#include "autopath.h"
#include "bond.h"
+#include "bundle.h"
#include "byte-order.h"
#include "connmgr.h"
#include "coverage.h"
struct cfm *cfm; /* Connectivity Fault Management, if any. */
tag_type tag; /* Tag associated with this port. */
uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
+ bool may_enable; /* May be enabled in bonds. */
};
static struct ofport_dpif *
/* Support for debugging async flow mods. */
struct list completions;
+
+ bool has_bundle_action; /* True when the first bundle action appears. */
};
/* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
}
static int
-construct(struct ofproto *ofproto_)
+construct(struct ofproto *ofproto_, int *n_tablesp)
{
struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
const char *name = ofproto->up.name;
list_init(&ofproto->completions);
- ofproto->up.tables = xmalloc(sizeof *ofproto->up.tables);
- classifier_init(&ofproto->up.tables[0]);
- ofproto->up.n_tables = 1;
-
ofproto_dpif_unixctl_init();
+ ofproto->has_bundle_action = false;
+
+ *n_tablesp = 255;
return 0;
}
{
struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
struct rule_dpif *rule, *next_rule;
- struct cls_cursor cursor;
+ struct classifier *table;
int i;
complete_operations(ofproto);
- cls_cursor_init(&cursor, &ofproto->up.tables[0], NULL);
- CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
- ofproto_rule_destroy(&rule->up);
+ OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
+ struct cls_cursor cursor;
+
+ cls_cursor_init(&cursor, table, NULL);
+ CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
+ ofproto_rule_destroy(&rule->up);
+ }
}
for (i = 0; i < MAX_MIRRORS; i++) {
port->bundle = NULL;
port->cfm = NULL;
port->tag = tag_create_random();
+ port->may_enable = true;
if (ofproto->sflow) {
dpif_sflow_add_port(ofproto->sflow, port->odp_port,
struct ofport_dpif *port;
LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
- bool may_enable = lacp_slave_may_enable(bundle->lacp, port);
-
- if (may_enable && port->cfm) {
- may_enable = !cfm_get_fault(port->cfm);
- }
- bond_slave_set_may_enable(bundle->bond, port, may_enable);
+ bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
}
bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
mirror->ofproto = ofproto;
mirror->idx = idx;
+ mirror->aux = aux;
mirror->out_vlan = -1;
mirror->name = NULL;
}
static void
port_run(struct ofport_dpif *ofport)
{
+ bool enable = netdev_get_carrier(ofport->up.netdev);
+
if (ofport->cfm) {
cfm_run(ofport->cfm);
ofport->odp_port, &packet);
ofpbuf_uninit(&packet);
}
+
+ enable = enable && !cfm_get_fault(ofport->cfm);
+ }
+
+ if (ofport->bundle) {
+ enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
+ }
+
+ if (ofport->may_enable != enable) {
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
+
+ if (ofproto->has_bundle_action) {
+ ofproto->need_revalidate = true;
+ }
}
+
+ ofport->may_enable = enable;
}
static void
expire(struct ofproto_dpif *ofproto)
{
struct rule_dpif *rule, *next_rule;
- struct cls_cursor cursor;
+ struct classifier *table;
int dp_max_idle;
/* Update stats for each flow in the datapath. */
expire_facets(ofproto, dp_max_idle);
/* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
- cls_cursor_init(&cursor, &ofproto->up.tables[0], NULL);
- CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
- rule_expire(rule);
+ OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
+ struct cls_cursor cursor;
+
+ cls_cursor_init(&cursor, table, NULL);
+ CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
+ rule_expire(rule);
+ }
}
/* All outstanding data in existing flows has been accounted, so it's a
* 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).
+ * the most-recently-used 1% of facets (but at least
+ * ofproto->up.flow_eviction_threshold 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 update_stats(), because the former function never looks
int i;
total = hmap_count(&ofproto->facets);
- if (total <= 1000) {
+ if (total <= ofproto->up.flow_eviction_threshold) {
return N_BUCKETS * BUCKET_WIDTH;
}
subtotal = bucket = 0;
do {
subtotal += buckets[bucket++];
- } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
+ } while (bucket < N_BUCKETS &&
+ subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
if (VLOG_IS_DBG_ENABLED()) {
struct ds s;
struct ofpbuf *packet)
{
if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
- && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) {
+ && odp_actions->nla_type == ODP_ACTION_ATTR_USERSPACE) {
/* 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. */
static void do_xlate_actions(const union ofp_action *in, size_t n_in,
struct action_xlate_ctx *ctx);
-static bool xlate_normal(struct action_xlate_ctx *);
+static void xlate_normal(struct action_xlate_ctx *);
static void
commit_odp_actions(struct action_xlate_ctx *ctx)
base->nw_dst = flow->nw_dst;
}
+ if (base->nw_tos != flow->nw_tos) {
+ nl_msg_put_u8(odp_actions, ODP_ACTION_ATTR_SET_NW_TOS, flow->nw_tos);
+ base->nw_tos = flow->nw_tos;
+ }
+
if (base->vlan_tci != flow->vlan_tci) {
if (!(flow->vlan_tci & htons(VLAN_CFI))) {
nl_msg_put_flag(odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
break;
case OFPP_CONTROLLER:
commit_odp_actions(ctx);
- nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len);
+ nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_USERSPACE, max_len);
break;
case OFPP_LOCAL:
add_output_action(ctx, OFPP_LOCAL);
break;
+ case OFPP_NONE:
+ break;
default:
if (port != ctx->flow.in_port) {
add_output_action(ctx, port);
autopath_execute(naa, &ctx->flow, ofp_port);
}
+static bool
+slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
+{
+ struct ofproto_dpif *ofproto = ofproto_;
+ struct ofport_dpif *port;
+
+ switch (ofp_port) {
+ case OFPP_IN_PORT:
+ case OFPP_TABLE:
+ case OFPP_NORMAL:
+ case OFPP_FLOOD:
+ case OFPP_ALL:
+ case OFPP_LOCAL:
+ return true;
+ case OFPP_CONTROLLER: /* Not supported by the bundle action. */
+ return false;
+ default:
+ port = get_ofp_port(ofproto, ofp_port);
+ return port ? port->may_enable : false;
+ }
+}
+
static void
do_xlate_actions(const union ofp_action *in, size_t n_in,
struct action_xlate_ctx *ctx)
const struct nx_action_set_queue *nasq;
const struct nx_action_multipath *nam;
const struct nx_action_autopath *naa;
+ const struct nx_action_bundle *nab;
enum ofputil_action_code code;
ovs_be64 tun_id;
break;
case OFPUTIL_OFPAT_SET_NW_TOS:
- ctx->flow.nw_tos = ia->nw_tos.nw_tos;
+ ctx->flow.nw_tos = ia->nw_tos.nw_tos & IP_DSCP_MASK;
break;
case OFPUTIL_OFPAT_SET_TP_SRC:
naa = (const struct nx_action_autopath *) ia;
xlate_autopath(ctx, naa);
break;
+
+ case OFPUTIL_NXAST_BUNDLE:
+ ctx->ofproto->has_bundle_action = true;
+ nab = (const struct nx_action_bundle *) ia;
+ xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
+ slave_enabled_cb,
+ ctx->ofproto), 0);
+ break;
+
+ case OFPUTIL_NXAST_BUNDLE_LOAD:
+ ctx->ofproto->has_bundle_action = true;
+ nab = (const struct nx_action_bundle *) ia;
+ bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
+ ctx->ofproto);
+ break;
}
}
}
static bool
ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
{
- return bundle->vlan < 0 && vlan_bitmap_contains(bundle->trunks, vlan);
+ return (bundle->vlan < 0
+ && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
}
static bool
static bool
vlan_is_mirrored(const struct ofmirror *m, int vlan)
{
- return vlan_bitmap_contains(m->vlans, vlan);
+ return !m->vlans || bitmap_is_set(m->vlans, vlan);
+}
+
+/* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
+ * to a VLAN. In general most packets may be mirrored but we want to drop
+ * protocols that may confuse switches. */
+static bool
+eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
+{
+ /* If you change this function's behavior, please update corresponding
+ * documentation in vswitch.xml at the same time. */
+ if (dst[0] != 0x01) {
+ /* All the currently banned MACs happen to start with 01 currently, so
+ * this is a quick way to eliminate most of the good ones. */
+ } else {
+ if (eth_addr_is_reserved(dst)) {
+ /* Drop STP, IEEE pause frames, and other reserved protocols
+ * (01-80-c2-00-00-0x). */
+ return false;
+ }
+
+ if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
+ /* Cisco OUI. */
+ if ((dst[3] & 0xfe) == 0xcc &&
+ (dst[4] & 0xfe) == 0xcc &&
+ (dst[5] & 0xfe) == 0xcc) {
+ /* Drop the following protocols plus others following the same
+ pattern:
+
+ CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
+ Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
+ STP Uplink Fast (01-00-0c-cd-cd-cd) */
+ return false;
+ }
+
+ if (!(dst[3] | dst[4] | dst[5])) {
+ /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
+ return false;
+ }
+ }
+ }
+ return true;
}
static void
&& !dst_is_duplicate(set, &dst)) {
dst_set_add(set, &dst);
}
- } else {
+ } else if (eth_dst_may_rspan(ctx->flow.dl_dst)) {
struct ofbundle *bundle;
HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
return true;
}
-/* If the composed actions may be applied to any packet in the given 'flow',
- * returns true. Otherwise, the actions should only be applied to 'packet', or
- * not at all, if 'packet' was NULL. */
-static bool
+static void
xlate_normal(struct action_xlate_ctx *ctx)
{
struct ofbundle *in_bundle;
* of time where we could learn from a packet reflected on a bond and
* blackhole packets before the learning table is updated to reflect
* the correct port. */
- return false;
+ ctx->may_set_up_flow = false;
+ return;
} else {
out_bundle = OFBUNDLE_FLOOD;
}
if (in_bundle) {
compose_actions(ctx, vlan, in_bundle, out_bundle);
}
-
- return true;
}
\f
static bool
};
static void
-trace_format_rule(struct ds *result, int level, const struct rule *rule)
+trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
{
ds_put_char_multiple(result, '\t', level);
if (!rule) {
}
ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
- ntohll(rule->flow_cookie));
- cls_rule_format(&rule->cr, result);
+ ntohll(rule->up.flow_cookie));
+ cls_rule_format(&rule->up.cr, result);
ds_put_char(result, '\n');
ds_put_char_multiple(result, '\t', level);
ds_put_cstr(result, "OpenFlow ");
- ofp_print_actions(result, rule->actions, rule->n_actions);
+ ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
ds_put_char(result, '\n');
}
ds_put_char(result, '\n');
trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
- trace_format_rule(result, ctx->recurse + 1, &rule->up);
+ trace_format_rule(result, ctx->recurse + 1, rule);
}
static void
ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
void *aux OVS_UNUSED)
{
- char *dpname, *in_port_s, *tun_id_s, *packet_s;
+ char *dpname, *arg1, *arg2, *arg3;
char *args = xstrdup(args_);
char *save_ptr = NULL;
struct ofproto_dpif *ofproto;
- struct ofpbuf packet;
+ struct ofpbuf odp_key;
+ struct ofpbuf *packet;
struct rule_dpif *rule;
struct ds result;
struct flow flow;
- uint16_t in_port;
- ovs_be64 tun_id;
char *s;
- ofpbuf_init(&packet, strlen(args) / 2);
+ packet = NULL;
+ ofpbuf_init(&odp_key, 0);
ds_init(&result);
dpname = strtok_r(args, " ", &save_ptr);
- tun_id_s = strtok_r(NULL, " ", &save_ptr);
- in_port_s = strtok_r(NULL, " ", &save_ptr);
- packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
- if (!dpname || !in_port_s || !packet_s) {
+ arg1 = strtok_r(NULL, " ", &save_ptr);
+ arg2 = strtok_r(NULL, " ", &save_ptr);
+ arg3 = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
+ if (dpname && arg1 && !arg2 && !arg3) {
+ /* ofproto/trace dpname flow */
+ int error;
+
+ /* Convert string to ODP key. */
+ ofpbuf_init(&odp_key, 0);
+ error = odp_flow_key_from_string(arg1, &odp_key);
+ if (error) {
+ unixctl_command_reply(conn, 501, "Bad flow syntax");
+ goto exit;
+ }
+
+ /* Convert odp_key to flow. */
+ error = odp_flow_key_to_flow(odp_key.data, odp_key.size, &flow);
+ if (error) {
+ unixctl_command_reply(conn, 501, "Invalid flow");
+ goto exit;
+ }
+ } else if (dpname && arg1 && arg2 && arg3) {
+ /* ofproto/trace dpname tun_id in_port packet */
+ uint16_t in_port;
+ ovs_be64 tun_id;
+
+ tun_id = htonll(strtoull(arg1, NULL, 0));
+ in_port = ofp_port_to_odp_port(atoi(arg2));
+
+ packet = ofpbuf_new(strlen(args) / 2);
+ arg3 = ofpbuf_put_hex(packet, arg3, NULL);
+ arg3 += strspn(arg3, " ");
+ if (*arg3 != '\0') {
+ unixctl_command_reply(conn, 501, "Trailing garbage in command");
+ goto exit;
+ }
+ if (packet->size < ETH_HEADER_LEN) {
+ unixctl_command_reply(conn, 501,
+ "Packet data too short for Ethernet");
+ goto exit;
+ }
+
+ ds_put_cstr(&result, "Packet: ");
+ s = ofp_packet_to_string(packet->data, packet->size, packet->size);
+ ds_put_cstr(&result, s);
+ free(s);
+
+ flow_extract(packet, tun_id, in_port, &flow);
+ } else {
unixctl_command_reply(conn, 501, "Bad command syntax");
goto exit;
}
goto exit;
}
- tun_id = htonll(strtoull(tun_id_s, NULL, 0));
- in_port = ofp_port_to_odp_port(atoi(in_port_s));
-
- packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
- packet_s += strspn(packet_s, " ");
- if (*packet_s != '\0') {
- unixctl_command_reply(conn, 501, "Trailing garbage in command");
- goto exit;
- }
- if (packet.size < ETH_HEADER_LEN) {
- unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
- goto exit;
- }
-
- ds_put_cstr(&result, "Packet: ");
- s = ofp_packet_to_string(packet.data, packet.size, packet.size);
- ds_put_cstr(&result, s);
- free(s);
-
- flow_extract(&packet, tun_id, in_port, &flow);
ds_put_cstr(&result, "Flow: ");
flow_format(&result, &flow);
ds_put_char(&result, '\n');
rule = rule_dpif_lookup(ofproto, &flow);
- trace_format_rule(&result, 0, &rule->up);
+ trace_format_rule(&result, 0, rule);
if (rule) {
struct ofproto_trace trace;
struct ofpbuf *odp_actions;
trace.result = &result;
trace.flow = flow;
- action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
+ action_xlate_ctx_init(&trace.ctx, ofproto, &flow, packet);
trace.ctx.resubmit_hook = trace_resubmit;
odp_actions = xlate_actions(&trace.ctx,
rule->up.actions, rule->up.n_actions);
ds_put_cstr(&result, "Datapath actions: ");
format_odp_actions(&result, odp_actions->data, odp_actions->size);
ofpbuf_delete(odp_actions);
+
+ if (!trace.ctx.may_set_up_flow) {
+ if (packet) {
+ ds_put_cstr(&result, "\nThis flow is not cachable.");
+ } else {
+ ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
+ "for complete actions, please supply a packet.");
+ }
+ }
}
unixctl_command_reply(conn, 200, ds_cstr(&result));
exit:
ds_destroy(&result);
- ofpbuf_uninit(&packet);
+ ofpbuf_delete(packet);
+ ofpbuf_uninit(&odp_key);
free(args);
}
git://git.onelab.eu / sliver-openvswitch.git / blobdiff