/*
- * Copyright (c) 2009, 2010 Nicira Networks.
+ * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks.
* Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
*
* Licensed under the Apache License, Version 2.0 (the "License");
#include "ofproto.h"
#include <errno.h>
#include <inttypes.h>
-#include <sys/socket.h>
-#include <net/if.h>
-#include <netinet/in.h>
#include <stdbool.h>
#include <stdlib.h>
+#include "bitmap.h"
+#include "byte-order.h"
#include "classifier.h"
+#include "connmgr.h"
#include "coverage.h"
-#include "discovery.h"
-#include "dpif.h"
#include "dynamic-string.h"
-#include "fail-open.h"
-#include "in-band.h"
-#include "mac-learning.h"
+#include "hash.h"
+#include "hmap.h"
+#include "meta-flow.h"
#include "netdev.h"
-#include "netflow.h"
-#include "odp-util.h"
+#include "nx-match.h"
+#include "ofp-errors.h"
#include "ofp-print.h"
#include "ofp-util.h"
-#include "ofproto-sflow.h"
#include "ofpbuf.h"
+#include "ofproto-provider.h"
#include "openflow/nicira-ext.h"
#include "openflow/openflow.h"
-#include "openvswitch/datapath-protocol.h"
#include "packets.h"
#include "pinsched.h"
#include "pktbuf.h"
#include "poll-loop.h"
-#include "port-array.h"
-#include "rconn.h"
+#include "random.h"
#include "shash.h"
-#include "status.h"
-#include "stp.h"
-#include "stream-ssl.h"
-#include "svec.h"
-#include "tag.h"
+#include "sset.h"
#include "timeval.h"
+#include "unaligned.h"
#include "unixctl.h"
-#include "vconn.h"
#include "vlog.h"
-#include "xtoxll.h"
-VLOG_DEFINE_THIS_MODULE(ofproto)
-
-#include "sflow_api.h"
-
-enum {
- TABLEID_HASH = 0,
- TABLEID_CLASSIFIER = 1
+VLOG_DEFINE_THIS_MODULE(ofproto);
+
+COVERAGE_DEFINE(ofproto_error);
+COVERAGE_DEFINE(ofproto_flush);
+COVERAGE_DEFINE(ofproto_no_packet_in);
+COVERAGE_DEFINE(ofproto_packet_out);
+COVERAGE_DEFINE(ofproto_queue_req);
+COVERAGE_DEFINE(ofproto_recv_openflow);
+COVERAGE_DEFINE(ofproto_reinit_ports);
+COVERAGE_DEFINE(ofproto_uninstallable);
+COVERAGE_DEFINE(ofproto_update_port);
+
+enum ofproto_state {
+ S_OPENFLOW, /* Processing OpenFlow commands. */
+ S_EVICT, /* Evicting flows from over-limit tables. */
+ S_FLUSH, /* Deleting all flow table rules. */
};
-struct ofport {
- struct netdev *netdev;
- struct ofp_phy_port opp; /* In host byte order. */
+enum ofoperation_type {
+ OFOPERATION_ADD,
+ OFOPERATION_DELETE,
+ OFOPERATION_MODIFY
};
-static void ofport_free(struct ofport *);
-static void hton_ofp_phy_port(struct ofp_phy_port *);
-
-static int xlate_actions(const union ofp_action *in, size_t n_in,
- const flow_t *flow, struct ofproto *ofproto,
- const struct ofpbuf *packet,
- struct odp_actions *out, tag_type *tags,
- bool *may_set_up_flow, uint16_t *nf_output_iface);
-
-struct rule {
- struct cls_rule cr;
-
- uint64_t flow_cookie; /* Controller-issued identifier.
- (Kept in network-byte order.) */
- uint16_t idle_timeout; /* In seconds from time of last use. */
- uint16_t hard_timeout; /* In seconds from time of creation. */
- bool send_flow_removed; /* Send a flow removed message? */
- long long int used; /* Last-used time (0 if never used). */
- long long int created; /* Creation time. */
- uint64_t packet_count; /* Number of packets received. */
- uint64_t byte_count; /* Number of bytes received. */
- uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
- tag_type tags; /* Tags (set only by hooks). */
- struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
-
- /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
- * exact-match rule (having cr.wc.wildcards of 0) generated from the
- * wildcard rule 'super'. In this case, 'list' is an element of the
- * super-rule's list.
- *
- * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
- * a list of subrules. A super-rule with no wildcards (where
- * cr.wc.wildcards is 0) will never have any subrules. */
- struct rule *super;
- struct list list;
-
- /* OpenFlow actions.
- *
- * 'n_actions' is the number of elements in the 'actions' array. A single
- * action may take up more more than one element's worth of space.
+/* A single OpenFlow request can execute any number of operations. The
+ * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
+ * ofconn to which an error reply should be sent if necessary.
+ *
+ * ofproto initiates some operations internally. These operations are still
+ * assigned to groups but will not have an associated ofconn. */
+struct ofopgroup {
+ struct ofproto *ofproto; /* Owning ofproto. */
+ struct list ofproto_node; /* In ofproto's "pending" list. */
+ struct list ops; /* List of "struct ofoperation"s. */
+
+ /* Data needed to send OpenFlow reply on failure or to send a buffered
+ * packet on success.
*
- * A subrule has no actions (it uses the super-rule's actions). */
- int n_actions;
- union ofp_action *actions;
+ * If list_is_empty(ofconn_node) then this ofopgroup never had an
+ * associated ofconn or its ofconn's connection dropped after it initiated
+ * the operation. In the latter case 'ofconn' is a wild pointer that
+ * refers to freed memory, so the 'ofconn' member must be used only if
+ * !list_is_empty(ofconn_node).
+ */
+ struct list ofconn_node; /* In ofconn's list of pending opgroups. */
+ struct ofconn *ofconn; /* ofconn for reply (but see note above). */
+ struct ofp_header *request; /* Original request (truncated at 64 bytes). */
+ uint32_t buffer_id; /* Buffer id from original request. */
+ int error; /* 0 if no error yet, otherwise error code. */
+};
- /* Datapath actions.
- *
- * A super-rule with wildcard fields never has ODP actions (since the
- * datapath only supports exact-match flows). */
- bool installed; /* Installed in datapath? */
- bool may_install; /* True ordinarily; false if actions must
- * be reassessed for every packet. */
- int n_odp_actions;
- union odp_action *odp_actions;
+static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
+static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
+ const struct ofp_header *,
+ uint32_t buffer_id);
+static void ofopgroup_submit(struct ofopgroup *);
+static void ofopgroup_destroy(struct ofopgroup *);
+
+/* A single flow table operation. */
+struct ofoperation {
+ struct ofopgroup *group; /* Owning group. */
+ struct list group_node; /* In ofopgroup's "ops" list. */
+ struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
+ struct rule *rule; /* Rule being operated upon. */
+ enum ofoperation_type type; /* Type of operation. */
+ int status; /* -1 if pending, otherwise 0 or error code. */
+ struct rule *victim; /* OFOPERATION_ADDING: Replaced rule. */
+ union ofp_action *actions; /* OFOPERATION_MODIFYING: Replaced actions. */
+ int n_actions; /* OFOPERATION_MODIFYING: # of old actions. */
+ ovs_be64 flow_cookie; /* Rule's old flow cookie. */
};
-static inline bool
-rule_is_hidden(const struct rule *rule)
-{
- /* Subrules are merely an implementation detail, so hide them from the
- * controller. */
- if (rule->super != NULL) {
- return true;
- }
+static void ofoperation_create(struct ofopgroup *, struct rule *,
+ enum ofoperation_type);
+static void ofoperation_destroy(struct ofoperation *);
- /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
- * (e.g. by in-band control) and are intentionally hidden from the
- * controller. */
- if (rule->cr.priority > UINT16_MAX) {
- return true;
- }
+/* oftable. */
+static void oftable_init(struct oftable *);
+static void oftable_destroy(struct oftable *);
- return false;
-}
+static void oftable_set_name(struct oftable *, const char *name);
+
+static void oftable_disable_eviction(struct oftable *);
+static void oftable_enable_eviction(struct oftable *,
+ const struct mf_subfield *fields,
+ size_t n_fields);
-static struct rule *rule_create(struct ofproto *, struct rule *super,
- const union ofp_action *, size_t n_actions,
- uint16_t idle_timeout, uint16_t hard_timeout,
- uint64_t flow_cookie, bool send_flow_removed);
-static void rule_free(struct rule *);
-static void rule_destroy(struct ofproto *, struct rule *);
-static struct rule *rule_from_cls_rule(const struct cls_rule *);
-static void rule_insert(struct ofproto *, struct rule *,
- struct ofpbuf *packet, uint16_t in_port);
-static void rule_remove(struct ofproto *, struct rule *);
-static bool rule_make_actions(struct ofproto *, struct rule *,
- const struct ofpbuf *packet);
-static void rule_install(struct ofproto *, struct rule *,
- struct rule *displaced_rule);
-static void rule_uninstall(struct ofproto *, struct rule *);
-static void rule_post_uninstall(struct ofproto *, struct rule *);
-static void send_flow_removed(struct ofproto *p, struct rule *rule,
- long long int now, uint8_t reason);
-
-/* ofproto supports two kinds of OpenFlow connections:
+static void oftable_remove_rule(struct rule *);
+static struct rule *oftable_replace_rule(struct rule *);
+static void oftable_substitute_rule(struct rule *old, struct rule *new);
+
+/* A set of rules within a single OpenFlow table (oftable) that have the same
+ * values for the oftable's eviction_fields. A rule to be evicted, when one is
+ * needed, is taken from the eviction group that contains the greatest number
+ * of rules.
*
- * - "Controller connections": Connections to ordinary OpenFlow controllers.
- * ofproto maintains persistent connections to these controllers and by
- * default sends them asynchronous messages such as packet-ins.
+ * An oftable owns any number of eviction groups, each of which contains any
+ * number of rules.
*
- * - "Transient connections", e.g. from ovs-ofctl. When these connections
- * drop, it is the other side's responsibility to reconnect them if
- * necessary. ofproto does not send them asynchronous messages by default.
- */
-enum ofconn_type {
- OFCONN_CONTROLLER, /* An OpenFlow controller. */
- OFCONN_TRANSIENT /* A transient connection. */
-};
-
-/* An OpenFlow connection. */
-struct ofconn {
- struct ofproto *ofproto; /* The ofproto that owns this connection. */
- struct list node; /* In struct ofproto's "all_conns" list. */
- struct rconn *rconn; /* OpenFlow connection. */
- enum ofconn_type type; /* Type. */
-
- /* OFPT_PACKET_IN related data. */
- struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
- struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
- struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
- int miss_send_len; /* Bytes to send of buffered packets. */
-
- /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
- * requests, and the maximum number before we stop reading OpenFlow
- * requests. */
-#define OFCONN_REPLY_MAX 100
- struct rconn_packet_counter *reply_counter;
-
- /* type == OFCONN_CONTROLLER only. */
- enum nx_role role; /* Role. */
- struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
- struct discovery *discovery; /* Controller discovery object, if enabled. */
- struct status_category *ss; /* Switch status category. */
- enum ofproto_band band; /* In-band or out-of-band? */
+ * Membership in an eviction group is imprecise, based on the hash of the
+ * oftable's eviction_fields (in the eviction_group's id_node.hash member).
+ * That is, if two rules have different eviction_fields, but those
+ * eviction_fields hash to the same value, then they will belong to the same
+ * eviction_group anyway.
+ *
+ * (When eviction is not enabled on an oftable, we don't track any eviction
+ * groups, to save time and space.) */
+struct eviction_group {
+ struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
+ struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
+ struct heap rules; /* Contains "struct rule"s. */
};
-/* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
- * "schedulers" array. Their values are 0 and 1, and their meanings and values
- * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
- * case anything ever changes, check their values here. */
-#define N_SCHEDULERS 2
-BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
-BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
-BUILD_ASSERT_DECL(OFPR_ACTION == 1);
-BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
-
-static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
- enum ofconn_type);
-static void ofconn_destroy(struct ofconn *);
-static void ofconn_run(struct ofconn *, struct ofproto *);
-static void ofconn_wait(struct ofconn *);
-static bool ofconn_receives_async_msgs(const struct ofconn *);
-static char *ofconn_make_name(const struct ofproto *, const char *target);
-
-static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
- struct rconn_packet_counter *counter);
-
-static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
-static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
-
-struct ofproto {
- /* Settings. */
- uint64_t datapath_id; /* Datapath ID. */
- uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
- char *mfr_desc; /* Manufacturer. */
- char *hw_desc; /* Hardware. */
- char *sw_desc; /* Software version. */
- char *serial_desc; /* Serial number. */
- char *dp_desc; /* Datapath description. */
-
- /* Datapath. */
- struct dpif *dpif;
- struct netdev_monitor *netdev_monitor;
- struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
- * OFP port nr. */
- struct shash port_by_name;
- uint32_t max_ports;
-
- /* Configuration. */
- struct switch_status *switch_status;
- struct fail_open *fail_open;
- struct netflow *netflow;
- struct ofproto_sflow *sflow;
-
- /* In-band control. */
- struct in_band *in_band;
- long long int next_in_band_update;
- struct sockaddr_in *extra_in_band_remotes;
- size_t n_extra_remotes;
-
- /* Flow table. */
- struct classifier cls;
- bool need_revalidate;
- long long int next_expiration;
- struct tag_set revalidate_set;
- bool tun_id_from_cookie;
-
- /* OpenFlow connections. */
- struct hmap controllers; /* Controller "struct ofconn"s. */
- struct list all_conns; /* Contains "struct ofconn"s. */
- enum ofproto_fail_mode fail_mode;
- struct pvconn **listeners;
- size_t n_listeners;
- struct pvconn **snoops;
- size_t n_snoops;
-
- /* Hooks for ovs-vswitchd. */
- const struct ofhooks *ofhooks;
- void *aux;
-
- /* Used by default ofhooks. */
- struct mac_learning *ml;
-};
+static struct rule *choose_rule_to_evict(struct oftable *);
+static void ofproto_evict(struct ofproto *);
+static uint32_t rule_eviction_priority(struct rule *);
-static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+/* ofport. */
+static void ofport_destroy__(struct ofport *);
+static void ofport_destroy(struct ofport *);
-static const struct ofhooks default_ofhooks;
+static void update_port(struct ofproto *, const char *devname);
+static int init_ports(struct ofproto *);
+static void reinit_ports(struct ofproto *);
+/* rule. */
+static void ofproto_rule_destroy__(struct rule *);
+static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
+static bool rule_is_modifiable(const struct rule *);
+static bool rule_is_hidden(const struct rule *);
+
+/* OpenFlow. */
+static enum ofperr add_flow(struct ofproto *, struct ofconn *,
+ const struct ofputil_flow_mod *,
+ const struct ofp_header *);
+static void delete_flow__(struct rule *, struct ofopgroup *);
+static bool handle_openflow(struct ofconn *, struct ofpbuf *);
+static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
+ const struct ofputil_flow_mod *,
+ const struct ofp_header *);
+
+/* ofproto. */
static uint64_t pick_datapath_id(const struct ofproto *);
static uint64_t pick_fallback_dpid(void);
+static void ofproto_destroy__(struct ofproto *);
+static void set_internal_devs_mtu(struct ofproto *);
-static void update_used(struct ofproto *);
-static void update_stats(struct ofproto *, struct rule *,
- const struct odp_flow_stats *);
-static void expire_rule(struct cls_rule *, void *ofproto);
-static void active_timeout(struct ofproto *ofproto, struct rule *rule);
-static bool revalidate_rule(struct ofproto *p, struct rule *rule);
-static void revalidate_cb(struct cls_rule *rule_, void *p_);
+/* unixctl. */
+static void ofproto_unixctl_init(void);
-static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
+/* All registered ofproto classes, in probe order. */
+static const struct ofproto_class **ofproto_classes;
+static size_t n_ofproto_classes;
+static size_t allocated_ofproto_classes;
-static void handle_openflow(struct ofconn *, struct ofproto *,
- struct ofpbuf *);
+/* Map from datapath name to struct ofproto, for use by unixctl commands. */
+static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
-static void refresh_port_groups(struct ofproto *);
-
-static void update_port(struct ofproto *, const char *devname);
-static int init_ports(struct ofproto *);
-static void reinit_ports(struct ofproto *);
+static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
-int
-ofproto_create(const char *datapath, const char *datapath_type,
- const struct ofhooks *ofhooks, void *aux,
- struct ofproto **ofprotop)
+static void
+ofproto_initialize(void)
{
- struct odp_stats stats;
- struct ofproto *p;
- struct dpif *dpif;
- int error;
-
- *ofprotop = NULL;
+ static bool inited;
- /* Connect to datapath and start listening for messages. */
- error = dpif_open(datapath, datapath_type, &dpif);
- if (error) {
- VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
- return error;
- }
- error = dpif_get_dp_stats(dpif, &stats);
- if (error) {
- VLOG_ERR("failed to obtain stats for datapath %s: %s",
- datapath, strerror(error));
- dpif_close(dpif);
- return error;
- }
- error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
- if (error) {
- VLOG_ERR("failed to listen on datapath %s: %s",
- datapath, strerror(error));
- dpif_close(dpif);
- return error;
- }
- dpif_flow_flush(dpif);
- dpif_recv_purge(dpif);
-
- /* Initialize settings. */
- p = xzalloc(sizeof *p);
- p->fallback_dpid = pick_fallback_dpid();
- p->datapath_id = p->fallback_dpid;
- p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
- p->hw_desc = xstrdup(DEFAULT_HW_DESC);
- p->sw_desc = xstrdup(DEFAULT_SW_DESC);
- p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
- p->dp_desc = xstrdup(DEFAULT_DP_DESC);
-
- /* Initialize datapath. */
- p->dpif = dpif;
- p->netdev_monitor = netdev_monitor_create();
- port_array_init(&p->ports);
- shash_init(&p->port_by_name);
- p->max_ports = stats.max_ports;
-
- /* Initialize submodules. */
- p->switch_status = switch_status_create(p);
- p->in_band = NULL;
- p->fail_open = NULL;
- p->netflow = NULL;
- p->sflow = NULL;
-
- /* Initialize flow table. */
- classifier_init(&p->cls);
- p->need_revalidate = false;
- p->next_expiration = time_msec() + 1000;
- tag_set_init(&p->revalidate_set);
-
- /* Initialize OpenFlow connections. */
- list_init(&p->all_conns);
- hmap_init(&p->controllers);
- p->listeners = NULL;
- p->n_listeners = 0;
- p->snoops = NULL;
- p->n_snoops = 0;
-
- /* Initialize hooks. */
- if (ofhooks) {
- p->ofhooks = ofhooks;
- p->aux = aux;
- p->ml = NULL;
- } else {
- p->ofhooks = &default_ofhooks;
- p->aux = p;
- p->ml = mac_learning_create();
+ if (!inited) {
+ inited = true;
+ ofproto_class_register(&ofproto_dpif_class);
}
-
- /* Pick final datapath ID. */
- p->datapath_id = pick_datapath_id(p);
- VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
-
- *ofprotop = p;
- return 0;
}
-void
-ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
+/* 'type' should be a normalized datapath type, as returned by
+ * ofproto_normalize_type(). Returns the corresponding ofproto_class
+ * structure, or a null pointer if there is none registered for 'type'. */
+static const struct ofproto_class *
+ofproto_class_find__(const char *type)
{
- uint64_t old_dpid = p->datapath_id;
- p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
- if (p->datapath_id != old_dpid) {
- VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
+ size_t i;
- /* Force all active connections to reconnect, since there is no way to
- * notify a controller that the datapath ID has changed. */
- ofproto_reconnect_controllers(p);
- }
-}
+ ofproto_initialize();
+ for (i = 0; i < n_ofproto_classes; i++) {
+ const struct ofproto_class *class = ofproto_classes[i];
+ struct sset types;
+ bool found;
-static bool
-is_discovery_controller(const struct ofproto_controller *c)
-{
- return !strcmp(c->target, "discover");
-}
+ sset_init(&types);
+ class->enumerate_types(&types);
+ found = sset_contains(&types, type);
+ sset_destroy(&types);
-static bool
-is_in_band_controller(const struct ofproto_controller *c)
-{
- return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
+ if (found) {
+ return class;
+ }
+ }
+ VLOG_WARN("unknown datapath type %s", type);
+ return NULL;
}
-/* Creates a new controller in 'ofproto'. Some of the settings are initially
- * drawn from 'c', but update_controller() needs to be called later to finish
- * the new ofconn's configuration. */
-static void
-add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
+/* Registers a new ofproto class. After successful registration, new ofprotos
+ * of that type can be created using ofproto_create(). */
+int
+ofproto_class_register(const struct ofproto_class *new_class)
{
- struct discovery *discovery;
- struct ofconn *ofconn;
+ size_t i;
- if (is_discovery_controller(c)) {
- int error = discovery_create(c->accept_re, c->update_resolv_conf,
- ofproto->dpif, ofproto->switch_status,
- &discovery);
- if (error) {
- return;
+ for (i = 0; i < n_ofproto_classes; i++) {
+ if (ofproto_classes[i] == new_class) {
+ return EEXIST;
}
- } else {
- discovery = NULL;
}
- ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_CONTROLLER);
- ofconn->pktbuf = pktbuf_create();
- ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
- if (discovery) {
- ofconn->discovery = discovery;
- } else {
- char *name = ofconn_make_name(ofproto, c->target);
- rconn_connect(ofconn->rconn, c->target, name);
- free(name);
+ if (n_ofproto_classes >= allocated_ofproto_classes) {
+ ofproto_classes = x2nrealloc(ofproto_classes,
+ &allocated_ofproto_classes,
+ sizeof *ofproto_classes);
}
- hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
- hash_string(c->target, 0));
+ ofproto_classes[n_ofproto_classes++] = new_class;
+ return 0;
}
-/* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
- * target or turn discovery on or off (these are done by creating new ofconns
- * and deleting old ones), but it can update the rest of an ofconn's
- * settings. */
-static void
-update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
+/* Unregisters a datapath provider. 'type' must have been previously
+ * registered and not currently be in use by any ofprotos. After
+ * unregistration new datapaths of that type cannot be opened using
+ * ofproto_create(). */
+int
+ofproto_class_unregister(const struct ofproto_class *class)
{
- struct ofproto *ofproto = ofconn->ofproto;
- int probe_interval;
- int i;
-
- ofconn->band = (is_in_band_controller(c)
- ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
-
- rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
-
- probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
- rconn_set_probe_interval(ofconn->rconn, probe_interval);
-
- if (ofconn->discovery) {
- discovery_set_update_resolv_conf(ofconn->discovery,
- c->update_resolv_conf);
- discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
- }
-
- for (i = 0; i < N_SCHEDULERS; i++) {
- struct pinsched **s = &ofconn->schedulers[i];
+ size_t i;
- if (c->rate_limit > 0) {
- if (!*s) {
- *s = pinsched_create(c->rate_limit, c->burst_limit,
- ofproto->switch_status);
- } else {
- pinsched_set_limits(*s, c->rate_limit, c->burst_limit);
+ for (i = 0; i < n_ofproto_classes; i++) {
+ if (ofproto_classes[i] == class) {
+ for (i++; i < n_ofproto_classes; i++) {
+ ofproto_classes[i - 1] = ofproto_classes[i];
}
- } else {
- pinsched_destroy(*s);
- *s = NULL;
+ n_ofproto_classes--;
+ return 0;
}
}
+ VLOG_WARN("attempted to unregister an ofproto class that is not "
+ "registered");
+ return EAFNOSUPPORT;
}
-static const char *
-ofconn_get_target(const struct ofconn *ofconn)
-{
- return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
-}
-
-static struct ofconn *
-find_controller_by_target(struct ofproto *ofproto, const char *target)
+/* Clears 'types' and enumerates all registered ofproto types into it. The
+ * caller must first initialize the sset. */
+void
+ofproto_enumerate_types(struct sset *types)
{
- struct ofconn *ofconn;
+ size_t i;
- HMAP_FOR_EACH_WITH_HASH (ofconn, struct ofconn, hmap_node,
- hash_string(target, 0), &ofproto->controllers) {
- if (!strcmp(ofconn_get_target(ofconn), target)) {
- return ofconn;
- }
+ ofproto_initialize();
+ for (i = 0; i < n_ofproto_classes; i++) {
+ ofproto_classes[i]->enumerate_types(types);
}
- return NULL;
}
-static void
-update_in_band_remotes(struct ofproto *ofproto)
+/* Returns the fully spelled out name for the given ofproto 'type'.
+ *
+ * Normalized type string can be compared with strcmp(). Unnormalized type
+ * string might be the same even if they have different spellings. */
+const char *
+ofproto_normalize_type(const char *type)
{
- const struct ofconn *ofconn;
- struct sockaddr_in *addrs;
- size_t max_addrs, n_addrs;
- bool discovery;
- size_t i;
+ return type && type[0] ? type : "system";
+}
- /* Allocate enough memory for as many remotes as we could possibly have. */
- max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
- addrs = xmalloc(max_addrs * sizeof *addrs);
- n_addrs = 0;
+/* Clears 'names' and enumerates the names of all known created ofprotos with
+ * the given 'type'. The caller must first initialize the sset. Returns 0 if
+ * successful, otherwise a positive errno value.
+ *
+ * Some kinds of datapaths might not be practically enumerable. This is not
+ * considered an error. */
+int
+ofproto_enumerate_names(const char *type, struct sset *names)
+{
+ const struct ofproto_class *class = ofproto_class_find__(type);
+ return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
+ }
- /* Add all the remotes. */
- discovery = false;
- HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &ofproto->controllers) {
- struct sockaddr_in *sin = &addrs[n_addrs];
+int
+ofproto_create(const char *datapath_name, const char *datapath_type,
+ struct ofproto **ofprotop)
+{
+ const struct ofproto_class *class;
+ struct ofproto *ofproto;
+ int error;
- if (ofconn->band == OFPROTO_OUT_OF_BAND) {
- continue;
- }
+ *ofprotop = NULL;
- sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
- if (sin->sin_addr.s_addr) {
- sin->sin_port = rconn_get_remote_port(ofconn->rconn);
- n_addrs++;
- }
- if (ofconn->discovery) {
- discovery = true;
- }
- }
- for (i = 0; i < ofproto->n_extra_remotes; i++) {
- addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
+ ofproto_initialize();
+ ofproto_unixctl_init();
+
+ datapath_type = ofproto_normalize_type(datapath_type);
+ class = ofproto_class_find__(datapath_type);
+ if (!class) {
+ VLOG_WARN("could not create datapath %s of unknown type %s",
+ datapath_name, datapath_type);
+ return EAFNOSUPPORT;
+ }
+
+ ofproto = class->alloc();
+ if (!ofproto) {
+ VLOG_ERR("failed to allocate datapath %s of type %s",
+ datapath_name, datapath_type);
+ return ENOMEM;
+ }
+
+ /* Initialize. */
+ memset(ofproto, 0, sizeof *ofproto);
+ ofproto->ofproto_class = class;
+ ofproto->name = xstrdup(datapath_name);
+ ofproto->type = xstrdup(datapath_type);
+ hmap_insert(&all_ofprotos, &ofproto->hmap_node,
+ hash_string(ofproto->name, 0));
+ ofproto->datapath_id = 0;
+ ofproto_set_flow_eviction_threshold(ofproto,
+ OFPROTO_FLOW_EVICTON_THRESHOLD_DEFAULT);
+ ofproto->forward_bpdu = false;
+ ofproto->fallback_dpid = pick_fallback_dpid();
+ ofproto->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
+ ofproto->hw_desc = xstrdup(DEFAULT_HW_DESC);
+ ofproto->sw_desc = xstrdup(DEFAULT_SW_DESC);
+ ofproto->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
+ ofproto->dp_desc = xstrdup(DEFAULT_DP_DESC);
+ ofproto->frag_handling = OFPC_FRAG_NORMAL;
+ hmap_init(&ofproto->ports);
+ shash_init(&ofproto->port_by_name);
+ ofproto->tables = NULL;
+ ofproto->n_tables = 0;
+ ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
+ ofproto->state = S_OPENFLOW;
+ list_init(&ofproto->pending);
+ ofproto->n_pending = 0;
+ hmap_init(&ofproto->deletions);
+ ofproto->vlan_bitmap = NULL;
+ ofproto->vlans_changed = false;
+
+ error = ofproto->ofproto_class->construct(ofproto);
+ if (error) {
+ VLOG_ERR("failed to open datapath %s: %s",
+ datapath_name, strerror(error));
+ ofproto_destroy__(ofproto);
+ return error;
}
- /* Create or update or destroy in-band.
- *
- * Ordinarily we only enable in-band if there's at least one remote
- * address, but discovery needs the in-band rules for DHCP to be installed
- * even before we know any remote addresses. */
- if (n_addrs || discovery) {
- if (!ofproto->in_band) {
- in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
- &ofproto->in_band);
- }
- if (ofproto->in_band) {
- in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
- }
- ofproto->next_in_band_update = time_msec() + 1000;
- } else {
- in_band_destroy(ofproto->in_band);
- ofproto->in_band = NULL;
- }
+ assert(ofproto->n_tables);
- /* Clean up. */
- free(addrs);
+ ofproto->datapath_id = pick_datapath_id(ofproto);
+ VLOG_INFO("using datapath ID %016"PRIx64, ofproto->datapath_id);
+ init_ports(ofproto);
+
+ *ofprotop = ofproto;
+ return 0;
}
-static void
-update_fail_open(struct ofproto *p)
+void
+ofproto_init_tables(struct ofproto *ofproto, int n_tables)
{
- struct ofconn *ofconn;
+ struct oftable *table;
- if (!hmap_is_empty(&p->controllers)
- && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
- struct rconn **rconns;
- size_t n;
+ assert(!ofproto->n_tables);
+ assert(n_tables >= 1 && n_tables <= 255);
- if (!p->fail_open) {
- p->fail_open = fail_open_create(p, p->switch_status);
- }
+ ofproto->n_tables = n_tables;
+ ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
+ OFPROTO_FOR_EACH_TABLE (table, ofproto) {
+ oftable_init(table);
+ }
+}
- n = 0;
- rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
- HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &p->controllers) {
- rconns[n++] = ofconn->rconn;
- }
+void
+ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
+{
+ uint64_t old_dpid = p->datapath_id;
+ p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
+ if (p->datapath_id != old_dpid) {
+ VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
- fail_open_set_controllers(p->fail_open, rconns, n);
- /* p->fail_open takes ownership of 'rconns'. */
- } else {
- fail_open_destroy(p->fail_open);
- p->fail_open = NULL;
+ /* Force all active connections to reconnect, since there is no way to
+ * notify a controller that the datapath ID has changed. */
+ ofproto_reconnect_controllers(p);
}
}
const struct ofproto_controller *controllers,
size_t n_controllers)
{
- struct shash new_controllers;
- struct ofconn *ofconn, *next;
- bool ss_exists;
- size_t i;
-
- shash_init(&new_controllers);
- for (i = 0; i < n_controllers; i++) {
- const struct ofproto_controller *c = &controllers[i];
-
- shash_add_once(&new_controllers, c->target, &controllers[i]);
- if (!find_controller_by_target(p, c->target)) {
- add_controller(p, c);
- }
- }
-
- ss_exists = false;
- HMAP_FOR_EACH_SAFE (ofconn, next, struct ofconn, hmap_node,
- &p->controllers) {
- struct ofproto_controller *c;
-
- c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
- if (!c) {
- ofconn_destroy(ofconn);
- } else {
- update_controller(ofconn, c);
- if (ofconn->ss) {
- ss_exists = true;
- }
- }
- }
- shash_destroy(&new_controllers);
-
- update_in_band_remotes(p);
-
- update_fail_open(p);
-
- if (!hmap_is_empty(&p->controllers) && !ss_exists) {
- ofconn = CONTAINER_OF(hmap_first(&p->controllers),
- struct ofconn, hmap_node);
- ofconn->ss = switch_status_register(p->switch_status, "remote",
- rconn_status_cb, ofconn->rconn);
- }
+ connmgr_set_controllers(p->connmgr, controllers, n_controllers);
}
void
ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
{
- p->fail_mode = fail_mode;
- update_fail_open(p);
+ connmgr_set_fail_mode(p->connmgr, fail_mode);
}
/* Drops the connections between 'ofproto' and all of its controllers, forcing
void
ofproto_reconnect_controllers(struct ofproto *ofproto)
{
- struct ofconn *ofconn;
+ connmgr_reconnect(ofproto->connmgr);
+}
- LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
- rconn_reconnect(ofconn->rconn);
- }
+/* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
+ * in-band control should guarantee access, in the same way that in-band
+ * control guarantees access to OpenFlow controllers. */
+void
+ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
+ const struct sockaddr_in *extras, size_t n)
+{
+ connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
}
-static bool
-any_extras_changed(const struct ofproto *ofproto,
- const struct sockaddr_in *extras, size_t n)
+/* Sets the OpenFlow queue used by flows set up by in-band control on
+ * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
+ * flows will use the default queue. */
+void
+ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
{
- size_t i;
+ connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
+}
- if (n != ofproto->n_extra_remotes) {
- return true;
+/* Sets the number of flows at which eviction from the kernel flow table
+ * will occur. */
+void
+ofproto_set_flow_eviction_threshold(struct ofproto *ofproto, unsigned threshold)
+{
+ if (threshold < OFPROTO_FLOW_EVICTION_THRESHOLD_MIN) {
+ ofproto->flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_MIN;
+ } else {
+ ofproto->flow_eviction_threshold = threshold;
}
+}
- for (i = 0; i < n; i++) {
- const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
- const struct sockaddr_in *new = &extras[i];
-
- if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
- old->sin_port != new->sin_port) {
- return true;
+/* If forward_bpdu is true, the NORMAL action will forward frames with
+ * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
+ * the NORMAL action will drop these frames. */
+void
+ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
+{
+ bool old_val = ofproto->forward_bpdu;
+ ofproto->forward_bpdu = forward_bpdu;
+ if (old_val != ofproto->forward_bpdu) {
+ if (ofproto->ofproto_class->forward_bpdu_changed) {
+ ofproto->ofproto_class->forward_bpdu_changed(ofproto);
}
}
-
- return false;
}
-/* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
- * in-band control should guarantee access, in the same way that in-band
- * control guarantees access to OpenFlow controllers. */
+/* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
+ * 'idle_time', in seconds. */
void
-ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
- const struct sockaddr_in *extras, size_t n)
+ofproto_set_mac_idle_time(struct ofproto *ofproto, unsigned idle_time)
{
- if (!any_extras_changed(ofproto, extras, n)) {
- return;
+ if (ofproto->ofproto_class->set_mac_idle_time) {
+ ofproto->ofproto_class->set_mac_idle_time(ofproto, idle_time);
}
-
- free(ofproto->extra_in_band_remotes);
- ofproto->n_extra_remotes = n;
- ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
-
- update_in_band_remotes(ofproto);
}
void
}
}
-static int
-set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
- const struct svec *svec)
-{
- struct pvconn **pvconns = *pvconnsp;
- size_t n_pvconns = *n_pvconnsp;
- int retval = 0;
- size_t i;
-
- for (i = 0; i < n_pvconns; i++) {
- pvconn_close(pvconns[i]);
- }
- free(pvconns);
-
- pvconns = xmalloc(svec->n * sizeof *pvconns);
- n_pvconns = 0;
- for (i = 0; i < svec->n; i++) {
- const char *name = svec->names[i];
- struct pvconn *pvconn;
- int error;
-
- error = pvconn_open(name, &pvconn);
- if (!error) {
- pvconns[n_pvconns++] = pvconn;
- } else {
- VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
- if (!retval) {
- retval = error;
- }
- }
- }
-
- *pvconnsp = pvconns;
- *n_pvconnsp = n_pvconns;
-
- return retval;
-}
-
-int
-ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
-{
- return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
-}
-
int
-ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
+ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
{
- return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
+ return connmgr_set_snoops(ofproto->connmgr, snoops);
}
int
ofproto_set_netflow(struct ofproto *ofproto,
const struct netflow_options *nf_options)
{
- if (nf_options && nf_options->collectors.n) {
- if (!ofproto->netflow) {
- ofproto->netflow = netflow_create();
- }
- return netflow_set_options(ofproto->netflow, nf_options);
+ if (nf_options && sset_is_empty(&nf_options->collectors)) {
+ nf_options = NULL;
+ }
+
+ if (ofproto->ofproto_class->set_netflow) {
+ return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
} else {
- netflow_destroy(ofproto->netflow);
- ofproto->netflow = NULL;
- return 0;
+ return nf_options ? EOPNOTSUPP : 0;
}
}
-void
+int
ofproto_set_sflow(struct ofproto *ofproto,
const struct ofproto_sflow_options *oso)
{
- struct ofproto_sflow *os = ofproto->sflow;
- if (oso) {
- if (!os) {
- struct ofport *ofport;
- unsigned int odp_port;
-
- os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
- refresh_port_groups(ofproto);
- PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) {
- ofproto_sflow_add_port(os, odp_port,
- netdev_get_name(ofport->netdev));
- }
- }
- ofproto_sflow_set_options(os, oso);
+ if (oso && sset_is_empty(&oso->targets)) {
+ oso = NULL;
+ }
+
+ if (ofproto->ofproto_class->set_sflow) {
+ return ofproto->ofproto_class->set_sflow(ofproto, oso);
} else {
- ofproto_sflow_destroy(os);
- ofproto->sflow = NULL;
+ return oso ? EOPNOTSUPP : 0;
}
}
+\f
+/* Spanning Tree Protocol (STP) configuration. */
+/* Configures STP on 'ofproto' using the settings defined in 's'. If
+ * 's' is NULL, disables STP.
+ *
+ * Returns 0 if successful, otherwise a positive errno value. */
int
-ofproto_set_stp(struct ofproto *ofproto OVS_UNUSED, bool enable_stp)
+ofproto_set_stp(struct ofproto *ofproto,
+ const struct ofproto_stp_settings *s)
{
- /* XXX */
- if (enable_stp) {
- VLOG_WARN("STP is not yet implemented");
- return EINVAL;
- } else {
- return 0;
- }
+ return (ofproto->ofproto_class->set_stp
+ ? ofproto->ofproto_class->set_stp(ofproto, s)
+ : EOPNOTSUPP);
}
-uint64_t
-ofproto_get_datapath_id(const struct ofproto *ofproto)
+/* Retrieves STP status of 'ofproto' and stores it in 's'. If the
+ * 'enabled' member of 's' is false, then the other members are not
+ * meaningful.
+ *
+ * Returns 0 if successful, otherwise a positive errno value. */
+int
+ofproto_get_stp_status(struct ofproto *ofproto,
+ struct ofproto_stp_status *s)
{
- return ofproto->datapath_id;
+ return (ofproto->ofproto_class->get_stp_status
+ ? ofproto->ofproto_class->get_stp_status(ofproto, s)
+ : EOPNOTSUPP);
}
-bool
-ofproto_has_controller(const struct ofproto *ofproto)
+/* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
+ * in 's'. The caller is responsible for assigning STP port numbers
+ * (using the 'port_num' member in the range of 1 through 255, inclusive)
+ * and ensuring there are no duplicates. If the 's' is NULL, then STP
+ * is disabled on the port.
+ *
+ * Returns 0 if successful, otherwise a positive errno value.*/
+int
+ofproto_port_set_stp(struct ofproto *ofproto, uint16_t ofp_port,
+ const struct ofproto_port_stp_settings *s)
{
- return !hmap_is_empty(&ofproto->controllers);
+ struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
+ if (!ofport) {
+ VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
+ ofproto->name, ofp_port);
+ return ENODEV;
+ }
+
+ return (ofproto->ofproto_class->set_stp_port
+ ? ofproto->ofproto_class->set_stp_port(ofport, s)
+ : EOPNOTSUPP);
}
-void
-ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
+/* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
+ * 's'. If the 'enabled' member in 's' is false, then the other members
+ * are not meaningful.
+ *
+ * Returns 0 if successful, otherwise a positive errno value.*/
+int
+ofproto_port_get_stp_status(struct ofproto *ofproto, uint16_t ofp_port,
+ struct ofproto_port_stp_status *s)
{
- size_t i;
+ struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
+ if (!ofport) {
+ VLOG_WARN("%s: cannot get STP status on nonexistent port %"PRIu16,
+ ofproto->name, ofp_port);
+ return ENODEV;
+ }
+
+ return (ofproto->ofproto_class->get_stp_port_status
+ ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
+ : EOPNOTSUPP);
+}
+\f
+/* Queue DSCP configuration. */
- for (i = 0; i < ofproto->n_listeners; i++) {
- svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
+/* Registers meta-data associated with the 'n_qdscp' Qualities of Service
+ * 'queues' attached to 'ofport'. This data is not intended to be sufficient
+ * to implement QoS. Instead, it is used to implement features which require
+ * knowledge of what queues exist on a port, and some basic information about
+ * them.
+ *
+ * Returns 0 if successful, otherwise a positive errno value. */
+int
+ofproto_port_set_queues(struct ofproto *ofproto, uint16_t ofp_port,
+ const struct ofproto_port_queue *queues,
+ size_t n_queues)
+{
+ struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
+
+ if (!ofport) {
+ VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
+ ofproto->name, ofp_port);
+ return ENODEV;
}
+
+ return (ofproto->ofproto_class->set_queues
+ ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
+ : EOPNOTSUPP);
}
+\f
+/* Connectivity Fault Management configuration. */
+/* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
void
-ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
+ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
{
- size_t i;
-
- for (i = 0; i < ofproto->n_snoops; i++) {
- svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
+ struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
+ if (ofport && ofproto->ofproto_class->set_cfm) {
+ ofproto->ofproto_class->set_cfm(ofport, NULL);
}
}
+/* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
+ * basic configuration from the configuration members in 'cfm', and the remote
+ * maintenance point ID from remote_mpid. Ignores the statistics members of
+ * 'cfm'.
+ *
+ * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
void
-ofproto_destroy(struct ofproto *p)
+ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
+ const struct cfm_settings *s)
{
- struct ofconn *ofconn, *next_ofconn;
struct ofport *ofport;
- unsigned int port_no;
- size_t i;
+ int error;
- if (!p) {
+ ofport = ofproto_get_port(ofproto, ofp_port);
+ if (!ofport) {
+ VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
+ ofproto->name, ofp_port);
return;
}
- /* Destroy fail-open and in-band early, since they touch the classifier. */
- fail_open_destroy(p->fail_open);
- p->fail_open = NULL;
-
- in_band_destroy(p->in_band);
- p->in_band = NULL;
- free(p->extra_in_band_remotes);
-
- ofproto_flush_flows(p);
- classifier_destroy(&p->cls);
-
- LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
- &p->all_conns) {
- ofconn_destroy(ofconn);
- }
- hmap_destroy(&p->controllers);
-
- dpif_close(p->dpif);
- netdev_monitor_destroy(p->netdev_monitor);
- PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
- ofport_free(ofport);
+ /* XXX: For configuration simplicity, we only support one remote_mpid
+ * outside of the CFM module. It's not clear if this is the correct long
+ * term solution or not. */
+ error = (ofproto->ofproto_class->set_cfm
+ ? ofproto->ofproto_class->set_cfm(ofport, s)
+ : EOPNOTSUPP);
+ if (error) {
+ VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
+ ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
+ strerror(error));
}
- shash_destroy(&p->port_by_name);
+}
- switch_status_destroy(p->switch_status);
- netflow_destroy(p->netflow);
- ofproto_sflow_destroy(p->sflow);
+/* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
+ * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
+ * 0 if LACP partner information is not current (generally indicating a
+ * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
+int
+ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
+{
+ struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
+ return (ofport && ofproto->ofproto_class->port_is_lacp_current
+ ? ofproto->ofproto_class->port_is_lacp_current(ofport)
+ : -1);
+}
+\f
+/* Bundles. */
- for (i = 0; i < p->n_listeners; i++) {
- pvconn_close(p->listeners[i]);
- }
- free(p->listeners);
+/* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
+ * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
+ * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
+ * configuration plus, if there is more than one slave, a bonding
+ * configuration.
+ *
+ * If 'aux' is already registered then this function updates its configuration
+ * to 's'. Otherwise, this function registers a new bundle.
+ *
+ * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
+ * port. */
+int
+ofproto_bundle_register(struct ofproto *ofproto, void *aux,
+ const struct ofproto_bundle_settings *s)
+{
+ return (ofproto->ofproto_class->bundle_set
+ ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
+ : EOPNOTSUPP);
+}
- for (i = 0; i < p->n_snoops; i++) {
- pvconn_close(p->snoops[i]);
- }
- free(p->snoops);
+/* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
+ * If no such bundle has been registered, this has no effect. */
+int
+ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
+{
+ return ofproto_bundle_register(ofproto, aux, NULL);
+}
- mac_learning_destroy(p->ml);
+\f
+/* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
+ * If 'aux' is already registered then this function updates its configuration
+ * to 's'. Otherwise, this function registers a new mirror. */
+int
+ofproto_mirror_register(struct ofproto *ofproto, void *aux,
+ const struct ofproto_mirror_settings *s)
+{
+ return (ofproto->ofproto_class->mirror_set
+ ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
+ : EOPNOTSUPP);
+}
- free(p->mfr_desc);
- free(p->hw_desc);
- free(p->sw_desc);
- free(p->serial_desc);
- free(p->dp_desc);
+/* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
+ * If no mirror has been registered, this has no effect. */
+int
+ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
+{
+ return ofproto_mirror_register(ofproto, aux, NULL);
+}
- port_array_destroy(&p->ports);
+/* Retrieves statistics from mirror associated with client data pointer
+ * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
+ * 'bytes', respectively. If a particular counters is not supported,
+ * the appropriate argument is set to UINT64_MAX. */
+int
+ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
+ uint64_t *packets, uint64_t *bytes)
+{
+ if (!ofproto->ofproto_class->mirror_get_stats) {
+ *packets = *bytes = UINT64_MAX;
+ return EOPNOTSUPP;
+ }
- free(p);
+ return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
+ packets, bytes);
}
+/* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
+ * which all packets are flooded, instead of using MAC learning. If
+ * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
+ *
+ * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
+ * port. */
int
-ofproto_run(struct ofproto *p)
+ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
{
- int error = ofproto_run1(p);
- if (!error) {
- error = ofproto_run2(p, false);
- }
- return error;
+ return (ofproto->ofproto_class->set_flood_vlans
+ ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
+ : EOPNOTSUPP);
}
-static void
-process_port_change(struct ofproto *ofproto, int error, char *devname)
+/* Returns true if 'aux' is a registered bundle that is currently in use as the
+ * output for a mirror. */
+bool
+ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
{
- if (error == ENOBUFS) {
- reinit_ports(ofproto);
- } else if (!error) {
- update_port(ofproto, devname);
- free(devname);
- }
+ return (ofproto->ofproto_class->is_mirror_output_bundle
+ ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
+ : false);
}
+\f
+/* Configuration of OpenFlow tables. */
-/* Returns a "preference level" for snooping 'ofconn'. A higher return value
- * means that 'ofconn' is more interesting for monitoring than a lower return
- * value. */
-static int
-snoop_preference(const struct ofconn *ofconn)
-{
- switch (ofconn->role) {
- case NX_ROLE_MASTER:
- return 3;
- case NX_ROLE_OTHER:
- return 2;
- case NX_ROLE_SLAVE:
- return 1;
- default:
- /* Shouldn't happen. */
- return 0;
- }
+/* Returns the number of OpenFlow tables in 'ofproto'. */
+int
+ofproto_get_n_tables(const struct ofproto *ofproto)
+{
+ return ofproto->n_tables;
}
-/* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
- * Connects this vconn to a controller. */
-static void
-add_snooper(struct ofproto *ofproto, struct vconn *vconn)
+/* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
+ * settings from 's'. 'table_id' must be in the range 0 through the number of
+ * OpenFlow tables in 'ofproto' minus 1, inclusive.
+ *
+ * For read-only tables, only the name may be configured. */
+void
+ofproto_configure_table(struct ofproto *ofproto, int table_id,
+ const struct ofproto_table_settings *s)
{
- struct ofconn *ofconn, *best;
+ struct oftable *table;
- /* Pick a controller for monitoring. */
- best = NULL;
- LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
- if (ofconn->type == OFCONN_CONTROLLER
- && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
- best = ofconn;
- }
+ assert(table_id >= 0 && table_id < ofproto->n_tables);
+ table = &ofproto->tables[table_id];
+
+ oftable_set_name(table, s->name);
+
+ if (table->flags & OFTABLE_READONLY) {
+ return;
}
- if (best) {
- rconn_add_monitor(best->rconn, vconn);
+ if (s->groups) {
+ oftable_enable_eviction(table, s->groups, s->n_groups);
} else {
- VLOG_INFO_RL(&rl, "no controller connection to snoop");
- vconn_close(vconn);
+ oftable_disable_eviction(table);
+ }
+
+ table->max_flows = s->max_flows;
+ if (classifier_count(&table->cls) > table->max_flows
+ && table->eviction_fields) {
+ /* 'table' contains more flows than allowed. We might not be able to
+ * evict them right away because of the asynchronous nature of flow
+ * table changes. Schedule eviction for later. */
+ switch (ofproto->state) {
+ case S_OPENFLOW:
+ ofproto->state = S_EVICT;
+ break;
+ case S_EVICT:
+ case S_FLUSH:
+ /* We're already deleting flows, nothing more to do. */
+ break;
+ }
}
}
+\f
+bool
+ofproto_has_snoops(const struct ofproto *ofproto)
+{
+ return connmgr_has_snoops(ofproto->connmgr);
+}
-int
-ofproto_run1(struct ofproto *p)
+void
+ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
{
- struct ofconn *ofconn, *next_ofconn;
- char *devname;
- int error;
- int i;
+ connmgr_get_snoops(ofproto->connmgr, snoops);
+}
+
+static void
+ofproto_flush__(struct ofproto *ofproto)
+{
+ struct ofopgroup *group;
+ struct oftable *table;
- if (shash_is_empty(&p->port_by_name)) {
- init_ports(p);
+ if (ofproto->ofproto_class->flush) {
+ ofproto->ofproto_class->flush(ofproto);
}
- for (i = 0; i < 50; i++) {
- struct ofpbuf *buf;
- int error;
+ group = ofopgroup_create_unattached(ofproto);
+ OFPROTO_FOR_EACH_TABLE (table, ofproto) {
+ struct rule *rule, *next_rule;
+ struct cls_cursor cursor;
- error = dpif_recv(p->dpif, &buf);
- if (error) {
- if (error == ENODEV) {
- /* Someone destroyed the datapath behind our back. The caller
- * better destroy us and give up, because we're just going to
- * spin from here on out. */
- static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
- VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
- dpif_name(p->dpif));
- return ENODEV;
- }
- break;
+ if (table->flags & OFTABLE_HIDDEN) {
+ continue;
}
- handle_odp_msg(p, buf);
+ cls_cursor_init(&cursor, &table->cls, NULL);
+ CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
+ if (!rule->pending) {
+ ofoperation_create(group, rule, OFOPERATION_DELETE);
+ oftable_remove_rule(rule);
+ ofproto->ofproto_class->rule_destruct(rule);
+ }
+ }
}
+ ofopgroup_submit(group);
+}
- while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
- process_port_change(p, error, devname);
- }
- while ((error = netdev_monitor_poll(p->netdev_monitor,
- &devname)) != EAGAIN) {
- process_port_change(p, error, devname);
- }
+static void
+ofproto_destroy__(struct ofproto *ofproto)
+{
+ struct oftable *table;
- if (p->in_band) {
- if (time_msec() >= p->next_in_band_update) {
- update_in_band_remotes(p);
- }
- in_band_run(p->in_band);
+ assert(list_is_empty(&ofproto->pending));
+ assert(!ofproto->n_pending);
+
+ connmgr_destroy(ofproto->connmgr);
+
+ hmap_remove(&all_ofprotos, &ofproto->hmap_node);
+ free(ofproto->name);
+ free(ofproto->type);
+ free(ofproto->mfr_desc);
+ free(ofproto->hw_desc);
+ free(ofproto->sw_desc);
+ free(ofproto->serial_desc);
+ free(ofproto->dp_desc);
+ hmap_destroy(&ofproto->ports);
+ shash_destroy(&ofproto->port_by_name);
+
+ OFPROTO_FOR_EACH_TABLE (table, ofproto) {
+ oftable_destroy(table);
}
+ free(ofproto->tables);
+
+ hmap_destroy(&ofproto->deletions);
+
+ free(ofproto->vlan_bitmap);
+
+ ofproto->ofproto_class->dealloc(ofproto);
+}
+
+void
+ofproto_destroy(struct ofproto *p)
+{
+ struct ofport *ofport, *next_ofport;
- LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
- &p->all_conns) {
- ofconn_run(ofconn, p);
+ if (!p) {
+ return;
}
- /* Fail-open maintenance. Do this after processing the ofconns since
- * fail-open checks the status of the controller rconn. */
- if (p->fail_open) {
- fail_open_run(p->fail_open);
+ ofproto_flush__(p);
+ HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
+ ofport_destroy(ofport);
}
- for (i = 0; i < p->n_listeners; i++) {
- struct vconn *vconn;
- int retval;
+ p->ofproto_class->destruct(p);
+ ofproto_destroy__(p);
+}
+
+/* Destroys the datapath with the respective 'name' and 'type'. With the Linux
+ * kernel datapath, for example, this destroys the datapath in the kernel, and
+ * with the netdev-based datapath, it tears down the data structures that
+ * represent the datapath.
+ *
+ * The datapath should not be currently open as an ofproto. */
+int
+ofproto_delete(const char *name, const char *type)
+{
+ const struct ofproto_class *class = ofproto_class_find__(type);
+ return (!class ? EAFNOSUPPORT
+ : !class->del ? EACCES
+ : class->del(type, name));
+}
+
+static void
+process_port_change(struct ofproto *ofproto, int error, char *devname)
+{
+ if (error == ENOBUFS) {
+ reinit_ports(ofproto);
+ } else if (!error) {
+ update_port(ofproto, devname);
+ free(devname);
+ }
+}
- retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
- if (!retval) {
- struct rconn *rconn;
- char *name;
+int
+ofproto_run(struct ofproto *p)
+{
+ struct ofport *ofport;
+ char *devname;
+ int error;
- rconn = rconn_create(60, 0);
- name = ofconn_make_name(p, vconn_get_name(vconn));
- rconn_connect_unreliably(rconn, vconn, name);
- free(name);
+ error = p->ofproto_class->run(p);
+ if (error && error != EAGAIN) {
+ VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, strerror(error));
+ }
- ofconn_create(p, rconn, OFCONN_TRANSIENT);
- } else if (retval != EAGAIN) {
- VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
+ if (p->ofproto_class->port_poll) {
+ while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
+ process_port_change(p, error, devname);
}
}
- for (i = 0; i < p->n_snoops; i++) {
- struct vconn *vconn;
- int retval;
-
- retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
- if (!retval) {
- add_snooper(p, vconn);
- } else if (retval != EAGAIN) {
- VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
+ HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
+ unsigned int change_seq = netdev_change_seq(ofport->netdev);
+ if (ofport->change_seq != change_seq) {
+ ofport->change_seq = change_seq;
+ update_port(p, netdev_get_name(ofport->netdev));
}
}
- if (time_msec() >= p->next_expiration) {
- COVERAGE_INC(ofproto_expiration);
- p->next_expiration = time_msec() + 1000;
- update_used(p);
+ switch (p->state) {
+ case S_OPENFLOW:
+ connmgr_run(p->connmgr, handle_openflow);
+ break;
- classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
+ case S_EVICT:
+ connmgr_run(p->connmgr, NULL);
+ ofproto_evict(p);
+ if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
+ p->state = S_OPENFLOW;
+ }
+ break;
- /* Let the hook know that we're at a stable point: all outstanding data
- * in existing flows has been accounted to the account_cb. Thus, the
- * hook can now reasonably do operations that depend on having accurate
- * flow volume accounting (currently, that's just bond rebalancing). */
- if (p->ofhooks->account_checkpoint_cb) {
- p->ofhooks->account_checkpoint_cb(p->aux);
+ case S_FLUSH:
+ connmgr_run(p->connmgr, NULL);
+ ofproto_flush__(p);
+ if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
+ connmgr_flushed(p->connmgr);
+ p->state = S_OPENFLOW;
}
- }
+ break;
- if (p->netflow) {
- netflow_run(p->netflow);
- }
- if (p->sflow) {
- ofproto_sflow_run(p->sflow);
+ default:
+ NOT_REACHED();
}
- return 0;
+ return error;
}
-struct revalidate_cbdata {
- struct ofproto *ofproto;
- bool revalidate_all; /* Revalidate all exact-match rules? */
- bool revalidate_subrules; /* Revalidate all exact-match subrules? */
- struct tag_set revalidate_set; /* Set of tags to revalidate. */
-};
-
+/* Performs periodic activity required by 'ofproto' that needs to be done
+ * with the least possible latency.
+ *
+ * It makes sense to call this function a couple of times per poll loop, to
+ * provide a significant performance boost on some benchmarks with the
+ * ofproto-dpif implementation. */
int
-ofproto_run2(struct ofproto *p, bool revalidate_all)
+ofproto_run_fast(struct ofproto *p)
{
- if (p->need_revalidate || revalidate_all
- || !tag_set_is_empty(&p->revalidate_set)) {
- struct revalidate_cbdata cbdata;
- cbdata.ofproto = p;
- cbdata.revalidate_all = revalidate_all;
- cbdata.revalidate_subrules = p->need_revalidate;
- cbdata.revalidate_set = p->revalidate_set;
- tag_set_init(&p->revalidate_set);
- COVERAGE_INC(ofproto_revalidate);
- classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
- p->need_revalidate = false;
- }
+ int error;
- return 0;
+ error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
+ if (error && error != EAGAIN) {
+ VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
+ p->name, strerror(error));
+ }
+ return error;
}
void
ofproto_wait(struct ofproto *p)
{
- struct ofconn *ofconn;
- size_t i;
+ struct ofport *ofport;
- dpif_recv_wait(p->dpif);
- dpif_port_poll_wait(p->dpif);
- netdev_monitor_poll_wait(p->netdev_monitor);
- LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
- ofconn_wait(ofconn);
- }
- if (p->in_band) {
- poll_timer_wait_until(p->next_in_band_update);
- in_band_wait(p->in_band);
+ p->ofproto_class->wait(p);
+ if (p->ofproto_class->port_poll_wait) {
+ p->ofproto_class->port_poll_wait(p);
}
- if (p->fail_open) {
- fail_open_wait(p->fail_open);
- }
- if (p->sflow) {
- ofproto_sflow_wait(p->sflow);
- }
- if (!tag_set_is_empty(&p->revalidate_set)) {
- poll_immediate_wake();
- }
- if (p->need_revalidate) {
- /* Shouldn't happen, but if it does just go around again. */
- VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
- poll_immediate_wake();
- } else if (p->next_expiration != LLONG_MAX) {
- poll_timer_wait_until(p->next_expiration);
- }
- for (i = 0; i < p->n_listeners; i++) {
- pvconn_wait(p->listeners[i]);
+
+ HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
+ if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
+ poll_immediate_wake();
+ }
}
- for (i = 0; i < p->n_snoops; i++) {
- pvconn_wait(p->snoops[i]);
+
+ switch (p->state) {
+ case S_OPENFLOW:
+ connmgr_wait(p->connmgr, true);
+ break;
+
+ case S_EVICT:
+ case S_FLUSH:
+ connmgr_wait(p->connmgr, false);
+ if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
+ poll_immediate_wake();
+ }
+ break;
}
}
+bool
+ofproto_is_alive(const struct ofproto *p)
+{
+ return connmgr_has_controllers(p->connmgr);
+}
+
void
-ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
+ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
+ struct shash *info)
{
- tag_set_add(&ofproto->revalidate_set, tag);
+ connmgr_get_controller_info(ofproto->connmgr, info);
}
-struct tag_set *
-ofproto_get_revalidate_set(struct ofproto *ofproto)
+void
+ofproto_free_ofproto_controller_info(struct shash *info)
+{
+ connmgr_free_controller_info(info);
+}
+
+/* Makes a deep copy of 'old' into 'port'. */
+void
+ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
+{
+ port->name = xstrdup(old->name);
+ port->type = xstrdup(old->type);
+ port->ofp_port = old->ofp_port;
+}
+
+/* Frees memory allocated to members of 'ofproto_port'.
+ *
+ * Do not call this function on an ofproto_port obtained from
+ * ofproto_port_dump_next(): that function retains ownership of the data in the
+ * ofproto_port. */
+void
+ofproto_port_destroy(struct ofproto_port *ofproto_port)
+{
+ free(ofproto_port->name);
+ free(ofproto_port->type);
+}
+
+/* Initializes 'dump' to begin dumping the ports in an ofproto.
+ *
+ * This function provides no status indication. An error status for the entire
+ * dump operation is provided when it is completed by calling
+ * ofproto_port_dump_done().
+ */
+void
+ofproto_port_dump_start(struct ofproto_port_dump *dump,
+ const struct ofproto *ofproto)
{
- return &ofproto->revalidate_set;
+ dump->ofproto = ofproto;
+ dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
+ &dump->state);
}
+/* Attempts to retrieve another port from 'dump', which must have been created
+ * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
+ * 'port' and returns true. On failure, returns false.
+ *
+ * Failure might indicate an actual error or merely that the last port has been
+ * dumped. An error status for the entire dump operation is provided when it
+ * is completed by calling ofproto_port_dump_done().
+ *
+ * The ofproto owns the data stored in 'port'. It will remain valid until at
+ * least the next time 'dump' is passed to ofproto_port_dump_next() or
+ * ofproto_port_dump_done(). */
bool
-ofproto_is_alive(const struct ofproto *p)
+ofproto_port_dump_next(struct ofproto_port_dump *dump,
+ struct ofproto_port *port)
+{
+ const struct ofproto *ofproto = dump->ofproto;
+
+ if (dump->error) {
+ return false;
+ }
+
+ dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
+ port);
+ if (dump->error) {
+ ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
+ return false;
+ }
+ return true;
+}
+
+/* Completes port table dump operation 'dump', which must have been created
+ * with ofproto_port_dump_start(). Returns 0 if the dump operation was
+ * error-free, otherwise a positive errno value describing the problem. */
+int
+ofproto_port_dump_done(struct ofproto_port_dump *dump)
{
- return !hmap_is_empty(&p->controllers);
+ const struct ofproto *ofproto = dump->ofproto;
+ if (!dump->error) {
+ dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
+ dump->state);
+ }
+ return dump->error == EOF ? 0 : dump->error;
}
+/* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
+ * and sets '*ofp_portp' to the new port's OpenFlow port number (if 'ofp_portp'
+ * is non-null). On failure, returns a positive errno value and sets
+ * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
int
-ofproto_send_packet(struct ofproto *p, const flow_t *flow,
- const union ofp_action *actions, size_t n_actions,
- const struct ofpbuf *packet)
+ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
+ uint16_t *ofp_portp)
{
- struct odp_actions odp_actions;
+ uint16_t ofp_port;
int error;
- error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
- NULL, NULL, NULL);
- if (error) {
- return error;
+ error = ofproto->ofproto_class->port_add(ofproto, netdev, &ofp_port);
+ if (!error) {
+ update_port(ofproto, netdev_get_name(netdev));
}
+ if (ofp_portp) {
+ *ofp_portp = error ? OFPP_NONE : ofp_port;
+ }
+ return error;
+}
- /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
- * error code? */
- dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
- odp_actions.n_actions, packet);
- return 0;
+/* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
+ * initializes '*port' appropriately; on failure, returns a positive errno
+ * value.
+ *
+ * The caller owns the data in 'ofproto_port' and must free it with
+ * ofproto_port_destroy() when it is no longer needed. */
+int
+ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
+ struct ofproto_port *port)
+{
+ int error;
+
+ error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
+ if (error) {
+ memset(port, 0, sizeof *port);
+ }
+ return error;
}
-void
-ofproto_add_flow(struct ofproto *p,
- const flow_t *flow, uint32_t wildcards, unsigned int priority,
- const union ofp_action *actions, size_t n_actions,
- int idle_timeout)
+/* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
+ * Returns 0 if successful, otherwise a positive errno. */
+int
+ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
{
- struct rule *rule;
- rule = rule_create(p, NULL, actions, n_actions,
- idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
- 0, 0, false);
- cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
- rule_insert(p, rule, NULL, 0);
+ struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
+ const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
+ int error;
+
+ error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
+ if (!error && ofport) {
+ /* 'name' is the netdev's name and update_port() is going to close the
+ * netdev. Just in case update_port() refers to 'name' after it
+ * destroys 'ofport', make a copy of it around the update_port()
+ * call. */
+ char *devname = xstrdup(name);
+ update_port(ofproto, devname);
+ free(devname);
+ }
+ return error;
}
+/* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
+ * performs the 'n_actions' actions in 'actions'. The new flow will not
+ * timeout.
+ *
+ * If cls_rule->priority is in the range of priorities supported by OpenFlow
+ * (0...65535, inclusive) then the flow will be visible to OpenFlow
+ * controllers; otherwise, it will be hidden.
+ *
+ * The caller retains ownership of 'cls_rule' and 'actions'.
+ *
+ * This is a helper function for in-band control and fail-open. */
void
-ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
- uint32_t wildcards, unsigned int priority)
+ofproto_add_flow(struct ofproto *ofproto, const struct cls_rule *cls_rule,
+ const union ofp_action *actions, size_t n_actions)
{
- struct rule *rule;
+ const struct rule *rule;
- rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
- flow, wildcards,
- priority));
- if (rule) {
- rule_remove(ofproto, rule);
+ rule = rule_from_cls_rule(classifier_find_rule_exactly(
+ &ofproto->tables[0].cls, cls_rule));
+ if (!rule || !ofputil_actions_equal(rule->actions, rule->n_actions,
+ actions, n_actions)) {
+ struct ofputil_flow_mod fm;
+
+ memset(&fm, 0, sizeof fm);
+ fm.cr = *cls_rule;
+ fm.buffer_id = UINT32_MAX;
+ fm.actions = (union ofp_action *) actions;
+ fm.n_actions = n_actions;
+ add_flow(ofproto, NULL, &fm, NULL);
}
}
-static void
-destroy_rule(struct cls_rule *rule_, void *ofproto_)
+/* Executes the flow modification specified in 'fm'. Returns 0 on success, an
+ * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
+ * operation cannot be initiated now but may be retried later.
+ *
+ * This is a helper function for in-band control and fail-open. */
+int
+ofproto_flow_mod(struct ofproto *ofproto, const struct ofputil_flow_mod *fm)
{
- struct rule *rule = rule_from_cls_rule(rule_);
- struct ofproto *ofproto = ofproto_;
+ return handle_flow_mod__(ofproto, NULL, fm, NULL);
+}
+
+/* Searches for a rule with matching criteria exactly equal to 'target' in
+ * ofproto's table 0 and, if it finds one, deletes it.
+ *
+ * This is a helper function for in-band control and fail-open. */
+bool
+ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
+{
+ struct rule *rule;
- /* Mark the flow as not installed, even though it might really be
- * installed, so that rule_remove() doesn't bother trying to uninstall it.
- * There is no point in uninstalling it individually since we are about to
- * blow away all the flows with dpif_flow_flush(). */
- rule->installed = false;
+ rule = rule_from_cls_rule(classifier_find_rule_exactly(
+ &ofproto->tables[0].cls, target));
+ if (!rule) {
+ /* No such rule -> success. */
+ return true;
+ } else if (rule->pending) {
+ /* An operation on the rule is already pending -> failure.
+ * Caller must retry later if it's important. */
+ return false;
+ } else {
+ /* Initiate deletion -> success. */
+ struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
+ ofoperation_create(group, rule, OFOPERATION_DELETE);
+ oftable_remove_rule(rule);
+ ofproto->ofproto_class->rule_destruct(rule);
+ ofopgroup_submit(group);
+ return true;
+ }
- rule_remove(ofproto, rule);
}
+/* Starts the process of deleting all of the flows from all of ofproto's flow
+ * tables and then reintroducing the flows required by in-band control and
+ * fail-open. The process will complete in a later call to ofproto_run(). */
void
ofproto_flush_flows(struct ofproto *ofproto)
{
COVERAGE_INC(ofproto_flush);
- classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
- dpif_flow_flush(ofproto->dpif);
- if (ofproto->in_band) {
- in_band_flushed(ofproto->in_band);
- }
- if (ofproto->fail_open) {
- fail_open_flushed(ofproto->fail_open);
- }
+ ofproto->state = S_FLUSH;
}
\f
static void
reinit_ports(struct ofproto *p)
{
- struct svec devnames;
+ struct ofproto_port_dump dump;
+ struct sset devnames;
struct ofport *ofport;
- unsigned int port_no;
- struct odp_port *odp_ports;
- size_t n_odp_ports;
- size_t i;
+ struct ofproto_port ofproto_port;
+ const char *devname;
- svec_init(&devnames);
- PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
- svec_add (&devnames, (char *) ofport->opp.name);
- }
- dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
- for (i = 0; i < n_odp_ports; i++) {
- svec_add (&devnames, odp_ports[i].devname);
- }
- free(odp_ports);
+ COVERAGE_INC(ofproto_reinit_ports);
- svec_sort_unique(&devnames);
- for (i = 0; i < devnames.n; i++) {
- update_port(p, devnames.names[i]);
+ sset_init(&devnames);
+ HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
+ sset_add(&devnames, netdev_get_name(ofport->netdev));
}
- svec_destroy(&devnames);
-}
-
-static size_t
-refresh_port_group(struct ofproto *p, unsigned int group)
-{
- uint16_t *ports;
- size_t n_ports;
- struct ofport *port;
- unsigned int port_no;
-
- assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
-
- ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
- n_ports = 0;
- PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
- if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
- ports[n_ports++] = port_no;
- }
+ OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
+ sset_add(&devnames, ofproto_port.name);
}
- dpif_port_group_set(p->dpif, group, ports, n_ports);
- free(ports);
-
- return n_ports;
-}
-static void
-refresh_port_groups(struct ofproto *p)
-{
- size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
- size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
- if (p->sflow) {
- ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
+ SSET_FOR_EACH (devname, &devnames) {
+ update_port(p, devname);
}
+ sset_destroy(&devnames);
}
-static struct ofport *
-make_ofport(const struct odp_port *odp_port)
+/* Opens and returns a netdev for 'ofproto_port', or a null pointer if the
+ * netdev cannot be opened. On success, also fills in 'opp'. */
+static struct netdev *
+ofport_open(const struct ofproto_port *ofproto_port,
+ struct ofputil_phy_port *pp)
{
- struct netdev_options netdev_options;
enum netdev_flags flags;
- struct ofport *ofport;
struct netdev *netdev;
- bool carrier;
int error;
- memset(&netdev_options, 0, sizeof netdev_options);
- netdev_options.name = odp_port->devname;
- netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
-
- error = netdev_open(&netdev_options, &netdev);
+ error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
if (error) {
VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
"cannot be opened (%s)",
- odp_port->devname, odp_port->port,
- odp_port->devname, strerror(error));
+ ofproto_port->name, ofproto_port->ofp_port,
+ ofproto_port->name, strerror(error));
return NULL;
}
- ofport = xmalloc(sizeof *ofport);
- ofport->netdev = netdev;
- ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
- netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
- memcpy(ofport->opp.name, odp_port->devname,
- MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
- ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
-
+ pp->port_no = ofproto_port->ofp_port;
+ netdev_get_etheraddr(netdev, pp->hw_addr);
+ ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
netdev_get_flags(netdev, &flags);
- ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
+ pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
+ pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
+ netdev_get_features(netdev, &pp->curr, &pp->advertised,
+ &pp->supported, &pp->peer);
+ pp->curr_speed = netdev_features_to_bps(pp->curr);
+ pp->max_speed = netdev_features_to_bps(pp->supported);
- netdev_get_carrier(netdev, &carrier);
- ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
-
- netdev_get_features(netdev,
- &ofport->opp.curr, &ofport->opp.advertised,
- &ofport->opp.supported, &ofport->opp.peer);
- return ofport;
+ return netdev;
}
+/* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
+ * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
+ * disregarded. */
static bool
-ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
-{
- if (port_array_get(&p->ports, odp_port->port)) {
- VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
- odp_port->port);
- return true;
- } else if (shash_find(&p->port_by_name, odp_port->devname)) {
- VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
- odp_port->devname);
- return true;
- } else {
- return false;
- }
-}
-
-static int
-ofport_equal(const struct ofport *a_, const struct ofport *b_)
+ofport_equal(const struct ofputil_phy_port *a,
+ const struct ofputil_phy_port *b)
{
- const struct ofp_phy_port *a = &a_->opp;
- const struct ofp_phy_port *b = &b_->opp;
-
- BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
- return (a->port_no == b->port_no
- && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
- && !strcmp((char *) a->name, (char *) b->name)
+ return (eth_addr_equals(a->hw_addr, b->hw_addr)
&& a->state == b->state
- && a->config == b->config
+ && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
&& a->curr == b->curr
&& a->advertised == b->advertised
&& a->supported == b->supported
- && a->peer == b->peer);
+ && a->peer == b->peer
+ && a->curr_speed == b->curr_speed
+ && a->max_speed == b->max_speed);
}
+/* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
+ * The caller must ensure that 'p' does not have a conflicting ofport (that is,
+ * one with the same name or port number). */
static void
-send_port_status(struct ofproto *p, const struct ofport *ofport,
- uint8_t reason)
+ofport_install(struct ofproto *p,
+ struct netdev *netdev, const struct ofputil_phy_port *pp)
{
- /* XXX Should limit the number of queued port status change messages. */
- struct ofconn *ofconn;
- LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
- struct ofp_port_status *ops;
- struct ofpbuf *b;
-
- if (!ofconn_receives_async_msgs(ofconn)) {
- continue;
- }
+ const char *netdev_name = netdev_get_name(netdev);
+ struct ofport *ofport;
+ int dev_mtu;
+ int error;
- ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
- ops->reason = reason;
- ops->desc = ofport->opp;
- hton_ofp_phy_port(&ops->desc);
- queue_tx(b, ofconn, NULL);
+ /* Create ofport. */
+ ofport = p->ofproto_class->port_alloc();
+ if (!ofport) {
+ error = ENOMEM;
+ goto error;
}
- if (p->ofhooks->port_changed_cb) {
- p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
+ ofport->ofproto = p;
+ ofport->netdev = netdev;
+ ofport->change_seq = netdev_change_seq(netdev);
+ ofport->pp = *pp;
+ ofport->ofp_port = pp->port_no;
+
+ /* Add port to 'p'. */
+ hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
+ shash_add(&p->port_by_name, netdev_name, ofport);
+
+ if (!netdev_get_mtu(netdev, &dev_mtu)) {
+ set_internal_devs_mtu(p);
+ ofport->mtu = dev_mtu;
+ } else {
+ ofport->mtu = 0;
}
-}
-static void
-ofport_install(struct ofproto *p, struct ofport *ofport)
-{
- uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
- const char *netdev_name = (const char *) ofport->opp.name;
+ /* Let the ofproto_class initialize its private data. */
+ error = p->ofproto_class->port_construct(ofport);
+ if (error) {
+ goto error;
+ }
+ connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
+ return;
- netdev_monitor_add(p->netdev_monitor, ofport->netdev);
- port_array_set(&p->ports, odp_port, ofport);
- shash_add(&p->port_by_name, netdev_name, ofport);
- if (p->sflow) {
- ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
+error:
+ VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
+ p->name, netdev_name, strerror(error));
+ if (ofport) {
+ ofport_destroy__(ofport);
+ } else {
+ netdev_close(netdev);
}
}
+/* Removes 'ofport' from 'p' and destroys it. */
static void
-ofport_remove(struct ofproto *p, struct ofport *ofport)
+ofport_remove(struct ofport *ofport)
{
- uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
-
- netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
- port_array_delete(&p->ports, odp_port);
- shash_delete(&p->port_by_name,
- shash_find(&p->port_by_name, (char *) ofport->opp.name));
- if (p->sflow) {
- ofproto_sflow_del_port(p->sflow, odp_port);
- }
+ connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
+ OFPPR_DELETE);
+ ofport_destroy(ofport);
}
+/* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
+ * destroys it. */
static void
-ofport_free(struct ofport *ofport)
+ofport_remove_with_name(struct ofproto *ofproto, const char *name)
{
- if (ofport) {
- netdev_close(ofport->netdev);
- free(ofport);
+ struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
+ if (port) {
+ ofport_remove(port);
}
}
+/* Updates 'port' with new 'pp' description.
+ *
+ * Does not handle a name or port number change. The caller must implement
+ * such a change as a delete followed by an add. */
static void
-update_port(struct ofproto *p, const char *devname)
+ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
{
- struct odp_port odp_port;
- struct ofport *old_ofport;
- struct ofport *new_ofport;
- int error;
-
- COVERAGE_INC(ofproto_update_port);
+ memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
+ port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
+ | (pp->config & OFPUTIL_PC_PORT_DOWN));
+ port->pp.state = pp->state;
+ port->pp.curr = pp->curr;
+ port->pp.advertised = pp->advertised;
+ port->pp.supported = pp->supported;
+ port->pp.peer = pp->peer;
+ port->pp.curr_speed = pp->curr_speed;
+ port->pp.max_speed = pp->max_speed;
- /* Query the datapath for port information. */
- error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
+ connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
+}
- /* Find the old ofport. */
- old_ofport = shash_find_data(&p->port_by_name, devname);
- if (!error) {
- if (!old_ofport) {
- /* There's no port named 'devname' but there might be a port with
- * the same port number. This could happen if a port is deleted
- * and then a new one added in its place very quickly, or if a port
- * is renamed. In the former case we want to send an OFPPR_DELETE
- * and an OFPPR_ADD, and in the latter case we want to send a
- * single OFPPR_MODIFY. We can distinguish the cases by comparing
- * the old port's ifindex against the new port, or perhaps less
- * reliably but more portably by comparing the old port's MAC
- * against the new port's MAC. However, this code isn't that smart
- * and always sends an OFPPR_MODIFY (XXX). */
- old_ofport = port_array_get(&p->ports, odp_port.port);
- }
- } else if (error != ENOENT && error != ENODEV) {
- VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
- "%s", strerror(error));
- return;
+/* Update OpenFlow 'state' in 'port' and notify controller. */
+void
+ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
+{
+ if (port->pp.state != state) {
+ port->pp.state = state;
+ connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
+ OFPPR_MODIFY);
}
+}
- /* Create a new ofport. */
- new_ofport = !error ? make_ofport(&odp_port) : NULL;
-
- /* Eliminate a few pathological cases. */
- if (!old_ofport && !new_ofport) {
- return;
- } else if (old_ofport && new_ofport) {
- /* Most of the 'config' bits are OpenFlow soft state, but
- * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
- * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
- * leaves the other bits 0.) */
- new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
-
- if (ofport_equal(old_ofport, new_ofport)) {
- /* False alarm--no change. */
- ofport_free(new_ofport);
- return;
+void
+ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
+{
+ struct ofport *port = ofproto_get_port(ofproto, ofp_port);
+ if (port) {
+ if (port->ofproto->ofproto_class->set_realdev) {
+ port->ofproto->ofproto_class->set_realdev(port, 0, 0);
+ }
+ if (port->ofproto->ofproto_class->set_stp_port) {
+ port->ofproto->ofproto_class->set_stp_port(port, NULL);
+ }
+ if (port->ofproto->ofproto_class->set_cfm) {
+ port->ofproto->ofproto_class->set_cfm(port, NULL);
+ }
+ if (port->ofproto->ofproto_class->bundle_remove) {
+ port->ofproto->ofproto_class->bundle_remove(port);
}
}
+}
- /* Now deal with the normal cases. */
- if (old_ofport) {
- ofport_remove(p, old_ofport);
- }
- if (new_ofport) {
- ofport_install(p, new_ofport);
- }
- send_port_status(p, new_ofport ? new_ofport : old_ofport,
- (!old_ofport ? OFPPR_ADD
- : !new_ofport ? OFPPR_DELETE
- : OFPPR_MODIFY));
- ofport_free(old_ofport);
+static void
+ofport_destroy__(struct ofport *port)
+{
+ struct ofproto *ofproto = port->ofproto;
+ const char *name = netdev_get_name(port->netdev);
+
+ hmap_remove(&ofproto->ports, &port->hmap_node);
+ shash_delete(&ofproto->port_by_name,
+ shash_find(&ofproto->port_by_name, name));
- /* Update port groups. */
- refresh_port_groups(p);
+ netdev_close(port->netdev);
+ ofproto->ofproto_class->port_dealloc(port);
}
-static int
-init_ports(struct ofproto *p)
+static void
+ofport_destroy(struct ofport *port)
{
- struct odp_port *ports;
- size_t n_ports;
- size_t i;
- int error;
+ if (port) {
+ port->ofproto->ofproto_class->port_destruct(port);
+ ofport_destroy__(port);
+ }
+}
- error = dpif_port_list(p->dpif, &ports, &n_ports);
- if (error) {
- return error;
- }
+struct ofport *
+ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
+{
+ struct ofport *port;
- for (i = 0; i < n_ports; i++) {
- const struct odp_port *odp_port = &ports[i];
- if (!ofport_conflicts(p, odp_port)) {
- struct ofport *ofport = make_ofport(odp_port);
- if (ofport) {
- ofport_install(p, ofport);
- }
+ HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
+ hash_int(ofp_port, 0), &ofproto->ports) {
+ if (port->ofp_port == ofp_port) {
+ return port;
}
}
- free(ports);
- refresh_port_groups(p);
- return 0;
-}
-\f
-static struct ofconn *
-ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
-{
- struct ofconn *ofconn = xzalloc(sizeof *ofconn);
- ofconn->ofproto = p;
- list_push_back(&p->all_conns, &ofconn->node);
- ofconn->rconn = rconn;
- ofconn->type = type;
- ofconn->role = NX_ROLE_OTHER;
- ofconn->packet_in_counter = rconn_packet_counter_create ();
- ofconn->pktbuf = NULL;
- ofconn->miss_send_len = 0;
- ofconn->reply_counter = rconn_packet_counter_create ();
- return ofconn;
+ return NULL;
}
-static void
-ofconn_destroy(struct ofconn *ofconn)
+int
+ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
{
- if (ofconn->type == OFCONN_CONTROLLER) {
- hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
+ struct ofproto *ofproto = port->ofproto;
+ int error;
+
+ if (ofproto->ofproto_class->port_get_stats) {
+ error = ofproto->ofproto_class->port_get_stats(port, stats);
+ } else {
+ error = EOPNOTSUPP;
}
- discovery_destroy(ofconn->discovery);
- list_remove(&ofconn->node);
- switch_status_unregister(ofconn->ss);
- rconn_destroy(ofconn->rconn);
- rconn_packet_counter_destroy(ofconn->packet_in_counter);
- rconn_packet_counter_destroy(ofconn->reply_counter);
- pktbuf_destroy(ofconn->pktbuf);
- free(ofconn);
+ return error;
}
static void
-ofconn_run(struct ofconn *ofconn, struct ofproto *p)
+update_port(struct ofproto *ofproto, const char *name)
{
- int iteration;
- size_t i;
+ struct ofproto_port ofproto_port;
+ struct ofputil_phy_port pp;
+ struct netdev *netdev;
+ struct ofport *port;
- if (ofconn->discovery) {
- char *controller_name;
- if (rconn_is_connectivity_questionable(ofconn->rconn)) {
- discovery_question_connectivity(ofconn->discovery);
- }
- if (discovery_run(ofconn->discovery, &controller_name)) {
- if (controller_name) {
- char *ofconn_name = ofconn_make_name(p, controller_name);
- rconn_connect(ofconn->rconn, controller_name, ofconn_name);
- free(ofconn_name);
- } else {
- rconn_disconnect(ofconn->rconn);
+ COVERAGE_INC(ofproto_update_port);
+
+ /* Fetch 'name''s location and properties from the datapath. */
+ netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
+ ? ofport_open(&ofproto_port, &pp)
+ : NULL);
+ if (netdev) {
+ port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
+ if (port && !strcmp(netdev_get_name(port->netdev), name)) {
+ struct netdev *old_netdev = port->netdev;
+ int dev_mtu;
+
+ /* 'name' hasn't changed location. Any properties changed? */
+ if (!ofport_equal(&port->pp, &pp)) {
+ ofport_modified(port, &pp);
}
- }
- }
- for (i = 0; i < N_SCHEDULERS; i++) {
- pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
- }
+ /* If this is a non-internal port and the MTU changed, check
+ * if the datapath's MTU needs to be updated. */
+ if (strcmp(netdev_get_type(netdev), "internal")
+ && !netdev_get_mtu(netdev, &dev_mtu)
+ && port->mtu != dev_mtu) {
+ set_internal_devs_mtu(ofproto);
+ port->mtu = dev_mtu;
+ }
- rconn_run(ofconn->rconn);
+ /* Install the newly opened netdev in case it has changed.
+ * Don't close the old netdev yet in case port_modified has to
+ * remove a retained reference to it.*/
+ port->netdev = netdev;
+ port->change_seq = netdev_change_seq(netdev);
- if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
- /* Limit the number of iterations to prevent other tasks from
- * starving. */
- for (iteration = 0; iteration < 50; iteration++) {
- struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
- if (!of_msg) {
- break;
+ if (port->ofproto->ofproto_class->port_modified) {
+ port->ofproto->ofproto_class->port_modified(port);
}
- if (p->fail_open) {
- fail_open_maybe_recover(p->fail_open);
+
+ netdev_close(old_netdev);
+ } else {
+ /* If 'port' is nonnull then its name differs from 'name' and thus
+ * we should delete it. If we think there's a port named 'name'
+ * then its port number must be wrong now so delete it too. */
+ if (port) {
+ ofport_remove(port);
}
- handle_openflow(ofconn, p, of_msg);
- ofpbuf_delete(of_msg);
+ ofport_remove_with_name(ofproto, name);
+ ofport_install(ofproto, netdev, &pp);
}
+ } else {
+ /* Any port named 'name' is gone now. */
+ ofport_remove_with_name(ofproto, name);
}
-
- if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
- ofconn_destroy(ofconn);
- }
+ ofproto_port_destroy(&ofproto_port);
}
-static void
-ofconn_wait(struct ofconn *ofconn)
+static int
+init_ports(struct ofproto *p)
{
- int i;
+ struct ofproto_port_dump dump;
+ struct ofproto_port ofproto_port;
+
+ OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
+ uint16_t ofp_port = ofproto_port.ofp_port;
+ if (ofproto_get_port(p, ofp_port)) {
+ VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
+ ofp_port);
+ } else if (shash_find(&p->port_by_name, ofproto_port.name)) {
+ VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
+ ofproto_port.name);
+ } else {
+ struct ofputil_phy_port pp;
+ struct netdev *netdev;
- if (ofconn->discovery) {
- discovery_wait(ofconn->discovery);
- }
- for (i = 0; i < N_SCHEDULERS; i++) {
- pinsched_wait(ofconn->schedulers[i]);
- }
- rconn_run_wait(ofconn->rconn);
- if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
- rconn_recv_wait(ofconn->rconn);
- } else {
- COVERAGE_INC(ofproto_ofconn_stuck);
+ netdev = ofport_open(&ofproto_port, &pp);
+ if (netdev) {
+ ofport_install(p, netdev, &pp);
+ }
+ }
}
-}
-/* Returns true if 'ofconn' should receive asynchronous messages. */
-static bool
-ofconn_receives_async_msgs(const struct ofconn *ofconn)
-{
- if (ofconn->type == OFCONN_CONTROLLER) {
- /* Ordinary controllers always get asynchronous messages unless they
- * have configured themselves as "slaves". */
- return ofconn->role != NX_ROLE_SLAVE;
- } else {
- /* Transient connections don't get asynchronous messages unless they
- * have explicitly asked for them by setting a nonzero miss send
- * length. */
- return ofconn->miss_send_len > 0;
- }
+ return 0;
}
-/* Returns a human-readable name for an OpenFlow connection between 'ofproto'
- * and 'target', suitable for use in log messages for identifying the
- * connection.
- *
- * The name is dynamically allocated. The caller should free it (with free())
- * when it is no longer needed. */
-static char *
-ofconn_make_name(const struct ofproto *ofproto, const char *target)
+/* Find the minimum MTU of all non-datapath devices attached to 'p'.
+ * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
+static int
+find_min_mtu(struct ofproto *p)
{
- return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
-}
-\f
-/* Caller is responsible for initializing the 'cr' member of the returned
- * rule. */
-static struct rule *
-rule_create(struct ofproto *ofproto, struct rule *super,
- const union ofp_action *actions, size_t n_actions,
- uint16_t idle_timeout, uint16_t hard_timeout,
- uint64_t flow_cookie, bool send_flow_removed)
-{
- struct rule *rule = xzalloc(sizeof *rule);
- rule->idle_timeout = idle_timeout;
- rule->hard_timeout = hard_timeout;
- rule->flow_cookie = flow_cookie;
- rule->used = rule->created = time_msec();
- rule->send_flow_removed = send_flow_removed;
- rule->super = super;
- if (super) {
- list_push_back(&super->list, &rule->list);
- } else {
- list_init(&rule->list);
+ struct ofport *ofport;
+ int mtu = 0;
+
+ HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
+ struct netdev *netdev = ofport->netdev;
+ int dev_mtu;
+
+ /* Skip any internal ports, since that's what we're trying to
+ * set. */
+ if (!strcmp(netdev_get_type(netdev), "internal")) {
+ continue;
+ }
+
+ if (netdev_get_mtu(netdev, &dev_mtu)) {
+ continue;
+ }
+ if (!mtu || dev_mtu < mtu) {
+ mtu = dev_mtu;
+ }
}
- rule->n_actions = n_actions;
- rule->actions = xmemdup(actions, n_actions * sizeof *actions);
- netflow_flow_clear(&rule->nf_flow);
- netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
- return rule;
+ return mtu ? mtu: ETH_PAYLOAD_MAX;
}
-static struct rule *
-rule_from_cls_rule(const struct cls_rule *cls_rule)
+/* Set the MTU of all datapath devices on 'p' to the minimum of the
+ * non-datapath ports. */
+static void
+set_internal_devs_mtu(struct ofproto *p)
{
- return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
-}
+ struct ofport *ofport;
+ int mtu = find_min_mtu(p);
+ HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
+ struct netdev *netdev = ofport->netdev;
+
+ if (!strcmp(netdev_get_type(netdev), "internal")) {
+ netdev_set_mtu(netdev, mtu);
+ }
+ }
+}
+\f
static void
-rule_free(struct rule *rule)
+ofproto_rule_destroy__(struct rule *rule)
{
- free(rule->actions);
- free(rule->odp_actions);
- free(rule);
+ if (rule) {
+ free(rule->actions);
+ rule->ofproto->ofproto_class->rule_dealloc(rule);
+ }
}
-/* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
- * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
- * through all of its subrules and revalidates them, destroying any that no
- * longer has a super-rule (which is probably all of them).
+/* This function allows an ofproto implementation to destroy any rules that
+ * remain when its ->destruct() function is called. The caller must have
+ * already uninitialized any derived members of 'rule' (step 5 described in the
+ * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
+ * This function implements steps 6 and 7.
*
- * Before calling this function, the caller must make have removed 'rule' from
- * the classifier. If 'rule' is an exact-match rule, the caller is also
- * responsible for ensuring that it has been uninstalled from the datapath. */
-static void
-rule_destroy(struct ofproto *ofproto, struct rule *rule)
+ * This function should only be called from an ofproto implementation's
+ * ->destruct() function. It is not suitable elsewhere. */
+void
+ofproto_rule_destroy(struct rule *rule)
{
- if (!rule->super) {
- struct rule *subrule, *next;
- LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
- revalidate_rule(ofproto, subrule);
- }
- } else {
- list_remove(&rule->list);
- }
- rule_free(rule);
+ assert(!rule->pending);
+ oftable_remove_rule(rule);
+ ofproto_rule_destroy__(rule);
}
+/* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
+ * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
+ * count). */
static bool
rule_has_out_port(const struct rule *rule, uint16_t out_port)
{
const union ofp_action *oa;
- struct actions_iterator i;
+ size_t left;
- if (out_port == htons(OFPP_NONE)) {
+ if (out_port == OFPP_NONE) {
return true;
}
- for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
- oa = actions_next(&i)) {
- if (action_outputs_to_port(oa, out_port)) {
+ OFPUTIL_ACTION_FOR_EACH_UNSAFE (oa, left, rule->actions, rule->n_actions) {
+ if (action_outputs_to_port(oa, htons(out_port))) {
return true;
}
}
return false;
}
-/* Executes the actions indicated by 'rule' on 'packet', which is in flow
- * 'flow' and is considered to have arrived on ODP port 'in_port'.
+/* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
+ * statistics appropriately. 'packet' must have at least sizeof(struct
+ * ofp_packet_in) bytes of headroom.
*
- * The flow that 'packet' actually contains does not need to actually match
- * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
- * the packet and byte counters for 'rule' will be credited for the packet sent
- * out whether or not the packet actually matches 'rule'.
+ * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
+ * with statistics for 'packet' either way.
*
- * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
- * the caller must already have accurately composed ODP actions for it given
- * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
- * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
- * function will compose a set of ODP actions based on 'rule''s OpenFlow
- * actions and apply them to 'packet'. */
-static void
-rule_execute(struct ofproto *ofproto, struct rule *rule,
- struct ofpbuf *packet, const flow_t *flow)
+ * Takes ownership of 'packet'. */
+static int
+rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
{
- const union odp_action *actions;
- size_t n_actions;
- struct odp_actions a;
+ struct flow flow;
- /* Grab or compose the ODP actions.
- *
- * The special case for an exact-match 'rule' where 'flow' is not the
- * rule's flow is important to avoid, e.g., sending a packet out its input
- * port simply because the ODP actions were composed for the wrong
- * scenario. */
- if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
- struct rule *super = rule->super ? rule->super : rule;
- if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
- packet, &a, NULL, 0, NULL)) {
- return;
- }
- actions = a.actions;
- n_actions = a.n_actions;
- } else {
- actions = rule->odp_actions;
- n_actions = rule->n_odp_actions;
- }
+ assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
- /* Execute the ODP actions. */
- if (!dpif_execute(ofproto->dpif, flow->in_port,
- actions, n_actions, packet)) {
- struct odp_flow_stats stats;
- flow_extract_stats(flow, packet, &stats);
- update_stats(ofproto, rule, &stats);
- rule->used = time_msec();
- netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
- }
+ flow_extract(packet, 0, 0, in_port, &flow);
+ return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
}
-static void
-rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
- uint16_t in_port)
+/* Returns true if 'rule' should be hidden from the controller.
+ *
+ * Rules with priority higher than UINT16_MAX are set up by ofproto itself
+ * (e.g. by in-band control) and are intentionally hidden from the
+ * controller. */
+static bool
+rule_is_hidden(const struct rule *rule)
{
- struct rule *displaced_rule;
+ return rule->cr.priority > UINT16_MAX;
+}
- /* Insert the rule in the classifier. */
- displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
- if (!rule->cr.wc.wildcards) {
- rule_make_actions(p, rule, packet);
- }
+static enum oftable_flags
+rule_get_flags(const struct rule *rule)
+{
+ return rule->ofproto->tables[rule->table_id].flags;
+}
- /* Send the packet and credit it to the rule. */
- if (packet) {
- flow_t flow;
- flow_extract(packet, 0, in_port, &flow);
- rule_execute(p, rule, packet, &flow);
- }
+static bool
+rule_is_modifiable(const struct rule *rule)
+{
+ return !(rule_get_flags(rule) & OFTABLE_READONLY);
+}
+\f
+static enum ofperr
+handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
+{
+ ofconn_send_reply(ofconn, make_echo_reply(oh));
+ return 0;
+}
- /* Install the rule in the datapath only after sending the packet, to
- * avoid packet reordering. */
- if (rule->cr.wc.wildcards) {
- COVERAGE_INC(ofproto_add_wc_flow);
- p->need_revalidate = true;
- } else {
- rule_install(p, rule, displaced_rule);
- }
+static enum ofperr
+handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
+{
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
+ struct ofputil_switch_features features;
+ struct ofport *port;
+ bool arp_match_ip;
+ struct ofpbuf *b;
- /* Free the rule that was displaced, if any. */
- if (displaced_rule) {
- rule_destroy(p, displaced_rule);
+ ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
+ &features.actions);
+ assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
+
+ features.datapath_id = ofproto->datapath_id;
+ features.n_buffers = pktbuf_capacity();
+ features.n_tables = ofproto->n_tables;
+ features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
+ OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
+ if (arp_match_ip) {
+ features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
}
-}
-static struct rule *
-rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
- const flow_t *flow)
-{
- struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
- rule->idle_timeout, rule->hard_timeout,
- 0, false);
- COVERAGE_INC(ofproto_subrule_create);
- cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
- : rule->cr.priority), &subrule->cr);
- classifier_insert_exact(&ofproto->cls, &subrule->cr);
+ b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
+ oh->xid);
+ HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
+ ofputil_put_switch_features_port(&port->pp, b);
+ }
- return subrule;
+ ofconn_send_reply(ofconn, b);
+ return 0;
}
-static void
-rule_remove(struct ofproto *ofproto, struct rule *rule)
+static enum ofperr
+handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
{
- if (rule->cr.wc.wildcards) {
- COVERAGE_INC(ofproto_del_wc_flow);
- ofproto->need_revalidate = true;
- } else {
- rule_uninstall(ofproto, rule);
- }
- classifier_remove(&ofproto->cls, &rule->cr);
- rule_destroy(ofproto, rule);
-}
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
+ struct ofp_switch_config *osc;
+ enum ofp_config_flags flags;
+ struct ofpbuf *buf;
-/* Returns true if the actions changed, false otherwise. */
-static bool
-rule_make_actions(struct ofproto *p, struct rule *rule,
- const struct ofpbuf *packet)
-{
- const struct rule *super;
- struct odp_actions a;
- size_t actions_len;
-
- assert(!rule->cr.wc.wildcards);
-
- super = rule->super ? rule->super : rule;
- rule->tags = 0;
- xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
- packet, &a, &rule->tags, &rule->may_install,
- &rule->nf_flow.output_iface);
-
- actions_len = a.n_actions * sizeof *a.actions;
- if (rule->n_odp_actions != a.n_actions
- || memcmp(rule->odp_actions, a.actions, actions_len)) {
- COVERAGE_INC(ofproto_odp_unchanged);
- free(rule->odp_actions);
- rule->n_odp_actions = a.n_actions;
- rule->odp_actions = xmemdup(a.actions, actions_len);
- return true;
- } else {
- return false;
+ /* Send reply. */
+ osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
+ flags = ofproto->frag_handling;
+ if (ofconn_get_invalid_ttl_to_controller(ofconn)) {
+ flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
}
+ osc->flags = htons(flags);
+ osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
+ ofconn_send_reply(ofconn, buf);
+
+ return 0;
}
-static int
-do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
- struct odp_flow_put *put)
+static enum ofperr
+handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
{
- memset(&put->flow.stats, 0, sizeof put->flow.stats);
- put->flow.key = rule->cr.flow;
- put->flow.actions = rule->odp_actions;
- put->flow.n_actions = rule->n_odp_actions;
- put->flow.flags = 0;
- put->flags = flags;
- return dpif_flow_put(ofproto->dpif, put);
-}
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
+ uint16_t flags = ntohs(osc->flags);
-static void
-rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
-{
- assert(!rule->cr.wc.wildcards);
-
- if (rule->may_install) {
- struct odp_flow_put put;
- if (!do_put_flow(p, rule,
- ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
- &put)) {
- rule->installed = true;
- if (displaced_rule) {
- update_stats(p, displaced_rule, &put.flow.stats);
- rule_post_uninstall(p, displaced_rule);
+ if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
+ || ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
+ enum ofp_config_flags cur = ofproto->frag_handling;
+ enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
+
+ assert((cur & OFPC_FRAG_MASK) == cur);
+ if (cur != next) {
+ if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
+ ofproto->frag_handling = next;
+ } else {
+ VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
+ ofproto->name,
+ ofputil_frag_handling_to_string(next));
}
}
- } else if (displaced_rule) {
- rule_uninstall(p, displaced_rule);
}
-}
+ ofconn_set_invalid_ttl_to_controller(ofconn,
+ (flags & OFPC_INVALID_TTL_TO_CONTROLLER));
-static void
-rule_reinstall(struct ofproto *ofproto, struct rule *rule)
-{
- if (rule->installed) {
- struct odp_flow_put put;
- COVERAGE_INC(ofproto_dp_missed);
- do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
- } else {
- rule_install(ofproto, rule, NULL);
- }
+ ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
+
+ return 0;
}
-static void
-rule_update_actions(struct ofproto *ofproto, struct rule *rule)
-{
- bool actions_changed;
- uint16_t new_out_iface, old_out_iface;
-
- old_out_iface = rule->nf_flow.output_iface;
- actions_changed = rule_make_actions(ofproto, rule, NULL);
-
- if (rule->may_install) {
- if (rule->installed) {
- if (actions_changed) {
- struct odp_flow_put put;
- do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
- | ODPPF_ZERO_STATS, &put);
- update_stats(ofproto, rule, &put.flow.stats);
-
- /* Temporarily set the old output iface so that NetFlow
- * messages have the correct output interface for the old
- * stats. */
- new_out_iface = rule->nf_flow.output_iface;
- rule->nf_flow.output_iface = old_out_iface;
- rule_post_uninstall(ofproto, rule);
- rule->nf_flow.output_iface = new_out_iface;
- }
- } else {
- rule_install(ofproto, rule, NULL);
- }
+/* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
+ * error message code for the caller to propagate upward. Otherwise, returns
+ * 0.
+ *
+ * The log message mentions 'msg_type'. */
+static enum ofperr
+reject_slave_controller(struct ofconn *ofconn)
+{
+ if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
+ && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
+ return OFPERR_OFPBRC_EPERM;
} else {
- rule_uninstall(ofproto, rule);
+ return 0;
}
}
-static void
-rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
+static enum ofperr
+handle_packet_out(struct ofconn *ofconn, const struct ofp_packet_out *opo)
{
- uint64_t total_bytes = rule->byte_count + extra_bytes;
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
+ struct ofputil_packet_out po;
+ struct ofpbuf *payload;
+ struct flow flow;
+ enum ofperr error;
- if (ofproto->ofhooks->account_flow_cb
- && total_bytes > rule->accounted_bytes)
- {
- ofproto->ofhooks->account_flow_cb(
- &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
- total_bytes - rule->accounted_bytes, ofproto->aux);
- rule->accounted_bytes = total_bytes;
+ COVERAGE_INC(ofproto_packet_out);
+
+ error = reject_slave_controller(ofconn);
+ if (error) {
+ return error;
}
-}
-static void
-rule_uninstall(struct ofproto *p, struct rule *rule)
-{
- assert(!rule->cr.wc.wildcards);
- if (rule->installed) {
- struct odp_flow odp_flow;
-
- odp_flow.key = rule->cr.flow;
- odp_flow.actions = NULL;
- odp_flow.n_actions = 0;
- odp_flow.flags = 0;
- if (!dpif_flow_del(p->dpif, &odp_flow)) {
- update_stats(p, rule, &odp_flow.stats);
- }
- rule->installed = false;
+ /* Decode message. */
+ error = ofputil_decode_packet_out(&po, opo);
+ if (error) {
+ return error;
+ }
- rule_post_uninstall(p, rule);
+ /* Get payload. */
+ if (po.buffer_id != UINT32_MAX) {
+ error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
+ if (error || !payload) {
+ return error;
+ }
+ } else {
+ payload = xmalloc(sizeof *payload);
+ ofpbuf_use_const(payload, po.packet, po.packet_len);
}
-}
-static bool
-is_controller_rule(struct rule *rule)
-{
- /* If the only action is send to the controller then don't report
- * NetFlow expiration messages since it is just part of the control
- * logic for the network and not real traffic. */
+ /* Send out packet. */
+ flow_extract(payload, 0, 0, po.in_port, &flow);
+ error = p->ofproto_class->packet_out(p, payload, &flow,
+ po.actions, po.n_actions);
+ ofpbuf_delete(payload);
- return (rule
- && rule->super
- && rule->super->n_actions == 1
- && action_outputs_to_port(&rule->super->actions[0],
- htons(OFPP_CONTROLLER)));
+ return error;
}
static void
-rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
+update_port_config(struct ofport *port,
+ enum ofputil_port_config config,
+ enum ofputil_port_config mask)
{
- struct rule *super = rule->super;
+ enum ofputil_port_config old_config = port->pp.config;
+ enum ofputil_port_config toggle;
- rule_account(ofproto, rule, 0);
-
- if (ofproto->netflow && !is_controller_rule(rule)) {
- struct ofexpired expired;
- expired.flow = rule->cr.flow;
- expired.packet_count = rule->packet_count;
- expired.byte_count = rule->byte_count;
- expired.used = rule->used;
- netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
+ toggle = (config ^ port->pp.config) & mask;
+ if (toggle & OFPUTIL_PC_PORT_DOWN) {
+ if (config & OFPUTIL_PC_PORT_DOWN) {
+ netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
+ } else {
+ netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
+ }
+ toggle &= ~OFPUTIL_PC_PORT_DOWN;
}
- if (super) {
- super->packet_count += rule->packet_count;
- super->byte_count += rule->byte_count;
-
- /* Reset counters to prevent double counting if the rule ever gets
- * reinstalled. */
- rule->packet_count = 0;
- rule->byte_count = 0;
- rule->accounted_bytes = 0;
- netflow_flow_clear(&rule->nf_flow);
- }
-}
-\f
-static void
-queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
- struct rconn_packet_counter *counter)
-{
- update_openflow_length(msg);
- if (rconn_send(ofconn->rconn, msg, counter)) {
- ofpbuf_delete(msg);
+ port->pp.config ^= toggle;
+ if (port->pp.config != old_config) {
+ port->ofproto->ofproto_class->port_reconfigured(port, old_config);
}
}
-static void
-send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
- int error, const void *data, size_t len)
+static enum ofperr
+handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
{
- struct ofpbuf *buf;
- struct ofp_error_msg *oem;
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
+ struct ofputil_port_mod pm;
+ struct ofport *port;
+ enum ofperr error;
- if (!(error >> 16)) {
- VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
- error);
- return;
+ error = reject_slave_controller(ofconn);
+ if (error) {
+ return error;
}
- COVERAGE_INC(ofproto_error);
- oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
- oh ? oh->xid : 0, &buf);
- oem->type = htons((unsigned int) error >> 16);
- oem->code = htons(error & 0xffff);
- memcpy(oem->data, data, len);
- queue_tx(buf, ofconn, ofconn->reply_counter);
-}
+ error = ofputil_decode_port_mod(oh, &pm);
+ if (error) {
+ return error;
+ }
-static void
-send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
- int error)
-{
- size_t oh_length = ntohs(oh->length);
- send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
-}
-
-static void
-hton_ofp_phy_port(struct ofp_phy_port *opp)
-{
- opp->port_no = htons(opp->port_no);
- opp->config = htonl(opp->config);
- opp->state = htonl(opp->state);
- opp->curr = htonl(opp->curr);
- opp->advertised = htonl(opp->advertised);
- opp->supported = htonl(opp->supported);
- opp->peer = htonl(opp->peer);
-}
-
-static int
-handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
-{
- struct ofp_header *rq = oh;
- queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
+ port = ofproto_get_port(p, pm.port_no);
+ if (!port) {
+ return OFPERR_OFPPMFC_BAD_PORT;
+ } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
+ return OFPERR_OFPPMFC_BAD_HW_ADDR;
+ } else {
+ update_port_config(port, pm.config, pm.mask);
+ if (pm.advertise) {
+ netdev_set_advertisements(port->netdev, pm.advertise);
+ }
+ }
return 0;
}
-static int
-handle_features_request(struct ofproto *p, struct ofconn *ofconn,
- struct ofp_header *oh)
+static enum ofperr
+handle_desc_stats_request(struct ofconn *ofconn,
+ const struct ofp_stats_msg *request)
{
- struct ofp_switch_features *osf;
- struct ofpbuf *buf;
- unsigned int port_no;
- struct ofport *port;
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
+ struct ofp_desc_stats *ods;
+ struct ofpbuf *msg;
+
+ ods = ofputil_make_stats_reply(sizeof *ods, request, &msg);
+ ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
+ ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
+ ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
+ ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
+ ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
+ ofconn_send_reply(ofconn, msg);
- osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
- osf->datapath_id = htonll(p->datapath_id);
- osf->n_buffers = htonl(pktbuf_capacity());
- osf->n_tables = 2;
- osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
- OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
- osf->actions = htonl((1u << OFPAT_OUTPUT) |
- (1u << OFPAT_SET_VLAN_VID) |
- (1u << OFPAT_SET_VLAN_PCP) |
- (1u << OFPAT_STRIP_VLAN) |
- (1u << OFPAT_SET_DL_SRC) |
- (1u << OFPAT_SET_DL_DST) |
- (1u << OFPAT_SET_NW_SRC) |
- (1u << OFPAT_SET_NW_DST) |
- (1u << OFPAT_SET_NW_TOS) |
- (1u << OFPAT_SET_TP_SRC) |
- (1u << OFPAT_SET_TP_DST) |
- (1u << OFPAT_ENQUEUE));
-
- PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
- hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
- }
-
- queue_tx(buf, ofconn, ofconn->reply_counter);
return 0;
}
-static int
-handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
- struct ofp_header *oh)
+static enum ofperr
+handle_table_stats_request(struct ofconn *ofconn,
+ const struct ofp_stats_msg *request)
{
- struct ofpbuf *buf;
- struct ofp_switch_config *osc;
- uint16_t flags;
- bool drop_frags;
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
+ struct ofp_table_stats *ots;
+ struct ofpbuf *msg;
+ size_t i;
- /* Figure out flags. */
- dpif_get_drop_frags(p->dpif, &drop_frags);
- flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
+ ofputil_make_stats_reply(sizeof(struct ofp_stats_msg), request, &msg);
- /* Send reply. */
- osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
- osc->flags = htons(flags);
- osc->miss_send_len = htons(ofconn->miss_send_len);
- queue_tx(buf, ofconn, ofconn->reply_counter);
+ ots = ofpbuf_put_zeros(msg, sizeof *ots * p->n_tables);
+ for (i = 0; i < p->n_tables; i++) {
+ ots[i].table_id = i;
+ sprintf(ots[i].name, "table%zu", i);
+ ots[i].wildcards = htonl(OFPFW_ALL);
+ ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
+ ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
+ }
- return 0;
-}
+ p->ofproto_class->get_tables(p, ots);
-static int
-handle_set_config(struct ofproto *p, struct ofconn *ofconn,
- struct ofp_switch_config *osc)
-{
- uint16_t flags;
- int error;
+ for (i = 0; i < p->n_tables; i++) {
+ const struct oftable *table = &p->tables[i];
- error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
- if (error) {
- return error;
- }
- flags = ntohs(osc->flags);
+ if (table->name) {
+ ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
+ }
- if (ofconn->type == OFCONN_CONTROLLER && ofconn->role != NX_ROLE_SLAVE) {
- switch (flags & OFPC_FRAG_MASK) {
- case OFPC_FRAG_NORMAL:
- dpif_set_drop_frags(p->dpif, false);
- break;
- case OFPC_FRAG_DROP:
- dpif_set_drop_frags(p->dpif, true);
- break;
- default:
- VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
- osc->flags);
- break;
+ if (table->max_flows < ntohl(ots[i].max_entries)) {
+ ots[i].max_entries = htonl(table->max_flows);
}
}
- ofconn->miss_send_len = ntohs(osc->miss_send_len);
-
+ ofconn_send_reply(ofconn, msg);
return 0;
}
static void
-add_output_group_action(struct odp_actions *actions, uint16_t group,
- uint16_t *nf_output_iface)
+append_port_stat(struct ofport *port, struct list *replies)
{
- odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
+ struct netdev_stats stats;
+ struct ofp_port_stats *ops;
+
+ /* Intentionally ignore return value, since errors will set
+ * 'stats' to all-1s, which is correct for OpenFlow, and
+ * netdev_get_stats() will log errors. */
+ ofproto_port_get_stats(port, &stats);
+
+ ops = ofputil_append_stats_reply(sizeof *ops, replies);
+ ops->port_no = htons(port->pp.port_no);
+ memset(ops->pad, 0, sizeof ops->pad);
+ put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
+ put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
+ put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
+ put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
+ put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
+ put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
+ put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
+ put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
+ put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
+ put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
+ put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
+ put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
+}
+
+static enum ofperr
+handle_port_stats_request(struct ofconn *ofconn,
+ const struct ofp_port_stats_request *psr)
+{
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
+ struct ofport *port;
+ struct list replies;
- if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
- *nf_output_iface = NF_OUT_FLOOD;
+ ofputil_start_stats_reply(&psr->osm, &replies);
+ if (psr->port_no != htons(OFPP_NONE)) {
+ port = ofproto_get_port(p, ntohs(psr->port_no));
+ if (port) {
+ append_port_stat(port, &replies);
+ }
+ } else {
+ HMAP_FOR_EACH (port, hmap_node, &p->ports) {
+ append_port_stat(port, &replies);
+ }
}
+
+ ofconn_send_replies(ofconn, &replies);
+ return 0;
}
static void
-add_controller_action(struct odp_actions *actions, uint16_t max_len)
+calc_flow_duration__(long long int start, long long int now,
+ uint32_t *sec, uint32_t *nsec)
{
- union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
- a->controller.arg = max_len;
+ long long int msecs = now - start;
+ *sec = msecs / 1000;
+ *nsec = (msecs % 1000) * (1000 * 1000);
}
-struct action_xlate_ctx {
- /* Input. */
- flow_t flow; /* Flow to which these actions correspond. */
- int recurse; /* Recursion level, via xlate_table_action. */
- struct ofproto *ofproto;
- const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
- * null pointer if we are revalidating
- * without a packet to refer to. */
-
- /* Output. */
- struct odp_actions *out; /* Datapath actions. */
- tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
- bool may_set_up_flow; /* True ordinarily; false if the actions must
- * be reassessed for every packet. */
- uint16_t nf_output_iface; /* Output interface index for NetFlow. */
-};
+/* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
+ * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
+static enum ofperr
+check_table_id(const struct ofproto *ofproto, uint8_t table_id)
+{
+ return (table_id == 0xff || table_id < ofproto->n_tables
+ ? 0
+ : OFPERR_NXBRC_BAD_TABLE_ID);
-static void do_xlate_actions(const union ofp_action *in, size_t n_in,
- struct action_xlate_ctx *ctx);
+}
-static void
-add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
+static struct oftable *
+next_visible_table(struct ofproto *ofproto, uint8_t table_id)
{
- const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
+ struct oftable *table;
- if (ofport) {
- if (ofport->opp.config & OFPPC_NO_FWD) {
- /* Forwarding disabled on port. */
- return;
+ for (table = &ofproto->tables[table_id];
+ table < &ofproto->tables[ofproto->n_tables];
+ table++) {
+ if (!(table->flags & OFTABLE_HIDDEN)) {
+ return table;
}
- } else {
- /*
- * We don't have an ofport record for this port, but it doesn't hurt to
- * allow forwarding to it anyhow. Maybe such a port will appear later
- * and we're pre-populating the flow table.
- */
}
- odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
- ctx->nf_output_iface = port;
+ return NULL;
}
-static struct rule *
-lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
+static struct oftable *
+first_matching_table(struct ofproto *ofproto, uint8_t table_id)
{
- struct rule *rule;
- rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
-
- /* The rule we found might not be valid, since we could be in need of
- * revalidation. If it is not valid, don't return it. */
- if (rule
- && rule->super
- && ofproto->need_revalidate
- && !revalidate_rule(ofproto, rule)) {
- COVERAGE_INC(ofproto_invalidated);
+ if (table_id == 0xff) {
+ return next_visible_table(ofproto, 0);
+ } else if (table_id < ofproto->n_tables) {
+ return &ofproto->tables[table_id];
+ } else {
return NULL;
}
+}
- return rule;
+static struct oftable *
+next_matching_table(struct ofproto *ofproto,
+ struct oftable *table, uint8_t table_id)
+{
+ return (table_id == 0xff
+ ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
+ : NULL);
}
-static void
-xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
+/* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
+ *
+ * - If TABLE_ID is 0xff, this iterates over every classifier table in
+ * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
+ *
+ * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
+ * only once, for that table. (This can be used to access tables marked
+ * OFTABLE_HIDDEN.)
+ *
+ * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
+ * entered at all. (Perhaps you should have validated TABLE_ID with
+ * check_table_id().)
+ *
+ * All parameters are evaluated multiple times.
+ */
+#define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
+ for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
+ (TABLE) != NULL; \
+ (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
+
+/* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
+ * 'table_id' is 0xff) that match 'match' in the "loose" way required for
+ * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
+ * 'rules'.
+ *
+ * If 'out_port' is anything other than OFPP_NONE, then only rules that output
+ * to 'out_port' are included.
+ *
+ * Hidden rules are always omitted.
+ *
+ * Returns 0 on success, otherwise an OpenFlow error code. */
+static enum ofperr
+collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
+ const struct cls_rule *match,
+ ovs_be64 cookie, ovs_be64 cookie_mask,
+ uint16_t out_port, struct list *rules)
{
- if (!ctx->recurse) {
- uint16_t old_in_port;
- struct rule *rule;
+ struct oftable *table;
+ enum ofperr error;
+
+ error = check_table_id(ofproto, table_id);
+ if (error) {
+ return error;
+ }
- /* Look up a flow with 'in_port' as the input port. Then restore the
- * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
- * have surprising behavior). */
- old_in_port = ctx->flow.in_port;
- ctx->flow.in_port = in_port;
- rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
- ctx->flow.in_port = old_in_port;
+ list_init(rules);
+ FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
+ struct cls_cursor cursor;
+ struct rule *rule;
- if (rule) {
- if (rule->super) {
- rule = rule->super;
+ cls_cursor_init(&cursor, &table->cls, match);
+ CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
+ if (rule->pending) {
+ return OFPROTO_POSTPONE;
+ }
+ if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)
+ && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
+ list_push_back(rules, &rule->ofproto_node);
}
-
- ctx->recurse++;
- do_xlate_actions(rule->actions, rule->n_actions, ctx);
- ctx->recurse--;
}
}
+ return 0;
}
-static void
-xlate_output_action__(struct action_xlate_ctx *ctx,
- uint16_t port, uint16_t max_len)
-{
- uint16_t odp_port;
- uint16_t prev_nf_output_iface = ctx->nf_output_iface;
-
- ctx->nf_output_iface = NF_OUT_DROP;
+/* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
+ * 'table_id' is 0xff) that match 'match' in the "strict" way required for
+ * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
+ * on list 'rules'.
+ *
+ * If 'out_port' is anything other than OFPP_NONE, then only rules that output
+ * to 'out_port' are included.
+ *
+ * Hidden rules are always omitted.
+ *
+ * Returns 0 on success, otherwise an OpenFlow error code. */
+static enum ofperr
+collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
+ const struct cls_rule *match,
+ ovs_be64 cookie, ovs_be64 cookie_mask,
+ uint16_t out_port, struct list *rules)
+{
+ struct oftable *table;
+ int error;
- switch (port) {
- case OFPP_IN_PORT:
- add_output_action(ctx, ctx->flow.in_port);
- break;
- case OFPP_TABLE:
- xlate_table_action(ctx, ctx->flow.in_port);
- break;
- case OFPP_NORMAL:
- if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
- ctx->out, ctx->tags,
- &ctx->nf_output_iface,
- ctx->ofproto->aux)) {
- COVERAGE_INC(ofproto_uninstallable);
- ctx->may_set_up_flow = false;
- }
- break;
- case OFPP_FLOOD:
- add_output_group_action(ctx->out, DP_GROUP_FLOOD,
- &ctx->nf_output_iface);
- break;
- case OFPP_ALL:
- add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
- break;
- case OFPP_CONTROLLER:
- add_controller_action(ctx->out, max_len);
- break;
- case OFPP_LOCAL:
- add_output_action(ctx, ODPP_LOCAL);
- break;
- default:
- odp_port = ofp_port_to_odp_port(port);
- if (odp_port != ctx->flow.in_port) {
- add_output_action(ctx, odp_port);
- }
- break;
+ error = check_table_id(ofproto, table_id);
+ if (error) {
+ return error;
}
- if (prev_nf_output_iface == NF_OUT_FLOOD) {
- ctx->nf_output_iface = NF_OUT_FLOOD;
- } else if (ctx->nf_output_iface == NF_OUT_DROP) {
- ctx->nf_output_iface = prev_nf_output_iface;
- } else if (prev_nf_output_iface != NF_OUT_DROP &&
- ctx->nf_output_iface != NF_OUT_FLOOD) {
- ctx->nf_output_iface = NF_OUT_MULTI;
- }
-}
+ list_init(rules);
+ FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
+ struct rule *rule;
-static void
-xlate_output_action(struct action_xlate_ctx *ctx,
- const struct ofp_action_output *oao)
-{
- xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
+ rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls,
+ match));
+ if (rule) {
+ if (rule->pending) {
+ return OFPROTO_POSTPONE;
+ }
+ if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)
+ && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
+ list_push_back(rules, &rule->ofproto_node);
+ }
+ }
+ }
+ return 0;
}
-/* If the final ODP action in 'ctx' is "pop priority", drop it, as an
- * optimization, because we're going to add another action that sets the
- * priority immediately after, or because there are no actions following the
- * pop. */
-static void
-remove_pop_action(struct action_xlate_ctx *ctx)
+/* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
+ * forced into the range of a uint16_t. */
+static int
+age_secs(long long int age_ms)
{
- size_t n = ctx->out->n_actions;
- if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) {
- ctx->out->n_actions--;
- }
+ return (age_ms < 0 ? 0
+ : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
+ : (unsigned int) age_ms / 1000);
}
-static void
-xlate_enqueue_action(struct action_xlate_ctx *ctx,
- const struct ofp_action_enqueue *oae)
+static enum ofperr
+handle_flow_stats_request(struct ofconn *ofconn,
+ const struct ofp_stats_msg *osm)
{
- uint16_t ofp_port, odp_port;
- uint32_t priority;
- int error;
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
+ struct ofputil_flow_stats_request fsr;
+ struct list replies;
+ struct list rules;
+ struct rule *rule;
+ enum ofperr error;
- error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
- &priority);
+ error = ofputil_decode_flow_stats_request(&fsr, &osm->header);
if (error) {
- /* Fall back to ordinary output action. */
- xlate_output_action__(ctx, ntohs(oae->port), 0);
- return;
+ return error;
}
- /* Figure out ODP output port. */
- ofp_port = ntohs(oae->port);
- if (ofp_port != OFPP_IN_PORT) {
- odp_port = ofp_port_to_odp_port(ofp_port);
- } else {
- odp_port = ctx->flow.in_port;
+ error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
+ fsr.cookie, fsr.cookie_mask,
+ fsr.out_port, &rules);
+ if (error) {
+ return error;
}
- /* Add ODP actions. */
- remove_pop_action(ctx);
- odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
- = priority;
- add_output_action(ctx, odp_port);
- odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
+ ofputil_start_stats_reply(osm, &replies);
+ LIST_FOR_EACH (rule, ofproto_node, &rules) {
+ long long int now = time_msec();
+ struct ofputil_flow_stats fs;
+
+ fs.rule = rule->cr;
+ fs.cookie = rule->flow_cookie;
+ fs.table_id = rule->table_id;
+ calc_flow_duration__(rule->created, now, &fs.duration_sec,
+ &fs.duration_nsec);
+ fs.idle_timeout = rule->idle_timeout;
+ fs.hard_timeout = rule->hard_timeout;
+ fs.idle_age = age_secs(now - rule->used);
+ fs.hard_age = age_secs(now - rule->modified);
+ ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
+ &fs.byte_count);
+ fs.actions = rule->actions;
+ fs.n_actions = rule->n_actions;
+ ofputil_append_flow_stats_reply(&fs, &replies);
+ }
+ ofconn_send_replies(ofconn, &replies);
- /* Update NetFlow output port. */
- if (ctx->nf_output_iface == NF_OUT_DROP) {
- ctx->nf_output_iface = odp_port;
- } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
- ctx->nf_output_iface = NF_OUT_MULTI;
- }
+ return 0;
}
static void
-xlate_nicira_action(struct action_xlate_ctx *ctx,
- const struct nx_action_header *nah)
-{
- const struct nx_action_resubmit *nar;
- const struct nx_action_set_tunnel *nast;
- union odp_action *oa;
- int subtype = ntohs(nah->subtype);
-
- assert(nah->vendor == htonl(NX_VENDOR_ID));
- switch (subtype) {
- case NXAST_RESUBMIT:
- nar = (const struct nx_action_resubmit *) nah;
- xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
- break;
-
- case NXAST_SET_TUNNEL:
- nast = (const struct nx_action_set_tunnel *) nah;
- oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
- ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
- break;
+flow_stats_ds(struct rule *rule, struct ds *results)
+{
+ uint64_t packet_count, byte_count;
- /* If you add a new action here that modifies flow data, don't forget to
- * update the flow key in ctx->flow at the same time. */
+ rule->ofproto->ofproto_class->rule_get_stats(rule,
+ &packet_count, &byte_count);
- default:
- VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
- break;
+ if (rule->table_id != 0) {
+ ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
}
+ ds_put_format(results, "duration=%llds, ",
+ (time_msec() - rule->created) / 1000);
+ ds_put_format(results, "priority=%u, ", rule->cr.priority);
+ ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
+ ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
+ cls_rule_format(&rule->cr, results);
+ ds_put_char(results, ',');
+ if (rule->n_actions > 0) {
+ ofp_print_actions(results, rule->actions, rule->n_actions);
+ } else {
+ ds_put_cstr(results, "drop");
+ }
+ ds_put_cstr(results, "\n");
}
-static void
-do_xlate_actions(const union ofp_action *in, size_t n_in,
- struct action_xlate_ctx *ctx)
+/* Adds a pretty-printed description of all flows to 'results', including
+ * hidden flows (e.g., set up by in-band control). */
+void
+ofproto_get_all_flows(struct ofproto *p, struct ds *results)
{
- struct actions_iterator iter;
- const union ofp_action *ia;
- const struct ofport *port;
-
- port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port);
- if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
- port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, stp_eth_addr)
- ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
- /* Drop this flow. */
- return;
- }
-
- for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
- uint16_t type = ntohs(ia->type);
- union odp_action *oa;
-
- switch (type) {
- case OFPAT_OUTPUT:
- xlate_output_action(ctx, &ia->output);
- break;
-
- case OFPAT_SET_VLAN_VID:
- oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
- ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
- break;
-
- case OFPAT_SET_VLAN_PCP:
- oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
- ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
- break;
-
- case OFPAT_STRIP_VLAN:
- odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
- ctx->flow.dl_vlan = htons(OFP_VLAN_NONE);
- ctx->flow.dl_vlan_pcp = 0;
- break;
+ struct oftable *table;
- case OFPAT_SET_DL_SRC:
- oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
- memcpy(oa->dl_addr.dl_addr,
- ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
- memcpy(ctx->flow.dl_src,
- ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
- break;
-
- case OFPAT_SET_DL_DST:
- oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
- memcpy(oa->dl_addr.dl_addr,
- ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
- memcpy(ctx->flow.dl_dst,
- ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
- break;
-
- case OFPAT_SET_NW_SRC:
- oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
- ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
- break;
-
- case OFPAT_SET_NW_DST:
- oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
- ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
- break;
-
- case OFPAT_SET_NW_TOS:
- oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
- ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
- break;
-
- case OFPAT_SET_TP_SRC:
- oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
- ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
- break;
-
- case OFPAT_SET_TP_DST:
- oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
- ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
- break;
-
- case OFPAT_VENDOR:
- xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
- break;
-
- case OFPAT_ENQUEUE:
- xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
- break;
+ OFPROTO_FOR_EACH_TABLE (table, p) {
+ struct cls_cursor cursor;
+ struct rule *rule;
- default:
- VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
- break;
+ cls_cursor_init(&cursor, &table->cls, NULL);
+ CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
+ flow_stats_ds(rule, results);
}
}
}
-static int
-xlate_actions(const union ofp_action *in, size_t n_in,
- const flow_t *flow, struct ofproto *ofproto,
- const struct ofpbuf *packet,
- struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
- uint16_t *nf_output_iface)
-{
- tag_type no_tags = 0;
- struct action_xlate_ctx ctx;
- COVERAGE_INC(ofproto_ofp2odp);
- odp_actions_init(out);
- ctx.flow = *flow;
- ctx.recurse = 0;
- ctx.ofproto = ofproto;
- ctx.packet = packet;
- ctx.out = out;
- ctx.tags = tags ? tags : &no_tags;
- ctx.may_set_up_flow = true;
- ctx.nf_output_iface = NF_OUT_DROP;
- do_xlate_actions(in, n_in, &ctx);
- remove_pop_action(&ctx);
-
- /* Check with in-band control to see if we're allowed to set up this
- * flow. */
- if (!in_band_rule_check(ofproto->in_band, flow, out)) {
- ctx.may_set_up_flow = false;
- }
-
- if (may_set_up_flow) {
- *may_set_up_flow = ctx.may_set_up_flow;
- }
- if (nf_output_iface) {
- *nf_output_iface = ctx.nf_output_iface;
- }
- if (odp_actions_overflow(out)) {
- odp_actions_init(out);
- return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
- }
- return 0;
+/* Obtains the NetFlow engine type and engine ID for 'ofproto' into
+ * '*engine_type' and '*engine_id', respectively. */
+void
+ofproto_get_netflow_ids(const struct ofproto *ofproto,
+ uint8_t *engine_type, uint8_t *engine_id)
+{
+ ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
}
-/* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
- * error message code (composed with ofp_mkerr()) for the caller to propagate
- * upward. Otherwise, returns 0.
- *
- * 'oh' is used to make log messages more informative. */
-static int
-reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh)
+/* Checks the fault status of CFM for 'ofp_port' within 'ofproto'. Returns a
+ * bitmask of 'cfm_fault_reason's to indicate a CFM fault (generally
+ * indicating a connectivity problem). Returns zero if CFM is not faulted,
+ * and -1 if CFM is not enabled on 'port'. */
+int
+ofproto_port_get_cfm_fault(const struct ofproto *ofproto, uint16_t ofp_port)
{
- if (ofconn->type == OFCONN_CONTROLLER && ofconn->role == NX_ROLE_SLAVE) {
- static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
- char *type_name;
-
- type_name = ofp_message_type_to_string(oh->type);
- VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
- type_name);
- free(type_name);
-
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
- } else {
- return 0;
- }
+ struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
+ return (ofport && ofproto->ofproto_class->get_cfm_fault
+ ? ofproto->ofproto_class->get_cfm_fault(ofport)
+ : -1);
}
-static int
-handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
- struct ofp_header *oh)
-{
- struct ofp_packet_out *opo;
- struct ofpbuf payload, *buffer;
- struct odp_actions actions;
- int n_actions;
- uint16_t in_port;
- flow_t flow;
- int error;
+/* Gets the MPIDs of the remote maintenance points broadcasting to 'ofp_port'
+ * within 'ofproto'. Populates 'rmps' with an array of MPIDs owned by
+ * 'ofproto', and 'n_rmps' with the number of MPIDs in 'rmps'. Returns a
+ * number less than 0 if CFM is not enabled on 'ofp_port'. */
+int
+ofproto_port_get_cfm_remote_mpids(const struct ofproto *ofproto,
+ uint16_t ofp_port, const uint64_t **rmps,
+ size_t *n_rmps)
+{
+ struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
+
+ *rmps = NULL;
+ *n_rmps = 0;
+ return (ofport && ofproto->ofproto_class->get_cfm_remote_mpids
+ ? ofproto->ofproto_class->get_cfm_remote_mpids(ofport, rmps,
+ n_rmps)
+ : -1);
+}
+
+static enum ofperr
+handle_aggregate_stats_request(struct ofconn *ofconn,
+ const struct ofp_stats_msg *osm)
+{
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
+ struct ofputil_flow_stats_request request;
+ struct ofputil_aggregate_stats stats;
+ bool unknown_packets, unknown_bytes;
+ struct ofpbuf *reply;
+ struct list rules;
+ struct rule *rule;
+ enum ofperr error;
- error = reject_slave_controller(ofconn, oh);
+ error = ofputil_decode_flow_stats_request(&request, &osm->header);
if (error) {
return error;
}
- error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
+ error = collect_rules_loose(ofproto, request.table_id, &request.match,
+ request.cookie, request.cookie_mask,
+ request.out_port, &rules);
if (error) {
return error;
}
- opo = (struct ofp_packet_out *) oh;
- COVERAGE_INC(ofproto_packet_out);
- if (opo->buffer_id != htonl(UINT32_MAX)) {
- error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
- &buffer, &in_port);
- if (error || !buffer) {
- return error;
- }
- payload = *buffer;
- } else {
- buffer = NULL;
- }
-
- flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
- error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
- &flow, p, &payload, &actions, NULL, NULL, NULL);
- if (error) {
- return error;
- }
+ memset(&stats, 0, sizeof stats);
+ unknown_packets = unknown_bytes = false;
+ LIST_FOR_EACH (rule, ofproto_node, &rules) {
+ uint64_t packet_count;
+ uint64_t byte_count;
- dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
- &payload);
- ofpbuf_delete(buffer);
+ ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
+ &byte_count);
- return 0;
-}
-
-static void
-update_port_config(struct ofproto *p, struct ofport *port,
- uint32_t config, uint32_t mask)
-{
- mask &= config ^ port->opp.config;
- if (mask & OFPPC_PORT_DOWN) {
- if (config & OFPPC_PORT_DOWN) {
- netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
+ if (packet_count == UINT64_MAX) {
+ unknown_packets = true;
} else {
- netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
+ stats.packet_count += packet_count;
}
- }
-#define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
- if (mask & REVALIDATE_BITS) {
- COVERAGE_INC(ofproto_costly_flags);
- port->opp.config ^= mask & REVALIDATE_BITS;
- p->need_revalidate = true;
- }
-#undef REVALIDATE_BITS
- if (mask & OFPPC_NO_FLOOD) {
- port->opp.config ^= OFPPC_NO_FLOOD;
- refresh_port_groups(p);
- }
- if (mask & OFPPC_NO_PACKET_IN) {
- port->opp.config ^= OFPPC_NO_PACKET_IN;
- }
-}
-static int
-handle_port_mod(struct ofproto *p, struct ofconn *ofconn,
- struct ofp_header *oh)
-{
- const struct ofp_port_mod *opm;
- struct ofport *port;
- int error;
+ if (byte_count == UINT64_MAX) {
+ unknown_bytes = true;
+ } else {
+ stats.byte_count += byte_count;
+ }
- error = reject_slave_controller(ofconn, oh);
- if (error) {
- return error;
+ stats.flow_count++;
}
- error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
- if (error) {
- return error;
+ if (unknown_packets) {
+ stats.packet_count = UINT64_MAX;
}
- opm = (struct ofp_port_mod *) oh;
-
- port = port_array_get(&p->ports,
- ofp_port_to_odp_port(ntohs(opm->port_no)));
- if (!port) {
- return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
- } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
- return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
- } else {
- update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
- if (opm->advertise) {
- netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
- }
+ if (unknown_bytes) {
+ stats.byte_count = UINT64_MAX;
}
- return 0;
-}
-static struct ofpbuf *
-make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
-{
- struct ofp_stats_reply *osr;
- struct ofpbuf *msg;
-
- msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
- osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
- osr->type = type;
- osr->flags = htons(0);
- return msg;
-}
+ reply = ofputil_encode_aggregate_stats_reply(&stats, osm);
+ ofconn_send_reply(ofconn, reply);
-static struct ofpbuf *
-start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
-{
- return make_stats_reply(request->header.xid, request->type, body_len);
+ return 0;
}
-static void *
-append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
-{
- struct ofpbuf *msg = *msgp;
- assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
- if (nbytes + msg->size > UINT16_MAX) {
- struct ofp_stats_reply *reply = msg->data;
- reply->flags = htons(OFPSF_REPLY_MORE);
- *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
- queue_tx(msg, ofconn, ofconn->reply_counter);
- }
- return ofpbuf_put_uninit(*msgp, nbytes);
-}
+struct queue_stats_cbdata {
+ struct ofport *ofport;
+ struct list replies;
+};
-static int
-handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
- struct ofp_stats_request *request)
+static void
+put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
+ const struct netdev_queue_stats *stats)
{
- struct ofp_desc_stats *ods;
- struct ofpbuf *msg;
-
- msg = start_stats_reply(request, sizeof *ods);
- ods = append_stats_reply(sizeof *ods, ofconn, &msg);
- memset(ods, 0, sizeof *ods);
- ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
- ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
- ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
- ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
- ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
- queue_tx(msg, ofconn, ofconn->reply_counter);
+ struct ofp_queue_stats *reply;
- return 0;
+ reply = ofputil_append_stats_reply(sizeof *reply, &cbdata->replies);
+ reply->port_no = htons(cbdata->ofport->pp.port_no);
+ memset(reply->pad, 0, sizeof reply->pad);
+ reply->queue_id = htonl(queue_id);
+ put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
+ put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
+ put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
}
static void
-count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
+handle_queue_stats_dump_cb(uint32_t queue_id,
+ struct netdev_queue_stats *stats,
+ void *cbdata_)
{
- struct rule *rule = rule_from_cls_rule(cls_rule);
- int *n_subrules = n_subrules_;
-
- if (rule->super) {
- (*n_subrules)++;
- }
-}
+ struct queue_stats_cbdata *cbdata = cbdata_;
-static int
-handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
- struct ofp_stats_request *request)
-{
- struct ofp_table_stats *ots;
- struct ofpbuf *msg;
- struct odp_stats dpstats;
- int n_exact, n_subrules, n_wild;
-
- msg = start_stats_reply(request, sizeof *ots * 2);
-
- /* Count rules of various kinds. */
- n_subrules = 0;
- classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
- n_exact = classifier_count_exact(&p->cls) - n_subrules;
- n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
-
- /* Hash table. */
- dpif_get_dp_stats(p->dpif, &dpstats);
- ots = append_stats_reply(sizeof *ots, ofconn, &msg);
- memset(ots, 0, sizeof *ots);
- ots->table_id = TABLEID_HASH;
- strcpy(ots->name, "hash");
- ots->wildcards = htonl(0);
- ots->max_entries = htonl(dpstats.max_capacity);
- ots->active_count = htonl(n_exact);
- ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
- dpstats.n_missed);
- ots->matched_count = htonll(dpstats.n_hit); /* XXX */
-
- /* Classifier table. */
- ots = append_stats_reply(sizeof *ots, ofconn, &msg);
- memset(ots, 0, sizeof *ots);
- ots->table_id = TABLEID_CLASSIFIER;
- strcpy(ots->name, "classifier");
- ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
- : htonl(OFPFW_ALL);
- ots->max_entries = htonl(65536);
- ots->active_count = htonl(n_wild);
- ots->lookup_count = htonll(0); /* XXX */
- ots->matched_count = htonll(0); /* XXX */
-
- queue_tx(msg, ofconn, ofconn->reply_counter);
- return 0;
+ put_queue_stats(cbdata, queue_id, stats);
}
static void
-append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
- struct ofpbuf **msgp)
+handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
+ struct queue_stats_cbdata *cbdata)
{
- struct netdev_stats stats;
- struct ofp_port_stats *ops;
-
- /* Intentionally ignore return value, since errors will set
- * 'stats' to all-1s, which is correct for OpenFlow, and
- * netdev_get_stats() will log errors. */
- netdev_get_stats(port->netdev, &stats);
+ cbdata->ofport = port;
+ if (queue_id == OFPQ_ALL) {
+ netdev_dump_queue_stats(port->netdev,
+ handle_queue_stats_dump_cb, cbdata);
+ } else {
+ struct netdev_queue_stats stats;
- ops = append_stats_reply(sizeof *ops, ofconn, msgp);
- ops->port_no = htons(odp_port_to_ofp_port(port_no));
- memset(ops->pad, 0, sizeof ops->pad);
- ops->rx_packets = htonll(stats.rx_packets);
- ops->tx_packets = htonll(stats.tx_packets);
- ops->rx_bytes = htonll(stats.rx_bytes);
- ops->tx_bytes = htonll(stats.tx_bytes);
- ops->rx_dropped = htonll(stats.rx_dropped);
- ops->tx_dropped = htonll(stats.tx_dropped);
- ops->rx_errors = htonll(stats.rx_errors);
- ops->tx_errors = htonll(stats.tx_errors);
- ops->rx_frame_err = htonll(stats.rx_frame_errors);
- ops->rx_over_err = htonll(stats.rx_over_errors);
- ops->rx_crc_err = htonll(stats.rx_crc_errors);
- ops->collisions = htonll(stats.collisions);
+ if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
+ put_queue_stats(cbdata, queue_id, &stats);
+ }
+ }
}
-static int
-handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
- struct ofp_stats_request *osr,
- size_t arg_size)
+static enum ofperr
+handle_queue_stats_request(struct ofconn *ofconn,
+ const struct ofp_queue_stats_request *qsr)
{
- struct ofp_port_stats_request *psr;
- struct ofp_port_stats *ops;
- struct ofpbuf *msg;
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
+ struct queue_stats_cbdata cbdata;
struct ofport *port;
unsigned int port_no;
+ uint32_t queue_id;
- if (arg_size != sizeof *psr) {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
- }
- psr = (struct ofp_port_stats_request *) osr->body;
+ COVERAGE_INC(ofproto_queue_req);
- msg = start_stats_reply(osr, sizeof *ops * 16);
- if (psr->port_no != htons(OFPP_NONE)) {
- port = port_array_get(&p->ports,
- ofp_port_to_odp_port(ntohs(psr->port_no)));
+ ofputil_start_stats_reply(&qsr->osm, &cbdata.replies);
+
+ port_no = ntohs(qsr->port_no);
+ queue_id = ntohl(qsr->queue_id);
+ if (port_no == OFPP_ALL) {
+ HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
+ handle_queue_stats_for_port(port, queue_id, &cbdata);
+ }
+ } else if (port_no < OFPP_MAX) {
+ port = ofproto_get_port(ofproto, port_no);
if (port) {
- append_port_stat(port, ntohs(psr->port_no), ofconn, &msg);
+ handle_queue_stats_for_port(port, queue_id, &cbdata);
}
} else {
- PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
- append_port_stat(port, port_no, ofconn, &msg);
- }
+ ofpbuf_list_delete(&cbdata.replies);
+ return OFPERR_OFPQOFC_BAD_PORT;
}
+ ofconn_send_replies(ofconn, &cbdata.replies);
- queue_tx(msg, ofconn, ofconn->reply_counter);
return 0;
}
-struct flow_stats_cbdata {
- struct ofproto *ofproto;
- struct ofconn *ofconn;
- uint16_t out_port;
- struct ofpbuf *msg;
-};
+static bool
+is_flow_deletion_pending(const struct ofproto *ofproto,
+ const struct cls_rule *cls_rule,
+ uint8_t table_id)
+{
+ if (!hmap_is_empty(&ofproto->deletions)) {
+ struct ofoperation *op;
+
+ HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
+ cls_rule_hash(cls_rule, table_id),
+ &ofproto->deletions) {
+ if (cls_rule_equal(cls_rule, &op->rule->cr)) {
+ return true;
+ }
+ }
+ }
-/* Obtains statistic counters for 'rule' within 'p' and stores them into
- * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
- * returned statistic include statistics for all of 'rule''s subrules. */
-static void
-query_stats(struct ofproto *p, struct rule *rule,
- uint64_t *packet_countp, uint64_t *byte_countp)
+ return false;
+}
+
+/* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
+ * in which no matching flow already exists in the flow table.
+ *
+ * Adds the flow specified by 'ofm', which is followed by 'n_actions'
+ * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
+ * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
+ * initiated now but may be retried later.
+ *
+ * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
+ * if any. */
+static enum ofperr
+add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
+ const struct ofputil_flow_mod *fm, const struct ofp_header *request)
{
- uint64_t packet_count, byte_count;
- struct rule *subrule;
- struct odp_flow *odp_flows;
- size_t n_odp_flows;
-
- /* Start from historical data for 'rule' itself that are no longer tracked
- * by the datapath. This counts, for example, subrules that have
- * expired. */
- packet_count = rule->packet_count;
- byte_count = rule->byte_count;
-
- /* Prepare to ask the datapath for statistics on 'rule', or if it is
- * wildcarded then on all of its subrules.
- *
- * Also, add any statistics that are not tracked by the datapath for each
- * subrule. This includes, for example, statistics for packets that were
- * executed "by hand" by ofproto via dpif_execute() but must be accounted
- * to a flow. */
- n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
- odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
- if (rule->cr.wc.wildcards) {
- size_t i = 0;
- LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
- odp_flows[i++].key = subrule->cr.flow;
- packet_count += subrule->packet_count;
- byte_count += subrule->byte_count;
- }
- } else {
- odp_flows[0].key = rule->cr.flow;
+ struct oftable *table;
+ struct ofopgroup *group;
+ struct rule *victim;
+ struct rule *rule;
+ int error;
+
+ error = check_table_id(ofproto, fm->table_id);
+ if (error) {
+ return error;
}
- /* Fetch up-to-date statistics from the datapath and add them in. */
- if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
- size_t i;
- for (i = 0; i < n_odp_flows; i++) {
- struct odp_flow *odp_flow = &odp_flows[i];
- packet_count += odp_flow->stats.n_packets;
- byte_count += odp_flow->stats.n_bytes;
+ /* Pick table. */
+ if (fm->table_id == 0xff) {
+ uint8_t table_id;
+ if (ofproto->ofproto_class->rule_choose_table) {
+ error = ofproto->ofproto_class->rule_choose_table(ofproto, &fm->cr,
+ &table_id);
+ if (error) {
+ return error;
+ }
+ assert(table_id < ofproto->n_tables);
+ table = &ofproto->tables[table_id];
+ } else {
+ table = &ofproto->tables[0];
}
+ } else if (fm->table_id < ofproto->n_tables) {
+ table = &ofproto->tables[fm->table_id];
+ } else {
+ return OFPERR_NXFMFC_BAD_TABLE_ID;
}
- free(odp_flows);
- /* Return the stats to the caller. */
- *packet_countp = packet_count;
- *byte_countp = byte_count;
-}
-
-static void
-flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
-{
- struct rule *rule = rule_from_cls_rule(rule_);
- struct flow_stats_cbdata *cbdata = cbdata_;
- struct ofp_flow_stats *ofs;
- uint64_t packet_count, byte_count;
- size_t act_len, len;
- long long int tdiff = time_msec() - rule->created;
- uint32_t sec = tdiff / 1000;
- uint32_t msec = tdiff - (sec * 1000);
+ if (table->flags & OFTABLE_READONLY) {
+ return OFPERR_OFPBRC_EPERM;
+ }
- if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
- return;
+ /* Check for overlap, if requested. */
+ if (fm->flags & OFPFF_CHECK_OVERLAP
+ && classifier_rule_overlaps(&table->cls, &fm->cr)) {
+ return OFPERR_OFPFMFC_OVERLAP;
}
- act_len = sizeof *rule->actions * rule->n_actions;
- len = offsetof(struct ofp_flow_stats, actions) + act_len;
+ /* Serialize against pending deletion. */
+ if (is_flow_deletion_pending(ofproto, &fm->cr, table - ofproto->tables)) {
+ return OFPROTO_POSTPONE;
+ }
- query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
+ /* Allocate new rule. */
+ rule = ofproto->ofproto_class->rule_alloc();
+ if (!rule) {
+ VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
+ ofproto->name, strerror(error));
+ return ENOMEM;
+ }
+ rule->ofproto = ofproto;
+ rule->cr = fm->cr;
+ rule->pending = NULL;
+ rule->flow_cookie = fm->cookie;
+ rule->created = rule->modified = rule->used = time_msec();
+ rule->idle_timeout = fm->idle_timeout;
+ rule->hard_timeout = fm->hard_timeout;
+ rule->table_id = table - ofproto->tables;
+ rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
+ rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions);
+ rule->n_actions = fm->n_actions;
+ rule->evictable = true;
+ rule->eviction_group = NULL;
+
+ /* Insert new rule. */
+ victim = oftable_replace_rule(rule);
+ if (victim && !rule_is_modifiable(victim)) {
+ error = OFPERR_OFPBRC_EPERM;
+ } else if (victim && victim->pending) {
+ error = OFPROTO_POSTPONE;
+ } else {
+ struct rule *evict;
+
+ if (classifier_count(&table->cls) > table->max_flows) {
+ bool was_evictable;
+
+ was_evictable = rule->evictable;
+ rule->evictable = false;
+ evict = choose_rule_to_evict(table);
+ rule->evictable = was_evictable;
+
+ if (!evict) {
+ error = OFPERR_OFPFMFC_ALL_TABLES_FULL;
+ goto exit;
+ } else if (evict->pending) {
+ error = OFPROTO_POSTPONE;
+ goto exit;
+ }
+ } else {
+ evict = NULL;
+ }
- ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
- ofs->length = htons(len);
- ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
- ofs->pad = 0;
- flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
- cbdata->ofproto->tun_id_from_cookie, &ofs->match);
- ofs->duration_sec = htonl(sec);
- ofs->duration_nsec = htonl(msec * 1000000);
- ofs->cookie = rule->flow_cookie;
- ofs->priority = htons(rule->cr.priority);
- ofs->idle_timeout = htons(rule->idle_timeout);
- ofs->hard_timeout = htons(rule->hard_timeout);
- memset(ofs->pad2, 0, sizeof ofs->pad2);
- ofs->packet_count = htonll(packet_count);
- ofs->byte_count = htonll(byte_count);
- memcpy(ofs->actions, rule->actions, act_len);
-}
+ group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
+ ofoperation_create(group, rule, OFOPERATION_ADD);
+ rule->pending->victim = victim;
-static int
-table_id_to_include(uint8_t table_id)
-{
- return (table_id == TABLEID_HASH ? CLS_INC_EXACT
- : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
- : table_id == 0xff ? CLS_INC_ALL
- : 0);
-}
+ error = ofproto->ofproto_class->rule_construct(rule);
+ if (error) {
+ ofoperation_destroy(rule->pending);
+ } else if (evict) {
+ delete_flow__(evict, group);
+ }
+ ofopgroup_submit(group);
+ }
-static int
-handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
- const struct ofp_stats_request *osr,
- size_t arg_size)
-{
- struct ofp_flow_stats_request *fsr;
- struct flow_stats_cbdata cbdata;
- struct cls_rule target;
-
- if (arg_size != sizeof *fsr) {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
- }
- fsr = (struct ofp_flow_stats_request *) osr->body;
-
- COVERAGE_INC(ofproto_flows_req);
- cbdata.ofproto = p;
- cbdata.ofconn = ofconn;
- cbdata.out_port = fsr->out_port;
- cbdata.msg = start_stats_reply(osr, 1024);
- cls_rule_from_match(&fsr->match, 0, false, 0, &target);
- classifier_for_each_match(&p->cls, &target,
- table_id_to_include(fsr->table_id),
- flow_stats_cb, &cbdata);
- queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
- return 0;
+exit:
+ /* Back out if an error occurred. */
+ if (error) {
+ oftable_substitute_rule(rule, victim);
+ ofproto_rule_destroy__(rule);
+ }
+ return error;
}
+\f
+/* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
-struct flow_stats_ds_cbdata {
- struct ofproto *ofproto;
- struct ds *results;
-};
-
-static void
-flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
+/* Modifies the rules listed in 'rules', changing their actions to match those
+ * in 'fm'.
+ *
+ * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
+ * if any.
+ *
+ * Returns 0 on success, otherwise an OpenFlow error code. */
+static enum ofperr
+modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
+ const struct ofputil_flow_mod *fm,
+ const struct ofp_header *request, struct list *rules)
{
- struct rule *rule = rule_from_cls_rule(rule_);
- struct flow_stats_ds_cbdata *cbdata = cbdata_;
- struct ds *results = cbdata->results;
- struct ofp_match match;
- uint64_t packet_count, byte_count;
- size_t act_len = sizeof *rule->actions * rule->n_actions;
+ struct ofopgroup *group;
+ struct rule *rule;
+ enum ofperr error;
+
+ group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
+ error = OFPERR_OFPBRC_EPERM;
+ LIST_FOR_EACH (rule, ofproto_node, rules) {
+ if (rule_is_modifiable(rule)) {
+ /* At least one rule is modifiable, don't report EPERM error. */
+ error = 0;
+ } else {
+ continue;
+ }
- /* Don't report on subrules. */
- if (rule->super != NULL) {
- return;
+ if (!ofputil_actions_equal(fm->actions, fm->n_actions,
+ rule->actions, rule->n_actions)) {
+ ofoperation_create(group, rule, OFOPERATION_MODIFY);
+ rule->pending->actions = rule->actions;
+ rule->pending->n_actions = rule->n_actions;
+ rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions);
+ rule->n_actions = fm->n_actions;
+ ofproto->ofproto_class->rule_modify_actions(rule);
+ } else {
+ rule->modified = time_msec();
+ }
+ rule->flow_cookie = fm->cookie;
}
+ ofopgroup_submit(group);
- query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
- flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
- cbdata->ofproto->tun_id_from_cookie, &match);
-
- ds_put_format(results, "duration=%llds, ",
- (time_msec() - rule->created) / 1000);
- ds_put_format(results, "priority=%u, ", rule->cr.priority);
- ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
- ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
- ofp_print_match(results, &match, true);
- ofp_print_actions(results, &rule->actions->header, act_len);
- ds_put_cstr(results, "\n");
+ return error;
}
-/* Adds a pretty-printed description of all flows to 'results', including
- * those marked hidden by secchan (e.g., by in-band control). */
-void
-ofproto_get_all_flows(struct ofproto *p, struct ds *results)
+/* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
+ * failure.
+ *
+ * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
+ * if any. */
+static enum ofperr
+modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
+ const struct ofputil_flow_mod *fm,
+ const struct ofp_header *request)
{
- struct ofp_match match;
- struct cls_rule target;
- struct flow_stats_ds_cbdata cbdata;
+ struct list rules;
+ int error;
- memset(&match, 0, sizeof match);
- match.wildcards = htonl(OVSFW_ALL);
+ error = collect_rules_loose(ofproto, fm->table_id, &fm->cr,
+ fm->cookie, fm->cookie_mask,
+ OFPP_NONE, &rules);
+ return (error ? error
+ : list_is_empty(&rules) ? add_flow(ofproto, ofconn, fm, request)
+ : modify_flows__(ofproto, ofconn, fm, request, &rules));
+}
- cbdata.ofproto = p;
- cbdata.results = results;
+/* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
+ * code on failure.
+ *
+ * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
+ * if any. */
+static enum ofperr
+modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
+ const struct ofputil_flow_mod *fm,
+ const struct ofp_header *request)
+{
+ struct list rules;
+ int error;
- cls_rule_from_match(&match, 0, false, 0, &target);
- classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
- flow_stats_ds_cb, &cbdata);
+ error = collect_rules_strict(ofproto, fm->table_id, &fm->cr,
+ fm->cookie, fm->cookie_mask,
+ OFPP_NONE, &rules);
+ return (error ? error
+ : list_is_empty(&rules) ? add_flow(ofproto, ofconn, fm, request)
+ : list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
+ fm, request, &rules)
+ : 0);
}
-
-struct aggregate_stats_cbdata {
- struct ofproto *ofproto;
- uint16_t out_port;
- uint64_t packet_count;
- uint64_t byte_count;
- uint32_t n_flows;
-};
+\f
+/* OFPFC_DELETE implementation. */
static void
-aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
+delete_flow__(struct rule *rule, struct ofopgroup *group)
{
- struct rule *rule = rule_from_cls_rule(rule_);
- struct aggregate_stats_cbdata *cbdata = cbdata_;
- uint64_t packet_count, byte_count;
-
- if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
- return;
- }
+ struct ofproto *ofproto = rule->ofproto;
- query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
+ ofproto_rule_send_removed(rule, OFPRR_DELETE);
- cbdata->packet_count += packet_count;
- cbdata->byte_count += byte_count;
- cbdata->n_flows++;
-}
-
-static int
-handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
- const struct ofp_stats_request *osr,
- size_t arg_size)
-{
- struct ofp_aggregate_stats_request *asr;
- struct ofp_aggregate_stats_reply *reply;
- struct aggregate_stats_cbdata cbdata;
- struct cls_rule target;
- struct ofpbuf *msg;
-
- if (arg_size != sizeof *asr) {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
- }
- asr = (struct ofp_aggregate_stats_request *) osr->body;
-
- COVERAGE_INC(ofproto_agg_request);
- cbdata.ofproto = p;
- cbdata.out_port = asr->out_port;
- cbdata.packet_count = 0;
- cbdata.byte_count = 0;
- cbdata.n_flows = 0;
- cls_rule_from_match(&asr->match, 0, false, 0, &target);
- classifier_for_each_match(&p->cls, &target,
- table_id_to_include(asr->table_id),
- aggregate_stats_cb, &cbdata);
-
- msg = start_stats_reply(osr, sizeof *reply);
- reply = append_stats_reply(sizeof *reply, ofconn, &msg);
- reply->flow_count = htonl(cbdata.n_flows);
- reply->packet_count = htonll(cbdata.packet_count);
- reply->byte_count = htonll(cbdata.byte_count);
- queue_tx(msg, ofconn, ofconn->reply_counter);
- return 0;
+ ofoperation_create(group, rule, OFOPERATION_DELETE);
+ oftable_remove_rule(rule);
+ ofproto->ofproto_class->rule_destruct(rule);
}
-struct queue_stats_cbdata {
- struct ofconn *ofconn;
- struct ofpbuf *msg;
- uint16_t port_no;
-};
-
-static void
-put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
- const struct netdev_queue_stats *stats)
+/* Deletes the rules listed in 'rules'.
+ *
+ * Returns 0 on success, otherwise an OpenFlow error code. */
+static enum ofperr
+delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
+ const struct ofp_header *request, struct list *rules)
{
- struct ofp_queue_stats *reply;
+ struct rule *rule, *next;
+ struct ofopgroup *group;
- reply = append_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
- reply->port_no = htons(cbdata->port_no);
- memset(reply->pad, 0, sizeof reply->pad);
- reply->queue_id = htonl(queue_id);
- reply->tx_bytes = htonll(stats->tx_bytes);
- reply->tx_packets = htonll(stats->tx_packets);
- reply->tx_errors = htonll(stats->tx_errors);
+ group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
+ LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
+ delete_flow__(rule, group);
+ }
+ ofopgroup_submit(group);
+
+ return 0;
}
-static void
-handle_queue_stats_dump_cb(uint32_t queue_id,
- struct netdev_queue_stats *stats,
- void *cbdata_)
-{
- struct queue_stats_cbdata *cbdata = cbdata_;
+/* Implements OFPFC_DELETE. */
+static enum ofperr
+delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
+ const struct ofputil_flow_mod *fm,
+ const struct ofp_header *request)
+{
+ struct list rules;
+ enum ofperr error;
+
+ error = collect_rules_loose(ofproto, fm->table_id, &fm->cr,
+ fm->cookie, fm->cookie_mask,
+ fm->out_port, &rules);
+ return (error ? error
+ : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
+ &rules)
+ : 0);
+}
- put_queue_stats(cbdata, queue_id, stats);
+/* Implements OFPFC_DELETE_STRICT. */
+static enum ofperr
+delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
+ const struct ofputil_flow_mod *fm,
+ const struct ofp_header *request)
+{
+ struct list rules;
+ enum ofperr error;
+
+ error = collect_rules_strict(ofproto, fm->table_id, &fm->cr,
+ fm->cookie, fm->cookie_mask,
+ fm->out_port, &rules);
+ return (error ? error
+ : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
+ request, &rules)
+ : 0);
}
static void
-handle_queue_stats_for_port(struct ofport *port, uint16_t port_no,
- uint32_t queue_id,
- struct queue_stats_cbdata *cbdata)
+ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
{
- cbdata->port_no = port_no;
- if (queue_id == OFPQ_ALL) {
- netdev_dump_queue_stats(port->netdev,
- handle_queue_stats_dump_cb, cbdata);
- } else {
- struct netdev_queue_stats stats;
+ struct ofputil_flow_removed fr;
- netdev_get_queue_stats(port->netdev, queue_id, &stats);
- put_queue_stats(cbdata, queue_id, &stats);
+ if (rule_is_hidden(rule) || !rule->send_flow_removed) {
+ return;
}
+
+ fr.rule = rule->cr;
+ fr.cookie = rule->flow_cookie;
+ fr.reason = reason;
+ calc_flow_duration__(rule->created, time_msec(),
+ &fr.duration_sec, &fr.duration_nsec);
+ fr.idle_timeout = rule->idle_timeout;
+ rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
+ &fr.byte_count);
+
+ connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
}
-static int
-handle_queue_stats_request(struct ofproto *ofproto, struct ofconn *ofconn,
- const struct ofp_stats_request *osr,
- size_t arg_size)
+void
+ofproto_rule_update_used(struct rule *rule, long long int used)
{
- struct ofp_queue_stats_request *qsr;
- struct queue_stats_cbdata cbdata;
- struct ofport *port;
- unsigned int port_no;
- uint32_t queue_id;
+ if (used > rule->used) {
+ struct eviction_group *evg = rule->eviction_group;
- if (arg_size != sizeof *qsr) {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
+ rule->used = used;
+ if (evg) {
+ heap_change(&evg->rules, &rule->evg_node,
+ rule_eviction_priority(rule));
+ }
}
- qsr = (struct ofp_queue_stats_request *) osr->body;
+}
- COVERAGE_INC(ofproto_queue_req);
+/* Sends an OpenFlow "flow removed" message with the given 'reason' (either
+ * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
+ * ofproto.
+ *
+ * ofproto implementation ->run() functions should use this function to expire
+ * OpenFlow flows. */
+void
+ofproto_rule_expire(struct rule *rule, uint8_t reason)
+{
+ struct ofproto *ofproto = rule->ofproto;
+ struct ofopgroup *group;
- cbdata.ofconn = ofconn;
- cbdata.msg = start_stats_reply(osr, 128);
+ assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
- port_no = ntohs(qsr->port_no);
- queue_id = ntohl(qsr->queue_id);
- if (port_no == OFPP_ALL) {
- PORT_ARRAY_FOR_EACH (port, &ofproto->ports, port_no) {
- handle_queue_stats_for_port(port, port_no, queue_id, &cbdata);
- }
- } else if (port_no < ofproto->max_ports) {
- port = port_array_get(&ofproto->ports, port_no);
- if (port) {
- handle_queue_stats_for_port(port, port_no, queue_id, &cbdata);
- }
- } else {
- ofpbuf_delete(cbdata.msg);
- return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
- }
- queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
+ ofproto_rule_send_removed(rule, reason);
- return 0;
+ group = ofopgroup_create_unattached(ofproto);
+ ofoperation_create(group, rule, OFOPERATION_DELETE);
+ oftable_remove_rule(rule);
+ ofproto->ofproto_class->rule_destruct(rule);
+ ofopgroup_submit(group);
}
-
-static int
-handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
- struct ofp_header *oh)
+\f
+static enum ofperr
+handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
{
- struct ofp_stats_request *osr;
- size_t arg_size;
- int error;
+ struct ofputil_flow_mod fm;
+ enum ofperr error;
+
+ error = reject_slave_controller(ofconn);
+ if (error) {
+ return error;
+ }
- error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
- 1, &arg_size);
+ error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn));
if (error) {
return error;
}
- osr = (struct ofp_stats_request *) oh;
- switch (ntohs(osr->type)) {
- case OFPST_DESC:
- return handle_desc_stats_request(p, ofconn, osr);
+ /* We do not support the emergency flow cache. It will hopefully get
+ * dropped from OpenFlow in the near future. */
+ if (fm.flags & OFPFF_EMERG) {
+ /* There isn't a good fit for an error code, so just state that the
+ * flow table is full. */
+ return OFPERR_OFPFMFC_ALL_TABLES_FULL;
+ }
+
+ return handle_flow_mod__(ofconn_get_ofproto(ofconn), ofconn, &fm, oh);
+}
- case OFPST_FLOW:
- return handle_flow_stats_request(p, ofconn, osr, arg_size);
+static enum ofperr
+handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
+ const struct ofputil_flow_mod *fm,
+ const struct ofp_header *oh)
+{
+ if (ofproto->n_pending >= 50) {
+ assert(!list_is_empty(&ofproto->pending));
+ return OFPROTO_POSTPONE;
+ }
- case OFPST_AGGREGATE:
- return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
+ switch (fm->command) {
+ case OFPFC_ADD:
+ return add_flow(ofproto, ofconn, fm, oh);
- case OFPST_TABLE:
- return handle_table_stats_request(p, ofconn, osr);
+ case OFPFC_MODIFY:
+ return modify_flows_loose(ofproto, ofconn, fm, oh);
- case OFPST_PORT:
- return handle_port_stats_request(p, ofconn, osr, arg_size);
+ case OFPFC_MODIFY_STRICT:
+ return modify_flow_strict(ofproto, ofconn, fm, oh);
- case OFPST_QUEUE:
- return handle_queue_stats_request(p, ofconn, osr, arg_size);
+ case OFPFC_DELETE:
+ return delete_flows_loose(ofproto, ofconn, fm, oh);
- case OFPST_VENDOR:
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
+ case OFPFC_DELETE_STRICT:
+ return delete_flow_strict(ofproto, ofconn, fm, oh);
default:
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
+ if (fm->command > 0xff) {
+ VLOG_WARN_RL(&rl, "flow_mod has explicit table_id but "
+ "flow_mod_table_id extension is not enabled");
+ }
+ return OFPERR_OFPFMFC_BAD_COMMAND;
}
}
-static long long int
-msec_from_nsec(uint64_t sec, uint32_t nsec)
+static enum ofperr
+handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
{
- return !sec ? 0 : sec * 1000 + nsec / 1000000;
-}
+ struct nx_role_request *nrr = (struct nx_role_request *) oh;
+ struct nx_role_request *reply;
+ struct ofpbuf *buf;
+ uint32_t role;
-static void
-update_time(struct ofproto *ofproto, struct rule *rule,
- const struct odp_flow_stats *stats)
-{
- long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
- if (used > rule->used) {
- rule->used = used;
- if (rule->super && used > rule->super->used) {
- rule->super->used = used;
- }
- netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
+ role = ntohl(nrr->role);
+ if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
+ && role != NX_ROLE_SLAVE) {
+ return OFPERR_NXBRC_BAD_ROLE;
}
-}
-static void
-update_stats(struct ofproto *ofproto, struct rule *rule,
- const struct odp_flow_stats *stats)
-{
- if (stats->n_packets) {
- update_time(ofproto, rule, stats);
- rule->packet_count += stats->n_packets;
- rule->byte_count += stats->n_bytes;
- netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
- stats->tcp_flags);
+ if (ofconn_get_role(ofconn) != role
+ && ofconn_has_pending_opgroups(ofconn)) {
+ return OFPROTO_POSTPONE;
}
-}
-/* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
- * in which no matching flow already exists in the flow table.
- *
- * Adds the flow specified by 'ofm', which is followed by 'n_actions'
- * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
- * code as encoded by ofp_mkerr() on failure.
- *
- * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
- * if any. */
-static int
-add_flow(struct ofproto *p, struct ofconn *ofconn,
- const struct ofp_flow_mod *ofm, size_t n_actions)
-{
- struct ofpbuf *packet;
- struct rule *rule;
- uint16_t in_port;
- int error;
+ ofconn_set_role(ofconn, role);
- if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
- flow_t flow;
- uint32_t wildcards;
+ reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
+ reply->role = htonl(role);
+ ofconn_send_reply(ofconn, buf);
- flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
- &flow, &wildcards);
- if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
- ntohs(ofm->priority))) {
- return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
- }
- }
+ return 0;
+}
- rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
- n_actions, ntohs(ofm->idle_timeout),
- ntohs(ofm->hard_timeout), ofm->cookie,
- ofm->flags & htons(OFPFF_SEND_FLOW_REM));
- cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
- p->tun_id_from_cookie, ofm->cookie, &rule->cr);
+static enum ofperr
+handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
+ const struct ofp_header *oh)
+{
+ const struct nx_flow_mod_table_id *msg
+ = (const struct nx_flow_mod_table_id *) oh;
+ enum ofputil_protocol cur, next;
- error = 0;
- if (ofm->buffer_id != htonl(UINT32_MAX)) {
- error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
- &packet, &in_port);
- } else {
- packet = NULL;
- in_port = UINT16_MAX;
- }
+ cur = ofconn_get_protocol(ofconn);
+ next = ofputil_protocol_set_tid(cur, msg->set != 0);
+ ofconn_set_protocol(ofconn, next);
- rule_insert(p, rule, packet, in_port);
- ofpbuf_delete(packet);
- return error;
+ return 0;
}
-static struct rule *
-find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
+static enum ofperr
+handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
{
- uint32_t wildcards;
- flow_t flow;
+ const struct nx_set_flow_format *msg
+ = (const struct nx_set_flow_format *) oh;
+ enum ofputil_protocol cur, next;
+ enum ofputil_protocol next_base;
- flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
- &flow, &wildcards);
- return rule_from_cls_rule(classifier_find_rule_exactly(
- &p->cls, &flow, wildcards,
- ntohs(ofm->priority)));
+ next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
+ if (!next_base) {
+ return OFPERR_OFPBRC_EPERM;
+ }
+
+ cur = ofconn_get_protocol(ofconn);
+ next = ofputil_protocol_set_base(cur, next_base);
+ if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
+ /* Avoid sending async messages in surprising protocol. */
+ return OFPROTO_POSTPONE;
+ }
+
+ ofconn_set_protocol(ofconn, next);
+ return 0;
}
-static int
-send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
- struct rule *rule, const struct ofp_flow_mod *ofm)
+static enum ofperr
+handle_nxt_set_packet_in_format(struct ofconn *ofconn,
+ const struct ofp_header *oh)
{
- struct ofpbuf *packet;
- uint16_t in_port;
- flow_t flow;
- int error;
+ const struct nx_set_packet_in_format *msg;
+ uint32_t format;
- if (ofm->buffer_id == htonl(UINT32_MAX)) {
- return 0;
+ msg = (const struct nx_set_packet_in_format *) oh;
+ format = ntohl(msg->format);
+ if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
+ return OFPERR_OFPBRC_EPERM;
}
- error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
- &packet, &in_port);
- if (error) {
- return error;
+ if (format != ofconn_get_packet_in_format(ofconn)
+ && ofconn_has_pending_opgroups(ofconn)) {
+ /* Avoid sending async message in surprsing packet in format. */
+ return OFPROTO_POSTPONE;
}
- flow_extract(packet, 0, in_port, &flow);
- rule_execute(ofproto, rule, packet, &flow);
- ofpbuf_delete(packet);
-
+ ofconn_set_packet_in_format(ofconn, format);
return 0;
}
-\f
-/* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
-struct modify_flows_cbdata {
- struct ofproto *ofproto;
- const struct ofp_flow_mod *ofm;
- size_t n_actions;
- struct rule *match;
-};
+static enum ofperr
+handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
+{
+ const struct nx_async_config *msg = (const struct nx_async_config *) oh;
+ uint32_t master[OAM_N_TYPES];
+ uint32_t slave[OAM_N_TYPES];
-static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
- size_t n_actions, struct rule *);
-static void modify_flows_cb(struct cls_rule *, void *cbdata_);
+ master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
+ master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
+ master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
-/* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
- * encoded by ofp_mkerr() on failure.
- *
- * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
- * if any. */
-static int
-modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
- const struct ofp_flow_mod *ofm, size_t n_actions)
-{
- struct modify_flows_cbdata cbdata;
- struct cls_rule target;
-
- cbdata.ofproto = p;
- cbdata.ofm = ofm;
- cbdata.n_actions = n_actions;
- cbdata.match = NULL;
-
- cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
- &target);
-
- classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
- modify_flows_cb, &cbdata);
- if (cbdata.match) {
- /* This credits the packet to whichever flow happened to happened to
- * match last. That's weird. Maybe we should do a lookup for the
- * flow that actually matches the packet? Who knows. */
- send_buffered_packet(p, ofconn, cbdata.match, ofm);
- return 0;
- } else {
- return add_flow(p, ofconn, ofm, n_actions);
- }
+ slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
+ slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
+ slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
+
+ ofconn_set_async_config(ofconn, master, slave);
+
+ return 0;
}
-/* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
- * code as encoded by ofp_mkerr() on failure.
- *
- * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
- * if any. */
-static int
-modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
- struct ofp_flow_mod *ofm, size_t n_actions)
+static enum ofperr
+handle_nxt_set_controller_id(struct ofconn *ofconn,
+ const struct ofp_header *oh)
{
- struct rule *rule = find_flow_strict(p, ofm);
- if (rule && !rule_is_hidden(rule)) {
- modify_flow(p, ofm, n_actions, rule);
- return send_buffered_packet(p, ofconn, rule, ofm);
- } else {
- return add_flow(p, ofconn, ofm, n_actions);
+ const struct nx_controller_id *nci;
+
+ nci = (const struct nx_controller_id *) oh;
+ if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
+ return OFPERR_NXBRC_MUST_BE_ZERO;
}
+
+ ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
+ return 0;
}
-/* Callback for modify_flows_loose(). */
-static void
-modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
+static enum ofperr
+handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
{
- struct rule *rule = rule_from_cls_rule(rule_);
- struct modify_flows_cbdata *cbdata = cbdata_;
+ struct ofp_header *ob;
+ struct ofpbuf *buf;
- if (!rule_is_hidden(rule)) {
- cbdata->match = rule;
- modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
+ if (ofconn_has_pending_opgroups(ofconn)) {
+ return OFPROTO_POSTPONE;
}
+
+ ob = make_openflow_xid(sizeof *ob, OFPT10_BARRIER_REPLY, oh->xid, &buf);
+ ofconn_send_reply(ofconn, buf);
+ return 0;
}
-/* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
- * been identified as a flow in 'p''s flow table to be modified, by changing
- * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
- * ofp_action[] structures). */
-static int
-modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
- size_t n_actions, struct rule *rule)
+static enum ofperr
+handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
{
- size_t actions_len = n_actions * sizeof *rule->actions;
-
- rule->flow_cookie = ofm->cookie;
+ const struct ofp_header *oh = msg->data;
+ const struct ofputil_msg_type *type;
+ enum ofperr error;
- /* If the actions are the same, do nothing. */
- if (n_actions == rule->n_actions
- && !memcmp(ofm->actions, rule->actions, actions_len))
- {
- return 0;
+ error = ofputil_decode_msg_type(oh, &type);
+ if (error) {
+ return error;
}
- /* Replace actions. */
- free(rule->actions);
- rule->actions = xmemdup(ofm->actions, actions_len);
- rule->n_actions = n_actions;
+ switch (ofputil_msg_type_code(type)) {
+ /* OpenFlow requests. */
+ case OFPUTIL_OFPT_ECHO_REQUEST:
+ return handle_echo_request(ofconn, oh);
- /* Make sure that the datapath gets updated properly. */
- if (rule->cr.wc.wildcards) {
- COVERAGE_INC(ofproto_mod_wc_flow);
- p->need_revalidate = true;
- } else {
- rule_update_actions(p, rule);
- }
+ case OFPUTIL_OFPT_FEATURES_REQUEST:
+ return handle_features_request(ofconn, oh);
- return 0;
-}
-\f
-/* OFPFC_DELETE implementation. */
+ case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
+ return handle_get_config_request(ofconn, oh);
-struct delete_flows_cbdata {
- struct ofproto *ofproto;
- uint16_t out_port;
-};
+ case OFPUTIL_OFPT_SET_CONFIG:
+ return handle_set_config(ofconn, msg->data);
-static void delete_flows_cb(struct cls_rule *, void *cbdata_);
-static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port);
+ case OFPUTIL_OFPT_PACKET_OUT:
+ return handle_packet_out(ofconn, msg->data);
-/* Implements OFPFC_DELETE. */
-static void
-delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
-{
- struct delete_flows_cbdata cbdata;
- struct cls_rule target;
+ case OFPUTIL_OFPT_PORT_MOD:
+ return handle_port_mod(ofconn, oh);
+
+ case OFPUTIL_OFPT_FLOW_MOD:
+ return handle_flow_mod(ofconn, oh);
+
+ case OFPUTIL_OFPT_BARRIER_REQUEST:
+ return handle_barrier_request(ofconn, oh);
+
+ /* OpenFlow replies. */
+ case OFPUTIL_OFPT_ECHO_REPLY:
+ return 0;
+
+ /* Nicira extension requests. */
+ case OFPUTIL_NXT_ROLE_REQUEST:
+ return handle_role_request(ofconn, oh);
+
+ case OFPUTIL_NXT_FLOW_MOD_TABLE_ID:
+ return handle_nxt_flow_mod_table_id(ofconn, oh);
+
+ case OFPUTIL_NXT_SET_FLOW_FORMAT:
+ return handle_nxt_set_flow_format(ofconn, oh);
- cbdata.ofproto = p;
- cbdata.out_port = ofm->out_port;
+ case OFPUTIL_NXT_SET_PACKET_IN_FORMAT:
+ return handle_nxt_set_packet_in_format(ofconn, oh);
- cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
- &target);
+ case OFPUTIL_NXT_SET_CONTROLLER_ID:
+ return handle_nxt_set_controller_id(ofconn, oh);
- classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
- delete_flows_cb, &cbdata);
+ case OFPUTIL_NXT_FLOW_MOD:
+ return handle_flow_mod(ofconn, oh);
+
+ case OFPUTIL_NXT_FLOW_AGE:
+ /* Nothing to do. */
+ return 0;
+
+ case OFPUTIL_NXT_SET_ASYNC_CONFIG:
+ return handle_nxt_set_async_config(ofconn, oh);
+
+ /* Statistics requests. */
+ case OFPUTIL_OFPST_DESC_REQUEST:
+ return handle_desc_stats_request(ofconn, msg->data);
+
+ case OFPUTIL_OFPST_FLOW_REQUEST:
+ case OFPUTIL_NXST_FLOW_REQUEST:
+ return handle_flow_stats_request(ofconn, msg->data);
+
+ case OFPUTIL_OFPST_AGGREGATE_REQUEST:
+ case OFPUTIL_NXST_AGGREGATE_REQUEST:
+ return handle_aggregate_stats_request(ofconn, msg->data);
+
+ case OFPUTIL_OFPST_TABLE_REQUEST:
+ return handle_table_stats_request(ofconn, msg->data);
+
+ case OFPUTIL_OFPST_PORT_REQUEST:
+ return handle_port_stats_request(ofconn, msg->data);
+
+ case OFPUTIL_OFPST_QUEUE_REQUEST:
+ return handle_queue_stats_request(ofconn, msg->data);
+
+ case OFPUTIL_MSG_INVALID:
+ case OFPUTIL_OFPT_HELLO:
+ case OFPUTIL_OFPT_ERROR:
+ case OFPUTIL_OFPT_FEATURES_REPLY:
+ case OFPUTIL_OFPT_GET_CONFIG_REPLY:
+ case OFPUTIL_OFPT_PACKET_IN:
+ case OFPUTIL_OFPT_FLOW_REMOVED:
+ case OFPUTIL_OFPT_PORT_STATUS:
+ case OFPUTIL_OFPT_BARRIER_REPLY:
+ case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
+ case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
+ case OFPUTIL_OFPST_DESC_REPLY:
+ case OFPUTIL_OFPST_FLOW_REPLY:
+ case OFPUTIL_OFPST_QUEUE_REPLY:
+ case OFPUTIL_OFPST_PORT_REPLY:
+ case OFPUTIL_OFPST_TABLE_REPLY:
+ case OFPUTIL_OFPST_AGGREGATE_REPLY:
+ case OFPUTIL_NXT_ROLE_REPLY:
+ case OFPUTIL_NXT_FLOW_REMOVED:
+ case OFPUTIL_NXT_PACKET_IN:
+ case OFPUTIL_NXST_FLOW_REPLY:
+ case OFPUTIL_NXST_AGGREGATE_REPLY:
+ default:
+ return (oh->type == OFPT10_STATS_REQUEST ||
+ oh->type == OFPT10_STATS_REPLY
+ ? OFPERR_OFPBRC_BAD_STAT
+ : OFPERR_OFPBRC_BAD_TYPE);
+ }
}
-/* Implements OFPFC_DELETE_STRICT. */
-static void
-delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
+static bool
+handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
{
- struct rule *rule = find_flow_strict(p, ofm);
- if (rule) {
- delete_flow(p, rule, ofm->out_port);
+ int error = handle_openflow__(ofconn, ofp_msg);
+ if (error && error != OFPROTO_POSTPONE) {
+ ofconn_send_error(ofconn, ofp_msg->data, error);
}
+ COVERAGE_INC(ofproto_recv_openflow);
+ return error != OFPROTO_POSTPONE;
}
+\f
+/* Asynchronous operations. */
-/* Callback for delete_flows_loose(). */
-static void
-delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
+/* Creates and returns a new ofopgroup that is not associated with any
+ * OpenFlow connection.
+ *
+ * The caller should add operations to the returned group with
+ * ofoperation_create() and then submit it with ofopgroup_submit(). */
+static struct ofopgroup *
+ofopgroup_create_unattached(struct ofproto *ofproto)
+{
+ struct ofopgroup *group = xzalloc(sizeof *group);
+ group->ofproto = ofproto;
+ list_init(&group->ofproto_node);
+ list_init(&group->ops);
+ list_init(&group->ofconn_node);
+ return group;
+}
+
+/* Creates and returns a new ofopgroup for 'ofproto'.
+ *
+ * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
+ * connection. The 'request' and 'buffer_id' arguments are ignored.
+ *
+ * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
+ * If the ofopgroup eventually fails, then the error reply will include
+ * 'request'. If the ofopgroup eventually succeeds, then the packet with
+ * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
+ *
+ * The caller should add operations to the returned group with
+ * ofoperation_create() and then submit it with ofopgroup_submit(). */
+static struct ofopgroup *
+ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
+ const struct ofp_header *request, uint32_t buffer_id)
{
- struct rule *rule = rule_from_cls_rule(rule_);
- struct delete_flows_cbdata *cbdata = cbdata_;
+ struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
+ if (ofconn) {
+ size_t request_len = ntohs(request->length);
+
+ assert(ofconn_get_ofproto(ofconn) == ofproto);
- delete_flow(cbdata->ofproto, rule, cbdata->out_port);
+ ofconn_add_opgroup(ofconn, &group->ofconn_node);
+ group->ofconn = ofconn;
+ group->request = xmemdup(request, MIN(request_len, 64));
+ group->buffer_id = buffer_id;
+ }
+ return group;
}
-/* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
- * been identified as a flow to delete from 'p''s flow table, by deleting the
- * flow and sending out a OFPT_FLOW_REMOVED message to any interested
- * controller.
+/* Submits 'group' for processing.
*
- * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
- * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
- * specified 'out_port'. */
+ * If 'group' contains no operations (e.g. none were ever added, or all of the
+ * ones that were added completed synchronously), then it is destroyed
+ * immediately. Otherwise it is added to the ofproto's list of pending
+ * groups. */
static void
-delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port)
+ofopgroup_submit(struct ofopgroup *group)
{
- if (rule_is_hidden(rule)) {
- return;
+ if (list_is_empty(&group->ops)) {
+ ofopgroup_destroy(group);
+ } else {
+ list_push_back(&group->ofproto->pending, &group->ofproto_node);
+ group->ofproto->n_pending++;
}
+}
- if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
- return;
+static void
+ofopgroup_destroy(struct ofopgroup *group)
+{
+ assert(list_is_empty(&group->ops));
+ if (!list_is_empty(&group->ofproto_node)) {
+ assert(group->ofproto->n_pending > 0);
+ group->ofproto->n_pending--;
+ list_remove(&group->ofproto_node);
+ }
+ if (!list_is_empty(&group->ofconn_node)) {
+ list_remove(&group->ofconn_node);
+ if (group->error) {
+ ofconn_send_error(group->ofconn, group->request, group->error);
+ }
+ connmgr_retry(group->ofproto->connmgr);
}
+ free(group->request);
+ free(group);
+}
+
+/* Initiates a new operation on 'rule', of the specified 'type', within
+ * 'group'. Prior to calling, 'rule' must not have any pending operation. */
+static void
+ofoperation_create(struct ofopgroup *group, struct rule *rule,
+ enum ofoperation_type type)
+{
+ struct ofoperation *op;
+
+ assert(!rule->pending);
- send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
- rule_remove(p, rule);
+ op = rule->pending = xzalloc(sizeof *op);
+ op->group = group;
+ list_push_back(&group->ops, &op->group_node);
+ op->rule = rule;
+ op->type = type;
+ op->status = -1;
+ op->flow_cookie = rule->flow_cookie;
+
+ if (type == OFOPERATION_DELETE) {
+ hmap_insert(&op->group->ofproto->deletions, &op->hmap_node,
+ cls_rule_hash(&rule->cr, rule->table_id));
+ }
}
-\f
-static int
-handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
- struct ofp_flow_mod *ofm)
+
+static void
+ofoperation_destroy(struct ofoperation *op)
{
- struct ofp_match orig_match;
- size_t n_actions;
- int error;
+ struct ofopgroup *group = op->group;
- error = reject_slave_controller(ofconn, &ofm->header);
- if (error) {
- return error;
+ if (op->rule) {
+ op->rule->pending = NULL;
}
- error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
- sizeof *ofm->actions, &n_actions);
- if (error) {
- return error;
+ if (op->type == OFOPERATION_DELETE) {
+ hmap_remove(&group->ofproto->deletions, &op->hmap_node);
}
+ list_remove(&op->group_node);
+ free(op->actions);
+ free(op);
- /* We do not support the emergency flow cache. It will hopefully
- * get dropped from OpenFlow in the near future. */
- if (ofm->flags & htons(OFPFF_EMERG)) {
- /* There isn't a good fit for an error code, so just state that the
- * flow table is full. */
- return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
- }
-
- /* Normalize ofp->match. If normalization actually changes anything, then
- * log the differences. */
- ofm->match.pad1[0] = ofm->match.pad2[0] = 0;
- orig_match = ofm->match;
- normalize_match(&ofm->match);
- if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) {
- static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1);
- if (!VLOG_DROP_INFO(&normal_rl)) {
- char *old = ofp_match_to_literal_string(&orig_match);
- char *new = ofp_match_to_literal_string(&ofm->match);
- VLOG_INFO("%s: normalization changed ofp_match, details:",
- rconn_get_name(ofconn->rconn));
- VLOG_INFO(" pre: %s", old);
- VLOG_INFO("post: %s", new);
- free(old);
- free(new);
- }
+ if (list_is_empty(&group->ops) && !list_is_empty(&group->ofproto_node)) {
+ ofopgroup_destroy(group);
}
+}
- if (!ofm->match.wildcards) {
- ofm->priority = htons(UINT16_MAX);
+/* Indicates that 'op' completed with status 'error', which is either 0 to
+ * indicate success or an OpenFlow error code on failure.
+ *
+ * If 'error' is 0, indicating success, the operation will be committed
+ * permanently to the flow table. There is one interesting subcase:
+ *
+ * - If 'op' is an "add flow" operation that is replacing an existing rule in
+ * the flow table (the "victim" rule) by a new one, then the caller must
+ * have uninitialized any derived state in the victim rule, as in step 5 in
+ * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
+ * performs steps 6 and 7 for the victim rule, most notably by calling its
+ * ->rule_dealloc() function.
+ *
+ * If 'error' is nonzero, then generally the operation will be rolled back:
+ *
+ * - If 'op' is an "add flow" operation, ofproto removes the new rule or
+ * restores the original rule. The caller must have uninitialized any
+ * derived state in the new rule, as in step 5 of in the "Life Cycle" in
+ * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
+ * and 7 for the new rule, calling its ->rule_dealloc() function.
+ *
+ * - If 'op' is a "modify flow" operation, ofproto restores the original
+ * actions.
+ *
+ * - 'op' must not be a "delete flow" operation. Removing a rule is not
+ * allowed to fail. It must always succeed.
+ *
+ * Please see the large comment in ofproto/ofproto-provider.h titled
+ * "Asynchronous Operation Support" for more information. */
+void
+ofoperation_complete(struct ofoperation *op, enum ofperr error)
+{
+ struct ofopgroup *group = op->group;
+ struct rule *rule = op->rule;
+ struct ofproto *ofproto = rule->ofproto;
+
+ assert(rule->pending == op);
+ assert(op->status < 0);
+
+ if (!error
+ && !group->error
+ && op->type != OFOPERATION_DELETE
+ && group->ofconn
+ && group->buffer_id != UINT32_MAX
+ && list_is_singleton(&op->group_node)) {
+ struct ofpbuf *packet;
+ uint16_t in_port;
+
+ error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
+ &packet, &in_port);
+ if (packet) {
+ assert(!error);
+ error = rule_execute(rule, in_port, packet);
+ }
}
-
- error = validate_actions((const union ofp_action *) ofm->actions,
- n_actions, p->max_ports);
- if (error) {
- return error;
+ if (!group->error) {
+ group->error = error;
}
- switch (ntohs(ofm->command)) {
- case OFPFC_ADD:
- return add_flow(p, ofconn, ofm, n_actions);
+ switch (op->type) {
+ case OFOPERATION_ADD:
+ if (!error) {
+ ofproto_rule_destroy__(op->victim);
+ if ((rule->cr.wc.vlan_tci_mask & htons(VLAN_VID_MASK))
+ == htons(VLAN_VID_MASK)) {
+ if (ofproto->vlan_bitmap) {
+ uint16_t vid = vlan_tci_to_vid(rule->cr.flow.vlan_tci);
+
+ if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
+ bitmap_set1(ofproto->vlan_bitmap, vid);
+ ofproto->vlans_changed = true;
+ }
+ } else {
+ ofproto->vlans_changed = true;
+ }
+ }
+ } else {
+ oftable_substitute_rule(rule, op->victim);
+ ofproto_rule_destroy__(rule);
+ }
+ break;
- case OFPFC_MODIFY:
- return modify_flows_loose(p, ofconn, ofm, n_actions);
+ case OFOPERATION_DELETE:
+ assert(!error);
+ ofproto_rule_destroy__(rule);
+ op->rule = NULL;
+ break;
- case OFPFC_MODIFY_STRICT:
- return modify_flow_strict(p, ofconn, ofm, n_actions);
+ case OFOPERATION_MODIFY:
+ if (!error) {
+ rule->modified = time_msec();
+ } else {
+ free(rule->actions);
+ rule->actions = op->actions;
+ rule->n_actions = op->n_actions;
+ op->actions = NULL;
+ }
+ break;
- case OFPFC_DELETE:
- delete_flows_loose(p, ofm);
- return 0;
+ default:
+ NOT_REACHED();
+ }
+ ofoperation_destroy(op);
+}
- case OFPFC_DELETE_STRICT:
- delete_flow_strict(p, ofm);
- return 0;
+struct rule *
+ofoperation_get_victim(struct ofoperation *op)
+{
+ assert(op->type == OFOPERATION_ADD);
+ return op->victim;
+}
+\f
+static uint64_t
+pick_datapath_id(const struct ofproto *ofproto)
+{
+ const struct ofport *port;
+
+ port = ofproto_get_port(ofproto, OFPP_LOCAL);
+ if (port) {
+ uint8_t ea[ETH_ADDR_LEN];
+ int error;
- default:
- return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
+ error = netdev_get_etheraddr(port->netdev, ea);
+ if (!error) {
+ return eth_addr_to_uint64(ea);
+ }
+ VLOG_WARN("could not get MAC address for %s (%s)",
+ netdev_get_name(port->netdev), strerror(error));
}
+ return ofproto->fallback_dpid;
}
-static int
-handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg)
+static uint64_t
+pick_fallback_dpid(void)
{
- int error;
-
- error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
- if (error) {
- return error;
- }
-
- p->tun_id_from_cookie = !!msg->set;
- return 0;
+ uint8_t ea[ETH_ADDR_LEN];
+ eth_addr_nicira_random(ea);
+ return eth_addr_to_uint64(ea);
}
+\f
+/* Table overflow policy. */
-static int
-handle_role_request(struct ofproto *ofproto,
- struct ofconn *ofconn, struct nicira_header *msg)
+/* Chooses and returns a rule to evict from 'table'. Returns NULL if the table
+ * is not configured to evict rules or if the table contains no evictable
+ * rules. (Rules with 'evictable' set to false or with no timeouts are not
+ * evictable.) */
+static struct rule *
+choose_rule_to_evict(struct oftable *table)
{
- struct nx_role_request *nrr;
- struct nx_role_request *reply;
- struct ofpbuf *buf;
- uint32_t role;
-
- if (ntohs(msg->header.length) != sizeof *nrr) {
- VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
- ntohs(msg->header.length), sizeof *nrr);
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
- }
- nrr = (struct nx_role_request *) msg;
-
- if (ofconn->type != OFCONN_CONTROLLER) {
- VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
- "connection");
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
- }
-
- role = ntohl(nrr->role);
- if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
- && role != NX_ROLE_SLAVE) {
- VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
+ struct eviction_group *evg;
- /* There's no good error code for this. */
- return ofp_mkerr(OFPET_BAD_REQUEST, -1);
+ if (!table->eviction_fields) {
+ return NULL;
}
- if (role == NX_ROLE_MASTER) {
- struct ofconn *other;
+ /* In the common case, the outer and inner loops here will each be entered
+ * exactly once:
+ *
+ * - The inner loop normally "return"s in its first iteration. If the
+ * eviction group has any evictable rules, then it always returns in
+ * some iteration.
+ *
+ * - The outer loop only iterates more than once if the largest eviction
+ * group has no evictable rules.
+ *
+ * - The outer loop can exit only if table's 'max_flows' is all filled up
+ * by unevictable rules'. */
+ HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
+ struct rule *rule;
- HMAP_FOR_EACH (other, struct ofconn, hmap_node,
- &ofproto->controllers) {
- if (other->role == NX_ROLE_MASTER) {
- other->role = NX_ROLE_SLAVE;
+ HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
+ if (rule->evictable) {
+ return rule;
}
}
}
- ofconn->role = role;
-
- reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid,
- &buf);
- reply->nxh.vendor = htonl(NX_VENDOR_ID);
- reply->nxh.subtype = htonl(NXT_ROLE_REPLY);
- reply->role = htonl(role);
- queue_tx(buf, ofconn, ofconn->reply_counter);
- return 0;
+ return NULL;
}
-static int
-handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
+/* Searches 'ofproto' for tables that have more flows than their configured
+ * maximum and that have flow eviction enabled, and evicts as many flows as
+ * necessary and currently feasible from them.
+ *
+ * This triggers only when an OpenFlow table has N flows in it and then the
+ * client configures a maximum number of flows less than N. */
+static void
+ofproto_evict(struct ofproto *ofproto)
{
- struct ofp_vendor_header *ovh = msg;
- struct nicira_header *nh;
-
- if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
- VLOG_WARN_RL(&rl, "received vendor message of length %u "
- "(expected at least %zu)",
- ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
- }
- if (ovh->vendor != htonl(NX_VENDOR_ID)) {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
- }
- if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
- VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
- "(expected at least %zu)",
- ntohs(ovh->header.length), sizeof(struct nicira_header));
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
- }
+ struct ofopgroup *group;
+ struct oftable *table;
- nh = msg;
- switch (ntohl(nh->subtype)) {
- case NXT_STATUS_REQUEST:
- return switch_status_handle_request(p->switch_status, ofconn->rconn,
- msg);
+ group = ofopgroup_create_unattached(ofproto);
+ OFPROTO_FOR_EACH_TABLE (table, ofproto) {
+ while (classifier_count(&table->cls) > table->max_flows
+ && table->eviction_fields) {
+ struct rule *rule;
- case NXT_TUN_ID_FROM_COOKIE:
- return handle_tun_id_from_cookie(p, msg);
+ rule = choose_rule_to_evict(table);
+ if (!rule || rule->pending) {
+ break;
+ }
- case NXT_ROLE_REQUEST:
- return handle_role_request(p, ofconn, msg);
+ ofoperation_create(group, rule, OFOPERATION_DELETE);
+ oftable_remove_rule(rule);
+ ofproto->ofproto_class->rule_destruct(rule);
+ }
}
-
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
+ ofopgroup_submit(group);
}
+\f
+/* Eviction groups. */
-static int
-handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
+/* Returns the priority to use for an eviction_group that contains 'n_rules'
+ * rules. The priority contains low-order random bits to ensure that eviction
+ * groups with the same number of rules are prioritized randomly. */
+static uint32_t
+eviction_group_priority(size_t n_rules)
{
- struct ofp_header *ob;
- struct ofpbuf *buf;
-
- /* Currently, everything executes synchronously, so we can just
- * immediately send the barrier reply. */
- ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
- queue_tx(buf, ofconn, ofconn->reply_counter);
- return 0;
+ uint16_t size = MIN(UINT16_MAX, n_rules);
+ return (size << 16) | random_uint16();
}
+/* Updates 'evg', an eviction_group within 'table', following a change that
+ * adds or removes rules in 'evg'. */
static void
-handle_openflow(struct ofconn *ofconn, struct ofproto *p,
- struct ofpbuf *ofp_msg)
+eviction_group_resized(struct oftable *table, struct eviction_group *evg)
{
- struct ofp_header *oh = ofp_msg->data;
- int error;
-
- COVERAGE_INC(ofproto_recv_openflow);
- switch (oh->type) {
- case OFPT_ECHO_REQUEST:
- error = handle_echo_request(ofconn, oh);
- break;
-
- case OFPT_ECHO_REPLY:
- error = 0;
- break;
-
- case OFPT_FEATURES_REQUEST:
- error = handle_features_request(p, ofconn, oh);
- break;
-
- case OFPT_GET_CONFIG_REQUEST:
- error = handle_get_config_request(p, ofconn, oh);
- break;
-
- case OFPT_SET_CONFIG:
- error = handle_set_config(p, ofconn, ofp_msg->data);
- break;
+ heap_change(&table->eviction_groups_by_size, &evg->size_node,
+ eviction_group_priority(heap_count(&evg->rules)));
+}
- case OFPT_PACKET_OUT:
- error = handle_packet_out(p, ofconn, ofp_msg->data);
- break;
+/* Destroys 'evg', an eviction_group within 'table':
+ *
+ * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
+ * rules themselves, just removes them from the eviction group.)
+ *
+ * - Removes 'evg' from 'table'.
+ *
+ * - Frees 'evg'. */
+static void
+eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
+{
+ while (!heap_is_empty(&evg->rules)) {
+ struct rule *rule;
- case OFPT_PORT_MOD:
- error = handle_port_mod(p, ofconn, oh);
- break;
+ rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
+ rule->eviction_group = NULL;
+ }
+ hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
+ heap_remove(&table->eviction_groups_by_size, &evg->size_node);
+ heap_destroy(&evg->rules);
+ free(evg);
+}
- case OFPT_FLOW_MOD:
- error = handle_flow_mod(p, ofconn, ofp_msg->data);
- break;
+/* Removes 'rule' from its eviction group, if any. */
+static void
+eviction_group_remove_rule(struct rule *rule)
+{
+ if (rule->eviction_group) {
+ struct oftable *table = &rule->ofproto->tables[rule->table_id];
+ struct eviction_group *evg = rule->eviction_group;
- case OFPT_STATS_REQUEST:
- error = handle_stats_request(p, ofconn, oh);
- break;
+ rule->eviction_group = NULL;
+ heap_remove(&evg->rules, &rule->evg_node);
+ if (heap_is_empty(&evg->rules)) {
+ eviction_group_destroy(table, evg);
+ } else {
+ eviction_group_resized(table, evg);
+ }
+ }
+}
- case OFPT_VENDOR:
- error = handle_vendor(p, ofconn, ofp_msg->data);
- break;
+/* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
+ * returns the hash value. */
+static uint32_t
+eviction_group_hash_rule(struct rule *rule)
+{
+ struct oftable *table = &rule->ofproto->tables[rule->table_id];
+ const struct mf_subfield *sf;
+ uint32_t hash;
- case OFPT_BARRIER_REQUEST:
- error = handle_barrier_request(ofconn, oh);
- break;
+ hash = table->eviction_group_id_basis;
+ for (sf = table->eviction_fields;
+ sf < &table->eviction_fields[table->n_eviction_fields];
+ sf++)
+ {
+ if (mf_are_prereqs_ok(sf->field, &rule->cr.flow)) {
+ union mf_value value;
- default:
- if (VLOG_IS_WARN_ENABLED()) {
- char *s = ofp_to_string(oh, ntohs(oh->length), 2);
- VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
- free(s);
+ mf_get_value(sf->field, &rule->cr.flow, &value);
+ if (sf->ofs) {
+ bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
+ }
+ if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
+ unsigned int start = sf->ofs + sf->n_bits;
+ bitwise_zero(&value, sf->field->n_bytes, start,
+ sf->field->n_bytes * 8 - start);
+ }
+ hash = hash_bytes(&value, sf->field->n_bytes, hash);
+ } else {
+ hash = hash_int(hash, 0);
}
- error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
- break;
}
- if (error) {
- send_error_oh(ofconn, ofp_msg->data, error);
- }
+ return hash;
}
-\f
-static void
-handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
+
+/* Returns an eviction group within 'table' with the given 'id', creating one
+ * if necessary. */
+static struct eviction_group *
+eviction_group_find(struct oftable *table, uint32_t id)
{
- struct odp_msg *msg = packet->data;
- struct rule *rule;
- struct ofpbuf payload;
- flow_t flow;
+ struct eviction_group *evg;
- payload.data = msg + 1;
- payload.size = msg->length - sizeof *msg;
- flow_extract(&payload, msg->arg, msg->port, &flow);
+ HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
+ return evg;
+ }
- /* Check with in-band control to see if this packet should be sent
- * to the local port regardless of the flow table. */
- if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
- union odp_action action;
+ evg = xmalloc(sizeof *evg);
+ hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
+ heap_insert(&table->eviction_groups_by_size, &evg->size_node,
+ eviction_group_priority(0));
+ heap_init(&evg->rules);
- memset(&action, 0, sizeof(action));
- action.output.type = ODPAT_OUTPUT;
- action.output.port = ODPP_LOCAL;
- dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
- }
+ return evg;
+}
- rule = lookup_valid_rule(p, &flow);
- if (!rule) {
- /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
- struct ofport *port = port_array_get(&p->ports, msg->port);
- if (port) {
- if (port->opp.config & OFPPC_NO_PACKET_IN) {
- COVERAGE_INC(ofproto_no_packet_in);
- /* XXX install 'drop' flow entry */
- ofpbuf_delete(packet);
- return;
- }
- } else {
- VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
- }
+/* Returns an eviction priority for 'rule'. The return value should be
+ * interpreted so that higher priorities make a rule more attractive candidates
+ * for eviction. */
+static uint32_t
+rule_eviction_priority(struct rule *rule)
+{
+ long long int hard_expiration;
+ long long int idle_expiration;
+ long long int expiration;
+ uint32_t expiration_offset;
- COVERAGE_INC(ofproto_packet_in);
- send_packet_in(p, packet);
- return;
+ /* Calculate time of expiration. */
+ hard_expiration = (rule->hard_timeout
+ ? rule->modified + rule->hard_timeout * 1000
+ : LLONG_MAX);
+ idle_expiration = (rule->idle_timeout
+ ? rule->used + rule->idle_timeout * 1000
+ : LLONG_MAX);
+ expiration = MIN(hard_expiration, idle_expiration);
+ if (expiration == LLONG_MAX) {
+ return 0;
}
- if (rule->cr.wc.wildcards) {
- rule = rule_create_subrule(p, rule, &flow);
- rule_make_actions(p, rule, packet);
- } else {
- if (!rule->may_install) {
- /* The rule is not installable, that is, we need to process every
- * packet, so process the current packet and set its actions into
- * 'subrule'. */
- rule_make_actions(p, rule, packet);
- } else {
- /* XXX revalidate rule if it needs it */
- }
- }
+ /* Calculate the time of expiration as a number of (approximate) seconds
+ * after program startup.
+ *
+ * This should work OK for program runs that last UINT32_MAX seconds or
+ * less. Therefore, please restart OVS at least once every 136 years. */
+ expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
- rule_execute(p, rule, &payload, &flow);
- rule_reinstall(p, rule);
-
- if (rule->super && rule->super->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(p, packet);
- } else {
- ofpbuf_delete(packet);
- }
+ /* Invert the expiration offset because we're using a max-heap. */
+ return UINT32_MAX - expiration_offset;
}
+/* Adds 'rule' to an appropriate eviction group for its oftable's
+ * configuration. Does nothing if 'rule''s oftable doesn't have eviction
+ * enabled, or if 'rule' is a permanent rule (one that will never expire on its
+ * own).
+ *
+ * The caller must ensure that 'rule' is not already in an eviction group. */
static void
-handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
+eviction_group_add_rule(struct rule *rule)
{
- struct odp_msg *msg = packet->data;
+ struct ofproto *ofproto = rule->ofproto;
+ struct oftable *table = &ofproto->tables[rule->table_id];
- switch (msg->type) {
- case _ODPL_ACTION_NR:
- COVERAGE_INC(ofproto_ctlr_action);
- send_packet_in(p, packet);
- break;
-
- case _ODPL_SFLOW_NR:
- if (p->sflow) {
- ofproto_sflow_received(p->sflow, msg);
- }
- ofpbuf_delete(packet);
- break;
+ if (table->eviction_fields
+ && (rule->hard_timeout || rule->idle_timeout)) {
+ struct eviction_group *evg;
- case _ODPL_MISS_NR:
- handle_odp_miss_msg(p, packet);
- break;
+ evg = eviction_group_find(table, eviction_group_hash_rule(rule));
- default:
- VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
- msg->type);
- break;
+ rule->eviction_group = evg;
+ heap_insert(&evg->rules, &rule->evg_node,
+ rule_eviction_priority(rule));
+ eviction_group_resized(table, evg);
}
}
\f
+/* oftables. */
+
+/* Initializes 'table'. */
static void
-revalidate_cb(struct cls_rule *sub_, void *cbdata_)
+oftable_init(struct oftable *table)
{
- struct rule *sub = rule_from_cls_rule(sub_);
- struct revalidate_cbdata *cbdata = cbdata_;
-
- if (cbdata->revalidate_all
- || (cbdata->revalidate_subrules && sub->super)
- || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
- revalidate_rule(cbdata->ofproto, sub);
- }
-}
-
-static bool
-revalidate_rule(struct ofproto *p, struct rule *rule)
-{
- const flow_t *flow = &rule->cr.flow;
-
- COVERAGE_INC(ofproto_revalidate_rule);
- if (rule->super) {
- struct rule *super;
- super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
- if (!super) {
- rule_remove(p, rule);
- return false;
- } else if (super != rule->super) {
- COVERAGE_INC(ofproto_revalidate_moved);
- list_remove(&rule->list);
- list_push_back(&super->list, &rule->list);
- rule->super = super;
- rule->hard_timeout = super->hard_timeout;
- rule->idle_timeout = super->idle_timeout;
- rule->created = super->created;
- rule->used = 0;
- }
- }
-
- rule_update_actions(p, rule);
- return true;
+ memset(table, 0, sizeof *table);
+ classifier_init(&table->cls);
}
-static struct ofpbuf *
-compose_flow_removed(struct ofproto *p, const struct rule *rule,
- long long int now, uint8_t reason)
+/* Destroys 'table', including its classifier and eviction groups.
+ *
+ * The caller is responsible for freeing 'table' itself. */
+static void
+oftable_destroy(struct oftable *table)
{
- struct ofp_flow_removed *ofr;
- struct ofpbuf *buf;
- long long int tdiff = now - rule->created;
- uint32_t sec = tdiff / 1000;
- uint32_t msec = tdiff - (sec * 1000);
-
- ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
- flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
- &ofr->match);
- ofr->cookie = rule->flow_cookie;
- ofr->priority = htons(rule->cr.priority);
- ofr->reason = reason;
- ofr->duration_sec = htonl(sec);
- ofr->duration_nsec = htonl(msec * 1000000);
- ofr->idle_timeout = htons(rule->idle_timeout);
- ofr->packet_count = htonll(rule->packet_count);
- ofr->byte_count = htonll(rule->byte_count);
-
- return buf;
+ assert(classifier_is_empty(&table->cls));
+ oftable_disable_eviction(table);
+ classifier_destroy(&table->cls);
+ free(table->name);
}
+/* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
+ * string, then 'table' will use its default name.
+ *
+ * This only affects the name exposed for a table exposed through the OpenFlow
+ * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
static void
-uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
+oftable_set_name(struct oftable *table, const char *name)
{
- assert(rule->installed);
- assert(!rule->cr.wc.wildcards);
-
- if (rule->super) {
- rule_remove(ofproto, rule);
+ if (name && name[0]) {
+ int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
+ if (!table->name || strncmp(name, table->name, len)) {
+ free(table->name);
+ table->name = xmemdup0(name, len);
+ }
} else {
- rule_uninstall(ofproto, rule);
+ free(table->name);
+ table->name = NULL;
}
}
+/* oftables support a choice of two policies when adding a rule would cause the
+ * number of flows in the table to exceed the configured maximum number: either
+ * they can refuse to add the new flow or they can evict some existing flow.
+ * This function configures the former policy on 'table'. */
static void
-send_flow_removed(struct ofproto *p, struct rule *rule,
- long long int now, uint8_t reason)
-{
- struct ofconn *ofconn;
- struct ofconn *prev;
- struct ofpbuf *buf = NULL;
-
- /* We limit the maximum number of queued flow expirations it by accounting
- * them under the counter for replies. That works because preventing
- * OpenFlow requests from being processed also prevents new flows from
- * being added (and expiring). (It also prevents processing OpenFlow
- * requests that would not add new flows, so it is imperfect.) */
-
- prev = NULL;
- LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
- if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)
- && ofconn_receives_async_msgs(ofconn)) {
- if (prev) {
- queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
- } else {
- buf = compose_flow_removed(p, rule, now, reason);
- }
- prev = ofconn;
+oftable_disable_eviction(struct oftable *table)
+{
+ if (table->eviction_fields) {
+ struct eviction_group *evg, *next;
+
+ HMAP_FOR_EACH_SAFE (evg, next, id_node,
+ &table->eviction_groups_by_id) {
+ eviction_group_destroy(table, evg);
}
- }
- if (prev) {
- queue_tx(buf, prev, prev->reply_counter);
+ hmap_destroy(&table->eviction_groups_by_id);
+ heap_destroy(&table->eviction_groups_by_size);
+
+ free(table->eviction_fields);
+ table->eviction_fields = NULL;
+ table->n_eviction_fields = 0;
}
}
-
+/* oftables support a choice of two policies when adding a rule would cause the
+ * number of flows in the table to exceed the configured maximum number: either
+ * they can refuse to add the new flow or they can evict some existing flow.
+ * This function configures the latter policy on 'table', with fairness based
+ * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
+ * 'n_fields' as 0 disables fairness.) */
static void
-expire_rule(struct cls_rule *cls_rule, void *p_)
-{
- struct ofproto *p = p_;
- struct rule *rule = rule_from_cls_rule(cls_rule);
- long long int hard_expire, idle_expire, expire, now;
-
- hard_expire = (rule->hard_timeout
- ? rule->created + rule->hard_timeout * 1000
- : LLONG_MAX);
- idle_expire = (rule->idle_timeout
- && (rule->super || list_is_empty(&rule->list))
- ? rule->used + rule->idle_timeout * 1000
- : LLONG_MAX);
- expire = MIN(hard_expire, idle_expire);
-
- now = time_msec();
- if (now < expire) {
- if (rule->installed && now >= rule->used + 5000) {
- uninstall_idle_flow(p, rule);
- } else if (!rule->cr.wc.wildcards) {
- active_timeout(p, rule);
- }
+oftable_enable_eviction(struct oftable *table,
+ const struct mf_subfield *fields, size_t n_fields)
+{
+ struct cls_cursor cursor;
+ struct rule *rule;
+ if (table->eviction_fields
+ && n_fields == table->n_eviction_fields
+ && (!n_fields
+ || !memcmp(fields, table->eviction_fields,
+ n_fields * sizeof *fields))) {
+ /* No change. */
return;
}
- COVERAGE_INC(ofproto_expired);
+ oftable_disable_eviction(table);
- /* Update stats. This code will be a no-op if the rule expired
- * due to an idle timeout. */
- if (rule->cr.wc.wildcards) {
- struct rule *subrule, *next;
- LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
- rule_remove(p, subrule);
- }
- } else {
- rule_uninstall(p, rule);
- }
+ table->n_eviction_fields = n_fields;
+ table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
- if (!rule_is_hidden(rule)) {
- send_flow_removed(p, rule, now,
- (now >= hard_expire
- ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
+ table->eviction_group_id_basis = random_uint32();
+ hmap_init(&table->eviction_groups_by_id);
+ heap_init(&table->eviction_groups_by_size);
+
+ cls_cursor_init(&cursor, &table->cls, NULL);
+ CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
+ eviction_group_add_rule(rule);
}
- rule_remove(p, rule);
}
+/* Removes 'rule' from the oftable that contains it. */
static void
-active_timeout(struct ofproto *ofproto, struct rule *rule)
-{
- if (ofproto->netflow && !is_controller_rule(rule) &&
- netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
- struct ofexpired expired;
- struct odp_flow odp_flow;
-
- /* Get updated flow stats. */
- memset(&odp_flow, 0, sizeof odp_flow);
- if (rule->installed) {
- odp_flow.key = rule->cr.flow;
- odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
- dpif_flow_get(ofproto->dpif, &odp_flow);
-
- if (odp_flow.stats.n_packets) {
- update_time(ofproto, rule, &odp_flow.stats);
- netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
- odp_flow.stats.tcp_flags);
- }
- }
+oftable_remove_rule(struct rule *rule)
+{
+ struct ofproto *ofproto = rule->ofproto;
+ struct oftable *table = &ofproto->tables[rule->table_id];
- expired.flow = rule->cr.flow;
- expired.packet_count = rule->packet_count +
- odp_flow.stats.n_packets;
- expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
- expired.used = rule->used;
+ classifier_remove(&table->cls, &rule->cr);
+ eviction_group_remove_rule(rule);
+}
- netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
+/* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
+ * oftable that has an identical cls_rule. Returns the rule that was removed,
+ * if any, and otherwise NULL. */
+static struct rule *
+oftable_replace_rule(struct rule *rule)
+{
+ struct ofproto *ofproto = rule->ofproto;
+ struct oftable *table = &ofproto->tables[rule->table_id];
+ struct rule *victim;
- /* Schedule us to send the accumulated records once we have
- * collected all of them. */
- poll_immediate_wake();
+ victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr));
+ if (victim) {
+ eviction_group_remove_rule(victim);
}
+ eviction_group_add_rule(rule);
+ return victim;
}
+/* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
static void
-update_used(struct ofproto *p)
+oftable_substitute_rule(struct rule *old, struct rule *new)
{
- struct odp_flow *flows;
- size_t n_flows;
- size_t i;
- int error;
-
- error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
- if (error) {
- return;
+ if (new) {
+ oftable_replace_rule(new);
+ } else {
+ oftable_remove_rule(old);
}
+}
+\f
+/* unixctl commands. */
- for (i = 0; i < n_flows; i++) {
- struct odp_flow *f = &flows[i];
- struct rule *rule;
+struct ofproto *
+ofproto_lookup(const char *name)
+{
+ struct ofproto *ofproto;
- rule = rule_from_cls_rule(
- classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
- if (!rule || !rule->installed) {
- COVERAGE_INC(ofproto_unexpected_rule);
- dpif_flow_del(p->dpif, f);
- continue;
+ HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
+ &all_ofprotos) {
+ if (!strcmp(ofproto->name, name)) {
+ return ofproto;
}
-
- update_time(p, rule, &f->stats);
- rule_account(p, rule, f->stats.n_bytes);
}
- free(flows);
+ return NULL;
}
-/* pinsched callback for sending 'packet' on 'ofconn'. */
static void
-do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
+ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
+ const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
{
- struct ofconn *ofconn = ofconn_;
+ struct ofproto *ofproto;
+ struct ds results;
- rconn_send_with_limit(ofconn->rconn, packet,
- ofconn->packet_in_counter, 100);
+ ds_init(&results);
+ HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
+ ds_put_format(&results, "%s\n", ofproto->name);
+ }
+ unixctl_command_reply(conn, ds_cstr(&results));
+ ds_destroy(&results);
}
-/* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
- * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
- * packet scheduler for sending.
- *
- * 'max_len' specifies the maximum number of bytes of the packet to send on
- * 'ofconn' (INT_MAX specifies no limit).
- *
- * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
- * ownership is transferred to this function. */
static void
-schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
- bool clone)
-{
- struct ofproto *ofproto = ofconn->ofproto;
- struct ofp_packet_in *opi = packet->data;
- uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
- int send_len, trim_size;
- uint32_t buffer_id;
-
- /* Get buffer. */
- if (opi->reason == OFPR_ACTION) {
- buffer_id = UINT32_MAX;
- } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
- buffer_id = pktbuf_get_null();
- } else if (!ofconn->pktbuf) {
- buffer_id = UINT32_MAX;
- } else {
- struct ofpbuf payload;
- payload.data = opi->data;
- payload.size = packet->size - offsetof(struct ofp_packet_in, data);
- buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
- }
-
- /* Figure out how much of the packet to send. */
- send_len = ntohs(opi->total_len);
- if (buffer_id != UINT32_MAX) {
- send_len = MIN(send_len, ofconn->miss_send_len);
- }
- send_len = MIN(send_len, max_len);
-
- /* Adjust packet length and clone if necessary. */
- trim_size = offsetof(struct ofp_packet_in, data) + send_len;
- if (clone) {
- packet = ofpbuf_clone_data(packet->data, trim_size);
- opi = packet->data;
- } else {
- packet->size = trim_size;
+ofproto_unixctl_init(void)
+{
+ static bool registered;
+ if (registered) {
+ return;
}
+ registered = true;
- /* Update packet headers. */
- opi->buffer_id = htonl(buffer_id);
- update_openflow_length(packet);
-
- /* Hand over to packet scheduler. It might immediately call into
- * do_send_packet_in() or it might buffer it for a while (until a later
- * call to pinsched_run()). */
- pinsched_send(ofconn->schedulers[opi->reason], in_port,
- packet, do_send_packet_in, ofconn);
+ unixctl_command_register("ofproto/list", "", 0, 0,
+ ofproto_unixctl_list, NULL);
}
-
-/* Replace struct odp_msg header in 'packet' by equivalent struct
- * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
- * returned by dpif_recv()).
- *
- * The conversion is not complete: the caller still needs to trim any unneeded
- * payload off the end of the buffer, set the length in the OpenFlow header,
- * and set buffer_id. Those require us to know the controller settings and so
- * must be done on a per-controller basis.
+\f
+/* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
*
- * Returns the maximum number of bytes of the packet that should be sent to
- * the controller (INT_MAX if no limit). */
-static int
-do_convert_to_packet_in(struct ofpbuf *packet)
-{
- struct odp_msg *msg = packet->data;
- struct ofp_packet_in *opi;
- uint8_t reason;
- uint16_t total_len;
- uint16_t in_port;
- int max_len;
-
- /* Extract relevant header fields */
- if (msg->type == _ODPL_ACTION_NR) {
- reason = OFPR_ACTION;
- max_len = msg->arg;
- } else {
- reason = OFPR_NO_MATCH;
- max_len = INT_MAX;
- }
- total_len = msg->length - sizeof *msg;
- in_port = odp_port_to_ofp_port(msg->port);
+ * 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. */
- /* Repurpose packet buffer by overwriting header. */
- ofpbuf_pull(packet, sizeof(struct odp_msg));
- opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
- opi->header.version = OFP_VERSION;
- opi->header.type = OFPT_PACKET_IN;
- opi->total_len = htons(total_len);
- opi->in_port = htons(in_port);
- opi->reason = reason;
+/* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
+ * (exactly) by an OpenFlow rule in 'ofproto'. */
+void
+ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
+{
+ const struct oftable *oftable;
- return max_len;
-}
+ free(ofproto->vlan_bitmap);
+ ofproto->vlan_bitmap = bitmap_allocate(4096);
+ ofproto->vlans_changed = false;
-/* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
- * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
- * as necessary according to their individual configurations.
- *
- * 'packet' must have sufficient headroom to convert it into a struct
- * ofp_packet_in (e.g. as returned by dpif_recv()).
- *
- * Takes ownership of 'packet'. */
-static void
-send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
-{
- struct ofconn *ofconn, *prev;
- int max_len;
+ OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
+ const struct cls_table *table;
- max_len = do_convert_to_packet_in(packet);
+ HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
+ if ((table->wc.vlan_tci_mask & htons(VLAN_VID_MASK))
+ == htons(VLAN_VID_MASK)) {
+ const struct cls_rule *rule;
- prev = NULL;
- LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
- if (ofconn_receives_async_msgs(ofconn)) {
- if (prev) {
- schedule_packet_in(prev, packet, max_len, true);
+ HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
+ uint16_t vid = vlan_tci_to_vid(rule->flow.vlan_tci);
+ bitmap_set1(vlan_bitmap, vid);
+ bitmap_set1(ofproto->vlan_bitmap, vid);
+ }
}
- prev = ofconn;
}
}
- if (prev) {
- schedule_packet_in(prev, packet, max_len, false);
- } else {
- ofpbuf_delete(packet);
- }
}
-static uint64_t
-pick_datapath_id(const struct ofproto *ofproto)
+/* Returns true if new VLANs have come into use by the flow table since the
+ * last call to ofproto_get_vlan_usage().
+ *
+ * We don't track when old VLANs stop being used. */
+bool
+ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
{
- const struct ofport *port;
-
- port = port_array_get(&ofproto->ports, ODPP_LOCAL);
- if (port) {
- uint8_t ea[ETH_ADDR_LEN];
- int error;
-
- error = netdev_get_etheraddr(port->netdev, ea);
- if (!error) {
- return eth_addr_to_uint64(ea);
- }
- VLOG_WARN("could not get MAC address for %s (%s)",
- netdev_get_name(port->netdev), strerror(error));
- }
- return ofproto->fallback_dpid;
+ return ofproto->vlans_changed;
}
-static uint64_t
-pick_fallback_dpid(void)
-{
- uint8_t ea[ETH_ADDR_LEN];
- eth_addr_nicira_random(ea);
- return eth_addr_to_uint64(ea);
-}
-\f
-static bool
-default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
- struct odp_actions *actions, tag_type *tags,
- uint16_t *nf_output_iface, void *ofproto_)
+/* Configures a VLAN splinter binding between the ports identified by OpenFlow
+ * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
+ * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
+ * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
+ * then the VLAN device is un-enslaved. */
+int
+ofproto_port_set_realdev(struct ofproto *ofproto, uint16_t vlandev_ofp_port,
+ uint16_t realdev_ofp_port, int vid)
{
- struct ofproto *ofproto = ofproto_;
- int out_port;
+ struct ofport *ofport;
+ int error;
- /* Drop frames for reserved multicast addresses. */
- if (eth_addr_is_reserved(flow->dl_dst)) {
- return true;
+ assert(vlandev_ofp_port != realdev_ofp_port);
+
+ ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
+ if (!ofport) {
+ VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
+ ofproto->name, vlandev_ofp_port);
+ return EINVAL;
}
- /* Learn source MAC (but don't try to learn from revalidation). */
- if (packet != NULL) {
- tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
- 0, flow->in_port,
- GRAT_ARP_LOCK_NONE);
- if (rev_tag) {
- /* The log messages here could actually be useful in debugging,
- * so keep the rate limit relatively high. */
- static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
- VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
- ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
- ofproto_revalidate(ofproto, rev_tag);
+ if (!ofproto->ofproto_class->set_realdev) {
+ if (!vlandev_ofp_port) {
+ return 0;
}
+ VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
+ return EOPNOTSUPP;
}
- /* Determine output port. */
- out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
- NULL);
- if (out_port < 0) {
- add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
- } else if (out_port != flow->in_port) {
- odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
- *nf_output_iface = out_port;
- } else {
- /* Drop. */
+ error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
+ if (error) {
+ VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
+ ofproto->name, vlandev_ofp_port,
+ netdev_get_name(ofport->netdev), strerror(error));
}
-
- return true;
+ return error;
}
-
-static const struct ofhooks default_ofhooks = {
- NULL,
- default_normal_ofhook_cb,
- NULL,
- NULL
-};