto done;
+ }
+
+ error = dpif_flow_del(backer->dpif, ofpbuf_data(&key), ofpbuf_size(&key),
+ NULL);
+ if (error) {
+ VLOG_WARN("%s: failed to delete recirculation feature probe flow",
+ dpif_name(backer->dpif));
+ }
+
+ enable_recirc = true;
+
+done:
+ if (enable_recirc) {
+ VLOG_INFO("%s: Datapath supports recirculation",
+ dpif_name(backer->dpif));
+ } else {
+ VLOG_INFO("%s: Datapath does not support recirculation",
+ dpif_name(backer->dpif));
+ }
+
+ return enable_recirc;
+}
+
+/* Tests whether 'backer''s datapath supports variable-length
+ * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions. We need
+ * to disable some features on older datapaths that don't support this
+ * feature.
+ *
+ * Returns false if 'backer' definitely does not support variable-length
+ * userdata, true if it seems to support them or if at least the error we get
+ * is ambiguous. */
+static bool
+check_variable_length_userdata(struct dpif_backer *backer)
+{
+ struct eth_header *eth;
+ struct ofpbuf actions;
+ struct dpif_execute execute;
+ struct ofpbuf packet;
+ size_t start;
+ int error;
+
+ /* Compose a userspace action that will cause an ERANGE error on older
+ * datapaths that don't support variable-length userdata.
+ *
+ * We really test for using userdata longer than 8 bytes, but older
+ * datapaths accepted these, silently truncating the userdata to 8 bytes.
+ * The same older datapaths rejected userdata shorter than 8 bytes, so we
+ * test for that instead as a proxy for longer userdata support. */
+ ofpbuf_init(&actions, 64);
+ start = nl_msg_start_nested(&actions, OVS_ACTION_ATTR_USERSPACE);
+ nl_msg_put_u32(&actions, OVS_USERSPACE_ATTR_PID,
+ dpif_port_get_pid(backer->dpif, ODPP_NONE, 0));
+ nl_msg_put_unspec_zero(&actions, OVS_USERSPACE_ATTR_USERDATA, 4);
+ nl_msg_end_nested(&actions, start);
+
+ /* Compose a dummy ethernet packet. */
+ ofpbuf_init(&packet, ETH_HEADER_LEN);
+ eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
+ eth->eth_type = htons(0x1234);
+
+ /* Execute the actions. On older datapaths this fails with ERANGE, on
+ * newer datapaths it succeeds. */
+ execute.actions = ofpbuf_data(&actions);
+ execute.actions_len = ofpbuf_size(&actions);
+ execute.packet = &packet;
+ execute.md = PKT_METADATA_INITIALIZER(0);
+ execute.needs_help = false;
+
+ error = dpif_execute(backer->dpif, &execute);
+
+ ofpbuf_uninit(&packet);
+ ofpbuf_uninit(&actions);
+
+ switch (error) {
+ case 0:
+ /* Variable-length userdata is supported.
+ *
+ * Purge received packets to avoid processing the nonsense packet we
+ * sent to userspace, then report success. */
+ dpif_recv_purge(backer->dpif);
+ return true;
+
+ case ERANGE:
+ /* Variable-length userdata is not supported. */
+ VLOG_WARN("%s: datapath does not support variable-length userdata "
+ "feature (needs Linux 3.10+ or kernel module from OVS "
+ "1..11+). The NXAST_SAMPLE action will be ignored.",
+ dpif_name(backer->dpif));
+ return false;
+
+ default:
+ /* Something odd happened. We're not sure whether variable-length
+ * userdata is supported. Default to "yes". */
+ VLOG_WARN("%s: variable-length userdata feature probe failed (%s)",
+ dpif_name(backer->dpif), ovs_strerror(error));
+ return true;
+ }
+}
+
+/* Tests the MPLS label stack depth supported by 'backer''s datapath.
+ *
+ * Returns the number of elements in a struct flow's mpls_lse field
+ * if the datapath supports at least that many entries in an
+ * MPLS label stack.
+ * Otherwise returns the number of MPLS push actions supported by
+ * the datapath. */
+static size_t
+check_max_mpls_depth(struct dpif_backer *backer)
+{
+ struct flow flow;
+ int n;
+
+ for (n = 0; n < FLOW_MAX_MPLS_LABELS; n++) {
+ struct odputil_keybuf keybuf;
+ struct ofpbuf key;
+ int error;
+
+ memset(&flow, 0, sizeof flow);
+ flow.dl_type = htons(ETH_TYPE_MPLS);
+ flow_set_mpls_bos(&flow, n, 1);
+
+ ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
+ odp_flow_key_from_flow(&key, &flow, NULL, 0);
+
+ error = dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
+ ofpbuf_data(&key), ofpbuf_size(&key), NULL, 0, NULL, 0, NULL);
+ if (error && error != EEXIST) {
+ if (error != EINVAL) {
+ VLOG_WARN("%s: MPLS stack length feature probe failed (%s)",
+ dpif_name(backer->dpif), ovs_strerror(error));
+ }
+ break;
+ }
+
+ error = dpif_flow_del(backer->dpif, ofpbuf_data(&key), ofpbuf_size(&key), NULL);
+ if (error) {
+ VLOG_WARN("%s: failed to delete MPLS feature probe flow",
+ dpif_name(backer->dpif));
+ }
+ }
+
+ VLOG_INFO("%s: MPLS label stack length probed as %d",
+ dpif_name(backer->dpif), n);
+ return n;
+}
+
static int
construct(struct ofproto *ofproto_)
{
struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
struct shash_node *node, *next;
- odp_port_t max_ports;
int error;
- int i;
error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
if (error) {
return error;
}
- max_ports = dpif_get_max_ports(ofproto->backer->dpif);
- ofproto_init_max_ports(ofproto_, u16_to_ofp(MIN(odp_to_u32(max_ports),
- ofp_to_u16(OFPP_MAX))));
-
ofproto->netflow = NULL;
ofproto->sflow = NULL;
ofproto->ipfix = NULL;
ofproto->stp = NULL;
+ ofproto->dump_seq = 0;
hmap_init(&ofproto->bundles);
ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
- for (i = 0; i < MAX_MIRRORS; i++) {
- ofproto->mirrors[i] = NULL;
- }
+ ofproto->mbridge = mbridge_create();
ofproto->has_bonded_bundles = false;
+ ofproto->lacp_enabled = false;
+ ovs_mutex_init_adaptive(&ofproto->stats_mutex);
+ ovs_mutex_init(&ofproto->vsp_mutex);
- classifier_init(&ofproto->facets);
- ofproto->consistency_rl = LLONG_MIN;
-
- for (i = 0; i < N_TABLES; i++) {
- struct table_dpif *table = &ofproto->tables[i];
-
- table->catchall_table = NULL;
- table->other_table = NULL;
- table->basis = random_uint32();
- }
-
- list_init(&ofproto->completions);
+ guarded_list_init(&ofproto->pins);
ofproto_dpif_unixctl_init();
- ofproto->has_mirrors = false;
hmap_init(&ofproto->vlandev_map);
hmap_init(&ofproto->realdev_vid_map);
sset_init(&ofproto->ghost_ports);
sset_init(&ofproto->port_poll_set);
ofproto->port_poll_errno = 0;
+ ofproto->change_seq = 0;
+ ofproto->pins_seq = seq_create();
+ ofproto->pins_seqno = seq_read(ofproto->pins_seq);
+
SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
struct iface_hint *iface_hint = node->data;
ofproto_init_tables(ofproto_, N_TABLES);
error = add_internal_flows(ofproto);
- ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
- ofproto->n_hit = 0;
- ofproto->n_missed = 0;
+ ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
return error;
}
static int
-add_internal_flow(struct ofproto_dpif *ofproto, int id,
+add_internal_miss_flow(struct ofproto_dpif *ofproto, int id,
const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
{
- struct ofputil_flow_mod fm;
+ struct match match;
int error;
+ struct rule *rule;
- 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;
- }
+ match_init_catchall(&match);
+ match_set_reg(&match, 0, id);
- *rulep = rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL,
- TBL_INTERNAL);
- ovs_assert(*rulep != NULL);
+ error = ofproto_dpif_add_internal_flow(ofproto, &match, 0, ofpacts, &rule);
+ *rulep = error ? NULL : rule_dpif_cast(rule);
- return 0;
+ return error;
}
static int
struct ofpact_controller *controller;
uint64_t ofpacts_stub[128 / 8];
struct ofpbuf ofpacts;
+ struct rule *unused_rulep OVS_UNUSED;
+ struct ofpact_resubmit *resubmit;
+ struct match match;
int error;
int id;
controller->reason = OFPR_NO_MATCH;
ofpact_pad(&ofpacts);
- error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
+ error = add_internal_miss_flow(ofproto, id++, &ofpacts,
+ &ofproto->miss_rule);
if (error) {
return error;
}
ofpbuf_clear(&ofpacts);
- error = add_internal_flow(ofproto, id++, &ofpacts,
+ error = add_internal_miss_flow(ofproto, id++, &ofpacts,
&ofproto->no_packet_in_rule);
if (error) {
return error;
}
- error = add_internal_flow(ofproto, id++, &ofpacts,
+ error = add_internal_miss_flow(ofproto, id++, &ofpacts,
&ofproto->drop_frags_rule);
- return error;
-}
+ if (error) {
+ return error;
+ }
-static void
-complete_operations(struct ofproto_dpif *ofproto)
-{
- struct dpif_completion *c, *next;
+ /* Continue non-recirculation rule lookups from table 0.
+ *
+ * (priority=2), recirc=0, actions=resubmit(, 0)
+ */
+ resubmit = ofpact_put_RESUBMIT(&ofpacts);
+ resubmit->ofpact.compat = 0;
+ resubmit->in_port = OFPP_IN_PORT;
+ resubmit->table_id = 0;
- LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
- ofoperation_complete(c->op, 0);
- list_remove(&c->list_node);
- free(c);
+ match_init_catchall(&match);
+ match_set_recirc_id(&match, 0);
+
+ error = ofproto_dpif_add_internal_flow(ofproto, &match, 2, &ofpacts,
+ &unused_rulep);
+ if (error) {
+ return error;
}
+
+ /* Drop any run away recirc rule lookups. Recirc_id has to be
+ * non-zero when reaching this rule.
+ *
+ * (priority=1), *, actions=drop
+ */
+ ofpbuf_clear(&ofpacts);
+ match_init_catchall(&match);
+ error = ofproto_dpif_add_internal_flow(ofproto, &match, 1, &ofpacts,
+ &unused_rulep);
+
+ return error;
}
static void
{
struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
struct rule_dpif *rule, *next_rule;
+ struct ofproto_packet_in *pin, *next_pin;
struct oftable *table;
- int i;
+ struct list pins;
ofproto->backer->need_revalidate = REV_RECONFIGURE;
+ ovs_rwlock_wrlock(&xlate_rwlock);
+ xlate_remove_ofproto(ofproto);
+ ovs_rwlock_unlock(&xlate_rwlock);
+
+ /* Ensure that the upcall processing threads have no remaining references
+ * to the ofproto or anything in it. */
+ udpif_synchronize(ofproto->backer->udpif);
+
hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
- complete_operations(ofproto);
OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
struct cls_cursor cursor;
+ fat_rwlock_rdlock(&table->cls.rwlock);
cls_cursor_init(&cursor, &table->cls, NULL);
+ fat_rwlock_unlock(&table->cls.rwlock);
CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
- ofproto_rule_destroy(&rule->up);
+ ofproto_rule_delete(&ofproto->up, &rule->up);
}
}
- for (i = 0; i < MAX_MIRRORS; i++) {
- mirror_destroy(ofproto->mirrors[i]);
+ guarded_list_pop_all(&ofproto->pins, &pins);
+ LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
+ list_remove(&pin->list_node);
+ free(CONST_CAST(void *, pin->up.packet));
+ free(pin);
}
+ guarded_list_destroy(&ofproto->pins);
- netflow_destroy(ofproto->netflow);
+ mbridge_unref(ofproto->mbridge);
+
+ netflow_unref(ofproto->netflow);
dpif_sflow_unref(ofproto->sflow);
hmap_destroy(&ofproto->bundles);
mac_learning_unref(ofproto->ml);
- classifier_destroy(&ofproto->facets);
-
hmap_destroy(&ofproto->vlandev_map);
hmap_destroy(&ofproto->realdev_vid_map);
sset_destroy(&ofproto->ghost_ports);
sset_destroy(&ofproto->port_poll_set);
- close_dpif_backer(ofproto->backer);
-}
-
-static int
-run_fast(struct ofproto *ofproto_)
-{
- struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
- struct ofport_dpif *ofport;
-
- /* Do not perform any periodic activity required by 'ofproto' while
- * waiting for flow restore to complete. */
- if (ofproto_get_flow_restore_wait()) {
- return 0;
- }
+ ovs_mutex_destroy(&ofproto->stats_mutex);
+ ovs_mutex_destroy(&ofproto->vsp_mutex);
- HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
- port_run_fast(ofport);
- }
+ seq_destroy(ofproto->pins_seq);
- return 0;
+ close_dpif_backer(ofproto->backer);
}
static int
run(struct ofproto *ofproto_)
{
struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
- struct ofport_dpif *ofport;
- struct ofbundle *bundle;
- int error;
+ uint64_t new_seq, new_dump_seq;
- if (!clogged) {
- complete_operations(ofproto);
+ if (mbridge_need_revalidate(ofproto->mbridge)) {
+ ofproto->backer->need_revalidate = REV_RECONFIGURE;
+ ovs_rwlock_wrlock(&ofproto->ml->rwlock);
+ mac_learning_flush(ofproto->ml);
+ ovs_rwlock_unlock(&ofproto->ml->rwlock);
}
- /* Do not perform any periodic activity below required by 'ofproto' while
+ /* Do not perform any periodic activity required by 'ofproto' while
* waiting for flow restore to complete. */
- if (ofproto_get_flow_restore_wait()) {
- return 0;
- }
+ if (!ofproto_get_flow_restore_wait()) {
+ struct ofproto_packet_in *pin, *next_pin;
+ struct list pins;
- error = run_fast(ofproto_);
- if (error) {
- return error;
+ guarded_list_pop_all(&ofproto->pins, &pins);
+ LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
+ connmgr_send_packet_in(ofproto->up.connmgr, pin);
+ list_remove(&pin->list_node);
+ free(CONST_CAST(void *, pin->up.packet));
+ free(pin);
+ }
}
+ /* Always updates the ofproto->pins_seqno to avoid frequent wakeup during
+ * flow restore. Even though nothing is processed during flow restore,
+ * all queued 'pins' will be handled immediately when flow restore
+ * completes. */
+ ofproto->pins_seqno = seq_read(ofproto->pins_seq);
+
if (ofproto->netflow) {
- if (netflow_run(ofproto->netflow)) {
- send_netflow_active_timeouts(ofproto);
- }
+ netflow_run(ofproto->netflow);
}
if (ofproto->sflow) {
dpif_sflow_run(ofproto->sflow);
}
+ if (ofproto->ipfix) {
+ dpif_ipfix_run(ofproto->ipfix);
+ }
+
+ new_seq = seq_read(connectivity_seq_get());
+ if (ofproto->change_seq != new_seq) {
+ struct ofport_dpif *ofport;
- HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
- port_run(ofport);
+ HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
+ port_run(ofport);
+ }
+
+ ofproto->change_seq = new_seq;
}
- HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
- bundle_run(bundle);
+ if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
+ struct ofbundle *bundle;
+
+ HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
+ bundle_run(bundle);
+ }
}
stp_run(ofproto);
- mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
-
- /* Check the consistency of a random facet, to aid debugging. */
- if (time_msec() >= ofproto->consistency_rl
- && !classifier_is_empty(&ofproto->facets)
- && !ofproto->backer->need_revalidate) {
- struct cls_table *table;
- struct cls_rule *cr;
- struct facet *facet;
-
- ofproto->consistency_rl = time_msec() + 250;
-
- table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
- struct cls_table, hmap_node);
- cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
- hmap_node);
- facet = CONTAINER_OF(cr, struct facet, cr);
-
- if (!tag_set_intersects(&ofproto->backer->revalidate_set,
- facet->xout.tags)) {
- if (!facet_check_consistency(facet)) {
- ofproto->backer->need_revalidate = REV_INCONSISTENCY;
+ ovs_rwlock_wrlock(&ofproto->ml->rwlock);
+ if (mac_learning_run(ofproto->ml)) {
+ ofproto->backer->need_revalidate = REV_MAC_LEARNING;
+ }
+ ovs_rwlock_unlock(&ofproto->ml->rwlock);
+
+ new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
+ if (ofproto->dump_seq != new_dump_seq) {
+ struct rule *rule, *next_rule;
+
+ /* We know stats are relatively fresh, so now is a good time to do some
+ * periodic work. */
+ ofproto->dump_seq = new_dump_seq;
+
+ /* Expire OpenFlow flows whose idle_timeout or hard_timeout
+ * has passed. */
+ ovs_mutex_lock(&ofproto_mutex);
+ LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
+ &ofproto->up.expirable) {
+ rule_expire(rule_dpif_cast(rule));
+ }
+ ovs_mutex_unlock(&ofproto_mutex);
+
+ /* 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);
+ }
}
}
}
wait(struct ofproto *ofproto_)
{
struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
- struct ofport_dpif *ofport;
- struct ofbundle *bundle;
-
- if (!clogged && !list_is_empty(&ofproto->completions)) {
- poll_immediate_wake();
- }
if (ofproto_get_flow_restore_wait()) {
return;
}
- dpif_wait(ofproto->backer->dpif);
- dpif_recv_wait(ofproto->backer->dpif);
if (ofproto->sflow) {
dpif_sflow_wait(ofproto->sflow);
}
- if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
- poll_immediate_wake();
- }
- HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
- port_wait(ofport);
+ if (ofproto->ipfix) {
+ dpif_ipfix_wait(ofproto->ipfix);
}
- HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
- bundle_wait(bundle);
+ if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
+ struct ofbundle *bundle;
+
+ HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
+ bundle_wait(bundle);
+ }
}
if (ofproto->netflow) {
netflow_wait(ofproto->netflow);
}
+ ovs_rwlock_rdlock(&ofproto->ml->rwlock);
mac_learning_wait(ofproto->ml);
+ ovs_rwlock_unlock(&ofproto->ml->rwlock);
stp_wait(ofproto);
if (ofproto->backer->need_revalidate) {
/* Shouldn't happen, but if it does just go around again. */
VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
poll_immediate_wake();
}
+
+ seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
+ seq_wait(ofproto->pins_seq, ofproto->pins_seqno);
}
static void
-get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
+type_get_memory_usage(const char *type, struct simap *usage)
{
- const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
- struct cls_cursor cursor;
- size_t n_subfacets = 0;
- struct facet *facet;
-
- simap_increase(usage, "facets", classifier_count(&ofproto->facets));
+ struct dpif_backer *backer;
- cls_cursor_init(&cursor, &ofproto->facets, NULL);
- CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
- n_subfacets += list_size(&facet->subfacets);
+ backer = shash_find_data(&all_dpif_backers, type);
+ if (backer) {
+ udpif_get_memory_usage(backer->udpif, usage);
}
- simap_increase(usage, "subfacets", n_subfacets);
}
static void
flush(struct ofproto *ofproto_)
{
struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
- 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->backer->subfacets) {
- if (ofproto_dpif_cast(subfacet->facet->rule->up.ofproto) != ofproto) {
- continue;
- }
-
- if (subfacet->path != SF_NOT_INSTALLED) {
- batch[n_batch++] = subfacet;
- if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
- subfacet_destroy_batch(ofproto->backer, batch, n_batch);
- n_batch = 0;
- }
- } else {
- subfacet_destroy(subfacet);
- }
- }
+ struct dpif_backer *backer = ofproto->backer;
- if (n_batch > 0) {
- subfacet_destroy_batch(ofproto->backer, batch, n_batch);
+ if (backer) {
+ udpif_flush(backer->udpif);
}
}
struct dpif_dp_stats s;
uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
uint64_t n_lookup;
+ long long int used;
strcpy(ots->name, "classifier");
dpif_get_dp_stats(ofproto->backer->dpif, &s);
- rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
- rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
- rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
-
+ rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes, &used);
+ rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes,
+ &used);
+ rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes,
+ &used);
n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
ots->lookup_count = htonll(n_lookup);
ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
port->bundle = NULL;
port->cfm = NULL;
port->bfd = NULL;
- port->tag = tag_create_random();
port->may_enable = true;
port->stp_port = NULL;
port->stp_state = STP_DISABLED;
port->is_tunnel = false;
port->peer = NULL;
- hmap_init(&port->priorities);
+ port->qdscp = NULL;
+ port->n_qdscp = 0;
port->realdev_ofp_port = 0;
port->vlandev_vid = 0;
port->carrier_seq = netdev_get_carrier_resets(netdev);
+ port->is_layer3 = netdev_vport_is_layer3(netdev);
if (netdev_vport_is_patch(netdev)) {
/* By bailing out here, we don't submit the port to the sFlow module
return EBUSY;
}
+ ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
hash_odp_port(port->odp_port));
+ ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
}
dpif_port_destroy(&dpif_port);
const char *dp_port_name;
ofproto->backer->need_revalidate = REV_RECONFIGURE;
+ ovs_rwlock_wrlock(&xlate_rwlock);
+ xlate_ofport_remove(port);
+ ovs_rwlock_unlock(&xlate_rwlock);
dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
sizeof namebuf);
}
if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
+ ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
+ ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
}
tnl_port_del(port);
bundle_remove(port_);
set_cfm(port_, NULL);
set_bfd(port_, NULL);
+ if (port->stp_port) {
+ stp_port_disable(port->stp_port);
+ }
if (ofproto->sflow) {
dpif_sflow_del_port(ofproto->sflow, port->odp_port);
}
- ofport_clear_priorities(port);
- hmap_destroy(&port->priorities);
+ free(port->qdscp);
}
static void
cfm_set_netdev(port->cfm, port->up.netdev);
}
+ if (port->bfd) {
+ bfd_set_netdev(port->bfd, port->up.netdev);
+ }
+
+ ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
+ port->up.pp.hw_addr);
+
if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
port->odp_port)) {
ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
{
struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
struct dpif_ipfix *di = ofproto->ipfix;
+ bool has_options = bridge_exporter_options || flow_exporters_options;
- if (bridge_exporter_options || flow_exporters_options) {
- if (!di) {
- di = ofproto->ipfix = dpif_ipfix_create();
- }
+ if (has_options && !di) {
+ di = ofproto->ipfix = dpif_ipfix_create();
+ }
+
+ if (di) {
+ /* Call set_options in any case to cleanly flush the flow
+ * caches in the last exporters that are to be destroyed. */
dpif_ipfix_set_options(
di, bridge_exporter_options, flow_exporters_options,
n_flow_exporters_options);
- } else {
- if (di) {
+
+ if (!has_options) {
dpif_ipfix_unref(di);
ofproto->ipfix = NULL;
}
}
+
return 0;
}
set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
{
struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
- int error;
+ int error = 0;
- if (!s) {
- error = 0;
- } else {
+ if (s) {
if (!ofport->cfm) {
struct ofproto_dpif *ofproto;
}
if (cfm_configure(ofport->cfm, s)) {
- return 0;
+ error = 0;
+ goto out;
}
error = EINVAL;
}
cfm_unref(ofport->cfm);
ofport->cfm = NULL;
+out:
+ ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
+ ofport->up.pp.hw_addr);
return error;
}
-static bool
+static int
get_cfm_status(const struct ofport *ofport_,
struct ofproto_cfm_status *status)
{
struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
+ int ret = 0;
if (ofport->cfm) {
- status->faults = cfm_get_fault(ofport->cfm);
- status->remote_opstate = cfm_get_opup(ofport->cfm);
- status->health = cfm_get_health(ofport->cfm);
- cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
- return true;
+ if (cfm_check_status_change(ofport->cfm)) {
+ status->faults = cfm_get_fault(ofport->cfm);
+ status->flap_count = cfm_get_flap_count(ofport->cfm);
+ status->remote_opstate = cfm_get_opup(ofport->cfm);
+ status->health = cfm_get_health(ofport->cfm);
+ cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
+ } else {
+ ret = NO_STATUS_CHANGE;
+ }
} else {
- return false;
+ ret = ENOENT;
}
+
+ return ret;
}
static int
struct bfd *old;
old = ofport->bfd;
- ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
+ ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
+ cfg, ofport->up.netdev);
if (ofport->bfd != old) {
ofproto->backer->need_revalidate = REV_RECONFIGURE;
}
-
+ ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
+ ofport->up.pp.hw_addr);
return 0;
}
get_bfd_status(struct ofport *ofport_, struct smap *smap)
{
struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
+ int ret = 0;
if (ofport->bfd) {
- bfd_get_status(ofport->bfd, smap);
- return 0;
+ if (bfd_check_status_change(ofport->bfd)) {
+ bfd_get_status(ofport->bfd, smap);
+ } else {
+ ret = NO_STATUS_CHANGE;
+ }
} else {
- return ENOENT;
+ ret = ENOENT;
}
+
+ return ret;
}
\f
/* Spanning Tree. */
VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
ofproto->up.name, port_num);
} else {
- struct eth_header *eth = pkt->l2;
+ struct eth_header *eth = ofpbuf_l2(pkt);
netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
if (eth_addr_is_zero(eth->eth_src)) {
VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
"with unknown MAC", ofproto->up.name, port_num);
} else {
- send_packet(ofport, pkt);
+ ofproto_dpif_send_packet(ofport, pkt);
}
}
ofpbuf_delete(pkt);
set_stp_port(ofport, NULL);
}
- stp_destroy(ofproto->stp);
+ stp_unref(ofproto->stp);
ofproto->stp = NULL;
}
if (stp_learn_in_state(ofport->stp_state)
!= stp_learn_in_state(state)) {
/* xxx Learning action flows should also be flushed. */
- mac_learning_flush(ofproto->ml,
- &ofproto->backer->revalidate_set);
+ ovs_rwlock_wrlock(&ofproto->ml->rwlock);
+ mac_learning_flush(ofproto->ml);
+ ovs_rwlock_unlock(&ofproto->ml->rwlock);
}
fwd_change = stp_forward_in_state(ofport->stp_state)
!= stp_forward_in_state(state);
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);
+
+ return 0;
+}
+
+static int
+get_stp_port_stats(struct ofport *ofport_,
+ struct ofproto_port_stp_stats *s)
+{
+ struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
+ struct stp_port *sp = ofport->stp_port;
+
+ if (!ofproto->stp || !sp) {
+ s->enabled = false;
+ return 0;
+ }
+
+ s->enabled = true;
stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
return 0;
}
if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
- mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
+ ovs_rwlock_wrlock(&ofproto->ml->rwlock);
+ mac_learning_flush(ofproto->ml);
+ ovs_rwlock_unlock(&ofproto->ml->rwlock);
}
}
}
poll_timer_wait(1000);
}
}
-
-/* Returns true if STP should process 'flow'. Sets fields in 'wc' that
- * were used to make the determination.*/
-bool
-stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
-{
- memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
- return eth_addr_equals(flow->dl_dst, eth_addr_stp);
-}
-
-void
-stp_process_packet(const struct ofport_dpif *ofport,
- const struct ofpbuf *packet)
-{
- struct ofpbuf payload = *packet;
- struct eth_header *eth = payload.data;
- struct stp_port *sp = ofport->stp_port;
-
- /* Sink packets on ports that have STP disabled when the bridge has
- * STP enabled. */
- if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
- return;
- }
-
- /* Trim off padding on payload. */
- if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
- payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
- }
-
- if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
- stp_received_bpdu(sp, payload.data, payload.size);
- }
-}
\f
-int
-ofproto_dpif_queue_to_priority(const struct ofproto_dpif *ofproto,
- uint32_t queue_id, uint32_t *priority)
-{
- return dpif_queue_to_priority(ofproto->backer->dpif, queue_id, priority);
-}
-
-static struct priority_to_dscp *
-get_priority(const struct ofport_dpif *ofport, uint32_t priority)
-{
- 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;
-}
-
-bool
-ofproto_dpif_dscp_from_priority(const struct ofport_dpif *ofport,
- uint32_t priority, uint8_t *dscp)
-{
- struct priority_to_dscp *pdscp = get_priority(ofport, priority);
- *dscp = pdscp ? pdscp->dscp : 0;
- return pdscp != NULL;
-}
-
-static void
-ofport_clear_priorities(struct ofport_dpif *ofport)
-{
- 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,
+set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
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->backer->need_revalidate = REV_RECONFIGURE;
- }
-
- if (pdscp->dscp != dscp) {
- pdscp->dscp = dscp;
- ofproto->backer->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);
+ if (ofport->n_qdscp != n_qdscp
+ || (n_qdscp && memcmp(ofport->qdscp, qdscp,
+ n_qdscp * sizeof *qdscp))) {
ofproto->backer->need_revalidate = REV_RECONFIGURE;
+ free(ofport->qdscp);
+ ofport->qdscp = n_qdscp
+ ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
+ : NULL;
+ ofport->n_qdscp = n_qdscp;
}
- hmap_swap(&new, &ofport->priorities);
- hmap_destroy(&new);
-
return 0;
}
\f
struct mac_entry *mac, *next_mac;
ofproto->backer->need_revalidate = REV_RECONFIGURE;
+ ovs_rwlock_wrlock(&ml->rwlock);
LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
if (mac->port.p == bundle) {
if (all_ofprotos) {
if (o != ofproto) {
struct mac_entry *e;
- e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
- NULL);
+ ovs_rwlock_wrlock(&o->ml->rwlock);
+ e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
if (e) {
mac_learning_expire(o->ml, e);
}
+ ovs_rwlock_unlock(&o->ml->rwlock);
}
}
}
mac_learning_expire(ml, mac);
}
}
+ ovs_rwlock_unlock(&ml->rwlock);
}
static struct ofbundle *
return NULL;
}
-/* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
- * ones that are found to 'bundles'. */
-static void
-bundle_lookup_multiple(struct ofproto_dpif *ofproto,
- void **auxes, size_t n_auxes,
- struct hmapx *bundles)
-{
- size_t i;
-
- hmapx_init(bundles);
- for (i = 0; i < n_auxes; i++) {
- struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
- if (bundle) {
- hmapx_add(bundles, bundle);
- }
- }
-}
-
static void
bundle_update(struct ofbundle *bundle)
{
bundle->floodable = true;
LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
+ || port->is_layer3
|| !stp_forward_in_state(port->stp_state)) {
bundle->floodable = false;
break;
if (port->bundle != bundle) {
bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
if (port->bundle) {
- bundle_del_port(port);
+ bundle_remove(&port->up);
}
port->bundle = bundle;
list_push_back(&bundle->ports, &port->bundle_node);
if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
+ || port->is_layer3
|| !stp_forward_in_state(port->stp_state)) {
bundle->floodable = false;
}
{
struct ofproto_dpif *ofproto;
struct ofport_dpif *port, *next_port;
- int i;
if (!bundle) {
return;
}
ofproto = bundle->ofproto;
- for (i = 0; i < MAX_MIRRORS; i++) {
- struct ofmirror *m = ofproto->mirrors[i];
- if (m) {
- if (m->out == bundle) {
- mirror_destroy(m);
- } else if (hmapx_find_and_delete(&m->srcs, bundle)
- || hmapx_find_and_delete(&m->dsts, bundle)) {
- ofproto->backer->need_revalidate = REV_RECONFIGURE;
- }
- }
- }
+ mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
+
+ ovs_rwlock_wrlock(&xlate_rwlock);
+ xlate_bundle_remove(bundle);
+ ovs_rwlock_unlock(&xlate_rwlock);
LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
bundle_del_port(port);
bundle->bond = NULL;
bundle->floodable = true;
-
- bundle->src_mirrors = 0;
- bundle->dst_mirrors = 0;
- bundle->mirror_out = 0;
+ mbridge_register_bundle(ofproto->mbridge, bundle);
}
if (!bundle->name || strcmp(s->name, bundle->name)) {
/* LACP. */
if (s->lacp) {
+ ofproto->lacp_enabled = true;
if (!bundle->lacp) {
ofproto->backer->need_revalidate = REV_RECONFIGURE;
bundle->lacp = lacp_create();
break;
default:
- NOT_REACHED();
+ OVS_NOT_REACHED();
}
if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
free(bundle->trunks);
ofproto->backer->need_revalidate = REV_RECONFIGURE;
}
} else {
- bundle->bond = bond_create(s->bond);
+ bundle->bond = bond_create(s->bond, ofproto);
ofproto->backer->need_revalidate = REV_RECONFIGURE;
}
LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
- bond_slave_register(bundle->bond, port, port->up.netdev);
+ bond_slave_register(bundle->bond, port,
+ port->up.ofp_port, port->up.netdev);
}
} else {
bond_unref(bundle->bond);
pdu_size);
memcpy(packet_pdu, pdu, pdu_size);
- send_packet(port, &packet);
+ ofproto_dpif_send_packet(port, &packet);
ofpbuf_uninit(&packet);
} else {
VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
bundle_send_learning_packets(struct ofbundle *bundle)
{
struct ofproto_dpif *ofproto = bundle->ofproto;
+ struct ofpbuf *learning_packet;
int error, n_packets, n_errors;
struct mac_entry *e;
+ struct list packets;
- error = n_packets = n_errors = 0;
+ list_init(&packets);
+ ovs_rwlock_rdlock(&ofproto->ml->rwlock);
LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
if (e->port.p != bundle) {
- struct ofpbuf *learning_packet;
- struct ofport_dpif *port;
void *port_void;
- int ret;
- /* 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;
- n_errors++;
- }
- n_packets++;
+ /* Temporarily use 'frame' as a private pointer (see below). */
+ ovs_assert(learning_packet->frame == ofpbuf_data(learning_packet));
+ learning_packet->frame = port_void;
+ list_push_back(&packets, &learning_packet->list_node);
}
}
+ ovs_rwlock_unlock(&ofproto->ml->rwlock);
+
+ error = n_packets = n_errors = 0;
+ LIST_FOR_EACH (learning_packet, list_node, &packets) {
+ int ret;
+ void *port_void = learning_packet->frame;
+
+ /* Restore 'frame'. */
+ learning_packet->frame = ofpbuf_data(learning_packet);
+ ret = ofproto_dpif_send_packet(port_void, learning_packet);
+ if (ret) {
+ error = ret;
+ n_errors++;
+ }
+ n_packets++;
+ }
+ ofpbuf_list_delete(&packets);
if (n_errors) {
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
}
- bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
- lacp_status(bundle->lacp));
+ if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
+ bundle->ofproto->backer->need_revalidate = REV_BOND;
+ }
+
if (bond_should_send_learning_packets(bundle->bond)) {
bundle_send_learning_packets(bundle);
}
/* Mirrors. */
static int
-mirror_scan(struct ofproto_dpif *ofproto)
+mirror_set__(struct ofproto *ofproto_, void *aux,
+ const struct ofproto_mirror_settings *s)
{
- int idx;
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+ struct ofbundle **srcs, **dsts;
+ int error;
+ size_t i;
- for (idx = 0; idx < MAX_MIRRORS; idx++) {
- if (!ofproto->mirrors[idx]) {
- return idx;
- }
+ if (!s) {
+ mirror_destroy(ofproto->mbridge, aux);
+ return 0;
}
- return -1;
-}
-static struct ofmirror *
-mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
-{
- int i;
+ srcs = xmalloc(s->n_srcs * sizeof *srcs);
+ dsts = xmalloc(s->n_dsts * sizeof *dsts);
- for (i = 0; i < MAX_MIRRORS; i++) {
- struct ofmirror *mirror = ofproto->mirrors[i];
- if (mirror && mirror->aux == aux) {
- return mirror;
- }
- }
-
- 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)
-{
- struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
- mirror_mask_t mirror_bit;
- struct ofbundle *bundle;
- struct ofmirror *mirror;
- struct ofbundle *out;
- struct hmapx srcs; /* Contains "struct ofbundle *"s. */
- struct hmapx dsts; /* Contains "struct ofbundle *"s. */
- int out_vlan;
-
- mirror = mirror_lookup(ofproto, aux);
- if (!s) {
- mirror_destroy(mirror);
- return 0;
- }
- if (!mirror) {
- int idx;
-
- idx = mirror_scan(ofproto);
- if (idx < 0) {
- VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
- "cannot create %s",
- ofproto->up.name, MAX_MIRRORS, s->name);
- return EFBIG;
- }
-
- mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
- mirror->ofproto = ofproto;
- mirror->idx = idx;
- mirror->aux = aux;
- mirror->out_vlan = -1;
- mirror->name = NULL;
- }
-
- if (!mirror->name || strcmp(s->name, mirror->name)) {
- free(mirror->name);
- mirror->name = xstrdup(s->name);
- }
-
- /* Get the new configuration. */
- if (s->out_bundle) {
- out = bundle_lookup(ofproto, s->out_bundle);
- if (!out) {
- mirror_destroy(mirror);
- return EINVAL;
- }
- out_vlan = -1;
- } else {
- out = NULL;
- out_vlan = s->out_vlan;
- }
- bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
- bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
-
- /* If the configuration has not changed, do nothing. */
- if (hmapx_equals(&srcs, &mirror->srcs)
- && hmapx_equals(&dsts, &mirror->dsts)
- && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
- && mirror->out == out
- && mirror->out_vlan == out_vlan)
- {
- hmapx_destroy(&srcs);
- hmapx_destroy(&dsts);
- return 0;
- }
-
- hmapx_swap(&srcs, &mirror->srcs);
- hmapx_destroy(&srcs);
-
- hmapx_swap(&dsts, &mirror->dsts);
- hmapx_destroy(&dsts);
-
- free(mirror->vlans);
- mirror->vlans = vlan_bitmap_clone(s->src_vlans);
-
- mirror->out = out;
- mirror->out_vlan = out_vlan;
-
- /* Update bundles. */
- mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
- HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
- if (hmapx_contains(&mirror->srcs, bundle)) {
- bundle->src_mirrors |= mirror_bit;
- } else {
- bundle->src_mirrors &= ~mirror_bit;
- }
-
- if (hmapx_contains(&mirror->dsts, bundle)) {
- bundle->dst_mirrors |= mirror_bit;
- } else {
- bundle->dst_mirrors &= ~mirror_bit;
- }
-
- if (mirror->out == bundle) {
- bundle->mirror_out |= mirror_bit;
- } else {
- bundle->mirror_out &= ~mirror_bit;
- }
- }
-
- ofproto->backer->need_revalidate = REV_RECONFIGURE;
- ofproto->has_mirrors = true;
- mac_learning_flush(ofproto->ml,
- &ofproto->backer->revalidate_set);
- mirror_update_dups(ofproto);
-
- return 0;
-}
-
-static void
-mirror_destroy(struct ofmirror *mirror)
-{
- struct ofproto_dpif *ofproto;
- mirror_mask_t mirror_bit;
- struct ofbundle *bundle;
- int i;
-
- if (!mirror) {
- return;
+ for (i = 0; i < s->n_srcs; i++) {
+ srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
}
- ofproto = mirror->ofproto;
- ofproto->backer->need_revalidate = REV_RECONFIGURE;
- mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
-
- mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
- HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
- bundle->src_mirrors &= ~mirror_bit;
- bundle->dst_mirrors &= ~mirror_bit;
- bundle->mirror_out &= ~mirror_bit;
+ for (i = 0; i < s->n_dsts; i++) {
+ dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
}
- hmapx_destroy(&mirror->srcs);
- hmapx_destroy(&mirror->dsts);
- free(mirror->vlans);
-
- ofproto->mirrors[mirror->idx] = NULL;
- free(mirror->name);
- free(mirror);
-
- mirror_update_dups(ofproto);
-
- ofproto->has_mirrors = false;
- for (i = 0; i < MAX_MIRRORS; i++) {
- if (ofproto->mirrors[i]) {
- ofproto->has_mirrors = true;
- break;
- }
- }
+ error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
+ s->n_dsts, s->src_vlans,
+ bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
+ free(srcs);
+ free(dsts);
+ return error;
}
static int
-mirror_get_stats(struct ofproto *ofproto_, void *aux,
- uint64_t *packets, uint64_t *bytes)
+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;
- }
-
- push_all_stats();
-
- *packets = mirror->packet_count;
- *bytes = mirror->byte_count;
-
- return 0;
+ return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
+ bytes);
}
static int
set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
{
struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+ ovs_rwlock_wrlock(&ofproto->ml->rwlock);
if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
- mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
+ mac_learning_flush(ofproto->ml);
}
+ ovs_rwlock_unlock(&ofproto->ml->rwlock);
return 0;
}
{
struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
struct ofbundle *bundle = bundle_lookup(ofproto, aux);
- return bundle && bundle->mirror_out != 0;
+ return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
}
static void
size_t max_entries)
{
struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+ ovs_rwlock_wrlock(&ofproto->ml->rwlock);
mac_learning_set_idle_time(ofproto->ml, idle_time);
mac_learning_set_max_entries(ofproto->ml, max_entries);
+ ovs_rwlock_unlock(&ofproto->ml->rwlock);
}
\f
/* Ports. */
-struct ofport_dpif *
+static struct ofport_dpif *
get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
{
struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
return ofport ? ofport_dpif_cast(ofport) : NULL;
}
-struct ofport_dpif *
-get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
-{
- struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
- return port && &ofproto->up == port->up.ofproto ? port : NULL;
-}
-
static void
ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
struct ofproto_port *ofproto_port,
{
const struct ofproto_dpif *ofproto;
struct dpif_backer *backer;
- const char *peer_name;
+ char *peer_name;
if (!netdev_vport_is_patch(ofport->up.netdev)) {
return;
HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
struct ofport *peer_ofport;
struct ofport_dpif *peer;
- const char *peer_peer;
+ char *peer_peer;
if (ofproto->backer != backer) {
continue;
ofport->peer = peer;
ofport->peer->peer = ofport;
}
+ free(peer_peer);
- return;
- }
-}
-
-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);
- }
-
- if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
- struct ofpbuf packet;
-
- ofpbuf_init(&packet, 0);
- bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
- send_packet(ofport, &packet);
- ofpbuf_uninit(&packet);
+ break;
}
+ free(peer_name);
}
static void
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);
+ bool cfm_enable = false;
+ bool bfd_enable = false;
ofport->carrier_seq = carrier_seq;
- port_run_fast(ofport);
-
if (ofport->cfm) {
int cfm_opup = cfm_get_opup(ofport->cfm);
- cfm_run(ofport->cfm);
- enable = enable && !cfm_get_fault(ofport->cfm);
+ cfm_enable = !cfm_get_fault(ofport->cfm);
if (cfm_opup >= 0) {
- enable = enable && cfm_opup;
+ cfm_enable = cfm_enable && cfm_opup;
}
}
if (ofport->bfd) {
- bfd_run(ofport->bfd);
- enable = enable && bfd_forwarding(ofport->bfd);
+ bfd_enable = bfd_forwarding(ofport->bfd);
+ }
+
+ if (ofport->bfd || ofport->cfm) {
+ enable = enable && (cfm_enable || bfd_enable);
}
if (ofport->bundle) {
ofport->may_enable = enable;
}
-static void
-port_wait(struct ofport_dpif *ofport)
-{
- if (ofport->cfm) {
- cfm_wait(ofport->cfm);
- }
-
- if (ofport->bfd) {
- bfd_wait(ofport->bfd);
- }
-}
-
static int
port_query_by_name(const struct ofproto *ofproto_, const char *devname,
struct ofproto_port *ofproto_port)
sset_find_and_delete(&ofproto->ghost_ports,
netdev_get_name(ofport->up.netdev));
ofproto->backer->need_revalidate = REV_RECONFIGURE;
- if (!ofport->is_tunnel) {
+ if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
if (!error) {
/* The caller is going to close ofport->up.netdev. If this is a
struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
int error;
- push_all_stats();
-
error = netdev_get_stats(ofport->up.netdev, stats);
if (!error && ofport_->ofp_port == OFPP_LOCAL) {
struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
+ ovs_mutex_lock(&ofproto->stats_mutex);
/* 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. */
+ * ofproto_dpif_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->tx_bytes != UINT64_MAX) {
stats->tx_bytes += ofproto->stats.rx_bytes;
}
+ ovs_mutex_unlock(&ofproto->stats_mutex);
}
return error;
: -1);
}
\f
-/* Upcall handling. */
-
-/* Flow miss batching.
- *
- * Some dpifs implement operations faster when you hand them off in a batch.
- * To allow batching, "struct flow_miss" queues the dpif-related work needed
- * for a given flow. Each "struct flow_miss" corresponds to sending one or
- * more packets, plus possibly installing the flow in the dpif.
- *
- * So far we only batch the operations that affect flow setup time the most.
- * 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;
- struct list packets;
- enum dpif_upcall_type upcall_type;
-};
-
-struct flow_miss_op {
- struct dpif_op dpif_op;
-
- uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
- struct xlate_out xout;
- bool xout_garbage; /* 'xout' needs to be uninitialized? */
-
- struct ofpbuf mask; /* Flow mask for "put" ops. */
- struct odputil_keybuf maskbuf;
-
- /* If this is a "put" op, then a pointer to the subfacet that should
- * be marked as uninstalled if the operation fails. */
- struct subfacet *subfacet;
-};
-
-/* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
- * OpenFlow controller as necessary according to their individual
- * configurations. */
+/* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
+ * then delete it entirely. */
static void
-send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
- const struct flow *flow)
+rule_expire(struct rule_dpif *rule)
+ OVS_REQUIRES(ofproto_mutex)
{
- struct ofputil_packet_in pin;
+ uint16_t hard_timeout, idle_timeout;
+ long long int now = time_msec();
+ int reason = -1;
- pin.packet = packet->data;
- pin.packet_len = packet->size;
- pin.reason = OFPR_NO_MATCH;
- pin.controller_id = 0;
+ ovs_assert(!rule->up.pending);
- pin.table_id = 0;
- pin.cookie = 0;
+ hard_timeout = rule->up.hard_timeout;
+ idle_timeout = rule->up.idle_timeout;
- pin.send_len = 0; /* not used for flow table misses */
+ /* Has 'rule' expired? */
+ if (hard_timeout) {
+ long long int modified;
- flow_get_metadata(flow, &pin.fmd);
+ ovs_mutex_lock(&rule->up.mutex);
+ modified = rule->up.modified;
+ ovs_mutex_unlock(&rule->up.mutex);
- connmgr_send_packet_in(ofproto->up.connmgr, &pin);
-}
+ if (now > modified + hard_timeout * 1000) {
+ reason = OFPRR_HARD_TIMEOUT;
+ }
+ }
-static struct flow_miss *
-flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
- const struct flow *flow, uint32_t hash)
-{
- struct flow_miss *miss;
+ if (reason < 0 && idle_timeout) {
+ long long int used;
+
+ ovs_mutex_lock(&rule->stats_mutex);
+ used = rule->stats.used;
+ ovs_mutex_unlock(&rule->stats_mutex);
- HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
- if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
- return miss;
+ if (now > used + idle_timeout * 1000) {
+ reason = OFPRR_IDLE_TIMEOUT;
}
}
- return NULL;
+ if (reason >= 0) {
+ COVERAGE_INC(ofproto_dpif_expired);
+ ofproto_rule_expire(&rule->up, reason);
+ }
}
-/* 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
-init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
- struct flow_miss_op *op)
-{
- if (miss->flow.in_port.ofp_port
- != vsp_realdev_to_vlandev(miss->ofproto, miss->flow.in_port.ofp_port,
- miss->flow.vlan_tci)) {
- /* This packet was received on a VLAN splinter port. We
- * added a VLAN to the packet to make the packet resemble
- * the flow, but the actions were composed assuming that
- * 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);
- }
-
- op->subfacet = NULL;
- op->xout_garbage = false;
- op->dpif_op.type = DPIF_OP_EXECUTE;
- op->dpif_op.u.execute.key = miss->key;
- op->dpif_op.u.execute.key_len = miss->key_len;
- op->dpif_op.u.execute.packet = packet;
- ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
-}
-
-/* Helper for handle_flow_miss_without_facet() and
- * handle_flow_miss_with_facet(). */
-static void
-handle_flow_miss_common(struct rule_dpif *rule,
- struct ofpbuf *packet, const struct flow *flow)
+/* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
+ * 'flow' must reflect the data in 'packet'. */
+int
+ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
+ const struct flow *flow,
+ struct rule_dpif *rule,
+ const struct ofpact *ofpacts, size_t ofpacts_len,
+ struct ofpbuf *packet)
{
- struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
+ struct dpif_flow_stats stats;
+ struct xlate_out xout;
+ struct xlate_in xin;
+ ofp_port_t in_port;
+ struct dpif_execute execute;
+ int error;
- 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);
- }
-}
+ ovs_assert((rule != NULL) != (ofpacts != NULL));
-/* Figures out whether a flow that missed in 'ofproto', whose details are in
- * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
- * and (usually) installing a datapath flow. The answer is usually "yes" (a
- * return value of true). However, for short flows the cost of bookkeeping is
- * much higher than the benefits, so when the datapath holds a large number of
- * flows we impose some heuristics to decide which flows are likely to be worth
- * tracking. */
-static bool
-flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc)
-{
- struct dpif_backer *backer = miss->ofproto->backer;
- uint32_t hash;
+ dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
- switch (flow_miss_model) {
- case OFPROTO_HANDLE_MISS_AUTO:
- break;
- case OFPROTO_HANDLE_MISS_WITH_FACETS:
- return true;
- case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
- return false;
+ if (rule) {
+ rule_dpif_credit_stats(rule, &stats);
}
- if (!backer->governor) {
- size_t n_subfacets;
-
- n_subfacets = hmap_count(&backer->subfacets);
- if (n_subfacets * 2 <= flow_eviction_threshold) {
- return true;
- }
+ xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
+ xin.ofpacts = ofpacts;
+ xin.ofpacts_len = ofpacts_len;
+ xin.resubmit_stats = &stats;
+ xlate_actions(&xin, &xout);
- backer->governor = governor_create();
+ in_port = flow->in_port.ofp_port;
+ if (in_port == OFPP_NONE) {
+ in_port = OFPP_LOCAL;
}
+ execute.actions = ofpbuf_data(&xout.odp_actions);
+ execute.actions_len = ofpbuf_size(&xout.odp_actions);
+ execute.packet = packet;
+ execute.md.tunnel = flow->tunnel;
+ execute.md.skb_priority = flow->skb_priority;
+ execute.md.pkt_mark = flow->pkt_mark;
+ execute.md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
+ execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
- hash = flow_hash_in_wildcards(&miss->flow, wc, 0);
- return governor_should_install_flow(backer->governor, hash,
- list_size(&miss->packets));
-}
+ error = dpif_execute(ofproto->backer->dpif, &execute);
-/* Handles 'miss' without creating a facet or subfacet or creating any datapath
- * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'.
- * May add an "execute" operation to 'ops' and increment '*n_ops'. */
-static void
-handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout,
- struct flow_miss *miss,
- struct flow_miss_op *ops, size_t *n_ops)
-{
- struct ofpbuf *packet;
+ xlate_out_uninit(&xout);
- LIST_FOR_EACH (packet, list_node, &miss->packets) {
+ return error;
+}
- COVERAGE_INC(facet_suppress);
+void
+rule_dpif_credit_stats(struct rule_dpif *rule,
+ const struct dpif_flow_stats *stats)
+{
+ ovs_mutex_lock(&rule->stats_mutex);
+ rule->stats.n_packets += stats->n_packets;
+ rule->stats.n_bytes += stats->n_bytes;
+ rule->stats.used = MAX(rule->stats.used, stats->used);
+ ovs_mutex_unlock(&rule->stats_mutex);
+}
- handle_flow_miss_common(rule, packet, &miss->flow);
+bool
+rule_dpif_is_fail_open(const struct rule_dpif *rule)
+{
+ return is_fail_open_rule(&rule->up);
+}
- if (xout->slow) {
- struct xlate_in xin;
+bool
+rule_dpif_is_table_miss(const struct rule_dpif *rule)
+{
+ return rule_is_table_miss(&rule->up);
+}
- xlate_in_init(&xin, miss->ofproto, &miss->flow, rule, 0, packet);
- xlate_actions_for_side_effects(&xin);
- }
+bool
+rule_dpif_is_internal(const struct rule_dpif *rule)
+{
+ return rule_is_internal(&rule->up);
+}
- if (xout->odp_actions.size) {
- struct flow_miss_op *op = &ops[*n_ops];
- struct dpif_execute *execute = &op->dpif_op.u.execute;
+ovs_be64
+rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
+ OVS_REQUIRES(rule->up.mutex)
+{
+ return rule->up.flow_cookie;
+}
- init_flow_miss_execute_op(miss, packet, op);
- xlate_out_copy(&op->xout, xout);
- execute->actions = op->xout.odp_actions.data;
- execute->actions_len = op->xout.odp_actions.size;
- op->xout_garbage = true;
+void
+rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
+ uint16_t hard_timeout)
+{
+ ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
+}
- (*n_ops)++;
- }
- }
+/* Returns 'rule''s actions. The caller owns a reference on the returned
+ * actions and must eventually release it (with rule_actions_unref()) to avoid
+ * a memory leak. */
+const struct rule_actions *
+rule_dpif_get_actions(const struct rule_dpif *rule)
+{
+ return rule_get_actions(&rule->up);
}
-/* Handles 'miss', which matches 'facet'. May add any required datapath
- * operations to 'ops', incrementing '*n_ops' for each new op.
+/* Lookup 'flow' in table 0 of 'ofproto''s classifier.
+ * If 'wc' is non-null, sets the fields that were relevant as part of
+ * the lookup. Returns the table_id where a match or miss occurred.
*
- * 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.
+ * The return value will be zero unless there was a miss and
+ * OFPTC11_TABLE_MISS_CONTINUE is in effect for the sequence of tables
+ * where misses occur.
*
- * If non-null, 'stats' will be folded into 'facet'. */
-static void
-handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
- long long int now, struct dpif_flow_stats *stats,
- struct flow_miss_op *ops, size_t *n_ops)
-{
- struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
- enum subfacet_path want_path;
- struct subfacet *subfacet;
- struct ofpbuf *packet;
-
- subfacet = subfacet_create(facet, miss, now);
- want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
- if (stats) {
- subfacet_update_stats(subfacet, stats);
- }
-
- LIST_FOR_EACH (packet, list_node, &miss->packets) {
- struct flow_miss_op *op = &ops[*n_ops];
-
- handle_flow_miss_common(facet->rule, packet, &miss->flow);
-
- if (want_path != SF_FAST_PATH) {
- struct xlate_in xin;
+ * The rule is returned in '*rule', which is valid at least until the next
+ * RCU quiescent period. If the '*rule' needs to stay around longer,
+ * a non-zero 'take_ref' must be passed in to cause a reference to be taken
+ * on it before this returns. */
+uint8_t
+rule_dpif_lookup(struct ofproto_dpif *ofproto, struct flow *flow,
+ struct flow_wildcards *wc, struct rule_dpif **rule,
+ bool take_ref)
+{
+ enum rule_dpif_lookup_verdict verdict;
+ enum ofputil_port_config config = 0;
+ uint8_t table_id;
- xlate_in_init(&xin, ofproto, &miss->flow, facet->rule, 0, packet);
- xlate_actions_for_side_effects(&xin);
+ if (ofproto_dpif_get_enable_recirc(ofproto)) {
+ /* Always exactly match recirc_id since datapath supports
+ * recirculation. */
+ if (wc) {
+ wc->masks.recirc_id = UINT32_MAX;
}
- if (facet->xout.odp_actions.size) {
- struct dpif_execute *execute = &op->dpif_op.u.execute;
-
- init_flow_miss_execute_op(miss, packet, op);
- execute->actions = facet->xout.odp_actions.data,
- execute->actions_len = facet->xout.odp_actions.size;
- (*n_ops)++;
- }
+ /* Start looking up from internal table for post recirculation flows
+ * or packets. We can also simply send all, including normal flows
+ * or packets to the internal table. They will not match any post
+ * recirculation rules except the 'catch all' rule that resubmit
+ * them to table 0.
+ *
+ * As an optimization, we send normal flows and packets to table 0
+ * directly, saving one table lookup. */
+ table_id = flow->recirc_id ? TBL_INTERNAL : 0;
+ } else {
+ table_id = 0;
}
- if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
- struct flow_miss_op *op = &ops[(*n_ops)++];
- struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
+ verdict = rule_dpif_lookup_from_table(ofproto, flow, wc, true,
+ &table_id, rule, take_ref);
- subfacet->path = want_path;
-
- ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
- if (enable_megaflows) {
- odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
- &miss->flow, UINT32_MAX);
- }
+ switch (verdict) {
+ case RULE_DPIF_LOOKUP_VERDICT_MATCH:
+ return table_id;
+ case RULE_DPIF_LOOKUP_VERDICT_CONTROLLER: {
+ struct ofport_dpif *port;
- op->xout_garbage = false;
- op->dpif_op.type = DPIF_OP_FLOW_PUT;
- op->subfacet = subfacet;
- put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
- put->key = miss->key;
- put->key_len = miss->key_len;
- put->mask = op->mask.data;
- put->mask_len = op->mask.size;
-
- if (want_path == SF_FAST_PATH) {
- put->actions = facet->xout.odp_actions.data;
- put->actions_len = facet->xout.odp_actions.size;
- } else {
- compose_slow_path(ofproto, &miss->flow, facet->xout.slow,
- op->slow_stub, sizeof op->slow_stub,
- &put->actions, &put->actions_len);
+ port = get_ofp_port(ofproto, flow->in_port.ofp_port);
+ if (!port) {
+ VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
+ flow->in_port.ofp_port);
}
- put->stats = NULL;
+ config = port ? port->up.pp.config : 0;
+ break;
}
-}
-
-/* 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 dpif_flow_stats stats__;
- struct dpif_flow_stats *stats = &stats__;
- struct ofpbuf *packet;
- struct facet *facet;
- long long int now;
-
- now = time_msec();
- memset(stats, 0, sizeof *stats);
- stats->used = now;
- LIST_FOR_EACH (packet, list_node, &miss->packets) {
- stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow);
- stats->n_bytes += packet->size;
- stats->n_packets++;
- }
-
- facet = facet_lookup_valid(ofproto, &miss->flow);
- if (!facet) {
- struct flow_wildcards wc;
- struct rule_dpif *rule;
- struct xlate_out xout;
- struct xlate_in xin;
-
- flow_wildcards_init_catchall(&wc);
- rule = rule_dpif_lookup(ofproto, &miss->flow, &wc);
- rule_credit_stats(rule, stats);
-
- xlate_in_init(&xin, ofproto, &miss->flow, rule, stats->tcp_flags,
- NULL);
- xin.resubmit_stats = stats;
- xin.may_learn = true;
- xlate_actions(&xin, &xout);
- flow_wildcards_or(&xout.wc, &xout.wc, &wc);
-
- /* There does not exist a bijection between 'struct flow' and datapath
- * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
- * assumption used throughout the facet and subfacet handling code.
- * Since we have to handle these misses in userspace anyway, we simply
- * skip facet creation, avoiding the problem altogether. */
- if (miss->key_fitness == ODP_FIT_TOO_LITTLE
- || !flow_miss_should_make_facet(miss, &xout.wc)) {
- handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops);
- return;
+ case RULE_DPIF_LOOKUP_VERDICT_DROP:
+ config = OFPUTIL_PC_NO_PACKET_IN;
+ break;
+ case RULE_DPIF_LOOKUP_VERDICT_DEFAULT:
+ if (!connmgr_wants_packet_in_on_miss(ofproto->up.connmgr)) {
+ config = OFPUTIL_PC_NO_PACKET_IN;
}
-
- facet = facet_create(miss, rule, &xout, stats);
- stats = NULL;
+ break;
+ default:
+ OVS_NOT_REACHED();
}
- handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops);
-}
-static struct drop_key *
-drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
- size_t key_len)
-{
- struct drop_key *drop_key;
-
- HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
- &backer->drop_keys) {
- if (drop_key->key_len == key_len
- && !memcmp(drop_key->key, key, key_len)) {
- return drop_key;
- }
- }
- return NULL;
+ choose_miss_rule(config, ofproto->miss_rule,
+ ofproto->no_packet_in_rule, rule, take_ref);
+ return table_id;
}
-static void
-drop_key_clear(struct dpif_backer *backer)
+/* The returned rule is valid at least until the next RCU quiescent period.
+ * If the '*rule' needs to stay around longer, a non-zero 'take_ref' must be
+ * passed in to cause a reference to be taken on it before this returns. */
+static struct rule_dpif *
+rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, uint8_t table_id,
+ const struct flow *flow, struct flow_wildcards *wc,
+ bool take_ref)
{
- static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
- struct drop_key *drop_key, *next;
+ struct classifier *cls = &ofproto->up.tables[table_id].cls;
+ const struct cls_rule *cls_rule;
+ struct rule_dpif *rule;
- HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
- int error;
+ fat_rwlock_rdlock(&cls->rwlock);
+ if (ofproto->up.frag_handling != OFPC_FRAG_NX_MATCH) {
+ if (wc) {
+ memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
+ if (is_ip_any(flow)) {
+ wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
+ }
+ }
- error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
- NULL);
- if (error && !VLOG_DROP_WARN(&rl)) {
- struct ds ds = DS_EMPTY_INITIALIZER;
- odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
- VLOG_WARN("Failed to delete drop key (%s) (%s)",
- ovs_strerror(error), ds_cstr(&ds));
- ds_destroy(&ds);
+ if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
+ if (ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
+ /* We must pretend that transport ports are unavailable. */
+ struct flow ofpc_normal_flow = *flow;
+ ofpc_normal_flow.tp_src = htons(0);
+ ofpc_normal_flow.tp_dst = htons(0);
+ cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
+ } else {
+ /* Must be OFPC_FRAG_DROP (we don't have OFPC_FRAG_REASM). */
+ cls_rule = &ofproto->drop_frags_rule->up.cr;
+ }
+ } else {
+ cls_rule = classifier_lookup(cls, flow, wc);
}
+ } else {
+ cls_rule = classifier_lookup(cls, flow, wc);
+ }
- hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
- free(drop_key->key);
- free(drop_key);
+ rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
+ if (take_ref) {
+ rule_dpif_ref(rule);
}
+ fat_rwlock_unlock(&cls->rwlock);
+
+ return rule;
}
-/* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
- * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
- * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
- * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
- * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
- * 'packet' ingressed.
+/* Look up 'flow' in 'ofproto''s classifier starting from table '*table_id'.
+ * Stores the rule that was found in '*rule', or NULL if none was found.
+ * Updates 'wc', if nonnull, to reflect the fields that were used during the
+ * lookup.
*
- * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
- * 'flow''s in_port to OFPP_NONE.
+ * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
+ * if none is found then the table miss configuration for that table is
+ * honored, which can result in additional lookups in other OpenFlow tables.
+ * In this case the function updates '*table_id' to reflect the final OpenFlow
+ * table that was searched.
*
- * 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).
+ * If 'honor_table_miss' is false, then only one table lookup occurs, in
+ * '*table_id'.
*
- * Similarly, this function also includes some logic to help with tunnels. It
- * may modify 'flow' as necessary to make the tunneling implementation
- * transparent to the upcall processing logic.
+ * Returns:
*
- * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
- * or some other positive errno if there are other problems. */
-static int
-ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
- const struct nlattr *key, size_t key_len,
- struct flow *flow, enum odp_key_fitness *fitnessp,
- struct ofproto_dpif **ofproto, odp_port_t *odp_in_port)
-{
- const struct ofport_dpif *port;
- enum odp_key_fitness fitness;
- int error = ENODEV;
-
- fitness = odp_flow_key_to_flow(key, key_len, flow);
- if (fitness == ODP_FIT_ERROR) {
- error = EINVAL;
- goto exit;
- }
-
- if (odp_in_port) {
- *odp_in_port = flow->in_port.odp_port;
- }
-
- port = (tnl_port_should_receive(flow)
- ? tnl_port_receive(flow)
- : odp_port_to_ofport(backer, flow->in_port.odp_port));
- flow->in_port.ofp_port = port ? port->up.ofp_port : OFPP_NONE;
- if (!port) {
- goto exit;
- }
-
- /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
- * it's theoretically possible that we'll receive an ofport belonging to an
- * entirely different datapath. In practice, this can't happen because no
- * platforms has two separate datapaths which each support tunneling. */
- ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
-
- if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
- if (packet) {
- /* Make the packet resemble the flow, so that it gets sent to
- * an OpenFlow controller properly, so that it looks correct
- * for sFlow, and so that flow_extract() will get the correct
- * vlan_tci if it is called on 'packet'.
- *
- * The allocated space inside 'packet' probably also contains
- * 'key', that is, both 'packet' and 'key' are probably part of
- * a struct dpif_upcall (see the large comment on that
- * structure definition), so pushing data on 'packet' is in
- * general not a good idea since it could overwrite 'key' or
- * free it as a side effect. However, it's OK in this special
- * case because we know that 'packet' is inside a Netlink
- * attribute: pushing 4 bytes will just overwrite the 4-byte
- * "struct nlattr", which is fine since we don't need that
- * header anymore. */
- eth_push_vlan(packet, flow->vlan_tci);
- }
- /* We can't reproduce 'key' from 'flow'. */
- fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
- }
- error = 0;
-
- if (ofproto) {
- *ofproto = ofproto_dpif_cast(port->up.ofproto);
- }
-
-exit:
- if (fitnessp) {
- *fitnessp = fitness;
- }
- return error;
-}
-
-static void
-handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
- size_t n_upcalls)
-{
- struct dpif_upcall *upcall;
- struct flow_miss *miss;
- struct flow_miss misses[FLOW_MISS_MAX_BATCH];
- struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
- struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
- struct hmap todo;
- int n_misses;
- size_t n_ops;
- size_t i;
-
- if (!n_upcalls) {
- return;
- }
-
- /* Construct the to-do list.
- *
- * This just amounts to extracting the flow from each packet and sticking
- * the packets that have the same flow in the same "flow_miss" structure so
- * that we can process them together. */
- hmap_init(&todo);
- n_misses = 0;
- for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
- struct flow_miss *miss = &misses[n_misses];
- struct flow_miss *existing_miss;
- struct ofproto_dpif *ofproto;
- odp_port_t odp_in_port;
- struct flow flow;
- uint32_t hash;
- int error;
-
- error = ofproto_receive(backer, upcall->packet, upcall->key,
- upcall->key_len, &flow, &miss->key_fitness,
- &ofproto, &odp_in_port);
- if (error == ENODEV) {
- struct drop_key *drop_key;
-
- /* Received packet on datapath port for which we couldn't
- * associate an ofproto. This can happen if a port is removed
- * while traffic is being received. Print a rate-limited message
- * in case it happens frequently. Install a drop flow so
- * that future packets of the flow are inexpensively dropped
- * in the kernel. */
- VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
- "%"PRIu32, odp_in_port);
-
- drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
- if (!drop_key) {
- drop_key = xmalloc(sizeof *drop_key);
- drop_key->key = xmemdup(upcall->key, upcall->key_len);
- drop_key->key_len = upcall->key_len;
-
- hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
- hash_bytes(drop_key->key, drop_key->key_len, 0));
- dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
- drop_key->key, drop_key->key_len,
- NULL, 0, NULL, 0, NULL);
- }
- continue;
- }
- if (error) {
- continue;
- }
-
- ofproto->n_missed++;
- flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
- &flow.tunnel, &flow.in_port, &miss->flow);
-
- /* Add other packets to a to-do list. */
- hash = flow_hash(&miss->flow, 0);
- existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
- if (!existing_miss) {
- hmap_insert(&todo, &miss->hmap_node, hash);
- miss->ofproto = ofproto;
- miss->key = upcall->key;
- miss->key_len = upcall->key_len;
- miss->upcall_type = upcall->type;
- 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 (miss, hmap_node, &todo) {
- handle_flow_miss(miss, flow_miss_ops, &n_ops);
- }
- ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
-
- /* Execute batch. */
- for (i = 0; i < n_ops; i++) {
- dpif_ops[i] = &flow_miss_ops[i].dpif_op;
- }
- dpif_operate(backer->dpif, dpif_ops, n_ops);
-
- for (i = 0; i < n_ops; i++) {
- if (dpif_ops[i]->error != 0
- && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
- && flow_miss_ops[i].subfacet) {
- struct subfacet *subfacet = flow_miss_ops[i].subfacet;
-
- COVERAGE_INC(subfacet_install_fail);
-
- subfacet->path = SF_NOT_INSTALLED;
- }
-
- /* Free memory. */
- if (flow_miss_ops[i].xout_garbage) {
- xlate_out_uninit(&flow_miss_ops[i].xout);
- }
- }
- hmap_destroy(&todo);
-}
-
-static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
- IPFIX_UPCALL }
-classify_upcall(const struct dpif_upcall *upcall)
-{
- size_t userdata_len;
- union user_action_cookie cookie;
-
- /* First look at the upcall type. */
- 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. */
- if (!upcall->userdata) {
- VLOG_WARN_RL(&rl, "action upcall missing cookie");
- return BAD_UPCALL;
- }
- userdata_len = nl_attr_get_size(upcall->userdata);
- if (userdata_len < sizeof cookie.type
- || userdata_len > sizeof cookie) {
- VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
- userdata_len);
- return BAD_UPCALL;
- }
- memset(&cookie, 0, sizeof cookie);
- memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
- if (userdata_len == sizeof cookie.sflow
- && cookie.type == USER_ACTION_COOKIE_SFLOW) {
- return SFLOW_UPCALL;
- } else if (userdata_len == sizeof cookie.slow_path
- && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
- return MISS_UPCALL;
- } else if (userdata_len == sizeof cookie.flow_sample
- && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
- return FLOW_SAMPLE_UPCALL;
- } else if (userdata_len == sizeof cookie.ipfix
- && cookie.type == USER_ACTION_COOKIE_IPFIX) {
- return IPFIX_UPCALL;
- } else {
- VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
- " and size %zu", cookie.type, userdata_len);
- return BAD_UPCALL;
- }
-}
-
-static void
-handle_sflow_upcall(struct dpif_backer *backer,
- const struct dpif_upcall *upcall)
-{
- struct ofproto_dpif *ofproto;
- union user_action_cookie cookie;
- struct flow flow;
- odp_port_t odp_in_port;
-
- if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
- &flow, NULL, &ofproto, &odp_in_port)
- || !ofproto->sflow) {
- return;
- }
-
- memset(&cookie, 0, sizeof cookie);
- memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
- dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
- odp_in_port, &cookie);
-}
-
-static void
-handle_flow_sample_upcall(struct dpif_backer *backer,
- const struct dpif_upcall *upcall)
-{
- struct ofproto_dpif *ofproto;
- union user_action_cookie cookie;
- struct flow flow;
-
- if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
- &flow, NULL, &ofproto, NULL)
- || !ofproto->ipfix) {
- return;
- }
-
- memset(&cookie, 0, sizeof cookie);
- memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
-
- /* The flow reflects exactly the contents of the packet. Sample
- * the packet using it. */
- dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
- cookie.flow_sample.collector_set_id,
- cookie.flow_sample.probability,
- cookie.flow_sample.obs_domain_id,
- cookie.flow_sample.obs_point_id);
-}
-
-static void
-handle_ipfix_upcall(struct dpif_backer *backer,
- const struct dpif_upcall *upcall)
-{
- struct ofproto_dpif *ofproto;
- struct flow flow;
-
- if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
- &flow, NULL, &ofproto, NULL)
- || !ofproto->ipfix) {
- return;
- }
-
- /* The flow reflects exactly the contents of the packet. Sample
- * the packet using it. */
- dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
-}
-
-static int
-handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
-{
- 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;
-
- ovs_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 FLOW_SAMPLE_UPCALL:
- handle_flow_sample_upcall(backer, upcall);
- ofpbuf_uninit(buf);
- break;
-
- case IPFIX_UPCALL:
- handle_ipfix_upcall(backer, upcall);
- ofpbuf_uninit(buf);
- break;
-
- case BAD_UPCALL:
- ofpbuf_uninit(buf);
- break;
- }
- }
-
- /* 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 subfacet_max_idle(const struct dpif_backer *);
-static void update_stats(struct dpif_backer *);
-static void rule_expire(struct rule_dpif *);
-static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
-
-/* This function is called periodically by run(). Its job is to collect
- * updates for the flows that have been installed into the datapath, most
- * importantly when they last were used, and then use that information to
- * expire flows that have not been used recently.
- *
- * Returns the number of milliseconds after which it should be called again. */
-static int
-expire(struct dpif_backer *backer)
-{
- struct ofproto_dpif *ofproto;
- size_t n_subfacets;
- int max_idle;
-
- /* Periodically clear out the drop keys in an effort to keep them
- * relatively few. */
- drop_key_clear(backer);
-
- /* Update stats for each flow in the backer. */
- update_stats(backer);
-
- n_subfacets = hmap_count(&backer->subfacets);
- if (n_subfacets) {
- struct subfacet *subfacet;
- long long int total, now;
-
- total = 0;
- now = time_msec();
- HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
- total += now - subfacet->created;
- }
- backer->avg_subfacet_life += total / n_subfacets;
- }
- backer->avg_subfacet_life /= 2;
-
- backer->avg_n_subfacet += n_subfacets;
- backer->avg_n_subfacet /= 2;
-
- backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
-
- max_idle = subfacet_max_idle(backer);
- expire_subfacets(backer, max_idle);
-
- HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
- struct rule *rule, *next_rule;
-
- if (ofproto->backer != backer) {
- continue;
- }
-
- /* Expire OpenFlow flows whose idle_timeout or hard_timeout
- * has passed. */
- LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
- &ofproto->up.expirable) {
- rule_expire(rule_dpif_cast(rule));
- }
-
- /* All outstanding data in existing flows has been accounted, so it's a
- * good time to do bond rebalancing. */
- if (ofproto->has_bonded_bundles) {
- struct ofbundle *bundle;
-
- HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
- if (bundle->bond) {
- bond_rebalance(bundle->bond, &backer->revalidate_set);
- }
- }
- }
- }
-
- 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;
- struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
- struct dpif_flow_stats diff;
-
- diff.tcp_flags = stats->tcp_flags;
- diff.used = stats->used;
-
- if (stats->n_packets >= subfacet->dp_packet_count) {
- diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
- } else {
- VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
- diff.n_packets = 0;
- }
-
- if (stats->n_bytes >= subfacet->dp_byte_count) {
- diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
- } else {
- VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
- diff.n_bytes = 0;
- }
-
- ofproto->n_hit += diff.n_packets;
- subfacet->dp_packet_count = stats->n_packets;
- subfacet->dp_byte_count = stats->n_bytes;
- subfacet_update_stats(subfacet, &diff);
-
- if (facet->accounted_bytes < facet->byte_count) {
- facet_learn(facet);
- facet_account(facet);
- facet->accounted_bytes = facet->byte_count;
- }
-}
-
-/* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
- * about, or a flow that shouldn't be installed but was anyway. Delete it. */
-static void
-delete_unexpected_flow(struct dpif_backer *backer,
- 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: %s", ds_cstr(&s));
- ds_destroy(&s);
- }
-
- COVERAGE_INC(facet_unexpected);
- dpif_flow_del(backer->dpif, key, key_len, NULL);
-}
-
-/* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
- *
- * This function also pushes statistics updates to rules which each facet
- * resubmits into. Generally these statistics will be accurate. However, if a
- * facet changes the rule it resubmits into at some time in between
- * update_stats() runs, it is possible that statistics accrued to the
- * old rule will be incorrectly attributed to the new rule. This could be
- * avoided by calling update_stats() whenever rules are created or
- * deleted. However, the performance impact of making so many calls to the
- * datapath do not justify the benefit of having perfectly accurate statistics.
- *
- * In addition, this function maintains per ofproto flow hit counts. The patch
- * port is not treated specially. e.g. A packet ingress from br0 patched into
- * br1 will increase the hit count of br0 by 1, however, does not affect
- * the hit or miss counts of br1.
- */
-static void
-update_stats(struct dpif_backer *backer)
-{
- const struct dpif_flow_stats *stats;
- struct dpif_flow_dump dump;
- const struct nlattr *key, *mask;
- size_t key_len, mask_len;
-
- dpif_flow_dump_start(&dump, backer->dpif);
- while (dpif_flow_dump_next(&dump, &key, &key_len,
- &mask, &mask_len, NULL, NULL, &stats)) {
- struct subfacet *subfacet;
- uint32_t key_hash;
-
- key_hash = odp_flow_key_hash(key, key_len);
- subfacet = subfacet_find(backer, key, key_len, key_hash);
- switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
- case SF_FAST_PATH:
- update_subfacet_stats(subfacet, stats);
- break;
-
- case SF_SLOW_PATH:
- /* Stats are updated per-packet. */
- break;
-
- case SF_NOT_INSTALLED:
- default:
- delete_unexpected_flow(backer, key, key_len);
- break;
- }
- run_fast_rl();
- }
- dpif_flow_dump_done(&dump);
-
- update_moving_averages(backer);
-}
-
-/* Calculates and returns the number of milliseconds of idle time after which
- * 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
-subfacet_max_idle(const struct dpif_backer *backer)
-{
- /*
- * Idle time histogram.
- *
- * 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 subfacets increases, the memory required to maintain
- * statistics about them in userspace and in the kernel becomes
- * 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 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 subfacets (but at least
- * flow_eviction_threshold flows) are kept cached. At least
- * the most-recently-used bucket of subfacets is kept, so actually an
- * arbitrary number of subfacets can be kept in any given expiration run
- * (though the next run will delete most of those unless they receive
- * additional data).
- *
- * 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 subfacet *subfacet;
- long long int now;
- int i;
-
- total = hmap_count(&backer->subfacets);
- if (total <= flow_eviction_threshold) {
- return N_BUCKETS * BUCKET_WIDTH;
- }
-
- /* Build histogram. */
- now = time_msec();
- HMAP_FOR_EACH (subfacet, hmap_node, &backer->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);
- buckets[bucket]++;
- }
-
- /* Find the first bucket whose flows should be expired. */
- subtotal = bucket = 0;
- do {
- subtotal += buckets[bucket++];
- } while (bucket < N_BUCKETS &&
- subtotal < MAX(flow_eviction_threshold, total / 100));
-
- if (VLOG_IS_DBG_ENABLED()) {
- struct ds s;
-
- ds_init(&s);
- ds_put_cstr(&s, "keep");
- for (i = 0; i < N_BUCKETS; i++) {
- if (i == bucket) {
- ds_put_cstr(&s, ", drop");
- }
- if (buckets[i]) {
- ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
- }
- }
- VLOG_INFO("%s (msec:count)", ds_cstr(&s));
- ds_destroy(&s);
- }
-
- return bucket * BUCKET_WIDTH;
-}
-
-static void
-expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
-{
- /* Cutoff time for most flows. */
- long long int normal_cutoff = time_msec() - dp_max_idle;
-
- /* 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,
- &backer->subfacets) {
- long long int cutoff;
-
- cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
- | SLOW_STP)
- ? special_cutoff
- : normal_cutoff);
- if (subfacet->used < cutoff) {
- if (subfacet->path != SF_NOT_INSTALLED) {
- batch[n_batch++] = subfacet;
- if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
- subfacet_destroy_batch(backer, batch, n_batch);
- n_batch = 0;
- }
- } else {
- subfacet_destroy(subfacet);
- }
- }
- }
-
- if (n_batch > 0) {
- subfacet_destroy_batch(backer, batch, n_batch);
- }
-}
-
-/* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
- * then delete it entirely. */
-static void
-rule_expire(struct rule_dpif *rule)
-{
- 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
- && now > rule->up.used + rule->up.idle_timeout * 1000) {
- reason = OFPRR_IDLE_TIMEOUT;
- } else {
- return;
- }
-
- COVERAGE_INC(ofproto_dpif_expired);
-
- /* Update stats. (This is a no-op if the rule expired due to an idle
- * timeout, because that only happens when the rule has no facets left.) */
- LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
- facet_remove(facet);
- }
-
- /* Get rid of the rule. */
- ofproto_rule_expire(&rule->up, reason);
-}
-\f
-/* Facets. */
-
-/* Creates and returns a new facet based on 'miss'.
- *
- * The caller must already have determined that no facet with an identical
- * 'miss->flow' exists in 'miss->ofproto'.
- *
- * 'rule' and 'xout' must have been created based on 'miss'.
- *
- * 'facet'' statistics are initialized based on 'stats'.
- *
- * The facet will initially have no subfacets. The caller should create (at
- * least) one subfacet with subfacet_create(). */
-static struct facet *
-facet_create(const struct flow_miss *miss, struct rule_dpif *rule,
- struct xlate_out *xout, struct dpif_flow_stats *stats)
-{
- struct ofproto_dpif *ofproto = miss->ofproto;
- struct facet *facet;
- struct match match;
-
- facet = xzalloc(sizeof *facet);
- facet->packet_count = facet->prev_packet_count = stats->n_packets;
- facet->byte_count = facet->prev_byte_count = stats->n_bytes;
- facet->tcp_flags = stats->tcp_flags;
- facet->used = stats->used;
- facet->flow = miss->flow;
- facet->learn_rl = time_msec() + 500;
- facet->rule = rule;
-
- list_push_back(&facet->rule->facets, &facet->list_node);
- list_init(&facet->subfacets);
- netflow_flow_init(&facet->nf_flow);
- netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
-
- xlate_out_copy(&facet->xout, xout);
-
- match_init(&match, &facet->flow, &facet->xout.wc);
- cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
- classifier_insert(&ofproto->facets, &facet->cr);
-
- facet->nf_flow.output_iface = facet->xout.nf_output_iface;
-
- return facet;
-}
-
-static void
-facet_free(struct facet *facet)
-{
- if (facet) {
- xlate_out_uninit(&facet->xout);
- free(facet);
- }
-}
-
-/* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
- * 'packet', which arrived on 'in_port'. */
-static bool
-execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
- const struct nlattr *odp_actions, size_t actions_len,
- struct ofpbuf *packet)
-{
- struct odputil_keybuf keybuf;
- struct ofpbuf key;
- int error;
-
- ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
- odp_flow_key_from_flow(&key, flow,
- ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
-
- error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
- odp_actions, actions_len, packet);
- return !error;
-}
-
-/* Remove 'facet' from its ofproto and free up the associated memory:
- *
- * - If 'facet' was installed in the datapath, uninstalls it and updates its
- * rule's statistics, via subfacet_uninstall().
- *
- * - Removes 'facet' from its rule and from ofproto->facets.
- */
-static void
-facet_remove(struct facet *facet)
-{
- struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
- struct subfacet *subfacet, *next_subfacet;
-
- ovs_assert(!list_is_empty(&facet->subfacets));
-
- /* First uninstall all of the subfacets to get final statistics. */
- LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
- subfacet_uninstall(subfacet);
- }
-
- /* 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);
-
- /* Now we're really all done so destroy everything. */
- LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
- &facet->subfacets) {
- subfacet_destroy__(subfacet);
- }
- classifier_remove(&ofproto->facets, &facet->cr);
- cls_rule_destroy(&facet->cr);
- list_remove(&facet->list_node);
- facet_free(facet);
-}
-
-/* Feed information from 'facet' back into the learning table to keep it in
- * sync with what is actually flowing through the datapath. */
-static void
-facet_learn(struct facet *facet)
-{
- long long int now = time_msec();
-
- if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
- return;
- }
-
- facet->learn_rl = now + 500;
-
- if (!facet->xout.has_learn
- && !facet->xout.has_normal
- && (!facet->xout.has_fin_timeout
- || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
- return;
- }
-
- facet_push_stats(facet, true);
-}
-
-static void
-facet_account(struct facet *facet)
-{
- struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
- const struct nlattr *a;
- unsigned int left;
- ovs_be16 vlan_tci;
- uint64_t n_bytes;
-
- if (!facet->xout.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.)
- *
- * We use the actions from an arbitrary subfacet because they should all
- * be equally valid for our purpose. */
- vlan_tci = facet->flow.vlan_tci;
- NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
- facet->xout.odp_actions.size) {
- const struct ovs_action_push_vlan *vlan;
- struct ofport_dpif *port;
-
- switch (nl_attr_type(a)) {
- case OVS_ACTION_ATTR_OUTPUT:
- port = get_odp_port(ofproto, nl_attr_get_odp_port(a));
- if (port && port->bundle && port->bundle->bond) {
- bond_account(port->bundle->bond, &facet->flow,
- vlan_tci_to_vid(vlan_tci), n_bytes);
- }
- break;
-
- case OVS_ACTION_ATTR_POP_VLAN:
- vlan_tci = htons(0);
- break;
-
- case OVS_ACTION_ATTR_PUSH_VLAN:
- vlan = nl_attr_get(a);
- vlan_tci = vlan->vlan_tci;
- break;
- }
- }
-}
-
-/* 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)
-{
- if (facet) {
- const struct rule *rule = &facet->rule->up;
- const struct ofpact *ofpacts = rule->ofpacts;
- size_t ofpacts_len = rule->ofpacts_len;
-
- if (ofpacts_len > 0 &&
- ofpacts->type == OFPACT_CONTROLLER &&
- ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
- return true;
- }
- }
- return false;
-}
-
-/* Folds all of 'facet''s statistics into its rule. Also updates the
- * accounting ofhook and emits a NetFlow expiration if appropriate. All of
- * '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 facet *facet)
-{
- struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
- struct subfacet *subfacet;
-
- LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
- ovs_assert(!subfacet->dp_byte_count);
- ovs_assert(!subfacet->dp_packet_count);
- }
-
- facet_push_stats(facet, false);
- if (facet->accounted_bytes < facet->byte_count) {
- facet_account(facet);
- facet->accounted_bytes = facet->byte_count;
- }
-
- if (ofproto->netflow && !facet_is_controller_flow(facet)) {
- struct ofexpired expired;
- 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);
- }
-
- /* Reset counters to prevent double counting if 'facet' ever gets
- * reinstalled. */
- facet_reset_counters(facet);
-
- netflow_flow_clear(&facet->nf_flow);
- facet->tcp_flags = 0;
-}
-
-/* Searches 'ofproto''s table of facets for one which would be responsible for
- * 'flow'. Returns it if found, otherwise a null pointer.
- *
- * The returned facet might need revalidation; use facet_lookup_valid()
- * instead if that is important. */
-static struct facet *
-facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
-{
- struct cls_rule *cr = classifier_lookup(&ofproto->facets, flow, NULL);
- return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
-}
-
-/* Searches 'ofproto''s table of facets for one capable that covers
- * 'flow'. Returns it if found, otherwise a null pointer.
- *
- * The returned facet is guaranteed to be valid. */
-static struct facet *
-facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
-{
- struct facet *facet;
-
- facet = facet_find(ofproto, flow);
- if (facet
- && (ofproto->backer->need_revalidate
- || tag_set_intersects(&ofproto->backer->revalidate_set,
- facet->xout.tags))
- && !facet_revalidate(facet)) {
- return NULL;
- }
-
- return facet;
-}
-
-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);
-
- struct xlate_out xout;
- struct xlate_in xin;
-
- struct rule_dpif *rule;
- bool ok;
-
- /* Check the rule for consistency. */
- rule = rule_dpif_lookup(ofproto, &facet->flow, NULL);
- if (rule != facet->rule) {
- if (!VLOG_DROP_WARN(&rl)) {
- struct ds s = DS_EMPTY_INITIALIZER;
-
- 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);
- }
- return false;
- }
-
- /* Check the datapath actions for consistency. */
- xlate_in_init(&xin, ofproto, &facet->flow, rule, 0, NULL);
- xlate_actions(&xin, &xout);
-
- ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
- && facet->xout.slow == xout.slow;
- if (!ok && !VLOG_DROP_WARN(&rl)) {
- struct ds s = DS_EMPTY_INITIALIZER;
-
- flow_format(&s, &facet->flow);
- ds_put_cstr(&s, ": inconsistency in facet");
-
- if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
- ds_put_cstr(&s, " (actions were: ");
- format_odp_actions(&s, facet->xout.odp_actions.data,
- facet->xout.odp_actions.size);
- ds_put_cstr(&s, ") (correct actions: ");
- format_odp_actions(&s, xout.odp_actions.data,
- xout.odp_actions.size);
- ds_put_char(&s, ')');
- }
-
- if (facet->xout.slow != xout.slow) {
- ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
- }
-
- VLOG_WARN("%s", ds_cstr(&s));
- ds_destroy(&s);
- }
- xlate_out_uninit(&xout);
-
- return ok;
-}
-
-/* Re-searches the classifier for 'facet':
+ * - RULE_DPIF_LOOKUP_VERDICT_MATCH if a rule (in '*rule') was found.
*
- * - If the rule found is different from 'facet''s current rule, moves
- * 'facet' to the new rule and recompiles its actions.
+ * - RULE_OFPTC_TABLE_MISS_CONTROLLER if no rule was found and either:
+ * + 'honor_table_miss' is false
+ * + a table miss configuration specified that the packet should be
+ * sent to the controller in this case.
*
- * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
- * where it is and recompiles its actions anyway.
+ * - RULE_DPIF_LOOKUP_VERDICT_DROP if no rule was found, 'honor_table_miss'
+ * is true and a table miss configuration specified that the packet
+ * should be dropped in this case.
*
- * - If any of 'facet''s subfacets correspond to a new flow according to
- * ofproto_receive(), 'facet' is removed.
+ * - RULE_DPIF_LOOKUP_VERDICT_DEFAULT if no rule was found,
+ * 'honor_table_miss' is true and a table miss configuration has
+ * not been specified in this case.
*
- * Returns true if 'facet' is still valid. False if 'facet' was removed. */
-static bool
-facet_revalidate(struct facet *facet)
-{
- struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
- struct rule_dpif *new_rule;
- struct subfacet *subfacet;
- struct flow_wildcards wc;
- struct xlate_out xout;
- struct xlate_in xin;
-
- COVERAGE_INC(facet_revalidate);
-
- /* Check that child subfacets still correspond to this facet. Tunnel
- * configuration changes could cause a subfacet's OpenFlow in_port to
- * change. */
- LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
- struct ofproto_dpif *recv_ofproto;
- struct flow recv_flow;
- int error;
-
- error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
- subfacet->key_len, &recv_flow, NULL,
- &recv_ofproto, NULL);
- if (error
- || recv_ofproto != ofproto
- || facet != facet_find(ofproto, &recv_flow)) {
- facet_remove(facet);
- return false;
- }
- }
-
- flow_wildcards_init_catchall(&wc);
- new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc);
-
- /* Calculate new datapath actions.
- *
- * We do not modify any 'facet' state yet, because we might need to, e.g.,
- * emit a NetFlow expiration and, if so, we need to have the old state
- * around to properly compose it. */
- xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
- xlate_actions(&xin, &xout);
- flow_wildcards_or(&xout.wc, &xout.wc, &wc);
-
- /* A facet's slow path reason should only change under dramatic
- * circumstances. Rather than try to update everything, it's simpler to
- * remove the facet and start over.
- *
- * More importantly, if a facet's wildcards change, it will be relatively
- * difficult to figure out if its subfacets still belong to it, and if not
- * which facet they may belong to. Again, to avoid the complexity, we
- * simply give up instead. */
- if (facet->xout.slow != xout.slow
- || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
- facet_remove(facet);
- xlate_out_uninit(&xout);
- return false;
- }
-
- if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
- LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
- if (subfacet->path == SF_FAST_PATH) {
- struct dpif_flow_stats stats;
-
- subfacet_install(subfacet, &xout.odp_actions, &stats);
- subfacet_update_stats(subfacet, &stats);
- }
- }
-
- facet_flush_stats(facet);
-
- ofpbuf_clear(&facet->xout.odp_actions);
- ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
- xout.odp_actions.size);
- }
-
- /* Update 'facet' now that we've taken care of all the old state. */
- facet->xout.tags = xout.tags;
- facet->xout.slow = xout.slow;
- facet->xout.has_learn = xout.has_learn;
- facet->xout.has_normal = xout.has_normal;
- facet->xout.has_fin_timeout = xout.has_fin_timeout;
- facet->xout.nf_output_iface = xout.nf_output_iface;
- facet->xout.mirrors = xout.mirrors;
- facet->nf_flow.output_iface = facet->xout.nf_output_iface;
-
- if (facet->rule != new_rule) {
- COVERAGE_INC(facet_changed_rule);
- list_remove(&facet->list_node);
- list_push_back(&new_rule->facets, &facet->list_node);
- facet->rule = new_rule;
- facet->used = new_rule->up.created;
- facet->prev_used = facet->used;
- }
-
- xlate_out_uninit(&xout);
- return true;
-}
-
-static void
-facet_reset_counters(struct facet *facet)
-{
- facet->packet_count = 0;
- facet->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, bool may_learn)
-{
- struct dpif_flow_stats stats;
-
- ovs_assert(facet->packet_count >= facet->prev_packet_count);
- ovs_assert(facet->byte_count >= facet->prev_byte_count);
- ovs_assert(facet->used >= facet->prev_used);
-
- stats.n_packets = facet->packet_count - facet->prev_packet_count;
- stats.n_bytes = facet->byte_count - facet->prev_byte_count;
- stats.used = facet->used;
- stats.tcp_flags = facet->tcp_flags;
-
- if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
- struct ofproto_dpif *ofproto =
- ofproto_dpif_cast(facet->rule->up.ofproto);
-
- struct ofport_dpif *in_port;
- struct xlate_in xin;
-
- facet->prev_packet_count = facet->packet_count;
- facet->prev_byte_count = facet->byte_count;
- facet->prev_used = facet->used;
-
- in_port = get_ofp_port(ofproto, facet->flow.in_port.ofp_port);
- if (in_port && in_port->is_tunnel) {
- netdev_vport_inc_rx(in_port->up.netdev, &stats);
- }
-
- rule_credit_stats(facet->rule, &stats);
- netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
- facet->used);
- netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
- update_mirror_stats(ofproto, facet->xout.mirrors, stats.n_packets,
- stats.n_bytes);
-
- xlate_in_init(&xin, ofproto, &facet->flow, facet->rule,
- stats.tcp_flags, NULL);
- xin.resubmit_stats = &stats;
- xin.may_learn = may_learn;
- xlate_actions_for_side_effects(&xin);
- }
-}
-
-static void
-push_all_stats__(bool run_fast)
-{
- static long long int rl = LLONG_MIN;
- struct ofproto_dpif *ofproto;
-
- if (time_msec() < rl) {
- return;
- }
-
- HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
- struct cls_cursor cursor;
- struct facet *facet;
-
- cls_cursor_init(&cursor, &ofproto->facets, NULL);
- CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
- facet_push_stats(facet, false);
- if (run_fast) {
- run_fast_rl();
- }
- }
- }
-
- rl = time_msec() + 100;
-}
-
-static void
-push_all_stats(void)
-{
- push_all_stats__(true);
-}
-
-void
-rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
-{
- rule->packet_count += stats->n_packets;
- rule->byte_count += stats->n_bytes;
- ofproto_rule_update_used(&rule->up, stats->used);
-}
-\f
-/* Subfacets. */
-
-static struct subfacet *
-subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
- size_t key_len, uint32_t key_hash)
-{
- struct subfacet *subfacet;
-
- HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
- &backer->subfacets) {
- if (subfacet->key_len == key_len
- && !memcmp(key, subfacet->key, key_len)) {
- return subfacet;
- }
- }
-
- return NULL;
-}
+ * The rule is returned in '*rule', which is valid at least until the next
+ * RCU quiescent period. If the '*rule' needs to stay around longer,
+ * a non-zero 'take_ref' must be passed in to cause a reference to be taken
+ * on it before this returns. */
+enum rule_dpif_lookup_verdict
+rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto,
+ const struct flow *flow,
+ struct flow_wildcards *wc,
+ bool honor_table_miss,
+ uint8_t *table_id, struct rule_dpif **rule,
+ bool take_ref)
+{
+ uint8_t next_id;
+
+ for (next_id = *table_id;
+ next_id < ofproto->up.n_tables;
+ next_id++, next_id += (next_id == TBL_INTERNAL))
+ {
+ *table_id = next_id;
+ *rule = rule_dpif_lookup_in_table(ofproto, *table_id, flow, wc,
+ take_ref);
+ if (*rule) {
+ return RULE_DPIF_LOOKUP_VERDICT_MATCH;
+ } else if (!honor_table_miss) {
+ return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
+ } else {
+ switch (ofproto_table_get_config(&ofproto->up, *table_id)) {
+ case OFPROTO_TABLE_MISS_CONTINUE:
+ break;
-/* 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. */
-static struct subfacet *
-subfacet_create(struct facet *facet, struct flow_miss *miss,
- long long int now)
-{
- struct dpif_backer *backer = miss->ofproto->backer;
- enum odp_key_fitness key_fitness = miss->key_fitness;
- const struct nlattr *key = miss->key;
- size_t key_len = miss->key_len;
- uint32_t key_hash;
- struct subfacet *subfacet;
+ case OFPROTO_TABLE_MISS_CONTROLLER:
+ return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
- key_hash = odp_flow_key_hash(key, key_len);
+ case OFPROTO_TABLE_MISS_DROP:
+ return RULE_DPIF_LOOKUP_VERDICT_DROP;
- if (list_is_empty(&facet->subfacets)) {
- subfacet = &facet->one_subfacet;
- } else {
- subfacet = subfacet_find(backer, key, key_len, key_hash);
- if (subfacet) {
- if (subfacet->facet == facet) {
- return subfacet;
+ case OFPROTO_TABLE_MISS_DEFAULT:
+ return RULE_DPIF_LOOKUP_VERDICT_DEFAULT;
}
-
- /* This shouldn't happen. */
- VLOG_ERR_RL(&rl, "subfacet with wrong facet");
- subfacet_destroy(subfacet);
- }
-
- subfacet = xmalloc(sizeof *subfacet);
- }
-
- hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
- list_push_back(&facet->subfacets, &subfacet->list_node);
- subfacet->facet = facet;
- subfacet->key_fitness = key_fitness;
- subfacet->key = xmemdup(key, key_len);
- subfacet->key_len = key_len;
- subfacet->used = now;
- subfacet->created = now;
- subfacet->dp_packet_count = 0;
- subfacet->dp_byte_count = 0;
- subfacet->path = SF_NOT_INSTALLED;
- subfacet->backer = backer;
-
- backer->subfacet_add_count++;
- 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);
-
- /* Update ofproto stats before uninstall the subfacet. */
- ofproto->backer->subfacet_del_count++;
-
- subfacet_uninstall(subfacet);
- hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
- list_remove(&subfacet->list_node);
- free(subfacet->key);
- 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 dpif_backer *backer,
- struct subfacet **subfacets, int n)
-{
- struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
- struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
- struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
- int i;
-
- for (i = 0; i < n; i++) {
- ops[i].type = DPIF_OP_FLOW_DEL;
- ops[i].u.flow_del.key = subfacets[i]->key;
- ops[i].u.flow_del.key_len = subfacets[i]->key_len;
- ops[i].u.flow_del.stats = &stats[i];
- opsp[i] = &ops[i];
- }
-
- dpif_operate(backer->dpif, opsp, n);
- for (i = 0; i < n; i++) {
- subfacet_reset_dp_stats(subfacets[i], &stats[i]);
- subfacets[i]->path = SF_NOT_INSTALLED;
- subfacet_destroy(subfacets[i]);
- run_fast_rl();
- }
-}
-
-/* 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 ofpbuf *odp_actions,
- struct dpif_flow_stats *stats)
-{
- struct facet *facet = subfacet->facet;
- struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
- enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
- const struct nlattr *actions = odp_actions->data;
- size_t actions_len = odp_actions->size;
- struct odputil_keybuf maskbuf;
- struct ofpbuf mask;
-
- uint64_t slow_path_stub[128 / 8];
- enum dpif_flow_put_flags flags;
- 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, facet->xout.slow,
- slow_path_stub, sizeof slow_path_stub,
- &actions, &actions_len);
- }
-
- ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
- if (enable_megaflows) {
- odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
- &facet->flow, UINT32_MAX);
- }
-
- ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
- subfacet->key_len, mask.data, mask.size,
- actions, actions_len, stats);
-
- if (stats) {
- subfacet_reset_dp_stats(subfacet, stats);
- }
-
- if (ret) {
- COVERAGE_INC(subfacet_install_fail);
- } else {
- subfacet->path = path;
- }
- return ret;
-}
-
-/* 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 dpif_flow_stats stats;
- int error;
-
- error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
- subfacet->key_len, &stats);
- subfacet_reset_dp_stats(subfacet, &stats);
- if (!error) {
- subfacet_update_stats(subfacet, &stats);
}
- subfacet->path = SF_NOT_INSTALLED;
- } else {
- ovs_assert(subfacet->dp_packet_count == 0);
- ovs_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;
+ return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
}
-/* Folds the statistics from 'stats' into the counters in 'subfacet'.
+/* Given a port configuration (specified as zero if there's no port), chooses
+ * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
+ * flow table miss.
*
- * 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)
+ * The rule is returned in '*rule', which is valid at least until the next
+ * RCU quiescent period. If the '*rule' needs to stay around longer,
+ * a reference must be taken on it (rule_dpif_ref()).
+ */
+void
+choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
+ struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule,
+ bool take_ref)
{
- if (stats->n_packets || stats->used > subfacet->used) {
- struct facet *facet = subfacet->facet;
-
- subfacet->used = MAX(subfacet->used, stats->used);
- facet->used = MAX(facet->used, stats->used);
- facet->packet_count += stats->n_packets;
- facet->byte_count += stats->n_bytes;
- facet->tcp_flags |= stats->tcp_flags;
+ *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
+ if (take_ref) {
+ rule_dpif_ref(*rule);
}
}
-\f
-/* Rules. */
-/* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
- * the fields that were relevant as part of the lookup. */
-static struct rule_dpif *
-rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
- struct flow_wildcards *wc)
+void
+rule_dpif_ref(struct rule_dpif *rule)
{
- struct rule_dpif *rule;
-
- rule = rule_dpif_lookup_in_table(ofproto, flow, wc, 0);
if (rule) {
- return rule;
- }
-
- return rule_dpif_miss_rule(ofproto, flow);
-}
-
-struct rule_dpif *
-rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
- const struct flow *flow, struct flow_wildcards *wc,
- uint8_t table_id)
-{
- struct cls_rule *cls_rule;
- struct classifier *cls;
- bool frag;
-
- if (table_id >= N_TABLES) {
- return NULL;
- }
-
- if (wc) {
- memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
- wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
+ ofproto_rule_ref(&rule->up);
}
-
- cls = &ofproto->up.tables[table_id].cls;
- frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
- if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
- /* We must pretend that transport ports are unavailable. */
- struct flow ofpc_normal_flow = *flow;
- ofpc_normal_flow.tp_src = htons(0);
- ofpc_normal_flow.tp_dst = htons(0);
- cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
- } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
- cls_rule = &ofproto->drop_frags_rule->up.cr;
- if (wc) {
- flow_wildcards_init_exact(wc);
- }
- } else {
- cls_rule = classifier_lookup(cls, flow, wc);
- }
- return rule_dpif_cast(rule_from_cls_rule(cls_rule));
}
-struct rule_dpif *
-rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
+void
+rule_dpif_unref(struct rule_dpif *rule)
{
- struct ofport_dpif *port;
-
- port = get_ofp_port(ofproto, flow->in_port.ofp_port);
- if (!port) {
- VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
- flow->in_port.ofp_port);
- return ofproto->miss_rule;
- }
-
- if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
- return ofproto->no_packet_in_rule;
+ if (rule) {
+ ofproto_rule_unref(&rule->up);
}
- return ofproto->miss_rule;
}
static void
complete_operation(struct rule_dpif *rule)
+ OVS_REQUIRES(ofproto_mutex)
{
struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
- rule_invalidate(rule);
- if (clogged) {
- struct dpif_completion *c = xmalloc(sizeof *c);
- c->op = rule->up.pending;
- list_push_back(&ofproto->completions, &c->list_node);
- } else {
- ofoperation_complete(rule->up.pending, 0);
- }
+ ofproto->backer->need_revalidate = REV_FLOW_TABLE;
+ ofoperation_complete(rule->up.pending, 0);
+}
+
+static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
+{
+ return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
}
static struct rule *
static enum ofperr
rule_construct(struct rule *rule_)
+ OVS_NO_THREAD_SAFETY_ANALYSIS
{
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;
-
- rule->packet_count = 0;
- rule->byte_count = 0;
-
- victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
- if (victim && !list_is_empty(&victim->facets)) {
- struct facet *facet;
-
- rule->facets = victim->facets;
- list_moved(&rule->facets);
- LIST_FOR_EACH (facet, list_node, &rule->facets) {
- /* XXX: We're only clearing our local counters here. It's possible
- * that quite a few packets are unaccounted for in the datapath
- * statistics. These will be accounted to the new rule instead of
- * cleared as required. This could be fixed by clearing out the
- * datapath statistics for this facet, but currently it doesn't
- * seem worth it. */
- facet_reset_counters(facet);
- facet->rule = rule;
- }
- } else {
- /* Must avoid list_moved() in this case. */
- list_init(&rule->facets);
- }
-
- table_id = rule->up.table_id;
- 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);
+ ovs_mutex_init_adaptive(&rule->stats_mutex);
+ rule->stats.n_packets = 0;
+ rule->stats.n_bytes = 0;
+ rule->stats.used = rule->up.modified;
return 0;
}
static void
-rule_destruct(struct rule *rule_)
+rule_insert(struct rule *rule_)
+ OVS_REQUIRES(ofproto_mutex)
{
struct rule_dpif *rule = rule_dpif_cast(rule_);
- struct facet *facet, *next_facet;
-
- LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
- facet_revalidate(facet);
- }
+ complete_operation(rule);
+}
+static void
+rule_delete(struct rule *rule_)
+ OVS_REQUIRES(ofproto_mutex)
+{
+ struct rule_dpif *rule = rule_dpif_cast(rule_);
complete_operation(rule);
}
static void
-rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
+rule_destruct(struct rule *rule_)
{
struct rule_dpif *rule = rule_dpif_cast(rule_);
+ ovs_mutex_destroy(&rule->stats_mutex);
+}
- /* push_all_stats() can handle flow misses which, when using the learn
- * action, can cause rules to be added and deleted. This can corrupt our
- * caller's datastructures which assume that rule_get_stats() doesn't have
- * an impact on the flow table. To be safe, we disable miss handling. */
- push_all_stats__(false);
+static void
+rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
+ long long int *used)
+{
+ struct rule_dpif *rule = rule_dpif_cast(rule_);
- /* Start from historical data for 'rule' itself that are no longer tracked
- * in facets. This counts, for example, facets that have expired. */
- *packets = rule->packet_count;
- *bytes = rule->byte_count;
+ ovs_mutex_lock(&rule->stats_mutex);
+ *packets = rule->stats.n_packets;
+ *bytes = rule->stats.n_bytes;
+ *used = rule->stats.used;
+ ovs_mutex_unlock(&rule->stats_mutex);
}
static void
rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
struct ofpbuf *packet)
-{
- struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
- struct dpif_flow_stats stats;
- struct xlate_out xout;
- struct xlate_in xin;
-
- dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
- rule_credit_stats(rule, &stats);
-
- xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
- xin.resubmit_stats = &stats;
- xlate_actions(&xin, &xout);
-
- execute_odp_actions(ofproto, flow, xout.odp_actions.data,
- xout.odp_actions.size, packet);
+{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
- xlate_out_uninit(&xout);
+ ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
}
static enum ofperr
}
static void
-rule_modify_actions(struct rule *rule_)
+rule_modify_actions(struct rule *rule_, bool reset_counters)
+ OVS_REQUIRES(ofproto_mutex)
{
struct rule_dpif *rule = rule_dpif_cast(rule_);
+ if (reset_counters) {
+ ovs_mutex_lock(&rule->stats_mutex);
+ rule->stats.n_packets = 0;
+ rule->stats.n_bytes = 0;
+ ovs_mutex_unlock(&rule->stats_mutex);
+ }
+
complete_operation(rule);
}
-\f
-/* Sends 'packet' out 'ofport'.
- * May modify 'packet'.
- * Returns 0 if successful, otherwise a positive errno value. */
-static int
-send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
-{
- struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
- uint64_t odp_actions_stub[1024 / 8];
- struct ofpbuf key, odp_actions;
- struct dpif_flow_stats stats;
- struct odputil_keybuf keybuf;
- struct ofpact_output output;
- struct xlate_out xout;
- struct xlate_in xin;
- struct flow flow;
- union flow_in_port in_port_;
- int error;
-
- ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
- ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
-
- /* Use OFPP_NONE as the in_port to avoid special packet processing. */
- in_port_.ofp_port = OFPP_NONE;
- flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
- odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
- OFPP_LOCAL));
- dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
-
- ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
- output.port = ofport->up.ofp_port;
- output.max_len = 0;
-
- xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
- xin.ofpacts_len = sizeof output;
- xin.ofpacts = &output.ofpact;
- xin.resubmit_stats = &stats;
- xlate_actions(&xin, &xout);
- error = dpif_execute(ofproto->backer->dpif,
- key.data, key.size,
- xout.odp_actions.data, xout.odp_actions.size,
- packet);
- xlate_out_uninit(&xout);
+static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
+{
+ return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
+}
- if (error) {
- VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
- ofproto->up.name, netdev_get_name(ofport->up.netdev),
- ovs_strerror(error));
- }
+static struct ofgroup *
+group_alloc(void)
+{
+ struct group_dpif *group = xzalloc(sizeof *group);
+ return &group->up;
+}
- ofproto->stats.tx_packets++;
- ofproto->stats.tx_bytes += packet->size;
- return error;
+static void
+group_dealloc(struct ofgroup *group_)
+{
+ struct group_dpif *group = group_dpif_cast(group_);
+ free(group);
}
-/* 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_BFD | SLOW_LACP | SLOW_STP)) {
- uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
- ODPP_NONE);
- odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
+group_construct_stats(struct group_dpif *group)
+ OVS_REQUIRES(group->stats_mutex)
+{
+ group->packet_count = 0;
+ group->byte_count = 0;
+ if (!group->bucket_stats) {
+ group->bucket_stats = xcalloc(group->up.n_buckets,
+ sizeof *group->bucket_stats);
} else {
- put_userspace_action(ofproto, &buf, flow, &cookie,
- sizeof cookie.slow_path);
+ memset(group->bucket_stats, 0, group->up.n_buckets *
+ sizeof *group->bucket_stats);
}
- *actionsp = buf.data;
- *actions_lenp = buf.size;
}
-size_t
-put_userspace_action(const struct ofproto_dpif *ofproto,
- struct ofpbuf *odp_actions,
- const struct flow *flow,
- const union user_action_cookie *cookie,
- const size_t cookie_size)
+static enum ofperr
+group_construct(struct ofgroup *group_)
{
- uint32_t pid;
+ struct group_dpif *group = group_dpif_cast(group_);
+ const struct ofputil_bucket *bucket;
- pid = dpif_port_get_pid(ofproto->backer->dpif,
- ofp_port_to_odp_port(ofproto,
- flow->in_port.ofp_port));
+ /* Prevent group chaining because our locking structure makes it hard to
+ * implement deadlock-free. (See xlate_group_resource_check().) */
+ LIST_FOR_EACH (bucket, list_node, &group->up.buckets) {
+ const struct ofpact *a;
- return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
+ OFPACT_FOR_EACH (a, bucket->ofpacts, bucket->ofpacts_len) {
+ if (a->type == OFPACT_GROUP) {
+ return OFPERR_OFPGMFC_CHAINING_UNSUPPORTED;
+ }
+ }
+ }
+
+ ovs_mutex_init_adaptive(&group->stats_mutex);
+ ovs_mutex_lock(&group->stats_mutex);
+ group_construct_stats(group);
+ ovs_mutex_unlock(&group->stats_mutex);
+ return 0;
}
+static void
+group_destruct__(struct group_dpif *group)
+ OVS_REQUIRES(group->stats_mutex)
+{
+ free(group->bucket_stats);
+ group->bucket_stats = NULL;
+}
static void
-update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
- uint64_t packets, uint64_t bytes)
+group_destruct(struct ofgroup *group_)
{
- if (!mirrors) {
- return;
- }
+ struct group_dpif *group = group_dpif_cast(group_);
+ ovs_mutex_lock(&group->stats_mutex);
+ group_destruct__(group);
+ ovs_mutex_unlock(&group->stats_mutex);
+ ovs_mutex_destroy(&group->stats_mutex);
+}
- for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
- struct ofmirror *m;
+static enum ofperr
+group_modify(struct ofgroup *group_, struct ofgroup *victim_)
+{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
+ struct group_dpif *group = group_dpif_cast(group_);
+ struct group_dpif *victim = group_dpif_cast(victim_);
- m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
+ ovs_mutex_lock(&group->stats_mutex);
+ if (victim->up.n_buckets < group->up.n_buckets) {
+ group_destruct__(group);
+ }
+ group_construct_stats(group);
+ ovs_mutex_unlock(&group->stats_mutex);
- 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;
- }
+ ofproto->backer->need_revalidate = REV_FLOW_TABLE;
- m->packet_count += packets;
- m->byte_count += bytes;
- }
+ return 0;
}
-tag_type
-calculate_flow_tag(struct ofproto_dpif *ofproto, const struct flow *flow,
- uint8_t table_id, struct rule_dpif *rule)
+static enum ofperr
+group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
{
- if (table_id > 0 && table_id < N_TABLES) {
- struct table_dpif *table = &ofproto->tables[table_id];
- if (table->other_table) {
- return (rule && rule->tag
- ? rule->tag
- : rule_calculate_tag(flow, &table->other_table->mask,
- table->basis));
- }
- }
+ struct group_dpif *group = group_dpif_cast(group_);
+
+ ovs_mutex_lock(&group->stats_mutex);
+ ogs->packet_count = group->packet_count;
+ ogs->byte_count = group->byte_count;
+ memcpy(ogs->bucket_stats, group->bucket_stats,
+ group->up.n_buckets * sizeof *group->bucket_stats);
+ ovs_mutex_unlock(&group->stats_mutex);
return 0;
}
-\f
-/* Optimized flow revalidation.
- *
- * It's a difficult problem, in general, to tell which facets need to have
- * their actions recalculated whenever the OpenFlow flow table changes. We
- * don't try to solve that general problem: for most kinds of OpenFlow flow
- * table changes, we recalculate the actions for every facet. This is
- * relatively expensive, but it's good enough if the OpenFlow flow table
- * doesn't change very often.
- *
- * However, we can expect one particular kind of OpenFlow flow table change to
- * happen frequently: changes caused by MAC learning. To avoid wasting a lot
- * of CPU on revalidating every facet whenever MAC learning modifies the flow
- * table, we add a special case that applies to flow tables in which every rule
- * has the same form (that is, the same wildcards), except that the table is
- * also allowed to have a single "catch-all" flow that matches all packets. We
- * optimize this case by tagging all of the facets that resubmit into the table
- * and invalidating the same tag whenever a flow changes in that table. The
- * end result is that we revalidate just the facets that need it (and sometimes
- * 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 mask 'mask' when it is inserted
- * into an OpenFlow table with the given 'basis'. */
-tag_type
-rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
- uint32_t secret)
-{
- if (minimask_is_catchall(mask)) {
- return 0;
- } else {
- uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
- return tag_create_deterministic(hash);
- }
-}
-/* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
- * taggability of that table.
- *
- * This function must be called after *each* change to a flow table. If you
- * skip calling it on some changes then the pointer comparisons at the end can
- * be invalid if you get unlucky. For example, if a flow removal causes a
- * cls_table to be destroyed and then a flow insertion causes a cls_table with
- * different wildcards to be created with the same address, then this function
- * will incorrectly skip revalidation. */
-static void
-table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
+bool
+group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
+ struct group_dpif **group)
+ OVS_TRY_RDLOCK(true, (*group)->up.rwlock)
{
- struct table_dpif *table = &ofproto->tables[table_id];
- const struct oftable *oftable = &ofproto->up.tables[table_id];
- struct cls_table *catchall, *other;
- struct cls_table *t;
+ struct ofgroup *ofgroup;
+ bool found;
- catchall = other = NULL;
-
- 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
- * yet. */
- break;
-
- case 1:
- case 2:
- HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
- if (cls_table_is_catchall(t)) {
- catchall = t;
- } else if (!other) {
- other = t;
- } else {
- /* Indicate that we can't tag this by setting both tables to
- * NULL. (We know that 'catchall' is already NULL.) */
- other = NULL;
- }
- }
- break;
+ *group = NULL;
+ found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
+ *group = found ? group_dpif_cast(ofgroup) : NULL;
- default:
- /* Can't tag this table. */
- break;
- }
+ return found;
+}
- if (table->catchall_table != catchall || table->other_table != other) {
- table->catchall_table = catchall;
- table->other_table = other;
- ofproto->backer->need_revalidate = REV_FLOW_TABLE;
- }
+void
+group_dpif_release(struct group_dpif *group)
+ OVS_RELEASES(group->up.rwlock)
+{
+ ofproto_group_release(&group->up);
}
-/* Given 'rule' that has changed in some way (either it is a rule being
- * inserted, a rule being deleted, or a rule whose actions are being
- * modified), marks facets for revalidation to ensure that packets will be
- * forwarded correctly according to the new state of the flow table.
- *
- * This function must be called after *each* change to a flow table. See
- * the comment on table_update_taggable() for more information. */
-static void
-rule_invalidate(const struct rule_dpif *rule)
+void
+group_dpif_get_buckets(const struct group_dpif *group,
+ const struct list **buckets)
{
- struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
+ *buckets = &group->up.buckets;
+}
- table_update_taggable(ofproto, rule->up.table_id);
+enum ofp11_group_type
+group_dpif_get_type(const struct group_dpif *group)
+{
+ return group->up.type;
+}
+\f
+/* Sends 'packet' out 'ofport'.
+ * May modify 'packet'.
+ * Returns 0 if successful, otherwise a positive errno value. */
+int
+ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
+{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
+ int error;
- if (!ofproto->backer->need_revalidate) {
- struct table_dpif *table = &ofproto->tables[rule->up.table_id];
+ error = xlate_send_packet(ofport, packet);
- if (table->other_table && rule->tag) {
- tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
- } else {
- ofproto->backer->need_revalidate = REV_FLOW_TABLE;
- }
- }
+ ovs_mutex_lock(&ofproto->stats_mutex);
+ ofproto->stats.tx_packets++;
+ ofproto->stats.tx_bytes += ofpbuf_size(packet);
+ ovs_mutex_unlock(&ofproto->stats_mutex);
+ return error;
}
\f
static bool
const struct ofpact *ofpacts, size_t ofpacts_len)
{
struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
- struct odputil_keybuf keybuf;
- struct dpif_flow_stats stats;
- struct xlate_out xout;
- struct xlate_in xin;
- struct ofpbuf key;
-
-
- ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
- odp_flow_key_from_flow(&key, flow,
- ofp_port_to_odp_port(ofproto,
- flow->in_port.ofp_port));
-
- dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
-
- xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
- xin.resubmit_stats = &stats;
- xin.ofpacts_len = ofpacts_len;
- xin.ofpacts = ofpacts;
-
- xlate_actions(&xin, &xout);
- dpif_execute(ofproto->backer->dpif, key.data, key.size,
- xout.odp_actions.data, xout.odp_actions.size, packet);
- xlate_out_uninit(&xout);
+ ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
+ ofpacts_len, packet);
return 0;
}
\f
return netflow_set_options(ofproto->netflow, netflow_options);
} else if (ofproto->netflow) {
ofproto->backer->need_revalidate = REV_RECONFIGURE;
- netflow_destroy(ofproto->netflow);
+ netflow_unref(ofproto->netflow);
ofproto->netflow = NULL;
}
dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
}
-
-static void
-send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
-{
- if (!facet_is_controller_flow(facet) &&
- netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
- struct subfacet *subfacet;
- struct ofexpired expired;
-
- LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
- if (subfacet->path == SF_FAST_PATH) {
- struct dpif_flow_stats stats;
-
- subfacet_install(subfacet, &facet->xout.odp_actions,
- &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 cls_cursor cursor;
- struct facet *facet;
-
- cls_cursor_init(&cursor, &ofproto->facets, NULL);
- CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
- send_active_timeout(ofproto, facet);
- }
-}
\f
static struct ofproto_dpif *
ofproto_dpif_lookup(const char *name)
unixctl_command_reply_error(conn, "no such bridge");
return;
}
- mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
+ ovs_rwlock_wrlock(&ofproto->ml->rwlock);
+ mac_learning_flush(ofproto->ml);
+ ovs_rwlock_unlock(&ofproto->ml->rwlock);
} else {
HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
- mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
+ ovs_rwlock_wrlock(&ofproto->ml->rwlock);
+ mac_learning_flush(ofproto->ml);
+ ovs_rwlock_unlock(&ofproto->ml->rwlock);
}
}
unixctl_command_reply(conn, "table successfully flushed");
}
+static struct ofport_dpif *
+ofbundle_get_a_port(const struct ofbundle *bundle)
+{
+ return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
+ bundle_node);
+}
+
static void
ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
const char *argv[], void *aux OVS_UNUSED)
}
ds_put_cstr(&ds, " port VLAN MAC Age\n");
+ ovs_rwlock_rdlock(&ofproto->ml->rwlock);
LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
struct ofbundle *bundle = e->port.p;
char name[OFP_MAX_PORT_NAME_LEN];
name, e->vlan, ETH_ADDR_ARGS(e->mac),
mac_entry_age(ofproto->ml, e));
}
+ ovs_rwlock_unlock(&ofproto->ml->rwlock);
unixctl_command_reply(conn, ds_cstr(&ds));
ds_destroy(&ds);
}
struct trace_ctx {
struct xlate_out xout;
struct xlate_in xin;
+ const struct flow *key;
struct flow flow;
+ struct flow_wildcards wc;
struct ds *result;
};
static void
trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
{
+ const struct rule_actions *actions;
+ ovs_be64 cookie;
+
ds_put_char_multiple(result, '\t', level);
if (!rule) {
ds_put_cstr(result, "No match\n");
return;
}
+ ovs_mutex_lock(&rule->up.mutex);
+ cookie = rule->up.flow_cookie;
+ ovs_mutex_unlock(&rule->up.mutex);
+
ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
- rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
+ rule ? rule->up.table_id : 0, ntohll(cookie));
cls_rule_format(&rule->up.cr, result);
ds_put_char(result, '\n');
+ actions = rule_dpif_get_actions(rule);
+
ds_put_char_multiple(result, '\t', level);
- ds_put_cstr(result, "OpenFlow ");
- ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
+ ds_put_cstr(result, "OpenFlow actions=");
+ ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
ds_put_char(result, '\n');
}
{
ds_put_char_multiple(result, '\t', level);
ds_put_format(result, "%s: ", title);
- if (flow_equal(&trace->xin.flow, &trace->flow)) {
+ /* Do not report unchanged flows for resubmits. */
+ if ((level > 0 && flow_equal(&trace->xin.flow, &trace->flow))
+ || (level == 0 && flow_equal(&trace->xin.flow, trace->key))) {
ds_put_cstr(result, "unchanged");
} else {
flow_format(result, &trace->xin.flow);
ds_put_char_multiple(result, '\t', level);
ds_put_format(result, "%s:", title);
for (i = 0; i < FLOW_N_REGS; i++) {
- ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
+ ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
}
ds_put_char(result, '\n');
}
ds_put_char_multiple(result, '\t', level);
ds_put_format(result, "%s: ", title);
- format_odp_actions(result, odp_actions->data, odp_actions->size);
+ format_odp_actions(result, ofpbuf_data(odp_actions),
+ ofpbuf_size(odp_actions));
+ ds_put_char(result, '\n');
+}
+
+static void
+trace_format_megaflow(struct ds *result, int level, const char *title,
+ struct trace_ctx *trace)
+{
+ struct match match;
+
+ ds_put_char_multiple(result, '\t', level);
+ ds_put_format(result, "%s: ", title);
+ flow_wildcards_or(&trace->wc, &trace->xout.wc, &trace->wc);
+ match_init(&match, trace->key, &trace->wc);
+ match_format(&match, result, OFP_DEFAULT_PRIORITY);
ds_put_char(result, '\n');
}
trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
+ trace_format_megaflow(result, recurse + 1, "Resubmitted megaflow", trace);
trace_format_rule(result, recurse + 1, rule);
}
ds_put_char(result, '\n');
}
-static void
-ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
- void *aux OVS_UNUSED)
-{
- const struct dpif_backer *backer;
- struct ofproto_dpif *ofproto;
- struct ofpbuf odp_key, odp_mask;
+/* Parses the 'argc' elements of 'argv', ignoring argv[0]. The following
+ * forms are supported:
+ *
+ * - [dpname] odp_flow [-generate | packet]
+ * - bridge br_flow [-generate | packet]
+ *
+ * On success, initializes '*ofprotop' and 'flow' and returns NULL. On failure
+ * returns a nonnull malloced error message. */
+static char * WARN_UNUSED_RESULT
+parse_flow_and_packet(int argc, const char *argv[],
+ struct ofproto_dpif **ofprotop, struct flow *flow,
+ struct ofpbuf **packetp)
+{
+ const struct dpif_backer *backer = NULL;
+ const char *error = NULL;
+ char *m_err = NULL;
+ struct simap port_names = SIMAP_INITIALIZER(&port_names);
struct ofpbuf *packet;
- struct ds result;
- struct flow flow;
- char *s;
+ struct ofpbuf odp_key;
+ struct ofpbuf odp_mask;
- packet = NULL;
- backer = NULL;
- ds_init(&result);
ofpbuf_init(&odp_key, 0);
ofpbuf_init(&odp_mask, 0);
packet = ofpbuf_new(0);
argc--;
} else {
- const char *error = eth_from_hex(argv[argc - 1], &packet);
+ error = eth_from_hex(argv[argc - 1], &packet);
if (!error) {
argc--;
} else if (argc == 4) {
/* The 3-argument form must end in "-generate' or a hex string. */
- unixctl_command_reply_error(conn, error);
goto exit;
}
+ error = NULL;
}
- /* Parse the flow and determine whether a datapath or
- * bridge is specified. If function odp_flow_key_from_string()
- * returns 0, the flow is a odp_flow. If function
- * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
- if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
- /* If the odp_flow is the second argument,
- * the datapath name is the first argument. */
- if (argc == 3) {
- const char *dp_type;
- if (!strncmp(argv[1], "ovs-", 4)) {
- dp_type = argv[1] + 4;
- } else {
- dp_type = argv[1];
- }
- backer = shash_find_data(&all_dpif_backers, dp_type);
- if (!backer) {
- unixctl_command_reply_error(conn, "Cannot find datapath "
- "of this name");
- goto exit;
- }
+ /* odp_flow can have its in_port specified as a name instead of port no.
+ * We do not yet know whether a given flow is a odp_flow or a br_flow.
+ * But, to know whether a flow is odp_flow through odp_flow_from_string(),
+ * we need to create a simap of name to port no. */
+ if (argc == 3) {
+ const char *dp_type;
+ if (!strncmp(argv[1], "ovs-", 4)) {
+ dp_type = argv[1] + 4;
} else {
- /* No datapath name specified, so there should be only one
- * datapath. */
- struct shash_node *node;
- if (shash_count(&all_dpif_backers) != 1) {
- unixctl_command_reply_error(conn, "Must specify datapath "
- "name, there is more than one type of datapath");
- goto exit;
- }
+ dp_type = argv[1];
+ }
+ backer = shash_find_data(&all_dpif_backers, dp_type);
+ } else if (argc == 2) {
+ struct shash_node *node;
+ if (shash_count(&all_dpif_backers) == 1) {
node = shash_first(&all_dpif_backers);
backer = node->data;
}
+ } else {
+ error = "Syntax error";
+ goto exit;
+ }
+ if (backer && backer->dpif) {
+ struct dpif_port dpif_port;
+ struct dpif_port_dump port_dump;
+ DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
+ simap_put(&port_names, dpif_port.name,
+ odp_to_u32(dpif_port.port_no));
+ }
+ }
- /* Extract the ofproto_dpif object from the ofproto_receive()
- * function. */
- if (ofproto_receive(backer, NULL, odp_key.data,
- odp_key.size, &flow, NULL, &ofproto, NULL)) {
- unixctl_command_reply_error(conn, "Invalid datapath flow");
+ /* Parse the flow and determine whether a datapath or
+ * bridge is specified. If function odp_flow_key_from_string()
+ * returns 0, the flow is a odp_flow. If function
+ * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
+ if (!odp_flow_from_string(argv[argc - 1], &port_names,
+ &odp_key, &odp_mask)) {
+ if (!backer) {
+ error = "Cannot find the datapath";
goto exit;
}
- ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
- } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
- if (argc != 3) {
- unixctl_command_reply_error(conn, "Must specify bridge name");
+
+ if (xlate_receive(backer, NULL, ofpbuf_data(&odp_key),
+ ofpbuf_size(&odp_key), flow,
+ ofprotop, NULL, NULL, NULL, NULL)) {
+ error = "Invalid datapath flow";
goto exit;
}
+ } else {
+ char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
- ofproto = ofproto_dpif_lookup(argv[1]);
- if (!ofproto) {
- unixctl_command_reply_error(conn, "Unknown bridge name");
+ if (err) {
+ m_err = xasprintf("Bad flow syntax: %s", err);
+ free(err);
goto exit;
+ } else {
+ if (argc != 3) {
+ error = "Must specify bridge name";
+ goto exit;
+ }
+
+ *ofprotop = ofproto_dpif_lookup(argv[1]);
+ if (!*ofprotop) {
+ error = "Unknown bridge name";
+ goto exit;
+ }
+ }
+ }
+
+ /* Generate a packet, if requested. */
+ if (packet) {
+ if (!ofpbuf_size(packet)) {
+ flow_compose(packet, flow);
+ } else {
+ struct pkt_metadata md = pkt_metadata_from_flow(flow);
+
+ /* Use the metadata from the flow and the packet argument
+ * to reconstruct the flow. */
+ flow_extract(packet, &md, flow);
}
+ }
+
+exit:
+ if (error && !m_err) {
+ m_err = xstrdup(error);
+ }
+ if (m_err) {
+ ofpbuf_delete(packet);
+ packet = NULL;
+ }
+ *packetp = packet;
+ ofpbuf_uninit(&odp_key);
+ ofpbuf_uninit(&odp_mask);
+ simap_destroy(&port_names);
+ return m_err;
+}
+
+static void
+ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
+ void *aux OVS_UNUSED)
+{
+ struct ofproto_dpif *ofproto;
+ struct ofpbuf *packet;
+ char *error;
+ struct flow flow;
+
+ error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
+ if (!error) {
+ struct ds result;
+
+ ds_init(&result);
+ ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
+ unixctl_command_reply(conn, ds_cstr(&result));
+ ds_destroy(&result);
+ ofpbuf_delete(packet);
+ } else {
+ unixctl_command_reply_error(conn, error);
+ free(error);
+ }
+}
+
+static void
+ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
+ const char *argv[], void *aux OVS_UNUSED)
+{
+ enum ofputil_protocol usable_protocols;
+ struct ofproto_dpif *ofproto;
+ bool enforce_consistency;
+ struct ofpbuf ofpacts;
+ struct ofpbuf *packet;
+ struct ds result;
+ struct flow flow;
+ uint16_t in_port;
+
+ /* Three kinds of error return values! */
+ enum ofperr retval;
+ char *error;
+
+ packet = NULL;
+ ds_init(&result);
+ ofpbuf_init(&ofpacts, 0);
+
+ /* Parse actions. */
+ error = parse_ofpacts(argv[--argc], &ofpacts, &usable_protocols);
+ if (error) {
+ unixctl_command_reply_error(conn, error);
+ free(error);
+ goto exit;
+ }
+
+ /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
+ * consistency between the flow and the actions. With -consistent, we
+ * enforce consistency even for a flow supported in OpenFlow 1.0. */
+ if (!strcmp(argv[1], "-consistent")) {
+ enforce_consistency = true;
+ argv++;
+ argc--;
} else {
- unixctl_command_reply_error(conn, "Bad flow syntax");
+ enforce_consistency = false;
+ }
+
+ error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
+ if (error) {
+ unixctl_command_reply_error(conn, error);
+ free(error);
+ goto exit;
+ }
+
+ /* Do the same checks as handle_packet_out() in ofproto.c.
+ *
+ * We pass a 'table_id' of 0 to ofproto_check_ofpacts(), which isn't
+ * strictly correct because these actions aren't in any table, but it's OK
+ * because it 'table_id' is used only to check goto_table instructions, but
+ * packet-outs take a list of actions and therefore it can't include
+ * instructions.
+ *
+ * We skip the "meter" check here because meter is an instruction, not an
+ * action, and thus cannot appear in ofpacts. */
+ in_port = ofp_to_u16(flow.in_port.ofp_port);
+ if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
+ unixctl_command_reply_error(conn, "invalid in_port");
goto exit;
}
+ if (enforce_consistency) {
+ retval = ofpacts_check_consistency(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts),
+ &flow, u16_to_ofp(ofproto->up.max_ports),
+ 0, 0, usable_protocols);
+ } else {
+ retval = ofpacts_check(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &flow,
+ u16_to_ofp(ofproto->up.max_ports), 0, 0,
+ &usable_protocols);
+ }
- /* Generate a packet, if requested. */
- if (packet) {
- if (!packet->size) {
- flow_compose(packet, &flow);
- } else {
- union flow_in_port in_port_;
-
- in_port_ = flow.in_port;
- ds_put_cstr(&result, "Packet: ");
- s = ofp_packet_to_string(packet->data, packet->size);
- ds_put_cstr(&result, s);
- free(s);
-
- /* Use the metadata from the flow and the packet argument
- * to reconstruct the flow. */
- flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
- &in_port_, &flow);
- }
+ if (retval) {
+ ds_clear(&result);
+ ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
+ unixctl_command_reply_error(conn, ds_cstr(&result));
+ goto exit;
}
- ofproto_trace(ofproto, &flow, packet, &result);
+ ofproto_trace(ofproto, &flow, packet,
+ ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &result);
unixctl_command_reply(conn, ds_cstr(&result));
exit:
ds_destroy(&result);
ofpbuf_delete(packet);
- ofpbuf_uninit(&odp_key);
- ofpbuf_uninit(&odp_mask);
+ ofpbuf_uninit(&ofpacts);
}
-void
-ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
- const struct ofpbuf *packet, struct ds *ds)
+/* Implements a "trace" through 'ofproto''s flow table, appending a textual
+ * description of the results to 'ds'.
+ *
+ * The trace follows a packet with the specified 'flow' through the flow
+ * table. 'packet' may be nonnull to trace an actual packet, with consequent
+ * side effects (if it is nonnull then its flow must be 'flow').
+ *
+ * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
+ * trace, otherwise the actions are determined by a flow table lookup. */
+static void
+ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
+ const struct ofpbuf *packet,
+ const struct ofpact ofpacts[], size_t ofpacts_len,
+ struct ds *ds)
{
struct rule_dpif *rule;
+ struct trace_ctx trace;
+ ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
ds_put_cstr(ds, "Flow: ");
flow_format(ds, flow);
ds_put_char(ds, '\n');
- rule = rule_dpif_lookup(ofproto, flow, NULL);
+ flow_wildcards_init_catchall(&trace.wc);
+ if (ofpacts) {
+ rule = NULL;
+ } else {
+ rule_dpif_lookup(ofproto, flow, &trace.wc, &rule, false);
- trace_format_rule(ds, 0, rule);
- if (rule == ofproto->miss_rule) {
- ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
- } else if (rule == ofproto->no_packet_in_rule) {
- ds_put_cstr(ds, "\nNo match, packets dropped because "
- "OFPPC_NO_PACKET_IN is set on in_port.\n");
- } else if (rule == ofproto->drop_frags_rule) {
- ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
- "and the fragment handling mode is \"drop\".\n");
+ trace_format_rule(ds, 0, rule);
+ if (rule == ofproto->miss_rule) {
+ ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
+ } else if (rule == ofproto->no_packet_in_rule) {
+ ds_put_cstr(ds, "\nNo match, packets dropped because "
+ "OFPPC_NO_PACKET_IN is set on in_port.\n");
+ } else if (rule == ofproto->drop_frags_rule) {
+ ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
+ "and the fragment handling mode is \"drop\".\n");
+ }
}
- if (rule) {
- uint64_t odp_actions_stub[1024 / 8];
- struct ofpbuf odp_actions;
- struct trace_ctx trace;
- struct match match;
- uint8_t tcp_flags;
-
- tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
+ if (rule || ofpacts) {
trace.result = ds;
- trace.flow = *flow;
- ofpbuf_use_stub(&odp_actions,
- odp_actions_stub, sizeof odp_actions_stub);
- xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
+ trace.key = flow; /* Original flow key, used for megaflow. */
+ trace.flow = *flow; /* May be modified by actions. */
+ xlate_in_init(&trace.xin, ofproto, flow, rule, ntohs(flow->tcp_flags),
+ packet);
+ if (ofpacts) {
+ trace.xin.ofpacts = ofpacts;
+ trace.xin.ofpacts_len = ofpacts_len;
+ }
trace.xin.resubmit_hook = trace_resubmit;
trace.xin.report_hook = trace_report;
ds_put_char(ds, '\n');
trace_format_flow(ds, 0, "Final flow", &trace);
-
- match_init(&match, flow, &trace.xout.wc);
- ds_put_cstr(ds, "Relevant fields: ");
- match_format(&match, ds, OFP_DEFAULT_PRIORITY);
- ds_put_char(ds, '\n');
+ trace_format_megaflow(ds, 0, "Megaflow", &trace);
ds_put_cstr(ds, "Datapath actions: ");
- format_odp_actions(ds, trace.xout.odp_actions.data,
- trace.xout.odp_actions.size);
+ format_odp_actions(ds, ofpbuf_data(&trace.xout.odp_actions),
+ ofpbuf_size(&trace.xout.odp_actions));
if (trace.xout.slow) {
+ enum slow_path_reason slow;
+
ds_put_cstr(ds, "\nThis flow is handled by the userspace "
"slow path because it:");
- switch (trace.xout.slow) {
- case SLOW_CFM:
- ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
- break;
- case SLOW_LACP:
- ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
- break;
- case SLOW_STP:
- ds_put_cstr(ds, "\n\t- Consists of STP packets.");
- break;
- case SLOW_BFD:
- ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
- break;
- case SLOW_CONTROLLER:
- ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
- "to the OpenFlow controller.");
- break;
- case __SLOW_MAX:
- NOT_REACHED();
- }
- }
-
- xlate_out_uninit(&trace.xout);
- }
-}
-
-static void
-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, NULL);
-}
-static void
-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, NULL);
-}
+ slow = trace.xout.slow;
+ while (slow) {
+ enum slow_path_reason bit = rightmost_1bit(slow);
-/* 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 cls_cursor cursor;
- struct facet *facet;
- int errors;
+ ds_put_format(ds, "\n\t- %s.",
+ slow_path_reason_to_explanation(bit));
- errors = 0;
- cls_cursor_init(&cursor, &ofproto->facets, NULL);
- CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
- if (!facet_check_consistency(facet)) {
- errors++;
+ slow &= ~bit;
+ }
}
- }
- if (errors) {
- ofproto->backer->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);
- }
+ xlate_out_uninit(&trace.xout);
}
-
- unixctl_command_reply(conn, ds_cstr(&reply));
- ds_destroy(&reply);
}
/* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
ds_destroy(&ds);
}
-static void
-show_dp_rates(struct ds *ds, const char *heading,
- const struct avg_subfacet_rates *rates)
-{
- ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
- heading, rates->add_rate, rates->del_rate);
-}
-
static void
dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
{
const struct shash_node **ofprotos;
- struct ofproto_dpif *ofproto;
+ struct dpif_dp_stats dp_stats;
struct shash ofproto_shash;
- uint64_t n_hit, n_missed;
- long long int minutes;
size_t i;
- n_hit = n_missed = 0;
- HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
- if (ofproto->backer == backer) {
- n_missed += ofproto->n_missed;
- n_hit += ofproto->n_hit;
- }
- }
+ dpif_get_dp_stats(backer->dpif, &dp_stats);
ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
- dpif_name(backer->dpif), n_hit, n_missed);
- ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
- " life span: %lldms\n", hmap_count(&backer->subfacets),
- backer->avg_n_subfacet, backer->max_n_subfacet,
- backer->avg_subfacet_life);
-
- minutes = (time_msec() - backer->created) / (1000 * 60);
- if (minutes >= 60) {
- show_dp_rates(ds, "\thourly avg:", &backer->hourly);
- }
- if (minutes >= 60 * 24) {
- show_dp_rates(ds, "\tdaily avg:", &backer->daily);
- }
- show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
+ dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
shash_init(&ofproto_shash);
ofprotos = get_ofprotos(&ofproto_shash);
continue;
}
- ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
- ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
+ ds_put_format(ds, "\t%s:\n", ofproto->up.name);
ports = shash_sort(&ofproto->up.port_by_name);
for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
ds_destroy(&ds);
}
-/* Dump the megaflow (facet) cache. This is useful to check the
- * correctness of flow wildcarding, since the same mechanism is used for
- * both xlate caching and kernel wildcarding.
- *
- * It's important to note that in the output the flow description uses
- * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
- *
- * This command is only needed for advanced debugging, so it's not
- * documented in the man page. */
-static void
-ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
- int argc OVS_UNUSED, const char *argv[],
- void *aux OVS_UNUSED)
-{
- struct ds ds = DS_EMPTY_INITIALIZER;
- const struct ofproto_dpif *ofproto;
- long long int now = time_msec();
- struct cls_cursor cursor;
- struct facet *facet;
-
- ofproto = ofproto_dpif_lookup(argv[1]);
- if (!ofproto) {
- unixctl_command_reply_error(conn, "no such bridge");
- return;
- }
-
- cls_cursor_init(&cursor, &ofproto->facets, NULL);
- CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
- cls_rule_format(&facet->cr, &ds);
- ds_put_cstr(&ds, ", ");
- ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
- ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
- ds_put_cstr(&ds, "Datapath actions: ");
- if (facet->xout.slow) {
- uint64_t slow_path_stub[128 / 8];
- const struct nlattr *actions;
- size_t actions_len;
-
- compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
- slow_path_stub, sizeof slow_path_stub,
- &actions, &actions_len);
- format_odp_actions(&ds, actions, actions_len);
- } else {
- format_odp_actions(&ds, facet->xout.odp_actions.data,
- facet->xout.odp_actions.size);
- }
- ds_put_cstr(&ds, "\n");
- }
-
- ds_chomp(&ds, '\n');
- unixctl_command_reply(conn, ds_cstr(&ds));
- ds_destroy(&ds);
-}
-
-/* Disable using the megaflows.
- *
- * This command is only needed for advanced debugging, so it's not
- * documented in the man page. */
-static void
-ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
- int argc OVS_UNUSED,
- const char *argv[] OVS_UNUSED,
- void *aux OVS_UNUSED)
-{
- struct ofproto_dpif *ofproto;
-
- enable_megaflows = false;
-
- HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
- flush(&ofproto->up);
- }
-
- unixctl_command_reply(conn, "megaflows disabled");
-}
-
-/* Re-enable using megaflows.
- *
- * This command is only needed for advanced debugging, so it's not
- * documented in the man page. */
-static void
-ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
- int argc OVS_UNUSED,
- const char *argv[] OVS_UNUSED,
- void *aux OVS_UNUSED)
+static bool
+ofproto_dpif_contains_flow(const struct ofproto_dpif *ofproto,
+ const struct nlattr *key, size_t key_len)
{
- struct ofproto_dpif *ofproto;
-
- enable_megaflows = true;
-
- HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
- flush(&ofproto->up);
- }
+ struct ofproto_dpif *ofp;
+ struct flow flow;
- unixctl_command_reply(conn, "megaflows enabled");
+ xlate_receive(ofproto->backer, NULL, key, key_len, &flow, &ofp,
+ NULL, NULL, NULL, NULL);
+ return ofp == ofproto;
}
static void
void *aux OVS_UNUSED)
{
struct ds ds = DS_EMPTY_INITIALIZER;
+ const struct dpif_flow_stats *stats;
const struct ofproto_dpif *ofproto;
- struct subfacet *subfacet;
+ struct dpif_flow_dump flow_dump;
+ const struct nlattr *actions;
+ const struct nlattr *mask;
+ const struct nlattr *key;
+ size_t actions_len;
+ size_t mask_len;
+ size_t key_len;
+ bool verbosity = false;
+ struct dpif_port dpif_port;
+ struct dpif_port_dump port_dump;
+ struct hmap portno_names;
+ void *state = NULL;
+ int error;
- ofproto = ofproto_dpif_lookup(argv[1]);
+ ofproto = ofproto_dpif_lookup(argv[argc - 1]);
if (!ofproto) {
unixctl_command_reply_error(conn, "no such bridge");
return;
}
- update_stats(ofproto->backer);
+ if (argc > 2 && !strcmp(argv[1], "-m")) {
+ verbosity = true;
+ }
- HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
- struct facet *facet = subfacet->facet;
+ hmap_init(&portno_names);
+ DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
+ odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
+ }
- if (ofproto_dpif_cast(facet->rule->up.ofproto) != ofproto) {
+ ds_init(&ds);
+ error = dpif_flow_dump_start(&flow_dump, ofproto->backer->dpif);
+ if (error) {
+ goto exit;
+ }
+ dpif_flow_dump_state_init(ofproto->backer->dpif, &state);
+ while (dpif_flow_dump_next(&flow_dump, state, &key, &key_len,
+ &mask, &mask_len, &actions, &actions_len,
+ &stats)) {
+ if (!ofproto_dpif_contains_flow(ofproto, key, key_len)) {
continue;
}
- odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
-
- ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
- subfacet->dp_packet_count, subfacet->dp_byte_count);
- if (subfacet->used) {
- ds_put_format(&ds, "%.3fs",
- (time_msec() - subfacet->used) / 1000.0);
- } else {
- ds_put_format(&ds, "never");
- }
- if (subfacet->facet->tcp_flags) {
- ds_put_cstr(&ds, ", flags:");
- packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
- }
-
+ odp_flow_format(key, key_len, mask, mask_len, &portno_names, &ds,
+ verbosity);
+ ds_put_cstr(&ds, ", ");
+ dpif_flow_stats_format(stats, &ds);
ds_put_cstr(&ds, ", actions:");
- if (facet->xout.slow) {
- uint64_t slow_path_stub[128 / 8];
- const struct nlattr *actions;
- size_t actions_len;
-
- compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
- slow_path_stub, sizeof slow_path_stub,
- &actions, &actions_len);
- format_odp_actions(&ds, actions, actions_len);
- } else {
- format_odp_actions(&ds, facet->xout.odp_actions.data,
- facet->xout.odp_actions.size);
- }
+ format_odp_actions(&ds, actions, actions_len);
ds_put_char(&ds, '\n');
}
+ dpif_flow_dump_state_uninit(ofproto->backer->dpif, state);
+ error = dpif_flow_dump_done(&flow_dump);
- 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;
+exit:
+ if (error) {
+ ds_clear(&ds);
+ ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
+ unixctl_command_reply_error(conn, ds_cstr(&ds));
+ } else {
+ unixctl_command_reply(conn, ds_cstr(&ds));
}
-
- flush(&ofproto->up);
-
- unixctl_command_reply(conn, ds_cstr(&ds));
+ odp_portno_names_destroy(&portno_names);
+ hmap_destroy(&portno_names);
ds_destroy(&ds);
}
unixctl_command_register(
"ofproto/trace",
- "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
+ "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
1, 3, ofproto_unixctl_trace, NULL);
+ unixctl_command_register(
+ "ofproto/trace-packet-out",
+ "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
+ 2, 6, ofproto_unixctl_trace_actions, 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", "", 0, 0, ofproto_unixctl_dpif_show,
NULL);
- unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
+ unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
ofproto_unixctl_dpif_dump_flows, NULL);
- unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
- ofproto_unixctl_dpif_del_flows, NULL);
- unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
- ofproto_unixctl_dpif_dump_megaflows, NULL);
- unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
- ofproto_unixctl_dpif_disable_megaflows, NULL);
- unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
- ofproto_unixctl_dpif_enable_megaflows, NULL);
+}
+
+
+/* Returns true if 'rule' is an internal rule, false otherwise. */
+bool
+rule_is_internal(const struct rule *rule)
+{
+ return rule->table_id == TBL_INTERNAL;
}
\f
/* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
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);
+ bundle_set(ofport_->ofproto, ofport->bundle, NULL);
}
ofport->realdev_ofp_port = realdev_ofp_port;
return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
}
-/* Returns the OFP port number of the Linux VLAN device that corresponds to
- * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
- * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
- * 'vlan_tci' 9, it would return the port number of eth0.9.
- *
- * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
- * function just returns its 'realdev_ofp_port' argument. */
-ofp_port_t
-vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
- ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
+bool
+ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
+ OVS_EXCLUDED(ofproto->vsp_mutex)
+{
+ /* hmap_is_empty is thread safe. */
+ return !hmap_is_empty(&ofproto->realdev_vid_map);
+}
+
+
+static ofp_port_t
+vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
+ ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
+ OVS_REQUIRES(ofproto->vsp_mutex)
{
if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
int vid = vlan_tci_to_vid(vlan_tci);
return realdev_ofp_port;
}
+/* Returns the OFP port number of the Linux VLAN device that corresponds to
+ * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
+ * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
+ * 'vlan_tci' 9, it would return the port number of eth0.9.
+ *
+ * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
+ * function just returns its 'realdev_ofp_port' argument. */
+ofp_port_t
+vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
+ ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
+ OVS_EXCLUDED(ofproto->vsp_mutex)
+{
+ ofp_port_t ret;
+
+ /* hmap_is_empty is thread safe, see if we can return immediately. */
+ if (hmap_is_empty(&ofproto->realdev_vid_map)) {
+ return realdev_ofp_port;
+ }
+ ovs_mutex_lock(&ofproto->vsp_mutex);
+ ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
+ ovs_mutex_unlock(&ofproto->vsp_mutex);
+ return ret;
+}
+
static struct vlan_splinter *
vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
{
static ofp_port_t
vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
ofp_port_t vlandev_ofp_port, int *vid)
+ OVS_REQUIRES(ofproto->vsp_mutex)
{
if (!hmap_is_empty(&ofproto->vlandev_map)) {
const struct vlan_splinter *vsp;
* '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
+bool
vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
+ OVS_EXCLUDED(ofproto->vsp_mutex)
{
ofp_port_t realdev;
int vid;
+ /* hmap_is_empty is thread safe. */
+ if (hmap_is_empty(&ofproto->vlandev_map)) {
+ return false;
+ }
+
+ ovs_mutex_lock(&ofproto->vsp_mutex);
realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
+ ovs_mutex_unlock(&ofproto->vsp_mutex);
if (!realdev) {
return false;
}
struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
struct vlan_splinter *vsp;
+ ovs_mutex_lock(&ofproto->vsp_mutex);
vsp = vlandev_find(ofproto, port->up.ofp_port);
if (vsp) {
hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
} else {
VLOG_ERR("missing vlan device record");
}
+ ovs_mutex_unlock(&ofproto->vsp_mutex);
}
static void
{
struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
+ ovs_mutex_lock(&ofproto->vsp_mutex);
if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
- && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
+ && (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_ofp_port(port->up.ofp_port));
- hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
- hash_realdev_vid(realdev_ofp_port, vid));
vsp->realdev_ofp_port = realdev_ofp_port;
vsp->vlandev_ofp_port = port->up.ofp_port;
vsp->vid = vid;
port->realdev_ofp_port = realdev_ofp_port;
+
+ hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
+ hash_ofp_port(port->up.ofp_port));
+ hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
+ hash_realdev_vid(realdev_ofp_port, vid));
} else {
VLOG_ERR("duplicate vlan device record");
}
+ ovs_mutex_unlock(&ofproto->vsp_mutex);
}
-odp_port_t
+static odp_port_t
ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
{
const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
return ofport ? ofport->odp_port : ODPP_NONE;
}
-static struct ofport_dpif *
+struct ofport_dpif *
odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
{
struct ofport_dpif *port;
+ ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
&backer->odp_to_ofport_map) {
if (port->odp_port == odp_port) {
+ ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
return port;
}
}
+ ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
return NULL;
}
}
}
-/* Compute exponentially weighted moving average, adding 'new' as the newest,
- * most heavily weighted element. 'base' designates the rate of decay: after
- * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
- * (about .37). */
-static void
-exp_mavg(double *avg, int base, double new)
+uint32_t
+ofproto_dpif_alloc_recirc_id(struct ofproto_dpif *ofproto)
{
- *avg = (*avg * (base - 1) + new) / base;
+ struct dpif_backer *backer = ofproto->backer;
+
+ return recirc_id_alloc(backer->rid_pool);
}
-static void
-update_moving_averages(struct dpif_backer *backer)
+void
+ofproto_dpif_free_recirc_id(struct ofproto_dpif *ofproto, uint32_t recirc_id)
{
- const int min_ms = 60 * 1000; /* milliseconds in one minute. */
- long long int minutes = (time_msec() - backer->created) / min_ms;
+ struct dpif_backer *backer = ofproto->backer;
+
+ recirc_id_free(backer->rid_pool, recirc_id);
+}
+
+int
+ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
+ struct match *match, int priority,
+ const struct ofpbuf *ofpacts,
+ struct rule **rulep)
+{
+ struct ofputil_flow_mod fm;
+ struct rule_dpif *rule;
+ int error;
- if (minutes > 0) {
- backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
- / minutes;
- backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
- / minutes;
+ fm.match = *match;
+ fm.priority = priority;
+ fm.new_cookie = htonll(0);
+ fm.cookie = htonll(0);
+ fm.cookie_mask = htonll(0);
+ fm.modify_cookie = false;
+ fm.table_id = TBL_INTERNAL;
+ fm.command = OFPFC_ADD;
+ fm.idle_timeout = 0;
+ fm.hard_timeout = 0;
+ fm.buffer_id = 0;
+ fm.out_port = 0;
+ fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
+ fm.ofpacts = ofpbuf_data(ofpacts);
+ fm.ofpacts_len = ofpbuf_size(ofpacts);
+
+ error = ofproto_flow_mod(&ofproto->up, &fm);
+ if (error) {
+ VLOG_ERR_RL(&rl, "failed to add internal flow (%s)",
+ ofperr_to_string(error));
+ *rulep = NULL;
+ return error;
+ }
+
+ rule = rule_dpif_lookup_in_table(ofproto, TBL_INTERNAL, &match->flow,
+ &match->wc, false);
+ if (rule) {
+ *rulep = &rule->up;
} else {
- backer->lifetime.add_rate = 0.0;
- backer->lifetime.del_rate = 0.0;
+ OVS_NOT_REACHED();
}
+ return 0;
+}
- /* Update hourly averages on the minute boundaries. */
- if (time_msec() - backer->last_minute >= min_ms) {
- exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
- exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
+int
+ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
+ struct match *match, int priority)
+{
+ struct ofputil_flow_mod fm;
+ int error;
- /* Update daily averages on the hour boundaries. */
- if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
- exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
- exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
- }
+ fm.match = *match;
+ fm.priority = priority;
+ fm.new_cookie = htonll(0);
+ fm.cookie = htonll(0);
+ fm.cookie_mask = htonll(0);
+ fm.modify_cookie = false;
+ fm.table_id = TBL_INTERNAL;
+ fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
+ fm.command = OFPFC_DELETE_STRICT;
- backer->total_subfacet_add_count += backer->subfacet_add_count;
- backer->total_subfacet_del_count += backer->subfacet_del_count;
- backer->subfacet_add_count = 0;
- backer->subfacet_del_count = 0;
- backer->last_minute += min_ms;
+ error = ofproto_flow_mod(&ofproto->up, &fm);
+ if (error) {
+ VLOG_ERR_RL(&rl, "failed to delete internal flow (%s)",
+ ofperr_to_string(error));
+ return error;
}
+
+ return 0;
}
const struct ofproto_class ofproto_dpif_class = {
del,
port_open_type,
type_run,
- type_run_fast,
type_wait,
alloc,
construct,
destruct,
dealloc,
run,
- run_fast,
wait,
- get_memory_usage,
+ NULL, /* get_memory_usage. */
+ type_get_memory_usage,
flush,
get_features,
get_tables,
NULL, /* rule_choose_table */
rule_alloc,
rule_construct,
+ rule_insert,
+ rule_delete,
rule_destruct,
rule_dealloc,
rule_get_stats,
get_stp_status,
set_stp_port,
get_stp_port_status,
+ get_stp_port_stats,
set_queues,
bundle_set,
bundle_remove,
- mirror_set,
- mirror_get_stats,
+ mirror_set__,
+ mirror_get_stats__,
set_flood_vlans,
is_mirror_output_bundle,
forward_bpdu_changed,
NULL, /* meter_set */
NULL, /* meter_get */
NULL, /* meter_del */
+ group_alloc, /* group_alloc */
+ group_construct, /* group_construct */
+ group_destruct, /* group_destruct */
+ group_dealloc, /* group_dealloc */
+ group_modify, /* group_modify */
+ group_get_stats, /* group_get_stats */
};
git://git.onelab.eu / sliver-openvswitch.git / blobdiff