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"
29 #include "openflow/nicira-ext.h"
30 #include "openflow/openflow.h"
32 #include "poll-loop.h"
33 #include "port-array.h"
40 #include "wdp-provider.h"
42 #include "xflow-util.h"
45 #define THIS_MODULE VLM_wdp_xflow
48 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
50 /* Maximum numbers of rules. */
51 #define WX_MAX_WILD 65536 /* Wildcarded rules. */
52 #define WX_MAX_EXACT 1048576 /* Exact-match rules. */
55 struct list list_node;
58 struct classifier cls;
59 struct netdev_monitor *netdev_monitor;
60 struct port_array ports; /* Index is ODP port nr; wdp_port->opp.port_no
62 struct shash port_by_name;
64 long long int next_expiration;
67 static struct list all_wx = LIST_INITIALIZER(&all_wx);
69 static int wx_port_init(struct wx *);
70 static void wx_port_run(struct wx *);
71 static void wx_port_refresh_groups(struct wx *);
79 wx_cast(const struct wdp *wdp)
81 return CONTAINER_OF(wdp, struct wx, wdp);
85 wx_xlate_actions(struct wx *, const union ofp_action *, size_t n,
86 const flow_t *flow, const struct ofpbuf *packet,
87 struct xflow_actions *out, bool *may_set_up_flow);
92 uint64_t packet_count; /* Number of packets received. */
93 uint64_t byte_count; /* Number of bytes received. */
94 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
95 long long int used; /* Last-used time (0 if never used). */
97 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
98 * exact-match rule (having cr.wc.wildcards of 0) generated from the
99 * wildcard rule 'super'. In this case, 'list' is an element of the
102 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
103 * a list of subrules. A super-rule with no wildcards (where
104 * cr.wc.wildcards is 0) will never have any subrules. */
105 struct wx_rule *super;
110 * A super-rule with wildcard fields never has XFLOW actions (since the
111 * datapath only supports exact-match flows). */
112 bool installed; /* Installed in datapath? */
113 bool may_install; /* True ordinarily; false if actions must
114 * be reassessed for every packet. */
116 union xflow_action *xflow_actions;
119 static void wx_rule_destroy(struct wx *, struct wx_rule *);
120 static void wx_rule_update_actions(struct wx *, struct wx_rule *);
121 static void wx_rule_execute(struct wx *, struct wx_rule *,
122 struct ofpbuf *packet, const flow_t *);
123 static bool wx_rule_make_actions(struct wx *, struct wx_rule *,
124 const struct ofpbuf *packet);
125 static void wx_rule_install(struct wx *, struct wx_rule *,
126 struct wx_rule *displaced_rule);
128 static struct wx_rule *
129 wx_rule_cast(const struct cls_rule *cls_rule)
131 return cls_rule ? CONTAINER_OF(cls_rule, struct wx_rule, wr.cr) : NULL;
134 /* Returns true if 'rule' is merely an implementation detail that should be
135 * hidden from the client. */
137 wx_rule_is_hidden(const struct wx_rule *rule)
139 return rule->super != NULL;
143 wx_rule_free(struct wx_rule *rule)
145 wdp_rule_uninit(&rule->wr);
146 free(rule->xflow_actions);
151 wx_rule_account(struct wx *wx OVS_UNUSED, struct wx_rule *rule OVS_UNUSED,
152 uint64_t extra_bytes OVS_UNUSED)
154 /* XXX call account_cb hook */
158 wx_rule_post_uninstall(struct wx *wx, struct wx_rule *rule)
160 struct wx_rule *super = rule->super;
162 wx_rule_account(wx, rule, 0);
164 /* XXX netflow expiration */
167 super->packet_count += rule->packet_count;
168 super->byte_count += rule->byte_count;
170 /* Reset counters to prevent double counting if the rule ever gets
172 rule->packet_count = 0;
173 rule->byte_count = 0;
174 rule->accounted_bytes = 0;
176 //XXX netflow_flow_clear(&rule->nf_flow);
181 xflow_flow_stats_to_msec(const struct xflow_flow_stats *stats)
183 return (stats->used_sec
184 ? stats->used_sec * 1000 + stats->used_nsec / 1000000
189 wx_rule_update_time(struct wx *wx OVS_UNUSED, struct wx_rule *rule,
190 const struct xflow_flow_stats *stats)
192 long long int used = xflow_flow_stats_to_msec(stats);
193 if (used > rule->used) {
195 if (rule->super && used > rule->super->used) {
196 rule->super->used = used;
198 //XXX netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
203 wx_rule_update_stats(struct wx *wx, struct wx_rule *rule,
204 const struct xflow_flow_stats *stats)
206 if (stats->n_packets) {
207 wx_rule_update_time(wx, rule, stats);
208 rule->packet_count += stats->n_packets;
209 rule->byte_count += stats->n_bytes;
210 /* XXX netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
211 stats->tcp_flags); */
216 wx_rule_uninstall(struct wx *wx, struct wx_rule *rule)
218 assert(!rule->wr.cr.flow.wildcards);
219 if (rule->installed) {
220 struct xflow_flow xflow_flow;
222 xflow_key_from_flow(&xflow_flow.key, &rule->wr.cr.flow);
223 xflow_flow.actions = NULL;
224 xflow_flow.n_actions = 0;
225 xflow_flow.flags = 0;
226 if (!xfif_flow_del(wx->xfif, &xflow_flow)) {
227 wx_rule_update_stats(wx, rule, &xflow_flow.stats);
229 rule->installed = false;
231 wx_rule_post_uninstall(wx, rule);
237 is_controller_rule(struct wx_rule *rule)
239 /* If the only action is send to the controller then don't report
240 * NetFlow expiration messages since it is just part of the control
241 * logic for the network and not real traffic. */
243 if (rule && rule->super) {
244 struct wdp_rule *super = &rule->super->wr;
246 return super->n_actions == 1 &&
247 super->actions[0].type == htons(OFPAT_OUTPUT) &&
248 super->actions[0].output.port == 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, 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(&subrule->wr.cr, flow);
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 = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
599 struct wx_xlate_ctx {
601 const 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 struct wx_rule *rule;
667 flow.in_port = in_port;
669 rule = wx_rule_lookup_valid(ctx->wx, &flow);
676 do_xlate_actions(rule->wr.actions, rule->wr.n_actions, ctx);
683 xlate_output_action(struct wx_xlate_ctx *ctx,
684 const struct ofp_action_output *oao)
687 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
689 ctx->nf_output_iface = NF_OUT_DROP;
691 switch (ntohs(oao->port)) {
693 add_output_action(ctx, ctx->flow->in_port);
696 xlate_table_action(ctx, ctx->flow->in_port);
700 if (!ctx->wx->ofhooks->normal_cb(ctx->flow, ctx->packet,
702 &ctx->nf_output_iface,
704 COVERAGE_INC(wx_uninstallable);
705 ctx->may_set_up_flow = false;
709 /* fall through to flood for now */
712 add_output_group_action(ctx->out, WX_GROUP_FLOOD,
713 &ctx->nf_output_iface);
716 add_output_group_action(ctx->out, WX_GROUP_ALL, &ctx->nf_output_iface);
718 case OFPP_CONTROLLER:
719 add_controller_action(ctx->out, oao);
722 add_output_action(ctx, XFLOWP_LOCAL);
725 xflow_port = ofp_port_to_xflow_port(ntohs(oao->port));
726 if (xflow_port != ctx->flow->in_port) {
727 add_output_action(ctx, xflow_port);
732 if (prev_nf_output_iface == NF_OUT_FLOOD) {
733 ctx->nf_output_iface = NF_OUT_FLOOD;
734 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
735 ctx->nf_output_iface = prev_nf_output_iface;
736 } else if (prev_nf_output_iface != NF_OUT_DROP &&
737 ctx->nf_output_iface != NF_OUT_FLOOD) {
738 ctx->nf_output_iface = NF_OUT_MULTI;
743 xlate_nicira_action(struct wx_xlate_ctx *ctx,
744 const struct nx_action_header *nah)
746 const struct nx_action_resubmit *nar;
747 int subtype = ntohs(nah->subtype);
749 assert(nah->vendor == htonl(NX_VENDOR_ID));
752 nar = (const struct nx_action_resubmit *) nah;
753 xlate_table_action(ctx, ofp_port_to_xflow_port(ntohs(nar->in_port)));
757 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
763 do_xlate_actions(const union ofp_action *in, size_t n_in,
764 struct wx_xlate_ctx *ctx)
766 struct actions_iterator iter;
767 const union ofp_action *ia;
768 const struct wdp_port *port;
770 port = port_array_get(&ctx->wx->ports, ctx->flow->in_port);
771 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
772 port->opp.config & (eth_addr_equals(ctx->flow->dl_dst, stp_eth_addr)
773 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
774 /* Drop this flow. */
778 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
779 uint16_t type = ntohs(ia->type);
780 union xflow_action *oa;
784 xlate_output_action(ctx, &ia->output);
787 case OFPAT_SET_VLAN_VID:
788 oa = xflow_actions_add(ctx->out, XFLOWAT_SET_DL_TCI);
789 oa->dl_tci.tci = ia->vlan_vid.vlan_vid & htons(VLAN_VID_MASK);
790 oa->dl_tci.mask = htons(VLAN_VID_MASK);
793 case OFPAT_SET_VLAN_PCP:
794 oa = xflow_actions_add(ctx->out, XFLOWAT_SET_DL_TCI);
795 oa->dl_tci.tci = htons((ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT)
797 oa->dl_tci.mask = htons(VLAN_PCP_MASK);
800 case OFPAT_STRIP_VLAN:
801 xflow_actions_add(ctx->out, XFLOWAT_STRIP_VLAN);
804 case OFPAT_SET_DL_SRC:
805 oa = xflow_actions_add(ctx->out, XFLOWAT_SET_DL_SRC);
806 memcpy(oa->dl_addr.dl_addr,
807 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
810 case OFPAT_SET_DL_DST:
811 oa = xflow_actions_add(ctx->out, XFLOWAT_SET_DL_DST);
812 memcpy(oa->dl_addr.dl_addr,
813 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
816 case OFPAT_SET_NW_SRC:
817 oa = xflow_actions_add(ctx->out, XFLOWAT_SET_NW_SRC);
818 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
821 case OFPAT_SET_NW_DST:
822 oa = xflow_actions_add(ctx->out, XFLOWAT_SET_NW_DST);
823 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
826 case OFPAT_SET_NW_TOS:
827 oa = xflow_actions_add(ctx->out, XFLOWAT_SET_NW_TOS);
828 oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
831 case OFPAT_SET_TP_SRC:
832 oa = xflow_actions_add(ctx->out, XFLOWAT_SET_TP_SRC);
833 oa->tp_port.tp_port = ia->tp_port.tp_port;
836 case OFPAT_SET_TP_DST:
837 oa = xflow_actions_add(ctx->out, XFLOWAT_SET_TP_DST);
838 oa->tp_port.tp_port = ia->tp_port.tp_port;
842 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
846 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
852 /* Returns true if 'flow' and 'actions' may be set up as a flow in the kernel.
853 * This is true most of the time, but we don't allow flows that would prevent
854 * DHCP replies from being seen by the local port to be set up in the
857 * We only need this, strictly speaking, when in-band control is turned on. */
859 wx_may_set_up(const flow_t *flow, const struct xflow_actions *actions)
861 if (flow->dl_type == htons(ETH_TYPE_IP)
862 && flow->nw_proto == IP_TYPE_UDP
863 && flow->tp_src == htons(DHCP_SERVER_PORT)
864 && flow->tp_dst == htons(DHCP_CLIENT_PORT)) {
867 for (i = 0; i < actions->n_actions; i++) {
868 const struct xflow_action_output *oao = &actions->actions[i].output;
869 if (oao->type == XFLOWAT_OUTPUT && oao->port == XFLOWP_LOCAL) {
880 wx_xlate_actions(struct wx *wx, const union ofp_action *in, size_t n_in,
881 const flow_t *flow, const struct ofpbuf *packet,
882 struct xflow_actions *out, bool *may_set_up_flow)
884 //tag_type no_tags = 0;
885 struct wx_xlate_ctx ctx;
886 COVERAGE_INC(wx_ofp2xflow);
887 xflow_actions_init(out);
893 //ctx.tags = tags ? tags : &no_tags;
894 ctx.may_set_up_flow = true;
895 ctx.nf_output_iface = NF_OUT_DROP;
896 do_xlate_actions(in, n_in, &ctx);
898 if (may_set_up_flow) {
899 *may_set_up_flow = ctx.may_set_up_flow && wx_may_set_up(flow, out);
902 if (nf_output_iface) {
903 *nf_output_iface = ctx.nf_output_iface;
906 if (xflow_actions_overflow(out)) {
907 xflow_actions_init(out);
908 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
914 update_used(struct wx *wx)
916 struct xflow_flow *flows;
921 error = xfif_flow_list_all(wx->xfif, &flows, &n_flows);
926 for (i = 0; i < n_flows; i++) {
927 struct xflow_flow *f = &flows[i];
928 struct wx_rule *rule;
931 xflow_key_to_flow(&f->key, &flow);
932 rule = wx_rule_cast(classifier_find_rule_exactly(&wx->cls, &flow));
933 if (!rule || !rule->installed) {
934 COVERAGE_INC(wx_unexpected_rule);
935 xfif_flow_del(wx->xfif, f);
939 wx_rule_update_time(wx, rule, &f->stats);
940 wx_rule_account(wx, rule, f->stats.n_bytes);
946 uninstall_idle_flow(struct wx *wx, struct wx_rule *rule)
948 assert(rule->installed);
949 assert(!rule->wr.cr.flow.wildcards);
952 wx_rule_remove(wx, rule);
954 wx_rule_uninstall(wx, rule);
959 expire_rule(struct cls_rule *cls_rule, void *wx_)
962 struct wx_rule *rule = wx_rule_cast(cls_rule);
963 long long int hard_expire, idle_expire, expire, now;
965 hard_expire = (rule->wr.hard_timeout
966 ? rule->wr.created + rule->wr.hard_timeout * 1000
968 idle_expire = (rule->wr.idle_timeout
969 && (rule->super || list_is_empty(&rule->list))
970 ? rule->used + rule->wr.idle_timeout * 1000
972 expire = MIN(hard_expire, idle_expire);
976 if (rule->installed && now >= rule->used + 5000) {
977 uninstall_idle_flow(wx, rule);
978 } else if (!rule->wr.cr.flow.wildcards) {
979 //XXX active_timeout(wx, rule);
985 COVERAGE_INC(wx_expired);
987 /* Update stats. This code will be a no-op if the rule expired
988 * due to an idle timeout. */
989 if (rule->wr.cr.flow.wildcards) {
990 struct wx_rule *subrule, *next;
991 LIST_FOR_EACH_SAFE (subrule, next, struct wx_rule, list, &rule->list) {
992 wx_rule_remove(wx, subrule);
995 wx_rule_uninstall(wx, rule);
999 if (!wx_rule_is_hidden(rule)) {
1000 send_flow_removed(wx, rule, now,
1002 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
1005 wx_rule_remove(wx, rule);
1008 struct revalidate_cbdata {
1010 bool revalidate_all; /* Revalidate all exact-match rules? */
1011 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
1012 //struct tag_set revalidate_set; /* Set of tags to revalidate. */
1016 revalidate_rule(struct wx *wx, struct wx_rule *rule)
1018 const flow_t *flow = &rule->wr.cr.flow;
1020 COVERAGE_INC(wx_revalidate_rule);
1022 struct wx_rule *super;
1023 super = wx_rule_cast(classifier_lookup_wild(&wx->cls, flow));
1025 wx_rule_remove(wx, rule);
1027 } else if (super != rule->super) {
1028 COVERAGE_INC(wx_revalidate_moved);
1029 list_remove(&rule->list);
1030 list_push_back(&super->list, &rule->list);
1031 rule->super = super;
1032 rule->wr.hard_timeout = super->wr.hard_timeout;
1033 rule->wr.idle_timeout = super->wr.idle_timeout;
1034 rule->wr.created = super->wr.created;
1039 wx_rule_update_actions(wx, rule);
1044 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
1046 struct wx_rule *sub = wx_rule_cast(sub_);
1047 struct revalidate_cbdata *cbdata = cbdata_;
1049 if (cbdata->revalidate_all
1050 || (cbdata->revalidate_subrules && sub->super)
1051 /*|| (tag_set_intersects(&cbdata->revalidate_set, sub->tags))*/) {
1052 revalidate_rule(cbdata->wx, sub);
1057 wx_run_one(struct wx *wx)
1061 if (time_msec() >= wx->next_expiration) {
1062 COVERAGE_INC(wx_expiration);
1063 wx->next_expiration = time_msec() + 1000;
1066 classifier_for_each(&wx->cls, CLS_INC_ALL, expire_rule, wx);
1068 /* XXX account_checkpoint_cb */
1071 if (wx->need_revalidate /*|| !tag_set_is_empty(&p->revalidate_set)*/) {
1072 struct revalidate_cbdata cbdata;
1074 cbdata.revalidate_all = false;
1075 cbdata.revalidate_subrules = wx->need_revalidate;
1076 //cbdata.revalidate_set = wx->revalidate_set;
1077 //tag_set_init(&wx->revalidate_set);
1078 COVERAGE_INC(wx_revalidate);
1079 classifier_for_each(&wx->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
1080 wx->need_revalidate = false;
1089 LIST_FOR_EACH (wx, struct wx, list_node, &all_wx) {
1096 wx_wait_one(struct wx *wx)
1098 xfif_port_poll_wait(wx->xfif);
1099 netdev_monitor_poll_wait(wx->netdev_monitor);
1100 if (wx->need_revalidate /*|| !tag_set_is_empty(&p->revalidate_set)*/) {
1101 poll_immediate_wake();
1102 } else if (wx->next_expiration != LLONG_MAX) {
1103 poll_timer_wait(wx->next_expiration - time_msec());
1112 LIST_FOR_EACH (wx, struct wx, list_node, &all_wx) {
1118 static int wx_flow_flush(struct wdp *);
1121 wx_enumerate(const struct wdp_class *wdp_class, struct svec *all_wdps)
1123 struct svec names = SVEC_EMPTY_INITIALIZER;
1124 int error = xf_enumerate_names(wdp_class->type, &names);
1125 svec_move(all_wdps, &names);
1130 wx_open(const struct wdp_class *wdp_class, const char *name, bool create,
1137 ? xfif_create_and_open(name, wdp_class->type, &xfif)
1138 : xfif_open(name, wdp_class->type, &xfif));
1142 wx = xmalloc(sizeof *wx);
1143 list_push_back(&all_wx, &wx->list_node);
1144 wdp_init(&wx->wdp, wdp_class, name, 0, 0);
1146 classifier_init(&wx->cls);
1147 wx->netdev_monitor = netdev_monitor_create();
1148 port_array_init(&wx->ports);
1149 shash_init(&wx->port_by_name);
1150 wx->next_expiration = time_msec() + 1000;
1161 wx_close(struct wdp *wdp)
1163 struct wx *wx = wx_cast(wdp);
1166 xfif_close(wx->xfif);
1167 classifier_destroy(&wx->cls);
1168 netdev_monitor_destroy(wx->netdev_monitor);
1169 list_remove(&wx->list_node);
1174 wx_get_all_names(const struct wdp *wdp, struct svec *all_names)
1176 struct wx *wx = wx_cast(wdp);
1178 return xfif_get_all_names(wx->xfif, all_names);
1182 wx_destroy(struct wdp *wdp)
1184 struct wx *wx = wx_cast(wdp);
1186 return xfif_delete(wx->xfif);
1190 hton_ofp_phy_port(struct ofp_phy_port *opp)
1192 opp->port_no = htons(opp->port_no);
1193 opp->config = htonl(opp->config);
1194 opp->state = htonl(opp->state);
1195 opp->curr = htonl(opp->curr);
1196 opp->advertised = htonl(opp->advertised);
1197 opp->supported = htonl(opp->supported);
1198 opp->peer = htonl(opp->peer);
1202 wx_get_features(const struct wdp *wdp, struct ofpbuf **featuresp)
1204 struct wx *wx = wx_cast(wdp);
1205 struct ofp_switch_features *osf;
1207 unsigned int port_no;
1208 struct wdp_port *port;
1210 buf = ofpbuf_new(sizeof *osf);
1211 osf = ofpbuf_put_zeros(buf, sizeof *osf);
1213 osf->capabilities = htonl(OFPC_ARP_MATCH_IP);
1214 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1215 (1u << OFPAT_SET_VLAN_VID) |
1216 (1u << OFPAT_SET_VLAN_PCP) |
1217 (1u << OFPAT_STRIP_VLAN) |
1218 (1u << OFPAT_SET_DL_SRC) |
1219 (1u << OFPAT_SET_DL_DST) |
1220 (1u << OFPAT_SET_NW_SRC) |
1221 (1u << OFPAT_SET_NW_DST) |
1222 (1u << OFPAT_SET_NW_TOS) |
1223 (1u << OFPAT_SET_TP_SRC) |
1224 (1u << OFPAT_SET_TP_DST));
1226 PORT_ARRAY_FOR_EACH (port, &wx->ports, port_no) {
1227 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1235 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
1237 struct wx_rule *rule = wx_rule_cast(cls_rule);
1238 int *n_subrules = n_subrules_;
1246 wx_get_stats(const struct wdp *wdp, struct wdp_stats *stats)
1248 struct wx *wx = wx_cast(wdp);
1249 struct xflow_stats xflow_stats;
1253 error = xfif_get_xf_stats(wx->xfif, &xflow_stats);
1256 classifier_for_each(&wx->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
1258 stats->exact.n_flows = classifier_count_exact(&wx->cls) - n_subrules;
1259 stats->exact.cur_capacity = xflow_stats.cur_capacity;
1260 stats->exact.max_capacity = MIN(WX_MAX_EXACT, xflow_stats.max_capacity);
1261 stats->exact.n_hit = xflow_stats.n_hit;
1262 stats->exact.n_missed = xflow_stats.n_missed;
1263 stats->exact.n_lost = xflow_stats.n_lost;
1265 stats->wild.n_flows = classifier_count_wild(&wx->cls);
1266 stats->wild.cur_capacity = WX_MAX_WILD;
1267 stats->wild.max_capacity = WX_MAX_WILD;
1268 stats->wild.n_hit = 0; /* XXX */
1269 stats->wild.n_missed = 0; /* XXX */
1270 stats->wild.n_lost = 0; /* XXX */
1272 stats->n_ports = xflow_stats.n_ports;
1273 stats->max_ports = xflow_stats.max_ports;
1275 stats->n_frags = xflow_stats.n_frags;
1277 stats->max_miss_queue = xflow_stats.max_miss_queue;
1278 stats->max_action_queue = xflow_stats.max_action_queue;
1279 stats->max_sflow_queue = xflow_stats.max_sflow_queue;
1285 wx_get_drop_frags(const struct wdp *wdp, bool *drop_frags)
1287 struct wx *wx = wx_cast(wdp);
1289 return xfif_get_drop_frags(wx->xfif, drop_frags);
1293 wx_set_drop_frags(struct wdp *wdp, bool drop_frags)
1295 struct wx *wx = wx_cast(wdp);
1297 return xfif_set_drop_frags(wx->xfif, drop_frags);
1301 wx_port_add(struct wdp *wdp, const char *devname,
1302 bool internal, uint16_t *port_no)
1304 struct wx *wx = wx_cast(wdp);
1305 uint16_t xflow_flags = internal ? XFLOW_PORT_INTERNAL : 0;
1306 return xfif_port_add(wx->xfif, devname, xflow_flags, port_no);
1310 wx_port_del(struct wdp *wdp, uint16_t port_no)
1312 struct wx *wx = wx_cast(wdp);
1314 return xfif_port_del(wx->xfif, port_no);
1318 wx_answer_port_query(const struct wdp_port *port, struct wdp_port *portp)
1321 wdp_port_copy(portp, port);
1329 wx_port_query_by_number(const struct wdp *wdp, uint16_t port_no,
1330 struct wdp_port *portp)
1332 struct wx *wx = wx_cast(wdp);
1333 const struct wdp_port *port;
1335 port = port_array_get(&wx->ports, ofp_port_to_xflow_port(port_no));
1336 return wx_answer_port_query(port, portp);
1340 wx_port_query_by_name(const struct wdp *wdp, const char *devname,
1341 struct wdp_port *portp)
1343 struct wx *wx = wx_cast(wdp);
1345 return wx_answer_port_query(shash_find_data(&wx->port_by_name, devname),
1350 wx_port_set_config(struct wdp *wdp, uint16_t port_no, uint32_t config)
1352 struct wx *wx = wx_cast(wdp);
1353 struct wdp_port *port;
1356 port = port_array_get(&wx->ports, ofp_port_to_xflow_port(port_no));
1360 changes = config ^ port->opp.config;
1362 if (changes & OFPPC_PORT_DOWN) {
1364 if (config & OFPPC_PORT_DOWN) {
1365 error = netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
1367 error = netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
1370 port->opp.config ^= OFPPC_PORT_DOWN;
1374 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
1375 if (changes & REVALIDATE_BITS) {
1376 COVERAGE_INC(wx_costly_flags);
1377 port->opp.config ^= changes & REVALIDATE_BITS;
1378 wx->need_revalidate = true;
1380 #undef REVALIDATE_BITS
1382 if (changes & OFPPC_NO_FLOOD) {
1383 port->opp.config ^= OFPPC_NO_FLOOD;
1384 wx_port_refresh_groups(wx);
1387 if (changes & OFPPC_NO_PACKET_IN) {
1388 port->opp.config ^= OFPPC_NO_PACKET_IN;
1395 wx_port_list(const struct wdp *wdp, struct wdp_port **portsp, size_t *n_portsp)
1397 struct wx *wx = wx_cast(wdp);
1398 struct wdp_port *ports, *port;
1399 unsigned int port_no;
1402 *n_portsp = n_ports = port_array_count(&wx->ports);
1403 *portsp = ports = xmalloc(n_ports * sizeof *ports);
1405 PORT_ARRAY_FOR_EACH (port, &wx->ports, port_no) {
1406 wdp_port_copy(&ports[i++], port);
1408 assert(i == n_ports);
1414 wx_port_poll(const struct wdp *wdp, char **devnamep)
1416 struct wx *wx = wx_cast(wdp);
1418 return xfif_port_poll(wx->xfif, devnamep);
1422 wx_port_poll_wait(const struct wdp *wdp)
1424 struct wx *wx = wx_cast(wdp);
1426 xfif_port_poll_wait(wx->xfif);
1429 static struct wdp_rule *
1430 wx_flow_get(const struct wdp *wdp, const flow_t *flow)
1432 struct wx *wx = wx_cast(wdp);
1433 struct wx_rule *rule;
1435 rule = wx_rule_cast(classifier_find_rule_exactly(&wx->cls, flow));
1436 return rule && !wx_rule_is_hidden(rule) ? &rule->wr : NULL;
1439 static struct wdp_rule *
1440 wx_flow_match(const struct wdp *wdp, const flow_t *flow)
1442 struct wx *wx = wx_cast(wdp);
1443 struct wx_rule *rule;
1445 rule = wx_rule_cast(classifier_lookup(&wx->cls, flow));
1447 if (wx_rule_is_hidden(rule)) {
1456 struct wx_for_each_thunk_aux {
1457 wdp_flow_cb_func *client_callback;
1462 wx_for_each_thunk(struct cls_rule *cls_rule, void *aux_)
1464 struct wx_for_each_thunk_aux *aux = aux_;
1465 struct wx_rule *rule = wx_rule_cast(cls_rule);
1467 if (!wx_rule_is_hidden(rule)) {
1468 aux->client_callback(&rule->wr, aux->client_aux);
1473 wx_flow_for_each_match(const struct wdp *wdp, const flow_t *target,
1475 wdp_flow_cb_func *client_callback, void *client_aux)
1477 struct wx *wx = wx_cast(wdp);
1478 struct wx_for_each_thunk_aux aux;
1480 aux.client_callback = client_callback;
1481 aux.client_aux = client_aux;
1482 classifier_for_each_match(&wx->cls, target, include,
1483 wx_for_each_thunk, &aux);
1486 /* Obtains statistic counters for 'rule' within 'wx' and stores them into
1487 * '*stats'. If 'rule' is a wildcarded rule, the returned statistic include
1488 * statistics for all of 'rule''s subrules. */
1490 query_stats(struct wx *wx, struct wx_rule *rule, struct wdp_flow_stats *stats)
1492 struct wx_rule *subrule;
1493 struct xflow_flow *xflow_flows;
1494 size_t n_xflow_flows;
1496 /* Start from historical data for 'rule' itself that are no longer tracked
1497 * by the datapath. This counts, for example, subrules that have
1499 stats->n_packets = rule->packet_count;
1500 stats->n_bytes = rule->byte_count;
1501 stats->inserted = rule->wr.created;
1502 stats->used = LLONG_MIN;
1503 stats->tcp_flags = 0;
1506 /* Prepare to ask the datapath for statistics on 'rule', or if it is
1507 * wildcarded then on all of its subrules.
1509 * Also, add any statistics that are not tracked by the datapath for each
1510 * subrule. This includes, for example, statistics for packets that were
1511 * executed "by hand" by ofproto via xfif_execute() but must be accounted
1513 n_xflow_flows = rule->wr.cr.flow.wildcards ? list_size(&rule->list) : 1;
1514 xflow_flows = xzalloc(n_xflow_flows * sizeof *xflow_flows);
1515 if (rule->wr.cr.flow.wildcards) {
1517 LIST_FOR_EACH (subrule, struct wx_rule, list, &rule->list) {
1518 xflow_key_from_flow(&xflow_flows[i++].key, &subrule->wr.cr.flow);
1519 stats->n_packets += subrule->packet_count;
1520 stats->n_bytes += subrule->byte_count;
1523 xflow_key_from_flow(&xflow_flows[0].key, &rule->wr.cr.flow);
1526 /* Fetch up-to-date statistics from the datapath and add them in. */
1527 if (!xfif_flow_get_multiple(wx->xfif, xflow_flows, n_xflow_flows)) {
1529 for (i = 0; i < n_xflow_flows; i++) {
1530 struct xflow_flow *xflow_flow = &xflow_flows[i];
1533 stats->n_packets += xflow_flow->stats.n_packets;
1534 stats->n_bytes += xflow_flow->stats.n_bytes;
1535 used = xflow_flow_stats_to_msec(&xflow_flow->stats);
1536 if (used > stats->used) {
1538 if (xflow_flow->key.dl_type == htons(ETH_TYPE_IP)
1539 && xflow_flow->key.nw_proto == IP_TYPE_TCP) {
1540 stats->ip_tos = xflow_flow->stats.ip_tos;
1543 stats->tcp_flags |= xflow_flow->stats.tcp_flags;
1550 wx_flow_get_stats(const struct wdp *wdp,
1551 const struct wdp_rule *wdp_rule,
1552 struct wdp_flow_stats *stats)
1554 struct wx *wx = wx_cast(wdp);
1555 struct wx_rule *rule = wx_rule_cast(&wdp_rule->cr);
1557 query_stats(wx, rule, stats);
1562 wx_flow_overlaps(const struct wdp *wdp, const flow_t *flow)
1564 struct wx *wx = wx_cast(wdp);
1566 /* XXX overlap with a subrule? */
1567 return classifier_rule_overlaps(&wx->cls, flow);
1571 wx_flow_put(struct wdp *wdp, const struct wdp_flow_put *put,
1572 struct wdp_flow_stats *old_stats, struct wdp_rule **rulep)
1574 struct wx *wx = wx_cast(wdp);
1575 struct wx_rule *rule;
1577 rule = wx_rule_cast(classifier_find_rule_exactly(&wx->cls, put->flow));
1578 if (rule && wx_rule_is_hidden(rule)) {
1583 if (!(put->flags & WDP_PUT_MODIFY)) {
1587 if (!(put->flags & WDP_PUT_CREATE)) {
1590 if ((put->flow->wildcards
1591 ? classifier_count_wild(&wx->cls) >= WX_MAX_WILD
1592 : classifier_count_exact(&wx->cls) >= WX_MAX_EXACT)) {
1593 /* XXX subrules should not count against exact-match limit */
1598 rule = wx_rule_create(NULL, put->actions, put->n_actions,
1599 put->idle_timeout, put->hard_timeout);
1600 cls_rule_from_flow(&rule->wr.cr, put->flow);
1601 wx_rule_insert(wx, rule, NULL, 0);
1605 memset(old_stats, 0, sizeof *old_stats);
1615 wx_flow_delete(struct wdp *wdp, struct wdp_rule *wdp_rule,
1616 struct wdp_flow_stats *final_stats)
1618 struct wx *wx = wx_cast(wdp);
1619 struct wx_rule *rule = wx_rule_cast(&wdp_rule->cr);
1621 wx_rule_remove(wx, rule);
1623 memset(final_stats, 0, sizeof *final_stats); /* XXX */
1629 wx_flush_rule(struct cls_rule *cls_rule, void *wx_)
1631 struct wx_rule *rule = wx_rule_cast(cls_rule);
1632 struct wx *wx = wx_;
1634 /* Mark the flow as not installed, even though it might really be
1635 * installed, so that wx_rule_remove() doesn't bother trying to uninstall
1636 * it. There is no point in uninstalling it individually since we are
1637 * about to blow away all the flows with xfif_flow_flush(). */
1638 rule->installed = false;
1640 wx_rule_remove(wx, rule);
1644 wx_flow_flush(struct wdp *wdp)
1646 struct wx *wx = wx_cast(wdp);
1648 COVERAGE_INC(wx_flow_flush);
1649 classifier_for_each(&wx->cls, CLS_INC_ALL, wx_flush_rule, wx);
1650 xfif_flow_flush(wx->xfif);
1655 wx_execute(struct wdp *wdp, uint16_t in_port,
1656 const union ofp_action actions[], int n_actions,
1657 const struct ofpbuf *packet)
1659 struct wx *wx = wx_cast(wdp);
1660 struct xflow_actions xflow_actions;
1664 flow_extract((struct ofpbuf *) packet, in_port, &flow);
1665 error = wx_xlate_actions(wx, actions, n_actions, &flow, packet,
1666 &xflow_actions, NULL);
1670 xfif_execute(wx->xfif, ofp_port_to_xflow_port(in_port),
1671 xflow_actions.actions, xflow_actions.n_actions, packet);
1676 wx_flow_inject(struct wdp *wdp, struct wdp_rule *wdp_rule,
1677 uint16_t in_port, const struct ofpbuf *packet)
1679 struct wx_rule *rule = wx_rule_cast(&wdp_rule->cr);
1682 error = wx_execute(wdp, in_port, rule->wr.actions, rule->wr.n_actions,
1685 rule->packet_count++;
1686 rule->byte_count += packet->size;
1687 rule->used = time_msec();
1693 wx_recv_get_mask(const struct wdp *wdp, int *listen_mask)
1695 struct wx *wx = wx_cast(wdp);
1696 int xflow_listen_mask;
1699 error = xfif_recv_get_mask(wx->xfif, &xflow_listen_mask);
1702 if (xflow_listen_mask & XFLOWL_MISS) {
1703 *listen_mask |= 1 << WDP_CHAN_MISS;
1705 if (xflow_listen_mask & XFLOWL_ACTION) {
1706 *listen_mask |= 1 << WDP_CHAN_ACTION;
1708 if (xflow_listen_mask & XFLOWL_SFLOW) {
1709 *listen_mask |= 1 << WDP_CHAN_SFLOW;
1716 wx_recv_set_mask(struct wdp *wdp, int listen_mask)
1718 struct wx *wx = wx_cast(wdp);
1719 int xflow_listen_mask;
1721 xflow_listen_mask = 0;
1722 if (listen_mask & (1 << WDP_CHAN_MISS)) {
1723 xflow_listen_mask |= XFLOWL_MISS;
1725 if (listen_mask & (1 << WDP_CHAN_ACTION)) {
1726 xflow_listen_mask |= XFLOWL_ACTION;
1728 if (listen_mask & (1 << WDP_CHAN_SFLOW)) {
1729 xflow_listen_mask |= XFLOWL_SFLOW;
1732 return xfif_recv_set_mask(wx->xfif, xflow_listen_mask);
1736 wx_get_sflow_probability(const struct wdp *wdp, uint32_t *probability)
1738 struct wx *wx = wx_cast(wdp);
1740 return xfif_get_sflow_probability(wx->xfif, probability);
1744 wx_set_sflow_probability(struct wdp *wdp, uint32_t probability)
1746 struct wx *wx = wx_cast(wdp);
1748 return xfif_set_sflow_probability(wx->xfif, probability);
1752 wx_translate_xflow_msg(struct xflow_msg *msg, struct ofpbuf *payload,
1753 struct wdp_packet *packet)
1755 packet->in_port = xflow_port_to_ofp_port(msg->port);
1756 packet->send_len = 0;
1758 switch (msg->type) {
1759 case _XFLOWL_MISS_NR:
1760 packet->channel = WDP_CHAN_MISS;
1761 packet->payload = payload;
1764 case _XFLOWL_ACTION_NR:
1765 packet->channel = WDP_CHAN_ACTION;
1766 packet->payload = payload;
1767 packet->send_len = msg->arg;
1770 case _XFLOWL_SFLOW_NR:
1772 ofpbuf_delete(payload);
1776 VLOG_WARN_RL(&rl, "received XFLOW message of unexpected type %"PRIu32,
1778 ofpbuf_delete(payload);
1783 static const uint8_t *
1784 get_local_mac(const struct wx *wx)
1786 const struct wdp_port *port = port_array_get(&wx->ports, XFLOWP_LOCAL);
1787 return port ? port->opp.hw_addr : NULL;
1790 /* Returns true if 'packet' is a DHCP reply to the local port. Such a reply
1791 * should be sent to the local port regardless of the flow table.
1793 * We only need this, strictly speaking, when in-band control is turned on. */
1795 wx_is_local_dhcp_reply(const struct wx *wx,
1796 const flow_t *flow, const struct ofpbuf *packet)
1798 if (flow->dl_type == htons(ETH_TYPE_IP)
1799 && flow->nw_proto == IP_TYPE_UDP
1800 && flow->tp_src == htons(DHCP_SERVER_PORT)
1801 && flow->tp_dst == htons(DHCP_CLIENT_PORT)
1804 const uint8_t *local_mac = get_local_mac(wx);
1805 struct dhcp_header *dhcp = ofpbuf_at(
1806 packet, (char *)packet->l7 - (char *)packet->data, sizeof *dhcp);
1807 return dhcp && local_mac && eth_addr_equals(dhcp->chaddr, local_mac);
1814 wx_explode_rule(struct wx *wx, struct xflow_msg *msg, struct ofpbuf *payload)
1816 struct wx_rule *rule;
1819 flow_extract(payload, xflow_port_to_ofp_port(msg->port), &flow);
1821 if (wx_is_local_dhcp_reply(wx, &flow, payload)) {
1822 union xflow_action action;
1824 memset(&action, 0, sizeof(action));
1825 action.output.type = XFLOWAT_OUTPUT;
1826 action.output.port = XFLOWP_LOCAL;
1827 xfif_execute(wx->xfif, msg->port, &action, 1, payload);
1830 rule = wx_rule_lookup_valid(wx, &flow);
1835 if (rule->wr.cr.flow.wildcards) {
1836 rule = wx_rule_create_subrule(wx, rule, &flow);
1837 wx_rule_make_actions(wx, rule, payload);
1839 if (!rule->may_install) {
1840 /* The rule is not installable, that is, we need to process every
1841 * packet, so process the current packet and set its actions into
1843 wx_rule_make_actions(wx, rule, payload);
1845 /* XXX revalidate rule if it needs it */
1849 wx_rule_execute(wx, rule, payload, &flow);
1850 wx_rule_reinstall(wx, rule);
1856 wx_recv(struct wdp *wdp, struct wdp_packet *packet)
1858 struct wx *wx = wx_cast(wdp);
1861 /* XXX need to avoid 50*50 potential cost for caller. */
1862 for (i = 0; i < 50; i++) {
1863 struct xflow_msg *msg;
1867 error = xfif_recv(wx->xfif, &buf);
1872 msg = ofpbuf_pull(buf, sizeof *msg);
1873 if (msg->type != _XFLOWL_MISS_NR || !wx_explode_rule(wx, msg, buf)) {
1874 return wx_translate_xflow_msg(msg, buf, packet);
1882 wx_recv_wait(struct wdp *wdp)
1884 struct wx *wx = wx_cast(wdp);
1886 xfif_recv_wait(wx->xfif);
1889 static void wx_port_update(struct wx *, const char *devname);
1890 static void wx_port_reinit(struct wx *);
1893 wx_port_process_change(struct wx *wx, int error, char *devname)
1895 if (error == ENOBUFS) {
1897 } else if (!error) {
1898 wx_port_update(wx, devname);
1904 wx_port_run(struct wx *wx)
1909 while ((error = xfif_port_poll(wx->xfif, &devname)) != EAGAIN) {
1910 wx_port_process_change(wx, error, devname);
1912 while ((error = netdev_monitor_poll(wx->netdev_monitor,
1913 &devname)) != EAGAIN) {
1914 wx_port_process_change(wx, error, devname);
1919 wx_port_refresh_group(struct wx *wx, unsigned int group)
1923 struct wdp_port *port;
1924 unsigned int port_no;
1926 assert(group == WX_GROUP_ALL || group == WX_GROUP_FLOOD);
1928 ports = xmalloc(port_array_count(&wx->ports) * sizeof *ports);
1930 PORT_ARRAY_FOR_EACH (port, &wx->ports, port_no) {
1931 if (group == WX_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1932 ports[n_ports++] = port_no;
1935 xfif_port_group_set(wx->xfif, group, ports, n_ports);
1942 wx_port_refresh_groups(struct wx *wx)
1944 wx_port_refresh_group(wx, WX_GROUP_FLOOD);
1945 wx_port_refresh_group(wx, WX_GROUP_ALL);
1949 wx_port_reinit(struct wx *wx)
1951 struct svec devnames;
1952 struct wdp_port *wdp_port;
1953 unsigned int port_no;
1954 struct xflow_port *xflow_ports;
1955 size_t n_xflow_ports;
1958 svec_init(&devnames);
1959 PORT_ARRAY_FOR_EACH (wdp_port, &wx->ports, port_no) {
1960 svec_add (&devnames, (char *) wdp_port->opp.name);
1962 xfif_port_list(wx->xfif, &xflow_ports, &n_xflow_ports);
1963 for (i = 0; i < n_xflow_ports; i++) {
1964 svec_add(&devnames, xflow_ports[i].devname);
1968 svec_sort_unique(&devnames);
1969 for (i = 0; i < devnames.n; i++) {
1970 wx_port_update(wx, devnames.names[i]);
1972 svec_destroy(&devnames);
1974 wx_port_refresh_groups(wx);
1977 static struct wdp_port *
1978 make_wdp_port(const struct xflow_port *xflow_port)
1980 struct netdev_options netdev_options;
1981 enum netdev_flags flags;
1982 struct wdp_port *wdp_port;
1983 struct netdev *netdev;
1987 memset(&netdev_options, 0, sizeof netdev_options);
1988 netdev_options.name = xflow_port->devname;
1989 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1990 netdev_options.may_create = true;
1991 netdev_options.may_open = true;
1993 error = netdev_open(&netdev_options, &netdev);
1995 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1996 "cannot be opened (%s)",
1997 xflow_port->devname, xflow_port->port,
1998 xflow_port->devname, strerror(error));
2002 wdp_port = xmalloc(sizeof *wdp_port);
2003 wdp_port->netdev = netdev;
2004 wdp_port->opp.port_no = xflow_port_to_ofp_port(xflow_port->port);
2005 netdev_get_etheraddr(netdev, wdp_port->opp.hw_addr);
2006 strncpy((char *) wdp_port->opp.name, xflow_port->devname,
2007 sizeof wdp_port->opp.name);
2008 wdp_port->opp.name[sizeof wdp_port->opp.name - 1] = '\0';
2010 netdev_get_flags(netdev, &flags);
2011 wdp_port->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
2013 netdev_get_carrier(netdev, &carrier);
2014 wdp_port->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
2016 netdev_get_features(netdev,
2017 &wdp_port->opp.curr, &wdp_port->opp.advertised,
2018 &wdp_port->opp.supported, &wdp_port->opp.peer);
2020 wdp_port->devname = xstrdup(xflow_port->devname);
2021 wdp_port->internal = (xflow_port->flags & XFLOW_PORT_INTERNAL) != 0;
2026 wx_port_conflicts(const struct wx *wx, const struct xflow_port *xflow_port)
2028 if (port_array_get(&wx->ports, xflow_port->port)) {
2029 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
2032 } else if (shash_find(&wx->port_by_name, xflow_port->devname)) {
2033 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
2034 xflow_port->devname);
2042 wdp_port_equal(const struct wdp_port *a_, const struct wdp_port *b_)
2044 const struct ofp_phy_port *a = &a_->opp;
2045 const struct ofp_phy_port *b = &b_->opp;
2047 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
2048 return (a->port_no == b->port_no
2049 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
2050 && !strcmp((char *) a->name, (char *) b->name)
2051 && a->state == b->state
2052 && a->config == b->config
2053 && a->curr == b->curr
2054 && a->advertised == b->advertised
2055 && a->supported == b->supported
2056 && a->peer == b->peer);
2060 wx_port_install(struct wx *wx, struct wdp_port *wdp_port)
2062 uint16_t xflow_port = ofp_port_to_xflow_port(wdp_port->opp.port_no);
2063 const char *netdev_name = (const char *) wdp_port->opp.name;
2065 netdev_monitor_add(wx->netdev_monitor, wdp_port->netdev);
2066 port_array_set(&wx->ports, xflow_port, wdp_port);
2067 shash_add(&wx->port_by_name, netdev_name, wdp_port);
2071 wx_port_remove(struct wx *wx, struct wdp_port *wdp_port)
2073 uint16_t xflow_port = ofp_port_to_xflow_port(wdp_port->opp.port_no);
2075 netdev_monitor_remove(wx->netdev_monitor, wdp_port->netdev);
2076 port_array_set(&wx->ports, xflow_port, NULL);
2077 shash_delete(&wx->port_by_name,
2078 shash_find(&wx->port_by_name, (char *) wdp_port->opp.name));
2082 wx_port_free(struct wdp_port *wdp_port)
2085 netdev_close(wdp_port->netdev);
2091 wx_port_update(struct wx *wx, const char *devname)
2093 struct xflow_port xflow_port;
2094 struct wdp_port *old_wdp_port;
2095 struct wdp_port *new_wdp_port;
2098 COVERAGE_INC(wx_update_port);
2100 /* Query the datapath for port information. */
2101 error = xfif_port_query_by_name(wx->xfif, devname, &xflow_port);
2103 /* Find the old wdp_port. */
2104 old_wdp_port = shash_find_data(&wx->port_by_name, devname);
2106 if (!old_wdp_port) {
2107 /* There's no port named 'devname' but there might be a port with
2108 * the same port number. This could happen if a port is deleted
2109 * and then a new one added in its place very quickly, or if a port
2110 * is renamed. In the former case we want to send an OFPPR_DELETE
2111 * and an OFPPR_ADD, and in the latter case we want to send a
2112 * single OFPPR_MODIFY. We can distinguish the cases by comparing
2113 * the old port's ifindex against the new port, or perhaps less
2114 * reliably but more portably by comparing the old port's MAC
2115 * against the new port's MAC. However, this code isn't that smart
2116 * and always sends an OFPPR_MODIFY (XXX). */
2117 old_wdp_port = port_array_get(&wx->ports, xflow_port.port);
2119 } else if (error != ENOENT && error != ENODEV) {
2120 VLOG_WARN_RL(&rl, "xfif_port_query_by_name returned unexpected error "
2121 "%s", strerror(error));
2125 /* Create a new wdp_port. */
2126 new_wdp_port = !error ? make_wdp_port(&xflow_port) : NULL;
2128 /* Eliminate a few pathological cases. */
2129 if (!old_wdp_port && !new_wdp_port) {
2131 } else if (old_wdp_port && new_wdp_port) {
2132 /* Most of the 'config' bits are OpenFlow soft state, but
2133 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
2134 * OpenFlow bits from old_wdp_port. (make_wdp_port() only sets
2135 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
2136 new_wdp_port->opp.config |= old_wdp_port->opp.config & ~OFPPC_PORT_DOWN;
2138 if (wdp_port_equal(old_wdp_port, new_wdp_port)) {
2139 /* False alarm--no change. */
2140 wx_port_free(new_wdp_port);
2145 /* Now deal with the normal cases. */
2147 wx_port_remove(wx, old_wdp_port);
2150 wx_port_install(wx, new_wdp_port);
2152 wx_port_free(old_wdp_port);
2156 wx_port_init(struct wx *wx)
2158 struct xflow_port *ports;
2163 error = xfif_port_list(wx->xfif, &ports, &n_ports);
2168 for (i = 0; i < n_ports; i++) {
2169 const struct xflow_port *xflow_port = &ports[i];
2170 if (!wx_port_conflicts(wx, xflow_port)) {
2171 struct wdp_port *wdp_port = make_wdp_port(xflow_port);
2173 wx_port_install(wx, wdp_port);
2178 wx_port_refresh_groups(wx);
2183 wdp_xflow_register(void)
2185 static const struct wdp_class wdp_xflow_class = {
2200 wx_port_query_by_number,
2201 wx_port_query_by_name,
2208 wx_flow_for_each_match,
2218 wx_get_sflow_probability,
2219 wx_set_sflow_probability,
2224 static bool inited = false;
2237 xf_enumerate_types(&types);
2240 SVEC_FOR_EACH (i, type, &types) {
2241 struct wdp_class *class;
2243 class = xmalloc(sizeof *class);
2244 *class = wdp_xflow_class;
2245 class->type = xstrdup(type);
2250 if (!wdp_register_provider(class)) {
2255 svec_destroy(&types);