2 * Copyright (c) 2009 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
22 #include <netinet/in.h>
25 #include "classifier.h"
27 #include "discovery.h"
29 #include "dynamic-string.h"
31 #include "fail-open.h"
33 #include "mac-learning.h"
37 #include "ofp-print.h"
39 #include "openflow/nicira-ext.h"
40 #include "openflow/openflow.h"
41 #include "openflow/openflow-mgmt.h"
42 #include "openvswitch/datapath-protocol.h"
46 #include "poll-loop.h"
47 #include "port-array.h"
57 #include "vconn-ssl.h"
60 #define THIS_MODULE VLM_ofproto
70 TABLEID_CLASSIFIER = 1
74 struct netdev *netdev;
75 struct ofp_phy_port opp; /* In host byte order. */
78 static void ofport_free(struct ofport *);
79 static void hton_ofp_phy_port(struct ofp_phy_port *);
81 static int xlate_actions(const union ofp_action *in, size_t n_in,
82 const flow_t *flow, struct ofproto *ofproto,
83 const struct ofpbuf *packet,
84 struct odp_actions *out, tag_type *tags,
85 bool *may_set_up_flow, uint16_t *nf_output_iface);
90 uint16_t idle_timeout; /* In seconds from time of last use. */
91 uint16_t hard_timeout; /* In seconds from time of creation. */
92 long long int used; /* Last-used time (0 if never used). */
93 long long int created; /* Creation time. */
94 uint64_t packet_count; /* Number of packets received. */
95 uint64_t byte_count; /* Number of bytes received. */
96 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
97 tag_type tags; /* Tags (set only by hooks). */
98 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
100 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
101 * exact-match rule (having cr.wc.wildcards of 0) generated from the
102 * wildcard rule 'super'. In this case, 'list' is an element of the
105 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
106 * a list of subrules. A super-rule with no wildcards (where
107 * cr.wc.wildcards is 0) will never have any subrules. */
113 * A subrule has no actions (it uses the super-rule's actions). */
115 union ofp_action *actions;
119 * A super-rule with wildcard fields never has ODP actions (since the
120 * datapath only supports exact-match flows). */
121 bool installed; /* Installed in datapath? */
122 bool may_install; /* True ordinarily; false if actions must
123 * be reassessed for every packet. */
125 union odp_action *odp_actions;
129 rule_is_hidden(const struct rule *rule)
131 /* Subrules are merely an implementation detail, so hide them from the
133 if (rule->super != NULL) {
137 /* Rules with priority higher than UINT16_MAX are set up by secchan itself
138 * (e.g. by in-band control) and are intentionally hidden from the
140 if (rule->cr.priority > UINT16_MAX) {
147 static struct rule *rule_create(struct ofproto *, struct rule *super,
148 const union ofp_action *, size_t n_actions,
149 uint16_t idle_timeout, uint16_t hard_timeout);
150 static void rule_free(struct rule *);
151 static void rule_destroy(struct ofproto *, struct rule *);
152 static struct rule *rule_from_cls_rule(const struct cls_rule *);
153 static void rule_insert(struct ofproto *, struct rule *,
154 struct ofpbuf *packet, uint16_t in_port);
155 static void rule_remove(struct ofproto *, struct rule *);
156 static bool rule_make_actions(struct ofproto *, struct rule *,
157 const struct ofpbuf *packet);
158 static void rule_install(struct ofproto *, struct rule *,
159 struct rule *displaced_rule);
160 static void rule_uninstall(struct ofproto *, struct rule *);
161 static void rule_post_uninstall(struct ofproto *, struct rule *);
166 struct pktbuf *pktbuf;
170 struct rconn_packet_counter *packet_in_counter;
172 /* Number of OpenFlow messages queued as replies to OpenFlow requests, and
173 * the maximum number before we stop reading OpenFlow requests. */
174 #define OFCONN_REPLY_MAX 100
175 struct rconn_packet_counter *reply_counter;
178 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *);
179 static void ofconn_destroy(struct ofconn *, struct ofproto *);
180 static void ofconn_run(struct ofconn *, struct ofproto *);
181 static void ofconn_wait(struct ofconn *);
182 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
183 struct rconn_packet_counter *counter);
187 uint64_t datapath_id; /* Datapath ID. */
188 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
189 uint64_t mgmt_id; /* Management channel identifier. */
190 char *manufacturer; /* Manufacturer. */
191 char *hardware; /* Hardware. */
192 char *software; /* Software version. */
193 char *serial; /* Serial number. */
197 struct dpifmon *dpifmon;
198 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
200 struct shash port_by_name;
204 struct switch_status *switch_status;
205 struct status_category *ss_cat;
206 struct in_band *in_band;
207 struct discovery *discovery;
208 struct fail_open *fail_open;
209 struct pinsched *miss_sched, *action_sched;
210 struct executer *executer;
211 struct netflow *netflow;
214 struct classifier cls;
215 bool need_revalidate;
216 long long int next_expiration;
217 struct tag_set revalidate_set;
219 /* OpenFlow connections. */
220 struct list all_conns;
221 struct ofconn *controller;
222 struct pvconn **listeners;
224 struct pvconn **snoops;
227 /* Hooks for ovs-vswitchd. */
228 const struct ofhooks *ofhooks;
231 /* Used by default ofhooks. */
232 struct mac_learning *ml;
235 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
237 static const struct ofhooks default_ofhooks;
239 static uint64_t pick_datapath_id(struct dpif *, uint64_t fallback_dpid);
240 static uint64_t pick_fallback_dpid(void);
241 static void send_packet_in_miss(struct ofpbuf *, void *ofproto);
242 static void send_packet_in_action(struct ofpbuf *, void *ofproto);
243 static void update_used(struct ofproto *);
244 static void update_stats(struct ofproto *, struct rule *,
245 const struct odp_flow_stats *);
246 static void expire_rule(struct cls_rule *, void *ofproto);
247 static void active_timeout(struct ofproto *ofproto, struct rule *rule);
248 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
249 static void revalidate_cb(struct cls_rule *rule_, void *p_);
251 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
253 static void handle_openflow(struct ofconn *, struct ofproto *,
256 static void refresh_port_group(struct ofproto *, unsigned int group);
257 static void update_port(struct ofproto *, const char *devname);
258 static int init_ports(struct ofproto *);
259 static void reinit_ports(struct ofproto *);
262 ofproto_create(const char *datapath, const struct ofhooks *ofhooks, void *aux,
263 struct ofproto **ofprotop)
265 struct dpifmon *dpifmon;
266 struct odp_stats stats;
273 /* Connect to datapath and start listening for messages. */
274 error = dpif_open(datapath, &dpif);
276 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
279 error = dpif_get_dp_stats(&dpif, &stats);
281 VLOG_ERR("failed to obtain stats for datapath %s: %s",
282 datapath, strerror(error));
286 error = dpif_set_listen_mask(&dpif, ODPL_MISS | ODPL_ACTION);
288 VLOG_ERR("failed to listen on datapath %s: %s",
289 datapath, strerror(error));
293 dpif_flow_flush(&dpif);
296 /* Start monitoring datapath ports for status changes. */
297 error = dpifmon_create(datapath, &dpifmon);
299 VLOG_ERR("failed to starting monitoring datapath %s: %s",
300 datapath, strerror(error));
305 /* Initialize settings. */
306 p = xcalloc(1, sizeof *p);
307 p->fallback_dpid = pick_fallback_dpid();
308 p->datapath_id = pick_datapath_id(&dpif, p->fallback_dpid);
309 VLOG_INFO("using datapath ID %012"PRIx64, p->datapath_id);
310 p->manufacturer = xstrdup("Nicira Networks, Inc.");
311 p->hardware = xstrdup("Reference Implementation");
312 p->software = xstrdup(VERSION BUILDNR);
313 p->serial = xstrdup("None");
315 /* Initialize datapath. */
317 p->dpifmon = dpifmon;
318 port_array_init(&p->ports);
319 shash_init(&p->port_by_name);
320 p->max_ports = stats.max_ports;
322 /* Initialize submodules. */
323 p->switch_status = switch_status_create(p);
327 p->miss_sched = p->action_sched = NULL;
331 /* Initialize flow table. */
332 classifier_init(&p->cls);
333 p->need_revalidate = false;
334 p->next_expiration = time_msec() + 1000;
335 tag_set_init(&p->revalidate_set);
337 /* Initialize OpenFlow connections. */
338 list_init(&p->all_conns);
339 p->controller = ofconn_create(p, rconn_create(5, 8));
340 p->controller->pktbuf = pktbuf_create();
341 p->controller->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
347 /* Initialize hooks. */
349 p->ofhooks = ofhooks;
353 p->ofhooks = &default_ofhooks;
355 p->ml = mac_learning_create();
358 /* Register switch status category. */
359 p->ss_cat = switch_status_register(p->switch_status, "remote",
360 rconn_status_cb, p->controller->rconn);
363 error = init_ports(p);
374 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
376 uint64_t old_dpid = p->datapath_id;
377 p->datapath_id = (datapath_id
379 : pick_datapath_id(&p->dpif, p->fallback_dpid));
380 if (p->datapath_id != old_dpid) {
381 VLOG_INFO("datapath ID changed to %012"PRIx64, p->datapath_id);
382 rconn_reconnect(p->controller->rconn);
387 ofproto_set_mgmt_id(struct ofproto *p, uint64_t mgmt_id)
389 p->mgmt_id = mgmt_id;
393 ofproto_set_probe_interval(struct ofproto *p, int probe_interval)
395 probe_interval = probe_interval ? MAX(probe_interval, 5) : 0;
396 rconn_set_probe_interval(p->controller->rconn, probe_interval);
398 int trigger_duration = probe_interval ? probe_interval * 3 : 15;
399 fail_open_set_trigger_duration(p->fail_open, trigger_duration);
404 ofproto_set_max_backoff(struct ofproto *p, int max_backoff)
406 rconn_set_max_backoff(p->controller->rconn, max_backoff);
410 ofproto_set_desc(struct ofproto *p,
411 const char *manufacturer, const char *hardware,
412 const char *software, const char *serial)
415 free(p->manufacturer);
416 p->manufacturer = xstrdup(manufacturer);
420 p->hardware = xstrdup(hardware);
424 p->software = xstrdup(software);
428 p->serial = xstrdup(serial);
433 ofproto_set_in_band(struct ofproto *p, bool in_band)
435 if (in_band != (p->in_band != NULL)) {
437 in_band_create(p, &p->dpif, p->switch_status,
438 p->controller->rconn, &p->in_band);
441 ofproto_set_discovery(p, false, NULL, true);
442 in_band_destroy(p->in_band);
445 rconn_reconnect(p->controller->rconn);
451 ofproto_set_discovery(struct ofproto *p, bool discovery,
452 const char *re, bool update_resolv_conf)
454 if (discovery != (p->discovery != NULL)) {
456 int error = ofproto_set_in_band(p, true);
460 error = discovery_create(re, update_resolv_conf,
461 &p->dpif, p->switch_status,
467 discovery_destroy(p->discovery);
470 rconn_disconnect(p->controller->rconn);
471 } else if (discovery) {
472 discovery_set_update_resolv_conf(p->discovery, update_resolv_conf);
473 return discovery_set_accept_controller_re(p->discovery, re);
479 ofproto_set_controller(struct ofproto *ofproto, const char *controller)
481 if (ofproto->discovery) {
483 } else if (controller) {
484 if (strcmp(rconn_get_name(ofproto->controller->rconn), controller)) {
485 return rconn_connect(ofproto->controller->rconn, controller);
490 rconn_disconnect(ofproto->controller->rconn);
496 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
497 const struct svec *svec)
499 struct pvconn **pvconns = *pvconnsp;
500 size_t n_pvconns = *n_pvconnsp;
504 for (i = 0; i < n_pvconns; i++) {
505 pvconn_close(pvconns[i]);
509 pvconns = xmalloc(svec->n * sizeof *pvconns);
511 for (i = 0; i < svec->n; i++) {
512 const char *name = svec->names[i];
513 struct pvconn *pvconn;
516 error = pvconn_open(name, &pvconn);
518 pvconns[n_pvconns++] = pvconn;
520 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
528 *n_pvconnsp = n_pvconns;
534 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
536 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
540 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
542 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
546 ofproto_set_netflow(struct ofproto *ofproto,
547 const struct netflow_options *nf_options)
549 if (nf_options->collectors.n) {
550 if (!ofproto->netflow) {
551 ofproto->netflow = netflow_create();
553 return netflow_set_options(ofproto->netflow, nf_options);
555 netflow_destroy(ofproto->netflow);
556 ofproto->netflow = NULL;
562 ofproto_set_failure(struct ofproto *ofproto, bool fail_open)
565 struct rconn *rconn = ofproto->controller->rconn;
566 int trigger_duration = rconn_get_probe_interval(rconn) * 3;
567 if (!ofproto->fail_open) {
568 ofproto->fail_open = fail_open_create(ofproto, trigger_duration,
569 ofproto->switch_status,
572 fail_open_set_trigger_duration(ofproto->fail_open,
576 fail_open_destroy(ofproto->fail_open);
577 ofproto->fail_open = NULL;
582 ofproto_set_rate_limit(struct ofproto *ofproto,
583 int rate_limit, int burst_limit)
585 if (rate_limit > 0) {
586 if (!ofproto->miss_sched) {
587 ofproto->miss_sched = pinsched_create(rate_limit, burst_limit,
588 ofproto->switch_status);
589 ofproto->action_sched = pinsched_create(rate_limit, burst_limit,
592 pinsched_set_limits(ofproto->miss_sched, rate_limit, burst_limit);
593 pinsched_set_limits(ofproto->action_sched,
594 rate_limit, burst_limit);
597 pinsched_destroy(ofproto->miss_sched);
598 ofproto->miss_sched = NULL;
599 pinsched_destroy(ofproto->action_sched);
600 ofproto->action_sched = NULL;
605 ofproto_set_stp(struct ofproto *ofproto UNUSED, bool enable_stp)
609 VLOG_WARN("STP is not yet implemented");
617 ofproto_set_remote_execution(struct ofproto *ofproto, const char *command_acl,
618 const char *command_dir)
621 if (!ofproto->executer) {
622 return executer_create(command_acl, command_dir,
625 executer_set_acl(ofproto->executer, command_acl, command_dir);
628 executer_destroy(ofproto->executer);
629 ofproto->executer = NULL;
635 ofproto_get_datapath_id(const struct ofproto *ofproto)
637 return ofproto->datapath_id;
641 ofproto_get_mgmt_id(const struct ofproto *ofproto)
643 return ofproto->mgmt_id;
647 ofproto_get_probe_interval(const struct ofproto *ofproto)
649 return rconn_get_probe_interval(ofproto->controller->rconn);
653 ofproto_get_max_backoff(const struct ofproto *ofproto)
655 return rconn_get_max_backoff(ofproto->controller->rconn);
659 ofproto_get_in_band(const struct ofproto *ofproto)
661 return ofproto->in_band != NULL;
665 ofproto_get_discovery(const struct ofproto *ofproto)
667 return ofproto->discovery != NULL;
671 ofproto_get_controller(const struct ofproto *ofproto)
673 return rconn_get_name(ofproto->controller->rconn);
677 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
681 for (i = 0; i < ofproto->n_listeners; i++) {
682 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
687 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
691 for (i = 0; i < ofproto->n_snoops; i++) {
692 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
697 ofproto_destroy(struct ofproto *p)
699 struct ofconn *ofconn, *next_ofconn;
700 struct ofport *ofport;
701 unsigned int port_no;
708 ofproto_flush_flows(p);
709 classifier_destroy(&p->cls);
711 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
713 ofconn_destroy(ofconn, p);
716 dpif_close(&p->dpif);
717 dpifmon_destroy(p->dpifmon);
718 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
721 shash_destroy(&p->port_by_name);
723 switch_status_destroy(p->switch_status);
724 in_band_destroy(p->in_band);
725 discovery_destroy(p->discovery);
726 fail_open_destroy(p->fail_open);
727 pinsched_destroy(p->miss_sched);
728 pinsched_destroy(p->action_sched);
729 executer_destroy(p->executer);
730 netflow_destroy(p->netflow);
732 switch_status_unregister(p->ss_cat);
734 for (i = 0; i < p->n_listeners; i++) {
735 pvconn_close(p->listeners[i]);
739 for (i = 0; i < p->n_snoops; i++) {
740 pvconn_close(p->snoops[i]);
744 mac_learning_destroy(p->ml);
750 ofproto_run(struct ofproto *p)
752 int error = ofproto_run1(p);
754 error = ofproto_run2(p, false);
760 ofproto_run1(struct ofproto *p)
762 struct ofconn *ofconn, *next_ofconn;
767 for (i = 0; i < 50; i++) {
771 error = dpif_recv(&p->dpif, &buf);
773 if (error == ENODEV) {
774 /* Someone destroyed the datapath behind our back. The caller
775 * better destroy us and give up, because we're just going to
776 * spin from here on out. */
777 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
778 VLOG_ERR_RL(&rl, "dp%u: datapath was destroyed externally",
785 handle_odp_msg(p, buf);
788 while ((error = dpifmon_poll(p->dpifmon, &devname)) != EAGAIN) {
789 if (error == ENOBUFS) {
792 update_port(p, devname);
798 in_band_run(p->in_band);
801 char *controller_name;
802 if (rconn_is_connectivity_questionable(p->controller->rconn)) {
803 discovery_question_connectivity(p->discovery);
805 if (discovery_run(p->discovery, &controller_name)) {
806 if (controller_name) {
807 rconn_connect(p->controller->rconn, controller_name);
809 rconn_disconnect(p->controller->rconn);
813 pinsched_run(p->miss_sched, send_packet_in_miss, p);
814 pinsched_run(p->action_sched, send_packet_in_action, p);
816 executer_run(p->executer);
819 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
821 ofconn_run(ofconn, p);
824 /* Fail-open maintenance. Do this after processing the ofconns since
825 * fail-open checks the status of the controller rconn. */
827 fail_open_run(p->fail_open);
830 for (i = 0; i < p->n_listeners; i++) {
834 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
836 ofconn_create(p, rconn_new_from_vconn("passive", vconn));
837 } else if (retval != EAGAIN) {
838 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
842 for (i = 0; i < p->n_snoops; i++) {
846 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
848 rconn_add_monitor(p->controller->rconn, vconn);
849 } else if (retval != EAGAIN) {
850 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
854 if (time_msec() >= p->next_expiration) {
855 COVERAGE_INC(ofproto_expiration);
856 p->next_expiration = time_msec() + 1000;
859 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
861 /* Let the hook know that we're at a stable point: all outstanding data
862 * in existing flows has been accounted to the account_cb. Thus, the
863 * hook can now reasonably do operations that depend on having accurate
864 * flow volume accounting (currently, that's just bond rebalancing). */
865 if (p->ofhooks->account_checkpoint_cb) {
866 p->ofhooks->account_checkpoint_cb(p->aux);
871 netflow_run(p->netflow);
877 struct revalidate_cbdata {
878 struct ofproto *ofproto;
879 bool revalidate_all; /* Revalidate all exact-match rules? */
880 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
881 struct tag_set revalidate_set; /* Set of tags to revalidate. */
885 ofproto_run2(struct ofproto *p, bool revalidate_all)
887 if (p->need_revalidate || revalidate_all
888 || !tag_set_is_empty(&p->revalidate_set)) {
889 struct revalidate_cbdata cbdata;
891 cbdata.revalidate_all = revalidate_all;
892 cbdata.revalidate_subrules = p->need_revalidate;
893 cbdata.revalidate_set = p->revalidate_set;
894 tag_set_init(&p->revalidate_set);
895 COVERAGE_INC(ofproto_revalidate);
896 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
897 p->need_revalidate = false;
904 ofproto_wait(struct ofproto *p)
906 struct ofconn *ofconn;
909 dpif_recv_wait(&p->dpif);
910 dpifmon_wait(p->dpifmon);
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(p, 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 memcpy(ofport->opp.hw_addr, netdev_get_etheraddr(netdev), ETH_ALEN);
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 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no),
1194 shash_add(&p->port_by_name, (char *) ofport->opp.name, ofport);
1198 ofport_remove(struct ofproto *p, struct ofport *ofport)
1200 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no), NULL);
1201 shash_delete(&p->port_by_name,
1202 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1206 ofport_free(struct ofport *ofport)
1209 netdev_close(ofport->netdev);
1215 update_port(struct ofproto *p, const char *devname)
1217 struct odp_port odp_port;
1218 struct ofport *old_ofport;
1219 struct ofport *new_ofport;
1222 COVERAGE_INC(ofproto_update_port);
1224 /* Query the datapath for port information. */
1225 error = dpif_port_query_by_name(&p->dpif, devname, &odp_port);
1227 /* Find the old ofport. */
1228 old_ofport = shash_find_data(&p->port_by_name, devname);
1231 /* There's no port named 'devname' but there might be a port with
1232 * the same port number. This could happen if a port is deleted
1233 * and then a new one added in its place very quickly, or if a port
1234 * is renamed. In the former case we want to send an OFPPR_DELETE
1235 * and an OFPPR_ADD, and in the latter case we want to send a
1236 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1237 * the old port's ifindex against the new port, or perhaps less
1238 * reliably but more portably by comparing the old port's MAC
1239 * against the new port's MAC. However, this code isn't that smart
1240 * and always sends an OFPPR_MODIFY (XXX). */
1241 old_ofport = port_array_get(&p->ports, odp_port.port);
1243 } else if (error != ENOENT && error != ENODEV) {
1244 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1245 "%s", strerror(error));
1249 /* Create a new ofport. */
1250 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1252 /* Eliminate a few pathological cases. */
1253 if (!old_ofport && !new_ofport) {
1255 } else if (old_ofport && new_ofport) {
1256 /* Most of the 'config' bits are OpenFlow soft state, but
1257 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1258 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1259 * leaves the other bits 0.) */
1260 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1262 if (ofport_equal(old_ofport, new_ofport)) {
1263 /* False alarm--no change. */
1264 ofport_free(new_ofport);
1269 /* Now deal with the normal cases. */
1271 ofport_remove(p, old_ofport);
1274 ofport_install(p, new_ofport);
1276 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1277 (!old_ofport ? OFPPR_ADD
1278 : !new_ofport ? OFPPR_DELETE
1280 ofport_free(old_ofport);
1282 /* Update port groups. */
1283 refresh_port_groups(p);
1287 init_ports(struct ofproto *p)
1289 struct odp_port *ports;
1294 error = dpif_port_list(&p->dpif, &ports, &n_ports);
1299 for (i = 0; i < n_ports; i++) {
1300 const struct odp_port *odp_port = &ports[i];
1301 if (!ofport_conflicts(p, odp_port)) {
1302 struct ofport *ofport = make_ofport(odp_port);
1304 ofport_install(p, ofport);
1309 refresh_port_groups(p);
1313 static struct ofconn *
1314 ofconn_create(struct ofproto *p, struct rconn *rconn)
1316 struct ofconn *ofconn = xmalloc(sizeof *ofconn);
1317 list_push_back(&p->all_conns, &ofconn->node);
1318 ofconn->rconn = rconn;
1319 ofconn->pktbuf = NULL;
1320 ofconn->send_flow_exp = false;
1321 ofconn->miss_send_len = 0;
1322 ofconn->packet_in_counter = rconn_packet_counter_create ();
1323 ofconn->reply_counter = rconn_packet_counter_create ();
1328 ofconn_destroy(struct ofconn *ofconn, struct ofproto *p)
1331 executer_rconn_closing(p->executer, ofconn->rconn);
1334 list_remove(&ofconn->node);
1335 rconn_destroy(ofconn->rconn);
1336 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1337 rconn_packet_counter_destroy(ofconn->reply_counter);
1338 pktbuf_destroy(ofconn->pktbuf);
1343 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1347 rconn_run(ofconn->rconn);
1349 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1350 /* Limit the number of iterations to prevent other tasks from
1352 for (iteration = 0; iteration < 50; iteration++) {
1353 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1358 fail_open_maybe_recover(p->fail_open);
1360 handle_openflow(ofconn, p, of_msg);
1361 ofpbuf_delete(of_msg);
1365 if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
1366 ofconn_destroy(ofconn, p);
1371 ofconn_wait(struct ofconn *ofconn)
1373 rconn_run_wait(ofconn->rconn);
1374 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1375 rconn_recv_wait(ofconn->rconn);
1377 COVERAGE_INC(ofproto_ofconn_stuck);
1381 /* Caller is responsible for initializing the 'cr' member of the returned
1383 static struct rule *
1384 rule_create(struct ofproto *ofproto, struct rule *super,
1385 const union ofp_action *actions, size_t n_actions,
1386 uint16_t idle_timeout, uint16_t hard_timeout)
1388 struct rule *rule = xcalloc(1, sizeof *rule);
1389 rule->idle_timeout = idle_timeout;
1390 rule->hard_timeout = hard_timeout;
1391 rule->used = rule->created = time_msec();
1392 rule->super = super;
1394 list_push_back(&super->list, &rule->list);
1396 list_init(&rule->list);
1398 rule->n_actions = n_actions;
1399 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1400 netflow_flow_clear(&rule->nf_flow);
1401 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1406 static struct rule *
1407 rule_from_cls_rule(const struct cls_rule *cls_rule)
1409 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1413 rule_free(struct rule *rule)
1415 free(rule->actions);
1416 free(rule->odp_actions);
1420 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1421 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1422 * through all of its subrules and revalidates them, destroying any that no
1423 * longer has a super-rule (which is probably all of them).
1425 * Before calling this function, the caller must make have removed 'rule' from
1426 * the classifier. If 'rule' is an exact-match rule, the caller is also
1427 * responsible for ensuring that it has been uninstalled from the datapath. */
1429 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1432 struct rule *subrule, *next;
1433 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1434 revalidate_rule(ofproto, subrule);
1437 list_remove(&rule->list);
1443 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1445 const union ofp_action *oa;
1446 struct actions_iterator i;
1448 if (out_port == htons(OFPP_NONE)) {
1451 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1452 oa = actions_next(&i)) {
1453 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1460 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1461 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1463 * The flow that 'packet' actually contains does not need to actually match
1464 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1465 * the packet and byte counters for 'rule' will be credited for the packet sent
1466 * out whether or not the packet actually matches 'rule'.
1468 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1469 * the caller must already have accurately composed ODP actions for it given
1470 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1471 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1472 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1473 * actions and apply them to 'packet'. */
1475 rule_execute(struct ofproto *ofproto, struct rule *rule,
1476 struct ofpbuf *packet, const flow_t *flow)
1478 const union odp_action *actions;
1480 struct odp_actions a;
1482 /* Grab or compose the ODP actions.
1484 * The special case for an exact-match 'rule' where 'flow' is not the
1485 * rule's flow is important to avoid, e.g., sending a packet out its input
1486 * port simply because the ODP actions were composed for the wrong
1488 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1489 struct rule *super = rule->super ? rule->super : rule;
1490 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1491 packet, &a, NULL, 0, NULL)) {
1494 actions = a.actions;
1495 n_actions = a.n_actions;
1497 actions = rule->odp_actions;
1498 n_actions = rule->n_odp_actions;
1501 /* Execute the ODP actions. */
1502 if (!dpif_execute(&ofproto->dpif, flow->in_port,
1503 actions, n_actions, packet)) {
1504 struct odp_flow_stats stats;
1505 flow_extract_stats(flow, packet, &stats);
1506 update_stats(ofproto, rule, &stats);
1507 rule->used = time_msec();
1508 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
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(ofproto, 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,
1589 &rule->nf_flow.output_iface);
1591 actions_len = a.n_actions * sizeof *a.actions;
1592 if (rule->n_odp_actions != a.n_actions
1593 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1594 COVERAGE_INC(ofproto_odp_unchanged);
1595 free(rule->odp_actions);
1596 rule->n_odp_actions = a.n_actions;
1597 rule->odp_actions = xmemdup(a.actions, actions_len);
1605 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1606 struct odp_flow_put *put)
1608 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1609 put->flow.key = rule->cr.flow;
1610 put->flow.actions = rule->odp_actions;
1611 put->flow.n_actions = rule->n_odp_actions;
1613 return dpif_flow_put(&ofproto->dpif, put);
1617 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1619 assert(!rule->cr.wc.wildcards);
1621 if (rule->may_install) {
1622 struct odp_flow_put put;
1623 if (!do_put_flow(p, rule,
1624 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1626 rule->installed = true;
1627 if (displaced_rule) {
1628 update_stats(p, rule, &put.flow.stats);
1629 rule_post_uninstall(p, displaced_rule);
1632 } else if (displaced_rule) {
1633 rule_uninstall(p, displaced_rule);
1638 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1640 if (rule->installed) {
1641 struct odp_flow_put put;
1642 COVERAGE_INC(ofproto_dp_missed);
1643 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1645 rule_install(ofproto, rule, NULL);
1650 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1652 bool actions_changed = rule_make_actions(ofproto, rule, NULL);
1653 if (rule->may_install) {
1654 if (rule->installed) {
1655 if (actions_changed) {
1656 /* XXX should really do rule_post_uninstall() for the *old* set
1657 * of actions, and distinguish the old stats from the new. */
1658 struct odp_flow_put put;
1659 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1662 rule_install(ofproto, rule, NULL);
1665 rule_uninstall(ofproto, rule);
1670 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1672 uint64_t total_bytes = rule->byte_count + extra_bytes;
1674 if (ofproto->ofhooks->account_flow_cb
1675 && total_bytes > rule->accounted_bytes)
1677 ofproto->ofhooks->account_flow_cb(
1678 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1679 total_bytes - rule->accounted_bytes, ofproto->aux);
1680 rule->accounted_bytes = total_bytes;
1685 rule_uninstall(struct ofproto *p, struct rule *rule)
1687 assert(!rule->cr.wc.wildcards);
1688 if (rule->installed) {
1689 struct odp_flow odp_flow;
1691 odp_flow.key = rule->cr.flow;
1692 odp_flow.actions = NULL;
1693 odp_flow.n_actions = 0;
1694 if (!dpif_flow_del(&p->dpif, &odp_flow)) {
1695 update_stats(p, rule, &odp_flow.stats);
1697 rule->installed = false;
1699 rule_post_uninstall(p, rule);
1704 is_controller_rule(struct rule *rule)
1706 /* If the only action is send to the controller then don't report
1707 * NetFlow expiration messages since it is just part of the control
1708 * logic for the network and not real traffic. */
1710 if (rule && rule->super) {
1711 struct rule *super = rule->super;
1713 return super->n_actions == 1 &&
1714 super->actions[0].type == htons(OFPAT_OUTPUT) &&
1715 super->actions[0].output.port == htons(OFPP_CONTROLLER);
1722 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1724 struct rule *super = rule->super;
1726 rule_account(ofproto, rule, 0);
1728 if (ofproto->netflow && !is_controller_rule(rule)) {
1729 struct ofexpired expired;
1730 expired.flow = rule->cr.flow;
1731 expired.packet_count = rule->packet_count;
1732 expired.byte_count = rule->byte_count;
1733 expired.used = rule->used;
1734 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
1737 super->packet_count += rule->packet_count;
1738 super->byte_count += rule->byte_count;
1740 /* Reset counters to prevent double counting if the rule ever gets
1742 rule->packet_count = 0;
1743 rule->byte_count = 0;
1744 rule->accounted_bytes = 0;
1746 netflow_flow_clear(&rule->nf_flow);
1751 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1752 struct rconn_packet_counter *counter)
1754 update_openflow_length(msg);
1755 if (rconn_send(ofconn->rconn, msg, counter)) {
1761 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1762 int error, const void *data, size_t len)
1765 struct ofp_error_msg *oem;
1767 if (!(error >> 16)) {
1768 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1773 COVERAGE_INC(ofproto_error);
1774 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1775 oh ? oh->xid : 0, &buf);
1776 oem->type = htons((unsigned int) error >> 16);
1777 oem->code = htons(error & 0xffff);
1778 memcpy(oem->data, data, len);
1779 queue_tx(buf, ofconn, ofconn->reply_counter);
1783 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1786 size_t oh_length = ntohs(oh->length);
1787 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1791 hton_ofp_phy_port(struct ofp_phy_port *opp)
1793 opp->port_no = htons(opp->port_no);
1794 opp->config = htonl(opp->config);
1795 opp->state = htonl(opp->state);
1796 opp->curr = htonl(opp->curr);
1797 opp->advertised = htonl(opp->advertised);
1798 opp->supported = htonl(opp->supported);
1799 opp->peer = htonl(opp->peer);
1803 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
1805 struct ofp_header *rq = oh;
1806 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
1811 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
1812 struct ofp_header *oh)
1814 struct ofp_switch_features *osf;
1816 unsigned int port_no;
1817 struct ofport *port;
1819 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1820 osf->datapath_id = htonll(p->datapath_id);
1821 osf->n_buffers = htonl(pktbuf_capacity());
1823 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1824 OFPC_PORT_STATS | OFPC_MULTI_PHY_TX);
1825 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1826 (1u << OFPAT_SET_VLAN_VID) |
1827 (1u << OFPAT_SET_VLAN_PCP) |
1828 (1u << OFPAT_STRIP_VLAN) |
1829 (1u << OFPAT_SET_DL_SRC) |
1830 (1u << OFPAT_SET_DL_DST) |
1831 (1u << OFPAT_SET_NW_SRC) |
1832 (1u << OFPAT_SET_NW_DST) |
1833 (1u << OFPAT_SET_TP_SRC) |
1834 (1u << OFPAT_SET_TP_DST));
1836 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1837 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1840 queue_tx(buf, ofconn, ofconn->reply_counter);
1845 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1846 struct ofp_header *oh)
1849 struct ofp_switch_config *osc;
1853 /* Figure out flags. */
1854 dpif_get_drop_frags(&p->dpif, &drop_frags);
1855 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1856 if (ofconn->send_flow_exp) {
1857 flags |= OFPC_SEND_FLOW_EXP;
1861 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1862 osc->flags = htons(flags);
1863 osc->miss_send_len = htons(ofconn->miss_send_len);
1864 queue_tx(buf, ofconn, ofconn->reply_counter);
1870 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1871 struct ofp_switch_config *osc)
1876 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1880 flags = ntohs(osc->flags);
1882 ofconn->send_flow_exp = (flags & OFPC_SEND_FLOW_EXP) != 0;
1884 if (ofconn == p->controller) {
1885 switch (flags & OFPC_FRAG_MASK) {
1886 case OFPC_FRAG_NORMAL:
1887 dpif_set_drop_frags(&p->dpif, false);
1889 case OFPC_FRAG_DROP:
1890 dpif_set_drop_frags(&p->dpif, true);
1893 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1899 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1900 if (ntohs(osc->miss_send_len) != 0) {
1901 ofconn->pktbuf = pktbuf_create();
1903 pktbuf_destroy(ofconn->pktbuf);
1907 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1913 add_output_group_action(struct odp_actions *actions, uint16_t group,
1914 uint16_t *nf_output_iface)
1916 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1918 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
1919 *nf_output_iface = NF_OUT_FLOOD;
1924 add_controller_action(struct odp_actions *actions,
1925 const struct ofp_action_output *oao)
1927 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
1928 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
1931 struct action_xlate_ctx {
1933 const flow_t *flow; /* Flow to which these actions correspond. */
1934 int recurse; /* Recursion level, via xlate_table_action. */
1935 struct ofproto *ofproto;
1936 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
1937 * null pointer if we are revalidating
1938 * without a packet to refer to. */
1941 struct odp_actions *out; /* Datapath actions. */
1942 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
1943 bool may_set_up_flow; /* True ordinarily; false if the actions must
1944 * be reassessed for every packet. */
1945 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
1948 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
1949 struct action_xlate_ctx *ctx);
1952 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
1954 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
1957 if (ofport->opp.config & OFPPC_NO_FWD) {
1958 /* Forwarding disabled on port. */
1963 * We don't have an ofport record for this port, but it doesn't hurt to
1964 * allow forwarding to it anyhow. Maybe such a port will appear later
1965 * and we're pre-populating the flow table.
1969 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
1970 ctx->nf_output_iface = port;
1973 static struct rule *
1974 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
1977 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
1979 /* The rule we found might not be valid, since we could be in need of
1980 * revalidation. If it is not valid, don't return it. */
1983 && ofproto->need_revalidate
1984 && !revalidate_rule(ofproto, rule)) {
1985 COVERAGE_INC(ofproto_invalidated);
1993 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
1995 if (!ctx->recurse) {
2000 flow.in_port = in_port;
2002 rule = lookup_valid_rule(ctx->ofproto, &flow);
2009 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2016 xlate_output_action(struct action_xlate_ctx *ctx,
2017 const struct ofp_action_output *oao)
2020 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2022 ctx->nf_output_iface = NF_OUT_DROP;
2024 switch (ntohs(oao->port)) {
2026 add_output_action(ctx, ctx->flow->in_port);
2029 xlate_table_action(ctx, ctx->flow->in_port);
2032 if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet,
2033 ctx->out, ctx->tags,
2034 &ctx->nf_output_iface,
2035 ctx->ofproto->aux)) {
2036 COVERAGE_INC(ofproto_uninstallable);
2037 ctx->may_set_up_flow = false;
2041 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2042 &ctx->nf_output_iface);
2045 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2047 case OFPP_CONTROLLER:
2048 add_controller_action(ctx->out, oao);
2051 add_output_action(ctx, ODPP_LOCAL);
2054 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2055 if (odp_port != ctx->flow->in_port) {
2056 add_output_action(ctx, odp_port);
2061 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2062 ctx->nf_output_iface = NF_OUT_FLOOD;
2063 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2064 ctx->nf_output_iface = prev_nf_output_iface;
2065 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2066 ctx->nf_output_iface != NF_OUT_FLOOD) {
2067 ctx->nf_output_iface = NF_OUT_MULTI;
2072 xlate_nicira_action(struct action_xlate_ctx *ctx,
2073 const struct nx_action_header *nah)
2075 const struct nx_action_resubmit *nar;
2076 int subtype = ntohs(nah->subtype);
2078 assert(nah->vendor == htonl(NX_VENDOR_ID));
2080 case NXAST_RESUBMIT:
2081 nar = (const struct nx_action_resubmit *) nah;
2082 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2086 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2092 do_xlate_actions(const union ofp_action *in, size_t n_in,
2093 struct action_xlate_ctx *ctx)
2095 struct actions_iterator iter;
2096 const union ofp_action *ia;
2097 const struct ofport *port;
2099 port = port_array_get(&ctx->ofproto->ports, ctx->flow->in_port);
2100 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2101 port->opp.config & (eth_addr_equals(ctx->flow->dl_dst, stp_eth_addr)
2102 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2103 /* Drop this flow. */
2107 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2108 uint16_t type = ntohs(ia->type);
2109 union odp_action *oa;
2113 xlate_output_action(ctx, &ia->output);
2116 case OFPAT_SET_VLAN_VID:
2117 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2118 oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2121 case OFPAT_SET_VLAN_PCP:
2122 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2123 oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2126 case OFPAT_STRIP_VLAN:
2127 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2130 case OFPAT_SET_DL_SRC:
2131 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2132 memcpy(oa->dl_addr.dl_addr,
2133 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2136 case OFPAT_SET_DL_DST:
2137 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2138 memcpy(oa->dl_addr.dl_addr,
2139 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2142 case OFPAT_SET_NW_SRC:
2143 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2144 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2147 case OFPAT_SET_TP_SRC:
2148 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2149 oa->tp_port.tp_port = ia->tp_port.tp_port;
2153 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2157 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2164 xlate_actions(const union ofp_action *in, size_t n_in,
2165 const flow_t *flow, struct ofproto *ofproto,
2166 const struct ofpbuf *packet,
2167 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2168 uint16_t *nf_output_iface)
2170 tag_type no_tags = 0;
2171 struct action_xlate_ctx ctx;
2172 COVERAGE_INC(ofproto_ofp2odp);
2173 odp_actions_init(out);
2176 ctx.ofproto = ofproto;
2177 ctx.packet = packet;
2179 ctx.tags = tags ? tags : &no_tags;
2180 ctx.may_set_up_flow = true;
2181 ctx.nf_output_iface = NF_OUT_DROP;
2182 do_xlate_actions(in, n_in, &ctx);
2184 /* Check with in-band control to see if we're allowed to set up this
2186 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2187 ctx.may_set_up_flow = false;
2190 if (may_set_up_flow) {
2191 *may_set_up_flow = ctx.may_set_up_flow;
2193 if (nf_output_iface) {
2194 *nf_output_iface = ctx.nf_output_iface;
2196 if (odp_actions_overflow(out)) {
2197 odp_actions_init(out);
2198 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2204 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2205 struct ofp_header *oh)
2207 struct ofp_packet_out *opo;
2208 struct ofpbuf payload, *buffer;
2209 struct odp_actions actions;
2215 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2219 opo = (struct ofp_packet_out *) oh;
2221 COVERAGE_INC(ofproto_packet_out);
2222 if (opo->buffer_id != htonl(UINT32_MAX)) {
2223 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2225 if (error || !buffer) {
2233 flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2234 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2235 &flow, p, &payload, &actions, NULL, NULL, NULL);
2240 dpif_execute(&p->dpif, flow.in_port, actions.actions, actions.n_actions,
2242 ofpbuf_delete(buffer);
2248 update_port_config(struct ofproto *p, struct ofport *port,
2249 uint32_t config, uint32_t mask)
2251 mask &= config ^ port->opp.config;
2252 if (mask & OFPPC_PORT_DOWN) {
2253 if (config & OFPPC_PORT_DOWN) {
2254 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2256 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2259 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2260 if (mask & REVALIDATE_BITS) {
2261 COVERAGE_INC(ofproto_costly_flags);
2262 port->opp.config ^= mask & REVALIDATE_BITS;
2263 p->need_revalidate = true;
2265 #undef REVALIDATE_BITS
2266 if (mask & OFPPC_NO_FLOOD) {
2267 port->opp.config ^= OFPPC_NO_FLOOD;
2268 refresh_port_group(p, DP_GROUP_FLOOD);
2270 if (mask & OFPPC_NO_PACKET_IN) {
2271 port->opp.config ^= OFPPC_NO_PACKET_IN;
2276 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2278 const struct ofp_port_mod *opm;
2279 struct ofport *port;
2282 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2286 opm = (struct ofp_port_mod *) oh;
2288 port = port_array_get(&p->ports,
2289 ofp_port_to_odp_port(ntohs(opm->port_no)));
2291 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2292 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2293 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2295 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2296 if (opm->advertise) {
2297 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2303 static struct ofpbuf *
2304 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2306 struct ofp_stats_reply *osr;
2309 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2310 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2312 osr->flags = htons(0);
2316 static struct ofpbuf *
2317 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2319 return make_stats_reply(request->header.xid, request->type, body_len);
2323 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2325 struct ofpbuf *msg = *msgp;
2326 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2327 if (nbytes + msg->size > UINT16_MAX) {
2328 struct ofp_stats_reply *reply = msg->data;
2329 reply->flags = htons(OFPSF_REPLY_MORE);
2330 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2331 queue_tx(msg, ofconn, ofconn->reply_counter);
2333 return ofpbuf_put_uninit(*msgp, nbytes);
2337 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2338 struct ofp_stats_request *request)
2340 struct ofp_desc_stats *ods;
2343 msg = start_stats_reply(request, sizeof *ods);
2344 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2345 strncpy(ods->mfr_desc, p->manufacturer, sizeof ods->mfr_desc);
2346 strncpy(ods->hw_desc, p->hardware, sizeof ods->hw_desc);
2347 strncpy(ods->sw_desc, p->software, sizeof ods->sw_desc);
2348 strncpy(ods->serial_num, p->serial, sizeof ods->serial_num);
2349 queue_tx(msg, ofconn, ofconn->reply_counter);
2355 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2357 struct rule *rule = rule_from_cls_rule(cls_rule);
2358 int *n_subrules = n_subrules_;
2366 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2367 struct ofp_stats_request *request)
2369 struct ofp_table_stats *ots;
2371 struct odp_stats dpstats;
2372 int n_exact, n_subrules, n_wild;
2374 msg = start_stats_reply(request, sizeof *ots * 2);
2376 /* Count rules of various kinds. */
2378 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2379 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2380 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2383 dpif_get_dp_stats(&p->dpif, &dpstats);
2384 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2385 memset(ots, 0, sizeof *ots);
2386 ots->table_id = TABLEID_HASH;
2387 strcpy(ots->name, "hash");
2388 ots->wildcards = htonl(0);
2389 ots->max_entries = htonl(dpstats.max_capacity);
2390 ots->active_count = htonl(n_exact);
2391 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2393 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2395 /* Classifier table. */
2396 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2397 memset(ots, 0, sizeof *ots);
2398 ots->table_id = TABLEID_CLASSIFIER;
2399 strcpy(ots->name, "classifier");
2400 ots->wildcards = htonl(OFPFW_ALL);
2401 ots->max_entries = htonl(65536);
2402 ots->active_count = htonl(n_wild);
2403 ots->lookup_count = htonll(0); /* XXX */
2404 ots->matched_count = htonll(0); /* XXX */
2406 queue_tx(msg, ofconn, ofconn->reply_counter);
2411 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2412 struct ofp_stats_request *request)
2414 struct ofp_port_stats *ops;
2416 struct ofport *port;
2417 unsigned int port_no;
2419 msg = start_stats_reply(request, sizeof *ops * 16);
2420 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2421 struct netdev_stats stats;
2423 /* Intentionally ignore return value, since errors will set 'stats' to
2424 * all-1s, which is correct for OpenFlow, and netdev_get_stats() will
2426 netdev_get_stats(port->netdev, &stats);
2428 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2429 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2430 memset(ops->pad, 0, sizeof ops->pad);
2431 ops->rx_packets = htonll(stats.rx_packets);
2432 ops->tx_packets = htonll(stats.tx_packets);
2433 ops->rx_bytes = htonll(stats.rx_bytes);
2434 ops->tx_bytes = htonll(stats.tx_bytes);
2435 ops->rx_dropped = htonll(stats.rx_dropped);
2436 ops->tx_dropped = htonll(stats.tx_dropped);
2437 ops->rx_errors = htonll(stats.rx_errors);
2438 ops->tx_errors = htonll(stats.tx_errors);
2439 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2440 ops->rx_over_err = htonll(stats.rx_over_errors);
2441 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2442 ops->collisions = htonll(stats.collisions);
2445 queue_tx(msg, ofconn, ofconn->reply_counter);
2449 struct flow_stats_cbdata {
2450 struct ofproto *ofproto;
2451 struct ofconn *ofconn;
2457 query_stats(struct ofproto *p, struct rule *rule,
2458 uint64_t *packet_countp, uint64_t *byte_countp)
2460 uint64_t packet_count, byte_count;
2461 struct rule *subrule;
2462 struct odp_flow *odp_flows;
2465 packet_count = rule->packet_count;
2466 byte_count = rule->byte_count;
2468 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2469 odp_flows = xcalloc(1, n_odp_flows * sizeof *odp_flows);
2470 if (rule->cr.wc.wildcards) {
2472 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2473 odp_flows[i++].key = subrule->cr.flow;
2474 packet_count += subrule->packet_count;
2475 byte_count += subrule->byte_count;
2478 odp_flows[0].key = rule->cr.flow;
2481 if (!dpif_flow_get_multiple(&p->dpif, odp_flows, n_odp_flows)) {
2483 for (i = 0; i < n_odp_flows; i++) {
2484 struct odp_flow *odp_flow = &odp_flows[i];
2485 packet_count += odp_flow->stats.n_packets;
2486 byte_count += odp_flow->stats.n_bytes;
2491 *packet_countp = packet_count;
2492 *byte_countp = byte_count;
2496 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2498 struct rule *rule = rule_from_cls_rule(rule_);
2499 struct flow_stats_cbdata *cbdata = cbdata_;
2500 struct ofp_flow_stats *ofs;
2501 uint64_t packet_count, byte_count;
2502 size_t act_len, len;
2504 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2508 act_len = sizeof *rule->actions * rule->n_actions;
2509 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2511 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2513 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2514 ofs->length = htons(len);
2515 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2517 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match);
2518 ofs->duration = htonl((time_msec() - rule->created) / 1000);
2519 ofs->priority = htons(rule->cr.priority);
2520 ofs->idle_timeout = htons(rule->idle_timeout);
2521 ofs->hard_timeout = htons(rule->hard_timeout);
2522 memset(ofs->pad2, 0, sizeof ofs->pad2);
2523 ofs->packet_count = htonll(packet_count);
2524 ofs->byte_count = htonll(byte_count);
2525 memcpy(ofs->actions, rule->actions, act_len);
2529 table_id_to_include(uint8_t table_id)
2531 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2532 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2533 : table_id == 0xff ? CLS_INC_ALL
2538 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2539 const struct ofp_stats_request *osr,
2542 struct ofp_flow_stats_request *fsr;
2543 struct flow_stats_cbdata cbdata;
2544 struct cls_rule target;
2546 if (arg_size != sizeof *fsr) {
2547 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2549 fsr = (struct ofp_flow_stats_request *) osr->body;
2551 COVERAGE_INC(ofproto_flows_req);
2553 cbdata.ofconn = ofconn;
2554 cbdata.out_port = fsr->out_port;
2555 cbdata.msg = start_stats_reply(osr, 1024);
2556 cls_rule_from_match(&target, &fsr->match, 0);
2557 classifier_for_each_match(&p->cls, &target,
2558 table_id_to_include(fsr->table_id),
2559 flow_stats_cb, &cbdata);
2560 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2564 struct flow_stats_ds_cbdata {
2565 struct ofproto *ofproto;
2570 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
2572 struct rule *rule = rule_from_cls_rule(rule_);
2573 struct flow_stats_ds_cbdata *cbdata = cbdata_;
2574 struct ds *results = cbdata->results;
2575 struct ofp_match match;
2576 uint64_t packet_count, byte_count;
2577 size_t act_len = sizeof *rule->actions * rule->n_actions;
2579 /* Don't report on subrules. */
2580 if (rule->super != NULL) {
2584 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2585 flow_to_ovs_match(&rule->cr.flow, rule->cr.wc.wildcards, &match);
2587 ds_put_format(results, "duration=%llds, ",
2588 (time_msec() - rule->created) / 1000);
2589 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2590 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2591 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2592 ofp_print_match(results, &match, true);
2593 ofp_print_actions(results, &rule->actions->header, act_len);
2594 ds_put_cstr(results, "\n");
2597 /* Adds a pretty-printed description of all flows to 'results', including
2598 * those marked hidden by secchan (e.g., by in-band control). */
2600 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2602 struct ofp_match match;
2603 struct cls_rule target;
2604 struct flow_stats_ds_cbdata cbdata;
2606 memset(&match, 0, sizeof match);
2607 match.wildcards = htonl(OFPFW_ALL);
2610 cbdata.results = results;
2612 cls_rule_from_match(&target, &match, 0);
2613 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2614 flow_stats_ds_cb, &cbdata);
2617 struct aggregate_stats_cbdata {
2618 struct ofproto *ofproto;
2620 uint64_t packet_count;
2621 uint64_t byte_count;
2626 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2628 struct rule *rule = rule_from_cls_rule(rule_);
2629 struct aggregate_stats_cbdata *cbdata = cbdata_;
2630 uint64_t packet_count, byte_count;
2632 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2636 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2638 cbdata->packet_count += packet_count;
2639 cbdata->byte_count += byte_count;
2644 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2645 const struct ofp_stats_request *osr,
2648 struct ofp_aggregate_stats_request *asr;
2649 struct ofp_aggregate_stats_reply *reply;
2650 struct aggregate_stats_cbdata cbdata;
2651 struct cls_rule target;
2654 if (arg_size != sizeof *asr) {
2655 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2657 asr = (struct ofp_aggregate_stats_request *) osr->body;
2659 COVERAGE_INC(ofproto_agg_request);
2661 cbdata.out_port = asr->out_port;
2662 cbdata.packet_count = 0;
2663 cbdata.byte_count = 0;
2665 cls_rule_from_match(&target, &asr->match, 0);
2666 classifier_for_each_match(&p->cls, &target,
2667 table_id_to_include(asr->table_id),
2668 aggregate_stats_cb, &cbdata);
2670 msg = start_stats_reply(osr, sizeof *reply);
2671 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2672 reply->flow_count = htonl(cbdata.n_flows);
2673 reply->packet_count = htonll(cbdata.packet_count);
2674 reply->byte_count = htonll(cbdata.byte_count);
2675 queue_tx(msg, ofconn, ofconn->reply_counter);
2680 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2681 struct ofp_header *oh)
2683 struct ofp_stats_request *osr;
2687 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2692 osr = (struct ofp_stats_request *) oh;
2694 switch (ntohs(osr->type)) {
2696 return handle_desc_stats_request(p, ofconn, osr);
2699 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2701 case OFPST_AGGREGATE:
2702 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2705 return handle_table_stats_request(p, ofconn, osr);
2708 return handle_port_stats_request(p, ofconn, osr);
2711 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2714 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2718 static long long int
2719 msec_from_nsec(uint64_t sec, uint32_t nsec)
2721 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2725 update_time(struct ofproto *ofproto, struct rule *rule,
2726 const struct odp_flow_stats *stats)
2728 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2729 if (used > rule->used) {
2731 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
2736 update_stats(struct ofproto *ofproto, struct rule *rule,
2737 const struct odp_flow_stats *stats)
2739 if (stats->n_packets) {
2740 update_time(ofproto, rule, stats);
2741 rule->packet_count += stats->n_packets;
2742 rule->byte_count += stats->n_bytes;
2743 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
2749 add_flow(struct ofproto *p, struct ofconn *ofconn,
2750 struct ofp_flow_mod *ofm, size_t n_actions)
2752 struct ofpbuf *packet;
2757 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
2758 n_actions, ntohs(ofm->idle_timeout),
2759 ntohs(ofm->hard_timeout));
2760 cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority));
2764 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2765 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2769 rule_insert(p, rule, packet, in_port);
2770 ofpbuf_delete(packet);
2775 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
2776 size_t n_actions, uint16_t command, struct rule *rule)
2778 if (rule_is_hidden(rule)) {
2782 if (command == OFPFC_DELETE) {
2783 rule_remove(p, rule);
2785 size_t actions_len = n_actions * sizeof *rule->actions;
2787 if (n_actions == rule->n_actions
2788 && !memcmp(ofm->actions, rule->actions, actions_len))
2793 free(rule->actions);
2794 rule->actions = xmemdup(ofm->actions, actions_len);
2795 rule->n_actions = n_actions;
2797 if (rule->cr.wc.wildcards) {
2798 COVERAGE_INC(ofproto_mod_wc_flow);
2799 p->need_revalidate = true;
2801 rule_update_actions(p, rule);
2809 modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm,
2810 size_t n_actions, uint16_t command)
2816 flow_from_match(&flow, &wildcards, &ofm->match);
2817 rule = rule_from_cls_rule(classifier_find_rule_exactly(
2818 &p->cls, &flow, wildcards,
2819 ntohs(ofm->priority)));
2822 if (command == OFPFC_DELETE
2823 && ofm->out_port != htons(OFPP_NONE)
2824 && !rule_has_out_port(rule, ofm->out_port)) {
2828 modify_flow(p, ofm, n_actions, command, rule);
2833 struct modify_flows_cbdata {
2834 struct ofproto *ofproto;
2835 const struct ofp_flow_mod *ofm;
2842 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
2844 struct rule *rule = rule_from_cls_rule(rule_);
2845 struct modify_flows_cbdata *cbdata = cbdata_;
2847 if (cbdata->out_port != htons(OFPP_NONE)
2848 && !rule_has_out_port(rule, cbdata->out_port)) {
2852 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions,
2853 cbdata->command, rule);
2857 modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm,
2858 size_t n_actions, uint16_t command)
2860 struct modify_flows_cbdata cbdata;
2861 struct cls_rule target;
2865 cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port
2866 : htons(OFPP_NONE));
2867 cbdata.n_actions = n_actions;
2868 cbdata.command = command;
2870 cls_rule_from_match(&target, &ofm->match, 0);
2872 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2873 modify_flows_cb, &cbdata);
2878 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
2879 struct ofp_flow_mod *ofm)
2884 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
2885 sizeof *ofm->actions, &n_actions);
2890 normalize_match(&ofm->match);
2891 if (!ofm->match.wildcards) {
2892 ofm->priority = htons(UINT16_MAX);
2895 error = validate_actions((const union ofp_action *) ofm->actions,
2896 n_actions, p->max_ports);
2901 switch (ntohs(ofm->command)) {
2903 return add_flow(p, ofconn, ofm, n_actions);
2906 return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY);
2908 case OFPFC_MODIFY_STRICT:
2909 return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY);
2912 return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE);
2914 case OFPFC_DELETE_STRICT:
2915 return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE);
2918 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
2923 send_capability_reply(struct ofproto *p, struct ofconn *ofconn, uint32_t xid)
2925 struct ofmp_capability_reply *ocr;
2927 char capabilities[] = "com.nicira.mgmt.manager=false\n";
2929 ocr = make_openflow_xid(sizeof(*ocr), OFPT_VENDOR, xid, &b);
2930 ocr->header.header.vendor = htonl(NX_VENDOR_ID);
2931 ocr->header.header.subtype = htonl(NXT_MGMT);
2932 ocr->header.type = htons(OFMPT_CAPABILITY_REPLY);
2934 ocr->format = htonl(OFMPCOF_SIMPLE);
2935 ocr->mgmt_id = htonll(p->mgmt_id);
2937 ofpbuf_put(b, capabilities, strlen(capabilities));
2939 queue_tx(b, ofconn, ofconn->reply_counter);
2943 handle_ofmp(struct ofproto *p, struct ofconn *ofconn,
2944 struct ofmp_header *ofmph)
2946 size_t msg_len = ntohs(ofmph->header.header.length);
2947 if (msg_len < sizeof(*ofmph)) {
2948 VLOG_WARN_RL(&rl, "dropping short managment message: %zu\n", msg_len);
2949 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2952 if (ofmph->type == htons(OFMPT_CAPABILITY_REQUEST)) {
2953 struct ofmp_capability_request *ofmpcr;
2955 if (msg_len < sizeof(struct ofmp_capability_request)) {
2956 VLOG_WARN_RL(&rl, "dropping short capability request: %zu\n",
2958 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2961 ofmpcr = (struct ofmp_capability_request *)ofmph;
2962 if (ofmpcr->format != htonl(OFMPCAF_SIMPLE)) {
2963 /* xxx Find a better type than bad subtype */
2964 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2967 send_capability_reply(p, ofconn, ofmph->header.header.xid);
2970 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2975 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
2977 struct ofp_vendor_header *ovh = msg;
2978 struct nicira_header *nh;
2980 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
2981 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2983 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
2984 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2986 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
2987 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2991 switch (ntohl(nh->subtype)) {
2992 case NXT_STATUS_REQUEST:
2993 return switch_status_handle_request(p->switch_status, ofconn->rconn,
2996 case NXT_ACT_SET_CONFIG:
2997 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
2999 case NXT_ACT_GET_CONFIG:
3000 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
3002 case NXT_COMMAND_REQUEST:
3004 return executer_handle_request(p->executer, ofconn->rconn, msg);
3009 return handle_ofmp(p, ofconn, msg);
3012 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3016 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3017 struct ofpbuf *ofp_msg)
3019 struct ofp_header *oh = ofp_msg->data;
3022 COVERAGE_INC(ofproto_recv_openflow);
3024 case OFPT_ECHO_REQUEST:
3025 error = handle_echo_request(ofconn, oh);
3028 case OFPT_ECHO_REPLY:
3032 case OFPT_FEATURES_REQUEST:
3033 error = handle_features_request(p, ofconn, oh);
3036 case OFPT_GET_CONFIG_REQUEST:
3037 error = handle_get_config_request(p, ofconn, oh);
3040 case OFPT_SET_CONFIG:
3041 error = handle_set_config(p, ofconn, ofp_msg->data);
3044 case OFPT_PACKET_OUT:
3045 error = handle_packet_out(p, ofconn, ofp_msg->data);
3049 error = handle_port_mod(p, oh);
3053 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3056 case OFPT_STATS_REQUEST:
3057 error = handle_stats_request(p, ofconn, oh);
3061 error = handle_vendor(p, ofconn, ofp_msg->data);
3065 if (VLOG_IS_WARN_ENABLED()) {
3066 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3067 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3070 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3075 send_error_oh(ofconn, ofp_msg->data, error);
3080 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3082 struct odp_msg *msg = packet->data;
3083 uint16_t in_port = odp_port_to_ofp_port(msg->port);
3085 struct ofpbuf payload;
3088 /* Handle controller actions. */
3089 if (msg->type == _ODPL_ACTION_NR) {
3090 COVERAGE_INC(ofproto_ctlr_action);
3091 pinsched_send(p->action_sched, in_port, packet,
3092 send_packet_in_action, p);
3096 payload.data = msg + 1;
3097 payload.size = msg->length - sizeof *msg;
3098 flow_extract(&payload, msg->port, &flow);
3100 /* Check with in-band control to see if this packet should be sent
3101 * to the local port regardless of the flow table. */
3102 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3103 union odp_action action;
3105 memset(&action, 0, sizeof(action));
3106 action.output.type = ODPAT_OUTPUT;
3107 action.output.port = ODPP_LOCAL;
3108 dpif_execute(&p->dpif, flow.in_port, &action, 1, &payload);
3111 rule = lookup_valid_rule(p, &flow);
3113 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3114 struct ofport *port = port_array_get(&p->ports, msg->port);
3116 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3117 COVERAGE_INC(ofproto_no_packet_in);
3118 /* XXX install 'drop' flow entry */
3119 ofpbuf_delete(packet);
3123 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3126 COVERAGE_INC(ofproto_packet_in);
3127 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3131 if (rule->cr.wc.wildcards) {
3132 rule = rule_create_subrule(p, rule, &flow);
3133 rule_make_actions(p, rule, packet);
3135 if (!rule->may_install) {
3136 /* The rule is not installable, that is, we need to process every
3137 * packet, so process the current packet and set its actions into
3139 rule_make_actions(p, rule, packet);
3141 /* XXX revalidate rule if it needs it */
3145 rule_execute(p, rule, &payload, &flow);
3146 rule_reinstall(p, rule);
3148 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY
3149 && rconn_is_connected(p->controller->rconn)) {
3151 * Extra-special case for fail-open mode.
3153 * We are in fail-open mode and the packet matched the fail-open rule,
3154 * but we are connected to a controller too. We should send the packet
3155 * up to the controller in the hope that it will try to set up a flow
3156 * and thereby allow us to exit fail-open.
3158 * See the top-level comment in fail-open.c for more information.
3160 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3162 ofpbuf_delete(packet);
3167 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3169 struct rule *sub = rule_from_cls_rule(sub_);
3170 struct revalidate_cbdata *cbdata = cbdata_;
3172 if (cbdata->revalidate_all
3173 || (cbdata->revalidate_subrules && sub->super)
3174 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3175 revalidate_rule(cbdata->ofproto, sub);
3180 revalidate_rule(struct ofproto *p, struct rule *rule)
3182 const flow_t *flow = &rule->cr.flow;
3184 COVERAGE_INC(ofproto_revalidate_rule);
3187 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3189 rule_remove(p, rule);
3191 } else if (super != rule->super) {
3192 COVERAGE_INC(ofproto_revalidate_moved);
3193 list_remove(&rule->list);
3194 list_push_back(&super->list, &rule->list);
3195 rule->super = super;
3196 rule->hard_timeout = super->hard_timeout;
3197 rule->idle_timeout = super->idle_timeout;
3198 rule->created = super->created;
3203 rule_update_actions(p, rule);
3207 static struct ofpbuf *
3208 compose_flow_exp(const struct rule *rule, long long int now, uint8_t reason)
3210 struct ofp_flow_expired *ofe;
3213 ofe = make_openflow(sizeof *ofe, OFPT_FLOW_EXPIRED, &buf);
3214 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofe->match);
3215 ofe->priority = htons(rule->cr.priority);
3216 ofe->reason = reason;
3217 ofe->duration = htonl((now - rule->created) / 1000);
3218 ofe->packet_count = htonll(rule->packet_count);
3219 ofe->byte_count = htonll(rule->byte_count);
3225 send_flow_exp(struct ofproto *p, struct rule *rule,
3226 long long int now, uint8_t reason)
3228 struct ofconn *ofconn;
3229 struct ofconn *prev;
3232 /* We limit the maximum number of queued flow expirations it by accounting
3233 * them under the counter for replies. That works because preventing
3234 * OpenFlow requests from being processed also prevents new flows from
3235 * being added (and expiring). (It also prevents processing OpenFlow
3236 * requests that would not add new flows, so it is imperfect.) */
3239 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3240 if (ofconn->send_flow_exp && rconn_is_connected(ofconn->rconn)) {
3242 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3244 buf = compose_flow_exp(rule, now, reason);
3250 queue_tx(buf, prev, prev->reply_counter);
3255 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3257 assert(rule->installed);
3258 assert(!rule->cr.wc.wildcards);
3261 rule_remove(ofproto, rule);
3263 rule_uninstall(ofproto, rule);
3268 expire_rule(struct cls_rule *cls_rule, void *p_)
3270 struct ofproto *p = p_;
3271 struct rule *rule = rule_from_cls_rule(cls_rule);
3272 long long int hard_expire, idle_expire, expire, now;
3274 hard_expire = (rule->hard_timeout
3275 ? rule->created + rule->hard_timeout * 1000
3277 idle_expire = (rule->idle_timeout
3278 && (rule->super || list_is_empty(&rule->list))
3279 ? rule->used + rule->idle_timeout * 1000
3281 expire = MIN(hard_expire, idle_expire);
3285 if (rule->installed && now >= rule->used + 5000) {
3286 uninstall_idle_flow(p, rule);
3287 } else if (!rule->cr.wc.wildcards) {
3288 active_timeout(p, rule);
3294 COVERAGE_INC(ofproto_expired);
3295 if (rule->cr.wc.wildcards) {
3296 /* Update stats. (This code will be a no-op if the rule expired
3297 * due to an idle timeout, because in that case the rule has no
3298 * subrules left.) */
3299 struct rule *subrule, *next;
3300 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3301 rule_remove(p, subrule);
3305 send_flow_exp(p, rule, now,
3307 ? OFPER_HARD_TIMEOUT : OFPER_IDLE_TIMEOUT));
3308 rule_remove(p, rule);
3312 active_timeout(struct ofproto *ofproto, struct rule *rule)
3314 if (ofproto->netflow && !is_controller_rule(rule) &&
3315 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
3316 struct ofexpired expired;
3317 struct odp_flow odp_flow;
3319 /* Get updated flow stats. */
3320 memset(&odp_flow, 0, sizeof odp_flow);
3321 if (rule->installed) {
3322 odp_flow.key = rule->cr.flow;
3323 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
3324 dpif_flow_get(&ofproto->dpif, &odp_flow);
3326 if (odp_flow.stats.n_packets) {
3327 update_time(ofproto, rule, &odp_flow.stats);
3328 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
3329 odp_flow.stats.tcp_flags);
3333 expired.flow = rule->cr.flow;
3334 expired.packet_count = rule->packet_count +
3335 odp_flow.stats.n_packets;
3336 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
3337 expired.used = rule->used;
3339 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
3341 /* Schedule us to send the accumulated records once we have
3342 * collected all of them. */
3343 poll_immediate_wake();
3348 update_used(struct ofproto *p)
3350 struct odp_flow *flows;
3355 error = dpif_flow_list_all(&p->dpif, &flows, &n_flows);
3360 for (i = 0; i < n_flows; i++) {
3361 struct odp_flow *f = &flows[i];
3364 rule = rule_from_cls_rule(
3365 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3366 if (!rule || !rule->installed) {
3367 COVERAGE_INC(ofproto_unexpected_rule);
3368 dpif_flow_del(&p->dpif, f);
3372 update_time(p, rule, &f->stats);
3373 rule_account(p, rule, f->stats.n_bytes);
3379 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3380 const struct ofpbuf *packet, int send_len)
3382 struct odp_msg *msg = packet->data;
3383 struct ofpbuf payload;
3387 /* Extract packet payload from 'msg'. */
3388 payload.data = msg + 1;
3389 payload.size = msg->length - sizeof *msg;
3391 /* Construct ofp_packet_in message. */
3392 reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3393 opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port), reason,
3394 &payload, send_len);
3397 rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100);
3401 send_packet_in_action(struct ofpbuf *packet, void *p_)
3403 struct ofproto *p = p_;
3404 struct ofconn *ofconn;
3405 struct odp_msg *msg;
3408 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3409 if (ofconn == p->controller || ofconn->miss_send_len) {
3410 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3413 ofpbuf_delete(packet);
3417 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3419 struct ofproto *p = p_;
3420 bool in_fail_open = p->fail_open && fail_open_is_active(p->fail_open);
3421 struct ofconn *ofconn;
3422 struct ofpbuf payload;
3423 struct odp_msg *msg;
3426 payload.data = msg + 1;
3427 payload.size = msg->length - sizeof *msg;
3428 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3429 if (ofconn->miss_send_len) {
3430 struct pktbuf *pb = ofconn->pktbuf;
3431 uint32_t buffer_id = (in_fail_open
3433 : pktbuf_save(pb, &payload, msg->port));
3434 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3436 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3439 ofpbuf_delete(packet);
3443 pick_datapath_id(struct dpif *dpif, uint64_t fallback_dpid)
3445 char local_name[IF_NAMESIZE];
3446 uint8_t ea[ETH_ADDR_LEN];
3449 error = dpif_get_name(dpif, local_name, sizeof local_name);
3451 error = netdev_nodev_get_etheraddr(local_name, ea);
3453 return eth_addr_to_uint64(ea);
3455 VLOG_WARN("could not get MAC address for %s (%s)",
3456 local_name, strerror(error));
3459 return fallback_dpid;
3463 pick_fallback_dpid(void)
3465 uint8_t ea[ETH_ADDR_LEN];
3466 eth_addr_random(ea);
3467 ea[0] = 0x00; /* Set Nicira OUI. */
3470 return eth_addr_to_uint64(ea);
3474 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3475 struct odp_actions *actions, tag_type *tags,
3476 uint16_t *nf_output_iface, void *ofproto_)
3478 struct ofproto *ofproto = ofproto_;
3481 /* Drop frames for reserved multicast addresses. */
3482 if (eth_addr_is_reserved(flow->dl_dst)) {
3486 /* Learn source MAC (but don't try to learn from revalidation). */
3487 if (packet != NULL) {
3488 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3491 /* The log messages here could actually be useful in debugging,
3492 * so keep the rate limit relatively high. */
3493 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3494 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3495 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3496 ofproto_revalidate(ofproto, rev_tag);
3500 /* Determine output port. */
3501 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3503 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
3504 } else if (out_port != flow->in_port) {
3505 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3506 *nf_output_iface = out_port;
3514 static const struct ofhooks default_ofhooks = {
3516 default_normal_ofhook_cb,