static void bundle_del_port(struct ofport_dpif *);
static void bundle_run(struct ofbundle *);
static void bundle_wait(struct ofbundle *);
-static struct ofport_dpif *lookup_input_bundle(struct ofproto_dpif *,
- uint16_t in_port, bool warn);
+static struct ofbundle *lookup_input_bundle(struct ofproto_dpif *,
+ uint16_t in_port, bool warn);
+
+/* A controller may use OFPP_NONE as the ingress port to indicate that
+ * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
+ * when an input bundle is needed for validation (e.g., mirroring or
+ * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
+ * any 'port' structs, so care must be taken when dealing with it. */
+static struct ofbundle ofpp_none_bundle = {
+ .name = "OFPP_NONE",
+ .vlan_mode = PORT_VLAN_TRUNK
+};
static void stp_run(struct ofproto_dpif *ofproto);
static void stp_wait(struct ofproto_dpif *ofproto);
+static int set_stp_port(struct ofport *,
+ const struct ofproto_port_stp_settings *);
static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
const struct flow *,
const struct nlattr *odp_actions,
size_t actions_len,
- struct ofpbuf *packet);
+ struct ofpbuf *packet, bool clone);
static void facet_flush_stats(struct ofproto_dpif *, struct facet *);
const struct nlattr *key,
size_t key_len, ovs_be16 initial_tci);
static struct subfacet *subfacet_find(struct ofproto_dpif *,
- const struct nlattr *key, size_t key_len,
- const struct flow *);
+ const struct nlattr *key, size_t key_len);
static void subfacet_destroy(struct ofproto_dpif *, struct subfacet *);
static void subfacet_destroy__(struct ofproto_dpif *, struct subfacet *);
static void subfacet_reset_dp_stats(struct subfacet *,
};
struct ofproto_dpif {
+ struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
struct ofproto up;
struct dpif *dpif;
int max_ports;
* for debugging the asynchronous flow_mod implementation.) */
static bool clogged;
+/* All existing ofproto_dpif instances, indexed by ->up.name. */
+static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
+
static void ofproto_dpif_unixctl_init(void);
static struct ofproto_dpif *
ofproto->sflow = NULL;
ofproto->stp = NULL;
hmap_init(&ofproto->bundles);
- ofproto->ml = mac_learning_create();
+ ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
for (i = 0; i < MAX_MIRRORS; i++) {
ofproto->mirrors[i] = NULL;
}
hmap_init(&ofproto->vlandev_map);
hmap_init(&ofproto->realdev_vid_map);
+ hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
+ hash_string(ofproto->up.name, 0));
+
*n_tablesp = N_TABLES;
return 0;
}
struct classifier *table;
int i;
+ hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
complete_operations(ofproto);
OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
stp_set_max_age(ofproto->stp, s->max_age);
stp_set_forward_delay(ofproto->stp, s->fwd_delay);
} else {
+ struct ofport *ofport;
+
+ HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
+ set_stp_port(ofport, NULL);
+ }
+
stp_destroy(ofproto->stp);
ofproto->stp = NULL;
}
\f
/* Bundles. */
-/* Expires all MAC learning entries associated with 'port' and forces ofproto
- * to revalidate every flow. */
+/* Expires all MAC learning entries associated with 'bundle' and forces its
+ * ofproto to revalidate every flow.
+ *
+ * Normally MAC learning entries are removed only from the ofproto associated
+ * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
+ * are removed from every ofproto. When patch ports and SLB bonds are in use
+ * and a VM migration happens and the gratuitous ARPs are somehow lost, this
+ * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
+ * with the host from which it migrated. */
static void
-bundle_flush_macs(struct ofbundle *bundle)
+bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
{
struct ofproto_dpif *ofproto = bundle->ofproto;
struct mac_learning *ml = ofproto->ml;
ofproto->need_revalidate = true;
LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
if (mac->port.p == bundle) {
+ if (all_ofprotos) {
+ struct ofproto_dpif *o;
+
+ HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
+ if (o != ofproto) {
+ struct mac_entry *e;
+
+ e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
+ NULL);
+ if (e) {
+ tag_set_add(&o->revalidate_set, e->tag);
+ mac_learning_expire(o->ml, e);
+ }
+ }
+ }
+ }
+
mac_learning_expire(ml, mac);
}
}
bundle_del_port(port);
}
- bundle_flush_macs(bundle);
+ bundle_flush_macs(bundle, true);
hmap_remove(&ofproto->bundles, &bundle->hmap_node);
free(bundle->name);
free(bundle->trunks);
/* If we changed something that would affect MAC learning, un-learn
* everything on this port and force flow revalidation. */
if (need_flush) {
- bundle_flush_macs(bundle);
+ bundle_flush_macs(bundle, false);
}
return 0;
/* Revalidate cached flows whenever forward_bpdu option changes. */
ofproto->need_revalidate = true;
}
+
+static void
+set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
+{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+ mac_learning_set_idle_time(ofproto->ml, idle_time);
+}
\f
/* Ports. */
miss->initial_tci);
LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
+ struct dpif_flow_stats stats;
+
list_remove(&packet->list_node);
ofproto->n_matches++;
if (!facet->may_install || !subfacet->actions) {
subfacet_make_actions(ofproto, subfacet, packet);
}
+
+ /* Credit statistics to subfacet for this packet. We must do this now
+ * because execute_controller_action() below may destroy 'packet'. */
+ dpif_flow_stats_extract(&facet->flow, packet, &stats);
+ subfacet_update_stats(ofproto, subfacet, &stats);
+
if (!execute_controller_action(ofproto, &facet->flow,
subfacet->actions,
- subfacet->actions_len, packet)) {
+ subfacet->actions_len, packet, true)) {
struct flow_miss_op *op = &ops[(*n_ops)++];
struct dpif_execute *execute = &op->dpif_op.execute;
+ if (flow->vlan_tci != subfacet->initial_tci) {
+ /* This packet was received on a VLAN splinter port. We added
+ * a VLAN to the packet to make the packet resemble the flow,
+ * but the actions were composed assuming that the packet
+ * contained no VLAN. So, we must remove the VLAN header from
+ * the packet before trying to execute the actions. */
+ eth_pop_vlan(packet);
+ }
+
op->subfacet = subfacet;
execute->type = DPIF_OP_EXECUTE;
execute->key = miss->key;
}
}
+/* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
+ * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
+ * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
+ * what a flow key should contain.
+ *
+ * This function also includes some logic to help make VLAN splinters
+ * transparent to the rest of the upcall processing logic. In particular, if
+ * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
+ * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
+ * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
+ *
+ * Sets '*initial_tci' to the VLAN TCI with which the packet was really
+ * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
+ * (This differs from the value returned in flow->vlan_tci only for packets
+ * received on VLAN splinters.)
+ */
static enum odp_key_fitness
ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
const struct nlattr *key, size_t key_len,
- struct flow *flow, ovs_be16 *initial_tci)
+ struct flow *flow, ovs_be16 *initial_tci,
+ struct ofpbuf *packet)
{
enum odp_key_fitness fitness;
uint16_t realdev;
* with the VLAN device's VLAN ID. */
flow->in_port = realdev;
flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
+ if (packet) {
+ /* Make the packet resemble the flow, so that it gets sent to an
+ * OpenFlow controller properly, so that it looks correct for
+ * sFlow, and so that flow_extract() will get the correct vlan_tci
+ * if it is called on 'packet'.
+ *
+ * The allocated space inside 'packet' probably also contains
+ * 'key', that is, both 'packet' and 'key' are probably part of a
+ * struct dpif_upcall (see the large comment on that structure
+ * definition), so pushing data on 'packet' is in general not a
+ * good idea since it could overwrite 'key' or free it as a side
+ * effect. However, it's OK in this special case because we know
+ * that 'packet' is inside a Netlink attribute: pushing 4 bytes
+ * will just overwrite the 4-byte "struct nlattr", which is fine
+ * since we don't need that header anymore. */
+ eth_push_vlan(packet, flow->vlan_tci);
+ }
/* Let the caller know that we can't reproduce 'key' from 'flow'. */
if (fitness == ODP_FIT_PERFECT) {
* then set 'flow''s header pointers. */
fitness = ofproto_dpif_extract_flow_key(ofproto,
upcall->key, upcall->key_len,
- &flow, &initial_tci);
+ &flow, &initial_tci,
+ upcall->packet);
if (fitness == ODP_FIT_ERROR) {
ofpbuf_delete(upcall->packet);
continue;
fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
upcall->key_len, &flow,
- &initial_tci);
+ &initial_tci, upcall->packet);
if (fitness == ODP_FIT_ERROR) {
ofpbuf_delete(upcall->packet);
return;
dpif_flow_dump_start(&dump, p->dpif);
while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
- enum odp_key_fitness fitness;
struct subfacet *subfacet;
- struct flow flow;
-
- fitness = odp_flow_key_to_flow(key, key_len, &flow);
- if (fitness == ODP_FIT_ERROR) {
- continue;
- }
- subfacet = subfacet_find(p, key, key_len, &flow);
+ subfacet = subfacet_find(p, key, key_len);
if (subfacet && subfacet->installed) {
struct facet *facet = subfacet->facet;
facet_account(p, facet);
facet_push_stats(facet);
} else {
+ if (!VLOG_DROP_WARN(&rl)) {
+ struct ds s;
+
+ ds_init(&s);
+ odp_flow_key_format(key, key_len, &s);
+ VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
+ ds_destroy(&s);
+ }
+
+ COVERAGE_INC(facet_unexpected);
/* There's a flow in the datapath that we know nothing about, or a
* flow that shouldn't be installed but was anyway. Delete it. */
- COVERAGE_INC(facet_unexpected);
dpif_flow_del(p->dpif, key, key_len, NULL);
}
}
free(facet);
}
+/* If the 'actions_len' bytes of actions in 'odp_actions' are just a single
+ * OVS_ACTION_ATTR_USERSPACE action, executes it internally and returns true.
+ * Otherwise, returns false without doing anything.
+ *
+ * If 'clone' is true, the caller always retains ownership of 'packet'.
+ * Otherwise, ownership is transferred to this function if it returns true. */
static bool
execute_controller_action(struct ofproto_dpif *ofproto,
const struct flow *flow,
const struct nlattr *odp_actions, size_t actions_len,
- struct ofpbuf *packet)
+ struct ofpbuf *packet, bool clone)
{
if (actions_len
&& odp_actions->nla_type == OVS_ACTION_ATTR_USERSPACE
nla = nl_attr_find_nested(odp_actions, OVS_USERSPACE_ATTR_USERDATA);
send_packet_in_action(ofproto, packet, nl_attr_get_u64(nla), flow,
- false);
+ clone);
return true;
} else {
return false;
int error;
if (execute_controller_action(ofproto, flow, odp_actions, actions_len,
- packet)) {
+ packet, false)) {
return true;
}
{
struct subfacet *subfacet, *next_subfacet;
+ assert(!list_is_empty(&facet->subfacets));
+
+ /* First uninstall all of the subfacets to get final statistics. */
+ LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
+ subfacet_uninstall(ofproto, subfacet);
+ }
+
+ /* Flush the final stats to the rule.
+ *
+ * This might require us to have at least one subfacet around so that we
+ * can use its actions for accounting in facet_account(), which is why we
+ * have uninstalled but not yet destroyed the subfacets. */
+ facet_flush_stats(ofproto, facet);
+
+ /* Now we're really all done so destroy everything. */
LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
&facet->subfacets) {
subfacet_destroy__(ofproto, subfacet);
}
-
- facet_flush_stats(ofproto, facet);
hmap_remove(&ofproto->facets, &facet->hmap_node);
list_remove(&facet->list_node);
facet_free(facet);
* 'flow'. Returns the subfacet if one exists, otherwise NULL. */
static struct subfacet *
subfacet_find(struct ofproto_dpif *ofproto,
- const struct nlattr *key, size_t key_len,
- const struct flow *flow)
+ const struct nlattr *key, size_t key_len)
{
uint32_t key_hash = odp_flow_key_hash(key, key_len);
+ enum odp_key_fitness fitness;
+ struct flow flow;
+
+ fitness = odp_flow_key_to_flow(key, key_len, &flow);
+ if (fitness == ODP_FIT_ERROR) {
+ return NULL;
+ }
- return subfacet_find__(ofproto, key, key_len, key_hash, flow);
+ return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
}
/* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
{
struct facet *facet = subfacet->facet;
- subfacet_destroy__(ofproto, subfacet);
- if (list_is_empty(&facet->subfacets)) {
+ if (list_is_singleton(&facet->subfacets)) {
+ /* facet_remove() needs at least one subfacet (it will remove it). */
facet_remove(ofproto, facet);
+ } else {
+ subfacet_destroy__(ofproto, subfacet);
}
}
case OFPP_CONTROLLER:
compose_controller_action(ctx, max_len);
break;
- case OFPP_LOCAL:
- compose_output_action(ctx, OFPP_LOCAL);
- break;
case OFPP_NONE:
break;
+ case OFPP_LOCAL:
default:
if (port != ctx->flow.in_port) {
compose_output_action(ctx, port);
static bool
input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
{
+ /* Allow any VID on the OFPP_NONE port. */
+ if (in_bundle == &ofpp_none_bundle) {
+ return true;
+ }
+
switch (in_bundle->vlan_mode) {
case PORT_VLAN_ACCESS:
if (vid) {
{
struct ofproto_dpif *ofproto = ctx->ofproto;
mirror_mask_t mirrors;
- struct ofport_dpif *in_port;
struct ofbundle *in_bundle;
uint16_t vlan;
uint16_t vid;
const struct nlattr *a;
size_t left;
- /* Obtain in_port from orig_flow.in_port.
- *
- * lookup_input_bundle() also ensures that in_port belongs to a bundle. */
- in_port = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
- ctx->packet != NULL);
- if (!in_port) {
+ in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
+ ctx->packet != NULL);
+ if (!in_bundle) {
return;
}
- in_bundle = in_port->bundle;
mirrors = in_bundle->src_mirrors;
/* Drop frames on bundles reserved for mirroring. */
}
ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
- mirrors |= ofport ? ofport->bundle->dst_mirrors : 0;
+ if (ofport && ofport->bundle) {
+ mirrors |= ofport->bundle->dst_mirrors;
+ }
}
if (!mirrors) {
{
struct mac_entry *mac;
+ /* Don't learn the OFPP_NONE port. */
+ if (in_bundle == &ofpp_none_bundle) {
+ return;
+ }
+
if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
return;
}
}
}
-static struct ofport_dpif *
+static struct ofbundle *
lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
{
struct ofport_dpif *ofport;
+ /* Special-case OFPP_NONE, which a controller may use as the ingress
+ * port for traffic that it is sourcing. */
+ if (in_port == OFPP_NONE) {
+ return &ofpp_none_bundle;
+ }
+
/* Find the port and bundle for the received packet. */
ofport = get_ofp_port(ofproto, in_port);
if (ofport && ofport->bundle) {
- return ofport;
+ return ofport->bundle;
}
/* Odd. A few possible reasons here:
ctx->has_normal = true;
- /* Obtain in_port from ctx->flow.in_port.
- *
- * lookup_input_bundle() also ensures that in_port belongs to a bundle. */
- in_port = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
+ in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
ctx->packet != NULL);
- if (!in_port) {
+ if (!in_bundle) {
return;
}
- in_bundle = in_port->bundle;
+
+ /* We know 'in_port' exists unless it is "ofpp_none_bundle",
+ * since lookup_input_bundle() succeeded. */
+ in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
/* Drop malformed frames. */
if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
vlan = input_vid_to_vlan(in_bundle, vid);
/* Check other admissibility requirements. */
- if (!is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
+ if (in_port &&
+ !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
return;
}
static struct ofproto_dpif *
ofproto_dpif_lookup(const char *name)
{
- struct ofproto *ofproto = ofproto_lookup(name);
- return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class
- ? ofproto_dpif_cast(ofproto)
- : NULL);
+ struct ofproto_dpif *ofproto;
+
+ HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
+ hash_string(name, 0), &all_ofproto_dpifs) {
+ if (!strcmp(ofproto->up.name, name)) {
+ return ofproto;
+ }
+ }
+ return NULL;
}
static void
ofproto_unixctl_fdb_flush(struct unixctl_conn *conn,
- const char *args, void *aux OVS_UNUSED)
+ const char *args, void *aux OVS_UNUSED)
{
- const struct ofproto_dpif *ofproto;
+ struct ofproto_dpif *ofproto;
- ofproto = ofproto_dpif_lookup(args);
- if (!ofproto) {
- unixctl_command_reply(conn, 501, "no such bridge");
- return;
+ if (args[0] != '\0') {
+ ofproto = ofproto_dpif_lookup(args);
+ if (!ofproto) {
+ unixctl_command_reply(conn, 501, "no such bridge");
+ return;
+ }
+ mac_learning_flush(ofproto->ml);
+ ofproto->need_revalidate = true;
+ } else {
+ HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
+ mac_learning_flush(ofproto->ml);
+ ofproto->need_revalidate = true;
+ }
}
- mac_learning_flush(ofproto->ml);
unixctl_command_reply(conn, 200, "table successfully flushed");
}
struct ofbundle *bundle = e->port.p;
ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
ofbundle_get_a_port(bundle)->odp_port,
- e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
+ e->vlan, ETH_ADDR_ARGS(e->mac),
+ mac_entry_age(ofproto->ml, e));
}
unixctl_command_reply(conn, 200, ds_cstr(&ds));
ds_destroy(&ds);
/* Convert odp_key to flow. */
error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
odp_key.size, &flow,
- &initial_tci);
+ &initial_tci, NULL);
if (error == ODP_FIT_ERROR) {
unixctl_command_reply(conn, 501, "Invalid flow");
goto exit;
unixctl_command_register("ofproto/trace",
"bridge {tun_id in_port packet | odp_flow [-generate]}",
ofproto_unixctl_trace, NULL);
- unixctl_command_register("fdb/flush", "bridge", ofproto_unixctl_fdb_flush,
- NULL);
+ unixctl_command_register("fdb/flush", "[bridge]",
+ ofproto_unixctl_fdb_flush, NULL);
unixctl_command_register("fdb/show", "bridge", ofproto_unixctl_fdb_show,
NULL);
unixctl_command_register("ofproto/clog", "", ofproto_dpif_clog, NULL);
set_flood_vlans,
is_mirror_output_bundle,
forward_bpdu_changed,
+ set_mac_idle_time,
set_realdev,
};