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 = xcalloc(1, 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 = xcalloc(1, 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_NW_TOS) |
1850 (1u << OFPAT_SET_TP_SRC) |
1851 (1u << OFPAT_SET_TP_DST));
1853 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1854 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1857 queue_tx(buf, ofconn, ofconn->reply_counter);
1862 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1863 struct ofp_header *oh)
1866 struct ofp_switch_config *osc;
1870 /* Figure out flags. */
1871 dpif_get_drop_frags(p->dpif, &drop_frags);
1872 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1873 if (ofconn->send_flow_exp) {
1874 flags |= OFPC_SEND_FLOW_EXP;
1878 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1879 osc->flags = htons(flags);
1880 osc->miss_send_len = htons(ofconn->miss_send_len);
1881 queue_tx(buf, ofconn, ofconn->reply_counter);
1887 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1888 struct ofp_switch_config *osc)
1893 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1897 flags = ntohs(osc->flags);
1899 ofconn->send_flow_exp = (flags & OFPC_SEND_FLOW_EXP) != 0;
1901 if (ofconn == p->controller) {
1902 switch (flags & OFPC_FRAG_MASK) {
1903 case OFPC_FRAG_NORMAL:
1904 dpif_set_drop_frags(p->dpif, false);
1906 case OFPC_FRAG_DROP:
1907 dpif_set_drop_frags(p->dpif, true);
1910 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1916 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1917 if (ntohs(osc->miss_send_len) != 0) {
1918 ofconn->pktbuf = pktbuf_create();
1920 pktbuf_destroy(ofconn->pktbuf);
1924 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1930 add_output_group_action(struct odp_actions *actions, uint16_t group,
1931 uint16_t *nf_output_iface)
1933 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1935 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
1936 *nf_output_iface = NF_OUT_FLOOD;
1941 add_controller_action(struct odp_actions *actions,
1942 const struct ofp_action_output *oao)
1944 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
1945 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
1948 struct action_xlate_ctx {
1950 const flow_t *flow; /* Flow to which these actions correspond. */
1951 int recurse; /* Recursion level, via xlate_table_action. */
1952 struct ofproto *ofproto;
1953 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
1954 * null pointer if we are revalidating
1955 * without a packet to refer to. */
1958 struct odp_actions *out; /* Datapath actions. */
1959 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
1960 bool may_set_up_flow; /* True ordinarily; false if the actions must
1961 * be reassessed for every packet. */
1962 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
1965 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
1966 struct action_xlate_ctx *ctx);
1969 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
1971 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
1974 if (ofport->opp.config & OFPPC_NO_FWD) {
1975 /* Forwarding disabled on port. */
1980 * We don't have an ofport record for this port, but it doesn't hurt to
1981 * allow forwarding to it anyhow. Maybe such a port will appear later
1982 * and we're pre-populating the flow table.
1986 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
1987 ctx->nf_output_iface = port;
1990 static struct rule *
1991 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
1994 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
1996 /* The rule we found might not be valid, since we could be in need of
1997 * revalidation. If it is not valid, don't return it. */
2000 && ofproto->need_revalidate
2001 && !revalidate_rule(ofproto, rule)) {
2002 COVERAGE_INC(ofproto_invalidated);
2010 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2012 if (!ctx->recurse) {
2017 flow.in_port = in_port;
2019 rule = lookup_valid_rule(ctx->ofproto, &flow);
2026 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2033 xlate_output_action(struct action_xlate_ctx *ctx,
2034 const struct ofp_action_output *oao)
2037 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2039 ctx->nf_output_iface = NF_OUT_DROP;
2041 switch (ntohs(oao->port)) {
2043 add_output_action(ctx, ctx->flow->in_port);
2046 xlate_table_action(ctx, ctx->flow->in_port);
2049 if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet,
2050 ctx->out, ctx->tags,
2051 &ctx->nf_output_iface,
2052 ctx->ofproto->aux)) {
2053 COVERAGE_INC(ofproto_uninstallable);
2054 ctx->may_set_up_flow = false;
2058 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2059 &ctx->nf_output_iface);
2062 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2064 case OFPP_CONTROLLER:
2065 add_controller_action(ctx->out, oao);
2068 add_output_action(ctx, ODPP_LOCAL);
2071 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2072 if (odp_port != ctx->flow->in_port) {
2073 add_output_action(ctx, odp_port);
2078 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2079 ctx->nf_output_iface = NF_OUT_FLOOD;
2080 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2081 ctx->nf_output_iface = prev_nf_output_iface;
2082 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2083 ctx->nf_output_iface != NF_OUT_FLOOD) {
2084 ctx->nf_output_iface = NF_OUT_MULTI;
2089 xlate_nicira_action(struct action_xlate_ctx *ctx,
2090 const struct nx_action_header *nah)
2092 const struct nx_action_resubmit *nar;
2093 int subtype = ntohs(nah->subtype);
2095 assert(nah->vendor == htonl(NX_VENDOR_ID));
2097 case NXAST_RESUBMIT:
2098 nar = (const struct nx_action_resubmit *) nah;
2099 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2103 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2109 do_xlate_actions(const union ofp_action *in, size_t n_in,
2110 struct action_xlate_ctx *ctx)
2112 struct actions_iterator iter;
2113 const union ofp_action *ia;
2114 const struct ofport *port;
2116 port = port_array_get(&ctx->ofproto->ports, ctx->flow->in_port);
2117 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2118 port->opp.config & (eth_addr_equals(ctx->flow->dl_dst, stp_eth_addr)
2119 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2120 /* Drop this flow. */
2124 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2125 uint16_t type = ntohs(ia->type);
2126 union odp_action *oa;
2130 xlate_output_action(ctx, &ia->output);
2133 case OFPAT_SET_VLAN_VID:
2134 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2135 oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2138 case OFPAT_SET_VLAN_PCP:
2139 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2140 oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2143 case OFPAT_STRIP_VLAN:
2144 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2147 case OFPAT_SET_DL_SRC:
2148 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2149 memcpy(oa->dl_addr.dl_addr,
2150 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2153 case OFPAT_SET_DL_DST:
2154 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2155 memcpy(oa->dl_addr.dl_addr,
2156 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2159 case OFPAT_SET_NW_SRC:
2160 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2161 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2164 case OFPAT_SET_NW_DST:
2165 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2166 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2168 case OFPAT_SET_NW_TOS:
2169 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2170 oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2173 case OFPAT_SET_TP_SRC:
2174 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2175 oa->tp_port.tp_port = ia->tp_port.tp_port;
2178 case OFPAT_SET_TP_DST:
2179 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2180 oa->tp_port.tp_port = ia->tp_port.tp_port;
2184 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2188 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2195 xlate_actions(const union ofp_action *in, size_t n_in,
2196 const flow_t *flow, struct ofproto *ofproto,
2197 const struct ofpbuf *packet,
2198 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2199 uint16_t *nf_output_iface)
2201 tag_type no_tags = 0;
2202 struct action_xlate_ctx ctx;
2203 COVERAGE_INC(ofproto_ofp2odp);
2204 odp_actions_init(out);
2207 ctx.ofproto = ofproto;
2208 ctx.packet = packet;
2210 ctx.tags = tags ? tags : &no_tags;
2211 ctx.may_set_up_flow = true;
2212 ctx.nf_output_iface = NF_OUT_DROP;
2213 do_xlate_actions(in, n_in, &ctx);
2215 /* Check with in-band control to see if we're allowed to set up this
2217 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2218 ctx.may_set_up_flow = false;
2221 if (may_set_up_flow) {
2222 *may_set_up_flow = ctx.may_set_up_flow;
2224 if (nf_output_iface) {
2225 *nf_output_iface = ctx.nf_output_iface;
2227 if (odp_actions_overflow(out)) {
2228 odp_actions_init(out);
2229 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2235 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2236 struct ofp_header *oh)
2238 struct ofp_packet_out *opo;
2239 struct ofpbuf payload, *buffer;
2240 struct odp_actions actions;
2246 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2250 opo = (struct ofp_packet_out *) oh;
2252 COVERAGE_INC(ofproto_packet_out);
2253 if (opo->buffer_id != htonl(UINT32_MAX)) {
2254 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2256 if (error || !buffer) {
2264 flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2265 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2266 &flow, p, &payload, &actions, NULL, NULL, NULL);
2271 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2273 ofpbuf_delete(buffer);
2279 update_port_config(struct ofproto *p, struct ofport *port,
2280 uint32_t config, uint32_t mask)
2282 mask &= config ^ port->opp.config;
2283 if (mask & OFPPC_PORT_DOWN) {
2284 if (config & OFPPC_PORT_DOWN) {
2285 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2287 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2290 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2291 if (mask & REVALIDATE_BITS) {
2292 COVERAGE_INC(ofproto_costly_flags);
2293 port->opp.config ^= mask & REVALIDATE_BITS;
2294 p->need_revalidate = true;
2296 #undef REVALIDATE_BITS
2297 if (mask & OFPPC_NO_FLOOD) {
2298 port->opp.config ^= OFPPC_NO_FLOOD;
2299 refresh_port_group(p, DP_GROUP_FLOOD);
2301 if (mask & OFPPC_NO_PACKET_IN) {
2302 port->opp.config ^= OFPPC_NO_PACKET_IN;
2307 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2309 const struct ofp_port_mod *opm;
2310 struct ofport *port;
2313 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2317 opm = (struct ofp_port_mod *) oh;
2319 port = port_array_get(&p->ports,
2320 ofp_port_to_odp_port(ntohs(opm->port_no)));
2322 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2323 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2324 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2326 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2327 if (opm->advertise) {
2328 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2334 static struct ofpbuf *
2335 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2337 struct ofp_stats_reply *osr;
2340 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2341 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2343 osr->flags = htons(0);
2347 static struct ofpbuf *
2348 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2350 return make_stats_reply(request->header.xid, request->type, body_len);
2354 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2356 struct ofpbuf *msg = *msgp;
2357 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2358 if (nbytes + msg->size > UINT16_MAX) {
2359 struct ofp_stats_reply *reply = msg->data;
2360 reply->flags = htons(OFPSF_REPLY_MORE);
2361 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2362 queue_tx(msg, ofconn, ofconn->reply_counter);
2364 return ofpbuf_put_uninit(*msgp, nbytes);
2368 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2369 struct ofp_stats_request *request)
2371 struct ofp_desc_stats *ods;
2374 msg = start_stats_reply(request, sizeof *ods);
2375 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2376 strncpy(ods->mfr_desc, p->manufacturer, sizeof ods->mfr_desc);
2377 strncpy(ods->hw_desc, p->hardware, sizeof ods->hw_desc);
2378 strncpy(ods->sw_desc, p->software, sizeof ods->sw_desc);
2379 strncpy(ods->serial_num, p->serial, sizeof ods->serial_num);
2380 queue_tx(msg, ofconn, ofconn->reply_counter);
2386 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2388 struct rule *rule = rule_from_cls_rule(cls_rule);
2389 int *n_subrules = n_subrules_;
2397 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2398 struct ofp_stats_request *request)
2400 struct ofp_table_stats *ots;
2402 struct odp_stats dpstats;
2403 int n_exact, n_subrules, n_wild;
2405 msg = start_stats_reply(request, sizeof *ots * 2);
2407 /* Count rules of various kinds. */
2409 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2410 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2411 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2414 dpif_get_dp_stats(p->dpif, &dpstats);
2415 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2416 memset(ots, 0, sizeof *ots);
2417 ots->table_id = TABLEID_HASH;
2418 strcpy(ots->name, "hash");
2419 ots->wildcards = htonl(0);
2420 ots->max_entries = htonl(dpstats.max_capacity);
2421 ots->active_count = htonl(n_exact);
2422 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2424 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2426 /* Classifier table. */
2427 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2428 memset(ots, 0, sizeof *ots);
2429 ots->table_id = TABLEID_CLASSIFIER;
2430 strcpy(ots->name, "classifier");
2431 ots->wildcards = htonl(OFPFW_ALL);
2432 ots->max_entries = htonl(65536);
2433 ots->active_count = htonl(n_wild);
2434 ots->lookup_count = htonll(0); /* XXX */
2435 ots->matched_count = htonll(0); /* XXX */
2437 queue_tx(msg, ofconn, ofconn->reply_counter);
2442 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2443 struct ofp_stats_request *request)
2445 struct ofp_port_stats *ops;
2447 struct ofport *port;
2448 unsigned int port_no;
2450 msg = start_stats_reply(request, sizeof *ops * 16);
2451 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2452 struct netdev_stats stats;
2454 /* Intentionally ignore return value, since errors will set 'stats' to
2455 * all-1s, which is correct for OpenFlow, and netdev_get_stats() will
2457 netdev_get_stats(port->netdev, &stats);
2459 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2460 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2461 memset(ops->pad, 0, sizeof ops->pad);
2462 ops->rx_packets = htonll(stats.rx_packets);
2463 ops->tx_packets = htonll(stats.tx_packets);
2464 ops->rx_bytes = htonll(stats.rx_bytes);
2465 ops->tx_bytes = htonll(stats.tx_bytes);
2466 ops->rx_dropped = htonll(stats.rx_dropped);
2467 ops->tx_dropped = htonll(stats.tx_dropped);
2468 ops->rx_errors = htonll(stats.rx_errors);
2469 ops->tx_errors = htonll(stats.tx_errors);
2470 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2471 ops->rx_over_err = htonll(stats.rx_over_errors);
2472 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2473 ops->collisions = htonll(stats.collisions);
2476 queue_tx(msg, ofconn, ofconn->reply_counter);
2480 struct flow_stats_cbdata {
2481 struct ofproto *ofproto;
2482 struct ofconn *ofconn;
2488 query_stats(struct ofproto *p, struct rule *rule,
2489 uint64_t *packet_countp, uint64_t *byte_countp)
2491 uint64_t packet_count, byte_count;
2492 struct rule *subrule;
2493 struct odp_flow *odp_flows;
2496 packet_count = rule->packet_count;
2497 byte_count = rule->byte_count;
2499 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2500 odp_flows = xcalloc(1, n_odp_flows * sizeof *odp_flows);
2501 if (rule->cr.wc.wildcards) {
2503 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2504 odp_flows[i++].key = subrule->cr.flow;
2505 packet_count += subrule->packet_count;
2506 byte_count += subrule->byte_count;
2509 odp_flows[0].key = rule->cr.flow;
2512 packet_count = rule->packet_count;
2513 byte_count = rule->byte_count;
2514 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2516 for (i = 0; i < n_odp_flows; i++) {
2517 struct odp_flow *odp_flow = &odp_flows[i];
2518 packet_count += odp_flow->stats.n_packets;
2519 byte_count += odp_flow->stats.n_bytes;
2524 *packet_countp = packet_count;
2525 *byte_countp = byte_count;
2529 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2531 struct rule *rule = rule_from_cls_rule(rule_);
2532 struct flow_stats_cbdata *cbdata = cbdata_;
2533 struct ofp_flow_stats *ofs;
2534 uint64_t packet_count, byte_count;
2535 size_t act_len, len;
2537 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2541 act_len = sizeof *rule->actions * rule->n_actions;
2542 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2544 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2546 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2547 ofs->length = htons(len);
2548 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2550 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match);
2551 ofs->duration = htonl((time_msec() - rule->created) / 1000);
2552 ofs->priority = htons(rule->cr.priority);
2553 ofs->idle_timeout = htons(rule->idle_timeout);
2554 ofs->hard_timeout = htons(rule->hard_timeout);
2556 ofs->packet_count = htonll(packet_count);
2557 ofs->byte_count = htonll(byte_count);
2558 memcpy(ofs->actions, rule->actions, act_len);
2562 table_id_to_include(uint8_t table_id)
2564 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2565 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2566 : table_id == 0xff ? CLS_INC_ALL
2571 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2572 const struct ofp_stats_request *osr,
2575 struct ofp_flow_stats_request *fsr;
2576 struct flow_stats_cbdata cbdata;
2577 struct cls_rule target;
2579 if (arg_size != sizeof *fsr) {
2580 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2582 fsr = (struct ofp_flow_stats_request *) osr->body;
2584 COVERAGE_INC(ofproto_flows_req);
2586 cbdata.ofconn = ofconn;
2587 cbdata.out_port = fsr->out_port;
2588 cbdata.msg = start_stats_reply(osr, 1024);
2589 cls_rule_from_match(&target, &fsr->match, 0);
2590 classifier_for_each_match(&p->cls, &target,
2591 table_id_to_include(fsr->table_id),
2592 flow_stats_cb, &cbdata);
2593 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2597 struct flow_stats_ds_cbdata {
2598 struct ofproto *ofproto;
2603 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
2605 struct rule *rule = rule_from_cls_rule(rule_);
2606 struct flow_stats_ds_cbdata *cbdata = cbdata_;
2607 struct ds *results = cbdata->results;
2608 struct ofp_match match;
2609 uint64_t packet_count, byte_count;
2610 size_t act_len = sizeof *rule->actions * rule->n_actions;
2612 /* Don't report on subrules. */
2613 if (rule->super != NULL) {
2617 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2618 flow_to_ovs_match(&rule->cr.flow, rule->cr.wc.wildcards, &match);
2620 ds_put_format(results, "duration=%llds, ",
2621 (time_msec() - rule->created) / 1000);
2622 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2623 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2624 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2625 ofp_print_match(results, &match, true);
2626 ofp_print_actions(results, &rule->actions->header, act_len);
2627 ds_put_cstr(results, "\n");
2630 /* Adds a pretty-printed description of all flows to 'results', including
2631 * those marked hidden by secchan (e.g., by in-band control). */
2633 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2635 struct ofp_match match;
2636 struct cls_rule target;
2637 struct flow_stats_ds_cbdata cbdata;
2639 memset(&match, 0, sizeof match);
2640 match.wildcards = htonl(OFPFW_ALL);
2643 cbdata.results = results;
2645 cls_rule_from_match(&target, &match, 0);
2646 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2647 flow_stats_ds_cb, &cbdata);
2650 struct aggregate_stats_cbdata {
2651 struct ofproto *ofproto;
2653 uint64_t packet_count;
2654 uint64_t byte_count;
2659 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2661 struct rule *rule = rule_from_cls_rule(rule_);
2662 struct aggregate_stats_cbdata *cbdata = cbdata_;
2663 uint64_t packet_count, byte_count;
2665 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2669 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2671 cbdata->packet_count += packet_count;
2672 cbdata->byte_count += byte_count;
2677 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2678 const struct ofp_stats_request *osr,
2681 struct ofp_aggregate_stats_request *asr;
2682 struct ofp_aggregate_stats_reply *reply;
2683 struct aggregate_stats_cbdata cbdata;
2684 struct cls_rule target;
2687 if (arg_size != sizeof *asr) {
2688 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2690 asr = (struct ofp_aggregate_stats_request *) osr->body;
2692 COVERAGE_INC(ofproto_agg_request);
2694 cbdata.out_port = asr->out_port;
2695 cbdata.packet_count = 0;
2696 cbdata.byte_count = 0;
2698 cls_rule_from_match(&target, &asr->match, 0);
2699 classifier_for_each_match(&p->cls, &target,
2700 table_id_to_include(asr->table_id),
2701 aggregate_stats_cb, &cbdata);
2703 msg = start_stats_reply(osr, sizeof *reply);
2704 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2705 reply->flow_count = htonl(cbdata.n_flows);
2706 reply->packet_count = htonll(cbdata.packet_count);
2707 reply->byte_count = htonll(cbdata.byte_count);
2708 queue_tx(msg, ofconn, ofconn->reply_counter);
2713 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2714 struct ofp_header *oh)
2716 struct ofp_stats_request *osr;
2720 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2725 osr = (struct ofp_stats_request *) oh;
2727 switch (ntohs(osr->type)) {
2729 return handle_desc_stats_request(p, ofconn, osr);
2732 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2734 case OFPST_AGGREGATE:
2735 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2738 return handle_table_stats_request(p, ofconn, osr);
2741 return handle_port_stats_request(p, ofconn, osr);
2744 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2747 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2751 static long long int
2752 msec_from_nsec(uint64_t sec, uint32_t nsec)
2754 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2758 update_time(struct ofproto *ofproto, struct rule *rule,
2759 const struct odp_flow_stats *stats)
2761 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2762 if (used > rule->used) {
2764 if (rule->super && used > rule->super->used) {
2765 rule->super->used = used;
2767 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
2772 update_stats(struct ofproto *ofproto, struct rule *rule,
2773 const struct odp_flow_stats *stats)
2775 if (stats->n_packets) {
2776 update_time(ofproto, rule, stats);
2777 rule->packet_count += stats->n_packets;
2778 rule->byte_count += stats->n_bytes;
2779 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
2785 add_flow(struct ofproto *p, struct ofconn *ofconn,
2786 struct ofp_flow_mod *ofm, size_t n_actions)
2788 struct ofpbuf *packet;
2793 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
2797 flow_from_match(&flow, &wildcards, &ofm->match);
2798 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
2799 ntohs(ofm->priority))) {
2800 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
2804 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
2805 n_actions, ntohs(ofm->idle_timeout),
2806 ntohs(ofm->hard_timeout));
2807 cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority));
2811 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2812 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2816 rule_insert(p, rule, packet, in_port);
2817 ofpbuf_delete(packet);
2822 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
2823 size_t n_actions, uint16_t command, struct rule *rule)
2825 if (rule_is_hidden(rule)) {
2829 if (command == OFPFC_DELETE) {
2830 rule_remove(p, rule);
2832 size_t actions_len = n_actions * sizeof *rule->actions;
2834 if (n_actions == rule->n_actions
2835 && !memcmp(ofm->actions, rule->actions, actions_len))
2840 free(rule->actions);
2841 rule->actions = xmemdup(ofm->actions, actions_len);
2842 rule->n_actions = n_actions;
2844 if (rule->cr.wc.wildcards) {
2845 COVERAGE_INC(ofproto_mod_wc_flow);
2846 p->need_revalidate = true;
2848 rule_update_actions(p, rule);
2856 modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm,
2857 size_t n_actions, uint16_t command)
2863 flow_from_match(&flow, &wildcards, &ofm->match);
2864 rule = rule_from_cls_rule(classifier_find_rule_exactly(
2865 &p->cls, &flow, wildcards,
2866 ntohs(ofm->priority)));
2869 if (command == OFPFC_DELETE
2870 && ofm->out_port != htons(OFPP_NONE)
2871 && !rule_has_out_port(rule, ofm->out_port)) {
2875 modify_flow(p, ofm, n_actions, command, rule);
2880 struct modify_flows_cbdata {
2881 struct ofproto *ofproto;
2882 const struct ofp_flow_mod *ofm;
2889 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
2891 struct rule *rule = rule_from_cls_rule(rule_);
2892 struct modify_flows_cbdata *cbdata = cbdata_;
2894 if (cbdata->out_port != htons(OFPP_NONE)
2895 && !rule_has_out_port(rule, cbdata->out_port)) {
2899 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions,
2900 cbdata->command, rule);
2904 modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm,
2905 size_t n_actions, uint16_t command)
2907 struct modify_flows_cbdata cbdata;
2908 struct cls_rule target;
2912 cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port
2913 : htons(OFPP_NONE));
2914 cbdata.n_actions = n_actions;
2915 cbdata.command = command;
2917 cls_rule_from_match(&target, &ofm->match, 0);
2919 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2920 modify_flows_cb, &cbdata);
2925 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
2926 struct ofp_flow_mod *ofm)
2931 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
2932 sizeof *ofm->actions, &n_actions);
2937 /* We do not support the emergency flow cache. It will hopefully
2938 * get dropped from OpenFlow in the near future. */
2939 if (ofm->flags & htons(OFPFF_EMERG)) {
2940 /* There isn't a good fit for an error code, so just state that the
2941 * flow table is full. */
2942 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
2945 normalize_match(&ofm->match);
2946 if (!ofm->match.wildcards) {
2947 ofm->priority = htons(UINT16_MAX);
2950 error = validate_actions((const union ofp_action *) ofm->actions,
2951 n_actions, p->max_ports);
2956 switch (ntohs(ofm->command)) {
2958 return add_flow(p, ofconn, ofm, n_actions);
2961 return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY);
2963 case OFPFC_MODIFY_STRICT:
2964 return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY);
2967 return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE);
2969 case OFPFC_DELETE_STRICT:
2970 return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE);
2973 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
2978 send_capability_reply(struct ofproto *p, struct ofconn *ofconn, uint32_t xid)
2980 struct ofmp_capability_reply *ocr;
2982 char capabilities[] = "com.nicira.mgmt.manager=false\n";
2984 ocr = make_openflow_xid(sizeof(*ocr), OFPT_VENDOR, xid, &b);
2985 ocr->header.header.vendor = htonl(NX_VENDOR_ID);
2986 ocr->header.header.subtype = htonl(NXT_MGMT);
2987 ocr->header.type = htons(OFMPT_CAPABILITY_REPLY);
2989 ocr->format = htonl(OFMPCOF_SIMPLE);
2990 ocr->mgmt_id = htonll(p->mgmt_id);
2992 ofpbuf_put(b, capabilities, strlen(capabilities));
2994 queue_tx(b, ofconn, ofconn->reply_counter);
2998 handle_ofmp(struct ofproto *p, struct ofconn *ofconn,
2999 struct ofmp_header *ofmph)
3001 size_t msg_len = ntohs(ofmph->header.header.length);
3002 if (msg_len < sizeof(*ofmph)) {
3003 VLOG_WARN_RL(&rl, "dropping short managment message: %zu\n", msg_len);
3004 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3007 if (ofmph->type == htons(OFMPT_CAPABILITY_REQUEST)) {
3008 struct ofmp_capability_request *ofmpcr;
3010 if (msg_len < sizeof(struct ofmp_capability_request)) {
3011 VLOG_WARN_RL(&rl, "dropping short capability request: %zu\n",
3013 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3016 ofmpcr = (struct ofmp_capability_request *)ofmph;
3017 if (ofmpcr->format != htonl(OFMPCAF_SIMPLE)) {
3018 /* xxx Find a better type than bad subtype */
3019 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3022 send_capability_reply(p, ofconn, ofmph->header.header.xid);
3025 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3030 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3032 struct ofp_vendor_header *ovh = msg;
3033 struct nicira_header *nh;
3035 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3036 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3038 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3039 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3041 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3042 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3046 switch (ntohl(nh->subtype)) {
3047 case NXT_STATUS_REQUEST:
3048 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3051 case NXT_ACT_SET_CONFIG:
3052 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
3054 case NXT_ACT_GET_CONFIG:
3055 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
3057 case NXT_COMMAND_REQUEST:
3059 return executer_handle_request(p->executer, ofconn->rconn, msg);
3064 return handle_ofmp(p, ofconn, msg);
3067 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3071 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3072 struct ofpbuf *ofp_msg)
3074 struct ofp_header *oh = ofp_msg->data;
3077 COVERAGE_INC(ofproto_recv_openflow);
3079 case OFPT_ECHO_REQUEST:
3080 error = handle_echo_request(ofconn, oh);
3083 case OFPT_ECHO_REPLY:
3087 case OFPT_FEATURES_REQUEST:
3088 error = handle_features_request(p, ofconn, oh);
3091 case OFPT_GET_CONFIG_REQUEST:
3092 error = handle_get_config_request(p, ofconn, oh);
3095 case OFPT_SET_CONFIG:
3096 error = handle_set_config(p, ofconn, ofp_msg->data);
3099 case OFPT_PACKET_OUT:
3100 error = handle_packet_out(p, ofconn, ofp_msg->data);
3104 error = handle_port_mod(p, oh);
3108 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3111 case OFPT_STATS_REQUEST:
3112 error = handle_stats_request(p, ofconn, oh);
3116 error = handle_vendor(p, ofconn, ofp_msg->data);
3120 if (VLOG_IS_WARN_ENABLED()) {
3121 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3122 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3125 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3130 send_error_oh(ofconn, ofp_msg->data, error);
3135 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3137 struct odp_msg *msg = packet->data;
3138 uint16_t in_port = odp_port_to_ofp_port(msg->port);
3140 struct ofpbuf payload;
3143 /* Handle controller actions. */
3144 if (msg->type == _ODPL_ACTION_NR) {
3145 COVERAGE_INC(ofproto_ctlr_action);
3146 pinsched_send(p->action_sched, in_port, packet,
3147 send_packet_in_action, p);
3151 payload.data = msg + 1;
3152 payload.size = msg->length - sizeof *msg;
3153 flow_extract(&payload, msg->port, &flow);
3155 /* Check with in-band control to see if this packet should be sent
3156 * to the local port regardless of the flow table. */
3157 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3158 union odp_action action;
3160 memset(&action, 0, sizeof(action));
3161 action.output.type = ODPAT_OUTPUT;
3162 action.output.port = ODPP_LOCAL;
3163 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3166 rule = lookup_valid_rule(p, &flow);
3168 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3169 struct ofport *port = port_array_get(&p->ports, msg->port);
3171 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3172 COVERAGE_INC(ofproto_no_packet_in);
3173 /* XXX install 'drop' flow entry */
3174 ofpbuf_delete(packet);
3178 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3181 COVERAGE_INC(ofproto_packet_in);
3182 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3186 if (rule->cr.wc.wildcards) {
3187 rule = rule_create_subrule(p, rule, &flow);
3188 rule_make_actions(p, rule, packet);
3190 if (!rule->may_install) {
3191 /* The rule is not installable, that is, we need to process every
3192 * packet, so process the current packet and set its actions into
3194 rule_make_actions(p, rule, packet);
3196 /* XXX revalidate rule if it needs it */
3200 rule_execute(p, rule, &payload, &flow);
3201 rule_reinstall(p, rule);
3203 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY
3204 && rconn_is_connected(p->controller->rconn)) {
3206 * Extra-special case for fail-open mode.
3208 * We are in fail-open mode and the packet matched the fail-open rule,
3209 * but we are connected to a controller too. We should send the packet
3210 * up to the controller in the hope that it will try to set up a flow
3211 * and thereby allow us to exit fail-open.
3213 * See the top-level comment in fail-open.c for more information.
3215 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3217 ofpbuf_delete(packet);
3222 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3224 struct rule *sub = rule_from_cls_rule(sub_);
3225 struct revalidate_cbdata *cbdata = cbdata_;
3227 if (cbdata->revalidate_all
3228 || (cbdata->revalidate_subrules && sub->super)
3229 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3230 revalidate_rule(cbdata->ofproto, sub);
3235 revalidate_rule(struct ofproto *p, struct rule *rule)
3237 const flow_t *flow = &rule->cr.flow;
3239 COVERAGE_INC(ofproto_revalidate_rule);
3242 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3244 rule_remove(p, rule);
3246 } else if (super != rule->super) {
3247 COVERAGE_INC(ofproto_revalidate_moved);
3248 list_remove(&rule->list);
3249 list_push_back(&super->list, &rule->list);
3250 rule->super = super;
3251 rule->hard_timeout = super->hard_timeout;
3252 rule->idle_timeout = super->idle_timeout;
3253 rule->created = super->created;
3258 rule_update_actions(p, rule);
3262 static struct ofpbuf *
3263 compose_flow_exp(const struct rule *rule, long long int now, uint8_t reason)
3265 struct ofp_flow_expired *ofe;
3267 long long int last_used = rule->used ? now - rule->used : 0;
3269 ofe = make_openflow(sizeof *ofe, OFPT_FLOW_EXPIRED, &buf);
3270 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofe->match);
3271 ofe->priority = htons(rule->cr.priority);
3272 ofe->reason = reason;
3273 ofe->duration = htonl((now - rule->created - last_used) / 1000);
3274 ofe->packet_count = htonll(rule->packet_count);
3275 ofe->byte_count = htonll(rule->byte_count);
3281 send_flow_exp(struct ofproto *p, struct rule *rule,
3282 long long int now, uint8_t reason)
3284 struct ofconn *ofconn;
3285 struct ofconn *prev;
3286 struct ofpbuf *buf = NULL;
3288 /* We limit the maximum number of queued flow expirations it by accounting
3289 * them under the counter for replies. That works because preventing
3290 * OpenFlow requests from being processed also prevents new flows from
3291 * being added (and expiring). (It also prevents processing OpenFlow
3292 * requests that would not add new flows, so it is imperfect.) */
3295 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3296 if (ofconn->send_flow_exp && rconn_is_connected(ofconn->rconn)) {
3298 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3300 buf = compose_flow_exp(rule, now, reason);
3306 queue_tx(buf, prev, prev->reply_counter);
3311 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3313 assert(rule->installed);
3314 assert(!rule->cr.wc.wildcards);
3317 rule_remove(ofproto, rule);
3319 rule_uninstall(ofproto, rule);
3324 expire_rule(struct cls_rule *cls_rule, void *p_)
3326 struct ofproto *p = p_;
3327 struct rule *rule = rule_from_cls_rule(cls_rule);
3328 long long int hard_expire, idle_expire, expire, now;
3330 hard_expire = (rule->hard_timeout
3331 ? rule->created + rule->hard_timeout * 1000
3333 idle_expire = (rule->idle_timeout
3334 && (rule->super || list_is_empty(&rule->list))
3335 ? rule->used + rule->idle_timeout * 1000
3337 expire = MIN(hard_expire, idle_expire);
3341 if (rule->installed && now >= rule->used + 5000) {
3342 uninstall_idle_flow(p, rule);
3343 } else if (!rule->cr.wc.wildcards) {
3344 active_timeout(p, rule);
3350 COVERAGE_INC(ofproto_expired);
3352 /* Update stats. This code will be a no-op if the rule expired
3353 * due to an idle timeout. */
3354 if (rule->cr.wc.wildcards) {
3355 struct rule *subrule, *next;
3356 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3357 rule_remove(p, subrule);
3360 rule_uninstall(p, rule);
3363 if (!rule_is_hidden(rule)) {
3364 send_flow_exp(p, rule, now,
3366 ? OFPER_HARD_TIMEOUT : OFPER_IDLE_TIMEOUT));
3368 rule_remove(p, rule);
3372 active_timeout(struct ofproto *ofproto, struct rule *rule)
3374 if (ofproto->netflow && !is_controller_rule(rule) &&
3375 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
3376 struct ofexpired expired;
3377 struct odp_flow odp_flow;
3379 /* Get updated flow stats. */
3380 memset(&odp_flow, 0, sizeof odp_flow);
3381 if (rule->installed) {
3382 odp_flow.key = rule->cr.flow;
3383 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
3384 dpif_flow_get(ofproto->dpif, &odp_flow);
3386 if (odp_flow.stats.n_packets) {
3387 update_time(ofproto, rule, &odp_flow.stats);
3388 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
3389 odp_flow.stats.tcp_flags);
3393 expired.flow = rule->cr.flow;
3394 expired.packet_count = rule->packet_count +
3395 odp_flow.stats.n_packets;
3396 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
3397 expired.used = rule->used;
3399 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
3401 /* Schedule us to send the accumulated records once we have
3402 * collected all of them. */
3403 poll_immediate_wake();
3408 update_used(struct ofproto *p)
3410 struct odp_flow *flows;
3415 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
3420 for (i = 0; i < n_flows; i++) {
3421 struct odp_flow *f = &flows[i];
3424 rule = rule_from_cls_rule(
3425 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3426 if (!rule || !rule->installed) {
3427 COVERAGE_INC(ofproto_unexpected_rule);
3428 dpif_flow_del(p->dpif, f);
3432 update_time(p, rule, &f->stats);
3433 rule_account(p, rule, f->stats.n_bytes);
3439 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3440 const struct ofpbuf *packet, int send_len)
3442 struct odp_msg *msg = packet->data;
3443 struct ofpbuf payload;
3447 /* Extract packet payload from 'msg'. */
3448 payload.data = msg + 1;
3449 payload.size = msg->length - sizeof *msg;
3451 /* Construct ofp_packet_in message. */
3452 reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3453 opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port), reason,
3454 &payload, send_len);
3457 rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100);
3461 send_packet_in_action(struct ofpbuf *packet, void *p_)
3463 struct ofproto *p = p_;
3464 struct ofconn *ofconn;
3465 struct odp_msg *msg;
3468 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3469 if (ofconn == p->controller || ofconn->miss_send_len) {
3470 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3473 ofpbuf_delete(packet);
3477 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3479 struct ofproto *p = p_;
3480 bool in_fail_open = p->fail_open && fail_open_is_active(p->fail_open);
3481 struct ofconn *ofconn;
3482 struct ofpbuf payload;
3483 struct odp_msg *msg;
3486 payload.data = msg + 1;
3487 payload.size = msg->length - sizeof *msg;
3488 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3489 if (ofconn->miss_send_len) {
3490 struct pktbuf *pb = ofconn->pktbuf;
3491 uint32_t buffer_id = (in_fail_open
3493 : pktbuf_save(pb, &payload, msg->port));
3494 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3496 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3499 ofpbuf_delete(packet);
3503 pick_datapath_id(const struct ofproto *ofproto)
3505 const struct ofport *port;
3507 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
3509 uint8_t ea[ETH_ADDR_LEN];
3512 error = netdev_get_etheraddr(port->netdev, ea);
3514 return eth_addr_to_uint64(ea);
3516 VLOG_WARN("could not get MAC address for %s (%s)",
3517 netdev_get_name(port->netdev), strerror(error));
3519 return ofproto->fallback_dpid;
3523 pick_fallback_dpid(void)
3525 uint8_t ea[ETH_ADDR_LEN];
3526 eth_addr_random(ea);
3527 ea[0] = 0x00; /* Set Nicira OUI. */
3530 return eth_addr_to_uint64(ea);
3534 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3535 struct odp_actions *actions, tag_type *tags,
3536 uint16_t *nf_output_iface, void *ofproto_)
3538 struct ofproto *ofproto = ofproto_;
3541 /* Drop frames for reserved multicast addresses. */
3542 if (eth_addr_is_reserved(flow->dl_dst)) {
3546 /* Learn source MAC (but don't try to learn from revalidation). */
3547 if (packet != NULL) {
3548 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3551 /* The log messages here could actually be useful in debugging,
3552 * so keep the rate limit relatively high. */
3553 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3554 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3555 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3556 ofproto_revalidate(ofproto, rev_tag);
3560 /* Determine output port. */
3561 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3563 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
3564 } else if (out_port != flow->in_port) {
3565 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3566 *nf_output_iface = out_port;
3574 static const struct ofhooks default_ofhooks = {
3576 default_normal_ofhook_cb,