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 ofproto 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 netdev_monitor *netdev_monitor;
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(const struct ofproto *);
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 odp_stats stats;
269 /* Connect to datapath and start listening for messages. */
270 error = dpif_open(datapath, &dpif);
272 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
275 error = dpif_get_dp_stats(dpif, &stats);
277 VLOG_ERR("failed to obtain stats for datapath %s: %s",
278 datapath, strerror(error));
282 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION);
284 VLOG_ERR("failed to listen on datapath %s: %s",
285 datapath, strerror(error));
289 dpif_flow_flush(dpif);
290 dpif_recv_purge(dpif);
292 /* Initialize settings. */
293 p = xcalloc(1, sizeof *p);
294 p->fallback_dpid = pick_fallback_dpid();
295 p->datapath_id = p->fallback_dpid;
296 p->manufacturer = xstrdup("Nicira Networks, Inc.");
297 p->hardware = xstrdup("Reference Implementation");
298 p->software = xstrdup(VERSION BUILDNR);
299 p->serial = xstrdup("None");
301 /* Initialize datapath. */
303 p->netdev_monitor = netdev_monitor_create();
304 port_array_init(&p->ports);
305 shash_init(&p->port_by_name);
306 p->max_ports = stats.max_ports;
308 /* Initialize submodules. */
309 p->switch_status = switch_status_create(p);
313 p->miss_sched = p->action_sched = NULL;
317 /* Initialize flow table. */
318 classifier_init(&p->cls);
319 p->need_revalidate = false;
320 p->next_expiration = time_msec() + 1000;
321 tag_set_init(&p->revalidate_set);
323 /* Initialize OpenFlow connections. */
324 list_init(&p->all_conns);
325 p->controller = ofconn_create(p, rconn_create(15, 15));
326 p->controller->pktbuf = pktbuf_create();
327 p->controller->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
333 /* Initialize hooks. */
335 p->ofhooks = ofhooks;
339 p->ofhooks = &default_ofhooks;
341 p->ml = mac_learning_create();
344 /* Register switch status category. */
345 p->ss_cat = switch_status_register(p->switch_status, "remote",
346 rconn_status_cb, p->controller->rconn);
349 error = init_ports(p);
355 /* Pick final datapath ID. */
356 p->datapath_id = pick_datapath_id(p);
357 VLOG_INFO("using datapath ID %012"PRIx64, p->datapath_id);
364 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
366 uint64_t old_dpid = p->datapath_id;
367 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
368 if (p->datapath_id != old_dpid) {
369 VLOG_INFO("datapath ID changed to %012"PRIx64, p->datapath_id);
370 rconn_reconnect(p->controller->rconn);
375 ofproto_set_mgmt_id(struct ofproto *p, uint64_t mgmt_id)
377 p->mgmt_id = mgmt_id;
381 ofproto_set_probe_interval(struct ofproto *p, int probe_interval)
383 probe_interval = probe_interval ? MAX(probe_interval, 5) : 0;
384 rconn_set_probe_interval(p->controller->rconn, probe_interval);
386 int trigger_duration = probe_interval ? probe_interval * 3 : 15;
387 fail_open_set_trigger_duration(p->fail_open, trigger_duration);
392 ofproto_set_max_backoff(struct ofproto *p, int max_backoff)
394 rconn_set_max_backoff(p->controller->rconn, max_backoff);
398 ofproto_set_desc(struct ofproto *p,
399 const char *manufacturer, const char *hardware,
400 const char *software, const char *serial)
403 free(p->manufacturer);
404 p->manufacturer = xstrdup(manufacturer);
408 p->hardware = xstrdup(hardware);
412 p->software = xstrdup(software);
416 p->serial = xstrdup(serial);
421 ofproto_set_in_band(struct ofproto *p, bool in_band)
423 if (in_band != (p->in_band != NULL)) {
425 in_band_create(p, p->switch_status, p->controller->rconn,
429 ofproto_set_discovery(p, false, NULL, true);
430 in_band_destroy(p->in_band);
433 rconn_reconnect(p->controller->rconn);
439 ofproto_set_discovery(struct ofproto *p, bool discovery,
440 const char *re, bool update_resolv_conf)
442 if (discovery != (p->discovery != NULL)) {
444 int error = ofproto_set_in_band(p, true);
448 error = discovery_create(re, update_resolv_conf,
449 p->dpif, p->switch_status,
455 discovery_destroy(p->discovery);
458 rconn_disconnect(p->controller->rconn);
459 } else if (discovery) {
460 discovery_set_update_resolv_conf(p->discovery, update_resolv_conf);
461 return discovery_set_accept_controller_re(p->discovery, re);
467 ofproto_set_controller(struct ofproto *ofproto, const char *controller)
469 if (ofproto->discovery) {
471 } else if (controller) {
472 if (strcmp(rconn_get_name(ofproto->controller->rconn), controller)) {
473 return rconn_connect(ofproto->controller->rconn, controller);
478 rconn_disconnect(ofproto->controller->rconn);
484 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
485 const struct svec *svec)
487 struct pvconn **pvconns = *pvconnsp;
488 size_t n_pvconns = *n_pvconnsp;
492 for (i = 0; i < n_pvconns; i++) {
493 pvconn_close(pvconns[i]);
497 pvconns = xmalloc(svec->n * sizeof *pvconns);
499 for (i = 0; i < svec->n; i++) {
500 const char *name = svec->names[i];
501 struct pvconn *pvconn;
504 error = pvconn_open(name, &pvconn);
506 pvconns[n_pvconns++] = pvconn;
508 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
516 *n_pvconnsp = n_pvconns;
522 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
524 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
528 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
530 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
534 ofproto_set_netflow(struct ofproto *ofproto, const struct svec *collectors,
535 uint8_t engine_type, uint8_t engine_id, bool add_id_to_iface)
537 if (collectors && collectors->n) {
538 if (!ofproto->netflow) {
539 ofproto->netflow = netflow_create();
541 netflow_set_engine(ofproto->netflow, engine_type, engine_id,
543 return netflow_set_collectors(ofproto->netflow, collectors);
545 netflow_destroy(ofproto->netflow);
546 ofproto->netflow = NULL;
552 ofproto_set_failure(struct ofproto *ofproto, bool fail_open)
555 struct rconn *rconn = ofproto->controller->rconn;
556 int trigger_duration = rconn_get_probe_interval(rconn) * 3;
557 if (!ofproto->fail_open) {
558 ofproto->fail_open = fail_open_create(ofproto, trigger_duration,
559 ofproto->switch_status,
562 fail_open_set_trigger_duration(ofproto->fail_open,
566 fail_open_destroy(ofproto->fail_open);
567 ofproto->fail_open = NULL;
572 ofproto_set_rate_limit(struct ofproto *ofproto,
573 int rate_limit, int burst_limit)
575 if (rate_limit > 0) {
576 if (!ofproto->miss_sched) {
577 ofproto->miss_sched = pinsched_create(rate_limit, burst_limit,
578 ofproto->switch_status);
579 ofproto->action_sched = pinsched_create(rate_limit, burst_limit,
582 pinsched_set_limits(ofproto->miss_sched, rate_limit, burst_limit);
583 pinsched_set_limits(ofproto->action_sched,
584 rate_limit, burst_limit);
587 pinsched_destroy(ofproto->miss_sched);
588 ofproto->miss_sched = NULL;
589 pinsched_destroy(ofproto->action_sched);
590 ofproto->action_sched = NULL;
595 ofproto_set_stp(struct ofproto *ofproto UNUSED, bool enable_stp)
599 VLOG_WARN("STP is not yet implemented");
607 ofproto_set_remote_execution(struct ofproto *ofproto, const char *command_acl,
608 const char *command_dir)
611 if (!ofproto->executer) {
612 return executer_create(command_acl, command_dir,
615 executer_set_acl(ofproto->executer, command_acl, command_dir);
618 executer_destroy(ofproto->executer);
619 ofproto->executer = NULL;
625 ofproto_get_datapath_id(const struct ofproto *ofproto)
627 return ofproto->datapath_id;
631 ofproto_get_mgmt_id(const struct ofproto *ofproto)
633 return ofproto->mgmt_id;
637 ofproto_get_probe_interval(const struct ofproto *ofproto)
639 return rconn_get_probe_interval(ofproto->controller->rconn);
643 ofproto_get_max_backoff(const struct ofproto *ofproto)
645 return rconn_get_max_backoff(ofproto->controller->rconn);
649 ofproto_get_in_band(const struct ofproto *ofproto)
651 return ofproto->in_band != NULL;
655 ofproto_get_discovery(const struct ofproto *ofproto)
657 return ofproto->discovery != NULL;
661 ofproto_get_controller(const struct ofproto *ofproto)
663 return rconn_get_name(ofproto->controller->rconn);
667 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
671 for (i = 0; i < ofproto->n_listeners; i++) {
672 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
677 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
681 for (i = 0; i < ofproto->n_snoops; i++) {
682 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
687 ofproto_destroy(struct ofproto *p)
689 struct ofconn *ofconn, *next_ofconn;
690 struct ofport *ofport;
691 unsigned int port_no;
698 ofproto_flush_flows(p);
699 classifier_destroy(&p->cls);
701 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
703 ofconn_destroy(ofconn, p);
707 netdev_monitor_destroy(p->netdev_monitor);
708 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
711 shash_destroy(&p->port_by_name);
713 switch_status_destroy(p->switch_status);
714 in_band_destroy(p->in_band);
715 discovery_destroy(p->discovery);
716 fail_open_destroy(p->fail_open);
717 pinsched_destroy(p->miss_sched);
718 pinsched_destroy(p->action_sched);
719 executer_destroy(p->executer);
720 netflow_destroy(p->netflow);
722 switch_status_unregister(p->ss_cat);
724 for (i = 0; i < p->n_listeners; i++) {
725 pvconn_close(p->listeners[i]);
729 for (i = 0; i < p->n_snoops; i++) {
730 pvconn_close(p->snoops[i]);
734 mac_learning_destroy(p->ml);
740 ofproto_run(struct ofproto *p)
742 int error = ofproto_run1(p);
744 error = ofproto_run2(p, false);
750 process_port_change(struct ofproto *ofproto, int error, char *devname)
752 if (error == ENOBUFS) {
753 reinit_ports(ofproto);
755 update_port(ofproto, devname);
761 ofproto_run1(struct ofproto *p)
763 struct ofconn *ofconn, *next_ofconn;
768 for (i = 0; i < 50; i++) {
772 error = dpif_recv(p->dpif, &buf);
774 if (error == ENODEV) {
775 /* Someone destroyed the datapath behind our back. The caller
776 * better destroy us and give up, because we're just going to
777 * spin from here on out. */
778 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
779 VLOG_ERR_RL(&rl, "%s: datapath was destroyed externally",
786 handle_odp_msg(p, buf);
789 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
790 process_port_change(p, error, devname);
792 while ((error = netdev_monitor_poll(p->netdev_monitor,
793 &devname)) != EAGAIN) {
794 process_port_change(p, error, devname);
798 in_band_run(p->in_band);
801 char *controller_name;
802 if (rconn_is_connectivity_questionable(p->controller->rconn)) {
803 discovery_question_connectivity(p->discovery);
805 if (discovery_run(p->discovery, &controller_name)) {
806 if (controller_name) {
807 rconn_connect(p->controller->rconn, controller_name);
809 rconn_disconnect(p->controller->rconn);
814 fail_open_run(p->fail_open);
816 pinsched_run(p->miss_sched, send_packet_in_miss, p);
817 pinsched_run(p->action_sched, send_packet_in_action, p);
819 executer_run(p->executer);
822 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
824 ofconn_run(ofconn, p);
827 for (i = 0; i < p->n_listeners; i++) {
831 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
833 ofconn_create(p, rconn_new_from_vconn("passive", vconn));
834 } else if (retval != EAGAIN) {
835 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
839 for (i = 0; i < p->n_snoops; i++) {
843 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
845 rconn_add_monitor(p->controller->rconn, vconn);
846 } else if (retval != EAGAIN) {
847 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
851 if (time_msec() >= p->next_expiration) {
852 COVERAGE_INC(ofproto_expiration);
853 p->next_expiration = time_msec() + 1000;
856 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
858 /* Let the hook know that we're at a stable point: all outstanding data
859 * in existing flows has been accounted to the account_cb. Thus, the
860 * hook can now reasonably do operations that depend on having accurate
861 * flow volume accounting (currently, that's just bond rebalancing). */
862 if (p->ofhooks->account_checkpoint_cb) {
863 p->ofhooks->account_checkpoint_cb(p->aux);
868 netflow_run(p->netflow);
874 struct revalidate_cbdata {
875 struct ofproto *ofproto;
876 bool revalidate_all; /* Revalidate all exact-match rules? */
877 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
878 struct tag_set revalidate_set; /* Set of tags to revalidate. */
882 ofproto_run2(struct ofproto *p, bool revalidate_all)
884 if (p->need_revalidate || revalidate_all
885 || !tag_set_is_empty(&p->revalidate_set)) {
886 struct revalidate_cbdata cbdata;
888 cbdata.revalidate_all = revalidate_all;
889 cbdata.revalidate_subrules = p->need_revalidate;
890 cbdata.revalidate_set = p->revalidate_set;
891 tag_set_init(&p->revalidate_set);
892 COVERAGE_INC(ofproto_revalidate);
893 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
894 p->need_revalidate = false;
901 ofproto_wait(struct ofproto *p)
903 struct ofconn *ofconn;
906 dpif_recv_wait(p->dpif);
907 dpif_port_poll_wait(p->dpif);
908 netdev_monitor_poll_wait(p->netdev_monitor);
909 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
913 in_band_wait(p->in_band);
916 discovery_wait(p->discovery);
919 fail_open_wait(p->fail_open);
921 pinsched_wait(p->miss_sched);
922 pinsched_wait(p->action_sched);
924 executer_wait(p->executer);
926 if (!tag_set_is_empty(&p->revalidate_set)) {
927 poll_immediate_wake();
929 if (p->need_revalidate) {
930 /* Shouldn't happen, but if it does just go around again. */
931 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
932 poll_immediate_wake();
933 } else if (p->next_expiration != LLONG_MAX) {
934 poll_timer_wait(p->next_expiration - time_msec());
936 for (i = 0; i < p->n_listeners; i++) {
937 pvconn_wait(p->listeners[i]);
939 for (i = 0; i < p->n_snoops; i++) {
940 pvconn_wait(p->snoops[i]);
945 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
947 tag_set_add(&ofproto->revalidate_set, tag);
951 ofproto_get_revalidate_set(struct ofproto *ofproto)
953 return &ofproto->revalidate_set;
957 ofproto_is_alive(const struct ofproto *p)
959 return p->discovery || rconn_is_alive(p->controller->rconn);
963 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
964 const union ofp_action *actions, size_t n_actions,
965 const struct ofpbuf *packet)
967 struct odp_actions odp_actions;
970 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
976 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
978 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
979 odp_actions.n_actions, packet);
984 ofproto_add_flow(struct ofproto *p,
985 const flow_t *flow, uint32_t wildcards, unsigned int priority,
986 const union ofp_action *actions, size_t n_actions,
990 rule = rule_create(NULL, actions, n_actions,
991 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */, 0);
992 cls_rule_from_flow(&rule->cr, flow, wildcards, priority);
993 rule_insert(p, rule, NULL, 0);
997 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
998 uint32_t wildcards, unsigned int priority)
1002 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1006 rule_remove(ofproto, rule);
1011 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1013 struct rule *rule = rule_from_cls_rule(rule_);
1014 struct ofproto *ofproto = ofproto_;
1016 /* Mark the flow as not installed, even though it might really be
1017 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1018 * There is no point in uninstalling it individually since we are about to
1019 * blow away all the flows with dpif_flow_flush(). */
1020 rule->installed = false;
1022 rule_remove(ofproto, rule);
1026 ofproto_flush_flows(struct ofproto *ofproto)
1028 COVERAGE_INC(ofproto_flush);
1029 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1030 dpif_flow_flush(ofproto->dpif);
1031 if (ofproto->in_band) {
1032 in_band_flushed(ofproto->in_band);
1034 if (ofproto->fail_open) {
1035 fail_open_flushed(ofproto->fail_open);
1040 reinit_ports(struct ofproto *p)
1042 struct svec devnames;
1043 struct ofport *ofport;
1044 unsigned int port_no;
1045 struct odp_port *odp_ports;
1049 svec_init(&devnames);
1050 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1051 svec_add (&devnames, (char *) ofport->opp.name);
1053 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1054 for (i = 0; i < n_odp_ports; i++) {
1055 svec_add (&devnames, odp_ports[i].devname);
1059 svec_sort_unique(&devnames);
1060 for (i = 0; i < devnames.n; i++) {
1061 update_port(p, devnames.names[i]);
1063 svec_destroy(&devnames);
1067 refresh_port_group(struct ofproto *p, unsigned int group)
1071 struct ofport *port;
1072 unsigned int port_no;
1074 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1076 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1078 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1079 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1080 ports[n_ports++] = port_no;
1083 dpif_port_group_set(p->dpif, group, ports, n_ports);
1088 refresh_port_groups(struct ofproto *p)
1090 refresh_port_group(p, DP_GROUP_FLOOD);
1091 refresh_port_group(p, DP_GROUP_ALL);
1094 static struct ofport *
1095 make_ofport(const struct odp_port *odp_port)
1097 enum netdev_flags flags;
1098 struct ofport *ofport;
1099 struct netdev *netdev;
1103 error = netdev_open(odp_port->devname, NETDEV_ETH_TYPE_NONE, &netdev);
1105 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1106 "cannot be opened (%s)",
1107 odp_port->devname, odp_port->port,
1108 odp_port->devname, strerror(error));
1112 ofport = xmalloc(sizeof *ofport);
1113 ofport->netdev = netdev;
1114 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1115 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1116 memcpy(ofport->opp.name, odp_port->devname,
1117 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1118 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1120 netdev_get_flags(netdev, &flags);
1121 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1123 netdev_get_carrier(netdev, &carrier);
1124 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1126 netdev_get_features(netdev,
1127 &ofport->opp.curr, &ofport->opp.advertised,
1128 &ofport->opp.supported, &ofport->opp.peer);
1133 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1135 if (port_array_get(&p->ports, odp_port->port)) {
1136 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1139 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1140 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1149 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1151 const struct ofp_phy_port *a = &a_->opp;
1152 const struct ofp_phy_port *b = &b_->opp;
1154 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1155 return (a->port_no == b->port_no
1156 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1157 && !strcmp((char *) a->name, (char *) b->name)
1158 && a->state == b->state
1159 && a->config == b->config
1160 && a->curr == b->curr
1161 && a->advertised == b->advertised
1162 && a->supported == b->supported
1163 && a->peer == b->peer);
1167 send_port_status(struct ofproto *p, const struct ofport *ofport,
1170 /* XXX Should limit the number of queued port status change messages. */
1171 struct ofconn *ofconn;
1172 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1173 struct ofp_port_status *ops;
1176 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1177 ops->reason = reason;
1178 ops->desc = ofport->opp;
1179 hton_ofp_phy_port(&ops->desc);
1180 queue_tx(b, ofconn, NULL);
1182 if (p->ofhooks->port_changed_cb) {
1183 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1188 ofport_install(struct ofproto *p, struct ofport *ofport)
1190 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1191 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no),
1193 shash_add(&p->port_by_name, (char *) ofport->opp.name, ofport);
1197 ofport_remove(struct ofproto *p, struct ofport *ofport)
1199 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1200 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no), NULL);
1201 shash_delete(&p->port_by_name,
1202 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1206 ofport_free(struct ofport *ofport)
1209 netdev_close(ofport->netdev);
1215 update_port(struct ofproto *p, const char *devname)
1217 struct odp_port odp_port;
1218 struct ofport *ofport;
1221 COVERAGE_INC(ofproto_update_port);
1222 ofport = shash_find_data(&p->port_by_name, devname);
1223 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1227 if (!ofport_conflicts(p, &odp_port)) {
1228 ofport = make_ofport(&odp_port);
1230 ofport_install(p, ofport);
1231 send_port_status(p, ofport, OFPPR_ADD);
1235 /* Modified port. */
1236 struct ofport *new_ofport = make_ofport(&odp_port);
1241 new_ofport->opp.config &= OFPPC_PORT_DOWN;
1242 new_ofport->opp.config |= ofport->opp.config & ~OFPPC_PORT_DOWN;
1243 if (ofport_equal(ofport, new_ofport)) {
1244 /* False alarm--no change. */
1245 ofport_free(new_ofport);
1247 ofport_remove(p, ofport);
1248 ofport_install(p, new_ofport);
1249 ofport_free(ofport);
1250 send_port_status(p, new_ofport, OFPPR_MODIFY);
1253 } else if (error == ENOENT || error == ENODEV) {
1256 send_port_status(p, ofport, OFPPR_DELETE);
1257 ofport_remove(p, ofport);
1258 ofport_free(ofport);
1261 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1262 "%s", strerror(error));
1265 refresh_port_groups(p);
1269 init_ports(struct ofproto *p)
1271 struct odp_port *ports;
1276 error = dpif_port_list(p->dpif, &ports, &n_ports);
1281 for (i = 0; i < n_ports; i++) {
1282 const struct odp_port *odp_port = &ports[i];
1283 if (!ofport_conflicts(p, odp_port)) {
1284 struct ofport *ofport = make_ofport(odp_port);
1286 ofport_install(p, ofport);
1291 refresh_port_groups(p);
1295 static struct ofconn *
1296 ofconn_create(struct ofproto *p, struct rconn *rconn)
1298 struct ofconn *ofconn = xmalloc(sizeof *ofconn);
1299 list_push_back(&p->all_conns, &ofconn->node);
1300 ofconn->rconn = rconn;
1301 ofconn->pktbuf = NULL;
1302 ofconn->send_flow_exp = false;
1303 ofconn->miss_send_len = 0;
1304 ofconn->packet_in_counter = rconn_packet_counter_create ();
1305 ofconn->reply_counter = rconn_packet_counter_create ();
1310 ofconn_destroy(struct ofconn *ofconn, struct ofproto *p)
1313 executer_rconn_closing(p->executer, ofconn->rconn);
1316 list_remove(&ofconn->node);
1317 rconn_destroy(ofconn->rconn);
1318 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1319 rconn_packet_counter_destroy(ofconn->reply_counter);
1320 pktbuf_destroy(ofconn->pktbuf);
1325 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1329 rconn_run(ofconn->rconn);
1331 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1332 /* Limit the number of iterations to prevent other tasks from
1334 for (iteration = 0; iteration < 50; iteration++) {
1335 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1339 handle_openflow(ofconn, p, of_msg);
1340 ofpbuf_delete(of_msg);
1344 if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
1345 ofconn_destroy(ofconn, p);
1350 ofconn_wait(struct ofconn *ofconn)
1352 rconn_run_wait(ofconn->rconn);
1353 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1354 rconn_recv_wait(ofconn->rconn);
1356 COVERAGE_INC(ofproto_ofconn_stuck);
1360 /* Caller is responsible for initializing the 'cr' member of the returned
1362 static struct rule *
1363 rule_create(struct rule *super,
1364 const union ofp_action *actions, size_t n_actions,
1365 uint16_t idle_timeout, uint16_t hard_timeout)
1367 struct rule *rule = xcalloc(1, sizeof *rule);
1368 rule->idle_timeout = idle_timeout;
1369 rule->hard_timeout = hard_timeout;
1370 rule->used = rule->created = time_msec();
1371 rule->super = super;
1373 list_push_back(&super->list, &rule->list);
1375 list_init(&rule->list);
1377 rule->n_actions = n_actions;
1378 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1382 static struct rule *
1383 rule_from_cls_rule(const struct cls_rule *cls_rule)
1385 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1389 rule_free(struct rule *rule)
1391 free(rule->actions);
1392 free(rule->odp_actions);
1396 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1397 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1398 * through all of its subrules and revalidates them, destroying any that no
1399 * longer has a super-rule (which is probably all of them).
1401 * Before calling this function, the caller must make have removed 'rule' from
1402 * the classifier. If 'rule' is an exact-match rule, the caller is also
1403 * responsible for ensuring that it has been uninstalled from the datapath. */
1405 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1408 struct rule *subrule, *next;
1409 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1410 revalidate_rule(ofproto, subrule);
1413 list_remove(&rule->list);
1419 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1421 const union ofp_action *oa;
1422 struct actions_iterator i;
1424 if (out_port == htons(OFPP_NONE)) {
1427 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1428 oa = actions_next(&i)) {
1429 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1436 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1437 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1439 * The flow that 'packet' actually contains does not need to actually match
1440 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1441 * the packet and byte counters for 'rule' will be credited for the packet sent
1442 * out whether or not the packet actually matches 'rule'.
1444 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1445 * the caller must already have accurately composed ODP actions for it given
1446 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1447 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1448 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1449 * actions and apply them to 'packet'. */
1451 rule_execute(struct ofproto *ofproto, struct rule *rule,
1452 struct ofpbuf *packet, const flow_t *flow)
1454 const union odp_action *actions;
1456 struct odp_actions a;
1458 /* Grab or compose the ODP actions.
1460 * The special case for an exact-match 'rule' where 'flow' is not the
1461 * rule's flow is important to avoid, e.g., sending a packet out its input
1462 * port simply because the ODP actions were composed for the wrong
1464 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1465 struct rule *super = rule->super ? rule->super : rule;
1466 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1467 packet, &a, NULL, 0)) {
1470 actions = a.actions;
1471 n_actions = a.n_actions;
1473 actions = rule->odp_actions;
1474 n_actions = rule->n_odp_actions;
1477 /* Execute the ODP actions. */
1478 if (!dpif_execute(ofproto->dpif, flow->in_port,
1479 actions, n_actions, packet)) {
1480 struct odp_flow_stats stats;
1481 flow_extract_stats(flow, packet, &stats);
1482 update_stats(rule, &stats);
1483 rule->used = time_msec();
1488 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1491 struct rule *displaced_rule;
1493 /* Insert the rule in the classifier. */
1494 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1495 if (!rule->cr.wc.wildcards) {
1496 rule_make_actions(p, rule, packet);
1499 /* Send the packet and credit it to the rule. */
1502 flow_extract(packet, in_port, &flow);
1503 rule_execute(p, rule, packet, &flow);
1506 /* Install the rule in the datapath only after sending the packet, to
1507 * avoid packet reordering. */
1508 if (rule->cr.wc.wildcards) {
1509 COVERAGE_INC(ofproto_add_wc_flow);
1510 p->need_revalidate = true;
1512 rule_install(p, rule, displaced_rule);
1515 /* Free the rule that was displaced, if any. */
1516 if (displaced_rule) {
1517 rule_destroy(p, displaced_rule);
1521 static struct rule *
1522 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1525 struct rule *subrule = rule_create(rule, NULL, 0,
1526 rule->idle_timeout, rule->hard_timeout);
1527 COVERAGE_INC(ofproto_subrule_create);
1528 cls_rule_from_flow(&subrule->cr, flow, 0,
1529 (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1530 : rule->cr.priority));
1531 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1537 rule_remove(struct ofproto *ofproto, struct rule *rule)
1539 if (rule->cr.wc.wildcards) {
1540 COVERAGE_INC(ofproto_del_wc_flow);
1541 ofproto->need_revalidate = true;
1543 rule_uninstall(ofproto, rule);
1545 classifier_remove(&ofproto->cls, &rule->cr);
1546 rule_destroy(ofproto, rule);
1549 /* Returns true if the actions changed, false otherwise. */
1551 rule_make_actions(struct ofproto *p, struct rule *rule,
1552 const struct ofpbuf *packet)
1554 const struct rule *super;
1555 struct odp_actions a;
1558 assert(!rule->cr.wc.wildcards);
1560 super = rule->super ? rule->super : rule;
1562 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1563 packet, &a, &rule->tags, &rule->may_install);
1565 actions_len = a.n_actions * sizeof *a.actions;
1566 if (rule->n_odp_actions != a.n_actions
1567 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1568 COVERAGE_INC(ofproto_odp_unchanged);
1569 free(rule->odp_actions);
1570 rule->n_odp_actions = a.n_actions;
1571 rule->odp_actions = xmemdup(a.actions, actions_len);
1579 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1580 struct odp_flow_put *put)
1582 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1583 put->flow.key = rule->cr.flow;
1584 put->flow.actions = rule->odp_actions;
1585 put->flow.n_actions = rule->n_odp_actions;
1587 return dpif_flow_put(ofproto->dpif, put);
1591 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1593 assert(!rule->cr.wc.wildcards);
1595 if (rule->may_install) {
1596 struct odp_flow_put put;
1597 if (!do_put_flow(p, rule,
1598 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1600 rule->installed = true;
1601 if (displaced_rule) {
1602 update_stats(rule, &put.flow.stats);
1603 rule_post_uninstall(p, displaced_rule);
1606 } else if (displaced_rule) {
1607 rule_uninstall(p, displaced_rule);
1612 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1614 if (rule->installed) {
1615 struct odp_flow_put put;
1616 COVERAGE_INC(ofproto_dp_missed);
1617 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1619 rule_install(ofproto, rule, NULL);
1624 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1626 bool actions_changed = rule_make_actions(ofproto, rule, NULL);
1627 if (rule->may_install) {
1628 if (rule->installed) {
1629 if (actions_changed) {
1630 /* XXX should really do rule_post_uninstall() for the *old* set
1631 * of actions, and distinguish the old stats from the new. */
1632 struct odp_flow_put put;
1633 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1636 rule_install(ofproto, rule, NULL);
1639 rule_uninstall(ofproto, rule);
1644 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1646 uint64_t total_bytes = rule->byte_count + extra_bytes;
1648 if (ofproto->ofhooks->account_flow_cb
1649 && total_bytes > rule->accounted_bytes)
1651 ofproto->ofhooks->account_flow_cb(
1652 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1653 total_bytes - rule->accounted_bytes, ofproto->aux);
1654 rule->accounted_bytes = total_bytes;
1659 rule_uninstall(struct ofproto *p, struct rule *rule)
1661 assert(!rule->cr.wc.wildcards);
1662 if (rule->installed) {
1663 struct odp_flow odp_flow;
1665 odp_flow.key = rule->cr.flow;
1666 odp_flow.actions = NULL;
1667 odp_flow.n_actions = 0;
1668 if (!dpif_flow_del(p->dpif, &odp_flow)) {
1669 update_stats(rule, &odp_flow.stats);
1671 rule->installed = false;
1673 rule_post_uninstall(p, rule);
1678 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1680 struct rule *super = rule->super;
1682 rule_account(ofproto, rule, 0);
1683 if (ofproto->netflow) {
1684 struct ofexpired expired;
1685 expired.flow = rule->cr.flow;
1686 expired.packet_count = rule->packet_count;
1687 expired.byte_count = rule->byte_count;
1688 expired.used = rule->used;
1689 expired.created = rule->created;
1690 expired.tcp_flags = rule->tcp_flags;
1691 expired.ip_tos = rule->ip_tos;
1692 netflow_expire(ofproto->netflow, &expired);
1695 super->packet_count += rule->packet_count;
1696 super->byte_count += rule->byte_count;
1697 super->tcp_flags |= rule->tcp_flags;
1698 if (rule->packet_count) {
1699 super->ip_tos = rule->ip_tos;
1703 /* Reset counters to prevent double counting if the rule ever gets
1705 rule->packet_count = 0;
1706 rule->byte_count = 0;
1707 rule->accounted_bytes = 0;
1708 rule->tcp_flags = 0;
1713 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1714 struct rconn_packet_counter *counter)
1716 update_openflow_length(msg);
1717 if (rconn_send(ofconn->rconn, msg, counter)) {
1723 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1724 int error, const void *data, size_t len)
1727 struct ofp_error_msg *oem;
1729 if (!(error >> 16)) {
1730 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1735 COVERAGE_INC(ofproto_error);
1736 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1737 oh ? oh->xid : 0, &buf);
1738 oem->type = htons((unsigned int) error >> 16);
1739 oem->code = htons(error & 0xffff);
1740 memcpy(oem->data, data, len);
1741 queue_tx(buf, ofconn, ofconn->reply_counter);
1745 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1748 size_t oh_length = ntohs(oh->length);
1749 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1753 hton_ofp_phy_port(struct ofp_phy_port *opp)
1755 opp->port_no = htons(opp->port_no);
1756 opp->config = htonl(opp->config);
1757 opp->state = htonl(opp->state);
1758 opp->curr = htonl(opp->curr);
1759 opp->advertised = htonl(opp->advertised);
1760 opp->supported = htonl(opp->supported);
1761 opp->peer = htonl(opp->peer);
1765 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
1767 struct ofp_header *rq = oh;
1768 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
1773 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
1774 struct ofp_header *oh)
1776 struct ofp_switch_features *osf;
1778 unsigned int port_no;
1779 struct ofport *port;
1781 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1782 osf->datapath_id = htonll(p->datapath_id);
1783 osf->n_buffers = htonl(pktbuf_capacity());
1785 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1786 OFPC_PORT_STATS | OFPC_MULTI_PHY_TX);
1787 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1788 (1u << OFPAT_SET_VLAN_VID) |
1789 (1u << OFPAT_SET_VLAN_PCP) |
1790 (1u << OFPAT_STRIP_VLAN) |
1791 (1u << OFPAT_SET_DL_SRC) |
1792 (1u << OFPAT_SET_DL_DST) |
1793 (1u << OFPAT_SET_NW_SRC) |
1794 (1u << OFPAT_SET_NW_DST) |
1795 (1u << OFPAT_SET_TP_SRC) |
1796 (1u << OFPAT_SET_TP_DST));
1798 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1799 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1802 queue_tx(buf, ofconn, ofconn->reply_counter);
1807 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1808 struct ofp_header *oh)
1811 struct ofp_switch_config *osc;
1815 /* Figure out flags. */
1816 dpif_get_drop_frags(p->dpif, &drop_frags);
1817 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1818 if (ofconn->send_flow_exp) {
1819 flags |= OFPC_SEND_FLOW_EXP;
1823 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1824 osc->flags = htons(flags);
1825 osc->miss_send_len = htons(ofconn->miss_send_len);
1826 queue_tx(buf, ofconn, ofconn->reply_counter);
1832 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1833 struct ofp_switch_config *osc)
1838 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1842 flags = ntohs(osc->flags);
1844 ofconn->send_flow_exp = (flags & OFPC_SEND_FLOW_EXP) != 0;
1846 if (ofconn == p->controller) {
1847 switch (flags & OFPC_FRAG_MASK) {
1848 case OFPC_FRAG_NORMAL:
1849 dpif_set_drop_frags(p->dpif, false);
1851 case OFPC_FRAG_DROP:
1852 dpif_set_drop_frags(p->dpif, true);
1855 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1861 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1862 if (ntohs(osc->miss_send_len) != 0) {
1863 ofconn->pktbuf = pktbuf_create();
1865 pktbuf_destroy(ofconn->pktbuf);
1869 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1875 add_output_group_action(struct odp_actions *actions, uint16_t group)
1877 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1881 add_controller_action(struct odp_actions *actions,
1882 const struct ofp_action_output *oao)
1884 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
1885 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
1888 struct action_xlate_ctx {
1890 const flow_t *flow; /* Flow to which these actions correspond. */
1891 int recurse; /* Recursion level, via xlate_table_action. */
1892 struct ofproto *ofproto;
1893 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
1894 * null pointer if we are revalidating
1895 * without a packet to refer to. */
1898 struct odp_actions *out; /* Datapath actions. */
1899 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
1900 bool may_setup_flow; /* True ordinarily; false if the actions must
1901 * be reassessed for every packet. */
1904 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
1905 struct action_xlate_ctx *ctx);
1908 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
1910 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
1911 if (!ofport || !(ofport->opp.config & OFPPC_NO_FWD)) {
1912 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
1916 static struct rule *
1917 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
1920 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
1922 /* The rule we found might not be valid, since we could be in need of
1923 * revalidation. If it is not valid, don't return it. */
1926 && ofproto->need_revalidate
1927 && !revalidate_rule(ofproto, rule)) {
1928 COVERAGE_INC(ofproto_invalidated);
1936 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
1938 if (!ctx->recurse) {
1943 flow.in_port = in_port;
1945 rule = lookup_valid_rule(ctx->ofproto, &flow);
1952 do_xlate_actions(rule->actions, rule->n_actions, ctx);
1959 xlate_output_action(struct action_xlate_ctx *ctx,
1960 const struct ofp_action_output *oao)
1964 switch (ntohs(oao->port)) {
1966 add_output_action(ctx, ctx->flow->in_port);
1969 xlate_table_action(ctx, ctx->flow->in_port);
1972 if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet,
1973 ctx->out, ctx->tags,
1974 ctx->ofproto->aux)) {
1975 COVERAGE_INC(ofproto_uninstallable);
1976 ctx->may_setup_flow = false;
1980 add_output_group_action(ctx->out, DP_GROUP_FLOOD);
1983 add_output_group_action(ctx->out, DP_GROUP_ALL);
1985 case OFPP_CONTROLLER:
1986 add_controller_action(ctx->out, oao);
1989 add_output_action(ctx, ODPP_LOCAL);
1992 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
1993 if (odp_port != ctx->flow->in_port) {
1994 add_output_action(ctx, odp_port);
2001 xlate_nicira_action(struct action_xlate_ctx *ctx,
2002 const struct nx_action_header *nah)
2004 const struct nx_action_resubmit *nar;
2005 int subtype = ntohs(nah->subtype);
2007 assert(nah->vendor == htonl(NX_VENDOR_ID));
2009 case NXAST_RESUBMIT:
2010 nar = (const struct nx_action_resubmit *) nah;
2011 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2015 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2021 do_xlate_actions(const union ofp_action *in, size_t n_in,
2022 struct action_xlate_ctx *ctx)
2024 struct actions_iterator iter;
2025 const union ofp_action *ia;
2026 const struct ofport *port;
2028 port = port_array_get(&ctx->ofproto->ports, ctx->flow->in_port);
2029 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2030 port->opp.config & (eth_addr_equals(ctx->flow->dl_dst, stp_eth_addr)
2031 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2032 /* Drop this flow. */
2036 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2037 uint16_t type = ntohs(ia->type);
2038 union odp_action *oa;
2042 xlate_output_action(ctx, &ia->output);
2045 case OFPAT_SET_VLAN_VID:
2046 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2047 oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2050 case OFPAT_SET_VLAN_PCP:
2051 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2052 oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2055 case OFPAT_STRIP_VLAN:
2056 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2059 case OFPAT_SET_DL_SRC:
2060 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2061 memcpy(oa->dl_addr.dl_addr,
2062 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2065 case OFPAT_SET_DL_DST:
2066 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2067 memcpy(oa->dl_addr.dl_addr,
2068 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2071 case OFPAT_SET_NW_SRC:
2072 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2073 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2076 case OFPAT_SET_TP_SRC:
2077 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2078 oa->tp_port.tp_port = ia->tp_port.tp_port;
2082 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2086 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2093 xlate_actions(const union ofp_action *in, size_t n_in,
2094 const flow_t *flow, struct ofproto *ofproto,
2095 const struct ofpbuf *packet,
2096 struct odp_actions *out, tag_type *tags, bool *may_setup_flow)
2098 tag_type no_tags = 0;
2099 struct action_xlate_ctx ctx;
2100 COVERAGE_INC(ofproto_ofp2odp);
2101 odp_actions_init(out);
2104 ctx.ofproto = ofproto;
2105 ctx.packet = packet;
2107 ctx.tags = tags ? tags : &no_tags;
2108 ctx.may_setup_flow = true;
2109 do_xlate_actions(in, n_in, &ctx);
2110 if (may_setup_flow) {
2111 *may_setup_flow = ctx.may_setup_flow;
2113 if (odp_actions_overflow(out)) {
2114 odp_actions_init(out);
2115 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2121 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2122 struct ofp_header *oh)
2124 struct ofp_packet_out *opo;
2125 struct ofpbuf payload, *buffer;
2126 struct odp_actions actions;
2132 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2136 opo = (struct ofp_packet_out *) oh;
2138 COVERAGE_INC(ofproto_packet_out);
2139 if (opo->buffer_id != htonl(UINT32_MAX)) {
2140 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2150 flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2151 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2152 &flow, p, &payload, &actions, NULL, NULL);
2157 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2159 ofpbuf_delete(buffer);
2165 update_port_config(struct ofproto *p, struct ofport *port,
2166 uint32_t config, uint32_t mask)
2168 mask &= config ^ port->opp.config;
2169 if (mask & OFPPC_PORT_DOWN) {
2170 if (config & OFPPC_PORT_DOWN) {
2171 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2173 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2176 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2177 if (mask & REVALIDATE_BITS) {
2178 COVERAGE_INC(ofproto_costly_flags);
2179 port->opp.config ^= mask & REVALIDATE_BITS;
2180 p->need_revalidate = true;
2182 #undef REVALIDATE_BITS
2183 if (mask & OFPPC_NO_FLOOD) {
2184 port->opp.config ^= OFPPC_NO_FLOOD;
2185 refresh_port_group(p, DP_GROUP_FLOOD);
2187 if (mask & OFPPC_NO_PACKET_IN) {
2188 port->opp.config ^= OFPPC_NO_PACKET_IN;
2193 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2195 const struct ofp_port_mod *opm;
2196 struct ofport *port;
2199 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2203 opm = (struct ofp_port_mod *) oh;
2205 port = port_array_get(&p->ports,
2206 ofp_port_to_odp_port(ntohs(opm->port_no)));
2208 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2209 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2210 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2212 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2213 if (opm->advertise) {
2214 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2220 static struct ofpbuf *
2221 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2223 struct ofp_stats_reply *osr;
2226 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2227 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2229 osr->flags = htons(0);
2233 static struct ofpbuf *
2234 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2236 return make_stats_reply(request->header.xid, request->type, body_len);
2240 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2242 struct ofpbuf *msg = *msgp;
2243 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2244 if (nbytes + msg->size > UINT16_MAX) {
2245 struct ofp_stats_reply *reply = msg->data;
2246 reply->flags = htons(OFPSF_REPLY_MORE);
2247 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2248 queue_tx(msg, ofconn, ofconn->reply_counter);
2250 return ofpbuf_put_uninit(*msgp, nbytes);
2254 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2255 struct ofp_stats_request *request)
2257 struct ofp_desc_stats *ods;
2260 msg = start_stats_reply(request, sizeof *ods);
2261 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2262 strncpy(ods->mfr_desc, p->manufacturer, sizeof ods->mfr_desc);
2263 strncpy(ods->hw_desc, p->hardware, sizeof ods->hw_desc);
2264 strncpy(ods->sw_desc, p->software, sizeof ods->sw_desc);
2265 strncpy(ods->serial_num, p->serial, sizeof ods->serial_num);
2266 queue_tx(msg, ofconn, ofconn->reply_counter);
2272 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2274 struct rule *rule = rule_from_cls_rule(cls_rule);
2275 int *n_subrules = n_subrules_;
2283 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2284 struct ofp_stats_request *request)
2286 struct ofp_table_stats *ots;
2288 struct odp_stats dpstats;
2289 int n_exact, n_subrules, n_wild;
2291 msg = start_stats_reply(request, sizeof *ots * 2);
2293 /* Count rules of various kinds. */
2295 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2296 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2297 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2300 dpif_get_dp_stats(p->dpif, &dpstats);
2301 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2302 memset(ots, 0, sizeof *ots);
2303 ots->table_id = TABLEID_HASH;
2304 strcpy(ots->name, "hash");
2305 ots->wildcards = htonl(0);
2306 ots->max_entries = htonl(dpstats.max_capacity);
2307 ots->active_count = htonl(n_exact);
2308 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2310 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2312 /* Classifier table. */
2313 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2314 memset(ots, 0, sizeof *ots);
2315 ots->table_id = TABLEID_CLASSIFIER;
2316 strcpy(ots->name, "classifier");
2317 ots->wildcards = htonl(OFPFW_ALL);
2318 ots->max_entries = htonl(65536);
2319 ots->active_count = htonl(n_wild);
2320 ots->lookup_count = htonll(0); /* XXX */
2321 ots->matched_count = htonll(0); /* XXX */
2323 queue_tx(msg, ofconn, ofconn->reply_counter);
2328 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2329 struct ofp_stats_request *request)
2331 struct ofp_port_stats *ops;
2333 struct ofport *port;
2334 unsigned int port_no;
2336 msg = start_stats_reply(request, sizeof *ops * 16);
2337 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2338 struct netdev_stats stats;
2340 /* Intentionally ignore return value, since errors will set 'stats' to
2341 * all-1s, which is correct for OpenFlow, and netdev_get_stats() will
2343 netdev_get_stats(port->netdev, &stats);
2345 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2346 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2347 memset(ops->pad, 0, sizeof ops->pad);
2348 ops->rx_packets = htonll(stats.rx_packets);
2349 ops->tx_packets = htonll(stats.tx_packets);
2350 ops->rx_bytes = htonll(stats.rx_bytes);
2351 ops->tx_bytes = htonll(stats.tx_bytes);
2352 ops->rx_dropped = htonll(stats.rx_dropped);
2353 ops->tx_dropped = htonll(stats.tx_dropped);
2354 ops->rx_errors = htonll(stats.rx_errors);
2355 ops->tx_errors = htonll(stats.tx_errors);
2356 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2357 ops->rx_over_err = htonll(stats.rx_over_errors);
2358 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2359 ops->collisions = htonll(stats.collisions);
2362 queue_tx(msg, ofconn, ofconn->reply_counter);
2366 struct flow_stats_cbdata {
2367 struct ofproto *ofproto;
2368 struct ofconn *ofconn;
2374 query_stats(struct ofproto *p, struct rule *rule,
2375 uint64_t *packet_countp, uint64_t *byte_countp)
2377 uint64_t packet_count, byte_count;
2378 struct rule *subrule;
2379 struct odp_flow *odp_flows;
2382 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2383 odp_flows = xcalloc(1, n_odp_flows * sizeof *odp_flows);
2384 if (rule->cr.wc.wildcards) {
2386 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2387 odp_flows[i++].key = subrule->cr.flow;
2390 odp_flows[0].key = rule->cr.flow;
2393 packet_count = rule->packet_count;
2394 byte_count = rule->byte_count;
2395 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2397 for (i = 0; i < n_odp_flows; i++) {
2398 struct odp_flow *odp_flow = &odp_flows[i];
2399 packet_count += odp_flow->stats.n_packets;
2400 byte_count += odp_flow->stats.n_bytes;
2405 *packet_countp = packet_count;
2406 *byte_countp = byte_count;
2410 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2412 struct rule *rule = rule_from_cls_rule(rule_);
2413 struct flow_stats_cbdata *cbdata = cbdata_;
2414 struct ofp_flow_stats *ofs;
2415 uint64_t packet_count, byte_count;
2416 size_t act_len, len;
2418 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2422 act_len = sizeof *rule->actions * rule->n_actions;
2423 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2425 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2427 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2428 ofs->length = htons(len);
2429 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2431 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match);
2432 ofs->duration = htonl((time_msec() - rule->created) / 1000);
2433 ofs->priority = htons(rule->cr.priority);
2434 ofs->idle_timeout = htons(rule->idle_timeout);
2435 ofs->hard_timeout = htons(rule->hard_timeout);
2436 memset(ofs->pad2, 0, sizeof ofs->pad2);
2437 ofs->packet_count = htonll(packet_count);
2438 ofs->byte_count = htonll(byte_count);
2439 memcpy(ofs->actions, rule->actions, act_len);
2443 table_id_to_include(uint8_t table_id)
2445 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2446 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2447 : table_id == 0xff ? CLS_INC_ALL
2452 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2453 const struct ofp_stats_request *osr,
2456 struct ofp_flow_stats_request *fsr;
2457 struct flow_stats_cbdata cbdata;
2458 struct cls_rule target;
2460 if (arg_size != sizeof *fsr) {
2461 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2463 fsr = (struct ofp_flow_stats_request *) osr->body;
2465 COVERAGE_INC(ofproto_flows_req);
2467 cbdata.ofconn = ofconn;
2468 cbdata.out_port = fsr->out_port;
2469 cbdata.msg = start_stats_reply(osr, 1024);
2470 cls_rule_from_match(&target, &fsr->match, 0);
2471 classifier_for_each_match(&p->cls, &target,
2472 table_id_to_include(fsr->table_id),
2473 flow_stats_cb, &cbdata);
2474 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2478 struct aggregate_stats_cbdata {
2479 struct ofproto *ofproto;
2481 uint64_t packet_count;
2482 uint64_t byte_count;
2487 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2489 struct rule *rule = rule_from_cls_rule(rule_);
2490 struct aggregate_stats_cbdata *cbdata = cbdata_;
2491 uint64_t packet_count, byte_count;
2493 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2497 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2499 cbdata->packet_count += packet_count;
2500 cbdata->byte_count += byte_count;
2505 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2506 const struct ofp_stats_request *osr,
2509 struct ofp_aggregate_stats_request *asr;
2510 struct ofp_aggregate_stats_reply *reply;
2511 struct aggregate_stats_cbdata cbdata;
2512 struct cls_rule target;
2515 if (arg_size != sizeof *asr) {
2516 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2518 asr = (struct ofp_aggregate_stats_request *) osr->body;
2520 COVERAGE_INC(ofproto_agg_request);
2522 cbdata.out_port = asr->out_port;
2523 cbdata.packet_count = 0;
2524 cbdata.byte_count = 0;
2526 cls_rule_from_match(&target, &asr->match, 0);
2527 classifier_for_each_match(&p->cls, &target,
2528 table_id_to_include(asr->table_id),
2529 aggregate_stats_cb, &cbdata);
2531 msg = start_stats_reply(osr, sizeof *reply);
2532 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2533 reply->flow_count = htonl(cbdata.n_flows);
2534 reply->packet_count = htonll(cbdata.packet_count);
2535 reply->byte_count = htonll(cbdata.byte_count);
2536 queue_tx(msg, ofconn, ofconn->reply_counter);
2541 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2542 struct ofp_header *oh)
2544 struct ofp_stats_request *osr;
2548 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2553 osr = (struct ofp_stats_request *) oh;
2555 switch (ntohs(osr->type)) {
2557 return handle_desc_stats_request(p, ofconn, osr);
2560 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2562 case OFPST_AGGREGATE:
2563 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2566 return handle_table_stats_request(p, ofconn, osr);
2569 return handle_port_stats_request(p, ofconn, osr);
2572 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2575 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2579 static long long int
2580 msec_from_nsec(uint64_t sec, uint32_t nsec)
2582 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2586 update_time(struct rule *rule, const struct odp_flow_stats *stats)
2588 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2589 if (used > rule->used) {
2595 update_stats(struct rule *rule, const struct odp_flow_stats *stats)
2597 update_time(rule, stats);
2598 rule->packet_count += stats->n_packets;
2599 rule->byte_count += stats->n_bytes;
2600 rule->tcp_flags |= stats->tcp_flags;
2601 if (stats->n_packets) {
2602 rule->ip_tos = stats->ip_tos;
2607 add_flow(struct ofproto *p, struct ofconn *ofconn,
2608 struct ofp_flow_mod *ofm, size_t n_actions)
2610 struct ofpbuf *packet;
2615 rule = rule_create(NULL, (const union ofp_action *) ofm->actions,
2616 n_actions, ntohs(ofm->idle_timeout),
2617 ntohs(ofm->hard_timeout));
2618 cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority));
2622 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2623 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2627 rule_insert(p, rule, packet, in_port);
2628 ofpbuf_delete(packet);
2633 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
2634 size_t n_actions, uint16_t command, struct rule *rule)
2636 if (rule_is_hidden(rule)) {
2640 if (command == OFPFC_DELETE) {
2641 rule_remove(p, rule);
2643 size_t actions_len = n_actions * sizeof *rule->actions;
2645 if (n_actions == rule->n_actions
2646 && !memcmp(ofm->actions, rule->actions, actions_len))
2651 free(rule->actions);
2652 rule->actions = xmemdup(ofm->actions, actions_len);
2653 rule->n_actions = n_actions;
2655 if (rule->cr.wc.wildcards) {
2656 COVERAGE_INC(ofproto_mod_wc_flow);
2657 p->need_revalidate = true;
2659 rule_update_actions(p, rule);
2667 modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm,
2668 size_t n_actions, uint16_t command)
2674 flow_from_match(&flow, &wildcards, &ofm->match);
2675 rule = rule_from_cls_rule(classifier_find_rule_exactly(
2676 &p->cls, &flow, wildcards,
2677 ntohs(ofm->priority)));
2680 if (command == OFPFC_DELETE
2681 && ofm->out_port != htons(OFPP_NONE)
2682 && !rule_has_out_port(rule, ofm->out_port)) {
2686 modify_flow(p, ofm, n_actions, command, rule);
2691 struct modify_flows_cbdata {
2692 struct ofproto *ofproto;
2693 const struct ofp_flow_mod *ofm;
2700 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
2702 struct rule *rule = rule_from_cls_rule(rule_);
2703 struct modify_flows_cbdata *cbdata = cbdata_;
2705 if (cbdata->out_port != htons(OFPP_NONE)
2706 && !rule_has_out_port(rule, cbdata->out_port)) {
2710 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions,
2711 cbdata->command, rule);
2715 modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm,
2716 size_t n_actions, uint16_t command)
2718 struct modify_flows_cbdata cbdata;
2719 struct cls_rule target;
2723 cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port
2724 : htons(OFPP_NONE));
2725 cbdata.n_actions = n_actions;
2726 cbdata.command = command;
2728 cls_rule_from_match(&target, &ofm->match, 0);
2730 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2731 modify_flows_cb, &cbdata);
2736 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
2737 struct ofp_flow_mod *ofm)
2742 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
2743 sizeof *ofm->actions, &n_actions);
2748 normalize_match(&ofm->match);
2749 if (!ofm->match.wildcards) {
2750 ofm->priority = htons(UINT16_MAX);
2753 error = validate_actions((const union ofp_action *) ofm->actions,
2754 n_actions, p->max_ports);
2759 switch (ntohs(ofm->command)) {
2761 return add_flow(p, ofconn, ofm, n_actions);
2764 return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY);
2766 case OFPFC_MODIFY_STRICT:
2767 return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY);
2770 return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE);
2772 case OFPFC_DELETE_STRICT:
2773 return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE);
2776 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
2781 send_capability_reply(struct ofproto *p, struct ofconn *ofconn, uint32_t xid)
2783 struct ofmp_capability_reply *ocr;
2785 char capabilities[] = "com.nicira.mgmt.manager=false\n";
2787 ocr = make_openflow_xid(sizeof(*ocr), OFPT_VENDOR, xid, &b);
2788 ocr->header.header.vendor = htonl(NX_VENDOR_ID);
2789 ocr->header.header.subtype = htonl(NXT_MGMT);
2790 ocr->header.type = htons(OFMPT_CAPABILITY_REPLY);
2792 ocr->format = htonl(OFMPCOF_SIMPLE);
2793 ocr->mgmt_id = htonll(p->mgmt_id);
2795 ofpbuf_put(b, capabilities, strlen(capabilities));
2797 queue_tx(b, ofconn, ofconn->reply_counter);
2801 handle_ofmp(struct ofproto *p, struct ofconn *ofconn,
2802 struct ofmp_header *ofmph)
2804 size_t msg_len = ntohs(ofmph->header.header.length);
2805 if (msg_len < sizeof(*ofmph)) {
2806 VLOG_WARN_RL(&rl, "dropping short managment message: %d\n", msg_len);
2807 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2810 if (ofmph->type == htons(OFMPT_CAPABILITY_REQUEST)) {
2811 struct ofmp_capability_request *ofmpcr;
2813 if (msg_len < sizeof(struct ofmp_capability_request)) {
2814 VLOG_WARN_RL(&rl, "dropping short capability request: %d\n",
2816 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2819 ofmpcr = (struct ofmp_capability_request *)ofmph;
2820 if (ofmpcr->format != htonl(OFMPCAF_SIMPLE)) {
2821 /* xxx Find a better type than bad subtype */
2822 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2825 send_capability_reply(p, ofconn, ofmph->header.header.xid);
2828 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2833 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
2835 struct ofp_vendor_header *ovh = msg;
2836 struct nicira_header *nh;
2838 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
2839 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2841 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
2842 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2844 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
2845 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2849 switch (ntohl(nh->subtype)) {
2850 case NXT_STATUS_REQUEST:
2851 return switch_status_handle_request(p->switch_status, ofconn->rconn,
2854 case NXT_ACT_SET_CONFIG:
2855 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
2857 case NXT_ACT_GET_CONFIG:
2858 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
2860 case NXT_COMMAND_REQUEST:
2862 return executer_handle_request(p->executer, ofconn->rconn, msg);
2867 return handle_ofmp(p, ofconn, msg);
2870 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2874 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
2875 struct ofpbuf *ofp_msg)
2877 struct ofp_header *oh = ofp_msg->data;
2880 COVERAGE_INC(ofproto_recv_openflow);
2882 case OFPT_ECHO_REQUEST:
2883 error = handle_echo_request(ofconn, oh);
2886 case OFPT_ECHO_REPLY:
2890 case OFPT_FEATURES_REQUEST:
2891 error = handle_features_request(p, ofconn, oh);
2894 case OFPT_GET_CONFIG_REQUEST:
2895 error = handle_get_config_request(p, ofconn, oh);
2898 case OFPT_SET_CONFIG:
2899 error = handle_set_config(p, ofconn, ofp_msg->data);
2902 case OFPT_PACKET_OUT:
2903 error = handle_packet_out(p, ofconn, ofp_msg->data);
2907 error = handle_port_mod(p, oh);
2911 error = handle_flow_mod(p, ofconn, ofp_msg->data);
2914 case OFPT_STATS_REQUEST:
2915 error = handle_stats_request(p, ofconn, oh);
2919 error = handle_vendor(p, ofconn, ofp_msg->data);
2923 if (VLOG_IS_WARN_ENABLED()) {
2924 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
2925 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
2928 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
2933 send_error_oh(ofconn, ofp_msg->data, error);
2938 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
2940 struct odp_msg *msg = packet->data;
2941 uint16_t in_port = odp_port_to_ofp_port(msg->port);
2943 struct ofpbuf payload;
2946 /* Handle controller actions. */
2947 if (msg->type == _ODPL_ACTION_NR) {
2948 COVERAGE_INC(ofproto_ctlr_action);
2949 pinsched_send(p->action_sched, in_port, packet,
2950 send_packet_in_action, p);
2954 payload.data = msg + 1;
2955 payload.size = msg->length - sizeof *msg;
2956 flow_extract(&payload, msg->port, &flow);
2958 rule = lookup_valid_rule(p, &flow);
2960 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2961 struct ofport *port = port_array_get(&p->ports, msg->port);
2963 if (port->opp.config & OFPPC_NO_PACKET_IN) {
2964 COVERAGE_INC(ofproto_no_packet_in);
2965 /* XXX install 'drop' flow entry */
2966 ofpbuf_delete(packet);
2970 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
2973 COVERAGE_INC(ofproto_packet_in);
2974 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
2978 if (rule->cr.wc.wildcards) {
2979 rule = rule_create_subrule(p, rule, &flow);
2980 rule_make_actions(p, rule, packet);
2982 if (!rule->may_install) {
2983 /* The rule is not installable, that is, we need to process every
2984 * packet, so process the current packet and set its actions into
2986 rule_make_actions(p, rule, packet);
2988 /* XXX revalidate rule if it needs it */
2992 rule_execute(p, rule, &payload, &flow);
2993 rule_reinstall(p, rule);
2994 ofpbuf_delete(packet);
2998 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3000 struct rule *sub = rule_from_cls_rule(sub_);
3001 struct revalidate_cbdata *cbdata = cbdata_;
3003 if (cbdata->revalidate_all
3004 || (cbdata->revalidate_subrules && sub->super)
3005 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3006 revalidate_rule(cbdata->ofproto, sub);
3011 revalidate_rule(struct ofproto *p, struct rule *rule)
3013 const flow_t *flow = &rule->cr.flow;
3015 COVERAGE_INC(ofproto_revalidate_rule);
3018 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3020 rule_remove(p, rule);
3022 } else if (super != rule->super) {
3023 COVERAGE_INC(ofproto_revalidate_moved);
3024 list_remove(&rule->list);
3025 list_push_back(&super->list, &rule->list);
3026 rule->super = super;
3027 rule->hard_timeout = super->hard_timeout;
3028 rule->idle_timeout = super->idle_timeout;
3029 rule->created = super->created;
3034 rule_update_actions(p, rule);
3038 static struct ofpbuf *
3039 compose_flow_exp(const struct rule *rule, long long int now, uint8_t reason)
3041 struct ofp_flow_expired *ofe;
3044 ofe = make_openflow(sizeof *ofe, OFPT_FLOW_EXPIRED, &buf);
3045 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofe->match);
3046 ofe->priority = htons(rule->cr.priority);
3047 ofe->reason = reason;
3048 ofe->duration = (now - rule->created) / 1000;
3049 ofe->packet_count = rule->packet_count;
3050 ofe->byte_count = rule->byte_count;
3056 send_flow_exp(struct ofproto *p, struct rule *rule,
3057 long long int now, uint8_t reason)
3059 struct ofconn *ofconn;
3060 struct ofconn *prev;
3063 /* We limit the maximum number of queued flow expirations it by accounting
3064 * them under the counter for replies. That works because preventing
3065 * OpenFlow requests from being processed also prevents new flows from
3066 * being added (and expiring). (It also prevents processing OpenFlow
3067 * requests that would not add new flows, so it is imperfect.) */
3070 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3071 if (ofconn->send_flow_exp && rconn_is_connected(ofconn->rconn)) {
3073 queue_tx(ofpbuf_clone(buf), prev, ofconn->reply_counter);
3075 buf = compose_flow_exp(rule, now, reason);
3081 queue_tx(buf, prev, ofconn->reply_counter);
3086 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3088 assert(rule->installed);
3089 assert(!rule->cr.wc.wildcards);
3092 rule_remove(ofproto, rule);
3094 rule_uninstall(ofproto, rule);
3099 expire_rule(struct cls_rule *cls_rule, void *p_)
3101 struct ofproto *p = p_;
3102 struct rule *rule = rule_from_cls_rule(cls_rule);
3103 long long int hard_expire, idle_expire, expire, now;
3105 hard_expire = (rule->hard_timeout
3106 ? rule->created + rule->hard_timeout * 1000
3108 idle_expire = (rule->idle_timeout
3109 && (rule->super || list_is_empty(&rule->list))
3110 ? rule->used + rule->idle_timeout * 1000
3112 expire = MIN(hard_expire, idle_expire);
3113 if (expire == LLONG_MAX) {
3114 if (rule->installed && time_msec() >= rule->used + 5000) {
3115 uninstall_idle_flow(p, rule);
3122 if (rule->installed && now >= rule->used + 5000) {
3123 uninstall_idle_flow(p, rule);
3128 COVERAGE_INC(ofproto_expired);
3129 if (rule->cr.wc.wildcards) {
3130 /* Update stats. (This code will be a no-op if the rule expired
3131 * due to an idle timeout, because in that case the rule has no
3132 * subrules left.) */
3133 struct rule *subrule, *next;
3134 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3135 rule_remove(p, subrule);
3139 send_flow_exp(p, rule, now,
3141 ? OFPER_HARD_TIMEOUT : OFPER_IDLE_TIMEOUT));
3142 rule_remove(p, rule);
3146 update_used(struct ofproto *p)
3148 struct odp_flow *flows;
3153 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
3158 for (i = 0; i < n_flows; i++) {
3159 struct odp_flow *f = &flows[i];
3162 rule = rule_from_cls_rule(
3163 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3164 if (!rule || !rule->installed) {
3165 COVERAGE_INC(ofproto_unexpected_rule);
3166 dpif_flow_del(p->dpif, f);
3170 update_time(rule, &f->stats);
3171 rule_account(p, rule, f->stats.n_bytes);
3177 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3178 const struct ofpbuf *packet, int send_len)
3180 struct ofp_packet_in *opi;
3181 struct ofpbuf payload, *buf;
3182 struct odp_msg *msg;
3185 payload.data = msg + 1;
3186 payload.size = msg->length - sizeof *msg;
3188 send_len = MIN(send_len, payload.size);
3189 buf = ofpbuf_new(sizeof *opi + send_len);
3190 opi = put_openflow_xid(offsetof(struct ofp_packet_in, data),
3191 OFPT_PACKET_IN, 0, buf);
3192 opi->buffer_id = htonl(buffer_id);
3193 opi->total_len = htons(payload.size);
3194 opi->in_port = htons(odp_port_to_ofp_port(msg->port));
3195 opi->reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3196 ofpbuf_put(buf, payload.data, MIN(send_len, payload.size));
3197 update_openflow_length(buf);
3198 rconn_send_with_limit(ofconn->rconn, buf, ofconn->packet_in_counter, 100);
3202 send_packet_in_action(struct ofpbuf *packet, void *p_)
3204 struct ofproto *p = p_;
3205 struct ofconn *ofconn;
3206 struct odp_msg *msg;
3209 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3210 if (ofconn == p->controller || ofconn->miss_send_len) {
3211 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3214 ofpbuf_delete(packet);
3218 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3220 struct ofproto *p = p_;
3221 struct ofconn *ofconn;
3222 struct ofpbuf payload;
3223 struct odp_msg *msg;
3226 payload.data = msg + 1;
3227 payload.size = msg->length - sizeof *msg;
3228 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3229 if (ofconn->miss_send_len) {
3230 uint32_t buffer_id = pktbuf_save(ofconn->pktbuf, &payload,
3232 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3234 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3237 ofpbuf_delete(packet);
3241 pick_datapath_id(const struct ofproto *ofproto)
3243 const struct ofport *port;
3245 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
3247 uint8_t ea[ETH_ADDR_LEN];
3250 error = netdev_get_etheraddr(port->netdev, ea);
3252 return eth_addr_to_uint64(ea);
3254 VLOG_WARN("could not get MAC address for %s (%s)",
3255 netdev_get_name(port->netdev), strerror(error));
3257 return ofproto->fallback_dpid;
3261 pick_fallback_dpid(void)
3263 uint8_t ea[ETH_ADDR_LEN];
3264 eth_addr_random(ea);
3265 ea[0] = 0x00; /* Set Nicira OUI. */
3268 return eth_addr_to_uint64(ea);
3272 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3273 struct odp_actions *actions, tag_type *tags,
3276 struct ofproto *ofproto = ofproto_;
3279 /* Drop frames for reserved multicast addresses. */
3280 if (eth_addr_is_reserved(flow->dl_dst)) {
3284 /* Learn source MAC (but don't try to learn from revalidation). */
3285 if (packet != NULL) {
3286 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3289 /* The log messages here could actually be useful in debugging,
3290 * so keep the rate limit relatively high. */
3291 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3292 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3293 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3294 ofproto_revalidate(ofproto, rev_tag);
3298 /* Determine output port. */
3299 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3301 add_output_group_action(actions, DP_GROUP_FLOOD);
3302 } else if (out_port != flow->in_port) {
3303 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3311 static const struct ofhooks default_ofhooks = {
3313 default_normal_ofhook_cb,