2 * Copyright (c) 2010 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "wdp-xflow.h"
30 #include "openflow/nicira-ext.h"
31 #include "openflow/openflow.h"
33 #include "poll-loop.h"
34 #include "port-array.h"
41 #include "wdp-provider.h"
43 #include "xflow-util.h"
46 #include <linux/types.h> /* XXX */
47 #include <linux/pkt_sched.h> /* XXX */
49 #define THIS_MODULE VLM_wdp_xflow
52 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
54 /* Maximum numbers of rules. */
55 #define WX_MAX_WILD 65536 /* Wildcarded rules. */
56 #define WX_MAX_EXACT 1048576 /* Exact-match rules. */
59 struct list list_node;
62 struct classifier cls;
63 struct netdev_monitor *netdev_monitor;
64 struct port_array ports; /* Index is xflow port nr;
65 * wdp_port->opp.port_no is OFP port nr. */
66 struct shash port_by_name;
68 long long int next_expiration;
71 static struct list all_wx = LIST_INITIALIZER(&all_wx);
73 static int wx_port_init(struct wx *);
74 static void wx_port_run(struct wx *);
75 static void wx_port_refresh_groups(struct wx *);
83 wx_cast(const struct wdp *wdp)
85 return CONTAINER_OF(wdp, struct wx, wdp);
89 wx_xlate_actions(struct wx *, const union ofp_action *, size_t n,
90 const flow_t *flow, const struct ofpbuf *packet,
91 struct xflow_actions *out, bool *may_set_up_flow);
96 uint64_t packet_count; /* Number of packets received. */
97 uint64_t byte_count; /* Number of bytes received. */
98 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
99 long long int used; /* Last-used time (0 if never used). */
101 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
102 * exact-match rule (having cr.wc.wildcards of 0) generated from the
103 * wildcard rule 'super'. In this case, 'list' is an element of the
106 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
107 * a list of subrules. A super-rule with no wildcards (where
108 * cr.wc.wildcards is 0) will never have any subrules. */
109 struct wx_rule *super;
114 * A super-rule with wildcard fields never has xflow actions (since the
115 * datapath only supports exact-match flows). */
116 bool installed; /* Installed in datapath? */
117 bool may_install; /* True ordinarily; false if actions must
118 * be reassessed for every packet. */
120 union xflow_action *xflow_actions;
123 static void wx_rule_destroy(struct wx *, struct wx_rule *);
124 static void wx_rule_update_actions(struct wx *, struct wx_rule *);
125 static void wx_rule_execute(struct wx *, struct wx_rule *,
126 struct ofpbuf *packet, const flow_t *);
127 static bool wx_rule_make_actions(struct wx *, struct wx_rule *,
128 const struct ofpbuf *packet);
129 static void wx_rule_install(struct wx *, struct wx_rule *,
130 struct wx_rule *displaced_rule);
132 static struct wx_rule *
133 wx_rule_cast(const struct cls_rule *cls_rule)
135 return cls_rule ? CONTAINER_OF(cls_rule, struct wx_rule, wr.cr) : NULL;
138 /* Returns true if 'rule' is merely an implementation detail that should be
139 * hidden from the client. */
141 wx_rule_is_hidden(const struct wx_rule *rule)
143 return rule->super != NULL;
147 wx_rule_free(struct wx_rule *rule)
149 wdp_rule_uninit(&rule->wr);
150 free(rule->xflow_actions);
155 wx_rule_account(struct wx *wx OVS_UNUSED, struct wx_rule *rule OVS_UNUSED,
156 uint64_t extra_bytes OVS_UNUSED)
158 /* XXX call account_cb hook */
162 wx_rule_post_uninstall(struct wx *wx, struct wx_rule *rule)
164 struct wx_rule *super = rule->super;
166 wx_rule_account(wx, rule, 0);
168 /* XXX netflow expiration */
171 super->packet_count += rule->packet_count;
172 super->byte_count += rule->byte_count;
174 /* Reset counters to prevent double counting if the rule ever gets
176 rule->packet_count = 0;
177 rule->byte_count = 0;
178 rule->accounted_bytes = 0;
180 //XXX netflow_flow_clear(&rule->nf_flow);
185 xflow_flow_stats_to_msec(const struct xflow_flow_stats *stats)
187 return (stats->used_sec
188 ? stats->used_sec * 1000 + stats->used_nsec / 1000000
193 wx_rule_update_time(struct wx *wx OVS_UNUSED, struct wx_rule *rule,
194 const struct xflow_flow_stats *stats)
196 long long int used = xflow_flow_stats_to_msec(stats);
197 if (used > rule->used) {
199 if (rule->super && used > rule->super->used) {
200 rule->super->used = used;
202 //XXX netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
207 wx_rule_update_stats(struct wx *wx, struct wx_rule *rule,
208 const struct xflow_flow_stats *stats)
210 if (stats->n_packets) {
211 wx_rule_update_time(wx, rule, stats);
212 rule->packet_count += stats->n_packets;
213 rule->byte_count += stats->n_bytes;
214 /* XXX netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
215 stats->tcp_flags); */
220 wx_rule_uninstall(struct wx *wx, struct wx_rule *rule)
222 assert(!rule->wr.cr.flow.wildcards);
223 if (rule->installed) {
224 struct xflow_flow xflow_flow;
226 xflow_key_from_flow(&xflow_flow.key, &rule->wr.cr.flow);
227 xflow_flow.actions = NULL;
228 xflow_flow.n_actions = 0;
229 xflow_flow.flags = 0;
230 if (!xfif_flow_del(wx->xfif, &xflow_flow)) {
231 wx_rule_update_stats(wx, rule, &xflow_flow.stats);
233 rule->installed = false;
235 wx_rule_post_uninstall(wx, rule);
241 is_controller_rule(struct wx_rule *rule)
243 /* If the only action is send to the controller then don't report
244 * NetFlow expiration messages since it is just part of the control
245 * logic for the network and not real traffic. */
249 && rule->super->n_actions == 1
250 && action_outputs_to_port(&rule->super->actions[0],
251 htons(OFPP_CONTROLLER)));
256 wx_rule_remove(struct wx *wx, struct wx_rule *rule)
258 if (rule->wr.cr.flow.wildcards) {
259 COVERAGE_INC(wx_del_wc_flow);
260 wx->need_revalidate = true;
262 wx_rule_uninstall(wx, rule);
264 classifier_remove(&wx->cls, &rule->wr.cr);
265 wx_rule_destroy(wx, rule);
269 wx_rule_revalidate(struct wx *wx, struct wx_rule *rule)
271 const flow_t *flow = &rule->wr.cr.flow;
273 COVERAGE_INC(wx_rule_revalidate);
275 struct wx_rule *super;
276 super = wx_rule_cast(classifier_lookup_wild(&wx->cls, flow));
278 wx_rule_remove(wx, rule);
280 } else if (super != rule->super) {
281 COVERAGE_INC(wx_revalidate_moved);
282 list_remove(&rule->list);
283 list_push_back(&super->list, &rule->list);
285 rule->wr.hard_timeout = super->wr.hard_timeout;
286 rule->wr.idle_timeout = super->wr.idle_timeout;
287 rule->wr.created = super->wr.created;
292 wx_rule_update_actions(wx, rule);
296 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
297 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
298 * through all of its subrules and revalidates them, destroying any that no
299 * longer has a super-rule (which is probably all of them).
301 * Before calling this function, the caller must make have removed 'rule' from
302 * the classifier. If 'rule' is an exact-match rule, the caller is also
303 * responsible for ensuring that it has been uninstalled from the datapath. */
305 wx_rule_destroy(struct wx *wx, struct wx_rule *rule)
308 struct wx_rule *subrule, *next;
309 LIST_FOR_EACH_SAFE (subrule, next, struct wx_rule, list, &rule->list) {
310 wx_rule_revalidate(wx, subrule);
313 list_remove(&rule->list);
320 wx_rule_has_out_port(const struct wx_rule *rule, uint16_t out_port)
322 const union ofp_action *oa;
323 struct actions_iterator i;
325 if (out_port == htons(OFPP_NONE)) {
328 for (oa = actions_first(&i, rule->wr.actions,
331 oa = actions_next(&i)) {
332 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
340 /* Caller is responsible for initializing the 'cr' member of the returned
342 static struct wx_rule *
343 wx_rule_create(struct wx_rule *super,
344 const union ofp_action *actions, size_t n_actions,
345 uint16_t idle_timeout, uint16_t hard_timeout)
347 struct wx_rule *rule = xzalloc(sizeof *rule);
348 wdp_rule_init(&rule->wr, actions, n_actions);
349 rule->wr.idle_timeout = idle_timeout;
350 rule->wr.hard_timeout = hard_timeout;
351 rule->used = rule->wr.created;
354 list_push_back(&super->list, &rule->list);
356 list_init(&rule->list);
359 netflow_flow_clear(&rule->nf_flow);
360 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
366 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
367 * 'flow' and is considered to have arrived on xflow port 'in_port'.
369 * The flow that 'packet' actually contains does not need to actually match
370 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
371 * the packet and byte counters for 'rule' will be credited for the packet sent
372 * out whether or not the packet actually matches 'rule'.
374 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
375 * the caller must already have accurately composed xflow actions for it given
376 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
377 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
378 * function will compose a set of xflow actions based on 'rule''s OpenFlow
379 * actions and apply them to 'packet'. */
381 wx_rule_execute(struct wx *wx, struct wx_rule *rule,
382 struct ofpbuf *packet, const flow_t *flow)
384 const union xflow_action *actions;
386 struct xflow_actions a;
388 /* Grab or compose the xflow actions.
390 * The special case for an exact-match 'rule' where 'flow' is not the
391 * rule's flow is important to avoid, e.g., sending a packet out its input
392 * port simply because the xflow actions were composed for the wrong
394 if (rule->wr.cr.flow.wildcards
395 || !flow_equal(flow, &rule->wr.cr.flow))
397 struct wx_rule *super = rule->super ? rule->super : rule;
398 if (wx_xlate_actions(wx, super->wr.actions, super->wr.n_actions, flow,
403 n_actions = a.n_actions;
405 actions = rule->xflow_actions;
406 n_actions = rule->n_xflow_actions;
409 /* Execute the xflow actions. */
410 if (!xfif_execute(wx->xfif, flow->in_port,
411 actions, n_actions, packet)) {
412 struct xflow_flow_stats stats;
413 flow_extract_stats(flow, packet, &stats);
414 wx_rule_update_stats(wx, rule, &stats);
415 rule->used = time_msec();
416 //XXX netflow_flow_update_time(wx->netflow, &rule->nf_flow, rule->used);
421 wx_rule_insert(struct wx *wx, struct wx_rule *rule, struct ofpbuf *packet,
424 struct wx_rule *displaced_rule;
426 /* Insert the rule in the classifier. */
427 displaced_rule = wx_rule_cast(classifier_insert(&wx->cls, &rule->wr.cr));
428 if (!rule->wr.cr.flow.wildcards) {
429 wx_rule_make_actions(wx, rule, packet);
432 /* Send the packet and credit it to the rule. */
435 flow_extract(packet, 0, in_port, &flow);
436 wx_rule_execute(wx, rule, packet, &flow);
439 /* Install the rule in the datapath only after sending the packet, to
440 * avoid packet reordering. */
441 if (rule->wr.cr.flow.wildcards) {
442 COVERAGE_INC(wx_add_wc_flow);
443 wx->need_revalidate = true;
445 wx_rule_install(wx, rule, displaced_rule);
448 /* Free the rule that was displaced, if any. */
449 if (displaced_rule) {
450 rule->wr.client_data = displaced_rule->wr.client_data;
451 wx_rule_destroy(wx, displaced_rule);
455 static struct wx_rule *
456 wx_rule_create_subrule(struct wx *wx, struct wx_rule *rule, const flow_t *flow)
458 struct wx_rule *subrule;
460 subrule = wx_rule_create(rule, NULL, 0,
461 rule->wr.idle_timeout,
462 rule->wr.hard_timeout);
463 COVERAGE_INC(wx_subrule_create);
464 cls_rule_from_flow(flow, &subrule->wr.cr);
465 classifier_insert_exact(&wx->cls, &subrule->wr.cr);
470 /* Returns true if the actions changed, false otherwise. */
472 wx_rule_make_actions(struct wx *wx, struct wx_rule *rule,
473 const struct ofpbuf *packet)
475 const struct wx_rule *super;
476 struct xflow_actions a;
479 assert(!rule->wr.cr.flow.wildcards);
481 super = rule->super ? rule->super : rule;
482 wx_xlate_actions(wx, super->wr.actions, super->wr.n_actions,
483 &rule->wr.cr.flow, packet, &a, &rule->may_install);
485 actions_len = a.n_actions * sizeof *a.actions;
486 if (rule->n_xflow_actions != a.n_actions
487 || memcmp(rule->xflow_actions, a.actions, actions_len)) {
488 COVERAGE_INC(wx_xflow_unchanged);
489 free(rule->xflow_actions);
490 rule->n_xflow_actions = a.n_actions;
491 rule->xflow_actions = xmemdup(a.actions, actions_len);
499 do_put_flow(struct wx *wx, struct wx_rule *rule, int flags,
500 struct xflow_flow_put *put)
502 memset(&put->flow.stats, 0, sizeof put->flow.stats);
503 xflow_key_from_flow(&put->flow.key, &rule->wr.cr.flow);
504 put->flow.actions = rule->xflow_actions;
505 put->flow.n_actions = rule->n_xflow_actions;
508 return xfif_flow_put(wx->xfif, put);
512 wx_rule_install(struct wx *wx, struct wx_rule *rule, struct wx_rule *displaced_rule)
514 assert(!rule->wr.cr.flow.wildcards);
516 if (rule->may_install) {
517 struct xflow_flow_put put;
518 if (!do_put_flow(wx, rule,
519 XFLOWPF_CREATE | XFLOWPF_MODIFY | XFLOWPF_ZERO_STATS,
521 rule->installed = true;
522 if (displaced_rule) {
523 wx_rule_update_stats(wx, displaced_rule, &put.flow.stats);
524 wx_rule_post_uninstall(wx, displaced_rule);
527 } else if (displaced_rule) {
528 wx_rule_uninstall(wx, displaced_rule);
533 wx_rule_reinstall(struct wx *wx, struct wx_rule *rule)
535 if (rule->installed) {
536 struct xflow_flow_put put;
537 COVERAGE_INC(wx_dp_missed);
538 do_put_flow(wx, rule, XFLOWPF_CREATE | XFLOWPF_MODIFY, &put);
540 wx_rule_install(wx, rule, NULL);
545 wx_rule_update_actions(struct wx *wx, struct wx_rule *rule)
547 bool actions_changed;
549 uint16_t new_out_iface, old_out_iface;
551 old_out_iface = rule->nf_flow.output_iface;
553 actions_changed = wx_rule_make_actions(wx, rule, NULL);
555 if (rule->may_install) {
556 if (rule->installed) {
557 if (actions_changed) {
558 struct xflow_flow_put put;
559 do_put_flow(wx, rule, XFLOWPF_CREATE | XFLOWPF_MODIFY
560 | XFLOWPF_ZERO_STATS, &put);
561 wx_rule_update_stats(wx, rule, &put.flow.stats);
563 /* Temporarily set the old output iface so that NetFlow
564 * messages have the correct output interface for the old
566 new_out_iface = rule->nf_flow.output_iface;
567 rule->nf_flow.output_iface = old_out_iface;
569 wx_rule_post_uninstall(wx, rule);
570 //rule->nf_flow.output_iface = new_out_iface;
573 wx_rule_install(wx, rule, NULL);
576 wx_rule_uninstall(wx, rule);
581 add_output_group_action(struct xflow_actions *actions, uint16_t group,
582 uint16_t *nf_output_iface)
584 xflow_actions_add(actions, XFLOWAT_OUTPUT_GROUP)->output_group.group = group;
586 if (group == WX_GROUP_ALL || group == WX_GROUP_FLOOD) {
587 *nf_output_iface = NF_OUT_FLOOD;
592 add_controller_action(struct xflow_actions *actions,
593 const struct ofp_action_output *oao)
595 union xflow_action *a = xflow_actions_add(actions, XFLOWAT_CONTROLLER);
596 a->controller.arg = ntohs(oao->max_len);
599 struct wx_xlate_ctx {
601 flow_t flow; /* Flow to which these actions correspond. */
602 int recurse; /* Recursion level, via xlate_table_action. */
604 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
605 * null pointer if we are revalidating
606 * without a packet to refer to. */
609 struct xflow_actions *out; /* Datapath actions. */
610 //tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
611 bool may_set_up_flow; /* True ordinarily; false if the actions must
612 * be reassessed for every packet. */
613 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
616 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
617 struct wx_xlate_ctx *ctx);
620 add_output_action(struct wx_xlate_ctx *ctx, uint16_t port)
622 const struct wdp_port *wdp_port = port_array_get(&ctx->wx->ports, port);
625 if (wdp_port->opp.config & OFPPC_NO_FWD) {
626 /* Forwarding disabled on port. */
631 * We don't have an ofport record for this port, but it doesn't hurt to
632 * allow forwarding to it anyhow. Maybe such a port will appear later
633 * and we're pre-populating the flow table.
637 xflow_actions_add(ctx->out, XFLOWAT_OUTPUT)->output.port = port;
638 //ctx->nf_output_iface = port;
641 static struct wx_rule *
642 wx_rule_lookup_valid(struct wx *wx, const flow_t *flow)
644 struct wx_rule *rule = wx_rule_cast(classifier_lookup(&wx->cls, flow));
646 /* The rule we found might not be valid, since we could be in need of
647 * revalidation. If it is not valid, don't return it. */
650 && wx->need_revalidate
651 && !wx_rule_revalidate(wx, rule)) {
652 COVERAGE_INC(wx_invalidated);
660 xlate_table_action(struct wx_xlate_ctx *ctx, uint16_t in_port)
663 uint16_t old_in_port;
664 struct wx_rule *rule;
666 /* Look up a flow with 'in_port' as the input port. Then restore the
667 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
668 * have surprising behavior). */
669 old_in_port = ctx->flow.in_port;
670 ctx->flow.in_port = in_port;
671 rule = wx_rule_lookup_valid(ctx->wx, &ctx->flow);
672 ctx->flow.in_port = old_in_port;
680 do_xlate_actions(rule->wr.actions, rule->wr.n_actions, ctx);
687 xlate_output_action(struct wx_xlate_ctx *ctx,
688 const struct ofp_action_output *oao)
691 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
693 ctx->nf_output_iface = NF_OUT_DROP;
695 switch (ntohs(oao->port)) {
697 add_output_action(ctx, ctx->flow.in_port);
700 xlate_table_action(ctx, ctx->flow.in_port);
704 if (!ctx->wx->ofhooks->normal_cb(ctx->flow, ctx->packet,
706 &ctx->nf_output_iface,
708 COVERAGE_INC(wx_uninstallable);
709 ctx->may_set_up_flow = false;
713 /* fall through to flood for now */
716 add_output_group_action(ctx->out, WX_GROUP_FLOOD,
717 &ctx->nf_output_iface);
720 add_output_group_action(ctx->out, WX_GROUP_ALL, &ctx->nf_output_iface);
722 case OFPP_CONTROLLER:
723 add_controller_action(ctx->out, oao);
726 add_output_action(ctx, XFLOWP_LOCAL);
729 xflow_port = ofp_port_to_xflow_port(ntohs(oao->port));
730 if (xflow_port != ctx->flow.in_port) {
731 add_output_action(ctx, xflow_port);
736 if (prev_nf_output_iface == NF_OUT_FLOOD) {
737 ctx->nf_output_iface = NF_OUT_FLOOD;
738 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
739 ctx->nf_output_iface = prev_nf_output_iface;
740 } else if (prev_nf_output_iface != NF_OUT_DROP &&
741 ctx->nf_output_iface != NF_OUT_FLOOD) {
742 ctx->nf_output_iface = NF_OUT_MULTI;
746 /* If the final xflow action in 'ctx' is "pop priority", drop it, as an
747 * optimization, because we're going to add another action that sets the
748 * priority immediately after, or because there are no actions following the
751 remove_pop_action(struct wx_xlate_ctx *ctx)
753 size_t n = ctx->out->n_actions;
754 if (n > 0 && ctx->out->actions[n - 1].type == XFLOWAT_POP_PRIORITY) {
755 ctx->out->n_actions--;
760 xlate_enqueue_action(struct wx_xlate_ctx *ctx,
761 const struct ofp_action_enqueue *oae)
763 uint16_t ofp_port, xflow_port;
765 /* Figure out xflow output port. */
766 ofp_port = ntohs(oae->port);
767 if (ofp_port != OFPP_IN_PORT) {
768 xflow_port = ofp_port_to_xflow_port(ofp_port);
770 xflow_port = ctx->flow.in_port;
773 /* Add xflow actions. */
774 remove_pop_action(ctx);
775 xflow_actions_add(ctx->out, XFLOWAT_SET_PRIORITY)->priority.priority
776 = TC_H_MAKE(1, ntohl(oae->queue_id)); /* XXX */
777 add_output_action(ctx, xflow_port);
778 xflow_actions_add(ctx->out, XFLOWAT_POP_PRIORITY);
780 /* Update NetFlow output port. */
781 if (ctx->nf_output_iface == NF_OUT_DROP) {
782 ctx->nf_output_iface = xflow_port;
783 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
784 ctx->nf_output_iface = NF_OUT_MULTI;
789 xlate_nicira_action(struct wx_xlate_ctx *ctx,
790 const struct nx_action_header *nah)
792 const struct nx_action_resubmit *nar;
793 const struct nx_action_set_tunnel *nast;
794 union xflow_action *oa;
795 int subtype = ntohs(nah->subtype);
797 assert(nah->vendor == htonl(NX_VENDOR_ID));
800 nar = (const struct nx_action_resubmit *) nah;
801 xlate_table_action(ctx, ofp_port_to_xflow_port(ntohs(nar->in_port)));
804 case NXAST_SET_TUNNEL:
805 nast = (const struct nx_action_set_tunnel *) nah;
806 oa = xflow_actions_add(ctx->out, XFLOWAT_SET_TUNNEL);
807 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
810 /* If you add a new action here that modifies flow data, don't forget to
811 * update the flow key in ctx->flow at the same time. */
814 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
820 do_xlate_actions(const union ofp_action *in, size_t n_in,
821 struct wx_xlate_ctx *ctx)
823 struct actions_iterator iter;
824 const union ofp_action *ia;
825 const struct wdp_port *port;
827 port = port_array_get(&ctx->wx->ports, ctx->flow.in_port);
828 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
829 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, stp_eth_addr)
830 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
831 /* Drop this flow. */
835 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
836 uint16_t type = ntohs(ia->type);
837 union xflow_action *oa;
841 xlate_output_action(ctx, &ia->output);
844 case OFPAT_SET_VLAN_VID:
845 oa = xflow_actions_add(ctx->out, XFLOWAT_SET_DL_TCI);
846 oa->dl_tci.tci = ia->vlan_vid.vlan_vid & htons(VLAN_VID_MASK);
847 oa->dl_tci.mask = htons(VLAN_VID_MASK);
848 ctx->flow.dl_vlan = ia->vlan_vid.vlan_vid;
851 case OFPAT_SET_VLAN_PCP:
852 oa = xflow_actions_add(ctx->out, XFLOWAT_SET_DL_TCI);
853 oa->dl_tci.tci = htons((ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT)
855 oa->dl_tci.mask = htons(VLAN_PCP_MASK);
857 if (ctx->flow.dl_vlan == htons(OFP_VLAN_NONE)) {
858 ctx->flow.dl_vlan = htons(0);
860 ctx->flow.dl_vlan_pcp = ia->vlan_pcp.vlan_pcp;
863 case OFPAT_STRIP_VLAN:
864 xflow_actions_add(ctx->out, XFLOWAT_STRIP_VLAN);
865 ctx->flow.dl_vlan = htons(OFP_VLAN_NONE);
866 ctx->flow.dl_vlan_pcp = 0;
869 case OFPAT_SET_DL_SRC:
870 oa = xflow_actions_add(ctx->out, XFLOWAT_SET_DL_SRC);
871 memcpy(oa->dl_addr.dl_addr,
872 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
873 memcpy(ctx->flow.dl_src,
874 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
877 case OFPAT_SET_DL_DST:
878 oa = xflow_actions_add(ctx->out, XFLOWAT_SET_DL_DST);
879 memcpy(oa->dl_addr.dl_addr,
880 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
881 memcpy(ctx->flow.dl_dst,
882 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
885 case OFPAT_SET_NW_SRC:
886 oa = xflow_actions_add(ctx->out, XFLOWAT_SET_NW_SRC);
887 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
890 case OFPAT_SET_NW_DST:
891 oa = xflow_actions_add(ctx->out, XFLOWAT_SET_NW_DST);
892 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
895 case OFPAT_SET_NW_TOS:
896 oa = xflow_actions_add(ctx->out, XFLOWAT_SET_NW_TOS);
897 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
900 case OFPAT_SET_TP_SRC:
901 oa = xflow_actions_add(ctx->out, XFLOWAT_SET_TP_SRC);
902 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
905 case OFPAT_SET_TP_DST:
906 oa = xflow_actions_add(ctx->out, XFLOWAT_SET_TP_DST);
907 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
911 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
915 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
919 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
925 /* Returns true if 'flow' and 'actions' may be set up as a flow in the kernel.
926 * This is true most of the time, but we don't allow flows that would prevent
927 * DHCP replies from being seen by the local port to be set up in the
930 * We only need this, strictly speaking, when in-band control is turned on. */
932 wx_may_set_up(const flow_t *flow, const struct xflow_actions *actions)
934 if (flow->dl_type == htons(ETH_TYPE_IP)
935 && flow->nw_proto == IP_TYPE_UDP
936 && flow->tp_src == htons(DHCP_SERVER_PORT)
937 && flow->tp_dst == htons(DHCP_CLIENT_PORT)) {
940 for (i = 0; i < actions->n_actions; i++) {
941 const struct xflow_action_output *oao = &actions->actions[i].output;
942 if (oao->type == XFLOWAT_OUTPUT && oao->port == XFLOWP_LOCAL) {
953 wx_xlate_actions(struct wx *wx, const union ofp_action *in, size_t n_in,
954 const flow_t *flow, const struct ofpbuf *packet,
955 struct xflow_actions *out, bool *may_set_up_flow)
957 //tag_type no_tags = 0;
958 struct wx_xlate_ctx ctx;
959 COVERAGE_INC(wx_ofp2xflow);
960 xflow_actions_init(out);
966 //ctx.tags = tags ? tags : &no_tags;
967 ctx.may_set_up_flow = true;
968 ctx.nf_output_iface = NF_OUT_DROP;
969 do_xlate_actions(in, n_in, &ctx);
970 remove_pop_action(&ctx);
972 if (may_set_up_flow) {
973 *may_set_up_flow = ctx.may_set_up_flow && wx_may_set_up(flow, out);
976 if (nf_output_iface) {
977 *nf_output_iface = ctx.nf_output_iface;
980 if (xflow_actions_overflow(out)) {
981 xflow_actions_init(out);
982 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
988 update_used(struct wx *wx)
990 struct xflow_flow *flows;
995 error = xfif_flow_list_all(wx->xfif, &flows, &n_flows);
1000 for (i = 0; i < n_flows; i++) {
1001 struct xflow_flow *f = &flows[i];
1002 struct wx_rule *rule;
1005 xflow_key_to_flow(&f->key, &flow);
1006 rule = wx_rule_cast(classifier_find_rule_exactly(&wx->cls, &flow));
1007 if (!rule || !rule->installed) {
1008 COVERAGE_INC(wx_unexpected_rule);
1009 xfif_flow_del(wx->xfif, f);
1013 wx_rule_update_time(wx, rule, &f->stats);
1014 wx_rule_account(wx, rule, f->stats.n_bytes);
1020 uninstall_idle_flow(struct wx *wx, struct wx_rule *rule)
1022 assert(rule->installed);
1023 assert(!rule->wr.cr.flow.wildcards);
1026 wx_rule_remove(wx, rule);
1028 wx_rule_uninstall(wx, rule);
1033 expire_rule(struct cls_rule *cls_rule, void *wx_)
1035 struct wx *wx = wx_;
1036 struct wx_rule *rule = wx_rule_cast(cls_rule);
1037 long long int hard_expire, idle_expire, expire, now;
1039 hard_expire = (rule->wr.hard_timeout
1040 ? rule->wr.created + rule->wr.hard_timeout * 1000
1042 idle_expire = (rule->wr.idle_timeout
1043 && (rule->super || list_is_empty(&rule->list))
1044 ? rule->used + rule->wr.idle_timeout * 1000
1046 expire = MIN(hard_expire, idle_expire);
1050 if (rule->installed && now >= rule->used + 5000) {
1051 uninstall_idle_flow(wx, rule);
1052 } else if (!rule->wr.cr.flow.wildcards) {
1053 //XXX active_timeout(wx, rule);
1059 COVERAGE_INC(wx_expired);
1061 /* Update stats. This code will be a no-op if the rule expired
1062 * due to an idle timeout. */
1063 if (rule->wr.cr.flow.wildcards) {
1064 struct wx_rule *subrule, *next;
1065 LIST_FOR_EACH_SAFE (subrule, next, struct wx_rule, list, &rule->list) {
1066 wx_rule_remove(wx, subrule);
1069 wx_rule_uninstall(wx, rule);
1073 if (!wx_rule_is_hidden(rule)) {
1074 send_flow_removed(wx, rule, now,
1076 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
1079 wx_rule_remove(wx, rule);
1082 struct revalidate_cbdata {
1084 bool revalidate_all; /* Revalidate all exact-match rules? */
1085 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
1086 //struct tag_set revalidate_set; /* Set of tags to revalidate. */
1090 revalidate_rule(struct wx *wx, struct wx_rule *rule)
1092 const flow_t *flow = &rule->wr.cr.flow;
1094 COVERAGE_INC(wx_revalidate_rule);
1096 struct wx_rule *super;
1097 super = wx_rule_cast(classifier_lookup_wild(&wx->cls, flow));
1099 wx_rule_remove(wx, rule);
1101 } else if (super != rule->super) {
1102 COVERAGE_INC(wx_revalidate_moved);
1103 list_remove(&rule->list);
1104 list_push_back(&super->list, &rule->list);
1105 rule->super = super;
1106 rule->wr.hard_timeout = super->wr.hard_timeout;
1107 rule->wr.idle_timeout = super->wr.idle_timeout;
1108 rule->wr.created = super->wr.created;
1113 wx_rule_update_actions(wx, rule);
1118 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
1120 struct wx_rule *sub = wx_rule_cast(sub_);
1121 struct revalidate_cbdata *cbdata = cbdata_;
1123 if (cbdata->revalidate_all
1124 || (cbdata->revalidate_subrules && sub->super)
1125 /*|| (tag_set_intersects(&cbdata->revalidate_set, sub->tags))*/) {
1126 revalidate_rule(cbdata->wx, sub);
1131 wx_run_one(struct wx *wx)
1135 if (time_msec() >= wx->next_expiration) {
1136 COVERAGE_INC(wx_expiration);
1137 wx->next_expiration = time_msec() + 1000;
1140 classifier_for_each(&wx->cls, CLS_INC_ALL, expire_rule, wx);
1142 /* XXX account_checkpoint_cb */
1145 if (wx->need_revalidate /*|| !tag_set_is_empty(&p->revalidate_set)*/) {
1146 struct revalidate_cbdata cbdata;
1148 cbdata.revalidate_all = false;
1149 cbdata.revalidate_subrules = wx->need_revalidate;
1150 //cbdata.revalidate_set = wx->revalidate_set;
1151 //tag_set_init(&wx->revalidate_set);
1152 COVERAGE_INC(wx_revalidate);
1153 classifier_for_each(&wx->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
1154 wx->need_revalidate = false;
1163 LIST_FOR_EACH (wx, struct wx, list_node, &all_wx) {
1170 wx_wait_one(struct wx *wx)
1172 xfif_port_poll_wait(wx->xfif);
1173 netdev_monitor_poll_wait(wx->netdev_monitor);
1174 if (wx->need_revalidate /*|| !tag_set_is_empty(&p->revalidate_set)*/) {
1175 poll_immediate_wake();
1176 } else if (wx->next_expiration != LLONG_MAX) {
1177 poll_timer_wait_until(wx->next_expiration);
1186 LIST_FOR_EACH (wx, struct wx, list_node, &all_wx) {
1192 static int wx_flow_flush(struct wdp *);
1195 wx_enumerate(const struct wdp_class *wdp_class, struct svec *all_wdps)
1197 struct svec names = SVEC_EMPTY_INITIALIZER;
1198 int error = xf_enumerate_names(wdp_class->type, &names);
1199 svec_move(all_wdps, &names);
1204 wx_open(const struct wdp_class *wdp_class, const char *name, bool create,
1211 ? xfif_create_and_open(name, wdp_class->type, &xfif)
1212 : xfif_open(name, wdp_class->type, &xfif));
1216 wx = xzalloc(sizeof *wx);
1217 list_push_back(&all_wx, &wx->list_node);
1218 wdp_init(&wx->wdp, wdp_class, name, 0, 0);
1220 classifier_init(&wx->cls);
1221 wx->netdev_monitor = netdev_monitor_create();
1222 port_array_init(&wx->ports);
1223 shash_init(&wx->port_by_name);
1224 wx->next_expiration = time_msec() + 1000;
1235 wx_close(struct wdp *wdp)
1237 struct wx *wx = wx_cast(wdp);
1240 xfif_close(wx->xfif);
1241 classifier_destroy(&wx->cls);
1242 netdev_monitor_destroy(wx->netdev_monitor);
1243 list_remove(&wx->list_node);
1248 wx_get_all_names(const struct wdp *wdp, struct svec *all_names)
1250 struct wx *wx = wx_cast(wdp);
1252 return xfif_get_all_names(wx->xfif, all_names);
1256 wx_destroy(struct wdp *wdp)
1258 struct wx *wx = wx_cast(wdp);
1260 return xfif_delete(wx->xfif);
1264 hton_ofp_phy_port(struct ofp_phy_port *opp)
1266 opp->port_no = htons(opp->port_no);
1267 opp->config = htonl(opp->config);
1268 opp->state = htonl(opp->state);
1269 opp->curr = htonl(opp->curr);
1270 opp->advertised = htonl(opp->advertised);
1271 opp->supported = htonl(opp->supported);
1272 opp->peer = htonl(opp->peer);
1276 wx_get_features(const struct wdp *wdp, struct ofpbuf **featuresp)
1278 struct wx *wx = wx_cast(wdp);
1279 struct ofp_switch_features *osf;
1281 unsigned int port_no;
1282 struct wdp_port *port;
1284 buf = ofpbuf_new(sizeof *osf);
1285 osf = ofpbuf_put_zeros(buf, sizeof *osf);
1287 osf->capabilities = htonl(OFPC_ARP_MATCH_IP);
1288 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1289 (1u << OFPAT_SET_VLAN_VID) |
1290 (1u << OFPAT_SET_VLAN_PCP) |
1291 (1u << OFPAT_STRIP_VLAN) |
1292 (1u << OFPAT_SET_DL_SRC) |
1293 (1u << OFPAT_SET_DL_DST) |
1294 (1u << OFPAT_SET_NW_SRC) |
1295 (1u << OFPAT_SET_NW_DST) |
1296 (1u << OFPAT_SET_NW_TOS) |
1297 (1u << OFPAT_SET_TP_SRC) |
1298 (1u << OFPAT_SET_TP_DST) |
1299 (1u << OFPAT_ENQUEUE));
1301 PORT_ARRAY_FOR_EACH (port, &wx->ports, port_no) {
1302 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1310 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
1312 struct wx_rule *rule = wx_rule_cast(cls_rule);
1313 int *n_subrules = n_subrules_;
1321 wx_get_stats(const struct wdp *wdp, struct wdp_stats *stats)
1323 struct wx *wx = wx_cast(wdp);
1324 struct xflow_stats xflow_stats;
1328 error = xfif_get_xf_stats(wx->xfif, &xflow_stats);
1331 classifier_for_each(&wx->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
1333 stats->exact.n_flows = classifier_count_exact(&wx->cls) - n_subrules;
1334 stats->exact.cur_capacity = xflow_stats.cur_capacity;
1335 stats->exact.max_capacity = MIN(WX_MAX_EXACT, xflow_stats.max_capacity);
1336 stats->exact.n_hit = xflow_stats.n_hit;
1337 stats->exact.n_missed = xflow_stats.n_missed;
1338 stats->exact.n_lost = xflow_stats.n_lost;
1340 stats->wild.n_flows = classifier_count_wild(&wx->cls);
1341 stats->wild.cur_capacity = WX_MAX_WILD;
1342 stats->wild.max_capacity = WX_MAX_WILD;
1343 stats->wild.n_hit = 0; /* XXX */
1344 stats->wild.n_missed = 0; /* XXX */
1345 stats->wild.n_lost = 0; /* XXX */
1347 stats->n_ports = xflow_stats.n_ports;
1348 stats->max_ports = xflow_stats.max_ports;
1350 stats->n_frags = xflow_stats.n_frags;
1352 stats->max_miss_queue = xflow_stats.max_miss_queue;
1353 stats->max_action_queue = xflow_stats.max_action_queue;
1354 stats->max_sflow_queue = xflow_stats.max_sflow_queue;
1360 wx_get_drop_frags(const struct wdp *wdp, bool *drop_frags)
1362 struct wx *wx = wx_cast(wdp);
1364 return xfif_get_drop_frags(wx->xfif, drop_frags);
1368 wx_set_drop_frags(struct wdp *wdp, bool drop_frags)
1370 struct wx *wx = wx_cast(wdp);
1372 return xfif_set_drop_frags(wx->xfif, drop_frags);
1376 wx_port_add(struct wdp *wdp, const char *devname,
1377 bool internal, uint16_t *port_no)
1379 struct wx *wx = wx_cast(wdp);
1380 uint16_t xflow_flags = internal ? XFLOW_PORT_INTERNAL : 0;
1381 return xfif_port_add(wx->xfif, devname, xflow_flags, port_no);
1385 wx_port_del(struct wdp *wdp, uint16_t port_no)
1387 struct wx *wx = wx_cast(wdp);
1389 return xfif_port_del(wx->xfif, port_no);
1393 wx_answer_port_query(const struct wdp_port *port, struct wdp_port *portp)
1396 wdp_port_copy(portp, port);
1404 wx_port_query_by_number(const struct wdp *wdp, uint16_t port_no,
1405 struct wdp_port *portp)
1407 struct wx *wx = wx_cast(wdp);
1408 const struct wdp_port *port;
1410 port = port_array_get(&wx->ports, ofp_port_to_xflow_port(port_no));
1411 return wx_answer_port_query(port, portp);
1415 wx_port_query_by_name(const struct wdp *wdp, const char *devname,
1416 struct wdp_port *portp)
1418 struct wx *wx = wx_cast(wdp);
1420 return wx_answer_port_query(shash_find_data(&wx->port_by_name, devname),
1425 wx_port_set_config(struct wdp *wdp, uint16_t port_no, uint32_t config)
1427 struct wx *wx = wx_cast(wdp);
1428 struct wdp_port *port;
1431 port = port_array_get(&wx->ports, ofp_port_to_xflow_port(port_no));
1435 changes = config ^ port->opp.config;
1437 if (changes & OFPPC_PORT_DOWN) {
1439 if (config & OFPPC_PORT_DOWN) {
1440 error = netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
1442 error = netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
1445 port->opp.config ^= OFPPC_PORT_DOWN;
1449 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
1450 if (changes & REVALIDATE_BITS) {
1451 COVERAGE_INC(wx_costly_flags);
1452 port->opp.config ^= changes & REVALIDATE_BITS;
1453 wx->need_revalidate = true;
1455 #undef REVALIDATE_BITS
1457 if (changes & OFPPC_NO_FLOOD) {
1458 port->opp.config ^= OFPPC_NO_FLOOD;
1459 wx_port_refresh_groups(wx);
1462 if (changes & OFPPC_NO_PACKET_IN) {
1463 port->opp.config ^= OFPPC_NO_PACKET_IN;
1470 wx_port_list(const struct wdp *wdp, struct wdp_port **portsp, size_t *n_portsp)
1472 struct wx *wx = wx_cast(wdp);
1473 struct wdp_port *ports, *port;
1474 unsigned int port_no;
1477 *n_portsp = n_ports = port_array_count(&wx->ports);
1478 *portsp = ports = xmalloc(n_ports * sizeof *ports);
1480 PORT_ARRAY_FOR_EACH (port, &wx->ports, port_no) {
1481 wdp_port_copy(&ports[i++], port);
1483 assert(i == n_ports);
1489 wx_port_poll(const struct wdp *wdp, char **devnamep)
1491 struct wx *wx = wx_cast(wdp);
1493 return xfif_port_poll(wx->xfif, devnamep);
1497 wx_port_poll_wait(const struct wdp *wdp)
1499 struct wx *wx = wx_cast(wdp);
1501 xfif_port_poll_wait(wx->xfif);
1504 static struct wdp_rule *
1505 wx_flow_get(const struct wdp *wdp, const flow_t *flow)
1507 struct wx *wx = wx_cast(wdp);
1508 struct wx_rule *rule;
1510 rule = wx_rule_cast(classifier_find_rule_exactly(&wx->cls, flow));
1511 return rule && !wx_rule_is_hidden(rule) ? &rule->wr : NULL;
1514 static struct wdp_rule *
1515 wx_flow_match(const struct wdp *wdp, const flow_t *flow)
1517 struct wx *wx = wx_cast(wdp);
1518 struct wx_rule *rule;
1520 rule = wx_rule_cast(classifier_lookup(&wx->cls, flow));
1522 if (wx_rule_is_hidden(rule)) {
1531 struct wx_for_each_thunk_aux {
1532 wdp_flow_cb_func *client_callback;
1537 wx_for_each_thunk(struct cls_rule *cls_rule, void *aux_)
1539 struct wx_for_each_thunk_aux *aux = aux_;
1540 struct wx_rule *rule = wx_rule_cast(cls_rule);
1542 if (!wx_rule_is_hidden(rule)) {
1543 aux->client_callback(&rule->wr, aux->client_aux);
1548 wx_flow_for_each_match(const struct wdp *wdp, const flow_t *target,
1550 wdp_flow_cb_func *client_callback, void *client_aux)
1552 struct wx *wx = wx_cast(wdp);
1553 struct wx_for_each_thunk_aux aux;
1555 aux.client_callback = client_callback;
1556 aux.client_aux = client_aux;
1557 classifier_for_each_match(&wx->cls, target, include,
1558 wx_for_each_thunk, &aux);
1561 /* Obtains statistic counters for 'rule' within 'wx' and stores them into
1562 * '*stats'. If 'rule' is a wildcarded rule, the returned statistic include
1563 * statistics for all of 'rule''s subrules. */
1565 query_stats(struct wx *wx, struct wx_rule *rule, struct wdp_flow_stats *stats)
1567 struct wx_rule *subrule;
1568 struct xflow_flow *xflow_flows;
1569 size_t n_xflow_flows;
1571 /* Start from historical data for 'rule' itself that are no longer tracked
1572 * by the datapath. This counts, for example, subrules that have
1574 stats->n_packets = rule->packet_count;
1575 stats->n_bytes = rule->byte_count;
1576 stats->inserted = rule->wr.created;
1577 stats->used = LLONG_MIN;
1578 stats->tcp_flags = 0;
1581 /* Prepare to ask the datapath for statistics on 'rule', or if it is
1582 * wildcarded then on all of its subrules.
1584 * Also, add any statistics that are not tracked by the datapath for each
1585 * subrule. This includes, for example, statistics for packets that were
1586 * executed "by hand" by ofproto via xfif_execute() but must be accounted
1588 n_xflow_flows = rule->wr.cr.flow.wildcards ? list_size(&rule->list) : 1;
1589 xflow_flows = xzalloc(n_xflow_flows * sizeof *xflow_flows);
1590 if (rule->wr.cr.flow.wildcards) {
1592 LIST_FOR_EACH (subrule, struct wx_rule, list, &rule->list) {
1593 xflow_key_from_flow(&xflow_flows[i++].key, &subrule->wr.cr.flow);
1594 stats->n_packets += subrule->packet_count;
1595 stats->n_bytes += subrule->byte_count;
1598 xflow_key_from_flow(&xflow_flows[0].key, &rule->wr.cr.flow);
1601 /* Fetch up-to-date statistics from the datapath and add them in. */
1602 if (!xfif_flow_get_multiple(wx->xfif, xflow_flows, n_xflow_flows)) {
1604 for (i = 0; i < n_xflow_flows; i++) {
1605 struct xflow_flow *xflow_flow = &xflow_flows[i];
1608 stats->n_packets += xflow_flow->stats.n_packets;
1609 stats->n_bytes += xflow_flow->stats.n_bytes;
1610 used = xflow_flow_stats_to_msec(&xflow_flow->stats);
1611 if (used > stats->used) {
1613 if (xflow_flow->key.dl_type == htons(ETH_TYPE_IP)
1614 && xflow_flow->key.nw_proto == IP_TYPE_TCP) {
1615 stats->ip_tos = xflow_flow->stats.ip_tos;
1618 stats->tcp_flags |= xflow_flow->stats.tcp_flags;
1625 wx_flow_get_stats(const struct wdp *wdp,
1626 const struct wdp_rule *wdp_rule,
1627 struct wdp_flow_stats *stats)
1629 struct wx *wx = wx_cast(wdp);
1630 struct wx_rule *rule = wx_rule_cast(&wdp_rule->cr);
1632 query_stats(wx, rule, stats);
1637 wx_flow_overlaps(const struct wdp *wdp, const flow_t *flow)
1639 struct wx *wx = wx_cast(wdp);
1641 /* XXX overlap with a subrule? */
1642 return classifier_rule_overlaps(&wx->cls, flow);
1646 wx_flow_put(struct wdp *wdp, const struct wdp_flow_put *put,
1647 struct wdp_flow_stats *old_stats, struct wdp_rule **rulep)
1649 struct wx *wx = wx_cast(wdp);
1650 struct wx_rule *rule;
1652 rule = wx_rule_cast(classifier_find_rule_exactly(&wx->cls, put->flow));
1653 if (rule && wx_rule_is_hidden(rule)) {
1658 if (!(put->flags & WDP_PUT_MODIFY)) {
1662 if (!(put->flags & WDP_PUT_CREATE)) {
1665 if ((put->flow->wildcards
1666 ? classifier_count_wild(&wx->cls) >= WX_MAX_WILD
1667 : classifier_count_exact(&wx->cls) >= WX_MAX_EXACT)) {
1668 /* XXX subrules should not count against exact-match limit */
1673 rule = wx_rule_create(NULL, put->actions, put->n_actions,
1674 put->idle_timeout, put->hard_timeout);
1675 cls_rule_from_flow(put->flow, &rule->wr.cr);
1676 wx_rule_insert(wx, rule, NULL, 0);
1680 memset(old_stats, 0, sizeof *old_stats);
1690 wx_flow_delete(struct wdp *wdp, struct wdp_rule *wdp_rule,
1691 struct wdp_flow_stats *final_stats)
1693 struct wx *wx = wx_cast(wdp);
1694 struct wx_rule *rule = wx_rule_cast(&wdp_rule->cr);
1696 wx_rule_remove(wx, rule);
1698 memset(final_stats, 0, sizeof *final_stats); /* XXX */
1704 wx_flush_rule(struct cls_rule *cls_rule, void *wx_)
1706 struct wx_rule *rule = wx_rule_cast(cls_rule);
1707 struct wx *wx = wx_;
1709 /* Mark the flow as not installed, even though it might really be
1710 * installed, so that wx_rule_remove() doesn't bother trying to uninstall
1711 * it. There is no point in uninstalling it individually since we are
1712 * about to blow away all the flows with xfif_flow_flush(). */
1713 rule->installed = false;
1715 wx_rule_remove(wx, rule);
1719 wx_flow_flush(struct wdp *wdp)
1721 struct wx *wx = wx_cast(wdp);
1723 COVERAGE_INC(wx_flow_flush);
1724 classifier_for_each(&wx->cls, CLS_INC_ALL, wx_flush_rule, wx);
1725 xfif_flow_flush(wx->xfif);
1730 wx_execute(struct wdp *wdp, uint16_t in_port,
1731 const union ofp_action actions[], int n_actions,
1732 const struct ofpbuf *packet)
1734 struct wx *wx = wx_cast(wdp);
1735 struct xflow_actions xflow_actions;
1739 flow_extract((struct ofpbuf *) packet, 0, in_port, &flow);
1740 error = wx_xlate_actions(wx, actions, n_actions, &flow, packet,
1741 &xflow_actions, NULL);
1745 xfif_execute(wx->xfif, ofp_port_to_xflow_port(in_port),
1746 xflow_actions.actions, xflow_actions.n_actions, packet);
1751 wx_flow_inject(struct wdp *wdp, struct wdp_rule *wdp_rule,
1752 uint16_t in_port, const struct ofpbuf *packet)
1754 struct wx_rule *rule = wx_rule_cast(&wdp_rule->cr);
1757 error = wx_execute(wdp, in_port, rule->wr.actions, rule->wr.n_actions,
1760 rule->packet_count++;
1761 rule->byte_count += packet->size;
1762 rule->used = time_msec();
1768 wx_recv_get_mask(const struct wdp *wdp, int *listen_mask)
1770 struct wx *wx = wx_cast(wdp);
1771 int xflow_listen_mask;
1774 error = xfif_recv_get_mask(wx->xfif, &xflow_listen_mask);
1777 if (xflow_listen_mask & XFLOWL_MISS) {
1778 *listen_mask |= 1 << WDP_CHAN_MISS;
1780 if (xflow_listen_mask & XFLOWL_ACTION) {
1781 *listen_mask |= 1 << WDP_CHAN_ACTION;
1783 if (xflow_listen_mask & XFLOWL_SFLOW) {
1784 *listen_mask |= 1 << WDP_CHAN_SFLOW;
1791 wx_recv_set_mask(struct wdp *wdp, int listen_mask)
1793 struct wx *wx = wx_cast(wdp);
1794 int xflow_listen_mask;
1796 xflow_listen_mask = 0;
1797 if (listen_mask & (1 << WDP_CHAN_MISS)) {
1798 xflow_listen_mask |= XFLOWL_MISS;
1800 if (listen_mask & (1 << WDP_CHAN_ACTION)) {
1801 xflow_listen_mask |= XFLOWL_ACTION;
1803 if (listen_mask & (1 << WDP_CHAN_SFLOW)) {
1804 xflow_listen_mask |= XFLOWL_SFLOW;
1807 return xfif_recv_set_mask(wx->xfif, xflow_listen_mask);
1811 wx_get_sflow_probability(const struct wdp *wdp, uint32_t *probability)
1813 struct wx *wx = wx_cast(wdp);
1815 return xfif_get_sflow_probability(wx->xfif, probability);
1819 wx_set_sflow_probability(struct wdp *wdp, uint32_t probability)
1821 struct wx *wx = wx_cast(wdp);
1823 return xfif_set_sflow_probability(wx->xfif, probability);
1827 wx_translate_xflow_msg(struct xflow_msg *msg, struct ofpbuf *payload,
1828 struct wdp_packet *packet)
1830 packet->in_port = xflow_port_to_ofp_port(msg->port);
1831 packet->send_len = 0;
1834 switch (msg->type) {
1835 case _XFLOWL_MISS_NR:
1836 packet->channel = WDP_CHAN_MISS;
1837 packet->payload = payload;
1838 packet->tun_id = msg->arg;
1841 case _XFLOWL_ACTION_NR:
1842 packet->channel = WDP_CHAN_ACTION;
1843 packet->payload = payload;
1844 packet->send_len = msg->arg;
1847 case _XFLOWL_SFLOW_NR:
1849 ofpbuf_delete(payload);
1853 VLOG_WARN_RL(&rl, "received XFLOW message of unexpected type %"PRIu32,
1855 ofpbuf_delete(payload);
1860 static const uint8_t *
1861 get_local_mac(const struct wx *wx)
1863 const struct wdp_port *port = port_array_get(&wx->ports, XFLOWP_LOCAL);
1864 return port ? port->opp.hw_addr : NULL;
1867 /* Returns true if 'packet' is a DHCP reply to the local port. Such a reply
1868 * should be sent to the local port regardless of the flow table.
1870 * We only need this, strictly speaking, when in-band control is turned on. */
1872 wx_is_local_dhcp_reply(const struct wx *wx,
1873 const flow_t *flow, const struct ofpbuf *packet)
1875 if (flow->dl_type == htons(ETH_TYPE_IP)
1876 && flow->nw_proto == IP_TYPE_UDP
1877 && flow->tp_src == htons(DHCP_SERVER_PORT)
1878 && flow->tp_dst == htons(DHCP_CLIENT_PORT)
1881 const uint8_t *local_mac = get_local_mac(wx);
1882 struct dhcp_header *dhcp = ofpbuf_at(
1883 packet, (char *)packet->l7 - (char *)packet->data, sizeof *dhcp);
1884 return dhcp && local_mac && eth_addr_equals(dhcp->chaddr, local_mac);
1891 wx_explode_rule(struct wx *wx, struct xflow_msg *msg, struct ofpbuf *payload)
1893 struct wx_rule *rule;
1896 flow_extract(payload, 0, xflow_port_to_ofp_port(msg->port), &flow);
1898 if (wx_is_local_dhcp_reply(wx, &flow, payload)) {
1899 union xflow_action action;
1901 memset(&action, 0, sizeof(action));
1902 action.output.type = XFLOWAT_OUTPUT;
1903 action.output.port = XFLOWP_LOCAL;
1904 xfif_execute(wx->xfif, msg->port, &action, 1, payload);
1907 rule = wx_rule_lookup_valid(wx, &flow);
1912 if (rule->wr.cr.flow.wildcards) {
1913 rule = wx_rule_create_subrule(wx, rule, &flow);
1914 wx_rule_make_actions(wx, rule, payload);
1916 if (!rule->may_install) {
1917 /* The rule is not installable, that is, we need to process every
1918 * packet, so process the current packet and set its actions into
1920 wx_rule_make_actions(wx, rule, payload);
1922 /* XXX revalidate rule if it needs it */
1926 wx_rule_execute(wx, rule, payload, &flow);
1927 wx_rule_reinstall(wx, rule);
1933 wx_recv(struct wdp *wdp, struct wdp_packet *packet)
1935 struct wx *wx = wx_cast(wdp);
1938 /* XXX need to avoid 50*50 potential cost for caller. */
1939 for (i = 0; i < 50; i++) {
1940 struct xflow_msg *msg;
1944 error = xfif_recv(wx->xfif, &buf);
1949 msg = ofpbuf_pull(buf, sizeof *msg);
1950 if (msg->type != _XFLOWL_MISS_NR || !wx_explode_rule(wx, msg, buf)) {
1951 return wx_translate_xflow_msg(msg, buf, packet);
1959 wx_recv_wait(struct wdp *wdp)
1961 struct wx *wx = wx_cast(wdp);
1963 xfif_recv_wait(wx->xfif);
1966 static void wx_port_update(struct wx *, const char *devname);
1967 static void wx_port_reinit(struct wx *);
1970 wx_port_process_change(struct wx *wx, int error, char *devname)
1972 if (error == ENOBUFS) {
1974 } else if (!error) {
1975 wx_port_update(wx, devname);
1981 wx_port_run(struct wx *wx)
1986 while ((error = xfif_port_poll(wx->xfif, &devname)) != EAGAIN) {
1987 wx_port_process_change(wx, error, devname);
1989 while ((error = netdev_monitor_poll(wx->netdev_monitor,
1990 &devname)) != EAGAIN) {
1991 wx_port_process_change(wx, error, devname);
1996 wx_port_refresh_group(struct wx *wx, unsigned int group)
2000 struct wdp_port *port;
2001 unsigned int port_no;
2003 assert(group == WX_GROUP_ALL || group == WX_GROUP_FLOOD);
2005 ports = xmalloc(port_array_count(&wx->ports) * sizeof *ports);
2007 PORT_ARRAY_FOR_EACH (port, &wx->ports, port_no) {
2008 if (group == WX_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
2009 ports[n_ports++] = port_no;
2012 xfif_port_group_set(wx->xfif, group, ports, n_ports);
2019 wx_port_refresh_groups(struct wx *wx)
2021 wx_port_refresh_group(wx, WX_GROUP_FLOOD);
2022 wx_port_refresh_group(wx, WX_GROUP_ALL);
2026 wx_port_reinit(struct wx *wx)
2028 struct svec devnames;
2029 struct wdp_port *wdp_port;
2030 unsigned int port_no;
2031 struct xflow_port *xflow_ports;
2032 size_t n_xflow_ports;
2035 svec_init(&devnames);
2036 PORT_ARRAY_FOR_EACH (wdp_port, &wx->ports, port_no) {
2037 svec_add (&devnames, (char *) wdp_port->opp.name);
2039 xfif_port_list(wx->xfif, &xflow_ports, &n_xflow_ports);
2040 for (i = 0; i < n_xflow_ports; i++) {
2041 svec_add(&devnames, xflow_ports[i].devname);
2045 svec_sort_unique(&devnames);
2046 for (i = 0; i < devnames.n; i++) {
2047 wx_port_update(wx, devnames.names[i]);
2049 svec_destroy(&devnames);
2051 wx_port_refresh_groups(wx);
2054 static struct wdp_port *
2055 make_wdp_port(const struct xflow_port *xflow_port)
2057 struct netdev_options netdev_options;
2058 enum netdev_flags flags;
2059 struct wdp_port *wdp_port;
2060 struct netdev *netdev;
2064 memset(&netdev_options, 0, sizeof netdev_options);
2065 netdev_options.name = xflow_port->devname;
2066 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
2068 error = netdev_open(&netdev_options, &netdev);
2070 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
2071 "cannot be opened (%s)",
2072 xflow_port->devname, xflow_port->port,
2073 xflow_port->devname, strerror(error));
2077 wdp_port = xmalloc(sizeof *wdp_port);
2078 wdp_port->netdev = netdev;
2079 wdp_port->opp.port_no = xflow_port_to_ofp_port(xflow_port->port);
2080 netdev_get_etheraddr(netdev, wdp_port->opp.hw_addr);
2081 strncpy((char *) wdp_port->opp.name, xflow_port->devname,
2082 sizeof wdp_port->opp.name);
2083 wdp_port->opp.name[sizeof wdp_port->opp.name - 1] = '\0';
2085 netdev_get_flags(netdev, &flags);
2086 wdp_port->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
2088 netdev_get_carrier(netdev, &carrier);
2089 wdp_port->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
2091 netdev_get_features(netdev,
2092 &wdp_port->opp.curr, &wdp_port->opp.advertised,
2093 &wdp_port->opp.supported, &wdp_port->opp.peer);
2095 wdp_port->devname = xstrdup(xflow_port->devname);
2096 wdp_port->internal = (xflow_port->flags & XFLOW_PORT_INTERNAL) != 0;
2101 wx_port_conflicts(const struct wx *wx, const struct xflow_port *xflow_port)
2103 if (port_array_get(&wx->ports, xflow_port->port)) {
2104 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
2107 } else if (shash_find(&wx->port_by_name, xflow_port->devname)) {
2108 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
2109 xflow_port->devname);
2117 wdp_port_equal(const struct wdp_port *a_, const struct wdp_port *b_)
2119 const struct ofp_phy_port *a = &a_->opp;
2120 const struct ofp_phy_port *b = &b_->opp;
2122 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
2123 return (a->port_no == b->port_no
2124 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
2125 && !strcmp((char *) a->name, (char *) b->name)
2126 && a->state == b->state
2127 && a->config == b->config
2128 && a->curr == b->curr
2129 && a->advertised == b->advertised
2130 && a->supported == b->supported
2131 && a->peer == b->peer);
2135 wx_port_install(struct wx *wx, struct wdp_port *wdp_port)
2137 uint16_t xflow_port = ofp_port_to_xflow_port(wdp_port->opp.port_no);
2138 const char *netdev_name = (const char *) wdp_port->opp.name;
2140 netdev_monitor_add(wx->netdev_monitor, wdp_port->netdev);
2141 port_array_set(&wx->ports, xflow_port, wdp_port);
2142 shash_add(&wx->port_by_name, netdev_name, wdp_port);
2146 wx_port_remove(struct wx *wx, struct wdp_port *wdp_port)
2148 uint16_t xflow_port = ofp_port_to_xflow_port(wdp_port->opp.port_no);
2150 netdev_monitor_remove(wx->netdev_monitor, wdp_port->netdev);
2151 port_array_delete(&wx->ports, xflow_port);
2152 shash_delete(&wx->port_by_name,
2153 shash_find(&wx->port_by_name, (char *) wdp_port->opp.name));
2157 wx_port_free(struct wdp_port *wdp_port)
2160 netdev_close(wdp_port->netdev);
2166 wx_port_update(struct wx *wx, const char *devname)
2168 struct xflow_port xflow_port;
2169 struct wdp_port *old_wdp_port;
2170 struct wdp_port *new_wdp_port;
2173 COVERAGE_INC(wx_update_port);
2175 /* Query the datapath for port information. */
2176 error = xfif_port_query_by_name(wx->xfif, devname, &xflow_port);
2178 /* Find the old wdp_port. */
2179 old_wdp_port = shash_find_data(&wx->port_by_name, devname);
2181 if (!old_wdp_port) {
2182 /* There's no port named 'devname' but there might be a port with
2183 * the same port number. This could happen if a port is deleted
2184 * and then a new one added in its place very quickly, or if a port
2185 * is renamed. In the former case we want to send an OFPPR_DELETE
2186 * and an OFPPR_ADD, and in the latter case we want to send a
2187 * single OFPPR_MODIFY. We can distinguish the cases by comparing
2188 * the old port's ifindex against the new port, or perhaps less
2189 * reliably but more portably by comparing the old port's MAC
2190 * against the new port's MAC. However, this code isn't that smart
2191 * and always sends an OFPPR_MODIFY (XXX). */
2192 old_wdp_port = port_array_get(&wx->ports, xflow_port.port);
2194 } else if (error != ENOENT && error != ENODEV) {
2195 VLOG_WARN_RL(&rl, "xfif_port_query_by_name returned unexpected error "
2196 "%s", strerror(error));
2200 /* Create a new wdp_port. */
2201 new_wdp_port = !error ? make_wdp_port(&xflow_port) : NULL;
2203 /* Eliminate a few pathological cases. */
2204 if (!old_wdp_port && !new_wdp_port) {
2206 } else if (old_wdp_port && new_wdp_port) {
2207 /* Most of the 'config' bits are OpenFlow soft state, but
2208 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
2209 * OpenFlow bits from old_wdp_port. (make_wdp_port() only sets
2210 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
2211 new_wdp_port->opp.config |= old_wdp_port->opp.config & ~OFPPC_PORT_DOWN;
2213 if (wdp_port_equal(old_wdp_port, new_wdp_port)) {
2214 /* False alarm--no change. */
2215 wx_port_free(new_wdp_port);
2220 /* Now deal with the normal cases. */
2222 wx_port_remove(wx, old_wdp_port);
2225 wx_port_install(wx, new_wdp_port);
2227 wx_port_free(old_wdp_port);
2229 /* Update port groups. */
2230 wx_port_refresh_groups(wx);
2234 wx_port_init(struct wx *wx)
2236 struct xflow_port *ports;
2241 error = xfif_port_list(wx->xfif, &ports, &n_ports);
2246 for (i = 0; i < n_ports; i++) {
2247 const struct xflow_port *xflow_port = &ports[i];
2248 if (!wx_port_conflicts(wx, xflow_port)) {
2249 struct wdp_port *wdp_port = make_wdp_port(xflow_port);
2251 wx_port_install(wx, wdp_port);
2256 wx_port_refresh_groups(wx);
2261 wdp_xflow_register(void)
2263 static const struct wdp_class wdp_xflow_class = {
2278 wx_port_query_by_number,
2279 wx_port_query_by_name,
2286 wx_flow_for_each_match,
2296 wx_get_sflow_probability,
2297 wx_set_sflow_probability,
2302 static bool inited = false;
2315 xf_enumerate_types(&types);
2318 SVEC_FOR_EACH (i, type, &types) {
2319 struct wdp_class *class;
2321 class = xmalloc(sizeof *class);
2322 *class = wdp_xflow_class;
2323 class->type = xstrdup(type);
2328 if (!wdp_register_provider(class)) {
2333 svec_destroy(&types);