/*
- * Copyright (c) 2009, 2010 Nicira Networks.
+ * Copyright (c) 2009, 2010, 2011 Nicira Networks.
* Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
*
* Licensed under the Apache License, Version 2.0 (the "License");
#include "ofproto.h"
#include <errno.h>
#include <inttypes.h>
-#include <sys/socket.h>
-#include <net/if.h>
-#include <netinet/in.h>
#include <stdbool.h>
#include <stdlib.h>
#include "byte-order.h"
#include "classifier.h"
+#include "connmgr.h"
#include "coverage.h"
-#include "discovery.h"
-#include "dpif.h"
#include "dynamic-string.h"
-#include "fail-open.h"
#include "hash.h"
#include "hmap.h"
-#include "in-band.h"
-#include "mac-learning.h"
#include "netdev.h"
-#include "netflow.h"
#include "nx-match.h"
-#include "odp-util.h"
#include "ofp-print.h"
#include "ofp-util.h"
-#include "ofproto-sflow.h"
#include "ofpbuf.h"
#include "openflow/nicira-ext.h"
#include "openflow/openflow.h"
-#include "openvswitch/datapath-protocol.h"
#include "packets.h"
#include "pinsched.h"
#include "pktbuf.h"
#include "poll-loop.h"
-#include "rconn.h"
+#include "private.h"
#include "shash.h"
-#include "status.h"
-#include "stream-ssl.h"
-#include "svec.h"
-#include "tag.h"
+#include "sset.h"
#include "timeval.h"
+#include "unaligned.h"
#include "unixctl.h"
-#include "vconn.h"
#include "vlog.h"
VLOG_DEFINE_THIS_MODULE(ofproto);
-COVERAGE_DEFINE(facet_changed_rule);
-COVERAGE_DEFINE(facet_revalidate);
-COVERAGE_DEFINE(odp_overflow);
-COVERAGE_DEFINE(ofproto_agg_request);
-COVERAGE_DEFINE(ofproto_costly_flags);
-COVERAGE_DEFINE(ofproto_ctlr_action);
-COVERAGE_DEFINE(ofproto_del_rule);
COVERAGE_DEFINE(ofproto_error);
-COVERAGE_DEFINE(ofproto_expiration);
-COVERAGE_DEFINE(ofproto_expired);
-COVERAGE_DEFINE(ofproto_flows_req);
COVERAGE_DEFINE(ofproto_flush);
-COVERAGE_DEFINE(ofproto_invalidated);
COVERAGE_DEFINE(ofproto_no_packet_in);
-COVERAGE_DEFINE(ofproto_ofconn_stuck);
-COVERAGE_DEFINE(ofproto_ofp2odp);
-COVERAGE_DEFINE(ofproto_packet_in);
COVERAGE_DEFINE(ofproto_packet_out);
COVERAGE_DEFINE(ofproto_queue_req);
COVERAGE_DEFINE(ofproto_recv_openflow);
COVERAGE_DEFINE(ofproto_reinit_ports);
-COVERAGE_DEFINE(ofproto_unexpected_rule);
COVERAGE_DEFINE(ofproto_uninstallable);
COVERAGE_DEFINE(ofproto_update_port);
-#include "sflow_api.h"
-
-struct ofport {
- struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
- struct netdev *netdev;
- struct ofp_phy_port opp; /* In host byte order. */
- uint16_t odp_port;
+enum ofproto_state {
+ S_OPENFLOW, /* Processing OpenFlow commands. */
+ S_FLUSH, /* Deleting all flow table rules. */
};
-static void ofport_free(struct ofport *);
-static void hton_ofp_phy_port(struct ofp_phy_port *);
-
-static int xlate_actions(const union ofp_action *in, size_t n_in,
- const struct flow *, struct ofproto *,
- const struct ofpbuf *packet,
- struct odp_actions *out, tag_type *tags,
- bool *may_set_up_flow, uint16_t *nf_output_iface);
-
-/* An OpenFlow flow. */
-struct rule {
- long long int used; /* Time last used; time created if not used. */
- long long int created; /* Creation time. */
-
- /* These statistics:
- *
- * - Do include packets and bytes from facets that have been deleted or
- * whose own statistics have been folded into the rule.
- *
- * - Do include packets and bytes sent "by hand" that were accounted to
- * the rule without any facet being involved (this is a rare corner
- * case in rule_execute()).
- *
- * - Do not include packet or bytes that can be obtained from any facet's
- * packet_count or byte_count member or that can be obtained from the
- * datapath by, e.g., dpif_flow_get() for any facet.
- */
- uint64_t packet_count; /* Number of packets received. */
- uint64_t byte_count; /* Number of bytes received. */
-
- ovs_be64 flow_cookie; /* Controller-issued identifier. */
-
- struct cls_rule cr; /* In owning ofproto's classifier. */
- uint16_t idle_timeout; /* In seconds from time of last use. */
- uint16_t hard_timeout; /* In seconds from time of creation. */
- bool send_flow_removed; /* Send a flow removed message? */
- int n_actions; /* Number of elements in actions[]. */
- union ofp_action *actions; /* OpenFlow actions. */
- struct list facets; /* List of "struct facet"s. */
+enum ofoperation_type {
+ OFOPERATION_ADD,
+ OFOPERATION_DELETE,
+ OFOPERATION_MODIFY
};
-static struct rule *rule_from_cls_rule(const struct cls_rule *);
-static bool rule_is_hidden(const struct rule *);
-
-static struct rule *rule_create(const struct cls_rule *,
- const union ofp_action *, size_t n_actions,
- uint16_t idle_timeout, uint16_t hard_timeout,
- ovs_be64 flow_cookie, bool send_flow_removed);
-static void rule_destroy(struct ofproto *, struct rule *);
-static void rule_free(struct rule *);
-
-static struct rule *rule_lookup(struct ofproto *, const struct flow *);
-static void rule_insert(struct ofproto *, struct rule *);
-static void rule_remove(struct ofproto *, struct rule *);
-
-static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason);
-
-/* An exact-match instantiation of an OpenFlow flow. */
-struct facet {
- long long int used; /* Time last used; time created if not used. */
-
- /* These statistics:
- *
- * - Do include packets and bytes sent "by hand", e.g. with
- * dpif_execute().
- *
- * - Do include packets and bytes that were obtained from the datapath
- * when a flow was deleted (e.g. dpif_flow_del()) or when its
- * statistics were reset (e.g. dpif_flow_put() with ODPPF_ZERO_STATS).
+/* A single OpenFlow request can execute any number of operations. The
+ * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
+ * ofconn to which an error reply should be sent if necessary.
+ *
+ * ofproto initiates some operations internally. These operations are still
+ * assigned to groups but will not have an associated ofconn. */
+struct ofopgroup {
+ struct ofproto *ofproto; /* Owning ofproto. */
+ struct list ofproto_node; /* In ofproto's "pending" list. */
+ struct list ops; /* List of "struct ofoperation"s. */
+
+ /* Data needed to send OpenFlow reply on failure or to send a buffered
+ * packet on success.
*
- * - Do not include any packets or bytes that can currently be obtained
- * from the datapath by, e.g., dpif_flow_get().
+ * If list_is_empty(ofconn_node) then this ofopgroup never had an
+ * associated ofconn or its ofconn's connection dropped after it initiated
+ * the operation. In the latter case 'ofconn' is a wild pointer that
+ * refers to freed memory, so the 'ofconn' member must be used only if
+ * !list_is_empty(ofconn_node).
*/
- uint64_t packet_count; /* Number of packets received. */
- uint64_t byte_count; /* Number of bytes received. */
-
- /* Number of bytes passed to account_cb. This may include bytes that can
- * currently obtained from the datapath (thus, it can be greater than
- * byte_count). */
- uint64_t accounted_bytes;
-
- struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
- struct list list_node; /* In owning rule's 'facets' list. */
- struct rule *rule; /* Owning rule. */
- struct flow flow; /* Exact-match flow. */
- bool installed; /* Installed in datapath? */
- bool may_install; /* True ordinarily; false if actions must
- * be reassessed for every packet. */
- int n_actions; /* Number of elements in actions[]. */
- union odp_action *actions; /* Datapath actions. */
- tag_type tags; /* Tags (set only by hooks). */
- struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
+ struct list ofconn_node; /* In ofconn's list of pending opgroups. */
+ struct ofconn *ofconn; /* ofconn for reply (but see note above). */
+ struct ofp_header *request; /* Original request (truncated at 64 bytes). */
+ uint32_t buffer_id; /* Buffer id from original request. */
+ int error; /* 0 if no error yet, otherwise error code. */
};
-static struct facet *facet_create(struct ofproto *, struct rule *,
- const struct flow *,
- const struct ofpbuf *packet);
-static void facet_remove(struct ofproto *, struct facet *);
-static void facet_free(struct facet *);
-
-static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *);
-static bool facet_revalidate(struct ofproto *, struct facet *);
-
-static void facet_install(struct ofproto *, struct facet *, bool zero_stats);
-static void facet_uninstall(struct ofproto *, struct facet *);
-static void facet_flush_stats(struct ofproto *, struct facet *);
-
-static void facet_make_actions(struct ofproto *, struct facet *,
- const struct ofpbuf *packet);
-static void facet_update_stats(struct ofproto *, struct facet *,
- const struct odp_flow_stats *);
-
-/* ofproto supports two kinds of OpenFlow connections:
- *
- * - "Primary" connections to ordinary OpenFlow controllers. ofproto
- * maintains persistent connections to these controllers and by default
- * sends them asynchronous messages such as packet-ins.
- *
- * - "Service" connections, e.g. from ovs-ofctl. When these connections
- * drop, it is the other side's responsibility to reconnect them if
- * necessary. ofproto does not send them asynchronous messages by default.
- *
- * Currently, active (tcp, ssl, unix) connections are always "primary"
- * connections and passive (ptcp, pssl, punix) connections are always "service"
- * connections. There is no inherent reason for this, but it reflects the
- * common case.
- */
-enum ofconn_type {
- OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */
- OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */
+static struct ofopgroup *ofopgroup_create(struct ofproto *);
+static struct ofopgroup *ofopgroup_create_for_ofconn(struct ofconn *,
+ const struct ofp_header *,
+ uint32_t buffer_id);
+static void ofopgroup_submit(struct ofopgroup *);
+static void ofopgroup_destroy(struct ofopgroup *);
+
+/* A single flow table operation. */
+struct ofoperation {
+ struct ofopgroup *group; /* Owning group. */
+ struct list group_node; /* In ofopgroup's "ops" list. */
+ struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
+ struct rule *rule; /* Rule being operated upon. */
+ enum ofoperation_type type; /* Type of operation. */
+ int status; /* -1 if pending, otherwise 0 or error code. */
+ struct rule *victim; /* OFOPERATION_ADDING: Replaced rule. */
+ union ofp_action *actions; /* OFOPERATION_MODIFYING: Replaced actions. */
+ int n_actions; /* OFOPERATION_MODIFYING: # of old actions. */
+ ovs_be64 flow_cookie; /* Rule's old flow cookie. */
};
-/* A listener for incoming OpenFlow "service" connections. */
-struct ofservice {
- struct hmap_node node; /* In struct ofproto's "services" hmap. */
- struct pvconn *pvconn; /* OpenFlow connection listener. */
-
- /* These are not used by ofservice directly. They are settings for
- * accepted "struct ofconn"s from the pvconn. */
- int probe_interval; /* Max idle time before probing, in seconds. */
- int rate_limit; /* Max packet-in rate in packets per second. */
- int burst_limit; /* Limit on accumulating packet credits. */
-};
+static void ofoperation_create(struct ofopgroup *, struct rule *,
+ enum ofoperation_type);
+static void ofoperation_destroy(struct ofoperation *);
-static struct ofservice *ofservice_lookup(struct ofproto *,
- const char *target);
-static int ofservice_create(struct ofproto *,
- const struct ofproto_controller *);
-static void ofservice_reconfigure(struct ofservice *,
- const struct ofproto_controller *);
-static void ofservice_destroy(struct ofproto *, struct ofservice *);
-
-/* An OpenFlow connection. */
-struct ofconn {
- struct ofproto *ofproto; /* The ofproto that owns this connection. */
- struct list node; /* In struct ofproto's "all_conns" list. */
- struct rconn *rconn; /* OpenFlow connection. */
- enum ofconn_type type; /* Type. */
- enum nx_flow_format flow_format; /* Currently selected flow format. */
-
- /* OFPT_PACKET_IN related data. */
- struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
- struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
- struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
- int miss_send_len; /* Bytes to send of buffered packets. */
-
- /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
- * requests, and the maximum number before we stop reading OpenFlow
- * requests. */
-#define OFCONN_REPLY_MAX 100
- struct rconn_packet_counter *reply_counter;
-
- /* type == OFCONN_PRIMARY only. */
- enum nx_role role; /* Role. */
- struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
- struct discovery *discovery; /* Controller discovery object, if enabled. */
- struct status_category *ss; /* Switch status category. */
- enum ofproto_band band; /* In-band or out-of-band? */
-};
+static void ofport_destroy__(struct ofport *);
+static void ofport_destroy(struct ofport *);
-/* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
- * "schedulers" array. Their values are 0 and 1, and their meanings and values
- * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
- * case anything ever changes, check their values here. */
-#define N_SCHEDULERS 2
-BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
-BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
-BUILD_ASSERT_DECL(OFPR_ACTION == 1);
-BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
-
-static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
- enum ofconn_type);
-static void ofconn_destroy(struct ofconn *);
-static void ofconn_run(struct ofconn *);
-static void ofconn_wait(struct ofconn *);
-static bool ofconn_receives_async_msgs(const struct ofconn *);
-static char *ofconn_make_name(const struct ofproto *, const char *target);
-static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
-
-static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
- struct rconn_packet_counter *counter);
-
-static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
-static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
-
-struct ofproto {
- /* Settings. */
- uint64_t datapath_id; /* Datapath ID. */
- uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
- char *mfr_desc; /* Manufacturer. */
- char *hw_desc; /* Hardware. */
- char *sw_desc; /* Software version. */
- char *serial_desc; /* Serial number. */
- char *dp_desc; /* Datapath description. */
-
- /* Datapath. */
- struct dpif *dpif;
- struct netdev_monitor *netdev_monitor;
- struct hmap ports; /* Contains "struct ofport"s. */
- struct shash port_by_name;
- uint32_t max_ports;
-
- /* Configuration. */
- struct switch_status *switch_status;
- struct fail_open *fail_open;
- struct netflow *netflow;
- struct ofproto_sflow *sflow;
-
- /* In-band control. */
- struct in_band *in_band;
- long long int next_in_band_update;
- struct sockaddr_in *extra_in_band_remotes;
- size_t n_extra_remotes;
- int in_band_queue;
-
- /* Flow table. */
- struct classifier cls;
- long long int next_expiration;
-
- /* Facets. */
- struct hmap facets;
- bool need_revalidate;
- struct tag_set revalidate_set;
-
- /* OpenFlow connections. */
- struct hmap controllers; /* Controller "struct ofconn"s. */
- struct list all_conns; /* Contains "struct ofconn"s. */
- enum ofproto_fail_mode fail_mode;
-
- /* OpenFlow listeners. */
- struct hmap services; /* Contains "struct ofservice"s. */
- struct pvconn **snoops;
- size_t n_snoops;
-
- /* Hooks for ovs-vswitchd. */
- const struct ofhooks *ofhooks;
- void *aux;
-
- /* Used by default ofhooks. */
- struct mac_learning *ml;
-};
+static uint64_t pick_datapath_id(const struct ofproto *);
+static uint64_t pick_fallback_dpid(void);
-static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+static void ofproto_destroy__(struct ofproto *);
-static const struct ofhooks default_ofhooks;
+static void ofproto_rule_destroy__(struct rule *);
+static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
-static uint64_t pick_datapath_id(const struct ofproto *);
-static uint64_t pick_fallback_dpid(void);
+static void ofopgroup_destroy(struct ofopgroup *);
-static int ofproto_expire(struct ofproto *);
+static int add_flow(struct ofproto *, struct ofconn *, struct flow_mod *,
+ const struct ofp_header *);
-static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
+/* This return value tells handle_openflow() that processing of the current
+ * OpenFlow message must be postponed until some ongoing operations have
+ * completed.
+ *
+ * This particular value is a good choice because it is negative (so it won't
+ * collide with any errno value or any value returned by ofp_mkerr()) and large
+ * (so it won't accidentally collide with EOF or a negative errno value). */
+enum { OFPROTO_POSTPONE = -100000 };
-static void handle_openflow(struct ofconn *, struct ofpbuf *);
+static bool handle_openflow(struct ofconn *, struct ofpbuf *);
-static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
static void update_port(struct ofproto *, const char *devname);
static int init_ports(struct ofproto *);
static void reinit_ports(struct ofproto *);
-int
-ofproto_create(const char *datapath, const char *datapath_type,
- const struct ofhooks *ofhooks, void *aux,
- struct ofproto **ofprotop)
-{
- struct odp_stats stats;
- struct ofproto *p;
- struct dpif *dpif;
- int error;
-
- *ofprotop = NULL;
+static void ofproto_unixctl_init(void);
- /* Connect to datapath and start listening for messages. */
- error = dpif_open(datapath, datapath_type, &dpif);
- if (error) {
- VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
- return error;
- }
- error = dpif_get_dp_stats(dpif, &stats);
- if (error) {
- VLOG_ERR("failed to obtain stats for datapath %s: %s",
- datapath, strerror(error));
- dpif_close(dpif);
- return error;
- }
- error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
- if (error) {
- VLOG_ERR("failed to listen on datapath %s: %s",
- datapath, strerror(error));
- dpif_close(dpif);
- return error;
- }
- dpif_flow_flush(dpif);
- dpif_recv_purge(dpif);
-
- /* Initialize settings. */
- p = xzalloc(sizeof *p);
- p->fallback_dpid = pick_fallback_dpid();
- p->datapath_id = p->fallback_dpid;
- p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
- p->hw_desc = xstrdup(DEFAULT_HW_DESC);
- p->sw_desc = xstrdup(DEFAULT_SW_DESC);
- p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
- p->dp_desc = xstrdup(DEFAULT_DP_DESC);
-
- /* Initialize datapath. */
- p->dpif = dpif;
- p->netdev_monitor = netdev_monitor_create();
- hmap_init(&p->ports);
- shash_init(&p->port_by_name);
- p->max_ports = stats.max_ports;
-
- /* Initialize submodules. */
- p->switch_status = switch_status_create(p);
- p->fail_open = NULL;
- p->netflow = NULL;
- p->sflow = NULL;
-
- /* Initialize in-band control. */
- p->in_band = NULL;
- p->in_band_queue = -1;
-
- /* Initialize flow table. */
- classifier_init(&p->cls);
- p->next_expiration = time_msec() + 1000;
-
- /* Initialize facet table. */
- hmap_init(&p->facets);
- p->need_revalidate = false;
- tag_set_init(&p->revalidate_set);
-
- /* Initialize OpenFlow connections. */
- list_init(&p->all_conns);
- hmap_init(&p->controllers);
- hmap_init(&p->services);
- p->snoops = NULL;
- p->n_snoops = 0;
-
- /* Initialize hooks. */
- if (ofhooks) {
- p->ofhooks = ofhooks;
- p->aux = aux;
- p->ml = NULL;
- } else {
- p->ofhooks = &default_ofhooks;
- p->aux = p;
- p->ml = mac_learning_create();
- }
+/* All registered ofproto classes, in probe order. */
+static const struct ofproto_class **ofproto_classes;
+static size_t n_ofproto_classes;
+static size_t allocated_ofproto_classes;
- /* Pick final datapath ID. */
- p->datapath_id = pick_datapath_id(p);
- VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
+/* Map from datapath name to struct ofproto, for use by unixctl commands. */
+static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
- *ofprotop = p;
- return 0;
-}
+static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
-void
-ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
+static void
+ofproto_initialize(void)
{
- 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);
+ static bool inited;
- /* 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);
+ if (!inited) {
+ inited = true;
+ ofproto_class_register(&ofproto_dpif_class);
}
}
-static bool
-is_discovery_controller(const struct ofproto_controller *c)
+/* 'type' should be a normalized datapath type, as returned by
+ * ofproto_normalize_type(). Returns the corresponding ofproto_class
+ * structure, or a null pointer if there is none registered for 'type'. */
+static const struct ofproto_class *
+ofproto_class_find__(const char *type)
{
- return !strcmp(c->target, "discover");
-}
+ size_t i;
-static bool
-is_in_band_controller(const struct ofproto_controller *c)
-{
- return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
-}
+ ofproto_initialize();
+ for (i = 0; i < n_ofproto_classes; i++) {
+ const struct ofproto_class *class = ofproto_classes[i];
+ struct sset types;
+ bool found;
-/* Creates a new controller in 'ofproto'. Some of the settings are initially
- * drawn from 'c', but update_controller() needs to be called later to finish
- * the new ofconn's configuration. */
-static void
-add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
-{
- struct discovery *discovery;
- struct ofconn *ofconn;
+ sset_init(&types);
+ class->enumerate_types(&types);
+ found = sset_contains(&types, type);
+ sset_destroy(&types);
- if (is_discovery_controller(c)) {
- int error = discovery_create(c->accept_re, c->update_resolv_conf,
- ofproto->dpif, ofproto->switch_status,
- &discovery);
- if (error) {
- return;
+ if (found) {
+ return class;
}
- } else {
- discovery = NULL;
- }
-
- ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
- ofconn->pktbuf = pktbuf_create();
- ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
- if (discovery) {
- ofconn->discovery = discovery;
- } else {
- char *name = ofconn_make_name(ofproto, c->target);
- rconn_connect(ofconn->rconn, c->target, name);
- free(name);
}
- hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
- hash_string(c->target, 0));
+ VLOG_WARN("unknown datapath type %s", type);
+ return NULL;
}
-/* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
- * target or turn discovery on or off (these are done by creating new ofconns
- * and deleting old ones), but it can update the rest of an ofconn's
- * settings. */
-static void
-update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
+/* Registers a new ofproto class. After successful registration, new ofprotos
+ * of that type can be created using ofproto_create(). */
+int
+ofproto_class_register(const struct ofproto_class *new_class)
{
- int probe_interval;
-
- ofconn->band = (is_in_band_controller(c)
- ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
-
- rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
-
- probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
- rconn_set_probe_interval(ofconn->rconn, probe_interval);
+ size_t i;
- if (ofconn->discovery) {
- discovery_set_update_resolv_conf(ofconn->discovery,
- c->update_resolv_conf);
- discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
+ for (i = 0; i < n_ofproto_classes; i++) {
+ if (ofproto_classes[i] == new_class) {
+ return EEXIST;
+ }
}
- ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
+ if (n_ofproto_classes >= allocated_ofproto_classes) {
+ ofproto_classes = x2nrealloc(ofproto_classes,
+ &allocated_ofproto_classes,
+ sizeof *ofproto_classes);
+ }
+ ofproto_classes[n_ofproto_classes++] = new_class;
+ return 0;
}
-static const char *
-ofconn_get_target(const struct ofconn *ofconn)
+/* Unregisters a datapath provider. 'type' must have been previously
+ * registered and not currently be in use by any ofprotos. After
+ * unregistration new datapaths of that type cannot be opened using
+ * ofproto_create(). */
+int
+ofproto_class_unregister(const struct ofproto_class *class)
{
- return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
+ size_t i;
+
+ for (i = 0; i < n_ofproto_classes; i++) {
+ if (ofproto_classes[i] == class) {
+ for (i++; i < n_ofproto_classes; i++) {
+ ofproto_classes[i - 1] = ofproto_classes[i];
+ }
+ n_ofproto_classes--;
+ return 0;
+ }
+ }
+ VLOG_WARN("attempted to unregister an ofproto class that is not "
+ "registered");
+ return EAFNOSUPPORT;
}
-static struct ofconn *
-find_controller_by_target(struct ofproto *ofproto, const char *target)
+/* Clears 'types' and enumerates all registered ofproto types into it. The
+ * caller must first initialize the sset. */
+void
+ofproto_enumerate_types(struct sset *types)
{
- struct ofconn *ofconn;
+ size_t i;
- HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
- hash_string(target, 0), &ofproto->controllers) {
- if (!strcmp(ofconn_get_target(ofconn), target)) {
- return ofconn;
- }
+ ofproto_initialize();
+ for (i = 0; i < n_ofproto_classes; i++) {
+ ofproto_classes[i]->enumerate_types(types);
}
- return NULL;
}
-static void
-update_in_band_remotes(struct ofproto *ofproto)
+/* Returns the fully spelled out name for the given ofproto 'type'.
+ *
+ * Normalized type string can be compared with strcmp(). Unnormalized type
+ * string might be the same even if they have different spellings. */
+const char *
+ofproto_normalize_type(const char *type)
{
- const struct ofconn *ofconn;
- struct sockaddr_in *addrs;
- size_t max_addrs, n_addrs;
- bool discovery;
- size_t i;
+ return type && type[0] ? type : "system";
+}
- /* Allocate enough memory for as many remotes as we could possibly have. */
- max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
- addrs = xmalloc(max_addrs * sizeof *addrs);
- n_addrs = 0;
+/* Clears 'names' and enumerates the names of all known created ofprotos with
+ * the given 'type'. The caller must first initialize the sset. Returns 0 if
+ * successful, otherwise a positive errno value.
+ *
+ * Some kinds of datapaths might not be practically enumerable. This is not
+ * considered an error. */
+int
+ofproto_enumerate_names(const char *type, struct sset *names)
+{
+ const struct ofproto_class *class = ofproto_class_find__(type);
+ return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
+ }
- /* Add all the remotes. */
- discovery = false;
- HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
- struct sockaddr_in *sin = &addrs[n_addrs];
+int
+ofproto_create(const char *datapath_name, const char *datapath_type,
+ struct ofproto **ofprotop)
+{
+ const struct ofproto_class *class;
+ struct ofproto *ofproto;
+ int error;
- if (ofconn->band == OFPROTO_OUT_OF_BAND) {
- continue;
- }
+ *ofprotop = NULL;
- sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
- if (sin->sin_addr.s_addr) {
- sin->sin_port = rconn_get_remote_port(ofconn->rconn);
- n_addrs++;
- }
- if (ofconn->discovery) {
- discovery = true;
- }
- }
- for (i = 0; i < ofproto->n_extra_remotes; i++) {
- addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
+ ofproto_initialize();
+ ofproto_unixctl_init();
+
+ datapath_type = ofproto_normalize_type(datapath_type);
+ class = ofproto_class_find__(datapath_type);
+ if (!class) {
+ VLOG_WARN("could not create datapath %s of unknown type %s",
+ datapath_name, datapath_type);
+ return EAFNOSUPPORT;
+ }
+
+ ofproto = class->alloc();
+ if (!ofproto) {
+ VLOG_ERR("failed to allocate datapath %s of type %s",
+ datapath_name, datapath_type);
+ return ENOMEM;
+ }
+
+ /* Initialize. */
+ memset(ofproto, 0, sizeof *ofproto);
+ ofproto->ofproto_class = class;
+ ofproto->name = xstrdup(datapath_name);
+ ofproto->type = xstrdup(datapath_type);
+ hmap_insert(&all_ofprotos, &ofproto->hmap_node,
+ hash_string(ofproto->name, 0));
+ ofproto->datapath_id = 0;
+ ofproto->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);
+ hmap_init(&ofproto->ports);
+ shash_init(&ofproto->port_by_name);
+ ofproto->tables = NULL;
+ ofproto->n_tables = 0;
+ ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
+ ofproto->state = S_OPENFLOW;
+ list_init(&ofproto->pending);
+ hmap_init(&ofproto->deletions);
+
+ error = ofproto->ofproto_class->construct(ofproto);
+ if (error) {
+ VLOG_ERR("failed to open datapath %s: %s",
+ datapath_name, strerror(error));
+ ofproto_destroy__(ofproto);
+ return error;
}
+ assert(ofproto->n_tables > 0);
- /* Create or update or destroy in-band.
- *
- * Ordinarily we only enable in-band if there's at least one remote
- * address, but discovery needs the in-band rules for DHCP to be installed
- * even before we know any remote addresses. */
- if (n_addrs || discovery) {
- if (!ofproto->in_band) {
- in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
- &ofproto->in_band);
- }
- if (ofproto->in_band) {
- in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
- }
- in_band_set_queue(ofproto->in_band, ofproto->in_band_queue);
- ofproto->next_in_band_update = time_msec() + 1000;
- } else {
- in_band_destroy(ofproto->in_band);
- ofproto->in_band = NULL;
- }
+ ofproto->datapath_id = pick_datapath_id(ofproto);
+ VLOG_INFO("using datapath ID %016"PRIx64, ofproto->datapath_id);
+ init_ports(ofproto);
- /* Clean up. */
- free(addrs);
+ *ofprotop = ofproto;
+ return 0;
}
-static void
-update_fail_open(struct ofproto *p)
+void
+ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
{
- struct ofconn *ofconn;
-
- if (!hmap_is_empty(&p->controllers)
- && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
- struct rconn **rconns;
- size_t n;
-
- if (!p->fail_open) {
- p->fail_open = fail_open_create(p, p->switch_status);
- }
-
- n = 0;
- rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
- HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) {
- rconns[n++] = ofconn->rconn;
- }
+ uint64_t old_dpid = p->datapath_id;
+ p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
+ if (p->datapath_id != old_dpid) {
+ VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
- fail_open_set_controllers(p->fail_open, rconns, n);
- /* p->fail_open takes ownership of 'rconns'. */
- } else {
- fail_open_destroy(p->fail_open);
- p->fail_open = NULL;
+ /* Force all active connections to reconnect, since there is no way to
+ * notify a controller that the datapath ID has changed. */
+ ofproto_reconnect_controllers(p);
}
}
const struct ofproto_controller *controllers,
size_t n_controllers)
{
- struct shash new_controllers;
- struct ofconn *ofconn, *next_ofconn;
- struct ofservice *ofservice, *next_ofservice;
- bool ss_exists;
- size_t i;
-
- /* Create newly configured controllers and services.
- * Create a name to ofproto_controller mapping in 'new_controllers'. */
- shash_init(&new_controllers);
- for (i = 0; i < n_controllers; i++) {
- const struct ofproto_controller *c = &controllers[i];
-
- if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) {
- if (!find_controller_by_target(p, c->target)) {
- add_controller(p, c);
- }
- } else if (!pvconn_verify_name(c->target)) {
- if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
- continue;
- }
- } else {
- VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
- dpif_name(p->dpif), c->target);
- continue;
- }
-
- shash_add_once(&new_controllers, c->target, &controllers[i]);
- }
-
- /* Delete controllers that are no longer configured.
- * Update configuration of all now-existing controllers. */
- ss_exists = false;
- HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) {
- struct ofproto_controller *c;
-
- c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
- if (!c) {
- ofconn_destroy(ofconn);
- } else {
- update_controller(ofconn, c);
- if (ofconn->ss) {
- ss_exists = true;
- }
- }
- }
-
- /* Delete services that are no longer configured.
- * Update configuration of all now-existing services. */
- HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
- struct ofproto_controller *c;
-
- c = shash_find_data(&new_controllers,
- pvconn_get_name(ofservice->pvconn));
- if (!c) {
- ofservice_destroy(p, ofservice);
- } else {
- ofservice_reconfigure(ofservice, c);
- }
- }
-
- shash_destroy(&new_controllers);
-
- update_in_band_remotes(p);
- update_fail_open(p);
-
- if (!hmap_is_empty(&p->controllers) && !ss_exists) {
- ofconn = CONTAINER_OF(hmap_first(&p->controllers),
- struct ofconn, hmap_node);
- ofconn->ss = switch_status_register(p->switch_status, "remote",
- rconn_status_cb, ofconn->rconn);
- }
+ connmgr_set_controllers(p->connmgr, controllers, n_controllers);
}
void
ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
{
- p->fail_mode = fail_mode;
- update_fail_open(p);
+ connmgr_set_fail_mode(p->connmgr, fail_mode);
}
/* Drops the connections between 'ofproto' and all of its controllers, forcing
void
ofproto_reconnect_controllers(struct ofproto *ofproto)
{
- struct ofconn *ofconn;
-
- LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
- rconn_reconnect(ofconn->rconn);
- }
-}
-
-static bool
-any_extras_changed(const struct ofproto *ofproto,
- const struct sockaddr_in *extras, size_t n)
-{
- size_t i;
-
- if (n != ofproto->n_extra_remotes) {
- return true;
- }
-
- for (i = 0; i < n; i++) {
- const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
- const struct sockaddr_in *new = &extras[i];
-
- if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
- old->sin_port != new->sin_port) {
- return true;
- }
- }
-
- return false;
+ connmgr_reconnect(ofproto->connmgr);
}
/* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
const struct sockaddr_in *extras, size_t n)
{
- if (!any_extras_changed(ofproto, extras, n)) {
- return;
- }
-
- free(ofproto->extra_in_band_remotes);
- ofproto->n_extra_remotes = n;
- ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
-
- update_in_band_remotes(ofproto);
+ connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
}
/* Sets the OpenFlow queue used by flows set up by in-band control on
void
ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
{
- if (queue_id != ofproto->in_band_queue) {
- ofproto->in_band_queue = queue_id;
- update_in_band_remotes(ofproto);
- }
+ connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
}
void
}
}
-static int
-set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
- const struct svec *svec)
-{
- struct pvconn **pvconns = *pvconnsp;
- size_t n_pvconns = *n_pvconnsp;
- int retval = 0;
- size_t i;
-
- for (i = 0; i < n_pvconns; i++) {
- pvconn_close(pvconns[i]);
- }
- free(pvconns);
-
- pvconns = xmalloc(svec->n * sizeof *pvconns);
- n_pvconns = 0;
- for (i = 0; i < svec->n; i++) {
- const char *name = svec->names[i];
- struct pvconn *pvconn;
- int error;
-
- error = pvconn_open(name, &pvconn);
- if (!error) {
- pvconns[n_pvconns++] = pvconn;
- } else {
- VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
- if (!retval) {
- retval = error;
- }
- }
- }
-
- *pvconnsp = pvconns;
- *n_pvconnsp = n_pvconns;
-
- return retval;
-}
-
int
-ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
+ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
{
- return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
+ return connmgr_set_snoops(ofproto->connmgr, snoops);
}
int
ofproto_set_netflow(struct ofproto *ofproto,
const struct netflow_options *nf_options)
{
- if (nf_options && nf_options->collectors.n) {
- if (!ofproto->netflow) {
- ofproto->netflow = netflow_create();
- }
- return netflow_set_options(ofproto->netflow, nf_options);
+ if (nf_options && sset_is_empty(&nf_options->collectors)) {
+ nf_options = NULL;
+ }
+
+ if (ofproto->ofproto_class->set_netflow) {
+ return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
} else {
- netflow_destroy(ofproto->netflow);
- ofproto->netflow = NULL;
- return 0;
+ return nf_options ? EOPNOTSUPP : 0;
}
}
-void
+int
ofproto_set_sflow(struct ofproto *ofproto,
const struct ofproto_sflow_options *oso)
{
- struct ofproto_sflow *os = ofproto->sflow;
- if (oso) {
- if (!os) {
- struct ofport *ofport;
-
- os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
- HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
- ofproto_sflow_add_port(os, ofport->odp_port,
- netdev_get_name(ofport->netdev));
- }
- }
- ofproto_sflow_set_options(os, oso);
- } else {
- ofproto_sflow_destroy(os);
- ofproto->sflow = NULL;
+ if (oso && sset_is_empty(&oso->targets)) {
+ oso = NULL;
}
-}
-
-uint64_t
-ofproto_get_datapath_id(const struct ofproto *ofproto)
-{
- return ofproto->datapath_id;
-}
-
-bool
-ofproto_has_primary_controller(const struct ofproto *ofproto)
-{
- return !hmap_is_empty(&ofproto->controllers);
-}
-enum ofproto_fail_mode
-ofproto_get_fail_mode(const struct ofproto *p)
-{
- return p->fail_mode;
+ if (ofproto->ofproto_class->set_sflow) {
+ return ofproto->ofproto_class->set_sflow(ofproto, oso);
+ } else {
+ return oso ? EOPNOTSUPP : 0;
+ }
}
+\f
+/* Connectivity Fault Management configuration. */
+/* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
void
-ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
+ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
{
- size_t i;
-
- for (i = 0; i < ofproto->n_snoops; i++) {
- svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
+ struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
+ if (ofport && ofproto->ofproto_class->set_cfm) {
+ ofproto->ofproto_class->set_cfm(ofport, NULL);
}
}
+/* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
+ * basic configuration from the configuration members in 'cfm', and the remote
+ * maintenance point ID from remote_mpid. Ignores the statistics members of
+ * 'cfm'.
+ *
+ * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
void
-ofproto_destroy(struct ofproto *p)
+ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
+ const struct cfm_settings *s)
{
- struct ofservice *ofservice, *next_ofservice;
- struct ofconn *ofconn, *next_ofconn;
- struct ofport *ofport, *next_ofport;
- size_t i;
+ struct ofport *ofport;
+ int error;
- if (!p) {
+ ofport = ofproto_get_port(ofproto, ofp_port);
+ if (!ofport) {
+ VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
+ ofproto->name, ofp_port);
return;
}
- /* Destroy fail-open and in-band early, since they touch the classifier. */
- fail_open_destroy(p->fail_open);
- p->fail_open = NULL;
-
- in_band_destroy(p->in_band);
- p->in_band = NULL;
- free(p->extra_in_band_remotes);
-
- ofproto_flush_flows(p);
- classifier_destroy(&p->cls);
- hmap_destroy(&p->facets);
-
- LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
- ofconn_destroy(ofconn);
- }
- hmap_destroy(&p->controllers);
-
- dpif_close(p->dpif);
- netdev_monitor_destroy(p->netdev_monitor);
- HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
- hmap_remove(&p->ports, &ofport->hmap_node);
- ofport_free(ofport);
- }
- shash_destroy(&p->port_by_name);
-
- switch_status_destroy(p->switch_status);
- netflow_destroy(p->netflow);
- ofproto_sflow_destroy(p->sflow);
-
- HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
- ofservice_destroy(p, ofservice);
- }
- hmap_destroy(&p->services);
-
- for (i = 0; i < p->n_snoops; i++) {
- pvconn_close(p->snoops[i]);
+ /* XXX: For configuration simplicity, we only support one remote_mpid
+ * outside of the CFM module. It's not clear if this is the correct long
+ * term solution or not. */
+ error = (ofproto->ofproto_class->set_cfm
+ ? ofproto->ofproto_class->set_cfm(ofport, s)
+ : EOPNOTSUPP);
+ if (error) {
+ VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
+ ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
+ strerror(error));
}
- free(p->snoops);
-
- mac_learning_destroy(p->ml);
-
- free(p->mfr_desc);
- free(p->hw_desc);
- free(p->sw_desc);
- free(p->serial_desc);
- free(p->dp_desc);
-
- hmap_destroy(&p->ports);
-
- free(p);
}
+/* 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_run(struct ofproto *p)
+ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
{
- int error = ofproto_run1(p);
- if (!error) {
- error = ofproto_run2(p, false);
- }
- return error;
+ struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
+ return (ofport && ofproto->ofproto_class->port_is_lacp_current
+ ? ofproto->ofproto_class->port_is_lacp_current(ofport)
+ : -1);
}
+\f
+/* Bundles. */
-static void
-process_port_change(struct ofproto *ofproto, int error, char *devname)
-{
- if (error == ENOBUFS) {
- reinit_ports(ofproto);
- } else if (!error) {
- update_port(ofproto, devname);
- free(devname);
- }
+/* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
+ * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
+ * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
+ * configuration plus, if there is more than one slave, a bonding
+ * configuration.
+ *
+ * If 'aux' is already registered then this function updates its configuration
+ * to 's'. Otherwise, this function registers a new bundle.
+ *
+ * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
+ * port. */
+int
+ofproto_bundle_register(struct ofproto *ofproto, void *aux,
+ const struct ofproto_bundle_settings *s)
+{
+ return (ofproto->ofproto_class->bundle_set
+ ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
+ : EOPNOTSUPP);
}
-/* Returns a "preference level" for snooping 'ofconn'. A higher return value
- * means that 'ofconn' is more interesting for monitoring than a lower return
- * value. */
-static int
-snoop_preference(const struct ofconn *ofconn)
+/* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
+ * If no such bundle has been registered, this has no effect. */
+int
+ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
{
- switch (ofconn->role) {
- case NX_ROLE_MASTER:
- return 3;
- case NX_ROLE_OTHER:
- return 2;
- case NX_ROLE_SLAVE:
- return 1;
- default:
- /* Shouldn't happen. */
- return 0;
- }
+ return ofproto_bundle_register(ofproto, aux, NULL);
}
-/* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
- * Connects this vconn to a controller. */
-static void
-add_snooper(struct ofproto *ofproto, struct vconn *vconn)
+\f
+/* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
+ * If 'aux' is already registered then this function updates its configuration
+ * to 's'. Otherwise, this function registers a new mirror.
+ *
+ * Mirrors affect only the treatment of packets output to the OFPP_NORMAL
+ * port. */
+int
+ofproto_mirror_register(struct ofproto *ofproto, void *aux,
+ const struct ofproto_mirror_settings *s)
{
- struct ofconn *ofconn, *best;
-
- /* Pick a controller for monitoring. */
- best = NULL;
- LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
- if (ofconn->type == OFCONN_PRIMARY
- && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
- best = ofconn;
- }
- }
+ return (ofproto->ofproto_class->mirror_set
+ ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
+ : EOPNOTSUPP);
+}
- if (best) {
- rconn_add_monitor(best->rconn, vconn);
- } else {
- VLOG_INFO_RL(&rl, "no controller connection to snoop");
- vconn_close(vconn);
- }
+/* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
+ * If no mirror has been registered, this has no effect. */
+int
+ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
+{
+ return ofproto_mirror_register(ofproto, aux, NULL);
}
+/* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
+ * which all packets are flooded, instead of using MAC learning. If
+ * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
+ *
+ * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
+ * port. */
int
-ofproto_run1(struct ofproto *p)
+ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
{
- struct ofconn *ofconn, *next_ofconn;
- struct ofservice *ofservice;
- char *devname;
- int error;
- int i;
+ return (ofproto->ofproto_class->set_flood_vlans
+ ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
+ : EOPNOTSUPP);
+}
+
+/* Returns true if 'aux' is a registered bundle that is currently in use as the
+ * output for a mirror. */
+bool
+ofproto_is_mirror_output_bundle(struct ofproto *ofproto, void *aux)
+{
+ return (ofproto->ofproto_class->is_mirror_output_bundle
+ ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
+ : false);
+}
+\f
+bool
+ofproto_has_snoops(const struct ofproto *ofproto)
+{
+ return connmgr_has_snoops(ofproto->connmgr);
+}
+
+void
+ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
+{
+ connmgr_get_snoops(ofproto->connmgr, snoops);
+}
+
+static void
+ofproto_flush__(struct ofproto *ofproto)
+{
+ struct classifier *table;
+ struct ofopgroup *group;
- if (shash_is_empty(&p->port_by_name)) {
- init_ports(p);
+ if (ofproto->ofproto_class->flush) {
+ ofproto->ofproto_class->flush(ofproto);
}
- for (i = 0; i < 50; i++) {
- struct ofpbuf *buf;
+ group = ofopgroup_create(ofproto);
+ for (table = ofproto->tables; table < &ofproto->tables[ofproto->n_tables];
+ table++) {
+ struct rule *rule, *next_rule;
+ struct cls_cursor cursor;
- error = dpif_recv(p->dpif, &buf);
- if (error) {
- if (error == ENODEV) {
- /* Someone destroyed the datapath behind our back. The caller
- * better destroy us and give up, because we're just going to
- * spin from here on out. */
- static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
- VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
- dpif_name(p->dpif));
- return ENODEV;
+ cls_cursor_init(&cursor, table, 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);
}
- break;
}
-
- handle_odp_msg(p, buf);
}
+ ofopgroup_submit(group);
+}
+
+static void
+ofproto_destroy__(struct ofproto *ofproto)
+{
+ size_t i;
+
+ assert(list_is_empty(&ofproto->pending));
+
+ connmgr_destroy(ofproto->connmgr);
- while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
- process_port_change(p, error, devname);
+ hmap_remove(&all_ofprotos, &ofproto->hmap_node);
+ free(ofproto->name);
+ free(ofproto->type);
+ free(ofproto->mfr_desc);
+ free(ofproto->hw_desc);
+ free(ofproto->sw_desc);
+ free(ofproto->serial_desc);
+ free(ofproto->dp_desc);
+ hmap_destroy(&ofproto->ports);
+ shash_destroy(&ofproto->port_by_name);
+
+ for (i = 0; i < ofproto->n_tables; i++) {
+ assert(classifier_is_empty(&ofproto->tables[i]));
+ classifier_destroy(&ofproto->tables[i]);
}
- while ((error = netdev_monitor_poll(p->netdev_monitor,
- &devname)) != EAGAIN) {
- process_port_change(p, error, devname);
+ free(ofproto->tables);
+
+ hmap_destroy(&ofproto->deletions);
+
+ ofproto->ofproto_class->dealloc(ofproto);
+}
+
+void
+ofproto_destroy(struct ofproto *p)
+{
+ struct ofport *ofport, *next_ofport;
+
+ if (!p) {
+ return;
}
- if (p->in_band) {
- if (time_msec() >= p->next_in_band_update) {
- update_in_band_remotes(p);
- }
- in_band_run(p->in_band);
+ ofproto_flush__(p);
+ HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
+ ofport_destroy(ofport);
}
- LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
- ofconn_run(ofconn);
+ p->ofproto_class->destruct(p);
+ ofproto_destroy__(p);
+}
+
+/* Destroys the datapath with the respective 'name' and 'type'. With the Linux
+ * kernel datapath, for example, this destroys the datapath in the kernel, and
+ * with the netdev-based datapath, it tears down the data structures that
+ * represent the datapath.
+ *
+ * The datapath should not be currently open as an ofproto. */
+int
+ofproto_delete(const char *name, const char *type)
+{
+ const struct ofproto_class *class = ofproto_class_find__(type);
+ return (!class ? EAFNOSUPPORT
+ : !class->del ? EACCES
+ : class->del(type, name));
+}
+
+static void
+process_port_change(struct ofproto *ofproto, int error, char *devname)
+{
+ if (error == ENOBUFS) {
+ reinit_ports(ofproto);
+ } else if (!error) {
+ update_port(ofproto, devname);
+ free(devname);
}
+}
+
+int
+ofproto_run(struct ofproto *p)
+{
+ struct ofport *ofport;
+ char *devname;
+ int error;
- /* Fail-open maintenance. Do this after processing the ofconns since
- * fail-open checks the status of the controller rconn. */
- if (p->fail_open) {
- fail_open_run(p->fail_open);
+ error = p->ofproto_class->run(p);
+ if (error == ENODEV) {
+ /* Someone destroyed the datapath behind our back. The caller
+ * better destroy us and give up, because we're just going to
+ * spin from here on out. */
+ static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
+ VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
+ p->name);
+ return ENODEV;
}
- HMAP_FOR_EACH (ofservice, node, &p->services) {
- struct vconn *vconn;
- int retval;
-
- retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
- if (!retval) {
- struct rconn *rconn;
- char *name;
-
- rconn = rconn_create(ofservice->probe_interval, 0);
- name = ofconn_make_name(p, vconn_get_name(vconn));
- rconn_connect_unreliably(rconn, vconn, name);
- free(name);
-
- ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
- ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
- ofservice->burst_limit);
- } else if (retval != EAGAIN) {
- VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
+ if (p->ofproto_class->port_poll) {
+ while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
+ process_port_change(p, error, devname);
}
}
- for (i = 0; i < p->n_snoops; i++) {
- struct vconn *vconn;
- int retval;
-
- retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
- if (!retval) {
- add_snooper(p, vconn);
- } else if (retval != EAGAIN) {
- VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
+ HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
+ unsigned int change_seq = netdev_change_seq(ofport->netdev);
+ if (ofport->change_seq != change_seq) {
+ ofport->change_seq = change_seq;
+ update_port(p, netdev_get_name(ofport->netdev));
}
}
- if (time_msec() >= p->next_expiration) {
- int delay = ofproto_expire(p);
- p->next_expiration = time_msec() + delay;
- COVERAGE_INC(ofproto_expiration);
- }
- if (p->netflow) {
- netflow_run(p->netflow);
- }
- if (p->sflow) {
- ofproto_sflow_run(p->sflow);
+ switch (p->state) {
+ case S_OPENFLOW:
+ connmgr_run(p->connmgr, handle_openflow);
+ break;
+
+ case S_FLUSH:
+ connmgr_run(p->connmgr, NULL);
+ ofproto_flush__(p);
+ if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
+ connmgr_flushed(p->connmgr);
+ p->state = S_OPENFLOW;
+ }
+ break;
+
+ default:
+ NOT_REACHED();
}
return 0;
}
-int
-ofproto_run2(struct ofproto *p, bool revalidate_all)
+void
+ofproto_wait(struct ofproto *p)
{
- /* Figure out what we need to revalidate now, if anything. */
- struct tag_set revalidate_set = p->revalidate_set;
- if (p->need_revalidate) {
- revalidate_all = true;
+ struct ofport *ofport;
+
+ p->ofproto_class->wait(p);
+ if (p->ofproto_class->port_poll_wait) {
+ p->ofproto_class->port_poll_wait(p);
}
- /* Clear the revalidation flags. */
- tag_set_init(&p->revalidate_set);
- p->need_revalidate = false;
+ HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
+ if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
+ poll_immediate_wake();
+ }
+ }
- /* Now revalidate if there's anything to do. */
- if (revalidate_all || !tag_set_is_empty(&revalidate_set)) {
- struct facet *facet, *next;
+ switch (p->state) {
+ case S_OPENFLOW:
+ connmgr_wait(p->connmgr, true);
+ break;
- HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) {
- if (revalidate_all
- || tag_set_intersects(&revalidate_set, facet->tags)) {
- facet_revalidate(p, facet);
- }
+ case S_FLUSH:
+ connmgr_wait(p->connmgr, false);
+ if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
+ poll_immediate_wake();
}
+ break;
}
+}
- return 0;
+bool
+ofproto_is_alive(const struct ofproto *p)
+{
+ return connmgr_has_controllers(p->connmgr);
}
void
-ofproto_wait(struct ofproto *p)
+ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
+ struct shash *info)
{
- struct ofservice *ofservice;
- struct ofconn *ofconn;
- size_t i;
+ connmgr_get_controller_info(ofproto->connmgr, info);
+}
- dpif_recv_wait(p->dpif);
- dpif_port_poll_wait(p->dpif);
- netdev_monitor_poll_wait(p->netdev_monitor);
- LIST_FOR_EACH (ofconn, node, &p->all_conns) {
- ofconn_wait(ofconn);
- }
- if (p->in_band) {
- poll_timer_wait_until(p->next_in_band_update);
- in_band_wait(p->in_band);
- }
- if (p->fail_open) {
- fail_open_wait(p->fail_open);
- }
- if (p->sflow) {
- ofproto_sflow_wait(p->sflow);
- }
- if (!tag_set_is_empty(&p->revalidate_set)) {
- poll_immediate_wake();
- }
- if (p->need_revalidate) {
- /* Shouldn't happen, but if it does just go around again. */
- VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
- poll_immediate_wake();
- } else if (p->next_expiration != LLONG_MAX) {
- poll_timer_wait_until(p->next_expiration);
- }
- HMAP_FOR_EACH (ofservice, node, &p->services) {
- pvconn_wait(ofservice->pvconn);
- }
- for (i = 0; i < p->n_snoops; i++) {
- pvconn_wait(p->snoops[i]);
+void
+ofproto_free_ofproto_controller_info(struct shash *info)
+{
+ struct shash_node *node;
+
+ SHASH_FOR_EACH (node, info) {
+ struct ofproto_controller_info *cinfo = node->data;
+ while (cinfo->pairs.n) {
+ free((char *) cinfo->pairs.values[--cinfo->pairs.n]);
+ }
+ free(cinfo);
}
+ shash_destroy(info);
}
+/* Makes a deep copy of 'old' into 'port'. */
void
-ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
+ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
{
- tag_set_add(&ofproto->revalidate_set, tag);
+ port->name = xstrdup(old->name);
+ port->type = xstrdup(old->type);
+ port->ofp_port = old->ofp_port;
}
-struct tag_set *
-ofproto_get_revalidate_set(struct ofproto *ofproto)
+/* Frees memory allocated to members of 'ofproto_port'.
+ *
+ * Do not call this function on an ofproto_port obtained from
+ * ofproto_port_dump_next(): that function retains ownership of the data in the
+ * ofproto_port. */
+void
+ofproto_port_destroy(struct ofproto_port *ofproto_port)
{
- return &ofproto->revalidate_set;
+ free(ofproto_port->name);
+ free(ofproto_port->type);
}
-bool
-ofproto_is_alive(const struct ofproto *p)
+/* Initializes 'dump' to begin dumping the ports in an ofproto.
+ *
+ * This function provides no status indication. An error status for the entire
+ * dump operation is provided when it is completed by calling
+ * ofproto_port_dump_done().
+ */
+void
+ofproto_port_dump_start(struct ofproto_port_dump *dump,
+ const struct ofproto *ofproto)
{
- return !hmap_is_empty(&p->controllers);
+ dump->ofproto = ofproto;
+ dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
+ &dump->state);
}
-/* Deletes port number 'odp_port' from the datapath for 'ofproto'.
+/* Attempts to retrieve another port from 'dump', which must have been created
+ * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
+ * 'port' and returns true. On failure, returns false.
*
- * This is almost the same as calling dpif_port_del() directly on the
- * datapath, but it also makes 'ofproto' close its open netdev for the port
- * (if any). This makes it possible to create a new netdev of a different
- * type under the same name, which otherwise the netdev library would refuse
- * to do because of the conflict. (The netdev would eventually get closed on
- * the next trip through ofproto_run(), but this interface is more direct.)
+ * Failure might indicate an actual error or merely that the last port has been
+ * dumped. An error status for the entire dump operation is provided when it
+ * is completed by calling ofproto_port_dump_done().
*
- * Returns 0 if successful, otherwise a positive errno. */
+ * The ofproto owns the data stored in 'port'. It will remain valid until at
+ * least the next time 'dump' is passed to ofproto_port_dump_next() or
+ * ofproto_port_dump_done(). */
+bool
+ofproto_port_dump_next(struct ofproto_port_dump *dump,
+ struct ofproto_port *port)
+{
+ const struct ofproto *ofproto = dump->ofproto;
+
+ if (dump->error) {
+ return false;
+ }
+
+ dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
+ port);
+ if (dump->error) {
+ ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
+ return false;
+ }
+ return true;
+}
+
+/* Completes port table dump operation 'dump', which must have been created
+ * with ofproto_port_dump_start(). Returns 0 if the dump operation was
+ * error-free, otherwise a positive errno value describing the problem. */
int
-ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
+ofproto_port_dump_done(struct ofproto_port_dump *dump)
{
- struct ofport *ofport = get_port(ofproto, odp_port);
- const char *name = ofport ? ofport->opp.name : "<unknown>";
+ const struct ofproto *ofproto = dump->ofproto;
+ if (!dump->error) {
+ dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
+ dump->state);
+ }
+ return dump->error == EOF ? 0 : dump->error;
+}
+
+/* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
+ * and sets '*ofp_portp' to the new port's OpenFlow port number (if 'ofp_portp'
+ * is non-null). On failure, returns a positive errno value and sets
+ * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
+int
+ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
+ uint16_t *ofp_portp)
+{
+ uint16_t ofp_port;
int error;
- error = dpif_port_del(ofproto->dpif, odp_port);
- if (error) {
- VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
- dpif_name(ofproto->dpif), odp_port, name, strerror(error));
- } else if (ofport) {
- /* 'name' is ofport->opp.name and update_port() is going to destroy
- * 'ofport'. Just in case update_port() refers to 'name' after it
- * destroys 'ofport', make a copy of it around the update_port()
- * call. */
- char *devname = xstrdup(name);
- update_port(ofproto, devname);
- free(devname);
+ error = ofproto->ofproto_class->port_add(ofproto, netdev, &ofp_port);
+ if (!error) {
+ update_port(ofproto, netdev_get_name(netdev));
+ }
+ if (ofp_portp) {
+ *ofp_portp = error ? OFPP_NONE : ofp_port;
}
return error;
}
-/* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
- * true if 'odp_port' exists and should be included, false otherwise. */
-bool
-ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
+/* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
+ * initializes '*port' appropriately; on failure, returns a positive errno
+ * value.
+ *
+ * The caller owns the data in 'ofproto_port' and must free it with
+ * ofproto_port_destroy() when it is no longer needed. */
+int
+ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
+ struct ofproto_port *port)
{
- struct ofport *ofport = get_port(ofproto, odp_port);
- return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD);
+ int error;
+
+ error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
+ if (error) {
+ memset(port, 0, sizeof *port);
+ }
+ return error;
}
+/* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
+ * Returns 0 if successful, otherwise a positive errno. */
int
-ofproto_send_packet(struct ofproto *p, const struct flow *flow,
- const union ofp_action *actions, size_t n_actions,
- const struct ofpbuf *packet)
+ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
{
- struct odp_actions odp_actions;
+ struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
+ const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
int error;
- error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
- NULL, NULL, NULL);
- if (error) {
- return error;
+ error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
+ if (!error && ofport) {
+ /* 'name' is the netdev's name and update_port() is going to close the
+ * netdev. Just in case update_port() refers to 'name' after it
+ * destroys 'ofport', make a copy of it around the update_port()
+ * call. */
+ char *devname = xstrdup(name);
+ update_port(ofproto, devname);
+ free(devname);
}
-
- /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
- * error code? */
- dpif_execute(p->dpif, odp_actions.actions, odp_actions.n_actions, packet);
- return 0;
+ return error;
}
-/* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
+/* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
* performs the 'n_actions' actions in 'actions'. The new flow will not
* timeout.
*
* (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 'actions'.
+ *
+ * This is a helper function for in-band control and fail-open. */
void
-ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
+ofproto_add_flow(struct ofproto *ofproto, const struct cls_rule *cls_rule,
const union ofp_action *actions, size_t n_actions)
{
- struct rule *rule;
- rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
- rule_insert(p, rule);
+ 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)) {
+ struct flow_mod fm;
+
+ memset(&fm, 0, sizeof fm);
+ fm.cr = *cls_rule;
+ fm.buffer_id = UINT32_MAX;
+ fm.actions = (union ofp_action *) actions;
+ fm.n_actions = n_actions;
+ add_flow(ofproto, NULL, &fm, NULL);
+ }
}
-void
+/* Searches for a rule with matching criteria exactly equal to 'target' in
+ * ofproto's table 0 and, if it finds one, deletes it.
+ *
+ * This is a helper function for in-band control and fail-open. */
+bool
ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
{
struct rule *rule;
- rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
- target));
- if (rule) {
- rule_remove(ofproto, rule);
+ rule = rule_from_cls_rule(classifier_find_rule_exactly(
+ &ofproto->tables[0], target));
+ if (!rule) {
+ /* No such rule -> success. */
+ return true;
+ } else if (rule->pending) {
+ /* An operation on the rule is already pending -> failure.
+ * Caller must retry later if it's important. */
+ return false;
+ } else {
+ /* Initiate deletion -> success. */
+ struct ofopgroup *group = ofopgroup_create(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);
+ return true;
}
+
}
+/* Starts the process of deleting all of the flows from all of ofproto's flow
+ * tables and then reintroducing the flows required by in-band control and
+ * fail-open. The process will complete in a later call to ofproto_run(). */
void
ofproto_flush_flows(struct ofproto *ofproto)
{
- struct facet *facet, *next_facet;
- struct rule *rule, *next_rule;
- struct cls_cursor cursor;
-
COVERAGE_INC(ofproto_flush);
-
- HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
- /* Mark the facet as not installed so that facet_remove() doesn't
- * bother trying to uninstall it. There is no point in uninstalling it
- * individually since we are about to blow away all the facets with
- * dpif_flow_flush(). */
- facet->installed = false;
- facet_remove(ofproto, facet);
- }
-
- cls_cursor_init(&cursor, &ofproto->cls, NULL);
- CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
- rule_remove(ofproto, rule);
- }
-
- dpif_flow_flush(ofproto->dpif);
- if (ofproto->in_band) {
- in_band_flushed(ofproto->in_band);
- }
- if (ofproto->fail_open) {
- fail_open_flushed(ofproto->fail_open);
- }
+ ofproto->state = S_FLUSH;
}
\f
static void
reinit_ports(struct ofproto *p)
{
- struct svec devnames;
+ struct ofproto_port_dump dump;
+ struct sset devnames;
struct ofport *ofport;
- struct odp_port *odp_ports;
- size_t n_odp_ports;
- size_t i;
+ struct ofproto_port ofproto_port;
+ const char *devname;
COVERAGE_INC(ofproto_reinit_ports);
- svec_init(&devnames);
+ sset_init(&devnames);
HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
- svec_add (&devnames, ofport->opp.name);
+ sset_add(&devnames, netdev_get_name(ofport->netdev));
}
- dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
- for (i = 0; i < n_odp_ports; i++) {
- svec_add (&devnames, odp_ports[i].devname);
+ OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
+ sset_add(&devnames, ofproto_port.name);
}
- free(odp_ports);
- svec_sort_unique(&devnames);
- for (i = 0; i < devnames.n; i++) {
- update_port(p, devnames.names[i]);
+ SSET_FOR_EACH (devname, &devnames) {
+ update_port(p, devname);
}
- svec_destroy(&devnames);
+ sset_destroy(&devnames);
}
-static struct ofport *
-make_ofport(const struct odp_port *odp_port)
+/* Opens and returns a netdev for 'ofproto_port', or a null pointer if the
+ * netdev cannot be opened. On success, also fills in 'opp'. */
+static struct netdev *
+ofport_open(const struct ofproto_port *ofproto_port, struct ofp_phy_port *opp)
{
+ uint32_t curr, advertised, supported, peer;
struct netdev_options netdev_options;
enum netdev_flags flags;
- struct ofport *ofport;
struct netdev *netdev;
int error;
memset(&netdev_options, 0, sizeof netdev_options);
- netdev_options.name = odp_port->devname;
- netdev_options.type = odp_port->type;
+ netdev_options.name = ofproto_port->name;
+ netdev_options.type = ofproto_port->type;
netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
error = netdev_open(&netdev_options, &netdev);
if (error) {
VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
"cannot be opened (%s)",
- odp_port->devname, odp_port->port,
- odp_port->devname, strerror(error));
+ ofproto_port->name, ofproto_port->ofp_port,
+ ofproto_port->name, strerror(error));
return NULL;
}
- ofport = xmalloc(sizeof *ofport);
- ofport->netdev = netdev;
- ofport->odp_port = odp_port->port;
- ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
- netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
- memcpy(ofport->opp.name, odp_port->devname,
- MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
- ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
-
netdev_get_flags(netdev, &flags);
- ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
+ netdev_get_features(netdev, &curr, &advertised, &supported, &peer);
- ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
+ 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);
- netdev_get_features(netdev,
- &ofport->opp.curr, &ofport->opp.advertised,
- &ofport->opp.supported, &ofport->opp.peer);
- return ofport;
+ return netdev;
}
+/* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
+ * port number, and 'config' bits other than OFPPC_PORT_DOWN are
+ * disregarded. */
static bool
-ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
-{
- if (get_port(p, odp_port->port)) {
- VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
- odp_port->port);
- return true;
- } else if (shash_find(&p->port_by_name, odp_port->devname)) {
- VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
- odp_port->devname);
- return true;
- } else {
- return false;
- }
-}
-
-static int
-ofport_equal(const struct ofport *a_, const struct ofport *b_)
+ofport_equal(const struct ofp_phy_port *a, const struct ofp_phy_port *b)
{
- const struct ofp_phy_port *a = &a_->opp;
- const struct ofp_phy_port *b = &b_->opp;
-
BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
- return (a->port_no == b->port_no
- && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
- && !strcmp(a->name, b->name)
+ return (!memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
&& a->state == b->state
- && a->config == b->config
+ && !((a->config ^ b->config) & htonl(OFPPC_PORT_DOWN))
&& a->curr == b->curr
&& a->advertised == b->advertised
&& a->supported == b->supported
&& a->peer == b->peer);
}
+/* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
+ * The caller must ensure that 'p' does not have a conflicting ofport (that is,
+ * one with the same name or port number). */
static void
-send_port_status(struct ofproto *p, const struct ofport *ofport,
- uint8_t reason)
+ofport_install(struct ofproto *p,
+ struct netdev *netdev, const struct ofp_phy_port *opp)
{
- /* XXX Should limit the number of queued port status change messages. */
- struct ofconn *ofconn;
- LIST_FOR_EACH (ofconn, node, &p->all_conns) {
- struct ofp_port_status *ops;
- struct ofpbuf *b;
-
- /* Primary controllers, even slaves, should always get port status
- updates. Otherwise obey ofconn_receives_async_msgs(). */
- if (ofconn->type != OFCONN_PRIMARY
- && !ofconn_receives_async_msgs(ofconn)) {
- continue;
- }
+ const char *netdev_name = netdev_get_name(netdev);
+ struct ofport *ofport;
+ int error;
+
+ /* Create ofport. */
+ ofport = p->ofproto_class->port_alloc();
+ if (!ofport) {
+ error = ENOMEM;
+ goto error;
+ }
+ ofport->ofproto = p;
+ ofport->netdev = netdev;
+ ofport->change_seq = netdev_change_seq(netdev);
+ ofport->opp = *opp;
+ ofport->ofp_port = ntohs(opp->port_no);
+
+ /* Add port to 'p'. */
+ hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
+ shash_add(&p->port_by_name, netdev_name, ofport);
+
+ /* 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);
+ return;
- ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
- ops->reason = reason;
- ops->desc = ofport->opp;
- hton_ofp_phy_port(&ops->desc);
- queue_tx(b, ofconn, NULL);
+error:
+ VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
+ p->name, netdev_name, strerror(error));
+ if (ofport) {
+ ofport_destroy__(ofport);
+ } else {
+ netdev_close(netdev);
}
}
+/* Removes 'ofport' from 'p' and destroys it. */
static void
-ofport_install(struct ofproto *p, struct ofport *ofport)
+ofport_remove(struct ofport *ofport)
{
- const char *netdev_name = ofport->opp.name;
+ connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->opp,
+ OFPPR_DELETE);
+ ofport_destroy(ofport);
+}
- netdev_monitor_add(p->netdev_monitor, ofport->netdev);
- hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
- shash_add(&p->port_by_name, netdev_name, ofport);
- if (p->sflow) {
- ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
+/* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
+ * destroys it. */
+static void
+ofport_remove_with_name(struct ofproto *ofproto, const char *name)
+{
+ struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
+ if (port) {
+ ofport_remove(port);
}
}
+/* Updates 'port' within 'ofproto' with the new 'netdev' and 'opp'.
+ *
+ * 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_remove(struct ofproto *p, struct ofport *ofport)
+ofport_modified(struct ofport *port, struct ofp_phy_port *opp)
+{
+ 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;
+
+ connmgr_send_port_status(port->ofproto->connmgr, &port->opp, OFPPR_MODIFY);
+}
+
+void
+ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
{
- netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
- hmap_remove(&p->ports, &ofport->hmap_node);
- shash_delete(&p->port_by_name,
- shash_find(&p->port_by_name, ofport->opp.name));
- if (p->sflow) {
- ofproto_sflow_del_port(p->sflow, ofport->odp_port);
+ struct ofport *port = ofproto_get_port(ofproto, ofp_port);
+ if (port) {
+ if (port->ofproto->ofproto_class->set_cfm) {
+ port->ofproto->ofproto_class->set_cfm(port, NULL);
+ }
+ if (port->ofproto->ofproto_class->bundle_remove) {
+ port->ofproto->ofproto_class->bundle_remove(port);
+ }
}
}
static void
-ofport_free(struct ofport *ofport)
+ofport_destroy__(struct ofport *port)
{
- if (ofport) {
- netdev_close(ofport->netdev);
- free(ofport);
- }
+ struct ofproto *ofproto = port->ofproto;
+ const char *name = netdev_get_name(port->netdev);
+
+ hmap_remove(&ofproto->ports, &port->hmap_node);
+ shash_delete(&ofproto->port_by_name,
+ shash_find(&ofproto->port_by_name, name));
+
+ netdev_close(port->netdev);
+ ofproto->ofproto_class->port_dealloc(port);
+}
+
+static void
+ofport_destroy(struct ofport *port)
+{
+ if (port) {
+ port->ofproto->ofproto_class->port_destruct(port);
+ ofport_destroy__(port);
+ }
}
-static struct ofport *
-get_port(const struct ofproto *ofproto, uint16_t odp_port)
+struct ofport *
+ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
{
struct ofport *port;
HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
- hash_int(odp_port, 0), &ofproto->ports) {
- if (port->odp_port == odp_port) {
+ hash_int(ofp_port, 0), &ofproto->ports) {
+ if (port->ofp_port == ofp_port) {
return port;
}
}
}
static void
-update_port(struct ofproto *p, const char *devname)
+update_port(struct ofproto *ofproto, const char *name)
{
- struct odp_port odp_port;
- struct ofport *old_ofport;
- struct ofport *new_ofport;
- int error;
-
- COVERAGE_INC(ofproto_update_port);
-
- /* Query the datapath for port information. */
- error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
-
- /* Find the old ofport. */
- old_ofport = shash_find_data(&p->port_by_name, devname);
- if (!error) {
- if (!old_ofport) {
- /* There's no port named 'devname' but there might be a port with
- * the same port number. This could happen if a port is deleted
- * and then a new one added in its place very quickly, or if a port
- * is renamed. In the former case we want to send an OFPPR_DELETE
- * and an OFPPR_ADD, and in the latter case we want to send a
- * single OFPPR_MODIFY. We can distinguish the cases by comparing
- * the old port's ifindex against the new port, or perhaps less
- * reliably but more portably by comparing the old port's MAC
- * against the new port's MAC. However, this code isn't that smart
- * and always sends an OFPPR_MODIFY (XXX). */
- old_ofport = get_port(p, odp_port.port);
- }
- } else if (error != ENOENT && error != ENODEV) {
- VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
- "%s", strerror(error));
- return;
- }
-
- /* Create a new ofport. */
- new_ofport = !error ? make_ofport(&odp_port) : NULL;
-
- /* Eliminate a few pathological cases. */
- if (!old_ofport && !new_ofport) {
- return;
- } else if (old_ofport && new_ofport) {
- /* Most of the 'config' bits are OpenFlow soft state, but
- * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
- * OpenFlow bits from old_ofport. (make_ofport() only sets
- * OFPPC_PORT_DOWN and leaves the other bits 0.) */
- new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
-
- if (ofport_equal(old_ofport, new_ofport)) {
- /* False alarm--no change. */
- ofport_free(new_ofport);
- return;
- }
- }
-
- /* Now deal with the normal cases. */
- if (old_ofport) {
- ofport_remove(p, old_ofport);
- }
- if (new_ofport) {
- ofport_install(p, new_ofport);
- }
- send_port_status(p, new_ofport ? new_ofport : old_ofport,
- (!old_ofport ? OFPPR_ADD
- : !new_ofport ? OFPPR_DELETE
- : OFPPR_MODIFY));
- ofport_free(old_ofport);
-}
-
-static int
-init_ports(struct ofproto *p)
-{
- struct odp_port *ports;
- size_t n_ports;
- size_t i;
- int error;
-
- error = dpif_port_list(p->dpif, &ports, &n_ports);
- if (error) {
- return error;
- }
-
- for (i = 0; i < n_ports; i++) {
- const struct odp_port *odp_port = &ports[i];
- if (!ofport_conflicts(p, odp_port)) {
- struct ofport *ofport = make_ofport(odp_port);
- if (ofport) {
- ofport_install(p, ofport);
- }
- }
- }
- free(ports);
- return 0;
-}
-\f
-static struct ofconn *
-ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
-{
- struct ofconn *ofconn = xzalloc(sizeof *ofconn);
- ofconn->ofproto = p;
- list_push_back(&p->all_conns, &ofconn->node);
- ofconn->rconn = rconn;
- ofconn->type = type;
- ofconn->flow_format = NXFF_OPENFLOW10;
- ofconn->role = NX_ROLE_OTHER;
- ofconn->packet_in_counter = rconn_packet_counter_create ();
- ofconn->pktbuf = NULL;
- ofconn->miss_send_len = 0;
- ofconn->reply_counter = rconn_packet_counter_create ();
- return ofconn;
-}
-
-static void
-ofconn_destroy(struct ofconn *ofconn)
-{
- if (ofconn->type == OFCONN_PRIMARY) {
- hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
- }
- discovery_destroy(ofconn->discovery);
-
- list_remove(&ofconn->node);
- switch_status_unregister(ofconn->ss);
- rconn_destroy(ofconn->rconn);
- rconn_packet_counter_destroy(ofconn->packet_in_counter);
- rconn_packet_counter_destroy(ofconn->reply_counter);
- pktbuf_destroy(ofconn->pktbuf);
- free(ofconn);
-}
-
-static void
-ofconn_run(struct ofconn *ofconn)
-{
- struct ofproto *p = ofconn->ofproto;
- int iteration;
- size_t i;
-
- if (ofconn->discovery) {
- char *controller_name;
- if (rconn_is_connectivity_questionable(ofconn->rconn)) {
- discovery_question_connectivity(ofconn->discovery);
- }
- if (discovery_run(ofconn->discovery, &controller_name)) {
- if (controller_name) {
- char *ofconn_name = ofconn_make_name(p, controller_name);
- rconn_connect(ofconn->rconn, controller_name, ofconn_name);
- free(ofconn_name);
- } else {
- rconn_disconnect(ofconn->rconn);
- }
- }
- }
-
- for (i = 0; i < N_SCHEDULERS; i++) {
- pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
- }
-
- rconn_run(ofconn->rconn);
-
- if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
- /* Limit the number of iterations to prevent other tasks from
- * starving. */
- for (iteration = 0; iteration < 50; iteration++) {
- struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
- if (!of_msg) {
- break;
- }
- if (p->fail_open) {
- fail_open_maybe_recover(p->fail_open);
- }
- handle_openflow(ofconn, of_msg);
- ofpbuf_delete(of_msg);
- }
- }
-
- if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
- ofconn_destroy(ofconn);
- }
-}
-
-static void
-ofconn_wait(struct ofconn *ofconn)
-{
- int i;
-
- if (ofconn->discovery) {
- discovery_wait(ofconn->discovery);
- }
- for (i = 0; i < N_SCHEDULERS; i++) {
- pinsched_wait(ofconn->schedulers[i]);
- }
- rconn_run_wait(ofconn->rconn);
- if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
- rconn_recv_wait(ofconn->rconn);
- } else {
- COVERAGE_INC(ofproto_ofconn_stuck);
- }
-}
-
-/* Returns true if 'ofconn' should receive asynchronous messages. */
-static bool
-ofconn_receives_async_msgs(const struct ofconn *ofconn)
-{
- if (ofconn->type == OFCONN_PRIMARY) {
- /* Primary controllers always get asynchronous messages unless they
- * have configured themselves as "slaves". */
- return ofconn->role != NX_ROLE_SLAVE;
- } else {
- /* Service connections don't get asynchronous messages unless they have
- * explicitly asked for them by setting a nonzero miss send length. */
- return ofconn->miss_send_len > 0;
- }
-}
-
-/* Returns a human-readable name for an OpenFlow connection between 'ofproto'
- * and 'target', suitable for use in log messages for identifying the
- * connection.
- *
- * The name is dynamically allocated. The caller should free it (with free())
- * when it is no longer needed. */
-static char *
-ofconn_make_name(const struct ofproto *ofproto, const char *target)
-{
- return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
-}
-
-static void
-ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
-{
- int i;
-
- for (i = 0; i < N_SCHEDULERS; i++) {
- struct pinsched **s = &ofconn->schedulers[i];
-
- if (rate > 0) {
- if (!*s) {
- *s = pinsched_create(rate, burst,
- ofconn->ofproto->switch_status);
- } else {
- pinsched_set_limits(*s, rate, burst);
- }
- } else {
- pinsched_destroy(*s);
- *s = NULL;
- }
- }
-}
-\f
-static void
-ofservice_reconfigure(struct ofservice *ofservice,
- const struct ofproto_controller *c)
-{
- ofservice->probe_interval = c->probe_interval;
- ofservice->rate_limit = c->rate_limit;
- ofservice->burst_limit = c->burst_limit;
-}
-
-/* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
- * positive errno value. */
-static int
-ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
-{
- struct ofservice *ofservice;
- struct pvconn *pvconn;
- int error;
-
- error = pvconn_open(c->target, &pvconn);
- if (error) {
- return error;
- }
-
- ofservice = xzalloc(sizeof *ofservice);
- hmap_insert(&ofproto->services, &ofservice->node,
- hash_string(c->target, 0));
- ofservice->pvconn = pvconn;
-
- ofservice_reconfigure(ofservice, c);
-
- return 0;
-}
-
-static void
-ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
-{
- hmap_remove(&ofproto->services, &ofservice->node);
- pvconn_close(ofservice->pvconn);
- free(ofservice);
-}
-
-/* Finds and returns the ofservice within 'ofproto' that has the given
- * 'target', or a null pointer if none exists. */
-static struct ofservice *
-ofservice_lookup(struct ofproto *ofproto, const char *target)
-{
- struct ofservice *ofservice;
-
- HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
- &ofproto->services) {
- if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
- return ofservice;
- }
- }
- return NULL;
-}
-\f
-/* Returns true if 'rule' should be hidden from the controller.
- *
- * Rules with priority higher than UINT16_MAX are set up by ofproto itself
- * (e.g. by in-band control) and are intentionally hidden from the
- * controller. */
-static bool
-rule_is_hidden(const struct rule *rule)
-{
- return rule->cr.priority > UINT16_MAX;
-}
-
-/* Creates and returns a new rule initialized as specified.
- *
- * The caller is responsible for inserting the rule into the classifier (with
- * rule_insert()). */
-static struct rule *
-rule_create(const struct cls_rule *cls_rule,
- const union ofp_action *actions, size_t n_actions,
- uint16_t idle_timeout, uint16_t hard_timeout,
- ovs_be64 flow_cookie, bool send_flow_removed)
-{
- struct rule *rule = xzalloc(sizeof *rule);
- rule->cr = *cls_rule;
- rule->idle_timeout = idle_timeout;
- rule->hard_timeout = hard_timeout;
- rule->flow_cookie = flow_cookie;
- rule->used = rule->created = time_msec();
- rule->send_flow_removed = send_flow_removed;
- list_init(&rule->facets);
- if (n_actions > 0) {
- rule->n_actions = n_actions;
- rule->actions = xmemdup(actions, n_actions * sizeof *actions);
- }
-
- return rule;
-}
-
-static struct rule *
-rule_from_cls_rule(const struct cls_rule *cls_rule)
-{
- return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
-}
-
-static void
-rule_free(struct rule *rule)
-{
- free(rule->actions);
- free(rule);
-}
-
-/* Destroys 'rule' and iterates through all of its facets and revalidates them,
- * destroying any that no longer has a rule (which is probably all of them).
- *
- * The caller must have already removed 'rule' from the classifier. */
-static void
-rule_destroy(struct ofproto *ofproto, struct rule *rule)
-{
- struct facet *facet, *next_facet;
- LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
- facet_revalidate(ofproto, facet);
- }
- rule_free(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, ovs_be16 out_port)
-{
- const union ofp_action *oa;
- struct actions_iterator i;
-
- if (out_port == htons(OFPP_NONE)) {
- return true;
- }
- for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
- oa = actions_next(&i)) {
- if (action_outputs_to_port(oa, out_port)) {
- return true;
- }
- }
- return false;
-}
-
-/* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
- * 'packet', which arrived on 'in_port'.
- *
- * Takes ownership of 'packet'. */
-static bool
-execute_odp_actions(struct ofproto *ofproto, uint16_t in_port,
- const union odp_action *actions, size_t n_actions,
- struct ofpbuf *packet)
-{
- if (n_actions == 1 && actions[0].type == ODPAT_CONTROLLER) {
- /* As an optimization, avoid a round-trip from userspace to kernel to
- * userspace. This also avoids possibly filling up kernel packet
- * buffers along the way. */
- struct odp_msg *msg;
-
- msg = ofpbuf_push_uninit(packet, sizeof *msg);
- msg->type = _ODPL_ACTION_NR;
- msg->length = sizeof(struct odp_msg) + packet->size;
- msg->port = in_port;
- msg->reserved = 0;
- msg->arg = actions[0].controller.arg;
-
- send_packet_in(ofproto, packet);
-
- return true;
- } else {
- int error;
-
- error = dpif_execute(ofproto->dpif, actions, n_actions, packet);
- ofpbuf_delete(packet);
- return !error;
- }
-}
-
-/* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
- * statistics appropriately. 'packet' must have at least sizeof(struct
- * ofp_packet_in) bytes of headroom.
- *
- * For correct results, 'packet' must actually be in 'facet''s flow; that is,
- * applying flow_extract() to 'packet' would yield the same flow as
- * 'facet->flow'.
- *
- * 'facet' must have accurately composed ODP actions; that is, it must not be
- * in need of revalidation.
- *
- * Takes ownership of 'packet'. */
-static void
-facet_execute(struct ofproto *ofproto, struct facet *facet,
- struct ofpbuf *packet)
-{
- struct odp_flow_stats stats;
-
- assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
-
- flow_extract_stats(&facet->flow, packet, &stats);
- if (execute_odp_actions(ofproto, facet->flow.in_port,
- facet->actions, facet->n_actions, packet)) {
- facet_update_stats(ofproto, facet, &stats);
- facet->used = time_msec();
- netflow_flow_update_time(ofproto->netflow,
- &facet->nf_flow, facet->used);
- }
-}
-
-/* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
- * statistics (or the statistics for one of its facets) appropriately.
- * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
- *
- * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
- * with statistics for 'packet' either way.
- *
- * Takes ownership of 'packet'. */
-static void
-rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
- struct ofpbuf *packet)
-{
- struct facet *facet;
- struct odp_actions a;
- struct flow flow;
- size_t size;
-
- assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
-
- flow_extract(packet, 0, in_port, &flow);
-
- /* First look for a related facet. If we find one, account it to that. */
- facet = facet_lookup_valid(ofproto, &flow);
- if (facet && facet->rule == rule) {
- facet_execute(ofproto, facet, packet);
- return;
- }
-
- /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
- * create a new facet for it and use that. */
- if (rule_lookup(ofproto, &flow) == rule) {
- facet = facet_create(ofproto, rule, &flow, packet);
- facet_execute(ofproto, facet, packet);
- facet_install(ofproto, facet, true);
- return;
- }
-
- /* We can't account anything to a facet. If we were to try, then that
- * facet would have a non-matching rule, busting our invariants. */
- if (xlate_actions(rule->actions, rule->n_actions, &flow, ofproto,
- packet, &a, NULL, 0, NULL)) {
- ofpbuf_delete(packet);
- return;
- }
- size = packet->size;
- if (execute_odp_actions(ofproto, in_port,
- a.actions, a.n_actions, packet)) {
- rule->used = time_msec();
- rule->packet_count++;
- rule->byte_count += size;
- }
-}
-
-/* Inserts 'rule' into 'p''s flow table. */
-static void
-rule_insert(struct ofproto *p, struct rule *rule)
-{
- struct rule *displaced_rule;
-
- displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
- if (displaced_rule) {
- rule_destroy(p, displaced_rule);
- }
- p->need_revalidate = true;
-}
-
-/* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
- * 'flow' and an example 'packet' within that flow.
- *
- * The caller must already have determined that no facet with an identical
- * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
- * 'ofproto''s classifier table. */
-static struct facet *
-facet_create(struct ofproto *ofproto, struct rule *rule,
- const struct flow *flow, const struct ofpbuf *packet)
-{
- struct facet *facet;
-
- facet = xzalloc(sizeof *facet);
- facet->used = time_msec();
- hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
- list_push_back(&rule->facets, &facet->list_node);
- facet->rule = rule;
- facet->flow = *flow;
- netflow_flow_init(&facet->nf_flow);
- netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
-
- facet_make_actions(ofproto, facet, packet);
-
- return facet;
-}
-
-static void
-facet_free(struct facet *facet)
-{
- free(facet->actions);
- free(facet);
-}
-
-/* Remove 'rule' from 'ofproto' and free up the associated memory:
- *
- * - Removes 'rule' from the classifier.
- *
- * - If 'rule' has facets, revalidates them (and possibly uninstalls and
- * destroys them), via rule_destroy().
- */
-static void
-rule_remove(struct ofproto *ofproto, struct rule *rule)
-{
- COVERAGE_INC(ofproto_del_rule);
- ofproto->need_revalidate = true;
- classifier_remove(&ofproto->cls, &rule->cr);
- rule_destroy(ofproto, rule);
-}
-
-/* Remove 'facet' from 'ofproto' and free up the associated memory:
- *
- * - If 'facet' was installed in the datapath, uninstalls it and updates its
- * rule's statistics, via facet_uninstall().
- *
- * - Removes 'facet' from its rule and from ofproto->facets.
- */
-static void
-facet_remove(struct ofproto *ofproto, struct facet *facet)
-{
- facet_uninstall(ofproto, facet);
- facet_flush_stats(ofproto, facet);
- hmap_remove(&ofproto->facets, &facet->hmap_node);
- list_remove(&facet->list_node);
- facet_free(facet);
-}
-
-/* Composes the ODP actions for 'facet' based on its rule's actions. */
-static void
-facet_make_actions(struct ofproto *p, struct facet *facet,
- const struct ofpbuf *packet)
-{
- const struct rule *rule = facet->rule;
- struct odp_actions a;
- size_t actions_len;
-
- xlate_actions(rule->actions, rule->n_actions, &facet->flow, p,
- packet, &a, &facet->tags, &facet->may_install,
- &facet->nf_flow.output_iface);
-
- actions_len = a.n_actions * sizeof *a.actions;
- if (facet->n_actions != a.n_actions
- || memcmp(facet->actions, a.actions, actions_len)) {
- free(facet->actions);
- facet->n_actions = a.n_actions;
- facet->actions = xmemdup(a.actions, actions_len);
- }
-}
-
-static int
-facet_put__(struct ofproto *ofproto, struct facet *facet, int flags,
- struct odp_flow_put *put)
-{
- memset(&put->flow.stats, 0, sizeof put->flow.stats);
- odp_flow_key_from_flow(&put->flow.key, &facet->flow);
- put->flow.actions = facet->actions;
- put->flow.n_actions = facet->n_actions;
- put->flow.flags = 0;
- put->flags = flags;
- return dpif_flow_put(ofproto->dpif, put);
-}
-
-/* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
- * 'zero_stats' is true, clears any existing statistics from the datapath for
- * 'facet'. */
-static void
-facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
-{
- if (facet->may_install) {
- struct odp_flow_put put;
- int flags;
-
- flags = ODPPF_CREATE | ODPPF_MODIFY;
- if (zero_stats) {
- flags |= ODPPF_ZERO_STATS;
- }
- if (!facet_put__(p, facet, flags, &put)) {
- facet->installed = true;
- }
- }
-}
-
-/* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
- * to the accounting hook function in the ofhooks structure. */
-static void
-facet_account(struct ofproto *ofproto,
- struct facet *facet, uint64_t extra_bytes)
-{
- uint64_t total_bytes = facet->byte_count + extra_bytes;
-
- if (ofproto->ofhooks->account_flow_cb
- && total_bytes > facet->accounted_bytes)
- {
- ofproto->ofhooks->account_flow_cb(
- &facet->flow, facet->tags, facet->actions, facet->n_actions,
- total_bytes - facet->accounted_bytes, ofproto->aux);
- facet->accounted_bytes = total_bytes;
- }
-}
-
-/* If 'rule' is installed in the datapath, uninstalls it. */
-static void
-facet_uninstall(struct ofproto *p, struct facet *facet)
-{
- if (facet->installed) {
- struct odp_flow odp_flow;
-
- odp_flow_key_from_flow(&odp_flow.key, &facet->flow);
- odp_flow.actions = NULL;
- odp_flow.n_actions = 0;
- odp_flow.flags = 0;
- if (!dpif_flow_del(p->dpif, &odp_flow)) {
- facet_update_stats(p, facet, &odp_flow.stats);
- }
- facet->installed = false;
- }
-}
-
-/* Returns true if the only action for 'facet' is to send to the controller.
- * (We don't report NetFlow expiration messages for such facets because they
- * are just part of the control logic for the network, not real traffic). */
-static bool
-facet_is_controller_flow(struct facet *facet)
-{
- return (facet
- && facet->rule->n_actions == 1
- && action_outputs_to_port(&facet->rule->actions[0],
- htons(OFPP_CONTROLLER)));
-}
-
-/* Folds all of 'facet''s statistics into its rule. Also updates the
- * accounting ofhook and emits a NetFlow expiration if appropriate. */
-static void
-facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
-{
- facet_account(ofproto, facet, 0);
-
- if (ofproto->netflow && !facet_is_controller_flow(facet)) {
- struct ofexpired expired;
- expired.flow = facet->flow;
- expired.packet_count = facet->packet_count;
- expired.byte_count = facet->byte_count;
- expired.used = facet->used;
- netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
- }
-
- facet->rule->packet_count += facet->packet_count;
- facet->rule->byte_count += facet->byte_count;
-
- /* Reset counters to prevent double counting if 'facet' ever gets
- * reinstalled. */
- facet->packet_count = 0;
- facet->byte_count = 0;
- facet->accounted_bytes = 0;
-
- netflow_flow_clear(&facet->nf_flow);
-}
-
-/* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
- * Returns it if found, otherwise a null pointer.
- *
- * The returned facet might need revalidation; use facet_lookup_valid()
- * instead if that is important. */
-static struct facet *
-facet_find(struct ofproto *ofproto, const struct flow *flow)
-{
- struct facet *facet;
-
- HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
- &ofproto->facets) {
- if (flow_equal(flow, &facet->flow)) {
- return facet;
- }
- }
-
- return NULL;
-}
-
-/* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
- * Returns it if found, otherwise a null pointer.
- *
- * The returned facet is guaranteed to be valid. */
-static struct facet *
-facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
-{
- struct facet *facet = facet_find(ofproto, flow);
-
- /* The facet we found might not be valid, since we could be in need of
- * revalidation. If it is not valid, don't return it. */
- if (facet
- && ofproto->need_revalidate
- && !facet_revalidate(ofproto, facet)) {
- COVERAGE_INC(ofproto_invalidated);
- return NULL;
- }
-
- return facet;
-}
-
-/* Re-searches 'ofproto''s classifier for a rule matching 'facet':
- *
- * - If the rule found is different from 'facet''s current rule, moves
- * 'facet' to the new rule and recompiles its actions.
- *
- * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
- * where it is and recompiles its actions anyway.
- *
- * - If there is none, destroys 'facet'.
- *
- * Returns true if 'facet' still exists, false if it has been destroyed. */
-static bool
-facet_revalidate(struct ofproto *ofproto, struct facet *facet)
-{
- struct rule *new_rule;
- struct odp_actions a;
- size_t actions_len;
- uint16_t new_nf_output_iface;
- bool actions_changed;
-
- COVERAGE_INC(facet_revalidate);
-
- /* Determine the new rule. */
- new_rule = rule_lookup(ofproto, &facet->flow);
- if (!new_rule) {
- /* No new rule, so delete the facet. */
- facet_remove(ofproto, facet);
- return false;
- }
-
- /* Calculate new ODP actions.
- *
- * We are very cautious about actually modifying 'facet' state at this
- * point, because we might need to, e.g., emit a NetFlow expiration and, if
- * so, we need to have the old state around to properly compose it. */
- xlate_actions(new_rule->actions, new_rule->n_actions, &facet->flow,
- ofproto, NULL, &a, &facet->tags, &facet->may_install,
- &new_nf_output_iface);
- actions_len = a.n_actions * sizeof *a.actions;
- actions_changed = (facet->n_actions != a.n_actions
- || memcmp(facet->actions, a.actions, actions_len));
-
- /* If the ODP actions changed or the installability changed, then we need
- * to talk to the datapath. */
- if (actions_changed || facet->may_install != facet->installed) {
- if (facet->may_install) {
- struct odp_flow_put put;
-
- memset(&put.flow.stats, 0, sizeof put.flow.stats);
- odp_flow_key_from_flow(&put.flow.key, &facet->flow);
- put.flow.actions = a.actions;
- put.flow.n_actions = a.n_actions;
- put.flow.flags = 0;
- put.flags = ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS;
- dpif_flow_put(ofproto->dpif, &put);
-
- facet_update_stats(ofproto, facet, &put.flow.stats);
- } else {
- facet_uninstall(ofproto, facet);
- }
-
- /* The datapath flow is gone or has zeroed stats, so push stats out of
- * 'facet' into 'rule'. */
- facet_flush_stats(ofproto, facet);
- }
-
- /* Update 'facet' now that we've taken care of all the old state. */
- facet->nf_flow.output_iface = new_nf_output_iface;
- if (actions_changed) {
- free(facet->actions);
- facet->n_actions = a.n_actions;
- facet->actions = xmemdup(a.actions, actions_len);
- }
- if (facet->rule != new_rule) {
- COVERAGE_INC(facet_changed_rule);
- list_remove(&facet->list_node);
- list_push_back(&new_rule->facets, &facet->list_node);
- facet->rule = new_rule;
- facet->used = new_rule->created;
- }
-
- return true;
-}
-\f
-static void
-queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
- struct rconn_packet_counter *counter)
-{
- update_openflow_length(msg);
- if (rconn_send(ofconn->rconn, msg, counter)) {
- ofpbuf_delete(msg);
- }
-}
-
-static void
-send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
- int error)
-{
- struct ofpbuf *buf = make_ofp_error_msg(error, oh);
- if (buf) {
- COVERAGE_INC(ofproto_error);
- queue_tx(buf, ofconn, ofconn->reply_counter);
- }
-}
-
-static void
-hton_ofp_phy_port(struct ofp_phy_port *opp)
-{
- opp->port_no = htons(opp->port_no);
- opp->config = htonl(opp->config);
- opp->state = htonl(opp->state);
- opp->curr = htonl(opp->curr);
- opp->advertised = htonl(opp->advertised);
- opp->supported = htonl(opp->supported);
- opp->peer = htonl(opp->peer);
-}
-
-static int
-handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
-{
- struct ofp_header *rq = oh;
- queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
- return 0;
-}
-
-static int
-handle_features_request(struct ofconn *ofconn, struct ofp_header *oh)
-{
- struct ofp_switch_features *osf;
- struct ofpbuf *buf;
- struct ofport *port;
-
- osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
- osf->datapath_id = htonll(ofconn->ofproto->datapath_id);
- osf->n_buffers = htonl(pktbuf_capacity());
- osf->n_tables = 2;
- osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
- OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
- osf->actions = htonl((1u << OFPAT_OUTPUT) |
- (1u << OFPAT_SET_VLAN_VID) |
- (1u << OFPAT_SET_VLAN_PCP) |
- (1u << OFPAT_STRIP_VLAN) |
- (1u << OFPAT_SET_DL_SRC) |
- (1u << OFPAT_SET_DL_DST) |
- (1u << OFPAT_SET_NW_SRC) |
- (1u << OFPAT_SET_NW_DST) |
- (1u << OFPAT_SET_NW_TOS) |
- (1u << OFPAT_SET_TP_SRC) |
- (1u << OFPAT_SET_TP_DST) |
- (1u << OFPAT_ENQUEUE));
-
- HMAP_FOR_EACH (port, hmap_node, &ofconn->ofproto->ports) {
- hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
- }
-
- queue_tx(buf, ofconn, ofconn->reply_counter);
- return 0;
-}
-
-static int
-handle_get_config_request(struct ofconn *ofconn, struct ofp_header *oh)
-{
- struct ofpbuf *buf;
- struct ofp_switch_config *osc;
- uint16_t flags;
- bool drop_frags;
-
- /* Figure out flags. */
- dpif_get_drop_frags(ofconn->ofproto->dpif, &drop_frags);
- flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
-
- /* Send reply. */
- osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
- osc->flags = htons(flags);
- osc->miss_send_len = htons(ofconn->miss_send_len);
- queue_tx(buf, ofconn, ofconn->reply_counter);
-
- return 0;
-}
-
-static int
-handle_set_config(struct ofconn *ofconn, struct ofp_switch_config *osc)
-{
- uint16_t flags;
- int error;
-
- error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
- if (error) {
- return error;
- }
- flags = ntohs(osc->flags);
-
- if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
- switch (flags & OFPC_FRAG_MASK) {
- case OFPC_FRAG_NORMAL:
- dpif_set_drop_frags(ofconn->ofproto->dpif, false);
- break;
- case OFPC_FRAG_DROP:
- dpif_set_drop_frags(ofconn->ofproto->dpif, true);
- break;
- default:
- VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
- osc->flags);
- break;
- }
- }
-
- ofconn->miss_send_len = ntohs(osc->miss_send_len);
-
- return 0;
-}
-
-static void
-add_controller_action(struct odp_actions *actions, uint16_t max_len)
-{
- union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
- a->controller.arg = max_len;
-}
-
-struct action_xlate_ctx {
- /* Input. */
- struct flow flow; /* Flow to which these actions correspond. */
- int recurse; /* Recursion level, via xlate_table_action. */
- struct ofproto *ofproto;
- const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
- * null pointer if we are revalidating
- * without a packet to refer to. */
-
- /* Output. */
- struct odp_actions *out; /* Datapath actions. */
- tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
- bool may_set_up_flow; /* True ordinarily; false if the actions must
- * be reassessed for every packet. */
- uint16_t nf_output_iface; /* Output interface index for NetFlow. */
-};
-
-/* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
- * flow translation. */
-#define MAX_RESUBMIT_RECURSION 8
-
-static void do_xlate_actions(const union ofp_action *in, size_t n_in,
- struct action_xlate_ctx *ctx);
-
-static void
-add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
-{
- const struct ofport *ofport = get_port(ctx->ofproto, port);
-
- if (ofport) {
- if (ofport->opp.config & OFPPC_NO_FWD) {
- /* Forwarding disabled on port. */
- return;
- }
- } else {
- /*
- * We don't have an ofport record for this port, but it doesn't hurt to
- * allow forwarding to it anyhow. Maybe such a port will appear later
- * and we're pre-populating the flow table.
- */
- }
-
- odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
- ctx->nf_output_iface = port;
-}
-
-static struct rule *
-rule_lookup(struct ofproto *ofproto, const struct flow *flow)
-{
- return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
-}
-
-static void
-xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
-{
- if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
- uint16_t old_in_port;
- struct rule *rule;
+ struct ofproto_port ofproto_port;
+ struct ofp_phy_port opp;
+ struct netdev *netdev;
+ struct ofport *port;
- /* Look up a flow with 'in_port' as the input port. Then restore the
- * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
- * have surprising behavior). */
- old_in_port = ctx->flow.in_port;
- ctx->flow.in_port = in_port;
- rule = rule_lookup(ctx->ofproto, &ctx->flow);
- ctx->flow.in_port = old_in_port;
+ COVERAGE_INC(ofproto_update_port);
- if (rule) {
- ctx->recurse++;
- do_xlate_actions(rule->actions, rule->n_actions, ctx);
- ctx->recurse--;
- }
- } else {
- struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
+ /* Fetch 'name''s location and properties from the datapath. */
+ netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
+ ? ofport_open(&ofproto_port, &opp)
+ : 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;
+
+ /* 'name' hasn't changed location. Any properties changed? */
+ if (!ofport_equal(&port->opp, &opp)) {
+ ofport_modified(port, &opp);
+ }
- VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
- MAX_RESUBMIT_RECURSION);
- }
-}
+ /* Install the newly opened netdev in case it has changed.
+ * Don't close the old netdev yet in case port_modified has to
+ * remove a retained reference to it.*/
+ port->netdev = netdev;
+ port->change_seq = netdev_change_seq(netdev);
-static void
-flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
- uint16_t *nf_output_iface, struct odp_actions *actions)
-{
- struct ofport *ofport;
+ if (port->ofproto->ofproto_class->port_modified) {
+ port->ofproto->ofproto_class->port_modified(port);
+ }
- HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
- uint16_t odp_port = ofport->odp_port;
- if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
- odp_actions_add(actions, ODPAT_OUTPUT)->output.port = odp_port;
+ netdev_close(old_netdev);
+ } else {
+ /* If 'port' is nonnull then its name differs from 'name' and thus
+ * we should delete it. If we think there's a port named 'name'
+ * then its port number must be wrong now so delete it too. */
+ if (port) {
+ ofport_remove(port);
+ }
+ ofport_remove_with_name(ofproto, name);
+ ofport_install(ofproto, netdev, &opp);
}
+ } else {
+ /* Any port named 'name' is gone now. */
+ ofport_remove_with_name(ofproto, name);
}
- *nf_output_iface = NF_OUT_FLOOD;
+ ofproto_port_destroy(&ofproto_port);
}
-static void
-xlate_output_action__(struct action_xlate_ctx *ctx,
- uint16_t port, uint16_t max_len)
+static int
+init_ports(struct ofproto *p)
{
- uint16_t odp_port;
- uint16_t prev_nf_output_iface = ctx->nf_output_iface;
-
- ctx->nf_output_iface = NF_OUT_DROP;
+ struct ofproto_port_dump dump;
+ struct ofproto_port ofproto_port;
+
+ OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
+ uint16_t ofp_port = ofproto_port.ofp_port;
+ if (ofproto_get_port(p, ofp_port)) {
+ VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
+ ofp_port);
+ } else if (shash_find(&p->port_by_name, ofproto_port.name)) {
+ VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
+ ofproto_port.name);
+ } else {
+ struct ofp_phy_port opp;
+ struct netdev *netdev;
- switch (port) {
- case OFPP_IN_PORT:
- add_output_action(ctx, ctx->flow.in_port);
- break;
- case OFPP_TABLE:
- xlate_table_action(ctx, ctx->flow.in_port);
- break;
- case OFPP_NORMAL:
- if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
- ctx->out, &ctx->tags,
- &ctx->nf_output_iface,
- ctx->ofproto->aux)) {
- COVERAGE_INC(ofproto_uninstallable);
- ctx->may_set_up_flow = false;
- }
- break;
- case OFPP_FLOOD:
- flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
- &ctx->nf_output_iface, ctx->out);
- break;
- case OFPP_ALL:
- flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
- &ctx->nf_output_iface, ctx->out);
- break;
- case OFPP_CONTROLLER:
- add_controller_action(ctx->out, max_len);
- break;
- case OFPP_LOCAL:
- add_output_action(ctx, ODPP_LOCAL);
- break;
- default:
- odp_port = ofp_port_to_odp_port(port);
- if (odp_port != ctx->flow.in_port) {
- add_output_action(ctx, odp_port);
+ netdev = ofport_open(&ofproto_port, &opp);
+ if (netdev) {
+ ofport_install(p, netdev, &opp);
+ }
}
- break;
}
- if (prev_nf_output_iface == NF_OUT_FLOOD) {
- ctx->nf_output_iface = NF_OUT_FLOOD;
- } else if (ctx->nf_output_iface == NF_OUT_DROP) {
- ctx->nf_output_iface = prev_nf_output_iface;
- } else if (prev_nf_output_iface != NF_OUT_DROP &&
- ctx->nf_output_iface != NF_OUT_FLOOD) {
- ctx->nf_output_iface = NF_OUT_MULTI;
- }
+ return 0;
}
-
+\f
static void
-xlate_output_action(struct action_xlate_ctx *ctx,
- const struct ofp_action_output *oao)
+ofproto_rule_destroy__(struct rule *rule)
{
- xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
+ free(rule->actions);
+ rule->ofproto->ofproto_class->rule_dealloc(rule);
}
-/* If the final ODP action in 'ctx' is "pop priority", drop it, as an
- * optimization, because we're going to add another action that sets the
- * priority immediately after, or because there are no actions following the
- * pop. */
-static void
-remove_pop_action(struct action_xlate_ctx *ctx)
+/* 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/private.h titled "Life Cycle"). This function
+ * implements steps 6 and 7.
+ *
+ * This function should only be called from an ofproto implementation's
+ * ->destruct() function. It is not suitable elsewhere. */
+void
+ofproto_rule_destroy(struct rule *rule)
{
- size_t n = ctx->out->n_actions;
- if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) {
- ctx->out->n_actions--;
- }
+ assert(!rule->pending);
+ classifier_remove(&rule->ofproto->tables[rule->table_id], &rule->cr);
+ ofproto_rule_destroy__(rule);
}
-static void
-xlate_enqueue_action(struct action_xlate_ctx *ctx,
- const struct ofp_action_enqueue *oae)
+/* 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)
{
- uint16_t ofp_port, odp_port;
- uint32_t priority;
- int error;
-
- error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
- &priority);
- if (error) {
- /* Fall back to ordinary output action. */
- xlate_output_action__(ctx, ntohs(oae->port), 0);
- return;
- }
+ const union ofp_action *oa;
+ struct actions_iterator i;
- /* Figure out ODP output port. */
- ofp_port = ntohs(oae->port);
- if (ofp_port != OFPP_IN_PORT) {
- odp_port = ofp_port_to_odp_port(ofp_port);
- } else {
- odp_port = ctx->flow.in_port;
+ if (out_port == OFPP_NONE) {
+ return true;
}
-
- /* Add ODP actions. */
- remove_pop_action(ctx);
- odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
- = priority;
- add_output_action(ctx, odp_port);
- odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
-
- /* Update NetFlow output port. */
- if (ctx->nf_output_iface == NF_OUT_DROP) {
- ctx->nf_output_iface = odp_port;
- } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
- ctx->nf_output_iface = NF_OUT_MULTI;
+ for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
+ oa = actions_next(&i)) {
+ if (action_outputs_to_port(oa, htons(out_port))) {
+ return true;
+ }
}
+ return false;
}
-static void
-xlate_set_queue_action(struct action_xlate_ctx *ctx,
- const struct nx_action_set_queue *nasq)
+/* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
+ * statistics appropriately. 'packet' must have at least sizeof(struct
+ * ofp_packet_in) bytes of headroom.
+ *
+ * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
+ * with statistics for 'packet' either way.
+ *
+ * Takes ownership of 'packet'. */
+static int
+rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
{
- uint32_t priority;
- int error;
+ struct flow flow;
- error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
- &priority);
- if (error) {
- /* Couldn't translate queue to a priority, so ignore. A warning
- * has already been logged. */
- return;
- }
+ assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
- remove_pop_action(ctx);
- odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
- = priority;
+ flow_extract(packet, 0, in_port, &flow);
+ return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
}
-static void
-xlate_set_dl_tci(struct action_xlate_ctx *ctx)
+/* Returns true if 'rule' should be hidden from the controller.
+ *
+ * Rules with priority higher than UINT16_MAX are set up by ofproto itself
+ * (e.g. by in-band control) and are intentionally hidden from the
+ * controller. */
+static bool
+rule_is_hidden(const struct rule *rule)
{
- ovs_be16 tci = ctx->flow.vlan_tci;
- if (!(tci & htons(VLAN_CFI))) {
- odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
- } else {
- union odp_action *oa = odp_actions_add(ctx->out, ODPAT_SET_DL_TCI);
- oa->dl_tci.tci = tci & ~htons(VLAN_CFI);
- }
+ return rule->cr.priority > UINT16_MAX;
}
-
-static void
-xlate_reg_move_action(struct action_xlate_ctx *ctx,
- const struct nx_action_reg_move *narm)
+\f
+static int
+handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
{
- ovs_be16 old_tci = ctx->flow.vlan_tci;
-
- nxm_execute_reg_move(narm, &ctx->flow);
-
- if (ctx->flow.vlan_tci != old_tci) {
- xlate_set_dl_tci(ctx);
- }
+ ofconn_send_reply(ofconn, make_echo_reply(oh));
+ return 0;
}
-static void
-xlate_nicira_action(struct action_xlate_ctx *ctx,
- const struct nx_action_header *nah)
+static int
+handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
{
- const struct nx_action_resubmit *nar;
- const struct nx_action_set_tunnel *nast;
- const struct nx_action_set_queue *nasq;
- union odp_action *oa;
- int subtype = ntohs(nah->subtype);
-
- assert(nah->vendor == htonl(NX_VENDOR_ID));
- switch (subtype) {
- case NXAST_RESUBMIT:
- nar = (const struct nx_action_resubmit *) nah;
- xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
- break;
-
- case NXAST_SET_TUNNEL:
- nast = (const struct nx_action_set_tunnel *) nah;
- oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
- ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
- break;
-
- case NXAST_DROP_SPOOFED_ARP:
- if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
- odp_actions_add(ctx->out, ODPAT_DROP_SPOOFED_ARP);
- }
- break;
-
- case NXAST_SET_QUEUE:
- nasq = (const struct nx_action_set_queue *) nah;
- xlate_set_queue_action(ctx, nasq);
- break;
-
- case NXAST_POP_QUEUE:
- odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
- break;
-
- case NXAST_REG_MOVE:
- xlate_reg_move_action(ctx, (const struct nx_action_reg_move *) nah);
- break;
-
- case NXAST_REG_LOAD:
- nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
- &ctx->flow);
-
- case NXAST_NOTE:
- /* Nothing to do. */
- break;
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
+ struct ofp_switch_features *osf;
+ struct ofpbuf *buf;
+ struct ofport *port;
+ bool arp_match_ip;
+ uint32_t actions;
- /* If you add a new action here that modifies flow data, don't forget to
- * update the flow key in ctx->flow at the same time. */
+ ofproto->ofproto_class->get_features(ofproto, &arp_match_ip, &actions);
+ assert(actions & (1 << OFPAT_OUTPUT)); /* sanity check */
- default:
- VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
- 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);
+ if (arp_match_ip) {
+ osf->capabilities |= htonl(OFPC_ARP_MATCH_IP);
}
-}
+ osf->actions = htonl(actions);
-static void
-do_xlate_actions(const union ofp_action *in, size_t n_in,
- struct action_xlate_ctx *ctx)
-{
- struct actions_iterator iter;
- const union ofp_action *ia;
- const struct ofport *port;
-
- port = get_port(ctx->ofproto, ctx->flow.in_port);
- if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
- port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
- ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
- /* Drop this flow. */
- return;
+ HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
+ ofpbuf_put(buf, &port->opp, sizeof port->opp);
}
- for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
- uint16_t type = ntohs(ia->type);
- union odp_action *oa;
-
- switch (type) {
- case OFPAT_OUTPUT:
- xlate_output_action(ctx, &ia->output);
- break;
-
- case OFPAT_SET_VLAN_VID:
- ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
- ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
- xlate_set_dl_tci(ctx);
- break;
-
- case OFPAT_SET_VLAN_PCP:
- ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
- ctx->flow.vlan_tci |= htons(
- (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
- xlate_set_dl_tci(ctx);
- break;
-
- case OFPAT_STRIP_VLAN:
- ctx->flow.vlan_tci = htons(0);
- xlate_set_dl_tci(ctx);
- break;
-
- case OFPAT_SET_DL_SRC:
- oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
- memcpy(oa->dl_addr.dl_addr,
- ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
- memcpy(ctx->flow.dl_src,
- ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
- break;
-
- case OFPAT_SET_DL_DST:
- oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
- memcpy(oa->dl_addr.dl_addr,
- ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
- memcpy(ctx->flow.dl_dst,
- ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
- break;
+ ofconn_send_reply(ofconn, buf);
+ return 0;
+}
- case OFPAT_SET_NW_SRC:
- oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
- ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
- break;
+static int
+handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
+{
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
+ struct ofpbuf *buf;
+ struct ofp_switch_config *osc;
+ uint16_t flags;
+ bool drop_frags;
- case OFPAT_SET_NW_DST:
- oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
- ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
- break;
+ /* Figure out flags. */
+ drop_frags = ofproto->ofproto_class->get_drop_frags(ofproto);
+ flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
- case OFPAT_SET_NW_TOS:
- oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
- ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
- break;
+ /* Send reply. */
+ osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
+ osc->flags = htons(flags);
+ osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
+ ofconn_send_reply(ofconn, buf);
- case OFPAT_SET_TP_SRC:
- oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
- ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
- break;
+ return 0;
+}
- case OFPAT_SET_TP_DST:
- oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
- ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
- break;
+static int
+handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
+{
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
+ uint16_t flags = ntohs(osc->flags);
- case OFPAT_VENDOR:
- xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
+ if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
+ && ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
+ switch (flags & OFPC_FRAG_MASK) {
+ case OFPC_FRAG_NORMAL:
+ ofproto->ofproto_class->set_drop_frags(ofproto, false);
break;
-
- case OFPAT_ENQUEUE:
- xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
+ case OFPC_FRAG_DROP:
+ ofproto->ofproto_class->set_drop_frags(ofproto, true);
break;
-
default:
- VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
+ VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
+ osc->flags);
break;
}
}
-}
-static int
-xlate_actions(const union ofp_action *in, size_t n_in,
- const struct flow *flow, struct ofproto *ofproto,
- const struct ofpbuf *packet,
- struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
- uint16_t *nf_output_iface)
-{
- struct action_xlate_ctx ctx;
-
- COVERAGE_INC(ofproto_ofp2odp);
- odp_actions_init(out);
- ctx.flow = *flow;
- ctx.recurse = 0;
- ctx.ofproto = ofproto;
- ctx.packet = packet;
- ctx.out = out;
- ctx.tags = 0;
- ctx.may_set_up_flow = true;
- ctx.nf_output_iface = NF_OUT_DROP;
- do_xlate_actions(in, n_in, &ctx);
- remove_pop_action(&ctx);
-
- /* Check with in-band control to see if we're allowed to set up this
- * flow. */
- if (!in_band_rule_check(ofproto->in_band, flow, out)) {
- ctx.may_set_up_flow = false;
- }
+ ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
- if (tags) {
- *tags = ctx.tags;
- }
- if (may_set_up_flow) {
- *may_set_up_flow = ctx.may_set_up_flow;
- }
- if (nf_output_iface) {
- *nf_output_iface = ctx.nf_output_iface;
- }
- if (odp_actions_overflow(out)) {
- COVERAGE_INC(odp_overflow);
- odp_actions_init(out);
- return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
- }
return 0;
}
*
* The log message mentions 'msg_type'. */
static int
-reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
+reject_slave_controller(struct ofconn *ofconn, const char *msg_type)
{
- if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
+ if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
+ && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
msg_type);
}
static int
-handle_packet_out(struct ofconn *ofconn, struct ofp_header *oh)
+handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
{
- struct ofproto *p = ofconn->ofproto;
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
struct ofp_packet_out *opo;
struct ofpbuf payload, *buffer;
union ofp_action *ofp_actions;
- struct odp_actions odp_actions;
struct ofpbuf request;
struct flow flow;
size_t n_ofp_actions;
}
/* Get ofp_packet_out. */
- request.data = oh;
- request.size = ntohs(oh->length);
- opo = ofpbuf_try_pull(&request, offsetof(struct ofp_packet_out, actions));
- if (!opo) {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
- }
+ 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),
/* Get payload. */
if (opo->buffer_id != htonl(UINT32_MAX)) {
- error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
- &buffer, &in_port);
+ error = ofconn_pktbuf_retrieve(ofconn, ntohl(opo->buffer_id),
+ &buffer, &in_port);
if (error || !buffer) {
return error;
}
buffer = NULL;
}
- /* Extract flow, check actions. */
- flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
- &flow);
- error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
- if (error) {
- goto exit;
- }
-
- /* Send. */
- error = xlate_actions(ofp_actions, n_ofp_actions, &flow, p, &payload,
- &odp_actions, NULL, NULL, NULL);
- if (!error) {
- dpif_execute(p->dpif, odp_actions.actions, odp_actions.n_actions,
- &payload);
- }
-
-exit:
+ /* Send out packet. */
+ flow_extract(&payload, 0, ntohs(opo->in_port), &flow);
+ error = p->ofproto_class->packet_out(p, &payload, &flow,
+ ofp_actions, n_ofp_actions);
ofpbuf_delete(buffer);
- return 0;
+
+ return error;
}
static void
-update_port_config(struct ofproto *p, struct ofport *port,
- uint32_t config, uint32_t mask)
+update_port_config(struct ofport *port, ovs_be32 config, ovs_be32 mask)
{
+ ovs_be32 old_config = port->opp.config;
+
mask &= config ^ port->opp.config;
- if (mask & OFPPC_PORT_DOWN) {
- if (config & OFPPC_PORT_DOWN) {
+ if (mask & htonl(OFPPC_PORT_DOWN)) {
+ if (config & htonl(OFPPC_PORT_DOWN)) {
netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
} else {
netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
}
}
-#define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
- OFPPC_NO_FWD | OFPPC_NO_FLOOD)
- if (mask & REVALIDATE_BITS) {
- COVERAGE_INC(ofproto_costly_flags);
- port->opp.config ^= mask & REVALIDATE_BITS;
- p->need_revalidate = true;
- }
-#undef REVALIDATE_BITS
- if (mask & OFPPC_NO_PACKET_IN) {
- port->opp.config ^= OFPPC_NO_PACKET_IN;
+
+ 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->ofproto->ofproto_class->port_reconfigured(port, old_config);
}
}
static int
-handle_port_mod(struct ofconn *ofconn, struct ofp_header *oh)
+handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
{
- struct ofproto *p = ofconn->ofproto;
- const struct ofp_port_mod *opm;
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
+ const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
struct ofport *port;
int error;
if (error) {
return error;
}
- error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
- if (error) {
- return error;
- }
- opm = (struct ofp_port_mod *) oh;
- port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
+ port = ofproto_get_port(p, ntohs(opm->port_no));
if (!port) {
return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
} else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
} else {
- update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
+ update_port_config(port, opm->config, opm->mask);
if (opm->advertise) {
netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
}
return 0;
}
-static struct ofpbuf *
-make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
-{
- struct ofp_stats_reply *osr;
- struct ofpbuf *msg;
-
- msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
- osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
- osr->type = type;
- osr->flags = htons(0);
- return msg;
-}
-
-static struct ofpbuf *
-start_ofp_stats_reply(const struct ofp_stats_request *request, size_t body_len)
-{
- return make_ofp_stats_reply(request->header.xid, request->type, body_len);
-}
-
-static void *
-append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
- struct ofpbuf **msgp)
-{
- struct ofpbuf *msg = *msgp;
- assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
- if (nbytes + msg->size > UINT16_MAX) {
- struct ofp_stats_reply *reply = msg->data;
- reply->flags = htons(OFPSF_REPLY_MORE);
- *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
- queue_tx(msg, ofconn, ofconn->reply_counter);
- }
- return ofpbuf_put_uninit(*msgp, nbytes);
-}
-
-static struct ofpbuf *
-make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
-{
- struct nicira_stats_msg *nsm;
- struct ofpbuf *msg;
-
- msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
- nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
- nsm->type = htons(OFPST_VENDOR);
- nsm->flags = htons(0);
- nsm->vendor = htonl(NX_VENDOR_ID);
- nsm->subtype = htonl(subtype);
- return msg;
-}
-
-static struct ofpbuf *
-start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
-{
- return make_nxstats_reply(request->header.xid, request->subtype, body_len);
-}
-
-static void
-append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
- struct ofpbuf **msgp)
-{
- struct ofpbuf *msg = *msgp;
- assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
- if (nbytes + msg->size > UINT16_MAX) {
- struct nicira_stats_msg *reply = msg->data;
- reply->flags = htons(OFPSF_REPLY_MORE);
- *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
- queue_tx(msg, ofconn, ofconn->reply_counter);
- }
- ofpbuf_prealloc_tailroom(*msgp, nbytes);
-}
-
static int
handle_desc_stats_request(struct ofconn *ofconn,
- struct ofp_stats_request *request)
+ const struct ofp_stats_msg *request)
{
- struct ofproto *p = ofconn->ofproto;
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
struct ofp_desc_stats *ods;
struct ofpbuf *msg;
- msg = start_ofp_stats_reply(request, sizeof *ods);
- ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
- memset(ods, 0, sizeof *ods);
+ 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);
- queue_tx(msg, ofconn, ofconn->reply_counter);
+ ofconn_send_reply(ofconn, msg);
return 0;
}
static int
handle_table_stats_request(struct ofconn *ofconn,
- struct ofp_stats_request *request)
+ const struct ofp_stats_msg *request)
{
- struct ofproto *p = ofconn->ofproto;
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
struct ofp_table_stats *ots;
struct ofpbuf *msg;
+ size_t i;
- msg = start_ofp_stats_reply(request, sizeof *ots * 2);
+ ofputil_make_stats_reply(sizeof(struct ofp_stats_msg), request, &msg);
- /* Classifier table. */
- ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
- memset(ots, 0, sizeof *ots);
- strcpy(ots->name, "classifier");
- ots->wildcards = (ofconn->flow_format == NXFF_OPENFLOW10
- ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
- ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
- ots->active_count = htonl(classifier_count(&p->cls));
- ots->lookup_count = htonll(0); /* XXX */
- ots->matched_count = htonll(0); /* XXX */
+ ots = ofpbuf_put_zeros(msg, sizeof *ots * p->n_tables);
+ for (i = 0; i < p->n_tables; i++) {
+ ots[i].table_id = i;
+ sprintf(ots[i].name, "table%zu", i);
+ ots[i].wildcards = htonl(OFPFW_ALL);
+ ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
+ ots[i].active_count = htonl(classifier_count(&p->tables[i]));
+ }
+
+ p->ofproto_class->get_tables(p, ots);
- queue_tx(msg, ofconn, ofconn->reply_counter);
+ ofconn_send_reply(ofconn, msg);
return 0;
}
static void
-append_port_stat(struct ofport *port, struct ofconn *ofconn,
- struct ofpbuf **msgp)
+append_port_stat(struct ofport *port, struct list *replies)
{
struct netdev_stats stats;
struct ofp_port_stats *ops;
* netdev_get_stats() will log errors. */
netdev_get_stats(port->netdev, &stats);
- ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
- ops->port_no = htons(port->opp.port_no);
+ ops = ofputil_append_stats_reply(sizeof *ops, replies);
+ ops->port_no = port->opp.port_no;
memset(ops->pad, 0, sizeof ops->pad);
- ops->rx_packets = htonll(stats.rx_packets);
- ops->tx_packets = htonll(stats.tx_packets);
- ops->rx_bytes = htonll(stats.rx_bytes);
- ops->tx_bytes = htonll(stats.tx_bytes);
- ops->rx_dropped = htonll(stats.rx_dropped);
- ops->tx_dropped = htonll(stats.tx_dropped);
- ops->rx_errors = htonll(stats.rx_errors);
- ops->tx_errors = htonll(stats.tx_errors);
- ops->rx_frame_err = htonll(stats.rx_frame_errors);
- ops->rx_over_err = htonll(stats.rx_over_errors);
- ops->rx_crc_err = htonll(stats.rx_crc_errors);
- ops->collisions = htonll(stats.collisions);
+ put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
+ put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
+ put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
+ put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
+ put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
+ put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
+ put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
+ put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
+ put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
+ put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
+ put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
+ put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
}
static int
-handle_port_stats_request(struct ofconn *ofconn, struct ofp_stats_request *osr,
- size_t arg_size)
+handle_port_stats_request(struct ofconn *ofconn,
+ const struct ofp_port_stats_request *psr)
{
- struct ofproto *p = ofconn->ofproto;
- struct ofp_port_stats_request *psr;
- struct ofp_port_stats *ops;
- struct ofpbuf *msg;
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
struct ofport *port;
+ struct list replies;
- if (arg_size != sizeof *psr) {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
- }
- psr = (struct ofp_port_stats_request *) osr->body;
-
- msg = start_ofp_stats_reply(osr, sizeof *ops * 16);
+ ofputil_start_stats_reply(&psr->osm, &replies);
if (psr->port_no != htons(OFPP_NONE)) {
- port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
+ port = ofproto_get_port(p, ntohs(psr->port_no));
if (port) {
- append_port_stat(port, ofconn, &msg);
+ append_port_stat(port, &replies);
}
} else {
HMAP_FOR_EACH (port, hmap_node, &p->ports) {
- append_port_stat(port, ofconn, &msg);
+ append_port_stat(port, &replies);
}
}
- queue_tx(msg, ofconn, ofconn->reply_counter);
+ ofconn_send_replies(ofconn, &replies);
return 0;
}
-/* Obtains statistic counters for 'rule' within 'p' and stores them into
- * '*packet_countp' and '*byte_countp'. The returned statistics include
- * statistics for all of 'rule''s facets. */
-static void
-query_stats(struct ofproto *p, struct rule *rule,
- uint64_t *packet_countp, uint64_t *byte_countp)
-{
- uint64_t packet_count, byte_count;
- struct facet *facet;
- struct odp_flow *odp_flows;
- size_t n_odp_flows;
-
- /* Start from historical data for 'rule' itself that are no longer tracked
- * by the datapath. This counts, for example, facets that have expired. */
- packet_count = rule->packet_count;
- byte_count = rule->byte_count;
-
- /* Prepare to ask the datapath for statistics on all of the rule's facets.
- *
- * Also, add any statistics that are not tracked by the datapath for each
- * facet. This includes, for example, statistics for packets that were
- * executed "by hand" by ofproto via dpif_execute() but must be accounted
- * to a rule. */
- odp_flows = xzalloc(list_size(&rule->facets) * sizeof *odp_flows);
- n_odp_flows = 0;
- LIST_FOR_EACH (facet, list_node, &rule->facets) {
- struct odp_flow *odp_flow = &odp_flows[n_odp_flows++];
- odp_flow_key_from_flow(&odp_flow->key, &facet->flow);
- packet_count += facet->packet_count;
- byte_count += facet->byte_count;
- }
-
- /* Fetch up-to-date statistics from the datapath and add them in. */
- if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
- size_t i;
-
- for (i = 0; i < n_odp_flows; i++) {
- struct odp_flow *odp_flow = &odp_flows[i];
- packet_count += odp_flow->stats.n_packets;
- byte_count += odp_flow->stats.n_bytes;
- }
- }
- free(odp_flows);
-
- /* Return the stats to the caller. */
- *packet_countp = packet_count;
- *byte_countp = byte_count;
-}
-
static void
-calc_flow_duration(long long int start, ovs_be32 *sec, ovs_be32 *nsec)
+calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec)
{
long long int msecs = time_msec() - start;
- *sec = htonl(msecs / 1000);
- *nsec = htonl((msecs % 1000) * (1000 * 1000));
+ *sec = msecs / 1000;
+ *nsec = (msecs % 1000) * (1000 * 1000);
}
-static void
-put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
- ovs_be16 out_port, struct ofpbuf **replyp)
+static struct classifier *
+first_matching_table(struct ofproto *ofproto, uint8_t table_id)
{
- struct ofp_flow_stats *ofs;
- uint64_t packet_count, byte_count;
- size_t act_len, len;
-
- if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
- return;
- }
-
- act_len = sizeof *rule->actions * rule->n_actions;
- len = offsetof(struct ofp_flow_stats, actions) + act_len;
-
- query_stats(ofconn->ofproto, rule, &packet_count, &byte_count);
-
- ofs = append_ofp_stats_reply(len, ofconn, replyp);
- ofs->length = htons(len);
- ofs->table_id = 0;
- ofs->pad = 0;
- ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofs->match);
- calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
- ofs->cookie = rule->flow_cookie;
- ofs->priority = htons(rule->cr.priority);
- ofs->idle_timeout = htons(rule->idle_timeout);
- ofs->hard_timeout = htons(rule->hard_timeout);
- memset(ofs->pad2, 0, sizeof ofs->pad2);
- ofs->packet_count = htonll(packet_count);
- ofs->byte_count = htonll(byte_count);
- if (rule->n_actions > 0) {
- memcpy(ofs->actions, rule->actions, act_len);
+ if (table_id == 0xff) {
+ return &ofproto->tables[0];
+ } else if (table_id < ofproto->n_tables) {
+ return &ofproto->tables[table_id];
+ } else {
+ /* It would probably be better to reply with an error but there doesn't
+ * seem to be any appropriate value, so that might just be
+ * confusing. */
+ VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8,
+ table_id);
+ return NULL;
}
}
-static bool
-is_valid_table(uint8_t table_id)
+static struct classifier *
+next_matching_table(struct ofproto *ofproto,
+ struct classifier *cls, uint8_t table_id)
{
- return table_id == 0 || table_id == 0xff;
+ return (table_id == 0xff && cls != &ofproto->tables[ofproto->n_tables - 1]
+ ? cls + 1
+ : NULL);
}
+/* Assigns CLS to each classifier table, in turn, that matches TABLE_ID in
+ * OFPROTO:
+ *
+ * - If TABLE_ID is 0xff, this iterates over every classifier table in
+ * OFPROTO.
+ *
+ * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
+ * only once, for that table.
+ *
+ * - Otherwise, TABLE_ID isn't valid for OFPROTO, so ofproto logs a warning
+ * and does not enter the loop at all.
+ *
+ * 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))
+
+/* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
+ * 'table_id' is 0xff) that match 'match' in the "loose" way required for
+ * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
+ * 'rules'.
+ *
+ * If 'out_port' is anything other than OFPP_NONE, then only rules that output
+ * to 'out_port' are included.
+ *
+ * Hidden rules are always omitted.
+ *
+ * Returns 0 on success, otherwise an OpenFlow error code. */
static int
-handle_flow_stats_request(struct ofconn *ofconn,
- const struct ofp_stats_request *osr, size_t arg_size)
+collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
+ const struct cls_rule *match, uint16_t out_port,
+ struct list *rules)
{
- struct ofp_flow_stats_request *fsr;
- struct ofpbuf *reply;
-
- if (arg_size != sizeof *fsr) {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
- }
- fsr = (struct ofp_flow_stats_request *) osr->body;
+ struct classifier *cls;
- COVERAGE_INC(ofproto_flows_req);
- reply = start_ofp_stats_reply(osr, 1024);
- if (is_valid_table(fsr->table_id)) {
+ list_init(rules);
+ FOR_EACH_MATCHING_TABLE (cls, table_id, ofproto) {
struct cls_cursor cursor;
- struct cls_rule target;
struct rule *rule;
- ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
- &target);
- cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
+ cls_cursor_init(&cursor, cls, match);
CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
- put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
+ if (rule->pending) {
+ return OFPROTO_POSTPONE;
+ }
+ if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
+ list_push_back(rules, &rule->ofproto_node);
+ }
}
}
- queue_tx(reply, ofconn, ofconn->reply_counter);
-
return 0;
}
-static void
-put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
- ovs_be16 out_port, struct ofpbuf **replyp)
+/* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
+ * 'table_id' is 0xff) that match 'match' in the "strict" way required for
+ * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
+ * on list 'rules'.
+ *
+ * If 'out_port' is anything other than OFPP_NONE, then only rules that output
+ * to 'out_port' are included.
+ *
+ * Hidden rules are always omitted.
+ *
+ * Returns 0 on success, otherwise an OpenFlow error code. */
+static int
+collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
+ const struct cls_rule *match, uint16_t out_port,
+ struct list *rules)
{
- struct nx_flow_stats *nfs;
- uint64_t packet_count, byte_count;
- size_t act_len, start_len;
- struct ofpbuf *reply;
+ struct classifier *cls;
- if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
- return;
- }
+ list_init(rules);
+ FOR_EACH_MATCHING_TABLE (cls, table_id, ofproto) {
+ struct rule *rule;
- query_stats(ofconn->ofproto, rule, &packet_count, &byte_count);
-
- act_len = sizeof *rule->actions * rule->n_actions;
-
- start_len = (*replyp)->size;
- append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
- reply = *replyp;
-
- nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
- nfs->table_id = 0;
- nfs->pad = 0;
- calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
- nfs->cookie = rule->flow_cookie;
- nfs->priority = htons(rule->cr.priority);
- nfs->idle_timeout = htons(rule->idle_timeout);
- nfs->hard_timeout = htons(rule->hard_timeout);
- nfs->match_len = htons(nx_put_match(reply, &rule->cr));
- memset(nfs->pad2, 0, sizeof nfs->pad2);
- nfs->packet_count = htonll(packet_count);
- nfs->byte_count = htonll(byte_count);
- if (rule->n_actions > 0) {
- ofpbuf_put(reply, rule->actions, act_len);
+ rule = rule_from_cls_rule(classifier_find_rule_exactly(cls, match));
+ if (rule) {
+ if (rule->pending) {
+ return OFPROTO_POSTPONE;
+ }
+ if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
+ list_push_back(rules, &rule->ofproto_node);
+ }
+ }
}
- nfs->length = htons(reply->size - start_len);
+ return 0;
}
static int
-handle_nxst_flow(struct ofconn *ofconn, struct ofpbuf *b)
+handle_flow_stats_request(struct ofconn *ofconn,
+ const struct ofp_stats_msg *osm)
{
- struct nx_flow_stats_request *nfsr;
- struct cls_rule target;
- struct ofpbuf *reply;
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
+ struct flow_stats_request fsr;
+ struct list replies;
+ struct list rules;
+ struct rule *rule;
int error;
- /* Dissect the message. */
- nfsr = ofpbuf_try_pull(b, sizeof *nfsr);
- if (!nfsr) {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
+ error = ofputil_decode_flow_stats_request(&fsr, &osm->header);
+ if (error) {
+ return error;
}
- error = nx_pull_match(b, ntohs(nfsr->match_len), 0, &target);
+
+ error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
+ fsr.out_port, &rules);
if (error) {
return error;
}
- COVERAGE_INC(ofproto_flows_req);
- reply = start_nxstats_reply(&nfsr->nsm, 1024);
- if (is_valid_table(nfsr->table_id)) {
- struct cls_cursor cursor;
- struct rule *rule;
+ ofputil_start_stats_reply(osm, &replies);
+ LIST_FOR_EACH (rule, ofproto_node, &rules) {
+ struct ofputil_flow_stats fs;
- cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
- CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
- put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
- }
+ fs.rule = rule->cr;
+ 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;
+ 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);
}
- queue_tx(reply, ofconn, ofconn->reply_counter);
+ ofconn_send_replies(ofconn, &replies);
return 0;
}
static void
-flow_stats_ds(struct ofproto *ofproto, struct rule *rule, struct ds *results)
+flow_stats_ds(struct rule *rule, struct ds *results)
{
- struct ofp_match match;
uint64_t packet_count, byte_count;
size_t act_len = sizeof *rule->actions * rule->n_actions;
- query_stats(ofproto, rule, &packet_count, &byte_count);
- ofputil_cls_rule_to_match(&rule->cr, NXFF_OPENFLOW10, &match);
+ rule->ofproto->ofproto_class->rule_get_stats(rule,
+ &packet_count, &byte_count);
+ if (rule->table_id != 0) {
+ ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
+ }
ds_put_format(results, "duration=%llds, ",
(time_msec() - rule->created) / 1000);
ds_put_format(results, "priority=%u, ", rule->cr.priority);
ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
- ofp_print_match(results, &match, true);
+ cls_rule_format(&rule->cr, results);
+ ds_put_char(results, ',');
if (act_len > 0) {
ofp_print_actions(results, &rule->actions->header, act_len);
} else {
}
/* Adds a pretty-printed description of all flows to 'results', including
- * those marked hidden by secchan (e.g., by in-band control). */
+ * hidden flows (e.g., set up by in-band control). */
void
ofproto_get_all_flows(struct ofproto *p, struct ds *results)
{
- struct cls_cursor cursor;
- struct rule *rule;
-
- cls_cursor_init(&cursor, &p->cls, NULL);
- CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
- flow_stats_ds(p, rule, results);
- }
-}
-
-static void
-query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
- ovs_be16 out_port, uint8_t table_id,
- struct ofp_aggregate_stats_reply *oasr)
-{
- uint64_t total_packets = 0;
- uint64_t total_bytes = 0;
- int n_flows = 0;
-
- COVERAGE_INC(ofproto_agg_request);
+ struct classifier *cls;
- if (is_valid_table(table_id)) {
+ for (cls = &p->tables[0]; cls < &p->tables[p->n_tables]; cls++) {
struct cls_cursor cursor;
struct rule *rule;
- cls_cursor_init(&cursor, &ofproto->cls, target);
+ cls_cursor_init(&cursor, cls, NULL);
CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
- if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
- uint64_t packet_count;
- uint64_t byte_count;
-
- query_stats(ofproto, rule, &packet_count, &byte_count);
-
- total_packets += packet_count;
- total_bytes += byte_count;
- n_flows++;
- }
+ flow_stats_ds(rule, results);
}
}
-
- oasr->flow_count = htonl(n_flows);
- oasr->packet_count = htonll(total_packets);
- oasr->byte_count = htonll(total_bytes);
- memset(oasr->pad, 0, sizeof oasr->pad);
}
-static int
-handle_aggregate_stats_request(struct ofconn *ofconn,
- const struct ofp_stats_request *osr,
- size_t arg_size)
+/* Obtains the NetFlow engine type and engine ID for 'ofproto' into
+ * '*engine_type' and '*engine_id', respectively. */
+void
+ofproto_get_netflow_ids(const struct ofproto *ofproto,
+ uint8_t *engine_type, uint8_t *engine_id)
{
- struct ofp_aggregate_stats_request *request;
- struct ofp_aggregate_stats_reply *reply;
- struct cls_rule target;
- struct ofpbuf *msg;
-
- if (arg_size != sizeof *request) {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
- }
- request = (struct ofp_aggregate_stats_request *) osr->body;
-
- ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
- &target);
+ ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
+}
- msg = start_ofp_stats_reply(osr, sizeof *reply);
- reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
- query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
- request->table_id, reply);
- queue_tx(msg, ofconn, ofconn->reply_counter);
- return 0;
+/* 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);
}
static int
-handle_nxst_aggregate(struct ofconn *ofconn, struct ofpbuf *b)
+handle_aggregate_stats_request(struct ofconn *ofconn,
+ const struct ofp_stats_msg *osm)
{
- struct nx_aggregate_stats_request *request;
- struct ofp_aggregate_stats_reply *reply;
- struct cls_rule target;
- struct ofpbuf *buf;
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
+ struct flow_stats_request request;
+ struct ofputil_aggregate_stats stats;
+ struct ofpbuf *reply;
+ struct list rules;
+ struct rule *rule;
int error;
- /* Dissect the message. */
- request = ofpbuf_try_pull(b, sizeof *request);
- if (!request) {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
+ error = ofputil_decode_flow_stats_request(&request, &osm->header);
+ if (error) {
+ return error;
}
- error = nx_pull_match(b, ntohs(request->match_len), 0, &target);
+
+ error = collect_rules_loose(ofproto, request.table_id, &request.match,
+ request.out_port, &rules);
if (error) {
return error;
}
- /* Reply. */
- COVERAGE_INC(ofproto_flows_req);
- buf = start_nxstats_reply(&request->nsm, sizeof *reply);
- reply = ofpbuf_put_uninit(buf, sizeof *reply);
- query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
- request->table_id, reply);
- queue_tx(buf, ofconn, ofconn->reply_counter);
+ memset(&stats, 0, sizeof stats);
+ LIST_FOR_EACH (rule, ofproto_node, &rules) {
+ uint64_t packet_count;
+ uint64_t byte_count;
+
+ ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
+ &byte_count);
+
+ stats.packet_count += packet_count;
+ stats.byte_count += byte_count;
+ stats.flow_count++;
+ }
+
+ reply = ofputil_encode_aggregate_stats_reply(&stats, osm);
+ ofconn_send_reply(ofconn, reply);
return 0;
}
struct queue_stats_cbdata {
- struct ofconn *ofconn;
struct ofport *ofport;
- struct ofpbuf *msg;
+ struct list replies;
};
static void
{
struct ofp_queue_stats *reply;
- reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
- reply->port_no = htons(cbdata->ofport->opp.port_no);
+ 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);
- reply->tx_bytes = htonll(stats->tx_bytes);
- reply->tx_packets = htonll(stats->tx_packets);
- reply->tx_errors = htonll(stats->tx_errors);
+ put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
+ put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
+ put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
}
static void
static int
handle_queue_stats_request(struct ofconn *ofconn,
- const struct ofp_stats_request *osr,
- size_t arg_size)
+ const struct ofp_queue_stats_request *qsr)
{
- struct ofproto *ofproto = ofconn->ofproto;
- struct ofp_queue_stats_request *qsr;
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
struct queue_stats_cbdata cbdata;
struct ofport *port;
unsigned int port_no;
uint32_t queue_id;
- if (arg_size != sizeof *qsr) {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
- }
- qsr = (struct ofp_queue_stats_request *) osr->body;
-
COVERAGE_INC(ofproto_queue_req);
- cbdata.ofconn = ofconn;
- cbdata.msg = start_ofp_stats_reply(osr, 128);
+ ofputil_start_stats_reply(&qsr->osm, &cbdata.replies);
port_no = ntohs(qsr->port_no);
queue_id = ntohl(qsr->queue_id);
HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
handle_queue_stats_for_port(port, queue_id, &cbdata);
}
- } else if (port_no < ofproto->max_ports) {
- port = get_port(ofproto, ofp_port_to_odp_port(port_no));
+ } else if (port_no < OFPP_MAX) {
+ port = ofproto_get_port(ofproto, port_no);
if (port) {
handle_queue_stats_for_port(port, queue_id, &cbdata);
}
} else {
- ofpbuf_delete(cbdata.msg);
+ ofpbuf_list_delete(&cbdata.replies);
return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
}
- queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
+ ofconn_send_replies(ofconn, &cbdata.replies);
return 0;
}
-static int
-handle_vendor_stats_request(struct ofconn *ofconn,
- struct ofp_stats_request *osr, size_t arg_size)
-{
- struct nicira_stats_msg *nsm;
- struct ofpbuf b;
- ovs_be32 vendor;
-
- if (arg_size < 4) {
- VLOG_WARN_RL(&rl, "truncated vendor stats request body");
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
- }
-
- memcpy(&vendor, osr->body, sizeof vendor);
- if (vendor != htonl(NX_VENDOR_ID)) {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
- }
-
- if (ntohs(osr->header.length) < sizeof(struct nicira_stats_msg)) {
- VLOG_WARN_RL(&rl, "truncated Nicira stats request");
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
- }
-
- nsm = (struct nicira_stats_msg *) osr;
- b.data = nsm;
- b.size = ntohs(nsm->header.length);
- switch (ntohl(nsm->subtype)) {
- case NXST_FLOW:
- return handle_nxst_flow(ofconn, &b);
-
- case NXST_AGGREGATE:
- return handle_nxst_aggregate(ofconn, &b);
-
- default:
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
- }
-}
-
-static int
-handle_stats_request(struct ofconn *ofconn, struct ofp_header *oh)
-{
- struct ofp_stats_request *osr;
- size_t arg_size;
- int error;
-
- error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
- 1, &arg_size);
- if (error) {
- return error;
- }
- osr = (struct ofp_stats_request *) oh;
-
- switch (ntohs(osr->type)) {
- case OFPST_DESC:
- return handle_desc_stats_request(ofconn, osr);
-
- case OFPST_FLOW:
- return handle_flow_stats_request(ofconn, osr, arg_size);
-
- case OFPST_AGGREGATE:
- return handle_aggregate_stats_request(ofconn, osr, arg_size);
-
- case OFPST_TABLE:
- return handle_table_stats_request(ofconn, osr);
-
- case OFPST_PORT:
- return handle_port_stats_request(ofconn, osr, arg_size);
-
- case OFPST_QUEUE:
- return handle_queue_stats_request(ofconn, osr, arg_size);
-
- case OFPST_VENDOR:
- return handle_vendor_stats_request(ofconn, osr, arg_size);
-
- default:
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
- }
-}
-
-static long long int
-msec_from_nsec(uint64_t sec, uint32_t nsec)
-{
- return !sec ? 0 : sec * 1000 + nsec / 1000000;
-}
-
-static void
-facet_update_time(struct ofproto *ofproto, struct facet *facet,
- const struct odp_flow_stats *stats)
-{
- long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
- if (used > facet->used) {
- facet->used = used;
- if (used > facet->rule->used) {
- facet->rule->used = used;
+static bool
+is_flow_deletion_pending(const struct ofproto *ofproto,
+ const struct cls_rule *cls_rule,
+ uint8_t table_id)
+{
+ if (!hmap_is_empty(&ofproto->deletions)) {
+ struct ofoperation *op;
+
+ HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
+ cls_rule_hash(cls_rule, table_id),
+ &ofproto->deletions) {
+ if (cls_rule_equal(cls_rule, &op->rule->cr)) {
+ return true;
+ }
}
- netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
}
-}
-/* Folds the statistics from 'stats' into the counters in 'facet'.
- *
- * Because of the meaning of a facet's counters, it only makes sense to do this
- * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
- * packet that was sent by hand or if it represents statistics that have been
- * cleared out of the datapath. */
-static void
-facet_update_stats(struct ofproto *ofproto, struct facet *facet,
- const struct odp_flow_stats *stats)
-{
- if (stats->n_packets) {
- facet_update_time(ofproto, facet, stats);
- facet->packet_count += stats->n_packets;
- facet->byte_count += stats->n_bytes;
- netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
- }
+ return false;
}
-struct flow_mod {
- struct cls_rule cr;
- ovs_be64 cookie;
- uint16_t command;
- uint16_t idle_timeout;
- uint16_t hard_timeout;
- uint32_t buffer_id;
- uint16_t out_port;
- uint16_t flags;
- union ofp_action *actions;
- size_t n_actions;
-};
-
/* 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 ofconn->ofproto's flow table. Returns 0 on success or an
+ * ofp_actions, to the ofproto's flow table. Returns 0 on success or an
* OpenFlow error code as encoded by ofp_mkerr() on failure.
*
* 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
* if any. */
static int
-add_flow(struct ofconn *ofconn, struct flow_mod *fm)
+add_flow(struct ofproto *ofproto, struct ofconn *ofconn, struct flow_mod *fm,
+ const struct ofp_header *request)
{
- struct ofproto *p = ofconn->ofproto;
- struct ofpbuf *packet;
+ struct classifier *table;
+ struct ofopgroup *group;
+ struct rule *victim;
struct rule *rule;
- uint16_t in_port;
int error;
- if (fm->flags & OFPFF_CHECK_OVERLAP
- && classifier_rule_overlaps(&p->cls, &fm->cr)) {
- return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
+ /* Check for overlap, if requested. */
+ if (fm->flags & OFPFF_CHECK_OVERLAP) {
+ struct classifier *cls;
+
+ FOR_EACH_MATCHING_TABLE (cls, fm->table_id, ofproto) {
+ if (classifier_rule_overlaps(cls, &fm->cr)) {
+ return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
+ }
+ }
}
- error = 0;
- if (fm->buffer_id != UINT32_MAX) {
- error = pktbuf_retrieve(ofconn->pktbuf, fm->buffer_id,
- &packet, &in_port);
+ /* Pick table. */
+ if (fm->table_id == 0xff) {
+ uint8_t table_id;
+ if (ofproto->n_tables > 1) {
+ error = ofproto->ofproto_class->rule_choose_table(ofproto, &fm->cr,
+ &table_id);
+ if (error) {
+ return error;
+ }
+ assert(table_id < ofproto->n_tables);
+ table = &ofproto->tables[table_id];
+ } else {
+ table = &ofproto->tables[0];
+ }
+ } else if (fm->table_id < ofproto->n_tables) {
+ table = &ofproto->tables[fm->table_id];
} else {
- packet = NULL;
- in_port = UINT16_MAX;
+ return ofp_mkerr_nicira(OFPET_FLOW_MOD_FAILED, NXFMFC_BAD_TABLE_ID);
}
- rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
- fm->idle_timeout, fm->hard_timeout, fm->cookie,
- fm->flags & OFPFF_SEND_FLOW_REM);
- rule_insert(p, rule);
- if (packet) {
- rule_execute(p, rule, in_port, packet);
+ /* Serialize against pending deletion. */
+ if (is_flow_deletion_pending(ofproto, &fm->cr, table - ofproto->tables)) {
+ return OFPROTO_POSTPONE;
}
- return error;
-}
-static struct rule *
-find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
-{
- return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
-}
+ /* Allocate new rule. */
+ rule = ofproto->ofproto_class->rule_alloc();
+ if (!rule) {
+ VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
+ ofproto->name, strerror(error));
+ return ENOMEM;
+ }
+ rule->ofproto = ofproto;
+ rule->cr = fm->cr;
+ rule->pending = NULL;
+ rule->flow_cookie = fm->cookie;
+ rule->created = time_msec();
+ rule->idle_timeout = fm->idle_timeout;
+ rule->hard_timeout = fm->hard_timeout;
+ rule->table_id = table - ofproto->tables;
+ rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
+ rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions);
+ rule->n_actions = fm->n_actions;
-static int
-send_buffered_packet(struct ofconn *ofconn,
- struct rule *rule, uint32_t buffer_id)
-{
- struct ofpbuf *packet;
- uint16_t in_port;
- int error;
+ /* Insert new rule. */
+ victim = rule_from_cls_rule(classifier_replace(table, &rule->cr));
+ if (victim && victim->pending) {
+ error = OFPROTO_POSTPONE;
+ } else {
+ group = (ofconn
+ ? ofopgroup_create_for_ofconn(ofconn, request, fm->buffer_id)
+ : ofopgroup_create(ofproto));
+ ofoperation_create(group, rule, OFOPERATION_ADD);
+ rule->pending->victim = victim;
- if (buffer_id == UINT32_MAX) {
- return 0;
+ error = ofproto->ofproto_class->rule_construct(rule);
+ if (error) {
+ ofoperation_destroy(rule->pending);
+ }
+ ofopgroup_submit(group);
}
- error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port);
+ /* Back out if an error occurred. */
if (error) {
- return error;
+ if (victim) {
+ classifier_replace(table, &victim->cr);
+ } else {
+ classifier_remove(table, &rule->cr);
+ }
+ ofproto_rule_destroy__(rule);
}
-
- rule_execute(ofconn->ofproto, rule, in_port, packet);
-
- return 0;
+ return error;
}
\f
/* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
-struct modify_flows_cbdata {
- struct ofproto *ofproto;
- const struct flow_mod *fm;
- struct rule *match;
-};
+/* Modifies the rules listed in 'rules', changing their actions to match those
+ * in 'fm'.
+ *
+ * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
+ * if any.
+ *
+ * Returns 0 on success, otherwise an OpenFlow error code. */
+static int
+modify_flows__(struct ofconn *ofconn, const struct flow_mod *fm,
+ const struct ofp_header *request, struct list *rules)
+{
+ struct ofopgroup *group;
+ struct rule *rule;
-static int modify_flow(struct ofproto *, const struct flow_mod *,
- struct rule *);
+ group = ofopgroup_create_for_ofconn(ofconn, request, fm->buffer_id);
+ 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);
+ }
+ rule->flow_cookie = fm->cookie;
+ }
+ ofopgroup_submit(group);
+
+ return 0;
+}
/* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
* encoded by ofp_mkerr() on failure.
*
- * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
+ * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
* if any. */
static int
-modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
+modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm,
+ const struct ofp_header *request)
{
- struct ofproto *p = ofconn->ofproto;
- struct rule *match = NULL;
- struct cls_cursor cursor;
- struct rule *rule;
-
- cls_cursor_init(&cursor, &p->cls, &fm->cr);
- CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
- if (!rule_is_hidden(rule)) {
- match = rule;
- modify_flow(p, fm, rule);
- }
- }
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
+ struct list rules;
+ int error;
- if (match) {
- /* This credits the packet to whichever flow happened to match last.
- * That's weird. Maybe we should do a lookup for the flow that
- * actually matches the packet? Who knows. */
- send_buffered_packet(ofconn, match, fm->buffer_id);
- return 0;
- } else {
- return add_flow(ofconn, fm);
- }
+ error = collect_rules_loose(p, fm->table_id, &fm->cr, OFPP_NONE, &rules);
+ return (error ? error
+ : list_is_empty(&rules) ? add_flow(p, ofconn, fm, request)
+ : modify_flows__(ofconn, fm, request, &rules));
}
/* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
* code as encoded by ofp_mkerr() on failure.
*
- * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
+ * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
* if any. */
static int
-modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
+modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm,
+ const struct ofp_header *request)
{
- struct ofproto *p = ofconn->ofproto;
- struct rule *rule = find_flow_strict(p, fm);
- if (rule && !rule_is_hidden(rule)) {
- modify_flow(p, fm, rule);
- return send_buffered_packet(ofconn, rule, fm->buffer_id);
- } else {
- return add_flow(ofconn, fm);
- }
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
+ struct list rules;
+ int error;
+
+ error = collect_rules_strict(p, fm->table_id, &fm->cr, OFPP_NONE, &rules);
+ return (error ? error
+ : list_is_empty(&rules) ? add_flow(p, ofconn, fm, request)
+ : list_is_singleton(&rules) ? modify_flows__(ofconn, fm, request,
+ &rules)
+ : 0);
}
+\f
+/* OFPFC_DELETE implementation. */
-/* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
- * been identified as a flow in 'p''s flow table to be modified, by changing
- * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
- * ofp_action[] structures). */
+/* Deletes the rules listed in 'rules'.
+ *
+ * Returns 0 on success, otherwise an OpenFlow error code. */
static int
-modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
+delete_flows__(struct ofconn *ofconn, const struct ofp_header *request,
+ struct list *rules)
{
- size_t actions_len = fm->n_actions * sizeof *rule->actions;
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
+ struct rule *rule, *next;
+ struct ofopgroup *group;
- rule->flow_cookie = fm->cookie;
+ group = ofopgroup_create_for_ofconn(ofconn, request, UINT32_MAX);
+ LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
+ ofproto_rule_send_removed(rule, OFPRR_DELETE);
- /* If the actions are the same, do nothing. */
- if (fm->n_actions == rule->n_actions
- && (!fm->n_actions
- || !memcmp(fm->actions, rule->actions, actions_len))) {
- return 0;
+ ofoperation_create(group, rule, OFOPERATION_DELETE);
+ classifier_remove(&ofproto->tables[rule->table_id], &rule->cr);
+ rule->ofproto->ofproto_class->rule_destruct(rule);
}
-
- /* Replace actions. */
- free(rule->actions);
- rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
- rule->n_actions = fm->n_actions;
-
- p->need_revalidate = true;
+ ofopgroup_submit(group);
return 0;
}
-\f
-/* OFPFC_DELETE implementation. */
-
-static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
/* Implements OFPFC_DELETE. */
-static void
-delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
+static int
+delete_flows_loose(struct ofconn *ofconn, const struct flow_mod *fm,
+ const struct ofp_header *request)
{
- struct rule *rule, *next_rule;
- struct cls_cursor cursor;
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
+ struct list rules;
+ int error;
- cls_cursor_init(&cursor, &p->cls, &fm->cr);
- CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
- delete_flow(p, rule, htons(fm->out_port));
- }
+ error = collect_rules_loose(p, fm->table_id, &fm->cr, fm->out_port,
+ &rules);
+ return (error ? error
+ : !list_is_empty(&rules) ? delete_flows__(ofconn, request, &rules)
+ : 0);
}
/* Implements OFPFC_DELETE_STRICT. */
-static void
-delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
+static int
+delete_flow_strict(struct ofconn *ofconn, struct flow_mod *fm,
+ const struct ofp_header *request)
{
- struct rule *rule = find_flow_strict(p, fm);
- if (rule) {
- delete_flow(p, rule, htons(fm->out_port));
- }
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
+ struct list rules;
+ int error;
+
+ error = collect_rules_strict(p, fm->table_id, &fm->cr, fm->out_port,
+ &rules);
+ return (error ? error
+ : list_is_singleton(&rules) ? delete_flows__(ofconn, request,
+ &rules)
+ : 0);
}
-/* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
- * been identified as a flow to delete from 'p''s flow table, by deleting the
- * flow and sending out a OFPT_FLOW_REMOVED message to any interested
- * controller.
- *
- * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
- * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
- * specified 'out_port'. */
static void
-delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
+ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
{
- if (rule_is_hidden(rule)) {
- return;
- }
+ struct ofputil_flow_removed fr;
- if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
+ if (rule_is_hidden(rule) || !rule->send_flow_removed) {
return;
}
- rule_send_removed(p, rule, OFPRR_DELETE);
- rule_remove(p, rule);
+ fr.rule = rule->cr;
+ fr.cookie = rule->flow_cookie;
+ fr.reason = reason;
+ calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec);
+ fr.idle_timeout = rule->idle_timeout;
+ rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
+ &fr.byte_count);
+
+ connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
+}
+
+/* Sends an OpenFlow "flow removed" message with the given 'reason' (either
+ * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
+ * ofproto.
+ *
+ * ofproto implementation ->run() functions should use this function to expire
+ * OpenFlow flows. */
+void
+ofproto_rule_expire(struct rule *rule, uint8_t reason)
+{
+ struct ofproto *ofproto = rule->ofproto;
+ struct ofopgroup *group;
+
+ assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
+
+ ofproto_rule_send_removed(rule, reason);
+
+ group = ofopgroup_create(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);
}
\f
static int
-flow_mod_core(struct ofconn *ofconn, struct flow_mod *fm)
+handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
{
- struct ofproto *p = ofconn->ofproto;
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
+ struct flow_mod fm;
int error;
error = reject_slave_controller(ofconn, "flow_mod");
return error;
}
- error = validate_actions(fm->actions, fm->n_actions,
- &fm->cr.flow, p->max_ports);
+ if (list_size(&ofproto->pending) >= 50) {
+ return OFPROTO_POSTPONE;
+ }
+
+ error = ofputil_decode_flow_mod(&fm, oh,
+ ofconn_get_flow_mod_table_id(ofconn));
if (error) {
return error;
}
- /* We do not support the emergency flow cache. It will hopefully
- * get dropped from OpenFlow in the near future. */
- if (fm->flags & OFPFF_EMERG) {
+ /* 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);
}
- switch (fm->command) {
- case OFPFC_ADD:
- return add_flow(ofconn, fm);
-
- case OFPFC_MODIFY:
- return modify_flows_loose(ofconn, fm);
-
- case OFPFC_MODIFY_STRICT:
- return modify_flow_strict(ofconn, fm);
-
- case OFPFC_DELETE:
- delete_flows_loose(p, fm);
- return 0;
-
- case OFPFC_DELETE_STRICT:
- delete_flow_strict(p, fm);
- return 0;
-
- default:
- return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
- }
-}
-
-static int
-handle_ofpt_flow_mod(struct ofconn *ofconn, struct ofp_header *oh)
-{
- struct ofp_match orig_match;
- struct ofp_flow_mod *ofm;
- struct flow_mod fm;
- struct ofpbuf b;
- int error;
-
- b.data = oh;
- b.size = ntohs(oh->length);
-
- /* Dissect the message. */
- ofm = ofpbuf_try_pull(&b, sizeof *ofm);
- if (!ofm) {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
- }
- error = ofputil_pull_actions(&b, b.size, &fm.actions, &fm.n_actions);
- if (error) {
- return error;
- }
-
- /* Normalize ofm->match. If normalization actually changes anything, then
- * log the differences. */
- ofm->match.pad1[0] = ofm->match.pad2[0] = 0;
- orig_match = ofm->match;
- normalize_match(&ofm->match);
- if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) {
- static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1);
- if (!VLOG_DROP_INFO(&normal_rl)) {
- char *old = ofp_match_to_literal_string(&orig_match);
- char *new = ofp_match_to_literal_string(&ofm->match);
- VLOG_INFO("%s: normalization changed ofp_match, details:",
- rconn_get_name(ofconn->rconn));
- VLOG_INFO(" pre: %s", old);
- VLOG_INFO("post: %s", new);
- free(old);
- free(new);
- }
- }
-
- /* Translate the message. */
- ofputil_cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
- ofconn->flow_format, ofm->cookie, &fm.cr);
- fm.cookie = ofm->cookie;
- fm.command = ntohs(ofm->command);
- fm.idle_timeout = ntohs(ofm->idle_timeout);
- fm.hard_timeout = ntohs(ofm->hard_timeout);
- fm.buffer_id = ntohl(ofm->buffer_id);
- fm.out_port = ntohs(ofm->out_port);
- fm.flags = ntohs(ofm->flags);
-
- /* Execute the command. */
- return flow_mod_core(ofconn, &fm);
-}
-
-static int
-handle_nxt_flow_mod(struct ofconn *ofconn, struct ofp_header *oh)
-{
- struct nx_flow_mod *nfm;
- struct flow_mod fm;
- struct ofpbuf b;
- int error;
+ switch (fm.command) {
+ case OFPFC_ADD:
+ return add_flow(ofproto, ofconn, &fm, oh);
- b.data = oh;
- b.size = ntohs(oh->length);
+ case OFPFC_MODIFY:
+ return modify_flows_loose(ofconn, &fm, oh);
- /* Dissect the message. */
- nfm = ofpbuf_try_pull(&b, sizeof *nfm);
- if (!nfm) {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
- }
- error = nx_pull_match(&b, ntohs(nfm->match_len), ntohs(nfm->priority),
- &fm.cr);
- if (error) {
- return error;
- }
- error = ofputil_pull_actions(&b, b.size, &fm.actions, &fm.n_actions);
- if (error) {
- return error;
- }
+ case OFPFC_MODIFY_STRICT:
+ return modify_flow_strict(ofconn, &fm, oh);
- /* Translate the message. */
- fm.cookie = nfm->cookie;
- fm.command = ntohs(nfm->command);
- fm.idle_timeout = ntohs(nfm->idle_timeout);
- fm.hard_timeout = ntohs(nfm->hard_timeout);
- fm.buffer_id = ntohl(nfm->buffer_id);
- fm.out_port = ntohs(nfm->out_port);
- fm.flags = ntohs(nfm->flags);
-
- /* Execute the command. */
- return flow_mod_core(ofconn, &fm);
-}
+ case OFPFC_DELETE:
+ return delete_flows_loose(ofconn, &fm, oh);
-static int
-handle_tun_id_from_cookie(struct ofconn *ofconn, struct nxt_tun_id_cookie *msg)
-{
- int error;
+ case OFPFC_DELETE_STRICT:
+ return delete_flow_strict(ofconn, &fm, oh);
- error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
- if (error) {
- return error;
+ default:
+ if (fm.command > 0xff) {
+ VLOG_WARN_RL(&rl, "flow_mod has explicit table_id but "
+ "flow_mod_table_id extension is not enabled");
+ }
+ return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
}
-
- ofconn->flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
- return 0;
}
static int
-handle_role_request(struct ofconn *ofconn, struct nicira_header *msg)
+handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
{
- struct nx_role_request *nrr;
+ struct nx_role_request *nrr = (struct nx_role_request *) oh;
struct nx_role_request *reply;
struct ofpbuf *buf;
uint32_t role;
- if (ntohs(msg->header.length) != sizeof *nrr) {
- VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
- ntohs(msg->header.length), sizeof *nrr);
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
- }
- nrr = (struct nx_role_request *) msg;
-
- if (ofconn->type != OFCONN_PRIMARY) {
- VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
- "connection");
+ if (ofconn_get_type(ofconn) != OFCONN_PRIMARY) {
+ VLOG_WARN_RL(&rl, "ignoring role request on service connection");
return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
}
return ofp_mkerr(OFPET_BAD_REQUEST, -1);
}
- if (role == NX_ROLE_MASTER) {
- struct ofconn *other;
-
- HMAP_FOR_EACH (other, hmap_node, &ofconn->ofproto->controllers) {
- if (other->role == NX_ROLE_MASTER) {
- other->role = NX_ROLE_SLAVE;
- }
- }
+ if (ofconn_get_role(ofconn) != role
+ && ofconn_has_pending_opgroups(ofconn)) {
+ return OFPROTO_POSTPONE;
}
- ofconn->role = role;
- reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, msg->header.xid,
- &buf);
+ ofconn_set_role(ofconn, role);
+
+ reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
reply->role = htonl(role);
- queue_tx(buf, ofconn, ofconn->reply_counter);
+ ofconn_send_reply(ofconn, buf);
return 0;
}
static int
-handle_nxt_set_flow_format(struct ofconn *ofconn,
- struct nxt_set_flow_format *msg)
+handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
+ const struct ofp_header *oh)
{
- uint32_t format;
- int error;
-
- error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
- if (error) {
- return error;
- }
+ const struct nxt_flow_mod_table_id *msg
+ = (const struct nxt_flow_mod_table_id *) oh;
- format = ntohl(msg->format);
- if (format == NXFF_OPENFLOW10
- || format == NXFF_TUN_ID_FROM_COOKIE
- || format == NXFF_NXM) {
- ofconn->flow_format = format;
- return 0;
- } else {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
- }
+ ofconn_set_flow_mod_table_id(ofconn, msg->set != 0);
+ return 0;
}
static int
-handle_vendor(struct ofconn *ofconn, void *msg)
+handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
{
- struct ofproto *p = ofconn->ofproto;
- struct ofp_vendor_header *ovh = msg;
- struct nicira_header *nh;
-
- if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
- VLOG_WARN_RL(&rl, "received vendor message of length %u "
- "(expected at least %zu)",
- ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
- }
- if (ovh->vendor != htonl(NX_VENDOR_ID)) {
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
- }
- if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
- VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
- "(expected at least %zu)",
- ntohs(ovh->header.length), sizeof(struct nicira_header));
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
- }
-
- nh = msg;
- switch (ntohl(nh->subtype)) {
- case NXT_STATUS_REQUEST:
- return switch_status_handle_request(p->switch_status, ofconn->rconn,
- msg);
-
- case NXT_TUN_ID_FROM_COOKIE:
- return handle_tun_id_from_cookie(ofconn, msg);
-
- case NXT_ROLE_REQUEST:
- return handle_role_request(ofconn, msg);
+ const struct nxt_set_flow_format *msg
+ = (const struct nxt_set_flow_format *) oh;
+ uint32_t format;
- case NXT_SET_FLOW_FORMAT:
- return handle_nxt_set_flow_format(ofconn, msg);
+ format = ntohl(msg->format);
+ if (format != NXFF_OPENFLOW10 && format != NXFF_NXM) {
+ return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
+ }
- case NXT_FLOW_MOD:
- return handle_nxt_flow_mod(ofconn, &ovh->header);
+ if (format != ofconn_get_flow_format(ofconn)
+ && ofconn_has_pending_opgroups(ofconn)) {
+ /* Avoid sending async messages in surprising flow format. */
+ return OFPROTO_POSTPONE;
}
- return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
+ ofconn_set_flow_format(ofconn, format);
+ return 0;
}
static int
-handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
+handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
{
struct ofp_header *ob;
struct ofpbuf *buf;
- /* Currently, everything executes synchronously, so we can just
- * immediately send the barrier reply. */
+ if (ofconn_has_pending_opgroups(ofconn)) {
+ return OFPROTO_POSTPONE;
+ }
+
ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
- queue_tx(buf, ofconn, ofconn->reply_counter);
+ ofconn_send_reply(ofconn, buf);
return 0;
}
-static void
-handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
+static int
+handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
{
- struct ofp_header *oh = ofp_msg->data;
+ const struct ofp_header *oh = msg->data;
+ const struct ofputil_msg_type *type;
int error;
- COVERAGE_INC(ofproto_recv_openflow);
- switch (oh->type) {
- case OFPT_ECHO_REQUEST:
- error = handle_echo_request(ofconn, oh);
- break;
-
- case OFPT_ECHO_REPLY:
- error = 0;
- break;
+ error = ofputil_decode_msg_type(oh, &type);
+ if (error) {
+ return error;
+ }
- case OFPT_FEATURES_REQUEST:
- error = handle_features_request(ofconn, oh);
- break;
+ switch (ofputil_msg_type_code(type)) {
+ /* OpenFlow requests. */
+ case OFPUTIL_OFPT_ECHO_REQUEST:
+ return handle_echo_request(ofconn, oh);
- case OFPT_GET_CONFIG_REQUEST:
- error = handle_get_config_request(ofconn, oh);
- break;
+ case OFPUTIL_OFPT_FEATURES_REQUEST:
+ return handle_features_request(ofconn, oh);
- case OFPT_SET_CONFIG:
- error = handle_set_config(ofconn, ofp_msg->data);
- break;
+ case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
+ return handle_get_config_request(ofconn, oh);
- case OFPT_PACKET_OUT:
- error = handle_packet_out(ofconn, ofp_msg->data);
- break;
+ case OFPUTIL_OFPT_SET_CONFIG:
+ return handle_set_config(ofconn, msg->data);
- case OFPT_PORT_MOD:
- error = handle_port_mod(ofconn, oh);
- break;
+ case OFPUTIL_OFPT_PACKET_OUT:
+ return handle_packet_out(ofconn, oh);
- case OFPT_FLOW_MOD:
- error = handle_ofpt_flow_mod(ofconn, ofp_msg->data);
- break;
+ case OFPUTIL_OFPT_PORT_MOD:
+ return handle_port_mod(ofconn, oh);
- case OFPT_STATS_REQUEST:
- error = handle_stats_request(ofconn, oh);
- break;
+ case OFPUTIL_OFPT_FLOW_MOD:
+ return handle_flow_mod(ofconn, oh);
- case OFPT_VENDOR:
- error = handle_vendor(ofconn, ofp_msg->data);
- break;
+ case OFPUTIL_OFPT_BARRIER_REQUEST:
+ return handle_barrier_request(ofconn, oh);
- case OFPT_BARRIER_REQUEST:
- error = handle_barrier_request(ofconn, oh);
- break;
+ /* OpenFlow replies. */
+ case OFPUTIL_OFPT_ECHO_REPLY:
+ return 0;
+ /* Nicira extension requests. */
+ case OFPUTIL_NXT_ROLE_REQUEST:
+ return handle_role_request(ofconn, oh);
+
+ case OFPUTIL_NXT_FLOW_MOD_TABLE_ID:
+ return handle_nxt_flow_mod_table_id(ofconn, oh);
+
+ case OFPUTIL_NXT_SET_FLOW_FORMAT:
+ return handle_nxt_set_flow_format(ofconn, oh);
+
+ case OFPUTIL_NXT_FLOW_MOD:
+ return handle_flow_mod(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_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:
default:
if (VLOG_IS_WARN_ENABLED()) {
char *s = ofp_to_string(oh, ntohs(oh->length), 2);
VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
free(s);
}
- error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
- break;
- }
-
- if (error) {
- send_error_oh(ofconn, ofp_msg->data, error);
- }
-}
-\f
-static void
-handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
-{
- struct odp_msg *msg = packet->data;
- struct ofpbuf payload;
- struct facet *facet;
- struct flow flow;
-
- payload.data = msg + 1;
- payload.size = msg->length - sizeof *msg;
- flow_extract(&payload, msg->arg, msg->port, &flow);
-
- packet->l2 = payload.l2;
- packet->l3 = payload.l3;
- packet->l4 = payload.l4;
- packet->l7 = payload.l7;
-
- /* Check with in-band control to see if this packet should be sent
- * to the local port regardless of the flow table. */
- if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
- union odp_action action;
-
- memset(&action, 0, sizeof(action));
- action.output.type = ODPAT_OUTPUT;
- action.output.port = ODPP_LOCAL;
- dpif_execute(p->dpif, &action, 1, &payload);
- }
-
- facet = facet_lookup_valid(p, &flow);
- if (!facet) {
- struct rule *rule = rule_lookup(p, &flow);
- if (!rule) {
- /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
- struct ofport *port = get_port(p, msg->port);
- if (port) {
- if (port->opp.config & OFPPC_NO_PACKET_IN) {
- COVERAGE_INC(ofproto_no_packet_in);
- /* XXX install 'drop' flow entry */
- ofpbuf_delete(packet);
- return;
- }
- } else {
- VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
- msg->port);
- }
-
- COVERAGE_INC(ofproto_packet_in);
- send_packet_in(p, packet);
- return;
+ 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);
}
-
- facet = facet_create(p, rule, &flow, packet);
- } else if (!facet->may_install) {
- /* The facet is not installable, that is, we need to process every
- * packet, so process the current packet's actions into 'facet'. */
- facet_make_actions(p, facet, packet);
}
-
- if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
- /*
- * Extra-special case for fail-open mode.
- *
- * We are in fail-open mode and the packet matched the fail-open rule,
- * but we are connected to a controller too. We should send the packet
- * up to the controller in the hope that it will try to set up a flow
- * and thereby allow us to exit fail-open.
- *
- * See the top-level comment in fail-open.c for more information.
- */
- send_packet_in(p, ofpbuf_clone_with_headroom(packet,
- DPIF_RECV_MSG_PADDING));
- }
-
- ofpbuf_pull(packet, sizeof *msg);
- facet_execute(p, facet, packet);
- facet_install(p, facet, false);
}
-static void
-handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
+static bool
+handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
{
- struct odp_msg *msg = packet->data;
-
- switch (msg->type) {
- case _ODPL_ACTION_NR:
- COVERAGE_INC(ofproto_ctlr_action);
- send_packet_in(p, packet);
- break;
-
- case _ODPL_SFLOW_NR:
- if (p->sflow) {
- ofproto_sflow_received(p->sflow, msg);
- }
- ofpbuf_delete(packet);
- break;
-
- case _ODPL_MISS_NR:
- handle_odp_miss_msg(p, packet);
- break;
-
- default:
- VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
- msg->type);
- break;
+ int error = handle_openflow__(ofconn, ofp_msg);
+ if (error && error != OFPROTO_POSTPONE) {
+ ofconn_send_error(ofconn, ofp_msg->data, error);
}
+ COVERAGE_INC(ofproto_recv_openflow);
+ return error != OFPROTO_POSTPONE;
}
\f
-/* Flow expiration. */
-
-static int ofproto_dp_max_idle(const struct ofproto *);
-static void ofproto_update_used(struct ofproto *);
-static void rule_expire(struct ofproto *, struct rule *);
-static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
+/* Asynchronous operations. */
-/* This function is called periodically by ofproto_run(). Its job is to
- * collect updates for the flows that have been installed into the datapath,
- * most importantly when they last were used, and then use that information to
- * expire flows that have not been used recently.
+/* Creates and returns a new ofopgroup that is not associated with any
+ * OpenFlow connection.
*
- * Returns the number of milliseconds after which it should be called again. */
-static int
-ofproto_expire(struct ofproto *ofproto)
+ * The caller should add operations to the returned group with
+ * ofoperation_create() and then submit it with ofopgroup_submit(). */
+static struct ofopgroup *
+ofopgroup_create(struct ofproto *ofproto)
+{
+ struct ofopgroup *group = xzalloc(sizeof *group);
+ group->ofproto = ofproto;
+ list_init(&group->ofproto_node);
+ list_init(&group->ops);
+ list_init(&group->ofconn_node);
+ return group;
+}
+
+/* Creates and returns a new ofopgroup that is associated with 'ofconn'. If
+ * the ofopgroup eventually fails, then the error reply will include 'request'.
+ * If the ofopgroup eventually succeeds, then the packet with buffer id
+ * 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
+ *
+ * The caller should add operations to the returned group with
+ * ofoperation_create() and then submit it with ofopgroup_submit(). */
+static struct ofopgroup *
+ofopgroup_create_for_ofconn(struct ofconn *ofconn,
+ const struct ofp_header *request,
+ uint32_t buffer_id)
{
- struct rule *rule, *next_rule;
- struct cls_cursor cursor;
- int dp_max_idle;
-
- /* Update 'used' for each flow in the datapath. */
- ofproto_update_used(ofproto);
-
- /* Expire facets that have been idle too long. */
- dp_max_idle = ofproto_dp_max_idle(ofproto);
- ofproto_expire_facets(ofproto, dp_max_idle);
-
- /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
- cls_cursor_init(&cursor, &ofproto->cls, NULL);
- CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
- rule_expire(ofproto, rule);
- }
+ struct ofopgroup *group = ofopgroup_create(ofconn_get_ofproto(ofconn));
+ size_t request_len = ntohs(request->length);
- /* Let the hook know that we're at a stable point: all outstanding data
- * in existing flows has been accounted to the account_cb. Thus, the
- * hook can now reasonably do operations that depend on having accurate
- * flow volume accounting (currently, that's just bond rebalancing). */
- if (ofproto->ofhooks->account_checkpoint_cb) {
- ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
- }
+ ofconn_add_opgroup(ofconn, &group->ofconn_node);
+ group->ofconn = ofconn;
+ group->request = xmemdup(request, MIN(request_len, 64));
+ group->buffer_id = buffer_id;
- return MIN(dp_max_idle, 1000);
+ return group;
}
-/* Update 'used' member of installed facets. */
+/* Submits 'group' for processing.
+ *
+ * If 'group' contains no operations (e.g. none were ever added, or all of the
+ * ones that were added completed synchronously), then it is destroyed
+ * immediately. Otherwise it is added to the ofproto's list of pending
+ * groups. */
static void
-ofproto_update_used(struct ofproto *p)
-{
- struct odp_flow *flows;
- size_t n_flows;
- size_t i;
- int error;
-
- error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
- if (error) {
- return;
- }
-
- for (i = 0; i < n_flows; i++) {
- struct odp_flow *f = &flows[i];
- struct facet *facet;
- struct flow flow;
-
- odp_flow_key_to_flow(&f->key, &flow);
- facet = facet_find(p, &flow);
-
- if (facet && facet->installed) {
- facet_update_time(p, facet, &f->stats);
- facet_account(p, facet, f->stats.n_bytes);
- } else {
- /* There's a flow in the datapath that we know nothing about.
- * Delete it. */
- COVERAGE_INC(ofproto_unexpected_rule);
- dpif_flow_del(p->dpif, f);
- }
-
- }
- free(flows);
-}
-
-/* Calculates and returns the number of milliseconds of idle time after which
- * facets should expire from the datapath and we should fold their statistics
- * into their parent rules in userspace. */
-static int
-ofproto_dp_max_idle(const struct ofproto *ofproto)
+ofopgroup_submit(struct ofopgroup *group)
{
- /*
- * Idle time histogram.
- *
- * Most of the time a switch has a relatively small number of facets. When
- * this is the case we might as well keep statistics for all of them in
- * userspace and to cache them in the kernel datapath for performance as
- * well.
- *
- * As the number of facets increases, the memory required to maintain
- * statistics about them in userspace and in the kernel becomes
- * significant. However, with a large number of facets it is likely that
- * only a few of them are "heavy hitters" that consume a large amount of
- * bandwidth. At this point, only heavy hitters are worth caching in the
- * kernel and maintaining in userspaces; other facets we can discard.
- *
- * The technique used to compute the idle time is to build a histogram with
- * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
- * that is installed in the kernel gets dropped in the appropriate bucket.
- * After the histogram has been built, we compute the cutoff so that only
- * the most-recently-used 1% of facets (but at least 1000 flows) are kept
- * cached. At least the most-recently-used bucket of facets is kept, so
- * actually an arbitrary number of facets can be kept in any given
- * expiration run (though the next run will delete most of those unless
- * they receive additional data).
- *
- * This requires a second pass through the facets, in addition to the pass
- * made by ofproto_update_used(), because the former function never looks
- * at uninstallable facets.
- */
- enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
- enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
- int buckets[N_BUCKETS] = { 0 };
- struct facet *facet;
- int total, bucket;
- long long int now;
- int i;
-
- total = hmap_count(&ofproto->facets);
- if (total <= 1000) {
- return N_BUCKETS * BUCKET_WIDTH;
- }
-
- /* Build histogram. */
- now = time_msec();
- HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
- long long int idle = now - facet->used;
- int bucket = (idle <= 0 ? 0
- : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
- : (unsigned int) idle / BUCKET_WIDTH);
- buckets[bucket]++;
- }
-
- /* Find the first bucket whose flows should be expired. */
- for (bucket = 0; bucket < N_BUCKETS; bucket++) {
- if (buckets[bucket]) {
- int subtotal = 0;
- do {
- subtotal += buckets[bucket++];
- } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
- break;
- }
- }
-
- if (VLOG_IS_DBG_ENABLED()) {
- struct ds s;
-
- ds_init(&s);
- ds_put_cstr(&s, "keep");
- for (i = 0; i < N_BUCKETS; i++) {
- if (i == bucket) {
- ds_put_cstr(&s, ", drop");
- }
- if (buckets[i]) {
- ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
- }
- }
- VLOG_INFO("%s: %s (msec:count)",
- dpif_name(ofproto->dpif), ds_cstr(&s));
- ds_destroy(&s);
+ if (list_is_empty(&group->ops)) {
+ ofopgroup_destroy(group);
+ } else {
+ list_push_back(&group->ofproto->pending, &group->ofproto_node);
}
-
- return bucket * BUCKET_WIDTH;
}
static void
-facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
+ofopgroup_destroy(struct ofopgroup *group)
{
- if (ofproto->netflow && !facet_is_controller_flow(facet) &&
- netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
- struct ofexpired expired;
- struct odp_flow odp_flow;
-
- /* Get updated flow stats.
- *
- * XXX We could avoid this call entirely if (1) ofproto_update_used()
- * updated TCP flags and (2) the dpif_flow_list_all() in
- * ofproto_update_used() zeroed TCP flags. */
- memset(&odp_flow, 0, sizeof odp_flow);
- if (facet->installed) {
- odp_flow_key_from_flow(&odp_flow.key, &facet->flow);
- odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
- dpif_flow_get(ofproto->dpif, &odp_flow);
-
- if (odp_flow.stats.n_packets) {
- facet_update_time(ofproto, facet, &odp_flow.stats);
- netflow_flow_update_flags(&facet->nf_flow,
- odp_flow.stats.tcp_flags);
- }
- }
-
- expired.flow = facet->flow;
- expired.packet_count = facet->packet_count +
- odp_flow.stats.n_packets;
- expired.byte_count = facet->byte_count + odp_flow.stats.n_bytes;
- expired.used = facet->used;
-
- netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
+ assert(list_is_empty(&group->ops));
+ if (!list_is_empty(&group->ofproto_node)) {
+ list_remove(&group->ofproto_node);
}
-}
-
-static void
-ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
-{
- long long int cutoff = time_msec() - dp_max_idle;
- struct facet *facet, *next_facet;
-
- HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
- facet_active_timeout(ofproto, facet);
- if (facet->used < cutoff) {
- facet_remove(ofproto, facet);
+ if (!list_is_empty(&group->ofconn_node)) {
+ list_remove(&group->ofconn_node);
+ if (group->error) {
+ ofconn_send_error(group->ofconn, group->request, group->error);
}
+ connmgr_retry(group->ofproto->connmgr);
}
+ free(group->request);
+ free(group);
}
-/* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
- * then delete it entirely. */
+/* Initiates a new operation on 'rule', of the specified 'type', within
+ * 'group'. Prior to calling, 'rule' must not have any pending operation. */
static void
-rule_expire(struct ofproto *ofproto, struct rule *rule)
+ofoperation_create(struct ofopgroup *group, struct rule *rule,
+ enum ofoperation_type type)
{
- struct facet *facet, *next_facet;
- long long int now;
- uint8_t reason;
-
- /* Has 'rule' expired? */
- now = time_msec();
- if (rule->hard_timeout
- && now > rule->created + rule->hard_timeout * 1000) {
- reason = OFPRR_HARD_TIMEOUT;
- } else if (rule->idle_timeout && list_is_empty(&rule->facets)
- && now >rule->used + rule->idle_timeout * 1000) {
- reason = OFPRR_IDLE_TIMEOUT;
- } else {
- return;
- }
+ struct ofoperation *op;
- COVERAGE_INC(ofproto_expired);
+ assert(!rule->pending);
- /* Update stats. (This is a no-op if the rule expired due to an idle
- * timeout, because that only happens when the rule has no facets left.) */
- LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
- facet_remove(ofproto, facet);
- }
+ op = rule->pending = xzalloc(sizeof *op);
+ op->group = group;
+ list_push_back(&group->ops, &op->group_node);
+ op->rule = rule;
+ op->type = type;
+ op->status = -1;
+ op->flow_cookie = rule->flow_cookie;
- /* Get rid of the rule. */
- if (!rule_is_hidden(rule)) {
- rule_send_removed(ofproto, rule, reason);
+ if (type == OFOPERATION_DELETE) {
+ hmap_insert(&op->group->ofproto->deletions, &op->hmap_node,
+ cls_rule_hash(&rule->cr, rule->table_id));
}
- rule_remove(ofproto, rule);
-}
-\f
-static struct ofpbuf *
-compose_ofp_flow_removed(struct ofconn *ofconn, const struct rule *rule,
- uint8_t reason)
-{
- struct ofp_flow_removed *ofr;
- struct ofpbuf *buf;
-
- ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
- ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofr->match);
- ofr->cookie = rule->flow_cookie;
- ofr->priority = htons(rule->cr.priority);
- ofr->reason = reason;
- calc_flow_duration(rule->created, &ofr->duration_sec, &ofr->duration_nsec);
- ofr->idle_timeout = htons(rule->idle_timeout);
- ofr->packet_count = htonll(rule->packet_count);
- ofr->byte_count = htonll(rule->byte_count);
-
- return buf;
-}
-
-static struct ofpbuf *
-compose_nx_flow_removed(const struct rule *rule, uint8_t reason)
-{
- struct nx_flow_removed *nfr;
- struct ofpbuf *buf;
- int match_len;
-
- nfr = make_nxmsg(sizeof *nfr, NXT_FLOW_REMOVED, &buf);
-
- match_len = nx_put_match(buf, &rule->cr);
-
- nfr->cookie = rule->flow_cookie;
- nfr->priority = htons(rule->cr.priority);
- nfr->reason = reason;
- calc_flow_duration(rule->created, &nfr->duration_sec, &nfr->duration_nsec);
- nfr->idle_timeout = htons(rule->idle_timeout);
- nfr->match_len = htons(match_len);
- nfr->packet_count = htonll(rule->packet_count);
- nfr->byte_count = htonll(rule->byte_count);
-
- return buf;
}
static void
-rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
+ofoperation_destroy(struct ofoperation *op)
{
- struct ofconn *ofconn;
+ struct ofopgroup *group = op->group;
- if (!rule->send_flow_removed) {
- return;
+ if (op->rule) {
+ op->rule->pending = NULL;
}
-
- LIST_FOR_EACH (ofconn, node, &p->all_conns) {
- struct ofpbuf *msg;
-
- if (!rconn_is_connected(ofconn->rconn)
- || !ofconn_receives_async_msgs(ofconn)) {
- continue;
- }
-
- msg = (ofconn->flow_format == NXFF_NXM
- ? compose_nx_flow_removed(rule, reason)
- : compose_ofp_flow_removed(ofconn, rule, reason));
-
- /* Account flow expirations under ofconn->reply_counter, the counter
- * for replies to OpenFlow requests. That works because preventing
- * OpenFlow requests from being processed also prevents new flows from
- * being added (and expiring). (It also prevents processing OpenFlow
- * requests that would not add new flows, so it is imperfect.) */
- queue_tx(msg, ofconn, ofconn->reply_counter);
+ if (op->type == OFOPERATION_DELETE) {
+ hmap_remove(&group->ofproto->deletions, &op->hmap_node);
}
-}
-
-/* pinsched callback for sending 'packet' on 'ofconn'. */
-static void
-do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
-{
- struct ofconn *ofconn = ofconn_;
+ list_remove(&op->group_node);
+ free(op->actions);
+ free(op);
- rconn_send_with_limit(ofconn->rconn, packet,
- ofconn->packet_in_counter, 100);
+ if (list_is_empty(&group->ops) && !list_is_empty(&group->ofproto_node)) {
+ ofopgroup_destroy(group);
+ }
}
-/* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
- * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
- * packet scheduler for sending.
+/* 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()).
*
- * 'max_len' specifies the maximum number of bytes of the packet to send on
- * 'ofconn' (INT_MAX specifies no limit).
+ * If 'op' is a "delete flow" operation, 'error' must be 0. That is, flow
+ * deletions are not allowed to fail.
*
- * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
- * ownership is transferred to this function. */
-static void
-schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
- bool clone)
+ * Please see the large comment in ofproto/private.h titled "Asynchronous
+ * Operation Support" for more information. */
+void
+ofoperation_complete(struct ofoperation *op, int error)
{
- struct ofproto *ofproto = ofconn->ofproto;
- struct ofp_packet_in *opi = packet->data;
- uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
- int send_len, trim_size;
- uint32_t buffer_id;
-
- /* Get buffer. */
- if (opi->reason == OFPR_ACTION) {
- buffer_id = UINT32_MAX;
- } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
- buffer_id = pktbuf_get_null();
- } else if (!ofconn->pktbuf) {
- buffer_id = UINT32_MAX;
- } else {
- struct ofpbuf payload;
- payload.data = opi->data;
- payload.size = packet->size - offsetof(struct ofp_packet_in, data);
- buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
- }
+ struct ofopgroup *group = op->group;
+ struct rule *rule = op->rule;
+ struct classifier *table = &rule->ofproto->tables[rule->table_id];
- /* Figure out how much of the packet to send. */
- send_len = ntohs(opi->total_len);
- if (buffer_id != UINT32_MAX) {
- send_len = MIN(send_len, ofconn->miss_send_len);
- }
- send_len = MIN(send_len, max_len);
+ assert(rule->pending == op);
+ assert(op->status < 0);
+ assert(error >= 0);
- /* Adjust packet length and clone if necessary. */
- trim_size = offsetof(struct ofp_packet_in, data) + send_len;
- if (clone) {
- packet = ofpbuf_clone_data(packet->data, trim_size);
- opi = packet->data;
- } else {
- packet->size = trim_size;
+ if (!error
+ && !group->error
+ && op->type != OFOPERATION_DELETE
+ && group->ofconn
+ && group->buffer_id != UINT32_MAX
+ && list_is_singleton(&op->group_node)) {
+ struct ofpbuf *packet;
+ uint16_t in_port;
+
+ error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
+ &packet, &in_port);
+ if (packet) {
+ assert(!error);
+ error = rule_execute(rule, in_port, packet);
+ }
+ }
+ if (!group->error) {
+ group->error = error;
}
- /* Update packet headers. */
- opi->buffer_id = htonl(buffer_id);
- update_openflow_length(packet);
+ switch (op->type) {
+ case OFOPERATION_ADD:
+ if (!error) {
+ if (op->victim) {
+ ofproto_rule_destroy__(op->victim);
+ }
+ } else {
+ if (op->victim) {
+ classifier_replace(table, &op->victim->cr);
+ op->victim = NULL;
+ } else {
+ classifier_remove(table, &rule->cr);
+ }
+ ofproto_rule_destroy__(rule);
+ }
+ op->victim = NULL;
+ break;
- /* Hand over to packet scheduler. It might immediately call into
- * do_send_packet_in() or it might buffer it for a while (until a later
- * call to pinsched_run()). */
- pinsched_send(ofconn->schedulers[opi->reason], in_port,
- packet, do_send_packet_in, ofconn);
-}
+ case OFOPERATION_DELETE:
+ assert(!error);
+ ofproto_rule_destroy__(rule);
+ op->rule = NULL;
+ break;
-/* Replace struct odp_msg header in 'packet' by equivalent struct
- * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
- * returned by dpif_recv()).
- *
- * The conversion is not complete: the caller still needs to trim any unneeded
- * payload off the end of the buffer, set the length in the OpenFlow header,
- * and set buffer_id. Those require us to know the controller settings and so
- * must be done on a per-controller basis.
- *
- * Returns the maximum number of bytes of the packet that should be sent to
- * the controller (INT_MAX if no limit). */
-static int
-do_convert_to_packet_in(struct ofpbuf *packet)
-{
- struct odp_msg *msg = packet->data;
- struct ofp_packet_in *opi;
- uint8_t reason;
- uint16_t total_len;
- uint16_t in_port;
- int max_len;
+ case OFOPERATION_MODIFY:
+ if (error) {
+ free(rule->actions);
+ rule->actions = op->actions;
+ rule->n_actions = op->n_actions;
+ op->actions = NULL;
+ }
+ break;
- /* Extract relevant header fields */
- if (msg->type == _ODPL_ACTION_NR) {
- reason = OFPR_ACTION;
- max_len = msg->arg;
- } else {
- reason = OFPR_NO_MATCH;
- max_len = INT_MAX;
+ default:
+ NOT_REACHED();
}
- total_len = msg->length - sizeof *msg;
- in_port = odp_port_to_ofp_port(msg->port);
-
- /* Repurpose packet buffer by overwriting header. */
- ofpbuf_pull(packet, sizeof(struct odp_msg));
- opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
- opi->header.version = OFP_VERSION;
- opi->header.type = OFPT_PACKET_IN;
- opi->total_len = htons(total_len);
- opi->in_port = htons(in_port);
- opi->reason = reason;
-
- return max_len;
+ ofoperation_destroy(op);
}
-/* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
- * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
- * as necessary according to their individual configurations.
- *
- * 'packet' must have sufficient headroom to convert it into a struct
- * ofp_packet_in (e.g. as returned by dpif_recv()).
- *
- * Takes ownership of 'packet'. */
-static void
-send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
+struct rule *
+ofoperation_get_victim(struct ofoperation *op)
{
- struct ofconn *ofconn, *prev;
- int max_len;
-
- max_len = do_convert_to_packet_in(packet);
-
- prev = NULL;
- LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
- if (ofconn_receives_async_msgs(ofconn)) {
- if (prev) {
- schedule_packet_in(prev, packet, max_len, true);
- }
- prev = ofconn;
- }
- }
- if (prev) {
- schedule_packet_in(prev, packet, max_len, false);
- } else {
- ofpbuf_delete(packet);
- }
+ assert(op->type == OFOPERATION_ADD);
+ return op->victim;
}
-
+\f
static uint64_t
pick_datapath_id(const struct ofproto *ofproto)
{
const struct ofport *port;
- port = get_port(ofproto, ODPP_LOCAL);
+ port = ofproto_get_port(ofproto, OFPP_LOCAL);
if (port) {
uint8_t ea[ETH_ADDR_LEN];
int error;
return eth_addr_to_uint64(ea);
}
\f
-static bool
-default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
- struct odp_actions *actions, tag_type *tags,
- uint16_t *nf_output_iface, void *ofproto_)
+/* unixctl commands. */
+
+struct ofproto *
+ofproto_lookup(const char *name)
{
- struct ofproto *ofproto = ofproto_;
- int out_port;
+ struct ofproto *ofproto;
- /* Drop frames for reserved multicast addresses. */
- if (eth_addr_is_reserved(flow->dl_dst)) {
- return true;
+ HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
+ &all_ofprotos) {
+ if (!strcmp(ofproto->name, name)) {
+ return ofproto;
+ }
}
+ return NULL;
+}
- /* Learn source MAC (but don't try to learn from revalidation). */
- if (packet != NULL) {
- tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
- 0, flow->in_port,
- GRAT_ARP_LOCK_NONE);
- if (rev_tag) {
- /* The log messages here could actually be useful in debugging,
- * so keep the rate limit relatively high. */
- static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
- VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
- ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
- ofproto_revalidate(ofproto, rev_tag);
- }
+static void
+ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
+ void *aux OVS_UNUSED)
+{
+ struct ofproto *ofproto;
+ struct ds results;
+
+ ds_init(&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));
+ ds_destroy(&results);
+}
- /* Determine output port. */
- out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
- NULL);
- if (out_port < 0) {
- flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
- nf_output_iface, actions);
- } else if (out_port != flow->in_port) {
- odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
- *nf_output_iface = out_port;
- } else {
- /* Drop. */
+static void
+ofproto_unixctl_init(void)
+{
+ static bool registered;
+ if (registered) {
+ return;
}
+ registered = true;
- return true;
+ unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);
}
-
-static const struct ofhooks default_ofhooks = {
- default_normal_ofhook_cb,
- NULL,
- NULL
-};