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 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). */
100 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
102 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
103 * exact-match rule (having cr.wc.wildcards of 0) generated from the
104 * wildcard rule 'super'. In this case, 'list' is an element of the
107 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
108 * a list of subrules. A super-rule with no wildcards (where
109 * cr.wc.wildcards is 0) will never have any subrules. */
115 * A subrule has no actions (it uses the super-rule's actions). */
117 union ofp_action *actions;
121 * A super-rule with wildcard fields never has ODP actions (since the
122 * datapath only supports exact-match flows). */
123 bool installed; /* Installed in datapath? */
124 bool may_install; /* True ordinarily; false if actions must
125 * be reassessed for every packet. */
127 union odp_action *odp_actions;
131 rule_is_hidden(const struct rule *rule)
133 /* Subrules are merely an implementation detail, so hide them from the
135 if (rule->super != NULL) {
139 /* Rules with priority higher than UINT16_MAX are set up by secchan itself
140 * (e.g. by in-band control) and are intentionally hidden from the
142 if (rule->cr.priority > UINT16_MAX) {
149 static struct rule *rule_create(struct rule *super, const union ofp_action *,
150 size_t n_actions, uint16_t idle_timeout,
151 uint16_t hard_timeout);
152 static void rule_free(struct rule *);
153 static void rule_destroy(struct ofproto *, struct rule *);
154 static struct rule *rule_from_cls_rule(const struct cls_rule *);
155 static void rule_insert(struct ofproto *, struct rule *,
156 struct ofpbuf *packet, uint16_t in_port);
157 static void rule_remove(struct ofproto *, struct rule *);
158 static bool rule_make_actions(struct ofproto *, struct rule *,
159 const struct ofpbuf *packet);
160 static void rule_install(struct ofproto *, struct rule *,
161 struct rule *displaced_rule);
162 static void rule_uninstall(struct ofproto *, struct rule *);
163 static void rule_post_uninstall(struct ofproto *, struct rule *);
168 struct pktbuf *pktbuf;
172 struct rconn_packet_counter *packet_in_counter;
174 /* Number of OpenFlow messages queued as replies to OpenFlow requests, and
175 * the maximum number before we stop reading OpenFlow requests. */
176 #define OFCONN_REPLY_MAX 100
177 struct rconn_packet_counter *reply_counter;
180 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *);
181 static void ofconn_destroy(struct ofconn *, struct ofproto *);
182 static void ofconn_run(struct ofconn *, struct ofproto *);
183 static void ofconn_wait(struct ofconn *);
184 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
185 struct rconn_packet_counter *counter);
189 uint64_t datapath_id; /* Datapath ID. */
190 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
191 uint64_t mgmt_id; /* Management channel identifier. */
192 char *manufacturer; /* Manufacturer. */
193 char *hardware; /* Hardware. */
194 char *software; /* Software version. */
195 char *serial; /* Serial number. */
199 struct dpifmon *dpifmon;
200 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
202 struct shash port_by_name;
206 struct switch_status *switch_status;
207 struct status_category *ss_cat;
208 struct in_band *in_band;
209 struct discovery *discovery;
210 struct fail_open *fail_open;
211 struct pinsched *miss_sched, *action_sched;
212 struct executer *executer;
213 struct netflow *netflow;
216 struct classifier cls;
217 bool need_revalidate;
218 long long int next_expiration;
219 struct tag_set revalidate_set;
221 /* OpenFlow connections. */
222 struct list all_conns;
223 struct ofconn *controller;
224 struct pvconn **listeners;
226 struct pvconn **snoops;
229 /* Hooks for ovs-vswitchd. */
230 const struct ofhooks *ofhooks;
233 /* Used by default ofhooks. */
234 struct mac_learning *ml;
237 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
239 static const struct ofhooks default_ofhooks;
241 static uint64_t pick_datapath_id(struct dpif *, uint64_t fallback_dpid);
242 static uint64_t pick_fallback_dpid(void);
243 static void send_packet_in_miss(struct ofpbuf *, void *ofproto);
244 static void send_packet_in_action(struct ofpbuf *, void *ofproto);
245 static void update_used(struct ofproto *);
246 static void update_stats(struct rule *, const struct odp_flow_stats *);
247 static void expire_rule(struct cls_rule *, void *ofproto);
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, const struct svec *collectors,
547 uint8_t engine_type, uint8_t engine_id, bool add_id_to_iface)
549 if (collectors && collectors->n) {
550 if (!ofproto->netflow) {
551 ofproto->netflow = netflow_create();
553 netflow_set_engine(ofproto->netflow, engine_type, engine_id,
555 return netflow_set_collectors(ofproto->netflow, collectors);
557 netflow_destroy(ofproto->netflow);
558 ofproto->netflow = NULL;
564 ofproto_set_failure(struct ofproto *ofproto, bool fail_open)
567 struct rconn *rconn = ofproto->controller->rconn;
568 int trigger_duration = rconn_get_probe_interval(rconn) * 3;
569 if (!ofproto->fail_open) {
570 ofproto->fail_open = fail_open_create(ofproto, trigger_duration,
571 ofproto->switch_status,
574 fail_open_set_trigger_duration(ofproto->fail_open,
578 fail_open_destroy(ofproto->fail_open);
579 ofproto->fail_open = NULL;
584 ofproto_set_rate_limit(struct ofproto *ofproto,
585 int rate_limit, int burst_limit)
587 if (rate_limit > 0) {
588 if (!ofproto->miss_sched) {
589 ofproto->miss_sched = pinsched_create(rate_limit, burst_limit,
590 ofproto->switch_status);
591 ofproto->action_sched = pinsched_create(rate_limit, burst_limit,
594 pinsched_set_limits(ofproto->miss_sched, rate_limit, burst_limit);
595 pinsched_set_limits(ofproto->action_sched,
596 rate_limit, burst_limit);
599 pinsched_destroy(ofproto->miss_sched);
600 ofproto->miss_sched = NULL;
601 pinsched_destroy(ofproto->action_sched);
602 ofproto->action_sched = NULL;
607 ofproto_set_stp(struct ofproto *ofproto UNUSED, bool enable_stp)
611 VLOG_WARN("STP is not yet implemented");
619 ofproto_set_remote_execution(struct ofproto *ofproto, const char *command_acl,
620 const char *command_dir)
623 if (!ofproto->executer) {
624 return executer_create(command_acl, command_dir,
627 executer_set_acl(ofproto->executer, command_acl, command_dir);
630 executer_destroy(ofproto->executer);
631 ofproto->executer = NULL;
637 ofproto_get_datapath_id(const struct ofproto *ofproto)
639 return ofproto->datapath_id;
643 ofproto_get_mgmt_id(const struct ofproto *ofproto)
645 return ofproto->mgmt_id;
649 ofproto_get_probe_interval(const struct ofproto *ofproto)
651 return rconn_get_probe_interval(ofproto->controller->rconn);
655 ofproto_get_max_backoff(const struct ofproto *ofproto)
657 return rconn_get_max_backoff(ofproto->controller->rconn);
661 ofproto_get_in_band(const struct ofproto *ofproto)
663 return ofproto->in_band != NULL;
667 ofproto_get_discovery(const struct ofproto *ofproto)
669 return ofproto->discovery != NULL;
673 ofproto_get_controller(const struct ofproto *ofproto)
675 return rconn_get_name(ofproto->controller->rconn);
679 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
683 for (i = 0; i < ofproto->n_listeners; i++) {
684 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
689 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
693 for (i = 0; i < ofproto->n_snoops; i++) {
694 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
699 ofproto_destroy(struct ofproto *p)
701 struct ofconn *ofconn, *next_ofconn;
702 struct ofport *ofport;
703 unsigned int port_no;
710 ofproto_flush_flows(p);
711 classifier_destroy(&p->cls);
713 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
715 ofconn_destroy(ofconn, p);
718 dpif_close(&p->dpif);
719 dpifmon_destroy(p->dpifmon);
720 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
723 shash_destroy(&p->port_by_name);
725 switch_status_destroy(p->switch_status);
726 in_band_destroy(p->in_band);
727 discovery_destroy(p->discovery);
728 fail_open_destroy(p->fail_open);
729 pinsched_destroy(p->miss_sched);
730 pinsched_destroy(p->action_sched);
731 executer_destroy(p->executer);
732 netflow_destroy(p->netflow);
734 switch_status_unregister(p->ss_cat);
736 for (i = 0; i < p->n_listeners; i++) {
737 pvconn_close(p->listeners[i]);
741 for (i = 0; i < p->n_snoops; i++) {
742 pvconn_close(p->snoops[i]);
746 mac_learning_destroy(p->ml);
752 ofproto_run(struct ofproto *p)
754 int error = ofproto_run1(p);
756 error = ofproto_run2(p, false);
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, "dp%u: datapath was destroyed externally",
787 handle_odp_msg(p, buf);
790 while ((error = dpifmon_poll(p->dpifmon, &devname)) != EAGAIN) {
791 if (error == ENOBUFS) {
794 update_port(p, devname);
800 in_band_run(p->in_band);
803 char *controller_name;
804 if (rconn_is_connectivity_questionable(p->controller->rconn)) {
805 discovery_question_connectivity(p->discovery);
807 if (discovery_run(p->discovery, &controller_name)) {
808 if (controller_name) {
809 rconn_connect(p->controller->rconn, controller_name);
811 rconn_disconnect(p->controller->rconn);
815 pinsched_run(p->miss_sched, send_packet_in_miss, p);
816 pinsched_run(p->action_sched, send_packet_in_action, p);
818 executer_run(p->executer);
821 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
823 ofconn_run(ofconn, p);
826 /* Fail-open maintenance. Do this after processing the ofconns since
827 * fail-open checks the status of the controller rconn. */
829 fail_open_run(p->fail_open);
832 for (i = 0; i < p->n_listeners; i++) {
836 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
838 ofconn_create(p, rconn_new_from_vconn("passive", vconn));
839 } else if (retval != EAGAIN) {
840 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
844 for (i = 0; i < p->n_snoops; i++) {
848 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
850 rconn_add_monitor(p->controller->rconn, vconn);
851 } else if (retval != EAGAIN) {
852 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
856 if (time_msec() >= p->next_expiration) {
857 COVERAGE_INC(ofproto_expiration);
858 p->next_expiration = time_msec() + 1000;
861 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
863 /* Let the hook know that we're at a stable point: all outstanding data
864 * in existing flows has been accounted to the account_cb. Thus, the
865 * hook can now reasonably do operations that depend on having accurate
866 * flow volume accounting (currently, that's just bond rebalancing). */
867 if (p->ofhooks->account_checkpoint_cb) {
868 p->ofhooks->account_checkpoint_cb(p->aux);
873 netflow_run(p->netflow);
879 struct revalidate_cbdata {
880 struct ofproto *ofproto;
881 bool revalidate_all; /* Revalidate all exact-match rules? */
882 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
883 struct tag_set revalidate_set; /* Set of tags to revalidate. */
887 ofproto_run2(struct ofproto *p, bool revalidate_all)
889 if (p->need_revalidate || revalidate_all
890 || !tag_set_is_empty(&p->revalidate_set)) {
891 struct revalidate_cbdata cbdata;
893 cbdata.revalidate_all = revalidate_all;
894 cbdata.revalidate_subrules = p->need_revalidate;
895 cbdata.revalidate_set = p->revalidate_set;
896 tag_set_init(&p->revalidate_set);
897 COVERAGE_INC(ofproto_revalidate);
898 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
899 p->need_revalidate = false;
906 ofproto_wait(struct ofproto *p)
908 struct ofconn *ofconn;
911 dpif_recv_wait(&p->dpif);
912 dpifmon_wait(p->dpifmon);
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 memcpy(ofport->opp.hw_addr, netdev_get_etheraddr(netdev), ETH_ALEN);
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 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no),
1196 shash_add(&p->port_by_name, (char *) ofport->opp.name, ofport);
1200 ofport_remove(struct ofproto *p, struct ofport *ofport)
1202 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no), NULL);
1203 shash_delete(&p->port_by_name,
1204 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1208 ofport_free(struct ofport *ofport)
1211 netdev_close(ofport->netdev);
1217 update_port(struct ofproto *p, const char *devname)
1219 struct odp_port odp_port;
1220 struct ofport *old_ofport;
1221 struct ofport *new_ofport;
1224 COVERAGE_INC(ofproto_update_port);
1226 /* Query the datapath for port information. */
1227 error = dpif_port_query_by_name(&p->dpif, devname, &odp_port);
1229 /* Find the old ofport. */
1230 old_ofport = shash_find_data(&p->port_by_name, devname);
1233 /* There's no port named 'devname' but there might be a port with
1234 * the same port number. This could happen if a port is deleted
1235 * and then a new one added in its place very quickly, or if a port
1236 * is renamed. In the former case we want to send an OFPPR_DELETE
1237 * and an OFPPR_ADD, and in the latter case we want to send a
1238 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1239 * the old port's ifindex against the new port, or perhaps less
1240 * reliably but more portably by comparing the old port's MAC
1241 * against the new port's MAC. However, this code isn't that smart
1242 * and always sends an OFPPR_MODIFY (XXX). */
1243 old_ofport = port_array_get(&p->ports, odp_port.port);
1245 } else if (error != ENOENT && error != ENODEV) {
1246 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1247 "%s", strerror(error));
1251 /* Create a new ofport. */
1252 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1254 /* Eliminate a few pathological cases. */
1255 if (!old_ofport && !new_ofport) {
1257 } else if (old_ofport && new_ofport) {
1258 /* Most of the 'config' bits are OpenFlow soft state, but
1259 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1260 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1261 * leaves the other bits 0.) */
1262 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1264 if (ofport_equal(old_ofport, new_ofport)) {
1265 /* False alarm--no change. */
1266 ofport_free(new_ofport);
1271 /* Now deal with the normal cases. */
1273 ofport_remove(p, old_ofport);
1276 ofport_install(p, new_ofport);
1278 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1279 (!old_ofport ? OFPPR_ADD
1280 : !new_ofport ? OFPPR_DELETE
1282 ofport_free(old_ofport);
1284 /* Update port groups. */
1285 refresh_port_groups(p);
1289 init_ports(struct ofproto *p)
1291 struct odp_port *ports;
1296 error = dpif_port_list(&p->dpif, &ports, &n_ports);
1301 for (i = 0; i < n_ports; i++) {
1302 const struct odp_port *odp_port = &ports[i];
1303 if (!ofport_conflicts(p, odp_port)) {
1304 struct ofport *ofport = make_ofport(odp_port);
1306 ofport_install(p, ofport);
1311 refresh_port_groups(p);
1315 static struct ofconn *
1316 ofconn_create(struct ofproto *p, struct rconn *rconn)
1318 struct ofconn *ofconn = xmalloc(sizeof *ofconn);
1319 list_push_back(&p->all_conns, &ofconn->node);
1320 ofconn->rconn = rconn;
1321 ofconn->pktbuf = NULL;
1322 ofconn->send_flow_exp = false;
1323 ofconn->miss_send_len = 0;
1324 ofconn->packet_in_counter = rconn_packet_counter_create ();
1325 ofconn->reply_counter = rconn_packet_counter_create ();
1330 ofconn_destroy(struct ofconn *ofconn, struct ofproto *p)
1333 executer_rconn_closing(p->executer, ofconn->rconn);
1336 list_remove(&ofconn->node);
1337 rconn_destroy(ofconn->rconn);
1338 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1339 rconn_packet_counter_destroy(ofconn->reply_counter);
1340 pktbuf_destroy(ofconn->pktbuf);
1345 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1349 rconn_run(ofconn->rconn);
1351 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1352 /* Limit the number of iterations to prevent other tasks from
1354 for (iteration = 0; iteration < 50; iteration++) {
1355 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1360 fail_open_maybe_recover(p->fail_open);
1362 handle_openflow(ofconn, p, of_msg);
1363 ofpbuf_delete(of_msg);
1367 if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
1368 ofconn_destroy(ofconn, p);
1373 ofconn_wait(struct ofconn *ofconn)
1375 rconn_run_wait(ofconn->rconn);
1376 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1377 rconn_recv_wait(ofconn->rconn);
1379 COVERAGE_INC(ofproto_ofconn_stuck);
1383 /* Caller is responsible for initializing the 'cr' member of the returned
1385 static struct rule *
1386 rule_create(struct rule *super,
1387 const union ofp_action *actions, size_t n_actions,
1388 uint16_t idle_timeout, uint16_t hard_timeout)
1390 struct rule *rule = xcalloc(1, sizeof *rule);
1391 rule->idle_timeout = idle_timeout;
1392 rule->hard_timeout = hard_timeout;
1393 rule->used = rule->created = time_msec();
1394 rule->super = super;
1396 list_push_back(&super->list, &rule->list);
1398 list_init(&rule->list);
1400 rule->n_actions = n_actions;
1401 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1405 static struct rule *
1406 rule_from_cls_rule(const struct cls_rule *cls_rule)
1408 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1412 rule_free(struct rule *rule)
1414 free(rule->actions);
1415 free(rule->odp_actions);
1419 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1420 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1421 * through all of its subrules and revalidates them, destroying any that no
1422 * longer has a super-rule (which is probably all of them).
1424 * Before calling this function, the caller must make have removed 'rule' from
1425 * the classifier. If 'rule' is an exact-match rule, the caller is also
1426 * responsible for ensuring that it has been uninstalled from the datapath. */
1428 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1431 struct rule *subrule, *next;
1432 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1433 revalidate_rule(ofproto, subrule);
1436 list_remove(&rule->list);
1442 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1444 const union ofp_action *oa;
1445 struct actions_iterator i;
1447 if (out_port == htons(OFPP_NONE)) {
1450 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1451 oa = actions_next(&i)) {
1452 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1459 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1460 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1462 * The flow that 'packet' actually contains does not need to actually match
1463 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1464 * the packet and byte counters for 'rule' will be credited for the packet sent
1465 * out whether or not the packet actually matches 'rule'.
1467 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1468 * the caller must already have accurately composed ODP actions for it given
1469 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1470 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1471 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1472 * actions and apply them to 'packet'. */
1474 rule_execute(struct ofproto *ofproto, struct rule *rule,
1475 struct ofpbuf *packet, const flow_t *flow)
1477 const union odp_action *actions;
1479 struct odp_actions a;
1481 /* Grab or compose the ODP actions.
1483 * The special case for an exact-match 'rule' where 'flow' is not the
1484 * rule's flow is important to avoid, e.g., sending a packet out its input
1485 * port simply because the ODP actions were composed for the wrong
1487 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1488 struct rule *super = rule->super ? rule->super : rule;
1489 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1490 packet, &a, NULL, 0, NULL)) {
1493 actions = a.actions;
1494 n_actions = a.n_actions;
1496 actions = rule->odp_actions;
1497 n_actions = rule->n_odp_actions;
1500 /* Execute the ODP actions. */
1501 if (!dpif_execute(&ofproto->dpif, flow->in_port,
1502 actions, n_actions, packet)) {
1503 struct odp_flow_stats stats;
1504 flow_extract_stats(flow, packet, &stats);
1505 update_stats(rule, &stats);
1506 rule->used = time_msec();
1511 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1514 struct rule *displaced_rule;
1516 /* Insert the rule in the classifier. */
1517 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1518 if (!rule->cr.wc.wildcards) {
1519 rule_make_actions(p, rule, packet);
1522 /* Send the packet and credit it to the rule. */
1525 flow_extract(packet, in_port, &flow);
1526 rule_execute(p, rule, packet, &flow);
1529 /* Install the rule in the datapath only after sending the packet, to
1530 * avoid packet reordering. */
1531 if (rule->cr.wc.wildcards) {
1532 COVERAGE_INC(ofproto_add_wc_flow);
1533 p->need_revalidate = true;
1535 rule_install(p, rule, displaced_rule);
1538 /* Free the rule that was displaced, if any. */
1539 if (displaced_rule) {
1540 rule_destroy(p, displaced_rule);
1544 static struct rule *
1545 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1548 struct rule *subrule = rule_create(rule, NULL, 0,
1549 rule->idle_timeout, rule->hard_timeout);
1550 COVERAGE_INC(ofproto_subrule_create);
1551 cls_rule_from_flow(&subrule->cr, flow, 0,
1552 (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1553 : rule->cr.priority));
1554 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1560 rule_remove(struct ofproto *ofproto, struct rule *rule)
1562 if (rule->cr.wc.wildcards) {
1563 COVERAGE_INC(ofproto_del_wc_flow);
1564 ofproto->need_revalidate = true;
1566 rule_uninstall(ofproto, rule);
1568 classifier_remove(&ofproto->cls, &rule->cr);
1569 rule_destroy(ofproto, rule);
1572 /* Returns true if the actions changed, false otherwise. */
1574 rule_make_actions(struct ofproto *p, struct rule *rule,
1575 const struct ofpbuf *packet)
1577 const struct rule *super;
1578 struct odp_actions a;
1581 assert(!rule->cr.wc.wildcards);
1583 super = rule->super ? rule->super : rule;
1585 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1586 packet, &a, &rule->tags, &rule->may_install,
1587 &rule->nf_output_iface);
1589 actions_len = a.n_actions * sizeof *a.actions;
1590 if (rule->n_odp_actions != a.n_actions
1591 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1592 COVERAGE_INC(ofproto_odp_unchanged);
1593 free(rule->odp_actions);
1594 rule->n_odp_actions = a.n_actions;
1595 rule->odp_actions = xmemdup(a.actions, actions_len);
1603 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1604 struct odp_flow_put *put)
1606 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1607 put->flow.key = rule->cr.flow;
1608 put->flow.actions = rule->odp_actions;
1609 put->flow.n_actions = rule->n_odp_actions;
1611 return dpif_flow_put(&ofproto->dpif, put);
1615 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1617 assert(!rule->cr.wc.wildcards);
1619 if (rule->may_install) {
1620 struct odp_flow_put put;
1621 if (!do_put_flow(p, rule,
1622 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1624 rule->installed = true;
1625 if (displaced_rule) {
1626 update_stats(rule, &put.flow.stats);
1627 rule_post_uninstall(p, displaced_rule);
1630 } else if (displaced_rule) {
1631 rule_uninstall(p, displaced_rule);
1636 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1638 if (rule->installed) {
1639 struct odp_flow_put put;
1640 COVERAGE_INC(ofproto_dp_missed);
1641 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1643 rule_install(ofproto, rule, NULL);
1648 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1650 bool actions_changed = rule_make_actions(ofproto, rule, NULL);
1651 if (rule->may_install) {
1652 if (rule->installed) {
1653 if (actions_changed) {
1654 /* XXX should really do rule_post_uninstall() for the *old* set
1655 * of actions, and distinguish the old stats from the new. */
1656 struct odp_flow_put put;
1657 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1660 rule_install(ofproto, rule, NULL);
1663 rule_uninstall(ofproto, rule);
1668 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1670 uint64_t total_bytes = rule->byte_count + extra_bytes;
1672 if (ofproto->ofhooks->account_flow_cb
1673 && total_bytes > rule->accounted_bytes)
1675 ofproto->ofhooks->account_flow_cb(
1676 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1677 total_bytes - rule->accounted_bytes, ofproto->aux);
1678 rule->accounted_bytes = total_bytes;
1683 rule_uninstall(struct ofproto *p, struct rule *rule)
1685 assert(!rule->cr.wc.wildcards);
1686 if (rule->installed) {
1687 struct odp_flow odp_flow;
1689 odp_flow.key = rule->cr.flow;
1690 odp_flow.actions = NULL;
1691 odp_flow.n_actions = 0;
1692 if (!dpif_flow_del(&p->dpif, &odp_flow)) {
1693 update_stats(rule, &odp_flow.stats);
1695 rule->installed = false;
1697 rule_post_uninstall(p, rule);
1702 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1704 struct rule *super = rule->super;
1705 bool controller_action;
1707 rule_account(ofproto, rule, 0);
1709 /* If the only action is send to the controller then don't report
1710 * NetFlow expiration messages since it is just part of the control
1711 * logic for the network and not real traffic. */
1712 controller_action = rule->n_odp_actions == 1 &&
1713 rule->odp_actions[0].type == ODPAT_CONTROLLER;
1715 if (ofproto->netflow && rule->byte_count && !controller_action) {
1716 struct ofexpired expired;
1717 expired.flow = rule->cr.flow;
1718 expired.packet_count = rule->packet_count;
1719 expired.byte_count = rule->byte_count;
1720 expired.used = rule->used;
1721 expired.created = rule->created;
1722 expired.tcp_flags = rule->tcp_flags;
1723 expired.ip_tos = rule->ip_tos;
1724 expired.output_iface = rule->nf_output_iface;
1725 netflow_expire(ofproto->netflow, &expired);
1728 super->packet_count += rule->packet_count;
1729 super->byte_count += rule->byte_count;
1730 super->tcp_flags |= rule->tcp_flags;
1731 if (rule->packet_count) {
1732 super->ip_tos = rule->ip_tos;
1735 /* Reset counters to prevent double counting if the rule ever gets
1737 rule->packet_count = 0;
1738 rule->byte_count = 0;
1739 rule->accounted_bytes = 0;
1740 rule->tcp_flags = 0;
1746 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1747 struct rconn_packet_counter *counter)
1749 update_openflow_length(msg);
1750 if (rconn_send(ofconn->rconn, msg, counter)) {
1756 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1757 int error, const void *data, size_t len)
1760 struct ofp_error_msg *oem;
1762 if (!(error >> 16)) {
1763 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1768 COVERAGE_INC(ofproto_error);
1769 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1770 oh ? oh->xid : 0, &buf);
1771 oem->type = htons((unsigned int) error >> 16);
1772 oem->code = htons(error & 0xffff);
1773 memcpy(oem->data, data, len);
1774 queue_tx(buf, ofconn, ofconn->reply_counter);
1778 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1781 size_t oh_length = ntohs(oh->length);
1782 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1786 hton_ofp_phy_port(struct ofp_phy_port *opp)
1788 opp->port_no = htons(opp->port_no);
1789 opp->config = htonl(opp->config);
1790 opp->state = htonl(opp->state);
1791 opp->curr = htonl(opp->curr);
1792 opp->advertised = htonl(opp->advertised);
1793 opp->supported = htonl(opp->supported);
1794 opp->peer = htonl(opp->peer);
1798 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
1800 struct ofp_header *rq = oh;
1801 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
1806 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
1807 struct ofp_header *oh)
1809 struct ofp_switch_features *osf;
1811 unsigned int port_no;
1812 struct ofport *port;
1814 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1815 osf->datapath_id = htonll(p->datapath_id);
1816 osf->n_buffers = htonl(pktbuf_capacity());
1818 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1819 OFPC_PORT_STATS | OFPC_MULTI_PHY_TX);
1820 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1821 (1u << OFPAT_SET_VLAN_VID) |
1822 (1u << OFPAT_SET_VLAN_PCP) |
1823 (1u << OFPAT_STRIP_VLAN) |
1824 (1u << OFPAT_SET_DL_SRC) |
1825 (1u << OFPAT_SET_DL_DST) |
1826 (1u << OFPAT_SET_NW_SRC) |
1827 (1u << OFPAT_SET_NW_DST) |
1828 (1u << OFPAT_SET_TP_SRC) |
1829 (1u << OFPAT_SET_TP_DST));
1831 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1832 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1835 queue_tx(buf, ofconn, ofconn->reply_counter);
1840 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1841 struct ofp_header *oh)
1844 struct ofp_switch_config *osc;
1848 /* Figure out flags. */
1849 dpif_get_drop_frags(&p->dpif, &drop_frags);
1850 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1851 if (ofconn->send_flow_exp) {
1852 flags |= OFPC_SEND_FLOW_EXP;
1856 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1857 osc->flags = htons(flags);
1858 osc->miss_send_len = htons(ofconn->miss_send_len);
1859 queue_tx(buf, ofconn, ofconn->reply_counter);
1865 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1866 struct ofp_switch_config *osc)
1871 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1875 flags = ntohs(osc->flags);
1877 ofconn->send_flow_exp = (flags & OFPC_SEND_FLOW_EXP) != 0;
1879 if (ofconn == p->controller) {
1880 switch (flags & OFPC_FRAG_MASK) {
1881 case OFPC_FRAG_NORMAL:
1882 dpif_set_drop_frags(&p->dpif, false);
1884 case OFPC_FRAG_DROP:
1885 dpif_set_drop_frags(&p->dpif, true);
1888 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1894 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1895 if (ntohs(osc->miss_send_len) != 0) {
1896 ofconn->pktbuf = pktbuf_create();
1898 pktbuf_destroy(ofconn->pktbuf);
1902 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1908 add_output_group_action(struct odp_actions *actions, uint16_t group,
1909 uint16_t *nf_output_iface)
1911 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1913 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
1914 *nf_output_iface = NF_OUT_FLOOD;
1919 add_controller_action(struct odp_actions *actions,
1920 const struct ofp_action_output *oao)
1922 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
1923 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
1926 struct action_xlate_ctx {
1928 const flow_t *flow; /* Flow to which these actions correspond. */
1929 int recurse; /* Recursion level, via xlate_table_action. */
1930 struct ofproto *ofproto;
1931 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
1932 * null pointer if we are revalidating
1933 * without a packet to refer to. */
1936 struct odp_actions *out; /* Datapath actions. */
1937 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
1938 bool may_set_up_flow; /* True ordinarily; false if the actions must
1939 * be reassessed for every packet. */
1940 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
1943 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
1944 struct action_xlate_ctx *ctx);
1947 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
1949 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
1952 if (ofport->opp.config & OFPPC_NO_FWD) {
1953 /* Forwarding disabled on port. */
1958 * We don't have an ofport record for this port, but it doesn't hurt to
1959 * allow forwarding to it anyhow. Maybe such a port will appear later
1960 * and we're pre-populating the flow table.
1964 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
1965 ctx->nf_output_iface = port;
1968 static struct rule *
1969 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
1972 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
1974 /* The rule we found might not be valid, since we could be in need of
1975 * revalidation. If it is not valid, don't return it. */
1978 && ofproto->need_revalidate
1979 && !revalidate_rule(ofproto, rule)) {
1980 COVERAGE_INC(ofproto_invalidated);
1988 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
1990 if (!ctx->recurse) {
1995 flow.in_port = in_port;
1997 rule = lookup_valid_rule(ctx->ofproto, &flow);
2004 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2011 xlate_output_action(struct action_xlate_ctx *ctx,
2012 const struct ofp_action_output *oao)
2015 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2017 ctx->nf_output_iface = NF_OUT_DROP;
2019 switch (ntohs(oao->port)) {
2021 add_output_action(ctx, ctx->flow->in_port);
2024 xlate_table_action(ctx, ctx->flow->in_port);
2027 if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet,
2028 ctx->out, ctx->tags,
2029 &ctx->nf_output_iface,
2030 ctx->ofproto->aux)) {
2031 COVERAGE_INC(ofproto_uninstallable);
2032 ctx->may_set_up_flow = false;
2036 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2037 &ctx->nf_output_iface);
2040 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2042 case OFPP_CONTROLLER:
2043 add_controller_action(ctx->out, oao);
2046 add_output_action(ctx, ODPP_LOCAL);
2049 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2050 if (odp_port != ctx->flow->in_port) {
2051 add_output_action(ctx, odp_port);
2056 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2057 ctx->nf_output_iface = NF_OUT_FLOOD;
2058 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2059 ctx->nf_output_iface = prev_nf_output_iface;
2060 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2061 ctx->nf_output_iface != NF_OUT_FLOOD) {
2062 ctx->nf_output_iface = NF_OUT_MULTI;
2067 xlate_nicira_action(struct action_xlate_ctx *ctx,
2068 const struct nx_action_header *nah)
2070 const struct nx_action_resubmit *nar;
2071 int subtype = ntohs(nah->subtype);
2073 assert(nah->vendor == htonl(NX_VENDOR_ID));
2075 case NXAST_RESUBMIT:
2076 nar = (const struct nx_action_resubmit *) nah;
2077 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2081 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2087 do_xlate_actions(const union ofp_action *in, size_t n_in,
2088 struct action_xlate_ctx *ctx)
2090 struct actions_iterator iter;
2091 const union ofp_action *ia;
2092 const struct ofport *port;
2094 port = port_array_get(&ctx->ofproto->ports, ctx->flow->in_port);
2095 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2096 port->opp.config & (eth_addr_equals(ctx->flow->dl_dst, stp_eth_addr)
2097 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2098 /* Drop this flow. */
2102 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2103 uint16_t type = ntohs(ia->type);
2104 union odp_action *oa;
2108 xlate_output_action(ctx, &ia->output);
2111 case OFPAT_SET_VLAN_VID:
2112 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2113 oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2116 case OFPAT_SET_VLAN_PCP:
2117 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2118 oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2121 case OFPAT_STRIP_VLAN:
2122 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2125 case OFPAT_SET_DL_SRC:
2126 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2127 memcpy(oa->dl_addr.dl_addr,
2128 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2131 case OFPAT_SET_DL_DST:
2132 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2133 memcpy(oa->dl_addr.dl_addr,
2134 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2137 case OFPAT_SET_NW_SRC:
2138 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2139 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2142 case OFPAT_SET_TP_SRC:
2143 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2144 oa->tp_port.tp_port = ia->tp_port.tp_port;
2148 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2152 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2159 xlate_actions(const union ofp_action *in, size_t n_in,
2160 const flow_t *flow, struct ofproto *ofproto,
2161 const struct ofpbuf *packet,
2162 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2163 uint16_t *nf_output_iface)
2165 tag_type no_tags = 0;
2166 struct action_xlate_ctx ctx;
2167 COVERAGE_INC(ofproto_ofp2odp);
2168 odp_actions_init(out);
2171 ctx.ofproto = ofproto;
2172 ctx.packet = packet;
2174 ctx.tags = tags ? tags : &no_tags;
2175 ctx.may_set_up_flow = true;
2176 ctx.nf_output_iface = NF_OUT_DROP;
2177 do_xlate_actions(in, n_in, &ctx);
2179 /* Check with in-band control to see if we're allowed to set up this
2181 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2182 ctx.may_set_up_flow = false;
2185 if (may_set_up_flow) {
2186 *may_set_up_flow = ctx.may_set_up_flow;
2188 if (nf_output_iface) {
2189 *nf_output_iface = ctx.nf_output_iface;
2191 if (odp_actions_overflow(out)) {
2192 odp_actions_init(out);
2193 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2199 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2200 struct ofp_header *oh)
2202 struct ofp_packet_out *opo;
2203 struct ofpbuf payload, *buffer;
2204 struct odp_actions actions;
2210 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2214 opo = (struct ofp_packet_out *) oh;
2216 COVERAGE_INC(ofproto_packet_out);
2217 if (opo->buffer_id != htonl(UINT32_MAX)) {
2218 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2220 if (error || !buffer) {
2228 flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2229 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2230 &flow, p, &payload, &actions, NULL, NULL, NULL);
2235 dpif_execute(&p->dpif, flow.in_port, actions.actions, actions.n_actions,
2237 ofpbuf_delete(buffer);
2243 update_port_config(struct ofproto *p, struct ofport *port,
2244 uint32_t config, uint32_t mask)
2246 mask &= config ^ port->opp.config;
2247 if (mask & OFPPC_PORT_DOWN) {
2248 if (config & OFPPC_PORT_DOWN) {
2249 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2251 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2254 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2255 if (mask & REVALIDATE_BITS) {
2256 COVERAGE_INC(ofproto_costly_flags);
2257 port->opp.config ^= mask & REVALIDATE_BITS;
2258 p->need_revalidate = true;
2260 #undef REVALIDATE_BITS
2261 if (mask & OFPPC_NO_FLOOD) {
2262 port->opp.config ^= OFPPC_NO_FLOOD;
2263 refresh_port_group(p, DP_GROUP_FLOOD);
2265 if (mask & OFPPC_NO_PACKET_IN) {
2266 port->opp.config ^= OFPPC_NO_PACKET_IN;
2271 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2273 const struct ofp_port_mod *opm;
2274 struct ofport *port;
2277 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2281 opm = (struct ofp_port_mod *) oh;
2283 port = port_array_get(&p->ports,
2284 ofp_port_to_odp_port(ntohs(opm->port_no)));
2286 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2287 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2288 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2290 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2291 if (opm->advertise) {
2292 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2298 static struct ofpbuf *
2299 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2301 struct ofp_stats_reply *osr;
2304 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2305 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2307 osr->flags = htons(0);
2311 static struct ofpbuf *
2312 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2314 return make_stats_reply(request->header.xid, request->type, body_len);
2318 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2320 struct ofpbuf *msg = *msgp;
2321 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2322 if (nbytes + msg->size > UINT16_MAX) {
2323 struct ofp_stats_reply *reply = msg->data;
2324 reply->flags = htons(OFPSF_REPLY_MORE);
2325 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2326 queue_tx(msg, ofconn, ofconn->reply_counter);
2328 return ofpbuf_put_uninit(*msgp, nbytes);
2332 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2333 struct ofp_stats_request *request)
2335 struct ofp_desc_stats *ods;
2338 msg = start_stats_reply(request, sizeof *ods);
2339 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2340 strncpy(ods->mfr_desc, p->manufacturer, sizeof ods->mfr_desc);
2341 strncpy(ods->hw_desc, p->hardware, sizeof ods->hw_desc);
2342 strncpy(ods->sw_desc, p->software, sizeof ods->sw_desc);
2343 strncpy(ods->serial_num, p->serial, sizeof ods->serial_num);
2344 queue_tx(msg, ofconn, ofconn->reply_counter);
2350 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2352 struct rule *rule = rule_from_cls_rule(cls_rule);
2353 int *n_subrules = n_subrules_;
2361 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2362 struct ofp_stats_request *request)
2364 struct ofp_table_stats *ots;
2366 struct odp_stats dpstats;
2367 int n_exact, n_subrules, n_wild;
2369 msg = start_stats_reply(request, sizeof *ots * 2);
2371 /* Count rules of various kinds. */
2373 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2374 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2375 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2378 dpif_get_dp_stats(&p->dpif, &dpstats);
2379 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2380 memset(ots, 0, sizeof *ots);
2381 ots->table_id = TABLEID_HASH;
2382 strcpy(ots->name, "hash");
2383 ots->wildcards = htonl(0);
2384 ots->max_entries = htonl(dpstats.max_capacity);
2385 ots->active_count = htonl(n_exact);
2386 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2388 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2390 /* Classifier table. */
2391 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2392 memset(ots, 0, sizeof *ots);
2393 ots->table_id = TABLEID_CLASSIFIER;
2394 strcpy(ots->name, "classifier");
2395 ots->wildcards = htonl(OFPFW_ALL);
2396 ots->max_entries = htonl(65536);
2397 ots->active_count = htonl(n_wild);
2398 ots->lookup_count = htonll(0); /* XXX */
2399 ots->matched_count = htonll(0); /* XXX */
2401 queue_tx(msg, ofconn, ofconn->reply_counter);
2406 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2407 struct ofp_stats_request *request)
2409 struct ofp_port_stats *ops;
2411 struct ofport *port;
2412 unsigned int port_no;
2414 msg = start_stats_reply(request, sizeof *ops * 16);
2415 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2416 struct netdev_stats stats;
2418 /* Intentionally ignore return value, since errors will set 'stats' to
2419 * all-1s, which is correct for OpenFlow, and netdev_get_stats() will
2421 netdev_get_stats(port->netdev, &stats);
2423 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2424 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2425 memset(ops->pad, 0, sizeof ops->pad);
2426 ops->rx_packets = htonll(stats.rx_packets);
2427 ops->tx_packets = htonll(stats.tx_packets);
2428 ops->rx_bytes = htonll(stats.rx_bytes);
2429 ops->tx_bytes = htonll(stats.tx_bytes);
2430 ops->rx_dropped = htonll(stats.rx_dropped);
2431 ops->tx_dropped = htonll(stats.tx_dropped);
2432 ops->rx_errors = htonll(stats.rx_errors);
2433 ops->tx_errors = htonll(stats.tx_errors);
2434 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2435 ops->rx_over_err = htonll(stats.rx_over_errors);
2436 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2437 ops->collisions = htonll(stats.collisions);
2440 queue_tx(msg, ofconn, ofconn->reply_counter);
2444 struct flow_stats_cbdata {
2445 struct ofproto *ofproto;
2446 struct ofconn *ofconn;
2452 query_stats(struct ofproto *p, struct rule *rule,
2453 uint64_t *packet_countp, uint64_t *byte_countp)
2455 uint64_t packet_count, byte_count;
2456 struct rule *subrule;
2457 struct odp_flow *odp_flows;
2460 packet_count = rule->packet_count;
2461 byte_count = rule->byte_count;
2463 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2464 odp_flows = xcalloc(1, n_odp_flows * sizeof *odp_flows);
2465 if (rule->cr.wc.wildcards) {
2467 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2468 odp_flows[i++].key = subrule->cr.flow;
2469 packet_count += subrule->packet_count;
2470 byte_count += subrule->byte_count;
2473 odp_flows[0].key = rule->cr.flow;
2476 if (!dpif_flow_get_multiple(&p->dpif, odp_flows, n_odp_flows)) {
2478 for (i = 0; i < n_odp_flows; i++) {
2479 struct odp_flow *odp_flow = &odp_flows[i];
2480 packet_count += odp_flow->stats.n_packets;
2481 byte_count += odp_flow->stats.n_bytes;
2486 *packet_countp = packet_count;
2487 *byte_countp = byte_count;
2491 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2493 struct rule *rule = rule_from_cls_rule(rule_);
2494 struct flow_stats_cbdata *cbdata = cbdata_;
2495 struct ofp_flow_stats *ofs;
2496 uint64_t packet_count, byte_count;
2497 size_t act_len, len;
2499 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2503 act_len = sizeof *rule->actions * rule->n_actions;
2504 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2506 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2508 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2509 ofs->length = htons(len);
2510 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2512 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match);
2513 ofs->duration = htonl((time_msec() - rule->created) / 1000);
2514 ofs->priority = htons(rule->cr.priority);
2515 ofs->idle_timeout = htons(rule->idle_timeout);
2516 ofs->hard_timeout = htons(rule->hard_timeout);
2517 memset(ofs->pad2, 0, sizeof ofs->pad2);
2518 ofs->packet_count = htonll(packet_count);
2519 ofs->byte_count = htonll(byte_count);
2520 memcpy(ofs->actions, rule->actions, act_len);
2524 table_id_to_include(uint8_t table_id)
2526 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2527 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2528 : table_id == 0xff ? CLS_INC_ALL
2533 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2534 const struct ofp_stats_request *osr,
2537 struct ofp_flow_stats_request *fsr;
2538 struct flow_stats_cbdata cbdata;
2539 struct cls_rule target;
2541 if (arg_size != sizeof *fsr) {
2542 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2544 fsr = (struct ofp_flow_stats_request *) osr->body;
2546 COVERAGE_INC(ofproto_flows_req);
2548 cbdata.ofconn = ofconn;
2549 cbdata.out_port = fsr->out_port;
2550 cbdata.msg = start_stats_reply(osr, 1024);
2551 cls_rule_from_match(&target, &fsr->match, 0);
2552 classifier_for_each_match(&p->cls, &target,
2553 table_id_to_include(fsr->table_id),
2554 flow_stats_cb, &cbdata);
2555 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2559 struct flow_stats_ds_cbdata {
2560 struct ofproto *ofproto;
2565 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
2567 struct rule *rule = rule_from_cls_rule(rule_);
2568 struct flow_stats_ds_cbdata *cbdata = cbdata_;
2569 struct ds *results = cbdata->results;
2570 struct ofp_match match;
2571 uint64_t packet_count, byte_count;
2572 size_t act_len = sizeof *rule->actions * rule->n_actions;
2574 /* Don't report on subrules. */
2575 if (rule->super != NULL) {
2579 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2580 flow_to_ovs_match(&rule->cr.flow, rule->cr.wc.wildcards, &match);
2582 ds_put_format(results, "duration=%llds, ",
2583 (time_msec() - rule->created) / 1000);
2584 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2585 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2586 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2587 ofp_print_match(results, &match, true);
2588 ofp_print_actions(results, &rule->actions->header, act_len);
2589 ds_put_cstr(results, "\n");
2592 /* Adds a pretty-printed description of all flows to 'results', including
2593 * those marked hidden by secchan (e.g., by in-band control). */
2595 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2597 struct ofp_match match;
2598 struct cls_rule target;
2599 struct flow_stats_ds_cbdata cbdata;
2601 memset(&match, 0, sizeof match);
2602 match.wildcards = htonl(OFPFW_ALL);
2605 cbdata.results = results;
2607 cls_rule_from_match(&target, &match, 0);
2608 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2609 flow_stats_ds_cb, &cbdata);
2612 struct aggregate_stats_cbdata {
2613 struct ofproto *ofproto;
2615 uint64_t packet_count;
2616 uint64_t byte_count;
2621 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2623 struct rule *rule = rule_from_cls_rule(rule_);
2624 struct aggregate_stats_cbdata *cbdata = cbdata_;
2625 uint64_t packet_count, byte_count;
2627 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2631 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2633 cbdata->packet_count += packet_count;
2634 cbdata->byte_count += byte_count;
2639 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2640 const struct ofp_stats_request *osr,
2643 struct ofp_aggregate_stats_request *asr;
2644 struct ofp_aggregate_stats_reply *reply;
2645 struct aggregate_stats_cbdata cbdata;
2646 struct cls_rule target;
2649 if (arg_size != sizeof *asr) {
2650 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2652 asr = (struct ofp_aggregate_stats_request *) osr->body;
2654 COVERAGE_INC(ofproto_agg_request);
2656 cbdata.out_port = asr->out_port;
2657 cbdata.packet_count = 0;
2658 cbdata.byte_count = 0;
2660 cls_rule_from_match(&target, &asr->match, 0);
2661 classifier_for_each_match(&p->cls, &target,
2662 table_id_to_include(asr->table_id),
2663 aggregate_stats_cb, &cbdata);
2665 msg = start_stats_reply(osr, sizeof *reply);
2666 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2667 reply->flow_count = htonl(cbdata.n_flows);
2668 reply->packet_count = htonll(cbdata.packet_count);
2669 reply->byte_count = htonll(cbdata.byte_count);
2670 queue_tx(msg, ofconn, ofconn->reply_counter);
2675 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2676 struct ofp_header *oh)
2678 struct ofp_stats_request *osr;
2682 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2687 osr = (struct ofp_stats_request *) oh;
2689 switch (ntohs(osr->type)) {
2691 return handle_desc_stats_request(p, ofconn, osr);
2694 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2696 case OFPST_AGGREGATE:
2697 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2700 return handle_table_stats_request(p, ofconn, osr);
2703 return handle_port_stats_request(p, ofconn, osr);
2706 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2709 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2713 static long long int
2714 msec_from_nsec(uint64_t sec, uint32_t nsec)
2716 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2720 update_time(struct rule *rule, const struct odp_flow_stats *stats)
2722 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2723 if (used > rule->used) {
2729 update_stats(struct rule *rule, const struct odp_flow_stats *stats)
2731 update_time(rule, stats);
2732 rule->packet_count += stats->n_packets;
2733 rule->byte_count += stats->n_bytes;
2734 rule->tcp_flags |= stats->tcp_flags;
2735 if (stats->n_packets) {
2736 rule->ip_tos = stats->ip_tos;
2741 add_flow(struct ofproto *p, struct ofconn *ofconn,
2742 struct ofp_flow_mod *ofm, size_t n_actions)
2744 struct ofpbuf *packet;
2749 rule = rule_create(NULL, (const union ofp_action *) ofm->actions,
2750 n_actions, ntohs(ofm->idle_timeout),
2751 ntohs(ofm->hard_timeout));
2752 cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority));
2756 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2757 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2761 rule_insert(p, rule, packet, in_port);
2762 ofpbuf_delete(packet);
2767 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
2768 size_t n_actions, uint16_t command, struct rule *rule)
2770 if (rule_is_hidden(rule)) {
2774 if (command == OFPFC_DELETE) {
2775 rule_remove(p, rule);
2777 size_t actions_len = n_actions * sizeof *rule->actions;
2779 if (n_actions == rule->n_actions
2780 && !memcmp(ofm->actions, rule->actions, actions_len))
2785 free(rule->actions);
2786 rule->actions = xmemdup(ofm->actions, actions_len);
2787 rule->n_actions = n_actions;
2789 if (rule->cr.wc.wildcards) {
2790 COVERAGE_INC(ofproto_mod_wc_flow);
2791 p->need_revalidate = true;
2793 rule_update_actions(p, rule);
2801 modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm,
2802 size_t n_actions, uint16_t command)
2808 flow_from_match(&flow, &wildcards, &ofm->match);
2809 rule = rule_from_cls_rule(classifier_find_rule_exactly(
2810 &p->cls, &flow, wildcards,
2811 ntohs(ofm->priority)));
2814 if (command == OFPFC_DELETE
2815 && ofm->out_port != htons(OFPP_NONE)
2816 && !rule_has_out_port(rule, ofm->out_port)) {
2820 modify_flow(p, ofm, n_actions, command, rule);
2825 struct modify_flows_cbdata {
2826 struct ofproto *ofproto;
2827 const struct ofp_flow_mod *ofm;
2834 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
2836 struct rule *rule = rule_from_cls_rule(rule_);
2837 struct modify_flows_cbdata *cbdata = cbdata_;
2839 if (cbdata->out_port != htons(OFPP_NONE)
2840 && !rule_has_out_port(rule, cbdata->out_port)) {
2844 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions,
2845 cbdata->command, rule);
2849 modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm,
2850 size_t n_actions, uint16_t command)
2852 struct modify_flows_cbdata cbdata;
2853 struct cls_rule target;
2857 cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port
2858 : htons(OFPP_NONE));
2859 cbdata.n_actions = n_actions;
2860 cbdata.command = command;
2862 cls_rule_from_match(&target, &ofm->match, 0);
2864 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2865 modify_flows_cb, &cbdata);
2870 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
2871 struct ofp_flow_mod *ofm)
2876 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
2877 sizeof *ofm->actions, &n_actions);
2882 normalize_match(&ofm->match);
2883 if (!ofm->match.wildcards) {
2884 ofm->priority = htons(UINT16_MAX);
2887 error = validate_actions((const union ofp_action *) ofm->actions,
2888 n_actions, p->max_ports);
2893 switch (ntohs(ofm->command)) {
2895 return add_flow(p, ofconn, ofm, n_actions);
2898 return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY);
2900 case OFPFC_MODIFY_STRICT:
2901 return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY);
2904 return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE);
2906 case OFPFC_DELETE_STRICT:
2907 return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE);
2910 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
2915 send_capability_reply(struct ofproto *p, struct ofconn *ofconn, uint32_t xid)
2917 struct ofmp_capability_reply *ocr;
2919 char capabilities[] = "com.nicira.mgmt.manager=false\n";
2921 ocr = make_openflow_xid(sizeof(*ocr), OFPT_VENDOR, xid, &b);
2922 ocr->header.header.vendor = htonl(NX_VENDOR_ID);
2923 ocr->header.header.subtype = htonl(NXT_MGMT);
2924 ocr->header.type = htons(OFMPT_CAPABILITY_REPLY);
2926 ocr->format = htonl(OFMPCOF_SIMPLE);
2927 ocr->mgmt_id = htonll(p->mgmt_id);
2929 ofpbuf_put(b, capabilities, strlen(capabilities));
2931 queue_tx(b, ofconn, ofconn->reply_counter);
2935 handle_ofmp(struct ofproto *p, struct ofconn *ofconn,
2936 struct ofmp_header *ofmph)
2938 size_t msg_len = ntohs(ofmph->header.header.length);
2939 if (msg_len < sizeof(*ofmph)) {
2940 VLOG_WARN_RL(&rl, "dropping short managment message: %d\n", msg_len);
2941 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2944 if (ofmph->type == htons(OFMPT_CAPABILITY_REQUEST)) {
2945 struct ofmp_capability_request *ofmpcr;
2947 if (msg_len < sizeof(struct ofmp_capability_request)) {
2948 VLOG_WARN_RL(&rl, "dropping short capability request: %d\n",
2950 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2953 ofmpcr = (struct ofmp_capability_request *)ofmph;
2954 if (ofmpcr->format != htonl(OFMPCAF_SIMPLE)) {
2955 /* xxx Find a better type than bad subtype */
2956 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2959 send_capability_reply(p, ofconn, ofmph->header.header.xid);
2962 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2967 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
2969 struct ofp_vendor_header *ovh = msg;
2970 struct nicira_header *nh;
2972 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
2973 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2975 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
2976 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2978 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
2979 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2983 switch (ntohl(nh->subtype)) {
2984 case NXT_STATUS_REQUEST:
2985 return switch_status_handle_request(p->switch_status, ofconn->rconn,
2988 case NXT_ACT_SET_CONFIG:
2989 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
2991 case NXT_ACT_GET_CONFIG:
2992 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
2994 case NXT_COMMAND_REQUEST:
2996 return executer_handle_request(p->executer, ofconn->rconn, msg);
3001 return handle_ofmp(p, ofconn, msg);
3004 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3008 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3009 struct ofpbuf *ofp_msg)
3011 struct ofp_header *oh = ofp_msg->data;
3014 COVERAGE_INC(ofproto_recv_openflow);
3016 case OFPT_ECHO_REQUEST:
3017 error = handle_echo_request(ofconn, oh);
3020 case OFPT_ECHO_REPLY:
3024 case OFPT_FEATURES_REQUEST:
3025 error = handle_features_request(p, ofconn, oh);
3028 case OFPT_GET_CONFIG_REQUEST:
3029 error = handle_get_config_request(p, ofconn, oh);
3032 case OFPT_SET_CONFIG:
3033 error = handle_set_config(p, ofconn, ofp_msg->data);
3036 case OFPT_PACKET_OUT:
3037 error = handle_packet_out(p, ofconn, ofp_msg->data);
3041 error = handle_port_mod(p, oh);
3045 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3048 case OFPT_STATS_REQUEST:
3049 error = handle_stats_request(p, ofconn, oh);
3053 error = handle_vendor(p, ofconn, ofp_msg->data);
3057 if (VLOG_IS_WARN_ENABLED()) {
3058 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3059 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3062 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3067 send_error_oh(ofconn, ofp_msg->data, error);
3072 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3074 struct odp_msg *msg = packet->data;
3075 uint16_t in_port = odp_port_to_ofp_port(msg->port);
3077 struct ofpbuf payload;
3080 /* Handle controller actions. */
3081 if (msg->type == _ODPL_ACTION_NR) {
3082 COVERAGE_INC(ofproto_ctlr_action);
3083 pinsched_send(p->action_sched, in_port, packet,
3084 send_packet_in_action, p);
3088 payload.data = msg + 1;
3089 payload.size = msg->length - sizeof *msg;
3090 flow_extract(&payload, msg->port, &flow);
3092 /* Check with in-band control to see if this packet should be sent
3093 * to the local port regardless of the flow table. */
3094 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3095 union odp_action action;
3097 memset(&action, 0, sizeof(action));
3098 action.output.type = ODPAT_OUTPUT;
3099 action.output.port = ODPP_LOCAL;
3100 dpif_execute(&p->dpif, flow.in_port, &action, 1, &payload);
3103 rule = lookup_valid_rule(p, &flow);
3105 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3106 struct ofport *port = port_array_get(&p->ports, msg->port);
3108 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3109 COVERAGE_INC(ofproto_no_packet_in);
3110 /* XXX install 'drop' flow entry */
3111 ofpbuf_delete(packet);
3115 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3118 COVERAGE_INC(ofproto_packet_in);
3119 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3123 if (rule->cr.wc.wildcards) {
3124 rule = rule_create_subrule(p, rule, &flow);
3125 rule_make_actions(p, rule, packet);
3127 if (!rule->may_install) {
3128 /* The rule is not installable, that is, we need to process every
3129 * packet, so process the current packet and set its actions into
3131 rule_make_actions(p, rule, packet);
3133 /* XXX revalidate rule if it needs it */
3137 rule_execute(p, rule, &payload, &flow);
3138 rule_reinstall(p, rule);
3140 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY
3141 && rconn_is_connected(p->controller->rconn)) {
3143 * Extra-special case for fail-open mode.
3145 * We are in fail-open mode and the packet matched the fail-open rule,
3146 * but we are connected to a controller too. We should send the packet
3147 * up to the controller in the hope that it will try to set up a flow
3148 * and thereby allow us to exit fail-open.
3150 * See the top-level comment in fail-open.c for more information.
3152 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3154 ofpbuf_delete(packet);
3159 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3161 struct rule *sub = rule_from_cls_rule(sub_);
3162 struct revalidate_cbdata *cbdata = cbdata_;
3164 if (cbdata->revalidate_all
3165 || (cbdata->revalidate_subrules && sub->super)
3166 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3167 revalidate_rule(cbdata->ofproto, sub);
3172 revalidate_rule(struct ofproto *p, struct rule *rule)
3174 const flow_t *flow = &rule->cr.flow;
3176 COVERAGE_INC(ofproto_revalidate_rule);
3179 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3181 rule_remove(p, rule);
3183 } else if (super != rule->super) {
3184 COVERAGE_INC(ofproto_revalidate_moved);
3185 list_remove(&rule->list);
3186 list_push_back(&super->list, &rule->list);
3187 rule->super = super;
3188 rule->hard_timeout = super->hard_timeout;
3189 rule->idle_timeout = super->idle_timeout;
3190 rule->created = super->created;
3195 rule_update_actions(p, rule);
3199 static struct ofpbuf *
3200 compose_flow_exp(const struct rule *rule, long long int now, uint8_t reason)
3202 struct ofp_flow_expired *ofe;
3205 ofe = make_openflow(sizeof *ofe, OFPT_FLOW_EXPIRED, &buf);
3206 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofe->match);
3207 ofe->priority = htons(rule->cr.priority);
3208 ofe->reason = reason;
3209 ofe->duration = htonl((now - rule->created) / 1000);
3210 ofe->packet_count = htonll(rule->packet_count);
3211 ofe->byte_count = htonll(rule->byte_count);
3217 send_flow_exp(struct ofproto *p, struct rule *rule,
3218 long long int now, uint8_t reason)
3220 struct ofconn *ofconn;
3221 struct ofconn *prev;
3224 /* We limit the maximum number of queued flow expirations it by accounting
3225 * them under the counter for replies. That works because preventing
3226 * OpenFlow requests from being processed also prevents new flows from
3227 * being added (and expiring). (It also prevents processing OpenFlow
3228 * requests that would not add new flows, so it is imperfect.) */
3231 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3232 if (ofconn->send_flow_exp && rconn_is_connected(ofconn->rconn)) {
3234 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3236 buf = compose_flow_exp(rule, now, reason);
3242 queue_tx(buf, prev, prev->reply_counter);
3247 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3249 assert(rule->installed);
3250 assert(!rule->cr.wc.wildcards);
3253 rule_remove(ofproto, rule);
3255 rule_uninstall(ofproto, rule);
3260 expire_rule(struct cls_rule *cls_rule, void *p_)
3262 struct ofproto *p = p_;
3263 struct rule *rule = rule_from_cls_rule(cls_rule);
3264 long long int hard_expire, idle_expire, expire, now;
3266 hard_expire = (rule->hard_timeout
3267 ? rule->created + rule->hard_timeout * 1000
3269 idle_expire = (rule->idle_timeout
3270 && (rule->super || list_is_empty(&rule->list))
3271 ? rule->used + rule->idle_timeout * 1000
3273 expire = MIN(hard_expire, idle_expire);
3274 if (expire == LLONG_MAX) {
3275 if (rule->installed && time_msec() >= rule->used + 5000) {
3276 uninstall_idle_flow(p, rule);
3283 if (rule->installed && now >= rule->used + 5000) {
3284 uninstall_idle_flow(p, rule);
3289 COVERAGE_INC(ofproto_expired);
3290 if (rule->cr.wc.wildcards) {
3291 /* Update stats. (This code will be a no-op if the rule expired
3292 * due to an idle timeout, because in that case the rule has no
3293 * subrules left.) */
3294 struct rule *subrule, *next;
3295 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3296 rule_remove(p, subrule);
3300 send_flow_exp(p, rule, now,
3302 ? OFPER_HARD_TIMEOUT : OFPER_IDLE_TIMEOUT));
3303 rule_remove(p, rule);
3307 update_used(struct ofproto *p)
3309 struct odp_flow *flows;
3314 error = dpif_flow_list_all(&p->dpif, &flows, &n_flows);
3319 for (i = 0; i < n_flows; i++) {
3320 struct odp_flow *f = &flows[i];
3323 rule = rule_from_cls_rule(
3324 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3325 if (!rule || !rule->installed) {
3326 COVERAGE_INC(ofproto_unexpected_rule);
3327 dpif_flow_del(&p->dpif, f);
3331 update_time(rule, &f->stats);
3332 rule_account(p, rule, f->stats.n_bytes);
3338 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3339 const struct ofpbuf *packet, int send_len)
3341 struct odp_msg *msg = packet->data;
3342 struct ofpbuf payload;
3346 /* Extract packet payload from 'msg'. */
3347 payload.data = msg + 1;
3348 payload.size = msg->length - sizeof *msg;
3350 /* Construct ofp_packet_in message. */
3351 reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3352 opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port), reason,
3353 &payload, send_len);
3356 rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100);
3360 send_packet_in_action(struct ofpbuf *packet, void *p_)
3362 struct ofproto *p = p_;
3363 struct ofconn *ofconn;
3364 struct odp_msg *msg;
3367 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3368 if (ofconn == p->controller || ofconn->miss_send_len) {
3369 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3372 ofpbuf_delete(packet);
3376 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3378 struct ofproto *p = p_;
3379 bool in_fail_open = p->fail_open && fail_open_is_active(p->fail_open);
3380 struct ofconn *ofconn;
3381 struct ofpbuf payload;
3382 struct odp_msg *msg;
3385 payload.data = msg + 1;
3386 payload.size = msg->length - sizeof *msg;
3387 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3388 if (ofconn->miss_send_len) {
3389 struct pktbuf *pb = ofconn->pktbuf;
3390 uint32_t buffer_id = (in_fail_open
3392 : pktbuf_save(pb, &payload, msg->port));
3393 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3395 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3398 ofpbuf_delete(packet);
3402 pick_datapath_id(struct dpif *dpif, uint64_t fallback_dpid)
3404 char local_name[IF_NAMESIZE];
3405 uint8_t ea[ETH_ADDR_LEN];
3408 error = dpif_get_name(dpif, local_name, sizeof local_name);
3410 error = netdev_nodev_get_etheraddr(local_name, ea);
3412 return eth_addr_to_uint64(ea);
3414 VLOG_WARN("could not get MAC address for %s (%s)",
3415 local_name, strerror(error));
3418 return fallback_dpid;
3422 pick_fallback_dpid(void)
3424 uint8_t ea[ETH_ADDR_LEN];
3425 eth_addr_random(ea);
3426 ea[0] = 0x00; /* Set Nicira OUI. */
3429 return eth_addr_to_uint64(ea);
3433 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3434 struct odp_actions *actions, tag_type *tags,
3435 uint16_t *nf_output_iface, void *ofproto_)
3437 struct ofproto *ofproto = ofproto_;
3440 /* Drop frames for reserved multicast addresses. */
3441 if (eth_addr_is_reserved(flow->dl_dst)) {
3445 /* Learn source MAC (but don't try to learn from revalidation). */
3446 if (packet != NULL) {
3447 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3450 /* The log messages here could actually be useful in debugging,
3451 * so keep the rate limit relatively high. */
3452 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3453 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3454 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3455 ofproto_revalidate(ofproto, rev_tag);
3459 /* Determine output port. */
3460 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3462 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
3463 } else if (out_port != flow->in_port) {
3464 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3465 *nf_output_iface = out_port;
3473 static const struct ofhooks default_ofhooks = {
3475 default_normal_ofhook_cb,