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_mgmt_id(const struct ofproto *ofproto)
641 return ofproto->mgmt_id;
645 ofproto_get_probe_interval(const struct ofproto *ofproto)
647 return rconn_get_probe_interval(ofproto->controller->rconn);
651 ofproto_get_max_backoff(const struct ofproto *ofproto)
653 return rconn_get_max_backoff(ofproto->controller->rconn);
657 ofproto_get_in_band(const struct ofproto *ofproto)
659 return ofproto->in_band != NULL;
663 ofproto_get_discovery(const struct ofproto *ofproto)
665 return ofproto->discovery != NULL;
669 ofproto_get_controller(const struct ofproto *ofproto)
671 return rconn_get_name(ofproto->controller->rconn);
675 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
679 for (i = 0; i < ofproto->n_listeners; i++) {
680 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
685 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
689 for (i = 0; i < ofproto->n_snoops; i++) {
690 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
695 ofproto_destroy(struct ofproto *p)
697 struct ofconn *ofconn, *next_ofconn;
698 struct ofport *ofport;
699 unsigned int port_no;
706 ofproto_flush_flows(p);
707 classifier_destroy(&p->cls);
709 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
711 ofconn_destroy(ofconn, p);
714 dpif_close(&p->dpif);
715 dpifmon_destroy(p->dpifmon);
716 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
719 shash_destroy(&p->port_by_name);
721 switch_status_destroy(p->switch_status);
722 in_band_destroy(p->in_band);
723 discovery_destroy(p->discovery);
724 fail_open_destroy(p->fail_open);
725 pinsched_destroy(p->miss_sched);
726 pinsched_destroy(p->action_sched);
727 executer_destroy(p->executer);
728 netflow_destroy(p->netflow);
730 switch_status_unregister(p->ss_cat);
732 for (i = 0; i < p->n_listeners; i++) {
733 pvconn_close(p->listeners[i]);
737 for (i = 0; i < p->n_snoops; i++) {
738 pvconn_close(p->snoops[i]);
742 mac_learning_destroy(p->ml);
748 ofproto_run(struct ofproto *p)
750 int error = ofproto_run1(p);
752 error = ofproto_run2(p, false);
758 ofproto_run1(struct ofproto *p)
760 struct ofconn *ofconn, *next_ofconn;
765 for (i = 0; i < 50; i++) {
769 error = dpif_recv(&p->dpif, &buf);
771 if (error == ENODEV) {
772 /* Someone destroyed the datapath behind our back. The caller
773 * better destroy us and give up, because we're just going to
774 * spin from here on out. */
775 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
776 VLOG_ERR_RL(&rl, "dp%u: datapath was destroyed externally",
783 handle_odp_msg(p, buf);
786 while ((error = dpifmon_poll(p->dpifmon, &devname)) != EAGAIN) {
787 if (error == ENOBUFS) {
790 update_port(p, devname);
796 in_band_run(p->in_band);
799 char *controller_name;
800 if (rconn_is_connectivity_questionable(p->controller->rconn)) {
801 discovery_question_connectivity(p->discovery);
803 if (discovery_run(p->discovery, &controller_name)) {
804 if (controller_name) {
805 rconn_connect(p->controller->rconn, controller_name);
807 rconn_disconnect(p->controller->rconn);
812 fail_open_run(p->fail_open);
814 pinsched_run(p->miss_sched, send_packet_in_miss, p);
815 pinsched_run(p->action_sched, send_packet_in_action, p);
817 executer_run(p->executer);
820 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
822 ofconn_run(ofconn, p);
825 for (i = 0; i < p->n_listeners; i++) {
829 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
831 ofconn_create(p, rconn_new_from_vconn("passive", vconn));
832 } else if (retval != EAGAIN) {
833 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
837 for (i = 0; i < p->n_snoops; i++) {
841 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
843 rconn_add_monitor(p->controller->rconn, vconn);
844 } else if (retval != EAGAIN) {
845 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
849 if (time_msec() >= p->next_expiration) {
850 COVERAGE_INC(ofproto_expiration);
851 p->next_expiration = time_msec() + 1000;
854 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
856 /* Let the hook know that we're at a stable point: all outstanding data
857 * in existing flows has been accounted to the account_cb. Thus, the
858 * hook can now reasonably do operations that depend on having accurate
859 * flow volume accounting (currently, that's just bond rebalancing). */
860 if (p->ofhooks->account_checkpoint_cb) {
861 p->ofhooks->account_checkpoint_cb(p->aux);
866 netflow_run(p->netflow);
872 struct revalidate_cbdata {
873 struct ofproto *ofproto;
874 bool revalidate_all; /* Revalidate all exact-match rules? */
875 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
876 struct tag_set revalidate_set; /* Set of tags to revalidate. */
880 ofproto_run2(struct ofproto *p, bool revalidate_all)
882 if (p->need_revalidate || revalidate_all
883 || !tag_set_is_empty(&p->revalidate_set)) {
884 struct revalidate_cbdata cbdata;
886 cbdata.revalidate_all = revalidate_all;
887 cbdata.revalidate_subrules = p->need_revalidate;
888 cbdata.revalidate_set = p->revalidate_set;
889 tag_set_init(&p->revalidate_set);
890 COVERAGE_INC(ofproto_revalidate);
891 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
892 p->need_revalidate = false;
899 ofproto_wait(struct ofproto *p)
901 struct ofconn *ofconn;
904 dpif_recv_wait(&p->dpif);
905 dpifmon_wait(p->dpifmon);
906 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
910 in_band_wait(p->in_band);
913 discovery_wait(p->discovery);
916 fail_open_wait(p->fail_open);
918 pinsched_wait(p->miss_sched);
919 pinsched_wait(p->action_sched);
921 executer_wait(p->executer);
923 if (!tag_set_is_empty(&p->revalidate_set)) {
924 poll_immediate_wake();
926 if (p->need_revalidate) {
927 /* Shouldn't happen, but if it does just go around again. */
928 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
929 poll_immediate_wake();
930 } else if (p->next_expiration != LLONG_MAX) {
931 poll_timer_wait(p->next_expiration - time_msec());
933 for (i = 0; i < p->n_listeners; i++) {
934 pvconn_wait(p->listeners[i]);
936 for (i = 0; i < p->n_snoops; i++) {
937 pvconn_wait(p->snoops[i]);
942 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
944 tag_set_add(&ofproto->revalidate_set, tag);
948 ofproto_get_revalidate_set(struct ofproto *ofproto)
950 return &ofproto->revalidate_set;
954 ofproto_is_alive(const struct ofproto *p)
956 return p->discovery || rconn_is_alive(p->controller->rconn);
960 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
961 const union ofp_action *actions, size_t n_actions,
962 const struct ofpbuf *packet)
964 struct odp_actions odp_actions;
967 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
973 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
975 dpif_execute(&p->dpif, flow->in_port, odp_actions.actions,
976 odp_actions.n_actions, packet);
981 ofproto_add_flow(struct ofproto *p,
982 const flow_t *flow, uint32_t wildcards, unsigned int priority,
983 const union ofp_action *actions, size_t n_actions,
987 rule = rule_create(NULL, actions, n_actions,
988 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */, 0);
989 cls_rule_from_flow(&rule->cr, flow, wildcards, priority);
990 rule_insert(p, rule, NULL, 0);
994 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
995 uint32_t wildcards, unsigned int priority)
999 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1003 rule_remove(ofproto, rule);
1008 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1010 struct rule *rule = rule_from_cls_rule(rule_);
1011 struct ofproto *ofproto = ofproto_;
1013 /* Mark the flow as not installed, even though it might really be
1014 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1015 * There is no point in uninstalling it individually since we are about to
1016 * blow away all the flows with dpif_flow_flush(). */
1017 rule->installed = false;
1019 rule_remove(ofproto, rule);
1023 ofproto_flush_flows(struct ofproto *ofproto)
1025 COVERAGE_INC(ofproto_flush);
1026 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1027 dpif_flow_flush(&ofproto->dpif);
1028 if (ofproto->in_band) {
1029 in_band_flushed(ofproto->in_band);
1031 if (ofproto->fail_open) {
1032 fail_open_flushed(ofproto->fail_open);
1037 reinit_ports(struct ofproto *p)
1039 struct svec devnames;
1040 struct ofport *ofport;
1041 unsigned int port_no;
1042 struct odp_port *odp_ports;
1046 svec_init(&devnames);
1047 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1048 svec_add (&devnames, (char *) ofport->opp.name);
1050 dpif_port_list(&p->dpif, &odp_ports, &n_odp_ports);
1051 for (i = 0; i < n_odp_ports; i++) {
1052 svec_add (&devnames, odp_ports[i].devname);
1056 svec_sort_unique(&devnames);
1057 for (i = 0; i < devnames.n; i++) {
1058 update_port(p, devnames.names[i]);
1060 svec_destroy(&devnames);
1064 refresh_port_group(struct ofproto *p, unsigned int group)
1068 struct ofport *port;
1069 unsigned int port_no;
1071 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1073 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1075 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1076 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1077 ports[n_ports++] = port_no;
1080 dpif_port_group_set(&p->dpif, group, ports, n_ports);
1085 refresh_port_groups(struct ofproto *p)
1087 refresh_port_group(p, DP_GROUP_FLOOD);
1088 refresh_port_group(p, DP_GROUP_ALL);
1091 static struct ofport *
1092 make_ofport(const struct odp_port *odp_port)
1094 enum netdev_flags flags;
1095 struct ofport *ofport;
1096 struct netdev *netdev;
1100 error = netdev_open(odp_port->devname, NETDEV_ETH_TYPE_NONE, &netdev);
1102 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1103 "cannot be opened (%s)",
1104 odp_port->devname, odp_port->port,
1105 odp_port->devname, strerror(error));
1109 ofport = xmalloc(sizeof *ofport);
1110 ofport->netdev = netdev;
1111 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1112 memcpy(ofport->opp.hw_addr, netdev_get_etheraddr(netdev), ETH_ALEN);
1113 memcpy(ofport->opp.name, odp_port->devname,
1114 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1115 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1117 netdev_get_flags(netdev, &flags);
1118 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1120 netdev_get_carrier(netdev, &carrier);
1121 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1123 netdev_get_features(netdev,
1124 &ofport->opp.curr, &ofport->opp.advertised,
1125 &ofport->opp.supported, &ofport->opp.peer);
1130 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1132 if (port_array_get(&p->ports, odp_port->port)) {
1133 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1136 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1137 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1146 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1148 const struct ofp_phy_port *a = &a_->opp;
1149 const struct ofp_phy_port *b = &b_->opp;
1151 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1152 return (a->port_no == b->port_no
1153 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1154 && !strcmp((char *) a->name, (char *) b->name)
1155 && a->state == b->state
1156 && a->config == b->config
1157 && a->curr == b->curr
1158 && a->advertised == b->advertised
1159 && a->supported == b->supported
1160 && a->peer == b->peer);
1164 send_port_status(struct ofproto *p, const struct ofport *ofport,
1167 /* XXX Should limit the number of queued port status change messages. */
1168 struct ofconn *ofconn;
1169 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1170 struct ofp_port_status *ops;
1173 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1174 ops->reason = reason;
1175 ops->desc = ofport->opp;
1176 hton_ofp_phy_port(&ops->desc);
1177 queue_tx(b, ofconn, NULL);
1179 if (p->ofhooks->port_changed_cb) {
1180 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1185 ofport_install(struct ofproto *p, struct ofport *ofport)
1187 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no),
1189 shash_add(&p->port_by_name, (char *) ofport->opp.name, ofport);
1193 ofport_remove(struct ofproto *p, struct ofport *ofport)
1195 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no), NULL);
1196 shash_delete(&p->port_by_name,
1197 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1201 ofport_free(struct ofport *ofport)
1204 netdev_close(ofport->netdev);
1210 update_port(struct ofproto *p, const char *devname)
1212 struct odp_port odp_port;
1213 struct ofport *ofport;
1216 COVERAGE_INC(ofproto_update_port);
1217 ofport = shash_find_data(&p->port_by_name, devname);
1218 error = dpif_port_query_by_name(&p->dpif, devname, &odp_port);
1222 if (!ofport_conflicts(p, &odp_port)) {
1223 ofport = make_ofport(&odp_port);
1225 ofport_install(p, ofport);
1226 send_port_status(p, ofport, OFPPR_ADD);
1230 /* Modified port. */
1231 struct ofport *new_ofport = make_ofport(&odp_port);
1236 new_ofport->opp.config &= OFPPC_PORT_DOWN;
1237 new_ofport->opp.config |= ofport->opp.config & ~OFPPC_PORT_DOWN;
1238 if (ofport_equal(ofport, new_ofport)) {
1239 /* False alarm--no change. */
1240 ofport_free(new_ofport);
1242 ofport_remove(p, ofport);
1243 ofport_install(p, new_ofport);
1244 ofport_free(ofport);
1245 send_port_status(p, new_ofport, OFPPR_MODIFY);
1248 } else if (error == ENOENT || error == ENODEV) {
1251 send_port_status(p, ofport, OFPPR_DELETE);
1252 ofport_remove(p, ofport);
1253 ofport_free(ofport);
1256 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1257 "%s", strerror(error));
1260 refresh_port_groups(p);
1264 init_ports(struct ofproto *p)
1266 struct odp_port *ports;
1271 error = dpif_port_list(&p->dpif, &ports, &n_ports);
1276 for (i = 0; i < n_ports; i++) {
1277 const struct odp_port *odp_port = &ports[i];
1278 if (!ofport_conflicts(p, odp_port)) {
1279 struct ofport *ofport = make_ofport(odp_port);
1281 ofport_install(p, ofport);
1286 refresh_port_groups(p);
1290 static struct ofconn *
1291 ofconn_create(struct ofproto *p, struct rconn *rconn)
1293 struct ofconn *ofconn = xmalloc(sizeof *ofconn);
1294 list_push_back(&p->all_conns, &ofconn->node);
1295 ofconn->rconn = rconn;
1296 ofconn->pktbuf = NULL;
1297 ofconn->send_flow_exp = false;
1298 ofconn->miss_send_len = 0;
1299 ofconn->packet_in_counter = rconn_packet_counter_create ();
1300 ofconn->reply_counter = rconn_packet_counter_create ();
1305 ofconn_destroy(struct ofconn *ofconn, struct ofproto *p)
1308 executer_rconn_closing(p->executer, ofconn->rconn);
1311 list_remove(&ofconn->node);
1312 rconn_destroy(ofconn->rconn);
1313 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1314 rconn_packet_counter_destroy(ofconn->reply_counter);
1315 pktbuf_destroy(ofconn->pktbuf);
1320 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1324 rconn_run(ofconn->rconn);
1326 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1327 /* Limit the number of iterations to prevent other tasks from
1329 for (iteration = 0; iteration < 50; iteration++) {
1330 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1334 handle_openflow(ofconn, p, of_msg);
1335 ofpbuf_delete(of_msg);
1339 if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
1340 ofconn_destroy(ofconn, p);
1345 ofconn_wait(struct ofconn *ofconn)
1347 rconn_run_wait(ofconn->rconn);
1348 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1349 rconn_recv_wait(ofconn->rconn);
1351 COVERAGE_INC(ofproto_ofconn_stuck);
1355 /* Caller is responsible for initializing the 'cr' member of the returned
1357 static struct rule *
1358 rule_create(struct rule *super,
1359 const union ofp_action *actions, size_t n_actions,
1360 uint16_t idle_timeout, uint16_t hard_timeout)
1362 struct rule *rule = xcalloc(1, sizeof *rule);
1363 rule->idle_timeout = idle_timeout;
1364 rule->hard_timeout = hard_timeout;
1365 rule->used = rule->created = time_msec();
1366 rule->super = super;
1368 list_push_back(&super->list, &rule->list);
1370 list_init(&rule->list);
1372 rule->n_actions = n_actions;
1373 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1377 static struct rule *
1378 rule_from_cls_rule(const struct cls_rule *cls_rule)
1380 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1384 rule_free(struct rule *rule)
1386 free(rule->actions);
1387 free(rule->odp_actions);
1391 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1392 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1393 * through all of its subrules and revalidates them, destroying any that no
1394 * longer has a super-rule (which is probably all of them).
1396 * Before calling this function, the caller must make have removed 'rule' from
1397 * the classifier. If 'rule' is an exact-match rule, the caller is also
1398 * responsible for ensuring that it has been uninstalled from the datapath. */
1400 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1403 struct rule *subrule, *next;
1404 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1405 revalidate_rule(ofproto, subrule);
1408 list_remove(&rule->list);
1414 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1416 const union ofp_action *oa;
1417 struct actions_iterator i;
1419 if (out_port == htons(OFPP_NONE)) {
1422 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1423 oa = actions_next(&i)) {
1424 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1431 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1432 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1434 * The flow that 'packet' actually contains does not need to actually match
1435 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1436 * the packet and byte counters for 'rule' will be credited for the packet sent
1437 * out whether or not the packet actually matches 'rule'.
1439 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1440 * the caller must already have accurately composed ODP actions for it given
1441 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1442 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1443 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1444 * actions and apply them to 'packet'. */
1446 rule_execute(struct ofproto *ofproto, struct rule *rule,
1447 struct ofpbuf *packet, const flow_t *flow)
1449 const union odp_action *actions;
1451 struct odp_actions a;
1453 /* Grab or compose the ODP actions.
1455 * The special case for an exact-match 'rule' where 'flow' is not the
1456 * rule's flow is important to avoid, e.g., sending a packet out its input
1457 * port simply because the ODP actions were composed for the wrong
1459 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1460 struct rule *super = rule->super ? rule->super : rule;
1461 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1462 packet, &a, NULL, 0)) {
1465 actions = a.actions;
1466 n_actions = a.n_actions;
1468 actions = rule->odp_actions;
1469 n_actions = rule->n_odp_actions;
1472 /* Execute the ODP actions. */
1473 if (!dpif_execute(&ofproto->dpif, flow->in_port,
1474 actions, n_actions, packet)) {
1475 struct odp_flow_stats stats;
1476 flow_extract_stats(flow, packet, &stats);
1477 update_stats(rule, &stats);
1478 rule->used = time_msec();
1483 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1486 struct rule *displaced_rule;
1488 /* Insert the rule in the classifier. */
1489 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1490 if (!rule->cr.wc.wildcards) {
1491 rule_make_actions(p, rule, packet);
1494 /* Send the packet and credit it to the rule. */
1497 flow_extract(packet, in_port, &flow);
1498 rule_execute(p, rule, packet, &flow);
1501 /* Install the rule in the datapath only after sending the packet, to
1502 * avoid packet reordering. */
1503 if (rule->cr.wc.wildcards) {
1504 COVERAGE_INC(ofproto_add_wc_flow);
1505 p->need_revalidate = true;
1507 rule_install(p, rule, displaced_rule);
1510 /* Free the rule that was displaced, if any. */
1511 if (displaced_rule) {
1512 rule_destroy(p, displaced_rule);
1516 static struct rule *
1517 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1520 struct rule *subrule = rule_create(rule, NULL, 0,
1521 rule->idle_timeout, rule->hard_timeout);
1522 COVERAGE_INC(ofproto_subrule_create);
1523 cls_rule_from_flow(&subrule->cr, flow, 0,
1524 (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1525 : rule->cr.priority));
1526 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1532 rule_remove(struct ofproto *ofproto, struct rule *rule)
1534 if (rule->cr.wc.wildcards) {
1535 COVERAGE_INC(ofproto_del_wc_flow);
1536 ofproto->need_revalidate = true;
1538 rule_uninstall(ofproto, rule);
1540 classifier_remove(&ofproto->cls, &rule->cr);
1541 rule_destroy(ofproto, rule);
1544 /* Returns true if the actions changed, false otherwise. */
1546 rule_make_actions(struct ofproto *p, struct rule *rule,
1547 const struct ofpbuf *packet)
1549 const struct rule *super;
1550 struct odp_actions a;
1553 assert(!rule->cr.wc.wildcards);
1555 super = rule->super ? rule->super : rule;
1557 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1558 packet, &a, &rule->tags, &rule->may_install);
1560 actions_len = a.n_actions * sizeof *a.actions;
1561 if (rule->n_odp_actions != a.n_actions
1562 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1563 COVERAGE_INC(ofproto_odp_unchanged);
1564 free(rule->odp_actions);
1565 rule->n_odp_actions = a.n_actions;
1566 rule->odp_actions = xmemdup(a.actions, actions_len);
1574 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1575 struct odp_flow_put *put)
1577 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1578 put->flow.key = rule->cr.flow;
1579 put->flow.actions = rule->odp_actions;
1580 put->flow.n_actions = rule->n_odp_actions;
1582 return dpif_flow_put(&ofproto->dpif, put);
1586 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1588 assert(!rule->cr.wc.wildcards);
1590 if (rule->may_install) {
1591 struct odp_flow_put put;
1592 if (!do_put_flow(p, rule,
1593 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1595 rule->installed = true;
1596 if (displaced_rule) {
1597 update_stats(rule, &put.flow.stats);
1598 rule_post_uninstall(p, displaced_rule);
1601 } else if (displaced_rule) {
1602 rule_uninstall(p, displaced_rule);
1607 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1609 if (rule->installed) {
1610 struct odp_flow_put put;
1611 COVERAGE_INC(ofproto_dp_missed);
1612 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1614 rule_install(ofproto, rule, NULL);
1619 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1621 bool actions_changed = rule_make_actions(ofproto, rule, NULL);
1622 if (rule->may_install) {
1623 if (rule->installed) {
1624 if (actions_changed) {
1625 /* XXX should really do rule_post_uninstall() for the *old* set
1626 * of actions, and distinguish the old stats from the new. */
1627 struct odp_flow_put put;
1628 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1631 rule_install(ofproto, rule, NULL);
1634 rule_uninstall(ofproto, rule);
1639 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1641 uint64_t total_bytes = rule->byte_count + extra_bytes;
1643 if (ofproto->ofhooks->account_flow_cb
1644 && total_bytes > rule->accounted_bytes)
1646 ofproto->ofhooks->account_flow_cb(
1647 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1648 total_bytes - rule->accounted_bytes, ofproto->aux);
1649 rule->accounted_bytes = total_bytes;
1654 rule_uninstall(struct ofproto *p, struct rule *rule)
1656 assert(!rule->cr.wc.wildcards);
1657 if (rule->installed) {
1658 struct odp_flow odp_flow;
1660 odp_flow.key = rule->cr.flow;
1661 odp_flow.actions = NULL;
1662 odp_flow.n_actions = 0;
1663 if (!dpif_flow_del(&p->dpif, &odp_flow)) {
1664 update_stats(rule, &odp_flow.stats);
1666 rule->installed = false;
1668 rule_post_uninstall(p, rule);
1673 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1675 struct rule *super = rule->super;
1677 rule_account(ofproto, rule, 0);
1678 if (ofproto->netflow) {
1679 struct ofexpired expired;
1680 expired.flow = rule->cr.flow;
1681 expired.packet_count = rule->packet_count;
1682 expired.byte_count = rule->byte_count;
1683 expired.used = rule->used;
1684 expired.created = rule->created;
1685 expired.tcp_flags = rule->tcp_flags;
1686 expired.ip_tos = rule->ip_tos;
1687 netflow_expire(ofproto->netflow, &expired);
1690 super->packet_count += rule->packet_count;
1691 super->byte_count += rule->byte_count;
1692 super->tcp_flags |= rule->tcp_flags;
1693 if (rule->packet_count) {
1694 super->ip_tos = rule->ip_tos;
1698 /* Reset counters to prevent double counting if the rule ever gets
1700 rule->packet_count = 0;
1701 rule->byte_count = 0;
1702 rule->accounted_bytes = 0;
1703 rule->tcp_flags = 0;
1708 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1709 struct rconn_packet_counter *counter)
1711 update_openflow_length(msg);
1712 if (rconn_send(ofconn->rconn, msg, counter)) {
1718 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1719 int error, const void *data, size_t len)
1722 struct ofp_error_msg *oem;
1724 if (!(error >> 16)) {
1725 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1730 COVERAGE_INC(ofproto_error);
1731 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1732 oh ? oh->xid : 0, &buf);
1733 oem->type = htons((unsigned int) error >> 16);
1734 oem->code = htons(error & 0xffff);
1735 memcpy(oem->data, data, len);
1736 queue_tx(buf, ofconn, ofconn->reply_counter);
1740 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1743 size_t oh_length = ntohs(oh->length);
1744 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1748 hton_ofp_phy_port(struct ofp_phy_port *opp)
1750 opp->port_no = htons(opp->port_no);
1751 opp->config = htonl(opp->config);
1752 opp->state = htonl(opp->state);
1753 opp->curr = htonl(opp->curr);
1754 opp->advertised = htonl(opp->advertised);
1755 opp->supported = htonl(opp->supported);
1756 opp->peer = htonl(opp->peer);
1760 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
1762 struct ofp_header *rq = oh;
1763 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
1768 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
1769 struct ofp_header *oh)
1771 struct ofp_switch_features *osf;
1773 unsigned int port_no;
1774 struct ofport *port;
1776 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1777 osf->datapath_id = htonll(p->datapath_id);
1778 osf->n_buffers = htonl(pktbuf_capacity());
1780 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1781 OFPC_PORT_STATS | OFPC_MULTI_PHY_TX);
1782 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1783 (1u << OFPAT_SET_VLAN_VID) |
1784 (1u << OFPAT_SET_VLAN_PCP) |
1785 (1u << OFPAT_STRIP_VLAN) |
1786 (1u << OFPAT_SET_DL_SRC) |
1787 (1u << OFPAT_SET_DL_DST) |
1788 (1u << OFPAT_SET_NW_SRC) |
1789 (1u << OFPAT_SET_NW_DST) |
1790 (1u << OFPAT_SET_TP_SRC) |
1791 (1u << OFPAT_SET_TP_DST));
1793 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1794 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1797 queue_tx(buf, ofconn, ofconn->reply_counter);
1802 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1803 struct ofp_header *oh)
1806 struct ofp_switch_config *osc;
1810 /* Figure out flags. */
1811 dpif_get_drop_frags(&p->dpif, &drop_frags);
1812 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1813 if (ofconn->send_flow_exp) {
1814 flags |= OFPC_SEND_FLOW_EXP;
1818 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1819 osc->flags = htons(flags);
1820 osc->miss_send_len = htons(ofconn->miss_send_len);
1821 queue_tx(buf, ofconn, ofconn->reply_counter);
1827 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1828 struct ofp_switch_config *osc)
1833 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1837 flags = ntohs(osc->flags);
1839 ofconn->send_flow_exp = (flags & OFPC_SEND_FLOW_EXP) != 0;
1841 if (ofconn == p->controller) {
1842 switch (flags & OFPC_FRAG_MASK) {
1843 case OFPC_FRAG_NORMAL:
1844 dpif_set_drop_frags(&p->dpif, false);
1846 case OFPC_FRAG_DROP:
1847 dpif_set_drop_frags(&p->dpif, true);
1850 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1856 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1857 if (ntohs(osc->miss_send_len) != 0) {
1858 ofconn->pktbuf = pktbuf_create();
1860 pktbuf_destroy(ofconn->pktbuf);
1864 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1870 add_output_group_action(struct odp_actions *actions, uint16_t group)
1872 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1876 add_controller_action(struct odp_actions *actions,
1877 const struct ofp_action_output *oao)
1879 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
1880 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
1883 struct action_xlate_ctx {
1885 const flow_t *flow; /* Flow to which these actions correspond. */
1886 int recurse; /* Recursion level, via xlate_table_action. */
1887 struct ofproto *ofproto;
1888 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
1889 * null pointer if we are revalidating
1890 * without a packet to refer to. */
1893 struct odp_actions *out; /* Datapath actions. */
1894 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
1895 bool may_setup_flow; /* True ordinarily; false if the actions must
1896 * be reassessed for every packet. */
1899 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
1900 struct action_xlate_ctx *ctx);
1903 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
1905 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
1906 if (!ofport || !(ofport->opp.config & OFPPC_NO_FWD)) {
1907 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
1911 static struct rule *
1912 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
1915 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
1917 /* The rule we found might not be valid, since we could be in need of
1918 * revalidation. If it is not valid, don't return it. */
1921 && ofproto->need_revalidate
1922 && !revalidate_rule(ofproto, rule)) {
1923 COVERAGE_INC(ofproto_invalidated);
1931 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
1933 if (!ctx->recurse) {
1938 flow.in_port = in_port;
1940 rule = lookup_valid_rule(ctx->ofproto, &flow);
1947 do_xlate_actions(rule->actions, rule->n_actions, ctx);
1954 xlate_output_action(struct action_xlate_ctx *ctx,
1955 const struct ofp_action_output *oao)
1959 switch (ntohs(oao->port)) {
1961 add_output_action(ctx, ctx->flow->in_port);
1964 xlate_table_action(ctx, ctx->flow->in_port);
1967 if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet,
1968 ctx->out, ctx->tags,
1969 ctx->ofproto->aux)) {
1970 COVERAGE_INC(ofproto_uninstallable);
1971 ctx->may_setup_flow = false;
1975 add_output_group_action(ctx->out, DP_GROUP_FLOOD);
1978 add_output_group_action(ctx->out, DP_GROUP_ALL);
1980 case OFPP_CONTROLLER:
1981 add_controller_action(ctx->out, oao);
1984 add_output_action(ctx, ODPP_LOCAL);
1987 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
1988 if (odp_port != ctx->flow->in_port) {
1989 add_output_action(ctx, odp_port);
1996 xlate_nicira_action(struct action_xlate_ctx *ctx,
1997 const struct nx_action_header *nah)
1999 const struct nx_action_resubmit *nar;
2000 int subtype = ntohs(nah->subtype);
2002 assert(nah->vendor == htonl(NX_VENDOR_ID));
2004 case NXAST_RESUBMIT:
2005 nar = (const struct nx_action_resubmit *) nah;
2006 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2010 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2016 do_xlate_actions(const union ofp_action *in, size_t n_in,
2017 struct action_xlate_ctx *ctx)
2019 struct actions_iterator iter;
2020 const union ofp_action *ia;
2021 const struct ofport *port;
2023 port = port_array_get(&ctx->ofproto->ports, ctx->flow->in_port);
2024 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2025 port->opp.config & (eth_addr_equals(ctx->flow->dl_dst, stp_eth_addr)
2026 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2027 /* Drop this flow. */
2031 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2032 uint16_t type = ntohs(ia->type);
2033 union odp_action *oa;
2037 xlate_output_action(ctx, &ia->output);
2040 case OFPAT_SET_VLAN_VID:
2041 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2042 oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2045 case OFPAT_SET_VLAN_PCP:
2046 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2047 oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2050 case OFPAT_STRIP_VLAN:
2051 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2054 case OFPAT_SET_DL_SRC:
2055 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2056 memcpy(oa->dl_addr.dl_addr,
2057 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2060 case OFPAT_SET_DL_DST:
2061 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2062 memcpy(oa->dl_addr.dl_addr,
2063 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2066 case OFPAT_SET_NW_SRC:
2067 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2068 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2071 case OFPAT_SET_TP_SRC:
2072 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2073 oa->tp_port.tp_port = ia->tp_port.tp_port;
2077 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2081 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2088 xlate_actions(const union ofp_action *in, size_t n_in,
2089 const flow_t *flow, struct ofproto *ofproto,
2090 const struct ofpbuf *packet,
2091 struct odp_actions *out, tag_type *tags, bool *may_setup_flow)
2093 tag_type no_tags = 0;
2094 struct action_xlate_ctx ctx;
2095 COVERAGE_INC(ofproto_ofp2odp);
2096 odp_actions_init(out);
2099 ctx.ofproto = ofproto;
2100 ctx.packet = packet;
2102 ctx.tags = tags ? tags : &no_tags;
2103 ctx.may_setup_flow = true;
2104 do_xlate_actions(in, n_in, &ctx);
2105 if (may_setup_flow) {
2106 *may_setup_flow = ctx.may_setup_flow;
2108 if (odp_actions_overflow(out)) {
2109 odp_actions_init(out);
2110 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2116 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2117 struct ofp_header *oh)
2119 struct ofp_packet_out *opo;
2120 struct ofpbuf payload, *buffer;
2121 struct odp_actions actions;
2127 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2131 opo = (struct ofp_packet_out *) oh;
2133 COVERAGE_INC(ofproto_packet_out);
2134 if (opo->buffer_id != htonl(UINT32_MAX)) {
2135 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2145 flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2146 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2147 &flow, p, &payload, &actions, NULL, NULL);
2152 dpif_execute(&p->dpif, flow.in_port, actions.actions, actions.n_actions,
2154 ofpbuf_delete(buffer);
2160 update_port_config(struct ofproto *p, struct ofport *port,
2161 uint32_t config, uint32_t mask)
2163 mask &= config ^ port->opp.config;
2164 if (mask & OFPPC_PORT_DOWN) {
2165 if (config & OFPPC_PORT_DOWN) {
2166 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2168 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2171 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2172 if (mask & REVALIDATE_BITS) {
2173 COVERAGE_INC(ofproto_costly_flags);
2174 port->opp.config ^= mask & REVALIDATE_BITS;
2175 p->need_revalidate = true;
2177 #undef REVALIDATE_BITS
2178 if (mask & OFPPC_NO_FLOOD) {
2179 port->opp.config ^= OFPPC_NO_FLOOD;
2180 refresh_port_group(p, DP_GROUP_FLOOD);
2182 if (mask & OFPPC_NO_PACKET_IN) {
2183 port->opp.config ^= OFPPC_NO_PACKET_IN;
2188 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2190 const struct ofp_port_mod *opm;
2191 struct ofport *port;
2194 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2198 opm = (struct ofp_port_mod *) oh;
2200 port = port_array_get(&p->ports,
2201 ofp_port_to_odp_port(ntohs(opm->port_no)));
2203 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2204 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2205 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2207 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2208 if (opm->advertise) {
2209 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2215 static struct ofpbuf *
2216 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2218 struct ofp_stats_reply *osr;
2221 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2222 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2224 osr->flags = htons(0);
2228 static struct ofpbuf *
2229 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2231 return make_stats_reply(request->header.xid, request->type, body_len);
2235 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2237 struct ofpbuf *msg = *msgp;
2238 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2239 if (nbytes + msg->size > UINT16_MAX) {
2240 struct ofp_stats_reply *reply = msg->data;
2241 reply->flags = htons(OFPSF_REPLY_MORE);
2242 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2243 queue_tx(msg, ofconn, ofconn->reply_counter);
2245 return ofpbuf_put_uninit(*msgp, nbytes);
2249 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2250 struct ofp_stats_request *request)
2252 struct ofp_desc_stats *ods;
2255 msg = start_stats_reply(request, sizeof *ods);
2256 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2257 strncpy(ods->mfr_desc, p->manufacturer, sizeof ods->mfr_desc);
2258 strncpy(ods->hw_desc, p->hardware, sizeof ods->hw_desc);
2259 strncpy(ods->sw_desc, p->software, sizeof ods->sw_desc);
2260 strncpy(ods->serial_num, p->serial, sizeof ods->serial_num);
2261 queue_tx(msg, ofconn, ofconn->reply_counter);
2267 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2269 struct rule *rule = rule_from_cls_rule(cls_rule);
2270 int *n_subrules = n_subrules_;
2278 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2279 struct ofp_stats_request *request)
2281 struct ofp_table_stats *ots;
2283 struct odp_stats dpstats;
2284 int n_exact, n_subrules, n_wild;
2286 msg = start_stats_reply(request, sizeof *ots * 2);
2288 /* Count rules of various kinds. */
2290 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2291 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2292 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2295 dpif_get_dp_stats(&p->dpif, &dpstats);
2296 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2297 memset(ots, 0, sizeof *ots);
2298 ots->table_id = TABLEID_HASH;
2299 strcpy(ots->name, "hash");
2300 ots->wildcards = htonl(0);
2301 ots->max_entries = htonl(dpstats.max_capacity);
2302 ots->active_count = htonl(n_exact);
2303 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2305 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2307 /* Classifier table. */
2308 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2309 memset(ots, 0, sizeof *ots);
2310 ots->table_id = TABLEID_CLASSIFIER;
2311 strcpy(ots->name, "classifier");
2312 ots->wildcards = htonl(OFPFW_ALL);
2313 ots->max_entries = htonl(65536);
2314 ots->active_count = htonl(n_wild);
2315 ots->lookup_count = htonll(0); /* XXX */
2316 ots->matched_count = htonll(0); /* XXX */
2318 queue_tx(msg, ofconn, ofconn->reply_counter);
2323 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2324 struct ofp_stats_request *request)
2326 struct ofp_port_stats *ops;
2328 struct ofport *port;
2329 unsigned int port_no;
2331 msg = start_stats_reply(request, sizeof *ops * 16);
2332 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2333 struct netdev_stats stats;
2335 /* Intentionally ignore return value, since errors will set 'stats' to
2336 * all-1s, which is correct for OpenFlow, and netdev_get_stats() will
2338 netdev_get_stats(port->netdev, &stats);
2340 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2341 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2342 memset(ops->pad, 0, sizeof ops->pad);
2343 ops->rx_packets = htonll(stats.rx_packets);
2344 ops->tx_packets = htonll(stats.tx_packets);
2345 ops->rx_bytes = htonll(stats.rx_bytes);
2346 ops->tx_bytes = htonll(stats.tx_bytes);
2347 ops->rx_dropped = htonll(stats.rx_dropped);
2348 ops->tx_dropped = htonll(stats.tx_dropped);
2349 ops->rx_errors = htonll(stats.rx_errors);
2350 ops->tx_errors = htonll(stats.tx_errors);
2351 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2352 ops->rx_over_err = htonll(stats.rx_over_errors);
2353 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2354 ops->collisions = htonll(stats.collisions);
2357 queue_tx(msg, ofconn, ofconn->reply_counter);
2361 struct flow_stats_cbdata {
2362 struct ofproto *ofproto;
2363 struct ofconn *ofconn;
2369 query_stats(struct ofproto *p, struct rule *rule,
2370 uint64_t *packet_countp, uint64_t *byte_countp)
2372 uint64_t packet_count, byte_count;
2373 struct rule *subrule;
2374 struct odp_flow *odp_flows;
2377 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2378 odp_flows = xcalloc(1, n_odp_flows * sizeof *odp_flows);
2379 if (rule->cr.wc.wildcards) {
2381 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2382 odp_flows[i++].key = subrule->cr.flow;
2385 odp_flows[0].key = rule->cr.flow;
2388 packet_count = rule->packet_count;
2389 byte_count = rule->byte_count;
2390 if (!dpif_flow_get_multiple(&p->dpif, odp_flows, n_odp_flows)) {
2392 for (i = 0; i < n_odp_flows; i++) {
2393 struct odp_flow *odp_flow = &odp_flows[i];
2394 packet_count += odp_flow->stats.n_packets;
2395 byte_count += odp_flow->stats.n_bytes;
2400 *packet_countp = packet_count;
2401 *byte_countp = byte_count;
2405 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2407 struct rule *rule = rule_from_cls_rule(rule_);
2408 struct flow_stats_cbdata *cbdata = cbdata_;
2409 struct ofp_flow_stats *ofs;
2410 uint64_t packet_count, byte_count;
2411 size_t act_len, len;
2413 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2417 act_len = sizeof *rule->actions * rule->n_actions;
2418 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2420 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2422 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2423 ofs->length = htons(len);
2424 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2426 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match);
2427 ofs->duration = htonl((time_msec() - rule->created) / 1000);
2428 ofs->priority = htons(rule->cr.priority);
2429 ofs->idle_timeout = htons(rule->idle_timeout);
2430 ofs->hard_timeout = htons(rule->hard_timeout);
2431 memset(ofs->pad2, 0, sizeof ofs->pad2);
2432 ofs->packet_count = htonll(packet_count);
2433 ofs->byte_count = htonll(byte_count);
2434 memcpy(ofs->actions, rule->actions, act_len);
2438 table_id_to_include(uint8_t table_id)
2440 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2441 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2442 : table_id == 0xff ? CLS_INC_ALL
2447 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2448 const struct ofp_stats_request *osr,
2451 struct ofp_flow_stats_request *fsr;
2452 struct flow_stats_cbdata cbdata;
2453 struct cls_rule target;
2455 if (arg_size != sizeof *fsr) {
2456 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2458 fsr = (struct ofp_flow_stats_request *) osr->body;
2460 COVERAGE_INC(ofproto_flows_req);
2462 cbdata.ofconn = ofconn;
2463 cbdata.out_port = fsr->out_port;
2464 cbdata.msg = start_stats_reply(osr, 1024);
2465 cls_rule_from_match(&target, &fsr->match, 0);
2466 classifier_for_each_match(&p->cls, &target,
2467 table_id_to_include(fsr->table_id),
2468 flow_stats_cb, &cbdata);
2469 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2473 struct aggregate_stats_cbdata {
2474 struct ofproto *ofproto;
2476 uint64_t packet_count;
2477 uint64_t byte_count;
2482 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2484 struct rule *rule = rule_from_cls_rule(rule_);
2485 struct aggregate_stats_cbdata *cbdata = cbdata_;
2486 uint64_t packet_count, byte_count;
2488 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2492 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2494 cbdata->packet_count += packet_count;
2495 cbdata->byte_count += byte_count;
2500 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2501 const struct ofp_stats_request *osr,
2504 struct ofp_aggregate_stats_request *asr;
2505 struct ofp_aggregate_stats_reply *reply;
2506 struct aggregate_stats_cbdata cbdata;
2507 struct cls_rule target;
2510 if (arg_size != sizeof *asr) {
2511 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2513 asr = (struct ofp_aggregate_stats_request *) osr->body;
2515 COVERAGE_INC(ofproto_agg_request);
2517 cbdata.out_port = asr->out_port;
2518 cbdata.packet_count = 0;
2519 cbdata.byte_count = 0;
2521 cls_rule_from_match(&target, &asr->match, 0);
2522 classifier_for_each_match(&p->cls, &target,
2523 table_id_to_include(asr->table_id),
2524 aggregate_stats_cb, &cbdata);
2526 msg = start_stats_reply(osr, sizeof *reply);
2527 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2528 reply->flow_count = htonl(cbdata.n_flows);
2529 reply->packet_count = htonll(cbdata.packet_count);
2530 reply->byte_count = htonll(cbdata.byte_count);
2531 queue_tx(msg, ofconn, ofconn->reply_counter);
2536 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2537 struct ofp_header *oh)
2539 struct ofp_stats_request *osr;
2543 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2548 osr = (struct ofp_stats_request *) oh;
2550 switch (ntohs(osr->type)) {
2552 return handle_desc_stats_request(p, ofconn, osr);
2555 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2557 case OFPST_AGGREGATE:
2558 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2561 return handle_table_stats_request(p, ofconn, osr);
2564 return handle_port_stats_request(p, ofconn, osr);
2567 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2570 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2574 static long long int
2575 msec_from_nsec(uint64_t sec, uint32_t nsec)
2577 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2581 update_time(struct rule *rule, const struct odp_flow_stats *stats)
2583 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2584 if (used > rule->used) {
2590 update_stats(struct rule *rule, const struct odp_flow_stats *stats)
2592 update_time(rule, stats);
2593 rule->packet_count += stats->n_packets;
2594 rule->byte_count += stats->n_bytes;
2595 rule->tcp_flags |= stats->tcp_flags;
2596 if (stats->n_packets) {
2597 rule->ip_tos = stats->ip_tos;
2602 add_flow(struct ofproto *p, struct ofconn *ofconn,
2603 struct ofp_flow_mod *ofm, size_t n_actions)
2605 struct ofpbuf *packet;
2610 rule = rule_create(NULL, (const union ofp_action *) ofm->actions,
2611 n_actions, ntohs(ofm->idle_timeout),
2612 ntohs(ofm->hard_timeout));
2613 cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority));
2617 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2618 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2622 rule_insert(p, rule, packet, in_port);
2623 ofpbuf_delete(packet);
2628 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
2629 size_t n_actions, uint16_t command, struct rule *rule)
2631 if (rule_is_hidden(rule)) {
2635 if (command == OFPFC_DELETE) {
2636 rule_remove(p, rule);
2638 size_t actions_len = n_actions * sizeof *rule->actions;
2640 if (n_actions == rule->n_actions
2641 && !memcmp(ofm->actions, rule->actions, actions_len))
2646 free(rule->actions);
2647 rule->actions = xmemdup(ofm->actions, actions_len);
2648 rule->n_actions = n_actions;
2650 if (rule->cr.wc.wildcards) {
2651 COVERAGE_INC(ofproto_mod_wc_flow);
2652 p->need_revalidate = true;
2654 rule_update_actions(p, rule);
2662 modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm,
2663 size_t n_actions, uint16_t command)
2669 flow_from_match(&flow, &wildcards, &ofm->match);
2670 rule = rule_from_cls_rule(classifier_find_rule_exactly(
2671 &p->cls, &flow, wildcards,
2672 ntohs(ofm->priority)));
2675 if (command == OFPFC_DELETE
2676 && ofm->out_port != htons(OFPP_NONE)
2677 && !rule_has_out_port(rule, ofm->out_port)) {
2681 modify_flow(p, ofm, n_actions, command, rule);
2686 struct modify_flows_cbdata {
2687 struct ofproto *ofproto;
2688 const struct ofp_flow_mod *ofm;
2695 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
2697 struct rule *rule = rule_from_cls_rule(rule_);
2698 struct modify_flows_cbdata *cbdata = cbdata_;
2700 if (cbdata->out_port != htons(OFPP_NONE)
2701 && !rule_has_out_port(rule, cbdata->out_port)) {
2705 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions,
2706 cbdata->command, rule);
2710 modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm,
2711 size_t n_actions, uint16_t command)
2713 struct modify_flows_cbdata cbdata;
2714 struct cls_rule target;
2718 cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port
2719 : htons(OFPP_NONE));
2720 cbdata.n_actions = n_actions;
2721 cbdata.command = command;
2723 cls_rule_from_match(&target, &ofm->match, 0);
2725 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2726 modify_flows_cb, &cbdata);
2731 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
2732 struct ofp_flow_mod *ofm)
2737 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
2738 sizeof *ofm->actions, &n_actions);
2743 normalize_match(&ofm->match);
2744 if (!ofm->match.wildcards) {
2745 ofm->priority = htons(UINT16_MAX);
2748 error = validate_actions((const union ofp_action *) ofm->actions,
2749 n_actions, p->max_ports);
2754 switch (ntohs(ofm->command)) {
2756 return add_flow(p, ofconn, ofm, n_actions);
2759 return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY);
2761 case OFPFC_MODIFY_STRICT:
2762 return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY);
2765 return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE);
2767 case OFPFC_DELETE_STRICT:
2768 return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE);
2771 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
2776 send_capability_reply(struct ofproto *p, struct ofconn *ofconn, uint32_t xid)
2778 struct ofmp_capability_reply *ocr;
2780 char capabilities[] = "com.nicira.mgmt.manager=false\n";
2782 ocr = make_openflow_xid(sizeof(*ocr), OFPT_VENDOR, xid, &b);
2783 ocr->header.header.vendor = htonl(NX_VENDOR_ID);
2784 ocr->header.header.subtype = htonl(NXT_MGMT);
2785 ocr->header.type = htons(OFMPT_CAPABILITY_REPLY);
2787 ocr->format = htonl(OFMPCOF_SIMPLE);
2788 ocr->mgmt_id = htonll(p->mgmt_id);
2790 ofpbuf_put(b, capabilities, strlen(capabilities));
2792 queue_tx(b, ofconn, ofconn->reply_counter);
2796 handle_ofmp(struct ofproto *p, struct ofconn *ofconn,
2797 struct ofmp_header *ofmph)
2799 size_t msg_len = ntohs(ofmph->header.header.length);
2800 if (msg_len < sizeof(*ofmph)) {
2801 VLOG_WARN_RL(&rl, "dropping short managment message: %d\n", msg_len);
2802 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2805 if (ofmph->type == htons(OFMPT_CAPABILITY_REQUEST)) {
2806 struct ofmp_capability_request *ofmpcr;
2808 if (msg_len < sizeof(struct ofmp_capability_request)) {
2809 VLOG_WARN_RL(&rl, "dropping short capability request: %d\n",
2811 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2814 ofmpcr = (struct ofmp_capability_request *)ofmph;
2815 if (ofmpcr->format != htonl(OFMPCAF_SIMPLE)) {
2816 /* xxx Find a better type than bad subtype */
2817 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2820 send_capability_reply(p, ofconn, ofmph->header.header.xid);
2823 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2828 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
2830 struct ofp_vendor_header *ovh = msg;
2831 struct nicira_header *nh;
2833 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
2834 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2836 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
2837 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2839 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
2840 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2844 switch (ntohl(nh->subtype)) {
2845 case NXT_STATUS_REQUEST:
2846 return switch_status_handle_request(p->switch_status, ofconn->rconn,
2849 case NXT_ACT_SET_CONFIG:
2850 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
2852 case NXT_ACT_GET_CONFIG:
2853 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
2855 case NXT_COMMAND_REQUEST:
2857 return executer_handle_request(p->executer, ofconn->rconn, msg);
2862 return handle_ofmp(p, ofconn, msg);
2865 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2869 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
2870 struct ofpbuf *ofp_msg)
2872 struct ofp_header *oh = ofp_msg->data;
2875 COVERAGE_INC(ofproto_recv_openflow);
2877 case OFPT_ECHO_REQUEST:
2878 error = handle_echo_request(ofconn, oh);
2881 case OFPT_ECHO_REPLY:
2885 case OFPT_FEATURES_REQUEST:
2886 error = handle_features_request(p, ofconn, oh);
2889 case OFPT_GET_CONFIG_REQUEST:
2890 error = handle_get_config_request(p, ofconn, oh);
2893 case OFPT_SET_CONFIG:
2894 error = handle_set_config(p, ofconn, ofp_msg->data);
2897 case OFPT_PACKET_OUT:
2898 error = handle_packet_out(p, ofconn, ofp_msg->data);
2902 error = handle_port_mod(p, oh);
2906 error = handle_flow_mod(p, ofconn, ofp_msg->data);
2909 case OFPT_STATS_REQUEST:
2910 error = handle_stats_request(p, ofconn, oh);
2914 error = handle_vendor(p, ofconn, ofp_msg->data);
2918 if (VLOG_IS_WARN_ENABLED()) {
2919 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
2920 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
2923 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
2928 send_error_oh(ofconn, ofp_msg->data, error);
2933 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
2935 struct odp_msg *msg = packet->data;
2936 uint16_t in_port = odp_port_to_ofp_port(msg->port);
2938 struct ofpbuf payload;
2941 /* Handle controller actions. */
2942 if (msg->type == _ODPL_ACTION_NR) {
2943 COVERAGE_INC(ofproto_ctlr_action);
2944 pinsched_send(p->action_sched, in_port, packet,
2945 send_packet_in_action, p);
2949 payload.data = msg + 1;
2950 payload.size = msg->length - sizeof *msg;
2951 flow_extract(&payload, msg->port, &flow);
2953 rule = lookup_valid_rule(p, &flow);
2955 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2956 struct ofport *port = port_array_get(&p->ports, msg->port);
2958 if (port->opp.config & OFPPC_NO_PACKET_IN) {
2959 COVERAGE_INC(ofproto_no_packet_in);
2960 /* XXX install 'drop' flow entry */
2961 ofpbuf_delete(packet);
2965 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
2968 COVERAGE_INC(ofproto_packet_in);
2969 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
2973 if (rule->cr.wc.wildcards) {
2974 rule = rule_create_subrule(p, rule, &flow);
2975 rule_make_actions(p, rule, packet);
2977 if (!rule->may_install) {
2978 /* The rule is not installable, that is, we need to process every
2979 * packet, so process the current packet and set its actions into
2981 rule_make_actions(p, rule, packet);
2983 /* XXX revalidate rule if it needs it */
2987 rule_execute(p, rule, &payload, &flow);
2988 rule_reinstall(p, rule);
2989 ofpbuf_delete(packet);
2993 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
2995 struct rule *sub = rule_from_cls_rule(sub_);
2996 struct revalidate_cbdata *cbdata = cbdata_;
2998 if (cbdata->revalidate_all
2999 || (cbdata->revalidate_subrules && sub->super)
3000 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3001 revalidate_rule(cbdata->ofproto, sub);
3006 revalidate_rule(struct ofproto *p, struct rule *rule)
3008 const flow_t *flow = &rule->cr.flow;
3010 COVERAGE_INC(ofproto_revalidate_rule);
3013 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3015 rule_remove(p, rule);
3017 } else if (super != rule->super) {
3018 COVERAGE_INC(ofproto_revalidate_moved);
3019 list_remove(&rule->list);
3020 list_push_back(&super->list, &rule->list);
3021 rule->super = super;
3022 rule->hard_timeout = super->hard_timeout;
3023 rule->idle_timeout = super->idle_timeout;
3024 rule->created = super->created;
3029 rule_update_actions(p, rule);
3033 static struct ofpbuf *
3034 compose_flow_exp(const struct rule *rule, long long int now, uint8_t reason)
3036 struct ofp_flow_expired *ofe;
3039 ofe = make_openflow(sizeof *ofe, OFPT_FLOW_EXPIRED, &buf);
3040 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofe->match);
3041 ofe->priority = htons(rule->cr.priority);
3042 ofe->reason = reason;
3043 ofe->duration = (now - rule->created) / 1000;
3044 ofe->packet_count = rule->packet_count;
3045 ofe->byte_count = rule->byte_count;
3051 send_flow_exp(struct ofproto *p, struct rule *rule,
3052 long long int now, uint8_t reason)
3054 struct ofconn *ofconn;
3055 struct ofconn *prev;
3058 /* We limit the maximum number of queued flow expirations it by accounting
3059 * them under the counter for replies. That works because preventing
3060 * OpenFlow requests from being processed also prevents new flows from
3061 * being added (and expiring). (It also prevents processing OpenFlow
3062 * requests that would not add new flows, so it is imperfect.) */
3065 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3066 if (ofconn->send_flow_exp && rconn_is_connected(ofconn->rconn)) {
3068 queue_tx(ofpbuf_clone(buf), prev, ofconn->reply_counter);
3070 buf = compose_flow_exp(rule, now, reason);
3076 queue_tx(buf, prev, ofconn->reply_counter);
3081 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3083 assert(rule->installed);
3084 assert(!rule->cr.wc.wildcards);
3087 rule_remove(ofproto, rule);
3089 rule_uninstall(ofproto, rule);
3094 expire_rule(struct cls_rule *cls_rule, void *p_)
3096 struct ofproto *p = p_;
3097 struct rule *rule = rule_from_cls_rule(cls_rule);
3098 long long int hard_expire, idle_expire, expire, now;
3100 hard_expire = (rule->hard_timeout
3101 ? rule->created + rule->hard_timeout * 1000
3103 idle_expire = (rule->idle_timeout
3104 && (rule->super || list_is_empty(&rule->list))
3105 ? rule->used + rule->idle_timeout * 1000
3107 expire = MIN(hard_expire, idle_expire);
3108 if (expire == LLONG_MAX) {
3109 if (rule->installed && time_msec() >= rule->used + 5000) {
3110 uninstall_idle_flow(p, rule);
3117 if (rule->installed && now >= rule->used + 5000) {
3118 uninstall_idle_flow(p, rule);
3123 COVERAGE_INC(ofproto_expired);
3124 if (rule->cr.wc.wildcards) {
3125 /* Update stats. (This code will be a no-op if the rule expired
3126 * due to an idle timeout, because in that case the rule has no
3127 * subrules left.) */
3128 struct rule *subrule, *next;
3129 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3130 rule_remove(p, subrule);
3134 send_flow_exp(p, rule, now,
3136 ? OFPER_HARD_TIMEOUT : OFPER_IDLE_TIMEOUT));
3137 rule_remove(p, rule);
3141 update_used(struct ofproto *p)
3143 struct odp_flow *flows;
3148 error = dpif_flow_list_all(&p->dpif, &flows, &n_flows);
3153 for (i = 0; i < n_flows; i++) {
3154 struct odp_flow *f = &flows[i];
3157 rule = rule_from_cls_rule(
3158 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3159 if (!rule || !rule->installed) {
3160 COVERAGE_INC(ofproto_unexpected_rule);
3161 dpif_flow_del(&p->dpif, f);
3165 update_time(rule, &f->stats);
3166 rule_account(p, rule, f->stats.n_bytes);
3172 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3173 const struct ofpbuf *packet, int send_len)
3175 struct ofp_packet_in *opi;
3176 struct ofpbuf payload, *buf;
3177 struct odp_msg *msg;
3180 payload.data = msg + 1;
3181 payload.size = msg->length - sizeof *msg;
3183 send_len = MIN(send_len, payload.size);
3184 buf = ofpbuf_new(sizeof *opi + send_len);
3185 opi = put_openflow_xid(offsetof(struct ofp_packet_in, data),
3186 OFPT_PACKET_IN, 0, buf);
3187 opi->buffer_id = htonl(buffer_id);
3188 opi->total_len = htons(payload.size);
3189 opi->in_port = htons(odp_port_to_ofp_port(msg->port));
3190 opi->reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3191 ofpbuf_put(buf, payload.data, MIN(send_len, payload.size));
3192 update_openflow_length(buf);
3193 rconn_send_with_limit(ofconn->rconn, buf, ofconn->packet_in_counter, 100);
3197 send_packet_in_action(struct ofpbuf *packet, void *p_)
3199 struct ofproto *p = p_;
3200 struct ofconn *ofconn;
3201 struct odp_msg *msg;
3204 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3205 if (ofconn == p->controller || ofconn->miss_send_len) {
3206 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3209 ofpbuf_delete(packet);
3213 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3215 struct ofproto *p = p_;
3216 struct ofconn *ofconn;
3217 struct ofpbuf payload;
3218 struct odp_msg *msg;
3221 payload.data = msg + 1;
3222 payload.size = msg->length - sizeof *msg;
3223 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3224 if (ofconn->miss_send_len) {
3225 uint32_t buffer_id = pktbuf_save(ofconn->pktbuf, &payload,
3227 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3229 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3232 ofpbuf_delete(packet);
3236 pick_datapath_id(struct dpif *dpif, uint64_t fallback_dpid)
3238 char local_name[IF_NAMESIZE];
3239 uint8_t ea[ETH_ADDR_LEN];
3242 error = dpif_get_name(dpif, local_name, sizeof local_name);
3244 error = netdev_nodev_get_etheraddr(local_name, ea);
3246 return eth_addr_to_uint64(ea);
3248 VLOG_WARN("could not get MAC address for %s (%s)",
3249 local_name, strerror(error));
3252 return fallback_dpid;
3256 pick_fallback_dpid(void)
3258 uint8_t ea[ETH_ADDR_LEN];
3259 eth_addr_random(ea);
3260 ea[0] = 0x00; /* Set Nicira OUI. */
3263 return eth_addr_to_uint64(ea);
3267 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3268 struct odp_actions *actions, tag_type *tags,
3271 struct ofproto *ofproto = ofproto_;
3274 /* Drop frames for reserved multicast addresses. */
3275 if (eth_addr_is_reserved(flow->dl_dst)) {
3279 /* Learn source MAC (but don't try to learn from revalidation). */
3280 if (packet != NULL) {
3281 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3284 /* The log messages here could actually be useful in debugging,
3285 * so keep the rate limit relatively high. */
3286 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3287 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3288 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3289 ofproto_revalidate(ofproto, rev_tag);
3293 /* Determine output port. */
3294 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3296 add_output_group_action(actions, DP_GROUP_FLOOD);
3297 } else if (out_port != flow->in_port) {
3298 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3306 static const struct ofhooks default_ofhooks = {
3308 default_normal_ofhook_cb,