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_setup_flow);
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 uint8_t tcp_flags; /* Bitwise-OR of all TCP flags seen. */
98 uint8_t ip_tos; /* Last-seen IP type-of-service. */
99 tag_type tags; /* Tags (set only by hooks). */
101 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
102 * exact-match rule (having cr.wc.wildcards of 0) generated from the
103 * wildcard rule 'super'. In this case, 'list' is an element of the
106 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
107 * a list of subrules. A super-rule with no wildcards (where
108 * cr.wc.wildcards is 0) will never have any subrules. */
114 * A subrule has no actions (it uses the super-rule's actions). */
116 union ofp_action *actions;
120 * A super-rule with wildcard fields never has ODP actions (since the
121 * datapath only supports exact-match flows). */
122 bool installed; /* Installed in datapath? */
123 bool may_install; /* True ordinarily; false if actions must
124 * be reassessed for every packet. */
126 union odp_action *odp_actions;
130 rule_is_hidden(const struct rule *rule)
132 /* Subrules are merely an implementation detail, so hide them from the
134 if (rule->super != NULL) {
138 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
139 * (e.g. by in-band control) and are intentionally hidden from the
141 if (rule->cr.priority > UINT16_MAX) {
148 static struct rule *rule_create(struct rule *super, const union ofp_action *,
149 size_t n_actions, uint16_t idle_timeout,
150 uint16_t hard_timeout);
151 static void rule_free(struct rule *);
152 static void rule_destroy(struct ofproto *, struct rule *);
153 static struct rule *rule_from_cls_rule(const struct cls_rule *);
154 static void rule_insert(struct ofproto *, struct rule *,
155 struct ofpbuf *packet, uint16_t in_port);
156 static void rule_remove(struct ofproto *, struct rule *);
157 static bool rule_make_actions(struct ofproto *, struct rule *,
158 const struct ofpbuf *packet);
159 static void rule_install(struct ofproto *, struct rule *,
160 struct rule *displaced_rule);
161 static void rule_uninstall(struct ofproto *, struct rule *);
162 static void rule_post_uninstall(struct ofproto *, struct rule *);
167 struct pktbuf *pktbuf;
171 struct rconn_packet_counter *packet_in_counter;
173 /* Number of OpenFlow messages queued as replies to OpenFlow requests, and
174 * the maximum number before we stop reading OpenFlow requests. */
175 #define OFCONN_REPLY_MAX 100
176 struct rconn_packet_counter *reply_counter;
179 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *);
180 static void ofconn_destroy(struct ofconn *, struct ofproto *);
181 static void ofconn_run(struct ofconn *, struct ofproto *);
182 static void ofconn_wait(struct ofconn *);
183 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
184 struct rconn_packet_counter *counter);
188 uint64_t datapath_id; /* Datapath ID. */
189 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
190 uint64_t mgmt_id; /* Management channel identifier. */
191 char *manufacturer; /* Manufacturer. */
192 char *hardware; /* Hardware. */
193 char *software; /* Software version. */
194 char *serial; /* Serial number. */
198 struct netdev_monitor *netdev_monitor;
199 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
201 struct shash port_by_name;
205 struct switch_status *switch_status;
206 struct status_category *ss_cat;
207 struct in_band *in_band;
208 struct discovery *discovery;
209 struct fail_open *fail_open;
210 struct pinsched *miss_sched, *action_sched;
211 struct executer *executer;
212 struct netflow *netflow;
215 struct classifier cls;
216 bool need_revalidate;
217 long long int next_expiration;
218 struct tag_set revalidate_set;
220 /* OpenFlow connections. */
221 struct list all_conns;
222 struct ofconn *controller;
223 struct pvconn **listeners;
225 struct pvconn **snoops;
228 /* Hooks for ovs-vswitchd. */
229 const struct ofhooks *ofhooks;
232 /* Used by default ofhooks. */
233 struct mac_learning *ml;
236 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
238 static const struct ofhooks default_ofhooks;
240 static uint64_t pick_datapath_id(const struct ofproto *);
241 static uint64_t pick_fallback_dpid(void);
242 static void send_packet_in_miss(struct ofpbuf *, void *ofproto);
243 static void send_packet_in_action(struct ofpbuf *, void *ofproto);
244 static void update_used(struct ofproto *);
245 static void update_stats(struct rule *, const struct odp_flow_stats *);
246 static void expire_rule(struct cls_rule *, void *ofproto);
247 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
248 static void revalidate_cb(struct cls_rule *rule_, void *p_);
250 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
252 static void handle_openflow(struct ofconn *, struct ofproto *,
255 static void refresh_port_group(struct ofproto *, unsigned int group);
256 static void update_port(struct ofproto *, const char *devname);
257 static int init_ports(struct ofproto *);
258 static void reinit_ports(struct ofproto *);
261 ofproto_create(const char *datapath, const struct ofhooks *ofhooks, void *aux,
262 struct ofproto **ofprotop)
264 struct odp_stats stats;
271 /* Connect to datapath and start listening for messages. */
272 error = dpif_open(datapath, &dpif);
274 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
277 error = dpif_get_dp_stats(dpif, &stats);
279 VLOG_ERR("failed to obtain stats for datapath %s: %s",
280 datapath, strerror(error));
284 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION);
286 VLOG_ERR("failed to listen on datapath %s: %s",
287 datapath, strerror(error));
291 dpif_flow_flush(dpif);
292 dpif_recv_purge(dpif);
294 /* Initialize settings. */
295 p = xcalloc(1, sizeof *p);
296 p->fallback_dpid = pick_fallback_dpid();
297 p->datapath_id = p->fallback_dpid;
298 p->manufacturer = xstrdup("Nicira Networks, Inc.");
299 p->hardware = xstrdup("Reference Implementation");
300 p->software = xstrdup(VERSION BUILDNR);
301 p->serial = xstrdup("None");
303 /* Initialize datapath. */
305 p->netdev_monitor = netdev_monitor_create();
306 port_array_init(&p->ports);
307 shash_init(&p->port_by_name);
308 p->max_ports = stats.max_ports;
310 /* Initialize submodules. */
311 p->switch_status = switch_status_create(p);
315 p->miss_sched = p->action_sched = NULL;
319 /* Initialize flow table. */
320 classifier_init(&p->cls);
321 p->need_revalidate = false;
322 p->next_expiration = time_msec() + 1000;
323 tag_set_init(&p->revalidate_set);
325 /* Initialize OpenFlow connections. */
326 list_init(&p->all_conns);
327 p->controller = ofconn_create(p, rconn_create(5, 8));
328 p->controller->pktbuf = pktbuf_create();
329 p->controller->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
335 /* Initialize hooks. */
337 p->ofhooks = ofhooks;
341 p->ofhooks = &default_ofhooks;
343 p->ml = mac_learning_create();
346 /* Register switch status category. */
347 p->ss_cat = switch_status_register(p->switch_status, "remote",
348 rconn_status_cb, p->controller->rconn);
351 error = init_ports(p);
357 /* Pick final datapath ID. */
358 p->datapath_id = pick_datapath_id(p);
359 VLOG_INFO("using datapath ID %012"PRIx64, p->datapath_id);
366 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
368 uint64_t old_dpid = p->datapath_id;
369 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
370 if (p->datapath_id != old_dpid) {
371 VLOG_INFO("datapath ID changed to %012"PRIx64, p->datapath_id);
372 rconn_reconnect(p->controller->rconn);
377 ofproto_set_mgmt_id(struct ofproto *p, uint64_t mgmt_id)
379 p->mgmt_id = mgmt_id;
383 ofproto_set_probe_interval(struct ofproto *p, int probe_interval)
385 probe_interval = probe_interval ? MAX(probe_interval, 5) : 0;
386 rconn_set_probe_interval(p->controller->rconn, probe_interval);
388 int trigger_duration = probe_interval ? probe_interval * 3 : 15;
389 fail_open_set_trigger_duration(p->fail_open, trigger_duration);
394 ofproto_set_max_backoff(struct ofproto *p, int max_backoff)
396 rconn_set_max_backoff(p->controller->rconn, max_backoff);
400 ofproto_set_desc(struct ofproto *p,
401 const char *manufacturer, const char *hardware,
402 const char *software, const char *serial)
405 free(p->manufacturer);
406 p->manufacturer = xstrdup(manufacturer);
410 p->hardware = xstrdup(hardware);
414 p->software = xstrdup(software);
418 p->serial = xstrdup(serial);
423 ofproto_set_in_band(struct ofproto *p, bool in_band)
425 if (in_band != (p->in_band != NULL)) {
427 in_band_create(p, p->switch_status, p->controller->rconn,
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, const struct svec *collectors,
537 uint8_t engine_type, uint8_t engine_id, bool add_id_to_iface)
539 if (collectors && collectors->n) {
540 if (!ofproto->netflow) {
541 ofproto->netflow = netflow_create();
543 netflow_set_engine(ofproto->netflow, engine_type, engine_id,
545 return netflow_set_collectors(ofproto->netflow, collectors);
547 netflow_destroy(ofproto->netflow);
548 ofproto->netflow = NULL;
554 ofproto_set_failure(struct ofproto *ofproto, bool fail_open)
557 struct rconn *rconn = ofproto->controller->rconn;
558 int trigger_duration = rconn_get_probe_interval(rconn) * 3;
559 if (!ofproto->fail_open) {
560 ofproto->fail_open = fail_open_create(ofproto, trigger_duration,
561 ofproto->switch_status,
564 fail_open_set_trigger_duration(ofproto->fail_open,
568 fail_open_destroy(ofproto->fail_open);
569 ofproto->fail_open = NULL;
574 ofproto_set_rate_limit(struct ofproto *ofproto,
575 int rate_limit, int burst_limit)
577 if (rate_limit > 0) {
578 if (!ofproto->miss_sched) {
579 ofproto->miss_sched = pinsched_create(rate_limit, burst_limit,
580 ofproto->switch_status);
581 ofproto->action_sched = pinsched_create(rate_limit, burst_limit,
584 pinsched_set_limits(ofproto->miss_sched, rate_limit, burst_limit);
585 pinsched_set_limits(ofproto->action_sched,
586 rate_limit, burst_limit);
589 pinsched_destroy(ofproto->miss_sched);
590 ofproto->miss_sched = NULL;
591 pinsched_destroy(ofproto->action_sched);
592 ofproto->action_sched = NULL;
597 ofproto_set_stp(struct ofproto *ofproto UNUSED, bool enable_stp)
601 VLOG_WARN("STP is not yet implemented");
609 ofproto_set_remote_execution(struct ofproto *ofproto, const char *command_acl,
610 const char *command_dir)
613 if (!ofproto->executer) {
614 return executer_create(command_acl, command_dir,
617 executer_set_acl(ofproto->executer, command_acl, command_dir);
620 executer_destroy(ofproto->executer);
621 ofproto->executer = NULL;
627 ofproto_get_datapath_id(const struct ofproto *ofproto)
629 return ofproto->datapath_id;
633 ofproto_get_mgmt_id(const struct ofproto *ofproto)
635 return ofproto->mgmt_id;
639 ofproto_get_probe_interval(const struct ofproto *ofproto)
641 return rconn_get_probe_interval(ofproto->controller->rconn);
645 ofproto_get_max_backoff(const struct ofproto *ofproto)
647 return rconn_get_max_backoff(ofproto->controller->rconn);
651 ofproto_get_in_band(const struct ofproto *ofproto)
653 return ofproto->in_band != NULL;
657 ofproto_get_discovery(const struct ofproto *ofproto)
659 return ofproto->discovery != NULL;
663 ofproto_get_controller(const struct ofproto *ofproto)
665 return rconn_get_name(ofproto->controller->rconn);
669 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
673 for (i = 0; i < ofproto->n_listeners; i++) {
674 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
679 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
683 for (i = 0; i < ofproto->n_snoops; i++) {
684 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
689 ofproto_destroy(struct ofproto *p)
691 struct ofconn *ofconn, *next_ofconn;
692 struct ofport *ofport;
693 unsigned int port_no;
700 ofproto_flush_flows(p);
701 classifier_destroy(&p->cls);
703 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
705 ofconn_destroy(ofconn, p);
709 netdev_monitor_destroy(p->netdev_monitor);
710 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
713 shash_destroy(&p->port_by_name);
715 switch_status_destroy(p->switch_status);
716 in_band_destroy(p->in_band);
717 discovery_destroy(p->discovery);
718 fail_open_destroy(p->fail_open);
719 pinsched_destroy(p->miss_sched);
720 pinsched_destroy(p->action_sched);
721 executer_destroy(p->executer);
722 netflow_destroy(p->netflow);
724 switch_status_unregister(p->ss_cat);
726 for (i = 0; i < p->n_listeners; i++) {
727 pvconn_close(p->listeners[i]);
731 for (i = 0; i < p->n_snoops; i++) {
732 pvconn_close(p->snoops[i]);
736 mac_learning_destroy(p->ml);
742 ofproto_run(struct ofproto *p)
744 int error = ofproto_run1(p);
746 error = ofproto_run2(p, false);
752 process_port_change(struct ofproto *ofproto, int error, char *devname)
754 if (error == ENOBUFS) {
755 reinit_ports(ofproto);
757 update_port(ofproto, devname);
763 ofproto_run1(struct ofproto *p)
765 struct ofconn *ofconn, *next_ofconn;
770 for (i = 0; i < 50; i++) {
774 error = dpif_recv(p->dpif, &buf);
776 if (error == ENODEV) {
777 /* Someone destroyed the datapath behind our back. The caller
778 * better destroy us and give up, because we're just going to
779 * spin from here on out. */
780 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
781 VLOG_ERR_RL(&rl, "%s: datapath was destroyed externally",
788 handle_odp_msg(p, buf);
791 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
792 process_port_change(p, error, devname);
794 while ((error = netdev_monitor_poll(p->netdev_monitor,
795 &devname)) != EAGAIN) {
796 process_port_change(p, error, devname);
800 in_band_run(p->in_band);
803 char *controller_name;
804 if (rconn_is_connectivity_questionable(p->controller->rconn)) {
805 discovery_question_connectivity(p->discovery);
807 if (discovery_run(p->discovery, &controller_name)) {
808 if (controller_name) {
809 rconn_connect(p->controller->rconn, controller_name);
811 rconn_disconnect(p->controller->rconn);
816 fail_open_run(p->fail_open);
818 pinsched_run(p->miss_sched, send_packet_in_miss, p);
819 pinsched_run(p->action_sched, send_packet_in_action, p);
821 executer_run(p->executer);
824 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
826 ofconn_run(ofconn, p);
829 for (i = 0; i < p->n_listeners; i++) {
833 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
835 ofconn_create(p, rconn_new_from_vconn("passive", vconn));
836 } else if (retval != EAGAIN) {
837 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
841 for (i = 0; i < p->n_snoops; i++) {
845 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
847 rconn_add_monitor(p->controller->rconn, vconn);
848 } else if (retval != EAGAIN) {
849 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
853 if (time_msec() >= p->next_expiration) {
854 COVERAGE_INC(ofproto_expiration);
855 p->next_expiration = time_msec() + 1000;
858 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
860 /* Let the hook know that we're at a stable point: all outstanding data
861 * in existing flows has been accounted to the account_cb. Thus, the
862 * hook can now reasonably do operations that depend on having accurate
863 * flow volume accounting (currently, that's just bond rebalancing). */
864 if (p->ofhooks->account_checkpoint_cb) {
865 p->ofhooks->account_checkpoint_cb(p->aux);
870 netflow_run(p->netflow);
876 struct revalidate_cbdata {
877 struct ofproto *ofproto;
878 bool revalidate_all; /* Revalidate all exact-match rules? */
879 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
880 struct tag_set revalidate_set; /* Set of tags to revalidate. */
884 ofproto_run2(struct ofproto *p, bool revalidate_all)
886 if (p->need_revalidate || revalidate_all
887 || !tag_set_is_empty(&p->revalidate_set)) {
888 struct revalidate_cbdata cbdata;
890 cbdata.revalidate_all = revalidate_all;
891 cbdata.revalidate_subrules = p->need_revalidate;
892 cbdata.revalidate_set = p->revalidate_set;
893 tag_set_init(&p->revalidate_set);
894 COVERAGE_INC(ofproto_revalidate);
895 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
896 p->need_revalidate = false;
903 ofproto_wait(struct ofproto *p)
905 struct ofconn *ofconn;
908 dpif_recv_wait(p->dpif);
909 dpif_port_poll_wait(p->dpif);
910 netdev_monitor_poll_wait(p->netdev_monitor);
911 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
915 in_band_wait(p->in_band);
918 discovery_wait(p->discovery);
921 fail_open_wait(p->fail_open);
923 pinsched_wait(p->miss_sched);
924 pinsched_wait(p->action_sched);
926 executer_wait(p->executer);
928 if (!tag_set_is_empty(&p->revalidate_set)) {
929 poll_immediate_wake();
931 if (p->need_revalidate) {
932 /* Shouldn't happen, but if it does just go around again. */
933 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
934 poll_immediate_wake();
935 } else if (p->next_expiration != LLONG_MAX) {
936 poll_timer_wait(p->next_expiration - time_msec());
938 for (i = 0; i < p->n_listeners; i++) {
939 pvconn_wait(p->listeners[i]);
941 for (i = 0; i < p->n_snoops; i++) {
942 pvconn_wait(p->snoops[i]);
947 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
949 tag_set_add(&ofproto->revalidate_set, tag);
953 ofproto_get_revalidate_set(struct ofproto *ofproto)
955 return &ofproto->revalidate_set;
959 ofproto_is_alive(const struct ofproto *p)
961 return p->discovery || rconn_is_alive(p->controller->rconn);
965 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
966 const union ofp_action *actions, size_t n_actions,
967 const struct ofpbuf *packet)
969 struct odp_actions odp_actions;
972 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
978 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
980 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
981 odp_actions.n_actions, packet);
986 ofproto_add_flow(struct ofproto *p,
987 const flow_t *flow, uint32_t wildcards, unsigned int priority,
988 const union ofp_action *actions, size_t n_actions,
992 rule = rule_create(NULL, actions, n_actions,
993 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */, 0);
994 cls_rule_from_flow(&rule->cr, flow, wildcards, priority);
995 rule_insert(p, rule, NULL, 0);
999 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1000 uint32_t wildcards, unsigned int priority)
1004 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1008 rule_remove(ofproto, rule);
1013 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1015 struct rule *rule = rule_from_cls_rule(rule_);
1016 struct ofproto *ofproto = ofproto_;
1018 /* Mark the flow as not installed, even though it might really be
1019 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1020 * There is no point in uninstalling it individually since we are about to
1021 * blow away all the flows with dpif_flow_flush(). */
1022 rule->installed = false;
1024 rule_remove(ofproto, rule);
1028 ofproto_flush_flows(struct ofproto *ofproto)
1030 COVERAGE_INC(ofproto_flush);
1031 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1032 dpif_flow_flush(ofproto->dpif);
1033 if (ofproto->in_band) {
1034 in_band_flushed(ofproto->in_band);
1036 if (ofproto->fail_open) {
1037 fail_open_flushed(ofproto->fail_open);
1042 reinit_ports(struct ofproto *p)
1044 struct svec devnames;
1045 struct ofport *ofport;
1046 unsigned int port_no;
1047 struct odp_port *odp_ports;
1051 svec_init(&devnames);
1052 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1053 svec_add (&devnames, (char *) ofport->opp.name);
1055 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1056 for (i = 0; i < n_odp_ports; i++) {
1057 svec_add (&devnames, odp_ports[i].devname);
1061 svec_sort_unique(&devnames);
1062 for (i = 0; i < devnames.n; i++) {
1063 update_port(p, devnames.names[i]);
1065 svec_destroy(&devnames);
1069 refresh_port_group(struct ofproto *p, unsigned int group)
1073 struct ofport *port;
1074 unsigned int port_no;
1076 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1078 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1080 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1081 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1082 ports[n_ports++] = port_no;
1085 dpif_port_group_set(p->dpif, group, ports, n_ports);
1090 refresh_port_groups(struct ofproto *p)
1092 refresh_port_group(p, DP_GROUP_FLOOD);
1093 refresh_port_group(p, DP_GROUP_ALL);
1096 static struct ofport *
1097 make_ofport(const struct odp_port *odp_port)
1099 enum netdev_flags flags;
1100 struct ofport *ofport;
1101 struct netdev *netdev;
1105 error = netdev_open(odp_port->devname, NETDEV_ETH_TYPE_NONE, &netdev);
1107 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1108 "cannot be opened (%s)",
1109 odp_port->devname, odp_port->port,
1110 odp_port->devname, strerror(error));
1114 ofport = xmalloc(sizeof *ofport);
1115 ofport->netdev = netdev;
1116 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1117 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1118 memcpy(ofport->opp.name, odp_port->devname,
1119 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1120 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1122 netdev_get_flags(netdev, &flags);
1123 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1125 netdev_get_carrier(netdev, &carrier);
1126 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1128 netdev_get_features(netdev,
1129 &ofport->opp.curr, &ofport->opp.advertised,
1130 &ofport->opp.supported, &ofport->opp.peer);
1135 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1137 if (port_array_get(&p->ports, odp_port->port)) {
1138 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1141 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1142 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1151 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1153 const struct ofp_phy_port *a = &a_->opp;
1154 const struct ofp_phy_port *b = &b_->opp;
1156 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1157 return (a->port_no == b->port_no
1158 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1159 && !strcmp((char *) a->name, (char *) b->name)
1160 && a->state == b->state
1161 && a->config == b->config
1162 && a->curr == b->curr
1163 && a->advertised == b->advertised
1164 && a->supported == b->supported
1165 && a->peer == b->peer);
1169 send_port_status(struct ofproto *p, const struct ofport *ofport,
1172 /* XXX Should limit the number of queued port status change messages. */
1173 struct ofconn *ofconn;
1174 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1175 struct ofp_port_status *ops;
1178 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1179 ops->reason = reason;
1180 ops->desc = ofport->opp;
1181 hton_ofp_phy_port(&ops->desc);
1182 queue_tx(b, ofconn, NULL);
1184 if (p->ofhooks->port_changed_cb) {
1185 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1190 ofport_install(struct ofproto *p, struct ofport *ofport)
1192 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1193 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no),
1195 shash_add(&p->port_by_name, (char *) ofport->opp.name, ofport);
1199 ofport_remove(struct ofproto *p, struct ofport *ofport)
1201 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1202 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no), NULL);
1203 shash_delete(&p->port_by_name,
1204 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1208 ofport_free(struct ofport *ofport)
1211 netdev_close(ofport->netdev);
1217 update_port(struct ofproto *p, const char *devname)
1219 struct odp_port odp_port;
1220 struct ofport *old_ofport;
1221 struct ofport *new_ofport;
1224 COVERAGE_INC(ofproto_update_port);
1226 /* Query the datapath for port information. */
1227 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1229 /* Find the old ofport. */
1230 old_ofport = shash_find_data(&p->port_by_name, devname);
1233 /* There's no port named 'devname' but there might be a port with
1234 * the same port number. This could happen if a port is deleted
1235 * and then a new one added in its place very quickly, or if a port
1236 * is renamed. In the former case we want to send an OFPPR_DELETE
1237 * and an OFPPR_ADD, and in the latter case we want to send a
1238 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1239 * the old port's ifindex against the new port, or perhaps less
1240 * reliably but more portably by comparing the old port's MAC
1241 * against the new port's MAC. However, this code isn't that smart
1242 * and always sends an OFPPR_MODIFY (XXX). */
1243 old_ofport = port_array_get(&p->ports, odp_port.port);
1245 } else if (error != ENOENT && error != ENODEV) {
1246 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1247 "%s", strerror(error));
1251 /* Create a new ofport. */
1252 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1254 /* Eliminate a few pathological cases. */
1255 if (!old_ofport && !new_ofport) {
1257 } else if (old_ofport && new_ofport) {
1258 /* Most of the 'config' bits are OpenFlow soft state, but
1259 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1260 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1261 * leaves the other bits 0.) */
1262 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1264 if (ofport_equal(old_ofport, new_ofport)) {
1265 /* False alarm--no change. */
1266 ofport_free(new_ofport);
1271 /* Now deal with the normal cases. */
1273 ofport_remove(p, old_ofport);
1276 ofport_install(p, new_ofport);
1278 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1279 (!old_ofport ? OFPPR_ADD
1280 : !new_ofport ? OFPPR_DELETE
1282 ofport_free(old_ofport);
1284 /* Update port groups. */
1285 refresh_port_groups(p);
1289 init_ports(struct ofproto *p)
1291 struct odp_port *ports;
1296 error = dpif_port_list(p->dpif, &ports, &n_ports);
1301 for (i = 0; i < n_ports; i++) {
1302 const struct odp_port *odp_port = &ports[i];
1303 if (!ofport_conflicts(p, odp_port)) {
1304 struct ofport *ofport = make_ofport(odp_port);
1306 ofport_install(p, ofport);
1311 refresh_port_groups(p);
1315 static struct ofconn *
1316 ofconn_create(struct ofproto *p, struct rconn *rconn)
1318 struct ofconn *ofconn = xmalloc(sizeof *ofconn);
1319 list_push_back(&p->all_conns, &ofconn->node);
1320 ofconn->rconn = rconn;
1321 ofconn->pktbuf = NULL;
1322 ofconn->send_flow_exp = false;
1323 ofconn->miss_send_len = 0;
1324 ofconn->packet_in_counter = rconn_packet_counter_create ();
1325 ofconn->reply_counter = rconn_packet_counter_create ();
1330 ofconn_destroy(struct ofconn *ofconn, struct ofproto *p)
1333 executer_rconn_closing(p->executer, ofconn->rconn);
1336 list_remove(&ofconn->node);
1337 rconn_destroy(ofconn->rconn);
1338 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1339 rconn_packet_counter_destroy(ofconn->reply_counter);
1340 pktbuf_destroy(ofconn->pktbuf);
1345 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1349 rconn_run(ofconn->rconn);
1351 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1352 /* Limit the number of iterations to prevent other tasks from
1354 for (iteration = 0; iteration < 50; iteration++) {
1355 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1359 handle_openflow(ofconn, p, of_msg);
1360 ofpbuf_delete(of_msg);
1364 if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
1365 ofconn_destroy(ofconn, p);
1370 ofconn_wait(struct ofconn *ofconn)
1372 rconn_run_wait(ofconn->rconn);
1373 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1374 rconn_recv_wait(ofconn->rconn);
1376 COVERAGE_INC(ofproto_ofconn_stuck);
1380 /* Caller is responsible for initializing the 'cr' member of the returned
1382 static struct rule *
1383 rule_create(struct rule *super,
1384 const union ofp_action *actions, size_t n_actions,
1385 uint16_t idle_timeout, uint16_t hard_timeout)
1387 struct rule *rule = xcalloc(1, sizeof *rule);
1388 rule->idle_timeout = idle_timeout;
1389 rule->hard_timeout = hard_timeout;
1390 rule->used = rule->created = time_msec();
1391 rule->super = super;
1393 list_push_back(&super->list, &rule->list);
1395 list_init(&rule->list);
1397 rule->n_actions = n_actions;
1398 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1402 static struct rule *
1403 rule_from_cls_rule(const struct cls_rule *cls_rule)
1405 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1409 rule_free(struct rule *rule)
1411 free(rule->actions);
1412 free(rule->odp_actions);
1416 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1417 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1418 * through all of its subrules and revalidates them, destroying any that no
1419 * longer has a super-rule (which is probably all of them).
1421 * Before calling this function, the caller must make have removed 'rule' from
1422 * the classifier. If 'rule' is an exact-match rule, the caller is also
1423 * responsible for ensuring that it has been uninstalled from the datapath. */
1425 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1428 struct rule *subrule, *next;
1429 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1430 revalidate_rule(ofproto, subrule);
1433 list_remove(&rule->list);
1439 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1441 const union ofp_action *oa;
1442 struct actions_iterator i;
1444 if (out_port == htons(OFPP_NONE)) {
1447 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1448 oa = actions_next(&i)) {
1449 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1456 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1457 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1459 * The flow that 'packet' actually contains does not need to actually match
1460 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1461 * the packet and byte counters for 'rule' will be credited for the packet sent
1462 * out whether or not the packet actually matches 'rule'.
1464 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1465 * the caller must already have accurately composed ODP actions for it given
1466 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1467 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1468 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1469 * actions and apply them to 'packet'. */
1471 rule_execute(struct ofproto *ofproto, struct rule *rule,
1472 struct ofpbuf *packet, const flow_t *flow)
1474 const union odp_action *actions;
1476 struct odp_actions a;
1478 /* Grab or compose the ODP actions.
1480 * The special case for an exact-match 'rule' where 'flow' is not the
1481 * rule's flow is important to avoid, e.g., sending a packet out its input
1482 * port simply because the ODP actions were composed for the wrong
1484 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1485 struct rule *super = rule->super ? rule->super : rule;
1486 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1487 packet, &a, NULL, 0)) {
1490 actions = a.actions;
1491 n_actions = a.n_actions;
1493 actions = rule->odp_actions;
1494 n_actions = rule->n_odp_actions;
1497 /* Execute the ODP actions. */
1498 if (!dpif_execute(ofproto->dpif, flow->in_port,
1499 actions, n_actions, packet)) {
1500 struct odp_flow_stats stats;
1501 flow_extract_stats(flow, packet, &stats);
1502 update_stats(rule, &stats);
1503 rule->used = time_msec();
1508 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1511 struct rule *displaced_rule;
1513 /* Insert the rule in the classifier. */
1514 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1515 if (!rule->cr.wc.wildcards) {
1516 rule_make_actions(p, rule, packet);
1519 /* Send the packet and credit it to the rule. */
1522 flow_extract(packet, in_port, &flow);
1523 rule_execute(p, rule, packet, &flow);
1526 /* Install the rule in the datapath only after sending the packet, to
1527 * avoid packet reordering. */
1528 if (rule->cr.wc.wildcards) {
1529 COVERAGE_INC(ofproto_add_wc_flow);
1530 p->need_revalidate = true;
1532 rule_install(p, rule, displaced_rule);
1535 /* Free the rule that was displaced, if any. */
1536 if (displaced_rule) {
1537 rule_destroy(p, displaced_rule);
1541 static struct rule *
1542 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1545 struct rule *subrule = rule_create(rule, NULL, 0,
1546 rule->idle_timeout, rule->hard_timeout);
1547 COVERAGE_INC(ofproto_subrule_create);
1548 cls_rule_from_flow(&subrule->cr, flow, 0,
1549 (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1550 : rule->cr.priority));
1551 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1557 rule_remove(struct ofproto *ofproto, struct rule *rule)
1559 if (rule->cr.wc.wildcards) {
1560 COVERAGE_INC(ofproto_del_wc_flow);
1561 ofproto->need_revalidate = true;
1563 rule_uninstall(ofproto, rule);
1565 classifier_remove(&ofproto->cls, &rule->cr);
1566 rule_destroy(ofproto, rule);
1569 /* Returns true if the actions changed, false otherwise. */
1571 rule_make_actions(struct ofproto *p, struct rule *rule,
1572 const struct ofpbuf *packet)
1574 const struct rule *super;
1575 struct odp_actions a;
1578 assert(!rule->cr.wc.wildcards);
1580 super = rule->super ? rule->super : rule;
1582 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1583 packet, &a, &rule->tags, &rule->may_install);
1585 actions_len = a.n_actions * sizeof *a.actions;
1586 if (rule->n_odp_actions != a.n_actions
1587 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1588 COVERAGE_INC(ofproto_odp_unchanged);
1589 free(rule->odp_actions);
1590 rule->n_odp_actions = a.n_actions;
1591 rule->odp_actions = xmemdup(a.actions, actions_len);
1599 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1600 struct odp_flow_put *put)
1602 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1603 put->flow.key = rule->cr.flow;
1604 put->flow.actions = rule->odp_actions;
1605 put->flow.n_actions = rule->n_odp_actions;
1607 return dpif_flow_put(ofproto->dpif, put);
1611 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1613 assert(!rule->cr.wc.wildcards);
1615 if (rule->may_install) {
1616 struct odp_flow_put put;
1617 if (!do_put_flow(p, rule,
1618 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1620 rule->installed = true;
1621 if (displaced_rule) {
1622 update_stats(rule, &put.flow.stats);
1623 rule_post_uninstall(p, displaced_rule);
1626 } else if (displaced_rule) {
1627 rule_uninstall(p, displaced_rule);
1632 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1634 if (rule->installed) {
1635 struct odp_flow_put put;
1636 COVERAGE_INC(ofproto_dp_missed);
1637 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1639 rule_install(ofproto, rule, NULL);
1644 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1646 bool actions_changed = rule_make_actions(ofproto, rule, NULL);
1647 if (rule->may_install) {
1648 if (rule->installed) {
1649 if (actions_changed) {
1650 /* XXX should really do rule_post_uninstall() for the *old* set
1651 * of actions, and distinguish the old stats from the new. */
1652 struct odp_flow_put put;
1653 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1656 rule_install(ofproto, rule, NULL);
1659 rule_uninstall(ofproto, rule);
1664 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1666 uint64_t total_bytes = rule->byte_count + extra_bytes;
1668 if (ofproto->ofhooks->account_flow_cb
1669 && total_bytes > rule->accounted_bytes)
1671 ofproto->ofhooks->account_flow_cb(
1672 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1673 total_bytes - rule->accounted_bytes, ofproto->aux);
1674 rule->accounted_bytes = total_bytes;
1679 rule_uninstall(struct ofproto *p, struct rule *rule)
1681 assert(!rule->cr.wc.wildcards);
1682 if (rule->installed) {
1683 struct odp_flow odp_flow;
1685 odp_flow.key = rule->cr.flow;
1686 odp_flow.actions = NULL;
1687 odp_flow.n_actions = 0;
1688 if (!dpif_flow_del(p->dpif, &odp_flow)) {
1689 update_stats(rule, &odp_flow.stats);
1691 rule->installed = false;
1693 rule_post_uninstall(p, rule);
1698 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1700 struct rule *super = rule->super;
1702 rule_account(ofproto, rule, 0);
1703 if (ofproto->netflow) {
1704 struct ofexpired expired;
1705 expired.flow = rule->cr.flow;
1706 expired.packet_count = rule->packet_count;
1707 expired.byte_count = rule->byte_count;
1708 expired.used = rule->used;
1709 expired.created = rule->created;
1710 expired.tcp_flags = rule->tcp_flags;
1711 expired.ip_tos = rule->ip_tos;
1712 netflow_expire(ofproto->netflow, &expired);
1715 super->packet_count += rule->packet_count;
1716 super->byte_count += rule->byte_count;
1717 super->tcp_flags |= rule->tcp_flags;
1718 if (rule->packet_count) {
1719 super->ip_tos = rule->ip_tos;
1723 /* Reset counters to prevent double counting if the rule ever gets
1725 rule->packet_count = 0;
1726 rule->byte_count = 0;
1727 rule->accounted_bytes = 0;
1728 rule->tcp_flags = 0;
1733 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1734 struct rconn_packet_counter *counter)
1736 update_openflow_length(msg);
1737 if (rconn_send(ofconn->rconn, msg, counter)) {
1743 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1744 int error, const void *data, size_t len)
1747 struct ofp_error_msg *oem;
1749 if (!(error >> 16)) {
1750 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1755 COVERAGE_INC(ofproto_error);
1756 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1757 oh ? oh->xid : 0, &buf);
1758 oem->type = htons((unsigned int) error >> 16);
1759 oem->code = htons(error & 0xffff);
1760 memcpy(oem->data, data, len);
1761 queue_tx(buf, ofconn, ofconn->reply_counter);
1765 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1768 size_t oh_length = ntohs(oh->length);
1769 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1773 hton_ofp_phy_port(struct ofp_phy_port *opp)
1775 opp->port_no = htons(opp->port_no);
1776 opp->config = htonl(opp->config);
1777 opp->state = htonl(opp->state);
1778 opp->curr = htonl(opp->curr);
1779 opp->advertised = htonl(opp->advertised);
1780 opp->supported = htonl(opp->supported);
1781 opp->peer = htonl(opp->peer);
1785 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
1787 struct ofp_header *rq = oh;
1788 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
1793 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
1794 struct ofp_header *oh)
1796 struct ofp_switch_features *osf;
1798 unsigned int port_no;
1799 struct ofport *port;
1801 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1802 osf->datapath_id = htonll(p->datapath_id);
1803 osf->n_buffers = htonl(pktbuf_capacity());
1805 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1806 OFPC_PORT_STATS | OFPC_MULTI_PHY_TX);
1807 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1808 (1u << OFPAT_SET_VLAN_VID) |
1809 (1u << OFPAT_SET_VLAN_PCP) |
1810 (1u << OFPAT_STRIP_VLAN) |
1811 (1u << OFPAT_SET_DL_SRC) |
1812 (1u << OFPAT_SET_DL_DST) |
1813 (1u << OFPAT_SET_NW_SRC) |
1814 (1u << OFPAT_SET_NW_DST) |
1815 (1u << OFPAT_SET_TP_SRC) |
1816 (1u << OFPAT_SET_TP_DST));
1818 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1819 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1822 queue_tx(buf, ofconn, ofconn->reply_counter);
1827 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1828 struct ofp_header *oh)
1831 struct ofp_switch_config *osc;
1835 /* Figure out flags. */
1836 dpif_get_drop_frags(p->dpif, &drop_frags);
1837 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1838 if (ofconn->send_flow_exp) {
1839 flags |= OFPC_SEND_FLOW_EXP;
1843 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1844 osc->flags = htons(flags);
1845 osc->miss_send_len = htons(ofconn->miss_send_len);
1846 queue_tx(buf, ofconn, ofconn->reply_counter);
1852 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1853 struct ofp_switch_config *osc)
1858 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1862 flags = ntohs(osc->flags);
1864 ofconn->send_flow_exp = (flags & OFPC_SEND_FLOW_EXP) != 0;
1866 if (ofconn == p->controller) {
1867 switch (flags & OFPC_FRAG_MASK) {
1868 case OFPC_FRAG_NORMAL:
1869 dpif_set_drop_frags(p->dpif, false);
1871 case OFPC_FRAG_DROP:
1872 dpif_set_drop_frags(p->dpif, true);
1875 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1881 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1882 if (ntohs(osc->miss_send_len) != 0) {
1883 ofconn->pktbuf = pktbuf_create();
1885 pktbuf_destroy(ofconn->pktbuf);
1889 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1895 add_output_group_action(struct odp_actions *actions, uint16_t group)
1897 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1901 add_controller_action(struct odp_actions *actions,
1902 const struct ofp_action_output *oao)
1904 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
1905 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
1908 struct action_xlate_ctx {
1910 const flow_t *flow; /* Flow to which these actions correspond. */
1911 int recurse; /* Recursion level, via xlate_table_action. */
1912 struct ofproto *ofproto;
1913 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
1914 * null pointer if we are revalidating
1915 * without a packet to refer to. */
1918 struct odp_actions *out; /* Datapath actions. */
1919 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
1920 bool may_setup_flow; /* True ordinarily; false if the actions must
1921 * be reassessed for every packet. */
1924 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
1925 struct action_xlate_ctx *ctx);
1928 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
1930 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
1931 if (!ofport || !(ofport->opp.config & OFPPC_NO_FWD)) {
1932 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
1936 static struct rule *
1937 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
1940 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
1942 /* The rule we found might not be valid, since we could be in need of
1943 * revalidation. If it is not valid, don't return it. */
1946 && ofproto->need_revalidate
1947 && !revalidate_rule(ofproto, rule)) {
1948 COVERAGE_INC(ofproto_invalidated);
1956 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
1958 if (!ctx->recurse) {
1963 flow.in_port = in_port;
1965 rule = lookup_valid_rule(ctx->ofproto, &flow);
1972 do_xlate_actions(rule->actions, rule->n_actions, ctx);
1979 xlate_output_action(struct action_xlate_ctx *ctx,
1980 const struct ofp_action_output *oao)
1984 switch (ntohs(oao->port)) {
1986 add_output_action(ctx, ctx->flow->in_port);
1989 xlate_table_action(ctx, ctx->flow->in_port);
1992 if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet,
1993 ctx->out, ctx->tags,
1994 ctx->ofproto->aux)) {
1995 COVERAGE_INC(ofproto_uninstallable);
1996 ctx->may_setup_flow = false;
2000 add_output_group_action(ctx->out, DP_GROUP_FLOOD);
2003 add_output_group_action(ctx->out, DP_GROUP_ALL);
2005 case OFPP_CONTROLLER:
2006 add_controller_action(ctx->out, oao);
2009 add_output_action(ctx, ODPP_LOCAL);
2012 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2013 if (odp_port != ctx->flow->in_port) {
2014 add_output_action(ctx, odp_port);
2021 xlate_nicira_action(struct action_xlate_ctx *ctx,
2022 const struct nx_action_header *nah)
2024 const struct nx_action_resubmit *nar;
2025 int subtype = ntohs(nah->subtype);
2027 assert(nah->vendor == htonl(NX_VENDOR_ID));
2029 case NXAST_RESUBMIT:
2030 nar = (const struct nx_action_resubmit *) nah;
2031 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2035 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2041 do_xlate_actions(const union ofp_action *in, size_t n_in,
2042 struct action_xlate_ctx *ctx)
2044 struct actions_iterator iter;
2045 const union ofp_action *ia;
2046 const struct ofport *port;
2048 port = port_array_get(&ctx->ofproto->ports, ctx->flow->in_port);
2049 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2050 port->opp.config & (eth_addr_equals(ctx->flow->dl_dst, stp_eth_addr)
2051 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2052 /* Drop this flow. */
2056 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2057 uint16_t type = ntohs(ia->type);
2058 union odp_action *oa;
2062 xlate_output_action(ctx, &ia->output);
2065 case OFPAT_SET_VLAN_VID:
2066 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2067 oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2070 case OFPAT_SET_VLAN_PCP:
2071 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2072 oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2075 case OFPAT_STRIP_VLAN:
2076 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2079 case OFPAT_SET_DL_SRC:
2080 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2081 memcpy(oa->dl_addr.dl_addr,
2082 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2085 case OFPAT_SET_DL_DST:
2086 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2087 memcpy(oa->dl_addr.dl_addr,
2088 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2091 case OFPAT_SET_NW_SRC:
2092 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2093 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2096 case OFPAT_SET_TP_SRC:
2097 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2098 oa->tp_port.tp_port = ia->tp_port.tp_port;
2102 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2106 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2113 xlate_actions(const union ofp_action *in, size_t n_in,
2114 const flow_t *flow, struct ofproto *ofproto,
2115 const struct ofpbuf *packet,
2116 struct odp_actions *out, tag_type *tags, bool *may_setup_flow)
2118 tag_type no_tags = 0;
2119 struct action_xlate_ctx ctx;
2120 COVERAGE_INC(ofproto_ofp2odp);
2121 odp_actions_init(out);
2124 ctx.ofproto = ofproto;
2125 ctx.packet = packet;
2127 ctx.tags = tags ? tags : &no_tags;
2128 ctx.may_setup_flow = true;
2129 do_xlate_actions(in, n_in, &ctx);
2130 if (may_setup_flow) {
2131 *may_setup_flow = ctx.may_setup_flow;
2133 if (odp_actions_overflow(out)) {
2134 odp_actions_init(out);
2135 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2141 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2142 struct ofp_header *oh)
2144 struct ofp_packet_out *opo;
2145 struct ofpbuf payload, *buffer;
2146 struct odp_actions actions;
2152 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2156 opo = (struct ofp_packet_out *) oh;
2158 COVERAGE_INC(ofproto_packet_out);
2159 if (opo->buffer_id != htonl(UINT32_MAX)) {
2160 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2170 flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2171 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2172 &flow, p, &payload, &actions, NULL, NULL);
2177 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2179 ofpbuf_delete(buffer);
2185 update_port_config(struct ofproto *p, struct ofport *port,
2186 uint32_t config, uint32_t mask)
2188 mask &= config ^ port->opp.config;
2189 if (mask & OFPPC_PORT_DOWN) {
2190 if (config & OFPPC_PORT_DOWN) {
2191 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2193 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2196 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2197 if (mask & REVALIDATE_BITS) {
2198 COVERAGE_INC(ofproto_costly_flags);
2199 port->opp.config ^= mask & REVALIDATE_BITS;
2200 p->need_revalidate = true;
2202 #undef REVALIDATE_BITS
2203 if (mask & OFPPC_NO_FLOOD) {
2204 port->opp.config ^= OFPPC_NO_FLOOD;
2205 refresh_port_group(p, DP_GROUP_FLOOD);
2207 if (mask & OFPPC_NO_PACKET_IN) {
2208 port->opp.config ^= OFPPC_NO_PACKET_IN;
2213 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2215 const struct ofp_port_mod *opm;
2216 struct ofport *port;
2219 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2223 opm = (struct ofp_port_mod *) oh;
2225 port = port_array_get(&p->ports,
2226 ofp_port_to_odp_port(ntohs(opm->port_no)));
2228 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2229 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2230 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2232 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2233 if (opm->advertise) {
2234 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2240 static struct ofpbuf *
2241 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2243 struct ofp_stats_reply *osr;
2246 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2247 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2249 osr->flags = htons(0);
2253 static struct ofpbuf *
2254 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2256 return make_stats_reply(request->header.xid, request->type, body_len);
2260 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2262 struct ofpbuf *msg = *msgp;
2263 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2264 if (nbytes + msg->size > UINT16_MAX) {
2265 struct ofp_stats_reply *reply = msg->data;
2266 reply->flags = htons(OFPSF_REPLY_MORE);
2267 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2268 queue_tx(msg, ofconn, ofconn->reply_counter);
2270 return ofpbuf_put_uninit(*msgp, nbytes);
2274 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2275 struct ofp_stats_request *request)
2277 struct ofp_desc_stats *ods;
2280 msg = start_stats_reply(request, sizeof *ods);
2281 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2282 strncpy(ods->mfr_desc, p->manufacturer, sizeof ods->mfr_desc);
2283 strncpy(ods->hw_desc, p->hardware, sizeof ods->hw_desc);
2284 strncpy(ods->sw_desc, p->software, sizeof ods->sw_desc);
2285 strncpy(ods->serial_num, p->serial, sizeof ods->serial_num);
2286 queue_tx(msg, ofconn, ofconn->reply_counter);
2292 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2294 struct rule *rule = rule_from_cls_rule(cls_rule);
2295 int *n_subrules = n_subrules_;
2303 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2304 struct ofp_stats_request *request)
2306 struct ofp_table_stats *ots;
2308 struct odp_stats dpstats;
2309 int n_exact, n_subrules, n_wild;
2311 msg = start_stats_reply(request, sizeof *ots * 2);
2313 /* Count rules of various kinds. */
2315 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2316 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2317 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2320 dpif_get_dp_stats(p->dpif, &dpstats);
2321 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2322 memset(ots, 0, sizeof *ots);
2323 ots->table_id = TABLEID_HASH;
2324 strcpy(ots->name, "hash");
2325 ots->wildcards = htonl(0);
2326 ots->max_entries = htonl(dpstats.max_capacity);
2327 ots->active_count = htonl(n_exact);
2328 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2330 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2332 /* Classifier table. */
2333 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2334 memset(ots, 0, sizeof *ots);
2335 ots->table_id = TABLEID_CLASSIFIER;
2336 strcpy(ots->name, "classifier");
2337 ots->wildcards = htonl(OFPFW_ALL);
2338 ots->max_entries = htonl(65536);
2339 ots->active_count = htonl(n_wild);
2340 ots->lookup_count = htonll(0); /* XXX */
2341 ots->matched_count = htonll(0); /* XXX */
2343 queue_tx(msg, ofconn, ofconn->reply_counter);
2348 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2349 struct ofp_stats_request *request)
2351 struct ofp_port_stats *ops;
2353 struct ofport *port;
2354 unsigned int port_no;
2356 msg = start_stats_reply(request, sizeof *ops * 16);
2357 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2358 struct netdev_stats stats;
2360 /* Intentionally ignore return value, since errors will set 'stats' to
2361 * all-1s, which is correct for OpenFlow, and netdev_get_stats() will
2363 netdev_get_stats(port->netdev, &stats);
2365 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2366 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2367 memset(ops->pad, 0, sizeof ops->pad);
2368 ops->rx_packets = htonll(stats.rx_packets);
2369 ops->tx_packets = htonll(stats.tx_packets);
2370 ops->rx_bytes = htonll(stats.rx_bytes);
2371 ops->tx_bytes = htonll(stats.tx_bytes);
2372 ops->rx_dropped = htonll(stats.rx_dropped);
2373 ops->tx_dropped = htonll(stats.tx_dropped);
2374 ops->rx_errors = htonll(stats.rx_errors);
2375 ops->tx_errors = htonll(stats.tx_errors);
2376 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2377 ops->rx_over_err = htonll(stats.rx_over_errors);
2378 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2379 ops->collisions = htonll(stats.collisions);
2382 queue_tx(msg, ofconn, ofconn->reply_counter);
2386 struct flow_stats_cbdata {
2387 struct ofproto *ofproto;
2388 struct ofconn *ofconn;
2394 query_stats(struct ofproto *p, struct rule *rule,
2395 uint64_t *packet_countp, uint64_t *byte_countp)
2397 uint64_t packet_count, byte_count;
2398 struct rule *subrule;
2399 struct odp_flow *odp_flows;
2402 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2403 odp_flows = xcalloc(1, n_odp_flows * sizeof *odp_flows);
2404 if (rule->cr.wc.wildcards) {
2406 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2407 odp_flows[i++].key = subrule->cr.flow;
2410 odp_flows[0].key = rule->cr.flow;
2413 packet_count = rule->packet_count;
2414 byte_count = rule->byte_count;
2415 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2417 for (i = 0; i < n_odp_flows; i++) {
2418 struct odp_flow *odp_flow = &odp_flows[i];
2419 packet_count += odp_flow->stats.n_packets;
2420 byte_count += odp_flow->stats.n_bytes;
2425 *packet_countp = packet_count;
2426 *byte_countp = byte_count;
2430 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2432 struct rule *rule = rule_from_cls_rule(rule_);
2433 struct flow_stats_cbdata *cbdata = cbdata_;
2434 struct ofp_flow_stats *ofs;
2435 uint64_t packet_count, byte_count;
2436 size_t act_len, len;
2438 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2442 act_len = sizeof *rule->actions * rule->n_actions;
2443 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2445 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2447 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2448 ofs->length = htons(len);
2449 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2451 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match);
2452 ofs->duration = htonl((time_msec() - rule->created) / 1000);
2453 ofs->priority = htons(rule->cr.priority);
2454 ofs->idle_timeout = htons(rule->idle_timeout);
2455 ofs->hard_timeout = htons(rule->hard_timeout);
2456 memset(ofs->pad2, 0, sizeof ofs->pad2);
2457 ofs->packet_count = htonll(packet_count);
2458 ofs->byte_count = htonll(byte_count);
2459 memcpy(ofs->actions, rule->actions, act_len);
2463 table_id_to_include(uint8_t table_id)
2465 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2466 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2467 : table_id == 0xff ? CLS_INC_ALL
2472 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2473 const struct ofp_stats_request *osr,
2476 struct ofp_flow_stats_request *fsr;
2477 struct flow_stats_cbdata cbdata;
2478 struct cls_rule target;
2480 if (arg_size != sizeof *fsr) {
2481 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2483 fsr = (struct ofp_flow_stats_request *) osr->body;
2485 COVERAGE_INC(ofproto_flows_req);
2487 cbdata.ofconn = ofconn;
2488 cbdata.out_port = fsr->out_port;
2489 cbdata.msg = start_stats_reply(osr, 1024);
2490 cls_rule_from_match(&target, &fsr->match, 0);
2491 classifier_for_each_match(&p->cls, &target,
2492 table_id_to_include(fsr->table_id),
2493 flow_stats_cb, &cbdata);
2494 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2498 struct flow_stats_ds_cbdata {
2499 struct ofproto *ofproto;
2504 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
2506 struct rule *rule = rule_from_cls_rule(rule_);
2507 struct flow_stats_ds_cbdata *cbdata = cbdata_;
2508 struct ds *results = cbdata->results;
2509 struct ofp_match match;
2510 uint64_t packet_count, byte_count;
2511 size_t act_len = sizeof *rule->actions * rule->n_actions;
2513 /* Don't report on subrules. */
2514 if (rule->super != NULL) {
2518 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2519 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &match);
2521 ds_put_format(results, "duration=%llds, ",
2522 (time_msec() - rule->created) / 1000);
2523 ds_put_format(results, "priority=%u", rule->cr.priority);
2524 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2525 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2526 ofp_print_match(results, &match, true);
2527 ofp_print_actions(results, &rule->actions->header, act_len);
2528 ds_put_cstr(results, "\n");
2531 /* Adds a pretty-printed description of all flows to 'results', including
2532 * those marked hidden by secchan (e.g., by in-band control). */
2534 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2536 struct ofp_match match;
2537 struct cls_rule target;
2538 struct flow_stats_ds_cbdata cbdata;
2540 memset(&match, 0, sizeof match);
2541 match.wildcards = htonl(OFPFW_ALL);
2544 cbdata.results = results;
2546 cls_rule_from_match(&target, &match, 0);
2547 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2548 flow_stats_ds_cb, &cbdata);
2551 struct aggregate_stats_cbdata {
2552 struct ofproto *ofproto;
2554 uint64_t packet_count;
2555 uint64_t byte_count;
2560 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2562 struct rule *rule = rule_from_cls_rule(rule_);
2563 struct aggregate_stats_cbdata *cbdata = cbdata_;
2564 uint64_t packet_count, byte_count;
2566 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2570 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2572 cbdata->packet_count += packet_count;
2573 cbdata->byte_count += byte_count;
2578 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2579 const struct ofp_stats_request *osr,
2582 struct ofp_aggregate_stats_request *asr;
2583 struct ofp_aggregate_stats_reply *reply;
2584 struct aggregate_stats_cbdata cbdata;
2585 struct cls_rule target;
2588 if (arg_size != sizeof *asr) {
2589 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2591 asr = (struct ofp_aggregate_stats_request *) osr->body;
2593 COVERAGE_INC(ofproto_agg_request);
2595 cbdata.out_port = asr->out_port;
2596 cbdata.packet_count = 0;
2597 cbdata.byte_count = 0;
2599 cls_rule_from_match(&target, &asr->match, 0);
2600 classifier_for_each_match(&p->cls, &target,
2601 table_id_to_include(asr->table_id),
2602 aggregate_stats_cb, &cbdata);
2604 msg = start_stats_reply(osr, sizeof *reply);
2605 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2606 reply->flow_count = htonl(cbdata.n_flows);
2607 reply->packet_count = htonll(cbdata.packet_count);
2608 reply->byte_count = htonll(cbdata.byte_count);
2609 queue_tx(msg, ofconn, ofconn->reply_counter);
2614 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2615 struct ofp_header *oh)
2617 struct ofp_stats_request *osr;
2621 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2626 osr = (struct ofp_stats_request *) oh;
2628 switch (ntohs(osr->type)) {
2630 return handle_desc_stats_request(p, ofconn, osr);
2633 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2635 case OFPST_AGGREGATE:
2636 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2639 return handle_table_stats_request(p, ofconn, osr);
2642 return handle_port_stats_request(p, ofconn, osr);
2645 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2648 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2652 static long long int
2653 msec_from_nsec(uint64_t sec, uint32_t nsec)
2655 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2659 update_time(struct rule *rule, const struct odp_flow_stats *stats)
2661 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2662 if (used > rule->used) {
2668 update_stats(struct rule *rule, const struct odp_flow_stats *stats)
2670 update_time(rule, stats);
2671 rule->packet_count += stats->n_packets;
2672 rule->byte_count += stats->n_bytes;
2673 rule->tcp_flags |= stats->tcp_flags;
2674 if (stats->n_packets) {
2675 rule->ip_tos = stats->ip_tos;
2680 add_flow(struct ofproto *p, struct ofconn *ofconn,
2681 struct ofp_flow_mod *ofm, size_t n_actions)
2683 struct ofpbuf *packet;
2688 rule = rule_create(NULL, (const union ofp_action *) ofm->actions,
2689 n_actions, ntohs(ofm->idle_timeout),
2690 ntohs(ofm->hard_timeout));
2691 cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority));
2695 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2696 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2700 rule_insert(p, rule, packet, in_port);
2701 ofpbuf_delete(packet);
2706 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
2707 size_t n_actions, uint16_t command, struct rule *rule)
2709 if (rule_is_hidden(rule)) {
2713 if (command == OFPFC_DELETE) {
2714 rule_remove(p, rule);
2716 size_t actions_len = n_actions * sizeof *rule->actions;
2718 if (n_actions == rule->n_actions
2719 && !memcmp(ofm->actions, rule->actions, actions_len))
2724 free(rule->actions);
2725 rule->actions = xmemdup(ofm->actions, actions_len);
2726 rule->n_actions = n_actions;
2728 if (rule->cr.wc.wildcards) {
2729 COVERAGE_INC(ofproto_mod_wc_flow);
2730 p->need_revalidate = true;
2732 rule_update_actions(p, rule);
2740 modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm,
2741 size_t n_actions, uint16_t command)
2747 flow_from_match(&flow, &wildcards, &ofm->match);
2748 rule = rule_from_cls_rule(classifier_find_rule_exactly(
2749 &p->cls, &flow, wildcards,
2750 ntohs(ofm->priority)));
2753 if (command == OFPFC_DELETE
2754 && ofm->out_port != htons(OFPP_NONE)
2755 && !rule_has_out_port(rule, ofm->out_port)) {
2759 modify_flow(p, ofm, n_actions, command, rule);
2764 struct modify_flows_cbdata {
2765 struct ofproto *ofproto;
2766 const struct ofp_flow_mod *ofm;
2773 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
2775 struct rule *rule = rule_from_cls_rule(rule_);
2776 struct modify_flows_cbdata *cbdata = cbdata_;
2778 if (cbdata->out_port != htons(OFPP_NONE)
2779 && !rule_has_out_port(rule, cbdata->out_port)) {
2783 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions,
2784 cbdata->command, rule);
2788 modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm,
2789 size_t n_actions, uint16_t command)
2791 struct modify_flows_cbdata cbdata;
2792 struct cls_rule target;
2796 cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port
2797 : htons(OFPP_NONE));
2798 cbdata.n_actions = n_actions;
2799 cbdata.command = command;
2801 cls_rule_from_match(&target, &ofm->match, 0);
2803 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2804 modify_flows_cb, &cbdata);
2809 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
2810 struct ofp_flow_mod *ofm)
2815 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
2816 sizeof *ofm->actions, &n_actions);
2821 normalize_match(&ofm->match);
2822 if (!ofm->match.wildcards) {
2823 ofm->priority = htons(UINT16_MAX);
2826 error = validate_actions((const union ofp_action *) ofm->actions,
2827 n_actions, p->max_ports);
2832 switch (ntohs(ofm->command)) {
2834 return add_flow(p, ofconn, ofm, n_actions);
2837 return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY);
2839 case OFPFC_MODIFY_STRICT:
2840 return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY);
2843 return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE);
2845 case OFPFC_DELETE_STRICT:
2846 return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE);
2849 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
2854 send_capability_reply(struct ofproto *p, struct ofconn *ofconn, uint32_t xid)
2856 struct ofmp_capability_reply *ocr;
2858 char capabilities[] = "com.nicira.mgmt.manager=false\n";
2860 ocr = make_openflow_xid(sizeof(*ocr), OFPT_VENDOR, xid, &b);
2861 ocr->header.header.vendor = htonl(NX_VENDOR_ID);
2862 ocr->header.header.subtype = htonl(NXT_MGMT);
2863 ocr->header.type = htons(OFMPT_CAPABILITY_REPLY);
2865 ocr->format = htonl(OFMPCOF_SIMPLE);
2866 ocr->mgmt_id = htonll(p->mgmt_id);
2868 ofpbuf_put(b, capabilities, strlen(capabilities));
2870 queue_tx(b, ofconn, ofconn->reply_counter);
2874 handle_ofmp(struct ofproto *p, struct ofconn *ofconn,
2875 struct ofmp_header *ofmph)
2877 size_t msg_len = ntohs(ofmph->header.header.length);
2878 if (msg_len < sizeof(*ofmph)) {
2879 VLOG_WARN_RL(&rl, "dropping short managment message: %d\n", msg_len);
2880 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2883 if (ofmph->type == htons(OFMPT_CAPABILITY_REQUEST)) {
2884 struct ofmp_capability_request *ofmpcr;
2886 if (msg_len < sizeof(struct ofmp_capability_request)) {
2887 VLOG_WARN_RL(&rl, "dropping short capability request: %d\n",
2889 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2892 ofmpcr = (struct ofmp_capability_request *)ofmph;
2893 if (ofmpcr->format != htonl(OFMPCAF_SIMPLE)) {
2894 /* xxx Find a better type than bad subtype */
2895 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2898 send_capability_reply(p, ofconn, ofmph->header.header.xid);
2901 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2906 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
2908 struct ofp_vendor_header *ovh = msg;
2909 struct nicira_header *nh;
2911 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
2912 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2914 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
2915 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2917 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
2918 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2922 switch (ntohl(nh->subtype)) {
2923 case NXT_STATUS_REQUEST:
2924 return switch_status_handle_request(p->switch_status, ofconn->rconn,
2927 case NXT_ACT_SET_CONFIG:
2928 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
2930 case NXT_ACT_GET_CONFIG:
2931 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
2933 case NXT_COMMAND_REQUEST:
2935 return executer_handle_request(p->executer, ofconn->rconn, msg);
2940 return handle_ofmp(p, ofconn, msg);
2943 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2947 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
2948 struct ofpbuf *ofp_msg)
2950 struct ofp_header *oh = ofp_msg->data;
2953 COVERAGE_INC(ofproto_recv_openflow);
2955 case OFPT_ECHO_REQUEST:
2956 error = handle_echo_request(ofconn, oh);
2959 case OFPT_ECHO_REPLY:
2963 case OFPT_FEATURES_REQUEST:
2964 error = handle_features_request(p, ofconn, oh);
2967 case OFPT_GET_CONFIG_REQUEST:
2968 error = handle_get_config_request(p, ofconn, oh);
2971 case OFPT_SET_CONFIG:
2972 error = handle_set_config(p, ofconn, ofp_msg->data);
2975 case OFPT_PACKET_OUT:
2976 error = handle_packet_out(p, ofconn, ofp_msg->data);
2980 error = handle_port_mod(p, oh);
2984 error = handle_flow_mod(p, ofconn, ofp_msg->data);
2987 case OFPT_STATS_REQUEST:
2988 error = handle_stats_request(p, ofconn, oh);
2992 error = handle_vendor(p, ofconn, ofp_msg->data);
2996 if (VLOG_IS_WARN_ENABLED()) {
2997 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
2998 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3001 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3006 send_error_oh(ofconn, ofp_msg->data, error);
3011 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3013 struct odp_msg *msg = packet->data;
3014 uint16_t in_port = odp_port_to_ofp_port(msg->port);
3016 struct ofpbuf payload;
3019 /* Handle controller actions. */
3020 if (msg->type == _ODPL_ACTION_NR) {
3021 COVERAGE_INC(ofproto_ctlr_action);
3022 pinsched_send(p->action_sched, in_port, packet,
3023 send_packet_in_action, p);
3027 payload.data = msg + 1;
3028 payload.size = msg->length - sizeof *msg;
3029 flow_extract(&payload, msg->port, &flow);
3031 rule = lookup_valid_rule(p, &flow);
3033 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3034 struct ofport *port = port_array_get(&p->ports, msg->port);
3036 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3037 COVERAGE_INC(ofproto_no_packet_in);
3038 /* XXX install 'drop' flow entry */
3039 ofpbuf_delete(packet);
3043 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3046 COVERAGE_INC(ofproto_packet_in);
3047 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3051 if (rule->cr.wc.wildcards) {
3052 rule = rule_create_subrule(p, rule, &flow);
3053 rule_make_actions(p, rule, packet);
3055 if (!rule->may_install) {
3056 /* The rule is not installable, that is, we need to process every
3057 * packet, so process the current packet and set its actions into
3059 rule_make_actions(p, rule, packet);
3061 /* XXX revalidate rule if it needs it */
3065 rule_execute(p, rule, &payload, &flow);
3066 rule_reinstall(p, rule);
3067 ofpbuf_delete(packet);
3071 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3073 struct rule *sub = rule_from_cls_rule(sub_);
3074 struct revalidate_cbdata *cbdata = cbdata_;
3076 if (cbdata->revalidate_all
3077 || (cbdata->revalidate_subrules && sub->super)
3078 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3079 revalidate_rule(cbdata->ofproto, sub);
3084 revalidate_rule(struct ofproto *p, struct rule *rule)
3086 const flow_t *flow = &rule->cr.flow;
3088 COVERAGE_INC(ofproto_revalidate_rule);
3091 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3093 rule_remove(p, rule);
3095 } else if (super != rule->super) {
3096 COVERAGE_INC(ofproto_revalidate_moved);
3097 list_remove(&rule->list);
3098 list_push_back(&super->list, &rule->list);
3099 rule->super = super;
3100 rule->hard_timeout = super->hard_timeout;
3101 rule->idle_timeout = super->idle_timeout;
3102 rule->created = super->created;
3107 rule_update_actions(p, rule);
3111 static struct ofpbuf *
3112 compose_flow_exp(const struct rule *rule, long long int now, uint8_t reason)
3114 struct ofp_flow_expired *ofe;
3117 ofe = make_openflow(sizeof *ofe, OFPT_FLOW_EXPIRED, &buf);
3118 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofe->match);
3119 ofe->priority = htons(rule->cr.priority);
3120 ofe->reason = reason;
3121 ofe->duration = (now - rule->created) / 1000;
3122 ofe->packet_count = rule->packet_count;
3123 ofe->byte_count = rule->byte_count;
3129 send_flow_exp(struct ofproto *p, struct rule *rule,
3130 long long int now, uint8_t reason)
3132 struct ofconn *ofconn;
3133 struct ofconn *prev;
3134 struct ofpbuf *buf = NULL;
3136 /* We limit the maximum number of queued flow expirations it by accounting
3137 * them under the counter for replies. That works because preventing
3138 * OpenFlow requests from being processed also prevents new flows from
3139 * being added (and expiring). (It also prevents processing OpenFlow
3140 * requests that would not add new flows, so it is imperfect.) */
3143 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3144 if (ofconn->send_flow_exp && rconn_is_connected(ofconn->rconn)) {
3146 queue_tx(ofpbuf_clone(buf), prev, ofconn->reply_counter);
3148 buf = compose_flow_exp(rule, now, reason);
3154 queue_tx(buf, prev, ofconn->reply_counter);
3159 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3161 assert(rule->installed);
3162 assert(!rule->cr.wc.wildcards);
3165 rule_remove(ofproto, rule);
3167 rule_uninstall(ofproto, rule);
3172 expire_rule(struct cls_rule *cls_rule, void *p_)
3174 struct ofproto *p = p_;
3175 struct rule *rule = rule_from_cls_rule(cls_rule);
3176 long long int hard_expire, idle_expire, expire, now;
3178 hard_expire = (rule->hard_timeout
3179 ? rule->created + rule->hard_timeout * 1000
3181 idle_expire = (rule->idle_timeout
3182 && (rule->super || list_is_empty(&rule->list))
3183 ? rule->used + rule->idle_timeout * 1000
3185 expire = MIN(hard_expire, idle_expire);
3186 if (expire == LLONG_MAX) {
3187 if (rule->installed && time_msec() >= rule->used + 5000) {
3188 uninstall_idle_flow(p, rule);
3195 if (rule->installed && now >= rule->used + 5000) {
3196 uninstall_idle_flow(p, rule);
3201 COVERAGE_INC(ofproto_expired);
3202 if (rule->cr.wc.wildcards) {
3203 /* Update stats. (This code will be a no-op if the rule expired
3204 * due to an idle timeout, because in that case the rule has no
3205 * subrules left.) */
3206 struct rule *subrule, *next;
3207 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3208 rule_remove(p, subrule);
3212 send_flow_exp(p, rule, now,
3214 ? OFPER_HARD_TIMEOUT : OFPER_IDLE_TIMEOUT));
3215 rule_remove(p, rule);
3219 update_used(struct ofproto *p)
3221 struct odp_flow *flows;
3226 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
3231 for (i = 0; i < n_flows; i++) {
3232 struct odp_flow *f = &flows[i];
3235 rule = rule_from_cls_rule(
3236 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3237 if (!rule || !rule->installed) {
3238 COVERAGE_INC(ofproto_unexpected_rule);
3239 dpif_flow_del(p->dpif, f);
3243 update_time(rule, &f->stats);
3244 rule_account(p, rule, f->stats.n_bytes);
3250 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3251 const struct ofpbuf *packet, int send_len)
3253 struct ofp_packet_in *opi;
3254 struct ofpbuf payload, *buf;
3255 struct odp_msg *msg;
3258 payload.data = msg + 1;
3259 payload.size = msg->length - sizeof *msg;
3261 send_len = MIN(send_len, payload.size);
3262 buf = ofpbuf_new(sizeof *opi + send_len);
3263 opi = put_openflow_xid(offsetof(struct ofp_packet_in, data),
3264 OFPT_PACKET_IN, 0, buf);
3265 opi->buffer_id = htonl(buffer_id);
3266 opi->total_len = htons(payload.size);
3267 opi->in_port = htons(odp_port_to_ofp_port(msg->port));
3268 opi->reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3269 ofpbuf_put(buf, payload.data, MIN(send_len, payload.size));
3270 update_openflow_length(buf);
3271 rconn_send_with_limit(ofconn->rconn, buf, ofconn->packet_in_counter, 100);
3275 send_packet_in_action(struct ofpbuf *packet, void *p_)
3277 struct ofproto *p = p_;
3278 struct ofconn *ofconn;
3279 struct odp_msg *msg;
3282 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3283 if (ofconn == p->controller || ofconn->miss_send_len) {
3284 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3287 ofpbuf_delete(packet);
3291 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3293 struct ofproto *p = p_;
3294 struct ofconn *ofconn;
3295 struct ofpbuf payload;
3296 struct odp_msg *msg;
3299 payload.data = msg + 1;
3300 payload.size = msg->length - sizeof *msg;
3301 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3302 if (ofconn->miss_send_len) {
3303 uint32_t buffer_id = pktbuf_save(ofconn->pktbuf, &payload,
3305 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3307 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3310 ofpbuf_delete(packet);
3314 pick_datapath_id(const struct ofproto *ofproto)
3316 const struct ofport *port;
3318 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
3320 uint8_t ea[ETH_ADDR_LEN];
3323 error = netdev_get_etheraddr(port->netdev, ea);
3325 return eth_addr_to_uint64(ea);
3327 VLOG_WARN("could not get MAC address for %s (%s)",
3328 netdev_get_name(port->netdev), strerror(error));
3330 return ofproto->fallback_dpid;
3334 pick_fallback_dpid(void)
3336 uint8_t ea[ETH_ADDR_LEN];
3337 eth_addr_random(ea);
3338 ea[0] = 0x00; /* Set Nicira OUI. */
3341 return eth_addr_to_uint64(ea);
3345 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3346 struct odp_actions *actions, tag_type *tags,
3349 struct ofproto *ofproto = ofproto_;
3352 /* Drop frames for reserved multicast addresses. */
3353 if (eth_addr_is_reserved(flow->dl_dst)) {
3357 /* Learn source MAC (but don't try to learn from revalidation). */
3358 if (packet != NULL) {
3359 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3362 /* The log messages here could actually be useful in debugging,
3363 * so keep the rate limit relatively high. */
3364 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3365 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3366 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3367 ofproto_revalidate(ofproto, rev_tag);
3371 /* Determine output port. */
3372 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3374 add_output_group_action(actions, DP_GROUP_FLOOD);
3375 } else if (out_port != flow->in_port) {
3376 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3384 static const struct ofhooks default_ofhooks = {
3386 default_normal_ofhook_cb,