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 = xzalloc(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 return in_band_create(p, p->dpif, p->switch_status,
428 p->controller->rconn, &p->in_band);
430 ofproto_set_discovery(p, false, NULL, true);
431 in_band_destroy(p->in_band);
434 rconn_reconnect(p->controller->rconn);
440 ofproto_set_discovery(struct ofproto *p, bool discovery,
441 const char *re, bool update_resolv_conf)
443 if (discovery != (p->discovery != NULL)) {
445 int error = ofproto_set_in_band(p, true);
449 error = discovery_create(re, update_resolv_conf,
450 p->dpif, p->switch_status,
456 discovery_destroy(p->discovery);
459 rconn_disconnect(p->controller->rconn);
460 } else if (discovery) {
461 discovery_set_update_resolv_conf(p->discovery, update_resolv_conf);
462 return discovery_set_accept_controller_re(p->discovery, re);
468 ofproto_set_controller(struct ofproto *ofproto, const char *controller)
470 if (ofproto->discovery) {
472 } else if (controller) {
473 if (strcmp(rconn_get_name(ofproto->controller->rconn), controller)) {
474 return rconn_connect(ofproto->controller->rconn, controller);
479 rconn_disconnect(ofproto->controller->rconn);
485 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
486 const struct svec *svec)
488 struct pvconn **pvconns = *pvconnsp;
489 size_t n_pvconns = *n_pvconnsp;
493 for (i = 0; i < n_pvconns; i++) {
494 pvconn_close(pvconns[i]);
498 pvconns = xmalloc(svec->n * sizeof *pvconns);
500 for (i = 0; i < svec->n; i++) {
501 const char *name = svec->names[i];
502 struct pvconn *pvconn;
505 error = pvconn_open(name, &pvconn);
507 pvconns[n_pvconns++] = pvconn;
509 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
517 *n_pvconnsp = n_pvconns;
523 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
525 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
529 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
531 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
535 ofproto_set_netflow(struct ofproto *ofproto, const struct svec *collectors,
536 uint8_t engine_type, uint8_t engine_id, bool add_id_to_iface)
538 if (collectors && collectors->n) {
539 if (!ofproto->netflow) {
540 ofproto->netflow = netflow_create();
542 netflow_set_engine(ofproto->netflow, engine_type, engine_id,
544 return netflow_set_collectors(ofproto->netflow, collectors);
546 netflow_destroy(ofproto->netflow);
547 ofproto->netflow = NULL;
553 ofproto_set_failure(struct ofproto *ofproto, bool fail_open)
556 struct rconn *rconn = ofproto->controller->rconn;
557 int trigger_duration = rconn_get_probe_interval(rconn) * 3;
558 if (!ofproto->fail_open) {
559 ofproto->fail_open = fail_open_create(ofproto, trigger_duration,
560 ofproto->switch_status,
563 fail_open_set_trigger_duration(ofproto->fail_open,
567 fail_open_destroy(ofproto->fail_open);
568 ofproto->fail_open = NULL;
573 ofproto_set_rate_limit(struct ofproto *ofproto,
574 int rate_limit, int burst_limit)
576 if (rate_limit > 0) {
577 if (!ofproto->miss_sched) {
578 ofproto->miss_sched = pinsched_create(rate_limit, burst_limit,
579 ofproto->switch_status);
580 ofproto->action_sched = pinsched_create(rate_limit, burst_limit,
583 pinsched_set_limits(ofproto->miss_sched, rate_limit, burst_limit);
584 pinsched_set_limits(ofproto->action_sched,
585 rate_limit, burst_limit);
588 pinsched_destroy(ofproto->miss_sched);
589 ofproto->miss_sched = NULL;
590 pinsched_destroy(ofproto->action_sched);
591 ofproto->action_sched = NULL;
596 ofproto_set_stp(struct ofproto *ofproto UNUSED, bool enable_stp)
600 VLOG_WARN("STP is not yet implemented");
608 ofproto_set_remote_execution(struct ofproto *ofproto, const char *command_acl,
609 const char *command_dir)
612 if (!ofproto->executer) {
613 return executer_create(command_acl, command_dir,
616 executer_set_acl(ofproto->executer, command_acl, command_dir);
619 executer_destroy(ofproto->executer);
620 ofproto->executer = NULL;
626 ofproto_get_datapath_id(const struct ofproto *ofproto)
628 return ofproto->datapath_id;
632 ofproto_get_mgmt_id(const struct ofproto *ofproto)
634 return ofproto->mgmt_id;
638 ofproto_get_probe_interval(const struct ofproto *ofproto)
640 return rconn_get_probe_interval(ofproto->controller->rconn);
644 ofproto_get_max_backoff(const struct ofproto *ofproto)
646 return rconn_get_max_backoff(ofproto->controller->rconn);
650 ofproto_get_in_band(const struct ofproto *ofproto)
652 return ofproto->in_band != NULL;
656 ofproto_get_discovery(const struct ofproto *ofproto)
658 return ofproto->discovery != NULL;
662 ofproto_get_controller(const struct ofproto *ofproto)
664 return rconn_get_name(ofproto->controller->rconn);
668 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
672 for (i = 0; i < ofproto->n_listeners; i++) {
673 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
678 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
682 for (i = 0; i < ofproto->n_snoops; i++) {
683 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
688 ofproto_destroy(struct ofproto *p)
690 struct ofconn *ofconn, *next_ofconn;
691 struct ofport *ofport;
692 unsigned int port_no;
699 ofproto_flush_flows(p);
700 classifier_destroy(&p->cls);
702 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
704 ofconn_destroy(ofconn, p);
708 netdev_monitor_destroy(p->netdev_monitor);
709 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
712 shash_destroy(&p->port_by_name);
714 switch_status_destroy(p->switch_status);
715 in_band_destroy(p->in_band);
716 discovery_destroy(p->discovery);
717 fail_open_destroy(p->fail_open);
718 pinsched_destroy(p->miss_sched);
719 pinsched_destroy(p->action_sched);
720 executer_destroy(p->executer);
721 netflow_destroy(p->netflow);
723 switch_status_unregister(p->ss_cat);
725 for (i = 0; i < p->n_listeners; i++) {
726 pvconn_close(p->listeners[i]);
730 for (i = 0; i < p->n_snoops; i++) {
731 pvconn_close(p->snoops[i]);
735 mac_learning_destroy(p->ml);
741 ofproto_run(struct ofproto *p)
743 int error = ofproto_run1(p);
745 error = ofproto_run2(p, false);
751 process_port_change(struct ofproto *ofproto, int error, char *devname)
753 if (error == ENOBUFS) {
754 reinit_ports(ofproto);
756 update_port(ofproto, devname);
762 ofproto_run1(struct ofproto *p)
764 struct ofconn *ofconn, *next_ofconn;
769 for (i = 0; i < 50; i++) {
773 error = dpif_recv(p->dpif, &buf);
775 if (error == ENODEV) {
776 /* Someone destroyed the datapath behind our back. The caller
777 * better destroy us and give up, because we're just going to
778 * spin from here on out. */
779 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
780 VLOG_ERR_RL(&rl, "%s: datapath was destroyed externally",
787 handle_odp_msg(p, buf);
790 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
791 process_port_change(p, error, devname);
793 while ((error = netdev_monitor_poll(p->netdev_monitor,
794 &devname)) != EAGAIN) {
795 process_port_change(p, error, devname);
799 in_band_run(p->in_band);
802 char *controller_name;
803 if (rconn_is_connectivity_questionable(p->controller->rconn)) {
804 discovery_question_connectivity(p->discovery);
806 if (discovery_run(p->discovery, &controller_name)) {
807 if (controller_name) {
808 rconn_connect(p->controller->rconn, controller_name);
810 rconn_disconnect(p->controller->rconn);
814 pinsched_run(p->miss_sched, send_packet_in_miss, p);
815 pinsched_run(p->action_sched, send_packet_in_action, p);
817 executer_run(p->executer);
820 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
822 ofconn_run(ofconn, p);
825 /* Fail-open maintenance. Do this after processing the ofconns since
826 * fail-open checks the status of the controller rconn. */
828 fail_open_run(p->fail_open);
831 for (i = 0; i < p->n_listeners; i++) {
835 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
837 ofconn_create(p, rconn_new_from_vconn("passive", vconn));
838 } else if (retval != EAGAIN) {
839 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
843 for (i = 0; i < p->n_snoops; i++) {
847 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
849 rconn_add_monitor(p->controller->rconn, vconn);
850 } else if (retval != EAGAIN) {
851 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
855 if (time_msec() >= p->next_expiration) {
856 COVERAGE_INC(ofproto_expiration);
857 p->next_expiration = time_msec() + 1000;
860 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
862 /* Let the hook know that we're at a stable point: all outstanding data
863 * in existing flows has been accounted to the account_cb. Thus, the
864 * hook can now reasonably do operations that depend on having accurate
865 * flow volume accounting (currently, that's just bond rebalancing). */
866 if (p->ofhooks->account_checkpoint_cb) {
867 p->ofhooks->account_checkpoint_cb(p->aux);
872 netflow_run(p->netflow);
878 struct revalidate_cbdata {
879 struct ofproto *ofproto;
880 bool revalidate_all; /* Revalidate all exact-match rules? */
881 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
882 struct tag_set revalidate_set; /* Set of tags to revalidate. */
886 ofproto_run2(struct ofproto *p, bool revalidate_all)
888 if (p->need_revalidate || revalidate_all
889 || !tag_set_is_empty(&p->revalidate_set)) {
890 struct revalidate_cbdata cbdata;
892 cbdata.revalidate_all = revalidate_all;
893 cbdata.revalidate_subrules = p->need_revalidate;
894 cbdata.revalidate_set = p->revalidate_set;
895 tag_set_init(&p->revalidate_set);
896 COVERAGE_INC(ofproto_revalidate);
897 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
898 p->need_revalidate = false;
905 ofproto_wait(struct ofproto *p)
907 struct ofconn *ofconn;
910 dpif_recv_wait(p->dpif);
911 dpif_port_poll_wait(p->dpif);
912 netdev_monitor_poll_wait(p->netdev_monitor);
913 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
917 in_band_wait(p->in_band);
920 discovery_wait(p->discovery);
923 fail_open_wait(p->fail_open);
925 pinsched_wait(p->miss_sched);
926 pinsched_wait(p->action_sched);
928 executer_wait(p->executer);
930 if (!tag_set_is_empty(&p->revalidate_set)) {
931 poll_immediate_wake();
933 if (p->need_revalidate) {
934 /* Shouldn't happen, but if it does just go around again. */
935 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
936 poll_immediate_wake();
937 } else if (p->next_expiration != LLONG_MAX) {
938 poll_timer_wait(p->next_expiration - time_msec());
940 for (i = 0; i < p->n_listeners; i++) {
941 pvconn_wait(p->listeners[i]);
943 for (i = 0; i < p->n_snoops; i++) {
944 pvconn_wait(p->snoops[i]);
949 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
951 tag_set_add(&ofproto->revalidate_set, tag);
955 ofproto_get_revalidate_set(struct ofproto *ofproto)
957 return &ofproto->revalidate_set;
961 ofproto_is_alive(const struct ofproto *p)
963 return p->discovery || rconn_is_alive(p->controller->rconn);
967 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
968 const union ofp_action *actions, size_t n_actions,
969 const struct ofpbuf *packet)
971 struct odp_actions odp_actions;
974 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
980 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
982 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
983 odp_actions.n_actions, packet);
988 ofproto_add_flow(struct ofproto *p,
989 const flow_t *flow, uint32_t wildcards, unsigned int priority,
990 const union ofp_action *actions, size_t n_actions,
994 rule = rule_create(NULL, actions, n_actions,
995 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */, 0);
996 cls_rule_from_flow(&rule->cr, flow, wildcards, priority);
997 rule_insert(p, rule, NULL, 0);
1001 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1002 uint32_t wildcards, unsigned int priority)
1006 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1010 rule_remove(ofproto, rule);
1015 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1017 struct rule *rule = rule_from_cls_rule(rule_);
1018 struct ofproto *ofproto = ofproto_;
1020 /* Mark the flow as not installed, even though it might really be
1021 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1022 * There is no point in uninstalling it individually since we are about to
1023 * blow away all the flows with dpif_flow_flush(). */
1024 rule->installed = false;
1026 rule_remove(ofproto, rule);
1030 ofproto_flush_flows(struct ofproto *ofproto)
1032 COVERAGE_INC(ofproto_flush);
1033 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1034 dpif_flow_flush(ofproto->dpif);
1035 if (ofproto->in_band) {
1036 in_band_flushed(ofproto->in_band);
1038 if (ofproto->fail_open) {
1039 fail_open_flushed(ofproto->fail_open);
1044 reinit_ports(struct ofproto *p)
1046 struct svec devnames;
1047 struct ofport *ofport;
1048 unsigned int port_no;
1049 struct odp_port *odp_ports;
1053 svec_init(&devnames);
1054 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1055 svec_add (&devnames, (char *) ofport->opp.name);
1057 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1058 for (i = 0; i < n_odp_ports; i++) {
1059 svec_add (&devnames, odp_ports[i].devname);
1063 svec_sort_unique(&devnames);
1064 for (i = 0; i < devnames.n; i++) {
1065 update_port(p, devnames.names[i]);
1067 svec_destroy(&devnames);
1071 refresh_port_group(struct ofproto *p, unsigned int group)
1075 struct ofport *port;
1076 unsigned int port_no;
1078 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1080 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1082 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1083 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1084 ports[n_ports++] = port_no;
1087 dpif_port_group_set(p->dpif, group, ports, n_ports);
1092 refresh_port_groups(struct ofproto *p)
1094 refresh_port_group(p, DP_GROUP_FLOOD);
1095 refresh_port_group(p, DP_GROUP_ALL);
1098 static struct ofport *
1099 make_ofport(const struct odp_port *odp_port)
1101 enum netdev_flags flags;
1102 struct ofport *ofport;
1103 struct netdev *netdev;
1107 error = netdev_open(odp_port->devname, NETDEV_ETH_TYPE_NONE, &netdev);
1109 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1110 "cannot be opened (%s)",
1111 odp_port->devname, odp_port->port,
1112 odp_port->devname, strerror(error));
1116 ofport = xmalloc(sizeof *ofport);
1117 ofport->netdev = netdev;
1118 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1119 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1120 memcpy(ofport->opp.name, odp_port->devname,
1121 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1122 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1124 netdev_get_flags(netdev, &flags);
1125 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1127 netdev_get_carrier(netdev, &carrier);
1128 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1130 netdev_get_features(netdev,
1131 &ofport->opp.curr, &ofport->opp.advertised,
1132 &ofport->opp.supported, &ofport->opp.peer);
1137 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1139 if (port_array_get(&p->ports, odp_port->port)) {
1140 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1143 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1144 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1153 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1155 const struct ofp_phy_port *a = &a_->opp;
1156 const struct ofp_phy_port *b = &b_->opp;
1158 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1159 return (a->port_no == b->port_no
1160 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1161 && !strcmp((char *) a->name, (char *) b->name)
1162 && a->state == b->state
1163 && a->config == b->config
1164 && a->curr == b->curr
1165 && a->advertised == b->advertised
1166 && a->supported == b->supported
1167 && a->peer == b->peer);
1171 send_port_status(struct ofproto *p, const struct ofport *ofport,
1174 /* XXX Should limit the number of queued port status change messages. */
1175 struct ofconn *ofconn;
1176 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1177 struct ofp_port_status *ops;
1180 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1181 ops->reason = reason;
1182 ops->desc = ofport->opp;
1183 hton_ofp_phy_port(&ops->desc);
1184 queue_tx(b, ofconn, NULL);
1186 if (p->ofhooks->port_changed_cb) {
1187 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1192 ofport_install(struct ofproto *p, struct ofport *ofport)
1194 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1195 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no),
1197 shash_add(&p->port_by_name, (char *) ofport->opp.name, ofport);
1201 ofport_remove(struct ofproto *p, struct ofport *ofport)
1203 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1204 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no), NULL);
1205 shash_delete(&p->port_by_name,
1206 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1210 ofport_free(struct ofport *ofport)
1213 netdev_close(ofport->netdev);
1219 update_port(struct ofproto *p, const char *devname)
1221 struct odp_port odp_port;
1222 struct ofport *old_ofport;
1223 struct ofport *new_ofport;
1226 COVERAGE_INC(ofproto_update_port);
1228 /* Query the datapath for port information. */
1229 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1231 /* Find the old ofport. */
1232 old_ofport = shash_find_data(&p->port_by_name, devname);
1235 /* There's no port named 'devname' but there might be a port with
1236 * the same port number. This could happen if a port is deleted
1237 * and then a new one added in its place very quickly, or if a port
1238 * is renamed. In the former case we want to send an OFPPR_DELETE
1239 * and an OFPPR_ADD, and in the latter case we want to send a
1240 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1241 * the old port's ifindex against the new port, or perhaps less
1242 * reliably but more portably by comparing the old port's MAC
1243 * against the new port's MAC. However, this code isn't that smart
1244 * and always sends an OFPPR_MODIFY (XXX). */
1245 old_ofport = port_array_get(&p->ports, odp_port.port);
1247 } else if (error != ENOENT && error != ENODEV) {
1248 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1249 "%s", strerror(error));
1253 /* Create a new ofport. */
1254 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1256 /* Eliminate a few pathological cases. */
1257 if (!old_ofport && !new_ofport) {
1259 } else if (old_ofport && new_ofport) {
1260 /* Most of the 'config' bits are OpenFlow soft state, but
1261 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1262 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1263 * leaves the other bits 0.) */
1264 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1266 if (ofport_equal(old_ofport, new_ofport)) {
1267 /* False alarm--no change. */
1268 ofport_free(new_ofport);
1273 /* Now deal with the normal cases. */
1275 ofport_remove(p, old_ofport);
1278 ofport_install(p, new_ofport);
1280 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1281 (!old_ofport ? OFPPR_ADD
1282 : !new_ofport ? OFPPR_DELETE
1284 ofport_free(old_ofport);
1286 /* Update port groups. */
1287 refresh_port_groups(p);
1291 init_ports(struct ofproto *p)
1293 struct odp_port *ports;
1298 error = dpif_port_list(p->dpif, &ports, &n_ports);
1303 for (i = 0; i < n_ports; i++) {
1304 const struct odp_port *odp_port = &ports[i];
1305 if (!ofport_conflicts(p, odp_port)) {
1306 struct ofport *ofport = make_ofport(odp_port);
1308 ofport_install(p, ofport);
1313 refresh_port_groups(p);
1317 static struct ofconn *
1318 ofconn_create(struct ofproto *p, struct rconn *rconn)
1320 struct ofconn *ofconn = xmalloc(sizeof *ofconn);
1321 list_push_back(&p->all_conns, &ofconn->node);
1322 ofconn->rconn = rconn;
1323 ofconn->pktbuf = NULL;
1324 ofconn->send_flow_exp = false;
1325 ofconn->miss_send_len = 0;
1326 ofconn->packet_in_counter = rconn_packet_counter_create ();
1327 ofconn->reply_counter = rconn_packet_counter_create ();
1332 ofconn_destroy(struct ofconn *ofconn, struct ofproto *p)
1335 executer_rconn_closing(p->executer, ofconn->rconn);
1338 list_remove(&ofconn->node);
1339 rconn_destroy(ofconn->rconn);
1340 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1341 rconn_packet_counter_destroy(ofconn->reply_counter);
1342 pktbuf_destroy(ofconn->pktbuf);
1347 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1351 rconn_run(ofconn->rconn);
1353 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1354 /* Limit the number of iterations to prevent other tasks from
1356 for (iteration = 0; iteration < 50; iteration++) {
1357 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1362 fail_open_maybe_recover(p->fail_open);
1364 handle_openflow(ofconn, p, of_msg);
1365 ofpbuf_delete(of_msg);
1369 if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
1370 ofconn_destroy(ofconn, p);
1375 ofconn_wait(struct ofconn *ofconn)
1377 rconn_run_wait(ofconn->rconn);
1378 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1379 rconn_recv_wait(ofconn->rconn);
1381 COVERAGE_INC(ofproto_ofconn_stuck);
1385 /* Caller is responsible for initializing the 'cr' member of the returned
1387 static struct rule *
1388 rule_create(struct rule *super,
1389 const union ofp_action *actions, size_t n_actions,
1390 uint16_t idle_timeout, uint16_t hard_timeout)
1392 struct rule *rule = xzalloc(sizeof *rule);
1393 rule->idle_timeout = idle_timeout;
1394 rule->hard_timeout = hard_timeout;
1395 rule->used = rule->created = time_msec();
1396 rule->super = super;
1398 list_push_back(&super->list, &rule->list);
1400 list_init(&rule->list);
1402 rule->n_actions = n_actions;
1403 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1407 static struct rule *
1408 rule_from_cls_rule(const struct cls_rule *cls_rule)
1410 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1414 rule_free(struct rule *rule)
1416 free(rule->actions);
1417 free(rule->odp_actions);
1421 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1422 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1423 * through all of its subrules and revalidates them, destroying any that no
1424 * longer has a super-rule (which is probably all of them).
1426 * Before calling this function, the caller must make have removed 'rule' from
1427 * the classifier. If 'rule' is an exact-match rule, the caller is also
1428 * responsible for ensuring that it has been uninstalled from the datapath. */
1430 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1433 struct rule *subrule, *next;
1434 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1435 revalidate_rule(ofproto, subrule);
1438 list_remove(&rule->list);
1444 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1446 const union ofp_action *oa;
1447 struct actions_iterator i;
1449 if (out_port == htons(OFPP_NONE)) {
1452 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1453 oa = actions_next(&i)) {
1454 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1461 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1462 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1464 * The flow that 'packet' actually contains does not need to actually match
1465 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1466 * the packet and byte counters for 'rule' will be credited for the packet sent
1467 * out whether or not the packet actually matches 'rule'.
1469 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1470 * the caller must already have accurately composed ODP actions for it given
1471 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1472 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1473 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1474 * actions and apply them to 'packet'. */
1476 rule_execute(struct ofproto *ofproto, struct rule *rule,
1477 struct ofpbuf *packet, const flow_t *flow)
1479 const union odp_action *actions;
1481 struct odp_actions a;
1483 /* Grab or compose the ODP actions.
1485 * The special case for an exact-match 'rule' where 'flow' is not the
1486 * rule's flow is important to avoid, e.g., sending a packet out its input
1487 * port simply because the ODP actions were composed for the wrong
1489 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1490 struct rule *super = rule->super ? rule->super : rule;
1491 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1492 packet, &a, NULL, 0)) {
1495 actions = a.actions;
1496 n_actions = a.n_actions;
1498 actions = rule->odp_actions;
1499 n_actions = rule->n_odp_actions;
1502 /* Execute the ODP actions. */
1503 if (!dpif_execute(ofproto->dpif, flow->in_port,
1504 actions, n_actions, packet)) {
1505 struct odp_flow_stats stats;
1506 flow_extract_stats(flow, packet, &stats);
1507 update_stats(rule, &stats);
1508 rule->used = time_msec();
1513 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1516 struct rule *displaced_rule;
1518 /* Insert the rule in the classifier. */
1519 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1520 if (!rule->cr.wc.wildcards) {
1521 rule_make_actions(p, rule, packet);
1524 /* Send the packet and credit it to the rule. */
1527 flow_extract(packet, in_port, &flow);
1528 rule_execute(p, rule, packet, &flow);
1531 /* Install the rule in the datapath only after sending the packet, to
1532 * avoid packet reordering. */
1533 if (rule->cr.wc.wildcards) {
1534 COVERAGE_INC(ofproto_add_wc_flow);
1535 p->need_revalidate = true;
1537 rule_install(p, rule, displaced_rule);
1540 /* Free the rule that was displaced, if any. */
1541 if (displaced_rule) {
1542 rule_destroy(p, displaced_rule);
1546 static struct rule *
1547 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1550 struct rule *subrule = rule_create(rule, NULL, 0,
1551 rule->idle_timeout, rule->hard_timeout);
1552 COVERAGE_INC(ofproto_subrule_create);
1553 cls_rule_from_flow(&subrule->cr, flow, 0,
1554 (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1555 : rule->cr.priority));
1556 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1562 rule_remove(struct ofproto *ofproto, struct rule *rule)
1564 if (rule->cr.wc.wildcards) {
1565 COVERAGE_INC(ofproto_del_wc_flow);
1566 ofproto->need_revalidate = true;
1568 rule_uninstall(ofproto, rule);
1570 classifier_remove(&ofproto->cls, &rule->cr);
1571 rule_destroy(ofproto, rule);
1574 /* Returns true if the actions changed, false otherwise. */
1576 rule_make_actions(struct ofproto *p, struct rule *rule,
1577 const struct ofpbuf *packet)
1579 const struct rule *super;
1580 struct odp_actions a;
1583 assert(!rule->cr.wc.wildcards);
1585 super = rule->super ? rule->super : rule;
1587 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1588 packet, &a, &rule->tags, &rule->may_install);
1590 actions_len = a.n_actions * sizeof *a.actions;
1591 if (rule->n_odp_actions != a.n_actions
1592 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1593 COVERAGE_INC(ofproto_odp_unchanged);
1594 free(rule->odp_actions);
1595 rule->n_odp_actions = a.n_actions;
1596 rule->odp_actions = xmemdup(a.actions, actions_len);
1604 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1605 struct odp_flow_put *put)
1607 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1608 put->flow.key = rule->cr.flow;
1609 put->flow.actions = rule->odp_actions;
1610 put->flow.n_actions = rule->n_odp_actions;
1612 return dpif_flow_put(ofproto->dpif, put);
1616 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1618 assert(!rule->cr.wc.wildcards);
1620 if (rule->may_install) {
1621 struct odp_flow_put put;
1622 if (!do_put_flow(p, rule,
1623 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1625 rule->installed = true;
1626 if (displaced_rule) {
1627 update_stats(rule, &put.flow.stats);
1628 rule_post_uninstall(p, displaced_rule);
1631 } else if (displaced_rule) {
1632 rule_uninstall(p, displaced_rule);
1637 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1639 if (rule->installed) {
1640 struct odp_flow_put put;
1641 COVERAGE_INC(ofproto_dp_missed);
1642 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1644 rule_install(ofproto, rule, NULL);
1649 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1651 bool actions_changed = rule_make_actions(ofproto, rule, NULL);
1652 if (rule->may_install) {
1653 if (rule->installed) {
1654 if (actions_changed) {
1655 /* XXX should really do rule_post_uninstall() for the *old* set
1656 * of actions, and distinguish the old stats from the new. */
1657 struct odp_flow_put put;
1658 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1661 rule_install(ofproto, rule, NULL);
1664 rule_uninstall(ofproto, rule);
1669 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1671 uint64_t total_bytes = rule->byte_count + extra_bytes;
1673 if (ofproto->ofhooks->account_flow_cb
1674 && total_bytes > rule->accounted_bytes)
1676 ofproto->ofhooks->account_flow_cb(
1677 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1678 total_bytes - rule->accounted_bytes, ofproto->aux);
1679 rule->accounted_bytes = total_bytes;
1684 rule_uninstall(struct ofproto *p, struct rule *rule)
1686 assert(!rule->cr.wc.wildcards);
1687 if (rule->installed) {
1688 struct odp_flow odp_flow;
1690 odp_flow.key = rule->cr.flow;
1691 odp_flow.actions = NULL;
1692 odp_flow.n_actions = 0;
1693 if (!dpif_flow_del(p->dpif, &odp_flow)) {
1694 update_stats(rule, &odp_flow.stats);
1696 rule->installed = false;
1698 rule_post_uninstall(p, rule);
1703 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1705 struct rule *super = rule->super;
1707 rule_account(ofproto, rule, 0);
1708 if (ofproto->netflow && rule->byte_count) {
1709 struct ofexpired expired;
1710 expired.flow = rule->cr.flow;
1711 expired.packet_count = rule->packet_count;
1712 expired.byte_count = rule->byte_count;
1713 expired.used = rule->used;
1714 expired.created = rule->created;
1715 expired.tcp_flags = rule->tcp_flags;
1716 expired.ip_tos = rule->ip_tos;
1717 netflow_expire(ofproto->netflow, &expired);
1720 super->packet_count += rule->packet_count;
1721 super->byte_count += rule->byte_count;
1722 super->tcp_flags |= rule->tcp_flags;
1723 if (rule->packet_count) {
1724 super->ip_tos = rule->ip_tos;
1728 /* Reset counters to prevent double counting if the rule ever gets
1730 rule->packet_count = 0;
1731 rule->byte_count = 0;
1732 rule->accounted_bytes = 0;
1733 rule->tcp_flags = 0;
1738 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1739 struct rconn_packet_counter *counter)
1741 update_openflow_length(msg);
1742 if (rconn_send(ofconn->rconn, msg, counter)) {
1748 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1749 int error, const void *data, size_t len)
1752 struct ofp_error_msg *oem;
1754 if (!(error >> 16)) {
1755 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1760 COVERAGE_INC(ofproto_error);
1761 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1762 oh ? oh->xid : 0, &buf);
1763 oem->type = htons((unsigned int) error >> 16);
1764 oem->code = htons(error & 0xffff);
1765 memcpy(oem->data, data, len);
1766 queue_tx(buf, ofconn, ofconn->reply_counter);
1770 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1773 size_t oh_length = ntohs(oh->length);
1774 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1778 hton_ofp_phy_port(struct ofp_phy_port *opp)
1780 opp->port_no = htons(opp->port_no);
1781 opp->config = htonl(opp->config);
1782 opp->state = htonl(opp->state);
1783 opp->curr = htonl(opp->curr);
1784 opp->advertised = htonl(opp->advertised);
1785 opp->supported = htonl(opp->supported);
1786 opp->peer = htonl(opp->peer);
1790 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
1792 struct ofp_header *rq = oh;
1793 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
1798 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
1799 struct ofp_header *oh)
1801 struct ofp_switch_features *osf;
1803 unsigned int port_no;
1804 struct ofport *port;
1806 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1807 osf->datapath_id = htonll(p->datapath_id);
1808 osf->n_buffers = htonl(pktbuf_capacity());
1810 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1811 OFPC_PORT_STATS | OFPC_MULTI_PHY_TX);
1812 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1813 (1u << OFPAT_SET_VLAN_VID) |
1814 (1u << OFPAT_SET_VLAN_PCP) |
1815 (1u << OFPAT_STRIP_VLAN) |
1816 (1u << OFPAT_SET_DL_SRC) |
1817 (1u << OFPAT_SET_DL_DST) |
1818 (1u << OFPAT_SET_NW_SRC) |
1819 (1u << OFPAT_SET_NW_DST) |
1820 (1u << OFPAT_SET_TP_SRC) |
1821 (1u << OFPAT_SET_TP_DST));
1823 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1824 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1827 queue_tx(buf, ofconn, ofconn->reply_counter);
1832 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1833 struct ofp_header *oh)
1836 struct ofp_switch_config *osc;
1840 /* Figure out flags. */
1841 dpif_get_drop_frags(p->dpif, &drop_frags);
1842 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1843 if (ofconn->send_flow_exp) {
1844 flags |= OFPC_SEND_FLOW_EXP;
1848 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1849 osc->flags = htons(flags);
1850 osc->miss_send_len = htons(ofconn->miss_send_len);
1851 queue_tx(buf, ofconn, ofconn->reply_counter);
1857 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1858 struct ofp_switch_config *osc)
1863 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1867 flags = ntohs(osc->flags);
1869 ofconn->send_flow_exp = (flags & OFPC_SEND_FLOW_EXP) != 0;
1871 if (ofconn == p->controller) {
1872 switch (flags & OFPC_FRAG_MASK) {
1873 case OFPC_FRAG_NORMAL:
1874 dpif_set_drop_frags(p->dpif, false);
1876 case OFPC_FRAG_DROP:
1877 dpif_set_drop_frags(p->dpif, true);
1880 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1886 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1887 if (ntohs(osc->miss_send_len) != 0) {
1888 ofconn->pktbuf = pktbuf_create();
1890 pktbuf_destroy(ofconn->pktbuf);
1894 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1900 add_output_group_action(struct odp_actions *actions, uint16_t group)
1902 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1906 add_controller_action(struct odp_actions *actions,
1907 const struct ofp_action_output *oao)
1909 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
1910 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
1913 struct action_xlate_ctx {
1915 const flow_t *flow; /* Flow to which these actions correspond. */
1916 int recurse; /* Recursion level, via xlate_table_action. */
1917 struct ofproto *ofproto;
1918 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
1919 * null pointer if we are revalidating
1920 * without a packet to refer to. */
1923 struct odp_actions *out; /* Datapath actions. */
1924 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
1925 bool may_setup_flow; /* True ordinarily; false if the actions must
1926 * be reassessed for every packet. */
1929 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
1930 struct action_xlate_ctx *ctx);
1933 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
1935 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
1938 if (ofport->opp.config & OFPPC_NO_FWD) {
1939 /* Forwarding disabled on port. */
1944 * We don't have an ofport record for this port, but it doesn't hurt to
1945 * allow forwarding to it anyhow. Maybe such a port will appear later
1946 * and we're pre-populating the flow table.
1950 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
1953 static struct rule *
1954 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
1957 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
1959 /* The rule we found might not be valid, since we could be in need of
1960 * revalidation. If it is not valid, don't return it. */
1963 && ofproto->need_revalidate
1964 && !revalidate_rule(ofproto, rule)) {
1965 COVERAGE_INC(ofproto_invalidated);
1973 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
1975 if (!ctx->recurse) {
1980 flow.in_port = in_port;
1982 rule = lookup_valid_rule(ctx->ofproto, &flow);
1989 do_xlate_actions(rule->actions, rule->n_actions, ctx);
1996 xlate_output_action(struct action_xlate_ctx *ctx,
1997 const struct ofp_action_output *oao)
2001 switch (ntohs(oao->port)) {
2003 add_output_action(ctx, ctx->flow->in_port);
2006 xlate_table_action(ctx, ctx->flow->in_port);
2009 if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet,
2010 ctx->out, ctx->tags,
2011 ctx->ofproto->aux)) {
2012 COVERAGE_INC(ofproto_uninstallable);
2013 ctx->may_setup_flow = false;
2017 add_output_group_action(ctx->out, DP_GROUP_FLOOD);
2020 add_output_group_action(ctx->out, DP_GROUP_ALL);
2022 case OFPP_CONTROLLER:
2023 add_controller_action(ctx->out, oao);
2026 add_output_action(ctx, ODPP_LOCAL);
2029 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2030 if (odp_port != ctx->flow->in_port) {
2031 add_output_action(ctx, odp_port);
2038 xlate_nicira_action(struct action_xlate_ctx *ctx,
2039 const struct nx_action_header *nah)
2041 const struct nx_action_resubmit *nar;
2042 int subtype = ntohs(nah->subtype);
2044 assert(nah->vendor == htonl(NX_VENDOR_ID));
2046 case NXAST_RESUBMIT:
2047 nar = (const struct nx_action_resubmit *) nah;
2048 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2052 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2058 do_xlate_actions(const union ofp_action *in, size_t n_in,
2059 struct action_xlate_ctx *ctx)
2061 struct actions_iterator iter;
2062 const union ofp_action *ia;
2063 const struct ofport *port;
2065 port = port_array_get(&ctx->ofproto->ports, ctx->flow->in_port);
2066 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2067 port->opp.config & (eth_addr_equals(ctx->flow->dl_dst, stp_eth_addr)
2068 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2069 /* Drop this flow. */
2073 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2074 uint16_t type = ntohs(ia->type);
2075 union odp_action *oa;
2079 xlate_output_action(ctx, &ia->output);
2082 case OFPAT_SET_VLAN_VID:
2083 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2084 oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2087 case OFPAT_SET_VLAN_PCP:
2088 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2089 oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2092 case OFPAT_STRIP_VLAN:
2093 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2096 case OFPAT_SET_DL_SRC:
2097 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2098 memcpy(oa->dl_addr.dl_addr,
2099 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2102 case OFPAT_SET_DL_DST:
2103 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2104 memcpy(oa->dl_addr.dl_addr,
2105 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2108 case OFPAT_SET_NW_SRC:
2109 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2110 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2113 case OFPAT_SET_TP_SRC:
2114 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2115 oa->tp_port.tp_port = ia->tp_port.tp_port;
2119 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2123 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2130 xlate_actions(const union ofp_action *in, size_t n_in,
2131 const flow_t *flow, struct ofproto *ofproto,
2132 const struct ofpbuf *packet,
2133 struct odp_actions *out, tag_type *tags, bool *may_setup_flow)
2135 tag_type no_tags = 0;
2136 struct action_xlate_ctx ctx;
2137 COVERAGE_INC(ofproto_ofp2odp);
2138 odp_actions_init(out);
2141 ctx.ofproto = ofproto;
2142 ctx.packet = packet;
2144 ctx.tags = tags ? tags : &no_tags;
2145 ctx.may_setup_flow = true;
2146 do_xlate_actions(in, n_in, &ctx);
2148 /* Check with in-band control to see if we're allowed to setup this
2150 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2151 ctx.may_setup_flow = false;
2154 if (may_setup_flow) {
2155 *may_setup_flow = ctx.may_setup_flow;
2157 if (odp_actions_overflow(out)) {
2158 odp_actions_init(out);
2159 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2165 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2166 struct ofp_header *oh)
2168 struct ofp_packet_out *opo;
2169 struct ofpbuf payload, *buffer;
2170 struct odp_actions actions;
2176 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2180 opo = (struct ofp_packet_out *) oh;
2182 COVERAGE_INC(ofproto_packet_out);
2183 if (opo->buffer_id != htonl(UINT32_MAX)) {
2184 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2186 if (error || !buffer) {
2194 flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2195 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2196 &flow, p, &payload, &actions, NULL, NULL);
2201 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2203 ofpbuf_delete(buffer);
2209 update_port_config(struct ofproto *p, struct ofport *port,
2210 uint32_t config, uint32_t mask)
2212 mask &= config ^ port->opp.config;
2213 if (mask & OFPPC_PORT_DOWN) {
2214 if (config & OFPPC_PORT_DOWN) {
2215 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2217 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2220 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2221 if (mask & REVALIDATE_BITS) {
2222 COVERAGE_INC(ofproto_costly_flags);
2223 port->opp.config ^= mask & REVALIDATE_BITS;
2224 p->need_revalidate = true;
2226 #undef REVALIDATE_BITS
2227 if (mask & OFPPC_NO_FLOOD) {
2228 port->opp.config ^= OFPPC_NO_FLOOD;
2229 refresh_port_group(p, DP_GROUP_FLOOD);
2231 if (mask & OFPPC_NO_PACKET_IN) {
2232 port->opp.config ^= OFPPC_NO_PACKET_IN;
2237 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2239 const struct ofp_port_mod *opm;
2240 struct ofport *port;
2243 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2247 opm = (struct ofp_port_mod *) oh;
2249 port = port_array_get(&p->ports,
2250 ofp_port_to_odp_port(ntohs(opm->port_no)));
2252 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2253 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2254 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2256 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2257 if (opm->advertise) {
2258 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2264 static struct ofpbuf *
2265 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2267 struct ofp_stats_reply *osr;
2270 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2271 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2273 osr->flags = htons(0);
2277 static struct ofpbuf *
2278 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2280 return make_stats_reply(request->header.xid, request->type, body_len);
2284 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2286 struct ofpbuf *msg = *msgp;
2287 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2288 if (nbytes + msg->size > UINT16_MAX) {
2289 struct ofp_stats_reply *reply = msg->data;
2290 reply->flags = htons(OFPSF_REPLY_MORE);
2291 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2292 queue_tx(msg, ofconn, ofconn->reply_counter);
2294 return ofpbuf_put_uninit(*msgp, nbytes);
2298 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2299 struct ofp_stats_request *request)
2301 struct ofp_desc_stats *ods;
2304 msg = start_stats_reply(request, sizeof *ods);
2305 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2306 strncpy(ods->mfr_desc, p->manufacturer, sizeof ods->mfr_desc);
2307 strncpy(ods->hw_desc, p->hardware, sizeof ods->hw_desc);
2308 strncpy(ods->sw_desc, p->software, sizeof ods->sw_desc);
2309 strncpy(ods->serial_num, p->serial, sizeof ods->serial_num);
2310 queue_tx(msg, ofconn, ofconn->reply_counter);
2316 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2318 struct rule *rule = rule_from_cls_rule(cls_rule);
2319 int *n_subrules = n_subrules_;
2327 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2328 struct ofp_stats_request *request)
2330 struct ofp_table_stats *ots;
2332 struct odp_stats dpstats;
2333 int n_exact, n_subrules, n_wild;
2335 msg = start_stats_reply(request, sizeof *ots * 2);
2337 /* Count rules of various kinds. */
2339 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2340 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2341 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2344 dpif_get_dp_stats(p->dpif, &dpstats);
2345 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2346 memset(ots, 0, sizeof *ots);
2347 ots->table_id = TABLEID_HASH;
2348 strcpy(ots->name, "hash");
2349 ots->wildcards = htonl(0);
2350 ots->max_entries = htonl(dpstats.max_capacity);
2351 ots->active_count = htonl(n_exact);
2352 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2354 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2356 /* Classifier table. */
2357 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2358 memset(ots, 0, sizeof *ots);
2359 ots->table_id = TABLEID_CLASSIFIER;
2360 strcpy(ots->name, "classifier");
2361 ots->wildcards = htonl(OFPFW_ALL);
2362 ots->max_entries = htonl(65536);
2363 ots->active_count = htonl(n_wild);
2364 ots->lookup_count = htonll(0); /* XXX */
2365 ots->matched_count = htonll(0); /* XXX */
2367 queue_tx(msg, ofconn, ofconn->reply_counter);
2372 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2373 struct ofp_stats_request *request)
2375 struct ofp_port_stats *ops;
2377 struct ofport *port;
2378 unsigned int port_no;
2380 msg = start_stats_reply(request, sizeof *ops * 16);
2381 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2382 struct netdev_stats stats;
2384 /* Intentionally ignore return value, since errors will set 'stats' to
2385 * all-1s, which is correct for OpenFlow, and netdev_get_stats() will
2387 netdev_get_stats(port->netdev, &stats);
2389 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2390 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2391 memset(ops->pad, 0, sizeof ops->pad);
2392 ops->rx_packets = htonll(stats.rx_packets);
2393 ops->tx_packets = htonll(stats.tx_packets);
2394 ops->rx_bytes = htonll(stats.rx_bytes);
2395 ops->tx_bytes = htonll(stats.tx_bytes);
2396 ops->rx_dropped = htonll(stats.rx_dropped);
2397 ops->tx_dropped = htonll(stats.tx_dropped);
2398 ops->rx_errors = htonll(stats.rx_errors);
2399 ops->tx_errors = htonll(stats.tx_errors);
2400 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2401 ops->rx_over_err = htonll(stats.rx_over_errors);
2402 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2403 ops->collisions = htonll(stats.collisions);
2406 queue_tx(msg, ofconn, ofconn->reply_counter);
2410 struct flow_stats_cbdata {
2411 struct ofproto *ofproto;
2412 struct ofconn *ofconn;
2418 query_stats(struct ofproto *p, struct rule *rule,
2419 uint64_t *packet_countp, uint64_t *byte_countp)
2421 uint64_t packet_count, byte_count;
2422 struct rule *subrule;
2423 struct odp_flow *odp_flows;
2426 packet_count = rule->packet_count;
2427 byte_count = rule->byte_count;
2429 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2430 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
2431 if (rule->cr.wc.wildcards) {
2433 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2434 odp_flows[i++].key = subrule->cr.flow;
2435 packet_count += subrule->packet_count;
2436 byte_count += subrule->byte_count;
2439 odp_flows[0].key = rule->cr.flow;
2442 packet_count = rule->packet_count;
2443 byte_count = rule->byte_count;
2444 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2446 for (i = 0; i < n_odp_flows; i++) {
2447 struct odp_flow *odp_flow = &odp_flows[i];
2448 packet_count += odp_flow->stats.n_packets;
2449 byte_count += odp_flow->stats.n_bytes;
2454 *packet_countp = packet_count;
2455 *byte_countp = byte_count;
2459 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2461 struct rule *rule = rule_from_cls_rule(rule_);
2462 struct flow_stats_cbdata *cbdata = cbdata_;
2463 struct ofp_flow_stats *ofs;
2464 uint64_t packet_count, byte_count;
2465 size_t act_len, len;
2467 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2471 act_len = sizeof *rule->actions * rule->n_actions;
2472 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2474 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2476 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2477 ofs->length = htons(len);
2478 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2480 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match);
2481 ofs->duration = htonl((time_msec() - rule->created) / 1000);
2482 ofs->priority = htons(rule->cr.priority);
2483 ofs->idle_timeout = htons(rule->idle_timeout);
2484 ofs->hard_timeout = htons(rule->hard_timeout);
2485 memset(ofs->pad2, 0, sizeof ofs->pad2);
2486 ofs->packet_count = htonll(packet_count);
2487 ofs->byte_count = htonll(byte_count);
2488 memcpy(ofs->actions, rule->actions, act_len);
2492 table_id_to_include(uint8_t table_id)
2494 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2495 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2496 : table_id == 0xff ? CLS_INC_ALL
2501 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2502 const struct ofp_stats_request *osr,
2505 struct ofp_flow_stats_request *fsr;
2506 struct flow_stats_cbdata cbdata;
2507 struct cls_rule target;
2509 if (arg_size != sizeof *fsr) {
2510 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2512 fsr = (struct ofp_flow_stats_request *) osr->body;
2514 COVERAGE_INC(ofproto_flows_req);
2516 cbdata.ofconn = ofconn;
2517 cbdata.out_port = fsr->out_port;
2518 cbdata.msg = start_stats_reply(osr, 1024);
2519 cls_rule_from_match(&target, &fsr->match, 0);
2520 classifier_for_each_match(&p->cls, &target,
2521 table_id_to_include(fsr->table_id),
2522 flow_stats_cb, &cbdata);
2523 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2527 struct flow_stats_ds_cbdata {
2528 struct ofproto *ofproto;
2533 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
2535 struct rule *rule = rule_from_cls_rule(rule_);
2536 struct flow_stats_ds_cbdata *cbdata = cbdata_;
2537 struct ds *results = cbdata->results;
2538 struct ofp_match match;
2539 uint64_t packet_count, byte_count;
2540 size_t act_len = sizeof *rule->actions * rule->n_actions;
2542 /* Don't report on subrules. */
2543 if (rule->super != NULL) {
2547 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2548 flow_to_ovs_match(&rule->cr.flow, rule->cr.wc.wildcards, &match);
2550 ds_put_format(results, "duration=%llds, ",
2551 (time_msec() - rule->created) / 1000);
2552 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2553 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2554 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2555 ofp_print_match(results, &match, true);
2556 ofp_print_actions(results, &rule->actions->header, act_len);
2557 ds_put_cstr(results, "\n");
2560 /* Adds a pretty-printed description of all flows to 'results', including
2561 * those marked hidden by secchan (e.g., by in-band control). */
2563 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2565 struct ofp_match match;
2566 struct cls_rule target;
2567 struct flow_stats_ds_cbdata cbdata;
2569 memset(&match, 0, sizeof match);
2570 match.wildcards = htonl(OFPFW_ALL);
2573 cbdata.results = results;
2575 cls_rule_from_match(&target, &match, 0);
2576 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2577 flow_stats_ds_cb, &cbdata);
2580 struct aggregate_stats_cbdata {
2581 struct ofproto *ofproto;
2583 uint64_t packet_count;
2584 uint64_t byte_count;
2589 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2591 struct rule *rule = rule_from_cls_rule(rule_);
2592 struct aggregate_stats_cbdata *cbdata = cbdata_;
2593 uint64_t packet_count, byte_count;
2595 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2599 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2601 cbdata->packet_count += packet_count;
2602 cbdata->byte_count += byte_count;
2607 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2608 const struct ofp_stats_request *osr,
2611 struct ofp_aggregate_stats_request *asr;
2612 struct ofp_aggregate_stats_reply *reply;
2613 struct aggregate_stats_cbdata cbdata;
2614 struct cls_rule target;
2617 if (arg_size != sizeof *asr) {
2618 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2620 asr = (struct ofp_aggregate_stats_request *) osr->body;
2622 COVERAGE_INC(ofproto_agg_request);
2624 cbdata.out_port = asr->out_port;
2625 cbdata.packet_count = 0;
2626 cbdata.byte_count = 0;
2628 cls_rule_from_match(&target, &asr->match, 0);
2629 classifier_for_each_match(&p->cls, &target,
2630 table_id_to_include(asr->table_id),
2631 aggregate_stats_cb, &cbdata);
2633 msg = start_stats_reply(osr, sizeof *reply);
2634 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2635 reply->flow_count = htonl(cbdata.n_flows);
2636 reply->packet_count = htonll(cbdata.packet_count);
2637 reply->byte_count = htonll(cbdata.byte_count);
2638 queue_tx(msg, ofconn, ofconn->reply_counter);
2643 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2644 struct ofp_header *oh)
2646 struct ofp_stats_request *osr;
2650 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2655 osr = (struct ofp_stats_request *) oh;
2657 switch (ntohs(osr->type)) {
2659 return handle_desc_stats_request(p, ofconn, osr);
2662 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2664 case OFPST_AGGREGATE:
2665 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2668 return handle_table_stats_request(p, ofconn, osr);
2671 return handle_port_stats_request(p, ofconn, osr);
2674 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2677 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2681 static long long int
2682 msec_from_nsec(uint64_t sec, uint32_t nsec)
2684 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2688 update_time(struct rule *rule, const struct odp_flow_stats *stats)
2690 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2691 if (used > rule->used) {
2697 update_stats(struct rule *rule, const struct odp_flow_stats *stats)
2699 update_time(rule, stats);
2700 rule->packet_count += stats->n_packets;
2701 rule->byte_count += stats->n_bytes;
2702 rule->tcp_flags |= stats->tcp_flags;
2703 if (stats->n_packets) {
2704 rule->ip_tos = stats->ip_tos;
2709 add_flow(struct ofproto *p, struct ofconn *ofconn,
2710 struct ofp_flow_mod *ofm, size_t n_actions)
2712 struct ofpbuf *packet;
2717 rule = rule_create(NULL, (const union ofp_action *) ofm->actions,
2718 n_actions, ntohs(ofm->idle_timeout),
2719 ntohs(ofm->hard_timeout));
2720 cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority));
2724 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2725 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2729 rule_insert(p, rule, packet, in_port);
2730 ofpbuf_delete(packet);
2735 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
2736 size_t n_actions, uint16_t command, struct rule *rule)
2738 if (rule_is_hidden(rule)) {
2742 if (command == OFPFC_DELETE) {
2743 rule_remove(p, rule);
2745 size_t actions_len = n_actions * sizeof *rule->actions;
2747 if (n_actions == rule->n_actions
2748 && !memcmp(ofm->actions, rule->actions, actions_len))
2753 free(rule->actions);
2754 rule->actions = xmemdup(ofm->actions, actions_len);
2755 rule->n_actions = n_actions;
2757 if (rule->cr.wc.wildcards) {
2758 COVERAGE_INC(ofproto_mod_wc_flow);
2759 p->need_revalidate = true;
2761 rule_update_actions(p, rule);
2769 modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm,
2770 size_t n_actions, uint16_t command)
2776 flow_from_match(&flow, &wildcards, &ofm->match);
2777 rule = rule_from_cls_rule(classifier_find_rule_exactly(
2778 &p->cls, &flow, wildcards,
2779 ntohs(ofm->priority)));
2782 if (command == OFPFC_DELETE
2783 && ofm->out_port != htons(OFPP_NONE)
2784 && !rule_has_out_port(rule, ofm->out_port)) {
2788 modify_flow(p, ofm, n_actions, command, rule);
2793 struct modify_flows_cbdata {
2794 struct ofproto *ofproto;
2795 const struct ofp_flow_mod *ofm;
2802 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
2804 struct rule *rule = rule_from_cls_rule(rule_);
2805 struct modify_flows_cbdata *cbdata = cbdata_;
2807 if (cbdata->out_port != htons(OFPP_NONE)
2808 && !rule_has_out_port(rule, cbdata->out_port)) {
2812 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions,
2813 cbdata->command, rule);
2817 modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm,
2818 size_t n_actions, uint16_t command)
2820 struct modify_flows_cbdata cbdata;
2821 struct cls_rule target;
2825 cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port
2826 : htons(OFPP_NONE));
2827 cbdata.n_actions = n_actions;
2828 cbdata.command = command;
2830 cls_rule_from_match(&target, &ofm->match, 0);
2832 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2833 modify_flows_cb, &cbdata);
2838 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
2839 struct ofp_flow_mod *ofm)
2844 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
2845 sizeof *ofm->actions, &n_actions);
2850 normalize_match(&ofm->match);
2851 if (!ofm->match.wildcards) {
2852 ofm->priority = htons(UINT16_MAX);
2855 error = validate_actions((const union ofp_action *) ofm->actions,
2856 n_actions, p->max_ports);
2861 switch (ntohs(ofm->command)) {
2863 return add_flow(p, ofconn, ofm, n_actions);
2866 return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY);
2868 case OFPFC_MODIFY_STRICT:
2869 return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY);
2872 return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE);
2874 case OFPFC_DELETE_STRICT:
2875 return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE);
2878 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
2883 send_capability_reply(struct ofproto *p, struct ofconn *ofconn, uint32_t xid)
2885 struct ofmp_capability_reply *ocr;
2887 char capabilities[] = "com.nicira.mgmt.manager=false\n";
2889 ocr = make_openflow_xid(sizeof(*ocr), OFPT_VENDOR, xid, &b);
2890 ocr->header.header.vendor = htonl(NX_VENDOR_ID);
2891 ocr->header.header.subtype = htonl(NXT_MGMT);
2892 ocr->header.type = htons(OFMPT_CAPABILITY_REPLY);
2894 ocr->format = htonl(OFMPCOF_SIMPLE);
2895 ocr->mgmt_id = htonll(p->mgmt_id);
2897 ofpbuf_put(b, capabilities, strlen(capabilities));
2899 queue_tx(b, ofconn, ofconn->reply_counter);
2903 handle_ofmp(struct ofproto *p, struct ofconn *ofconn,
2904 struct ofmp_header *ofmph)
2906 size_t msg_len = ntohs(ofmph->header.header.length);
2907 if (msg_len < sizeof(*ofmph)) {
2908 VLOG_WARN_RL(&rl, "dropping short managment message: %d\n", msg_len);
2909 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2912 if (ofmph->type == htons(OFMPT_CAPABILITY_REQUEST)) {
2913 struct ofmp_capability_request *ofmpcr;
2915 if (msg_len < sizeof(struct ofmp_capability_request)) {
2916 VLOG_WARN_RL(&rl, "dropping short capability request: %d\n",
2918 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2921 ofmpcr = (struct ofmp_capability_request *)ofmph;
2922 if (ofmpcr->format != htonl(OFMPCAF_SIMPLE)) {
2923 /* xxx Find a better type than bad subtype */
2924 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2927 send_capability_reply(p, ofconn, ofmph->header.header.xid);
2930 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2935 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
2937 struct ofp_vendor_header *ovh = msg;
2938 struct nicira_header *nh;
2940 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
2941 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2943 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
2944 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2946 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
2947 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2951 switch (ntohl(nh->subtype)) {
2952 case NXT_STATUS_REQUEST:
2953 return switch_status_handle_request(p->switch_status, ofconn->rconn,
2956 case NXT_ACT_SET_CONFIG:
2957 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
2959 case NXT_ACT_GET_CONFIG:
2960 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
2962 case NXT_COMMAND_REQUEST:
2964 return executer_handle_request(p->executer, ofconn->rconn, msg);
2969 return handle_ofmp(p, ofconn, msg);
2972 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2976 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
2977 struct ofpbuf *ofp_msg)
2979 struct ofp_header *oh = ofp_msg->data;
2982 COVERAGE_INC(ofproto_recv_openflow);
2984 case OFPT_ECHO_REQUEST:
2985 error = handle_echo_request(ofconn, oh);
2988 case OFPT_ECHO_REPLY:
2992 case OFPT_FEATURES_REQUEST:
2993 error = handle_features_request(p, ofconn, oh);
2996 case OFPT_GET_CONFIG_REQUEST:
2997 error = handle_get_config_request(p, ofconn, oh);
3000 case OFPT_SET_CONFIG:
3001 error = handle_set_config(p, ofconn, ofp_msg->data);
3004 case OFPT_PACKET_OUT:
3005 error = handle_packet_out(p, ofconn, ofp_msg->data);
3009 error = handle_port_mod(p, oh);
3013 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3016 case OFPT_STATS_REQUEST:
3017 error = handle_stats_request(p, ofconn, oh);
3021 error = handle_vendor(p, ofconn, ofp_msg->data);
3025 if (VLOG_IS_WARN_ENABLED()) {
3026 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3027 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3030 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3035 send_error_oh(ofconn, ofp_msg->data, error);
3040 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3042 struct odp_msg *msg = packet->data;
3043 uint16_t in_port = odp_port_to_ofp_port(msg->port);
3045 struct ofpbuf payload;
3048 /* Handle controller actions. */
3049 if (msg->type == _ODPL_ACTION_NR) {
3050 COVERAGE_INC(ofproto_ctlr_action);
3051 pinsched_send(p->action_sched, in_port, packet,
3052 send_packet_in_action, p);
3056 payload.data = msg + 1;
3057 payload.size = msg->length - sizeof *msg;
3058 flow_extract(&payload, msg->port, &flow);
3060 /* Check with in-band control to see if this packet should be sent
3061 * to the local port regardless of the flow table. */
3062 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3063 union odp_action action;
3065 memset(&action, 0, sizeof(action));
3066 action.output.type = ODPAT_OUTPUT;
3067 action.output.port = ODPP_LOCAL;
3068 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3071 rule = lookup_valid_rule(p, &flow);
3073 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3074 struct ofport *port = port_array_get(&p->ports, msg->port);
3076 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3077 COVERAGE_INC(ofproto_no_packet_in);
3078 /* XXX install 'drop' flow entry */
3079 ofpbuf_delete(packet);
3083 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3086 COVERAGE_INC(ofproto_packet_in);
3087 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3091 if (rule->cr.wc.wildcards) {
3092 rule = rule_create_subrule(p, rule, &flow);
3093 rule_make_actions(p, rule, packet);
3095 if (!rule->may_install) {
3096 /* The rule is not installable, that is, we need to process every
3097 * packet, so process the current packet and set its actions into
3099 rule_make_actions(p, rule, packet);
3101 /* XXX revalidate rule if it needs it */
3105 rule_execute(p, rule, &payload, &flow);
3106 rule_reinstall(p, rule);
3108 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY
3109 && rconn_is_connected(p->controller->rconn)) {
3111 * Extra-special case for fail-open mode.
3113 * We are in fail-open mode and the packet matched the fail-open rule,
3114 * but we are connected to a controller too. We should send the packet
3115 * up to the controller in the hope that it will try to set up a flow
3116 * and thereby allow us to exit fail-open.
3118 * See the top-level comment in fail-open.c for more information.
3120 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3122 ofpbuf_delete(packet);
3127 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3129 struct rule *sub = rule_from_cls_rule(sub_);
3130 struct revalidate_cbdata *cbdata = cbdata_;
3132 if (cbdata->revalidate_all
3133 || (cbdata->revalidate_subrules && sub->super)
3134 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3135 revalidate_rule(cbdata->ofproto, sub);
3140 revalidate_rule(struct ofproto *p, struct rule *rule)
3142 const flow_t *flow = &rule->cr.flow;
3144 COVERAGE_INC(ofproto_revalidate_rule);
3147 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3149 rule_remove(p, rule);
3151 } else if (super != rule->super) {
3152 COVERAGE_INC(ofproto_revalidate_moved);
3153 list_remove(&rule->list);
3154 list_push_back(&super->list, &rule->list);
3155 rule->super = super;
3156 rule->hard_timeout = super->hard_timeout;
3157 rule->idle_timeout = super->idle_timeout;
3158 rule->created = super->created;
3163 rule_update_actions(p, rule);
3167 static struct ofpbuf *
3168 compose_flow_exp(const struct rule *rule, long long int now, uint8_t reason)
3170 struct ofp_flow_expired *ofe;
3173 ofe = make_openflow(sizeof *ofe, OFPT_FLOW_EXPIRED, &buf);
3174 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofe->match);
3175 ofe->priority = htons(rule->cr.priority);
3176 ofe->reason = reason;
3177 ofe->duration = (now - rule->created) / 1000;
3178 ofe->packet_count = rule->packet_count;
3179 ofe->byte_count = rule->byte_count;
3185 send_flow_exp(struct ofproto *p, struct rule *rule,
3186 long long int now, uint8_t reason)
3188 struct ofconn *ofconn;
3189 struct ofconn *prev;
3190 struct ofpbuf *buf = NULL;
3192 /* We limit the maximum number of queued flow expirations it by accounting
3193 * them under the counter for replies. That works because preventing
3194 * OpenFlow requests from being processed also prevents new flows from
3195 * being added (and expiring). (It also prevents processing OpenFlow
3196 * requests that would not add new flows, so it is imperfect.) */
3199 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3200 if (ofconn->send_flow_exp && rconn_is_connected(ofconn->rconn)) {
3202 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3204 buf = compose_flow_exp(rule, now, reason);
3210 queue_tx(buf, prev, prev->reply_counter);
3215 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3217 assert(rule->installed);
3218 assert(!rule->cr.wc.wildcards);
3221 rule_remove(ofproto, rule);
3223 rule_uninstall(ofproto, rule);
3228 expire_rule(struct cls_rule *cls_rule, void *p_)
3230 struct ofproto *p = p_;
3231 struct rule *rule = rule_from_cls_rule(cls_rule);
3232 long long int hard_expire, idle_expire, expire, now;
3234 hard_expire = (rule->hard_timeout
3235 ? rule->created + rule->hard_timeout * 1000
3237 idle_expire = (rule->idle_timeout
3238 && (rule->super || list_is_empty(&rule->list))
3239 ? rule->used + rule->idle_timeout * 1000
3241 expire = MIN(hard_expire, idle_expire);
3242 if (expire == LLONG_MAX) {
3243 if (rule->installed && time_msec() >= rule->used + 5000) {
3244 uninstall_idle_flow(p, rule);
3251 if (rule->installed && now >= rule->used + 5000) {
3252 uninstall_idle_flow(p, rule);
3257 COVERAGE_INC(ofproto_expired);
3258 if (rule->cr.wc.wildcards) {
3259 /* Update stats. (This code will be a no-op if the rule expired
3260 * due to an idle timeout, because in that case the rule has no
3261 * subrules left.) */
3262 struct rule *subrule, *next;
3263 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3264 rule_remove(p, subrule);
3268 send_flow_exp(p, rule, now,
3270 ? OFPER_HARD_TIMEOUT : OFPER_IDLE_TIMEOUT));
3271 rule_remove(p, rule);
3275 update_used(struct ofproto *p)
3277 struct odp_flow *flows;
3282 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
3287 for (i = 0; i < n_flows; i++) {
3288 struct odp_flow *f = &flows[i];
3291 rule = rule_from_cls_rule(
3292 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3293 if (!rule || !rule->installed) {
3294 COVERAGE_INC(ofproto_unexpected_rule);
3295 dpif_flow_del(p->dpif, f);
3299 update_time(rule, &f->stats);
3300 rule_account(p, rule, f->stats.n_bytes);
3306 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3307 const struct ofpbuf *packet, int send_len)
3309 struct odp_msg *msg = packet->data;
3310 struct ofpbuf payload;
3314 /* Extract packet payload from 'msg'. */
3315 payload.data = msg + 1;
3316 payload.size = msg->length - sizeof *msg;
3318 /* Construct ofp_packet_in message. */
3319 reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3320 opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port), reason,
3321 &payload, send_len);
3324 rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100);
3328 send_packet_in_action(struct ofpbuf *packet, void *p_)
3330 struct ofproto *p = p_;
3331 struct ofconn *ofconn;
3332 struct odp_msg *msg;
3335 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3336 if (ofconn == p->controller || ofconn->miss_send_len) {
3337 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3340 ofpbuf_delete(packet);
3344 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3346 struct ofproto *p = p_;
3347 bool in_fail_open = p->fail_open && fail_open_is_active(p->fail_open);
3348 struct ofconn *ofconn;
3349 struct ofpbuf payload;
3350 struct odp_msg *msg;
3353 payload.data = msg + 1;
3354 payload.size = msg->length - sizeof *msg;
3355 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3356 if (ofconn->miss_send_len) {
3357 struct pktbuf *pb = ofconn->pktbuf;
3358 uint32_t buffer_id = (in_fail_open
3360 : pktbuf_save(pb, &payload, msg->port));
3361 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3363 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3366 ofpbuf_delete(packet);
3370 pick_datapath_id(const struct ofproto *ofproto)
3372 const struct ofport *port;
3374 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
3376 uint8_t ea[ETH_ADDR_LEN];
3379 error = netdev_get_etheraddr(port->netdev, ea);
3381 return eth_addr_to_uint64(ea);
3383 VLOG_WARN("could not get MAC address for %s (%s)",
3384 netdev_get_name(port->netdev), strerror(error));
3386 return ofproto->fallback_dpid;
3390 pick_fallback_dpid(void)
3392 uint8_t ea[ETH_ADDR_LEN];
3393 eth_addr_random(ea);
3394 ea[0] = 0x00; /* Set Nicira OUI. */
3397 return eth_addr_to_uint64(ea);
3401 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3402 struct odp_actions *actions, tag_type *tags,
3405 struct ofproto *ofproto = ofproto_;
3408 /* Drop frames for reserved multicast addresses. */
3409 if (eth_addr_is_reserved(flow->dl_dst)) {
3413 /* Learn source MAC (but don't try to learn from revalidation). */
3414 if (packet != NULL) {
3415 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3418 /* The log messages here could actually be useful in debugging,
3419 * so keep the rate limit relatively high. */
3420 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3421 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3422 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3423 ofproto_revalidate(ofproto, rev_tag);
3427 /* Determine output port. */
3428 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3430 add_output_group_action(actions, DP_GROUP_FLOOD);
3431 } else if (out_port != flow->in_port) {
3432 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3440 static const struct ofhooks default_ofhooks = {
3442 default_normal_ofhook_cb,