2 * Copyright (c) 2009, 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.
22 #include <netinet/in.h>
25 #include "classifier.h"
27 #include "discovery.h"
29 #include "dynamic-string.h"
30 #include "fail-open.h"
32 #include "mac-learning.h"
36 #include "ofp-print.h"
38 #include "openflow/nicira-ext.h"
39 #include "openflow/openflow.h"
40 #include "openvswitch/datapath-protocol.h"
44 #include "poll-loop.h"
45 #include "port-array.h"
50 #include "stream-ssl.h"
58 #define THIS_MODULE VLM_ofproto
68 TABLEID_CLASSIFIER = 1
72 struct netdev *netdev;
73 struct ofp_phy_port opp; /* In host byte order. */
76 static void ofport_free(struct ofport *);
77 static void hton_ofp_phy_port(struct ofp_phy_port *);
79 static int xlate_actions(const union ofp_action *in, size_t n_in,
80 const flow_t *flow, struct ofproto *ofproto,
81 const struct ofpbuf *packet,
82 struct odp_actions *out, tag_type *tags,
83 bool *may_set_up_flow, uint16_t *nf_output_iface);
88 uint16_t idle_timeout; /* In seconds from time of last use. */
89 uint16_t hard_timeout; /* In seconds from time of creation. */
90 long long int used; /* Last-used time (0 if never used). */
91 long long int created; /* Creation time. */
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 tag_type tags; /* Tags (set only by hooks). */
96 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
98 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
99 * exact-match rule (having cr.wc.wildcards of 0) generated from the
100 * wildcard rule 'super'. In this case, 'list' is an element of the
103 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
104 * a list of subrules. A super-rule with no wildcards (where
105 * cr.wc.wildcards is 0) will never have any subrules. */
111 * A subrule has no actions (it uses the super-rule's actions). */
113 union ofp_action *actions;
117 * A super-rule with wildcard fields never has ODP actions (since the
118 * datapath only supports exact-match flows). */
119 bool installed; /* Installed in datapath? */
120 bool may_install; /* True ordinarily; false if actions must
121 * be reassessed for every packet. */
123 union odp_action *odp_actions;
127 rule_is_hidden(const struct rule *rule)
129 /* Subrules are merely an implementation detail, so hide them from the
131 if (rule->super != NULL) {
135 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
136 * (e.g. by in-band control) and are intentionally hidden from the
138 if (rule->cr.priority > UINT16_MAX) {
145 static struct rule *rule_create(struct ofproto *, struct rule *super,
146 const union ofp_action *, size_t n_actions,
147 uint16_t idle_timeout, uint16_t hard_timeout);
148 static void rule_free(struct rule *);
149 static void rule_destroy(struct ofproto *, struct rule *);
150 static struct rule *rule_from_cls_rule(const struct cls_rule *);
151 static void rule_insert(struct ofproto *, struct rule *,
152 struct ofpbuf *packet, uint16_t in_port);
153 static void rule_remove(struct ofproto *, struct rule *);
154 static bool rule_make_actions(struct ofproto *, struct rule *,
155 const struct ofpbuf *packet);
156 static void rule_install(struct ofproto *, struct rule *,
157 struct rule *displaced_rule);
158 static void rule_uninstall(struct ofproto *, struct rule *);
159 static void rule_post_uninstall(struct ofproto *, struct rule *);
164 struct pktbuf *pktbuf;
168 struct rconn_packet_counter *packet_in_counter;
170 /* Number of OpenFlow messages queued as replies to OpenFlow requests, and
171 * the maximum number before we stop reading OpenFlow requests. */
172 #define OFCONN_REPLY_MAX 100
173 struct rconn_packet_counter *reply_counter;
176 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *);
177 static void ofconn_destroy(struct ofconn *);
178 static void ofconn_run(struct ofconn *, struct ofproto *);
179 static void ofconn_wait(struct ofconn *);
180 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
181 struct rconn_packet_counter *counter);
185 uint64_t datapath_id; /* Datapath ID. */
186 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
187 char *manufacturer; /* Manufacturer. */
188 char *hardware; /* Hardware. */
189 char *software; /* Software version. */
190 char *serial; /* Serial number. */
194 struct netdev_monitor *netdev_monitor;
195 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
197 struct shash port_by_name;
201 struct switch_status *switch_status;
202 struct status_category *ss_cat;
203 struct in_band *in_band;
204 struct discovery *discovery;
205 struct fail_open *fail_open;
206 struct pinsched *miss_sched, *action_sched;
207 struct netflow *netflow;
210 struct classifier cls;
211 bool need_revalidate;
212 long long int next_expiration;
213 struct tag_set revalidate_set;
215 /* OpenFlow connections. */
216 struct list all_conns;
217 struct ofconn *controller;
218 struct pvconn **listeners;
220 struct pvconn **snoops;
223 /* Hooks for ovs-vswitchd. */
224 const struct ofhooks *ofhooks;
227 /* Used by default ofhooks. */
228 struct mac_learning *ml;
231 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
233 static const struct ofhooks default_ofhooks;
235 static uint64_t pick_datapath_id(const struct ofproto *);
236 static uint64_t pick_fallback_dpid(void);
237 static void send_packet_in_miss(struct ofpbuf *, void *ofproto);
238 static void send_packet_in_action(struct ofpbuf *, void *ofproto);
239 static void update_used(struct ofproto *);
240 static void update_stats(struct ofproto *, struct rule *,
241 const struct odp_flow_stats *);
242 static void expire_rule(struct cls_rule *, void *ofproto);
243 static void active_timeout(struct ofproto *ofproto, struct rule *rule);
244 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
245 static void revalidate_cb(struct cls_rule *rule_, void *p_);
247 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
249 static void handle_openflow(struct ofconn *, struct ofproto *,
252 static void refresh_port_group(struct ofproto *, unsigned int group);
253 static void update_port(struct ofproto *, const char *devname);
254 static int init_ports(struct ofproto *);
255 static void reinit_ports(struct ofproto *);
258 ofproto_create(const char *datapath, const char *datapath_type,
259 const struct ofhooks *ofhooks, void *aux,
260 struct ofproto **ofprotop)
262 struct odp_stats stats;
269 /* Connect to datapath and start listening for messages. */
270 error = dpif_open(datapath, datapath_type, &dpif);
272 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
275 error = dpif_get_dp_stats(dpif, &stats);
277 VLOG_ERR("failed to obtain stats for datapath %s: %s",
278 datapath, strerror(error));
282 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION);
284 VLOG_ERR("failed to listen on datapath %s: %s",
285 datapath, strerror(error));
289 dpif_flow_flush(dpif);
290 dpif_recv_purge(dpif);
292 /* Initialize settings. */
293 p = xzalloc(sizeof *p);
294 p->fallback_dpid = pick_fallback_dpid();
295 p->datapath_id = p->fallback_dpid;
296 p->manufacturer = xstrdup("Nicira Networks, Inc.");
297 p->hardware = xstrdup("Reference Implementation");
298 p->software = xstrdup(VERSION BUILDNR);
299 p->serial = xstrdup("None");
301 /* Initialize datapath. */
303 p->netdev_monitor = netdev_monitor_create();
304 port_array_init(&p->ports);
305 shash_init(&p->port_by_name);
306 p->max_ports = stats.max_ports;
308 /* Initialize submodules. */
309 p->switch_status = switch_status_create(p);
313 p->miss_sched = p->action_sched = NULL;
316 /* Initialize flow table. */
317 classifier_init(&p->cls);
318 p->need_revalidate = false;
319 p->next_expiration = time_msec() + 1000;
320 tag_set_init(&p->revalidate_set);
322 /* Initialize OpenFlow connections. */
323 list_init(&p->all_conns);
324 p->controller = ofconn_create(p, rconn_create(5, 8));
325 p->controller->pktbuf = pktbuf_create();
326 p->controller->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
332 /* Initialize hooks. */
334 p->ofhooks = ofhooks;
338 p->ofhooks = &default_ofhooks;
340 p->ml = mac_learning_create();
343 /* Register switch status category. */
344 p->ss_cat = switch_status_register(p->switch_status, "remote",
345 rconn_status_cb, p->controller->rconn);
347 /* Pick final datapath ID. */
348 p->datapath_id = pick_datapath_id(p);
349 VLOG_INFO("using datapath ID %012"PRIx64, p->datapath_id);
356 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
358 uint64_t old_dpid = p->datapath_id;
359 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
360 if (p->datapath_id != old_dpid) {
361 VLOG_INFO("datapath ID changed to %012"PRIx64, p->datapath_id);
362 rconn_reconnect(p->controller->rconn);
367 ofproto_set_probe_interval(struct ofproto *p, int probe_interval)
369 probe_interval = probe_interval ? MAX(probe_interval, 5) : 0;
370 rconn_set_probe_interval(p->controller->rconn, probe_interval);
372 int trigger_duration = probe_interval ? probe_interval * 3 : 15;
373 fail_open_set_trigger_duration(p->fail_open, trigger_duration);
378 ofproto_set_max_backoff(struct ofproto *p, int max_backoff)
380 rconn_set_max_backoff(p->controller->rconn, max_backoff);
384 ofproto_set_desc(struct ofproto *p,
385 const char *manufacturer, const char *hardware,
386 const char *software, const char *serial)
389 free(p->manufacturer);
390 p->manufacturer = xstrdup(manufacturer);
394 p->hardware = xstrdup(hardware);
398 p->software = xstrdup(software);
402 p->serial = xstrdup(serial);
407 ofproto_set_in_band(struct ofproto *p, bool in_band)
409 if (in_band != (p->in_band != NULL)) {
411 return in_band_create(p, p->dpif, p->switch_status,
412 p->controller->rconn, &p->in_band);
414 ofproto_set_discovery(p, false, NULL, true);
415 in_band_destroy(p->in_band);
418 rconn_reconnect(p->controller->rconn);
424 ofproto_set_discovery(struct ofproto *p, bool discovery,
425 const char *re, bool update_resolv_conf)
427 if (discovery != (p->discovery != NULL)) {
429 int error = ofproto_set_in_band(p, true);
433 error = discovery_create(re, update_resolv_conf,
434 p->dpif, p->switch_status,
440 discovery_destroy(p->discovery);
443 rconn_disconnect(p->controller->rconn);
444 } else if (discovery) {
445 discovery_set_update_resolv_conf(p->discovery, update_resolv_conf);
446 return discovery_set_accept_controller_re(p->discovery, re);
452 ofproto_set_controller(struct ofproto *ofproto, const char *controller)
454 if (ofproto->discovery) {
456 } else if (controller) {
457 if (strcmp(rconn_get_name(ofproto->controller->rconn), controller)) {
458 return rconn_connect(ofproto->controller->rconn, controller);
463 rconn_disconnect(ofproto->controller->rconn);
469 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
470 const struct svec *svec)
472 struct pvconn **pvconns = *pvconnsp;
473 size_t n_pvconns = *n_pvconnsp;
477 for (i = 0; i < n_pvconns; i++) {
478 pvconn_close(pvconns[i]);
482 pvconns = xmalloc(svec->n * sizeof *pvconns);
484 for (i = 0; i < svec->n; i++) {
485 const char *name = svec->names[i];
486 struct pvconn *pvconn;
489 error = pvconn_open(name, &pvconn);
491 pvconns[n_pvconns++] = pvconn;
493 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
501 *n_pvconnsp = n_pvconns;
507 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
509 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
513 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
515 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
519 ofproto_set_netflow(struct ofproto *ofproto,
520 const struct netflow_options *nf_options)
522 if (nf_options && nf_options->collectors.n) {
523 if (!ofproto->netflow) {
524 ofproto->netflow = netflow_create();
526 return netflow_set_options(ofproto->netflow, nf_options);
528 netflow_destroy(ofproto->netflow);
529 ofproto->netflow = NULL;
535 ofproto_set_failure(struct ofproto *ofproto, bool fail_open)
538 struct rconn *rconn = ofproto->controller->rconn;
539 int trigger_duration = rconn_get_probe_interval(rconn) * 3;
540 if (!ofproto->fail_open) {
541 ofproto->fail_open = fail_open_create(ofproto, trigger_duration,
542 ofproto->switch_status,
545 fail_open_set_trigger_duration(ofproto->fail_open,
549 fail_open_destroy(ofproto->fail_open);
550 ofproto->fail_open = NULL;
555 ofproto_set_rate_limit(struct ofproto *ofproto,
556 int rate_limit, int burst_limit)
558 if (rate_limit > 0) {
559 if (!ofproto->miss_sched) {
560 ofproto->miss_sched = pinsched_create(rate_limit, burst_limit,
561 ofproto->switch_status);
562 ofproto->action_sched = pinsched_create(rate_limit, burst_limit,
565 pinsched_set_limits(ofproto->miss_sched, rate_limit, burst_limit);
566 pinsched_set_limits(ofproto->action_sched,
567 rate_limit, burst_limit);
570 pinsched_destroy(ofproto->miss_sched);
571 ofproto->miss_sched = NULL;
572 pinsched_destroy(ofproto->action_sched);
573 ofproto->action_sched = NULL;
578 ofproto_set_stp(struct ofproto *ofproto UNUSED, bool enable_stp)
582 VLOG_WARN("STP is not yet implemented");
590 ofproto_get_datapath_id(const struct ofproto *ofproto)
592 return ofproto->datapath_id;
596 ofproto_get_probe_interval(const struct ofproto *ofproto)
598 return rconn_get_probe_interval(ofproto->controller->rconn);
602 ofproto_get_max_backoff(const struct ofproto *ofproto)
604 return rconn_get_max_backoff(ofproto->controller->rconn);
608 ofproto_get_in_band(const struct ofproto *ofproto)
610 return ofproto->in_band != NULL;
614 ofproto_get_discovery(const struct ofproto *ofproto)
616 return ofproto->discovery != NULL;
620 ofproto_get_controller(const struct ofproto *ofproto)
622 return rconn_get_name(ofproto->controller->rconn);
626 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
630 for (i = 0; i < ofproto->n_listeners; i++) {
631 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
636 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
640 for (i = 0; i < ofproto->n_snoops; i++) {
641 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
646 ofproto_destroy(struct ofproto *p)
648 struct ofconn *ofconn, *next_ofconn;
649 struct ofport *ofport;
650 unsigned int port_no;
657 ofproto_flush_flows(p);
658 classifier_destroy(&p->cls);
660 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
662 ofconn_destroy(ofconn);
666 netdev_monitor_destroy(p->netdev_monitor);
667 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
670 shash_destroy(&p->port_by_name);
672 switch_status_destroy(p->switch_status);
673 in_band_destroy(p->in_band);
674 discovery_destroy(p->discovery);
675 fail_open_destroy(p->fail_open);
676 pinsched_destroy(p->miss_sched);
677 pinsched_destroy(p->action_sched);
678 netflow_destroy(p->netflow);
680 switch_status_unregister(p->ss_cat);
682 for (i = 0; i < p->n_listeners; i++) {
683 pvconn_close(p->listeners[i]);
687 for (i = 0; i < p->n_snoops; i++) {
688 pvconn_close(p->snoops[i]);
692 mac_learning_destroy(p->ml);
698 ofproto_run(struct ofproto *p)
700 int error = ofproto_run1(p);
702 error = ofproto_run2(p, false);
708 process_port_change(struct ofproto *ofproto, int error, char *devname)
710 if (error == ENOBUFS) {
711 reinit_ports(ofproto);
713 update_port(ofproto, devname);
719 ofproto_run1(struct ofproto *p)
721 struct ofconn *ofconn, *next_ofconn;
726 if (shash_is_empty(&p->port_by_name)) {
730 for (i = 0; i < 50; i++) {
734 error = dpif_recv(p->dpif, &buf);
736 if (error == ENODEV) {
737 /* Someone destroyed the datapath behind our back. The caller
738 * better destroy us and give up, because we're just going to
739 * spin from here on out. */
740 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
741 VLOG_ERR_RL(&rl, "%s: datapath was destroyed externally",
748 handle_odp_msg(p, buf);
751 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
752 process_port_change(p, error, devname);
754 while ((error = netdev_monitor_poll(p->netdev_monitor,
755 &devname)) != EAGAIN) {
756 process_port_change(p, error, devname);
760 in_band_run(p->in_band);
763 char *controller_name;
764 if (rconn_is_connectivity_questionable(p->controller->rconn)) {
765 discovery_question_connectivity(p->discovery);
767 if (discovery_run(p->discovery, &controller_name)) {
768 if (controller_name) {
769 rconn_connect(p->controller->rconn, controller_name);
771 rconn_disconnect(p->controller->rconn);
775 pinsched_run(p->miss_sched, send_packet_in_miss, p);
776 pinsched_run(p->action_sched, send_packet_in_action, p);
778 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
780 ofconn_run(ofconn, p);
783 /* Fail-open maintenance. Do this after processing the ofconns since
784 * fail-open checks the status of the controller rconn. */
786 fail_open_run(p->fail_open);
789 for (i = 0; i < p->n_listeners; i++) {
793 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
795 ofconn_create(p, rconn_new_from_vconn("passive", vconn));
796 } else if (retval != EAGAIN) {
797 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
801 for (i = 0; i < p->n_snoops; i++) {
805 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
807 rconn_add_monitor(p->controller->rconn, vconn);
808 } else if (retval != EAGAIN) {
809 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
813 if (time_msec() >= p->next_expiration) {
814 COVERAGE_INC(ofproto_expiration);
815 p->next_expiration = time_msec() + 1000;
818 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
820 /* Let the hook know that we're at a stable point: all outstanding data
821 * in existing flows has been accounted to the account_cb. Thus, the
822 * hook can now reasonably do operations that depend on having accurate
823 * flow volume accounting (currently, that's just bond rebalancing). */
824 if (p->ofhooks->account_checkpoint_cb) {
825 p->ofhooks->account_checkpoint_cb(p->aux);
830 netflow_run(p->netflow);
836 struct revalidate_cbdata {
837 struct ofproto *ofproto;
838 bool revalidate_all; /* Revalidate all exact-match rules? */
839 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
840 struct tag_set revalidate_set; /* Set of tags to revalidate. */
844 ofproto_run2(struct ofproto *p, bool revalidate_all)
846 if (p->need_revalidate || revalidate_all
847 || !tag_set_is_empty(&p->revalidate_set)) {
848 struct revalidate_cbdata cbdata;
850 cbdata.revalidate_all = revalidate_all;
851 cbdata.revalidate_subrules = p->need_revalidate;
852 cbdata.revalidate_set = p->revalidate_set;
853 tag_set_init(&p->revalidate_set);
854 COVERAGE_INC(ofproto_revalidate);
855 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
856 p->need_revalidate = false;
863 ofproto_wait(struct ofproto *p)
865 struct ofconn *ofconn;
868 dpif_recv_wait(p->dpif);
869 dpif_port_poll_wait(p->dpif);
870 netdev_monitor_poll_wait(p->netdev_monitor);
871 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
875 in_band_wait(p->in_band);
878 discovery_wait(p->discovery);
881 fail_open_wait(p->fail_open);
883 pinsched_wait(p->miss_sched);
884 pinsched_wait(p->action_sched);
885 if (!tag_set_is_empty(&p->revalidate_set)) {
886 poll_immediate_wake();
888 if (p->need_revalidate) {
889 /* Shouldn't happen, but if it does just go around again. */
890 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
891 poll_immediate_wake();
892 } else if (p->next_expiration != LLONG_MAX) {
893 poll_timer_wait(p->next_expiration - time_msec());
895 for (i = 0; i < p->n_listeners; i++) {
896 pvconn_wait(p->listeners[i]);
898 for (i = 0; i < p->n_snoops; i++) {
899 pvconn_wait(p->snoops[i]);
904 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
906 tag_set_add(&ofproto->revalidate_set, tag);
910 ofproto_get_revalidate_set(struct ofproto *ofproto)
912 return &ofproto->revalidate_set;
916 ofproto_is_alive(const struct ofproto *p)
918 return p->discovery || rconn_is_alive(p->controller->rconn);
922 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
923 const union ofp_action *actions, size_t n_actions,
924 const struct ofpbuf *packet)
926 struct odp_actions odp_actions;
929 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
935 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
937 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
938 odp_actions.n_actions, packet);
943 ofproto_add_flow(struct ofproto *p,
944 const flow_t *flow, uint32_t wildcards, unsigned int priority,
945 const union ofp_action *actions, size_t n_actions,
949 rule = rule_create(p, NULL, actions, n_actions,
950 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */, 0);
951 cls_rule_from_flow(&rule->cr, flow, wildcards, priority);
952 rule_insert(p, rule, NULL, 0);
956 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
957 uint32_t wildcards, unsigned int priority)
961 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
965 rule_remove(ofproto, rule);
970 destroy_rule(struct cls_rule *rule_, void *ofproto_)
972 struct rule *rule = rule_from_cls_rule(rule_);
973 struct ofproto *ofproto = ofproto_;
975 /* Mark the flow as not installed, even though it might really be
976 * installed, so that rule_remove() doesn't bother trying to uninstall it.
977 * There is no point in uninstalling it individually since we are about to
978 * blow away all the flows with dpif_flow_flush(). */
979 rule->installed = false;
981 rule_remove(ofproto, rule);
985 ofproto_flush_flows(struct ofproto *ofproto)
987 COVERAGE_INC(ofproto_flush);
988 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
989 dpif_flow_flush(ofproto->dpif);
990 if (ofproto->in_band) {
991 in_band_flushed(ofproto->in_band);
993 if (ofproto->fail_open) {
994 fail_open_flushed(ofproto->fail_open);
999 reinit_ports(struct ofproto *p)
1001 struct svec devnames;
1002 struct ofport *ofport;
1003 unsigned int port_no;
1004 struct odp_port *odp_ports;
1008 svec_init(&devnames);
1009 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1010 svec_add (&devnames, (char *) ofport->opp.name);
1012 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1013 for (i = 0; i < n_odp_ports; i++) {
1014 svec_add (&devnames, odp_ports[i].devname);
1018 svec_sort_unique(&devnames);
1019 for (i = 0; i < devnames.n; i++) {
1020 update_port(p, devnames.names[i]);
1022 svec_destroy(&devnames);
1026 refresh_port_group(struct ofproto *p, unsigned int group)
1030 struct ofport *port;
1031 unsigned int port_no;
1033 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1035 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1037 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1038 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1039 ports[n_ports++] = port_no;
1042 dpif_port_group_set(p->dpif, group, ports, n_ports);
1047 refresh_port_groups(struct ofproto *p)
1049 refresh_port_group(p, DP_GROUP_FLOOD);
1050 refresh_port_group(p, DP_GROUP_ALL);
1053 static struct ofport *
1054 make_ofport(const struct odp_port *odp_port)
1056 struct netdev_options netdev_options;
1057 enum netdev_flags flags;
1058 struct ofport *ofport;
1059 struct netdev *netdev;
1063 memset(&netdev_options, 0, sizeof netdev_options);
1064 netdev_options.name = odp_port->devname;
1065 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1066 netdev_options.may_open = true;
1068 error = netdev_open(&netdev_options, &netdev);
1070 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1071 "cannot be opened (%s)",
1072 odp_port->devname, odp_port->port,
1073 odp_port->devname, strerror(error));
1077 ofport = xmalloc(sizeof *ofport);
1078 ofport->netdev = netdev;
1079 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1080 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1081 memcpy(ofport->opp.name, odp_port->devname,
1082 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1083 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1085 netdev_get_flags(netdev, &flags);
1086 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1088 netdev_get_carrier(netdev, &carrier);
1089 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1091 netdev_get_features(netdev,
1092 &ofport->opp.curr, &ofport->opp.advertised,
1093 &ofport->opp.supported, &ofport->opp.peer);
1098 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1100 if (port_array_get(&p->ports, odp_port->port)) {
1101 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1104 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1105 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1114 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1116 const struct ofp_phy_port *a = &a_->opp;
1117 const struct ofp_phy_port *b = &b_->opp;
1119 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1120 return (a->port_no == b->port_no
1121 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1122 && !strcmp((char *) a->name, (char *) b->name)
1123 && a->state == b->state
1124 && a->config == b->config
1125 && a->curr == b->curr
1126 && a->advertised == b->advertised
1127 && a->supported == b->supported
1128 && a->peer == b->peer);
1132 send_port_status(struct ofproto *p, const struct ofport *ofport,
1135 /* XXX Should limit the number of queued port status change messages. */
1136 struct ofconn *ofconn;
1137 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1138 struct ofp_port_status *ops;
1141 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1142 ops->reason = reason;
1143 ops->desc = ofport->opp;
1144 hton_ofp_phy_port(&ops->desc);
1145 queue_tx(b, ofconn, NULL);
1147 if (p->ofhooks->port_changed_cb) {
1148 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1153 ofport_install(struct ofproto *p, struct ofport *ofport)
1155 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1156 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no),
1158 shash_add(&p->port_by_name, (char *) ofport->opp.name, ofport);
1162 ofport_remove(struct ofproto *p, struct ofport *ofport)
1164 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1165 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no), NULL);
1166 shash_delete(&p->port_by_name,
1167 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1171 ofport_free(struct ofport *ofport)
1174 netdev_close(ofport->netdev);
1180 update_port(struct ofproto *p, const char *devname)
1182 struct odp_port odp_port;
1183 struct ofport *old_ofport;
1184 struct ofport *new_ofport;
1187 COVERAGE_INC(ofproto_update_port);
1189 /* Query the datapath for port information. */
1190 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1192 /* Find the old ofport. */
1193 old_ofport = shash_find_data(&p->port_by_name, devname);
1196 /* There's no port named 'devname' but there might be a port with
1197 * the same port number. This could happen if a port is deleted
1198 * and then a new one added in its place very quickly, or if a port
1199 * is renamed. In the former case we want to send an OFPPR_DELETE
1200 * and an OFPPR_ADD, and in the latter case we want to send a
1201 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1202 * the old port's ifindex against the new port, or perhaps less
1203 * reliably but more portably by comparing the old port's MAC
1204 * against the new port's MAC. However, this code isn't that smart
1205 * and always sends an OFPPR_MODIFY (XXX). */
1206 old_ofport = port_array_get(&p->ports, odp_port.port);
1208 } else if (error != ENOENT && error != ENODEV) {
1209 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1210 "%s", strerror(error));
1214 /* Create a new ofport. */
1215 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1217 /* Eliminate a few pathological cases. */
1218 if (!old_ofport && !new_ofport) {
1220 } else if (old_ofport && new_ofport) {
1221 /* Most of the 'config' bits are OpenFlow soft state, but
1222 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1223 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1224 * leaves the other bits 0.) */
1225 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1227 if (ofport_equal(old_ofport, new_ofport)) {
1228 /* False alarm--no change. */
1229 ofport_free(new_ofport);
1234 /* Now deal with the normal cases. */
1236 ofport_remove(p, old_ofport);
1239 ofport_install(p, new_ofport);
1241 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1242 (!old_ofport ? OFPPR_ADD
1243 : !new_ofport ? OFPPR_DELETE
1245 ofport_free(old_ofport);
1247 /* Update port groups. */
1248 refresh_port_groups(p);
1252 init_ports(struct ofproto *p)
1254 struct odp_port *ports;
1259 error = dpif_port_list(p->dpif, &ports, &n_ports);
1264 for (i = 0; i < n_ports; i++) {
1265 const struct odp_port *odp_port = &ports[i];
1266 if (!ofport_conflicts(p, odp_port)) {
1267 struct ofport *ofport = make_ofport(odp_port);
1269 ofport_install(p, ofport);
1274 refresh_port_groups(p);
1278 static struct ofconn *
1279 ofconn_create(struct ofproto *p, struct rconn *rconn)
1281 struct ofconn *ofconn = xmalloc(sizeof *ofconn);
1282 list_push_back(&p->all_conns, &ofconn->node);
1283 ofconn->rconn = rconn;
1284 ofconn->pktbuf = NULL;
1285 ofconn->send_flow_exp = false;
1286 ofconn->miss_send_len = 0;
1287 ofconn->packet_in_counter = rconn_packet_counter_create ();
1288 ofconn->reply_counter = rconn_packet_counter_create ();
1293 ofconn_destroy(struct ofconn *ofconn)
1295 list_remove(&ofconn->node);
1296 rconn_destroy(ofconn->rconn);
1297 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1298 rconn_packet_counter_destroy(ofconn->reply_counter);
1299 pktbuf_destroy(ofconn->pktbuf);
1304 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1308 rconn_run(ofconn->rconn);
1310 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1311 /* Limit the number of iterations to prevent other tasks from
1313 for (iteration = 0; iteration < 50; iteration++) {
1314 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1319 fail_open_maybe_recover(p->fail_open);
1321 handle_openflow(ofconn, p, of_msg);
1322 ofpbuf_delete(of_msg);
1326 if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
1327 ofconn_destroy(ofconn);
1332 ofconn_wait(struct ofconn *ofconn)
1334 rconn_run_wait(ofconn->rconn);
1335 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1336 rconn_recv_wait(ofconn->rconn);
1338 COVERAGE_INC(ofproto_ofconn_stuck);
1342 /* Caller is responsible for initializing the 'cr' member of the returned
1344 static struct rule *
1345 rule_create(struct ofproto *ofproto, struct rule *super,
1346 const union ofp_action *actions, size_t n_actions,
1347 uint16_t idle_timeout, uint16_t hard_timeout)
1349 struct rule *rule = xzalloc(sizeof *rule);
1350 rule->idle_timeout = idle_timeout;
1351 rule->hard_timeout = hard_timeout;
1352 rule->used = rule->created = time_msec();
1353 rule->super = super;
1355 list_push_back(&super->list, &rule->list);
1357 list_init(&rule->list);
1359 rule->n_actions = n_actions;
1360 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1361 netflow_flow_clear(&rule->nf_flow);
1362 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1367 static struct rule *
1368 rule_from_cls_rule(const struct cls_rule *cls_rule)
1370 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1374 rule_free(struct rule *rule)
1376 free(rule->actions);
1377 free(rule->odp_actions);
1381 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1382 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1383 * through all of its subrules and revalidates them, destroying any that no
1384 * longer has a super-rule (which is probably all of them).
1386 * Before calling this function, the caller must make have removed 'rule' from
1387 * the classifier. If 'rule' is an exact-match rule, the caller is also
1388 * responsible for ensuring that it has been uninstalled from the datapath. */
1390 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1393 struct rule *subrule, *next;
1394 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1395 revalidate_rule(ofproto, subrule);
1398 list_remove(&rule->list);
1404 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1406 const union ofp_action *oa;
1407 struct actions_iterator i;
1409 if (out_port == htons(OFPP_NONE)) {
1412 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1413 oa = actions_next(&i)) {
1414 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1421 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1422 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1424 * The flow that 'packet' actually contains does not need to actually match
1425 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1426 * the packet and byte counters for 'rule' will be credited for the packet sent
1427 * out whether or not the packet actually matches 'rule'.
1429 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1430 * the caller must already have accurately composed ODP actions for it given
1431 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1432 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1433 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1434 * actions and apply them to 'packet'. */
1436 rule_execute(struct ofproto *ofproto, struct rule *rule,
1437 struct ofpbuf *packet, const flow_t *flow)
1439 const union odp_action *actions;
1441 struct odp_actions a;
1443 /* Grab or compose the ODP actions.
1445 * The special case for an exact-match 'rule' where 'flow' is not the
1446 * rule's flow is important to avoid, e.g., sending a packet out its input
1447 * port simply because the ODP actions were composed for the wrong
1449 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1450 struct rule *super = rule->super ? rule->super : rule;
1451 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1452 packet, &a, NULL, 0, NULL)) {
1455 actions = a.actions;
1456 n_actions = a.n_actions;
1458 actions = rule->odp_actions;
1459 n_actions = rule->n_odp_actions;
1462 /* Execute the ODP actions. */
1463 if (!dpif_execute(ofproto->dpif, flow->in_port,
1464 actions, n_actions, packet)) {
1465 struct odp_flow_stats stats;
1466 flow_extract_stats(flow, packet, &stats);
1467 update_stats(ofproto, rule, &stats);
1468 rule->used = time_msec();
1469 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1474 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1477 struct rule *displaced_rule;
1479 /* Insert the rule in the classifier. */
1480 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1481 if (!rule->cr.wc.wildcards) {
1482 rule_make_actions(p, rule, packet);
1485 /* Send the packet and credit it to the rule. */
1488 flow_extract(packet, in_port, &flow);
1489 rule_execute(p, rule, packet, &flow);
1492 /* Install the rule in the datapath only after sending the packet, to
1493 * avoid packet reordering. */
1494 if (rule->cr.wc.wildcards) {
1495 COVERAGE_INC(ofproto_add_wc_flow);
1496 p->need_revalidate = true;
1498 rule_install(p, rule, displaced_rule);
1501 /* Free the rule that was displaced, if any. */
1502 if (displaced_rule) {
1503 rule_destroy(p, displaced_rule);
1507 static struct rule *
1508 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1511 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1512 rule->idle_timeout, rule->hard_timeout);
1513 COVERAGE_INC(ofproto_subrule_create);
1514 cls_rule_from_flow(&subrule->cr, flow, 0,
1515 (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1516 : rule->cr.priority));
1517 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1523 rule_remove(struct ofproto *ofproto, struct rule *rule)
1525 if (rule->cr.wc.wildcards) {
1526 COVERAGE_INC(ofproto_del_wc_flow);
1527 ofproto->need_revalidate = true;
1529 rule_uninstall(ofproto, rule);
1531 classifier_remove(&ofproto->cls, &rule->cr);
1532 rule_destroy(ofproto, rule);
1535 /* Returns true if the actions changed, false otherwise. */
1537 rule_make_actions(struct ofproto *p, struct rule *rule,
1538 const struct ofpbuf *packet)
1540 const struct rule *super;
1541 struct odp_actions a;
1544 assert(!rule->cr.wc.wildcards);
1546 super = rule->super ? rule->super : rule;
1548 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1549 packet, &a, &rule->tags, &rule->may_install,
1550 &rule->nf_flow.output_iface);
1552 actions_len = a.n_actions * sizeof *a.actions;
1553 if (rule->n_odp_actions != a.n_actions
1554 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1555 COVERAGE_INC(ofproto_odp_unchanged);
1556 free(rule->odp_actions);
1557 rule->n_odp_actions = a.n_actions;
1558 rule->odp_actions = xmemdup(a.actions, actions_len);
1566 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1567 struct odp_flow_put *put)
1569 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1570 put->flow.key = rule->cr.flow;
1571 put->flow.actions = rule->odp_actions;
1572 put->flow.n_actions = rule->n_odp_actions;
1574 return dpif_flow_put(ofproto->dpif, put);
1578 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1580 assert(!rule->cr.wc.wildcards);
1582 if (rule->may_install) {
1583 struct odp_flow_put put;
1584 if (!do_put_flow(p, rule,
1585 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1587 rule->installed = true;
1588 if (displaced_rule) {
1589 update_stats(p, displaced_rule, &put.flow.stats);
1590 rule_post_uninstall(p, displaced_rule);
1593 } else if (displaced_rule) {
1594 rule_uninstall(p, displaced_rule);
1599 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1601 if (rule->installed) {
1602 struct odp_flow_put put;
1603 COVERAGE_INC(ofproto_dp_missed);
1604 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1606 rule_install(ofproto, rule, NULL);
1611 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1613 bool actions_changed;
1614 uint16_t new_out_iface, old_out_iface;
1616 old_out_iface = rule->nf_flow.output_iface;
1617 actions_changed = rule_make_actions(ofproto, rule, NULL);
1619 if (rule->may_install) {
1620 if (rule->installed) {
1621 if (actions_changed) {
1622 struct odp_flow_put put;
1623 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
1624 | ODPPF_ZERO_STATS, &put);
1625 update_stats(ofproto, rule, &put.flow.stats);
1627 /* Temporarily set the old output iface so that NetFlow
1628 * messages have the correct output interface for the old
1630 new_out_iface = rule->nf_flow.output_iface;
1631 rule->nf_flow.output_iface = old_out_iface;
1632 rule_post_uninstall(ofproto, rule);
1633 rule->nf_flow.output_iface = new_out_iface;
1636 rule_install(ofproto, rule, NULL);
1639 rule_uninstall(ofproto, rule);
1644 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1646 uint64_t total_bytes = rule->byte_count + extra_bytes;
1648 if (ofproto->ofhooks->account_flow_cb
1649 && total_bytes > rule->accounted_bytes)
1651 ofproto->ofhooks->account_flow_cb(
1652 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1653 total_bytes - rule->accounted_bytes, ofproto->aux);
1654 rule->accounted_bytes = total_bytes;
1659 rule_uninstall(struct ofproto *p, struct rule *rule)
1661 assert(!rule->cr.wc.wildcards);
1662 if (rule->installed) {
1663 struct odp_flow odp_flow;
1665 odp_flow.key = rule->cr.flow;
1666 odp_flow.actions = NULL;
1667 odp_flow.n_actions = 0;
1668 if (!dpif_flow_del(p->dpif, &odp_flow)) {
1669 update_stats(p, rule, &odp_flow.stats);
1671 rule->installed = false;
1673 rule_post_uninstall(p, rule);
1678 is_controller_rule(struct rule *rule)
1680 /* If the only action is send to the controller then don't report
1681 * NetFlow expiration messages since it is just part of the control
1682 * logic for the network and not real traffic. */
1684 if (rule && rule->super) {
1685 struct rule *super = rule->super;
1687 return super->n_actions == 1 &&
1688 super->actions[0].type == htons(OFPAT_OUTPUT) &&
1689 super->actions[0].output.port == htons(OFPP_CONTROLLER);
1696 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1698 struct rule *super = rule->super;
1700 rule_account(ofproto, rule, 0);
1702 if (ofproto->netflow && !is_controller_rule(rule)) {
1703 struct ofexpired expired;
1704 expired.flow = rule->cr.flow;
1705 expired.packet_count = rule->packet_count;
1706 expired.byte_count = rule->byte_count;
1707 expired.used = rule->used;
1708 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
1711 super->packet_count += rule->packet_count;
1712 super->byte_count += rule->byte_count;
1714 /* Reset counters to prevent double counting if the rule ever gets
1716 rule->packet_count = 0;
1717 rule->byte_count = 0;
1718 rule->accounted_bytes = 0;
1720 netflow_flow_clear(&rule->nf_flow);
1725 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1726 struct rconn_packet_counter *counter)
1728 update_openflow_length(msg);
1729 if (rconn_send(ofconn->rconn, msg, counter)) {
1735 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1736 int error, const void *data, size_t len)
1739 struct ofp_error_msg *oem;
1741 if (!(error >> 16)) {
1742 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1747 COVERAGE_INC(ofproto_error);
1748 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1749 oh ? oh->xid : 0, &buf);
1750 oem->type = htons((unsigned int) error >> 16);
1751 oem->code = htons(error & 0xffff);
1752 memcpy(oem->data, data, len);
1753 queue_tx(buf, ofconn, ofconn->reply_counter);
1757 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1760 size_t oh_length = ntohs(oh->length);
1761 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1765 hton_ofp_phy_port(struct ofp_phy_port *opp)
1767 opp->port_no = htons(opp->port_no);
1768 opp->config = htonl(opp->config);
1769 opp->state = htonl(opp->state);
1770 opp->curr = htonl(opp->curr);
1771 opp->advertised = htonl(opp->advertised);
1772 opp->supported = htonl(opp->supported);
1773 opp->peer = htonl(opp->peer);
1777 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
1779 struct ofp_header *rq = oh;
1780 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
1785 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
1786 struct ofp_header *oh)
1788 struct ofp_switch_features *osf;
1790 unsigned int port_no;
1791 struct ofport *port;
1793 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1794 osf->datapath_id = htonll(p->datapath_id);
1795 osf->n_buffers = htonl(pktbuf_capacity());
1797 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1798 OFPC_PORT_STATS | OFPC_MULTI_PHY_TX);
1799 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1800 (1u << OFPAT_SET_VLAN_VID) |
1801 (1u << OFPAT_SET_VLAN_PCP) |
1802 (1u << OFPAT_STRIP_VLAN) |
1803 (1u << OFPAT_SET_DL_SRC) |
1804 (1u << OFPAT_SET_DL_DST) |
1805 (1u << OFPAT_SET_NW_SRC) |
1806 (1u << OFPAT_SET_NW_DST) |
1807 (1u << OFPAT_SET_TP_SRC) |
1808 (1u << OFPAT_SET_TP_DST));
1810 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1811 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1814 queue_tx(buf, ofconn, ofconn->reply_counter);
1819 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1820 struct ofp_header *oh)
1823 struct ofp_switch_config *osc;
1827 /* Figure out flags. */
1828 dpif_get_drop_frags(p->dpif, &drop_frags);
1829 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1830 if (ofconn->send_flow_exp) {
1831 flags |= OFPC_SEND_FLOW_EXP;
1835 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1836 osc->flags = htons(flags);
1837 osc->miss_send_len = htons(ofconn->miss_send_len);
1838 queue_tx(buf, ofconn, ofconn->reply_counter);
1844 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1845 struct ofp_switch_config *osc)
1850 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1854 flags = ntohs(osc->flags);
1856 ofconn->send_flow_exp = (flags & OFPC_SEND_FLOW_EXP) != 0;
1858 if (ofconn == p->controller) {
1859 switch (flags & OFPC_FRAG_MASK) {
1860 case OFPC_FRAG_NORMAL:
1861 dpif_set_drop_frags(p->dpif, false);
1863 case OFPC_FRAG_DROP:
1864 dpif_set_drop_frags(p->dpif, true);
1867 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1873 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1874 if (ntohs(osc->miss_send_len) != 0) {
1875 ofconn->pktbuf = pktbuf_create();
1877 pktbuf_destroy(ofconn->pktbuf);
1881 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1887 add_output_group_action(struct odp_actions *actions, uint16_t group,
1888 uint16_t *nf_output_iface)
1890 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1892 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
1893 *nf_output_iface = NF_OUT_FLOOD;
1898 add_controller_action(struct odp_actions *actions,
1899 const struct ofp_action_output *oao)
1901 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
1902 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
1905 struct action_xlate_ctx {
1907 const flow_t *flow; /* Flow to which these actions correspond. */
1908 int recurse; /* Recursion level, via xlate_table_action. */
1909 struct ofproto *ofproto;
1910 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
1911 * null pointer if we are revalidating
1912 * without a packet to refer to. */
1915 struct odp_actions *out; /* Datapath actions. */
1916 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
1917 bool may_set_up_flow; /* True ordinarily; false if the actions must
1918 * be reassessed for every packet. */
1919 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
1922 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
1923 struct action_xlate_ctx *ctx);
1926 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
1928 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
1931 if (ofport->opp.config & OFPPC_NO_FWD) {
1932 /* Forwarding disabled on port. */
1937 * We don't have an ofport record for this port, but it doesn't hurt to
1938 * allow forwarding to it anyhow. Maybe such a port will appear later
1939 * and we're pre-populating the flow table.
1943 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
1944 ctx->nf_output_iface = port;
1947 static struct rule *
1948 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
1951 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
1953 /* The rule we found might not be valid, since we could be in need of
1954 * revalidation. If it is not valid, don't return it. */
1957 && ofproto->need_revalidate
1958 && !revalidate_rule(ofproto, rule)) {
1959 COVERAGE_INC(ofproto_invalidated);
1967 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
1969 if (!ctx->recurse) {
1974 flow.in_port = in_port;
1976 rule = lookup_valid_rule(ctx->ofproto, &flow);
1983 do_xlate_actions(rule->actions, rule->n_actions, ctx);
1990 xlate_output_action(struct action_xlate_ctx *ctx,
1991 const struct ofp_action_output *oao)
1994 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
1996 ctx->nf_output_iface = NF_OUT_DROP;
1998 switch (ntohs(oao->port)) {
2000 add_output_action(ctx, ctx->flow->in_port);
2003 xlate_table_action(ctx, ctx->flow->in_port);
2006 if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet,
2007 ctx->out, ctx->tags,
2008 &ctx->nf_output_iface,
2009 ctx->ofproto->aux)) {
2010 COVERAGE_INC(ofproto_uninstallable);
2011 ctx->may_set_up_flow = false;
2015 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2016 &ctx->nf_output_iface);
2019 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2021 case OFPP_CONTROLLER:
2022 add_controller_action(ctx->out, oao);
2025 add_output_action(ctx, ODPP_LOCAL);
2028 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2029 if (odp_port != ctx->flow->in_port) {
2030 add_output_action(ctx, odp_port);
2035 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2036 ctx->nf_output_iface = NF_OUT_FLOOD;
2037 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2038 ctx->nf_output_iface = prev_nf_output_iface;
2039 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2040 ctx->nf_output_iface != NF_OUT_FLOOD) {
2041 ctx->nf_output_iface = NF_OUT_MULTI;
2046 xlate_nicira_action(struct action_xlate_ctx *ctx,
2047 const struct nx_action_header *nah)
2049 const struct nx_action_resubmit *nar;
2050 int subtype = ntohs(nah->subtype);
2052 assert(nah->vendor == htonl(NX_VENDOR_ID));
2054 case NXAST_RESUBMIT:
2055 nar = (const struct nx_action_resubmit *) nah;
2056 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2060 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2066 do_xlate_actions(const union ofp_action *in, size_t n_in,
2067 struct action_xlate_ctx *ctx)
2069 struct actions_iterator iter;
2070 const union ofp_action *ia;
2071 const struct ofport *port;
2073 port = port_array_get(&ctx->ofproto->ports, ctx->flow->in_port);
2074 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2075 port->opp.config & (eth_addr_equals(ctx->flow->dl_dst, stp_eth_addr)
2076 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2077 /* Drop this flow. */
2081 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2082 uint16_t type = ntohs(ia->type);
2083 union odp_action *oa;
2087 xlate_output_action(ctx, &ia->output);
2090 case OFPAT_SET_VLAN_VID:
2091 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2092 oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2095 case OFPAT_SET_VLAN_PCP:
2096 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2097 oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2100 case OFPAT_STRIP_VLAN:
2101 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2104 case OFPAT_SET_DL_SRC:
2105 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2106 memcpy(oa->dl_addr.dl_addr,
2107 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2110 case OFPAT_SET_DL_DST:
2111 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2112 memcpy(oa->dl_addr.dl_addr,
2113 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2116 case OFPAT_SET_NW_SRC:
2117 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2118 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2121 case OFPAT_SET_NW_DST:
2122 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2123 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2126 case OFPAT_SET_TP_SRC:
2127 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2128 oa->tp_port.tp_port = ia->tp_port.tp_port;
2131 case OFPAT_SET_TP_DST:
2132 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2133 oa->tp_port.tp_port = ia->tp_port.tp_port;
2137 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2141 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2148 xlate_actions(const union ofp_action *in, size_t n_in,
2149 const flow_t *flow, struct ofproto *ofproto,
2150 const struct ofpbuf *packet,
2151 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2152 uint16_t *nf_output_iface)
2154 tag_type no_tags = 0;
2155 struct action_xlate_ctx ctx;
2156 COVERAGE_INC(ofproto_ofp2odp);
2157 odp_actions_init(out);
2160 ctx.ofproto = ofproto;
2161 ctx.packet = packet;
2163 ctx.tags = tags ? tags : &no_tags;
2164 ctx.may_set_up_flow = true;
2165 ctx.nf_output_iface = NF_OUT_DROP;
2166 do_xlate_actions(in, n_in, &ctx);
2168 /* Check with in-band control to see if we're allowed to set up this
2170 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2171 ctx.may_set_up_flow = false;
2174 if (may_set_up_flow) {
2175 *may_set_up_flow = ctx.may_set_up_flow;
2177 if (nf_output_iface) {
2178 *nf_output_iface = ctx.nf_output_iface;
2180 if (odp_actions_overflow(out)) {
2181 odp_actions_init(out);
2182 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2188 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2189 struct ofp_header *oh)
2191 struct ofp_packet_out *opo;
2192 struct ofpbuf payload, *buffer;
2193 struct odp_actions actions;
2199 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2203 opo = (struct ofp_packet_out *) oh;
2205 COVERAGE_INC(ofproto_packet_out);
2206 if (opo->buffer_id != htonl(UINT32_MAX)) {
2207 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2209 if (error || !buffer) {
2217 flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2218 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2219 &flow, p, &payload, &actions, NULL, NULL, NULL);
2224 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2226 ofpbuf_delete(buffer);
2232 update_port_config(struct ofproto *p, struct ofport *port,
2233 uint32_t config, uint32_t mask)
2235 mask &= config ^ port->opp.config;
2236 if (mask & OFPPC_PORT_DOWN) {
2237 if (config & OFPPC_PORT_DOWN) {
2238 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2240 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2243 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2244 if (mask & REVALIDATE_BITS) {
2245 COVERAGE_INC(ofproto_costly_flags);
2246 port->opp.config ^= mask & REVALIDATE_BITS;
2247 p->need_revalidate = true;
2249 #undef REVALIDATE_BITS
2250 if (mask & OFPPC_NO_FLOOD) {
2251 port->opp.config ^= OFPPC_NO_FLOOD;
2252 refresh_port_group(p, DP_GROUP_FLOOD);
2254 if (mask & OFPPC_NO_PACKET_IN) {
2255 port->opp.config ^= OFPPC_NO_PACKET_IN;
2260 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2262 const struct ofp_port_mod *opm;
2263 struct ofport *port;
2266 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2270 opm = (struct ofp_port_mod *) oh;
2272 port = port_array_get(&p->ports,
2273 ofp_port_to_odp_port(ntohs(opm->port_no)));
2275 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2276 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2277 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2279 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2280 if (opm->advertise) {
2281 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2287 static struct ofpbuf *
2288 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2290 struct ofp_stats_reply *osr;
2293 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2294 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2296 osr->flags = htons(0);
2300 static struct ofpbuf *
2301 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2303 return make_stats_reply(request->header.xid, request->type, body_len);
2307 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2309 struct ofpbuf *msg = *msgp;
2310 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2311 if (nbytes + msg->size > UINT16_MAX) {
2312 struct ofp_stats_reply *reply = msg->data;
2313 reply->flags = htons(OFPSF_REPLY_MORE);
2314 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2315 queue_tx(msg, ofconn, ofconn->reply_counter);
2317 return ofpbuf_put_uninit(*msgp, nbytes);
2321 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2322 struct ofp_stats_request *request)
2324 struct ofp_desc_stats *ods;
2327 msg = start_stats_reply(request, sizeof *ods);
2328 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2329 strncpy(ods->mfr_desc, p->manufacturer, sizeof ods->mfr_desc);
2330 strncpy(ods->hw_desc, p->hardware, sizeof ods->hw_desc);
2331 strncpy(ods->sw_desc, p->software, sizeof ods->sw_desc);
2332 strncpy(ods->serial_num, p->serial, sizeof ods->serial_num);
2333 queue_tx(msg, ofconn, ofconn->reply_counter);
2339 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2341 struct rule *rule = rule_from_cls_rule(cls_rule);
2342 int *n_subrules = n_subrules_;
2350 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2351 struct ofp_stats_request *request)
2353 struct ofp_table_stats *ots;
2355 struct odp_stats dpstats;
2356 int n_exact, n_subrules, n_wild;
2358 msg = start_stats_reply(request, sizeof *ots * 2);
2360 /* Count rules of various kinds. */
2362 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2363 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2364 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2367 dpif_get_dp_stats(p->dpif, &dpstats);
2368 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2369 memset(ots, 0, sizeof *ots);
2370 ots->table_id = TABLEID_HASH;
2371 strcpy(ots->name, "hash");
2372 ots->wildcards = htonl(0);
2373 ots->max_entries = htonl(dpstats.max_capacity);
2374 ots->active_count = htonl(n_exact);
2375 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2377 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2379 /* Classifier table. */
2380 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2381 memset(ots, 0, sizeof *ots);
2382 ots->table_id = TABLEID_CLASSIFIER;
2383 strcpy(ots->name, "classifier");
2384 ots->wildcards = htonl(OFPFW_ALL);
2385 ots->max_entries = htonl(65536);
2386 ots->active_count = htonl(n_wild);
2387 ots->lookup_count = htonll(0); /* XXX */
2388 ots->matched_count = htonll(0); /* XXX */
2390 queue_tx(msg, ofconn, ofconn->reply_counter);
2395 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2396 struct ofp_stats_request *request)
2398 struct ofp_port_stats *ops;
2400 struct ofport *port;
2401 unsigned int port_no;
2403 msg = start_stats_reply(request, sizeof *ops * 16);
2404 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2405 struct netdev_stats stats;
2407 /* Intentionally ignore return value, since errors will set 'stats' to
2408 * all-1s, which is correct for OpenFlow, and netdev_get_stats() will
2410 netdev_get_stats(port->netdev, &stats);
2412 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2413 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2414 memset(ops->pad, 0, sizeof ops->pad);
2415 ops->rx_packets = htonll(stats.rx_packets);
2416 ops->tx_packets = htonll(stats.tx_packets);
2417 ops->rx_bytes = htonll(stats.rx_bytes);
2418 ops->tx_bytes = htonll(stats.tx_bytes);
2419 ops->rx_dropped = htonll(stats.rx_dropped);
2420 ops->tx_dropped = htonll(stats.tx_dropped);
2421 ops->rx_errors = htonll(stats.rx_errors);
2422 ops->tx_errors = htonll(stats.tx_errors);
2423 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2424 ops->rx_over_err = htonll(stats.rx_over_errors);
2425 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2426 ops->collisions = htonll(stats.collisions);
2429 queue_tx(msg, ofconn, ofconn->reply_counter);
2433 struct flow_stats_cbdata {
2434 struct ofproto *ofproto;
2435 struct ofconn *ofconn;
2441 query_stats(struct ofproto *p, struct rule *rule,
2442 uint64_t *packet_countp, uint64_t *byte_countp)
2444 uint64_t packet_count, byte_count;
2445 struct rule *subrule;
2446 struct odp_flow *odp_flows;
2449 packet_count = rule->packet_count;
2450 byte_count = rule->byte_count;
2452 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2453 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
2454 if (rule->cr.wc.wildcards) {
2456 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2457 odp_flows[i++].key = subrule->cr.flow;
2458 packet_count += subrule->packet_count;
2459 byte_count += subrule->byte_count;
2462 odp_flows[0].key = rule->cr.flow;
2465 packet_count = rule->packet_count;
2466 byte_count = rule->byte_count;
2467 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2469 for (i = 0; i < n_odp_flows; i++) {
2470 struct odp_flow *odp_flow = &odp_flows[i];
2471 packet_count += odp_flow->stats.n_packets;
2472 byte_count += odp_flow->stats.n_bytes;
2477 *packet_countp = packet_count;
2478 *byte_countp = byte_count;
2482 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2484 struct rule *rule = rule_from_cls_rule(rule_);
2485 struct flow_stats_cbdata *cbdata = cbdata_;
2486 struct ofp_flow_stats *ofs;
2487 uint64_t packet_count, byte_count;
2488 size_t act_len, len;
2490 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2494 act_len = sizeof *rule->actions * rule->n_actions;
2495 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2497 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2499 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2500 ofs->length = htons(len);
2501 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2503 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match);
2504 ofs->duration = htonl((time_msec() - rule->created) / 1000);
2505 ofs->priority = htons(rule->cr.priority);
2506 ofs->idle_timeout = htons(rule->idle_timeout);
2507 ofs->hard_timeout = htons(rule->hard_timeout);
2508 memset(ofs->pad2, 0, sizeof ofs->pad2);
2509 ofs->packet_count = htonll(packet_count);
2510 ofs->byte_count = htonll(byte_count);
2511 memcpy(ofs->actions, rule->actions, act_len);
2515 table_id_to_include(uint8_t table_id)
2517 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2518 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2519 : table_id == 0xff ? CLS_INC_ALL
2524 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2525 const struct ofp_stats_request *osr,
2528 struct ofp_flow_stats_request *fsr;
2529 struct flow_stats_cbdata cbdata;
2530 struct cls_rule target;
2532 if (arg_size != sizeof *fsr) {
2533 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2535 fsr = (struct ofp_flow_stats_request *) osr->body;
2537 COVERAGE_INC(ofproto_flows_req);
2539 cbdata.ofconn = ofconn;
2540 cbdata.out_port = fsr->out_port;
2541 cbdata.msg = start_stats_reply(osr, 1024);
2542 cls_rule_from_match(&target, &fsr->match, 0);
2543 classifier_for_each_match(&p->cls, &target,
2544 table_id_to_include(fsr->table_id),
2545 flow_stats_cb, &cbdata);
2546 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2550 struct flow_stats_ds_cbdata {
2551 struct ofproto *ofproto;
2556 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
2558 struct rule *rule = rule_from_cls_rule(rule_);
2559 struct flow_stats_ds_cbdata *cbdata = cbdata_;
2560 struct ds *results = cbdata->results;
2561 struct ofp_match match;
2562 uint64_t packet_count, byte_count;
2563 size_t act_len = sizeof *rule->actions * rule->n_actions;
2565 /* Don't report on subrules. */
2566 if (rule->super != NULL) {
2570 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2571 flow_to_ovs_match(&rule->cr.flow, rule->cr.wc.wildcards, &match);
2573 ds_put_format(results, "duration=%llds, ",
2574 (time_msec() - rule->created) / 1000);
2575 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2576 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2577 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2578 ofp_print_match(results, &match, true);
2579 ofp_print_actions(results, &rule->actions->header, act_len);
2580 ds_put_cstr(results, "\n");
2583 /* Adds a pretty-printed description of all flows to 'results', including
2584 * those marked hidden by secchan (e.g., by in-band control). */
2586 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2588 struct ofp_match match;
2589 struct cls_rule target;
2590 struct flow_stats_ds_cbdata cbdata;
2592 memset(&match, 0, sizeof match);
2593 match.wildcards = htonl(OFPFW_ALL);
2596 cbdata.results = results;
2598 cls_rule_from_match(&target, &match, 0);
2599 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2600 flow_stats_ds_cb, &cbdata);
2603 struct aggregate_stats_cbdata {
2604 struct ofproto *ofproto;
2606 uint64_t packet_count;
2607 uint64_t byte_count;
2612 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2614 struct rule *rule = rule_from_cls_rule(rule_);
2615 struct aggregate_stats_cbdata *cbdata = cbdata_;
2616 uint64_t packet_count, byte_count;
2618 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2622 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2624 cbdata->packet_count += packet_count;
2625 cbdata->byte_count += byte_count;
2630 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2631 const struct ofp_stats_request *osr,
2634 struct ofp_aggregate_stats_request *asr;
2635 struct ofp_aggregate_stats_reply *reply;
2636 struct aggregate_stats_cbdata cbdata;
2637 struct cls_rule target;
2640 if (arg_size != sizeof *asr) {
2641 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2643 asr = (struct ofp_aggregate_stats_request *) osr->body;
2645 COVERAGE_INC(ofproto_agg_request);
2647 cbdata.out_port = asr->out_port;
2648 cbdata.packet_count = 0;
2649 cbdata.byte_count = 0;
2651 cls_rule_from_match(&target, &asr->match, 0);
2652 classifier_for_each_match(&p->cls, &target,
2653 table_id_to_include(asr->table_id),
2654 aggregate_stats_cb, &cbdata);
2656 msg = start_stats_reply(osr, sizeof *reply);
2657 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2658 reply->flow_count = htonl(cbdata.n_flows);
2659 reply->packet_count = htonll(cbdata.packet_count);
2660 reply->byte_count = htonll(cbdata.byte_count);
2661 queue_tx(msg, ofconn, ofconn->reply_counter);
2666 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2667 struct ofp_header *oh)
2669 struct ofp_stats_request *osr;
2673 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2678 osr = (struct ofp_stats_request *) oh;
2680 switch (ntohs(osr->type)) {
2682 return handle_desc_stats_request(p, ofconn, osr);
2685 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2687 case OFPST_AGGREGATE:
2688 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2691 return handle_table_stats_request(p, ofconn, osr);
2694 return handle_port_stats_request(p, ofconn, osr);
2697 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2700 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2704 static long long int
2705 msec_from_nsec(uint64_t sec, uint32_t nsec)
2707 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2711 update_time(struct ofproto *ofproto, struct rule *rule,
2712 const struct odp_flow_stats *stats)
2714 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2715 if (used > rule->used) {
2717 if (rule->super && used > rule->super->used) {
2718 rule->super->used = used;
2720 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
2725 update_stats(struct ofproto *ofproto, struct rule *rule,
2726 const struct odp_flow_stats *stats)
2728 if (stats->n_packets) {
2729 update_time(ofproto, rule, stats);
2730 rule->packet_count += stats->n_packets;
2731 rule->byte_count += stats->n_bytes;
2732 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
2738 add_flow(struct ofproto *p, struct ofconn *ofconn,
2739 struct ofp_flow_mod *ofm, size_t n_actions)
2741 struct ofpbuf *packet;
2746 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
2747 n_actions, ntohs(ofm->idle_timeout),
2748 ntohs(ofm->hard_timeout));
2749 cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority));
2753 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2754 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2758 rule_insert(p, rule, packet, in_port);
2759 ofpbuf_delete(packet);
2764 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
2765 size_t n_actions, uint16_t command, struct rule *rule)
2767 if (rule_is_hidden(rule)) {
2771 if (command == OFPFC_DELETE) {
2772 rule_remove(p, rule);
2774 size_t actions_len = n_actions * sizeof *rule->actions;
2776 if (n_actions == rule->n_actions
2777 && !memcmp(ofm->actions, rule->actions, actions_len))
2782 free(rule->actions);
2783 rule->actions = xmemdup(ofm->actions, actions_len);
2784 rule->n_actions = n_actions;
2786 if (rule->cr.wc.wildcards) {
2787 COVERAGE_INC(ofproto_mod_wc_flow);
2788 p->need_revalidate = true;
2790 rule_update_actions(p, rule);
2798 modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm,
2799 size_t n_actions, uint16_t command)
2805 flow_from_match(&flow, &wildcards, &ofm->match);
2806 rule = rule_from_cls_rule(classifier_find_rule_exactly(
2807 &p->cls, &flow, wildcards,
2808 ntohs(ofm->priority)));
2811 if (command == OFPFC_DELETE
2812 && ofm->out_port != htons(OFPP_NONE)
2813 && !rule_has_out_port(rule, ofm->out_port)) {
2817 modify_flow(p, ofm, n_actions, command, rule);
2822 struct modify_flows_cbdata {
2823 struct ofproto *ofproto;
2824 const struct ofp_flow_mod *ofm;
2831 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
2833 struct rule *rule = rule_from_cls_rule(rule_);
2834 struct modify_flows_cbdata *cbdata = cbdata_;
2836 if (cbdata->out_port != htons(OFPP_NONE)
2837 && !rule_has_out_port(rule, cbdata->out_port)) {
2841 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions,
2842 cbdata->command, rule);
2846 modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm,
2847 size_t n_actions, uint16_t command)
2849 struct modify_flows_cbdata cbdata;
2850 struct cls_rule target;
2854 cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port
2855 : htons(OFPP_NONE));
2856 cbdata.n_actions = n_actions;
2857 cbdata.command = command;
2859 cls_rule_from_match(&target, &ofm->match, 0);
2861 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2862 modify_flows_cb, &cbdata);
2867 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
2868 struct ofp_flow_mod *ofm)
2873 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
2874 sizeof *ofm->actions, &n_actions);
2879 normalize_match(&ofm->match);
2880 if (!ofm->match.wildcards) {
2881 ofm->priority = htons(UINT16_MAX);
2884 error = validate_actions((const union ofp_action *) ofm->actions,
2885 n_actions, p->max_ports);
2890 switch (ntohs(ofm->command)) {
2892 return add_flow(p, ofconn, ofm, n_actions);
2895 return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY);
2897 case OFPFC_MODIFY_STRICT:
2898 return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY);
2901 return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE);
2903 case OFPFC_DELETE_STRICT:
2904 return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE);
2907 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
2912 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
2914 struct ofp_vendor_header *ovh = msg;
2915 struct nicira_header *nh;
2917 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
2918 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2920 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
2921 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2923 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
2924 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2928 switch (ntohl(nh->subtype)) {
2929 case NXT_STATUS_REQUEST:
2930 return switch_status_handle_request(p->switch_status, ofconn->rconn,
2933 case NXT_ACT_SET_CONFIG:
2934 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
2936 case NXT_ACT_GET_CONFIG:
2937 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
2940 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2944 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
2945 struct ofpbuf *ofp_msg)
2947 struct ofp_header *oh = ofp_msg->data;
2950 COVERAGE_INC(ofproto_recv_openflow);
2952 case OFPT_ECHO_REQUEST:
2953 error = handle_echo_request(ofconn, oh);
2956 case OFPT_ECHO_REPLY:
2960 case OFPT_FEATURES_REQUEST:
2961 error = handle_features_request(p, ofconn, oh);
2964 case OFPT_GET_CONFIG_REQUEST:
2965 error = handle_get_config_request(p, ofconn, oh);
2968 case OFPT_SET_CONFIG:
2969 error = handle_set_config(p, ofconn, ofp_msg->data);
2972 case OFPT_PACKET_OUT:
2973 error = handle_packet_out(p, ofconn, ofp_msg->data);
2977 error = handle_port_mod(p, oh);
2981 error = handle_flow_mod(p, ofconn, ofp_msg->data);
2984 case OFPT_STATS_REQUEST:
2985 error = handle_stats_request(p, ofconn, oh);
2989 error = handle_vendor(p, ofconn, ofp_msg->data);
2993 if (VLOG_IS_WARN_ENABLED()) {
2994 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
2995 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
2998 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3003 send_error_oh(ofconn, ofp_msg->data, error);
3008 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3010 struct odp_msg *msg = packet->data;
3011 uint16_t in_port = odp_port_to_ofp_port(msg->port);
3013 struct ofpbuf payload;
3016 /* Handle controller actions. */
3017 if (msg->type == _ODPL_ACTION_NR) {
3018 COVERAGE_INC(ofproto_ctlr_action);
3019 pinsched_send(p->action_sched, in_port, packet,
3020 send_packet_in_action, p);
3024 payload.data = msg + 1;
3025 payload.size = msg->length - sizeof *msg;
3026 flow_extract(&payload, msg->port, &flow);
3028 /* Check with in-band control to see if this packet should be sent
3029 * to the local port regardless of the flow table. */
3030 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3031 union odp_action action;
3033 memset(&action, 0, sizeof(action));
3034 action.output.type = ODPAT_OUTPUT;
3035 action.output.port = ODPP_LOCAL;
3036 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3039 rule = lookup_valid_rule(p, &flow);
3041 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3042 struct ofport *port = port_array_get(&p->ports, msg->port);
3044 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3045 COVERAGE_INC(ofproto_no_packet_in);
3046 /* XXX install 'drop' flow entry */
3047 ofpbuf_delete(packet);
3051 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3054 COVERAGE_INC(ofproto_packet_in);
3055 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3059 if (rule->cr.wc.wildcards) {
3060 rule = rule_create_subrule(p, rule, &flow);
3061 rule_make_actions(p, rule, packet);
3063 if (!rule->may_install) {
3064 /* The rule is not installable, that is, we need to process every
3065 * packet, so process the current packet and set its actions into
3067 rule_make_actions(p, rule, packet);
3069 /* XXX revalidate rule if it needs it */
3073 rule_execute(p, rule, &payload, &flow);
3074 rule_reinstall(p, rule);
3076 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY
3077 && rconn_is_connected(p->controller->rconn)) {
3079 * Extra-special case for fail-open mode.
3081 * We are in fail-open mode and the packet matched the fail-open rule,
3082 * but we are connected to a controller too. We should send the packet
3083 * up to the controller in the hope that it will try to set up a flow
3084 * and thereby allow us to exit fail-open.
3086 * See the top-level comment in fail-open.c for more information.
3088 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3090 ofpbuf_delete(packet);
3095 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3097 struct rule *sub = rule_from_cls_rule(sub_);
3098 struct revalidate_cbdata *cbdata = cbdata_;
3100 if (cbdata->revalidate_all
3101 || (cbdata->revalidate_subrules && sub->super)
3102 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3103 revalidate_rule(cbdata->ofproto, sub);
3108 revalidate_rule(struct ofproto *p, struct rule *rule)
3110 const flow_t *flow = &rule->cr.flow;
3112 COVERAGE_INC(ofproto_revalidate_rule);
3115 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3117 rule_remove(p, rule);
3119 } else if (super != rule->super) {
3120 COVERAGE_INC(ofproto_revalidate_moved);
3121 list_remove(&rule->list);
3122 list_push_back(&super->list, &rule->list);
3123 rule->super = super;
3124 rule->hard_timeout = super->hard_timeout;
3125 rule->idle_timeout = super->idle_timeout;
3126 rule->created = super->created;
3131 rule_update_actions(p, rule);
3135 static struct ofpbuf *
3136 compose_flow_exp(const struct rule *rule, long long int now, uint8_t reason)
3138 struct ofp_flow_expired *ofe;
3141 ofe = make_openflow(sizeof *ofe, OFPT_FLOW_EXPIRED, &buf);
3142 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofe->match);
3143 ofe->priority = htons(rule->cr.priority);
3144 ofe->reason = reason;
3145 ofe->duration = htonl((now - rule->created) / 1000);
3146 ofe->packet_count = htonll(rule->packet_count);
3147 ofe->byte_count = htonll(rule->byte_count);
3153 send_flow_exp(struct ofproto *p, struct rule *rule,
3154 long long int now, uint8_t reason)
3156 struct ofconn *ofconn;
3157 struct ofconn *prev;
3158 struct ofpbuf *buf = NULL;
3160 /* We limit the maximum number of queued flow expirations it by accounting
3161 * them under the counter for replies. That works because preventing
3162 * OpenFlow requests from being processed also prevents new flows from
3163 * being added (and expiring). (It also prevents processing OpenFlow
3164 * requests that would not add new flows, so it is imperfect.) */
3167 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3168 if (ofconn->send_flow_exp && rconn_is_connected(ofconn->rconn)) {
3170 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3172 buf = compose_flow_exp(rule, now, reason);
3178 queue_tx(buf, prev, prev->reply_counter);
3183 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3185 assert(rule->installed);
3186 assert(!rule->cr.wc.wildcards);
3189 rule_remove(ofproto, rule);
3191 rule_uninstall(ofproto, rule);
3196 expire_rule(struct cls_rule *cls_rule, void *p_)
3198 struct ofproto *p = p_;
3199 struct rule *rule = rule_from_cls_rule(cls_rule);
3200 long long int hard_expire, idle_expire, expire, now;
3202 hard_expire = (rule->hard_timeout
3203 ? rule->created + rule->hard_timeout * 1000
3205 idle_expire = (rule->idle_timeout
3206 && (rule->super || list_is_empty(&rule->list))
3207 ? rule->used + rule->idle_timeout * 1000
3209 expire = MIN(hard_expire, idle_expire);
3213 if (rule->installed && now >= rule->used + 5000) {
3214 uninstall_idle_flow(p, rule);
3215 } else if (!rule->cr.wc.wildcards) {
3216 active_timeout(p, rule);
3222 COVERAGE_INC(ofproto_expired);
3224 /* Update stats. This code will be a no-op if the rule expired
3225 * due to an idle timeout. */
3226 if (rule->cr.wc.wildcards) {
3227 struct rule *subrule, *next;
3228 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3229 rule_remove(p, subrule);
3232 rule_uninstall(p, rule);
3235 if (!rule_is_hidden(rule)) {
3236 send_flow_exp(p, rule, now,
3238 ? OFPER_HARD_TIMEOUT : OFPER_IDLE_TIMEOUT));
3240 rule_remove(p, rule);
3244 active_timeout(struct ofproto *ofproto, struct rule *rule)
3246 if (ofproto->netflow && !is_controller_rule(rule) &&
3247 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
3248 struct ofexpired expired;
3249 struct odp_flow odp_flow;
3251 /* Get updated flow stats. */
3252 memset(&odp_flow, 0, sizeof odp_flow);
3253 if (rule->installed) {
3254 odp_flow.key = rule->cr.flow;
3255 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
3256 dpif_flow_get(ofproto->dpif, &odp_flow);
3258 if (odp_flow.stats.n_packets) {
3259 update_time(ofproto, rule, &odp_flow.stats);
3260 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
3261 odp_flow.stats.tcp_flags);
3265 expired.flow = rule->cr.flow;
3266 expired.packet_count = rule->packet_count +
3267 odp_flow.stats.n_packets;
3268 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
3269 expired.used = rule->used;
3271 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
3273 /* Schedule us to send the accumulated records once we have
3274 * collected all of them. */
3275 poll_immediate_wake();
3280 update_used(struct ofproto *p)
3282 struct odp_flow *flows;
3287 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
3292 for (i = 0; i < n_flows; i++) {
3293 struct odp_flow *f = &flows[i];
3296 rule = rule_from_cls_rule(
3297 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3298 if (!rule || !rule->installed) {
3299 COVERAGE_INC(ofproto_unexpected_rule);
3300 dpif_flow_del(p->dpif, f);
3304 update_time(p, rule, &f->stats);
3305 rule_account(p, rule, f->stats.n_bytes);
3311 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3312 const struct ofpbuf *packet, int send_len)
3314 struct odp_msg *msg = packet->data;
3315 struct ofpbuf payload;
3319 /* Extract packet payload from 'msg'. */
3320 payload.data = msg + 1;
3321 payload.size = msg->length - sizeof *msg;
3323 /* Construct ofp_packet_in message. */
3324 reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3325 opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port), reason,
3326 &payload, send_len);
3329 rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100);
3333 send_packet_in_action(struct ofpbuf *packet, void *p_)
3335 struct ofproto *p = p_;
3336 struct ofconn *ofconn;
3337 struct odp_msg *msg;
3340 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3341 if (ofconn == p->controller || ofconn->miss_send_len) {
3342 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3345 ofpbuf_delete(packet);
3349 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3351 struct ofproto *p = p_;
3352 bool in_fail_open = p->fail_open && fail_open_is_active(p->fail_open);
3353 struct ofconn *ofconn;
3354 struct ofpbuf payload;
3355 struct odp_msg *msg;
3358 payload.data = msg + 1;
3359 payload.size = msg->length - sizeof *msg;
3360 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3361 if (ofconn->miss_send_len) {
3362 struct pktbuf *pb = ofconn->pktbuf;
3363 uint32_t buffer_id = (in_fail_open
3365 : pktbuf_save(pb, &payload, msg->port));
3366 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3368 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3371 ofpbuf_delete(packet);
3375 pick_datapath_id(const struct ofproto *ofproto)
3377 const struct ofport *port;
3379 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
3381 uint8_t ea[ETH_ADDR_LEN];
3384 error = netdev_get_etheraddr(port->netdev, ea);
3386 return eth_addr_to_uint64(ea);
3388 VLOG_WARN("could not get MAC address for %s (%s)",
3389 netdev_get_name(port->netdev), strerror(error));
3391 return ofproto->fallback_dpid;
3395 pick_fallback_dpid(void)
3397 uint8_t ea[ETH_ADDR_LEN];
3398 eth_addr_random(ea);
3399 ea[0] = 0x00; /* Set Nicira OUI. */
3402 return eth_addr_to_uint64(ea);
3406 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3407 struct odp_actions *actions, tag_type *tags,
3408 uint16_t *nf_output_iface, void *ofproto_)
3410 struct ofproto *ofproto = ofproto_;
3413 /* Drop frames for reserved multicast addresses. */
3414 if (eth_addr_is_reserved(flow->dl_dst)) {
3418 /* Learn source MAC (but don't try to learn from revalidation). */
3419 if (packet != NULL) {
3420 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3423 /* The log messages here could actually be useful in debugging,
3424 * so keep the rate limit relatively high. */
3425 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3426 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3427 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3428 ofproto_revalidate(ofproto, rev_tag);
3432 /* Determine output port. */
3433 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3435 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
3436 } else if (out_port != flow->in_port) {
3437 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3438 *nf_output_iface = out_port;
3446 static const struct ofhooks default_ofhooks = {
3448 default_normal_ofhook_cb,