2 * Copyright (c) 2009, 2010 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
22 #include <netinet/in.h>
25 #include "classifier.h"
27 #include "discovery.h"
29 #include "dynamic-string.h"
31 #include "fail-open.h"
33 #include "mac-learning.h"
37 #include "ofp-print.h"
39 #include "openflow/nicira-ext.h"
40 #include "openflow/openflow.h"
41 #include "openflow/openflow-mgmt.h"
42 #include "openvswitch/datapath-protocol.h"
46 #include "poll-loop.h"
47 #include "port-array.h"
57 #include "vconn-ssl.h"
60 #define THIS_MODULE VLM_ofproto
70 TABLEID_CLASSIFIER = 1
74 struct netdev *netdev;
75 struct ofp_phy_port opp; /* In host byte order. */
78 static void ofport_free(struct ofport *);
79 static void hton_ofp_phy_port(struct ofp_phy_port *);
81 static int xlate_actions(const union ofp_action *in, size_t n_in,
82 const flow_t *flow, struct ofproto *ofproto,
83 const struct ofpbuf *packet,
84 struct odp_actions *out, tag_type *tags,
85 bool *may_set_up_flow, uint16_t *nf_output_iface);
90 uint64_t flow_cookie; /* Controller-issued identifier.
91 (Kept in network-byte order.) */
92 uint16_t idle_timeout; /* In seconds from time of last use. */
93 uint16_t hard_timeout; /* In seconds from time of creation. */
94 bool send_flow_removed; /* Send a flow removed message? */
95 long long int used; /* Last-used time (0 if never used). */
96 long long int created; /* Creation time. */
97 uint64_t packet_count; /* Number of packets received. */
98 uint64_t byte_count; /* Number of bytes received. */
99 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
100 tag_type tags; /* Tags (set only by hooks). */
101 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
103 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
104 * exact-match rule (having cr.wc.wildcards of 0) generated from the
105 * wildcard rule 'super'. In this case, 'list' is an element of the
108 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
109 * a list of subrules. A super-rule with no wildcards (where
110 * cr.wc.wildcards is 0) will never have any subrules. */
116 * A subrule has no actions (it uses the super-rule's actions). */
118 union ofp_action *actions;
122 * A super-rule with wildcard fields never has ODP actions (since the
123 * datapath only supports exact-match flows). */
124 bool installed; /* Installed in datapath? */
125 bool may_install; /* True ordinarily; false if actions must
126 * be reassessed for every packet. */
128 union odp_action *odp_actions;
132 rule_is_hidden(const struct rule *rule)
134 /* Subrules are merely an implementation detail, so hide them from the
136 if (rule->super != NULL) {
140 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
141 * (e.g. by in-band control) and are intentionally hidden from the
143 if (rule->cr.priority > UINT16_MAX) {
150 static struct rule *rule_create(struct ofproto *, struct rule *super,
151 const union ofp_action *, size_t n_actions,
152 uint16_t idle_timeout, uint16_t hard_timeout,
153 uint64_t flow_cookie, bool send_flow_removed);
154 static void rule_free(struct rule *);
155 static void rule_destroy(struct ofproto *, struct rule *);
156 static struct rule *rule_from_cls_rule(const struct cls_rule *);
157 static void rule_insert(struct ofproto *, struct rule *,
158 struct ofpbuf *packet, uint16_t in_port);
159 static void rule_remove(struct ofproto *, struct rule *);
160 static bool rule_make_actions(struct ofproto *, struct rule *,
161 const struct ofpbuf *packet);
162 static void rule_install(struct ofproto *, struct rule *,
163 struct rule *displaced_rule);
164 static void rule_uninstall(struct ofproto *, struct rule *);
165 static void rule_post_uninstall(struct ofproto *, struct rule *);
166 static void send_flow_removed(struct ofproto *p, struct rule *rule,
167 long long int now, uint8_t reason);
172 struct pktbuf *pktbuf;
175 struct rconn_packet_counter *packet_in_counter;
177 /* Number of OpenFlow messages queued as replies to OpenFlow requests, and
178 * the maximum number before we stop reading OpenFlow requests. */
179 #define OFCONN_REPLY_MAX 100
180 struct rconn_packet_counter *reply_counter;
183 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *);
184 static void ofconn_destroy(struct ofconn *, struct ofproto *);
185 static void ofconn_run(struct ofconn *, struct ofproto *);
186 static void ofconn_wait(struct ofconn *);
187 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
188 struct rconn_packet_counter *counter);
192 uint64_t datapath_id; /* Datapath ID. */
193 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
194 uint64_t mgmt_id; /* Management channel identifier. */
195 char *manufacturer; /* Manufacturer. */
196 char *hardware; /* Hardware. */
197 char *software; /* Software version. */
198 char *serial; /* Serial number. */
199 char *dp_desc; /* Datapath description. */
203 struct netdev_monitor *netdev_monitor;
204 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
206 struct shash port_by_name;
210 struct switch_status *switch_status;
211 struct status_category *ss_cat;
212 struct in_band *in_band;
213 struct discovery *discovery;
214 struct fail_open *fail_open;
215 struct pinsched *miss_sched, *action_sched;
216 struct executer *executer;
217 struct netflow *netflow;
220 struct classifier cls;
221 bool need_revalidate;
222 long long int next_expiration;
223 struct tag_set revalidate_set;
225 /* OpenFlow connections. */
226 struct list all_conns;
227 struct ofconn *controller;
228 struct pvconn **listeners;
230 struct pvconn **snoops;
233 /* Hooks for ovs-vswitchd. */
234 const struct ofhooks *ofhooks;
237 /* Used by default ofhooks. */
238 struct mac_learning *ml;
241 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
243 static const struct ofhooks default_ofhooks;
245 static uint64_t pick_datapath_id(const struct ofproto *);
246 static uint64_t pick_fallback_dpid(void);
247 static void send_packet_in_miss(struct ofpbuf *, void *ofproto);
248 static void send_packet_in_action(struct ofpbuf *, void *ofproto);
249 static void update_used(struct ofproto *);
250 static void update_stats(struct ofproto *, struct rule *,
251 const struct odp_flow_stats *);
252 static void expire_rule(struct cls_rule *, void *ofproto);
253 static void active_timeout(struct ofproto *ofproto, struct rule *rule);
254 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
255 static void revalidate_cb(struct cls_rule *rule_, void *p_);
257 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
259 static void handle_openflow(struct ofconn *, struct ofproto *,
262 static void refresh_port_group(struct ofproto *, unsigned int group);
263 static void update_port(struct ofproto *, const char *devname);
264 static int init_ports(struct ofproto *);
265 static void reinit_ports(struct ofproto *);
268 ofproto_create(const char *datapath, const struct ofhooks *ofhooks, void *aux,
269 struct ofproto **ofprotop)
271 struct odp_stats stats;
278 /* Connect to datapath and start listening for messages. */
279 error = dpif_open(datapath, &dpif);
281 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
284 error = dpif_get_dp_stats(dpif, &stats);
286 VLOG_ERR("failed to obtain stats for datapath %s: %s",
287 datapath, strerror(error));
291 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION);
293 VLOG_ERR("failed to listen on datapath %s: %s",
294 datapath, strerror(error));
298 dpif_flow_flush(dpif);
299 dpif_recv_purge(dpif);
301 /* Initialize settings. */
302 p = xcalloc(1, sizeof *p);
303 p->fallback_dpid = pick_fallback_dpid();
304 p->datapath_id = p->fallback_dpid;
305 p->manufacturer = xstrdup("Nicira Networks, Inc.");
306 p->hardware = xstrdup("Reference Implementation");
307 p->software = xstrdup(VERSION BUILDNR);
308 p->serial = xstrdup("None");
309 p->dp_desc = xstrdup("None");
311 /* Initialize datapath. */
313 p->netdev_monitor = netdev_monitor_create();
314 port_array_init(&p->ports);
315 shash_init(&p->port_by_name);
316 p->max_ports = stats.max_ports;
318 /* Initialize submodules. */
319 p->switch_status = switch_status_create(p);
323 p->miss_sched = p->action_sched = NULL;
327 /* Initialize flow table. */
328 classifier_init(&p->cls);
329 p->need_revalidate = false;
330 p->next_expiration = time_msec() + 1000;
331 tag_set_init(&p->revalidate_set);
333 /* Initialize OpenFlow connections. */
334 list_init(&p->all_conns);
335 p->controller = ofconn_create(p, rconn_create(5, 8));
336 p->controller->pktbuf = pktbuf_create();
337 p->controller->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
343 /* Initialize hooks. */
345 p->ofhooks = ofhooks;
349 p->ofhooks = &default_ofhooks;
351 p->ml = mac_learning_create();
354 /* Register switch status category. */
355 p->ss_cat = switch_status_register(p->switch_status, "remote",
356 rconn_status_cb, p->controller->rconn);
359 error = init_ports(p);
365 /* Pick final datapath ID. */
366 p->datapath_id = pick_datapath_id(p);
367 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
374 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
376 uint64_t old_dpid = p->datapath_id;
377 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
378 if (p->datapath_id != old_dpid) {
379 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
380 rconn_reconnect(p->controller->rconn);
385 ofproto_set_mgmt_id(struct ofproto *p, uint64_t mgmt_id)
387 p->mgmt_id = mgmt_id;
391 ofproto_set_probe_interval(struct ofproto *p, int probe_interval)
393 probe_interval = probe_interval ? MAX(probe_interval, 5) : 0;
394 rconn_set_probe_interval(p->controller->rconn, probe_interval);
396 int trigger_duration = probe_interval ? probe_interval * 3 : 15;
397 fail_open_set_trigger_duration(p->fail_open, trigger_duration);
402 ofproto_set_max_backoff(struct ofproto *p, int max_backoff)
404 rconn_set_max_backoff(p->controller->rconn, max_backoff);
408 ofproto_set_desc(struct ofproto *p,
409 const char *manufacturer, const char *hardware,
410 const char *software, const char *serial,
414 free(p->manufacturer);
415 p->manufacturer = xstrdup(manufacturer);
419 p->hardware = xstrdup(hardware);
423 p->software = xstrdup(software);
427 p->serial = xstrdup(serial);
431 p->dp_desc = xstrdup(dp_desc);
436 ofproto_set_in_band(struct ofproto *p, bool in_band)
438 if (in_band != (p->in_band != NULL)) {
440 return in_band_create(p, p->dpif, p->switch_status,
441 p->controller->rconn, &p->in_band);
443 ofproto_set_discovery(p, false, NULL, true);
444 in_band_destroy(p->in_band);
447 rconn_reconnect(p->controller->rconn);
453 ofproto_set_discovery(struct ofproto *p, bool discovery,
454 const char *re, bool update_resolv_conf)
456 if (discovery != (p->discovery != NULL)) {
458 int error = ofproto_set_in_band(p, true);
462 error = discovery_create(re, update_resolv_conf,
463 p->dpif, p->switch_status,
469 discovery_destroy(p->discovery);
472 rconn_disconnect(p->controller->rconn);
473 } else if (discovery) {
474 discovery_set_update_resolv_conf(p->discovery, update_resolv_conf);
475 return discovery_set_accept_controller_re(p->discovery, re);
481 ofproto_set_controller(struct ofproto *ofproto, const char *controller)
483 if (ofproto->discovery) {
485 } else if (controller) {
486 if (strcmp(rconn_get_name(ofproto->controller->rconn), controller)) {
487 return rconn_connect(ofproto->controller->rconn, controller);
492 rconn_disconnect(ofproto->controller->rconn);
498 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
499 const struct svec *svec)
501 struct pvconn **pvconns = *pvconnsp;
502 size_t n_pvconns = *n_pvconnsp;
506 for (i = 0; i < n_pvconns; i++) {
507 pvconn_close(pvconns[i]);
511 pvconns = xmalloc(svec->n * sizeof *pvconns);
513 for (i = 0; i < svec->n; i++) {
514 const char *name = svec->names[i];
515 struct pvconn *pvconn;
518 error = pvconn_open(name, &pvconn);
520 pvconns[n_pvconns++] = pvconn;
522 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
530 *n_pvconnsp = n_pvconns;
536 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
538 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
542 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
544 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
548 ofproto_set_netflow(struct ofproto *ofproto,
549 const struct netflow_options *nf_options)
551 if (nf_options->collectors.n) {
552 if (!ofproto->netflow) {
553 ofproto->netflow = netflow_create();
555 return netflow_set_options(ofproto->netflow, nf_options);
557 netflow_destroy(ofproto->netflow);
558 ofproto->netflow = NULL;
564 ofproto_set_failure(struct ofproto *ofproto, bool fail_open)
567 struct rconn *rconn = ofproto->controller->rconn;
568 int trigger_duration = rconn_get_probe_interval(rconn) * 3;
569 if (!ofproto->fail_open) {
570 ofproto->fail_open = fail_open_create(ofproto, trigger_duration,
571 ofproto->switch_status,
574 fail_open_set_trigger_duration(ofproto->fail_open,
578 fail_open_destroy(ofproto->fail_open);
579 ofproto->fail_open = NULL;
584 ofproto_set_rate_limit(struct ofproto *ofproto,
585 int rate_limit, int burst_limit)
587 if (rate_limit > 0) {
588 if (!ofproto->miss_sched) {
589 ofproto->miss_sched = pinsched_create(rate_limit, burst_limit,
590 ofproto->switch_status);
591 ofproto->action_sched = pinsched_create(rate_limit, burst_limit,
594 pinsched_set_limits(ofproto->miss_sched, rate_limit, burst_limit);
595 pinsched_set_limits(ofproto->action_sched,
596 rate_limit, burst_limit);
599 pinsched_destroy(ofproto->miss_sched);
600 ofproto->miss_sched = NULL;
601 pinsched_destroy(ofproto->action_sched);
602 ofproto->action_sched = NULL;
607 ofproto_set_stp(struct ofproto *ofproto UNUSED, bool enable_stp)
611 VLOG_WARN("STP is not yet implemented");
619 ofproto_set_remote_execution(struct ofproto *ofproto, const char *command_acl,
620 const char *command_dir)
623 if (!ofproto->executer) {
624 return executer_create(command_acl, command_dir,
627 executer_set_acl(ofproto->executer, command_acl, command_dir);
630 executer_destroy(ofproto->executer);
631 ofproto->executer = NULL;
637 ofproto_get_datapath_id(const struct ofproto *ofproto)
639 return ofproto->datapath_id;
643 ofproto_get_mgmt_id(const struct ofproto *ofproto)
645 return ofproto->mgmt_id;
649 ofproto_get_probe_interval(const struct ofproto *ofproto)
651 return rconn_get_probe_interval(ofproto->controller->rconn);
655 ofproto_get_max_backoff(const struct ofproto *ofproto)
657 return rconn_get_max_backoff(ofproto->controller->rconn);
661 ofproto_get_in_band(const struct ofproto *ofproto)
663 return ofproto->in_band != NULL;
667 ofproto_get_discovery(const struct ofproto *ofproto)
669 return ofproto->discovery != NULL;
673 ofproto_get_controller(const struct ofproto *ofproto)
675 return rconn_get_name(ofproto->controller->rconn);
679 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
683 for (i = 0; i < ofproto->n_listeners; i++) {
684 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
689 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
693 for (i = 0; i < ofproto->n_snoops; i++) {
694 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
699 ofproto_destroy(struct ofproto *p)
701 struct ofconn *ofconn, *next_ofconn;
702 struct ofport *ofport;
703 unsigned int port_no;
710 ofproto_flush_flows(p);
711 classifier_destroy(&p->cls);
713 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
715 ofconn_destroy(ofconn, p);
719 netdev_monitor_destroy(p->netdev_monitor);
720 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
723 shash_destroy(&p->port_by_name);
725 switch_status_destroy(p->switch_status);
726 in_band_destroy(p->in_band);
727 discovery_destroy(p->discovery);
728 fail_open_destroy(p->fail_open);
729 pinsched_destroy(p->miss_sched);
730 pinsched_destroy(p->action_sched);
731 executer_destroy(p->executer);
732 netflow_destroy(p->netflow);
734 switch_status_unregister(p->ss_cat);
736 for (i = 0; i < p->n_listeners; i++) {
737 pvconn_close(p->listeners[i]);
741 for (i = 0; i < p->n_snoops; i++) {
742 pvconn_close(p->snoops[i]);
746 mac_learning_destroy(p->ml);
748 free(p->manufacturer);
758 ofproto_run(struct ofproto *p)
760 int error = ofproto_run1(p);
762 error = ofproto_run2(p, false);
768 process_port_change(struct ofproto *ofproto, int error, char *devname)
770 if (error == ENOBUFS) {
771 reinit_ports(ofproto);
773 update_port(ofproto, devname);
779 ofproto_run1(struct ofproto *p)
781 struct ofconn *ofconn, *next_ofconn;
786 for (i = 0; i < 50; i++) {
790 error = dpif_recv(p->dpif, &buf);
792 if (error == ENODEV) {
793 /* Someone destroyed the datapath behind our back. The caller
794 * better destroy us and give up, because we're just going to
795 * spin from here on out. */
796 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
797 VLOG_ERR_RL(&rl, "%s: datapath was destroyed externally",
804 handle_odp_msg(p, buf);
807 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
808 process_port_change(p, error, devname);
810 while ((error = netdev_monitor_poll(p->netdev_monitor,
811 &devname)) != EAGAIN) {
812 process_port_change(p, error, devname);
816 in_band_run(p->in_band);
819 char *controller_name;
820 if (rconn_is_connectivity_questionable(p->controller->rconn)) {
821 discovery_question_connectivity(p->discovery);
823 if (discovery_run(p->discovery, &controller_name)) {
824 if (controller_name) {
825 rconn_connect(p->controller->rconn, controller_name);
827 rconn_disconnect(p->controller->rconn);
831 pinsched_run(p->miss_sched, send_packet_in_miss, p);
832 pinsched_run(p->action_sched, send_packet_in_action, p);
834 executer_run(p->executer);
837 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
839 ofconn_run(ofconn, p);
842 /* Fail-open maintenance. Do this after processing the ofconns since
843 * fail-open checks the status of the controller rconn. */
845 fail_open_run(p->fail_open);
848 for (i = 0; i < p->n_listeners; i++) {
852 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
854 ofconn_create(p, rconn_new_from_vconn("passive", vconn));
855 } else if (retval != EAGAIN) {
856 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
860 for (i = 0; i < p->n_snoops; i++) {
864 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
866 rconn_add_monitor(p->controller->rconn, vconn);
867 } else if (retval != EAGAIN) {
868 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
872 if (time_msec() >= p->next_expiration) {
873 COVERAGE_INC(ofproto_expiration);
874 p->next_expiration = time_msec() + 1000;
877 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
879 /* Let the hook know that we're at a stable point: all outstanding data
880 * in existing flows has been accounted to the account_cb. Thus, the
881 * hook can now reasonably do operations that depend on having accurate
882 * flow volume accounting (currently, that's just bond rebalancing). */
883 if (p->ofhooks->account_checkpoint_cb) {
884 p->ofhooks->account_checkpoint_cb(p->aux);
889 netflow_run(p->netflow);
895 struct revalidate_cbdata {
896 struct ofproto *ofproto;
897 bool revalidate_all; /* Revalidate all exact-match rules? */
898 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
899 struct tag_set revalidate_set; /* Set of tags to revalidate. */
903 ofproto_run2(struct ofproto *p, bool revalidate_all)
905 if (p->need_revalidate || revalidate_all
906 || !tag_set_is_empty(&p->revalidate_set)) {
907 struct revalidate_cbdata cbdata;
909 cbdata.revalidate_all = revalidate_all;
910 cbdata.revalidate_subrules = p->need_revalidate;
911 cbdata.revalidate_set = p->revalidate_set;
912 tag_set_init(&p->revalidate_set);
913 COVERAGE_INC(ofproto_revalidate);
914 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
915 p->need_revalidate = false;
922 ofproto_wait(struct ofproto *p)
924 struct ofconn *ofconn;
927 dpif_recv_wait(p->dpif);
928 dpif_port_poll_wait(p->dpif);
929 netdev_monitor_poll_wait(p->netdev_monitor);
930 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
934 in_band_wait(p->in_band);
937 discovery_wait(p->discovery);
940 fail_open_wait(p->fail_open);
942 pinsched_wait(p->miss_sched);
943 pinsched_wait(p->action_sched);
945 executer_wait(p->executer);
947 if (!tag_set_is_empty(&p->revalidate_set)) {
948 poll_immediate_wake();
950 if (p->need_revalidate) {
951 /* Shouldn't happen, but if it does just go around again. */
952 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
953 poll_immediate_wake();
954 } else if (p->next_expiration != LLONG_MAX) {
955 poll_timer_wait(p->next_expiration - time_msec());
957 for (i = 0; i < p->n_listeners; i++) {
958 pvconn_wait(p->listeners[i]);
960 for (i = 0; i < p->n_snoops; i++) {
961 pvconn_wait(p->snoops[i]);
966 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
968 tag_set_add(&ofproto->revalidate_set, tag);
972 ofproto_get_revalidate_set(struct ofproto *ofproto)
974 return &ofproto->revalidate_set;
978 ofproto_is_alive(const struct ofproto *p)
980 return p->discovery || rconn_is_alive(p->controller->rconn);
984 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
985 const union ofp_action *actions, size_t n_actions,
986 const struct ofpbuf *packet)
988 struct odp_actions odp_actions;
991 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
997 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
999 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
1000 odp_actions.n_actions, packet);
1005 ofproto_add_flow(struct ofproto *p,
1006 const flow_t *flow, uint32_t wildcards, unsigned int priority,
1007 const union ofp_action *actions, size_t n_actions,
1011 rule = rule_create(p, NULL, actions, n_actions,
1012 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1014 cls_rule_from_flow(&rule->cr, flow, wildcards, priority);
1015 rule_insert(p, rule, NULL, 0);
1019 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1020 uint32_t wildcards, unsigned int priority)
1024 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1028 rule_remove(ofproto, rule);
1033 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1035 struct rule *rule = rule_from_cls_rule(rule_);
1036 struct ofproto *ofproto = ofproto_;
1038 /* Mark the flow as not installed, even though it might really be
1039 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1040 * There is no point in uninstalling it individually since we are about to
1041 * blow away all the flows with dpif_flow_flush(). */
1042 rule->installed = false;
1044 rule_remove(ofproto, rule);
1048 ofproto_flush_flows(struct ofproto *ofproto)
1050 COVERAGE_INC(ofproto_flush);
1051 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1052 dpif_flow_flush(ofproto->dpif);
1053 if (ofproto->in_band) {
1054 in_band_flushed(ofproto->in_band);
1056 if (ofproto->fail_open) {
1057 fail_open_flushed(ofproto->fail_open);
1062 reinit_ports(struct ofproto *p)
1064 struct svec devnames;
1065 struct ofport *ofport;
1066 unsigned int port_no;
1067 struct odp_port *odp_ports;
1071 svec_init(&devnames);
1072 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1073 svec_add (&devnames, (char *) ofport->opp.name);
1075 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1076 for (i = 0; i < n_odp_ports; i++) {
1077 svec_add (&devnames, odp_ports[i].devname);
1081 svec_sort_unique(&devnames);
1082 for (i = 0; i < devnames.n; i++) {
1083 update_port(p, devnames.names[i]);
1085 svec_destroy(&devnames);
1089 refresh_port_group(struct ofproto *p, unsigned int group)
1093 struct ofport *port;
1094 unsigned int port_no;
1096 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1098 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1100 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1101 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1102 ports[n_ports++] = port_no;
1105 dpif_port_group_set(p->dpif, group, ports, n_ports);
1110 refresh_port_groups(struct ofproto *p)
1112 refresh_port_group(p, DP_GROUP_FLOOD);
1113 refresh_port_group(p, DP_GROUP_ALL);
1116 static struct ofport *
1117 make_ofport(const struct odp_port *odp_port)
1119 enum netdev_flags flags;
1120 struct ofport *ofport;
1121 struct netdev *netdev;
1125 error = netdev_open(odp_port->devname, NETDEV_ETH_TYPE_NONE, &netdev);
1127 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1128 "cannot be opened (%s)",
1129 odp_port->devname, odp_port->port,
1130 odp_port->devname, strerror(error));
1134 ofport = xmalloc(sizeof *ofport);
1135 ofport->netdev = netdev;
1136 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1137 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1138 memcpy(ofport->opp.name, odp_port->devname,
1139 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1140 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1142 netdev_get_flags(netdev, &flags);
1143 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1145 netdev_get_carrier(netdev, &carrier);
1146 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1148 netdev_get_features(netdev,
1149 &ofport->opp.curr, &ofport->opp.advertised,
1150 &ofport->opp.supported, &ofport->opp.peer);
1155 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1157 if (port_array_get(&p->ports, odp_port->port)) {
1158 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1161 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1162 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1171 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1173 const struct ofp_phy_port *a = &a_->opp;
1174 const struct ofp_phy_port *b = &b_->opp;
1176 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1177 return (a->port_no == b->port_no
1178 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1179 && !strcmp((char *) a->name, (char *) b->name)
1180 && a->state == b->state
1181 && a->config == b->config
1182 && a->curr == b->curr
1183 && a->advertised == b->advertised
1184 && a->supported == b->supported
1185 && a->peer == b->peer);
1189 send_port_status(struct ofproto *p, const struct ofport *ofport,
1192 /* XXX Should limit the number of queued port status change messages. */
1193 struct ofconn *ofconn;
1194 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1195 struct ofp_port_status *ops;
1198 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1199 ops->reason = reason;
1200 ops->desc = ofport->opp;
1201 hton_ofp_phy_port(&ops->desc);
1202 queue_tx(b, ofconn, NULL);
1204 if (p->ofhooks->port_changed_cb) {
1205 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1210 ofport_install(struct ofproto *p, struct ofport *ofport)
1212 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1213 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no),
1215 shash_add(&p->port_by_name, (char *) ofport->opp.name, ofport);
1219 ofport_remove(struct ofproto *p, struct ofport *ofport)
1221 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1222 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no), NULL);
1223 shash_delete(&p->port_by_name,
1224 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1228 ofport_free(struct ofport *ofport)
1231 netdev_close(ofport->netdev);
1237 update_port(struct ofproto *p, const char *devname)
1239 struct odp_port odp_port;
1240 struct ofport *old_ofport;
1241 struct ofport *new_ofport;
1244 COVERAGE_INC(ofproto_update_port);
1246 /* Query the datapath for port information. */
1247 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1249 /* Find the old ofport. */
1250 old_ofport = shash_find_data(&p->port_by_name, devname);
1253 /* There's no port named 'devname' but there might be a port with
1254 * the same port number. This could happen if a port is deleted
1255 * and then a new one added in its place very quickly, or if a port
1256 * is renamed. In the former case we want to send an OFPPR_DELETE
1257 * and an OFPPR_ADD, and in the latter case we want to send a
1258 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1259 * the old port's ifindex against the new port, or perhaps less
1260 * reliably but more portably by comparing the old port's MAC
1261 * against the new port's MAC. However, this code isn't that smart
1262 * and always sends an OFPPR_MODIFY (XXX). */
1263 old_ofport = port_array_get(&p->ports, odp_port.port);
1265 } else if (error != ENOENT && error != ENODEV) {
1266 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1267 "%s", strerror(error));
1271 /* Create a new ofport. */
1272 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1274 /* Eliminate a few pathological cases. */
1275 if (!old_ofport && !new_ofport) {
1277 } else if (old_ofport && new_ofport) {
1278 /* Most of the 'config' bits are OpenFlow soft state, but
1279 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1280 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1281 * leaves the other bits 0.) */
1282 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1284 if (ofport_equal(old_ofport, new_ofport)) {
1285 /* False alarm--no change. */
1286 ofport_free(new_ofport);
1291 /* Now deal with the normal cases. */
1293 ofport_remove(p, old_ofport);
1296 ofport_install(p, new_ofport);
1298 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1299 (!old_ofport ? OFPPR_ADD
1300 : !new_ofport ? OFPPR_DELETE
1302 ofport_free(old_ofport);
1304 /* Update port groups. */
1305 refresh_port_groups(p);
1309 init_ports(struct ofproto *p)
1311 struct odp_port *ports;
1316 error = dpif_port_list(p->dpif, &ports, &n_ports);
1321 for (i = 0; i < n_ports; i++) {
1322 const struct odp_port *odp_port = &ports[i];
1323 if (!ofport_conflicts(p, odp_port)) {
1324 struct ofport *ofport = make_ofport(odp_port);
1326 ofport_install(p, ofport);
1331 refresh_port_groups(p);
1335 static struct ofconn *
1336 ofconn_create(struct ofproto *p, struct rconn *rconn)
1338 struct ofconn *ofconn = xmalloc(sizeof *ofconn);
1339 list_push_back(&p->all_conns, &ofconn->node);
1340 ofconn->rconn = rconn;
1341 ofconn->pktbuf = NULL;
1342 ofconn->miss_send_len = 0;
1343 ofconn->packet_in_counter = rconn_packet_counter_create ();
1344 ofconn->reply_counter = rconn_packet_counter_create ();
1349 ofconn_destroy(struct ofconn *ofconn, struct ofproto *p)
1352 executer_rconn_closing(p->executer, ofconn->rconn);
1355 list_remove(&ofconn->node);
1356 rconn_destroy(ofconn->rconn);
1357 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1358 rconn_packet_counter_destroy(ofconn->reply_counter);
1359 pktbuf_destroy(ofconn->pktbuf);
1364 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1368 rconn_run(ofconn->rconn);
1370 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1371 /* Limit the number of iterations to prevent other tasks from
1373 for (iteration = 0; iteration < 50; iteration++) {
1374 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1379 fail_open_maybe_recover(p->fail_open);
1381 handle_openflow(ofconn, p, of_msg);
1382 ofpbuf_delete(of_msg);
1386 if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
1387 ofconn_destroy(ofconn, p);
1392 ofconn_wait(struct ofconn *ofconn)
1394 rconn_run_wait(ofconn->rconn);
1395 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1396 rconn_recv_wait(ofconn->rconn);
1398 COVERAGE_INC(ofproto_ofconn_stuck);
1402 /* Caller is responsible for initializing the 'cr' member of the returned
1404 static struct rule *
1405 rule_create(struct ofproto *ofproto, struct rule *super,
1406 const union ofp_action *actions, size_t n_actions,
1407 uint16_t idle_timeout, uint16_t hard_timeout,
1408 uint64_t flow_cookie, bool send_flow_removed)
1410 struct rule *rule = xcalloc(1, sizeof *rule);
1411 rule->idle_timeout = idle_timeout;
1412 rule->hard_timeout = hard_timeout;
1413 rule->flow_cookie = flow_cookie;
1414 rule->used = rule->created = time_msec();
1415 rule->send_flow_removed = send_flow_removed;
1416 rule->super = super;
1418 list_push_back(&super->list, &rule->list);
1420 list_init(&rule->list);
1422 rule->n_actions = n_actions;
1423 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1424 netflow_flow_clear(&rule->nf_flow);
1425 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1430 static struct rule *
1431 rule_from_cls_rule(const struct cls_rule *cls_rule)
1433 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1437 rule_free(struct rule *rule)
1439 free(rule->actions);
1440 free(rule->odp_actions);
1444 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1445 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1446 * through all of its subrules and revalidates them, destroying any that no
1447 * longer has a super-rule (which is probably all of them).
1449 * Before calling this function, the caller must make have removed 'rule' from
1450 * the classifier. If 'rule' is an exact-match rule, the caller is also
1451 * responsible for ensuring that it has been uninstalled from the datapath. */
1453 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1456 struct rule *subrule, *next;
1457 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1458 revalidate_rule(ofproto, subrule);
1461 list_remove(&rule->list);
1467 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1469 const union ofp_action *oa;
1470 struct actions_iterator i;
1472 if (out_port == htons(OFPP_NONE)) {
1475 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1476 oa = actions_next(&i)) {
1477 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1484 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1485 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1487 * The flow that 'packet' actually contains does not need to actually match
1488 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1489 * the packet and byte counters for 'rule' will be credited for the packet sent
1490 * out whether or not the packet actually matches 'rule'.
1492 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1493 * the caller must already have accurately composed ODP actions for it given
1494 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1495 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1496 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1497 * actions and apply them to 'packet'. */
1499 rule_execute(struct ofproto *ofproto, struct rule *rule,
1500 struct ofpbuf *packet, const flow_t *flow)
1502 const union odp_action *actions;
1504 struct odp_actions a;
1506 /* Grab or compose the ODP actions.
1508 * The special case for an exact-match 'rule' where 'flow' is not the
1509 * rule's flow is important to avoid, e.g., sending a packet out its input
1510 * port simply because the ODP actions were composed for the wrong
1512 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1513 struct rule *super = rule->super ? rule->super : rule;
1514 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1515 packet, &a, NULL, 0, NULL)) {
1518 actions = a.actions;
1519 n_actions = a.n_actions;
1521 actions = rule->odp_actions;
1522 n_actions = rule->n_odp_actions;
1525 /* Execute the ODP actions. */
1526 if (!dpif_execute(ofproto->dpif, flow->in_port,
1527 actions, n_actions, packet)) {
1528 struct odp_flow_stats stats;
1529 flow_extract_stats(flow, packet, &stats);
1530 update_stats(ofproto, rule, &stats);
1531 rule->used = time_msec();
1532 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1537 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1540 struct rule *displaced_rule;
1542 /* Insert the rule in the classifier. */
1543 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1544 if (!rule->cr.wc.wildcards) {
1545 rule_make_actions(p, rule, packet);
1548 /* Send the packet and credit it to the rule. */
1551 flow_extract(packet, in_port, &flow);
1552 rule_execute(p, rule, packet, &flow);
1555 /* Install the rule in the datapath only after sending the packet, to
1556 * avoid packet reordering. */
1557 if (rule->cr.wc.wildcards) {
1558 COVERAGE_INC(ofproto_add_wc_flow);
1559 p->need_revalidate = true;
1561 rule_install(p, rule, displaced_rule);
1564 /* Free the rule that was displaced, if any. */
1565 if (displaced_rule) {
1566 rule_destroy(p, displaced_rule);
1570 static struct rule *
1571 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1574 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1575 rule->idle_timeout, rule->hard_timeout,
1577 COVERAGE_INC(ofproto_subrule_create);
1578 cls_rule_from_flow(&subrule->cr, flow, 0,
1579 (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1580 : rule->cr.priority));
1581 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1587 rule_remove(struct ofproto *ofproto, struct rule *rule)
1589 if (rule->cr.wc.wildcards) {
1590 COVERAGE_INC(ofproto_del_wc_flow);
1591 ofproto->need_revalidate = true;
1593 rule_uninstall(ofproto, rule);
1595 classifier_remove(&ofproto->cls, &rule->cr);
1596 rule_destroy(ofproto, rule);
1599 /* Returns true if the actions changed, false otherwise. */
1601 rule_make_actions(struct ofproto *p, struct rule *rule,
1602 const struct ofpbuf *packet)
1604 const struct rule *super;
1605 struct odp_actions a;
1608 assert(!rule->cr.wc.wildcards);
1610 super = rule->super ? rule->super : rule;
1612 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1613 packet, &a, &rule->tags, &rule->may_install,
1614 &rule->nf_flow.output_iface);
1616 actions_len = a.n_actions * sizeof *a.actions;
1617 if (rule->n_odp_actions != a.n_actions
1618 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1619 COVERAGE_INC(ofproto_odp_unchanged);
1620 free(rule->odp_actions);
1621 rule->n_odp_actions = a.n_actions;
1622 rule->odp_actions = xmemdup(a.actions, actions_len);
1630 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1631 struct odp_flow_put *put)
1633 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1634 put->flow.key = rule->cr.flow;
1635 put->flow.actions = rule->odp_actions;
1636 put->flow.n_actions = rule->n_odp_actions;
1638 return dpif_flow_put(ofproto->dpif, put);
1642 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1644 assert(!rule->cr.wc.wildcards);
1646 if (rule->may_install) {
1647 struct odp_flow_put put;
1648 if (!do_put_flow(p, rule,
1649 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1651 rule->installed = true;
1652 if (displaced_rule) {
1653 update_stats(p, displaced_rule, &put.flow.stats);
1654 rule_post_uninstall(p, displaced_rule);
1657 } else if (displaced_rule) {
1658 rule_uninstall(p, displaced_rule);
1663 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1665 if (rule->installed) {
1666 struct odp_flow_put put;
1667 COVERAGE_INC(ofproto_dp_missed);
1668 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1670 rule_install(ofproto, rule, NULL);
1675 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1677 bool actions_changed;
1678 uint16_t new_out_iface, old_out_iface;
1680 old_out_iface = rule->nf_flow.output_iface;
1681 actions_changed = rule_make_actions(ofproto, rule, NULL);
1683 if (rule->may_install) {
1684 if (rule->installed) {
1685 if (actions_changed) {
1686 struct odp_flow_put put;
1687 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
1688 | ODPPF_ZERO_STATS, &put);
1689 update_stats(ofproto, rule, &put.flow.stats);
1691 /* Temporarily set the old output iface so that NetFlow
1692 * messages have the correct output interface for the old
1694 new_out_iface = rule->nf_flow.output_iface;
1695 rule->nf_flow.output_iface = old_out_iface;
1696 rule_post_uninstall(ofproto, rule);
1697 rule->nf_flow.output_iface = new_out_iface;
1700 rule_install(ofproto, rule, NULL);
1703 rule_uninstall(ofproto, rule);
1708 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1710 uint64_t total_bytes = rule->byte_count + extra_bytes;
1712 if (ofproto->ofhooks->account_flow_cb
1713 && total_bytes > rule->accounted_bytes)
1715 ofproto->ofhooks->account_flow_cb(
1716 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1717 total_bytes - rule->accounted_bytes, ofproto->aux);
1718 rule->accounted_bytes = total_bytes;
1723 rule_uninstall(struct ofproto *p, struct rule *rule)
1725 assert(!rule->cr.wc.wildcards);
1726 if (rule->installed) {
1727 struct odp_flow odp_flow;
1729 odp_flow.key = rule->cr.flow;
1730 odp_flow.actions = NULL;
1731 odp_flow.n_actions = 0;
1732 if (!dpif_flow_del(p->dpif, &odp_flow)) {
1733 update_stats(p, rule, &odp_flow.stats);
1735 rule->installed = false;
1737 rule_post_uninstall(p, rule);
1742 is_controller_rule(struct rule *rule)
1744 /* If the only action is send to the controller then don't report
1745 * NetFlow expiration messages since it is just part of the control
1746 * logic for the network and not real traffic. */
1748 if (rule && rule->super) {
1749 struct rule *super = rule->super;
1751 return super->n_actions == 1 &&
1752 super->actions[0].type == htons(OFPAT_OUTPUT) &&
1753 super->actions[0].output.port == htons(OFPP_CONTROLLER);
1760 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1762 struct rule *super = rule->super;
1764 rule_account(ofproto, rule, 0);
1766 if (ofproto->netflow && !is_controller_rule(rule)) {
1767 struct ofexpired expired;
1768 expired.flow = rule->cr.flow;
1769 expired.packet_count = rule->packet_count;
1770 expired.byte_count = rule->byte_count;
1771 expired.used = rule->used;
1772 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
1775 super->packet_count += rule->packet_count;
1776 super->byte_count += rule->byte_count;
1778 /* Reset counters to prevent double counting if the rule ever gets
1780 rule->packet_count = 0;
1781 rule->byte_count = 0;
1782 rule->accounted_bytes = 0;
1784 netflow_flow_clear(&rule->nf_flow);
1789 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1790 struct rconn_packet_counter *counter)
1792 update_openflow_length(msg);
1793 if (rconn_send(ofconn->rconn, msg, counter)) {
1799 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1800 int error, const void *data, size_t len)
1803 struct ofp_error_msg *oem;
1805 if (!(error >> 16)) {
1806 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1811 COVERAGE_INC(ofproto_error);
1812 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1813 oh ? oh->xid : 0, &buf);
1814 oem->type = htons((unsigned int) error >> 16);
1815 oem->code = htons(error & 0xffff);
1816 memcpy(oem->data, data, len);
1817 queue_tx(buf, ofconn, ofconn->reply_counter);
1821 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1824 size_t oh_length = ntohs(oh->length);
1825 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1829 hton_ofp_phy_port(struct ofp_phy_port *opp)
1831 opp->port_no = htons(opp->port_no);
1832 opp->config = htonl(opp->config);
1833 opp->state = htonl(opp->state);
1834 opp->curr = htonl(opp->curr);
1835 opp->advertised = htonl(opp->advertised);
1836 opp->supported = htonl(opp->supported);
1837 opp->peer = htonl(opp->peer);
1841 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
1843 struct ofp_header *rq = oh;
1844 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
1849 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
1850 struct ofp_header *oh)
1852 struct ofp_switch_features *osf;
1854 unsigned int port_no;
1855 struct ofport *port;
1857 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1858 osf->datapath_id = htonll(p->datapath_id);
1859 osf->n_buffers = htonl(pktbuf_capacity());
1861 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1862 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
1863 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1864 (1u << OFPAT_SET_VLAN_VID) |
1865 (1u << OFPAT_SET_VLAN_PCP) |
1866 (1u << OFPAT_STRIP_VLAN) |
1867 (1u << OFPAT_SET_DL_SRC) |
1868 (1u << OFPAT_SET_DL_DST) |
1869 (1u << OFPAT_SET_NW_SRC) |
1870 (1u << OFPAT_SET_NW_DST) |
1871 (1u << OFPAT_SET_NW_TOS) |
1872 (1u << OFPAT_SET_TP_SRC) |
1873 (1u << OFPAT_SET_TP_DST));
1875 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1876 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1879 queue_tx(buf, ofconn, ofconn->reply_counter);
1884 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1885 struct ofp_header *oh)
1888 struct ofp_switch_config *osc;
1892 /* Figure out flags. */
1893 dpif_get_drop_frags(p->dpif, &drop_frags);
1894 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1897 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1898 osc->flags = htons(flags);
1899 osc->miss_send_len = htons(ofconn->miss_send_len);
1900 queue_tx(buf, ofconn, ofconn->reply_counter);
1906 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1907 struct ofp_switch_config *osc)
1912 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1916 flags = ntohs(osc->flags);
1918 if (ofconn == p->controller) {
1919 switch (flags & OFPC_FRAG_MASK) {
1920 case OFPC_FRAG_NORMAL:
1921 dpif_set_drop_frags(p->dpif, false);
1923 case OFPC_FRAG_DROP:
1924 dpif_set_drop_frags(p->dpif, true);
1927 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1933 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1934 if (ntohs(osc->miss_send_len) != 0) {
1935 ofconn->pktbuf = pktbuf_create();
1937 pktbuf_destroy(ofconn->pktbuf);
1941 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1947 add_output_group_action(struct odp_actions *actions, uint16_t group,
1948 uint16_t *nf_output_iface)
1950 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1952 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
1953 *nf_output_iface = NF_OUT_FLOOD;
1958 add_controller_action(struct odp_actions *actions,
1959 const struct ofp_action_output *oao)
1961 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
1962 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
1965 struct action_xlate_ctx {
1967 const flow_t *flow; /* Flow to which these actions correspond. */
1968 int recurse; /* Recursion level, via xlate_table_action. */
1969 struct ofproto *ofproto;
1970 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
1971 * null pointer if we are revalidating
1972 * without a packet to refer to. */
1975 struct odp_actions *out; /* Datapath actions. */
1976 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
1977 bool may_set_up_flow; /* True ordinarily; false if the actions must
1978 * be reassessed for every packet. */
1979 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
1982 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
1983 struct action_xlate_ctx *ctx);
1986 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
1988 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
1991 if (ofport->opp.config & OFPPC_NO_FWD) {
1992 /* Forwarding disabled on port. */
1997 * We don't have an ofport record for this port, but it doesn't hurt to
1998 * allow forwarding to it anyhow. Maybe such a port will appear later
1999 * and we're pre-populating the flow table.
2003 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2004 ctx->nf_output_iface = port;
2007 static struct rule *
2008 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2011 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2013 /* The rule we found might not be valid, since we could be in need of
2014 * revalidation. If it is not valid, don't return it. */
2017 && ofproto->need_revalidate
2018 && !revalidate_rule(ofproto, rule)) {
2019 COVERAGE_INC(ofproto_invalidated);
2027 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2029 if (!ctx->recurse) {
2034 flow.in_port = in_port;
2036 rule = lookup_valid_rule(ctx->ofproto, &flow);
2043 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2050 xlate_output_action(struct action_xlate_ctx *ctx,
2051 const struct ofp_action_output *oao)
2054 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2056 ctx->nf_output_iface = NF_OUT_DROP;
2058 switch (ntohs(oao->port)) {
2060 add_output_action(ctx, ctx->flow->in_port);
2063 xlate_table_action(ctx, ctx->flow->in_port);
2066 if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet,
2067 ctx->out, ctx->tags,
2068 &ctx->nf_output_iface,
2069 ctx->ofproto->aux)) {
2070 COVERAGE_INC(ofproto_uninstallable);
2071 ctx->may_set_up_flow = false;
2075 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2076 &ctx->nf_output_iface);
2079 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2081 case OFPP_CONTROLLER:
2082 add_controller_action(ctx->out, oao);
2085 add_output_action(ctx, ODPP_LOCAL);
2088 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2089 if (odp_port != ctx->flow->in_port) {
2090 add_output_action(ctx, odp_port);
2095 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2096 ctx->nf_output_iface = NF_OUT_FLOOD;
2097 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2098 ctx->nf_output_iface = prev_nf_output_iface;
2099 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2100 ctx->nf_output_iface != NF_OUT_FLOOD) {
2101 ctx->nf_output_iface = NF_OUT_MULTI;
2106 xlate_nicira_action(struct action_xlate_ctx *ctx,
2107 const struct nx_action_header *nah)
2109 const struct nx_action_resubmit *nar;
2110 int subtype = ntohs(nah->subtype);
2112 assert(nah->vendor == htonl(NX_VENDOR_ID));
2114 case NXAST_RESUBMIT:
2115 nar = (const struct nx_action_resubmit *) nah;
2116 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2120 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2126 do_xlate_actions(const union ofp_action *in, size_t n_in,
2127 struct action_xlate_ctx *ctx)
2129 struct actions_iterator iter;
2130 const union ofp_action *ia;
2131 const struct ofport *port;
2133 port = port_array_get(&ctx->ofproto->ports, ctx->flow->in_port);
2134 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2135 port->opp.config & (eth_addr_equals(ctx->flow->dl_dst, stp_eth_addr)
2136 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2137 /* Drop this flow. */
2141 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2142 uint16_t type = ntohs(ia->type);
2143 union odp_action *oa;
2147 xlate_output_action(ctx, &ia->output);
2150 case OFPAT_SET_VLAN_VID:
2151 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2152 oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2155 case OFPAT_SET_VLAN_PCP:
2156 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2157 oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2160 case OFPAT_STRIP_VLAN:
2161 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2164 case OFPAT_SET_DL_SRC:
2165 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2166 memcpy(oa->dl_addr.dl_addr,
2167 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2170 case OFPAT_SET_DL_DST:
2171 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2172 memcpy(oa->dl_addr.dl_addr,
2173 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2176 case OFPAT_SET_NW_SRC:
2177 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2178 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2181 case OFPAT_SET_NW_DST:
2182 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2183 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2186 case OFPAT_SET_NW_TOS:
2187 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2188 oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2191 case OFPAT_SET_TP_SRC:
2192 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2193 oa->tp_port.tp_port = ia->tp_port.tp_port;
2196 case OFPAT_SET_TP_DST:
2197 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2198 oa->tp_port.tp_port = ia->tp_port.tp_port;
2202 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2206 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2213 xlate_actions(const union ofp_action *in, size_t n_in,
2214 const flow_t *flow, struct ofproto *ofproto,
2215 const struct ofpbuf *packet,
2216 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2217 uint16_t *nf_output_iface)
2219 tag_type no_tags = 0;
2220 struct action_xlate_ctx ctx;
2221 COVERAGE_INC(ofproto_ofp2odp);
2222 odp_actions_init(out);
2225 ctx.ofproto = ofproto;
2226 ctx.packet = packet;
2228 ctx.tags = tags ? tags : &no_tags;
2229 ctx.may_set_up_flow = true;
2230 ctx.nf_output_iface = NF_OUT_DROP;
2231 do_xlate_actions(in, n_in, &ctx);
2233 /* Check with in-band control to see if we're allowed to set up this
2235 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2236 ctx.may_set_up_flow = false;
2239 if (may_set_up_flow) {
2240 *may_set_up_flow = ctx.may_set_up_flow;
2242 if (nf_output_iface) {
2243 *nf_output_iface = ctx.nf_output_iface;
2245 if (odp_actions_overflow(out)) {
2246 odp_actions_init(out);
2247 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2253 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2254 struct ofp_header *oh)
2256 struct ofp_packet_out *opo;
2257 struct ofpbuf payload, *buffer;
2258 struct odp_actions actions;
2264 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2268 opo = (struct ofp_packet_out *) oh;
2270 COVERAGE_INC(ofproto_packet_out);
2271 if (opo->buffer_id != htonl(UINT32_MAX)) {
2272 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2274 if (error || !buffer) {
2282 flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2283 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2284 &flow, p, &payload, &actions, NULL, NULL, NULL);
2289 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2291 ofpbuf_delete(buffer);
2297 update_port_config(struct ofproto *p, struct ofport *port,
2298 uint32_t config, uint32_t mask)
2300 mask &= config ^ port->opp.config;
2301 if (mask & OFPPC_PORT_DOWN) {
2302 if (config & OFPPC_PORT_DOWN) {
2303 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2305 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2308 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2309 if (mask & REVALIDATE_BITS) {
2310 COVERAGE_INC(ofproto_costly_flags);
2311 port->opp.config ^= mask & REVALIDATE_BITS;
2312 p->need_revalidate = true;
2314 #undef REVALIDATE_BITS
2315 if (mask & OFPPC_NO_FLOOD) {
2316 port->opp.config ^= OFPPC_NO_FLOOD;
2317 refresh_port_group(p, DP_GROUP_FLOOD);
2319 if (mask & OFPPC_NO_PACKET_IN) {
2320 port->opp.config ^= OFPPC_NO_PACKET_IN;
2325 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2327 const struct ofp_port_mod *opm;
2328 struct ofport *port;
2331 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2335 opm = (struct ofp_port_mod *) oh;
2337 port = port_array_get(&p->ports,
2338 ofp_port_to_odp_port(ntohs(opm->port_no)));
2340 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2341 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2342 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2344 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2345 if (opm->advertise) {
2346 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2352 static struct ofpbuf *
2353 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2355 struct ofp_stats_reply *osr;
2358 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2359 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2361 osr->flags = htons(0);
2365 static struct ofpbuf *
2366 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2368 return make_stats_reply(request->header.xid, request->type, body_len);
2372 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2374 struct ofpbuf *msg = *msgp;
2375 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2376 if (nbytes + msg->size > UINT16_MAX) {
2377 struct ofp_stats_reply *reply = msg->data;
2378 reply->flags = htons(OFPSF_REPLY_MORE);
2379 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2380 queue_tx(msg, ofconn, ofconn->reply_counter);
2382 return ofpbuf_put_uninit(*msgp, nbytes);
2386 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2387 struct ofp_stats_request *request)
2389 struct ofp_desc_stats *ods;
2392 msg = start_stats_reply(request, sizeof *ods);
2393 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2394 strncpy(ods->mfr_desc, p->manufacturer, sizeof ods->mfr_desc);
2395 strncpy(ods->hw_desc, p->hardware, sizeof ods->hw_desc);
2396 strncpy(ods->sw_desc, p->software, sizeof ods->sw_desc);
2397 strncpy(ods->serial_num, p->serial, sizeof ods->serial_num);
2398 strncpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2399 queue_tx(msg, ofconn, ofconn->reply_counter);
2405 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2407 struct rule *rule = rule_from_cls_rule(cls_rule);
2408 int *n_subrules = n_subrules_;
2416 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2417 struct ofp_stats_request *request)
2419 struct ofp_table_stats *ots;
2421 struct odp_stats dpstats;
2422 int n_exact, n_subrules, n_wild;
2424 msg = start_stats_reply(request, sizeof *ots * 2);
2426 /* Count rules of various kinds. */
2428 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2429 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2430 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2433 dpif_get_dp_stats(p->dpif, &dpstats);
2434 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2435 memset(ots, 0, sizeof *ots);
2436 ots->table_id = TABLEID_HASH;
2437 strcpy(ots->name, "hash");
2438 ots->wildcards = htonl(0);
2439 ots->max_entries = htonl(dpstats.max_capacity);
2440 ots->active_count = htonl(n_exact);
2441 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2443 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2445 /* Classifier table. */
2446 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2447 memset(ots, 0, sizeof *ots);
2448 ots->table_id = TABLEID_CLASSIFIER;
2449 strcpy(ots->name, "classifier");
2450 ots->wildcards = htonl(OFPFW_ALL);
2451 ots->max_entries = htonl(65536);
2452 ots->active_count = htonl(n_wild);
2453 ots->lookup_count = htonll(0); /* XXX */
2454 ots->matched_count = htonll(0); /* XXX */
2456 queue_tx(msg, ofconn, ofconn->reply_counter);
2461 append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
2464 struct netdev_stats stats;
2465 struct ofp_port_stats *ops;
2467 /* Intentionally ignore return value, since errors will set
2468 * 'stats' to all-1s, which is correct for OpenFlow, and
2469 * netdev_get_stats() will log errors. */
2470 netdev_get_stats(port->netdev, &stats);
2472 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2473 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2474 memset(ops->pad, 0, sizeof ops->pad);
2475 ops->rx_packets = htonll(stats.rx_packets);
2476 ops->tx_packets = htonll(stats.tx_packets);
2477 ops->rx_bytes = htonll(stats.rx_bytes);
2478 ops->tx_bytes = htonll(stats.tx_bytes);
2479 ops->rx_dropped = htonll(stats.rx_dropped);
2480 ops->tx_dropped = htonll(stats.tx_dropped);
2481 ops->rx_errors = htonll(stats.rx_errors);
2482 ops->tx_errors = htonll(stats.tx_errors);
2483 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2484 ops->rx_over_err = htonll(stats.rx_over_errors);
2485 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2486 ops->collisions = htonll(stats.collisions);
2490 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2491 struct ofp_stats_request *osr,
2494 struct ofp_port_stats_request *psr;
2495 struct ofp_port_stats *ops;
2497 struct ofport *port;
2498 unsigned int port_no;
2500 if (arg_size != sizeof *psr) {
2501 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2503 psr = (struct ofp_port_stats_request *) osr->body;
2505 msg = start_stats_reply(osr, sizeof *ops * 16);
2506 if (psr->port_no != htons(OFPP_NONE)) {
2507 port = port_array_get(&p->ports,
2508 ofp_port_to_odp_port(ntohs(psr->port_no)));
2510 append_port_stat(port, ntohs(psr->port_no), ofconn, msg);
2513 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2514 append_port_stat(port, port_no, ofconn, msg);
2518 queue_tx(msg, ofconn, ofconn->reply_counter);
2522 struct flow_stats_cbdata {
2523 struct ofproto *ofproto;
2524 struct ofconn *ofconn;
2530 query_stats(struct ofproto *p, struct rule *rule,
2531 uint64_t *packet_countp, uint64_t *byte_countp)
2533 uint64_t packet_count, byte_count;
2534 struct rule *subrule;
2535 struct odp_flow *odp_flows;
2538 packet_count = rule->packet_count;
2539 byte_count = rule->byte_count;
2541 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2542 odp_flows = xcalloc(1, n_odp_flows * sizeof *odp_flows);
2543 if (rule->cr.wc.wildcards) {
2545 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2546 odp_flows[i++].key = subrule->cr.flow;
2547 packet_count += subrule->packet_count;
2548 byte_count += subrule->byte_count;
2551 odp_flows[0].key = rule->cr.flow;
2554 packet_count = rule->packet_count;
2555 byte_count = rule->byte_count;
2556 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2558 for (i = 0; i < n_odp_flows; i++) {
2559 struct odp_flow *odp_flow = &odp_flows[i];
2560 packet_count += odp_flow->stats.n_packets;
2561 byte_count += odp_flow->stats.n_bytes;
2566 *packet_countp = packet_count;
2567 *byte_countp = byte_count;
2571 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2573 struct rule *rule = rule_from_cls_rule(rule_);
2574 struct flow_stats_cbdata *cbdata = cbdata_;
2575 struct ofp_flow_stats *ofs;
2576 uint64_t packet_count, byte_count;
2577 size_t act_len, len;
2578 long long int tdiff = time_msec() - rule->created;
2579 uint32_t sec = tdiff / 1000;
2580 uint32_t msec = tdiff - (sec * 1000);
2582 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2586 act_len = sizeof *rule->actions * rule->n_actions;
2587 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2589 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2591 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2592 ofs->length = htons(len);
2593 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2595 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match);
2596 ofs->duration_sec = htonl(sec);
2597 ofs->duration_nsec = htonl(msec * 1000000);
2598 ofs->cookie = rule->flow_cookie;
2599 ofs->priority = htons(rule->cr.priority);
2600 ofs->idle_timeout = htons(rule->idle_timeout);
2601 ofs->hard_timeout = htons(rule->hard_timeout);
2602 memset(ofs->pad2, 0, sizeof ofs->pad2);
2603 ofs->packet_count = htonll(packet_count);
2604 ofs->byte_count = htonll(byte_count);
2605 memcpy(ofs->actions, rule->actions, act_len);
2609 table_id_to_include(uint8_t table_id)
2611 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2612 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2613 : table_id == 0xff ? CLS_INC_ALL
2618 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2619 const struct ofp_stats_request *osr,
2622 struct ofp_flow_stats_request *fsr;
2623 struct flow_stats_cbdata cbdata;
2624 struct cls_rule target;
2626 if (arg_size != sizeof *fsr) {
2627 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2629 fsr = (struct ofp_flow_stats_request *) osr->body;
2631 COVERAGE_INC(ofproto_flows_req);
2633 cbdata.ofconn = ofconn;
2634 cbdata.out_port = fsr->out_port;
2635 cbdata.msg = start_stats_reply(osr, 1024);
2636 cls_rule_from_match(&target, &fsr->match, 0);
2637 classifier_for_each_match(&p->cls, &target,
2638 table_id_to_include(fsr->table_id),
2639 flow_stats_cb, &cbdata);
2640 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2644 struct flow_stats_ds_cbdata {
2645 struct ofproto *ofproto;
2650 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
2652 struct rule *rule = rule_from_cls_rule(rule_);
2653 struct flow_stats_ds_cbdata *cbdata = cbdata_;
2654 struct ds *results = cbdata->results;
2655 struct ofp_match match;
2656 uint64_t packet_count, byte_count;
2657 size_t act_len = sizeof *rule->actions * rule->n_actions;
2659 /* Don't report on subrules. */
2660 if (rule->super != NULL) {
2664 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2665 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &match);
2667 ds_put_format(results, "duration=%llds, ",
2668 (time_msec() - rule->created) / 1000);
2669 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2670 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2671 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2672 ofp_print_match(results, &match, true);
2673 ofp_print_actions(results, &rule->actions->header, act_len);
2674 ds_put_cstr(results, "\n");
2677 /* Adds a pretty-printed description of all flows to 'results', including
2678 * those marked hidden by secchan (e.g., by in-band control). */
2680 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2682 struct ofp_match match;
2683 struct cls_rule target;
2684 struct flow_stats_ds_cbdata cbdata;
2686 memset(&match, 0, sizeof match);
2687 match.wildcards = htonl(OFPFW_ALL);
2690 cbdata.results = results;
2692 cls_rule_from_match(&target, &match, 0);
2693 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2694 flow_stats_ds_cb, &cbdata);
2697 struct aggregate_stats_cbdata {
2698 struct ofproto *ofproto;
2700 uint64_t packet_count;
2701 uint64_t byte_count;
2706 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2708 struct rule *rule = rule_from_cls_rule(rule_);
2709 struct aggregate_stats_cbdata *cbdata = cbdata_;
2710 uint64_t packet_count, byte_count;
2712 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2716 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2718 cbdata->packet_count += packet_count;
2719 cbdata->byte_count += byte_count;
2724 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2725 const struct ofp_stats_request *osr,
2728 struct ofp_aggregate_stats_request *asr;
2729 struct ofp_aggregate_stats_reply *reply;
2730 struct aggregate_stats_cbdata cbdata;
2731 struct cls_rule target;
2734 if (arg_size != sizeof *asr) {
2735 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2737 asr = (struct ofp_aggregate_stats_request *) osr->body;
2739 COVERAGE_INC(ofproto_agg_request);
2741 cbdata.out_port = asr->out_port;
2742 cbdata.packet_count = 0;
2743 cbdata.byte_count = 0;
2745 cls_rule_from_match(&target, &asr->match, 0);
2746 classifier_for_each_match(&p->cls, &target,
2747 table_id_to_include(asr->table_id),
2748 aggregate_stats_cb, &cbdata);
2750 msg = start_stats_reply(osr, sizeof *reply);
2751 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2752 reply->flow_count = htonl(cbdata.n_flows);
2753 reply->packet_count = htonll(cbdata.packet_count);
2754 reply->byte_count = htonll(cbdata.byte_count);
2755 queue_tx(msg, ofconn, ofconn->reply_counter);
2760 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2761 struct ofp_header *oh)
2763 struct ofp_stats_request *osr;
2767 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2772 osr = (struct ofp_stats_request *) oh;
2774 switch (ntohs(osr->type)) {
2776 return handle_desc_stats_request(p, ofconn, osr);
2779 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2781 case OFPST_AGGREGATE:
2782 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2785 return handle_table_stats_request(p, ofconn, osr);
2788 return handle_port_stats_request(p, ofconn, osr, arg_size);
2791 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2794 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2798 static long long int
2799 msec_from_nsec(uint64_t sec, uint32_t nsec)
2801 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2805 update_time(struct ofproto *ofproto, struct rule *rule,
2806 const struct odp_flow_stats *stats)
2808 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2809 if (used > rule->used) {
2811 if (rule->super && used > rule->super->used) {
2812 rule->super->used = used;
2814 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
2819 update_stats(struct ofproto *ofproto, struct rule *rule,
2820 const struct odp_flow_stats *stats)
2822 if (stats->n_packets) {
2823 update_time(ofproto, rule, stats);
2824 rule->packet_count += stats->n_packets;
2825 rule->byte_count += stats->n_bytes;
2826 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
2832 add_flow(struct ofproto *p, struct ofconn *ofconn,
2833 struct ofp_flow_mod *ofm, size_t n_actions)
2835 struct ofpbuf *packet;
2840 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
2844 flow_from_match(&flow, &wildcards, &ofm->match);
2845 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
2846 ntohs(ofm->priority))) {
2847 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
2851 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
2852 n_actions, ntohs(ofm->idle_timeout),
2853 ntohs(ofm->hard_timeout), ofm->cookie,
2854 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
2855 cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority));
2859 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2860 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2864 rule_insert(p, rule, packet, in_port);
2865 ofpbuf_delete(packet);
2870 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
2871 size_t n_actions, uint16_t command, struct rule *rule)
2873 if (rule_is_hidden(rule)) {
2877 if (command == OFPFC_DELETE) {
2878 long long int now = time_msec();
2879 send_flow_removed(p, rule, now, OFPRR_DELETE);
2880 rule_remove(p, rule);
2882 size_t actions_len = n_actions * sizeof *rule->actions;
2884 if (n_actions == rule->n_actions
2885 && !memcmp(ofm->actions, rule->actions, actions_len))
2890 free(rule->actions);
2891 rule->actions = xmemdup(ofm->actions, actions_len);
2892 rule->n_actions = n_actions;
2893 rule->flow_cookie = ofm->cookie;
2895 if (rule->cr.wc.wildcards) {
2896 COVERAGE_INC(ofproto_mod_wc_flow);
2897 p->need_revalidate = true;
2899 rule_update_actions(p, rule);
2907 modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm,
2908 size_t n_actions, uint16_t command)
2914 flow_from_match(&flow, &wildcards, &ofm->match);
2915 rule = rule_from_cls_rule(classifier_find_rule_exactly(
2916 &p->cls, &flow, wildcards,
2917 ntohs(ofm->priority)));
2920 if (command == OFPFC_DELETE
2921 && ofm->out_port != htons(OFPP_NONE)
2922 && !rule_has_out_port(rule, ofm->out_port)) {
2926 modify_flow(p, ofm, n_actions, command, rule);
2931 struct modify_flows_cbdata {
2932 struct ofproto *ofproto;
2933 const struct ofp_flow_mod *ofm;
2940 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
2942 struct rule *rule = rule_from_cls_rule(rule_);
2943 struct modify_flows_cbdata *cbdata = cbdata_;
2945 if (cbdata->out_port != htons(OFPP_NONE)
2946 && !rule_has_out_port(rule, cbdata->out_port)) {
2950 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions,
2951 cbdata->command, rule);
2955 modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm,
2956 size_t n_actions, uint16_t command)
2958 struct modify_flows_cbdata cbdata;
2959 struct cls_rule target;
2963 cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port
2964 : htons(OFPP_NONE));
2965 cbdata.n_actions = n_actions;
2966 cbdata.command = command;
2968 cls_rule_from_match(&target, &ofm->match, 0);
2970 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2971 modify_flows_cb, &cbdata);
2976 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
2977 struct ofp_flow_mod *ofm)
2982 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
2983 sizeof *ofm->actions, &n_actions);
2988 /* We do not support the emergency flow cache. It will hopefully
2989 * get dropped from OpenFlow in the near future. */
2990 if (ofm->flags & htons(OFPFF_EMERG)) {
2991 /* There isn't a good fit for an error code, so just state that the
2992 * flow table is full. */
2993 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
2996 normalize_match(&ofm->match);
2997 if (!ofm->match.wildcards) {
2998 ofm->priority = htons(UINT16_MAX);
3001 error = validate_actions((const union ofp_action *) ofm->actions,
3002 n_actions, p->max_ports);
3007 switch (ntohs(ofm->command)) {
3009 return add_flow(p, ofconn, ofm, n_actions);
3012 return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY);
3014 case OFPFC_MODIFY_STRICT:
3015 return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY);
3018 return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE);
3020 case OFPFC_DELETE_STRICT:
3021 return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE);
3024 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3029 send_capability_reply(struct ofproto *p, struct ofconn *ofconn, uint32_t xid)
3031 struct ofmp_capability_reply *ocr;
3033 char capabilities[] = "com.nicira.mgmt.manager=false\n";
3035 ocr = make_openflow_xid(sizeof(*ocr), OFPT_VENDOR, xid, &b);
3036 ocr->header.header.vendor = htonl(NX_VENDOR_ID);
3037 ocr->header.header.subtype = htonl(NXT_MGMT);
3038 ocr->header.type = htons(OFMPT_CAPABILITY_REPLY);
3040 ocr->format = htonl(OFMPCOF_SIMPLE);
3041 ocr->mgmt_id = htonll(p->mgmt_id);
3043 ofpbuf_put(b, capabilities, strlen(capabilities));
3045 queue_tx(b, ofconn, ofconn->reply_counter);
3049 handle_ofmp(struct ofproto *p, struct ofconn *ofconn,
3050 struct ofmp_header *ofmph)
3052 size_t msg_len = ntohs(ofmph->header.header.length);
3053 if (msg_len < sizeof(*ofmph)) {
3054 VLOG_WARN_RL(&rl, "dropping short managment message: %zu\n", msg_len);
3055 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3058 if (ofmph->type == htons(OFMPT_CAPABILITY_REQUEST)) {
3059 struct ofmp_capability_request *ofmpcr;
3061 if (msg_len < sizeof(struct ofmp_capability_request)) {
3062 VLOG_WARN_RL(&rl, "dropping short capability request: %zu\n",
3064 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3067 ofmpcr = (struct ofmp_capability_request *)ofmph;
3068 if (ofmpcr->format != htonl(OFMPCAF_SIMPLE)) {
3069 /* xxx Find a better type than bad subtype */
3070 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3073 send_capability_reply(p, ofconn, ofmph->header.header.xid);
3076 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3081 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3083 struct ofp_vendor_header *ovh = msg;
3084 struct nicira_header *nh;
3086 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3087 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3089 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3090 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3092 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3093 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3097 switch (ntohl(nh->subtype)) {
3098 case NXT_STATUS_REQUEST:
3099 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3102 case NXT_ACT_SET_CONFIG:
3103 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
3105 case NXT_ACT_GET_CONFIG:
3106 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
3108 case NXT_COMMAND_REQUEST:
3110 return executer_handle_request(p->executer, ofconn->rconn, msg);
3115 return handle_ofmp(p, ofconn, msg);
3118 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3122 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
3124 struct ofp_header *ob;
3127 /* Currently, everything executes synchronously, so we can just
3128 * immediately send the barrier reply. */
3129 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3130 queue_tx(buf, ofconn, ofconn->reply_counter);
3135 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3136 struct ofpbuf *ofp_msg)
3138 struct ofp_header *oh = ofp_msg->data;
3141 COVERAGE_INC(ofproto_recv_openflow);
3143 case OFPT_ECHO_REQUEST:
3144 error = handle_echo_request(ofconn, oh);
3147 case OFPT_ECHO_REPLY:
3151 case OFPT_FEATURES_REQUEST:
3152 error = handle_features_request(p, ofconn, oh);
3155 case OFPT_GET_CONFIG_REQUEST:
3156 error = handle_get_config_request(p, ofconn, oh);
3159 case OFPT_SET_CONFIG:
3160 error = handle_set_config(p, ofconn, ofp_msg->data);
3163 case OFPT_PACKET_OUT:
3164 error = handle_packet_out(p, ofconn, ofp_msg->data);
3168 error = handle_port_mod(p, oh);
3172 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3175 case OFPT_STATS_REQUEST:
3176 error = handle_stats_request(p, ofconn, oh);
3180 error = handle_vendor(p, ofconn, ofp_msg->data);
3183 case OFPT_BARRIER_REQUEST:
3184 error = handle_barrier_request(ofconn, oh);
3188 if (VLOG_IS_WARN_ENABLED()) {
3189 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3190 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3193 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3198 send_error_oh(ofconn, ofp_msg->data, error);
3203 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3205 struct odp_msg *msg = packet->data;
3206 uint16_t in_port = odp_port_to_ofp_port(msg->port);
3208 struct ofpbuf payload;
3211 /* Handle controller actions. */
3212 if (msg->type == _ODPL_ACTION_NR) {
3213 COVERAGE_INC(ofproto_ctlr_action);
3214 pinsched_send(p->action_sched, in_port, packet,
3215 send_packet_in_action, p);
3219 payload.data = msg + 1;
3220 payload.size = msg->length - sizeof *msg;
3221 flow_extract(&payload, msg->port, &flow);
3223 /* Check with in-band control to see if this packet should be sent
3224 * to the local port regardless of the flow table. */
3225 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3226 union odp_action action;
3228 memset(&action, 0, sizeof(action));
3229 action.output.type = ODPAT_OUTPUT;
3230 action.output.port = ODPP_LOCAL;
3231 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3234 rule = lookup_valid_rule(p, &flow);
3236 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3237 struct ofport *port = port_array_get(&p->ports, msg->port);
3239 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3240 COVERAGE_INC(ofproto_no_packet_in);
3241 /* XXX install 'drop' flow entry */
3242 ofpbuf_delete(packet);
3246 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3249 COVERAGE_INC(ofproto_packet_in);
3250 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3254 if (rule->cr.wc.wildcards) {
3255 rule = rule_create_subrule(p, rule, &flow);
3256 rule_make_actions(p, rule, packet);
3258 if (!rule->may_install) {
3259 /* The rule is not installable, that is, we need to process every
3260 * packet, so process the current packet and set its actions into
3262 rule_make_actions(p, rule, packet);
3264 /* XXX revalidate rule if it needs it */
3268 rule_execute(p, rule, &payload, &flow);
3269 rule_reinstall(p, rule);
3271 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY
3272 && rconn_is_connected(p->controller->rconn)) {
3274 * Extra-special case for fail-open mode.
3276 * We are in fail-open mode and the packet matched the fail-open rule,
3277 * but we are connected to a controller too. We should send the packet
3278 * up to the controller in the hope that it will try to set up a flow
3279 * and thereby allow us to exit fail-open.
3281 * See the top-level comment in fail-open.c for more information.
3283 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3285 ofpbuf_delete(packet);
3290 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3292 struct rule *sub = rule_from_cls_rule(sub_);
3293 struct revalidate_cbdata *cbdata = cbdata_;
3295 if (cbdata->revalidate_all
3296 || (cbdata->revalidate_subrules && sub->super)
3297 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3298 revalidate_rule(cbdata->ofproto, sub);
3303 revalidate_rule(struct ofproto *p, struct rule *rule)
3305 const flow_t *flow = &rule->cr.flow;
3307 COVERAGE_INC(ofproto_revalidate_rule);
3310 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3312 rule_remove(p, rule);
3314 } else if (super != rule->super) {
3315 COVERAGE_INC(ofproto_revalidate_moved);
3316 list_remove(&rule->list);
3317 list_push_back(&super->list, &rule->list);
3318 rule->super = super;
3319 rule->hard_timeout = super->hard_timeout;
3320 rule->idle_timeout = super->idle_timeout;
3321 rule->created = super->created;
3326 rule_update_actions(p, rule);
3330 static struct ofpbuf *
3331 compose_flow_removed(const struct rule *rule, long long int now, uint8_t reason)
3333 struct ofp_flow_removed *ofr;
3335 long long int last_used = rule->used ? now - rule->used : 0;
3336 long long int tdiff = time_msec() - rule->created - last_used;
3337 uint32_t sec = tdiff / 1000;
3338 uint32_t msec = tdiff - (sec * 1000);
3340 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
3341 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofr->match);
3342 ofr->cookie = rule->flow_cookie;
3343 ofr->priority = htons(rule->cr.priority);
3344 ofr->reason = reason;
3345 ofr->duration_sec = htonl(sec);
3346 ofr->duration_nsec = htonl(msec * 1000000);
3347 ofr->idle_timeout = htons(rule->idle_timeout);
3348 ofr->packet_count = htonll(rule->packet_count);
3349 ofr->byte_count = htonll(rule->byte_count);
3355 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3357 assert(rule->installed);
3358 assert(!rule->cr.wc.wildcards);
3361 rule_remove(ofproto, rule);
3363 rule_uninstall(ofproto, rule);
3367 send_flow_removed(struct ofproto *p, struct rule *rule,
3368 long long int now, uint8_t reason)
3370 struct ofconn *ofconn;
3371 struct ofconn *prev;
3372 struct ofpbuf *buf = NULL;
3374 /* We limit the maximum number of queued flow expirations it by accounting
3375 * them under the counter for replies. That works because preventing
3376 * OpenFlow requests from being processed also prevents new flows from
3377 * being added (and expiring). (It also prevents processing OpenFlow
3378 * requests that would not add new flows, so it is imperfect.) */
3381 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3382 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)) {
3384 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3386 buf = compose_flow_removed(rule, now, reason);
3392 queue_tx(buf, prev, prev->reply_counter);
3398 expire_rule(struct cls_rule *cls_rule, void *p_)
3400 struct ofproto *p = p_;
3401 struct rule *rule = rule_from_cls_rule(cls_rule);
3402 long long int hard_expire, idle_expire, expire, now;
3404 hard_expire = (rule->hard_timeout
3405 ? rule->created + rule->hard_timeout * 1000
3407 idle_expire = (rule->idle_timeout
3408 && (rule->super || list_is_empty(&rule->list))
3409 ? rule->used + rule->idle_timeout * 1000
3411 expire = MIN(hard_expire, idle_expire);
3415 if (rule->installed && now >= rule->used + 5000) {
3416 uninstall_idle_flow(p, rule);
3417 } else if (!rule->cr.wc.wildcards) {
3418 active_timeout(p, rule);
3424 COVERAGE_INC(ofproto_expired);
3426 /* Update stats. This code will be a no-op if the rule expired
3427 * due to an idle timeout. */
3428 if (rule->cr.wc.wildcards) {
3429 struct rule *subrule, *next;
3430 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3431 rule_remove(p, subrule);
3434 rule_uninstall(p, rule);
3437 if (!rule_is_hidden(rule)) {
3438 send_flow_removed(p, rule, now,
3440 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
3442 rule_remove(p, rule);
3446 active_timeout(struct ofproto *ofproto, struct rule *rule)
3448 if (ofproto->netflow && !is_controller_rule(rule) &&
3449 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
3450 struct ofexpired expired;
3451 struct odp_flow odp_flow;
3453 /* Get updated flow stats. */
3454 memset(&odp_flow, 0, sizeof odp_flow);
3455 if (rule->installed) {
3456 odp_flow.key = rule->cr.flow;
3457 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
3458 dpif_flow_get(ofproto->dpif, &odp_flow);
3460 if (odp_flow.stats.n_packets) {
3461 update_time(ofproto, rule, &odp_flow.stats);
3462 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
3463 odp_flow.stats.tcp_flags);
3467 expired.flow = rule->cr.flow;
3468 expired.packet_count = rule->packet_count +
3469 odp_flow.stats.n_packets;
3470 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
3471 expired.used = rule->used;
3473 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
3475 /* Schedule us to send the accumulated records once we have
3476 * collected all of them. */
3477 poll_immediate_wake();
3482 update_used(struct ofproto *p)
3484 struct odp_flow *flows;
3489 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
3494 for (i = 0; i < n_flows; i++) {
3495 struct odp_flow *f = &flows[i];
3498 rule = rule_from_cls_rule(
3499 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3500 if (!rule || !rule->installed) {
3501 COVERAGE_INC(ofproto_unexpected_rule);
3502 dpif_flow_del(p->dpif, f);
3506 update_time(p, rule, &f->stats);
3507 rule_account(p, rule, f->stats.n_bytes);
3513 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3514 const struct ofpbuf *packet, int send_len)
3516 struct odp_msg *msg = packet->data;
3517 struct ofpbuf payload;
3521 /* Extract packet payload from 'msg'. */
3522 payload.data = msg + 1;
3523 payload.size = msg->length - sizeof *msg;
3525 /* Construct ofp_packet_in message. */
3526 reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3527 opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port), reason,
3528 &payload, send_len);
3531 rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100);
3535 send_packet_in_action(struct ofpbuf *packet, void *p_)
3537 struct ofproto *p = p_;
3538 struct ofconn *ofconn;
3539 struct odp_msg *msg;
3542 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3543 if (ofconn == p->controller || ofconn->miss_send_len) {
3544 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3547 ofpbuf_delete(packet);
3551 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3553 struct ofproto *p = p_;
3554 bool in_fail_open = p->fail_open && fail_open_is_active(p->fail_open);
3555 struct ofconn *ofconn;
3556 struct ofpbuf payload;
3557 struct odp_msg *msg;
3560 payload.data = msg + 1;
3561 payload.size = msg->length - sizeof *msg;
3562 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3563 if (ofconn->miss_send_len) {
3564 struct pktbuf *pb = ofconn->pktbuf;
3565 uint32_t buffer_id = (in_fail_open
3567 : pktbuf_save(pb, &payload, msg->port));
3568 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3570 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3573 ofpbuf_delete(packet);
3577 pick_datapath_id(const struct ofproto *ofproto)
3579 const struct ofport *port;
3581 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
3583 uint8_t ea[ETH_ADDR_LEN];
3586 error = netdev_get_etheraddr(port->netdev, ea);
3588 return eth_addr_to_uint64(ea);
3590 VLOG_WARN("could not get MAC address for %s (%s)",
3591 netdev_get_name(port->netdev), strerror(error));
3593 return ofproto->fallback_dpid;
3597 pick_fallback_dpid(void)
3599 uint8_t ea[ETH_ADDR_LEN];
3600 eth_addr_random(ea);
3601 ea[0] = 0x00; /* Set Nicira OUI. */
3604 return eth_addr_to_uint64(ea);
3608 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3609 struct odp_actions *actions, tag_type *tags,
3610 uint16_t *nf_output_iface, void *ofproto_)
3612 struct ofproto *ofproto = ofproto_;
3615 /* Drop frames for reserved multicast addresses. */
3616 if (eth_addr_is_reserved(flow->dl_dst)) {
3620 /* Learn source MAC (but don't try to learn from revalidation). */
3621 if (packet != NULL) {
3622 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3625 /* The log messages here could actually be useful in debugging,
3626 * so keep the rate limit relatively high. */
3627 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3628 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3629 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3630 ofproto_revalidate(ofproto, rev_tag);
3634 /* Determine output port. */
3635 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3637 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
3638 } else if (out_port != flow->in_port) {
3639 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3640 *nf_output_iface = out_port;
3648 static const struct ofhooks default_ofhooks = {
3650 default_normal_ofhook_cb,