long long int prev_used; /* Used time from last stats push. */
/* Accounting. */
- uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
uint16_t tcp_flags; /* TCP flags seen for this 'rule'. */
struct xlate_out xout;
struct dpif_flow_stats *, bool may_learn);
static void facet_push_stats(struct facet *, bool may_learn);
static void facet_learn(struct facet *);
-static void facet_account(struct facet *);
static void push_all_stats(void);
static bool facet_is_controller_flow(struct facet *);
static int set_bfd(struct ofport *, const struct smap *);
static int set_cfm(struct ofport *, const struct cfm_settings *);
static void ofport_update_peer(struct ofport_dpif *);
-static void run_fast_rl(void);
-static int run_fast(struct ofproto *);
struct dpif_completion {
struct list list_node;
* performance in new situations. */
unsigned max_n_subfacet; /* Maximum number of flows */
unsigned avg_n_subfacet; /* Average number of flows. */
- long long int avg_subfacet_life; /* Average life span of subfacets. */
-
- /* Number of upcall handling threads. */
- unsigned int n_handler_threads;
};
/* All existing ofproto_backer instances, indexed by ofproto->up.type. */
dpif_run(backer->dpif);
+ handle_upcalls(backer);
+
/* The most natural place to push facet statistics is when they're pulled
* from the datapath. However, when there are many flows in the datapath,
* this expensive operation can occur so frequently, that it reduces our
error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
if (error) {
- udpif_recv_set(backer->udpif, 0, false);
VLOG_ERR("Failed to enable receiving packets in dpif.");
return error;
}
- udpif_recv_set(backer->udpif, n_handler_threads,
- backer->recv_set_enable);
dpif_flow_flush(backer->dpif);
backer->need_revalidate = REV_RECONFIGURE;
}
- /* If the n_handler_threads is reconfigured, call udpif_recv_set()
- * to reset the handler threads. */
- if (backer->n_handler_threads != n_handler_threads) {
- udpif_recv_set(backer->udpif, n_handler_threads,
- backer->recv_set_enable);
- backer->n_handler_threads = n_handler_threads;
+ if (backer->recv_set_enable) {
+ udpif_set_threads(backer->udpif, n_handlers);
}
if (backer->need_revalidate) {
ofproto->no_packet_in_rule, ofproto->ml,
ofproto->stp, ofproto->mbridge,
ofproto->sflow, ofproto->ipfix,
- ofproto->up.frag_handling,
+ ofproto->netflow, ofproto->up.frag_handling,
ofproto->up.forward_bpdu,
- connmgr_has_in_band(ofproto->up.connmgr),
- ofproto->netflow != NULL);
+ connmgr_has_in_band(ofproto->up.connmgr));
HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
xlate_bundle_set(ofproto, bundle, bundle->name,
ovs_rwlock_unlock(&ofproto->facets.rwlock);
CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
facet_revalidate(facet);
- run_fast_rl();
}
}
}
}
-static int
-dpif_backer_run_fast(struct dpif_backer *backer)
-{
- handle_upcalls(backer);
-
- return 0;
-}
-
-static int
-type_run_fast(const char *type)
-{
- struct dpif_backer *backer;
-
- backer = shash_find_data(&all_dpif_backers, type);
- if (!backer) {
- /* This is not necessarily a problem, since backers are only
- * created on demand. */
- return 0;
- }
-
- return dpif_backer_run_fast(backer);
-}
-
-static void
-run_fast_rl(void)
-{
- static long long int port_rl = LLONG_MIN;
-
- if (time_msec() >= port_rl) {
- struct ofproto_dpif *ofproto;
-
- HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
- run_fast(&ofproto->up);
- }
- port_rl = time_msec() + 200;
- }
-}
-
static void
type_wait(const char *type)
{
close_dpif_backer(backer);
return error;
}
- udpif_recv_set(backer->udpif, n_handler_threads,
- backer->recv_set_enable);
- backer->n_handler_threads = n_handler_threads;
+
+ if (backer->recv_set_enable) {
+ udpif_set_threads(backer->udpif, n_handlers);
+ }
backer->max_n_subfacet = 0;
backer->avg_n_subfacet = 0;
- backer->avg_subfacet_life = 0;
return error;
}
rulep)) {
rule_dpif_unref(*rulep);
} else {
- NOT_REACHED();
+ OVS_NOT_REACHED();
}
return 0;
close_dpif_backer(ofproto->backer);
}
-static int
-run_fast(struct ofproto *ofproto_)
-{
- struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
- struct ofproto_packet_in *pin, *next_pin;
- struct list pins;
-
- /* Do not perform any periodic activity required by 'ofproto' while
- * waiting for flow restore to complete. */
- if (ofproto_get_flow_restore_wait()) {
- return 0;
- }
-
- guarded_list_pop_all(&ofproto->pins, &pins);
- LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
- connmgr_send_packet_in(ofproto->up.connmgr, pin);
- list_remove(&pin->list_node);
- free(CONST_CAST(void *, pin->up.packet));
- free(pin);
- }
-
- return 0;
-}
-
static int
run(struct ofproto *ofproto_)
{
struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
uint64_t new_seq;
- int error;
if (mbridge_need_revalidate(ofproto->mbridge)) {
ofproto->backer->need_revalidate = REV_RECONFIGURE;
ovs_rwlock_unlock(&ofproto->ml->rwlock);
}
- /* Do not perform any periodic activity below required by 'ofproto' while
+ /* Do not perform any periodic activity required by 'ofproto' while
* waiting for flow restore to complete. */
- if (ofproto_get_flow_restore_wait()) {
- return 0;
- }
+ if (!ofproto_get_flow_restore_wait()) {
+ struct ofproto_packet_in *pin, *next_pin;
+ struct list pins;
- error = run_fast(ofproto_);
- if (error) {
- return error;
+ guarded_list_pop_all(&ofproto->pins, &pins);
+ LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
+ connmgr_send_packet_in(ofproto->up.connmgr, pin);
+ list_remove(&pin->list_node);
+ free(CONST_CAST(void *, pin->up.packet));
+ free(pin);
+ }
}
if (ofproto->netflow) {
simap_increase(usage, "subfacets", n_subfacets);
}
+static void
+type_get_memory_usage(const char *type, struct simap *usage)
+{
+ struct dpif_backer *backer;
+
+ backer = shash_find_data(&all_dpif_backers, type);
+ if (backer) {
+ udpif_get_memory_usage(backer->udpif, usage);
+ }
+}
+
static void
flush(struct ofproto *ofproto_)
{
break;
default:
- NOT_REACHED();
+ OVS_NOT_REACHED();
}
if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
free(bundle->trunks);
return ofport ? ofport_dpif_cast(ofport) : NULL;
}
-static struct ofport_dpif *
-get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
-{
- struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
- return port && &ofproto->up == port->up.ofproto ? port : NULL;
-}
-
static void
ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
struct ofproto_port *ofproto_port,
update_stats(backer);
n_subfacets = hmap_count(&backer->subfacets);
- if (n_subfacets) {
- struct subfacet *subfacet;
- long long int total, now;
-
- total = 0;
- now = time_msec();
- HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
- total += now - subfacet->created;
- }
- backer->avg_subfacet_life += total / n_subfacets;
- }
- backer->avg_subfacet_life /= 2;
-
backer->avg_n_subfacet += n_subfacets;
backer->avg_n_subfacet /= 2;
subfacet->dp_byte_count = stats->n_bytes;
subfacet_update_stats(subfacet, &diff);
- if (facet->accounted_bytes < facet->byte_count) {
+ if (diff.n_packets) {
facet_learn(facet);
- facet_account(facet);
- facet->accounted_bytes = facet->byte_count;
}
}
delete_unexpected_flow(backer, key, key_len);
break;
}
- run_fast_rl();
}
dpif_flow_dump_done(&dump);
}
classifier_insert(&ofproto->facets, &facet->cr);
ovs_rwlock_unlock(&ofproto->facets.rwlock);
- if (ofproto->netflow && !facet_is_controller_flow(facet)) {
- netflow_flow_update(ofproto->netflow, &facet->flow,
- facet->xout.nf_output_iface, &miss->stats);
- }
-
return facet;
}
facet_push_stats(facet, true);
}
-static void
-facet_account(struct facet *facet)
-{
- const struct nlattr *a;
- unsigned int left;
- ovs_be16 vlan_tci;
- uint64_t n_bytes;
-
- if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
- return;
- }
- n_bytes = facet->byte_count - facet->accounted_bytes;
-
- /* This loop feeds byte counters to bond_account() for rebalancing to use
- * as a basis. We also need to track the actual VLAN on which the packet
- * is going to be sent to ensure that it matches the one passed to
- * bond_choose_output_slave(). (Otherwise, we will account to the wrong
- * hash bucket.)
- *
- * We use the actions from an arbitrary subfacet because they should all
- * be equally valid for our purpose. */
- vlan_tci = facet->flow.vlan_tci;
- NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
- facet->xout.odp_actions.size) {
- const struct ovs_action_push_vlan *vlan;
- struct ofport_dpif *port;
-
- switch (nl_attr_type(a)) {
- case OVS_ACTION_ATTR_OUTPUT:
- port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
- if (port && port->bundle && port->bundle->bond) {
- bond_account(port->bundle->bond, &facet->flow,
- vlan_tci_to_vid(vlan_tci), n_bytes);
- }
- break;
-
- case OVS_ACTION_ATTR_POP_VLAN:
- vlan_tci = htons(0);
- break;
-
- case OVS_ACTION_ATTR_PUSH_VLAN:
- vlan = nl_attr_get(a);
- vlan_tci = vlan->vlan_tci;
- break;
- }
- }
-}
-
/* 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). */
}
facet_push_stats(facet, false);
- if (facet->accounted_bytes < facet->byte_count) {
- facet_account(facet);
- facet->accounted_bytes = facet->byte_count;
- }
if (ofproto->netflow && !facet_is_controller_flow(facet)) {
netflow_expire(ofproto->netflow, &facet->flow);
error = xlate_receive(ofproto->backer, NULL, subfacet->key,
subfacet->key_len, &recv_flow, NULL,
- &recv_ofproto, NULL);
+ &recv_ofproto, NULL, NULL, NULL, NULL);
if (error
|| recv_ofproto != ofproto
|| facet != facet_find(ofproto, &recv_flow)) {
facet->byte_count = 0;
facet->prev_packet_count = 0;
facet->prev_byte_count = 0;
- facet->accounted_bytes = 0;
}
static void
flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
struct dpif_flow_stats *stats, bool may_learn)
{
- struct ofport_dpif *in_port;
struct xlate_in xin;
- in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
- if (in_port && in_port->is_tunnel) {
- netdev_vport_inc_rx(in_port->up.netdev, stats);
- if (in_port->bfd) {
- bfd_account_rx(in_port->bfd, stats);
- }
- }
-
xlate_in_init(&xin, ofproto, flow, NULL, stats->tcp_flags, NULL);
xin.resubmit_stats = stats;
xin.may_learn = may_learn;
facet->prev_packet_count = facet->packet_count;
facet->prev_byte_count = facet->byte_count;
facet->prev_used = facet->used;
-
- if (facet->ofproto->netflow && !facet_is_controller_flow(facet)) {
- netflow_flow_update(facet->ofproto->netflow, &facet->flow,
- facet->xout.nf_output_iface, &stats);
- }
- mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
- stats.n_packets, stats.n_bytes);
flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
}
}
static void
-push_all_stats__(bool run_fast)
+push_all_stats(void)
{
static long long int rl = LLONG_MIN;
struct ofproto_dpif *ofproto;
cls_cursor_init(&cursor, &ofproto->facets, NULL);
CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
facet_push_stats(facet, false);
- if (run_fast) {
- run_fast_rl();
- }
}
ovs_rwlock_unlock(&ofproto->facets.rwlock);
}
rl = time_msec() + 100;
}
-static void
-push_all_stats(void)
-{
- push_all_stats__(true);
-}
-
void
rule_dpif_credit_stats(struct rule_dpif *rule,
const struct dpif_flow_stats *stats)
subfacet_reset_dp_stats(subfacets[i], &stats[i]);
subfacets[i]->path = SF_NOT_INSTALLED;
subfacet_destroy(subfacets[i]);
- run_fast_rl();
}
}
{
struct rule_dpif *rule = rule_dpif_cast(rule_);
- /* push_all_stats() can handle flow misses which, when using the learn
- * action, can cause rules to be added and deleted. This can corrupt our
- * caller's datastructures which assume that rule_get_stats() doesn't have
- * an impact on the flow table. To be safe, we disable miss handling. */
- push_all_stats__(false);
+ push_all_stats();
/* Start from historical data for 'rule' itself that are no longer tracked
* in facets. This counts, for example, facets that have expired. */
* - bridge br_flow [-generate | packet]
*
* On success, initializes '*ofprotop' and 'flow' and returns NULL. On failure
- * returns a nonnull error message. */
-static const char *
+ * returns a nonnull malloced error message. */
+static char * WARN_UNUSED_RESULT
parse_flow_and_packet(int argc, const char *argv[],
struct ofproto_dpif **ofprotop, struct flow *flow,
struct ofpbuf **packetp)
{
const struct dpif_backer *backer = NULL;
const char *error = NULL;
+ char *m_err = NULL;
struct simap port_names = SIMAP_INITIALIZER(&port_names);
struct ofpbuf *packet;
struct ofpbuf odp_key;
/* The 3-argument form must end in "-generate' or a hex string. */
goto exit;
}
+ error = NULL;
}
/* odp_flow can have its in_port specified as a name instead of port no.
/* Parse the flow and determine whether a datapath or
* bridge is specified. If function odp_flow_key_from_string()
* returns 0, the flow is a odp_flow. If function
- * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
+ * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
if (!odp_flow_from_string(argv[argc - 1], &port_names,
&odp_key, &odp_mask)) {
if (!backer) {
}
if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, flow,
- NULL, ofprotop, NULL)) {
+ NULL, ofprotop, NULL, NULL, NULL, NULL)) {
error = "Invalid datapath flow";
goto exit;
}
- } else if (!parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL)) {
- if (argc != 3) {
- error = "Must specify bridge name";
- goto exit;
- }
+ } else {
+ char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
- *ofprotop = ofproto_dpif_lookup(argv[1]);
- if (!*ofprotop) {
- error = "Unknown bridge name";
+ if (err) {
+ m_err = xasprintf("Bad flow syntax: %s", err);
+ free(err);
goto exit;
+ } else {
+ if (argc != 3) {
+ error = "Must specify bridge name";
+ goto exit;
+ }
+
+ *ofprotop = ofproto_dpif_lookup(argv[1]);
+ if (!*ofprotop) {
+ error = "Unknown bridge name";
+ goto exit;
+ }
}
- } else {
- error = "Bad flow syntax";
- goto exit;
}
/* Generate a packet, if requested. */
}
}
- error = NULL;
-
exit:
- if (error) {
+ if (error && !m_err) {
+ m_err = xstrdup(error);
+ }
+ if (m_err) {
ofpbuf_delete(packet);
packet = NULL;
}
ofpbuf_uninit(&odp_key);
ofpbuf_uninit(&odp_mask);
simap_destroy(&port_names);
- return error;
+ return m_err;
}
static void
{
struct ofproto_dpif *ofproto;
struct ofpbuf *packet;
- const char *error;
+ char *error;
struct flow flow;
error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
ofpbuf_delete(packet);
} else {
unixctl_command_reply_error(conn, error);
+ free(error);
}
}
/* Three kinds of error return values! */
enum ofperr retval;
- const char *error;
- char *rw_error;
+ char *error;
packet = NULL;
ds_init(&result);
ofpbuf_init(&ofpacts, 0);
/* Parse actions. */
- rw_error = parse_ofpacts(argv[--argc], &ofpacts, &usable_protocols);
- if (rw_error) {
- unixctl_command_reply_error(conn, rw_error);
- free(rw_error);
+ error = parse_ofpacts(argv[--argc], &ofpacts, &usable_protocols);
+ if (error) {
+ unixctl_command_reply_error(conn, error);
+ free(error);
goto exit;
}
error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
if (error) {
unixctl_command_reply_error(conn, error);
+ free(error);
goto exit;
}
ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
dpif_name(backer->dpif), n_hit, n_missed);
- ds_put_format(ds, "\tflows: cur: %"PRIuSIZE", avg: %u, max: %u,"
- " life span: %lldms\n", hmap_count(&backer->subfacets),
- backer->avg_n_subfacet, backer->max_n_subfacet,
- backer->avg_subfacet_life);
+
+ ds_put_format(ds, "\tflows: cur: %"PRIuSIZE", avg: %u, max: %u\n",
+ hmap_count(&backer->subfacets), backer->avg_n_subfacet,
+ backer->max_n_subfacet);
shash_init(&ofproto_shash);
ofprotos = get_ofprotos(&ofproto_shash);
unixctl_command_reply(conn, "megaflows enabled");
}
+static bool
+ofproto_dpif_contains_flow(const struct ofproto_dpif *ofproto,
+ const struct nlattr *key, size_t key_len)
+{
+ enum odp_key_fitness fitness;
+ struct ofproto_dpif *ofp;
+ struct flow flow;
+
+ xlate_receive(ofproto->backer, NULL, key, key_len, &flow, &fitness, &ofp,
+ NULL, NULL, NULL, NULL);
+ return ofp == ofproto;
+}
+
static void
ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
int argc OVS_UNUSED, const char *argv[],
void *aux OVS_UNUSED)
{
struct ds ds = DS_EMPTY_INITIALIZER;
+ const struct dpif_flow_stats *stats;
const struct ofproto_dpif *ofproto;
- struct subfacet *subfacet;
+ struct dpif_flow_dump flow_dump;
+ const struct nlattr *actions;
+ const struct nlattr *mask;
+ const struct nlattr *key;
+ size_t actions_len;
+ size_t mask_len;
+ size_t key_len;
ofproto = ofproto_dpif_lookup(argv[1]);
if (!ofproto) {
return;
}
- update_stats(ofproto->backer);
-
- HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
- struct facet *facet = subfacet->facet;
- struct odputil_keybuf maskbuf;
- struct ofpbuf mask;
-
- if (facet->ofproto != ofproto) {
+ ds_init(&ds);
+ dpif_flow_dump_start(&flow_dump, ofproto->backer->dpif);
+ while (dpif_flow_dump_next(&flow_dump, &key, &key_len, &mask, &mask_len,
+ &actions, &actions_len, &stats)) {
+ if (!ofproto_dpif_contains_flow(ofproto, key, key_len)) {
continue;
}
- ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
- if (enable_megaflows) {
- odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
- &facet->flow, UINT32_MAX);
- }
-
- odp_flow_format(subfacet->key, subfacet->key_len,
- mask.data, mask.size, NULL, &ds, false);
-
- ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
- subfacet->dp_packet_count, subfacet->dp_byte_count);
- if (subfacet->used) {
- ds_put_format(&ds, "%.3fs",
- (time_msec() - subfacet->used) / 1000.0);
- } else {
- ds_put_format(&ds, "never");
- }
- if (subfacet->facet->tcp_flags) {
- ds_put_cstr(&ds, ", flags:");
- packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
- }
-
+ odp_flow_format(key, key_len, mask, mask_len, NULL, &ds, false);
+ ds_put_cstr(&ds, ", ");
+ dpif_flow_stats_format(stats, &ds);
ds_put_cstr(&ds, ", actions:");
- if (facet->xout.slow) {
- uint64_t slow_path_stub[128 / 8];
- const struct nlattr *actions;
- size_t actions_len;
-
- compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
- slow_path_stub, sizeof slow_path_stub,
- &actions, &actions_len);
- format_odp_actions(&ds, actions, actions_len);
- } else {
- format_odp_actions(&ds, facet->xout.odp_actions.data,
- facet->xout.odp_actions.size);
- }
+ format_odp_actions(&ds, actions, actions_len);
ds_put_char(&ds, '\n');
}
- unixctl_command_reply(conn, ds_cstr(&ds));
- ds_destroy(&ds);
-}
-
-static void
-ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
- int argc OVS_UNUSED, const char *argv[],
- void *aux OVS_UNUSED)
-{
- struct ds ds = DS_EMPTY_INITIALIZER;
- struct ofproto_dpif *ofproto;
-
- ofproto = ofproto_dpif_lookup(argv[1]);
- if (!ofproto) {
- unixctl_command_reply_error(conn, "no such bridge");
- return;
+ if (dpif_flow_dump_done(&flow_dump)) {
+ ds_clear(&ds);
+ ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
+ unixctl_command_reply_error(conn, ds_cstr(&ds));
+ } else {
+ unixctl_command_reply(conn, ds_cstr(&ds));
}
-
- flush(&ofproto->up);
-
- unixctl_command_reply(conn, ds_cstr(&ds));
ds_destroy(&ds);
}
NULL);
unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
ofproto_unixctl_dpif_dump_flows, NULL);
- unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
- ofproto_unixctl_dpif_del_flows, NULL);
unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
ofproto_unixctl_dpif_dump_megaflows, NULL);
unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
del,
port_open_type,
type_run,
- type_run_fast,
type_wait,
alloc,
construct,
destruct,
dealloc,
run,
- run_fast,
wait,
get_memory_usage,
+ type_get_memory_usage,
flush,
get_features,
get_tables,