#include <netinet/in.h>
#include <stdbool.h>
#include <stdlib.h>
+#include "byte-order.h"
#include "classifier.h"
#include "coverage.h"
#include "discovery.h"
#include "unixctl.h"
#include "vconn.h"
#include "vlog.h"
-#include "xtoxll.h"
-VLOG_DEFINE_THIS_MODULE(ofproto)
+VLOG_DEFINE_THIS_MODULE(ofproto);
#include "sflow_api.h"
-enum {
- TABLEID_HASH = 0,
- TABLEID_CLASSIFIER = 1
-};
-
-
struct ofport {
struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
struct netdev *netdev;
static void hton_ofp_phy_port(struct ofp_phy_port *);
static int xlate_actions(const union ofp_action *in, size_t n_in,
- const flow_t *flow, struct ofproto *ofproto,
+ const struct flow *, struct ofproto *,
const struct ofpbuf *packet,
struct odp_actions *out, tag_type *tags,
bool *may_set_up_flow, uint16_t *nf_output_iface);
struct rule {
struct cls_rule cr;
- uint64_t flow_cookie; /* Controller-issued identifier.
- (Kept in network-byte order.) */
+ ovs_be64 flow_cookie; /* Controller-issued identifier. */
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? */
static struct rule *rule_create(struct ofproto *, struct rule *super,
const union ofp_action *, size_t n_actions,
uint16_t idle_timeout, uint16_t hard_timeout,
- uint64_t flow_cookie, bool send_flow_removed);
+ ovs_be64 flow_cookie, bool send_flow_removed);
static void rule_free(struct rule *);
static void rule_destroy(struct ofproto *, struct rule *);
static struct rule *rule_from_cls_rule(const struct cls_rule *);
static uint64_t pick_datapath_id(const struct ofproto *);
static uint64_t pick_fallback_dpid(void);
-static void update_used(struct ofproto *);
+static int ofproto_expire(struct ofproto *);
+
static void update_stats(struct ofproto *, struct rule *,
const struct odp_flow_stats *);
-static void expire_rule(struct cls_rule *, void *ofproto);
-static void active_timeout(struct ofproto *ofproto, struct rule *rule);
static bool revalidate_rule(struct ofproto *p, struct rule *rule);
static void revalidate_cb(struct cls_rule *rule_, void *p_);
static void handle_openflow(struct ofconn *, struct ofproto *,
struct ofpbuf *);
-static void refresh_port_groups(struct ofproto *);
-
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 *);
struct ofport *ofport;
os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
- refresh_port_groups(ofproto);
HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
ofproto_sflow_add_port(os, ofport->odp_port,
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);
- p->next_expiration = time_msec() + 1000;
- update_used(p);
-
- classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
-
- /* Let the hook know that we're at a stable point: all outstanding data
- * in existing flows has been accounted to the account_cb. Thus, the
- * hook can now reasonably do operations that depend on having accurate
- * flow volume accounting (currently, that's just bond rebalancing). */
- if (p->ofhooks->account_checkpoint_cb) {
- p->ofhooks->account_checkpoint_cb(p->aux);
- }
}
if (p->netflow) {
return !hmap_is_empty(&p->controllers);
}
+/* Deletes port number 'odp_port' from the datapath for 'ofproto'.
+ *
+ * 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.)
+ *
+ * Returns 0 if successful, otherwise a positive errno. */
int
-ofproto_send_packet(struct ofproto *p, const flow_t *flow,
+ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
+{
+ struct ofport *ofport = get_port(ofproto, odp_port);
+ const char *name = ofport ? (char *) ofport->opp.name : "<unknown>";
+ 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);
+ }
+ 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)
+{
+ struct ofport *ofport = get_port(ofproto, odp_port);
+ return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD);
+}
+
+int
+ofproto_send_packet(struct ofproto *p, const struct flow *flow,
const union ofp_action *actions, size_t n_actions,
const struct ofpbuf *packet)
{
/* XXX Should we translate the dpif_execute() errno value into an OpenFlow
* error code? */
- dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
- odp_actions.n_actions, packet);
+ dpif_execute(p->dpif, odp_actions.actions, odp_actions.n_actions, packet);
return 0;
}
void
-ofproto_add_flow(struct ofproto *p,
- const flow_t *flow, uint32_t wildcards, unsigned int priority,
+ofproto_add_flow(struct ofproto *p, const struct flow *flow,
+ uint32_t wildcards, unsigned int priority,
const union ofp_action *actions, size_t n_actions,
int idle_timeout)
{
}
void
-ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
+ofproto_delete_flow(struct ofproto *ofproto, const struct flow *flow,
uint32_t wildcards, unsigned int priority)
{
+ struct cls_rule target;
struct rule *rule;
+ cls_rule_from_flow(flow, wildcards, priority, &target);
rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
- flow, wildcards,
- priority));
+ &target));
if (rule) {
rule_remove(ofproto, rule);
}
size_t n_odp_ports;
size_t i;
+ COVERAGE_INC(ofproto_reinit_ports);
+
svec_init(&devnames);
HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
svec_add (&devnames, (char *) ofport->opp.name);
svec_destroy(&devnames);
}
-static size_t
-refresh_port_group(struct ofproto *p, unsigned int group)
-{
- uint16_t *ports;
- size_t n_ports;
- struct ofport *port;
-
- assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
-
- ports = xmalloc(hmap_count(&p->ports) * sizeof *ports);
- n_ports = 0;
- HMAP_FOR_EACH (port, hmap_node, &p->ports) {
- if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
- ports[n_ports++] = port->odp_port;
- }
- }
- dpif_port_group_set(p->dpif, group, ports, n_ports);
- free(ports);
-
- return n_ports;
-}
-
-static void
-refresh_port_groups(struct ofproto *p)
-{
- size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
- size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
- if (p->sflow) {
- ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
- }
-}
-
static struct ofport *
make_ofport(const struct odp_port *odp_port)
{
enum netdev_flags flags;
struct ofport *ofport;
struct netdev *netdev;
- bool carrier;
int error;
memset(&netdev_options, 0, sizeof netdev_options);
netdev_get_flags(netdev, &flags);
ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
- netdev_get_carrier(netdev, &carrier);
- ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
+ ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
netdev_get_features(netdev,
&ofport->opp.curr, &ofport->opp.advertised,
hton_ofp_phy_port(&ops->desc);
queue_tx(b, ofconn, NULL);
}
- if (p->ofhooks->port_changed_cb) {
- p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
- }
}
static void
: !new_ofport ? OFPPR_DELETE
: OFPPR_MODIFY));
ofport_free(old_ofport);
-
- /* Update port groups. */
- refresh_port_groups(p);
}
static int
}
}
free(ports);
- refresh_port_groups(p);
return 0;
}
\f
rule_create(struct ofproto *ofproto, struct rule *super,
const union ofp_action *actions, size_t n_actions,
uint16_t idle_timeout, uint16_t hard_timeout,
- uint64_t flow_cookie, bool send_flow_removed)
+ ovs_be64 flow_cookie, bool send_flow_removed)
{
struct rule *rule = xzalloc(sizeof *rule);
rule->idle_timeout = idle_timeout;
} else {
list_init(&rule->list);
}
- rule->n_actions = n_actions;
- rule->actions = xmemdup(actions, n_actions * sizeof *actions);
+ if (n_actions > 0) {
+ rule->n_actions = n_actions;
+ rule->actions = xmemdup(actions, n_actions * sizeof *actions);
+ }
netflow_flow_clear(&rule->nf_flow);
netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
}
static bool
-rule_has_out_port(const struct rule *rule, uint16_t out_port)
+rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
{
const union ofp_action *oa;
struct actions_iterator i;
} else {
int error;
- error = dpif_execute(ofproto->dpif, in_port,
- actions, n_actions, packet);
+ error = dpif_execute(ofproto->dpif, actions, n_actions, packet);
ofpbuf_delete(packet);
return !error;
}
* Takes ownership of 'packet'. */
static void
rule_execute(struct ofproto *ofproto, struct rule *rule,
- struct ofpbuf *packet, const flow_t *flow)
+ struct ofpbuf *packet, const struct flow *flow)
{
const union odp_action *actions;
struct odp_flow_stats stats;
/* Send the packet and credit it to the rule. */
if (packet) {
- flow_t flow;
+ struct flow flow;
flow_extract(packet, 0, in_port, &flow);
rule_execute(p, rule, packet, &flow);
}
static struct rule *
rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
- const flow_t *flow)
+ const struct flow *flow)
{
struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
rule->idle_timeout, rule->hard_timeout,
COVERAGE_INC(ofproto_subrule_create);
cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
: rule->cr.priority), &subrule->cr);
- classifier_insert_exact(&ofproto->cls, &subrule->cr);
+
+ if (classifier_insert(&ofproto->cls, &subrule->cr)) {
+ /* Can't happen, */
+ NOT_REACHED();
+ }
return subrule;
}
+/* Remove 'rule' from 'ofproto' and free up the associated memory:
+ *
+ * - If 'rule' was installed in the datapath, uninstalls it and updates
+ * 'rule''s statistics (or its super-rule's statistics, if it is a
+ * subrule), via rule_uninstall().
+ *
+ * - Removes 'rule' from the classifier.
+ *
+ * - If 'rule' is a super-rule that has subrules, revalidates (and possibly
+ * uninstalls and destroys) its subrules, via rule_destroy().
+ */
static void
rule_remove(struct ofproto *ofproto, struct rule *rule)
{
struct odp_flow_put *put)
{
memset(&put->flow.stats, 0, sizeof put->flow.stats);
- put->flow.key = rule->cr.flow;
+ odp_flow_key_from_flow(&put->flow.key, &rule->cr.flow);
put->flow.actions = rule->odp_actions;
put->flow.n_actions = rule->n_odp_actions;
put->flow.flags = 0;
}
}
+/* 'rule' must be an exact-match rule in 'p'.
+ *
+ * If 'rule' is installed in the datapath, uninstalls it and updates's
+ * statistics. If 'rule' is a subrule, the statistics that are updated are
+ * actually its super-rule's statistics; otherwise 'rule''s own statistics are
+ * updated.
+ *
+ * If 'rule' is not installed, this function has no effect. */
static void
rule_uninstall(struct ofproto *p, struct rule *rule)
{
if (rule->installed) {
struct odp_flow odp_flow;
- odp_flow.key = rule->cr.flow;
+ odp_flow_key_from_flow(&odp_flow.key, &rule->cr.flow);
odp_flow.actions = NULL;
odp_flow.n_actions = 0;
odp_flow.flags = 0;
}
}
-static void
-send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
- int error, const void *data, size_t len)
-{
- struct ofpbuf *buf;
- struct ofp_error_msg *oem;
-
- if (!(error >> 16)) {
- VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
- error);
- return;
- }
-
- COVERAGE_INC(ofproto_error);
- oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
- oh ? oh->xid : 0, &buf);
- oem->type = htons((unsigned int) error >> 16);
- oem->code = htons(error & 0xffff);
- memcpy(oem->data, data, len);
- queue_tx(buf, ofconn, ofconn->reply_counter);
-}
-
static void
send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
int error)
{
- size_t oh_length = ntohs(oh->length);
- send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
+ struct ofpbuf *buf = make_ofp_error_msg(error, oh);
+ if (buf) {
+ COVERAGE_INC(ofproto_error);
+ queue_tx(buf, ofconn, ofconn->reply_counter);
+ }
}
static void
return 0;
}
-static void
-add_output_group_action(struct odp_actions *actions, uint16_t group,
- uint16_t *nf_output_iface)
-{
- odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
-
- if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
- *nf_output_iface = NF_OUT_FLOOD;
- }
-}
-
static void
add_controller_action(struct odp_actions *actions, uint16_t max_len)
{
struct action_xlate_ctx {
/* Input. */
- flow_t flow; /* Flow to which these actions correspond. */
+ 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
}
static struct rule *
-lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
+lookup_valid_rule(struct ofproto *ofproto, const struct flow *flow)
{
struct rule *rule;
- rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
+ rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow,
+ CLS_INC_ALL));
/* The rule we found might not be valid, since we could be in need of
* revalidation. If it is not valid, don't return it. */
}
}
+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;
+
+ 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;
+ }
+ }
+ *nf_output_iface = NF_OUT_FLOOD;
+}
+
static void
xlate_output_action__(struct action_xlate_ctx *ctx,
uint16_t port, uint16_t max_len)
}
break;
case OFPP_FLOOD:
- add_output_group_action(ctx->out, DP_GROUP_FLOOD,
- &ctx->nf_output_iface);
+ flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
+ &ctx->nf_output_iface, ctx->out);
break;
case OFPP_ALL:
- add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
+ 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);
}
}
+static void
+xlate_set_queue_action(struct action_xlate_ctx *ctx,
+ const struct nx_action_set_queue *nasq)
+{
+ uint32_t priority;
+ int error;
+
+ 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;
+ }
+
+ remove_pop_action(ctx);
+ odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
+ = priority;
+}
+
static void
xlate_nicira_action(struct action_xlate_ctx *ctx,
const struct nx_action_header *nah)
{
const struct nx_action_resubmit *nar;
const struct nx_action_set_tunnel *nast;
+ const struct nx_action_set_queue *nasq;
union odp_action *oa;
int subtype = ntohs(nah->subtype);
}
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;
+
/* 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. */
break;
case OFPAT_SET_VLAN_VID:
- oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
- ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
+ oa = odp_actions_add(ctx->out, ODPAT_SET_DL_TCI);
+ oa->dl_tci.tci = ia->vlan_vid.vlan_vid;
+ oa->dl_tci.tci |= htons(ctx->flow.dl_vlan_pcp << VLAN_PCP_SHIFT);
+ ctx->flow.dl_vlan = ia->vlan_vid.vlan_vid;
break;
case OFPAT_SET_VLAN_PCP:
- oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
- ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
+ oa = odp_actions_add(ctx->out, ODPAT_SET_DL_TCI);
+ oa->dl_tci.tci = htons(ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT);
+ oa->dl_tci.tci |= ctx->flow.dl_vlan;
+ ctx->flow.dl_vlan_pcp = ia->vlan_pcp.vlan_pcp;
break;
case OFPAT_STRIP_VLAN:
static int
xlate_actions(const union ofp_action *in, size_t n_in,
- const flow_t *flow, struct ofproto *ofproto,
+ 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 ofp_packet_out *opo;
struct ofpbuf payload, *buffer;
struct odp_actions actions;
+ struct flow flow;
int n_actions;
uint16_t in_port;
- flow_t flow;
int error;
error = reject_slave_controller(ofconn, oh);
flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
&flow, p, &payload, &actions, NULL, NULL, NULL);
- if (error) {
- return error;
+ if (!error) {
+ dpif_execute(p->dpif, actions.actions, actions.n_actions, &payload);
}
-
- dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
- &payload);
ofpbuf_delete(buffer);
- return 0;
+ return error;
}
static void
netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
}
}
-#define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
+#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_FLOOD) {
- port->opp.config ^= OFPPC_NO_FLOOD;
- refresh_port_groups(p);
- }
if (mask & OFPPC_NO_PACKET_IN) {
port->opp.config ^= OFPPC_NO_PACKET_IN;
}
}
static struct ofpbuf *
-make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
+make_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
{
struct ofp_stats_reply *osr;
struct ofpbuf *msg;
return 0;
}
-static void
-count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
-{
- struct rule *rule = rule_from_cls_rule(cls_rule);
- int *n_subrules = n_subrules_;
-
- if (rule->super) {
- (*n_subrules)++;
- }
-}
-
static int
handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
struct ofp_stats_request *request)
{
struct ofp_table_stats *ots;
struct ofpbuf *msg;
- struct odp_stats dpstats;
- int n_exact, n_subrules, n_wild;
+ struct rule *rule;
+ int n_rules;
msg = start_stats_reply(request, sizeof *ots * 2);
- /* Count rules of various kinds. */
- n_subrules = 0;
- classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
- n_exact = classifier_count_exact(&p->cls) - n_subrules;
- n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
-
- /* Hash table. */
- dpif_get_dp_stats(p->dpif, &dpstats);
- ots = append_stats_reply(sizeof *ots, ofconn, &msg);
- memset(ots, 0, sizeof *ots);
- ots->table_id = TABLEID_HASH;
- strcpy(ots->name, "hash");
- ots->wildcards = htonl(0);
- ots->max_entries = htonl(dpstats.max_capacity);
- ots->active_count = htonl(n_exact);
- ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
- dpstats.n_missed);
- ots->matched_count = htonll(dpstats.n_hit); /* XXX */
+ /* Count rules other than subrules. */
+ n_rules = classifier_count(&p->cls);
+ CLASSIFIER_FOR_EACH_EXACT_RULE (rule, cr, &p->cls) {
+ if (rule->super) {
+ n_rules--;
+ }
+ }
/* Classifier table. */
ots = append_stats_reply(sizeof *ots, ofconn, &msg);
memset(ots, 0, sizeof *ots);
- ots->table_id = TABLEID_CLASSIFIER;
strcpy(ots->name, "classifier");
ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
: htonl(OFPFW_ALL);
- ots->max_entries = htonl(65536);
- ots->active_count = htonl(n_wild);
+ ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
+ ots->active_count = htonl(n_rules);
ots->lookup_count = htonll(0); /* XXX */
ots->matched_count = htonll(0); /* XXX */
struct flow_stats_cbdata {
struct ofproto *ofproto;
struct ofconn *ofconn;
- uint16_t out_port;
+ ovs_be16 out_port;
struct ofpbuf *msg;
};
if (rule->cr.wc.wildcards) {
size_t i = 0;
LIST_FOR_EACH (subrule, list, &rule->list) {
- odp_flows[i++].key = subrule->cr.flow;
+ odp_flow_key_from_flow(&odp_flows[i++].key, &subrule->cr.flow);
packet_count += subrule->packet_count;
byte_count += subrule->byte_count;
}
} else {
- odp_flows[0].key = rule->cr.flow;
+ odp_flow_key_from_flow(&odp_flows[0].key, &rule->cr.flow);
}
/* Fetch up-to-date statistics from the datapath and add them in. */
ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
ofs->length = htons(len);
- ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
+ ofs->table_id = 0;
ofs->pad = 0;
flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
cbdata->ofproto->tun_id_from_cookie, &ofs->match);
memset(ofs->pad2, 0, sizeof ofs->pad2);
ofs->packet_count = htonll(packet_count);
ofs->byte_count = htonll(byte_count);
- memcpy(ofs->actions, rule->actions, act_len);
+ if (rule->n_actions > 0) {
+ memcpy(ofs->actions, rule->actions, act_len);
+ }
}
static int
table_id_to_include(uint8_t table_id)
{
- return (table_id == TABLEID_HASH ? CLS_INC_EXACT
- : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
- : table_id == 0xff ? CLS_INC_ALL
- : 0);
+ return table_id == 0 || table_id == 0xff ? CLS_INC_ALL : 0;
}
static int
ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
ofp_print_match(results, &match, true);
- ofp_print_actions(results, &rule->actions->header, act_len);
+ if (act_len > 0) {
+ ofp_print_actions(results, &rule->actions->header, act_len);
+ } else {
+ ds_put_cstr(results, "drop");
+ }
ds_put_cstr(results, "\n");
}
struct aggregate_stats_cbdata {
struct ofproto *ofproto;
- uint16_t out_port;
+ ovs_be16 out_port;
uint64_t packet_count;
uint64_t byte_count;
uint32_t n_flows;
int error;
if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
- flow_t flow;
- uint32_t wildcards;
+ struct cls_rule cr;
- flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
- &flow, &wildcards);
- if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
- ntohs(ofm->priority))) {
+ cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
+ p->tun_id_from_cookie, ofm->cookie, &cr);
+ if (classifier_rule_overlaps(&p->cls, &cr)) {
return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
}
}
static struct rule *
find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
{
- uint32_t wildcards;
- flow_t flow;
+ struct cls_rule target;
- flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
- &flow, &wildcards);
- return rule_from_cls_rule(classifier_find_rule_exactly(
- &p->cls, &flow, wildcards,
- ntohs(ofm->priority)));
+ cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
+ p->tun_id_from_cookie, ofm->cookie, &target);
+ return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &target));
}
static int
{
struct ofpbuf *packet;
uint16_t in_port;
- flow_t flow;
+ struct flow flow;
int error;
if (ofm->buffer_id == htonl(UINT32_MAX)) {
/* If the actions are the same, do nothing. */
if (n_actions == rule->n_actions
- && !memcmp(ofm->actions, rule->actions, actions_len))
+ && (!n_actions || !memcmp(ofm->actions, rule->actions, actions_len)))
{
return 0;
}
/* Replace actions. */
free(rule->actions);
- rule->actions = xmemdup(ofm->actions, actions_len);
+ rule->actions = n_actions ? xmemdup(ofm->actions, actions_len) : NULL;
rule->n_actions = n_actions;
/* Make sure that the datapath gets updated properly. */
struct delete_flows_cbdata {
struct ofproto *ofproto;
- uint16_t out_port;
+ ovs_be16 out_port;
};
static void delete_flows_cb(struct cls_rule *, void *cbdata_);
-static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port);
+static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
/* Implements OFPFC_DELETE. */
static void
* '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, uint16_t out_port)
+delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
{
if (rule_is_hidden(rule)) {
return;
}
ofconn->role = role;
- reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid,
- &buf);
- reply->nxh.vendor = htonl(NX_VENDOR_ID);
- reply->nxh.subtype = htonl(NXT_ROLE_REPLY);
+ reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, msg->header.xid,
+ &buf);
reply->role = htonl(role);
queue_tx(buf, ofconn, ofconn->reply_counter);
struct odp_msg *msg = packet->data;
struct rule *rule;
struct ofpbuf payload;
- flow_t flow;
+ struct flow flow;
payload.data = msg + 1;
payload.size = msg->length - sizeof *msg;
memset(&action, 0, sizeof(action));
action.output.type = ODPAT_OUTPUT;
action.output.port = ODPP_LOCAL;
- dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
+ dpif_execute(p->dpif, &action, 1, &payload);
}
rule = lookup_valid_rule(p, &flow);
}
}
\f
+/* Flow expiration. */
+
+struct expire_cbdata {
+ struct ofproto *ofproto;
+ int dp_max_idle;
+};
+
+static int ofproto_dp_max_idle(const struct ofproto *);
+static void ofproto_update_used(struct ofproto *);
+static void rule_expire(struct cls_rule *, void *cbdata);
+
+/* 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.
+ *
+ * Returns the number of milliseconds after which it should be called again. */
+static int
+ofproto_expire(struct ofproto *ofproto)
+{
+ struct expire_cbdata cbdata;
+
+ /* Update 'used' for each flow in the datapath. */
+ ofproto_update_used(ofproto);
+
+ /* Expire idle flows.
+ *
+ * A wildcarded flow is idle only when all of its subrules have expired due
+ * to becoming idle, so iterate through the exact-match flows first. */
+ cbdata.ofproto = ofproto;
+ cbdata.dp_max_idle = ofproto_dp_max_idle(ofproto);
+ classifier_for_each(&ofproto->cls, CLS_INC_EXACT, rule_expire, &cbdata);
+ classifier_for_each(&ofproto->cls, CLS_INC_WILD, rule_expire, &cbdata);
+
+ /* 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);
+ }
+
+ return MIN(cbdata.dp_max_idle, 1000);
+}
+
+/* Update 'used' member of each flow currently installed into the datapath. */
+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 cls_rule target;
+ struct rule *rule;
+ struct flow flow;
+
+ odp_flow_key_to_flow(&f->key, &flow);
+ cls_rule_from_flow(&flow, 0, UINT16_MAX, &target);
+
+ rule = rule_from_cls_rule(classifier_find_rule_exactly(&p->cls,
+ &target));
+
+ if (rule && rule->installed) {
+ update_time(p, rule, &f->stats);
+ rule_account(p, rule, 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
+ * flows 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)
+{
+ /*
+ * Idle time histogram.
+ *
+ * Most of the time a switch has a relatively small number of flows. 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 flows increases, the memory required to maintain
+ * statistics about them in userspace and in the kernel becomes
+ * significant. However, with a large number of flows 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 flows we can discard.
+ *
+ * The technique used to compute the idle time is to build a histogram with
+ * N_BUCKETS bucket whose width is BUCKET_WIDTH msecs each. Each flow 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 flows (but at least 1000 flows) are kept
+ * cached. At least the most-recently-used bucket of flows is kept, so
+ * actually an arbitrary number of flows 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 exact-match flows, in addition
+ * to the pass made by ofproto_update_used(), because the former function
+ * never looks at uninstallable flows.
+ */
+ enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
+ enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
+ int buckets[N_BUCKETS] = { 0 };
+ int total, bucket;
+ struct rule *rule;
+ long long int now;
+ int i;
+
+ total = classifier_count_exact(&ofproto->cls);
+ if (total <= 1000) {
+ return N_BUCKETS * BUCKET_WIDTH;
+ }
+
+ /* Build histogram. */
+ now = time_msec();
+ CLASSIFIER_FOR_EACH_EXACT_RULE (rule, cr, &ofproto->cls) {
+ long long int idle = now - rule->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);
+ }
+
+ return bucket * BUCKET_WIDTH;
+}
+
+static void
+rule_active_timeout(struct ofproto *ofproto, struct rule *rule)
+{
+ if (ofproto->netflow && !is_controller_rule(rule) &&
+ netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
+ struct ofexpired expired;
+ struct odp_flow odp_flow;
+
+ /* Get updated flow stats.
+ *
+ * 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 (rule->installed) {
+ odp_flow_key_from_flow(&odp_flow.key, &rule->cr.flow);
+ odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
+ dpif_flow_get(ofproto->dpif, &odp_flow);
+
+ if (odp_flow.stats.n_packets) {
+ update_time(ofproto, rule, &odp_flow.stats);
+ netflow_flow_update_flags(&rule->nf_flow,
+ odp_flow.stats.tcp_flags);
+ }
+ }
+
+ expired.flow = rule->cr.flow;
+ expired.packet_count = rule->packet_count +
+ odp_flow.stats.n_packets;
+ expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
+ expired.used = rule->used;
+
+ netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
+ }
+}
+
+/* If 'cls_rule' is an OpenFlow rule, that has expired according to OpenFlow
+ * rules, then delete it entirely.
+ *
+ * If 'cls_rule' is a subrule, that has not been used recently, remove it from
+ * the datapath and fold its statistics back into its super-rule.
+ *
+ * (This is a callback function for classifier_for_each().) */
+static void
+rule_expire(struct cls_rule *cls_rule, void *cbdata_)
+{
+ struct expire_cbdata *cbdata = cbdata_;
+ struct ofproto *ofproto = cbdata->ofproto;
+ struct rule *rule = rule_from_cls_rule(cls_rule);
+ long long int hard_expire, idle_expire, expire, now;
+
+ /* Calculate OpenFlow expiration times for 'rule'. */
+ hard_expire = (rule->hard_timeout
+ ? rule->created + rule->hard_timeout * 1000
+ : LLONG_MAX);
+ idle_expire = (rule->idle_timeout
+ && (rule->super || list_is_empty(&rule->list))
+ ? rule->used + rule->idle_timeout * 1000
+ : LLONG_MAX);
+ expire = MIN(hard_expire, idle_expire);
+
+ now = time_msec();
+ if (now < expire) {
+ /* 'rule' has not expired according to OpenFlow rules. */
+ if (!rule->cr.wc.wildcards) {
+ if (now >= rule->used + cbdata->dp_max_idle) {
+ /* This rule is idle, so drop it to free up resources. */
+ if (rule->super) {
+ /* It's not part of the OpenFlow flow table, so we can
+ * delete it entirely and fold its statistics into its
+ * super-rule. */
+ rule_remove(ofproto, rule);
+ } else {
+ /* It is part of the OpenFlow flow table, so we have to
+ * keep the rule but we can at least uninstall it from the
+ * datapath. */
+ rule_uninstall(ofproto, rule);
+ }
+ } else {
+ /* Send NetFlow active timeout if appropriate. */
+ rule_active_timeout(cbdata->ofproto, rule);
+ }
+ }
+ } else {
+ /* 'rule' has expired according to OpenFlow rules. */
+ COVERAGE_INC(ofproto_expired);
+
+ /* Update stats. (This is a no-op if the rule expired due to an idle
+ * timeout, because that only happens when the rule has no subrules
+ * left.) */
+ if (rule->cr.wc.wildcards) {
+ struct rule *subrule, *next;
+ LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) {
+ rule_remove(cbdata->ofproto, subrule);
+ }
+ } else {
+ rule_uninstall(cbdata->ofproto, rule);
+ }
+
+ /* Get rid of the rule. */
+ if (!rule_is_hidden(rule)) {
+ send_flow_removed(cbdata->ofproto, rule, now,
+ (now >= hard_expire
+ ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
+ }
+ rule_remove(cbdata->ofproto, rule);
+ }
+}
+\f
static void
revalidate_cb(struct cls_rule *sub_, void *cbdata_)
{
static bool
revalidate_rule(struct ofproto *p, struct rule *rule)
{
- const flow_t *flow = &rule->cr.flow;
+ const struct flow *flow = &rule->cr.flow;
COVERAGE_INC(ofproto_revalidate_rule);
if (rule->super) {
struct rule *super;
- super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
+ super = rule_from_cls_rule(classifier_lookup(&p->cls, flow,
+ CLS_INC_WILD));
if (!super) {
rule_remove(p, rule);
return false;
return buf;
}
-static void
-uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
-{
- assert(rule->installed);
- assert(!rule->cr.wc.wildcards);
-
- if (rule->super) {
- rule_remove(ofproto, rule);
- } else {
- rule_uninstall(ofproto, rule);
- }
-}
-
static void
send_flow_removed(struct ofproto *p, struct rule *rule,
long long int now, uint8_t reason)
struct ofconn *prev;
struct ofpbuf *buf = NULL;
+ if (!rule->send_flow_removed) {
+ return;
+ }
+
/* We limit the maximum number of queued flow expirations it by accounting
* them under the counter for replies. That works because preventing
* OpenFlow requests from being processed also prevents new flows from
prev = NULL;
LIST_FOR_EACH (ofconn, node, &p->all_conns) {
- if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)
+ if (rconn_is_connected(ofconn->rconn)
&& ofconn_receives_async_msgs(ofconn)) {
if (prev) {
queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
}
}
-
-static void
-expire_rule(struct cls_rule *cls_rule, void *p_)
-{
- struct ofproto *p = p_;
- struct rule *rule = rule_from_cls_rule(cls_rule);
- long long int hard_expire, idle_expire, expire, now;
-
- hard_expire = (rule->hard_timeout
- ? rule->created + rule->hard_timeout * 1000
- : LLONG_MAX);
- idle_expire = (rule->idle_timeout
- && (rule->super || list_is_empty(&rule->list))
- ? rule->used + rule->idle_timeout * 1000
- : LLONG_MAX);
- expire = MIN(hard_expire, idle_expire);
-
- now = time_msec();
- if (now < expire) {
- if (rule->installed && now >= rule->used + 5000) {
- uninstall_idle_flow(p, rule);
- } else if (!rule->cr.wc.wildcards) {
- active_timeout(p, rule);
- }
-
- return;
- }
-
- COVERAGE_INC(ofproto_expired);
-
- /* Update stats. This code will be a no-op if the rule expired
- * due to an idle timeout. */
- if (rule->cr.wc.wildcards) {
- struct rule *subrule, *next;
- LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) {
- rule_remove(p, subrule);
- }
- } else {
- rule_uninstall(p, rule);
- }
-
- if (!rule_is_hidden(rule)) {
- send_flow_removed(p, rule, now,
- (now >= hard_expire
- ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
- }
- rule_remove(p, rule);
-}
-
-static void
-active_timeout(struct ofproto *ofproto, struct rule *rule)
-{
- if (ofproto->netflow && !is_controller_rule(rule) &&
- netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
- struct ofexpired expired;
- struct odp_flow odp_flow;
-
- /* Get updated flow stats. */
- memset(&odp_flow, 0, sizeof odp_flow);
- if (rule->installed) {
- odp_flow.key = rule->cr.flow;
- odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
- dpif_flow_get(ofproto->dpif, &odp_flow);
-
- if (odp_flow.stats.n_packets) {
- update_time(ofproto, rule, &odp_flow.stats);
- netflow_flow_update_flags(&rule->nf_flow,
- odp_flow.stats.tcp_flags);
- }
- }
-
- expired.flow = rule->cr.flow;
- expired.packet_count = rule->packet_count +
- odp_flow.stats.n_packets;
- expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
- expired.used = rule->used;
-
- netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
-
- /* Schedule us to send the accumulated records once we have
- * collected all of them. */
- poll_immediate_wake();
- }
-}
-
-static void
-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 rule *rule;
-
- rule = rule_from_cls_rule(
- classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
- if (!rule || !rule->installed) {
- COVERAGE_INC(ofproto_unexpected_rule);
- dpif_flow_del(p->dpif, f);
- continue;
- }
-
- update_time(p, rule, &f->stats);
- rule_account(p, rule, f->stats.n_bytes);
- }
- free(flows);
-}
-
/* pinsched callback for sending 'packet' on 'ofconn'. */
static void
do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
}
\f
static bool
-default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
+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_)
{
out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
NULL);
if (out_port < 0) {
- add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
+ 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;
}
static const struct ofhooks default_ofhooks = {
- NULL,
default_normal_ofhook_cb,
NULL,
NULL
git://git.onelab.eu / sliver-openvswitch.git / blobdiff