/*
- * Copyright (c) 2009, 2010, 2011 Nicira Networks.
+ * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
* Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
*
* Licensed under the Apache License, Version 2.0 (the "License");
#include <inttypes.h>
#include <stdbool.h>
#include <stdlib.h>
+#include <unistd.h>
#include "bitmap.h"
#include "byte-order.h"
#include "classifier.h"
#include "dynamic-string.h"
#include "hash.h"
#include "hmap.h"
+#include "meta-flow.h"
#include "netdev.h"
#include "nx-match.h"
+#include "ofp-actions.h"
+#include "ofp-errors.h"
+#include "ofp-msgs.h"
#include "ofp-print.h"
#include "ofp-util.h"
#include "ofpbuf.h"
#include "pinsched.h"
#include "pktbuf.h"
#include "poll-loop.h"
+#include "random.h"
#include "shash.h"
+#include "simap.h"
#include "sset.h"
#include "timeval.h"
#include "unaligned.h"
enum ofproto_state {
S_OPENFLOW, /* Processing OpenFlow commands. */
+ S_EVICT, /* Evicting flows from over-limit tables. */
S_FLUSH, /* Deleting all flow table rules. */
};
enum ofoperation_type {
OFOPERATION_ADD,
OFOPERATION_DELETE,
- OFOPERATION_MODIFY
+ OFOPERATION_MODIFY,
+ OFOPERATION_REPLACE
};
/* A single OpenFlow request can execute any number of operations. The
struct ofproto *ofproto; /* Owning ofproto. */
struct list ofproto_node; /* In ofproto's "pending" list. */
struct list ops; /* List of "struct ofoperation"s. */
+ int n_running; /* Number of ops still pending. */
/* Data needed to send OpenFlow reply on failure or to send a buffered
* packet on success.
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 struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
const struct ofp_header *,
uint32_t buffer_id);
static void ofopgroup_submit(struct ofopgroup *);
-static void ofopgroup_destroy(struct ofopgroup *);
+static void ofopgroup_complete(struct ofopgroup *);
/* A single flow table operation. */
struct ofoperation {
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. */
+
+ /* OFOPERATION_MODIFY, OFOPERATION_REPLACE: The old actions, if the actions
+ * are changing. */
+ struct ofpact *ofpacts;
+ size_t ofpacts_len;
+ uint32_t meter_id;
+
+ /* OFOPERATION_DELETE. */
+ enum ofp_flow_removed_reason reason; /* Reason flow was removed. */
+
ovs_be64 flow_cookie; /* Rule's old flow cookie. */
+ uint16_t idle_timeout; /* Rule's old idle timeout. */
+ uint16_t hard_timeout; /* Rule's old hard timeout. */
+ bool send_flow_removed; /* Rule's old 'send_flow_removed'. */
+ enum ofperr error; /* 0 if no error. */
};
-static void ofoperation_create(struct ofopgroup *, struct rule *,
- enum ofoperation_type);
+static struct ofoperation *ofoperation_create(struct ofopgroup *,
+ struct rule *,
+ enum ofoperation_type,
+ enum ofp_flow_removed_reason);
static void ofoperation_destroy(struct ofoperation *);
-static void ofport_destroy__(struct ofport *);
-static void ofport_destroy(struct ofport *);
+/* oftable. */
+static void oftable_init(struct oftable *);
+static void oftable_destroy(struct oftable *);
-static uint64_t pick_datapath_id(const struct ofproto *);
-static uint64_t pick_fallback_dpid(void);
+static void oftable_set_name(struct oftable *, const char *name);
-static void ofproto_destroy__(struct ofproto *);
+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 ofproto_rule_destroy__(struct rule *);
-static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
+static void oftable_remove_rule(struct rule *rule) OVS_RELEASES(rule->rwlock);
+static void oftable_remove_rule__(struct ofproto *ofproto,
+ struct classifier *cls, struct rule *rule)
+ OVS_REQ_WRLOCK(cls->rwlock) OVS_RELEASES(rule->rwlock);
+static void oftable_insert_rule(struct rule *);
-static void ofopgroup_destroy(struct ofopgroup *);
+/* 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 int add_flow(struct ofproto *, struct ofconn *,
- const struct ofputil_flow_mod *,
- const struct ofp_header *);
+static bool choose_rule_to_evict(struct oftable *table, struct rule **rulep)
+ OVS_TRY_WRLOCK(true, (*rulep)->rwlock);
+static void ofproto_evict(struct ofproto *);
+static uint32_t rule_eviction_priority(struct rule *);
+static void eviction_group_add_rule(struct rule *);
+static void eviction_group_remove_rule(struct rule *);
-static bool handle_openflow(struct ofconn *, struct ofpbuf *);
-static int handle_flow_mod__(struct ofproto *, struct ofconn *,
- const struct ofputil_flow_mod *,
- const struct ofp_header *);
+/* 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 *);
-static void set_internal_devs_mtu(struct ofproto *);
+/* rule. */
+static void ofproto_rule_destroy(struct 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 *);
+
+/* OpenFlow. */
+static enum ofperr add_flow(struct ofproto *, struct ofconn *,
+ struct ofputil_flow_mod *,
+ const struct ofp_header *);
+static enum ofperr modify_flows__(struct ofproto *, struct ofconn *,
+ struct ofputil_flow_mod *,
+ const struct ofp_header *, struct list *);
+static void delete_flow__(struct rule *rule, struct ofopgroup *,
+ enum ofp_flow_removed_reason)
+ OVS_RELEASES(rule->rwlock);
+static enum ofperr add_group(struct ofproto *, struct ofputil_group_mod *);
+static bool handle_openflow(struct ofconn *, const struct ofpbuf *);
+static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
+ struct ofputil_flow_mod *,
+ const struct ofp_header *);
+static void calc_duration(long long int start, long long int now,
+ uint32_t *sec, uint32_t *nsec);
+
+/* 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 update_mtu(struct ofproto *, struct ofport *);
+static void meter_delete(struct ofproto *, uint32_t first, uint32_t last);
+
+/* unixctl. */
static void ofproto_unixctl_init(void);
/* All registered ofproto classes, in probe order. */
static size_t n_ofproto_classes;
static size_t allocated_ofproto_classes;
+unsigned flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT;
+unsigned n_handler_threads;
+enum ofproto_flow_miss_model flow_miss_model = OFPROTO_HANDLE_MISS_AUTO;
+
/* Map from datapath name to struct ofproto, for use by unixctl commands. */
static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
+/* Initial mappings of port to OpenFlow number mappings. */
+static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
+
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
-static void
-ofproto_initialize(void)
+/* The default value of true waits for flow restore. */
+static bool flow_restore_wait = true;
+
+/* Must be called to initialize the ofproto library.
+ *
+ * The caller may pass in 'iface_hints', which contains an shash of
+ * "iface_hint" elements indexed by the interface's name. The provider
+ * may use these hints to describe the startup configuration in order to
+ * reinitialize its state. The caller owns the provided data, so a
+ * provider will make copies of anything required. An ofproto provider
+ * will remove any existing state that is not described by the hint, and
+ * may choose to remove it all. */
+void
+ofproto_init(const struct shash *iface_hints)
{
- static bool inited;
+ struct shash_node *node;
+ size_t i;
+
+ ofproto_class_register(&ofproto_dpif_class);
+
+ /* Make a local copy, since we don't own 'iface_hints' elements. */
+ SHASH_FOR_EACH(node, iface_hints) {
+ const struct iface_hint *orig_hint = node->data;
+ struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
+ const char *br_type = ofproto_normalize_type(orig_hint->br_type);
+
+ new_hint->br_name = xstrdup(orig_hint->br_name);
+ new_hint->br_type = xstrdup(br_type);
+ new_hint->ofp_port = orig_hint->ofp_port;
- if (!inited) {
- inited = true;
- ofproto_class_register(&ofproto_dpif_class);
+ shash_add(&init_ofp_ports, node->name, new_hint);
+ }
+
+ for (i = 0; i < n_ofproto_classes; i++) {
+ ofproto_classes[i]->init(&init_ofp_ports);
}
}
{
size_t i;
- ofproto_initialize();
for (i = 0; i < n_ofproto_classes; i++) {
const struct ofproto_class *class = ofproto_classes[i];
struct sset types;
{
size_t i;
- ofproto_initialize();
for (i = 0; i < n_ofproto_classes; i++) {
ofproto_classes[i]->enumerate_types(types);
}
struct ofproto **ofprotop)
{
const struct ofproto_class *class;
- struct classifier *table;
struct ofproto *ofproto;
- int n_tables;
int error;
+ int i;
*ofprotop = NULL;
- ofproto_initialize();
ofproto_unixctl_init();
datapath_type = ofproto_normalize_type(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->mfr_desc = NULL;
+ ofproto->hw_desc = NULL;
+ ofproto->sw_desc = NULL;
+ ofproto->serial_desc = NULL;
+ ofproto->dp_desc = NULL;
ofproto->frag_handling = OFPC_FRAG_NORMAL;
hmap_init(&ofproto->ports);
shash_init(&ofproto->port_by_name);
+ simap_init(&ofproto->ofp_requests);
+ ofproto->max_ports = ofp_to_u16(OFPP_MAX);
+ ofproto->eviction_group_timer = LLONG_MIN;
ofproto->tables = NULL;
ofproto->n_tables = 0;
+ hindex_init(&ofproto->cookies);
+ list_init(&ofproto->expirable);
+ ovs_mutex_init_recursive(&ofproto->expirable_mutex);
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->n_add = ofproto->n_delete = ofproto->n_modify = 0;
+ ofproto->first_op = ofproto->last_op = LLONG_MIN;
+ ofproto->next_op_report = LLONG_MAX;
+ ofproto->op_backoff = LLONG_MIN;
ofproto->vlan_bitmap = NULL;
ofproto->vlans_changed = false;
+ ofproto->min_mtu = INT_MAX;
+ ovs_rwlock_init(&ofproto->groups_rwlock);
+ hmap_init(&ofproto->groups);
- error = ofproto->ofproto_class->construct(ofproto, &n_tables);
+ error = ofproto->ofproto_class->construct(ofproto);
if (error) {
VLOG_ERR("failed to open datapath %s: %s",
- datapath_name, strerror(error));
+ datapath_name, ovs_strerror(error));
ofproto_destroy__(ofproto);
return error;
}
- 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) {
- classifier_init(table);
+ /* The "max_ports" member should have been set by ->construct(ofproto).
+ * Port 0 is not a valid OpenFlow port, so mark that as unavailable. */
+ ofproto->ofp_port_ids = bitmap_allocate(ofproto->max_ports);
+ bitmap_set1(ofproto->ofp_port_ids, 0);
+
+ /* Check that hidden tables, if any, are at the end. */
+ ovs_assert(ofproto->n_tables);
+ for (i = 0; i + 1 < ofproto->n_tables; i++) {
+ enum oftable_flags flags = ofproto->tables[i].flags;
+ enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
+
+ ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
}
ofproto->datapath_id = pick_datapath_id(ofproto);
- VLOG_INFO("using datapath ID %016"PRIx64, ofproto->datapath_id);
init_ports(ofproto);
+ /* Initialize meters table. */
+ if (ofproto->ofproto_class->meter_get_features) {
+ ofproto->ofproto_class->meter_get_features(ofproto,
+ &ofproto->meter_features);
+ } else {
+ memset(&ofproto->meter_features, 0, sizeof ofproto->meter_features);
+ }
+ ofproto->meters = xzalloc((ofproto->meter_features.max_meters + 1)
+ * sizeof(struct meter *));
+
*ofprotop = ofproto;
return 0;
}
+/* Must be called (only) by an ofproto implementation in its constructor
+ * function. See the large comment on 'construct' in struct ofproto_class for
+ * details. */
+void
+ofproto_init_tables(struct ofproto *ofproto, int n_tables)
+{
+ struct oftable *table;
+
+ ovs_assert(!ofproto->n_tables);
+ ovs_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);
+ }
+}
+
+/* To be optionally called (only) by an ofproto implementation in its
+ * constructor function. See the large comment on 'construct' in struct
+ * ofproto_class for details.
+ *
+ * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
+ * will then ensure that actions passed into the ofproto implementation will
+ * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
+ * function is not called, there will be no such restriction.
+ *
+ * Reserved ports numbered OFPP_MAX and higher are special and not subject to
+ * the 'max_ports' restriction. */
+void
+ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
+{
+ ovs_assert(max_ports <= ofp_to_u16(OFPP_MAX));
+ ofproto->max_ports = max_ports;
+}
+
+uint64_t
+ofproto_get_datapath_id(const struct ofproto *ofproto)
+{
+ return ofproto->datapath_id;
+}
+
void
ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
{
uint64_t old_dpid = p->datapath_id;
p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
if (p->datapath_id != old_dpid) {
- VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
-
/* 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_controllers(struct ofproto *p,
const struct ofproto_controller *controllers,
- size_t n_controllers)
+ size_t n_controllers, uint32_t allowed_versions)
{
- connmgr_set_controllers(p->connmgr, controllers, n_controllers);
+ connmgr_set_controllers(p->connmgr, controllers, n_controllers,
+ allowed_versions);
}
void
/* 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)
+ofproto_set_flow_eviction_threshold(unsigned threshold)
{
- if (threshold < OFPROTO_FLOW_EVICTION_THRESHOLD_MIN) {
- ofproto->flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_MIN;
- } else {
- ofproto->flow_eviction_threshold = threshold;
- }
+ flow_eviction_threshold = MAX(OFPROTO_FLOW_EVICTION_THRESHOLD_MIN,
+ threshold);
+}
+
+/* Sets the path for handling flow misses. */
+void
+ofproto_set_flow_miss_model(unsigned model)
+{
+ flow_miss_model = model;
}
/* If forward_bpdu is true, the NORMAL action will forward frames with
}
}
+/* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
+ * 'idle_time', in seconds, and the maximum number of MAC table entries to
+ * 'max_entries'. */
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)
+ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
+ size_t max_entries)
{
- struct ofp_desc_stats *ods;
-
- 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 (ofproto->ofproto_class->set_mac_table_config) {
+ ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
+ max_entries);
}
- 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);
+}
+
+/* Sets number of upcall handler threads. The default is
+ * (number of online cores - 2). */
+void
+ofproto_set_n_handler_threads(unsigned limit)
+{
+ if (limit) {
+ n_handler_threads = limit;
+ } else {
+ int n_proc = sysconf(_SC_NPROCESSORS_ONLN);
+ n_handler_threads = n_proc > 2 ? n_proc - 2 : 1;
}
}
+void
+ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
+{
+ free(p->dp_desc);
+ p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
+}
+
int
ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
{
return oso ? EOPNOTSUPP : 0;
}
}
+
+int
+ofproto_set_ipfix(struct ofproto *ofproto,
+ const struct ofproto_ipfix_bridge_exporter_options *bo,
+ const struct ofproto_ipfix_flow_exporter_options *fo,
+ size_t n_fo)
+{
+ if (ofproto->ofproto_class->set_ipfix) {
+ return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
+ } else {
+ return (bo || fo) ? EOPNOTSUPP : 0;
+ }
+}
+
+void
+ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
+{
+ flow_restore_wait = flow_restore_wait_db;
+}
+
+bool
+ofproto_get_flow_restore_wait(void)
+{
+ return flow_restore_wait;
+}
+
\f
/* Spanning Tree Protocol (STP) configuration. */
*
* Returns 0 if successful, otherwise a positive errno value.*/
int
-ofproto_port_set_stp(struct ofproto *ofproto, uint16_t ofp_port,
+ofproto_port_set_stp(struct ofproto *ofproto, ofp_port_t ofp_port,
const struct ofproto_port_stp_settings *s)
{
struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
*
* Returns 0 if successful, otherwise a positive errno value.*/
int
-ofproto_port_get_stp_status(struct ofproto *ofproto, uint16_t ofp_port,
+ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
struct ofproto_port_stp_status *s)
{
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);
+ VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
+ "port %"PRIu16, ofproto->name, ofp_port);
return ENODEV;
}
*
* Returns 0 if successful, otherwise a positive errno value. */
int
-ofproto_port_set_queues(struct ofproto *ofproto, uint16_t ofp_port,
+ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port,
const struct ofproto_port_queue *queues,
size_t n_queues)
{
/* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
void
-ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
+ofproto_port_clear_cfm(struct ofproto *ofproto, ofp_port_t ofp_port)
{
struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
if (ofport && ofproto->ofproto_class->set_cfm) {
*
* This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
void
-ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
+ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port,
const struct cfm_settings *s)
{
struct ofport *ofport;
if (error) {
VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
- strerror(error));
+ ovs_strerror(error));
+ }
+}
+
+/* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
+ * 'ofproto' does not have a port 'ofp_port'. */
+void
+ofproto_port_set_bfd(struct ofproto *ofproto, ofp_port_t ofp_port,
+ const struct smap *cfg)
+{
+ struct ofport *ofport;
+ int error;
+
+ ofport = ofproto_get_port(ofproto, ofp_port);
+ if (!ofport) {
+ VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
+ ofproto->name, ofp_port);
+ return;
+ }
+
+ error = (ofproto->ofproto_class->set_bfd
+ ? ofproto->ofproto_class->set_bfd(ofport, cfg)
+ : EOPNOTSUPP);
+ if (error) {
+ VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
+ ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
+ ovs_strerror(error));
}
}
+/* Populates 'status' with key value pairs indicating the status of the BFD
+ * session on 'ofp_port'. This information is intended to be populated in the
+ * OVS database. Has no effect if 'ofp_port' is not na OpenFlow port in
+ * 'ofproto'. */
+int
+ofproto_port_get_bfd_status(struct ofproto *ofproto, ofp_port_t ofp_port,
+ struct smap *status)
+{
+ struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
+ return (ofport && ofproto->ofproto_class->get_bfd_status
+ ? ofproto->ofproto_class->get_bfd_status(ofport, status)
+ : EOPNOTSUPP);
+}
+
/* 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)
+ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port)
{
struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
return (ofport && ofproto->ofproto_class->port_is_lacp_current
: false);
}
\f
+/* Configuration of OpenFlow tables. */
+
+/* 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;
+
+ ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
+ table = &ofproto->tables[table_id];
+
+ oftable_set_name(table, s->name);
+
+ if (table->flags & OFTABLE_READONLY) {
+ return;
+ }
+
+ if (s->groups) {
+ oftable_enable_eviction(table, s->groups, s->n_groups);
+ } else {
+ oftable_disable_eviction(table);
+ }
+
+ table->max_flows = s->max_flows;
+ ovs_rwlock_rdlock(&table->cls.rwlock);
+ 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;
+ }
+ }
+ ovs_rwlock_unlock(&table->cls.rwlock);
+}
+\f
bool
ofproto_has_snoops(const struct ofproto *ofproto)
{
connmgr_get_snoops(ofproto->connmgr, snoops);
}
+/* Deletes 'rule' from 'cls' within 'ofproto'.
+ *
+ * The 'cls' argument is redundant (it is &ofproto->tables[rule->table_id].cls)
+ * but it allows Clang to do better checking. */
static void
-ofproto_flush__(struct ofproto *ofproto)
+ofproto_delete_rule(struct ofproto *ofproto, struct classifier *cls,
+ struct rule *rule)
+ OVS_REQ_WRLOCK(cls->rwlock)
{
- struct classifier *table;
struct ofopgroup *group;
+ ovs_assert(!rule->pending);
+ ovs_assert(cls == &ofproto->tables[rule->table_id].cls);
+
+ group = ofopgroup_create_unattached(ofproto);
+ ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
+ ovs_rwlock_wrlock(&rule->rwlock);
+ oftable_remove_rule__(ofproto, cls, rule);
+ ofproto->ofproto_class->rule_delete(rule);
+ ofopgroup_submit(group);
+}
+
+static void
+ofproto_flush__(struct ofproto *ofproto)
+{
+ struct oftable *table;
+
if (ofproto->ofproto_class->flush) {
ofproto->ofproto_class->flush(ofproto);
}
- group = ofopgroup_create_unattached(ofproto);
OFPROTO_FOR_EACH_TABLE (table, ofproto) {
struct rule *rule, *next_rule;
struct cls_cursor cursor;
- cls_cursor_init(&cursor, table, NULL);
+ if (table->flags & OFTABLE_HIDDEN) {
+ continue;
+ }
+
+ ovs_rwlock_wrlock(&table->cls.rwlock);
+ 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);
- classifier_remove(table, &rule->cr);
- ofproto->ofproto_class->rule_destruct(rule);
+ ofproto_delete_rule(ofproto, &table->cls, rule);
}
}
+ ovs_rwlock_unlock(&table->cls.rwlock);
}
- ofopgroup_submit(group);
}
+static void delete_group(struct ofproto *ofproto, uint32_t group_id);
+
static void
ofproto_destroy__(struct ofproto *ofproto)
{
- struct classifier *table;
+ struct oftable *table;
- assert(list_is_empty(&ofproto->pending));
- assert(!ofproto->n_pending);
+ ovs_assert(list_is_empty(&ofproto->pending));
+ ovs_assert(!ofproto->n_pending);
+
+ if (ofproto->meters) {
+ meter_delete(ofproto, 1, ofproto->meter_features.max_meters);
+ free(ofproto->meters);
+ }
+
+ delete_group(ofproto, OFPG_ALL);
+ ovs_rwlock_destroy(&ofproto->groups_rwlock);
+ hmap_destroy(&ofproto->groups);
connmgr_destroy(ofproto->connmgr);
free(ofproto->dp_desc);
hmap_destroy(&ofproto->ports);
shash_destroy(&ofproto->port_by_name);
+ bitmap_free(ofproto->ofp_port_ids);
+ simap_destroy(&ofproto->ofp_requests);
OFPROTO_FOR_EACH_TABLE (table, ofproto) {
- assert(classifier_is_empty(table));
- classifier_destroy(table);
+ oftable_destroy(table);
}
free(ofproto->tables);
free(ofproto->vlan_bitmap);
+ ovs_mutex_destroy(&ofproto->expirable_mutex);
ofproto->ofproto_class->dealloc(ofproto);
}
}
}
+int
+ofproto_type_run(const char *datapath_type)
+{
+ const struct ofproto_class *class;
+ int error;
+
+ datapath_type = ofproto_normalize_type(datapath_type);
+ class = ofproto_class_find__(datapath_type);
+
+ error = class->type_run ? class->type_run(datapath_type) : 0;
+ if (error && error != EAGAIN) {
+ VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
+ datapath_type, ovs_strerror(error));
+ }
+ return error;
+}
+
+int
+ofproto_type_run_fast(const char *datapath_type)
+{
+ const struct ofproto_class *class;
+ int error;
+
+ datapath_type = ofproto_normalize_type(datapath_type);
+ class = ofproto_class_find__(datapath_type);
+
+ error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
+ if (error && error != EAGAIN) {
+ VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
+ datapath_type, ovs_strerror(error));
+ }
+ return error;
+}
+
+void
+ofproto_type_wait(const char *datapath_type)
+{
+ const struct ofproto_class *class;
+
+ datapath_type = ofproto_normalize_type(datapath_type);
+ class = ofproto_class_find__(datapath_type);
+
+ if (class->type_wait) {
+ class->type_wait(datapath_type);
+ }
+}
+
int
ofproto_run(struct ofproto *p)
{
+ struct sset changed_netdevs;
+ const char *changed_netdev;
struct ofport *ofport;
- char *devname;
int error;
error = p->ofproto_class->run(p);
if (error && error != EAGAIN) {
- VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, strerror(error));
+ VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, ovs_strerror(error));
+ }
+
+ /* Restore the eviction group heap invariant occasionally. */
+ if (p->eviction_group_timer < time_msec()) {
+ size_t i;
+
+ p->eviction_group_timer = time_msec() + 1000;
+
+ for (i = 0; i < p->n_tables; i++) {
+ struct oftable *table = &p->tables[i];
+ struct eviction_group *evg;
+ struct cls_cursor cursor;
+ struct rule *rule;
+
+ if (!table->eviction_fields) {
+ continue;
+ }
+
+ HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
+ heap_rebuild(&evg->rules);
+ }
+
+ ovs_rwlock_rdlock(&table->cls.rwlock);
+ cls_cursor_init(&cursor, &table->cls, NULL);
+ CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
+ if (!rule->eviction_group
+ && (rule->idle_timeout || rule->hard_timeout)) {
+ eviction_group_add_rule(rule);
+ }
+ }
+ ovs_rwlock_unlock(&table->cls.rwlock);
+ }
}
if (p->ofproto_class->port_poll) {
+ char *devname;
+
while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
process_port_change(p, error, devname);
}
}
+ /* Update OpenFlow port status for any port whose netdev has changed.
+ *
+ * Refreshing a given 'ofport' can cause an arbitrary ofport to be
+ * destroyed, so it's not safe to update ports directly from the
+ * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
+ * need this two-phase approach. */
+ sset_init(&changed_netdevs);
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));
+ sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
}
}
-
+ SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
+ update_port(p, changed_netdev);
+ }
+ sset_destroy(&changed_netdevs);
switch (p->state) {
case S_OPENFLOW:
connmgr_run(p->connmgr, handle_openflow);
break;
+ 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;
+
case S_FLUSH:
connmgr_run(p->connmgr, NULL);
ofproto_flush__(p);
NOT_REACHED();
}
+ if (time_msec() >= p->next_op_report) {
+ long long int ago = (time_msec() - p->first_op) / 1000;
+ long long int interval = (p->last_op - p->first_op) / 1000;
+ struct ds s;
+
+ ds_init(&s);
+ ds_put_format(&s, "%d flow_mods ",
+ p->n_add + p->n_delete + p->n_modify);
+ if (interval == ago) {
+ ds_put_format(&s, "in the last %lld s", ago);
+ } else if (interval) {
+ ds_put_format(&s, "in the %lld s starting %lld s ago",
+ interval, ago);
+ } else {
+ ds_put_format(&s, "%lld s ago", ago);
+ }
+
+ ds_put_cstr(&s, " (");
+ if (p->n_add) {
+ ds_put_format(&s, "%d adds, ", p->n_add);
+ }
+ if (p->n_delete) {
+ ds_put_format(&s, "%d deletes, ", p->n_delete);
+ }
+ if (p->n_modify) {
+ ds_put_format(&s, "%d modifications, ", p->n_modify);
+ }
+ s.length -= 2;
+ ds_put_char(&s, ')');
+
+ VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
+ ds_destroy(&s);
+
+ p->n_add = p->n_delete = p->n_modify = 0;
+ p->next_op_report = LLONG_MAX;
+ }
+
return error;
}
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));
+ p->name, ovs_strerror(error));
}
return error;
}
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)) {
return connmgr_has_controllers(p->connmgr);
}
+/* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
+ * memory_report(). */
+void
+ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
+{
+ const struct oftable *table;
+ unsigned int n_rules;
+
+ simap_increase(usage, "ports", hmap_count(&ofproto->ports));
+ simap_increase(usage, "ops",
+ ofproto->n_pending + hmap_count(&ofproto->deletions));
+
+ n_rules = 0;
+ OFPROTO_FOR_EACH_TABLE (table, ofproto) {
+ ovs_rwlock_rdlock(&table->cls.rwlock);
+ n_rules += classifier_count(&table->cls);
+ ovs_rwlock_unlock(&table->cls.rwlock);
+ }
+ simap_increase(usage, "rules", n_rules);
+
+ if (ofproto->ofproto_class->get_memory_usage) {
+ ofproto->ofproto_class->get_memory_usage(ofproto, usage);
+ }
+
+ connmgr_get_memory_usage(ofproto->connmgr, usage);
+}
+
void
ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
struct shash *info)
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_port_add(struct ofproto *ofproto, struct netdev *netdev,
- uint16_t *ofp_portp)
+/* Returns the type to pass to netdev_open() when a datapath of type
+ * 'datapath_type' has a port of type 'port_type', for a few special
+ * cases when a netdev type differs from a port type. For example, when
+ * using the userspace datapath, a port of type "internal" needs to be
+ * opened as "tap".
+ *
+ * Returns either 'type' itself or a string literal, which must not be
+ * freed. */
+const char *
+ofproto_port_open_type(const char *datapath_type, const char *port_type)
{
- uint16_t ofp_port;
- int error;
+ const struct ofproto_class *class;
- 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;
+ datapath_type = ofproto_normalize_type(datapath_type);
+ class = ofproto_class_find__(datapath_type);
+ if (!class) {
+ return port_type;
+ }
+
+ return (class->port_open_type
+ ? class->port_open_type(datapath_type, port_type)
+ : port_type);
+}
+
+/* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
+ * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
+ * the port's OpenFlow port number.
+ *
+ * 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_port_add(struct ofproto *ofproto, struct netdev *netdev,
+ ofp_port_t *ofp_portp)
+{
+ ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
+ int error;
+
+ error = ofproto->ofproto_class->port_add(ofproto, netdev);
+ if (!error) {
+ const char *netdev_name = netdev_get_name(netdev);
+
+ simap_put(&ofproto->ofp_requests, netdev_name,
+ ofp_to_u16(ofp_port));
+ update_port(ofproto, netdev_name);
+ }
+ if (ofp_portp) {
+ struct ofproto_port ofproto_port;
+
+ ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
+ &ofproto_port);
+ *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
+ ofproto_port_destroy(&ofproto_port);
}
return error;
}
/* 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)
+ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port)
{
struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
+ struct simap_node *ofp_request_node;
int error;
+ ofp_request_node = simap_find(&ofproto->ofp_requests, name);
+ if (ofp_request_node) {
+ simap_delete(&ofproto->ofp_requests, ofp_request_node);
+ }
+
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
* (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'.
+ * The caller retains ownership of 'cls_rule' and 'ofpacts'.
*
* This is a helper function for in-band control and fail-open. */
void
-ofproto_add_flow(struct ofproto *ofproto, const struct cls_rule *cls_rule,
- const union ofp_action *actions, size_t n_actions)
+ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
+ unsigned int priority,
+ const struct ofpact *ofpacts, size_t ofpacts_len)
{
const struct rule *rule;
- rule = rule_from_cls_rule(classifier_find_rule_exactly(
- &ofproto->tables[0], cls_rule));
- if (!rule || !ofputil_actions_equal(rule->actions, rule->n_actions,
- actions, n_actions)) {
+ ovs_rwlock_rdlock(&ofproto->tables[0].cls.rwlock);
+ rule = rule_from_cls_rule(classifier_find_match_exactly(
+ &ofproto->tables[0].cls, match, priority));
+ ovs_rwlock_unlock(&ofproto->tables[0].cls.rwlock);
+ if (!rule || !ofpacts_equal(rule->ofpacts, rule->ofpacts_len,
+ ofpacts, ofpacts_len)) {
struct ofputil_flow_mod fm;
memset(&fm, 0, sizeof fm);
- fm.cr = *cls_rule;
+ fm.match = *match;
+ fm.priority = priority;
fm.buffer_id = UINT32_MAX;
- fm.actions = (union ofp_action *) actions;
- fm.n_actions = n_actions;
+ fm.ofpacts = xmemdup(ofpacts, ofpacts_len);
+ fm.ofpacts_len = ofpacts_len;
add_flow(ofproto, NULL, &fm, NULL);
+ free(fm.ofpacts);
}
}
/* Executes the flow modification specified in 'fm'. Returns 0 on success, an
- * OpenFlow error code as encoded by ofp_mkerr() on failure, or
- * OFPROTO_POSTPONE if the operation cannot be initiated now but may be retried
- * later.
+ * 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)
+ofproto_flow_mod(struct ofproto *ofproto, struct ofputil_flow_mod *fm)
{
return handle_flow_mod__(ofproto, NULL, fm, NULL);
}
*
* This is a helper function for in-band control and fail-open. */
bool
-ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
+ofproto_delete_flow(struct ofproto *ofproto,
+ const struct match *target, unsigned int priority)
{
+ struct classifier *cls = &ofproto->tables[0].cls;
struct rule *rule;
- rule = rule_from_cls_rule(classifier_find_rule_exactly(
- &ofproto->tables[0], target));
+ ovs_rwlock_rdlock(&cls->rwlock);
+ rule = rule_from_cls_rule(classifier_find_match_exactly(cls, target,
+ priority));
+ ovs_rwlock_unlock(&cls->rwlock);
if (!rule) {
/* No such rule -> success. */
return true;
return false;
} else {
/* Initiate deletion -> success. */
- struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
- ofoperation_create(group, rule, OFOPERATION_DELETE);
- classifier_remove(&ofproto->tables[rule->table_id], &rule->cr);
- rule->ofproto->ofproto_class->rule_destruct(rule);
- ofopgroup_submit(group);
+ ovs_rwlock_wrlock(&cls->rwlock);
+ ofproto_delete_rule(ofproto, cls, rule);
+ ovs_rwlock_unlock(&cls->rwlock);
+
return true;
}
sset_destroy(&devnames);
}
-/* 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 ofp_port_t
+alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
+{
+ uint16_t port_idx;
+
+ port_idx = simap_get(&ofproto->ofp_requests, netdev_name);
+ port_idx = port_idx ? port_idx : UINT16_MAX;
+
+ if (port_idx >= ofproto->max_ports
+ || bitmap_is_set(ofproto->ofp_port_ids, port_idx)) {
+ uint16_t end_port_no = ofproto->alloc_port_no;
+
+ /* Search for a free OpenFlow port number. We try not to
+ * immediately reuse them to prevent problems due to old
+ * flows. */
+ for (;;) {
+ if (++ofproto->alloc_port_no >= ofproto->max_ports) {
+ ofproto->alloc_port_no = 0;
+ }
+ if (!bitmap_is_set(ofproto->ofp_port_ids,
+ ofproto->alloc_port_no)) {
+ port_idx = ofproto->alloc_port_no;
+ break;
+ }
+ if (ofproto->alloc_port_no == end_port_no) {
+ return OFPP_NONE;
+ }
+ }
+ }
+ bitmap_set1(ofproto->ofp_port_ids, port_idx);
+ return u16_to_ofp(port_idx);
+}
+
+static void
+dealloc_ofp_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
+{
+ if (ofp_to_u16(ofp_port) < ofproto->max_ports) {
+ bitmap_set0(ofproto->ofp_port_ids, ofp_to_u16(ofp_port));
+ }
+}
+
+/* Opens and returns a netdev for 'ofproto_port' in 'ofproto', 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 ofp_phy_port *opp)
+ofport_open(struct ofproto *ofproto,
+ struct ofproto_port *ofproto_port,
+ struct ofputil_phy_port *pp)
{
- uint32_t curr, advertised, supported, peer;
enum netdev_flags flags;
struct netdev *netdev;
int error;
error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
if (error) {
- VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
+ VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
"cannot be opened (%s)",
+ ofproto->name,
ofproto_port->name, ofproto_port->ofp_port,
- ofproto_port->name, strerror(error));
+ ofproto_port->name, ovs_strerror(error));
return NULL;
}
+ if (ofproto_port->ofp_port == OFPP_NONE) {
+ if (!strcmp(ofproto->name, ofproto_port->name)) {
+ ofproto_port->ofp_port = OFPP_LOCAL;
+ } else {
+ ofproto_port->ofp_port = alloc_ofp_port(ofproto,
+ ofproto_port->name);
+ }
+ }
+ 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);
- netdev_get_features(netdev, &curr, &advertised, &supported, &peer);
-
- opp->port_no = htons(ofproto_port->ofp_port);
- netdev_get_etheraddr(netdev, opp->hw_addr);
- ovs_strzcpy(opp->name, ofproto_port->name, sizeof opp->name);
- opp->config = flags & NETDEV_UP ? 0 : htonl(OFPPC_PORT_DOWN);
- opp->state = netdev_get_carrier(netdev) ? 0 : htonl(OFPPS_LINK_DOWN);
- opp->curr = htonl(curr);
- opp->advertised = htonl(advertised);
- opp->supported = htonl(supported);
- opp->peer = htonl(peer);
+ 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, 0) / 1000;
+ pp->max_speed = netdev_features_to_bps(pp->supported, 0) / 1000;
return netdev;
}
/* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
- * port number, and 'config' bits other than OFPPC_PORT_DOWN are
+ * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
* disregarded. */
static bool
-ofport_equal(const struct ofp_phy_port *a, const struct ofp_phy_port *b)
+ofport_equal(const struct ofputil_phy_port *a,
+ const struct ofputil_phy_port *b)
{
- BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
- return (!memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
+ return (eth_addr_equals(a->hw_addr, b->hw_addr)
&& a->state == b->state
- && !((a->config ^ b->config) & htonl(OFPPC_PORT_DOWN))
+ && !((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'.
* one with the same name or port number). */
static void
ofport_install(struct ofproto *p,
- struct netdev *netdev, const struct ofp_phy_port *opp)
+ struct netdev *netdev, const struct ofputil_phy_port *pp)
{
const char *netdev_name = netdev_get_name(netdev);
struct ofport *ofport;
- int dev_mtu;
int error;
/* Create ofport. */
ofport->ofproto = p;
ofport->netdev = netdev;
ofport->change_seq = netdev_change_seq(netdev);
- ofport->opp = *opp;
- ofport->ofp_port = ntohs(opp->port_no);
+ ofport->pp = *pp;
+ ofport->ofp_port = pp->port_no;
+ ofport->created = time_msec();
/* Add port to 'p'. */
- hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
+ hmap_insert(&p->ports, &ofport->hmap_node,
+ hash_ofp_port(ofport->ofp_port));
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;
- }
+ update_mtu(p, ofport);
/* Let the ofproto_class initialize its private data. */
error = p->ofproto_class->port_construct(ofport);
if (error) {
goto error;
}
- connmgr_send_port_status(p->connmgr, opp, OFPPR_ADD);
+ 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));
+ p->name, netdev_name, ovs_strerror(error));
if (ofport) {
ofport_destroy__(ofport);
} else {
static void
ofport_remove(struct ofport *ofport)
{
- connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->opp,
+ connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
OFPPR_DELETE);
ofport_destroy(ofport);
}
}
}
-/* Updates 'port' with new 'opp' description.
+/* 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_modified(struct ofport *port, struct ofp_phy_port *opp)
+ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
{
- memcpy(port->opp.hw_addr, opp->hw_addr, ETH_ADDR_LEN);
- port->opp.config = ((port->opp.config & ~htonl(OFPPC_PORT_DOWN))
- | (opp->config & htonl(OFPPC_PORT_DOWN)));
- port->opp.state = opp->state;
- port->opp.curr = opp->curr;
- port->opp.advertised = opp->advertised;
- port->opp.supported = opp->supported;
- port->opp.peer = opp->peer;
+ 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->opp, OFPPR_MODIFY);
+ 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, ovs_be32 state)
+ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
{
- if (port->opp.state != state) {
- port->opp.state = state;
- connmgr_send_port_status(port->ofproto->connmgr, &port->opp,
+ if (port->pp.state != state) {
+ port->pp.state = state;
+ connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
OFPPR_MODIFY);
}
}
void
-ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
+ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port)
{
struct ofport *port = ofproto_get_port(ofproto, ofp_port);
if (port) {
ofport_destroy(struct ofport *port)
{
if (port) {
+ dealloc_ofp_port(port->ofproto, port->ofp_port);
port->ofproto->ofproto_class->port_destruct(port);
ofport_destroy__(port);
}
}
struct ofport *
-ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
+ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
{
struct ofport *port;
- HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
- hash_int(ofp_port, 0), &ofproto->ports) {
+ HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_ofp_port(ofp_port),
+ &ofproto->ports) {
if (port->ofp_port == ofp_port) {
return port;
}
update_port(struct ofproto *ofproto, const char *name)
{
struct ofproto_port ofproto_port;
- struct ofp_phy_port opp;
+ struct ofputil_phy_port pp;
struct netdev *netdev;
struct ofport *port;
/* Fetch 'name''s location and properties from the datapath. */
netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
- ? ofport_open(&ofproto_port, &opp)
+ ? ofport_open(ofproto, &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->opp, &opp)) {
- ofport_modified(port, &opp);
+ if (!ofport_equal(&port->pp, &pp)) {
+ ofport_modified(port, &pp);
}
- /* 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;
- }
+ update_mtu(ofproto, port);
/* Install the newly opened netdev in case it has changed.
* Don't close the old netdev yet in case port_modified has to
ofport_remove(port);
}
ofport_remove_with_name(ofproto, name);
- ofport_install(ofproto, netdev, &opp);
+ ofport_install(ofproto, netdev, &pp);
}
} else {
/* Any port named 'name' is gone now. */
{
struct ofproto_port_dump dump;
struct ofproto_port ofproto_port;
+ struct shash_node *node, *next;
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);
+ const char *name = ofproto_port.name;
+
+ if (shash_find(&p->port_by_name, name)) {
+ VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
+ p->name, name);
} else {
- struct ofp_phy_port opp;
+ struct ofputil_phy_port pp;
struct netdev *netdev;
- netdev = ofport_open(&ofproto_port, &opp);
+ /* Check if an OpenFlow port number had been requested. */
+ node = shash_find(&init_ofp_ports, name);
+ if (node) {
+ const struct iface_hint *iface_hint = node->data;
+ simap_put(&p->ofp_requests, name,
+ ofp_to_u16(iface_hint->ofp_port));
+ }
+
+ netdev = ofport_open(p, &ofproto_port, &pp);
if (netdev) {
- ofport_install(p, netdev, &opp);
+ ofport_install(p, netdev, &pp);
+ if (ofp_to_u16(ofproto_port.ofp_port) < p->max_ports) {
+ p->alloc_port_no = MAX(p->alloc_port_no,
+ ofp_to_u16(ofproto_port.ofp_port));
+ }
}
}
}
+ SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
+ struct iface_hint *iface_hint = node->data;
+
+ if (!strcmp(iface_hint->br_name, p->name)) {
+ free(iface_hint->br_name);
+ free(iface_hint->br_type);
+ free(iface_hint);
+ shash_delete(&init_ofp_ports, node);
+ }
+ }
+
return 0;
}
return mtu ? mtu: ETH_PAYLOAD_MAX;
}
-/* Set the MTU of all datapath devices on 'p' to the minimum of the
- * non-datapath ports. */
+/* Update MTU of all datapath devices on 'p' to the minimum of the
+ * non-datapath ports in event of 'port' added or changed. */
static void
-set_internal_devs_mtu(struct ofproto *p)
+update_mtu(struct ofproto *p, struct ofport *port)
{
struct ofport *ofport;
- int mtu = find_min_mtu(p);
+ struct netdev *netdev = port->netdev;
+ int dev_mtu, old_min;
+
+ if (netdev_get_mtu(netdev, &dev_mtu)) {
+ port->mtu = 0;
+ return;
+ }
+ if (!strcmp(netdev_get_type(port->netdev), "internal")) {
+ if (dev_mtu > p->min_mtu) {
+ if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
+ dev_mtu = p->min_mtu;
+ }
+ }
+ port->mtu = dev_mtu;
+ return;
+ }
+
+ /* For non-internal port find new min mtu. */
+ old_min = p->min_mtu;
+ port->mtu = dev_mtu;
+ p->min_mtu = find_min_mtu(p);
+ if (p->min_mtu == old_min) {
+ return;
+ }
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);
+ if (!netdev_set_mtu(netdev, p->min_mtu)) {
+ ofport->mtu = p->min_mtu;
+ }
}
}
}
\f
+static void
+ofproto_rule_destroy(struct rule *rule)
+{
+ if (rule) {
+ rule->ofproto->ofproto_class->rule_destruct(rule);
+ ofproto_rule_destroy__(rule);
+ }
+}
+
static void
ofproto_rule_destroy__(struct rule *rule)
{
- free(rule->actions);
+ cls_rule_destroy(&rule->cr);
+ free(rule->ofpacts);
+ ovs_mutex_destroy(&rule->timeout_mutex);
+ ovs_rwlock_destroy(&rule->rwlock);
rule->ofproto->ofproto_class->rule_dealloc(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.
+ * remain when its ->destruct() function is called.. This function implements
+ * steps 4.4 and 4.5 in the section titled "Rule Life Cycle" in
+ * ofproto-provider.h.
*
* 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)
+ofproto_rule_delete(struct ofproto *ofproto, struct classifier *cls,
+ struct rule *rule)
+ OVS_REQ_WRLOCK(cls->rwlock)
{
- assert(!rule->pending);
- classifier_remove(&rule->ofproto->tables[rule->table_id], &rule->cr);
- ofproto_rule_destroy__(rule);
+ ofproto_delete_rule(ofproto, 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)
+ * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
+bool
+ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port)
+{
+ return (port == OFPP_ANY
+ || ofpacts_output_to_port(rule->ofpacts, rule->ofpacts_len, port));
+}
+
+/* Returns true if 'rule' has group and equals group_id. */
+bool
+ofproto_rule_has_out_group(const struct rule *rule, uint32_t group_id)
{
- const union ofp_action *oa;
- size_t left;
+ return (group_id == OFPG11_ANY
+ || ofpacts_output_to_group(rule->ofpacts, rule->ofpacts_len, group_id));
+}
- if (out_port == OFPP_NONE) {
+/* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
+ * OFPAT_ENQUEUE action that outputs to 'out_port'. */
+bool
+ofoperation_has_out_port(const struct ofoperation *op, ofp_port_t out_port)
+{
+ if (ofproto_rule_has_out_port(op->rule, out_port)) {
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;
- }
+
+ switch (op->type) {
+ case OFOPERATION_ADD:
+ case OFOPERATION_DELETE:
+ return false;
+
+ case OFOPERATION_MODIFY:
+ case OFOPERATION_REPLACE:
+ return ofpacts_output_to_port(op->ofpacts, op->ofpacts_len, out_port);
}
- return false;
+
+ NOT_REACHED();
}
/* 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.
+ * statistics appropriately.
*
* '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)
+rule_execute(struct rule *rule, ofp_port_t in_port, struct ofpbuf *packet)
{
struct flow flow;
+ union flow_in_port in_port_;
- assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
-
- flow_extract(packet, 0, 0, in_port, &flow);
+ in_port_.ofp_port = in_port;
+ flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
}
* 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)
+bool
+ofproto_rule_is_hidden(const struct rule *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_is_modifiable(const struct rule *rule)
+{
+ return !(rule_get_flags(rule) & OFTABLE_READONLY);
+}
\f
-static int
+static enum ofperr
handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
{
ofconn_send_reply(ofconn, make_echo_reply(oh));
return 0;
}
-static int
+static enum ofperr
handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
{
struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
- struct ofp_switch_features *osf;
- struct ofpbuf *buf;
+ struct ofputil_switch_features features;
struct ofport *port;
bool arp_match_ip;
- uint32_t actions;
-
- ofproto->ofproto_class->get_features(ofproto, &arp_match_ip, &actions);
- assert(actions & (1 << OFPAT_OUTPUT)); /* sanity check */
+ struct ofpbuf *b;
+ int n_tables;
+ int i;
+
+ ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
+ &features.actions);
+ ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
+
+ /* Count only non-hidden tables in the number of tables. (Hidden tables,
+ * if present, are always at the end.) */
+ n_tables = ofproto->n_tables;
+ for (i = 0; i < ofproto->n_tables; i++) {
+ if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
+ n_tables = i;
+ break;
+ }
+ }
- osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
- osf->datapath_id = htonll(ofproto->datapath_id);
- osf->n_buffers = htonl(pktbuf_capacity());
- osf->n_tables = ofproto->n_tables;
- osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
- OFPC_PORT_STATS | OFPC_QUEUE_STATS);
+ features.datapath_id = ofproto->datapath_id;
+ features.n_buffers = pktbuf_capacity();
+ features.n_tables = n_tables;
+ features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
+ OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
if (arp_match_ip) {
- osf->capabilities |= htonl(OFPC_ARP_MATCH_IP);
+ features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
}
- osf->actions = htonl(actions);
-
+ /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
+ features.auxiliary_id = 0;
+ b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
+ oh->xid);
HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
- ofpbuf_put(buf, &port->opp, sizeof port->opp);
+ ofputil_put_switch_features_port(&port->pp, b);
}
- ofconn_send_reply(ofconn, buf);
+ ofconn_send_reply(ofconn, b);
return 0;
}
-static int
+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);
- osc->flags = htons(ofproto->frag_handling);
+ buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
+ osc = ofpbuf_put_uninit(buf, sizeof *osc);
+ flags = ofproto->frag_handling;
+ /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
+ if (oh->version < OFP13_VERSION
+ && ofconn_get_invalid_ttl_to_controller(ofconn)) {
+ flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
+ }
+ osc->flags = htons(flags);
osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
ofconn_send_reply(ofconn, buf);
return 0;
}
-static int
-handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
+static enum ofperr
+handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
{
+ const struct ofp_switch_config *osc = ofpmsg_body(oh);
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) {
+ || ofconn_get_role(ofconn) != OFPCR12_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);
+ ovs_assert((cur & OFPC_FRAG_MASK) == cur);
if (cur != next) {
if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
ofproto->frag_handling = next;
}
}
}
+ /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
+ ofconn_set_invalid_ttl_to_controller(ofconn,
+ (oh->version < OFP13_VERSION
+ && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
}
/* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
- * error message code (composed with ofp_mkerr()) for the caller to propagate
- * upward. Otherwise, returns 0. */
-static int
-reject_slave_controller(const struct ofconn *ofconn)
+ * error message code for the caller to propagate upward. Otherwise, returns
+ * 0.
+ *
+ * The log message mentions 'msg_type'. */
+static enum ofperr
+reject_slave_controller(struct ofconn *ofconn)
{
if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
- && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
+ && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
+ return OFPERR_OFPBRC_EPERM;
} else {
return 0;
}
}
-static int
+/* Finds the OFPACT_METER action, if any, in the 'ofpacts_len' bytes of
+ * 'ofpacts'. If found, returns its meter ID; if not, returns 0.
+ *
+ * This function relies on the order of 'ofpacts' being correct (as checked by
+ * ofpacts_verify()). */
+static uint32_t
+find_meter(const struct ofpact ofpacts[], size_t ofpacts_len)
+{
+ const struct ofpact *a;
+
+ OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
+ enum ovs_instruction_type inst;
+
+ inst = ovs_instruction_type_from_ofpact_type(a->type);
+ if (a->type == OFPACT_METER) {
+ return ofpact_get_METER(a)->meter_id;
+ } else if (inst > OVSINST_OFPIT13_METER) {
+ break;
+ }
+ }
+
+ return 0;
+}
+
+/* Checks that the 'ofpacts_len' bytes of actions in 'ofpacts' are appropriate
+ * for a packet with the prerequisites satisfied by 'flow' in table 'table_id'.
+ * 'flow' may be temporarily modified, but is restored at return.
+ */
+static enum ofperr
+ofproto_check_ofpacts(struct ofproto *ofproto,
+ const struct ofpact ofpacts[], size_t ofpacts_len,
+ struct flow *flow, uint8_t table_id)
+{
+ enum ofperr error;
+ uint32_t mid;
+
+ error = ofpacts_check(ofpacts, ofpacts_len, flow,
+ u16_to_ofp(ofproto->max_ports), table_id);
+ if (error) {
+ return error;
+ }
+
+ mid = find_meter(ofpacts, ofpacts_len);
+ if (mid && ofproto_get_provider_meter_id(ofproto, mid) == UINT32_MAX) {
+ return OFPERR_OFPMMFC_INVALID_METER;
+ }
+ return 0;
+}
+
+static enum ofperr
handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
{
struct ofproto *p = ofconn_get_ofproto(ofconn);
- struct ofp_packet_out *opo;
- struct ofpbuf payload, *buffer;
- union ofp_action *ofp_actions;
- struct ofpbuf request;
+ struct ofputil_packet_out po;
+ struct ofpbuf *payload;
+ uint64_t ofpacts_stub[1024 / 8];
+ struct ofpbuf ofpacts;
struct flow flow;
- size_t n_ofp_actions;
- uint16_t in_port;
- int error;
+ union flow_in_port in_port_;
+ enum ofperr error;
COVERAGE_INC(ofproto_packet_out);
error = reject_slave_controller(ofconn);
if (error) {
- return error;
+ goto exit;
}
- /* Get ofp_packet_out. */
- ofpbuf_use_const(&request, oh, ntohs(oh->length));
- opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
-
- /* Get actions. */
- error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
- &ofp_actions, &n_ofp_actions);
+ /* Decode message. */
+ ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
+ error = ofputil_decode_packet_out(&po, oh, &ofpacts);
if (error) {
- return error;
+ goto exit_free_ofpacts;
+ }
+ if (ofp_to_u16(po.in_port) >= p->max_ports
+ && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) {
+ error = OFPERR_OFPBRC_BAD_PORT;
+ goto exit_free_ofpacts;
}
+
/* Get payload. */
- if (opo->buffer_id != htonl(UINT32_MAX)) {
- error = ofconn_pktbuf_retrieve(ofconn, ntohl(opo->buffer_id),
- &buffer, NULL);
- if (error || !buffer) {
- return error;
+ if (po.buffer_id != UINT32_MAX) {
+ error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
+ if (error || !payload) {
+ goto exit_free_ofpacts;
}
- payload = *buffer;
} else {
- payload = request;
- buffer = NULL;
+ /* Ensure that the L3 header is 32-bit aligned. */
+ payload = ofpbuf_clone_data_with_headroom(po.packet, po.packet_len, 2);
}
- /* Get in_port and partially validate it.
- *
- * We don't know what range of ports the ofproto actually implements, but
- * we do know that only certain reserved ports (numbered OFPP_MAX and
- * above) are valid. */
- in_port = ntohs(opo->in_port);
- if (in_port >= OFPP_MAX && in_port != OFPP_LOCAL && in_port != OFPP_NONE) {
- return ofp_mkerr_nicira(OFPET_BAD_REQUEST, NXBRC_BAD_IN_PORT);
+ /* Verify actions against packet, then send packet if successful. */
+ in_port_.ofp_port = po.in_port;
+ flow_extract(payload, 0, 0, NULL, &in_port_, &flow);
+ error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len, &flow, 0);
+ if (!error) {
+ error = p->ofproto_class->packet_out(p, payload, &flow,
+ po.ofpacts, po.ofpacts_len);
}
+ ofpbuf_delete(payload);
- /* Send out packet. */
- flow_extract(&payload, 0, 0, in_port, &flow);
- error = p->ofproto_class->packet_out(p, &payload, &flow,
- ofp_actions, n_ofp_actions);
- ofpbuf_delete(buffer);
-
+exit_free_ofpacts:
+ ofpbuf_uninit(&ofpacts);
+exit:
return error;
}
static void
-update_port_config(struct ofport *port, ovs_be32 config, ovs_be32 mask)
+update_port_config(struct ofport *port,
+ enum ofputil_port_config config,
+ enum ofputil_port_config mask)
{
- ovs_be32 old_config = port->opp.config;
+ enum ofputil_port_config old_config = port->pp.config;
+ enum ofputil_port_config toggle;
- mask &= config ^ port->opp.config;
- if (mask & htonl(OFPPC_PORT_DOWN)) {
- if (config & htonl(OFPPC_PORT_DOWN)) {
- netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
+ 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, NULL);
} else {
- netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
+ netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL);
}
+ toggle &= ~OFPUTIL_PC_PORT_DOWN;
}
- port->opp.config ^= mask & (htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
- OFPPC_NO_FLOOD | OFPPC_NO_FWD |
- OFPPC_NO_PACKET_IN));
- if (port->opp.config != old_config) {
+ port->pp.config ^= toggle;
+ if (port->pp.config != old_config) {
port->ofproto->ofproto_class->port_reconfigured(port, old_config);
}
}
-static int
+static enum ofperr
handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
{
struct ofproto *p = ofconn_get_ofproto(ofconn);
- const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
+ struct ofputil_port_mod pm;
struct ofport *port;
- int error;
+ enum ofperr error;
error = reject_slave_controller(ofconn);
if (error) {
return error;
}
- port = ofproto_get_port(p, ntohs(opm->port_no));
+ error = ofputil_decode_port_mod(oh, &pm);
+ if (error) {
+ return error;
+ }
+
+ port = ofproto_get_port(p, pm.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);
+ return OFPERR_OFPPMFC_BAD_PORT;
+ } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
+ return OFPERR_OFPPMFC_BAD_HW_ADDR;
} else {
- update_port_config(port, opm->config, opm->mask);
- if (opm->advertise) {
- netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
+ update_port_config(port, pm.config, pm.mask);
+ if (pm.advertise) {
+ netdev_set_advertisements(port->netdev, pm.advertise);
}
}
return 0;
}
-static int
+static enum ofperr
handle_desc_stats_request(struct ofconn *ofconn,
- const struct ofp_stats_msg *request)
+ const struct ofp_header *request)
{
+ static const char *default_mfr_desc = "Nicira, Inc.";
+ static const char *default_hw_desc = "Open vSwitch";
+ static const char *default_sw_desc = VERSION;
+ static const char *default_serial_desc = "None";
+ static const char *default_dp_desc = "None";
+
struct ofproto *p = ofconn_get_ofproto(ofconn);
struct ofp_desc_stats *ods;
struct ofpbuf *msg;
- ods = ofputil_make_stats_reply(sizeof *ods, request, &msg);
- ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
- ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
- ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
- ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
- ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
+ msg = ofpraw_alloc_stats_reply(request, 0);
+ ods = ofpbuf_put_zeros(msg, sizeof *ods);
+ ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
+ sizeof ods->mfr_desc);
+ ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
+ sizeof ods->hw_desc);
+ ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
+ sizeof ods->sw_desc);
+ ovs_strlcpy(ods->serial_num,
+ p->serial_desc ? p->serial_desc : default_serial_desc,
+ sizeof ods->serial_num);
+ ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
+ sizeof ods->dp_desc);
ofconn_send_reply(ofconn, msg);
return 0;
}
-static int
+static enum ofperr
handle_table_stats_request(struct ofconn *ofconn,
- const struct ofp_stats_msg *request)
+ const struct ofp_header *request)
{
struct ofproto *p = ofconn_get_ofproto(ofconn);
- struct ofp_table_stats *ots;
+ struct ofp12_table_stats *ots;
struct ofpbuf *msg;
+ int n_tables;
size_t i;
- ofputil_make_stats_reply(sizeof(struct ofp_stats_msg), request, &msg);
-
- ots = ofpbuf_put_zeros(msg, sizeof *ots * p->n_tables);
+ /* Set up default values.
+ *
+ * ofp12_table_stats is used as a generic structure as
+ * it is able to hold all the fields for ofp10_table_stats
+ * and ofp11_table_stats (and of course itself).
+ */
+ ots = xcalloc(p->n_tables, sizeof *ots);
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].match = htonll(OFPXMT12_MASK);
+ ots[i].wildcards = htonll(OFPXMT12_MASK);
+ ots[i].write_actions = htonl(OFPAT11_OUTPUT);
+ ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
+ ots[i].write_setfields = htonll(OFPXMT12_MASK);
+ ots[i].apply_setfields = htonll(OFPXMT12_MASK);
+ ots[i].metadata_match = htonll(UINT64_MAX);
+ ots[i].metadata_write = htonll(UINT64_MAX);
+ ots[i].instructions = htonl(OFPIT11_ALL);
+ ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
- ots[i].active_count = htonl(classifier_count(&p->tables[i]));
+ ovs_rwlock_rdlock(&p->tables[i].cls.rwlock);
+ ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
+ ovs_rwlock_unlock(&p->tables[i].cls.rwlock);
}
p->ofproto_class->get_tables(p, ots);
+ /* Post-process the tables, dropping hidden tables. */
+ n_tables = p->n_tables;
+ for (i = 0; i < p->n_tables; i++) {
+ const struct oftable *table = &p->tables[i];
+
+ if (table->flags & OFTABLE_HIDDEN) {
+ n_tables = i;
+ break;
+ }
+
+ if (table->name) {
+ ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
+ }
+
+ if (table->max_flows < ntohl(ots[i].max_entries)) {
+ ots[i].max_entries = htonl(table->max_flows);
+ }
+ }
+
+ msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
ofconn_send_reply(ofconn, msg);
+
+ free(ots);
+
return 0;
}
static void
append_port_stat(struct ofport *port, struct list *replies)
{
- struct netdev_stats stats;
- struct ofp_port_stats *ops;
+ struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
+
+ calc_duration(port->created, time_msec(),
+ &ops.duration_sec, &ops.duration_nsec);
/* 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 = port->opp.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));
+ ofproto_port_get_stats(port, &ops.stats);
+
+ ofputil_append_port_stat(replies, &ops);
}
-static int
+static enum ofperr
handle_port_stats_request(struct ofconn *ofconn,
- const struct ofp_port_stats_request *psr)
+ const struct ofp_header *request)
{
struct ofproto *p = ofconn_get_ofproto(ofconn);
struct ofport *port;
struct list replies;
+ ofp_port_t port_no;
+ enum ofperr error;
- ofputil_start_stats_reply(&psr->osm, &replies);
- if (psr->port_no != htons(OFPP_NONE)) {
- port = ofproto_get_port(p, ntohs(psr->port_no));
+ error = ofputil_decode_port_stats_request(request, &port_no);
+ if (error) {
+ return error;
+ }
+
+ ofpmp_init(&replies, request);
+ if (port_no != OFPP_ANY) {
+ port = ofproto_get_port(p, port_no);
if (port) {
append_port_stat(port, &replies);
}
return 0;
}
+static enum ofperr
+handle_port_desc_stats_request(struct ofconn *ofconn,
+ const struct ofp_header *request)
+{
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
+ enum ofp_version version;
+ struct ofport *port;
+ struct list replies;
+
+ ofpmp_init(&replies, request);
+
+ version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
+ HMAP_FOR_EACH (port, hmap_node, &p->ports) {
+ ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
+ }
+
+ ofconn_send_replies(ofconn, &replies);
+ return 0;
+}
+
+static uint32_t
+hash_cookie(ovs_be64 cookie)
+{
+ return hash_2words((OVS_FORCE uint64_t)cookie >> 32,
+ (OVS_FORCE uint64_t)cookie);
+}
+
+static void
+cookies_insert(struct ofproto *ofproto, struct rule *rule)
+{
+ hindex_insert(&ofproto->cookies, &rule->cookie_node,
+ hash_cookie(rule->flow_cookie));
+}
+
+static void
+cookies_remove(struct ofproto *ofproto, struct rule *rule)
+{
+ hindex_remove(&ofproto->cookies, &rule->cookie_node);
+}
+
+static void
+ofproto_rule_change_cookie(struct ofproto *ofproto, struct rule *rule,
+ ovs_be64 new_cookie)
+{
+ if (new_cookie != rule->flow_cookie) {
+ cookies_remove(ofproto, rule);
+
+ ovs_rwlock_wrlock(&rule->rwlock);
+ rule->flow_cookie = new_cookie;
+ ovs_rwlock_unlock(&rule->rwlock);
+
+ cookies_insert(ofproto, rule);
+ }
+}
+
static void
-calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec)
+calc_duration(long long int start, long long int now,
+ uint32_t *sec, uint32_t *nsec)
{
- long long int msecs = time_msec() - start;
+ long long int msecs = now - start;
*sec = msecs / 1000;
*nsec = (msecs % 1000) * (1000 * 1000);
}
/* 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 int
+static enum ofperr
check_table_id(const struct ofproto *ofproto, uint8_t table_id)
{
return (table_id == 0xff || table_id < ofproto->n_tables
? 0
- : ofp_mkerr_nicira(OFPET_BAD_REQUEST, NXBRC_BAD_TABLE_ID));
+ : OFPERR_OFPBRC_BAD_TABLE_ID);
+
+}
+
+static struct oftable *
+next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
+{
+ struct oftable *table;
+
+ for (table = &ofproto->tables[table_id];
+ table < &ofproto->tables[ofproto->n_tables];
+ table++) {
+ if (!(table->flags & OFTABLE_HIDDEN)) {
+ return table;
+ }
+ }
+ return NULL;
}
-static struct classifier *
-first_matching_table(struct ofproto *ofproto, uint8_t table_id)
+static struct oftable *
+first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
{
if (table_id == 0xff) {
- return &ofproto->tables[0];
+ return next_visible_table(ofproto, 0);
} else if (table_id < ofproto->n_tables) {
return &ofproto->tables[table_id];
} else {
}
}
-static struct classifier *
-next_matching_table(struct ofproto *ofproto,
- struct classifier *cls, uint8_t table_id)
+static struct oftable *
+next_matching_table(const struct ofproto *ofproto,
+ const struct oftable *table, uint8_t table_id)
{
- return (table_id == 0xff && cls != &ofproto->tables[ofproto->n_tables - 1]
- ? cls + 1
+ return (table_id == 0xff
+ ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
: NULL);
}
-/* Assigns CLS to each classifier table, in turn, that matches TABLE_ID in
- * OFPROTO:
+/* 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.
+ * 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.
+ * 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
*
* All parameters are evaluated multiple times.
*/
-#define FOR_EACH_MATCHING_TABLE(CLS, TABLE_ID, OFPROTO) \
- for ((CLS) = first_matching_table(OFPROTO, TABLE_ID); \
- (CLS) != NULL; \
- (CLS) = next_matching_table(OFPROTO, CLS, TABLE_ID))
+#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
+ * If 'out_port' is anything other than OFPP_ANY, 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 int
+static enum ofperr
collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
- const struct cls_rule *match, uint16_t out_port,
+ const struct match *match,
+ ovs_be64 cookie, ovs_be64 cookie_mask,
+ ofp_port_t out_port, uint32_t out_group,
struct list *rules)
{
- struct classifier *cls;
- int error;
+ struct oftable *table;
+ struct cls_rule cr;
+ enum ofperr error;
error = check_table_id(ofproto, table_id);
if (error) {
}
list_init(rules);
- FOR_EACH_MATCHING_TABLE (cls, table_id, ofproto) {
+ cls_rule_init(&cr, match, 0);
+
+ if (cookie_mask == htonll(UINT64_MAX)) {
+ struct rule *rule;
+
+ HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node, hash_cookie(cookie),
+ &ofproto->cookies) {
+ if (table_id != rule->table_id && table_id != 0xff) {
+ continue;
+ }
+ if (ofproto_rule_is_hidden(rule)) {
+ continue;
+ }
+ if (cls_rule_is_loose_match(&rule->cr, &cr.match)) {
+ if (rule->pending) {
+ error = OFPROTO_POSTPONE;
+ goto exit;
+ }
+ if (rule->flow_cookie == cookie /* Hash collisions possible. */
+ && ofproto_rule_has_out_port(rule, out_port)
+ && ofproto_rule_has_out_group(rule, out_group)) {
+ list_push_back(rules, &rule->ofproto_node);
+ }
+ }
+ }
+ goto exit;
+ }
+
+ FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
struct cls_cursor cursor;
struct rule *rule;
- cls_cursor_init(&cursor, cls, match);
+ ovs_rwlock_rdlock(&table->cls.rwlock);
+ cls_cursor_init(&cursor, &table->cls, &cr);
CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
if (rule->pending) {
- return OFPROTO_POSTPONE;
+ ovs_rwlock_unlock(&table->cls.rwlock);
+ error = OFPROTO_POSTPONE;
+ goto exit;
}
- if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
+ if (!ofproto_rule_is_hidden(rule)
+ && ofproto_rule_has_out_port(rule, out_port)
+ && ofproto_rule_has_out_group(rule, out_group)
+ && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
list_push_back(rules, &rule->ofproto_node);
}
}
+ ovs_rwlock_unlock(&table->cls.rwlock);
}
- return 0;
+
+exit:
+ cls_rule_destroy(&cr);
+ return error;
}
/* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
* 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
+ * If 'out_port' is anything other than OFPP_ANY, 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 int
+static enum ofperr
collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
- const struct cls_rule *match, uint16_t out_port,
+ const struct match *match, unsigned int priority,
+ ovs_be64 cookie, ovs_be64 cookie_mask,
+ ofp_port_t out_port, uint32_t out_group,
struct list *rules)
{
- struct classifier *cls;
+ struct oftable *table;
+ struct cls_rule cr;
int error;
error = check_table_id(ofproto, table_id);
}
list_init(rules);
- FOR_EACH_MATCHING_TABLE (cls, table_id, ofproto) {
+ cls_rule_init(&cr, match, priority);
+
+ if (cookie_mask == htonll(UINT64_MAX)) {
+ struct rule *rule;
+
+ HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node, hash_cookie(cookie),
+ &ofproto->cookies) {
+ if (table_id != rule->table_id && table_id != 0xff) {
+ continue;
+ }
+ if (ofproto_rule_is_hidden(rule)) {
+ continue;
+ }
+ if (cls_rule_equal(&rule->cr, &cr)) {
+ if (rule->pending) {
+ error = OFPROTO_POSTPONE;
+ goto exit;
+ }
+ if (rule->flow_cookie == cookie /* Hash collisions possible. */
+ && ofproto_rule_has_out_port(rule, out_port)
+ && ofproto_rule_has_out_group(rule, out_group)) {
+ list_push_back(rules, &rule->ofproto_node);
+ }
+ }
+ }
+ goto exit;
+ }
+
+ FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
struct rule *rule;
- rule = rule_from_cls_rule(classifier_find_rule_exactly(cls, match));
+ ovs_rwlock_rdlock(&table->cls.rwlock);
+ rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls,
+ &cr));
+ ovs_rwlock_unlock(&table->cls.rwlock);
if (rule) {
if (rule->pending) {
- return OFPROTO_POSTPONE;
+ error = OFPROTO_POSTPONE;
+ goto exit;
}
- if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
+ if (!ofproto_rule_is_hidden(rule)
+ && ofproto_rule_has_out_port(rule, out_port)
+ && ofproto_rule_has_out_group(rule, out_group)
+ && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
list_push_back(rules, &rule->ofproto_node);
}
}
}
+
+exit:
+ cls_rule_destroy(&cr);
return 0;
}
+/* 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)
+ const struct ofp_header *request)
{
struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
struct ofputil_flow_stats_request fsr;
struct list replies;
struct list rules;
struct rule *rule;
- int error;
+ enum ofperr error;
- error = ofputil_decode_flow_stats_request(&fsr, &osm->header);
+ error = ofputil_decode_flow_stats_request(&fsr, request);
if (error) {
return error;
}
error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
- fsr.out_port, &rules);
+ fsr.cookie, fsr.cookie_mask,
+ fsr.out_port, fsr.out_group, &rules);
if (error) {
return error;
}
- ofputil_start_stats_reply(osm, &replies);
+ ofpmp_init(&replies, request);
LIST_FOR_EACH (rule, ofproto_node, &rules) {
+ long long int now = time_msec();
struct ofputil_flow_stats fs;
- fs.rule = rule->cr;
+ minimatch_expand(&rule->cr.match, &fs.match);
+ fs.priority = rule->cr.priority;
fs.cookie = rule->flow_cookie;
fs.table_id = rule->table_id;
- calc_flow_duration__(rule->created, &fs.duration_sec,
- &fs.duration_nsec);
- fs.idle_timeout = rule->idle_timeout;
- fs.hard_timeout = rule->hard_timeout;
+ calc_duration(rule->created, now, &fs.duration_sec, &fs.duration_nsec);
+ 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);
- }
+ fs.ofpacts = rule->ofpacts;
+ fs.ofpacts_len = rule->ofpacts_len;
+
+ ovs_mutex_lock(&rule->timeout_mutex);
+ fs.idle_timeout = rule->idle_timeout;
+ fs.hard_timeout = rule->hard_timeout;
+ ovs_mutex_unlock(&rule->timeout_mutex);
+
+ fs.flags = 0;
+ if (rule->send_flow_removed) {
+ fs.flags |= OFPUTIL_FF_SEND_FLOW_REM;
+ /* FIXME: Implement OFPUTIL_FF_NO_PKT_COUNTS and
+ OFPUTIL_FF_NO_BYT_COUNTS. */
+ }
+ ofputil_append_flow_stats_reply(&fs, &replies);
+ }
ofconn_send_replies(ofconn, &replies);
return 0;
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");
- }
+ ofpacts_format(rule->ofpacts, rule->ofpacts_len, results);
ds_put_cstr(results, "\n");
}
void
ofproto_get_all_flows(struct ofproto *p, struct ds *results)
{
- struct classifier *cls;
+ struct oftable *table;
- OFPROTO_FOR_EACH_TABLE (cls, p) {
+ OFPROTO_FOR_EACH_TABLE (table, p) {
struct cls_cursor cursor;
struct rule *rule;
- cls_cursor_init(&cursor, cls, NULL);
+ ovs_rwlock_rdlock(&table->cls.rwlock);
+ cls_cursor_init(&cursor, &table->cls, NULL);
CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
flow_stats_ds(rule, results);
}
+ ovs_rwlock_unlock(&table->cls.rwlock);
}
}
ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
}
-/* Checks the fault status of CFM for 'ofp_port' within 'ofproto'. Returns 1
- * if CFM is faulted (generally indiciating a connectivity problem), 0 if CFM
- * is not faulted, and -1 if CFM is not enabled on 'ofp_port'. */
-int
-ofproto_port_get_cfm_fault(const struct ofproto *ofproto, uint16_t ofp_port)
-{
- 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);
-}
-
-/* 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)
+/* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
+ * true if the port's CFM status was successfully stored into '*status'.
+ * Returns false if the port did not have CFM configured, in which case
+ * '*status' is indeterminate.
+ *
+ * The caller must provide and owns '*status', and must free 'status->rmps'. */
+bool
+ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port,
+ struct ofproto_cfm_status *status)
{
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);
+ return (ofport
+ && ofproto->ofproto_class->get_cfm_status
+ && ofproto->ofproto_class->get_cfm_status(ofport, status));
}
-static int
+static enum ofperr
handle_aggregate_stats_request(struct ofconn *ofconn,
- const struct ofp_stats_msg *osm)
+ const struct ofp_header *oh)
{
struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
struct ofputil_flow_stats_request request;
struct ofpbuf *reply;
struct list rules;
struct rule *rule;
- int error;
+ enum ofperr error;
- error = ofputil_decode_flow_stats_request(&request, &osm->header);
+ error = ofputil_decode_flow_stats_request(&request, oh);
if (error) {
return error;
}
error = collect_rules_loose(ofproto, request.table_id, &request.match,
- request.out_port, &rules);
+ request.cookie, request.cookie_mask,
+ request.out_port, request.out_group, &rules);
if (error) {
return error;
}
stats.byte_count = UINT64_MAX;
}
- reply = ofputil_encode_aggregate_stats_reply(&stats, osm);
+ reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
ofconn_send_reply(ofconn, reply);
return 0;
struct queue_stats_cbdata {
struct ofport *ofport;
struct list replies;
+ long long int now;
};
static void
put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
const struct netdev_queue_stats *stats)
{
- struct ofp_queue_stats *reply;
+ struct ofputil_queue_stats oqs;
- reply = ofputil_append_stats_reply(sizeof *reply, &cbdata->replies);
- reply->port_no = cbdata->ofport->opp.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));
+ oqs.port_no = cbdata->ofport->pp.port_no;
+ oqs.queue_id = queue_id;
+ oqs.tx_bytes = stats->tx_bytes;
+ oqs.tx_packets = stats->tx_packets;
+ oqs.tx_errors = stats->tx_errors;
+ if (stats->created != LLONG_MIN) {
+ calc_duration(stats->created, cbdata->now,
+ &oqs.duration_sec, &oqs.duration_nsec);
+ } else {
+ oqs.duration_sec = oqs.duration_nsec = UINT32_MAX;
+ }
+ ofputil_append_queue_stat(&cbdata->replies, &oqs);
}
static void
put_queue_stats(cbdata, queue_id, stats);
}
-static void
+static enum ofperr
handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
struct queue_stats_cbdata *cbdata)
{
if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
put_queue_stats(cbdata, queue_id, &stats);
+ } else {
+ return OFPERR_OFPQOFC_BAD_QUEUE;
}
}
+ return 0;
}
-static int
+static enum ofperr
handle_queue_stats_request(struct ofconn *ofconn,
- const struct ofp_queue_stats_request *qsr)
+ const struct ofp_header *rq)
{
struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
struct queue_stats_cbdata cbdata;
struct ofport *port;
- unsigned int port_no;
- uint32_t queue_id;
+ enum ofperr error;
+ struct ofputil_queue_stats_request oqsr;
COVERAGE_INC(ofproto_queue_req);
- ofputil_start_stats_reply(&qsr->osm, &cbdata.replies);
+ ofpmp_init(&cbdata.replies, rq);
+ cbdata.now = time_msec();
+
+ error = ofputil_decode_queue_stats_request(rq, &oqsr);
+ if (error) {
+ return error;
+ }
- port_no = ntohs(qsr->port_no);
- queue_id = ntohl(qsr->queue_id);
- if (port_no == OFPP_ALL) {
+ if (oqsr.port_no == OFPP_ANY) {
+ error = OFPERR_OFPQOFC_BAD_QUEUE;
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);
+ if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
+ error = 0;
+ }
}
+ } else {
+ port = ofproto_get_port(ofproto, oqsr.port_no);
+ error = (port
+ ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
+ : OFPERR_OFPQOFC_BAD_PORT);
+ }
+ if (!error) {
+ ofconn_send_replies(ofconn, &cbdata.replies);
} else {
ofpbuf_list_delete(&cbdata.replies);
- return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
}
- ofconn_send_replies(ofconn, &cbdata.replies);
- return 0;
+ return error;
}
static bool
return false;
}
+static enum ofperr
+evict_rule_from_table(struct ofproto *ofproto, struct oftable *table)
+{
+ struct rule *rule;
+ size_t n_rules;
+
+ ovs_rwlock_rdlock(&table->cls.rwlock);
+ n_rules = classifier_count(&table->cls);
+ ovs_rwlock_unlock(&table->cls.rwlock);
+
+ if (n_rules < table->max_flows) {
+ return 0;
+ } else if (!choose_rule_to_evict(table, &rule)) {
+ return OFPERR_OFPFMFC_TABLE_FULL;
+ } else if (rule->pending) {
+ ovs_rwlock_unlock(&rule->rwlock);
+ return OFPROTO_POSTPONE;
+ } else {
+ struct ofopgroup *group;
+
+ group = ofopgroup_create_unattached(ofproto);
+ delete_flow__(rule, group, OFPRR_EVICTION);
+ ofopgroup_submit(group);
+
+ return 0;
+ }
+}
+
/* 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 as encoded by ofp_mkerr() on failure, or OFPROTO_POSTPONE if the
- * operation cannot be initiated now but may be retried later.
+ * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
+ * initiated now but may be retried later.
+ *
+ * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
+ * ownership remains with the caller.
*
* 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
* if any. */
-static int
+static enum ofperr
add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
- const struct ofputil_flow_mod *fm, const struct ofp_header *request)
+ struct ofputil_flow_mod *fm, const struct ofp_header *request)
{
- struct classifier *table;
+ struct oftable *table;
struct ofopgroup *group;
- struct rule *victim;
+ struct cls_rule cr;
struct rule *rule;
+ uint8_t table_id;
int error;
error = check_table_id(ofproto, fm->table_id);
/* 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,
+ error = ofproto->ofproto_class->rule_choose_table(ofproto,
+ &fm->match,
&table_id);
if (error) {
return error;
}
- assert(table_id < ofproto->n_tables);
- table = &ofproto->tables[table_id];
+ ovs_assert(table_id < ofproto->n_tables);
} else {
- table = &ofproto->tables[0];
+ table_id = 0;
}
} else if (fm->table_id < ofproto->n_tables) {
- table = &ofproto->tables[fm->table_id];
+ table_id = fm->table_id;
} else {
- return ofp_mkerr_nicira(OFPET_FLOW_MOD_FAILED, NXFMFC_BAD_TABLE_ID);
+ return OFPERR_OFPBRC_BAD_TABLE_ID;
}
- /* Check for overlap, if requested. */
- if (fm->flags & OFPFF_CHECK_OVERLAP
- && classifier_rule_overlaps(table, &fm->cr)) {
- return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
+ table = &ofproto->tables[table_id];
+
+ if (table->flags & OFTABLE_READONLY) {
+ return OFPERR_OFPBRC_EPERM;
+ }
+
+ cls_rule_init(&cr, &fm->match, fm->priority);
+
+ /* Transform "add" into "modify" if there's an existing identical flow. */
+ ovs_rwlock_rdlock(&table->cls.rwlock);
+ rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr));
+ ovs_rwlock_unlock(&table->cls.rwlock);
+ if (rule) {
+ cls_rule_destroy(&cr);
+ if (!rule_is_modifiable(rule)) {
+ return OFPERR_OFPBRC_EPERM;
+ } else if (rule->pending) {
+ return OFPROTO_POSTPONE;
+ } else {
+ struct list rules;
+
+ list_init(&rules);
+ list_push_back(&rules, &rule->ofproto_node);
+ fm->modify_cookie = true;
+ return modify_flows__(ofproto, ofconn, fm, request, &rules);
+ }
+ }
+
+ /* Verify actions. */
+ error = ofproto_check_ofpacts(ofproto, fm->ofpacts, fm->ofpacts_len,
+ &fm->match.flow, table_id);
+ if (error) {
+ cls_rule_destroy(&cr);
+ return error;
}
/* Serialize against pending deletion. */
- if (is_flow_deletion_pending(ofproto, &fm->cr, table - ofproto->tables)) {
+ if (is_flow_deletion_pending(ofproto, &cr, table_id)) {
+ cls_rule_destroy(&cr);
return OFPROTO_POSTPONE;
}
+ /* Check for overlap, if requested. */
+ if (fm->flags & OFPUTIL_FF_CHECK_OVERLAP) {
+ bool overlaps;
+
+ ovs_rwlock_rdlock(&table->cls.rwlock);
+ overlaps = classifier_rule_overlaps(&table->cls, &cr);
+ ovs_rwlock_unlock(&table->cls.rwlock);
+
+ if (overlaps) {
+ cls_rule_destroy(&cr);
+ return OFPERR_OFPFMFC_OVERLAP;
+ }
+ }
+
+ /* If necessary, evict an existing rule to clear out space. */
+ error = evict_rule_from_table(ofproto, table);
+ if (error) {
+ cls_rule_destroy(&cr);
+ return error;
+ }
+
/* Allocate new rule. */
rule = ofproto->ofproto_class->rule_alloc();
if (!rule) {
+ cls_rule_destroy(&cr);
VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
- ofproto->name, strerror(error));
+ ofproto->name, ovs_strerror(error));
return ENOMEM;
}
+
+ /* Initialize base state. */
rule->ofproto = ofproto;
- rule->cr = fm->cr;
+ cls_rule_move(&rule->cr, &cr);
rule->pending = NULL;
- rule->flow_cookie = fm->cookie;
- rule->created = rule->modified = time_msec();
+ rule->flow_cookie = fm->new_cookie;
+ rule->created = rule->modified = rule->used = time_msec();
+
+ ovs_mutex_init(&rule->timeout_mutex);
+ ovs_mutex_lock(&rule->timeout_mutex);
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;
-
- /* Insert new rule. */
- victim = rule_from_cls_rule(classifier_replace(table, &rule->cr));
- if (victim && victim->pending) {
- error = OFPROTO_POSTPONE;
- } else {
- 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);
- }
- ofopgroup_submit(group);
- }
+ ovs_mutex_unlock(&rule->timeout_mutex);
- /* Back out if an error occurred. */
+ rule->table_id = table - ofproto->tables;
+ rule->send_flow_removed = (fm->flags & OFPUTIL_FF_SEND_FLOW_REM) != 0;
+ rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
+ rule->ofpacts_len = fm->ofpacts_len;
+ rule->meter_id = find_meter(rule->ofpacts, rule->ofpacts_len);
+ list_init(&rule->meter_list_node);
+ rule->eviction_group = NULL;
+ list_init(&rule->expirable);
+ rule->monitor_flags = 0;
+ rule->add_seqno = 0;
+ rule->modify_seqno = 0;
+ ovs_rwlock_init(&rule->rwlock);
+
+ /* Construct rule, initializing derived state. */
+ error = ofproto->ofproto_class->rule_construct(rule);
if (error) {
- if (victim) {
- classifier_replace(table, &victim->cr);
- } else {
- classifier_remove(table, &rule->cr);
- }
ofproto_rule_destroy__(rule);
+ return error;
}
+
+ /* Insert rule. */
+ oftable_insert_rule(rule);
+
+ group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
+ ofoperation_create(group, rule, OFOPERATION_ADD, 0);
+ ofproto->ofproto_class->rule_insert(rule);
+ ofopgroup_submit(group);
+
return error;
}
\f
* if any.
*
* Returns 0 on success, otherwise an OpenFlow error code. */
-static int
+static enum ofperr
modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
- const struct ofputil_flow_mod *fm,
- const struct ofp_header *request, struct list *rules)
+ struct ofputil_flow_mod *fm, const struct ofp_header *request,
+ struct list *rules)
{
+ enum ofoperation_type type;
struct ofopgroup *group;
struct rule *rule;
+ enum ofperr error;
+ type = fm->command == OFPFC_ADD ? OFOPERATION_REPLACE : OFOPERATION_MODIFY;
group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
+ error = OFPERR_OFPBRC_EPERM;
LIST_FOR_EACH (rule, ofproto_node, rules) {
- 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;
- rule->ofproto->ofproto_class->rule_modify_actions(rule);
+ struct ofoperation *op;
+ bool actions_changed;
+ bool reset_counters;
+
+ /* FIXME: Implement OFPFUTIL_FF_RESET_COUNTS */
+
+ if (rule_is_modifiable(rule)) {
+ /* At least one rule is modifiable, don't report EPERM error. */
+ error = 0;
+ } else {
+ continue;
+ }
+
+ /* Verify actions. */
+ error = ofpacts_check(fm->ofpacts, fm->ofpacts_len, &fm->match.flow,
+ u16_to_ofp(ofproto->max_ports), rule->table_id);
+ if (error) {
+ return error;
+ }
+
+ actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
+ rule->ofpacts, rule->ofpacts_len);
+
+ op = ofoperation_create(group, rule, type, 0);
+
+ if (fm->modify_cookie && fm->new_cookie != htonll(UINT64_MAX)) {
+ ofproto_rule_change_cookie(ofproto, rule, fm->new_cookie);
+ }
+ if (type == OFOPERATION_REPLACE) {
+ ovs_mutex_lock(&rule->timeout_mutex);
+ rule->idle_timeout = fm->idle_timeout;
+ rule->hard_timeout = fm->hard_timeout;
+ ovs_mutex_unlock(&rule->timeout_mutex);
+
+ rule->send_flow_removed = (fm->flags
+ & OFPUTIL_FF_SEND_FLOW_REM) != 0;
+
+ if (fm->idle_timeout || fm->hard_timeout) {
+ if (!rule->eviction_group) {
+ eviction_group_add_rule(rule);
+ }
+ } else {
+ eviction_group_remove_rule(rule);
+ }
+ }
+
+ reset_counters = (fm->flags & OFPUTIL_FF_RESET_COUNTS) != 0;
+ if (actions_changed || reset_counters) {
+ op->ofpacts = rule->ofpacts;
+ op->ofpacts_len = rule->ofpacts_len;
+ op->meter_id = rule->meter_id;
+
+ ovs_rwlock_wrlock(&rule->rwlock);
+ rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
+ rule->ofpacts_len = fm->ofpacts_len;
+ ovs_rwlock_unlock(&rule->rwlock);
+
+ rule->meter_id = find_meter(rule->ofpacts, rule->ofpacts_len);
+ rule->ofproto->ofproto_class->rule_modify_actions(rule,
+ reset_counters);
} else {
- rule->modified = time_msec();
+ ofoperation_complete(op, 0);
}
- rule->flow_cookie = fm->cookie;
}
ofopgroup_submit(group);
- return 0;
+ return error;
+}
+
+static enum ofperr
+modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
+ struct ofputil_flow_mod *fm, const struct ofp_header *request)
+{
+ if (fm->cookie_mask != htonll(0) || fm->new_cookie == htonll(UINT64_MAX)) {
+ return 0;
+ }
+ return add_flow(ofproto, ofconn, fm, request);
}
-/* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
- * encoded by ofp_mkerr() on failure.
+/* 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 int
+static enum ofperr
modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
- const struct ofputil_flow_mod *fm,
+ struct ofputil_flow_mod *fm,
const struct ofp_header *request)
{
struct list rules;
int error;
- error = collect_rules_loose(ofproto, fm->table_id, &fm->cr, OFPP_NONE,
- &rules);
- return (error ? error
- : list_is_empty(&rules) ? add_flow(ofproto, ofconn, fm, request)
- : modify_flows__(ofproto, ofconn, fm, request, &rules));
+ error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
+ fm->cookie, fm->cookie_mask,
+ OFPP_ANY, OFPG11_ANY, &rules);
+ if (error) {
+ return error;
+ } else if (list_is_empty(&rules)) {
+ return modify_flows_add(ofproto, ofconn, fm, request);
+ } else {
+ return modify_flows__(ofproto, ofconn, fm, request, &rules);
+ }
}
/* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
- * code as encoded by ofp_mkerr() on failure.
+ * code on failure.
*
* 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
* if any. */
-static int
+static enum ofperr
modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
- const struct ofputil_flow_mod *fm,
+ struct ofputil_flow_mod *fm,
const struct ofp_header *request)
{
struct list rules;
int error;
- error = collect_rules_strict(ofproto, fm->table_id, &fm->cr, 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);
+ error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
+ fm->priority, fm->cookie, fm->cookie_mask,
+ OFPP_ANY, OFPG11_ANY, &rules);
+ if (error) {
+ return error;
+ } else if (list_is_empty(&rules)) {
+ return modify_flows_add(ofproto, ofconn, fm, request);
+ } else {
+ return list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
+ fm, request, &rules)
+ : 0;
+ }
}
\f
/* OFPFC_DELETE implementation. */
+static void
+delete_flow__(struct rule *rule, struct ofopgroup *group,
+ enum ofp_flow_removed_reason reason)
+{
+ struct ofproto *ofproto = rule->ofproto;
+
+ ofproto_rule_send_removed(rule, reason);
+
+ ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
+ oftable_remove_rule(rule);
+ ofproto->ofproto_class->rule_delete(rule);
+}
+
/* Deletes the rules listed in 'rules'.
*
* Returns 0 on success, otherwise an OpenFlow error code. */
-static int
+static enum ofperr
delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
- const struct ofp_header *request, struct list *rules)
+ const struct ofp_header *request, struct list *rules,
+ enum ofp_flow_removed_reason reason)
{
struct rule *rule, *next;
struct ofopgroup *group;
group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
- ofproto_rule_send_removed(rule, OFPRR_DELETE);
-
- ofoperation_create(group, rule, OFOPERATION_DELETE);
- classifier_remove(&ofproto->tables[rule->table_id], &rule->cr);
- rule->ofproto->ofproto_class->rule_destruct(rule);
+ ovs_rwlock_wrlock(&rule->rwlock);
+ delete_flow__(rule, group, reason);
}
ofopgroup_submit(group);
}
/* Implements OFPFC_DELETE. */
-static int
+static enum ofperr
delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
const struct ofputil_flow_mod *fm,
const struct ofp_header *request)
{
struct list rules;
- int error;
+ enum ofperr error;
- error = collect_rules_loose(ofproto, fm->table_id, &fm->cr, fm->out_port,
- &rules);
+ error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
+ fm->cookie, fm->cookie_mask,
+ fm->out_port, fm->out_group, &rules);
return (error ? error
: !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
- &rules)
+ &rules, OFPRR_DELETE)
: 0);
}
/* Implements OFPFC_DELETE_STRICT. */
-static int
+static enum ofperr
delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
const struct ofputil_flow_mod *fm,
const struct ofp_header *request)
{
struct list rules;
- int error;
+ enum ofperr error;
- error = collect_rules_strict(ofproto, fm->table_id, &fm->cr, fm->out_port,
- &rules);
+ error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
+ fm->priority, fm->cookie, fm->cookie_mask,
+ fm->out_port, fm->out_group, &rules);
return (error ? error
: list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
- request, &rules)
+ request, &rules,
+ OFPRR_DELETE)
: 0);
}
{
struct ofputil_flow_removed fr;
- if (rule_is_hidden(rule) || !rule->send_flow_removed) {
+ if (ofproto_rule_is_hidden(rule) || !rule->send_flow_removed) {
return;
}
- fr.rule = rule->cr;
+ minimatch_expand(&rule->cr.match, &fr.match);
+ fr.priority = rule->cr.priority;
fr.cookie = rule->flow_cookie;
fr.reason = reason;
- calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec);
+ fr.table_id = rule->table_id;
+ calc_duration(rule->created, time_msec(),
+ &fr.duration_sec, &fr.duration_nsec);
+ ovs_mutex_lock(&rule->timeout_mutex);
fr.idle_timeout = rule->idle_timeout;
+ fr.hard_timeout = rule->hard_timeout;
+ ovs_mutex_unlock(&rule->timeout_mutex);
rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
&fr.byte_count);
* OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
* ofproto.
*
+ * 'rule' must not have a pending operation (that is, 'rule->pending' must be
+ * NULL).
+ *
* ofproto implementation ->run() functions should use this function to expire
* OpenFlow flows. */
void
ofproto_rule_expire(struct rule *rule, uint8_t reason)
{
struct ofproto *ofproto = rule->ofproto;
- struct ofopgroup *group;
-
- assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
+ struct classifier *cls = &ofproto->tables[rule->table_id].cls;
+ ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT
+ || reason == OFPRR_DELETE || reason == OFPRR_GROUP_DELETE);
ofproto_rule_send_removed(rule, reason);
- group = ofopgroup_create_unattached(ofproto);
- ofoperation_create(group, rule, OFOPERATION_DELETE);
- classifier_remove(&ofproto->tables[rule->table_id], &rule->cr);
- rule->ofproto->ofproto_class->rule_destruct(rule);
- ofopgroup_submit(group);
+ ovs_rwlock_wrlock(&cls->rwlock);
+ ofproto_delete_rule(ofproto, cls, rule);
+ ovs_rwlock_unlock(&cls->rwlock);
+}
+
+/* Reduces '*timeout' to no more than 'max'. A value of zero in either case
+ * means "infinite". */
+static void
+reduce_timeout(uint16_t max, uint16_t *timeout)
+{
+ if (max && (!*timeout || *timeout > max)) {
+ *timeout = max;
+ }
+}
+
+/* If 'idle_timeout' is nonzero, and 'rule' has no idle timeout or an idle
+ * timeout greater than 'idle_timeout', lowers 'rule''s idle timeout to
+ * 'idle_timeout' seconds. Similarly for 'hard_timeout'.
+ *
+ * Suitable for implementing OFPACT_FIN_TIMEOUT. */
+void
+ofproto_rule_reduce_timeouts(struct rule *rule,
+ uint16_t idle_timeout, uint16_t hard_timeout)
+ OVS_EXCLUDED(rule->ofproto->expirable_mutex, rule->timeout_mutex)
+{
+ if (!idle_timeout && !hard_timeout) {
+ return;
+ }
+
+ ovs_mutex_lock(&rule->ofproto->expirable_mutex);
+ if (list_is_empty(&rule->expirable)) {
+ list_insert(&rule->ofproto->expirable, &rule->expirable);
+ }
+ ovs_mutex_unlock(&rule->ofproto->expirable_mutex);
+
+ ovs_mutex_lock(&rule->timeout_mutex);
+ reduce_timeout(idle_timeout, &rule->idle_timeout);
+ reduce_timeout(hard_timeout, &rule->hard_timeout);
+ ovs_mutex_unlock(&rule->timeout_mutex);
}
\f
-static int
+static enum ofperr
handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
{
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
struct ofputil_flow_mod fm;
- int error;
+ uint64_t ofpacts_stub[1024 / 8];
+ struct ofpbuf ofpacts;
+ enum ofperr error;
+ long long int now;
error = reject_slave_controller(ofconn);
if (error) {
- return error;
+ goto exit;
}
- error = ofputil_decode_flow_mod(&fm, oh,
- ofconn_get_flow_mod_table_id(ofconn));
+ ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
+ error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
+ &ofpacts);
+ if (!error) {
+ error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
+ }
if (error) {
- return error;
+ goto exit_free_ofpacts;
+ }
+
+ /* Record the operation for logging a summary report. */
+ switch (fm.command) {
+ case OFPFC_ADD:
+ ofproto->n_add++;
+ break;
+
+ case OFPFC_MODIFY:
+ case OFPFC_MODIFY_STRICT:
+ ofproto->n_modify++;
+ break;
+
+ case OFPFC_DELETE:
+ case OFPFC_DELETE_STRICT:
+ ofproto->n_delete++;
+ break;
}
- /* 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 ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
+ now = time_msec();
+ if (ofproto->next_op_report == LLONG_MAX) {
+ ofproto->first_op = now;
+ ofproto->next_op_report = MAX(now + 10 * 1000,
+ ofproto->op_backoff);
+ ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
}
+ ofproto->last_op = now;
- return handle_flow_mod__(ofconn_get_ofproto(ofconn), ofconn, &fm, oh);
+exit_free_ofpacts:
+ ofpbuf_uninit(&ofpacts);
+exit:
+ return error;
}
-static int
+static enum ofperr
handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
- const struct ofputil_flow_mod *fm,
- const struct ofp_header *oh)
+ struct ofputil_flow_mod *fm, const struct ofp_header *oh)
{
if (ofproto->n_pending >= 50) {
- assert(!list_is_empty(&ofproto->pending));
+ ovs_assert(!list_is_empty(&ofproto->pending));
return OFPROTO_POSTPONE;
}
default:
if (fm->command > 0xff) {
- VLOG_WARN_RL(&rl, "flow_mod has explicit table_id but "
- "flow_mod_table_id extension is not enabled");
+ VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
+ "flow_mod_table_id extension is not enabled",
+ ofproto->name);
}
- return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
+ return OFPERR_OFPFMFC_BAD_COMMAND;
}
}
-static int
+static enum ofperr
handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
{
- struct nx_role_request *nrr = (struct nx_role_request *) oh;
- struct nx_role_request *reply;
+ struct ofputil_role_request request;
+ struct ofputil_role_request reply;
struct ofpbuf *buf;
- uint32_t role;
+ enum ofperr error;
- if (ofconn_get_type(ofconn) != OFCONN_PRIMARY) {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
+ error = ofputil_decode_role_message(oh, &request);
+ if (error) {
+ return error;
}
- role = ntohl(nrr->role);
- if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
- && role != NX_ROLE_SLAVE) {
- return ofp_mkerr_nicira(OFPET_BAD_REQUEST, NXBRC_BAD_ROLE);
- }
+ if (request.role != OFPCR12_ROLE_NOCHANGE) {
+ if (ofconn_get_role(ofconn) != request.role
+ && ofconn_has_pending_opgroups(ofconn)) {
+ return OFPROTO_POSTPONE;
+ }
- if (ofconn_get_role(ofconn) != role
- && ofconn_has_pending_opgroups(ofconn)) {
- return OFPROTO_POSTPONE;
- }
+ if (request.have_generation_id
+ && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
+ return OFPERR_OFPRRFC_STALE;
+ }
- ofconn_set_role(ofconn, role);
+ ofconn_set_role(ofconn, request.role);
+ }
- reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
- reply->role = htonl(role);
+ reply.role = ofconn_get_role(ofconn);
+ reply.have_generation_id = ofconn_get_master_election_id(
+ ofconn, &reply.generation_id);
+ buf = ofputil_encode_role_reply(oh, &reply);
ofconn_send_reply(ofconn, buf);
return 0;
}
-static int
+static enum ofperr
handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
const struct ofp_header *oh)
{
- const struct nxt_flow_mod_table_id *msg
- = (const struct nxt_flow_mod_table_id *) oh;
+ const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
+ enum ofputil_protocol cur, next;
+
+ cur = ofconn_get_protocol(ofconn);
+ next = ofputil_protocol_set_tid(cur, msg->set != 0);
+ ofconn_set_protocol(ofconn, next);
- ofconn_set_flow_mod_table_id(ofconn, msg->set != 0);
return 0;
}
-static int
+static enum ofperr
handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
{
- const struct nxt_set_flow_format *msg
- = (const struct nxt_set_flow_format *) oh;
- uint32_t format;
+ const struct nx_set_flow_format *msg = ofpmsg_body(oh);
+ enum ofputil_protocol cur, next;
+ enum ofputil_protocol next_base;
- format = ntohl(msg->format);
- if (format != NXFF_OPENFLOW10 && format != NXFF_NXM) {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
+ next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
+ if (!next_base) {
+ return OFPERR_OFPBRC_EPERM;
}
- if (format != ofconn_get_flow_format(ofconn)
- && ofconn_has_pending_opgroups(ofconn)) {
- /* Avoid sending async messages in surprising flow format. */
+ cur = ofconn_get_protocol(ofconn);
+ next = ofputil_protocol_set_base(cur, next_base);
+ if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
+ /* Avoid sending async messages in surprising protocol. */
return OFPROTO_POSTPONE;
}
- ofconn_set_flow_format(ofconn, format);
+ ofconn_set_protocol(ofconn, next);
return 0;
}
-static int
-handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
+static enum ofperr
+handle_nxt_set_packet_in_format(struct ofconn *ofconn,
+ const struct ofp_header *oh)
{
- struct ofp_header *ob;
- struct ofpbuf *buf;
+ const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
+ uint32_t format;
- if (ofconn_has_pending_opgroups(ofconn)) {
+ 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;
}
- ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
- ofconn_send_reply(ofconn, buf);
+ ofconn_set_packet_in_format(ofconn, format);
return 0;
}
-static int
-handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
+static enum ofperr
+handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
{
- const struct ofp_header *oh = msg->data;
- const struct ofputil_msg_type *type;
- int error;
+ const struct nx_async_config *msg = ofpmsg_body(oh);
+ uint32_t master[OAM_N_TYPES];
+ uint32_t slave[OAM_N_TYPES];
- error = ofputil_decode_msg_type(oh, &type);
- if (error) {
- return error;
- }
+ 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]);
- switch (ofputil_msg_type_code(type)) {
- /* OpenFlow requests. */
- case OFPUTIL_OFPT_ECHO_REQUEST:
- return handle_echo_request(ofconn, oh);
+ 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]);
- case OFPUTIL_OFPT_FEATURES_REQUEST:
- return handle_features_request(ofconn, oh);
+ ofconn_set_async_config(ofconn, master, slave);
+ if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
+ !ofconn_get_miss_send_len(ofconn)) {
+ ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
+ }
- case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
- return handle_get_config_request(ofconn, oh);
+ return 0;
+}
- case OFPUTIL_OFPT_SET_CONFIG:
- return handle_set_config(ofconn, msg->data);
+static enum ofperr
+handle_nxt_get_async_request(struct ofconn *ofconn, const struct ofp_header *oh)
+{
+ struct ofpbuf *buf;
+ uint32_t master[OAM_N_TYPES];
+ uint32_t slave[OAM_N_TYPES];
+ struct nx_async_config *msg;
- case OFPUTIL_OFPT_PACKET_OUT:
- return handle_packet_out(ofconn, oh);
+ ofconn_get_async_config(ofconn, master, slave);
+ buf = ofpraw_alloc_reply(OFPRAW_OFPT13_GET_ASYNC_REPLY, oh, 0);
+ msg = ofpbuf_put_zeros(buf, sizeof *msg);
- case OFPUTIL_OFPT_PORT_MOD:
- return handle_port_mod(ofconn, oh);
+ msg->packet_in_mask[0] = htonl(master[OAM_PACKET_IN]);
+ msg->port_status_mask[0] = htonl(master[OAM_PORT_STATUS]);
+ msg->flow_removed_mask[0] = htonl(master[OAM_FLOW_REMOVED]);
- case OFPUTIL_OFPT_FLOW_MOD:
- return handle_flow_mod(ofconn, oh);
+ msg->packet_in_mask[1] = htonl(slave[OAM_PACKET_IN]);
+ msg->port_status_mask[1] = htonl(slave[OAM_PORT_STATUS]);
+ msg->flow_removed_mask[1] = htonl(slave[OAM_FLOW_REMOVED]);
- case OFPUTIL_OFPT_BARRIER_REQUEST:
- return handle_barrier_request(ofconn, oh);
+ ofconn_send_reply(ofconn, buf);
- /* OpenFlow replies. */
- case OFPUTIL_OFPT_ECHO_REPLY:
- return 0;
+ return 0;
+}
- /* Nicira extension requests. */
- case OFPUTIL_NXT_ROLE_REQUEST:
- return handle_role_request(ofconn, oh);
+static enum ofperr
+handle_nxt_set_controller_id(struct ofconn *ofconn,
+ const struct ofp_header *oh)
+{
+ const struct nx_controller_id *nci = ofpmsg_body(oh);
- case OFPUTIL_NXT_FLOW_MOD_TABLE_ID:
- return handle_nxt_flow_mod_table_id(ofconn, oh);
+ if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
+ return OFPERR_NXBRC_MUST_BE_ZERO;
+ }
- case OFPUTIL_NXT_SET_FLOW_FORMAT:
- return handle_nxt_set_flow_format(ofconn, oh);
+ ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
+ return 0;
+}
+
+static enum ofperr
+handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
+{
+ struct ofpbuf *buf;
+
+ if (ofconn_has_pending_opgroups(ofconn)) {
+ return OFPROTO_POSTPONE;
+ }
+
+ buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
+ ? OFPRAW_OFPT10_BARRIER_REPLY
+ : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
+ ofconn_send_reply(ofconn, buf);
+ return 0;
+}
+
+static void
+ofproto_compose_flow_refresh_update(const struct rule *rule,
+ enum nx_flow_monitor_flags flags,
+ struct list *msgs)
+{
+ struct ofoperation *op = rule->pending;
+ struct ofputil_flow_update fu;
+ struct match match;
+
+ if (op && op->type == OFOPERATION_ADD) {
+ /* We'll report the final flow when the operation completes. Reporting
+ * it now would cause a duplicate report later. */
+ return;
+ }
+
+ fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
+ ? NXFME_ADDED : NXFME_MODIFIED);
+ fu.reason = 0;
+ ovs_mutex_lock(&rule->timeout_mutex);
+ fu.idle_timeout = rule->idle_timeout;
+ fu.hard_timeout = rule->hard_timeout;
+ ovs_mutex_unlock(&rule->timeout_mutex);
+ fu.table_id = rule->table_id;
+ fu.cookie = rule->flow_cookie;
+ minimatch_expand(&rule->cr.match, &match);
+ fu.match = &match;
+ fu.priority = rule->cr.priority;
+ if (!(flags & NXFMF_ACTIONS)) {
+ fu.ofpacts = NULL;
+ fu.ofpacts_len = 0;
+ } else if (!op) {
+ fu.ofpacts = rule->ofpacts;
+ fu.ofpacts_len = rule->ofpacts_len;
+ } else {
+ /* An operation is in progress. Use the previous version of the flow's
+ * actions, so that when the operation commits we report the change. */
+ switch (op->type) {
+ case OFOPERATION_ADD:
+ NOT_REACHED();
+
+ case OFOPERATION_MODIFY:
+ case OFOPERATION_REPLACE:
+ if (op->ofpacts) {
+ fu.ofpacts = op->ofpacts;
+ fu.ofpacts_len = op->ofpacts_len;
+ } else {
+ fu.ofpacts = rule->ofpacts;
+ fu.ofpacts_len = rule->ofpacts_len;
+ }
+ break;
+
+ case OFOPERATION_DELETE:
+ fu.ofpacts = rule->ofpacts;
+ fu.ofpacts_len = rule->ofpacts_len;
+ break;
+
+ default:
+ NOT_REACHED();
+ }
+ }
+
+ if (list_is_empty(msgs)) {
+ ofputil_start_flow_update(msgs);
+ }
+ ofputil_append_flow_update(&fu, msgs);
+}
+
+void
+ofmonitor_compose_refresh_updates(struct list *rules, struct list *msgs)
+{
+ struct rule *rule;
+
+ LIST_FOR_EACH (rule, ofproto_node, rules) {
+ enum nx_flow_monitor_flags flags = rule->monitor_flags;
+ rule->monitor_flags = 0;
+
+ ofproto_compose_flow_refresh_update(rule, flags, msgs);
+ }
+}
+
+static void
+ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
+ struct rule *rule, uint64_t seqno,
+ struct list *rules)
+{
+ enum nx_flow_monitor_flags update;
+
+ if (ofproto_rule_is_hidden(rule)) {
+ return;
+ }
+
+ if (!(rule->pending
+ ? ofoperation_has_out_port(rule->pending, m->out_port)
+ : ofproto_rule_has_out_port(rule, m->out_port))) {
+ return;
+ }
+
+ if (seqno) {
+ if (rule->add_seqno > seqno) {
+ update = NXFMF_ADD | NXFMF_MODIFY;
+ } else if (rule->modify_seqno > seqno) {
+ update = NXFMF_MODIFY;
+ } else {
+ return;
+ }
+
+ if (!(m->flags & update)) {
+ return;
+ }
+ } else {
+ update = NXFMF_INITIAL;
+ }
+
+ if (!rule->monitor_flags) {
+ list_push_back(rules, &rule->ofproto_node);
+ }
+ rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
+}
+
+static void
+ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
+ uint64_t seqno,
+ struct list *rules)
+{
+ const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
+ const struct ofoperation *op;
+ const struct oftable *table;
+ struct cls_rule target;
+
+ cls_rule_init_from_minimatch(&target, &m->match, 0);
+ FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
+ struct cls_cursor cursor;
+ struct rule *rule;
+
+ ovs_rwlock_rdlock(&table->cls.rwlock);
+ cls_cursor_init(&cursor, &table->cls, &target);
+ CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
+ ovs_assert(!rule->pending); /* XXX */
+ ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
+ }
+ ovs_rwlock_unlock(&table->cls.rwlock);
+ }
+
+ HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
+ struct rule *rule = op->rule;
+
+ if (((m->table_id == 0xff
+ ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
+ : m->table_id == rule->table_id))
+ && cls_rule_is_loose_match(&rule->cr, &target.match)) {
+ ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
+ }
+ }
+ cls_rule_destroy(&target);
+}
+
+static void
+ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
+ struct list *rules)
+{
+ if (m->flags & NXFMF_INITIAL) {
+ ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
+ }
+}
+
+void
+ofmonitor_collect_resume_rules(struct ofmonitor *m,
+ uint64_t seqno, struct list *rules)
+{
+ ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
+}
+
+static enum ofperr
+handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
+{
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
+ struct ofmonitor **monitors;
+ size_t n_monitors, allocated_monitors;
+ struct list replies;
+ enum ofperr error;
+ struct list rules;
+ struct ofpbuf b;
+ size_t i;
+
+ error = 0;
+ ofpbuf_use_const(&b, oh, ntohs(oh->length));
+ monitors = NULL;
+ n_monitors = allocated_monitors = 0;
+ for (;;) {
+ struct ofputil_flow_monitor_request request;
+ struct ofmonitor *m;
+ int retval;
+
+ retval = ofputil_decode_flow_monitor_request(&request, &b);
+ if (retval == EOF) {
+ break;
+ } else if (retval) {
+ error = retval;
+ goto error;
+ }
+
+ if (request.table_id != 0xff
+ && request.table_id >= ofproto->n_tables) {
+ error = OFPERR_OFPBRC_BAD_TABLE_ID;
+ goto error;
+ }
+
+ error = ofmonitor_create(&request, ofconn, &m);
+ if (error) {
+ goto error;
+ }
+
+ if (n_monitors >= allocated_monitors) {
+ monitors = x2nrealloc(monitors, &allocated_monitors,
+ sizeof *monitors);
+ }
+ monitors[n_monitors++] = m;
+ }
+
+ list_init(&rules);
+ for (i = 0; i < n_monitors; i++) {
+ ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
+ }
+
+ ofpmp_init(&replies, oh);
+ ofmonitor_compose_refresh_updates(&rules, &replies);
+ ofconn_send_replies(ofconn, &replies);
+
+ free(monitors);
+
+ return 0;
+
+error:
+ for (i = 0; i < n_monitors; i++) {
+ ofmonitor_destroy(monitors[i]);
+ }
+ free(monitors);
+ return error;
+}
+
+static enum ofperr
+handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
+{
+ struct ofmonitor *m;
+ uint32_t id;
+
+ id = ofputil_decode_flow_monitor_cancel(oh);
+ m = ofmonitor_lookup(ofconn, id);
+ if (!m) {
+ return OFPERR_NXBRC_FM_BAD_ID;
+ }
+
+ ofmonitor_destroy(m);
+ return 0;
+}
+
+/* Meters implementation.
+ *
+ * Meter table entry, indexed by the OpenFlow meter_id.
+ * These are always dynamically allocated to allocate enough space for
+ * the bands.
+ * 'created' is used to compute the duration for meter stats.
+ * 'list rules' is needed so that we can delete the dependent rules when the
+ * meter table entry is deleted.
+ * 'provider_meter_id' is for the provider's private use.
+ */
+struct meter {
+ long long int created; /* Time created. */
+ struct list rules; /* List of "struct rule_dpif"s. */
+ ofproto_meter_id provider_meter_id;
+ uint16_t flags; /* Meter flags. */
+ uint16_t n_bands; /* Number of meter bands. */
+ struct ofputil_meter_band *bands;
+};
+
+/*
+ * This is used in instruction validation at flow set-up time,
+ * as flows may not use non-existing meters.
+ * This is also used by ofproto-providers to translate OpenFlow meter_ids
+ * in METER instructions to the corresponding provider meter IDs.
+ * Return value of UINT32_MAX signifies an invalid meter.
+ */
+uint32_t
+ofproto_get_provider_meter_id(const struct ofproto * ofproto,
+ uint32_t of_meter_id)
+{
+ if (of_meter_id && of_meter_id <= ofproto->meter_features.max_meters) {
+ const struct meter *meter = ofproto->meters[of_meter_id];
+ if (meter) {
+ return meter->provider_meter_id.uint32;
+ }
+ }
+ return UINT32_MAX;
+}
+
+static void
+meter_update(struct meter *meter, const struct ofputil_meter_config *config)
+{
+ free(meter->bands);
+
+ meter->flags = config->flags;
+ meter->n_bands = config->n_bands;
+ meter->bands = xmemdup(config->bands,
+ config->n_bands * sizeof *meter->bands);
+}
+
+static struct meter *
+meter_create(const struct ofputil_meter_config *config,
+ ofproto_meter_id provider_meter_id)
+{
+ struct meter *meter;
+
+ meter = xzalloc(sizeof *meter);
+ meter->provider_meter_id = provider_meter_id;
+ meter->created = time_msec();
+ list_init(&meter->rules);
+
+ meter_update(meter, config);
+
+ return meter;
+}
+
+static void
+meter_delete(struct ofproto *ofproto, uint32_t first, uint32_t last)
+{
+ uint32_t mid;
+ for (mid = first; mid <= last; ++mid) {
+ struct meter *meter = ofproto->meters[mid];
+ if (meter) {
+ ofproto->meters[mid] = NULL;
+ ofproto->ofproto_class->meter_del(ofproto,
+ meter->provider_meter_id);
+ free(meter->bands);
+ free(meter);
+ }
+ }
+}
+
+static enum ofperr
+handle_add_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
+{
+ ofproto_meter_id provider_meter_id = { UINT32_MAX };
+ struct meter **meterp = &ofproto->meters[mm->meter.meter_id];
+ enum ofperr error;
+
+ if (*meterp) {
+ return OFPERR_OFPMMFC_METER_EXISTS;
+ }
+
+ error = ofproto->ofproto_class->meter_set(ofproto, &provider_meter_id,
+ &mm->meter);
+ if (!error) {
+ ovs_assert(provider_meter_id.uint32 != UINT32_MAX);
+ *meterp = meter_create(&mm->meter, provider_meter_id);
+ }
+ return 0;
+}
+
+static enum ofperr
+handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
+{
+ struct meter *meter = ofproto->meters[mm->meter.meter_id];
+ enum ofperr error;
+
+ if (!meter) {
+ return OFPERR_OFPMMFC_UNKNOWN_METER;
+ }
+
+ error = ofproto->ofproto_class->meter_set(ofproto,
+ &meter->provider_meter_id,
+ &mm->meter);
+ ovs_assert(meter->provider_meter_id.uint32 != UINT32_MAX);
+ if (!error) {
+ meter_update(meter, &mm->meter);
+ }
+ return error;
+}
+
+static enum ofperr
+handle_delete_meter(struct ofconn *ofconn, const struct ofp_header *oh,
+ struct ofputil_meter_mod *mm)
+{
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
+ uint32_t meter_id = mm->meter.meter_id;
+ uint32_t first, last;
+ struct list rules;
+
+ if (meter_id == OFPM13_ALL) {
+ first = 1;
+ last = ofproto->meter_features.max_meters;
+ } else {
+ if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
+ return 0;
+ }
+ first = last = meter_id;
+ }
+
+ /* First delete the rules that use this meter. If any of those rules are
+ * currently being modified, postpone the whole operation until later. */
+ list_init(&rules);
+ for (meter_id = first; meter_id <= last; ++meter_id) {
+ struct meter *meter = ofproto->meters[meter_id];
+ if (meter && !list_is_empty(&meter->rules)) {
+ struct rule *rule;
+
+ LIST_FOR_EACH (rule, meter_list_node, &meter->rules) {
+ if (rule->pending) {
+ return OFPROTO_POSTPONE;
+ }
+ list_push_back(&rules, &rule->ofproto_node);
+ }
+ }
+ }
+ if (!list_is_empty(&rules)) {
+ delete_flows__(ofproto, ofconn, oh, &rules, OFPRR_METER_DELETE);
+ }
+
+ /* Delete the meters. */
+ meter_delete(ofproto, first, last);
+
+ return 0;
+}
+
+static enum ofperr
+handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh)
+{
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
+ struct ofputil_meter_mod mm;
+ uint64_t bands_stub[256 / 8];
+ struct ofpbuf bands;
+ uint32_t meter_id;
+ enum ofperr error;
+
+ error = reject_slave_controller(ofconn);
+ if (error) {
+ return error;
+ }
+
+ ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
+
+ error = ofputil_decode_meter_mod(oh, &mm, &bands);
+ if (error) {
+ goto exit_free_bands;
+ }
+
+ meter_id = mm.meter.meter_id;
+
+ if (mm.command != OFPMC13_DELETE) {
+ /* Fails also when meters are not implemented by the provider. */
+ if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
+ error = OFPERR_OFPMMFC_INVALID_METER;
+ goto exit_free_bands;
+ }
+ if (mm.meter.n_bands > ofproto->meter_features.max_bands) {
+ error = OFPERR_OFPMMFC_OUT_OF_BANDS;
+ goto exit_free_bands;
+ }
+ }
+
+ switch (mm.command) {
+ case OFPMC13_ADD:
+ error = handle_add_meter(ofproto, &mm);
+ break;
+
+ case OFPMC13_MODIFY:
+ error = handle_modify_meter(ofproto, &mm);
+ break;
+
+ case OFPMC13_DELETE:
+ error = handle_delete_meter(ofconn, oh, &mm);
+ break;
+
+ default:
+ error = OFPERR_OFPMMFC_BAD_COMMAND;
+ break;
+ }
+
+exit_free_bands:
+ ofpbuf_uninit(&bands);
+ return error;
+}
+
+static enum ofperr
+handle_meter_features_request(struct ofconn *ofconn,
+ const struct ofp_header *request)
+{
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
+ struct ofputil_meter_features features;
+ struct ofpbuf *b;
+
+ if (ofproto->ofproto_class->meter_get_features) {
+ ofproto->ofproto_class->meter_get_features(ofproto, &features);
+ } else {
+ memset(&features, 0, sizeof features);
+ }
+ b = ofputil_encode_meter_features_reply(&features, request);
+
+ ofconn_send_reply(ofconn, b);
+ return 0;
+}
+
+static enum ofperr
+handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request,
+ enum ofptype type)
+{
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
+ struct list replies;
+ uint64_t bands_stub[256 / 8];
+ struct ofpbuf bands;
+ uint32_t meter_id, first, last;
+
+ ofputil_decode_meter_request(request, &meter_id);
+
+ if (meter_id == OFPM13_ALL) {
+ first = 1;
+ last = ofproto->meter_features.max_meters;
+ } else {
+ if (!meter_id || meter_id > ofproto->meter_features.max_meters ||
+ !ofproto->meters[meter_id]) {
+ return OFPERR_OFPMMFC_UNKNOWN_METER;
+ }
+ first = last = meter_id;
+ }
+
+ ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
+ ofpmp_init(&replies, request);
+
+ for (meter_id = first; meter_id <= last; ++meter_id) {
+ struct meter *meter = ofproto->meters[meter_id];
+ if (!meter) {
+ continue; /* Skip non-existing meters. */
+ }
+ if (type == OFPTYPE_METER_STATS_REQUEST) {
+ struct ofputil_meter_stats stats;
+
+ stats.meter_id = meter_id;
+
+ /* Provider sets the packet and byte counts, we do the rest. */
+ stats.flow_count = list_size(&meter->rules);
+ calc_duration(meter->created, time_msec(),
+ &stats.duration_sec, &stats.duration_nsec);
+ stats.n_bands = meter->n_bands;
+ ofpbuf_clear(&bands);
+ stats.bands
+ = ofpbuf_put_uninit(&bands,
+ meter->n_bands * sizeof *stats.bands);
+
+ if (!ofproto->ofproto_class->meter_get(ofproto,
+ meter->provider_meter_id,
+ &stats)) {
+ ofputil_append_meter_stats(&replies, &stats);
+ }
+ } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */
+ struct ofputil_meter_config config;
+
+ config.meter_id = meter_id;
+ config.flags = meter->flags;
+ config.n_bands = meter->n_bands;
+ config.bands = meter->bands;
+ ofputil_append_meter_config(&replies, &config);
+ }
+ }
+
+ ofconn_send_replies(ofconn, &replies);
+ ofpbuf_uninit(&bands);
+ return 0;
+}
+
+bool
+ofproto_group_lookup(const struct ofproto *ofproto, uint32_t group_id,
+ struct ofgroup **group)
+ OVS_TRY_RDLOCK(true, (*group)->rwlock)
+{
+ ovs_rwlock_rdlock(&ofproto->groups_rwlock);
+ HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
+ hash_int(group_id, 0), &ofproto->groups) {
+ if ((*group)->group_id == group_id) {
+ ovs_rwlock_rdlock(&(*group)->rwlock);
+ ovs_rwlock_unlock(&ofproto->groups_rwlock);
+ return true;
+ }
+ }
+ ovs_rwlock_unlock(&ofproto->groups_rwlock);
+ return false;
+}
+
+void
+ofproto_group_release(struct ofgroup *group)
+ OVS_RELEASES(group->rwlock)
+{
+ ovs_rwlock_unlock(&group->rwlock);
+}
+
+static bool
+ofproto_group_write_lookup(const struct ofproto *ofproto, uint32_t group_id,
+ struct ofgroup **group)
+ OVS_TRY_WRLOCK(true, ofproto->groups_rwlock)
+ OVS_TRY_WRLOCK(true, (*group)->rwlock)
+{
+ ovs_rwlock_wrlock(&ofproto->groups_rwlock);
+ HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
+ hash_int(group_id, 0), &ofproto->groups) {
+ if ((*group)->group_id == group_id) {
+ ovs_rwlock_wrlock(&(*group)->rwlock);
+ return true;
+ }
+ }
+ ovs_rwlock_unlock(&ofproto->groups_rwlock);
+ return false;
+}
+
+static bool
+ofproto_group_exists(const struct ofproto *ofproto, uint32_t group_id)
+ OVS_REQ_RDLOCK(ofproto->groups_rwlock)
+{
+ struct ofgroup *grp;
+
+ HMAP_FOR_EACH_IN_BUCKET (grp, hmap_node,
+ hash_int(group_id, 0), &ofproto->groups) {
+ if (grp->group_id == group_id) {
+ return true;
+ }
+ }
+ return false;
+}
+
+static void
+append_group_stats(struct ofgroup *group, struct list *replies)
+ OVS_REQ_RDLOCK(group->rwlock)
+{
+ struct ofputil_group_stats ogs;
+ struct ofproto *ofproto = group->ofproto;
+ long long int now = time_msec();
+ int error;
+
+ ogs.bucket_stats = xmalloc(group->n_buckets * sizeof *ogs.bucket_stats);
+
+ error = (ofproto->ofproto_class->group_get_stats
+ ? ofproto->ofproto_class->group_get_stats(group, &ogs)
+ : EOPNOTSUPP);
+ if (error) {
+ ogs.ref_count = UINT32_MAX;
+ ogs.packet_count = UINT64_MAX;
+ ogs.byte_count = UINT64_MAX;
+ ogs.n_buckets = group->n_buckets;
+ memset(ogs.bucket_stats, 0xff,
+ ogs.n_buckets * sizeof *ogs.bucket_stats);
+ }
+
+ ogs.group_id = group->group_id;
+ calc_duration(group->created, now, &ogs.duration_sec, &ogs.duration_nsec);
+
+ ofputil_append_group_stats(replies, &ogs);
+
+ free(ogs.bucket_stats);
+}
+
+static enum ofperr
+handle_group_stats_request(struct ofconn *ofconn,
+ const struct ofp_header *request)
+{
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
+ struct list replies;
+ enum ofperr error;
+ struct ofgroup *group;
+ uint32_t group_id;
+
+ error = ofputil_decode_group_stats_request(request, &group_id);
+ if (error) {
+ return error;
+ }
+
+ ofpmp_init(&replies, request);
+
+ if (group_id == OFPG_ALL) {
+ ovs_rwlock_rdlock(&ofproto->groups_rwlock);
+ HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
+ ovs_rwlock_rdlock(&group->rwlock);
+ append_group_stats(group, &replies);
+ ovs_rwlock_unlock(&group->rwlock);
+ }
+ ovs_rwlock_unlock(&ofproto->groups_rwlock);
+ } else {
+ if (ofproto_group_lookup(ofproto, group_id, &group)) {
+ append_group_stats(group, &replies);
+ ofproto_group_release(group);
+ }
+ }
+
+ ofconn_send_replies(ofconn, &replies);
+
+ return 0;
+}
+
+static enum ofperr
+handle_group_desc_stats_request(struct ofconn *ofconn,
+ const struct ofp_header *request)
+{
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
+ struct list replies;
+ struct ofputil_group_desc gds;
+ struct ofgroup *group;
+
+ ofpmp_init(&replies, request);
+
+ ovs_rwlock_rdlock(&ofproto->groups_rwlock);
+ HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
+ gds.group_id = group->group_id;
+ gds.type = group->type;
+ ofputil_append_group_desc_reply(&gds, &group->buckets, &replies);
+ }
+ ovs_rwlock_unlock(&ofproto->groups_rwlock);
+
+ ofconn_send_replies(ofconn, &replies);
+
+ return 0;
+}
+
+static enum ofperr
+handle_group_features_stats_request(struct ofconn *ofconn,
+ const struct ofp_header *request)
+{
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
+ struct ofpbuf *msg;
+
+ msg = ofputil_encode_group_features_reply(&p->ogf, request);
+ if (msg) {
+ ofconn_send_reply(ofconn, msg);
+ }
+
+ return 0;
+}
+
+/* Implements OFPGC11_ADD
+ * 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.
+ *
+ * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
+ * ownership remains with the caller.
+ *
+ * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
+ * if any. */
+static enum ofperr
+add_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
+{
+ struct ofgroup *ofgroup;
+ enum ofperr error;
+
+ if (gm->group_id > OFPG_MAX) {
+ return OFPERR_OFPGMFC_INVALID_GROUP;
+ }
+ if (gm->type > OFPGT11_FF) {
+ return OFPERR_OFPGMFC_BAD_TYPE;
+ }
+
+ /* Allocate new group and initialize it. */
+ ofgroup = ofproto->ofproto_class->group_alloc();
+ if (!ofgroup) {
+ VLOG_WARN_RL(&rl, "%s: failed to create group", ofproto->name);
+ return OFPERR_OFPGMFC_OUT_OF_GROUPS;
+ }
+
+ ovs_rwlock_init(&ofgroup->rwlock);
+ ofgroup->ofproto = ofproto;
+ ofgroup->group_id = gm->group_id;
+ ofgroup->type = gm->type;
+ ofgroup->created = ofgroup->modified = time_msec();
+
+ list_move(&ofgroup->buckets, &gm->buckets);
+ ofgroup->n_buckets = list_size(&ofgroup->buckets);
+
+ /* Construct called BEFORE any locks are held. */
+ error = ofproto->ofproto_class->group_construct(ofgroup);
+ if (error) {
+ goto free_out;
+ }
+
+ /* We wrlock as late as possible to minimize the time we jam any other
+ * threads: No visible state changes before acquiring the lock. */
+ ovs_rwlock_wrlock(&ofproto->groups_rwlock);
+
+ if (ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
+ error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
+ goto unlock_out;
+ }
+
+ if (ofproto_group_exists(ofproto, gm->group_id)) {
+ error = OFPERR_OFPGMFC_GROUP_EXISTS;
+ goto unlock_out;
+ }
+
+ if (!error) {
+ /* Insert new group. */
+ hmap_insert(&ofproto->groups, &ofgroup->hmap_node,
+ hash_int(ofgroup->group_id, 0));
+ ofproto->n_groups[ofgroup->type]++;
+
+ ovs_rwlock_unlock(&ofproto->groups_rwlock);
+ return error;
+ }
+
+ unlock_out:
+ ovs_rwlock_unlock(&ofproto->groups_rwlock);
+ ofproto->ofproto_class->group_destruct(ofgroup);
+ free_out:
+ ofputil_bucket_list_destroy(&ofgroup->buckets);
+ ofproto->ofproto_class->group_dealloc(ofgroup);
+
+ return error;
+}
+
+/* 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_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
+{
+ struct ofgroup *ofgroup;
+ struct ofgroup *victim;
+ enum ofperr error;
+
+ if (gm->group_id > OFPG_MAX) {
+ return OFPERR_OFPGMFC_INVALID_GROUP;
+ }
+
+ if (gm->type > OFPGT11_FF) {
+ return OFPERR_OFPGMFC_BAD_TYPE;
+ }
+
+ victim = ofproto->ofproto_class->group_alloc();
+ if (!victim) {
+ VLOG_WARN_RL(&rl, "%s: failed to allocate group", ofproto->name);
+ return OFPERR_OFPGMFC_OUT_OF_GROUPS;
+ }
+
+ if (!ofproto_group_write_lookup(ofproto, gm->group_id, &ofgroup)) {
+ error = OFPERR_OFPGMFC_UNKNOWN_GROUP;
+ goto free_out;
+ }
+ /* Both group's and its container's write locks held now.
+ * Also, n_groups[] is protected by ofproto->groups_rwlock. */
+ if (ofgroup->type != gm->type
+ && ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
+ error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
+ goto unlock_out;
+ }
+
+ *victim = *ofgroup;
+ list_move(&victim->buckets, &ofgroup->buckets);
+
+ ofgroup->type = gm->type;
+ list_move(&ofgroup->buckets, &gm->buckets);
+ ofgroup->n_buckets = list_size(&ofgroup->buckets);
+
+ error = ofproto->ofproto_class->group_modify(ofgroup, victim);
+ if (!error) {
+ ofputil_bucket_list_destroy(&victim->buckets);
+ ofproto->n_groups[victim->type]--;
+ ofproto->n_groups[ofgroup->type]++;
+ ofgroup->modified = time_msec();
+ } else {
+ ofputil_bucket_list_destroy(&ofgroup->buckets);
+
+ *ofgroup = *victim;
+ list_move(&ofgroup->buckets, &victim->buckets);
+ }
+
+ unlock_out:
+ ovs_rwlock_unlock(&ofgroup->rwlock);
+ ovs_rwlock_unlock(&ofproto->groups_rwlock);
+ free_out:
+ ofproto->ofproto_class->group_dealloc(victim);
+ return error;
+}
+
+static void
+delete_group__(struct ofproto *ofproto, struct ofgroup *ofgroup)
+ OVS_RELEASES(ofproto->groups_rwlock)
+{
+ /* Must wait until existing readers are done,
+ * while holding the container's write lock at the same time. */
+ ovs_rwlock_wrlock(&ofgroup->rwlock);
+ hmap_remove(&ofproto->groups, &ofgroup->hmap_node);
+ /* No-one can find this group any more. */
+ ofproto->n_groups[ofgroup->type]--;
+ ovs_rwlock_unlock(&ofproto->groups_rwlock);
+
+ ofproto->ofproto_class->group_destruct(ofgroup);
+ ofputil_bucket_list_destroy(&ofgroup->buckets);
+ ovs_rwlock_unlock(&ofgroup->rwlock);
+ ovs_rwlock_destroy(&ofgroup->rwlock);
+ ofproto->ofproto_class->group_dealloc(ofgroup);
+}
+
+/* Implements OFPGC_DELETE. */
+static void
+delete_group(struct ofproto *ofproto, uint32_t group_id)
+{
+ struct ofgroup *ofgroup;
+
+ ovs_rwlock_wrlock(&ofproto->groups_rwlock);
+ if (group_id == OFPG_ALL) {
+ for (;;) {
+ struct hmap_node *node = hmap_first(&ofproto->groups);
+ if (!node) {
+ break;
+ }
+ ofgroup = CONTAINER_OF(node, struct ofgroup, hmap_node);
+ delete_group__(ofproto, ofgroup);
+ /* Lock for each node separately, so that we will not jam the
+ * other threads for too long time. */
+ ovs_rwlock_wrlock(&ofproto->groups_rwlock);
+ }
+ } else {
+ HMAP_FOR_EACH_IN_BUCKET (ofgroup, hmap_node,
+ hash_int(group_id, 0), &ofproto->groups) {
+ if (ofgroup->group_id == group_id) {
+ delete_group__(ofproto, ofgroup);
+ return;
+ }
+ }
+ }
+ ovs_rwlock_unlock(&ofproto->groups_rwlock);
+}
+
+static enum ofperr
+handle_group_mod(struct ofconn *ofconn, const struct ofp_header *oh)
+{
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
+ struct ofputil_group_mod gm;
+ enum ofperr error;
+
+ error = reject_slave_controller(ofconn);
+ if (error) {
+ return error;
+ }
+
+ error = ofputil_decode_group_mod(oh, &gm);
+ if (error) {
+ return error;
+ }
+
+ switch (gm.command) {
+ case OFPGC11_ADD:
+ return add_group(ofproto, &gm);
+
+ case OFPGC11_MODIFY:
+ return modify_group(ofproto, &gm);
+
+ case OFPGC11_DELETE:
+ delete_group(ofproto, gm.group_id);
+ return 0;
+
+ default:
+ if (gm.command > OFPGC11_DELETE) {
+ VLOG_WARN_RL(&rl, "%s: Invalid group_mod command type %d",
+ ofproto->name, gm.command);
+ }
+ return OFPERR_OFPGMFC_BAD_COMMAND;
+ }
+}
+
+static enum ofperr
+handle_table_mod(struct ofconn *ofconn, const struct ofp_header *oh)
+{
+ struct ofputil_table_mod tm;
+ enum ofperr error;
+
+ error = reject_slave_controller(ofconn);
+ if (error) {
+ return error;
+ }
+
+ error = ofputil_decode_table_mod(oh, &tm);
+ if (error) {
+ return error;
+ }
+
+ /* XXX Actual table mod support is not implemented yet. */
+ return 0;
+}
+
+static enum ofperr
+handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
+{
+ const struct ofp_header *oh = msg->data;
+ enum ofptype type;
+ enum ofperr error;
+
+ error = ofptype_decode(&type, oh);
+ if (error) {
+ return error;
+ }
+
+ switch (type) {
+ /* OpenFlow requests. */
+ case OFPTYPE_ECHO_REQUEST:
+ return handle_echo_request(ofconn, oh);
+
+ case OFPTYPE_FEATURES_REQUEST:
+ return handle_features_request(ofconn, oh);
+
+ case OFPTYPE_GET_CONFIG_REQUEST:
+ return handle_get_config_request(ofconn, oh);
+
+ case OFPTYPE_SET_CONFIG:
+ return handle_set_config(ofconn, oh);
+
+ case OFPTYPE_PACKET_OUT:
+ return handle_packet_out(ofconn, oh);
+
+ case OFPTYPE_PORT_MOD:
+ return handle_port_mod(ofconn, oh);
+
+ case OFPTYPE_FLOW_MOD:
+ return handle_flow_mod(ofconn, oh);
+
+ case OFPTYPE_GROUP_MOD:
+ return handle_group_mod(ofconn, oh);
+
+ case OFPTYPE_TABLE_MOD:
+ return handle_table_mod(ofconn, oh);
+
+ case OFPTYPE_METER_MOD:
+ return handle_meter_mod(ofconn, oh);
+
+ case OFPTYPE_BARRIER_REQUEST:
+ return handle_barrier_request(ofconn, oh);
+
+ case OFPTYPE_ROLE_REQUEST:
+ return handle_role_request(ofconn, oh);
+
+ /* OpenFlow replies. */
+ case OFPTYPE_ECHO_REPLY:
+ return 0;
+
+ /* Nicira extension requests. */
+ case OFPTYPE_FLOW_MOD_TABLE_ID:
+ return handle_nxt_flow_mod_table_id(ofconn, oh);
- case OFPUTIL_NXT_FLOW_MOD:
- return handle_flow_mod(ofconn, oh);
+ case OFPTYPE_SET_FLOW_FORMAT:
+ return handle_nxt_set_flow_format(ofconn, oh);
+
+ case OFPTYPE_SET_PACKET_IN_FORMAT:
+ return handle_nxt_set_packet_in_format(ofconn, oh);
+
+ case OFPTYPE_SET_CONTROLLER_ID:
+ return handle_nxt_set_controller_id(ofconn, oh);
+
+ case OFPTYPE_FLOW_AGE:
+ /* Nothing to do. */
+ return 0;
+
+ case OFPTYPE_FLOW_MONITOR_CANCEL:
+ return handle_flow_monitor_cancel(ofconn, oh);
+
+ case OFPTYPE_SET_ASYNC_CONFIG:
+ return handle_nxt_set_async_config(ofconn, oh);
+
+ case OFPTYPE_GET_ASYNC_REQUEST:
+ return handle_nxt_get_async_request(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_NXST_FLOW_REPLY:
- case OFPUTIL_NXST_AGGREGATE_REPLY:
+ case OFPTYPE_DESC_STATS_REQUEST:
+ return handle_desc_stats_request(ofconn, oh);
+
+ case OFPTYPE_FLOW_STATS_REQUEST:
+ return handle_flow_stats_request(ofconn, oh);
+
+ case OFPTYPE_AGGREGATE_STATS_REQUEST:
+ return handle_aggregate_stats_request(ofconn, oh);
+
+ case OFPTYPE_TABLE_STATS_REQUEST:
+ return handle_table_stats_request(ofconn, oh);
+
+ case OFPTYPE_PORT_STATS_REQUEST:
+ return handle_port_stats_request(ofconn, oh);
+
+ case OFPTYPE_QUEUE_STATS_REQUEST:
+ return handle_queue_stats_request(ofconn, oh);
+
+ case OFPTYPE_PORT_DESC_STATS_REQUEST:
+ return handle_port_desc_stats_request(ofconn, oh);
+
+ case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
+ return handle_flow_monitor_request(ofconn, oh);
+
+ case OFPTYPE_METER_STATS_REQUEST:
+ case OFPTYPE_METER_CONFIG_STATS_REQUEST:
+ return handle_meter_request(ofconn, oh, type);
+
+ case OFPTYPE_METER_FEATURES_STATS_REQUEST:
+ return handle_meter_features_request(ofconn, oh);
+
+ case OFPTYPE_GROUP_STATS_REQUEST:
+ return handle_group_stats_request(ofconn, oh);
+
+ case OFPTYPE_GROUP_DESC_STATS_REQUEST:
+ return handle_group_desc_stats_request(ofconn, oh);
+
+ case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
+ return handle_group_features_stats_request(ofconn, oh);
+
+ /* FIXME: Change the following once they are implemented: */
+ case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
+ case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
+ return OFPERR_OFPBRC_BAD_TYPE;
+
+ case OFPTYPE_HELLO:
+ case OFPTYPE_ERROR:
+ case OFPTYPE_FEATURES_REPLY:
+ case OFPTYPE_GET_CONFIG_REPLY:
+ case OFPTYPE_PACKET_IN:
+ case OFPTYPE_FLOW_REMOVED:
+ case OFPTYPE_PORT_STATUS:
+ case OFPTYPE_BARRIER_REPLY:
+ case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
+ case OFPTYPE_DESC_STATS_REPLY:
+ case OFPTYPE_FLOW_STATS_REPLY:
+ case OFPTYPE_QUEUE_STATS_REPLY:
+ case OFPTYPE_PORT_STATS_REPLY:
+ case OFPTYPE_TABLE_STATS_REPLY:
+ case OFPTYPE_AGGREGATE_STATS_REPLY:
+ case OFPTYPE_PORT_DESC_STATS_REPLY:
+ case OFPTYPE_ROLE_REPLY:
+ case OFPTYPE_FLOW_MONITOR_PAUSED:
+ case OFPTYPE_FLOW_MONITOR_RESUMED:
+ case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
+ case OFPTYPE_GET_ASYNC_REPLY:
+ case OFPTYPE_GROUP_STATS_REPLY:
+ case OFPTYPE_GROUP_DESC_STATS_REPLY:
+ case OFPTYPE_GROUP_FEATURES_STATS_REPLY:
+ case OFPTYPE_METER_STATS_REPLY:
+ case OFPTYPE_METER_CONFIG_STATS_REPLY:
+ case OFPTYPE_METER_FEATURES_STATS_REPLY:
+ case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
default:
- if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
- } else {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
- }
+ return OFPERR_OFPBRC_BAD_TYPE;
}
}
static bool
-handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
+handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
{
int error = handle_openflow__(ofconn, ofp_msg);
if (error && error != OFPROTO_POSTPONE) {
if (ofconn) {
size_t request_len = ntohs(request->length);
- assert(ofconn_get_ofproto(ofconn) == ofproto);
+ ovs_assert(ofconn_get_ofproto(ofconn) == ofproto);
ofconn_add_opgroup(ofconn, &group->ofconn_node);
group->ofconn = ofconn;
static void
ofopgroup_submit(struct ofopgroup *group)
{
- if (list_is_empty(&group->ops)) {
- ofopgroup_destroy(group);
+ if (!group->n_running) {
+ ofopgroup_complete(group);
} else {
list_push_back(&group->ofproto->pending, &group->ofproto_node);
group->ofproto->n_pending++;
}
static void
-ofopgroup_destroy(struct ofopgroup *group)
+ofopgroup_complete(struct ofopgroup *group)
{
- assert(list_is_empty(&group->ops));
+ struct ofproto *ofproto = group->ofproto;
+
+ struct ofconn *abbrev_ofconn;
+ ovs_be32 abbrev_xid;
+
+ struct ofoperation *op, *next_op;
+ int error;
+
+ ovs_assert(!group->n_running);
+
+ error = 0;
+ LIST_FOR_EACH (op, group_node, &group->ops) {
+ if (op->error) {
+ error = op->error;
+ break;
+ }
+ }
+
+ if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
+ LIST_FOR_EACH (op, group_node, &group->ops) {
+ if (op->type != OFOPERATION_DELETE) {
+ struct ofpbuf *packet;
+ ofp_port_t in_port;
+
+ error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
+ &packet, &in_port);
+ if (packet) {
+ ovs_assert(!error);
+ error = rule_execute(op->rule, in_port, packet);
+ }
+ break;
+ }
+ }
+ }
+
+ if (!error && !list_is_empty(&group->ofconn_node)) {
+ abbrev_ofconn = group->ofconn;
+ abbrev_xid = group->request->xid;
+ } else {
+ abbrev_ofconn = NULL;
+ abbrev_xid = htonl(0);
+ }
+ LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
+ struct rule *rule = op->rule;
+
+ /* We generally want to report the change to active OpenFlow flow
+ monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
+
+ - The operation failed.
+
+ - The affected rule is not visible to controllers.
+
+ - The operation's only effect was to update rule->modified. */
+ if (!(op->error
+ || ofproto_rule_is_hidden(rule)
+ || (op->type == OFOPERATION_MODIFY
+ && op->ofpacts
+ && rule->flow_cookie == op->flow_cookie))) {
+ /* Check that we can just cast from ofoperation_type to
+ * nx_flow_update_event. */
+ enum nx_flow_update_event event_type;
+
+ switch (op->type) {
+ case OFOPERATION_ADD:
+ case OFOPERATION_REPLACE:
+ event_type = NXFME_ADDED;
+ break;
+
+ case OFOPERATION_DELETE:
+ event_type = NXFME_DELETED;
+ break;
+
+ case OFOPERATION_MODIFY:
+ event_type = NXFME_MODIFIED;
+ break;
+
+ default:
+ NOT_REACHED();
+ }
+
+ ofmonitor_report(ofproto->connmgr, rule, event_type,
+ op->reason, abbrev_ofconn, abbrev_xid);
+ }
+
+ rule->pending = NULL;
+
+ switch (op->type) {
+ case OFOPERATION_ADD:
+ if (!op->error) {
+ uint16_t vid_mask;
+
+ vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
+ if (vid_mask == VLAN_VID_MASK) {
+ if (ofproto->vlan_bitmap) {
+ uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
+ if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
+ bitmap_set1(ofproto->vlan_bitmap, vid);
+ ofproto->vlans_changed = true;
+ }
+ } else {
+ ofproto->vlans_changed = true;
+ }
+ }
+ } else {
+ ovs_rwlock_wrlock(&rule->rwlock);
+ oftable_remove_rule(rule);
+ ofproto_rule_destroy(rule);
+ }
+ break;
+
+ case OFOPERATION_DELETE:
+ ovs_assert(!op->error);
+ ofproto_rule_destroy(rule);
+ op->rule = NULL;
+ break;
+
+ case OFOPERATION_MODIFY:
+ case OFOPERATION_REPLACE:
+ if (!op->error) {
+ long long int now = time_msec();
+
+ rule->modified = now;
+ if (op->type == OFOPERATION_REPLACE) {
+ rule->created = rule->used = now;
+ }
+ } else {
+ ofproto_rule_change_cookie(ofproto, rule, op->flow_cookie);
+ ovs_mutex_lock(&rule->timeout_mutex);
+ rule->idle_timeout = op->idle_timeout;
+ rule->hard_timeout = op->hard_timeout;
+ ovs_mutex_unlock(&rule->timeout_mutex);
+ if (op->ofpacts) {
+ free(rule->ofpacts);
+
+ ovs_rwlock_wrlock(&rule->rwlock);
+ rule->ofpacts = op->ofpacts;
+ rule->ofpacts_len = op->ofpacts_len;
+ ovs_rwlock_unlock(&rule->rwlock);
+
+ op->ofpacts = NULL;
+ op->ofpacts_len = 0;
+ }
+ rule->send_flow_removed = op->send_flow_removed;
+ }
+ break;
+
+ default:
+ NOT_REACHED();
+ }
+
+ ofoperation_destroy(op);
+ }
+
+ ofmonitor_flush(ofproto->connmgr);
+
if (!list_is_empty(&group->ofproto_node)) {
- assert(group->ofproto->n_pending > 0);
- group->ofproto->n_pending--;
+ ovs_assert(ofproto->n_pending > 0);
+ 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);
+ if (error) {
+ ofconn_send_error(group->ofconn, group->request, error);
}
- connmgr_retry(group->ofproto->connmgr);
+ connmgr_retry(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
+ * 'group'. Prior to calling, 'rule' must not have any pending operation.
+ *
+ * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
+ * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
+ *
+ * Returns the newly created ofoperation (which is also available as
+ * rule->pending). */
+static struct ofoperation *
ofoperation_create(struct ofopgroup *group, struct rule *rule,
- enum ofoperation_type type)
+ enum ofoperation_type type,
+ enum ofp_flow_removed_reason reason)
{
+ struct ofproto *ofproto = group->ofproto;
struct ofoperation *op;
- assert(!rule->pending);
+ ovs_assert(!rule->pending);
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->reason = reason;
op->flow_cookie = rule->flow_cookie;
+ ovs_mutex_lock(&rule->timeout_mutex);
+ op->idle_timeout = rule->idle_timeout;
+ op->hard_timeout = rule->hard_timeout;
+ ovs_mutex_unlock(&rule->timeout_mutex);
+ op->send_flow_removed = rule->send_flow_removed;
+
+ group->n_running++;
if (type == OFOPERATION_DELETE) {
- hmap_insert(&op->group->ofproto->deletions, &op->hmap_node,
+ hmap_insert(&ofproto->deletions, &op->hmap_node,
cls_rule_hash(&rule->cr, rule->table_id));
}
+
+ return op;
}
static void
hmap_remove(&group->ofproto->deletions, &op->hmap_node);
}
list_remove(&op->group_node);
- free(op->actions);
+ free(op->ofpacts);
free(op);
-
- if (list_is_empty(&group->ops) && !list_is_empty(&group->ofproto_node)) {
- ofopgroup_destroy(group);
- }
}
/* Indicates that 'op' completed with status 'error', which is either 0 to
- * indicate success or an OpenFlow error code (constructed with
- * e.g. ofp_mkerr()).
+ * 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.
+ * permanently to the flow table.
*
* If 'error' is nonzero, then generally the operation will be rolled back:
*
* Please see the large comment in ofproto/ofproto-provider.h titled
* "Asynchronous Operation Support" for more information. */
void
-ofoperation_complete(struct ofoperation *op, int error)
+ofoperation_complete(struct ofoperation *op, enum ofperr error)
{
struct ofopgroup *group = op->group;
- struct rule *rule = op->rule;
- struct ofproto *ofproto = rule->ofproto;
- struct classifier *table = &ofproto->tables[rule->table_id];
- assert(rule->pending == op);
- assert(op->status < 0);
- assert(error >= 0);
+ ovs_assert(op->rule->pending == op);
+ ovs_assert(group->n_running > 0);
+ ovs_assert(!error || op->type != OFOPERATION_DELETE);
+
+ op->error = error;
+ if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
+ ofopgroup_complete(group);
+ }
+}
+\f
+static uint64_t
+pick_datapath_id(const struct ofproto *ofproto)
+{
+ const struct ofport *port;
- 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;
+ port = ofproto_get_port(ofproto, OFPP_LOCAL);
+ if (port) {
+ uint8_t ea[ETH_ADDR_LEN];
+ int error;
- error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
- &packet, &in_port);
- if (packet) {
- assert(!error);
- error = rule_execute(rule, in_port, packet);
+ error = netdev_get_etheraddr(port->netdev, ea);
+ if (!error) {
+ return eth_addr_to_uint64(ea);
}
+ VLOG_WARN("%s: could not get MAC address for %s (%s)",
+ ofproto->name, netdev_get_name(port->netdev),
+ ovs_strerror(error));
}
- if (!group->error) {
- group->error = error;
+ return ofproto->fallback_dpid;
+}
+
+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. */
+
+/* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep'
+ * to NULL if the table is not configured to evict rules or if the table
+ * contains no evictable rules. (Rules with a readlock on their evict rwlock,
+ * or with no timeouts are not evictable.) */
+static bool
+choose_rule_to_evict(struct oftable *table, struct rule **rulep)
+{
+ struct eviction_group *evg;
+
+ *rulep = NULL;
+ if (!table->eviction_fields) {
+ return false;
}
- switch (op->type) {
- case OFOPERATION_ADD:
- if (!error) {
- if (op->victim) {
- ofproto_rule_destroy__(op->victim);
+ /* 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 (!ovs_rwlock_trywrlock(&rule->rwlock)) {
+ *rulep = rule;
+ return true;
}
- if (!(rule->cr.wc.vlan_tci_mask & htons(VLAN_VID_MASK))
- && 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;
- }
+ return false;
+}
+
+/* Searches 'ofproto' for tables that have more flows than their configured
+ * maximum and that have flow eviction enabled, and evicts as many flows as
+ * necessary and currently feasible from them.
+ *
+ * This triggers only when an OpenFlow table has N flows in it and then the
+ * client configures a maximum number of flows less than N. */
+static void
+ofproto_evict(struct ofproto *ofproto)
+{
+ struct ofopgroup *group;
+ struct oftable *table;
+
+ group = ofopgroup_create_unattached(ofproto);
+ OFPROTO_FOR_EACH_TABLE (table, ofproto) {
+ while (table->eviction_fields) {
+ struct rule *rule;
+ size_t n_rules;
+
+ ovs_rwlock_rdlock(&table->cls.rwlock);
+ n_rules = classifier_count(&table->cls);
+ ovs_rwlock_unlock(&table->cls.rwlock);
+
+ if (n_rules <= table->max_flows) {
+ break;
}
- } else {
- if (op->victim) {
- classifier_replace(table, &op->victim->cr);
- op->victim = NULL;
- } else {
- classifier_remove(table, &rule->cr);
+
+ if (!choose_rule_to_evict(table, &rule)) {
+ break;
+ }
+
+ if (rule->pending) {
+ ovs_rwlock_unlock(&rule->rwlock);
+ break;
}
- ofproto_rule_destroy__(rule);
+
+ ofoperation_create(group, rule,
+ OFOPERATION_DELETE, OFPRR_EVICTION);
+ oftable_remove_rule(rule);
+ ofproto->ofproto_class->rule_delete(rule);
}
- op->victim = NULL;
- break;
+ }
+ ofopgroup_submit(group);
+}
+\f
+/* Eviction groups. */
- case OFOPERATION_DELETE:
- assert(!error);
- ofproto_rule_destroy__(rule);
- op->rule = NULL;
- break;
+/* 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();
+}
- case OFOPERATION_MODIFY:
- if (!error) {
- rule->modified = time_msec();
+/* 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)
+{
+ heap_change(&table->eviction_groups_by_size, &evg->size_node,
+ eviction_group_priority(heap_count(&evg->rules)));
+}
+
+/* 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;
+
+ 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
+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 {
- free(rule->actions);
- rule->actions = op->actions;
- rule->n_actions = op->n_actions;
- op->actions = NULL;
+ eviction_group_resized(table, evg);
}
- break;
+ }
+}
- default:
- NOT_REACHED();
+/* 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;
+ struct flow flow;
+ uint32_t hash;
+
+ hash = table->eviction_group_id_basis;
+ miniflow_expand(&rule->cr.match.flow, &flow);
+ for (sf = table->eviction_fields;
+ sf < &table->eviction_fields[table->n_eviction_fields];
+ sf++)
+ {
+ if (mf_are_prereqs_ok(sf->field, &flow)) {
+ union mf_value value;
+
+ mf_get_value(sf->field, &flow, &value);
+ if (sf->ofs) {
+ bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
+ }
+ if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
+ unsigned int start = sf->ofs + sf->n_bits;
+ bitwise_zero(&value, sf->field->n_bytes, start,
+ sf->field->n_bytes * 8 - start);
+ }
+ hash = hash_bytes(&value, sf->field->n_bytes, hash);
+ } else {
+ hash = hash_int(hash, 0);
+ }
+ }
+
+ 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;
+}
+
+/* Returns an eviction priority for 'rule'. The return value should be
+ * interpreted so that higher priorities make a rule more attractive candidates
+ * for eviction. */
+static uint32_t
+rule_eviction_priority(struct rule *rule)
+{
+ long long int hard_expiration;
+ long long int idle_expiration;
+ long long int expiration;
+ uint32_t expiration_offset;
+
+ /* Calculate time of expiration. */
+ ovs_mutex_lock(&rule->timeout_mutex);
+ 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);
+ ovs_mutex_unlock(&rule->timeout_mutex);
+ if (expiration == LLONG_MAX) {
+ return 0;
}
- ofoperation_destroy(op);
+
+ /* 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;
}
-struct rule *
-ofoperation_get_victim(struct ofoperation *op)
+/* 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
+eviction_group_add_rule(struct rule *rule)
{
- assert(op->type == OFOPERATION_ADD);
- return op->victim;
+ struct ofproto *ofproto = rule->ofproto;
+ struct oftable *table = &ofproto->tables[rule->table_id];
+ bool has_timeout;
+
+ ovs_mutex_lock(&rule->timeout_mutex);
+ has_timeout = rule->hard_timeout || rule->idle_timeout;
+ ovs_mutex_unlock(&rule->timeout_mutex);
+
+ if (table->eviction_fields && has_timeout) {
+ struct eviction_group *evg;
+
+ evg = eviction_group_find(table, eviction_group_hash_rule(rule));
+
+ rule->eviction_group = evg;
+ heap_insert(&evg->rules, &rule->evg_node,
+ rule_eviction_priority(rule));
+ eviction_group_resized(table, evg);
+ }
}
\f
-static uint64_t
-pick_datapath_id(const struct ofproto *ofproto)
+/* oftables. */
+
+/* Initializes 'table'. */
+static void
+oftable_init(struct oftable *table)
{
- const struct ofport *port;
+ memset(table, 0, sizeof *table);
+ classifier_init(&table->cls);
+ table->max_flows = UINT_MAX;
+}
- port = ofproto_get_port(ofproto, OFPP_LOCAL);
- if (port) {
- uint8_t ea[ETH_ADDR_LEN];
- int error;
+/* Destroys 'table', including its classifier and eviction groups.
+ *
+ * The caller is responsible for freeing 'table' itself. */
+static void
+oftable_destroy(struct oftable *table)
+{
+ ovs_rwlock_rdlock(&table->cls.rwlock);
+ ovs_assert(classifier_is_empty(&table->cls));
+ ovs_rwlock_unlock(&table->cls.rwlock);
+ oftable_disable_eviction(table);
+ classifier_destroy(&table->cls);
+ free(table->name);
+}
- error = netdev_get_etheraddr(port->netdev, ea);
- if (!error) {
- return eth_addr_to_uint64(ea);
+/* 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
+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);
}
- VLOG_WARN("could not get MAC address for %s (%s)",
- netdev_get_name(port->netdev), strerror(error));
+ } else {
+ free(table->name);
+ table->name = NULL;
}
- return ofproto->fallback_dpid;
}
-static uint64_t
-pick_fallback_dpid(void)
+/* 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)
{
- uint8_t ea[ETH_ADDR_LEN];
- eth_addr_nicira_random(ea);
- return eth_addr_to_uint64(ea);
+ if (table->eviction_fields) {
+ struct eviction_group *evg, *next;
+
+ HMAP_FOR_EACH_SAFE (evg, next, id_node,
+ &table->eviction_groups_by_id) {
+ eviction_group_destroy(table, evg);
+ }
+ hmap_destroy(&table->eviction_groups_by_id);
+ heap_destroy(&table->eviction_groups_by_size);
+
+ free(table->eviction_fields);
+ table->eviction_fields = NULL;
+ table->n_eviction_fields = 0;
+ }
+}
+
+/* oftables support a choice of two policies when adding a rule would cause the
+ * number of flows in the table to exceed the configured maximum number: either
+ * they can refuse to add the new flow or they can evict some existing flow.
+ * This function configures the latter policy on 'table', with fairness based
+ * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
+ * 'n_fields' as 0 disables fairness.) */
+static void
+oftable_enable_eviction(struct oftable *table,
+ const struct mf_subfield *fields, size_t n_fields)
+{
+ struct cls_cursor cursor;
+ struct rule *rule;
+
+ if (table->eviction_fields
+ && n_fields == table->n_eviction_fields
+ && (!n_fields
+ || !memcmp(fields, table->eviction_fields,
+ n_fields * sizeof *fields))) {
+ /* No change. */
+ return;
+ }
+
+ oftable_disable_eviction(table);
+
+ table->n_eviction_fields = n_fields;
+ table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
+
+ table->eviction_group_id_basis = random_uint32();
+ hmap_init(&table->eviction_groups_by_id);
+ heap_init(&table->eviction_groups_by_size);
+
+ ovs_rwlock_rdlock(&table->cls.rwlock);
+ cls_cursor_init(&cursor, &table->cls, NULL);
+ CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
+ eviction_group_add_rule(rule);
+ }
+ ovs_rwlock_unlock(&table->cls.rwlock);
+}
+
+/* Removes 'rule' from the oftable that contains it. */
+static void
+oftable_remove_rule__(struct ofproto *ofproto, struct classifier *cls,
+ struct rule *rule)
+ OVS_REQ_WRLOCK(cls->rwlock) OVS_RELEASES(rule->rwlock)
+{
+ classifier_remove(cls, &rule->cr);
+ cookies_remove(ofproto, rule);
+ eviction_group_remove_rule(rule);
+ ovs_mutex_lock(&ofproto->expirable_mutex);
+ if (!list_is_empty(&rule->expirable)) {
+ list_remove(&rule->expirable);
+ }
+ ovs_mutex_unlock(&ofproto->expirable_mutex);
+ if (!list_is_empty(&rule->meter_list_node)) {
+ list_remove(&rule->meter_list_node);
+ list_init(&rule->meter_list_node);
+ }
+ ovs_rwlock_unlock(&rule->rwlock);
+}
+
+static void
+oftable_remove_rule(struct rule *rule)
+{
+ struct ofproto *ofproto = rule->ofproto;
+ struct oftable *table = &ofproto->tables[rule->table_id];
+
+ ovs_rwlock_wrlock(&table->cls.rwlock);
+ oftable_remove_rule__(ofproto, &table->cls, rule);
+ ovs_rwlock_unlock(&table->cls.rwlock);
+}
+
+/* Inserts 'rule' into its oftable, which must not already contain any rule for
+ * the same cls_rule. */
+static void
+oftable_insert_rule(struct rule *rule)
+{
+ struct ofproto *ofproto = rule->ofproto;
+ struct oftable *table = &ofproto->tables[rule->table_id];
+ bool may_expire;
+
+ ovs_mutex_lock(&rule->timeout_mutex);
+ may_expire = rule->hard_timeout || rule->idle_timeout;
+ ovs_mutex_unlock(&rule->timeout_mutex);
+
+ if (may_expire) {
+ ovs_mutex_lock(&ofproto->expirable_mutex);
+ list_insert(&ofproto->expirable, &rule->expirable);
+ ovs_mutex_unlock(&ofproto->expirable_mutex);
+ }
+ cookies_insert(ofproto, rule);
+ if (rule->meter_id) {
+ struct meter *meter = ofproto->meters[rule->meter_id];
+ list_insert(&meter->rules, &rule->meter_list_node);
+ }
+ ovs_rwlock_wrlock(&table->cls.rwlock);
+ classifier_insert(&table->cls, &rule->cr);
+ ovs_rwlock_unlock(&table->cls.rwlock);
+ eviction_group_add_rule(rule);
}
\f
/* unixctl commands. */
}
static void
-ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
- void *aux OVS_UNUSED)
+ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
+ const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
{
struct ofproto *ofproto;
struct ds results;
HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
ds_put_format(&results, "%s\n", ofproto->name);
}
- unixctl_command_reply(conn, 200, ds_cstr(&results));
+ unixctl_command_reply(conn, ds_cstr(&results));
ds_destroy(&results);
}
}
registered = true;
- unixctl_command_register("ofproto/list", "", ofproto_unixctl_list, NULL);
+ unixctl_command_register("ofproto/list", "", 0, 0,
+ ofproto_unixctl_list, NULL);
}
\f
/* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
void
ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
{
- const struct classifier *cls;
+ const struct oftable *oftable;
free(ofproto->vlan_bitmap);
ofproto->vlan_bitmap = bitmap_allocate(4096);
ofproto->vlans_changed = false;
- OFPROTO_FOR_EACH_TABLE (cls, ofproto) {
+ OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
const struct cls_table *table;
- HMAP_FOR_EACH (table, hmap_node, &cls->tables) {
- if (!(table->wc.vlan_tci_mask & htons(VLAN_VID_MASK))) {
+ HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
+ if (minimask_get_vid_mask(&table->mask) == 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);
+ uint16_t vid = miniflow_get_vid(&rule->match.flow);
bitmap_set1(vlan_bitmap, vid);
bitmap_set1(ofproto->vlan_bitmap, vid);
}
* 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)
+ofproto_port_set_realdev(struct ofproto *ofproto, ofp_port_t vlandev_ofp_port,
+ ofp_port_t realdev_ofp_port, int vid)
{
struct ofport *ofport;
int error;
- assert(vlandev_ofp_port != realdev_ofp_port);
+ ovs_assert(vlandev_ofp_port != realdev_ofp_port);
ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
if (!ofport) {
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));
+ netdev_get_name(ofport->netdev), ovs_strerror(error));
}
return error;
}
git://git.onelab.eu / sliver-openvswitch.git / blobdiff