/*
- * Copyright (c) 2009 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");
* you may not use this file except in compliance with the License.
#include "ofproto.h"
#include <errno.h>
#include <inttypes.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 "executer.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 "ofpbuf.h"
+#include "ofproto-provider.h"
#include "openflow/nicira-ext.h"
#include "openflow/openflow.h"
-#include "openflow/openflow-mgmt.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 "svec.h"
-#include "tag.h"
+#include "sset.h"
#include "timeval.h"
+#include "unaligned.h"
#include "unixctl.h"
-#include "vconn.h"
-#include "vconn-ssl.h"
-#include "xtoxll.h"
-
-#define THIS_MODULE VLM_ofproto
#include "vlog.h"
-enum {
- DP_GROUP_FLOOD = 0,
- DP_GROUP_ALL = 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. */
};
-enum {
- TABLEID_HASH = 0,
- TABLEID_CLASSIFIER = 1
+enum ofoperation_type {
+ OFOPERATION_ADD,
+ OFOPERATION_DELETE,
+ OFOPERATION_MODIFY
};
-struct ofport {
- struct netdev *netdev;
- struct ofp_phy_port opp; /* In host byte order. */
+/* 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.
+ *
+ * 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. */
};
-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;
-
- uint16_t idle_timeout; /* In seconds from time of last use. */
- uint16_t hard_timeout; /* In seconds from time of creation. */
- 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;
+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. */
+};
- /* OpenFlow actions.
- *
- * A subrule has no actions (it uses the super-rule's actions). */
- int n_actions;
- union ofp_action *actions;
+static void ofoperation_create(struct ofopgroup *, struct rule *,
+ enum ofoperation_type);
+static void ofoperation_destroy(struct ofoperation *);
- /* 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;
+/* oftable. */
+static void oftable_init(struct oftable *);
+static void oftable_destroy(struct oftable *);
+
+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 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.
+ *
+ * An oftable owns any number of eviction groups, each of which contains any
+ * number of rules.
+ *
+ * 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. */
};
-static inline bool
-rule_is_hidden(const struct rule *rule)
+static struct rule *choose_rule_to_evict(struct oftable *);
+static void ofproto_evict(struct ofproto *);
+static uint32_t rule_eviction_priority(struct rule *);
+
+/* ofport. */
+static void ofport_destroy__(struct ofport *);
+static void ofport_destroy(struct ofport *);
+
+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 *);
+
+/* unixctl. */
+static void ofproto_unixctl_init(void);
+
+/* 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;
+
+/* Map from datapath name to struct ofproto, for use by unixctl commands. */
+static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
+
+static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+
+static void
+ofproto_initialize(void)
{
- /* Subrules are merely an implementation detail, so hide them from the
- * controller. */
- if (rule->super != NULL) {
- return true;
+ static bool inited;
+
+ if (!inited) {
+ inited = true;
+ ofproto_class_register(&ofproto_dpif_class);
}
+}
- /* 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;
+/* '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)
+{
+ size_t i;
+
+ ofproto_initialize();
+ for (i = 0; i < n_ofproto_classes; i++) {
+ const struct ofproto_class *class = ofproto_classes[i];
+ struct sset types;
+ bool found;
+
+ sset_init(&types);
+ class->enumerate_types(&types);
+ found = sset_contains(&types, type);
+ sset_destroy(&types);
+
+ if (found) {
+ return class;
+ }
}
+ VLOG_WARN("unknown datapath type %s", type);
+ return NULL;
+}
- return false;
+/* 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)
+{
+ size_t i;
+
+ for (i = 0; i < n_ofproto_classes; i++) {
+ if (ofproto_classes[i] == new_class) {
+ return EEXIST;
+ }
+ }
+
+ if (n_ofproto_classes >= allocated_ofproto_classes) {
+ ofproto_classes = x2nrealloc(ofproto_classes,
+ &allocated_ofproto_classes,
+ sizeof *ofproto_classes);
+ }
+ ofproto_classes[n_ofproto_classes++] = new_class;
+ return 0;
}
-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);
-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 *);
-
-struct ofconn {
- struct list node;
- struct rconn *rconn;
- struct pktbuf *pktbuf;
- bool send_flow_exp;
- int miss_send_len;
-
- struct rconn_packet_counter *packet_in_counter;
-
- /* Number of OpenFlow messages queued 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;
-};
+/* 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)
+{
+ size_t i;
-static struct ofconn *ofconn_create(struct ofproto *, struct rconn *);
-static void ofconn_destroy(struct ofconn *, struct ofproto *);
-static void ofconn_run(struct ofconn *, struct ofproto *);
-static void ofconn_wait(struct ofconn *);
-static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
- struct rconn_packet_counter *counter);
-
-struct ofproto {
- /* Settings. */
- uint64_t datapath_id; /* Datapath ID. */
- uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
- uint64_t mgmt_id; /* Management channel identifier. */
- char *manufacturer; /* Manufacturer. */
- char *hardware; /* Hardware. */
- char *software; /* Software version. */
- char *serial; /* Serial number. */
-
- /* 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 status_category *ss_cat;
- struct in_band *in_band;
- struct discovery *discovery;
- struct fail_open *fail_open;
- struct pinsched *miss_sched, *action_sched;
- struct executer *executer;
- struct netflow *netflow;
-
- /* Flow table. */
- struct classifier cls;
- bool need_revalidate;
- long long int next_expiration;
- struct tag_set revalidate_set;
-
- /* OpenFlow connections. */
- struct list all_conns;
- struct ofconn *controller;
- 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;
-};
+ 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];
+ }
+ n_ofproto_classes--;
+ return 0;
+ }
+ }
+ VLOG_WARN("attempted to unregister an ofproto class that is not "
+ "registered");
+ return EAFNOSUPPORT;
+}
-static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+/* Clears 'types' and enumerates all registered ofproto types into it. The
+ * caller must first initialize the sset. */
+void
+ofproto_enumerate_types(struct sset *types)
+{
+ size_t i;
-static const struct ofhooks default_ofhooks;
+ ofproto_initialize();
+ for (i = 0; i < n_ofproto_classes; i++) {
+ ofproto_classes[i]->enumerate_types(types);
+ }
+}
-static uint64_t pick_datapath_id(const struct ofproto *);
-static uint64_t pick_fallback_dpid(void);
-static void send_packet_in_miss(struct ofpbuf *, void *ofproto);
-static void send_packet_in_action(struct ofpbuf *, void *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_);
-
-static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
-
-static void handle_openflow(struct ofconn *, struct ofproto *,
- struct ofpbuf *);
-
-static void refresh_port_group(struct ofproto *, unsigned int group);
-static void update_port(struct ofproto *, const char *devname);
-static int init_ports(struct ofproto *);
-static void reinit_ports(struct 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)
+{
+ return type && type[0] ? type : "system";
+}
+
+/* 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;
+ }
int
-ofproto_create(const char *datapath, const struct ofhooks *ofhooks, void *aux,
+ofproto_create(const char *datapath_name, const char *datapath_type,
struct ofproto **ofprotop)
{
- struct odp_stats stats;
- struct ofproto *p;
- struct dpif *dpif;
+ const struct ofproto_class *class;
+ struct ofproto *ofproto;
int error;
*ofprotop = NULL;
- /* Connect to datapath and start listening for messages. */
- error = dpif_open(datapath, &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);
+ 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 listen on datapath %s: %s",
- datapath, strerror(error));
- dpif_close(dpif);
+ VLOG_ERR("failed to open datapath %s: %s",
+ datapath_name, strerror(error));
+ ofproto_destroy__(ofproto);
return error;
}
- dpif_flow_flush(dpif);
- dpif_recv_purge(dpif);
-
- /* Initialize settings. */
- p = xcalloc(1, sizeof *p);
- p->fallback_dpid = pick_fallback_dpid();
- p->datapath_id = p->fallback_dpid;
- p->manufacturer = xstrdup("Nicira Networks, Inc.");
- p->hardware = xstrdup("Reference Implementation");
- p->software = xstrdup(VERSION BUILDNR);
- p->serial = xstrdup("None");
-
- /* 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->discovery = NULL;
- p->fail_open = NULL;
- p->miss_sched = p->action_sched = NULL;
- p->executer = NULL;
- p->netflow = 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);
- p->controller = ofconn_create(p, rconn_create(5, 8));
- p->controller->pktbuf = pktbuf_create();
- p->controller->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
- 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();
- }
- /* Register switch status category. */
- p->ss_cat = switch_status_register(p->switch_status, "remote",
- rconn_status_cb, p->controller->rconn);
-
- /* Almost done... */
- error = init_ports(p);
- if (error) {
- ofproto_destroy(p);
- return error;
- }
+ assert(ofproto->n_tables);
- /* Pick final datapath ID. */
- p->datapath_id = pick_datapath_id(p);
- VLOG_INFO("using datapath ID %012"PRIx64, p->datapath_id);
+ ofproto->datapath_id = pick_datapath_id(ofproto);
+ VLOG_INFO("using datapath ID %016"PRIx64, ofproto->datapath_id);
+ init_ports(ofproto);
- *ofprotop = p;
+ *ofprotop = ofproto;
return 0;
}
+void
+ofproto_init_tables(struct ofproto *ofproto, int n_tables)
+{
+ struct oftable *table;
+
+ assert(!ofproto->n_tables);
+ assert(n_tables >= 1 && n_tables <= 255);
+
+ ofproto->n_tables = n_tables;
+ ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
+ OFPROTO_FOR_EACH_TABLE (table, ofproto) {
+ oftable_init(table);
+ }
+}
+
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 %012"PRIx64, p->datapath_id);
- rconn_reconnect(p->controller->rconn);
+ VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
+
+ /* 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);
}
}
void
-ofproto_set_mgmt_id(struct ofproto *p, uint64_t mgmt_id)
+ofproto_set_controllers(struct ofproto *p,
+ const struct ofproto_controller *controllers,
+ size_t n_controllers)
{
- p->mgmt_id = mgmt_id;
+ connmgr_set_controllers(p->connmgr, controllers, n_controllers);
}
void
-ofproto_set_probe_interval(struct ofproto *p, int probe_interval)
+ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
{
- probe_interval = probe_interval ? MAX(probe_interval, 5) : 0;
- rconn_set_probe_interval(p->controller->rconn, probe_interval);
- if (p->fail_open) {
- int trigger_duration = probe_interval ? probe_interval * 3 : 15;
- fail_open_set_trigger_duration(p->fail_open, trigger_duration);
- }
+ connmgr_set_fail_mode(p->connmgr, fail_mode);
}
+/* Drops the connections between 'ofproto' and all of its controllers, forcing
+ * them to reconnect. */
void
-ofproto_set_max_backoff(struct ofproto *p, int max_backoff)
+ofproto_reconnect_controllers(struct ofproto *ofproto)
{
- rconn_set_max_backoff(p->controller->rconn, max_backoff);
+ connmgr_reconnect(ofproto->connmgr);
}
+/* 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_desc(struct ofproto *p,
- const char *manufacturer, const char *hardware,
- const char *software, const char *serial)
+ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
+ const struct sockaddr_in *extras, size_t n)
{
- if (manufacturer) {
- free(p->manufacturer);
- p->manufacturer = xstrdup(manufacturer);
- }
- if (hardware) {
- free(p->hardware);
- p->hardware = xstrdup(hardware);
- }
- if (software) {
- free(p->software);
- p->software = xstrdup(software);
- }
- if (serial) {
- free(p->serial);
- p->serial = xstrdup(serial);
- }
+ connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
}
-int
-ofproto_set_in_band(struct ofproto *p, bool in_band)
+/* 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)
{
- if (in_band != (p->in_band != NULL)) {
- if (in_band) {
- return in_band_create(p, p->dpif, p->switch_status,
- p->controller->rconn, &p->in_band);
- } else {
- ofproto_set_discovery(p, false, NULL, true);
- in_band_destroy(p->in_band);
- p->in_band = NULL;
- }
- rconn_reconnect(p->controller->rconn);
- }
- return 0;
+ connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
}
-int
-ofproto_set_discovery(struct ofproto *p, bool discovery,
- const char *re, bool update_resolv_conf)
+/* 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 (discovery != (p->discovery != NULL)) {
- if (discovery) {
- int error = ofproto_set_in_band(p, true);
- if (error) {
- return error;
- }
- error = discovery_create(re, update_resolv_conf,
- p->dpif, p->switch_status,
- &p->discovery);
- if (error) {
- return error;
- }
- } else {
- discovery_destroy(p->discovery);
- p->discovery = NULL;
- }
- rconn_disconnect(p->controller->rconn);
- } else if (discovery) {
- discovery_set_update_resolv_conf(p->discovery, update_resolv_conf);
- return discovery_set_accept_controller_re(p->discovery, re);
+ if (threshold < OFPROTO_FLOW_EVICTION_THRESHOLD_MIN) {
+ ofproto->flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_MIN;
+ } else {
+ ofproto->flow_eviction_threshold = threshold;
}
- return 0;
}
-int
-ofproto_set_controller(struct ofproto *ofproto, const char *controller)
+/* 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)
{
- if (ofproto->discovery) {
- return EINVAL;
- } else if (controller) {
- if (strcmp(rconn_get_name(ofproto->controller->rconn), controller)) {
- return rconn_connect(ofproto->controller->rconn, controller);
- } else {
- return 0;
+ 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);
}
- } else {
- rconn_disconnect(ofproto->controller->rconn);
- return 0;
}
}
-static int
-set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
- const struct svec *svec)
+/* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
+ * 'idle_time', in seconds. */
+void
+ofproto_set_mac_idle_time(struct ofproto *ofproto, unsigned idle_time)
{
- 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]);
+ if (ofproto->ofproto_class->set_mac_idle_time) {
+ ofproto->ofproto_class->set_mac_idle_time(ofproto, idle_time);
}
- 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;
+void
+ofproto_set_desc(struct ofproto *p,
+ const char *mfr_desc, const char *hw_desc,
+ const char *sw_desc, const char *serial_desc,
+ const char *dp_desc)
+{
+ struct ofp_desc_stats *ods;
- 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;
- }
+ if (mfr_desc) {
+ if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
+ VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
+ sizeof ods->mfr_desc);
}
+ free(p->mfr_desc);
+ p->mfr_desc = xstrdup(mfr_desc);
+ }
+ if (hw_desc) {
+ if (strlen(hw_desc) >= sizeof ods->hw_desc) {
+ VLOG_WARN("truncating hw_desc, must be less than %zu characters",
+ sizeof ods->hw_desc);
+ }
+ free(p->hw_desc);
+ p->hw_desc = xstrdup(hw_desc);
+ }
+ if (sw_desc) {
+ if (strlen(sw_desc) >= sizeof ods->sw_desc) {
+ VLOG_WARN("truncating sw_desc, must be less than %zu characters",
+ sizeof ods->sw_desc);
+ }
+ free(p->sw_desc);
+ p->sw_desc = xstrdup(sw_desc);
+ }
+ if (serial_desc) {
+ if (strlen(serial_desc) >= sizeof ods->serial_num) {
+ VLOG_WARN("truncating serial_desc, must be less than %zu "
+ "characters",
+ sizeof ods->serial_num);
+ }
+ free(p->serial_desc);
+ p->serial_desc = xstrdup(serial_desc);
+ }
+ if (dp_desc) {
+ if (strlen(dp_desc) >= sizeof ods->dp_desc) {
+ VLOG_WARN("truncating dp_desc, must be less than %zu characters",
+ sizeof ods->dp_desc);
+ }
+ free(p->dp_desc);
+ p->dp_desc = xstrdup(dp_desc);
}
-
- *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->collectors.n) {
- if (!ofproto->netflow) {
- ofproto->netflow = netflow_create();
- }
- return netflow_set_options(ofproto->netflow, nf_options);
- } else {
- netflow_destroy(ofproto->netflow);
- ofproto->netflow = NULL;
- return 0;
- }
-}
-
-void
-ofproto_set_failure(struct ofproto *ofproto, bool fail_open)
-{
- if (fail_open) {
- struct rconn *rconn = ofproto->controller->rconn;
- int trigger_duration = rconn_get_probe_interval(rconn) * 3;
- if (!ofproto->fail_open) {
- ofproto->fail_open = fail_open_create(ofproto, trigger_duration,
- ofproto->switch_status,
- rconn);
- } else {
- fail_open_set_trigger_duration(ofproto->fail_open,
- trigger_duration);
- }
- } else {
- fail_open_destroy(ofproto->fail_open);
- ofproto->fail_open = NULL;
+ if (nf_options && sset_is_empty(&nf_options->collectors)) {
+ nf_options = NULL;
}
-}
-void
-ofproto_set_rate_limit(struct ofproto *ofproto,
- int rate_limit, int burst_limit)
-{
- if (rate_limit > 0) {
- if (!ofproto->miss_sched) {
- ofproto->miss_sched = pinsched_create(rate_limit, burst_limit,
- ofproto->switch_status);
- ofproto->action_sched = pinsched_create(rate_limit, burst_limit,
- NULL);
- } else {
- pinsched_set_limits(ofproto->miss_sched, rate_limit, burst_limit);
- pinsched_set_limits(ofproto->action_sched,
- rate_limit, burst_limit);
- }
+ if (ofproto->ofproto_class->set_netflow) {
+ return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
} else {
- pinsched_destroy(ofproto->miss_sched);
- ofproto->miss_sched = NULL;
- pinsched_destroy(ofproto->action_sched);
- ofproto->action_sched = NULL;
+ return nf_options ? EOPNOTSUPP : 0;
}
}
int
-ofproto_set_stp(struct ofproto *ofproto UNUSED, bool enable_stp)
+ofproto_set_sflow(struct ofproto *ofproto,
+ const struct ofproto_sflow_options *oso)
{
- /* XXX */
- if (enable_stp) {
- VLOG_WARN("STP is not yet implemented");
- return EINVAL;
- } else {
- return 0;
+ if (oso && sset_is_empty(&oso->targets)) {
+ oso = NULL;
}
-}
-int
-ofproto_set_remote_execution(struct ofproto *ofproto, const char *command_acl,
- const char *command_dir)
-{
- if (command_acl) {
- if (!ofproto->executer) {
- return executer_create(command_acl, command_dir,
- &ofproto->executer);
- } else {
- executer_set_acl(ofproto->executer, command_acl, command_dir);
- }
+ if (ofproto->ofproto_class->set_sflow) {
+ return ofproto->ofproto_class->set_sflow(ofproto, oso);
} else {
- executer_destroy(ofproto->executer);
- ofproto->executer = NULL;
+ return oso ? EOPNOTSUPP : 0;
}
- return 0;
-}
-
-uint64_t
-ofproto_get_datapath_id(const struct ofproto *ofproto)
-{
- return ofproto->datapath_id;
}
+\f
+/* Spanning Tree Protocol (STP) configuration. */
-uint64_t
-ofproto_get_mgmt_id(const struct ofproto *ofproto)
+/* 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,
+ const struct ofproto_stp_settings *s)
{
- return ofproto->mgmt_id;
+ return (ofproto->ofproto_class->set_stp
+ ? ofproto->ofproto_class->set_stp(ofproto, s)
+ : EOPNOTSUPP);
}
+/* 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_probe_interval(const struct ofproto *ofproto)
+ofproto_get_stp_status(struct ofproto *ofproto,
+ struct ofproto_stp_status *s)
{
- return rconn_get_probe_interval(ofproto->controller->rconn);
+ return (ofproto->ofproto_class->get_stp_status
+ ? ofproto->ofproto_class->get_stp_status(ofproto, s)
+ : EOPNOTSUPP);
}
+/* 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_get_max_backoff(const struct ofproto *ofproto)
+ofproto_port_set_stp(struct ofproto *ofproto, uint16_t ofp_port,
+ const struct ofproto_port_stp_settings *s)
{
- return rconn_get_max_backoff(ofproto->controller->rconn);
-}
+ 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;
+ }
-bool
-ofproto_get_in_band(const struct ofproto *ofproto)
-{
- return ofproto->in_band != NULL;
+ return (ofproto->ofproto_class->set_stp_port
+ ? ofproto->ofproto_class->set_stp_port(ofport, s)
+ : EOPNOTSUPP);
}
-bool
-ofproto_get_discovery(const struct ofproto *ofproto)
+/* 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)
{
- return ofproto->discovery != NULL;
-}
+ 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;
+ }
-const char *
-ofproto_get_controller(const struct ofproto *ofproto)
-{
- return rconn_get_name(ofproto->controller->rconn);
+ return (ofproto->ofproto_class->get_stp_port_status
+ ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
+ : EOPNOTSUPP);
}
+\f
+/* Queue DSCP configuration. */
-void
-ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
+/* 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)
{
- size_t i;
+ struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
- for (i = 0; i < ofproto->n_listeners; i++) {
- svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
+ 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;
}
- ofproto_flush_flows(p);
- classifier_destroy(&p->cls);
-
- LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
- &p->all_conns) {
- ofconn_destroy(ofconn, p);
- }
-
- dpif_close(p->dpif);
- netdev_monitor_destroy(p->netdev_monitor);
- PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
- ofport_free(ofport);
- }
- shash_destroy(&p->port_by_name);
-
- switch_status_destroy(p->switch_status);
- in_band_destroy(p->in_band);
- discovery_destroy(p->discovery);
- fail_open_destroy(p->fail_open);
- pinsched_destroy(p->miss_sched);
- pinsched_destroy(p->action_sched);
- executer_destroy(p->executer);
- netflow_destroy(p->netflow);
-
- switch_status_unregister(p->ss_cat);
-
- for (i = 0; i < p->n_listeners; i++) {
- pvconn_close(p->listeners[i]);
+ /* 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));
}
- free(p->listeners);
+}
- for (i = 0; i < p->n_snoops; i++) {
- pvconn_close(p->snoops[i]);
- }
- free(p->snoops);
+/* 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. */
- mac_learning_destroy(p->ml);
+/* 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);
+}
- free(p);
+/* 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);
}
+\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_run(struct ofproto *p)
+ofproto_mirror_register(struct ofproto *ofproto, void *aux,
+ const struct ofproto_mirror_settings *s)
{
- int error = ofproto_run1(p);
- if (!error) {
- error = ofproto_run2(p, false);
- }
- return error;
+ return (ofproto->ofproto_class->mirror_set
+ ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
+ : EOPNOTSUPP);
}
-static void
-process_port_change(struct ofproto *ofproto, int error, char *devname)
+/* 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)
{
- if (error == ENOBUFS) {
- reinit_ports(ofproto);
- } else if (!error) {
- update_port(ofproto, devname);
- free(devname);
- }
+ return ofproto_mirror_register(ofproto, aux, NULL);
}
+/* 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_run1(struct ofproto *p)
+ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
+ uint64_t *packets, uint64_t *bytes)
{
- struct ofconn *ofconn, *next_ofconn;
- char *devname;
- int error;
- int i;
+ if (!ofproto->ofproto_class->mirror_get_stats) {
+ *packets = *bytes = UINT64_MAX;
+ return EOPNOTSUPP;
+ }
- for (i = 0; i < 50; i++) {
- struct ofpbuf *buf;
- int error;
+ return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
+ packets, bytes);
+}
- 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 rl = VLOG_RATE_LIMIT_INIT(1, 5);
- VLOG_ERR_RL(&rl, "%s: datapath was destroyed externally",
- dpif_name(p->dpif));
- return ENODEV;
- }
- break;
- }
+/* 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_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
+{
+ return (ofproto->ofproto_class->set_flood_vlans
+ ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
+ : EOPNOTSUPP);
+}
- handle_odp_msg(p, buf);
- }
+/* 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)
+{
+ return (ofproto->ofproto_class->is_mirror_output_bundle
+ ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
+ : false);
+}
+\f
+/* Configuration of OpenFlow tables. */
- while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
- process_port_change(p, error, devname);
+/* Returns the number of OpenFlow tables in 'ofproto'. */
+int
+ofproto_get_n_tables(const struct ofproto *ofproto)
+{
+ return ofproto->n_tables;
+}
+
+/* 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 oftable *table;
+
+ 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;
}
- while ((error = netdev_monitor_poll(p->netdev_monitor,
- &devname)) != EAGAIN) {
- process_port_change(p, error, devname);
+
+ if (s->groups) {
+ oftable_enable_eviction(table, s->groups, s->n_groups);
+ } else {
+ 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);
+}
+
+void
+ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
+{
+ connmgr_get_snoops(ofproto->connmgr, snoops);
+}
+
+static void
+ofproto_flush__(struct ofproto *ofproto)
+{
+ struct ofopgroup *group;
+ struct oftable *table;
- if (p->in_band) {
- in_band_run(p->in_band);
+ if (ofproto->ofproto_class->flush) {
+ ofproto->ofproto_class->flush(ofproto);
}
- if (p->discovery) {
- char *controller_name;
- if (rconn_is_connectivity_questionable(p->controller->rconn)) {
- discovery_question_connectivity(p->discovery);
+
+ group = ofopgroup_create_unattached(ofproto);
+ OFPROTO_FOR_EACH_TABLE (table, ofproto) {
+ struct rule *rule, *next_rule;
+ struct cls_cursor cursor;
+
+ if (table->flags & OFTABLE_HIDDEN) {
+ continue;
}
- if (discovery_run(p->discovery, &controller_name)) {
- if (controller_name) {
- rconn_connect(p->controller->rconn, controller_name);
- } else {
- rconn_disconnect(p->controller->rconn);
+
+ 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);
}
}
}
- pinsched_run(p->miss_sched, send_packet_in_miss, p);
- pinsched_run(p->action_sched, send_packet_in_action, p);
- if (p->executer) {
- executer_run(p->executer);
+ ofopgroup_submit(group);
+}
+
+static void
+ofproto_destroy__(struct ofproto *ofproto)
+{
+ struct oftable *table;
+
+ 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;
+
+ if (!p) {
+ return;
}
- LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
- &p->all_conns) {
- ofconn_run(ofconn, p);
+ ofproto_flush__(p);
+ HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
+ ofport_destroy(ofport);
}
- /* 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);
+ 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);
}
+}
- for (i = 0; i < p->n_listeners; i++) {
- struct vconn *vconn;
- int retval;
+int
+ofproto_run(struct ofproto *p)
+{
+ struct ofport *ofport;
+ char *devname;
+ int error;
- retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
- if (!retval) {
- ofconn_create(p, rconn_new_from_vconn("passive", vconn));
- } else if (retval != EAGAIN) {
- VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
- }
+ error = p->ofproto_class->run(p);
+ if (error && error != EAGAIN) {
+ VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, strerror(error));
}
- for (i = 0; i < p->n_snoops; i++) {
- struct vconn *vconn;
- int retval;
+ if (p->ofproto_class->port_poll) {
+ while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
+ process_port_change(p, error, devname);
+ }
+ }
- retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
- if (!retval) {
- rconn_add_monitor(p->controller->rconn, 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);
+ 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) {
- 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->discovery) {
- discovery_wait(p->discovery);
- }
- if (p->fail_open) {
- fail_open_wait(p->fail_open);
- }
- pinsched_wait(p->miss_sched);
- pinsched_wait(p->action_sched);
- if (p->executer) {
- executer_wait(p->executer);
- }
- 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(p->next_expiration - time_msec());
- }
- 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_get_ofproto_controller_info(const struct ofproto *ofproto,
+ struct shash *info)
+{
+ connmgr_get_controller_info(ofproto->connmgr, info);
+}
+
void
-ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
+ofproto_free_ofproto_controller_info(struct shash *info)
{
- tag_set_add(&ofproto->revalidate_set, tag);
+ connmgr_free_controller_info(info);
}
-struct tag_set *
-ofproto_get_revalidate_set(struct ofproto *ofproto)
+/* 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 p->discovery || rconn_is_alive(p->controller->rconn);
+ 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);
- cls_rule_from_flow(&rule->cr, flow, wildcards, priority);
- 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)
+{
+ 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 = rule_from_cls_rule(rule_);
- struct ofproto *ofproto = ofproto_;
+ 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 void
-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);
-}
-static void
-refresh_port_groups(struct ofproto *p)
-{
- refresh_port_group(p, DP_GROUP_FLOOD);
- refresh_port_group(p, DP_GROUP_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)
{
enum netdev_flags flags;
- struct ofport *ofport;
struct netdev *netdev;
- bool carrier;
int error;
- error = netdev_open(odp_port->devname, NETDEV_ETH_TYPE_NONE, &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;
+ const char *netdev_name = netdev_get_name(netdev);
+ struct ofport *ofport;
+ int dev_mtu;
+ int error;
+
+ /* Create ofport. */
+ ofport = p->ofproto_class->port_alloc();
+ if (!ofport) {
+ error = ENOMEM;
+ goto error;
+ }
+ 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;
+ }
- 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);
+ /* Let the ofproto_class initialize its private data. */
+ error = p->ofproto_class->port_construct(ofport);
+ if (error) {
+ goto error;
}
- if (p->ofhooks->port_changed_cb) {
- p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
+ connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
+ return;
+
+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_install(struct ofproto *p, struct ofport *ofport)
+ofport_remove(struct ofport *ofport)
{
- netdev_monitor_add(p->netdev_monitor, ofport->netdev);
- port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no),
- ofport);
- shash_add(&p->port_by_name, (char *) ofport->opp.name, ofport);
+ 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_remove(struct ofproto *p, struct ofport *ofport)
+ofport_remove_with_name(struct ofproto *ofproto, const char *name)
{
- netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
- port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no), NULL);
- shash_delete(&p->port_by_name,
- shash_find(&p->port_by_name, (char *) ofport->opp.name));
+ 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
-ofport_free(struct ofport *ofport)
+ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
{
- if (ofport) {
- netdev_close(ofport->netdev);
- free(ofport);
+ 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;
+
+ connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
+}
+
+/* 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);
}
}
-static void
-update_port(struct ofproto *p, const char *devname)
+void
+ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
{
- struct odp_port odp_port;
- struct ofport *old_ofport;
- struct ofport *new_ofport;
- int error;
+ 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);
+ }
+ }
+}
- COVERAGE_INC(ofproto_update_port);
+static void
+ofport_destroy__(struct ofport *port)
+{
+ struct ofproto *ofproto = port->ofproto;
+ const char *name = netdev_get_name(port->netdev);
- /* Query the datapath for port information. */
- error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
+ hmap_remove(&ofproto->ports, &port->hmap_node);
+ shash_delete(&ofproto->port_by_name,
+ shash_find(&ofproto->port_by_name, name));
- /* 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;
- }
+ netdev_close(port->netdev);
+ ofproto->ofproto_class->port_dealloc(port);
+}
- /* Create a new ofport. */
- new_ofport = !error ? make_ofport(&odp_port) : NULL;
+static void
+ofport_destroy(struct ofport *port)
+{
+ if (port) {
+ port->ofproto->ofproto_class->port_destruct(port);
+ ofport_destroy__(port);
+ }
+}
- /* 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;
+struct ofport *
+ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
+{
+ struct ofport *port;
- if (ofport_equal(old_ofport, new_ofport)) {
- /* False alarm--no change. */
- ofport_free(new_ofport);
- return;
+ HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
+ hash_int(ofp_port, 0), &ofproto->ports) {
+ if (port->ofp_port == ofp_port) {
+ return port;
}
}
+ return NULL;
+}
- /* Now deal with the normal cases. */
- if (old_ofport) {
- ofport_remove(p, old_ofport);
- }
- if (new_ofport) {
- ofport_install(p, new_ofport);
+int
+ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
+{
+ 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;
}
- send_port_status(p, new_ofport ? new_ofport : old_ofport,
- (!old_ofport ? OFPPR_ADD
- : !new_ofport ? OFPPR_DELETE
- : OFPPR_MODIFY));
- ofport_free(old_ofport);
- /* Update port groups. */
- refresh_port_groups(p);
+ return error;
}
-static int
-init_ports(struct ofproto *p)
+static void
+update_port(struct ofproto *ofproto, const char *name)
{
- struct odp_port *ports;
- size_t n_ports;
- size_t i;
- int error;
+ struct ofproto_port ofproto_port;
+ struct ofputil_phy_port pp;
+ struct netdev *netdev;
+ struct ofport *port;
- error = dpif_port_list(p->dpif, &ports, &n_ports);
- if (error) {
- return error;
- }
+ 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_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);
+ /* 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;
+ }
+
+ /* 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 (port->ofproto->ofproto_class->port_modified) {
+ port->ofproto->ofproto_class->port_modified(port);
+ }
+
+ 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);
}
+ 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);
}
- free(ports);
- refresh_port_groups(p);
- return 0;
-}
-\f
-static struct ofconn *
-ofconn_create(struct ofproto *p, struct rconn *rconn)
-{
- struct ofconn *ofconn = xmalloc(sizeof *ofconn);
- list_push_back(&p->all_conns, &ofconn->node);
- ofconn->rconn = rconn;
- ofconn->pktbuf = NULL;
- ofconn->send_flow_exp = false;
- ofconn->miss_send_len = 0;
- ofconn->packet_in_counter = rconn_packet_counter_create ();
- ofconn->reply_counter = rconn_packet_counter_create ();
- return ofconn;
+ ofproto_port_destroy(&ofproto_port);
}
-static void
-ofconn_destroy(struct ofconn *ofconn, struct ofproto *p)
+static int
+init_ports(struct ofproto *p)
{
- if (p->executer) {
- executer_rconn_closing(p->executer, ofconn->rconn);
+ 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;
+
+ netdev = ofport_open(&ofproto_port, &pp);
+ if (netdev) {
+ ofport_install(p, netdev, &pp);
+ }
+ }
}
- list_remove(&ofconn->node);
- 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 0;
}
-static void
-ofconn_run(struct ofconn *ofconn, struct ofproto *p)
+/* 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)
{
- int iteration;
+ struct ofport *ofport;
+ int mtu = 0;
- rconn_run(ofconn->rconn);
+ HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
+ struct netdev *netdev = ofport->netdev;
+ int dev_mtu;
- 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 (p->fail_open) {
- fail_open_maybe_recover(p->fail_open);
- }
- handle_openflow(ofconn, p, of_msg);
- ofpbuf_delete(of_msg);
+ /* Skip any internal ports, since that's what we're trying to
+ * set. */
+ if (!strcmp(netdev_get_type(netdev), "internal")) {
+ continue;
}
- }
- if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
- ofconn_destroy(ofconn, p);
+ if (netdev_get_mtu(netdev, &dev_mtu)) {
+ continue;
+ }
+ if (!mtu || dev_mtu < mtu) {
+ mtu = dev_mtu;
+ }
}
+
+ return mtu ? mtu: ETH_PAYLOAD_MAX;
}
+/* Set the MTU of all datapath devices on 'p' to the minimum of the
+ * non-datapath ports. */
static void
-ofconn_wait(struct ofconn *ofconn)
+set_internal_devs_mtu(struct ofproto *p)
{
- 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);
+ 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
-/* 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)
-{
- struct rule *rule = xcalloc(1, sizeof *rule);
- rule->idle_timeout = idle_timeout;
- rule->hard_timeout = hard_timeout;
- rule->used = rule->created = time_msec();
- rule->super = super;
- if (super) {
- list_push_back(&super->list, &rule->list);
- } else {
- list_init(&rule->list);
+static void
+ofproto_rule_destroy__(struct rule *rule)
+{
+ if (rule) {
+ free(rule->actions);
+ rule->ofproto->ofproto_class->rule_dealloc(rule);
}
- 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;
+/* 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.
+ *
+ * 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)
+{
+ assert(!rule->pending);
+ oftable_remove_rule(rule);
+ ofproto_rule_destroy__(rule);
}
-static struct rule *
-rule_from_cls_rule(const struct cls_rule *cls_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)
{
- return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
+ const union ofp_action *oa;
+ size_t left;
+
+ if (out_port == OFPP_NONE) {
+ return true;
+ }
+ 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;
}
-static void
-rule_free(struct rule *rule)
+/* 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.
+ *
+ * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
+ * with statistics for 'packet' either way.
+ *
+ * Takes ownership of 'packet'. */
+static int
+rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
{
- free(rule->actions);
- free(rule->odp_actions);
- free(rule);
+ struct flow flow;
+
+ assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
+
+ flow_extract(packet, 0, 0, in_port, &flow);
+ return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
}
-/* 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).
+/* Returns true if 'rule' should be hidden from the controller.
*
- * 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)
+ * 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)
{
- 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);
+ return rule->cr.priority > UINT16_MAX;
+}
+
+static enum oftable_flags
+rule_get_flags(const struct rule *rule)
+{
+ return rule->ofproto->tables[rule->table_id].flags;
}
static bool
-rule_has_out_port(const struct rule *rule, uint16_t out_port)
+rule_is_modifiable(const struct rule *rule)
{
- const union ofp_action *oa;
- struct actions_iterator i;
+ 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;
+}
- if (out_port == htons(OFPP_NONE)) {
- return true;
+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;
+
+ 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;
}
- for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
- oa = actions_next(&i)) {
- if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
- return true;
+
+ 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);
+ }
+
+ ofconn_send_reply(ofconn, b);
+ return 0;
+}
+
+static enum ofperr
+handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
+{
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
+ struct ofp_switch_config *osc;
+ enum ofp_config_flags flags;
+ struct ofpbuf *buf;
+
+ /* 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 enum ofperr
+handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
+{
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
+ uint16_t flags = ntohs(osc->flags);
+
+ 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));
+ }
}
}
- return false;
+ ofconn_set_invalid_ttl_to_controller(ofconn,
+ (flags & OFPC_INVALID_TTL_TO_CONTROLLER));
+
+ ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
+
+ return 0;
}
-/* Executes the actions indicated by 'rule' on 'packet', which is in flow
- * 'flow' and is considered to have arrived on ODP port 'in_port'.
- *
- * 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'.
+/* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
+ * error message code for the caller to propagate upward. Otherwise, returns
+ * 0.
*
- * 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)
+ * The log message mentions 'msg_type'. */
+static enum ofperr
+reject_slave_controller(struct ofconn *ofconn)
{
- const union odp_action *actions;
- size_t n_actions;
- struct odp_actions a;
-
- /* 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;
+ if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
+ && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
+ return OFPERR_OFPBRC_EPERM;
} else {
- actions = rule->odp_actions;
- n_actions = rule->n_odp_actions;
- }
-
- /* 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);
+ return 0;
}
}
-static void
-rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
- uint16_t in_port)
+static enum ofperr
+handle_packet_out(struct ofconn *ofconn, const struct ofp_packet_out *opo)
{
- struct rule *displaced_rule;
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
+ struct ofputil_packet_out po;
+ struct ofpbuf *payload;
+ struct flow flow;
+ enum ofperr error;
- /* 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);
- }
+ COVERAGE_INC(ofproto_packet_out);
- /* Send the packet and credit it to the rule. */
- if (packet) {
- flow_t flow;
- flow_extract(packet, in_port, &flow);
- rule_execute(p, rule, packet, &flow);
+ error = reject_slave_controller(ofconn);
+ if (error) {
+ return error;
}
- /* 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);
+ /* Decode message. */
+ error = ofputil_decode_packet_out(&po, opo);
+ if (error) {
+ return error;
}
- /* Free the rule that was displaced, if any. */
- if (displaced_rule) {
- rule_destroy(p, displaced_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 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);
- COVERAGE_INC(ofproto_subrule_create);
- cls_rule_from_flow(&subrule->cr, flow, 0,
- (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
- : rule->cr.priority));
- classifier_insert_exact(&ofproto->cls, &subrule->cr);
+ /* 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 subrule;
+ return error;
}
static void
-rule_remove(struct ofproto *ofproto, struct rule *rule)
+update_port_config(struct ofport *port,
+ enum ofputil_port_config config,
+ enum ofputil_port_config mask)
{
- if (rule->cr.wc.wildcards) {
- COVERAGE_INC(ofproto_del_wc_flow);
- ofproto->need_revalidate = true;
- } else {
- rule_uninstall(ofproto, rule);
+ enum ofputil_port_config old_config = port->pp.config;
+ enum ofputil_port_config toggle;
+
+ 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;
}
- classifier_remove(&ofproto->cls, &rule->cr);
- rule_destroy(ofproto, rule);
-}
-/* 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;
+ port->pp.config ^= toggle;
+ if (port->pp.config != old_config) {
+ port->ofproto->ofproto_class->port_reconfigured(port, old_config);
}
}
-static int
-do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
- struct odp_flow_put *put)
+static enum ofperr
+handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
{
- 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->flags = flags;
- return dpif_flow_put(ofproto->dpif, put);
-}
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
+ struct ofputil_port_mod pm;
+ struct ofport *port;
+ enum ofperr error;
-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, rule, &put.flow.stats);
- rule_post_uninstall(p, displaced_rule);
- }
- }
- } else if (displaced_rule) {
- rule_uninstall(p, displaced_rule);
+ error = reject_slave_controller(ofconn);
+ if (error) {
+ return error;
}
-}
-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);
+ error = ofputil_decode_port_mod(oh, &pm);
+ if (error) {
+ return error;
}
-}
-static void
-rule_update_actions(struct ofproto *ofproto, struct rule *rule)
-{
- bool actions_changed = rule_make_actions(ofproto, rule, NULL);
- if (rule->may_install) {
- if (rule->installed) {
- if (actions_changed) {
- /* XXX should really do rule_post_uninstall() for the *old* set
- * of actions, and distinguish the old stats from the new. */
- struct odp_flow_put put;
- do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
- }
- } else {
- rule_install(ofproto, rule, NULL);
- }
+ 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 {
- rule_uninstall(ofproto, rule);
+ update_port_config(port, pm.config, pm.mask);
+ if (pm.advertise) {
+ netdev_set_advertisements(port->netdev, pm.advertise);
+ }
}
+ return 0;
}
-static void
-rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
+static enum ofperr
+handle_desc_stats_request(struct ofconn *ofconn,
+ const struct ofp_stats_msg *request)
{
- uint64_t total_bytes = rule->byte_count + extra_bytes;
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
+ struct ofp_desc_stats *ods;
+ struct ofpbuf *msg;
- 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;
- }
+ 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);
+
+ return 0;
}
-static void
-rule_uninstall(struct ofproto *p, struct rule *rule)
+static enum ofperr
+handle_table_stats_request(struct ofconn *ofconn,
+ const struct ofp_stats_msg *request)
{
- assert(!rule->cr.wc.wildcards);
- if (rule->installed) {
- struct odp_flow odp_flow;
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
+ struct ofp_table_stats *ots;
+ struct ofpbuf *msg;
+ size_t i;
- odp_flow.key = rule->cr.flow;
- odp_flow.actions = NULL;
- odp_flow.n_actions = 0;
- if (!dpif_flow_del(p->dpif, &odp_flow)) {
- update_stats(p, rule, &odp_flow.stats);
- }
- rule->installed = false;
+ ofputil_make_stats_reply(sizeof(struct ofp_stats_msg), request, &msg);
- rule_post_uninstall(p, rule);
+ 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));
}
-}
-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. */
+ p->ofproto_class->get_tables(p, ots);
+
+ for (i = 0; i < p->n_tables; i++) {
+ const struct oftable *table = &p->tables[i];
- if (rule && rule->super) {
- struct rule *super = rule->super;
+ if (table->name) {
+ ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
+ }
- return super->n_actions == 1 &&
- super->actions[0].type == htons(OFPAT_OUTPUT) &&
- super->actions[0].output.port == htons(OFPP_CONTROLLER);
+ if (table->max_flows < ntohl(ots[i].max_entries)) {
+ ots[i].max_entries = htonl(table->max_flows);
+ }
}
- return false;
+ ofconn_send_reply(ofconn, msg);
+ return 0;
}
static void
-rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
+append_port_stat(struct ofport *port, struct list *replies)
{
- struct rule *super = rule->super;
+ struct netdev_stats stats;
+ struct ofp_port_stats *ops;
- rule_account(ofproto, rule, 0);
+ /* 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 (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);
+ 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);
+ }
}
- 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);
- }
+ ofconn_send_replies(ofconn, &replies);
+ return 0;
}
-\f
+
static void
-queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
- struct rconn_packet_counter *counter)
+calc_flow_duration__(long long int start, long long int now,
+ uint32_t *sec, uint32_t *nsec)
{
- update_openflow_length(msg);
- if (rconn_send(ofconn->rconn, msg, counter)) {
- ofpbuf_delete(msg);
- }
+ long long int msecs = now - start;
+ *sec = msecs / 1000;
+ *nsec = (msecs % 1000) * (1000 * 1000);
}
-static void
-send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
- int error, const void *data, size_t len)
+/* 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)
{
- struct ofpbuf *buf;
- struct ofp_error_msg *oem;
+ return (table_id == 0xff || table_id < ofproto->n_tables
+ ? 0
+ : OFPERR_NXBRC_BAD_TABLE_ID);
- if (!(error >> 16)) {
- VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
- error);
- return;
- }
-
- 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);
}
-static void
-send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
- int error)
+static struct oftable *
+next_visible_table(struct ofproto *ofproto, uint8_t table_id)
{
- size_t oh_length = ntohs(oh->length);
- send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
-}
+ struct oftable *table;
-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);
+ for (table = &ofproto->tables[table_id];
+ table < &ofproto->tables[ofproto->n_tables];
+ table++) {
+ if (!(table->flags & OFTABLE_HIDDEN)) {
+ return table;
+ }
+ }
+
+ return NULL;
}
-static int
-handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
+static struct oftable *
+first_matching_table(struct ofproto *ofproto, uint8_t table_id)
{
- struct ofp_header *rq = oh;
- queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
- return 0;
+ 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;
+ }
}
-static int
-handle_features_request(struct ofproto *p, struct ofconn *ofconn,
- struct ofp_header *oh)
+static struct oftable *
+next_matching_table(struct ofproto *ofproto,
+ struct oftable *table, uint8_t table_id)
{
- struct ofp_switch_features *osf;
- struct ofpbuf *buf;
- unsigned int port_no;
- struct ofport *port;
-
- 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_MULTI_PHY_TX);
- 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_TP_SRC) |
- (1u << OFPAT_SET_TP_DST));
-
- 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;
+ return (table_id == 0xff
+ ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
+ : NULL);
}
-static int
-handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
- struct ofp_header *oh)
+/* 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)
{
- struct ofpbuf *buf;
- struct ofp_switch_config *osc;
- uint16_t flags;
- bool drop_frags;
+ struct oftable *table;
+ enum ofperr error;
- /* Figure out flags. */
- dpif_get_drop_frags(p->dpif, &drop_frags);
- flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
- if (ofconn->send_flow_exp) {
- flags |= OFPC_SEND_FLOW_EXP;
+ error = check_table_id(ofproto, table_id);
+ if (error) {
+ return error;
}
- /* 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);
+ list_init(rules);
+ FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
+ struct cls_cursor cursor;
+ struct rule *rule;
+ 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);
+ }
+ }
+ }
return 0;
}
-static int
-handle_set_config(struct ofproto *p, struct ofconn *ofconn,
- struct ofp_switch_config *osc)
-{
- uint16_t flags;
+/* 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;
- error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
+ error = check_table_id(ofproto, table_id);
if (error) {
return error;
}
- flags = ntohs(osc->flags);
- ofconn->send_flow_exp = (flags & OFPC_SEND_FLOW_EXP) != 0;
+ list_init(rules);
+ FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
+ struct rule *rule;
- if (ofconn == p->controller) {
- 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;
+ 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 ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
- if (ntohs(osc->miss_send_len) != 0) {
- ofconn->pktbuf = pktbuf_create();
- } else {
- pktbuf_destroy(ofconn->pktbuf);
- }
+/* 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)
+{
+ return (age_ms < 0 ? 0
+ : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
+ : (unsigned int) age_ms / 1000);
+}
+
+static enum ofperr
+handle_flow_stats_request(struct ofconn *ofconn,
+ const struct ofp_stats_msg *osm)
+{
+ 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 = ofputil_decode_flow_stats_request(&fsr, &osm->header);
+ if (error) {
+ return error;
+ }
+
+ error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
+ fsr.cookie, fsr.cookie_mask,
+ fsr.out_port, &rules);
+ if (error) {
+ return error;
}
- ofconn->miss_send_len = ntohs(osc->miss_send_len);
+ 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);
return 0;
}
static void
-add_output_group_action(struct odp_actions *actions, uint16_t group,
- uint16_t *nf_output_iface)
+flow_stats_ds(struct rule *rule, struct ds *results)
{
- odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
+ uint64_t packet_count, byte_count;
- if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
- *nf_output_iface = NF_OUT_FLOOD;
+ rule->ofproto->ofproto_class->rule_get_stats(rule,
+ &packet_count, &byte_count);
+
+ 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
-add_controller_action(struct odp_actions *actions,
- const struct ofp_action_output *oao)
+/* 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)
{
- union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
- a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
-}
-
-struct action_xlate_ctx {
- /* Input. */
- const 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. */
-};
+ struct oftable *table;
-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)
-{
- const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
+ OFPROTO_FOR_EACH_TABLE (table, p) {
+ struct cls_cursor cursor;
+ struct rule *rule;
- if (ofport) {
- if (ofport->opp.config & OFPPC_NO_FWD) {
- /* Forwarding disabled on port. */
- return;
+ cls_cursor_init(&cursor, &table->cls, NULL);
+ CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
+ flow_stats_ds(rule, results);
}
- } 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;
}
-static struct rule *
-lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
+/* 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)
{
- 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);
- return NULL;
- }
-
- return rule;
+ ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
}
-static void
-xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
+/* 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 (!ctx->recurse) {
- struct rule *rule;
- flow_t flow;
+ 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);
+}
- flow = *ctx->flow;
- flow.in_port = in_port;
+/* 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;
- rule = lookup_valid_rule(ctx->ofproto, &flow);
- if (rule) {
- if (rule->super) {
- rule = rule->super;
- }
+ error = ofputil_decode_flow_stats_request(&request, &osm->header);
+ if (error) {
+ return error;
+ }
- ctx->recurse++;
- do_xlate_actions(rule->actions, rule->n_actions, ctx);
- ctx->recurse--;
- }
+ error = collect_rules_loose(ofproto, request.table_id, &request.match,
+ request.cookie, request.cookie_mask,
+ request.out_port, &rules);
+ if (error) {
+ return error;
}
-}
-static void
-xlate_output_action(struct action_xlate_ctx *ctx,
- const struct ofp_action_output *oao)
-{
- uint16_t odp_port;
- uint16_t prev_nf_output_iface = ctx->nf_output_iface;
+ 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;
- ctx->nf_output_iface = NF_OUT_DROP;
+ ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
+ &byte_count);
- switch (ntohs(oao->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;
+ if (packet_count == UINT64_MAX) {
+ unknown_packets = true;
+ } else {
+ stats.packet_count += packet_count;
}
- 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, oao);
- break;
- case OFPP_LOCAL:
- add_output_action(ctx, ODPP_LOCAL);
- break;
- default:
- odp_port = ofp_port_to_odp_port(ntohs(oao->port));
- if (odp_port != ctx->flow->in_port) {
- add_output_action(ctx, odp_port);
+
+ if (byte_count == UINT64_MAX) {
+ unknown_bytes = true;
+ } else {
+ stats.byte_count += byte_count;
}
- break;
- }
- if (prev_nf_output_iface == NF_OUT_FLOOD) {
- ctx->nf_output_iface = NF_OUT_FLOOD;
- } else if (ctx->nf_output_iface == NF_OUT_DROP) {
- ctx->nf_output_iface = prev_nf_output_iface;
- } else if (prev_nf_output_iface != NF_OUT_DROP &&
- ctx->nf_output_iface != NF_OUT_FLOOD) {
- ctx->nf_output_iface = NF_OUT_MULTI;
+ stats.flow_count++;
+ }
+ if (unknown_packets) {
+ stats.packet_count = UINT64_MAX;
}
+ if (unknown_bytes) {
+ stats.byte_count = UINT64_MAX;
+ }
+
+ reply = ofputil_encode_aggregate_stats_reply(&stats, osm);
+ ofconn_send_reply(ofconn, reply);
+
+ return 0;
}
+struct queue_stats_cbdata {
+ struct ofport *ofport;
+ struct list replies;
+};
+
static void
-xlate_nicira_action(struct action_xlate_ctx *ctx,
- const struct nx_action_header *nah)
+put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
+ const struct netdev_queue_stats *stats)
{
- const struct nx_action_resubmit *nar;
- 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;
+ struct ofp_queue_stats *reply;
- default:
- VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
- break;
- }
+ 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
-do_xlate_actions(const union ofp_action *in, size_t n_in,
- struct action_xlate_ctx *ctx)
+handle_queue_stats_dump_cb(uint32_t queue_id,
+ struct netdev_queue_stats *stats,
+ void *cbdata_)
{
- 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;
- }
+ struct queue_stats_cbdata *cbdata = cbdata_;
- 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);
- 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);
- oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
- break;
-
- case OFPAT_STRIP_VLAN:
- odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
- break;
+ put_queue_stats(cbdata, queue_id, stats);
+}
- 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);
- break;
+static void
+handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
+ struct queue_stats_cbdata *cbdata)
+{
+ 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;
- 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);
- break;
+ if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
+ put_queue_stats(cbdata, queue_id, &stats);
+ }
+ }
+}
- case OFPAT_SET_NW_SRC:
- oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
- oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
- break;
+static enum ofperr
+handle_queue_stats_request(struct ofconn *ofconn,
+ const struct ofp_queue_stats_request *qsr)
+{
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
+ struct queue_stats_cbdata cbdata;
+ struct ofport *port;
+ unsigned int port_no;
+ uint32_t queue_id;
- case OFPAT_SET_TP_SRC:
- oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
- oa->tp_port.tp_port = ia->tp_port.tp_port;
- break;
+ COVERAGE_INC(ofproto_queue_req);
- case OFPAT_VENDOR:
- xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
- break;
+ ofputil_start_stats_reply(&qsr->osm, &cbdata.replies);
- default:
- VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
- break;
+ 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) {
+ handle_queue_stats_for_port(port, queue_id, &cbdata);
+ }
+ } else {
+ ofpbuf_list_delete(&cbdata.replies);
+ return OFPERR_OFPQOFC_BAD_PORT;
}
+ ofconn_send_replies(ofconn, &cbdata.replies);
+
+ return 0;
}
-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);
-
- /* 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);
+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;
+ }
+ }
}
- return 0;
+
+ return false;
}
-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;
+/* 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)
+{
+ struct oftable *table;
+ struct ofopgroup *group;
+ struct rule *victim;
+ struct rule *rule;
int error;
- error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
+ error = check_table_id(ofproto, fm->table_id);
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;
+ /* 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];
}
- payload = *buffer;
+ } else if (fm->table_id < ofproto->n_tables) {
+ table = &ofproto->tables[fm->table_id];
} else {
- buffer = NULL;
+ return OFPERR_NXFMFC_BAD_TABLE_ID;
}
- flow_extract(&payload, 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;
+ if (table->flags & OFTABLE_READONLY) {
+ return OFPERR_OFPBRC_EPERM;
}
- dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
- &payload);
- ofpbuf_delete(buffer);
+ /* Check for overlap, if requested. */
+ if (fm->flags & OFPFF_CHECK_OVERLAP
+ && classifier_rule_overlaps(&table->cls, &fm->cr)) {
+ return OFPERR_OFPFMFC_OVERLAP;
+ }
- return 0;
-}
+ /* Serialize against pending deletion. */
+ if (is_flow_deletion_pending(ofproto, &fm->cr, table - ofproto->tables)) {
+ return OFPROTO_POSTPONE;
+ }
-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);
+ /* 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 {
- netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
+ evict = NULL;
}
+
+ group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
+ ofoperation_create(group, rule, OFOPERATION_ADD);
+ rule->pending->victim = victim;
+
+ error = ofproto->ofproto_class->rule_construct(rule);
+ if (error) {
+ ofoperation_destroy(rule->pending);
+ } else if (evict) {
+ delete_flow__(evict, group);
+ }
+ ofopgroup_submit(group);
}
-#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_group(p, DP_GROUP_FLOOD);
- }
- if (mask & OFPPC_NO_PACKET_IN) {
- port->opp.config ^= OFPPC_NO_PACKET_IN;
+
+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. */
-static int
-handle_port_mod(struct ofproto *p, struct ofp_header *oh)
+/* 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)
{
- const struct ofp_port_mod *opm;
- struct ofport *port;
- int error;
-
- error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
- if (error) {
- return error;
- }
- opm = (struct ofp_port_mod *) oh;
+ 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;
+ }
- 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 (!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;
}
- return 0;
+ ofopgroup_submit(group);
+
+ return error;
}
-static struct ofpbuf *
-make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
-{
- struct ofp_stats_reply *osr;
- struct ofpbuf *msg;
+/* 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 list rules;
+ int error;
- 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;
+ 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));
}
-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);
-}
+/* 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;
-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);
+ 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);
}
+\f
+/* OFPFC_DELETE implementation. */
-static int
-handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
- struct ofp_stats_request *request)
+static void
+delete_flow__(struct rule *rule, struct ofopgroup *group)
{
- struct ofp_desc_stats *ods;
- struct ofpbuf *msg;
+ struct ofproto *ofproto = rule->ofproto;
- msg = start_stats_reply(request, sizeof *ods);
- ods = append_stats_reply(sizeof *ods, ofconn, &msg);
- strncpy(ods->mfr_desc, p->manufacturer, sizeof ods->mfr_desc);
- strncpy(ods->hw_desc, p->hardware, sizeof ods->hw_desc);
- strncpy(ods->sw_desc, p->software, sizeof ods->sw_desc);
- strncpy(ods->serial_num, p->serial, sizeof ods->serial_num);
- queue_tx(msg, ofconn, ofconn->reply_counter);
+ ofproto_rule_send_removed(rule, OFPRR_DELETE);
- return 0;
+ ofoperation_create(group, rule, OFOPERATION_DELETE);
+ oftable_remove_rule(rule);
+ ofproto->ofproto_class->rule_destruct(rule);
}
-static void
-count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
+/* 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 rule *rule = rule_from_cls_rule(cls_rule);
- int *n_subrules = n_subrules_;
+ struct rule *rule, *next;
+ struct ofopgroup *group;
- if (rule->super) {
- (*n_subrules)++;
+ group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
+ LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
+ delete_flow__(rule, group);
}
-}
+ ofopgroup_submit(group);
-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 = 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;
}
-static int
-handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
- struct ofp_stats_request *request)
+/* 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 ofp_port_stats *ops;
- struct ofpbuf *msg;
- struct ofport *port;
- unsigned int port_no;
+ struct list rules;
+ enum ofperr error;
- msg = start_stats_reply(request, sizeof *ops * 16);
- PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
- struct netdev_stats stats;
-
- /* 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);
-
- ops = append_stats_reply(sizeof *ops, ofconn, &msg);
- 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);
- }
-
- queue_tx(msg, ofconn, ofconn->reply_counter);
- return 0;
+ 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);
}
-struct flow_stats_cbdata {
- struct ofproto *ofproto;
- struct ofconn *ofconn;
- uint16_t out_port;
- struct ofpbuf *msg;
-};
-
-static void
-query_stats(struct ofproto *p, struct rule *rule,
- uint64_t *packet_countp, uint64_t *byte_countp)
+/* 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)
{
- uint64_t packet_count, byte_count;
- struct rule *subrule;
- struct odp_flow *odp_flows;
- size_t n_odp_flows;
-
- packet_count = rule->packet_count;
- byte_count = rule->byte_count;
-
- n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
- odp_flows = xcalloc(1, 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 list rules;
+ enum ofperr error;
- packet_count = rule->packet_count;
- byte_count = rule->byte_count;
- 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;
- }
- }
- free(odp_flows);
-
- *packet_countp = packet_count;
- *byte_countp = byte_count;
+ 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
-flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
+ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
{
- 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;
+ struct ofputil_flow_removed fr;
- if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
+ if (rule_is_hidden(rule) || !rule->send_flow_removed) {
return;
}
- act_len = sizeof *rule->actions * rule->n_actions;
- len = offsetof(struct ofp_flow_stats, actions) + act_len;
-
- query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
-
- 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, &ofs->match);
- ofs->duration = htonl((time_msec() - rule->created) / 1000);
- 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);
-}
-
-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);
-}
+ 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);
-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_LENGTH);
- }
- 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(&target, &fsr->match, 0);
- 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;
+ connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
}
-struct flow_stats_ds_cbdata {
- struct ofproto *ofproto;
- struct ds *results;
-};
-
-static void
-flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
+void
+ofproto_rule_update_used(struct rule *rule, long long int used)
{
- 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;
+ if (used > rule->used) {
+ struct eviction_group *evg = rule->eviction_group;
- /* Don't report on subrules. */
- if (rule->super != NULL) {
- return;
+ rule->used = used;
+ if (evg) {
+ heap_change(&evg->rules, &rule->evg_node,
+ rule_eviction_priority(rule));
+ }
}
-
- query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
- flow_to_ovs_match(&rule->cr.flow, rule->cr.wc.wildcards, &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");
}
-/* Adds a pretty-printed description of all flows to 'results', including
- * those marked hidden by secchan (e.g., by in-band control). */
+/* 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_get_all_flows(struct ofproto *p, struct ds *results)
+ofproto_rule_expire(struct rule *rule, uint8_t reason)
{
- struct ofp_match match;
- struct cls_rule target;
- struct flow_stats_ds_cbdata cbdata;
+ struct ofproto *ofproto = rule->ofproto;
+ struct ofopgroup *group;
- memset(&match, 0, sizeof match);
- match.wildcards = htonl(OFPFW_ALL);
+ assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
- cbdata.ofproto = p;
- cbdata.results = results;
+ ofproto_rule_send_removed(rule, reason);
- cls_rule_from_match(&target, &match, 0);
- classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
- flow_stats_ds_cb, &cbdata);
+ group = ofopgroup_create_unattached(ofproto);
+ ofoperation_create(group, rule, OFOPERATION_DELETE);
+ oftable_remove_rule(rule);
+ ofproto->ofproto_class->rule_destruct(rule);
+ ofopgroup_submit(group);
}
-
-struct aggregate_stats_cbdata {
- struct ofproto *ofproto;
- uint16_t out_port;
- uint64_t packet_count;
- uint64_t byte_count;
- uint32_t n_flows;
-};
-
-static void
-aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
+\f
+static enum ofperr
+handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
{
- struct rule *rule = rule_from_cls_rule(rule_);
- struct aggregate_stats_cbdata *cbdata = cbdata_;
- uint64_t packet_count, byte_count;
+ struct ofputil_flow_mod fm;
+ enum ofperr error;
- if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
- return;
+ error = reject_slave_controller(ofconn);
+ if (error) {
+ return error;
}
- query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
-
- cbdata->packet_count += packet_count;
- cbdata->byte_count += byte_count;
- cbdata->n_flows++;
-}
+ error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn));
+ if (error) {
+ return error;
+ }
-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;
+ /* 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;
+ }
- if (arg_size != sizeof *asr) {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
- }
- 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(&target, &asr->match, 0);
- 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;
+ return handle_flow_mod__(ofconn_get_ofproto(ofconn), ofconn, &fm, oh);
}
-static int
-handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
- struct ofp_header *oh)
+static enum ofperr
+handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
+ const struct ofputil_flow_mod *fm,
+ const struct ofp_header *oh)
{
- struct ofp_stats_request *osr;
- size_t arg_size;
- int error;
-
- error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
- 1, &arg_size);
- if (error) {
- return error;
+ if (ofproto->n_pending >= 50) {
+ assert(!list_is_empty(&ofproto->pending));
+ return OFPROTO_POSTPONE;
}
- osr = (struct ofp_stats_request *) oh;
-
- switch (ntohs(osr->type)) {
- case OFPST_DESC:
- return handle_desc_stats_request(p, ofconn, osr);
- case OFPST_FLOW:
- return handle_flow_stats_request(p, ofconn, osr, arg_size);
+ switch (fm->command) {
+ case OFPFC_ADD:
+ return add_flow(ofproto, ofconn, fm, oh);
- case OFPST_AGGREGATE:
- return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
+ case OFPFC_MODIFY:
+ return modify_flows_loose(ofproto, ofconn, fm, oh);
- case OFPST_TABLE:
- return handle_table_stats_request(p, ofconn, osr);
+ case OFPFC_MODIFY_STRICT:
+ return modify_flow_strict(ofproto, ofconn, fm, oh);
- case OFPST_PORT:
- return handle_port_stats_request(p, ofconn, osr);
+ 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;
}
+
+ ofconn_set_role(ofconn, role);
+
+ reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
+ reply->role = htonl(role);
+ ofconn_send_reply(ofconn, buf);
+
+ return 0;
}
-static int
-add_flow(struct ofproto *p, struct ofconn *ofconn,
- struct ofp_flow_mod *ofm, size_t n_actions)
+static enum ofperr
+handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
+ const struct ofp_header *oh)
{
- struct ofpbuf *packet;
- struct rule *rule;
- uint16_t in_port;
- int error;
-
- rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
- n_actions, ntohs(ofm->idle_timeout),
- ntohs(ofm->hard_timeout));
- cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority));
+ const struct nx_flow_mod_table_id *msg
+ = (const struct nx_flow_mod_table_id *) oh;
+ enum ofputil_protocol cur, next;
- packet = NULL;
- error = 0;
- if (ofm->buffer_id != htonl(UINT32_MAX)) {
- error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
- &packet, &in_port);
- }
+ 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 int
-modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
- size_t n_actions, uint16_t command, struct rule *rule)
+static enum ofperr
+handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
{
- if (rule_is_hidden(rule)) {
- return 0;
+ const struct nx_set_flow_format *msg
+ = (const struct nx_set_flow_format *) oh;
+ enum ofputil_protocol cur, next;
+ enum ofputil_protocol next_base;
+
+ next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
+ if (!next_base) {
+ return OFPERR_OFPBRC_EPERM;
}
- if (command == OFPFC_DELETE) {
- rule_remove(p, rule);
- } else {
- size_t actions_len = n_actions * sizeof *rule->actions;
+ 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;
+ }
- if (n_actions == rule->n_actions
- && !memcmp(ofm->actions, rule->actions, actions_len))
- {
- return 0;
- }
+ ofconn_set_protocol(ofconn, next);
+ return 0;
+}
- free(rule->actions);
- rule->actions = xmemdup(ofm->actions, actions_len);
- rule->n_actions = n_actions;
+static enum ofperr
+handle_nxt_set_packet_in_format(struct ofconn *ofconn,
+ const struct ofp_header *oh)
+{
+ const struct nx_set_packet_in_format *msg;
+ uint32_t format;
- if (rule->cr.wc.wildcards) {
- COVERAGE_INC(ofproto_mod_wc_flow);
- p->need_revalidate = true;
- } else {
- rule_update_actions(p, rule);
- }
+ msg = (const struct nx_set_packet_in_format *) oh;
+ format = ntohl(msg->format);
+ if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
+ return OFPERR_OFPBRC_EPERM;
}
+ 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;
+ }
+
+ ofconn_set_packet_in_format(ofconn, format);
return 0;
}
-static int
-modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm,
- size_t n_actions, uint16_t command)
+static enum ofperr
+handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
{
- struct rule *rule;
- uint32_t wildcards;
- flow_t flow;
+ const struct nx_async_config *msg = (const struct nx_async_config *) oh;
+ uint32_t master[OAM_N_TYPES];
+ uint32_t slave[OAM_N_TYPES];
- flow_from_match(&flow, &wildcards, &ofm->match);
- rule = rule_from_cls_rule(classifier_find_rule_exactly(
- &p->cls, &flow, wildcards,
- ntohs(ofm->priority)));
+ 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]);
- if (rule) {
- if (command == OFPFC_DELETE
- && ofm->out_port != htons(OFPP_NONE)
- && !rule_has_out_port(rule, ofm->out_port)) {
- return 0;
- }
+ 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);
- modify_flow(p, ofm, n_actions, command, rule);
- }
return 0;
}
-struct modify_flows_cbdata {
- struct ofproto *ofproto;
- const struct ofp_flow_mod *ofm;
- uint16_t out_port;
- size_t n_actions;
- uint16_t command;
-};
-
-static void
-modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
+static enum ofperr
+handle_nxt_set_controller_id(struct ofconn *ofconn,
+ const struct ofp_header *oh)
{
- struct rule *rule = rule_from_cls_rule(rule_);
- struct modify_flows_cbdata *cbdata = cbdata_;
+ const struct nx_controller_id *nci;
- if (cbdata->out_port != htons(OFPP_NONE)
- && !rule_has_out_port(rule, cbdata->out_port)) {
- return;
+ nci = (const struct nx_controller_id *) oh;
+ if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
+ return OFPERR_NXBRC_MUST_BE_ZERO;
}
- modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions,
- cbdata->command, rule);
+ ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
+ return 0;
}
-static int
-modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm,
- size_t n_actions, uint16_t command)
+static enum ofperr
+handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
{
- struct modify_flows_cbdata cbdata;
- struct cls_rule target;
-
- cbdata.ofproto = p;
- cbdata.ofm = ofm;
- cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port
- : htons(OFPP_NONE));
- cbdata.n_actions = n_actions;
- cbdata.command = command;
+ struct ofp_header *ob;
+ struct ofpbuf *buf;
- cls_rule_from_match(&target, &ofm->match, 0);
+ if (ofconn_has_pending_opgroups(ofconn)) {
+ return OFPROTO_POSTPONE;
+ }
- classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
- modify_flows_cb, &cbdata);
+ ob = make_openflow_xid(sizeof *ob, OFPT10_BARRIER_REPLY, oh->xid, &buf);
+ ofconn_send_reply(ofconn, buf);
return 0;
}
-static int
-handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
- struct ofp_flow_mod *ofm)
+static enum ofperr
+handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
{
- size_t n_actions;
- int error;
-
- error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
- sizeof *ofm->actions, &n_actions);
- if (error) {
- return error;
- }
-
- normalize_match(&ofm->match);
- if (!ofm->match.wildcards) {
- ofm->priority = htons(UINT16_MAX);
- }
+ const struct ofp_header *oh = msg->data;
+ const struct ofputil_msg_type *type;
+ enum ofperr error;
- error = validate_actions((const union ofp_action *) ofm->actions,
- n_actions, p->max_ports);
+ error = ofputil_decode_msg_type(oh, &type);
if (error) {
return error;
}
- switch (ntohs(ofm->command)) {
- case OFPFC_ADD:
- return add_flow(p, ofconn, ofm, n_actions);
+ switch (ofputil_msg_type_code(type)) {
+ /* OpenFlow requests. */
+ case OFPUTIL_OFPT_ECHO_REQUEST:
+ return handle_echo_request(ofconn, oh);
- case OFPFC_MODIFY:
- return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY);
+ case OFPUTIL_OFPT_FEATURES_REQUEST:
+ return handle_features_request(ofconn, oh);
- case OFPFC_MODIFY_STRICT:
- return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY);
+ case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
+ return handle_get_config_request(ofconn, oh);
- case OFPFC_DELETE:
- return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE);
+ case OFPUTIL_OFPT_SET_CONFIG:
+ return handle_set_config(ofconn, msg->data);
- case OFPFC_DELETE_STRICT:
- return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE);
+ case OFPUTIL_OFPT_PACKET_OUT:
+ return handle_packet_out(ofconn, msg->data);
- default:
- return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
- }
-}
+ case OFPUTIL_OFPT_PORT_MOD:
+ return handle_port_mod(ofconn, oh);
-static void
-send_capability_reply(struct ofproto *p, struct ofconn *ofconn, uint32_t xid)
-{
- struct ofmp_capability_reply *ocr;
- struct ofpbuf *b;
- char capabilities[] = "com.nicira.mgmt.manager=false\n";
+ case OFPUTIL_OFPT_FLOW_MOD:
+ return handle_flow_mod(ofconn, oh);
- ocr = make_openflow_xid(sizeof(*ocr), OFPT_VENDOR, xid, &b);
- ocr->header.header.vendor = htonl(NX_VENDOR_ID);
- ocr->header.header.subtype = htonl(NXT_MGMT);
- ocr->header.type = htons(OFMPT_CAPABILITY_REPLY);
+ case OFPUTIL_OFPT_BARRIER_REQUEST:
+ return handle_barrier_request(ofconn, oh);
- ocr->format = htonl(OFMPCOF_SIMPLE);
- ocr->mgmt_id = htonll(p->mgmt_id);
+ /* OpenFlow replies. */
+ case OFPUTIL_OFPT_ECHO_REPLY:
+ return 0;
- ofpbuf_put(b, capabilities, strlen(capabilities));
+ /* Nicira extension requests. */
+ case OFPUTIL_NXT_ROLE_REQUEST:
+ return handle_role_request(ofconn, oh);
- queue_tx(b, ofconn, ofconn->reply_counter);
-}
+ case OFPUTIL_NXT_FLOW_MOD_TABLE_ID:
+ return handle_nxt_flow_mod_table_id(ofconn, oh);
-static int
-handle_ofmp(struct ofproto *p, struct ofconn *ofconn,
- struct ofmp_header *ofmph)
-{
- size_t msg_len = ntohs(ofmph->header.header.length);
- if (msg_len < sizeof(*ofmph)) {
- VLOG_WARN_RL(&rl, "dropping short managment message: %zu\n", msg_len);
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
- }
+ case OFPUTIL_NXT_SET_FLOW_FORMAT:
+ return handle_nxt_set_flow_format(ofconn, oh);
- if (ofmph->type == htons(OFMPT_CAPABILITY_REQUEST)) {
- struct ofmp_capability_request *ofmpcr;
+ case OFPUTIL_NXT_SET_PACKET_IN_FORMAT:
+ return handle_nxt_set_packet_in_format(ofconn, oh);
- if (msg_len < sizeof(struct ofmp_capability_request)) {
- VLOG_WARN_RL(&rl, "dropping short capability request: %zu\n",
- msg_len);
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
- }
+ case OFPUTIL_NXT_SET_CONTROLLER_ID:
+ return handle_nxt_set_controller_id(ofconn, oh);
- ofmpcr = (struct ofmp_capability_request *)ofmph;
- if (ofmpcr->format != htonl(OFMPCAF_SIMPLE)) {
- /* xxx Find a better type than bad subtype */
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
- }
+ case OFPUTIL_NXT_FLOW_MOD:
+ return handle_flow_mod(ofconn, oh);
- send_capability_reply(p, ofconn, ofmph->header.header.xid);
+ case OFPUTIL_NXT_FLOW_AGE:
+ /* Nothing to do. */
return 0;
- } else {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
+
+ 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);
}
}
-static int
-handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
+static bool
+handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
{
- struct ofp_vendor_header *ovh = msg;
- struct nicira_header *nh;
-
- if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
- }
- 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)) {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
+ 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. */
- nh = msg;
- switch (ntohl(nh->subtype)) {
- case NXT_STATUS_REQUEST:
- return switch_status_handle_request(p->switch_status, ofconn->rconn,
- msg);
+/* 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;
+}
- case NXT_ACT_SET_CONFIG:
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
+/* 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 ofopgroup *group = ofopgroup_create_unattached(ofproto);
+ if (ofconn) {
+ size_t request_len = ntohs(request->length);
- case NXT_ACT_GET_CONFIG:
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
+ assert(ofconn_get_ofproto(ofconn) == ofproto);
- case NXT_COMMAND_REQUEST:
- if (p->executer) {
- return executer_handle_request(p->executer, ofconn->rconn, msg);
- }
- break;
+ 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;
+}
- case NXT_MGMT:
- return handle_ofmp(p, ofconn, msg);
+/* Submits 'group' for processing.
+ *
+ * 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
+ofopgroup_submit(struct ofopgroup *group)
+{
+ if (list_is_empty(&group->ops)) {
+ ofopgroup_destroy(group);
+ } else {
+ list_push_back(&group->ofproto->pending, &group->ofproto_node);
+ group->ofproto->n_pending++;
}
+}
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
+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
-handle_openflow(struct ofconn *ofconn, struct ofproto *p,
- struct ofpbuf *ofp_msg)
+ofoperation_create(struct ofopgroup *group, struct rule *rule,
+ enum ofoperation_type type)
{
- struct ofp_header *oh = ofp_msg->data;
- int error;
+ struct ofoperation *op;
- COVERAGE_INC(ofproto_recv_openflow);
- switch (oh->type) {
- case OFPT_ECHO_REQUEST:
- error = handle_echo_request(ofconn, oh);
- break;
+ assert(!rule->pending);
- case OFPT_ECHO_REPLY:
- error = 0;
- break;
+ 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;
- case OFPT_FEATURES_REQUEST:
- error = handle_features_request(p, ofconn, oh);
- break;
+ if (type == OFOPERATION_DELETE) {
+ hmap_insert(&op->group->ofproto->deletions, &op->hmap_node,
+ cls_rule_hash(&rule->cr, rule->table_id));
+ }
+}
- case OFPT_GET_CONFIG_REQUEST:
- error = handle_get_config_request(p, ofconn, oh);
- break;
+static void
+ofoperation_destroy(struct ofoperation *op)
+{
+ struct ofopgroup *group = op->group;
- case OFPT_SET_CONFIG:
- error = handle_set_config(p, ofconn, ofp_msg->data);
- break;
+ if (op->rule) {
+ op->rule->pending = NULL;
+ }
+ if (op->type == OFOPERATION_DELETE) {
+ hmap_remove(&group->ofproto->deletions, &op->hmap_node);
+ }
+ list_remove(&op->group_node);
+ free(op->actions);
+ free(op);
- case OFPT_PACKET_OUT:
- error = handle_packet_out(p, ofconn, ofp_msg->data);
- break;
+ if (list_is_empty(&group->ops) && !list_is_empty(&group->ofproto_node)) {
+ ofopgroup_destroy(group);
+ }
+}
- case OFPT_PORT_MOD:
- error = handle_port_mod(p, oh);
- break;
+/* 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);
+ }
+ }
+ if (!group->error) {
+ group->error = error;
+ }
- case OFPT_FLOW_MOD:
- error = handle_flow_mod(p, ofconn, ofp_msg->data);
+ 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 OFPT_STATS_REQUEST:
- error = handle_stats_request(p, ofconn, oh);
+ case OFOPERATION_DELETE:
+ assert(!error);
+ ofproto_rule_destroy__(rule);
+ op->rule = NULL;
break;
- case OFPT_VENDOR:
- error = handle_vendor(p, ofconn, ofp_msg->data);
+ 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;
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);
- }
- error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
- break;
+ NOT_REACHED();
}
+ ofoperation_destroy(op);
+}
- if (error) {
- send_error_oh(ofconn, ofp_msg->data, error);
- }
+struct rule *
+ofoperation_get_victim(struct ofoperation *op)
+{
+ assert(op->type == OFOPERATION_ADD);
+ return op->victim;
}
\f
-static void
-handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
+static uint64_t
+pick_datapath_id(const struct ofproto *ofproto)
{
- struct odp_msg *msg = packet->data;
- uint16_t in_port = odp_port_to_ofp_port(msg->port);
- struct rule *rule;
- struct ofpbuf payload;
- flow_t flow;
-
- /* Handle controller actions. */
- if (msg->type == _ODPL_ACTION_NR) {
- COVERAGE_INC(ofproto_ctlr_action);
- pinsched_send(p->action_sched, in_port, packet,
- send_packet_in_action, p);
- return;
+ const struct ofport *port;
+
+ port = ofproto_get_port(ofproto, OFPP_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;
+}
- payload.data = msg + 1;
- payload.size = msg->length - sizeof *msg;
- flow_extract(&payload, msg->port, &flow);
+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
+/* Table overflow policy. */
- /* 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;
+/* 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 eviction_group *evg;
- 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);
+ if (!table->eviction_fields) {
+ return NULL;
}
- 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;
+ /* 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;
+
+ HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
+ if (rule->evictable) {
+ return rule;
}
- } else {
- VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
}
-
- COVERAGE_INC(ofproto_packet_in);
- pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
- return;
}
- 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 */
- }
- }
-
- rule_execute(p, rule, &payload, &flow);
- rule_reinstall(p, rule);
-
- if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY
- && rconn_is_connected(p->controller->rconn)) {
- /*
- * 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.
- */
- pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
- } else {
- ofpbuf_delete(packet);
- }
+ return NULL;
}
-\f
+
+/* 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
-revalidate_cb(struct cls_rule *sub_, void *cbdata_)
+ofproto_evict(struct ofproto *ofproto)
{
- struct rule *sub = rule_from_cls_rule(sub_);
- struct revalidate_cbdata *cbdata = cbdata_;
+ struct ofopgroup *group;
+ struct oftable *table;
+
+ 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;
- if (cbdata->revalidate_all
- || (cbdata->revalidate_subrules && sub->super)
- || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
- revalidate_rule(cbdata->ofproto, sub);
+ rule = choose_rule_to_evict(table);
+ if (!rule || rule->pending) {
+ break;
+ }
+
+ ofoperation_create(group, rule, OFOPERATION_DELETE);
+ oftable_remove_rule(rule);
+ ofproto->ofproto_class->rule_destruct(rule);
+ }
}
+ ofopgroup_submit(group);
}
+\f
+/* Eviction groups. */
-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;
+/* 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)
+{
+ uint16_t size = MIN(UINT16_MAX, n_rules);
+ return (size << 16) | random_uint16();
}
-static struct ofpbuf *
-compose_flow_exp(const struct rule *rule, long long int now, uint8_t reason)
+/* Updates 'evg', an eviction_group within 'table', following a change that
+ * adds or removes rules in 'evg'. */
+static void
+eviction_group_resized(struct oftable *table, struct eviction_group *evg)
{
- struct ofp_flow_expired *ofe;
- struct ofpbuf *buf;
+ heap_change(&table->eviction_groups_by_size, &evg->size_node,
+ eviction_group_priority(heap_count(&evg->rules)));
+}
- ofe = make_openflow(sizeof *ofe, OFPT_FLOW_EXPIRED, &buf);
- flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofe->match);
- ofe->priority = htons(rule->cr.priority);
- ofe->reason = reason;
- ofe->duration = htonl((now - rule->created) / 1000);
- ofe->packet_count = htonll(rule->packet_count);
- ofe->byte_count = htonll(rule->byte_count);
+/* 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;
- return buf;
+ 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);
}
+/* Removes 'rule' from its eviction group, if any. */
static void
-send_flow_exp(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 (ofconn->send_flow_exp && rconn_is_connected(ofconn->rconn)) {
- if (prev) {
- queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
- } else {
- buf = compose_flow_exp(rule, now, reason);
+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;
+
+ 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);
+ }
+ }
+}
+
+/* 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;
+
+ 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;
+
+ mf_get_value(sf->field, &rule->cr.flow, &value);
+ if (sf->ofs) {
+ bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
}
- prev = ofconn;
+ 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);
}
}
- if (prev) {
- queue_tx(buf, prev, prev->reply_counter);
+
+ return hash;
+}
+
+/* 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 eviction_group *evg;
+
+ HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
+ return evg;
}
+
+ 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);
+
+ return evg;
}
-static void
-uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
+/* 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)
{
- assert(rule->installed);
- assert(!rule->cr.wc.wildcards);
+ long long int hard_expiration;
+ long long int idle_expiration;
+ long long int expiration;
+ uint32_t expiration_offset;
- if (rule->super) {
- rule_remove(ofproto, rule);
- } else {
- rule_uninstall(ofproto, rule);
+ /* 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;
}
+
+ /* 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);
+
+ /* 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
-expire_rule(struct cls_rule *cls_rule, void *p_)
+eviction_group_add_rule(struct rule *rule)
{
- struct ofproto *p = p_;
- struct rule *rule = rule_from_cls_rule(cls_rule);
- long long int hard_expire, idle_expire, expire, now;
+ struct ofproto *ofproto = rule->ofproto;
+ struct oftable *table = &ofproto->tables[rule->table_id];
- 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);
+ if (table->eviction_fields
+ && (rule->hard_timeout || rule->idle_timeout)) {
+ struct eviction_group *evg;
- 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);
- }
+ evg = eviction_group_find(table, eviction_group_hash_rule(rule));
- return;
+ rule->eviction_group = evg;
+ heap_insert(&evg->rules, &rule->evg_node,
+ rule_eviction_priority(rule));
+ eviction_group_resized(table, evg);
}
+}
+\f
+/* oftables. */
- COVERAGE_INC(ofproto_expired);
- if (rule->cr.wc.wildcards) {
- /* Update stats. (This code will be a no-op if the rule expired
- * due to an idle timeout, because in that case the rule has no
- * subrules left.) */
- struct rule *subrule, *next;
- LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
- rule_remove(p, subrule);
- }
- }
+/* Initializes 'table'. */
+static void
+oftable_init(struct oftable *table)
+{
+ memset(table, 0, sizeof *table);
+ classifier_init(&table->cls);
+}
- send_flow_exp(p, rule, now,
- (now >= hard_expire
- ? OFPER_HARD_TIMEOUT : OFPER_IDLE_TIMEOUT));
- rule_remove(p, rule);
+/* Destroys 'table', including its classifier and eviction groups.
+ *
+ * The caller is responsible for freeing 'table' itself. */
+static void
+oftable_destroy(struct oftable *table)
+{
+ 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
-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_set_name(struct oftable *table, const char *name)
+{
+ 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 {
+ free(table->name);
+ table->name = NULL;
+ }
+}
- 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;
+/* 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
+oftable_disable_eviction(struct oftable *table)
+{
+ if (table->eviction_fields) {
+ struct eviction_group *evg, *next;
- netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
+ HMAP_FOR_EACH_SAFE (evg, next, id_node,
+ &table->eviction_groups_by_id) {
+ eviction_group_destroy(table, evg);
+ }
+ hmap_destroy(&table->eviction_groups_by_id);
+ heap_destroy(&table->eviction_groups_by_size);
- /* Schedule us to send the accumulated records once we have
- * collected all of them. */
- poll_immediate_wake();
+ 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
-update_used(struct ofproto *p)
+oftable_enable_eviction(struct oftable *table,
+ const struct mf_subfield *fields, size_t n_fields)
{
- struct odp_flow *flows;
- size_t n_flows;
- size_t i;
- int error;
+ struct cls_cursor cursor;
+ struct rule *rule;
- error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
- if (error) {
+ if (table->eviction_fields
+ && n_fields == table->n_eviction_fields
+ && (!n_fields
+ || !memcmp(fields, table->eviction_fields,
+ n_fields * sizeof *fields))) {
+ /* No change. */
return;
}
- for (i = 0; i < n_flows; i++) {
- struct odp_flow *f = &flows[i];
- struct rule *rule;
+ oftable_disable_eviction(table);
- 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;
- }
+ table->n_eviction_fields = n_fields;
+ table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
- update_time(p, rule, &f->stats);
- rule_account(p, rule, f->stats.n_bytes);
+ 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);
}
- free(flows);
}
+/* Removes 'rule' from the oftable that contains it. */
static void
-do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
- const struct ofpbuf *packet, int send_len)
+oftable_remove_rule(struct rule *rule)
{
- struct odp_msg *msg = packet->data;
- struct ofpbuf payload;
- struct ofpbuf *opi;
- uint8_t reason;
+ struct ofproto *ofproto = rule->ofproto;
+ struct oftable *table = &ofproto->tables[rule->table_id];
- /* Extract packet payload from 'msg'. */
- payload.data = msg + 1;
- payload.size = msg->length - sizeof *msg;
+ classifier_remove(&table->cls, &rule->cr);
+ eviction_group_remove_rule(rule);
+}
- /* Construct ofp_packet_in message. */
- reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
- opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port), reason,
- &payload, send_len);
+/* 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;
- /* Send. */
- rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100);
+ 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
-send_packet_in_action(struct ofpbuf *packet, void *p_)
+oftable_substitute_rule(struct rule *old, struct rule *new)
+{
+ if (new) {
+ oftable_replace_rule(new);
+ } else {
+ oftable_remove_rule(old);
+ }
+}
+\f
+/* unixctl commands. */
+
+struct ofproto *
+ofproto_lookup(const char *name)
{
- struct ofproto *p = p_;
- struct ofconn *ofconn;
- struct odp_msg *msg;
+ struct ofproto *ofproto;
- msg = packet->data;
- LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
- if (ofconn == p->controller || ofconn->miss_send_len) {
- do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
+ HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
+ &all_ofprotos) {
+ if (!strcmp(ofproto->name, name)) {
+ return ofproto;
}
}
- ofpbuf_delete(packet);
+ return NULL;
}
static void
-send_packet_in_miss(struct ofpbuf *packet, void *p_)
+ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
+ const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
{
- struct ofproto *p = p_;
- bool in_fail_open = p->fail_open && fail_open_is_active(p->fail_open);
- struct ofconn *ofconn;
- struct ofpbuf payload;
- struct odp_msg *msg;
+ struct ofproto *ofproto;
+ struct ds results;
- msg = packet->data;
- payload.data = msg + 1;
- payload.size = msg->length - sizeof *msg;
- LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
- if (ofconn->miss_send_len) {
- struct pktbuf *pb = ofconn->pktbuf;
- uint32_t buffer_id = (in_fail_open
- ? pktbuf_get_null()
- : pktbuf_save(pb, &payload, msg->port));
- int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
- : UINT32_MAX);
- do_send_packet_in(ofconn, buffer_id, packet, send_len);
- }
+ ds_init(&results);
+ HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
+ ds_put_format(&results, "%s\n", ofproto->name);
}
- ofpbuf_delete(packet);
+ unixctl_command_reply(conn, ds_cstr(&results));
+ ds_destroy(&results);
}
-static uint64_t
-pick_datapath_id(const struct ofproto *ofproto)
+static void
+ofproto_unixctl_init(void)
{
- const struct ofport *port;
+ static bool registered;
+ if (registered) {
+ return;
+ }
+ registered = true;
- port = port_array_get(&ofproto->ports, ODPP_LOCAL);
- if (port) {
- uint8_t ea[ETH_ADDR_LEN];
- int error;
+ unixctl_command_register("ofproto/list", "", 0, 0,
+ ofproto_unixctl_list, NULL);
+}
+\f
+/* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
+ *
+ * This is deprecated. It is only for compatibility with broken device drivers
+ * in old versions of Linux that do not properly support VLANs when VLAN
+ * devices are not used. When broken device drivers are no longer in
+ * widespread use, we will delete these interfaces. */
- error = netdev_get_etheraddr(port->netdev, ea);
- if (!error) {
- return eth_addr_to_uint64(ea);
+/* 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;
+
+ free(ofproto->vlan_bitmap);
+ ofproto->vlan_bitmap = bitmap_allocate(4096);
+ ofproto->vlans_changed = false;
+
+ OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
+ const struct cls_table *table;
+
+ 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;
+
+ 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);
+ }
+ }
}
- VLOG_WARN("could not get MAC address for %s (%s)",
- netdev_get_name(port->netdev), strerror(error));
}
- return ofproto->fallback_dpid;
}
-static uint64_t
-pick_fallback_dpid(void)
+/* 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)
{
- uint8_t ea[ETH_ADDR_LEN];
- eth_addr_random(ea);
- ea[0] = 0x00; /* Set Nicira OUI. */
- ea[1] = 0x23;
- ea[2] = 0x20;
- return eth_addr_to_uint64(ea);
+ return ofproto->vlans_changed;
}
-\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);
- 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);
- 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
-};