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"
30 #include "fail-open.h"
32 #include "mac-learning.h"
36 #include "ofp-print.h"
38 #include "openflow/nicira-ext.h"
39 #include "openflow/openflow.h"
40 #include "openflow/openflow-mgmt.h"
41 #include "openvswitch/datapath-protocol.h"
45 #include "poll-loop.h"
46 #include "port-array.h"
55 #include "vconn-ssl.h"
58 #define THIS_MODULE VLM_ofproto
68 TABLEID_CLASSIFIER = 1
72 struct netdev *netdev;
73 struct ofp_phy_port opp; /* In host byte order. */
76 static void ofport_free(struct ofport *);
77 static void hton_ofp_phy_port(struct ofp_phy_port *);
79 static int xlate_actions(const union ofp_action *in, size_t n_in,
80 const flow_t *flow, struct ofproto *ofproto,
81 const struct ofpbuf *packet,
82 struct odp_actions *out, tag_type *tags,
83 bool *may_setup_flow);
88 uint16_t idle_timeout; /* In seconds from time of last use. */
89 uint16_t hard_timeout; /* In seconds from time of creation. */
90 long long int used; /* Last-used time (0 if never used). */
91 long long int created; /* Creation time. */
92 uint64_t packet_count; /* Number of packets received. */
93 uint64_t byte_count; /* Number of bytes received. */
94 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
95 uint8_t tcp_flags; /* Bitwise-OR of all TCP flags seen. */
96 uint8_t ip_tos; /* Last-seen IP type-of-service. */
97 tag_type tags; /* Tags (set only by hooks). */
99 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
100 * exact-match rule (having cr.wc.wildcards of 0) generated from the
101 * wildcard rule 'super'. In this case, 'list' is an element of the
104 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
105 * a list of subrules. A super-rule with no wildcards (where
106 * cr.wc.wildcards is 0) will never have any subrules. */
112 * A subrule has no actions (it uses the super-rule's actions). */
114 union ofp_action *actions;
118 * A super-rule with wildcard fields never has ODP actions (since the
119 * datapath only supports exact-match flows). */
120 bool installed; /* Installed in datapath? */
121 bool may_install; /* True ordinarily; false if actions must
122 * be reassessed for every packet. */
124 union odp_action *odp_actions;
128 rule_is_hidden(const struct rule *rule)
130 /* Subrules are merely an implementation detail, so hide them from the
132 if (rule->super != NULL) {
136 /* Rules with priority higher than UINT16_MAX are set up by secchan itself
137 * (e.g. by in-band control) and are intentionally hidden from the
139 if (rule->cr.priority > UINT16_MAX) {
146 static struct rule *rule_create(struct rule *super, const union ofp_action *,
147 size_t n_actions, uint16_t idle_timeout,
148 uint16_t hard_timeout);
149 static void rule_free(struct rule *);
150 static void rule_destroy(struct ofproto *, struct rule *);
151 static struct rule *rule_from_cls_rule(const struct cls_rule *);
152 static void rule_insert(struct ofproto *, struct rule *,
153 struct ofpbuf *packet, uint16_t in_port);
154 static void rule_remove(struct ofproto *, struct rule *);
155 static bool rule_make_actions(struct ofproto *, struct rule *,
156 const struct ofpbuf *packet);
157 static void rule_install(struct ofproto *, struct rule *,
158 struct rule *displaced_rule);
159 static void rule_uninstall(struct ofproto *, struct rule *);
160 static void rule_post_uninstall(struct ofproto *, struct rule *);
165 struct pktbuf *pktbuf;
169 struct rconn_packet_counter *packet_in_counter;
171 /* Number of OpenFlow messages queued as replies to OpenFlow requests, and
172 * the maximum number before we stop reading OpenFlow requests. */
173 #define OFCONN_REPLY_MAX 100
174 struct rconn_packet_counter *reply_counter;
177 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *);
178 static void ofconn_destroy(struct ofconn *, struct ofproto *);
179 static void ofconn_run(struct ofconn *, struct ofproto *);
180 static void ofconn_wait(struct ofconn *);
181 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
182 struct rconn_packet_counter *counter);
186 uint64_t datapath_id; /* Datapath ID. */
187 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
188 uint64_t mgmt_id; /* Management channel identifier. */
189 char *manufacturer; /* Manufacturer. */
190 char *hardware; /* Hardware. */
191 char *software; /* Software version. */
192 char *serial; /* Serial number. */
196 struct dpifmon *dpifmon;
197 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
199 struct shash port_by_name;
203 struct switch_status *switch_status;
204 struct status_category *ss_cat;
205 struct in_band *in_band;
206 struct discovery *discovery;
207 struct fail_open *fail_open;
208 struct pinsched *miss_sched, *action_sched;
209 struct executer *executer;
210 struct netflow *netflow;
213 struct classifier cls;
214 bool need_revalidate;
215 long long int next_expiration;
216 struct tag_set revalidate_set;
218 /* OpenFlow connections. */
219 struct list all_conns;
220 struct ofconn *controller;
221 struct pvconn **listeners;
223 struct pvconn **snoops;
226 /* Hooks for ovs-vswitchd. */
227 const struct ofhooks *ofhooks;
230 /* Used by default ofhooks. */
231 struct mac_learning *ml;
234 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
236 static const struct ofhooks default_ofhooks;
238 static uint64_t pick_datapath_id(struct dpif *, uint64_t fallback_dpid);
239 static uint64_t pick_fallback_dpid(void);
240 static void send_packet_in_miss(struct ofpbuf *, void *ofproto);
241 static void send_packet_in_action(struct ofpbuf *, void *ofproto);
242 static void update_used(struct ofproto *);
243 static void update_stats(struct rule *, const struct odp_flow_stats *);
244 static void expire_rule(struct cls_rule *, void *ofproto);
245 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
246 static void revalidate_cb(struct cls_rule *rule_, void *p_);
248 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
250 static void handle_openflow(struct ofconn *, struct ofproto *,
253 static void refresh_port_group(struct ofproto *, unsigned int group);
254 static void update_port(struct ofproto *, const char *devname);
255 static int init_ports(struct ofproto *);
256 static void reinit_ports(struct ofproto *);
259 ofproto_create(const char *datapath, const struct ofhooks *ofhooks, void *aux,
260 struct ofproto **ofprotop)
262 struct dpifmon *dpifmon;
263 struct odp_stats stats;
270 /* Connect to datapath and start listening for messages. */
271 error = dpif_open(datapath, &dpif);
273 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
276 error = dpif_get_dp_stats(&dpif, &stats);
278 VLOG_ERR("failed to obtain stats for datapath %s: %s",
279 datapath, strerror(error));
283 error = dpif_set_listen_mask(&dpif, ODPL_MISS | ODPL_ACTION);
285 VLOG_ERR("failed to listen on datapath %s: %s",
286 datapath, strerror(error));
290 dpif_flow_flush(&dpif);
293 /* Start monitoring datapath ports for status changes. */
294 error = dpifmon_create(datapath, &dpifmon);
296 VLOG_ERR("failed to starting monitoring datapath %s: %s",
297 datapath, strerror(error));
302 /* Initialize settings. */
303 p = xcalloc(1, sizeof *p);
304 p->fallback_dpid = pick_fallback_dpid();
305 p->datapath_id = pick_datapath_id(&dpif, p->fallback_dpid);
306 VLOG_INFO("using datapath ID %012"PRIx64, p->datapath_id);
307 p->manufacturer = xstrdup("Nicira Networks, Inc.");
308 p->hardware = xstrdup("Reference Implementation");
309 p->software = xstrdup(VERSION BUILDNR);
310 p->serial = xstrdup("None");
312 /* Initialize datapath. */
314 p->dpifmon = dpifmon;
315 port_array_init(&p->ports);
316 shash_init(&p->port_by_name);
317 p->max_ports = stats.max_ports;
319 /* Initialize submodules. */
320 p->switch_status = switch_status_create(p);
324 p->miss_sched = p->action_sched = NULL;
328 /* Initialize flow table. */
329 classifier_init(&p->cls);
330 p->need_revalidate = false;
331 p->next_expiration = time_msec() + 1000;
332 tag_set_init(&p->revalidate_set);
334 /* Initialize OpenFlow connections. */
335 list_init(&p->all_conns);
336 p->controller = ofconn_create(p, rconn_create(15, 15));
337 p->controller->pktbuf = pktbuf_create();
338 p->controller->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
344 /* Initialize hooks. */
346 p->ofhooks = ofhooks;
350 p->ofhooks = &default_ofhooks;
352 p->ml = mac_learning_create();
355 /* Register switch status category. */
356 p->ss_cat = switch_status_register(p->switch_status, "remote",
357 rconn_status_cb, p->controller->rconn);
360 error = init_ports(p);
371 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
373 uint64_t old_dpid = p->datapath_id;
374 p->datapath_id = (datapath_id
376 : pick_datapath_id(&p->dpif, p->fallback_dpid));
377 if (p->datapath_id != old_dpid) {
378 VLOG_INFO("datapath ID changed to %012"PRIx64, p->datapath_id);
379 rconn_reconnect(p->controller->rconn);
384 ofproto_set_mgmt_id(struct ofproto *p, uint64_t mgmt_id)
386 p->mgmt_id = mgmt_id;
390 ofproto_set_probe_interval(struct ofproto *p, int probe_interval)
392 probe_interval = probe_interval ? MAX(probe_interval, 5) : 0;
393 rconn_set_probe_interval(p->controller->rconn, probe_interval);
395 int trigger_duration = probe_interval ? probe_interval * 3 : 15;
396 fail_open_set_trigger_duration(p->fail_open, trigger_duration);
401 ofproto_set_max_backoff(struct ofproto *p, int max_backoff)
403 rconn_set_max_backoff(p->controller->rconn, max_backoff);
407 ofproto_set_desc(struct ofproto *p,
408 const char *manufacturer, const char *hardware,
409 const char *software, const char *serial)
412 free(p->manufacturer);
413 p->manufacturer = xstrdup(manufacturer);
417 p->hardware = xstrdup(hardware);
421 p->software = xstrdup(software);
425 p->serial = xstrdup(serial);
430 ofproto_set_in_band(struct ofproto *p, bool in_band)
432 if (in_band != (p->in_band != NULL)) {
434 return in_band_create(p, &p->dpif, p->switch_status,
435 p->controller->rconn, &p->in_band);
437 ofproto_set_discovery(p, false, NULL, true);
438 in_band_destroy(p->in_band);
441 rconn_reconnect(p->controller->rconn);
447 ofproto_set_discovery(struct ofproto *p, bool discovery,
448 const char *re, bool update_resolv_conf)
450 if (discovery != (p->discovery != NULL)) {
452 int error = ofproto_set_in_band(p, true);
456 error = discovery_create(re, update_resolv_conf,
457 &p->dpif, p->switch_status,
463 discovery_destroy(p->discovery);
466 rconn_disconnect(p->controller->rconn);
467 } else if (discovery) {
468 discovery_set_update_resolv_conf(p->discovery, update_resolv_conf);
469 return discovery_set_accept_controller_re(p->discovery, re);
475 ofproto_set_controller(struct ofproto *ofproto, const char *controller)
477 if (ofproto->discovery) {
479 } else if (controller) {
480 if (strcmp(rconn_get_name(ofproto->controller->rconn), controller)) {
481 return rconn_connect(ofproto->controller->rconn, controller);
486 rconn_disconnect(ofproto->controller->rconn);
492 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
493 const struct svec *svec)
495 struct pvconn **pvconns = *pvconnsp;
496 size_t n_pvconns = *n_pvconnsp;
500 for (i = 0; i < n_pvconns; i++) {
501 pvconn_close(pvconns[i]);
505 pvconns = xmalloc(svec->n * sizeof *pvconns);
507 for (i = 0; i < svec->n; i++) {
508 const char *name = svec->names[i];
509 struct pvconn *pvconn;
512 error = pvconn_open(name, &pvconn);
514 pvconns[n_pvconns++] = pvconn;
516 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
524 *n_pvconnsp = n_pvconns;
530 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
532 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
536 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
538 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
542 ofproto_set_netflow(struct ofproto *ofproto, const struct svec *collectors,
543 uint8_t engine_type, uint8_t engine_id, bool add_id_to_iface)
545 if (collectors && collectors->n) {
546 if (!ofproto->netflow) {
547 ofproto->netflow = netflow_create();
549 netflow_set_engine(ofproto->netflow, engine_type, engine_id,
551 return netflow_set_collectors(ofproto->netflow, collectors);
553 netflow_destroy(ofproto->netflow);
554 ofproto->netflow = NULL;
560 ofproto_set_failure(struct ofproto *ofproto, bool fail_open)
563 struct rconn *rconn = ofproto->controller->rconn;
564 int trigger_duration = rconn_get_probe_interval(rconn) * 3;
565 if (!ofproto->fail_open) {
566 ofproto->fail_open = fail_open_create(ofproto, trigger_duration,
567 ofproto->switch_status,
570 fail_open_set_trigger_duration(ofproto->fail_open,
574 fail_open_destroy(ofproto->fail_open);
575 ofproto->fail_open = NULL;
580 ofproto_set_rate_limit(struct ofproto *ofproto,
581 int rate_limit, int burst_limit)
583 if (rate_limit > 0) {
584 if (!ofproto->miss_sched) {
585 ofproto->miss_sched = pinsched_create(rate_limit, burst_limit,
586 ofproto->switch_status);
587 ofproto->action_sched = pinsched_create(rate_limit, burst_limit,
590 pinsched_set_limits(ofproto->miss_sched, rate_limit, burst_limit);
591 pinsched_set_limits(ofproto->action_sched,
592 rate_limit, burst_limit);
595 pinsched_destroy(ofproto->miss_sched);
596 ofproto->miss_sched = NULL;
597 pinsched_destroy(ofproto->action_sched);
598 ofproto->action_sched = NULL;
603 ofproto_set_stp(struct ofproto *ofproto UNUSED, bool enable_stp)
607 VLOG_WARN("STP is not yet implemented");
615 ofproto_set_remote_execution(struct ofproto *ofproto, const char *command_acl,
616 const char *command_dir)
619 if (!ofproto->executer) {
620 return executer_create(command_acl, command_dir,
623 executer_set_acl(ofproto->executer, command_acl, command_dir);
626 executer_destroy(ofproto->executer);
627 ofproto->executer = NULL;
633 ofproto_get_datapath_id(const struct ofproto *ofproto)
635 return ofproto->datapath_id;
639 ofproto_get_probe_interval(const struct ofproto *ofproto)
641 return rconn_get_probe_interval(ofproto->controller->rconn);
645 ofproto_get_max_backoff(const struct ofproto *ofproto)
647 return rconn_get_max_backoff(ofproto->controller->rconn);
651 ofproto_get_in_band(const struct ofproto *ofproto)
653 return ofproto->in_band != NULL;
657 ofproto_get_discovery(const struct ofproto *ofproto)
659 return ofproto->discovery != NULL;
663 ofproto_get_controller(const struct ofproto *ofproto)
665 return rconn_get_name(ofproto->controller->rconn);
669 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
673 for (i = 0; i < ofproto->n_listeners; i++) {
674 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
679 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
683 for (i = 0; i < ofproto->n_snoops; i++) {
684 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
689 ofproto_destroy(struct ofproto *p)
691 struct ofconn *ofconn, *next_ofconn;
692 struct ofport *ofport;
693 unsigned int port_no;
700 ofproto_flush_flows(p);
701 classifier_destroy(&p->cls);
703 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
705 ofconn_destroy(ofconn, p);
708 dpif_close(&p->dpif);
709 dpifmon_destroy(p->dpifmon);
710 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
713 shash_destroy(&p->port_by_name);
715 switch_status_destroy(p->switch_status);
716 in_band_destroy(p->in_band);
717 discovery_destroy(p->discovery);
718 fail_open_destroy(p->fail_open);
719 pinsched_destroy(p->miss_sched);
720 pinsched_destroy(p->action_sched);
721 executer_destroy(p->executer);
722 netflow_destroy(p->netflow);
724 switch_status_unregister(p->ss_cat);
726 for (i = 0; i < p->n_listeners; i++) {
727 pvconn_close(p->listeners[i]);
731 for (i = 0; i < p->n_snoops; i++) {
732 pvconn_close(p->snoops[i]);
736 mac_learning_destroy(p->ml);
742 ofproto_run(struct ofproto *p)
744 int error = ofproto_run1(p);
746 error = ofproto_run2(p, false);
752 ofproto_run1(struct ofproto *p)
754 struct ofconn *ofconn, *next_ofconn;
759 for (i = 0; i < 50; i++) {
763 error = dpif_recv(&p->dpif, &buf);
765 if (error == ENODEV) {
766 /* Someone destroyed the datapath behind our back. The caller
767 * better destroy us and give up, because we're just going to
768 * spin from here on out. */
769 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
770 VLOG_ERR_RL(&rl, "%s: datapath was destroyed externally",
771 dpif_name(&p->dpif));
777 handle_odp_msg(p, buf);
780 while ((error = dpifmon_poll(p->dpifmon, &devname)) != EAGAIN) {
781 if (error == ENOBUFS) {
784 update_port(p, devname);
790 in_band_run(p->in_band);
793 char *controller_name;
794 if (rconn_is_connectivity_questionable(p->controller->rconn)) {
795 discovery_question_connectivity(p->discovery);
797 if (discovery_run(p->discovery, &controller_name)) {
798 if (controller_name) {
799 rconn_connect(p->controller->rconn, controller_name);
801 rconn_disconnect(p->controller->rconn);
806 fail_open_run(p->fail_open);
808 pinsched_run(p->miss_sched, send_packet_in_miss, p);
809 pinsched_run(p->action_sched, send_packet_in_action, p);
811 executer_run(p->executer);
814 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
816 ofconn_run(ofconn, p);
819 for (i = 0; i < p->n_listeners; i++) {
823 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
825 ofconn_create(p, rconn_new_from_vconn("passive", vconn));
826 } else if (retval != EAGAIN) {
827 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
831 for (i = 0; i < p->n_snoops; i++) {
835 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
837 rconn_add_monitor(p->controller->rconn, vconn);
838 } else if (retval != EAGAIN) {
839 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
843 if (time_msec() >= p->next_expiration) {
844 COVERAGE_INC(ofproto_expiration);
845 p->next_expiration = time_msec() + 1000;
848 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
850 /* Let the hook know that we're at a stable point: all outstanding data
851 * in existing flows has been accounted to the account_cb. Thus, the
852 * hook can now reasonably do operations that depend on having accurate
853 * flow volume accounting (currently, that's just bond rebalancing). */
854 if (p->ofhooks->account_checkpoint_cb) {
855 p->ofhooks->account_checkpoint_cb(p->aux);
860 netflow_run(p->netflow);
866 struct revalidate_cbdata {
867 struct ofproto *ofproto;
868 bool revalidate_all; /* Revalidate all exact-match rules? */
869 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
870 struct tag_set revalidate_set; /* Set of tags to revalidate. */
874 ofproto_run2(struct ofproto *p, bool revalidate_all)
876 if (p->need_revalidate || revalidate_all
877 || !tag_set_is_empty(&p->revalidate_set)) {
878 struct revalidate_cbdata cbdata;
880 cbdata.revalidate_all = revalidate_all;
881 cbdata.revalidate_subrules = p->need_revalidate;
882 cbdata.revalidate_set = p->revalidate_set;
883 tag_set_init(&p->revalidate_set);
884 COVERAGE_INC(ofproto_revalidate);
885 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
886 p->need_revalidate = false;
893 ofproto_wait(struct ofproto *p)
895 struct ofconn *ofconn;
898 dpif_recv_wait(&p->dpif);
899 dpifmon_wait(p->dpifmon);
900 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
904 in_band_wait(p->in_band);
907 discovery_wait(p->discovery);
910 fail_open_wait(p->fail_open);
912 pinsched_wait(p->miss_sched);
913 pinsched_wait(p->action_sched);
915 executer_wait(p->executer);
917 if (!tag_set_is_empty(&p->revalidate_set)) {
918 poll_immediate_wake();
920 if (p->need_revalidate) {
921 /* Shouldn't happen, but if it does just go around again. */
922 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
923 poll_immediate_wake();
924 } else if (p->next_expiration != LLONG_MAX) {
925 poll_timer_wait(p->next_expiration - time_msec());
927 for (i = 0; i < p->n_listeners; i++) {
928 pvconn_wait(p->listeners[i]);
930 for (i = 0; i < p->n_snoops; i++) {
931 pvconn_wait(p->snoops[i]);
936 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
938 tag_set_add(&ofproto->revalidate_set, tag);
942 ofproto_get_revalidate_set(struct ofproto *ofproto)
944 return &ofproto->revalidate_set;
948 ofproto_is_alive(const struct ofproto *p)
950 return p->discovery || rconn_is_alive(p->controller->rconn);
954 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
955 const union ofp_action *actions, size_t n_actions,
956 const struct ofpbuf *packet)
958 struct odp_actions odp_actions;
961 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
967 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
969 dpif_execute(&p->dpif, flow->in_port, odp_actions.actions,
970 odp_actions.n_actions, packet);
975 ofproto_add_flow(struct ofproto *p,
976 const flow_t *flow, uint32_t wildcards, unsigned int priority,
977 const union ofp_action *actions, size_t n_actions,
981 rule = rule_create(NULL, actions, n_actions,
982 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */, 0);
983 cls_rule_from_flow(&rule->cr, flow, wildcards, priority);
984 rule_insert(p, rule, NULL, 0);
988 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
989 uint32_t wildcards, unsigned int priority)
993 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
997 rule_remove(ofproto, rule);
1002 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1004 struct rule *rule = rule_from_cls_rule(rule_);
1005 struct ofproto *ofproto = ofproto_;
1007 /* Mark the flow as not installed, even though it might really be
1008 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1009 * There is no point in uninstalling it individually since we are about to
1010 * blow away all the flows with dpif_flow_flush(). */
1011 rule->installed = false;
1013 rule_remove(ofproto, rule);
1017 ofproto_flush_flows(struct ofproto *ofproto)
1019 COVERAGE_INC(ofproto_flush);
1020 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1021 dpif_flow_flush(&ofproto->dpif);
1022 if (ofproto->in_band) {
1023 in_band_flushed(ofproto->in_band);
1025 if (ofproto->fail_open) {
1026 fail_open_flushed(ofproto->fail_open);
1031 reinit_ports(struct ofproto *p)
1033 struct svec devnames;
1034 struct ofport *ofport;
1035 unsigned int port_no;
1036 struct odp_port *odp_ports;
1040 svec_init(&devnames);
1041 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1042 svec_add (&devnames, (char *) ofport->opp.name);
1044 dpif_port_list(&p->dpif, &odp_ports, &n_odp_ports);
1045 for (i = 0; i < n_odp_ports; i++) {
1046 svec_add (&devnames, odp_ports[i].devname);
1050 svec_sort_unique(&devnames);
1051 for (i = 0; i < devnames.n; i++) {
1052 update_port(p, devnames.names[i]);
1054 svec_destroy(&devnames);
1058 refresh_port_group(struct ofproto *p, unsigned int group)
1062 struct ofport *port;
1063 unsigned int port_no;
1065 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1067 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1069 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1070 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1071 ports[n_ports++] = port_no;
1074 dpif_port_group_set(&p->dpif, group, ports, n_ports);
1079 refresh_port_groups(struct ofproto *p)
1081 refresh_port_group(p, DP_GROUP_FLOOD);
1082 refresh_port_group(p, DP_GROUP_ALL);
1085 static struct ofport *
1086 make_ofport(const struct odp_port *odp_port)
1088 enum netdev_flags flags;
1089 struct ofport *ofport;
1090 struct netdev *netdev;
1094 error = netdev_open(odp_port->devname, NETDEV_ETH_TYPE_NONE, &netdev);
1096 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1097 "cannot be opened (%s)",
1098 odp_port->devname, odp_port->port,
1099 odp_port->devname, strerror(error));
1103 ofport = xmalloc(sizeof *ofport);
1104 ofport->netdev = netdev;
1105 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1106 memcpy(ofport->opp.hw_addr, netdev_get_etheraddr(netdev), ETH_ALEN);
1107 memcpy(ofport->opp.name, odp_port->devname,
1108 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1109 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1111 netdev_get_flags(netdev, &flags);
1112 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1114 netdev_get_carrier(netdev, &carrier);
1115 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1117 netdev_get_features(netdev,
1118 &ofport->opp.curr, &ofport->opp.advertised,
1119 &ofport->opp.supported, &ofport->opp.peer);
1124 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1126 if (port_array_get(&p->ports, odp_port->port)) {
1127 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1130 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1131 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1140 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1142 const struct ofp_phy_port *a = &a_->opp;
1143 const struct ofp_phy_port *b = &b_->opp;
1145 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1146 return (a->port_no == b->port_no
1147 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1148 && !strcmp((char *) a->name, (char *) b->name)
1149 && a->state == b->state
1150 && a->config == b->config
1151 && a->curr == b->curr
1152 && a->advertised == b->advertised
1153 && a->supported == b->supported
1154 && a->peer == b->peer);
1158 send_port_status(struct ofproto *p, const struct ofport *ofport,
1161 /* XXX Should limit the number of queued port status change messages. */
1162 struct ofconn *ofconn;
1163 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1164 struct ofp_port_status *ops;
1167 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1168 ops->reason = reason;
1169 ops->desc = ofport->opp;
1170 hton_ofp_phy_port(&ops->desc);
1171 queue_tx(b, ofconn, NULL);
1173 if (p->ofhooks->port_changed_cb) {
1174 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1179 ofport_install(struct ofproto *p, struct ofport *ofport)
1181 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no),
1183 shash_add(&p->port_by_name, (char *) ofport->opp.name, ofport);
1187 ofport_remove(struct ofproto *p, struct ofport *ofport)
1189 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no), NULL);
1190 shash_delete(&p->port_by_name,
1191 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1195 ofport_free(struct ofport *ofport)
1198 netdev_close(ofport->netdev);
1204 update_port(struct ofproto *p, const char *devname)
1206 struct odp_port odp_port;
1207 struct ofport *ofport;
1210 COVERAGE_INC(ofproto_update_port);
1211 ofport = shash_find_data(&p->port_by_name, devname);
1212 error = dpif_port_query_by_name(&p->dpif, devname, &odp_port);
1216 if (!ofport_conflicts(p, &odp_port)) {
1217 ofport = make_ofport(&odp_port);
1219 ofport_install(p, ofport);
1220 send_port_status(p, ofport, OFPPR_ADD);
1224 /* Modified port. */
1225 struct ofport *new_ofport = make_ofport(&odp_port);
1230 new_ofport->opp.config &= OFPPC_PORT_DOWN;
1231 new_ofport->opp.config |= ofport->opp.config & ~OFPPC_PORT_DOWN;
1232 if (ofport_equal(ofport, new_ofport)) {
1233 /* False alarm--no change. */
1234 ofport_free(new_ofport);
1236 ofport_remove(p, ofport);
1237 ofport_install(p, new_ofport);
1238 ofport_free(ofport);
1239 send_port_status(p, new_ofport, OFPPR_MODIFY);
1242 } else if (error == ENOENT || error == ENODEV) {
1245 send_port_status(p, ofport, OFPPR_DELETE);
1246 ofport_remove(p, ofport);
1247 ofport_free(ofport);
1250 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1251 "%s", strerror(error));
1254 refresh_port_groups(p);
1258 init_ports(struct ofproto *p)
1260 struct odp_port *ports;
1265 error = dpif_port_list(&p->dpif, &ports, &n_ports);
1270 for (i = 0; i < n_ports; i++) {
1271 const struct odp_port *odp_port = &ports[i];
1272 if (!ofport_conflicts(p, odp_port)) {
1273 struct ofport *ofport = make_ofport(odp_port);
1275 ofport_install(p, ofport);
1280 refresh_port_groups(p);
1284 static struct ofconn *
1285 ofconn_create(struct ofproto *p, struct rconn *rconn)
1287 struct ofconn *ofconn = xmalloc(sizeof *ofconn);
1288 list_push_back(&p->all_conns, &ofconn->node);
1289 ofconn->rconn = rconn;
1290 ofconn->pktbuf = NULL;
1291 ofconn->send_flow_exp = false;
1292 ofconn->miss_send_len = 0;
1293 ofconn->packet_in_counter = rconn_packet_counter_create ();
1294 ofconn->reply_counter = rconn_packet_counter_create ();
1299 ofconn_destroy(struct ofconn *ofconn, struct ofproto *p)
1302 executer_rconn_closing(p->executer, ofconn->rconn);
1305 list_remove(&ofconn->node);
1306 rconn_destroy(ofconn->rconn);
1307 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1308 rconn_packet_counter_destroy(ofconn->reply_counter);
1309 pktbuf_destroy(ofconn->pktbuf);
1314 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1318 rconn_run(ofconn->rconn);
1320 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1321 /* Limit the number of iterations to prevent other tasks from
1323 for (iteration = 0; iteration < 50; iteration++) {
1324 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1328 handle_openflow(ofconn, p, of_msg);
1329 ofpbuf_delete(of_msg);
1333 if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
1334 ofconn_destroy(ofconn, p);
1339 ofconn_wait(struct ofconn *ofconn)
1341 rconn_run_wait(ofconn->rconn);
1342 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1343 rconn_recv_wait(ofconn->rconn);
1345 COVERAGE_INC(ofproto_ofconn_stuck);
1349 /* Caller is responsible for initializing the 'cr' member of the returned
1351 static struct rule *
1352 rule_create(struct rule *super,
1353 const union ofp_action *actions, size_t n_actions,
1354 uint16_t idle_timeout, uint16_t hard_timeout)
1356 struct rule *rule = xcalloc(1, sizeof *rule);
1357 rule->idle_timeout = idle_timeout;
1358 rule->hard_timeout = hard_timeout;
1359 rule->used = rule->created = time_msec();
1360 rule->super = super;
1362 list_push_back(&super->list, &rule->list);
1364 list_init(&rule->list);
1366 rule->n_actions = n_actions;
1367 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1371 static struct rule *
1372 rule_from_cls_rule(const struct cls_rule *cls_rule)
1374 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1378 rule_free(struct rule *rule)
1380 free(rule->actions);
1381 free(rule->odp_actions);
1385 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1386 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1387 * through all of its subrules and revalidates them, destroying any that no
1388 * longer has a super-rule (which is probably all of them).
1390 * Before calling this function, the caller must make have removed 'rule' from
1391 * the classifier. If 'rule' is an exact-match rule, the caller is also
1392 * responsible for ensuring that it has been uninstalled from the datapath. */
1394 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1397 struct rule *subrule, *next;
1398 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1399 revalidate_rule(ofproto, subrule);
1402 list_remove(&rule->list);
1408 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1410 const union ofp_action *oa;
1411 struct actions_iterator i;
1413 if (out_port == htons(OFPP_NONE)) {
1416 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1417 oa = actions_next(&i)) {
1418 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1425 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1426 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1428 * The flow that 'packet' actually contains does not need to actually match
1429 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1430 * the packet and byte counters for 'rule' will be credited for the packet sent
1431 * out whether or not the packet actually matches 'rule'.
1433 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1434 * the caller must already have accurately composed ODP actions for it given
1435 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1436 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1437 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1438 * actions and apply them to 'packet'. */
1440 rule_execute(struct ofproto *ofproto, struct rule *rule,
1441 struct ofpbuf *packet, const flow_t *flow)
1443 const union odp_action *actions;
1445 struct odp_actions a;
1447 /* Grab or compose the ODP actions.
1449 * The special case for an exact-match 'rule' where 'flow' is not the
1450 * rule's flow is important to avoid, e.g., sending a packet out its input
1451 * port simply because the ODP actions were composed for the wrong
1453 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1454 struct rule *super = rule->super ? rule->super : rule;
1455 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1456 packet, &a, NULL, 0)) {
1459 actions = a.actions;
1460 n_actions = a.n_actions;
1462 actions = rule->odp_actions;
1463 n_actions = rule->n_odp_actions;
1466 /* Execute the ODP actions. */
1467 if (!dpif_execute(&ofproto->dpif, flow->in_port,
1468 actions, n_actions, packet)) {
1469 struct odp_flow_stats stats;
1470 flow_extract_stats(flow, packet, &stats);
1471 update_stats(rule, &stats);
1472 rule->used = time_msec();
1477 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1480 struct rule *displaced_rule;
1482 /* Insert the rule in the classifier. */
1483 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1484 if (!rule->cr.wc.wildcards) {
1485 rule_make_actions(p, rule, packet);
1488 /* Send the packet and credit it to the rule. */
1491 flow_extract(packet, in_port, &flow);
1492 rule_execute(p, rule, packet, &flow);
1495 /* Install the rule in the datapath only after sending the packet, to
1496 * avoid packet reordering. */
1497 if (rule->cr.wc.wildcards) {
1498 COVERAGE_INC(ofproto_add_wc_flow);
1499 p->need_revalidate = true;
1501 rule_install(p, rule, displaced_rule);
1504 /* Free the rule that was displaced, if any. */
1505 if (displaced_rule) {
1506 rule_destroy(p, displaced_rule);
1510 static struct rule *
1511 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1514 struct rule *subrule = rule_create(rule, NULL, 0,
1515 rule->idle_timeout, rule->hard_timeout);
1516 COVERAGE_INC(ofproto_subrule_create);
1517 cls_rule_from_flow(&subrule->cr, flow, 0,
1518 (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1519 : rule->cr.priority));
1520 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1526 rule_remove(struct ofproto *ofproto, struct rule *rule)
1528 if (rule->cr.wc.wildcards) {
1529 COVERAGE_INC(ofproto_del_wc_flow);
1530 ofproto->need_revalidate = true;
1532 rule_uninstall(ofproto, rule);
1534 classifier_remove(&ofproto->cls, &rule->cr);
1535 rule_destroy(ofproto, rule);
1538 /* Returns true if the actions changed, false otherwise. */
1540 rule_make_actions(struct ofproto *p, struct rule *rule,
1541 const struct ofpbuf *packet)
1543 const struct rule *super;
1544 struct odp_actions a;
1547 assert(!rule->cr.wc.wildcards);
1549 super = rule->super ? rule->super : rule;
1551 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1552 packet, &a, &rule->tags, &rule->may_install);
1554 actions_len = a.n_actions * sizeof *a.actions;
1555 if (rule->n_odp_actions != a.n_actions
1556 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1557 COVERAGE_INC(ofproto_odp_unchanged);
1558 free(rule->odp_actions);
1559 rule->n_odp_actions = a.n_actions;
1560 rule->odp_actions = xmemdup(a.actions, actions_len);
1568 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1569 struct odp_flow_put *put)
1571 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1572 put->flow.key = rule->cr.flow;
1573 put->flow.actions = rule->odp_actions;
1574 put->flow.n_actions = rule->n_odp_actions;
1576 return dpif_flow_put(&ofproto->dpif, put);
1580 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1582 assert(!rule->cr.wc.wildcards);
1584 if (rule->may_install) {
1585 struct odp_flow_put put;
1586 if (!do_put_flow(p, rule,
1587 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1589 rule->installed = true;
1590 if (displaced_rule) {
1591 update_stats(rule, &put.flow.stats);
1592 rule_post_uninstall(p, displaced_rule);
1595 } else if (displaced_rule) {
1596 rule_uninstall(p, displaced_rule);
1601 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1603 if (rule->installed) {
1604 struct odp_flow_put put;
1605 COVERAGE_INC(ofproto_dp_missed);
1606 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1608 rule_install(ofproto, rule, NULL);
1613 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1615 bool actions_changed = rule_make_actions(ofproto, rule, NULL);
1616 if (rule->may_install) {
1617 if (rule->installed) {
1618 if (actions_changed) {
1619 /* XXX should really do rule_post_uninstall() for the *old* set
1620 * of actions, and distinguish the old stats from the new. */
1621 struct odp_flow_put put;
1622 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1625 rule_install(ofproto, rule, NULL);
1628 rule_uninstall(ofproto, rule);
1633 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1635 uint64_t total_bytes = rule->byte_count + extra_bytes;
1637 if (ofproto->ofhooks->account_flow_cb
1638 && total_bytes > rule->accounted_bytes)
1640 ofproto->ofhooks->account_flow_cb(
1641 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1642 total_bytes - rule->accounted_bytes, ofproto->aux);
1643 rule->accounted_bytes = total_bytes;
1648 rule_uninstall(struct ofproto *p, struct rule *rule)
1650 assert(!rule->cr.wc.wildcards);
1651 if (rule->installed) {
1652 struct odp_flow odp_flow;
1654 odp_flow.key = rule->cr.flow;
1655 odp_flow.actions = NULL;
1656 odp_flow.n_actions = 0;
1657 if (!dpif_flow_del(&p->dpif, &odp_flow)) {
1658 update_stats(rule, &odp_flow.stats);
1660 rule->installed = false;
1662 rule_post_uninstall(p, rule);
1667 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1669 struct rule *super = rule->super;
1671 rule_account(ofproto, rule, 0);
1672 if (ofproto->netflow) {
1673 struct ofexpired expired;
1674 expired.flow = rule->cr.flow;
1675 expired.packet_count = rule->packet_count;
1676 expired.byte_count = rule->byte_count;
1677 expired.used = rule->used;
1678 expired.created = rule->created;
1679 expired.tcp_flags = rule->tcp_flags;
1680 expired.ip_tos = rule->ip_tos;
1681 netflow_expire(ofproto->netflow, &expired);
1684 super->packet_count += rule->packet_count;
1685 super->byte_count += rule->byte_count;
1686 super->tcp_flags |= rule->tcp_flags;
1687 if (rule->packet_count) {
1688 super->ip_tos = rule->ip_tos;
1692 /* Reset counters to prevent double counting if the rule ever gets
1694 rule->packet_count = 0;
1695 rule->byte_count = 0;
1696 rule->accounted_bytes = 0;
1697 rule->tcp_flags = 0;
1702 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1703 struct rconn_packet_counter *counter)
1705 update_openflow_length(msg);
1706 if (rconn_send(ofconn->rconn, msg, counter)) {
1712 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1713 int error, const void *data, size_t len)
1716 struct ofp_error_msg *oem;
1718 if (!(error >> 16)) {
1719 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1724 COVERAGE_INC(ofproto_error);
1725 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1726 oh ? oh->xid : 0, &buf);
1727 oem->type = htons((unsigned int) error >> 16);
1728 oem->code = htons(error & 0xffff);
1729 memcpy(oem->data, data, len);
1730 queue_tx(buf, ofconn, ofconn->reply_counter);
1734 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1737 size_t oh_length = ntohs(oh->length);
1738 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1742 hton_ofp_phy_port(struct ofp_phy_port *opp)
1744 opp->port_no = htons(opp->port_no);
1745 opp->config = htonl(opp->config);
1746 opp->state = htonl(opp->state);
1747 opp->curr = htonl(opp->curr);
1748 opp->advertised = htonl(opp->advertised);
1749 opp->supported = htonl(opp->supported);
1750 opp->peer = htonl(opp->peer);
1754 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
1756 struct ofp_header *rq = oh;
1757 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
1762 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
1763 struct ofp_header *oh)
1765 struct ofp_switch_features *osf;
1767 unsigned int port_no;
1768 struct ofport *port;
1770 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1771 osf->datapath_id = htonll(p->datapath_id);
1772 osf->n_buffers = htonl(pktbuf_capacity());
1774 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1775 OFPC_PORT_STATS | OFPC_MULTI_PHY_TX);
1776 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1777 (1u << OFPAT_SET_VLAN_VID) |
1778 (1u << OFPAT_SET_VLAN_PCP) |
1779 (1u << OFPAT_STRIP_VLAN) |
1780 (1u << OFPAT_SET_DL_SRC) |
1781 (1u << OFPAT_SET_DL_DST) |
1782 (1u << OFPAT_SET_NW_SRC) |
1783 (1u << OFPAT_SET_NW_DST) |
1784 (1u << OFPAT_SET_TP_SRC) |
1785 (1u << OFPAT_SET_TP_DST));
1787 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1788 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1791 queue_tx(buf, ofconn, ofconn->reply_counter);
1796 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1797 struct ofp_header *oh)
1800 struct ofp_switch_config *osc;
1804 /* Figure out flags. */
1805 dpif_get_drop_frags(&p->dpif, &drop_frags);
1806 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1807 if (ofconn->send_flow_exp) {
1808 flags |= OFPC_SEND_FLOW_EXP;
1812 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1813 osc->flags = htons(flags);
1814 osc->miss_send_len = htons(ofconn->miss_send_len);
1815 queue_tx(buf, ofconn, ofconn->reply_counter);
1821 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1822 struct ofp_switch_config *osc)
1827 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1831 flags = ntohs(osc->flags);
1833 ofconn->send_flow_exp = (flags & OFPC_SEND_FLOW_EXP) != 0;
1835 if (ofconn == p->controller) {
1836 switch (flags & OFPC_FRAG_MASK) {
1837 case OFPC_FRAG_NORMAL:
1838 dpif_set_drop_frags(&p->dpif, false);
1840 case OFPC_FRAG_DROP:
1841 dpif_set_drop_frags(&p->dpif, true);
1844 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1850 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1851 if (ntohs(osc->miss_send_len) != 0) {
1852 ofconn->pktbuf = pktbuf_create();
1854 pktbuf_destroy(ofconn->pktbuf);
1858 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1864 add_output_group_action(struct odp_actions *actions, uint16_t group)
1866 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1870 add_controller_action(struct odp_actions *actions,
1871 const struct ofp_action_output *oao)
1873 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
1874 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
1877 struct action_xlate_ctx {
1879 const flow_t *flow; /* Flow to which these actions correspond. */
1880 int recurse; /* Recursion level, via xlate_table_action. */
1881 struct ofproto *ofproto;
1882 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
1883 * null pointer if we are revalidating
1884 * without a packet to refer to. */
1887 struct odp_actions *out; /* Datapath actions. */
1888 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
1889 bool may_setup_flow; /* True ordinarily; false if the actions must
1890 * be reassessed for every packet. */
1893 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
1894 struct action_xlate_ctx *ctx);
1897 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
1899 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
1900 if (!ofport || !(ofport->opp.config & OFPPC_NO_FWD)) {
1901 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
1905 static struct rule *
1906 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
1909 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
1911 /* The rule we found might not be valid, since we could be in need of
1912 * revalidation. If it is not valid, don't return it. */
1915 && ofproto->need_revalidate
1916 && !revalidate_rule(ofproto, rule)) {
1917 COVERAGE_INC(ofproto_invalidated);
1925 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
1927 if (!ctx->recurse) {
1932 flow.in_port = in_port;
1934 rule = lookup_valid_rule(ctx->ofproto, &flow);
1941 do_xlate_actions(rule->actions, rule->n_actions, ctx);
1948 xlate_output_action(struct action_xlate_ctx *ctx,
1949 const struct ofp_action_output *oao)
1953 switch (ntohs(oao->port)) {
1955 add_output_action(ctx, ctx->flow->in_port);
1958 xlate_table_action(ctx, ctx->flow->in_port);
1961 if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet,
1962 ctx->out, ctx->tags,
1963 ctx->ofproto->aux)) {
1964 COVERAGE_INC(ofproto_uninstallable);
1965 ctx->may_setup_flow = false;
1969 add_output_group_action(ctx->out, DP_GROUP_FLOOD);
1972 add_output_group_action(ctx->out, DP_GROUP_ALL);
1974 case OFPP_CONTROLLER:
1975 add_controller_action(ctx->out, oao);
1978 add_output_action(ctx, ODPP_LOCAL);
1981 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
1982 if (odp_port != ctx->flow->in_port) {
1983 add_output_action(ctx, odp_port);
1990 xlate_nicira_action(struct action_xlate_ctx *ctx,
1991 const struct nx_action_header *nah)
1993 const struct nx_action_resubmit *nar;
1994 int subtype = ntohs(nah->subtype);
1996 assert(nah->vendor == htonl(NX_VENDOR_ID));
1998 case NXAST_RESUBMIT:
1999 nar = (const struct nx_action_resubmit *) nah;
2000 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2004 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2010 do_xlate_actions(const union ofp_action *in, size_t n_in,
2011 struct action_xlate_ctx *ctx)
2013 struct actions_iterator iter;
2014 const union ofp_action *ia;
2015 const struct ofport *port;
2017 port = port_array_get(&ctx->ofproto->ports, ctx->flow->in_port);
2018 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2019 port->opp.config & (eth_addr_equals(ctx->flow->dl_dst, stp_eth_addr)
2020 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2021 /* Drop this flow. */
2025 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2026 uint16_t type = ntohs(ia->type);
2027 union odp_action *oa;
2031 xlate_output_action(ctx, &ia->output);
2034 case OFPAT_SET_VLAN_VID:
2035 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2036 oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2039 case OFPAT_SET_VLAN_PCP:
2040 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2041 oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2044 case OFPAT_STRIP_VLAN:
2045 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2048 case OFPAT_SET_DL_SRC:
2049 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2050 memcpy(oa->dl_addr.dl_addr,
2051 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2054 case OFPAT_SET_DL_DST:
2055 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2056 memcpy(oa->dl_addr.dl_addr,
2057 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2060 case OFPAT_SET_NW_SRC:
2061 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2062 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2065 case OFPAT_SET_TP_SRC:
2066 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2067 oa->tp_port.tp_port = ia->tp_port.tp_port;
2071 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2075 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2082 xlate_actions(const union ofp_action *in, size_t n_in,
2083 const flow_t *flow, struct ofproto *ofproto,
2084 const struct ofpbuf *packet,
2085 struct odp_actions *out, tag_type *tags, bool *may_setup_flow)
2087 tag_type no_tags = 0;
2088 struct action_xlate_ctx ctx;
2089 COVERAGE_INC(ofproto_ofp2odp);
2090 odp_actions_init(out);
2093 ctx.ofproto = ofproto;
2094 ctx.packet = packet;
2096 ctx.tags = tags ? tags : &no_tags;
2097 ctx.may_setup_flow = true;
2098 do_xlate_actions(in, n_in, &ctx);
2099 if (may_setup_flow) {
2100 *may_setup_flow = ctx.may_setup_flow;
2102 if (odp_actions_overflow(out)) {
2103 odp_actions_init(out);
2104 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2110 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2111 struct ofp_header *oh)
2113 struct ofp_packet_out *opo;
2114 struct ofpbuf payload, *buffer;
2115 struct odp_actions actions;
2121 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2125 opo = (struct ofp_packet_out *) oh;
2127 COVERAGE_INC(ofproto_packet_out);
2128 if (opo->buffer_id != htonl(UINT32_MAX)) {
2129 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2139 flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2140 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2141 &flow, p, &payload, &actions, NULL, NULL);
2146 dpif_execute(&p->dpif, flow.in_port, actions.actions, actions.n_actions,
2148 ofpbuf_delete(buffer);
2154 update_port_config(struct ofproto *p, struct ofport *port,
2155 uint32_t config, uint32_t mask)
2157 mask &= config ^ port->opp.config;
2158 if (mask & OFPPC_PORT_DOWN) {
2159 if (config & OFPPC_PORT_DOWN) {
2160 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2162 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2165 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2166 if (mask & REVALIDATE_BITS) {
2167 COVERAGE_INC(ofproto_costly_flags);
2168 port->opp.config ^= mask & REVALIDATE_BITS;
2169 p->need_revalidate = true;
2171 #undef REVALIDATE_BITS
2172 if (mask & OFPPC_NO_FLOOD) {
2173 port->opp.config ^= OFPPC_NO_FLOOD;
2174 refresh_port_group(p, DP_GROUP_FLOOD);
2176 if (mask & OFPPC_NO_PACKET_IN) {
2177 port->opp.config ^= OFPPC_NO_PACKET_IN;
2182 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2184 const struct ofp_port_mod *opm;
2185 struct ofport *port;
2188 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2192 opm = (struct ofp_port_mod *) oh;
2194 port = port_array_get(&p->ports,
2195 ofp_port_to_odp_port(ntohs(opm->port_no)));
2197 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2198 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2199 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2201 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2202 if (opm->advertise) {
2203 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2209 static struct ofpbuf *
2210 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2212 struct ofp_stats_reply *osr;
2215 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2216 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2218 osr->flags = htons(0);
2222 static struct ofpbuf *
2223 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2225 return make_stats_reply(request->header.xid, request->type, body_len);
2229 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2231 struct ofpbuf *msg = *msgp;
2232 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2233 if (nbytes + msg->size > UINT16_MAX) {
2234 struct ofp_stats_reply *reply = msg->data;
2235 reply->flags = htons(OFPSF_REPLY_MORE);
2236 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2237 queue_tx(msg, ofconn, ofconn->reply_counter);
2239 return ofpbuf_put_uninit(*msgp, nbytes);
2243 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2244 struct ofp_stats_request *request)
2246 struct ofp_desc_stats *ods;
2249 msg = start_stats_reply(request, sizeof *ods);
2250 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2251 strncpy(ods->mfr_desc, p->manufacturer, sizeof ods->mfr_desc);
2252 strncpy(ods->hw_desc, p->hardware, sizeof ods->hw_desc);
2253 strncpy(ods->sw_desc, p->software, sizeof ods->sw_desc);
2254 strncpy(ods->serial_num, p->serial, sizeof ods->serial_num);
2255 queue_tx(msg, ofconn, ofconn->reply_counter);
2261 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2263 struct rule *rule = rule_from_cls_rule(cls_rule);
2264 int *n_subrules = n_subrules_;
2272 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2273 struct ofp_stats_request *request)
2275 struct ofp_table_stats *ots;
2277 struct odp_stats dpstats;
2278 int n_exact, n_subrules, n_wild;
2280 msg = start_stats_reply(request, sizeof *ots * 2);
2282 /* Count rules of various kinds. */
2284 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2285 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2286 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2289 dpif_get_dp_stats(&p->dpif, &dpstats);
2290 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2291 memset(ots, 0, sizeof *ots);
2292 ots->table_id = TABLEID_HASH;
2293 strcpy(ots->name, "hash");
2294 ots->wildcards = htonl(0);
2295 ots->max_entries = htonl(dpstats.max_capacity);
2296 ots->active_count = htonl(n_exact);
2297 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2299 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2301 /* Classifier table. */
2302 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2303 memset(ots, 0, sizeof *ots);
2304 ots->table_id = TABLEID_CLASSIFIER;
2305 strcpy(ots->name, "classifier");
2306 ots->wildcards = htonl(OFPFW_ALL);
2307 ots->max_entries = htonl(65536);
2308 ots->active_count = htonl(n_wild);
2309 ots->lookup_count = htonll(0); /* XXX */
2310 ots->matched_count = htonll(0); /* XXX */
2312 queue_tx(msg, ofconn, ofconn->reply_counter);
2317 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2318 struct ofp_stats_request *request)
2320 struct ofp_port_stats *ops;
2322 struct ofport *port;
2323 unsigned int port_no;
2325 msg = start_stats_reply(request, sizeof *ops * 16);
2326 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2327 struct netdev_stats stats;
2329 /* Intentionally ignore return value, since errors will set 'stats' to
2330 * all-1s, which is correct for OpenFlow, and netdev_get_stats() will
2332 netdev_get_stats(port->netdev, &stats);
2334 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2335 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2336 memset(ops->pad, 0, sizeof ops->pad);
2337 ops->rx_packets = htonll(stats.rx_packets);
2338 ops->tx_packets = htonll(stats.tx_packets);
2339 ops->rx_bytes = htonll(stats.rx_bytes);
2340 ops->tx_bytes = htonll(stats.tx_bytes);
2341 ops->rx_dropped = htonll(stats.rx_dropped);
2342 ops->tx_dropped = htonll(stats.tx_dropped);
2343 ops->rx_errors = htonll(stats.rx_errors);
2344 ops->tx_errors = htonll(stats.tx_errors);
2345 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2346 ops->rx_over_err = htonll(stats.rx_over_errors);
2347 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2348 ops->collisions = htonll(stats.collisions);
2351 queue_tx(msg, ofconn, ofconn->reply_counter);
2355 struct flow_stats_cbdata {
2356 struct ofproto *ofproto;
2357 struct ofconn *ofconn;
2363 query_stats(struct ofproto *p, struct rule *rule,
2364 uint64_t *packet_countp, uint64_t *byte_countp)
2366 uint64_t packet_count, byte_count;
2367 struct rule *subrule;
2368 struct odp_flow *odp_flows;
2371 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2372 odp_flows = xcalloc(1, n_odp_flows * sizeof *odp_flows);
2373 if (rule->cr.wc.wildcards) {
2375 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2376 odp_flows[i++].key = subrule->cr.flow;
2379 odp_flows[0].key = rule->cr.flow;
2382 packet_count = rule->packet_count;
2383 byte_count = rule->byte_count;
2384 if (!dpif_flow_get_multiple(&p->dpif, odp_flows, n_odp_flows)) {
2386 for (i = 0; i < n_odp_flows; i++) {
2387 struct odp_flow *odp_flow = &odp_flows[i];
2388 packet_count += odp_flow->stats.n_packets;
2389 byte_count += odp_flow->stats.n_bytes;
2394 *packet_countp = packet_count;
2395 *byte_countp = byte_count;
2399 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2401 struct rule *rule = rule_from_cls_rule(rule_);
2402 struct flow_stats_cbdata *cbdata = cbdata_;
2403 struct ofp_flow_stats *ofs;
2404 uint64_t packet_count, byte_count;
2405 size_t act_len, len;
2407 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2411 act_len = sizeof *rule->actions * rule->n_actions;
2412 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2414 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2416 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2417 ofs->length = htons(len);
2418 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2420 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match);
2421 ofs->duration = htonl((time_msec() - rule->created) / 1000);
2422 ofs->priority = htons(rule->cr.priority);
2423 ofs->idle_timeout = htons(rule->idle_timeout);
2424 ofs->hard_timeout = htons(rule->hard_timeout);
2425 memset(ofs->pad2, 0, sizeof ofs->pad2);
2426 ofs->packet_count = htonll(packet_count);
2427 ofs->byte_count = htonll(byte_count);
2428 memcpy(ofs->actions, rule->actions, act_len);
2432 table_id_to_include(uint8_t table_id)
2434 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2435 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2436 : table_id == 0xff ? CLS_INC_ALL
2441 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2442 const struct ofp_stats_request *osr,
2445 struct ofp_flow_stats_request *fsr;
2446 struct flow_stats_cbdata cbdata;
2447 struct cls_rule target;
2449 if (arg_size != sizeof *fsr) {
2450 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2452 fsr = (struct ofp_flow_stats_request *) osr->body;
2454 COVERAGE_INC(ofproto_flows_req);
2456 cbdata.ofconn = ofconn;
2457 cbdata.out_port = fsr->out_port;
2458 cbdata.msg = start_stats_reply(osr, 1024);
2459 cls_rule_from_match(&target, &fsr->match, 0);
2460 classifier_for_each_match(&p->cls, &target,
2461 table_id_to_include(fsr->table_id),
2462 flow_stats_cb, &cbdata);
2463 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2467 struct aggregate_stats_cbdata {
2468 struct ofproto *ofproto;
2470 uint64_t packet_count;
2471 uint64_t byte_count;
2476 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2478 struct rule *rule = rule_from_cls_rule(rule_);
2479 struct aggregate_stats_cbdata *cbdata = cbdata_;
2480 uint64_t packet_count, byte_count;
2482 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2486 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2488 cbdata->packet_count += packet_count;
2489 cbdata->byte_count += byte_count;
2494 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2495 const struct ofp_stats_request *osr,
2498 struct ofp_aggregate_stats_request *asr;
2499 struct ofp_aggregate_stats_reply *reply;
2500 struct aggregate_stats_cbdata cbdata;
2501 struct cls_rule target;
2504 if (arg_size != sizeof *asr) {
2505 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2507 asr = (struct ofp_aggregate_stats_request *) osr->body;
2509 COVERAGE_INC(ofproto_agg_request);
2511 cbdata.out_port = asr->out_port;
2512 cbdata.packet_count = 0;
2513 cbdata.byte_count = 0;
2515 cls_rule_from_match(&target, &asr->match, 0);
2516 classifier_for_each_match(&p->cls, &target,
2517 table_id_to_include(asr->table_id),
2518 aggregate_stats_cb, &cbdata);
2520 msg = start_stats_reply(osr, sizeof *reply);
2521 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2522 reply->flow_count = htonl(cbdata.n_flows);
2523 reply->packet_count = htonll(cbdata.packet_count);
2524 reply->byte_count = htonll(cbdata.byte_count);
2525 queue_tx(msg, ofconn, ofconn->reply_counter);
2530 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2531 struct ofp_header *oh)
2533 struct ofp_stats_request *osr;
2537 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2542 osr = (struct ofp_stats_request *) oh;
2544 switch (ntohs(osr->type)) {
2546 return handle_desc_stats_request(p, ofconn, osr);
2549 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2551 case OFPST_AGGREGATE:
2552 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2555 return handle_table_stats_request(p, ofconn, osr);
2558 return handle_port_stats_request(p, ofconn, osr);
2561 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2564 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2568 static long long int
2569 msec_from_nsec(uint64_t sec, uint32_t nsec)
2571 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2575 update_time(struct rule *rule, const struct odp_flow_stats *stats)
2577 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2578 if (used > rule->used) {
2584 update_stats(struct rule *rule, const struct odp_flow_stats *stats)
2586 update_time(rule, stats);
2587 rule->packet_count += stats->n_packets;
2588 rule->byte_count += stats->n_bytes;
2589 rule->tcp_flags |= stats->tcp_flags;
2590 if (stats->n_packets) {
2591 rule->ip_tos = stats->ip_tos;
2596 add_flow(struct ofproto *p, struct ofconn *ofconn,
2597 struct ofp_flow_mod *ofm, size_t n_actions)
2599 struct ofpbuf *packet;
2604 rule = rule_create(NULL, (const union ofp_action *) ofm->actions,
2605 n_actions, ntohs(ofm->idle_timeout),
2606 ntohs(ofm->hard_timeout));
2607 cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority));
2611 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2612 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2616 rule_insert(p, rule, packet, in_port);
2617 ofpbuf_delete(packet);
2622 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
2623 size_t n_actions, uint16_t command, struct rule *rule)
2625 if (rule_is_hidden(rule)) {
2629 if (command == OFPFC_DELETE) {
2630 rule_remove(p, rule);
2632 size_t actions_len = n_actions * sizeof *rule->actions;
2634 if (n_actions == rule->n_actions
2635 && !memcmp(ofm->actions, rule->actions, actions_len))
2640 free(rule->actions);
2641 rule->actions = xmemdup(ofm->actions, actions_len);
2642 rule->n_actions = n_actions;
2644 if (rule->cr.wc.wildcards) {
2645 COVERAGE_INC(ofproto_mod_wc_flow);
2646 p->need_revalidate = true;
2648 rule_update_actions(p, rule);
2656 modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm,
2657 size_t n_actions, uint16_t command)
2663 flow_from_match(&flow, &wildcards, &ofm->match);
2664 rule = rule_from_cls_rule(classifier_find_rule_exactly(
2665 &p->cls, &flow, wildcards,
2666 ntohs(ofm->priority)));
2669 if (command == OFPFC_DELETE
2670 && ofm->out_port != htons(OFPP_NONE)
2671 && !rule_has_out_port(rule, ofm->out_port)) {
2675 modify_flow(p, ofm, n_actions, command, rule);
2680 struct modify_flows_cbdata {
2681 struct ofproto *ofproto;
2682 const struct ofp_flow_mod *ofm;
2689 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
2691 struct rule *rule = rule_from_cls_rule(rule_);
2692 struct modify_flows_cbdata *cbdata = cbdata_;
2694 if (cbdata->out_port != htons(OFPP_NONE)
2695 && !rule_has_out_port(rule, cbdata->out_port)) {
2699 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions,
2700 cbdata->command, rule);
2704 modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm,
2705 size_t n_actions, uint16_t command)
2707 struct modify_flows_cbdata cbdata;
2708 struct cls_rule target;
2712 cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port
2713 : htons(OFPP_NONE));
2714 cbdata.n_actions = n_actions;
2715 cbdata.command = command;
2717 cls_rule_from_match(&target, &ofm->match, 0);
2719 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2720 modify_flows_cb, &cbdata);
2725 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
2726 struct ofp_flow_mod *ofm)
2731 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
2732 sizeof *ofm->actions, &n_actions);
2737 normalize_match(&ofm->match);
2738 if (!ofm->match.wildcards) {
2739 ofm->priority = htons(UINT16_MAX);
2742 error = validate_actions((const union ofp_action *) ofm->actions,
2743 n_actions, p->max_ports);
2748 switch (ntohs(ofm->command)) {
2750 return add_flow(p, ofconn, ofm, n_actions);
2753 return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY);
2755 case OFPFC_MODIFY_STRICT:
2756 return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY);
2759 return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE);
2761 case OFPFC_DELETE_STRICT:
2762 return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE);
2765 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
2770 send_capability_reply(struct ofproto *p, struct ofconn *ofconn, uint32_t xid)
2772 struct ofmp_capability_reply *ocr;
2774 char capabilities[] = "com.nicira.mgmt.manager=false\n";
2776 ocr = make_openflow_xid(sizeof(*ocr), OFPT_VENDOR, xid, &b);
2777 ocr->header.header.vendor = htonl(NX_VENDOR_ID);
2778 ocr->header.header.subtype = htonl(NXT_MGMT);
2779 ocr->header.type = htons(OFMPT_CAPABILITY_REPLY);
2781 ocr->format = htonl(OFMPCOF_SIMPLE);
2782 ocr->mgmt_id = htonll(p->mgmt_id);
2784 ofpbuf_put(b, capabilities, strlen(capabilities));
2786 queue_tx(b, ofconn, ofconn->reply_counter);
2790 handle_ofmp(struct ofproto *p, struct ofconn *ofconn,
2791 struct ofmp_header *ofmph)
2793 size_t msg_len = ntohs(ofmph->header.header.length);
2794 if (msg_len < sizeof(*ofmph)) {
2795 VLOG_WARN_RL(&rl, "dropping short managment message: %d\n", msg_len);
2796 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2799 if (ofmph->type == htons(OFMPT_CAPABILITY_REQUEST)) {
2800 struct ofmp_capability_request *ofmpcr;
2802 if (msg_len < sizeof(struct ofmp_capability_request)) {
2803 VLOG_WARN_RL(&rl, "dropping short capability request: %d\n",
2805 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2808 ofmpcr = (struct ofmp_capability_request *)ofmph;
2809 if (ofmpcr->format != htonl(OFMPCAF_SIMPLE)) {
2810 /* xxx Find a better type than bad subtype */
2811 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2814 send_capability_reply(p, ofconn, ofmph->header.header.xid);
2817 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2822 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
2824 struct ofp_vendor_header *ovh = msg;
2825 struct nicira_header *nh;
2827 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
2828 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2830 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
2831 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2833 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
2834 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2838 switch (ntohl(nh->subtype)) {
2839 case NXT_STATUS_REQUEST:
2840 return switch_status_handle_request(p->switch_status, ofconn->rconn,
2843 case NXT_ACT_SET_CONFIG:
2844 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
2846 case NXT_ACT_GET_CONFIG:
2847 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
2849 case NXT_COMMAND_REQUEST:
2851 return executer_handle_request(p->executer, ofconn->rconn, msg);
2856 return handle_ofmp(p, ofconn, msg);
2859 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2863 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
2864 struct ofpbuf *ofp_msg)
2866 struct ofp_header *oh = ofp_msg->data;
2869 COVERAGE_INC(ofproto_recv_openflow);
2871 case OFPT_ECHO_REQUEST:
2872 error = handle_echo_request(ofconn, oh);
2875 case OFPT_ECHO_REPLY:
2879 case OFPT_FEATURES_REQUEST:
2880 error = handle_features_request(p, ofconn, oh);
2883 case OFPT_GET_CONFIG_REQUEST:
2884 error = handle_get_config_request(p, ofconn, oh);
2887 case OFPT_SET_CONFIG:
2888 error = handle_set_config(p, ofconn, ofp_msg->data);
2891 case OFPT_PACKET_OUT:
2892 error = handle_packet_out(p, ofconn, ofp_msg->data);
2896 error = handle_port_mod(p, oh);
2900 error = handle_flow_mod(p, ofconn, ofp_msg->data);
2903 case OFPT_STATS_REQUEST:
2904 error = handle_stats_request(p, ofconn, oh);
2908 error = handle_vendor(p, ofconn, ofp_msg->data);
2912 if (VLOG_IS_WARN_ENABLED()) {
2913 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
2914 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
2917 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
2922 send_error_oh(ofconn, ofp_msg->data, error);
2927 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
2929 struct odp_msg *msg = packet->data;
2930 uint16_t in_port = odp_port_to_ofp_port(msg->port);
2932 struct ofpbuf payload;
2935 /* Handle controller actions. */
2936 if (msg->type == _ODPL_ACTION_NR) {
2937 COVERAGE_INC(ofproto_ctlr_action);
2938 pinsched_send(p->action_sched, in_port, packet,
2939 send_packet_in_action, p);
2943 payload.data = msg + 1;
2944 payload.size = msg->length - sizeof *msg;
2945 flow_extract(&payload, msg->port, &flow);
2947 rule = lookup_valid_rule(p, &flow);
2949 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2950 struct ofport *port = port_array_get(&p->ports, msg->port);
2952 if (port->opp.config & OFPPC_NO_PACKET_IN) {
2953 COVERAGE_INC(ofproto_no_packet_in);
2954 /* XXX install 'drop' flow entry */
2955 ofpbuf_delete(packet);
2959 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
2962 COVERAGE_INC(ofproto_packet_in);
2963 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
2967 if (rule->cr.wc.wildcards) {
2968 rule = rule_create_subrule(p, rule, &flow);
2969 rule_make_actions(p, rule, packet);
2971 if (!rule->may_install) {
2972 /* The rule is not installable, that is, we need to process every
2973 * packet, so process the current packet and set its actions into
2975 rule_make_actions(p, rule, packet);
2977 /* XXX revalidate rule if it needs it */
2981 rule_execute(p, rule, &payload, &flow);
2982 rule_reinstall(p, rule);
2983 ofpbuf_delete(packet);
2987 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
2989 struct rule *sub = rule_from_cls_rule(sub_);
2990 struct revalidate_cbdata *cbdata = cbdata_;
2992 if (cbdata->revalidate_all
2993 || (cbdata->revalidate_subrules && sub->super)
2994 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
2995 revalidate_rule(cbdata->ofproto, sub);
3000 revalidate_rule(struct ofproto *p, struct rule *rule)
3002 const flow_t *flow = &rule->cr.flow;
3004 COVERAGE_INC(ofproto_revalidate_rule);
3007 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3009 rule_remove(p, rule);
3011 } else if (super != rule->super) {
3012 COVERAGE_INC(ofproto_revalidate_moved);
3013 list_remove(&rule->list);
3014 list_push_back(&super->list, &rule->list);
3015 rule->super = super;
3016 rule->hard_timeout = super->hard_timeout;
3017 rule->idle_timeout = super->idle_timeout;
3018 rule->created = super->created;
3023 rule_update_actions(p, rule);
3027 static struct ofpbuf *
3028 compose_flow_exp(const struct rule *rule, long long int now, uint8_t reason)
3030 struct ofp_flow_expired *ofe;
3033 ofe = make_openflow(sizeof *ofe, OFPT_FLOW_EXPIRED, &buf);
3034 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofe->match);
3035 ofe->priority = htons(rule->cr.priority);
3036 ofe->reason = reason;
3037 ofe->duration = (now - rule->created) / 1000;
3038 ofe->packet_count = rule->packet_count;
3039 ofe->byte_count = rule->byte_count;
3045 send_flow_exp(struct ofproto *p, struct rule *rule,
3046 long long int now, uint8_t reason)
3048 struct ofconn *ofconn;
3049 struct ofconn *prev;
3052 /* We limit the maximum number of queued flow expirations it by accounting
3053 * them under the counter for replies. That works because preventing
3054 * OpenFlow requests from being processed also prevents new flows from
3055 * being added (and expiring). (It also prevents processing OpenFlow
3056 * requests that would not add new flows, so it is imperfect.) */
3059 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3060 if (ofconn->send_flow_exp && rconn_is_connected(ofconn->rconn)) {
3062 queue_tx(ofpbuf_clone(buf), prev, ofconn->reply_counter);
3064 buf = compose_flow_exp(rule, now, reason);
3070 queue_tx(buf, prev, ofconn->reply_counter);
3075 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3077 assert(rule->installed);
3078 assert(!rule->cr.wc.wildcards);
3081 rule_remove(ofproto, rule);
3083 rule_uninstall(ofproto, rule);
3088 expire_rule(struct cls_rule *cls_rule, void *p_)
3090 struct ofproto *p = p_;
3091 struct rule *rule = rule_from_cls_rule(cls_rule);
3092 long long int hard_expire, idle_expire, expire, now;
3094 hard_expire = (rule->hard_timeout
3095 ? rule->created + rule->hard_timeout * 1000
3097 idle_expire = (rule->idle_timeout
3098 && (rule->super || list_is_empty(&rule->list))
3099 ? rule->used + rule->idle_timeout * 1000
3101 expire = MIN(hard_expire, idle_expire);
3102 if (expire == LLONG_MAX) {
3103 if (rule->installed && time_msec() >= rule->used + 5000) {
3104 uninstall_idle_flow(p, rule);
3111 if (rule->installed && now >= rule->used + 5000) {
3112 uninstall_idle_flow(p, rule);
3117 COVERAGE_INC(ofproto_expired);
3118 if (rule->cr.wc.wildcards) {
3119 /* Update stats. (This code will be a no-op if the rule expired
3120 * due to an idle timeout, because in that case the rule has no
3121 * subrules left.) */
3122 struct rule *subrule, *next;
3123 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3124 rule_remove(p, subrule);
3128 send_flow_exp(p, rule, now,
3130 ? OFPER_HARD_TIMEOUT : OFPER_IDLE_TIMEOUT));
3131 rule_remove(p, rule);
3135 update_used(struct ofproto *p)
3137 struct odp_flow *flows;
3142 error = dpif_flow_list_all(&p->dpif, &flows, &n_flows);
3147 for (i = 0; i < n_flows; i++) {
3148 struct odp_flow *f = &flows[i];
3151 rule = rule_from_cls_rule(
3152 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3153 if (!rule || !rule->installed) {
3154 COVERAGE_INC(ofproto_unexpected_rule);
3155 dpif_flow_del(&p->dpif, f);
3159 update_time(rule, &f->stats);
3160 rule_account(p, rule, f->stats.n_bytes);
3166 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3167 const struct ofpbuf *packet, int send_len)
3169 struct ofp_packet_in *opi;
3170 struct ofpbuf payload, *buf;
3171 struct odp_msg *msg;
3174 payload.data = msg + 1;
3175 payload.size = msg->length - sizeof *msg;
3177 send_len = MIN(send_len, payload.size);
3178 buf = ofpbuf_new(sizeof *opi + send_len);
3179 opi = put_openflow_xid(offsetof(struct ofp_packet_in, data),
3180 OFPT_PACKET_IN, 0, buf);
3181 opi->buffer_id = htonl(buffer_id);
3182 opi->total_len = htons(payload.size);
3183 opi->in_port = htons(odp_port_to_ofp_port(msg->port));
3184 opi->reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3185 ofpbuf_put(buf, payload.data, MIN(send_len, payload.size));
3186 update_openflow_length(buf);
3187 rconn_send_with_limit(ofconn->rconn, buf, ofconn->packet_in_counter, 100);
3191 send_packet_in_action(struct ofpbuf *packet, void *p_)
3193 struct ofproto *p = p_;
3194 struct ofconn *ofconn;
3195 struct odp_msg *msg;
3198 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3199 if (ofconn == p->controller || ofconn->miss_send_len) {
3200 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3203 ofpbuf_delete(packet);
3207 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3209 struct ofproto *p = p_;
3210 struct ofconn *ofconn;
3211 struct ofpbuf payload;
3212 struct odp_msg *msg;
3215 payload.data = msg + 1;
3216 payload.size = msg->length - sizeof *msg;
3217 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3218 if (ofconn->miss_send_len) {
3219 uint32_t buffer_id = pktbuf_save(ofconn->pktbuf, &payload,
3221 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3223 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3226 ofpbuf_delete(packet);
3230 pick_datapath_id(struct dpif *dpif, uint64_t fallback_dpid)
3232 char local_name[IF_NAMESIZE];
3233 uint8_t ea[ETH_ADDR_LEN];
3236 error = dpif_port_get_name(dpif, ODPP_LOCAL,
3237 local_name, sizeof local_name);
3239 error = netdev_nodev_get_etheraddr(local_name, ea);
3241 return eth_addr_to_uint64(ea);
3243 VLOG_WARN("could not get MAC address for %s (%s)",
3244 local_name, strerror(error));
3247 return fallback_dpid;
3251 pick_fallback_dpid(void)
3253 uint8_t ea[ETH_ADDR_LEN];
3254 eth_addr_random(ea);
3255 ea[0] = 0x00; /* Set Nicira OUI. */
3258 return eth_addr_to_uint64(ea);
3262 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3263 struct odp_actions *actions, tag_type *tags,
3266 struct ofproto *ofproto = ofproto_;
3269 /* Drop frames for reserved multicast addresses. */
3270 if (eth_addr_is_reserved(flow->dl_dst)) {
3274 /* Learn source MAC (but don't try to learn from revalidation). */
3275 if (packet != NULL) {
3276 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3279 /* The log messages here could actually be useful in debugging,
3280 * so keep the rate limit relatively high. */
3281 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3282 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3283 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3284 ofproto_revalidate(ofproto, rev_tag);
3288 /* Determine output port. */
3289 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3291 add_output_group_action(actions, DP_GROUP_FLOOD);
3292 } else if (out_port != flow->in_port) {
3293 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3301 static const struct ofhooks default_ofhooks = {
3303 default_normal_ofhook_cb,