2 * Copyright (c) 2009 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"
31 #include "fail-open.h"
33 #include "mac-learning.h"
37 #include "ofp-print.h"
39 #include "openflow/nicira-ext.h"
40 #include "openflow/openflow.h"
41 #include "openflow/openflow-mgmt.h"
42 #include "openvswitch/datapath-protocol.h"
46 #include "poll-loop.h"
47 #include "port-array.h"
57 #include "vconn-ssl.h"
60 #define THIS_MODULE VLM_ofproto
70 TABLEID_CLASSIFIER = 1
74 struct netdev *netdev;
75 struct ofp_phy_port opp; /* In host byte order. */
78 static void ofport_free(struct ofport *);
79 static void hton_ofp_phy_port(struct ofp_phy_port *);
81 static int xlate_actions(const union ofp_action *in, size_t n_in,
82 const flow_t *flow, struct ofproto *ofproto,
83 const struct ofpbuf *packet,
84 struct odp_actions *out, tag_type *tags,
85 bool *may_set_up_flow, uint16_t *nf_output_iface);
90 uint16_t idle_timeout; /* In seconds from time of last use. */
91 uint16_t hard_timeout; /* In seconds from time of creation. */
92 long long int used; /* Last-used time (0 if never used). */
93 long long int created; /* Creation time. */
94 uint64_t packet_count; /* Number of packets received. */
95 uint64_t byte_count; /* Number of bytes received. */
96 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
97 tag_type tags; /* Tags (set only by hooks). */
98 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
100 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
101 * exact-match rule (having cr.wc.wildcards of 0) generated from the
102 * wildcard rule 'super'. In this case, 'list' is an element of the
105 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
106 * a list of subrules. A super-rule with no wildcards (where
107 * cr.wc.wildcards is 0) will never have any subrules. */
113 * A subrule has no actions (it uses the super-rule's actions). */
115 union ofp_action *actions;
119 * A super-rule with wildcard fields never has ODP actions (since the
120 * datapath only supports exact-match flows). */
121 bool installed; /* Installed in datapath? */
122 bool may_install; /* True ordinarily; false if actions must
123 * be reassessed for every packet. */
125 union odp_action *odp_actions;
129 rule_is_hidden(const struct rule *rule)
131 /* Subrules are merely an implementation detail, so hide them from the
133 if (rule->super != NULL) {
137 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
138 * (e.g. by in-band control) and are intentionally hidden from the
140 if (rule->cr.priority > UINT16_MAX) {
147 static struct rule *rule_create(struct ofproto *, struct rule *super,
148 const union ofp_action *, size_t n_actions,
149 uint16_t idle_timeout, uint16_t hard_timeout);
150 static void rule_free(struct rule *);
151 static void rule_destroy(struct ofproto *, struct rule *);
152 static struct rule *rule_from_cls_rule(const struct cls_rule *);
153 static void rule_insert(struct ofproto *, struct rule *,
154 struct ofpbuf *packet, uint16_t in_port);
155 static void rule_remove(struct ofproto *, struct rule *);
156 static bool rule_make_actions(struct ofproto *, struct rule *,
157 const struct ofpbuf *packet);
158 static void rule_install(struct ofproto *, struct rule *,
159 struct rule *displaced_rule);
160 static void rule_uninstall(struct ofproto *, struct rule *);
161 static void rule_post_uninstall(struct ofproto *, struct rule *);
166 struct pktbuf *pktbuf;
170 struct rconn_packet_counter *packet_in_counter;
172 /* Number of OpenFlow messages queued as replies to OpenFlow requests, and
173 * the maximum number before we stop reading OpenFlow requests. */
174 #define OFCONN_REPLY_MAX 100
175 struct rconn_packet_counter *reply_counter;
178 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *);
179 static void ofconn_destroy(struct ofconn *, struct ofproto *);
180 static void ofconn_run(struct ofconn *, struct ofproto *);
181 static void ofconn_wait(struct ofconn *);
182 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
183 struct rconn_packet_counter *counter);
187 uint64_t datapath_id; /* Datapath ID. */
188 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
189 uint64_t mgmt_id; /* Management channel identifier. */
190 char *manufacturer; /* Manufacturer. */
191 char *hardware; /* Hardware. */
192 char *software; /* Software version. */
193 char *serial; /* Serial number. */
197 struct netdev_monitor *netdev_monitor;
198 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
200 struct shash port_by_name;
204 struct switch_status *switch_status;
205 struct status_category *ss_cat;
206 struct in_band *in_band;
207 struct discovery *discovery;
208 struct fail_open *fail_open;
209 struct pinsched *miss_sched, *action_sched;
210 struct executer *executer;
211 struct netflow *netflow;
214 struct classifier cls;
215 bool need_revalidate;
216 long long int next_expiration;
217 struct tag_set revalidate_set;
219 /* OpenFlow connections. */
220 struct list all_conns;
221 struct ofconn *controller;
222 struct pvconn **listeners;
224 struct pvconn **snoops;
227 /* Hooks for ovs-vswitchd. */
228 const struct ofhooks *ofhooks;
231 /* Used by default ofhooks. */
232 struct mac_learning *ml;
235 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
237 static const struct ofhooks default_ofhooks;
239 static uint64_t pick_datapath_id(const struct ofproto *);
240 static uint64_t pick_fallback_dpid(void);
241 static void send_packet_in_miss(struct ofpbuf *, void *ofproto);
242 static void send_packet_in_action(struct ofpbuf *, void *ofproto);
243 static void update_used(struct ofproto *);
244 static void update_stats(struct ofproto *, struct rule *,
245 const struct odp_flow_stats *);
246 static void expire_rule(struct cls_rule *, void *ofproto);
247 static void active_timeout(struct ofproto *ofproto, struct rule *rule);
248 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
249 static void revalidate_cb(struct cls_rule *rule_, void *p_);
251 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
253 static void handle_openflow(struct ofconn *, struct ofproto *,
256 static void refresh_port_group(struct ofproto *, unsigned int group);
257 static void update_port(struct ofproto *, const char *devname);
258 static int init_ports(struct ofproto *);
259 static void reinit_ports(struct ofproto *);
262 ofproto_create(const char *datapath, const struct ofhooks *ofhooks, void *aux,
263 struct ofproto **ofprotop)
265 struct odp_stats stats;
272 /* Connect to datapath and start listening for messages. */
273 error = dpif_open(datapath, &dpif);
275 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
278 error = dpif_get_dp_stats(dpif, &stats);
280 VLOG_ERR("failed to obtain stats for datapath %s: %s",
281 datapath, strerror(error));
285 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION);
287 VLOG_ERR("failed to listen on datapath %s: %s",
288 datapath, strerror(error));
292 dpif_flow_flush(dpif);
293 dpif_recv_purge(dpif);
295 /* Initialize settings. */
296 p = xzalloc(sizeof *p);
297 p->fallback_dpid = pick_fallback_dpid();
298 p->datapath_id = p->fallback_dpid;
299 p->manufacturer = xstrdup("Nicira Networks, Inc.");
300 p->hardware = xstrdup("Reference Implementation");
301 p->software = xstrdup(VERSION BUILDNR);
302 p->serial = xstrdup("None");
304 /* Initialize datapath. */
306 p->netdev_monitor = netdev_monitor_create();
307 port_array_init(&p->ports);
308 shash_init(&p->port_by_name);
309 p->max_ports = stats.max_ports;
311 /* Initialize submodules. */
312 p->switch_status = switch_status_create(p);
316 p->miss_sched = p->action_sched = NULL;
320 /* Initialize flow table. */
321 classifier_init(&p->cls);
322 p->need_revalidate = false;
323 p->next_expiration = time_msec() + 1000;
324 tag_set_init(&p->revalidate_set);
326 /* Initialize OpenFlow connections. */
327 list_init(&p->all_conns);
328 p->controller = ofconn_create(p, rconn_create(5, 8));
329 p->controller->pktbuf = pktbuf_create();
330 p->controller->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
336 /* Initialize hooks. */
338 p->ofhooks = ofhooks;
342 p->ofhooks = &default_ofhooks;
344 p->ml = mac_learning_create();
347 /* Register switch status category. */
348 p->ss_cat = switch_status_register(p->switch_status, "remote",
349 rconn_status_cb, p->controller->rconn);
352 error = init_ports(p);
358 /* Pick final datapath ID. */
359 p->datapath_id = pick_datapath_id(p);
360 VLOG_INFO("using datapath ID %012"PRIx64, p->datapath_id);
367 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
369 uint64_t old_dpid = p->datapath_id;
370 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
371 if (p->datapath_id != old_dpid) {
372 VLOG_INFO("datapath ID changed to %012"PRIx64, p->datapath_id);
373 rconn_reconnect(p->controller->rconn);
378 ofproto_set_mgmt_id(struct ofproto *p, uint64_t mgmt_id)
380 p->mgmt_id = mgmt_id;
384 ofproto_set_probe_interval(struct ofproto *p, int probe_interval)
386 probe_interval = probe_interval ? MAX(probe_interval, 5) : 0;
387 rconn_set_probe_interval(p->controller->rconn, probe_interval);
389 int trigger_duration = probe_interval ? probe_interval * 3 : 15;
390 fail_open_set_trigger_duration(p->fail_open, trigger_duration);
395 ofproto_set_max_backoff(struct ofproto *p, int max_backoff)
397 rconn_set_max_backoff(p->controller->rconn, max_backoff);
401 ofproto_set_desc(struct ofproto *p,
402 const char *manufacturer, const char *hardware,
403 const char *software, const char *serial)
406 free(p->manufacturer);
407 p->manufacturer = xstrdup(manufacturer);
411 p->hardware = xstrdup(hardware);
415 p->software = xstrdup(software);
419 p->serial = xstrdup(serial);
424 ofproto_set_in_band(struct ofproto *p, bool in_band)
426 if (in_band != (p->in_band != NULL)) {
428 return in_band_create(p, p->dpif, p->switch_status,
429 p->controller->rconn, &p->in_band);
431 ofproto_set_discovery(p, false, NULL, true);
432 in_band_destroy(p->in_band);
435 rconn_reconnect(p->controller->rconn);
441 ofproto_set_discovery(struct ofproto *p, bool discovery,
442 const char *re, bool update_resolv_conf)
444 if (discovery != (p->discovery != NULL)) {
446 int error = ofproto_set_in_band(p, true);
450 error = discovery_create(re, update_resolv_conf,
451 p->dpif, p->switch_status,
457 discovery_destroy(p->discovery);
460 rconn_disconnect(p->controller->rconn);
461 } else if (discovery) {
462 discovery_set_update_resolv_conf(p->discovery, update_resolv_conf);
463 return discovery_set_accept_controller_re(p->discovery, re);
469 ofproto_set_controller(struct ofproto *ofproto, const char *controller)
471 if (ofproto->discovery) {
473 } else if (controller) {
474 if (strcmp(rconn_get_name(ofproto->controller->rconn), controller)) {
475 return rconn_connect(ofproto->controller->rconn, controller);
480 rconn_disconnect(ofproto->controller->rconn);
486 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
487 const struct svec *svec)
489 struct pvconn **pvconns = *pvconnsp;
490 size_t n_pvconns = *n_pvconnsp;
494 for (i = 0; i < n_pvconns; i++) {
495 pvconn_close(pvconns[i]);
499 pvconns = xmalloc(svec->n * sizeof *pvconns);
501 for (i = 0; i < svec->n; i++) {
502 const char *name = svec->names[i];
503 struct pvconn *pvconn;
506 error = pvconn_open(name, &pvconn);
508 pvconns[n_pvconns++] = pvconn;
510 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
518 *n_pvconnsp = n_pvconns;
524 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
526 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
530 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
532 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
536 ofproto_set_netflow(struct ofproto *ofproto,
537 const struct netflow_options *nf_options)
539 if (nf_options->collectors.n) {
540 if (!ofproto->netflow) {
541 ofproto->netflow = netflow_create();
543 return netflow_set_options(ofproto->netflow, nf_options);
545 netflow_destroy(ofproto->netflow);
546 ofproto->netflow = NULL;
552 ofproto_set_failure(struct ofproto *ofproto, bool fail_open)
555 struct rconn *rconn = ofproto->controller->rconn;
556 int trigger_duration = rconn_get_probe_interval(rconn) * 3;
557 if (!ofproto->fail_open) {
558 ofproto->fail_open = fail_open_create(ofproto, trigger_duration,
559 ofproto->switch_status,
562 fail_open_set_trigger_duration(ofproto->fail_open,
566 fail_open_destroy(ofproto->fail_open);
567 ofproto->fail_open = NULL;
572 ofproto_set_rate_limit(struct ofproto *ofproto,
573 int rate_limit, int burst_limit)
575 if (rate_limit > 0) {
576 if (!ofproto->miss_sched) {
577 ofproto->miss_sched = pinsched_create(rate_limit, burst_limit,
578 ofproto->switch_status);
579 ofproto->action_sched = pinsched_create(rate_limit, burst_limit,
582 pinsched_set_limits(ofproto->miss_sched, rate_limit, burst_limit);
583 pinsched_set_limits(ofproto->action_sched,
584 rate_limit, burst_limit);
587 pinsched_destroy(ofproto->miss_sched);
588 ofproto->miss_sched = NULL;
589 pinsched_destroy(ofproto->action_sched);
590 ofproto->action_sched = NULL;
595 ofproto_set_stp(struct ofproto *ofproto UNUSED, bool enable_stp)
599 VLOG_WARN("STP is not yet implemented");
607 ofproto_set_remote_execution(struct ofproto *ofproto, const char *command_acl,
608 const char *command_dir)
611 if (!ofproto->executer) {
612 return executer_create(command_acl, command_dir,
615 executer_set_acl(ofproto->executer, command_acl, command_dir);
618 executer_destroy(ofproto->executer);
619 ofproto->executer = NULL;
625 ofproto_get_datapath_id(const struct ofproto *ofproto)
627 return ofproto->datapath_id;
631 ofproto_get_mgmt_id(const struct ofproto *ofproto)
633 return ofproto->mgmt_id;
637 ofproto_get_probe_interval(const struct ofproto *ofproto)
639 return rconn_get_probe_interval(ofproto->controller->rconn);
643 ofproto_get_max_backoff(const struct ofproto *ofproto)
645 return rconn_get_max_backoff(ofproto->controller->rconn);
649 ofproto_get_in_band(const struct ofproto *ofproto)
651 return ofproto->in_band != NULL;
655 ofproto_get_discovery(const struct ofproto *ofproto)
657 return ofproto->discovery != NULL;
661 ofproto_get_controller(const struct ofproto *ofproto)
663 return rconn_get_name(ofproto->controller->rconn);
667 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
671 for (i = 0; i < ofproto->n_listeners; i++) {
672 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
677 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
681 for (i = 0; i < ofproto->n_snoops; i++) {
682 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
687 ofproto_destroy(struct ofproto *p)
689 struct ofconn *ofconn, *next_ofconn;
690 struct ofport *ofport;
691 unsigned int port_no;
698 ofproto_flush_flows(p);
699 classifier_destroy(&p->cls);
701 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
703 ofconn_destroy(ofconn, p);
707 netdev_monitor_destroy(p->netdev_monitor);
708 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
711 shash_destroy(&p->port_by_name);
713 switch_status_destroy(p->switch_status);
714 in_band_destroy(p->in_band);
715 discovery_destroy(p->discovery);
716 fail_open_destroy(p->fail_open);
717 pinsched_destroy(p->miss_sched);
718 pinsched_destroy(p->action_sched);
719 executer_destroy(p->executer);
720 netflow_destroy(p->netflow);
722 switch_status_unregister(p->ss_cat);
724 for (i = 0; i < p->n_listeners; i++) {
725 pvconn_close(p->listeners[i]);
729 for (i = 0; i < p->n_snoops; i++) {
730 pvconn_close(p->snoops[i]);
734 mac_learning_destroy(p->ml);
740 ofproto_run(struct ofproto *p)
742 int error = ofproto_run1(p);
744 error = ofproto_run2(p, false);
750 process_port_change(struct ofproto *ofproto, int error, char *devname)
752 if (error == ENOBUFS) {
753 reinit_ports(ofproto);
755 update_port(ofproto, devname);
761 ofproto_run1(struct ofproto *p)
763 struct ofconn *ofconn, *next_ofconn;
768 for (i = 0; i < 50; i++) {
772 error = dpif_recv(p->dpif, &buf);
774 if (error == ENODEV) {
775 /* Someone destroyed the datapath behind our back. The caller
776 * better destroy us and give up, because we're just going to
777 * spin from here on out. */
778 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
779 VLOG_ERR_RL(&rl, "%s: datapath was destroyed externally",
786 handle_odp_msg(p, buf);
789 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
790 process_port_change(p, error, devname);
792 while ((error = netdev_monitor_poll(p->netdev_monitor,
793 &devname)) != EAGAIN) {
794 process_port_change(p, error, devname);
798 in_band_run(p->in_band);
801 char *controller_name;
802 if (rconn_is_connectivity_questionable(p->controller->rconn)) {
803 discovery_question_connectivity(p->discovery);
805 if (discovery_run(p->discovery, &controller_name)) {
806 if (controller_name) {
807 rconn_connect(p->controller->rconn, controller_name);
809 rconn_disconnect(p->controller->rconn);
813 pinsched_run(p->miss_sched, send_packet_in_miss, p);
814 pinsched_run(p->action_sched, send_packet_in_action, p);
816 executer_run(p->executer);
819 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
821 ofconn_run(ofconn, p);
824 /* Fail-open maintenance. Do this after processing the ofconns since
825 * fail-open checks the status of the controller rconn. */
827 fail_open_run(p->fail_open);
830 for (i = 0; i < p->n_listeners; i++) {
834 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
836 ofconn_create(p, rconn_new_from_vconn("passive", vconn));
837 } else if (retval != EAGAIN) {
838 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
842 for (i = 0; i < p->n_snoops; i++) {
846 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
848 rconn_add_monitor(p->controller->rconn, vconn);
849 } else if (retval != EAGAIN) {
850 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
854 if (time_msec() >= p->next_expiration) {
855 COVERAGE_INC(ofproto_expiration);
856 p->next_expiration = time_msec() + 1000;
859 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
861 /* Let the hook know that we're at a stable point: all outstanding data
862 * in existing flows has been accounted to the account_cb. Thus, the
863 * hook can now reasonably do operations that depend on having accurate
864 * flow volume accounting (currently, that's just bond rebalancing). */
865 if (p->ofhooks->account_checkpoint_cb) {
866 p->ofhooks->account_checkpoint_cb(p->aux);
871 netflow_run(p->netflow);
877 struct revalidate_cbdata {
878 struct ofproto *ofproto;
879 bool revalidate_all; /* Revalidate all exact-match rules? */
880 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
881 struct tag_set revalidate_set; /* Set of tags to revalidate. */
885 ofproto_run2(struct ofproto *p, bool revalidate_all)
887 if (p->need_revalidate || revalidate_all
888 || !tag_set_is_empty(&p->revalidate_set)) {
889 struct revalidate_cbdata cbdata;
891 cbdata.revalidate_all = revalidate_all;
892 cbdata.revalidate_subrules = p->need_revalidate;
893 cbdata.revalidate_set = p->revalidate_set;
894 tag_set_init(&p->revalidate_set);
895 COVERAGE_INC(ofproto_revalidate);
896 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
897 p->need_revalidate = false;
904 ofproto_wait(struct ofproto *p)
906 struct ofconn *ofconn;
909 dpif_recv_wait(p->dpif);
910 dpif_port_poll_wait(p->dpif);
911 netdev_monitor_poll_wait(p->netdev_monitor);
912 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
916 in_band_wait(p->in_band);
919 discovery_wait(p->discovery);
922 fail_open_wait(p->fail_open);
924 pinsched_wait(p->miss_sched);
925 pinsched_wait(p->action_sched);
927 executer_wait(p->executer);
929 if (!tag_set_is_empty(&p->revalidate_set)) {
930 poll_immediate_wake();
932 if (p->need_revalidate) {
933 /* Shouldn't happen, but if it does just go around again. */
934 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
935 poll_immediate_wake();
936 } else if (p->next_expiration != LLONG_MAX) {
937 poll_timer_wait(p->next_expiration - time_msec());
939 for (i = 0; i < p->n_listeners; i++) {
940 pvconn_wait(p->listeners[i]);
942 for (i = 0; i < p->n_snoops; i++) {
943 pvconn_wait(p->snoops[i]);
948 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
950 tag_set_add(&ofproto->revalidate_set, tag);
954 ofproto_get_revalidate_set(struct ofproto *ofproto)
956 return &ofproto->revalidate_set;
960 ofproto_is_alive(const struct ofproto *p)
962 return p->discovery || rconn_is_alive(p->controller->rconn);
966 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
967 const union ofp_action *actions, size_t n_actions,
968 const struct ofpbuf *packet)
970 struct odp_actions odp_actions;
973 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
979 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
981 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
982 odp_actions.n_actions, packet);
987 ofproto_add_flow(struct ofproto *p,
988 const flow_t *flow, uint32_t wildcards, unsigned int priority,
989 const union ofp_action *actions, size_t n_actions,
993 rule = rule_create(p, NULL, actions, n_actions,
994 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */, 0);
995 cls_rule_from_flow(&rule->cr, flow, wildcards, priority);
996 rule_insert(p, rule, NULL, 0);
1000 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1001 uint32_t wildcards, unsigned int priority)
1005 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1009 rule_remove(ofproto, rule);
1014 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1016 struct rule *rule = rule_from_cls_rule(rule_);
1017 struct ofproto *ofproto = ofproto_;
1019 /* Mark the flow as not installed, even though it might really be
1020 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1021 * There is no point in uninstalling it individually since we are about to
1022 * blow away all the flows with dpif_flow_flush(). */
1023 rule->installed = false;
1025 rule_remove(ofproto, rule);
1029 ofproto_flush_flows(struct ofproto *ofproto)
1031 COVERAGE_INC(ofproto_flush);
1032 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1033 dpif_flow_flush(ofproto->dpif);
1034 if (ofproto->in_band) {
1035 in_band_flushed(ofproto->in_band);
1037 if (ofproto->fail_open) {
1038 fail_open_flushed(ofproto->fail_open);
1043 reinit_ports(struct ofproto *p)
1045 struct svec devnames;
1046 struct ofport *ofport;
1047 unsigned int port_no;
1048 struct odp_port *odp_ports;
1052 svec_init(&devnames);
1053 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1054 svec_add (&devnames, (char *) ofport->opp.name);
1056 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1057 for (i = 0; i < n_odp_ports; i++) {
1058 svec_add (&devnames, odp_ports[i].devname);
1062 svec_sort_unique(&devnames);
1063 for (i = 0; i < devnames.n; i++) {
1064 update_port(p, devnames.names[i]);
1066 svec_destroy(&devnames);
1070 refresh_port_group(struct ofproto *p, unsigned int group)
1074 struct ofport *port;
1075 unsigned int port_no;
1077 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1079 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1081 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1082 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1083 ports[n_ports++] = port_no;
1086 dpif_port_group_set(p->dpif, group, ports, n_ports);
1091 refresh_port_groups(struct ofproto *p)
1093 refresh_port_group(p, DP_GROUP_FLOOD);
1094 refresh_port_group(p, DP_GROUP_ALL);
1097 static struct ofport *
1098 make_ofport(const struct odp_port *odp_port)
1100 enum netdev_flags flags;
1101 struct ofport *ofport;
1102 struct netdev *netdev;
1106 error = netdev_open(odp_port->devname, NETDEV_ETH_TYPE_NONE, &netdev);
1108 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1109 "cannot be opened (%s)",
1110 odp_port->devname, odp_port->port,
1111 odp_port->devname, strerror(error));
1115 ofport = xmalloc(sizeof *ofport);
1116 ofport->netdev = netdev;
1117 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1118 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1119 memcpy(ofport->opp.name, odp_port->devname,
1120 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1121 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1123 netdev_get_flags(netdev, &flags);
1124 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1126 netdev_get_carrier(netdev, &carrier);
1127 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1129 netdev_get_features(netdev,
1130 &ofport->opp.curr, &ofport->opp.advertised,
1131 &ofport->opp.supported, &ofport->opp.peer);
1136 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1138 if (port_array_get(&p->ports, odp_port->port)) {
1139 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1142 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1143 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1152 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1154 const struct ofp_phy_port *a = &a_->opp;
1155 const struct ofp_phy_port *b = &b_->opp;
1157 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1158 return (a->port_no == b->port_no
1159 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1160 && !strcmp((char *) a->name, (char *) b->name)
1161 && a->state == b->state
1162 && a->config == b->config
1163 && a->curr == b->curr
1164 && a->advertised == b->advertised
1165 && a->supported == b->supported
1166 && a->peer == b->peer);
1170 send_port_status(struct ofproto *p, const struct ofport *ofport,
1173 /* XXX Should limit the number of queued port status change messages. */
1174 struct ofconn *ofconn;
1175 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1176 struct ofp_port_status *ops;
1179 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1180 ops->reason = reason;
1181 ops->desc = ofport->opp;
1182 hton_ofp_phy_port(&ops->desc);
1183 queue_tx(b, ofconn, NULL);
1185 if (p->ofhooks->port_changed_cb) {
1186 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1191 ofport_install(struct ofproto *p, struct ofport *ofport)
1193 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1194 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no),
1196 shash_add(&p->port_by_name, (char *) ofport->opp.name, ofport);
1200 ofport_remove(struct ofproto *p, struct ofport *ofport)
1202 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1203 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no), NULL);
1204 shash_delete(&p->port_by_name,
1205 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1209 ofport_free(struct ofport *ofport)
1212 netdev_close(ofport->netdev);
1218 update_port(struct ofproto *p, const char *devname)
1220 struct odp_port odp_port;
1221 struct ofport *old_ofport;
1222 struct ofport *new_ofport;
1225 COVERAGE_INC(ofproto_update_port);
1227 /* Query the datapath for port information. */
1228 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1230 /* Find the old ofport. */
1231 old_ofport = shash_find_data(&p->port_by_name, devname);
1234 /* There's no port named 'devname' but there might be a port with
1235 * the same port number. This could happen if a port is deleted
1236 * and then a new one added in its place very quickly, or if a port
1237 * is renamed. In the former case we want to send an OFPPR_DELETE
1238 * and an OFPPR_ADD, and in the latter case we want to send a
1239 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1240 * the old port's ifindex against the new port, or perhaps less
1241 * reliably but more portably by comparing the old port's MAC
1242 * against the new port's MAC. However, this code isn't that smart
1243 * and always sends an OFPPR_MODIFY (XXX). */
1244 old_ofport = port_array_get(&p->ports, odp_port.port);
1246 } else if (error != ENOENT && error != ENODEV) {
1247 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1248 "%s", strerror(error));
1252 /* Create a new ofport. */
1253 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1255 /* Eliminate a few pathological cases. */
1256 if (!old_ofport && !new_ofport) {
1258 } else if (old_ofport && new_ofport) {
1259 /* Most of the 'config' bits are OpenFlow soft state, but
1260 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1261 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1262 * leaves the other bits 0.) */
1263 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1265 if (ofport_equal(old_ofport, new_ofport)) {
1266 /* False alarm--no change. */
1267 ofport_free(new_ofport);
1272 /* Now deal with the normal cases. */
1274 ofport_remove(p, old_ofport);
1277 ofport_install(p, new_ofport);
1279 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1280 (!old_ofport ? OFPPR_ADD
1281 : !new_ofport ? OFPPR_DELETE
1283 ofport_free(old_ofport);
1285 /* Update port groups. */
1286 refresh_port_groups(p);
1290 init_ports(struct ofproto *p)
1292 struct odp_port *ports;
1297 error = dpif_port_list(p->dpif, &ports, &n_ports);
1302 for (i = 0; i < n_ports; i++) {
1303 const struct odp_port *odp_port = &ports[i];
1304 if (!ofport_conflicts(p, odp_port)) {
1305 struct ofport *ofport = make_ofport(odp_port);
1307 ofport_install(p, ofport);
1312 refresh_port_groups(p);
1316 static struct ofconn *
1317 ofconn_create(struct ofproto *p, struct rconn *rconn)
1319 struct ofconn *ofconn = xmalloc(sizeof *ofconn);
1320 list_push_back(&p->all_conns, &ofconn->node);
1321 ofconn->rconn = rconn;
1322 ofconn->pktbuf = NULL;
1323 ofconn->send_flow_exp = false;
1324 ofconn->miss_send_len = 0;
1325 ofconn->packet_in_counter = rconn_packet_counter_create ();
1326 ofconn->reply_counter = rconn_packet_counter_create ();
1331 ofconn_destroy(struct ofconn *ofconn, struct ofproto *p)
1334 executer_rconn_closing(p->executer, ofconn->rconn);
1337 list_remove(&ofconn->node);
1338 rconn_destroy(ofconn->rconn);
1339 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1340 rconn_packet_counter_destroy(ofconn->reply_counter);
1341 pktbuf_destroy(ofconn->pktbuf);
1346 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1350 rconn_run(ofconn->rconn);
1352 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1353 /* Limit the number of iterations to prevent other tasks from
1355 for (iteration = 0; iteration < 50; iteration++) {
1356 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1361 fail_open_maybe_recover(p->fail_open);
1363 handle_openflow(ofconn, p, of_msg);
1364 ofpbuf_delete(of_msg);
1368 if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
1369 ofconn_destroy(ofconn, p);
1374 ofconn_wait(struct ofconn *ofconn)
1376 rconn_run_wait(ofconn->rconn);
1377 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1378 rconn_recv_wait(ofconn->rconn);
1380 COVERAGE_INC(ofproto_ofconn_stuck);
1384 /* Caller is responsible for initializing the 'cr' member of the returned
1386 static struct rule *
1387 rule_create(struct ofproto *ofproto, struct rule *super,
1388 const union ofp_action *actions, size_t n_actions,
1389 uint16_t idle_timeout, uint16_t hard_timeout)
1391 struct rule *rule = xzalloc(sizeof *rule);
1392 rule->idle_timeout = idle_timeout;
1393 rule->hard_timeout = hard_timeout;
1394 rule->used = rule->created = time_msec();
1395 rule->super = super;
1397 list_push_back(&super->list, &rule->list);
1399 list_init(&rule->list);
1401 rule->n_actions = n_actions;
1402 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1403 netflow_flow_clear(&rule->nf_flow);
1404 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1409 static struct rule *
1410 rule_from_cls_rule(const struct cls_rule *cls_rule)
1412 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1416 rule_free(struct rule *rule)
1418 free(rule->actions);
1419 free(rule->odp_actions);
1423 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1424 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1425 * through all of its subrules and revalidates them, destroying any that no
1426 * longer has a super-rule (which is probably all of them).
1428 * Before calling this function, the caller must make have removed 'rule' from
1429 * the classifier. If 'rule' is an exact-match rule, the caller is also
1430 * responsible for ensuring that it has been uninstalled from the datapath. */
1432 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1435 struct rule *subrule, *next;
1436 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1437 revalidate_rule(ofproto, subrule);
1440 list_remove(&rule->list);
1446 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1448 const union ofp_action *oa;
1449 struct actions_iterator i;
1451 if (out_port == htons(OFPP_NONE)) {
1454 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1455 oa = actions_next(&i)) {
1456 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1463 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1464 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1466 * The flow that 'packet' actually contains does not need to actually match
1467 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1468 * the packet and byte counters for 'rule' will be credited for the packet sent
1469 * out whether or not the packet actually matches 'rule'.
1471 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1472 * the caller must already have accurately composed ODP actions for it given
1473 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1474 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1475 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1476 * actions and apply them to 'packet'. */
1478 rule_execute(struct ofproto *ofproto, struct rule *rule,
1479 struct ofpbuf *packet, const flow_t *flow)
1481 const union odp_action *actions;
1483 struct odp_actions a;
1485 /* Grab or compose the ODP actions.
1487 * The special case for an exact-match 'rule' where 'flow' is not the
1488 * rule's flow is important to avoid, e.g., sending a packet out its input
1489 * port simply because the ODP actions were composed for the wrong
1491 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1492 struct rule *super = rule->super ? rule->super : rule;
1493 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1494 packet, &a, NULL, 0, NULL)) {
1497 actions = a.actions;
1498 n_actions = a.n_actions;
1500 actions = rule->odp_actions;
1501 n_actions = rule->n_odp_actions;
1504 /* Execute the ODP actions. */
1505 if (!dpif_execute(ofproto->dpif, flow->in_port,
1506 actions, n_actions, packet)) {
1507 struct odp_flow_stats stats;
1508 flow_extract_stats(flow, packet, &stats);
1509 update_stats(ofproto, rule, &stats);
1510 rule->used = time_msec();
1511 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1516 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1519 struct rule *displaced_rule;
1521 /* Insert the rule in the classifier. */
1522 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1523 if (!rule->cr.wc.wildcards) {
1524 rule_make_actions(p, rule, packet);
1527 /* Send the packet and credit it to the rule. */
1530 flow_extract(packet, in_port, &flow);
1531 rule_execute(p, rule, packet, &flow);
1534 /* Install the rule in the datapath only after sending the packet, to
1535 * avoid packet reordering. */
1536 if (rule->cr.wc.wildcards) {
1537 COVERAGE_INC(ofproto_add_wc_flow);
1538 p->need_revalidate = true;
1540 rule_install(p, rule, displaced_rule);
1543 /* Free the rule that was displaced, if any. */
1544 if (displaced_rule) {
1545 rule_destroy(p, displaced_rule);
1549 static struct rule *
1550 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1553 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1554 rule->idle_timeout, rule->hard_timeout);
1555 COVERAGE_INC(ofproto_subrule_create);
1556 cls_rule_from_flow(&subrule->cr, flow, 0,
1557 (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1558 : rule->cr.priority));
1559 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1565 rule_remove(struct ofproto *ofproto, struct rule *rule)
1567 if (rule->cr.wc.wildcards) {
1568 COVERAGE_INC(ofproto_del_wc_flow);
1569 ofproto->need_revalidate = true;
1571 rule_uninstall(ofproto, rule);
1573 classifier_remove(&ofproto->cls, &rule->cr);
1574 rule_destroy(ofproto, rule);
1577 /* Returns true if the actions changed, false otherwise. */
1579 rule_make_actions(struct ofproto *p, struct rule *rule,
1580 const struct ofpbuf *packet)
1582 const struct rule *super;
1583 struct odp_actions a;
1586 assert(!rule->cr.wc.wildcards);
1588 super = rule->super ? rule->super : rule;
1590 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1591 packet, &a, &rule->tags, &rule->may_install,
1592 &rule->nf_flow.output_iface);
1594 actions_len = a.n_actions * sizeof *a.actions;
1595 if (rule->n_odp_actions != a.n_actions
1596 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1597 COVERAGE_INC(ofproto_odp_unchanged);
1598 free(rule->odp_actions);
1599 rule->n_odp_actions = a.n_actions;
1600 rule->odp_actions = xmemdup(a.actions, actions_len);
1608 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1609 struct odp_flow_put *put)
1611 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1612 put->flow.key = rule->cr.flow;
1613 put->flow.actions = rule->odp_actions;
1614 put->flow.n_actions = rule->n_odp_actions;
1616 return dpif_flow_put(ofproto->dpif, put);
1620 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1622 assert(!rule->cr.wc.wildcards);
1624 if (rule->may_install) {
1625 struct odp_flow_put put;
1626 if (!do_put_flow(p, rule,
1627 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1629 rule->installed = true;
1630 if (displaced_rule) {
1631 update_stats(p, displaced_rule, &put.flow.stats);
1632 rule_post_uninstall(p, displaced_rule);
1635 } else if (displaced_rule) {
1636 rule_uninstall(p, displaced_rule);
1641 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1643 if (rule->installed) {
1644 struct odp_flow_put put;
1645 COVERAGE_INC(ofproto_dp_missed);
1646 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1648 rule_install(ofproto, rule, NULL);
1653 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1655 bool actions_changed;
1656 uint16_t new_out_iface, old_out_iface;
1658 old_out_iface = rule->nf_flow.output_iface;
1659 actions_changed = rule_make_actions(ofproto, rule, NULL);
1661 if (rule->may_install) {
1662 if (rule->installed) {
1663 if (actions_changed) {
1664 struct odp_flow_put put;
1665 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
1666 | ODPPF_ZERO_STATS, &put);
1667 update_stats(ofproto, rule, &put.flow.stats);
1669 /* Temporarily set the old output iface so that NetFlow
1670 * messages have the correct output interface for the old
1672 new_out_iface = rule->nf_flow.output_iface;
1673 rule->nf_flow.output_iface = old_out_iface;
1674 rule_post_uninstall(ofproto, rule);
1675 rule->nf_flow.output_iface = new_out_iface;
1678 rule_install(ofproto, rule, NULL);
1681 rule_uninstall(ofproto, rule);
1686 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1688 uint64_t total_bytes = rule->byte_count + extra_bytes;
1690 if (ofproto->ofhooks->account_flow_cb
1691 && total_bytes > rule->accounted_bytes)
1693 ofproto->ofhooks->account_flow_cb(
1694 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1695 total_bytes - rule->accounted_bytes, ofproto->aux);
1696 rule->accounted_bytes = total_bytes;
1701 rule_uninstall(struct ofproto *p, struct rule *rule)
1703 assert(!rule->cr.wc.wildcards);
1704 if (rule->installed) {
1705 struct odp_flow odp_flow;
1707 odp_flow.key = rule->cr.flow;
1708 odp_flow.actions = NULL;
1709 odp_flow.n_actions = 0;
1710 if (!dpif_flow_del(p->dpif, &odp_flow)) {
1711 update_stats(p, rule, &odp_flow.stats);
1713 rule->installed = false;
1715 rule_post_uninstall(p, rule);
1720 is_controller_rule(struct rule *rule)
1722 /* If the only action is send to the controller then don't report
1723 * NetFlow expiration messages since it is just part of the control
1724 * logic for the network and not real traffic. */
1726 if (rule && rule->super) {
1727 struct rule *super = rule->super;
1729 return super->n_actions == 1 &&
1730 super->actions[0].type == htons(OFPAT_OUTPUT) &&
1731 super->actions[0].output.port == htons(OFPP_CONTROLLER);
1738 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1740 struct rule *super = rule->super;
1742 rule_account(ofproto, rule, 0);
1744 if (ofproto->netflow && !is_controller_rule(rule)) {
1745 struct ofexpired expired;
1746 expired.flow = rule->cr.flow;
1747 expired.packet_count = rule->packet_count;
1748 expired.byte_count = rule->byte_count;
1749 expired.used = rule->used;
1750 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
1753 super->packet_count += rule->packet_count;
1754 super->byte_count += rule->byte_count;
1756 /* Reset counters to prevent double counting if the rule ever gets
1758 rule->packet_count = 0;
1759 rule->byte_count = 0;
1760 rule->accounted_bytes = 0;
1762 netflow_flow_clear(&rule->nf_flow);
1767 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1768 struct rconn_packet_counter *counter)
1770 update_openflow_length(msg);
1771 if (rconn_send(ofconn->rconn, msg, counter)) {
1777 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1778 int error, const void *data, size_t len)
1781 struct ofp_error_msg *oem;
1783 if (!(error >> 16)) {
1784 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1789 COVERAGE_INC(ofproto_error);
1790 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1791 oh ? oh->xid : 0, &buf);
1792 oem->type = htons((unsigned int) error >> 16);
1793 oem->code = htons(error & 0xffff);
1794 memcpy(oem->data, data, len);
1795 queue_tx(buf, ofconn, ofconn->reply_counter);
1799 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1802 size_t oh_length = ntohs(oh->length);
1803 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1807 hton_ofp_phy_port(struct ofp_phy_port *opp)
1809 opp->port_no = htons(opp->port_no);
1810 opp->config = htonl(opp->config);
1811 opp->state = htonl(opp->state);
1812 opp->curr = htonl(opp->curr);
1813 opp->advertised = htonl(opp->advertised);
1814 opp->supported = htonl(opp->supported);
1815 opp->peer = htonl(opp->peer);
1819 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
1821 struct ofp_header *rq = oh;
1822 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
1827 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
1828 struct ofp_header *oh)
1830 struct ofp_switch_features *osf;
1832 unsigned int port_no;
1833 struct ofport *port;
1835 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1836 osf->datapath_id = htonll(p->datapath_id);
1837 osf->n_buffers = htonl(pktbuf_capacity());
1839 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1840 OFPC_PORT_STATS | OFPC_MULTI_PHY_TX);
1841 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1842 (1u << OFPAT_SET_VLAN_VID) |
1843 (1u << OFPAT_SET_VLAN_PCP) |
1844 (1u << OFPAT_STRIP_VLAN) |
1845 (1u << OFPAT_SET_DL_SRC) |
1846 (1u << OFPAT_SET_DL_DST) |
1847 (1u << OFPAT_SET_NW_SRC) |
1848 (1u << OFPAT_SET_NW_DST) |
1849 (1u << OFPAT_SET_TP_SRC) |
1850 (1u << OFPAT_SET_TP_DST));
1852 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1853 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1856 queue_tx(buf, ofconn, ofconn->reply_counter);
1861 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1862 struct ofp_header *oh)
1865 struct ofp_switch_config *osc;
1869 /* Figure out flags. */
1870 dpif_get_drop_frags(p->dpif, &drop_frags);
1871 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1872 if (ofconn->send_flow_exp) {
1873 flags |= OFPC_SEND_FLOW_EXP;
1877 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1878 osc->flags = htons(flags);
1879 osc->miss_send_len = htons(ofconn->miss_send_len);
1880 queue_tx(buf, ofconn, ofconn->reply_counter);
1886 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1887 struct ofp_switch_config *osc)
1892 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1896 flags = ntohs(osc->flags);
1898 ofconn->send_flow_exp = (flags & OFPC_SEND_FLOW_EXP) != 0;
1900 if (ofconn == p->controller) {
1901 switch (flags & OFPC_FRAG_MASK) {
1902 case OFPC_FRAG_NORMAL:
1903 dpif_set_drop_frags(p->dpif, false);
1905 case OFPC_FRAG_DROP:
1906 dpif_set_drop_frags(p->dpif, true);
1909 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1915 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1916 if (ntohs(osc->miss_send_len) != 0) {
1917 ofconn->pktbuf = pktbuf_create();
1919 pktbuf_destroy(ofconn->pktbuf);
1923 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1929 add_output_group_action(struct odp_actions *actions, uint16_t group,
1930 uint16_t *nf_output_iface)
1932 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1934 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
1935 *nf_output_iface = NF_OUT_FLOOD;
1940 add_controller_action(struct odp_actions *actions,
1941 const struct ofp_action_output *oao)
1943 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
1944 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
1947 struct action_xlate_ctx {
1949 const flow_t *flow; /* Flow to which these actions correspond. */
1950 int recurse; /* Recursion level, via xlate_table_action. */
1951 struct ofproto *ofproto;
1952 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
1953 * null pointer if we are revalidating
1954 * without a packet to refer to. */
1957 struct odp_actions *out; /* Datapath actions. */
1958 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
1959 bool may_set_up_flow; /* True ordinarily; false if the actions must
1960 * be reassessed for every packet. */
1961 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
1964 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
1965 struct action_xlate_ctx *ctx);
1968 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
1970 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
1973 if (ofport->opp.config & OFPPC_NO_FWD) {
1974 /* Forwarding disabled on port. */
1979 * We don't have an ofport record for this port, but it doesn't hurt to
1980 * allow forwarding to it anyhow. Maybe such a port will appear later
1981 * and we're pre-populating the flow table.
1985 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
1986 ctx->nf_output_iface = port;
1989 static struct rule *
1990 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
1993 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
1995 /* The rule we found might not be valid, since we could be in need of
1996 * revalidation. If it is not valid, don't return it. */
1999 && ofproto->need_revalidate
2000 && !revalidate_rule(ofproto, rule)) {
2001 COVERAGE_INC(ofproto_invalidated);
2009 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2011 if (!ctx->recurse) {
2016 flow.in_port = in_port;
2018 rule = lookup_valid_rule(ctx->ofproto, &flow);
2025 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2032 xlate_output_action(struct action_xlate_ctx *ctx,
2033 const struct ofp_action_output *oao)
2036 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2038 ctx->nf_output_iface = NF_OUT_DROP;
2040 switch (ntohs(oao->port)) {
2042 add_output_action(ctx, ctx->flow->in_port);
2045 xlate_table_action(ctx, ctx->flow->in_port);
2048 if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet,
2049 ctx->out, ctx->tags,
2050 &ctx->nf_output_iface,
2051 ctx->ofproto->aux)) {
2052 COVERAGE_INC(ofproto_uninstallable);
2053 ctx->may_set_up_flow = false;
2057 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2058 &ctx->nf_output_iface);
2061 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2063 case OFPP_CONTROLLER:
2064 add_controller_action(ctx->out, oao);
2067 add_output_action(ctx, ODPP_LOCAL);
2070 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2071 if (odp_port != ctx->flow->in_port) {
2072 add_output_action(ctx, odp_port);
2077 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2078 ctx->nf_output_iface = NF_OUT_FLOOD;
2079 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2080 ctx->nf_output_iface = prev_nf_output_iface;
2081 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2082 ctx->nf_output_iface != NF_OUT_FLOOD) {
2083 ctx->nf_output_iface = NF_OUT_MULTI;
2088 xlate_nicira_action(struct action_xlate_ctx *ctx,
2089 const struct nx_action_header *nah)
2091 const struct nx_action_resubmit *nar;
2092 int subtype = ntohs(nah->subtype);
2094 assert(nah->vendor == htonl(NX_VENDOR_ID));
2096 case NXAST_RESUBMIT:
2097 nar = (const struct nx_action_resubmit *) nah;
2098 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2102 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2108 do_xlate_actions(const union ofp_action *in, size_t n_in,
2109 struct action_xlate_ctx *ctx)
2111 struct actions_iterator iter;
2112 const union ofp_action *ia;
2113 const struct ofport *port;
2115 port = port_array_get(&ctx->ofproto->ports, ctx->flow->in_port);
2116 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2117 port->opp.config & (eth_addr_equals(ctx->flow->dl_dst, stp_eth_addr)
2118 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2119 /* Drop this flow. */
2123 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2124 uint16_t type = ntohs(ia->type);
2125 union odp_action *oa;
2129 xlate_output_action(ctx, &ia->output);
2132 case OFPAT_SET_VLAN_VID:
2133 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2134 oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2137 case OFPAT_SET_VLAN_PCP:
2138 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2139 oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2142 case OFPAT_STRIP_VLAN:
2143 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2146 case OFPAT_SET_DL_SRC:
2147 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2148 memcpy(oa->dl_addr.dl_addr,
2149 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2152 case OFPAT_SET_DL_DST:
2153 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2154 memcpy(oa->dl_addr.dl_addr,
2155 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2158 case OFPAT_SET_NW_SRC:
2159 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2160 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2163 case OFPAT_SET_NW_DST:
2164 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2165 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2168 case OFPAT_SET_TP_SRC:
2169 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2170 oa->tp_port.tp_port = ia->tp_port.tp_port;
2173 case OFPAT_SET_TP_DST:
2174 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2175 oa->tp_port.tp_port = ia->tp_port.tp_port;
2179 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2183 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2190 xlate_actions(const union ofp_action *in, size_t n_in,
2191 const flow_t *flow, struct ofproto *ofproto,
2192 const struct ofpbuf *packet,
2193 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2194 uint16_t *nf_output_iface)
2196 tag_type no_tags = 0;
2197 struct action_xlate_ctx ctx;
2198 COVERAGE_INC(ofproto_ofp2odp);
2199 odp_actions_init(out);
2202 ctx.ofproto = ofproto;
2203 ctx.packet = packet;
2205 ctx.tags = tags ? tags : &no_tags;
2206 ctx.may_set_up_flow = true;
2207 ctx.nf_output_iface = NF_OUT_DROP;
2208 do_xlate_actions(in, n_in, &ctx);
2210 /* Check with in-band control to see if we're allowed to set up this
2212 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2213 ctx.may_set_up_flow = false;
2216 if (may_set_up_flow) {
2217 *may_set_up_flow = ctx.may_set_up_flow;
2219 if (nf_output_iface) {
2220 *nf_output_iface = ctx.nf_output_iface;
2222 if (odp_actions_overflow(out)) {
2223 odp_actions_init(out);
2224 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2230 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2231 struct ofp_header *oh)
2233 struct ofp_packet_out *opo;
2234 struct ofpbuf payload, *buffer;
2235 struct odp_actions actions;
2241 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2245 opo = (struct ofp_packet_out *) oh;
2247 COVERAGE_INC(ofproto_packet_out);
2248 if (opo->buffer_id != htonl(UINT32_MAX)) {
2249 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2251 if (error || !buffer) {
2259 flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2260 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2261 &flow, p, &payload, &actions, NULL, NULL, NULL);
2266 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2268 ofpbuf_delete(buffer);
2274 update_port_config(struct ofproto *p, struct ofport *port,
2275 uint32_t config, uint32_t mask)
2277 mask &= config ^ port->opp.config;
2278 if (mask & OFPPC_PORT_DOWN) {
2279 if (config & OFPPC_PORT_DOWN) {
2280 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2282 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2285 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2286 if (mask & REVALIDATE_BITS) {
2287 COVERAGE_INC(ofproto_costly_flags);
2288 port->opp.config ^= mask & REVALIDATE_BITS;
2289 p->need_revalidate = true;
2291 #undef REVALIDATE_BITS
2292 if (mask & OFPPC_NO_FLOOD) {
2293 port->opp.config ^= OFPPC_NO_FLOOD;
2294 refresh_port_group(p, DP_GROUP_FLOOD);
2296 if (mask & OFPPC_NO_PACKET_IN) {
2297 port->opp.config ^= OFPPC_NO_PACKET_IN;
2302 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2304 const struct ofp_port_mod *opm;
2305 struct ofport *port;
2308 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2312 opm = (struct ofp_port_mod *) oh;
2314 port = port_array_get(&p->ports,
2315 ofp_port_to_odp_port(ntohs(opm->port_no)));
2317 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2318 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2319 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2321 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2322 if (opm->advertise) {
2323 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2329 static struct ofpbuf *
2330 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2332 struct ofp_stats_reply *osr;
2335 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2336 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2338 osr->flags = htons(0);
2342 static struct ofpbuf *
2343 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2345 return make_stats_reply(request->header.xid, request->type, body_len);
2349 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2351 struct ofpbuf *msg = *msgp;
2352 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2353 if (nbytes + msg->size > UINT16_MAX) {
2354 struct ofp_stats_reply *reply = msg->data;
2355 reply->flags = htons(OFPSF_REPLY_MORE);
2356 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2357 queue_tx(msg, ofconn, ofconn->reply_counter);
2359 return ofpbuf_put_uninit(*msgp, nbytes);
2363 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2364 struct ofp_stats_request *request)
2366 struct ofp_desc_stats *ods;
2369 msg = start_stats_reply(request, sizeof *ods);
2370 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2371 strncpy(ods->mfr_desc, p->manufacturer, sizeof ods->mfr_desc);
2372 strncpy(ods->hw_desc, p->hardware, sizeof ods->hw_desc);
2373 strncpy(ods->sw_desc, p->software, sizeof ods->sw_desc);
2374 strncpy(ods->serial_num, p->serial, sizeof ods->serial_num);
2375 queue_tx(msg, ofconn, ofconn->reply_counter);
2381 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2383 struct rule *rule = rule_from_cls_rule(cls_rule);
2384 int *n_subrules = n_subrules_;
2392 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2393 struct ofp_stats_request *request)
2395 struct ofp_table_stats *ots;
2397 struct odp_stats dpstats;
2398 int n_exact, n_subrules, n_wild;
2400 msg = start_stats_reply(request, sizeof *ots * 2);
2402 /* Count rules of various kinds. */
2404 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2405 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2406 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2409 dpif_get_dp_stats(p->dpif, &dpstats);
2410 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2411 memset(ots, 0, sizeof *ots);
2412 ots->table_id = TABLEID_HASH;
2413 strcpy(ots->name, "hash");
2414 ots->wildcards = htonl(0);
2415 ots->max_entries = htonl(dpstats.max_capacity);
2416 ots->active_count = htonl(n_exact);
2417 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2419 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2421 /* Classifier table. */
2422 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2423 memset(ots, 0, sizeof *ots);
2424 ots->table_id = TABLEID_CLASSIFIER;
2425 strcpy(ots->name, "classifier");
2426 ots->wildcards = htonl(OFPFW_ALL);
2427 ots->max_entries = htonl(65536);
2428 ots->active_count = htonl(n_wild);
2429 ots->lookup_count = htonll(0); /* XXX */
2430 ots->matched_count = htonll(0); /* XXX */
2432 queue_tx(msg, ofconn, ofconn->reply_counter);
2437 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2438 struct ofp_stats_request *request)
2440 struct ofp_port_stats *ops;
2442 struct ofport *port;
2443 unsigned int port_no;
2445 msg = start_stats_reply(request, sizeof *ops * 16);
2446 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2447 struct netdev_stats stats;
2449 /* Intentionally ignore return value, since errors will set 'stats' to
2450 * all-1s, which is correct for OpenFlow, and netdev_get_stats() will
2452 netdev_get_stats(port->netdev, &stats);
2454 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2455 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2456 memset(ops->pad, 0, sizeof ops->pad);
2457 ops->rx_packets = htonll(stats.rx_packets);
2458 ops->tx_packets = htonll(stats.tx_packets);
2459 ops->rx_bytes = htonll(stats.rx_bytes);
2460 ops->tx_bytes = htonll(stats.tx_bytes);
2461 ops->rx_dropped = htonll(stats.rx_dropped);
2462 ops->tx_dropped = htonll(stats.tx_dropped);
2463 ops->rx_errors = htonll(stats.rx_errors);
2464 ops->tx_errors = htonll(stats.tx_errors);
2465 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2466 ops->rx_over_err = htonll(stats.rx_over_errors);
2467 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2468 ops->collisions = htonll(stats.collisions);
2471 queue_tx(msg, ofconn, ofconn->reply_counter);
2475 struct flow_stats_cbdata {
2476 struct ofproto *ofproto;
2477 struct ofconn *ofconn;
2483 query_stats(struct ofproto *p, struct rule *rule,
2484 uint64_t *packet_countp, uint64_t *byte_countp)
2486 uint64_t packet_count, byte_count;
2487 struct rule *subrule;
2488 struct odp_flow *odp_flows;
2491 packet_count = rule->packet_count;
2492 byte_count = rule->byte_count;
2494 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2495 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
2496 if (rule->cr.wc.wildcards) {
2498 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2499 odp_flows[i++].key = subrule->cr.flow;
2500 packet_count += subrule->packet_count;
2501 byte_count += subrule->byte_count;
2504 odp_flows[0].key = rule->cr.flow;
2507 packet_count = rule->packet_count;
2508 byte_count = rule->byte_count;
2509 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2511 for (i = 0; i < n_odp_flows; i++) {
2512 struct odp_flow *odp_flow = &odp_flows[i];
2513 packet_count += odp_flow->stats.n_packets;
2514 byte_count += odp_flow->stats.n_bytes;
2519 *packet_countp = packet_count;
2520 *byte_countp = byte_count;
2524 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2526 struct rule *rule = rule_from_cls_rule(rule_);
2527 struct flow_stats_cbdata *cbdata = cbdata_;
2528 struct ofp_flow_stats *ofs;
2529 uint64_t packet_count, byte_count;
2530 size_t act_len, len;
2532 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2536 act_len = sizeof *rule->actions * rule->n_actions;
2537 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2539 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2541 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2542 ofs->length = htons(len);
2543 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2545 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match);
2546 ofs->duration = htonl((time_msec() - rule->created) / 1000);
2547 ofs->priority = htons(rule->cr.priority);
2548 ofs->idle_timeout = htons(rule->idle_timeout);
2549 ofs->hard_timeout = htons(rule->hard_timeout);
2550 memset(ofs->pad2, 0, sizeof ofs->pad2);
2551 ofs->packet_count = htonll(packet_count);
2552 ofs->byte_count = htonll(byte_count);
2553 memcpy(ofs->actions, rule->actions, act_len);
2557 table_id_to_include(uint8_t table_id)
2559 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2560 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2561 : table_id == 0xff ? CLS_INC_ALL
2566 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2567 const struct ofp_stats_request *osr,
2570 struct ofp_flow_stats_request *fsr;
2571 struct flow_stats_cbdata cbdata;
2572 struct cls_rule target;
2574 if (arg_size != sizeof *fsr) {
2575 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2577 fsr = (struct ofp_flow_stats_request *) osr->body;
2579 COVERAGE_INC(ofproto_flows_req);
2581 cbdata.ofconn = ofconn;
2582 cbdata.out_port = fsr->out_port;
2583 cbdata.msg = start_stats_reply(osr, 1024);
2584 cls_rule_from_match(&target, &fsr->match, 0);
2585 classifier_for_each_match(&p->cls, &target,
2586 table_id_to_include(fsr->table_id),
2587 flow_stats_cb, &cbdata);
2588 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2592 struct flow_stats_ds_cbdata {
2593 struct ofproto *ofproto;
2598 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
2600 struct rule *rule = rule_from_cls_rule(rule_);
2601 struct flow_stats_ds_cbdata *cbdata = cbdata_;
2602 struct ds *results = cbdata->results;
2603 struct ofp_match match;
2604 uint64_t packet_count, byte_count;
2605 size_t act_len = sizeof *rule->actions * rule->n_actions;
2607 /* Don't report on subrules. */
2608 if (rule->super != NULL) {
2612 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2613 flow_to_ovs_match(&rule->cr.flow, rule->cr.wc.wildcards, &match);
2615 ds_put_format(results, "duration=%llds, ",
2616 (time_msec() - rule->created) / 1000);
2617 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2618 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2619 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2620 ofp_print_match(results, &match, true);
2621 ofp_print_actions(results, &rule->actions->header, act_len);
2622 ds_put_cstr(results, "\n");
2625 /* Adds a pretty-printed description of all flows to 'results', including
2626 * those marked hidden by secchan (e.g., by in-band control). */
2628 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2630 struct ofp_match match;
2631 struct cls_rule target;
2632 struct flow_stats_ds_cbdata cbdata;
2634 memset(&match, 0, sizeof match);
2635 match.wildcards = htonl(OFPFW_ALL);
2638 cbdata.results = results;
2640 cls_rule_from_match(&target, &match, 0);
2641 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2642 flow_stats_ds_cb, &cbdata);
2645 struct aggregate_stats_cbdata {
2646 struct ofproto *ofproto;
2648 uint64_t packet_count;
2649 uint64_t byte_count;
2654 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2656 struct rule *rule = rule_from_cls_rule(rule_);
2657 struct aggregate_stats_cbdata *cbdata = cbdata_;
2658 uint64_t packet_count, byte_count;
2660 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2664 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2666 cbdata->packet_count += packet_count;
2667 cbdata->byte_count += byte_count;
2672 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2673 const struct ofp_stats_request *osr,
2676 struct ofp_aggregate_stats_request *asr;
2677 struct ofp_aggregate_stats_reply *reply;
2678 struct aggregate_stats_cbdata cbdata;
2679 struct cls_rule target;
2682 if (arg_size != sizeof *asr) {
2683 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2685 asr = (struct ofp_aggregate_stats_request *) osr->body;
2687 COVERAGE_INC(ofproto_agg_request);
2689 cbdata.out_port = asr->out_port;
2690 cbdata.packet_count = 0;
2691 cbdata.byte_count = 0;
2693 cls_rule_from_match(&target, &asr->match, 0);
2694 classifier_for_each_match(&p->cls, &target,
2695 table_id_to_include(asr->table_id),
2696 aggregate_stats_cb, &cbdata);
2698 msg = start_stats_reply(osr, sizeof *reply);
2699 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2700 reply->flow_count = htonl(cbdata.n_flows);
2701 reply->packet_count = htonll(cbdata.packet_count);
2702 reply->byte_count = htonll(cbdata.byte_count);
2703 queue_tx(msg, ofconn, ofconn->reply_counter);
2708 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2709 struct ofp_header *oh)
2711 struct ofp_stats_request *osr;
2715 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2720 osr = (struct ofp_stats_request *) oh;
2722 switch (ntohs(osr->type)) {
2724 return handle_desc_stats_request(p, ofconn, osr);
2727 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2729 case OFPST_AGGREGATE:
2730 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2733 return handle_table_stats_request(p, ofconn, osr);
2736 return handle_port_stats_request(p, ofconn, osr);
2739 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2742 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2746 static long long int
2747 msec_from_nsec(uint64_t sec, uint32_t nsec)
2749 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2753 update_time(struct ofproto *ofproto, struct rule *rule,
2754 const struct odp_flow_stats *stats)
2756 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2757 if (used > rule->used) {
2759 if (rule->super && used > rule->super->used) {
2760 rule->super->used = used;
2762 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
2767 update_stats(struct ofproto *ofproto, struct rule *rule,
2768 const struct odp_flow_stats *stats)
2770 if (stats->n_packets) {
2771 update_time(ofproto, rule, stats);
2772 rule->packet_count += stats->n_packets;
2773 rule->byte_count += stats->n_bytes;
2774 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
2780 add_flow(struct ofproto *p, struct ofconn *ofconn,
2781 struct ofp_flow_mod *ofm, size_t n_actions)
2783 struct ofpbuf *packet;
2788 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
2789 n_actions, ntohs(ofm->idle_timeout),
2790 ntohs(ofm->hard_timeout));
2791 cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority));
2795 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2796 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2800 rule_insert(p, rule, packet, in_port);
2801 ofpbuf_delete(packet);
2806 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
2807 size_t n_actions, uint16_t command, struct rule *rule)
2809 if (rule_is_hidden(rule)) {
2813 if (command == OFPFC_DELETE) {
2814 rule_remove(p, rule);
2816 size_t actions_len = n_actions * sizeof *rule->actions;
2818 if (n_actions == rule->n_actions
2819 && !memcmp(ofm->actions, rule->actions, actions_len))
2824 free(rule->actions);
2825 rule->actions = xmemdup(ofm->actions, actions_len);
2826 rule->n_actions = n_actions;
2828 if (rule->cr.wc.wildcards) {
2829 COVERAGE_INC(ofproto_mod_wc_flow);
2830 p->need_revalidate = true;
2832 rule_update_actions(p, rule);
2840 modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm,
2841 size_t n_actions, uint16_t command)
2847 flow_from_match(&flow, &wildcards, &ofm->match);
2848 rule = rule_from_cls_rule(classifier_find_rule_exactly(
2849 &p->cls, &flow, wildcards,
2850 ntohs(ofm->priority)));
2853 if (command == OFPFC_DELETE
2854 && ofm->out_port != htons(OFPP_NONE)
2855 && !rule_has_out_port(rule, ofm->out_port)) {
2859 modify_flow(p, ofm, n_actions, command, rule);
2864 struct modify_flows_cbdata {
2865 struct ofproto *ofproto;
2866 const struct ofp_flow_mod *ofm;
2873 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
2875 struct rule *rule = rule_from_cls_rule(rule_);
2876 struct modify_flows_cbdata *cbdata = cbdata_;
2878 if (cbdata->out_port != htons(OFPP_NONE)
2879 && !rule_has_out_port(rule, cbdata->out_port)) {
2883 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions,
2884 cbdata->command, rule);
2888 modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm,
2889 size_t n_actions, uint16_t command)
2891 struct modify_flows_cbdata cbdata;
2892 struct cls_rule target;
2896 cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port
2897 : htons(OFPP_NONE));
2898 cbdata.n_actions = n_actions;
2899 cbdata.command = command;
2901 cls_rule_from_match(&target, &ofm->match, 0);
2903 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2904 modify_flows_cb, &cbdata);
2909 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
2910 struct ofp_flow_mod *ofm)
2915 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
2916 sizeof *ofm->actions, &n_actions);
2921 normalize_match(&ofm->match);
2922 if (!ofm->match.wildcards) {
2923 ofm->priority = htons(UINT16_MAX);
2926 error = validate_actions((const union ofp_action *) ofm->actions,
2927 n_actions, p->max_ports);
2932 switch (ntohs(ofm->command)) {
2934 return add_flow(p, ofconn, ofm, n_actions);
2937 return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY);
2939 case OFPFC_MODIFY_STRICT:
2940 return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY);
2943 return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE);
2945 case OFPFC_DELETE_STRICT:
2946 return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE);
2949 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
2954 send_capability_reply(struct ofproto *p, struct ofconn *ofconn, uint32_t xid)
2956 struct ofmp_capability_reply *ocr;
2958 char capabilities[] = "com.nicira.mgmt.manager=false\n";
2960 ocr = make_openflow_xid(sizeof(*ocr), OFPT_VENDOR, xid, &b);
2961 ocr->header.header.vendor = htonl(NX_VENDOR_ID);
2962 ocr->header.header.subtype = htonl(NXT_MGMT);
2963 ocr->header.type = htons(OFMPT_CAPABILITY_REPLY);
2965 ocr->format = htonl(OFMPCOF_SIMPLE);
2966 ocr->mgmt_id = htonll(p->mgmt_id);
2968 ofpbuf_put(b, capabilities, strlen(capabilities));
2970 queue_tx(b, ofconn, ofconn->reply_counter);
2974 handle_ofmp(struct ofproto *p, struct ofconn *ofconn,
2975 struct ofmp_header *ofmph)
2977 size_t msg_len = ntohs(ofmph->header.header.length);
2978 if (msg_len < sizeof(*ofmph)) {
2979 VLOG_WARN_RL(&rl, "dropping short managment message: %zu\n", msg_len);
2980 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2983 if (ofmph->type == htons(OFMPT_CAPABILITY_REQUEST)) {
2984 struct ofmp_capability_request *ofmpcr;
2986 if (msg_len < sizeof(struct ofmp_capability_request)) {
2987 VLOG_WARN_RL(&rl, "dropping short capability request: %zu\n",
2989 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2992 ofmpcr = (struct ofmp_capability_request *)ofmph;
2993 if (ofmpcr->format != htonl(OFMPCAF_SIMPLE)) {
2994 /* xxx Find a better type than bad subtype */
2995 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2998 send_capability_reply(p, ofconn, ofmph->header.header.xid);
3001 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3006 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3008 struct ofp_vendor_header *ovh = msg;
3009 struct nicira_header *nh;
3011 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3012 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
3014 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3015 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3017 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3018 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
3022 switch (ntohl(nh->subtype)) {
3023 case NXT_STATUS_REQUEST:
3024 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3027 case NXT_ACT_SET_CONFIG:
3028 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
3030 case NXT_ACT_GET_CONFIG:
3031 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
3033 case NXT_COMMAND_REQUEST:
3035 return executer_handle_request(p->executer, ofconn->rconn, msg);
3040 return handle_ofmp(p, ofconn, msg);
3043 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3047 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3048 struct ofpbuf *ofp_msg)
3050 struct ofp_header *oh = ofp_msg->data;
3053 COVERAGE_INC(ofproto_recv_openflow);
3055 case OFPT_ECHO_REQUEST:
3056 error = handle_echo_request(ofconn, oh);
3059 case OFPT_ECHO_REPLY:
3063 case OFPT_FEATURES_REQUEST:
3064 error = handle_features_request(p, ofconn, oh);
3067 case OFPT_GET_CONFIG_REQUEST:
3068 error = handle_get_config_request(p, ofconn, oh);
3071 case OFPT_SET_CONFIG:
3072 error = handle_set_config(p, ofconn, ofp_msg->data);
3075 case OFPT_PACKET_OUT:
3076 error = handle_packet_out(p, ofconn, ofp_msg->data);
3080 error = handle_port_mod(p, oh);
3084 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3087 case OFPT_STATS_REQUEST:
3088 error = handle_stats_request(p, ofconn, oh);
3092 error = handle_vendor(p, ofconn, ofp_msg->data);
3096 if (VLOG_IS_WARN_ENABLED()) {
3097 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3098 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3101 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3106 send_error_oh(ofconn, ofp_msg->data, error);
3111 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3113 struct odp_msg *msg = packet->data;
3114 uint16_t in_port = odp_port_to_ofp_port(msg->port);
3116 struct ofpbuf payload;
3119 /* Handle controller actions. */
3120 if (msg->type == _ODPL_ACTION_NR) {
3121 COVERAGE_INC(ofproto_ctlr_action);
3122 pinsched_send(p->action_sched, in_port, packet,
3123 send_packet_in_action, p);
3127 payload.data = msg + 1;
3128 payload.size = msg->length - sizeof *msg;
3129 flow_extract(&payload, msg->port, &flow);
3131 /* Check with in-band control to see if this packet should be sent
3132 * to the local port regardless of the flow table. */
3133 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3134 union odp_action action;
3136 memset(&action, 0, sizeof(action));
3137 action.output.type = ODPAT_OUTPUT;
3138 action.output.port = ODPP_LOCAL;
3139 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3142 rule = lookup_valid_rule(p, &flow);
3144 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3145 struct ofport *port = port_array_get(&p->ports, msg->port);
3147 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3148 COVERAGE_INC(ofproto_no_packet_in);
3149 /* XXX install 'drop' flow entry */
3150 ofpbuf_delete(packet);
3154 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3157 COVERAGE_INC(ofproto_packet_in);
3158 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3162 if (rule->cr.wc.wildcards) {
3163 rule = rule_create_subrule(p, rule, &flow);
3164 rule_make_actions(p, rule, packet);
3166 if (!rule->may_install) {
3167 /* The rule is not installable, that is, we need to process every
3168 * packet, so process the current packet and set its actions into
3170 rule_make_actions(p, rule, packet);
3172 /* XXX revalidate rule if it needs it */
3176 rule_execute(p, rule, &payload, &flow);
3177 rule_reinstall(p, rule);
3179 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY
3180 && rconn_is_connected(p->controller->rconn)) {
3182 * Extra-special case for fail-open mode.
3184 * We are in fail-open mode and the packet matched the fail-open rule,
3185 * but we are connected to a controller too. We should send the packet
3186 * up to the controller in the hope that it will try to set up a flow
3187 * and thereby allow us to exit fail-open.
3189 * See the top-level comment in fail-open.c for more information.
3191 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3193 ofpbuf_delete(packet);
3198 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3200 struct rule *sub = rule_from_cls_rule(sub_);
3201 struct revalidate_cbdata *cbdata = cbdata_;
3203 if (cbdata->revalidate_all
3204 || (cbdata->revalidate_subrules && sub->super)
3205 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3206 revalidate_rule(cbdata->ofproto, sub);
3211 revalidate_rule(struct ofproto *p, struct rule *rule)
3213 const flow_t *flow = &rule->cr.flow;
3215 COVERAGE_INC(ofproto_revalidate_rule);
3218 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3220 rule_remove(p, rule);
3222 } else if (super != rule->super) {
3223 COVERAGE_INC(ofproto_revalidate_moved);
3224 list_remove(&rule->list);
3225 list_push_back(&super->list, &rule->list);
3226 rule->super = super;
3227 rule->hard_timeout = super->hard_timeout;
3228 rule->idle_timeout = super->idle_timeout;
3229 rule->created = super->created;
3234 rule_update_actions(p, rule);
3238 static struct ofpbuf *
3239 compose_flow_exp(const struct rule *rule, long long int now, uint8_t reason)
3241 struct ofp_flow_expired *ofe;
3244 ofe = make_openflow(sizeof *ofe, OFPT_FLOW_EXPIRED, &buf);
3245 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofe->match);
3246 ofe->priority = htons(rule->cr.priority);
3247 ofe->reason = reason;
3248 ofe->duration = htonl((now - rule->created) / 1000);
3249 ofe->packet_count = htonll(rule->packet_count);
3250 ofe->byte_count = htonll(rule->byte_count);
3256 send_flow_exp(struct ofproto *p, struct rule *rule,
3257 long long int now, uint8_t reason)
3259 struct ofconn *ofconn;
3260 struct ofconn *prev;
3261 struct ofpbuf *buf = NULL;
3263 /* We limit the maximum number of queued flow expirations it by accounting
3264 * them under the counter for replies. That works because preventing
3265 * OpenFlow requests from being processed also prevents new flows from
3266 * being added (and expiring). (It also prevents processing OpenFlow
3267 * requests that would not add new flows, so it is imperfect.) */
3270 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3271 if (ofconn->send_flow_exp && rconn_is_connected(ofconn->rconn)) {
3273 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3275 buf = compose_flow_exp(rule, now, reason);
3281 queue_tx(buf, prev, prev->reply_counter);
3286 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3288 assert(rule->installed);
3289 assert(!rule->cr.wc.wildcards);
3292 rule_remove(ofproto, rule);
3294 rule_uninstall(ofproto, rule);
3299 expire_rule(struct cls_rule *cls_rule, void *p_)
3301 struct ofproto *p = p_;
3302 struct rule *rule = rule_from_cls_rule(cls_rule);
3303 long long int hard_expire, idle_expire, expire, now;
3305 hard_expire = (rule->hard_timeout
3306 ? rule->created + rule->hard_timeout * 1000
3308 idle_expire = (rule->idle_timeout
3309 && (rule->super || list_is_empty(&rule->list))
3310 ? rule->used + rule->idle_timeout * 1000
3312 expire = MIN(hard_expire, idle_expire);
3316 if (rule->installed && now >= rule->used + 5000) {
3317 uninstall_idle_flow(p, rule);
3318 } else if (!rule->cr.wc.wildcards) {
3319 active_timeout(p, rule);
3325 COVERAGE_INC(ofproto_expired);
3327 /* Update stats. This code will be a no-op if the rule expired
3328 * due to an idle timeout. */
3329 if (rule->cr.wc.wildcards) {
3330 struct rule *subrule, *next;
3331 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3332 rule_remove(p, subrule);
3335 rule_uninstall(p, rule);
3338 if (!rule_is_hidden(rule)) {
3339 send_flow_exp(p, rule, now,
3341 ? OFPER_HARD_TIMEOUT : OFPER_IDLE_TIMEOUT));
3343 rule_remove(p, rule);
3347 active_timeout(struct ofproto *ofproto, struct rule *rule)
3349 if (ofproto->netflow && !is_controller_rule(rule) &&
3350 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
3351 struct ofexpired expired;
3352 struct odp_flow odp_flow;
3354 /* Get updated flow stats. */
3355 memset(&odp_flow, 0, sizeof odp_flow);
3356 if (rule->installed) {
3357 odp_flow.key = rule->cr.flow;
3358 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
3359 dpif_flow_get(ofproto->dpif, &odp_flow);
3361 if (odp_flow.stats.n_packets) {
3362 update_time(ofproto, rule, &odp_flow.stats);
3363 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
3364 odp_flow.stats.tcp_flags);
3368 expired.flow = rule->cr.flow;
3369 expired.packet_count = rule->packet_count +
3370 odp_flow.stats.n_packets;
3371 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
3372 expired.used = rule->used;
3374 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
3376 /* Schedule us to send the accumulated records once we have
3377 * collected all of them. */
3378 poll_immediate_wake();
3383 update_used(struct ofproto *p)
3385 struct odp_flow *flows;
3390 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
3395 for (i = 0; i < n_flows; i++) {
3396 struct odp_flow *f = &flows[i];
3399 rule = rule_from_cls_rule(
3400 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3401 if (!rule || !rule->installed) {
3402 COVERAGE_INC(ofproto_unexpected_rule);
3403 dpif_flow_del(p->dpif, f);
3407 update_time(p, rule, &f->stats);
3408 rule_account(p, rule, f->stats.n_bytes);
3414 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3415 const struct ofpbuf *packet, int send_len)
3417 struct odp_msg *msg = packet->data;
3418 struct ofpbuf payload;
3422 /* Extract packet payload from 'msg'. */
3423 payload.data = msg + 1;
3424 payload.size = msg->length - sizeof *msg;
3426 /* Construct ofp_packet_in message. */
3427 reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3428 opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port), reason,
3429 &payload, send_len);
3432 rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100);
3436 send_packet_in_action(struct ofpbuf *packet, void *p_)
3438 struct ofproto *p = p_;
3439 struct ofconn *ofconn;
3440 struct odp_msg *msg;
3443 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3444 if (ofconn == p->controller || ofconn->miss_send_len) {
3445 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3448 ofpbuf_delete(packet);
3452 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3454 struct ofproto *p = p_;
3455 bool in_fail_open = p->fail_open && fail_open_is_active(p->fail_open);
3456 struct ofconn *ofconn;
3457 struct ofpbuf payload;
3458 struct odp_msg *msg;
3461 payload.data = msg + 1;
3462 payload.size = msg->length - sizeof *msg;
3463 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3464 if (ofconn->miss_send_len) {
3465 struct pktbuf *pb = ofconn->pktbuf;
3466 uint32_t buffer_id = (in_fail_open
3468 : pktbuf_save(pb, &payload, msg->port));
3469 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3471 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3474 ofpbuf_delete(packet);
3478 pick_datapath_id(const struct ofproto *ofproto)
3480 const struct ofport *port;
3482 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
3484 uint8_t ea[ETH_ADDR_LEN];
3487 error = netdev_get_etheraddr(port->netdev, ea);
3489 return eth_addr_to_uint64(ea);
3491 VLOG_WARN("could not get MAC address for %s (%s)",
3492 netdev_get_name(port->netdev), strerror(error));
3494 return ofproto->fallback_dpid;
3498 pick_fallback_dpid(void)
3500 uint8_t ea[ETH_ADDR_LEN];
3501 eth_addr_random(ea);
3502 ea[0] = 0x00; /* Set Nicira OUI. */
3505 return eth_addr_to_uint64(ea);
3509 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3510 struct odp_actions *actions, tag_type *tags,
3511 uint16_t *nf_output_iface, void *ofproto_)
3513 struct ofproto *ofproto = ofproto_;
3516 /* Drop frames for reserved multicast addresses. */
3517 if (eth_addr_is_reserved(flow->dl_dst)) {
3521 /* Learn source MAC (but don't try to learn from revalidation). */
3522 if (packet != NULL) {
3523 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3526 /* The log messages here could actually be useful in debugging,
3527 * so keep the rate limit relatively high. */
3528 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3529 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3530 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3531 ofproto_revalidate(ofproto, rev_tag);
3535 /* Determine output port. */
3536 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3538 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
3539 } else if (out_port != flow->in_port) {
3540 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3541 *nf_output_iface = out_port;
3549 static const struct ofhooks default_ofhooks = {
3551 default_normal_ofhook_cb,