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"
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"
51 #include "stream-ssl.h"
59 #define THIS_MODULE VLM_ofproto
69 TABLEID_CLASSIFIER = 1
73 struct netdev *netdev;
74 struct ofp_phy_port opp; /* In host byte order. */
77 static void ofport_free(struct ofport *);
78 static void hton_ofp_phy_port(struct ofp_phy_port *);
80 static int xlate_actions(const union ofp_action *in, size_t n_in,
81 const flow_t *flow, struct ofproto *ofproto,
82 const struct ofpbuf *packet,
83 struct odp_actions *out, tag_type *tags,
84 bool *may_set_up_flow, uint16_t *nf_output_iface);
89 uint16_t idle_timeout; /* In seconds from time of last use. */
90 uint16_t hard_timeout; /* In seconds from time of creation. */
91 long long int used; /* Last-used time (0 if never used). */
92 long long int created; /* Creation time. */
93 uint64_t packet_count; /* Number of packets received. */
94 uint64_t byte_count; /* Number of bytes received. */
95 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
96 tag_type tags; /* Tags (set only by hooks). */
97 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
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 ofproto *, struct rule *super,
147 const union ofp_action *, size_t n_actions,
148 uint16_t idle_timeout, 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 *);
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 netflow *netflow;
212 struct classifier cls;
213 bool need_revalidate;
214 long long int next_expiration;
215 struct tag_set revalidate_set;
217 /* OpenFlow connections. */
218 struct list all_conns;
219 struct ofconn *controller;
220 struct pvconn **listeners;
222 struct pvconn **snoops;
225 /* Hooks for ovs-vswitchd. */
226 const struct ofhooks *ofhooks;
229 /* Used by default ofhooks. */
230 struct mac_learning *ml;
233 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
235 static const struct ofhooks default_ofhooks;
237 static uint64_t pick_datapath_id(const struct ofproto *);
238 static uint64_t pick_fallback_dpid(void);
239 static void send_packet_in_miss(struct ofpbuf *, void *ofproto);
240 static void send_packet_in_action(struct ofpbuf *, void *ofproto);
241 static void update_used(struct ofproto *);
242 static void update_stats(struct ofproto *, struct rule *,
243 const struct odp_flow_stats *);
244 static void expire_rule(struct cls_rule *, void *ofproto);
245 static void active_timeout(struct ofproto *ofproto, struct rule *rule);
246 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
247 static void revalidate_cb(struct cls_rule *rule_, void *p_);
249 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
251 static void handle_openflow(struct ofconn *, struct ofproto *,
254 static void refresh_port_group(struct ofproto *, unsigned int group);
255 static void update_port(struct ofproto *, const char *devname);
256 static int init_ports(struct ofproto *);
257 static void reinit_ports(struct ofproto *);
260 ofproto_create(const char *datapath, const char *datapath_type,
261 const struct ofhooks *ofhooks, void *aux,
262 struct ofproto **ofprotop)
264 struct odp_stats stats;
271 /* Connect to datapath and start listening for messages. */
272 error = dpif_open(datapath, datapath_type, &dpif);
274 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
277 error = dpif_get_dp_stats(dpif, &stats);
279 VLOG_ERR("failed to obtain stats for datapath %s: %s",
280 datapath, strerror(error));
284 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION);
286 VLOG_ERR("failed to listen on datapath %s: %s",
287 datapath, strerror(error));
291 dpif_flow_flush(dpif);
292 dpif_recv_purge(dpif);
294 /* Initialize settings. */
295 p = xzalloc(sizeof *p);
296 p->fallback_dpid = pick_fallback_dpid();
297 p->datapath_id = p->fallback_dpid;
298 p->manufacturer = xstrdup("Nicira Networks, Inc.");
299 p->hardware = xstrdup("Reference Implementation");
300 p->software = xstrdup(VERSION BUILDNR);
301 p->serial = xstrdup("None");
303 /* Initialize datapath. */
305 p->netdev_monitor = netdev_monitor_create();
306 port_array_init(&p->ports);
307 shash_init(&p->port_by_name);
308 p->max_ports = stats.max_ports;
310 /* Initialize submodules. */
311 p->switch_status = switch_status_create(p);
315 p->miss_sched = p->action_sched = NULL;
318 /* Initialize flow table. */
319 classifier_init(&p->cls);
320 p->need_revalidate = false;
321 p->next_expiration = time_msec() + 1000;
322 tag_set_init(&p->revalidate_set);
324 /* Initialize OpenFlow connections. */
325 list_init(&p->all_conns);
326 p->controller = ofconn_create(p, rconn_create(5, 8));
327 p->controller->pktbuf = pktbuf_create();
328 p->controller->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
334 /* Initialize hooks. */
336 p->ofhooks = ofhooks;
340 p->ofhooks = &default_ofhooks;
342 p->ml = mac_learning_create();
345 /* Register switch status category. */
346 p->ss_cat = switch_status_register(p->switch_status, "remote",
347 rconn_status_cb, p->controller->rconn);
349 /* Pick final datapath ID. */
350 p->datapath_id = pick_datapath_id(p);
351 VLOG_INFO("using datapath ID %012"PRIx64, p->datapath_id);
358 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
360 uint64_t old_dpid = p->datapath_id;
361 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
362 if (p->datapath_id != old_dpid) {
363 VLOG_INFO("datapath ID changed to %012"PRIx64, p->datapath_id);
364 rconn_reconnect(p->controller->rconn);
369 ofproto_set_mgmt_id(struct ofproto *p, uint64_t mgmt_id)
371 p->mgmt_id = mgmt_id;
375 ofproto_set_probe_interval(struct ofproto *p, int probe_interval)
377 probe_interval = probe_interval ? MAX(probe_interval, 5) : 0;
378 rconn_set_probe_interval(p->controller->rconn, probe_interval);
380 int trigger_duration = probe_interval ? probe_interval * 3 : 15;
381 fail_open_set_trigger_duration(p->fail_open, trigger_duration);
386 ofproto_set_max_backoff(struct ofproto *p, int max_backoff)
388 rconn_set_max_backoff(p->controller->rconn, max_backoff);
392 ofproto_set_desc(struct ofproto *p,
393 const char *manufacturer, const char *hardware,
394 const char *software, const char *serial)
397 free(p->manufacturer);
398 p->manufacturer = xstrdup(manufacturer);
402 p->hardware = xstrdup(hardware);
406 p->software = xstrdup(software);
410 p->serial = xstrdup(serial);
415 ofproto_set_in_band(struct ofproto *p, bool in_band)
417 if (in_band != (p->in_band != NULL)) {
419 return in_band_create(p, p->dpif, p->switch_status,
420 p->controller->rconn, &p->in_band);
422 ofproto_set_discovery(p, false, NULL, true);
423 in_band_destroy(p->in_band);
426 rconn_reconnect(p->controller->rconn);
432 ofproto_set_discovery(struct ofproto *p, bool discovery,
433 const char *re, bool update_resolv_conf)
435 if (discovery != (p->discovery != NULL)) {
437 int error = ofproto_set_in_band(p, true);
441 error = discovery_create(re, update_resolv_conf,
442 p->dpif, p->switch_status,
448 discovery_destroy(p->discovery);
451 rconn_disconnect(p->controller->rconn);
452 } else if (discovery) {
453 discovery_set_update_resolv_conf(p->discovery, update_resolv_conf);
454 return discovery_set_accept_controller_re(p->discovery, re);
460 ofproto_set_controller(struct ofproto *ofproto, const char *controller)
462 if (ofproto->discovery) {
464 } else if (controller) {
465 if (strcmp(rconn_get_name(ofproto->controller->rconn), controller)) {
466 return rconn_connect(ofproto->controller->rconn, controller);
471 rconn_disconnect(ofproto->controller->rconn);
477 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
478 const struct svec *svec)
480 struct pvconn **pvconns = *pvconnsp;
481 size_t n_pvconns = *n_pvconnsp;
485 for (i = 0; i < n_pvconns; i++) {
486 pvconn_close(pvconns[i]);
490 pvconns = xmalloc(svec->n * sizeof *pvconns);
492 for (i = 0; i < svec->n; i++) {
493 const char *name = svec->names[i];
494 struct pvconn *pvconn;
497 error = pvconn_open(name, &pvconn);
499 pvconns[n_pvconns++] = pvconn;
501 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
509 *n_pvconnsp = n_pvconns;
515 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
517 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
521 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
523 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
527 ofproto_set_netflow(struct ofproto *ofproto,
528 const struct netflow_options *nf_options)
530 if (nf_options && nf_options->collectors.n) {
531 if (!ofproto->netflow) {
532 ofproto->netflow = netflow_create();
534 return netflow_set_options(ofproto->netflow, nf_options);
536 netflow_destroy(ofproto->netflow);
537 ofproto->netflow = NULL;
543 ofproto_set_failure(struct ofproto *ofproto, bool fail_open)
546 struct rconn *rconn = ofproto->controller->rconn;
547 int trigger_duration = rconn_get_probe_interval(rconn) * 3;
548 if (!ofproto->fail_open) {
549 ofproto->fail_open = fail_open_create(ofproto, trigger_duration,
550 ofproto->switch_status,
553 fail_open_set_trigger_duration(ofproto->fail_open,
557 fail_open_destroy(ofproto->fail_open);
558 ofproto->fail_open = NULL;
563 ofproto_set_rate_limit(struct ofproto *ofproto,
564 int rate_limit, int burst_limit)
566 if (rate_limit > 0) {
567 if (!ofproto->miss_sched) {
568 ofproto->miss_sched = pinsched_create(rate_limit, burst_limit,
569 ofproto->switch_status);
570 ofproto->action_sched = pinsched_create(rate_limit, burst_limit,
573 pinsched_set_limits(ofproto->miss_sched, rate_limit, burst_limit);
574 pinsched_set_limits(ofproto->action_sched,
575 rate_limit, burst_limit);
578 pinsched_destroy(ofproto->miss_sched);
579 ofproto->miss_sched = NULL;
580 pinsched_destroy(ofproto->action_sched);
581 ofproto->action_sched = NULL;
586 ofproto_set_stp(struct ofproto *ofproto UNUSED, bool enable_stp)
590 VLOG_WARN("STP is not yet implemented");
598 ofproto_get_datapath_id(const struct ofproto *ofproto)
600 return ofproto->datapath_id;
604 ofproto_get_mgmt_id(const struct ofproto *ofproto)
606 return ofproto->mgmt_id;
610 ofproto_get_probe_interval(const struct ofproto *ofproto)
612 return rconn_get_probe_interval(ofproto->controller->rconn);
616 ofproto_get_max_backoff(const struct ofproto *ofproto)
618 return rconn_get_max_backoff(ofproto->controller->rconn);
622 ofproto_get_in_band(const struct ofproto *ofproto)
624 return ofproto->in_band != NULL;
628 ofproto_get_discovery(const struct ofproto *ofproto)
630 return ofproto->discovery != NULL;
634 ofproto_get_controller(const struct ofproto *ofproto)
636 return rconn_get_name(ofproto->controller->rconn);
640 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
644 for (i = 0; i < ofproto->n_listeners; i++) {
645 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
650 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
654 for (i = 0; i < ofproto->n_snoops; i++) {
655 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
660 ofproto_destroy(struct ofproto *p)
662 struct ofconn *ofconn, *next_ofconn;
663 struct ofport *ofport;
664 unsigned int port_no;
671 ofproto_flush_flows(p);
672 classifier_destroy(&p->cls);
674 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
676 ofconn_destroy(ofconn);
680 netdev_monitor_destroy(p->netdev_monitor);
681 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
684 shash_destroy(&p->port_by_name);
686 switch_status_destroy(p->switch_status);
687 in_band_destroy(p->in_band);
688 discovery_destroy(p->discovery);
689 fail_open_destroy(p->fail_open);
690 pinsched_destroy(p->miss_sched);
691 pinsched_destroy(p->action_sched);
692 netflow_destroy(p->netflow);
694 switch_status_unregister(p->ss_cat);
696 for (i = 0; i < p->n_listeners; i++) {
697 pvconn_close(p->listeners[i]);
701 for (i = 0; i < p->n_snoops; i++) {
702 pvconn_close(p->snoops[i]);
706 mac_learning_destroy(p->ml);
712 ofproto_run(struct ofproto *p)
714 int error = ofproto_run1(p);
716 error = ofproto_run2(p, false);
722 process_port_change(struct ofproto *ofproto, int error, char *devname)
724 if (error == ENOBUFS) {
725 reinit_ports(ofproto);
727 update_port(ofproto, devname);
733 ofproto_run1(struct ofproto *p)
735 struct ofconn *ofconn, *next_ofconn;
740 if (shash_is_empty(&p->port_by_name)) {
744 for (i = 0; i < 50; i++) {
748 error = dpif_recv(p->dpif, &buf);
750 if (error == ENODEV) {
751 /* Someone destroyed the datapath behind our back. The caller
752 * better destroy us and give up, because we're just going to
753 * spin from here on out. */
754 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
755 VLOG_ERR_RL(&rl, "%s: datapath was destroyed externally",
762 handle_odp_msg(p, buf);
765 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
766 process_port_change(p, error, devname);
768 while ((error = netdev_monitor_poll(p->netdev_monitor,
769 &devname)) != EAGAIN) {
770 process_port_change(p, error, devname);
774 in_band_run(p->in_band);
777 char *controller_name;
778 if (rconn_is_connectivity_questionable(p->controller->rconn)) {
779 discovery_question_connectivity(p->discovery);
781 if (discovery_run(p->discovery, &controller_name)) {
782 if (controller_name) {
783 rconn_connect(p->controller->rconn, controller_name);
785 rconn_disconnect(p->controller->rconn);
789 pinsched_run(p->miss_sched, send_packet_in_miss, p);
790 pinsched_run(p->action_sched, send_packet_in_action, p);
792 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
794 ofconn_run(ofconn, p);
797 /* Fail-open maintenance. Do this after processing the ofconns since
798 * fail-open checks the status of the controller rconn. */
800 fail_open_run(p->fail_open);
803 for (i = 0; i < p->n_listeners; i++) {
807 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
809 ofconn_create(p, rconn_new_from_vconn("passive", vconn));
810 } else if (retval != EAGAIN) {
811 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
815 for (i = 0; i < p->n_snoops; i++) {
819 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
821 rconn_add_monitor(p->controller->rconn, vconn);
822 } else if (retval != EAGAIN) {
823 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
827 if (time_msec() >= p->next_expiration) {
828 COVERAGE_INC(ofproto_expiration);
829 p->next_expiration = time_msec() + 1000;
832 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
834 /* Let the hook know that we're at a stable point: all outstanding data
835 * in existing flows has been accounted to the account_cb. Thus, the
836 * hook can now reasonably do operations that depend on having accurate
837 * flow volume accounting (currently, that's just bond rebalancing). */
838 if (p->ofhooks->account_checkpoint_cb) {
839 p->ofhooks->account_checkpoint_cb(p->aux);
844 netflow_run(p->netflow);
850 struct revalidate_cbdata {
851 struct ofproto *ofproto;
852 bool revalidate_all; /* Revalidate all exact-match rules? */
853 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
854 struct tag_set revalidate_set; /* Set of tags to revalidate. */
858 ofproto_run2(struct ofproto *p, bool revalidate_all)
860 if (p->need_revalidate || revalidate_all
861 || !tag_set_is_empty(&p->revalidate_set)) {
862 struct revalidate_cbdata cbdata;
864 cbdata.revalidate_all = revalidate_all;
865 cbdata.revalidate_subrules = p->need_revalidate;
866 cbdata.revalidate_set = p->revalidate_set;
867 tag_set_init(&p->revalidate_set);
868 COVERAGE_INC(ofproto_revalidate);
869 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
870 p->need_revalidate = false;
877 ofproto_wait(struct ofproto *p)
879 struct ofconn *ofconn;
882 dpif_recv_wait(p->dpif);
883 dpif_port_poll_wait(p->dpif);
884 netdev_monitor_poll_wait(p->netdev_monitor);
885 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
889 in_band_wait(p->in_band);
892 discovery_wait(p->discovery);
895 fail_open_wait(p->fail_open);
897 pinsched_wait(p->miss_sched);
898 pinsched_wait(p->action_sched);
899 if (!tag_set_is_empty(&p->revalidate_set)) {
900 poll_immediate_wake();
902 if (p->need_revalidate) {
903 /* Shouldn't happen, but if it does just go around again. */
904 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
905 poll_immediate_wake();
906 } else if (p->next_expiration != LLONG_MAX) {
907 poll_timer_wait(p->next_expiration - time_msec());
909 for (i = 0; i < p->n_listeners; i++) {
910 pvconn_wait(p->listeners[i]);
912 for (i = 0; i < p->n_snoops; i++) {
913 pvconn_wait(p->snoops[i]);
918 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
920 tag_set_add(&ofproto->revalidate_set, tag);
924 ofproto_get_revalidate_set(struct ofproto *ofproto)
926 return &ofproto->revalidate_set;
930 ofproto_is_alive(const struct ofproto *p)
932 return p->discovery || rconn_is_alive(p->controller->rconn);
936 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
937 const union ofp_action *actions, size_t n_actions,
938 const struct ofpbuf *packet)
940 struct odp_actions odp_actions;
943 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
949 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
951 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
952 odp_actions.n_actions, packet);
957 ofproto_add_flow(struct ofproto *p,
958 const flow_t *flow, uint32_t wildcards, unsigned int priority,
959 const union ofp_action *actions, size_t n_actions,
963 rule = rule_create(p, NULL, actions, n_actions,
964 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */, 0);
965 cls_rule_from_flow(&rule->cr, flow, wildcards, priority);
966 rule_insert(p, rule, NULL, 0);
970 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
971 uint32_t wildcards, unsigned int priority)
975 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
979 rule_remove(ofproto, rule);
984 destroy_rule(struct cls_rule *rule_, void *ofproto_)
986 struct rule *rule = rule_from_cls_rule(rule_);
987 struct ofproto *ofproto = ofproto_;
989 /* Mark the flow as not installed, even though it might really be
990 * installed, so that rule_remove() doesn't bother trying to uninstall it.
991 * There is no point in uninstalling it individually since we are about to
992 * blow away all the flows with dpif_flow_flush(). */
993 rule->installed = false;
995 rule_remove(ofproto, rule);
999 ofproto_flush_flows(struct ofproto *ofproto)
1001 COVERAGE_INC(ofproto_flush);
1002 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1003 dpif_flow_flush(ofproto->dpif);
1004 if (ofproto->in_band) {
1005 in_band_flushed(ofproto->in_band);
1007 if (ofproto->fail_open) {
1008 fail_open_flushed(ofproto->fail_open);
1013 reinit_ports(struct ofproto *p)
1015 struct svec devnames;
1016 struct ofport *ofport;
1017 unsigned int port_no;
1018 struct odp_port *odp_ports;
1022 svec_init(&devnames);
1023 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1024 svec_add (&devnames, (char *) ofport->opp.name);
1026 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1027 for (i = 0; i < n_odp_ports; i++) {
1028 svec_add (&devnames, odp_ports[i].devname);
1032 svec_sort_unique(&devnames);
1033 for (i = 0; i < devnames.n; i++) {
1034 update_port(p, devnames.names[i]);
1036 svec_destroy(&devnames);
1040 refresh_port_group(struct ofproto *p, unsigned int group)
1044 struct ofport *port;
1045 unsigned int port_no;
1047 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1049 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1051 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1052 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1053 ports[n_ports++] = port_no;
1056 dpif_port_group_set(p->dpif, group, ports, n_ports);
1061 refresh_port_groups(struct ofproto *p)
1063 refresh_port_group(p, DP_GROUP_FLOOD);
1064 refresh_port_group(p, DP_GROUP_ALL);
1067 static struct ofport *
1068 make_ofport(const struct odp_port *odp_port)
1070 struct netdev_options netdev_options;
1071 enum netdev_flags flags;
1072 struct ofport *ofport;
1073 struct netdev *netdev;
1077 memset(&netdev_options, 0, sizeof netdev_options);
1078 netdev_options.name = odp_port->devname;
1079 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1080 netdev_options.may_open = true;
1082 error = netdev_open(&netdev_options, &netdev);
1084 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1085 "cannot be opened (%s)",
1086 odp_port->devname, odp_port->port,
1087 odp_port->devname, strerror(error));
1091 ofport = xmalloc(sizeof *ofport);
1092 ofport->netdev = netdev;
1093 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1094 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1095 memcpy(ofport->opp.name, odp_port->devname,
1096 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1097 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1099 netdev_get_flags(netdev, &flags);
1100 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1102 netdev_get_carrier(netdev, &carrier);
1103 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1105 netdev_get_features(netdev,
1106 &ofport->opp.curr, &ofport->opp.advertised,
1107 &ofport->opp.supported, &ofport->opp.peer);
1112 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1114 if (port_array_get(&p->ports, odp_port->port)) {
1115 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1118 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1119 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1128 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1130 const struct ofp_phy_port *a = &a_->opp;
1131 const struct ofp_phy_port *b = &b_->opp;
1133 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1134 return (a->port_no == b->port_no
1135 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1136 && !strcmp((char *) a->name, (char *) b->name)
1137 && a->state == b->state
1138 && a->config == b->config
1139 && a->curr == b->curr
1140 && a->advertised == b->advertised
1141 && a->supported == b->supported
1142 && a->peer == b->peer);
1146 send_port_status(struct ofproto *p, const struct ofport *ofport,
1149 /* XXX Should limit the number of queued port status change messages. */
1150 struct ofconn *ofconn;
1151 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1152 struct ofp_port_status *ops;
1155 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1156 ops->reason = reason;
1157 ops->desc = ofport->opp;
1158 hton_ofp_phy_port(&ops->desc);
1159 queue_tx(b, ofconn, NULL);
1161 if (p->ofhooks->port_changed_cb) {
1162 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1167 ofport_install(struct ofproto *p, struct ofport *ofport)
1169 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1170 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no),
1172 shash_add(&p->port_by_name, (char *) ofport->opp.name, ofport);
1176 ofport_remove(struct ofproto *p, struct ofport *ofport)
1178 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1179 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no), NULL);
1180 shash_delete(&p->port_by_name,
1181 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1185 ofport_free(struct ofport *ofport)
1188 netdev_close(ofport->netdev);
1194 update_port(struct ofproto *p, const char *devname)
1196 struct odp_port odp_port;
1197 struct ofport *old_ofport;
1198 struct ofport *new_ofport;
1201 COVERAGE_INC(ofproto_update_port);
1203 /* Query the datapath for port information. */
1204 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1206 /* Find the old ofport. */
1207 old_ofport = shash_find_data(&p->port_by_name, devname);
1210 /* There's no port named 'devname' but there might be a port with
1211 * the same port number. This could happen if a port is deleted
1212 * and then a new one added in its place very quickly, or if a port
1213 * is renamed. In the former case we want to send an OFPPR_DELETE
1214 * and an OFPPR_ADD, and in the latter case we want to send a
1215 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1216 * the old port's ifindex against the new port, or perhaps less
1217 * reliably but more portably by comparing the old port's MAC
1218 * against the new port's MAC. However, this code isn't that smart
1219 * and always sends an OFPPR_MODIFY (XXX). */
1220 old_ofport = port_array_get(&p->ports, odp_port.port);
1222 } else if (error != ENOENT && error != ENODEV) {
1223 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1224 "%s", strerror(error));
1228 /* Create a new ofport. */
1229 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1231 /* Eliminate a few pathological cases. */
1232 if (!old_ofport && !new_ofport) {
1234 } else if (old_ofport && new_ofport) {
1235 /* Most of the 'config' bits are OpenFlow soft state, but
1236 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1237 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1238 * leaves the other bits 0.) */
1239 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1241 if (ofport_equal(old_ofport, new_ofport)) {
1242 /* False alarm--no change. */
1243 ofport_free(new_ofport);
1248 /* Now deal with the normal cases. */
1250 ofport_remove(p, old_ofport);
1253 ofport_install(p, new_ofport);
1255 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1256 (!old_ofport ? OFPPR_ADD
1257 : !new_ofport ? OFPPR_DELETE
1259 ofport_free(old_ofport);
1261 /* Update port groups. */
1262 refresh_port_groups(p);
1266 init_ports(struct ofproto *p)
1268 struct odp_port *ports;
1273 error = dpif_port_list(p->dpif, &ports, &n_ports);
1278 for (i = 0; i < n_ports; i++) {
1279 const struct odp_port *odp_port = &ports[i];
1280 if (!ofport_conflicts(p, odp_port)) {
1281 struct ofport *ofport = make_ofport(odp_port);
1283 ofport_install(p, ofport);
1288 refresh_port_groups(p);
1292 static struct ofconn *
1293 ofconn_create(struct ofproto *p, struct rconn *rconn)
1295 struct ofconn *ofconn = xmalloc(sizeof *ofconn);
1296 list_push_back(&p->all_conns, &ofconn->node);
1297 ofconn->rconn = rconn;
1298 ofconn->pktbuf = NULL;
1299 ofconn->send_flow_exp = false;
1300 ofconn->miss_send_len = 0;
1301 ofconn->packet_in_counter = rconn_packet_counter_create ();
1302 ofconn->reply_counter = rconn_packet_counter_create ();
1307 ofconn_destroy(struct ofconn *ofconn)
1309 list_remove(&ofconn->node);
1310 rconn_destroy(ofconn->rconn);
1311 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1312 rconn_packet_counter_destroy(ofconn->reply_counter);
1313 pktbuf_destroy(ofconn->pktbuf);
1318 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1322 rconn_run(ofconn->rconn);
1324 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1325 /* Limit the number of iterations to prevent other tasks from
1327 for (iteration = 0; iteration < 50; iteration++) {
1328 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1333 fail_open_maybe_recover(p->fail_open);
1335 handle_openflow(ofconn, p, of_msg);
1336 ofpbuf_delete(of_msg);
1340 if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
1341 ofconn_destroy(ofconn);
1346 ofconn_wait(struct ofconn *ofconn)
1348 rconn_run_wait(ofconn->rconn);
1349 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1350 rconn_recv_wait(ofconn->rconn);
1352 COVERAGE_INC(ofproto_ofconn_stuck);
1356 /* Caller is responsible for initializing the 'cr' member of the returned
1358 static struct rule *
1359 rule_create(struct ofproto *ofproto, struct rule *super,
1360 const union ofp_action *actions, size_t n_actions,
1361 uint16_t idle_timeout, uint16_t hard_timeout)
1363 struct rule *rule = xzalloc(sizeof *rule);
1364 rule->idle_timeout = idle_timeout;
1365 rule->hard_timeout = hard_timeout;
1366 rule->used = rule->created = time_msec();
1367 rule->super = super;
1369 list_push_back(&super->list, &rule->list);
1371 list_init(&rule->list);
1373 rule->n_actions = n_actions;
1374 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1375 netflow_flow_clear(&rule->nf_flow);
1376 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1381 static struct rule *
1382 rule_from_cls_rule(const struct cls_rule *cls_rule)
1384 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1388 rule_free(struct rule *rule)
1390 free(rule->actions);
1391 free(rule->odp_actions);
1395 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1396 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1397 * through all of its subrules and revalidates them, destroying any that no
1398 * longer has a super-rule (which is probably all of them).
1400 * Before calling this function, the caller must make have removed 'rule' from
1401 * the classifier. If 'rule' is an exact-match rule, the caller is also
1402 * responsible for ensuring that it has been uninstalled from the datapath. */
1404 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1407 struct rule *subrule, *next;
1408 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1409 revalidate_rule(ofproto, subrule);
1412 list_remove(&rule->list);
1418 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1420 const union ofp_action *oa;
1421 struct actions_iterator i;
1423 if (out_port == htons(OFPP_NONE)) {
1426 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1427 oa = actions_next(&i)) {
1428 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1435 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1436 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1438 * The flow that 'packet' actually contains does not need to actually match
1439 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1440 * the packet and byte counters for 'rule' will be credited for the packet sent
1441 * out whether or not the packet actually matches 'rule'.
1443 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1444 * the caller must already have accurately composed ODP actions for it given
1445 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1446 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1447 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1448 * actions and apply them to 'packet'. */
1450 rule_execute(struct ofproto *ofproto, struct rule *rule,
1451 struct ofpbuf *packet, const flow_t *flow)
1453 const union odp_action *actions;
1455 struct odp_actions a;
1457 /* Grab or compose the ODP actions.
1459 * The special case for an exact-match 'rule' where 'flow' is not the
1460 * rule's flow is important to avoid, e.g., sending a packet out its input
1461 * port simply because the ODP actions were composed for the wrong
1463 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1464 struct rule *super = rule->super ? rule->super : rule;
1465 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1466 packet, &a, NULL, 0, NULL)) {
1469 actions = a.actions;
1470 n_actions = a.n_actions;
1472 actions = rule->odp_actions;
1473 n_actions = rule->n_odp_actions;
1476 /* Execute the ODP actions. */
1477 if (!dpif_execute(ofproto->dpif, flow->in_port,
1478 actions, n_actions, packet)) {
1479 struct odp_flow_stats stats;
1480 flow_extract_stats(flow, packet, &stats);
1481 update_stats(ofproto, rule, &stats);
1482 rule->used = time_msec();
1483 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
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(ofproto, 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,
1564 &rule->nf_flow.output_iface);
1566 actions_len = a.n_actions * sizeof *a.actions;
1567 if (rule->n_odp_actions != a.n_actions
1568 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1569 COVERAGE_INC(ofproto_odp_unchanged);
1570 free(rule->odp_actions);
1571 rule->n_odp_actions = a.n_actions;
1572 rule->odp_actions = xmemdup(a.actions, actions_len);
1580 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1581 struct odp_flow_put *put)
1583 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1584 put->flow.key = rule->cr.flow;
1585 put->flow.actions = rule->odp_actions;
1586 put->flow.n_actions = rule->n_odp_actions;
1588 return dpif_flow_put(ofproto->dpif, put);
1592 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1594 assert(!rule->cr.wc.wildcards);
1596 if (rule->may_install) {
1597 struct odp_flow_put put;
1598 if (!do_put_flow(p, rule,
1599 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1601 rule->installed = true;
1602 if (displaced_rule) {
1603 update_stats(p, displaced_rule, &put.flow.stats);
1604 rule_post_uninstall(p, displaced_rule);
1607 } else if (displaced_rule) {
1608 rule_uninstall(p, displaced_rule);
1613 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1615 if (rule->installed) {
1616 struct odp_flow_put put;
1617 COVERAGE_INC(ofproto_dp_missed);
1618 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1620 rule_install(ofproto, rule, NULL);
1625 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1627 bool actions_changed;
1628 uint16_t new_out_iface, old_out_iface;
1630 old_out_iface = rule->nf_flow.output_iface;
1631 actions_changed = rule_make_actions(ofproto, rule, NULL);
1633 if (rule->may_install) {
1634 if (rule->installed) {
1635 if (actions_changed) {
1636 struct odp_flow_put put;
1637 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
1638 | ODPPF_ZERO_STATS, &put);
1639 update_stats(ofproto, rule, &put.flow.stats);
1641 /* Temporarily set the old output iface so that NetFlow
1642 * messages have the correct output interface for the old
1644 new_out_iface = rule->nf_flow.output_iface;
1645 rule->nf_flow.output_iface = old_out_iface;
1646 rule_post_uninstall(ofproto, rule);
1647 rule->nf_flow.output_iface = new_out_iface;
1650 rule_install(ofproto, rule, NULL);
1653 rule_uninstall(ofproto, rule);
1658 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1660 uint64_t total_bytes = rule->byte_count + extra_bytes;
1662 if (ofproto->ofhooks->account_flow_cb
1663 && total_bytes > rule->accounted_bytes)
1665 ofproto->ofhooks->account_flow_cb(
1666 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1667 total_bytes - rule->accounted_bytes, ofproto->aux);
1668 rule->accounted_bytes = total_bytes;
1673 rule_uninstall(struct ofproto *p, struct rule *rule)
1675 assert(!rule->cr.wc.wildcards);
1676 if (rule->installed) {
1677 struct odp_flow odp_flow;
1679 odp_flow.key = rule->cr.flow;
1680 odp_flow.actions = NULL;
1681 odp_flow.n_actions = 0;
1682 if (!dpif_flow_del(p->dpif, &odp_flow)) {
1683 update_stats(p, rule, &odp_flow.stats);
1685 rule->installed = false;
1687 rule_post_uninstall(p, rule);
1692 is_controller_rule(struct rule *rule)
1694 /* If the only action is send to the controller then don't report
1695 * NetFlow expiration messages since it is just part of the control
1696 * logic for the network and not real traffic. */
1698 if (rule && rule->super) {
1699 struct rule *super = rule->super;
1701 return super->n_actions == 1 &&
1702 super->actions[0].type == htons(OFPAT_OUTPUT) &&
1703 super->actions[0].output.port == htons(OFPP_CONTROLLER);
1710 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1712 struct rule *super = rule->super;
1714 rule_account(ofproto, rule, 0);
1716 if (ofproto->netflow && !is_controller_rule(rule)) {
1717 struct ofexpired expired;
1718 expired.flow = rule->cr.flow;
1719 expired.packet_count = rule->packet_count;
1720 expired.byte_count = rule->byte_count;
1721 expired.used = rule->used;
1722 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
1725 super->packet_count += rule->packet_count;
1726 super->byte_count += rule->byte_count;
1728 /* Reset counters to prevent double counting if the rule ever gets
1730 rule->packet_count = 0;
1731 rule->byte_count = 0;
1732 rule->accounted_bytes = 0;
1734 netflow_flow_clear(&rule->nf_flow);
1739 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1740 struct rconn_packet_counter *counter)
1742 update_openflow_length(msg);
1743 if (rconn_send(ofconn->rconn, msg, counter)) {
1749 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1750 int error, const void *data, size_t len)
1753 struct ofp_error_msg *oem;
1755 if (!(error >> 16)) {
1756 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1761 COVERAGE_INC(ofproto_error);
1762 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1763 oh ? oh->xid : 0, &buf);
1764 oem->type = htons((unsigned int) error >> 16);
1765 oem->code = htons(error & 0xffff);
1766 memcpy(oem->data, data, len);
1767 queue_tx(buf, ofconn, ofconn->reply_counter);
1771 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1774 size_t oh_length = ntohs(oh->length);
1775 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1779 hton_ofp_phy_port(struct ofp_phy_port *opp)
1781 opp->port_no = htons(opp->port_no);
1782 opp->config = htonl(opp->config);
1783 opp->state = htonl(opp->state);
1784 opp->curr = htonl(opp->curr);
1785 opp->advertised = htonl(opp->advertised);
1786 opp->supported = htonl(opp->supported);
1787 opp->peer = htonl(opp->peer);
1791 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
1793 struct ofp_header *rq = oh;
1794 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
1799 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
1800 struct ofp_header *oh)
1802 struct ofp_switch_features *osf;
1804 unsigned int port_no;
1805 struct ofport *port;
1807 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1808 osf->datapath_id = htonll(p->datapath_id);
1809 osf->n_buffers = htonl(pktbuf_capacity());
1811 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1812 OFPC_PORT_STATS | OFPC_MULTI_PHY_TX);
1813 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1814 (1u << OFPAT_SET_VLAN_VID) |
1815 (1u << OFPAT_SET_VLAN_PCP) |
1816 (1u << OFPAT_STRIP_VLAN) |
1817 (1u << OFPAT_SET_DL_SRC) |
1818 (1u << OFPAT_SET_DL_DST) |
1819 (1u << OFPAT_SET_NW_SRC) |
1820 (1u << OFPAT_SET_NW_DST) |
1821 (1u << OFPAT_SET_TP_SRC) |
1822 (1u << OFPAT_SET_TP_DST));
1824 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1825 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1828 queue_tx(buf, ofconn, ofconn->reply_counter);
1833 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1834 struct ofp_header *oh)
1837 struct ofp_switch_config *osc;
1841 /* Figure out flags. */
1842 dpif_get_drop_frags(p->dpif, &drop_frags);
1843 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1844 if (ofconn->send_flow_exp) {
1845 flags |= OFPC_SEND_FLOW_EXP;
1849 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1850 osc->flags = htons(flags);
1851 osc->miss_send_len = htons(ofconn->miss_send_len);
1852 queue_tx(buf, ofconn, ofconn->reply_counter);
1858 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1859 struct ofp_switch_config *osc)
1864 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1868 flags = ntohs(osc->flags);
1870 ofconn->send_flow_exp = (flags & OFPC_SEND_FLOW_EXP) != 0;
1872 if (ofconn == p->controller) {
1873 switch (flags & OFPC_FRAG_MASK) {
1874 case OFPC_FRAG_NORMAL:
1875 dpif_set_drop_frags(p->dpif, false);
1877 case OFPC_FRAG_DROP:
1878 dpif_set_drop_frags(p->dpif, true);
1881 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1887 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1888 if (ntohs(osc->miss_send_len) != 0) {
1889 ofconn->pktbuf = pktbuf_create();
1891 pktbuf_destroy(ofconn->pktbuf);
1895 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1901 add_output_group_action(struct odp_actions *actions, uint16_t group,
1902 uint16_t *nf_output_iface)
1904 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1906 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
1907 *nf_output_iface = NF_OUT_FLOOD;
1912 add_controller_action(struct odp_actions *actions,
1913 const struct ofp_action_output *oao)
1915 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
1916 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
1919 struct action_xlate_ctx {
1921 const flow_t *flow; /* Flow to which these actions correspond. */
1922 int recurse; /* Recursion level, via xlate_table_action. */
1923 struct ofproto *ofproto;
1924 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
1925 * null pointer if we are revalidating
1926 * without a packet to refer to. */
1929 struct odp_actions *out; /* Datapath actions. */
1930 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
1931 bool may_set_up_flow; /* True ordinarily; false if the actions must
1932 * be reassessed for every packet. */
1933 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
1936 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
1937 struct action_xlate_ctx *ctx);
1940 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
1942 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
1945 if (ofport->opp.config & OFPPC_NO_FWD) {
1946 /* Forwarding disabled on port. */
1951 * We don't have an ofport record for this port, but it doesn't hurt to
1952 * allow forwarding to it anyhow. Maybe such a port will appear later
1953 * and we're pre-populating the flow table.
1957 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
1958 ctx->nf_output_iface = port;
1961 static struct rule *
1962 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
1965 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
1967 /* The rule we found might not be valid, since we could be in need of
1968 * revalidation. If it is not valid, don't return it. */
1971 && ofproto->need_revalidate
1972 && !revalidate_rule(ofproto, rule)) {
1973 COVERAGE_INC(ofproto_invalidated);
1981 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
1983 if (!ctx->recurse) {
1988 flow.in_port = in_port;
1990 rule = lookup_valid_rule(ctx->ofproto, &flow);
1997 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2004 xlate_output_action(struct action_xlate_ctx *ctx,
2005 const struct ofp_action_output *oao)
2008 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2010 ctx->nf_output_iface = NF_OUT_DROP;
2012 switch (ntohs(oao->port)) {
2014 add_output_action(ctx, ctx->flow->in_port);
2017 xlate_table_action(ctx, ctx->flow->in_port);
2020 if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet,
2021 ctx->out, ctx->tags,
2022 &ctx->nf_output_iface,
2023 ctx->ofproto->aux)) {
2024 COVERAGE_INC(ofproto_uninstallable);
2025 ctx->may_set_up_flow = false;
2029 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2030 &ctx->nf_output_iface);
2033 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2035 case OFPP_CONTROLLER:
2036 add_controller_action(ctx->out, oao);
2039 add_output_action(ctx, ODPP_LOCAL);
2042 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2043 if (odp_port != ctx->flow->in_port) {
2044 add_output_action(ctx, odp_port);
2049 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2050 ctx->nf_output_iface = NF_OUT_FLOOD;
2051 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2052 ctx->nf_output_iface = prev_nf_output_iface;
2053 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2054 ctx->nf_output_iface != NF_OUT_FLOOD) {
2055 ctx->nf_output_iface = NF_OUT_MULTI;
2060 xlate_nicira_action(struct action_xlate_ctx *ctx,
2061 const struct nx_action_header *nah)
2063 const struct nx_action_resubmit *nar;
2064 int subtype = ntohs(nah->subtype);
2066 assert(nah->vendor == htonl(NX_VENDOR_ID));
2068 case NXAST_RESUBMIT:
2069 nar = (const struct nx_action_resubmit *) nah;
2070 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2074 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2080 do_xlate_actions(const union ofp_action *in, size_t n_in,
2081 struct action_xlate_ctx *ctx)
2083 struct actions_iterator iter;
2084 const union ofp_action *ia;
2085 const struct ofport *port;
2087 port = port_array_get(&ctx->ofproto->ports, ctx->flow->in_port);
2088 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2089 port->opp.config & (eth_addr_equals(ctx->flow->dl_dst, stp_eth_addr)
2090 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2091 /* Drop this flow. */
2095 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2096 uint16_t type = ntohs(ia->type);
2097 union odp_action *oa;
2101 xlate_output_action(ctx, &ia->output);
2104 case OFPAT_SET_VLAN_VID:
2105 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2106 oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2109 case OFPAT_SET_VLAN_PCP:
2110 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2111 oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2114 case OFPAT_STRIP_VLAN:
2115 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2118 case OFPAT_SET_DL_SRC:
2119 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2120 memcpy(oa->dl_addr.dl_addr,
2121 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2124 case OFPAT_SET_DL_DST:
2125 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2126 memcpy(oa->dl_addr.dl_addr,
2127 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2130 case OFPAT_SET_NW_SRC:
2131 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2132 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2135 case OFPAT_SET_NW_DST:
2136 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2137 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2140 case OFPAT_SET_TP_SRC:
2141 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2142 oa->tp_port.tp_port = ia->tp_port.tp_port;
2145 case OFPAT_SET_TP_DST:
2146 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2147 oa->tp_port.tp_port = ia->tp_port.tp_port;
2151 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2155 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2162 xlate_actions(const union ofp_action *in, size_t n_in,
2163 const flow_t *flow, struct ofproto *ofproto,
2164 const struct ofpbuf *packet,
2165 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2166 uint16_t *nf_output_iface)
2168 tag_type no_tags = 0;
2169 struct action_xlate_ctx ctx;
2170 COVERAGE_INC(ofproto_ofp2odp);
2171 odp_actions_init(out);
2174 ctx.ofproto = ofproto;
2175 ctx.packet = packet;
2177 ctx.tags = tags ? tags : &no_tags;
2178 ctx.may_set_up_flow = true;
2179 ctx.nf_output_iface = NF_OUT_DROP;
2180 do_xlate_actions(in, n_in, &ctx);
2182 /* Check with in-band control to see if we're allowed to set up this
2184 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2185 ctx.may_set_up_flow = false;
2188 if (may_set_up_flow) {
2189 *may_set_up_flow = ctx.may_set_up_flow;
2191 if (nf_output_iface) {
2192 *nf_output_iface = ctx.nf_output_iface;
2194 if (odp_actions_overflow(out)) {
2195 odp_actions_init(out);
2196 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2202 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2203 struct ofp_header *oh)
2205 struct ofp_packet_out *opo;
2206 struct ofpbuf payload, *buffer;
2207 struct odp_actions actions;
2213 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2217 opo = (struct ofp_packet_out *) oh;
2219 COVERAGE_INC(ofproto_packet_out);
2220 if (opo->buffer_id != htonl(UINT32_MAX)) {
2221 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2223 if (error || !buffer) {
2231 flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2232 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2233 &flow, p, &payload, &actions, NULL, NULL, NULL);
2238 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2240 ofpbuf_delete(buffer);
2246 update_port_config(struct ofproto *p, struct ofport *port,
2247 uint32_t config, uint32_t mask)
2249 mask &= config ^ port->opp.config;
2250 if (mask & OFPPC_PORT_DOWN) {
2251 if (config & OFPPC_PORT_DOWN) {
2252 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2254 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2257 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2258 if (mask & REVALIDATE_BITS) {
2259 COVERAGE_INC(ofproto_costly_flags);
2260 port->opp.config ^= mask & REVALIDATE_BITS;
2261 p->need_revalidate = true;
2263 #undef REVALIDATE_BITS
2264 if (mask & OFPPC_NO_FLOOD) {
2265 port->opp.config ^= OFPPC_NO_FLOOD;
2266 refresh_port_group(p, DP_GROUP_FLOOD);
2268 if (mask & OFPPC_NO_PACKET_IN) {
2269 port->opp.config ^= OFPPC_NO_PACKET_IN;
2274 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2276 const struct ofp_port_mod *opm;
2277 struct ofport *port;
2280 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2284 opm = (struct ofp_port_mod *) oh;
2286 port = port_array_get(&p->ports,
2287 ofp_port_to_odp_port(ntohs(opm->port_no)));
2289 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2290 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2291 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2293 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2294 if (opm->advertise) {
2295 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2301 static struct ofpbuf *
2302 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2304 struct ofp_stats_reply *osr;
2307 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2308 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2310 osr->flags = htons(0);
2314 static struct ofpbuf *
2315 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2317 return make_stats_reply(request->header.xid, request->type, body_len);
2321 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2323 struct ofpbuf *msg = *msgp;
2324 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2325 if (nbytes + msg->size > UINT16_MAX) {
2326 struct ofp_stats_reply *reply = msg->data;
2327 reply->flags = htons(OFPSF_REPLY_MORE);
2328 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2329 queue_tx(msg, ofconn, ofconn->reply_counter);
2331 return ofpbuf_put_uninit(*msgp, nbytes);
2335 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2336 struct ofp_stats_request *request)
2338 struct ofp_desc_stats *ods;
2341 msg = start_stats_reply(request, sizeof *ods);
2342 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2343 strncpy(ods->mfr_desc, p->manufacturer, sizeof ods->mfr_desc);
2344 strncpy(ods->hw_desc, p->hardware, sizeof ods->hw_desc);
2345 strncpy(ods->sw_desc, p->software, sizeof ods->sw_desc);
2346 strncpy(ods->serial_num, p->serial, sizeof ods->serial_num);
2347 queue_tx(msg, ofconn, ofconn->reply_counter);
2353 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2355 struct rule *rule = rule_from_cls_rule(cls_rule);
2356 int *n_subrules = n_subrules_;
2364 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2365 struct ofp_stats_request *request)
2367 struct ofp_table_stats *ots;
2369 struct odp_stats dpstats;
2370 int n_exact, n_subrules, n_wild;
2372 msg = start_stats_reply(request, sizeof *ots * 2);
2374 /* Count rules of various kinds. */
2376 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2377 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2378 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2381 dpif_get_dp_stats(p->dpif, &dpstats);
2382 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2383 memset(ots, 0, sizeof *ots);
2384 ots->table_id = TABLEID_HASH;
2385 strcpy(ots->name, "hash");
2386 ots->wildcards = htonl(0);
2387 ots->max_entries = htonl(dpstats.max_capacity);
2388 ots->active_count = htonl(n_exact);
2389 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2391 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2393 /* Classifier table. */
2394 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2395 memset(ots, 0, sizeof *ots);
2396 ots->table_id = TABLEID_CLASSIFIER;
2397 strcpy(ots->name, "classifier");
2398 ots->wildcards = htonl(OFPFW_ALL);
2399 ots->max_entries = htonl(65536);
2400 ots->active_count = htonl(n_wild);
2401 ots->lookup_count = htonll(0); /* XXX */
2402 ots->matched_count = htonll(0); /* XXX */
2404 queue_tx(msg, ofconn, ofconn->reply_counter);
2409 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2410 struct ofp_stats_request *request)
2412 struct ofp_port_stats *ops;
2414 struct ofport *port;
2415 unsigned int port_no;
2417 msg = start_stats_reply(request, sizeof *ops * 16);
2418 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2419 struct netdev_stats stats;
2421 /* Intentionally ignore return value, since errors will set 'stats' to
2422 * all-1s, which is correct for OpenFlow, and netdev_get_stats() will
2424 netdev_get_stats(port->netdev, &stats);
2426 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2427 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2428 memset(ops->pad, 0, sizeof ops->pad);
2429 ops->rx_packets = htonll(stats.rx_packets);
2430 ops->tx_packets = htonll(stats.tx_packets);
2431 ops->rx_bytes = htonll(stats.rx_bytes);
2432 ops->tx_bytes = htonll(stats.tx_bytes);
2433 ops->rx_dropped = htonll(stats.rx_dropped);
2434 ops->tx_dropped = htonll(stats.tx_dropped);
2435 ops->rx_errors = htonll(stats.rx_errors);
2436 ops->tx_errors = htonll(stats.tx_errors);
2437 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2438 ops->rx_over_err = htonll(stats.rx_over_errors);
2439 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2440 ops->collisions = htonll(stats.collisions);
2443 queue_tx(msg, ofconn, ofconn->reply_counter);
2447 struct flow_stats_cbdata {
2448 struct ofproto *ofproto;
2449 struct ofconn *ofconn;
2455 query_stats(struct ofproto *p, struct rule *rule,
2456 uint64_t *packet_countp, uint64_t *byte_countp)
2458 uint64_t packet_count, byte_count;
2459 struct rule *subrule;
2460 struct odp_flow *odp_flows;
2463 packet_count = rule->packet_count;
2464 byte_count = rule->byte_count;
2466 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2467 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
2468 if (rule->cr.wc.wildcards) {
2470 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2471 odp_flows[i++].key = subrule->cr.flow;
2472 packet_count += subrule->packet_count;
2473 byte_count += subrule->byte_count;
2476 odp_flows[0].key = rule->cr.flow;
2479 packet_count = rule->packet_count;
2480 byte_count = rule->byte_count;
2481 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2483 for (i = 0; i < n_odp_flows; i++) {
2484 struct odp_flow *odp_flow = &odp_flows[i];
2485 packet_count += odp_flow->stats.n_packets;
2486 byte_count += odp_flow->stats.n_bytes;
2491 *packet_countp = packet_count;
2492 *byte_countp = byte_count;
2496 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2498 struct rule *rule = rule_from_cls_rule(rule_);
2499 struct flow_stats_cbdata *cbdata = cbdata_;
2500 struct ofp_flow_stats *ofs;
2501 uint64_t packet_count, byte_count;
2502 size_t act_len, len;
2504 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2508 act_len = sizeof *rule->actions * rule->n_actions;
2509 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2511 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2513 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2514 ofs->length = htons(len);
2515 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2517 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match);
2518 ofs->duration = htonl((time_msec() - rule->created) / 1000);
2519 ofs->priority = htons(rule->cr.priority);
2520 ofs->idle_timeout = htons(rule->idle_timeout);
2521 ofs->hard_timeout = htons(rule->hard_timeout);
2522 memset(ofs->pad2, 0, sizeof ofs->pad2);
2523 ofs->packet_count = htonll(packet_count);
2524 ofs->byte_count = htonll(byte_count);
2525 memcpy(ofs->actions, rule->actions, act_len);
2529 table_id_to_include(uint8_t table_id)
2531 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2532 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2533 : table_id == 0xff ? CLS_INC_ALL
2538 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2539 const struct ofp_stats_request *osr,
2542 struct ofp_flow_stats_request *fsr;
2543 struct flow_stats_cbdata cbdata;
2544 struct cls_rule target;
2546 if (arg_size != sizeof *fsr) {
2547 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2549 fsr = (struct ofp_flow_stats_request *) osr->body;
2551 COVERAGE_INC(ofproto_flows_req);
2553 cbdata.ofconn = ofconn;
2554 cbdata.out_port = fsr->out_port;
2555 cbdata.msg = start_stats_reply(osr, 1024);
2556 cls_rule_from_match(&target, &fsr->match, 0);
2557 classifier_for_each_match(&p->cls, &target,
2558 table_id_to_include(fsr->table_id),
2559 flow_stats_cb, &cbdata);
2560 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2564 struct flow_stats_ds_cbdata {
2565 struct ofproto *ofproto;
2570 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
2572 struct rule *rule = rule_from_cls_rule(rule_);
2573 struct flow_stats_ds_cbdata *cbdata = cbdata_;
2574 struct ds *results = cbdata->results;
2575 struct ofp_match match;
2576 uint64_t packet_count, byte_count;
2577 size_t act_len = sizeof *rule->actions * rule->n_actions;
2579 /* Don't report on subrules. */
2580 if (rule->super != NULL) {
2584 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2585 flow_to_ovs_match(&rule->cr.flow, rule->cr.wc.wildcards, &match);
2587 ds_put_format(results, "duration=%llds, ",
2588 (time_msec() - rule->created) / 1000);
2589 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2590 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2591 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2592 ofp_print_match(results, &match, true);
2593 ofp_print_actions(results, &rule->actions->header, act_len);
2594 ds_put_cstr(results, "\n");
2597 /* Adds a pretty-printed description of all flows to 'results', including
2598 * those marked hidden by secchan (e.g., by in-band control). */
2600 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2602 struct ofp_match match;
2603 struct cls_rule target;
2604 struct flow_stats_ds_cbdata cbdata;
2606 memset(&match, 0, sizeof match);
2607 match.wildcards = htonl(OFPFW_ALL);
2610 cbdata.results = results;
2612 cls_rule_from_match(&target, &match, 0);
2613 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2614 flow_stats_ds_cb, &cbdata);
2617 struct aggregate_stats_cbdata {
2618 struct ofproto *ofproto;
2620 uint64_t packet_count;
2621 uint64_t byte_count;
2626 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2628 struct rule *rule = rule_from_cls_rule(rule_);
2629 struct aggregate_stats_cbdata *cbdata = cbdata_;
2630 uint64_t packet_count, byte_count;
2632 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2636 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2638 cbdata->packet_count += packet_count;
2639 cbdata->byte_count += byte_count;
2644 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2645 const struct ofp_stats_request *osr,
2648 struct ofp_aggregate_stats_request *asr;
2649 struct ofp_aggregate_stats_reply *reply;
2650 struct aggregate_stats_cbdata cbdata;
2651 struct cls_rule target;
2654 if (arg_size != sizeof *asr) {
2655 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2657 asr = (struct ofp_aggregate_stats_request *) osr->body;
2659 COVERAGE_INC(ofproto_agg_request);
2661 cbdata.out_port = asr->out_port;
2662 cbdata.packet_count = 0;
2663 cbdata.byte_count = 0;
2665 cls_rule_from_match(&target, &asr->match, 0);
2666 classifier_for_each_match(&p->cls, &target,
2667 table_id_to_include(asr->table_id),
2668 aggregate_stats_cb, &cbdata);
2670 msg = start_stats_reply(osr, sizeof *reply);
2671 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2672 reply->flow_count = htonl(cbdata.n_flows);
2673 reply->packet_count = htonll(cbdata.packet_count);
2674 reply->byte_count = htonll(cbdata.byte_count);
2675 queue_tx(msg, ofconn, ofconn->reply_counter);
2680 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2681 struct ofp_header *oh)
2683 struct ofp_stats_request *osr;
2687 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2692 osr = (struct ofp_stats_request *) oh;
2694 switch (ntohs(osr->type)) {
2696 return handle_desc_stats_request(p, ofconn, osr);
2699 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2701 case OFPST_AGGREGATE:
2702 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2705 return handle_table_stats_request(p, ofconn, osr);
2708 return handle_port_stats_request(p, ofconn, osr);
2711 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2714 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2718 static long long int
2719 msec_from_nsec(uint64_t sec, uint32_t nsec)
2721 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2725 update_time(struct ofproto *ofproto, struct rule *rule,
2726 const struct odp_flow_stats *stats)
2728 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2729 if (used > rule->used) {
2731 if (rule->super && used > rule->super->used) {
2732 rule->super->used = used;
2734 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
2739 update_stats(struct ofproto *ofproto, struct rule *rule,
2740 const struct odp_flow_stats *stats)
2742 if (stats->n_packets) {
2743 update_time(ofproto, rule, stats);
2744 rule->packet_count += stats->n_packets;
2745 rule->byte_count += stats->n_bytes;
2746 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
2752 add_flow(struct ofproto *p, struct ofconn *ofconn,
2753 struct ofp_flow_mod *ofm, size_t n_actions)
2755 struct ofpbuf *packet;
2760 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
2761 n_actions, ntohs(ofm->idle_timeout),
2762 ntohs(ofm->hard_timeout));
2763 cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority));
2767 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2768 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2772 rule_insert(p, rule, packet, in_port);
2773 ofpbuf_delete(packet);
2778 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
2779 size_t n_actions, uint16_t command, struct rule *rule)
2781 if (rule_is_hidden(rule)) {
2785 if (command == OFPFC_DELETE) {
2786 rule_remove(p, rule);
2788 size_t actions_len = n_actions * sizeof *rule->actions;
2790 if (n_actions == rule->n_actions
2791 && !memcmp(ofm->actions, rule->actions, actions_len))
2796 free(rule->actions);
2797 rule->actions = xmemdup(ofm->actions, actions_len);
2798 rule->n_actions = n_actions;
2800 if (rule->cr.wc.wildcards) {
2801 COVERAGE_INC(ofproto_mod_wc_flow);
2802 p->need_revalidate = true;
2804 rule_update_actions(p, rule);
2812 modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm,
2813 size_t n_actions, uint16_t command)
2819 flow_from_match(&flow, &wildcards, &ofm->match);
2820 rule = rule_from_cls_rule(classifier_find_rule_exactly(
2821 &p->cls, &flow, wildcards,
2822 ntohs(ofm->priority)));
2825 if (command == OFPFC_DELETE
2826 && ofm->out_port != htons(OFPP_NONE)
2827 && !rule_has_out_port(rule, ofm->out_port)) {
2831 modify_flow(p, ofm, n_actions, command, rule);
2836 struct modify_flows_cbdata {
2837 struct ofproto *ofproto;
2838 const struct ofp_flow_mod *ofm;
2845 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
2847 struct rule *rule = rule_from_cls_rule(rule_);
2848 struct modify_flows_cbdata *cbdata = cbdata_;
2850 if (cbdata->out_port != htons(OFPP_NONE)
2851 && !rule_has_out_port(rule, cbdata->out_port)) {
2855 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions,
2856 cbdata->command, rule);
2860 modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm,
2861 size_t n_actions, uint16_t command)
2863 struct modify_flows_cbdata cbdata;
2864 struct cls_rule target;
2868 cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port
2869 : htons(OFPP_NONE));
2870 cbdata.n_actions = n_actions;
2871 cbdata.command = command;
2873 cls_rule_from_match(&target, &ofm->match, 0);
2875 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2876 modify_flows_cb, &cbdata);
2881 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
2882 struct ofp_flow_mod *ofm)
2887 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
2888 sizeof *ofm->actions, &n_actions);
2893 normalize_match(&ofm->match);
2894 if (!ofm->match.wildcards) {
2895 ofm->priority = htons(UINT16_MAX);
2898 error = validate_actions((const union ofp_action *) ofm->actions,
2899 n_actions, p->max_ports);
2904 switch (ntohs(ofm->command)) {
2906 return add_flow(p, ofconn, ofm, n_actions);
2909 return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY);
2911 case OFPFC_MODIFY_STRICT:
2912 return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY);
2915 return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE);
2917 case OFPFC_DELETE_STRICT:
2918 return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE);
2921 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
2926 send_capability_reply(struct ofproto *p, struct ofconn *ofconn, uint32_t xid)
2928 struct ofmp_capability_reply *ocr;
2930 char capabilities[] = "com.nicira.mgmt.manager=false\n";
2932 ocr = make_openflow_xid(sizeof(*ocr), OFPT_VENDOR, xid, &b);
2933 ocr->header.header.vendor = htonl(NX_VENDOR_ID);
2934 ocr->header.header.subtype = htonl(NXT_MGMT);
2935 ocr->header.type = htons(OFMPT_CAPABILITY_REPLY);
2937 ocr->format = htonl(OFMPCOF_SIMPLE);
2938 ocr->mgmt_id = htonll(p->mgmt_id);
2940 ofpbuf_put(b, capabilities, strlen(capabilities));
2942 queue_tx(b, ofconn, ofconn->reply_counter);
2946 handle_ofmp(struct ofproto *p, struct ofconn *ofconn,
2947 struct ofmp_header *ofmph)
2949 size_t msg_len = ntohs(ofmph->header.header.length);
2950 if (msg_len < sizeof(*ofmph)) {
2951 VLOG_WARN_RL(&rl, "dropping short managment message: %zu\n", msg_len);
2952 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2955 if (ofmph->type == htons(OFMPT_CAPABILITY_REQUEST)) {
2956 struct ofmp_capability_request *ofmpcr;
2958 if (msg_len < sizeof(struct ofmp_capability_request)) {
2959 VLOG_WARN_RL(&rl, "dropping short capability request: %zu\n",
2961 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2964 ofmpcr = (struct ofmp_capability_request *)ofmph;
2965 if (ofmpcr->format != htonl(OFMPCAF_SIMPLE)) {
2966 /* xxx Find a better type than bad subtype */
2967 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2970 send_capability_reply(p, ofconn, ofmph->header.header.xid);
2973 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2978 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
2980 struct ofp_vendor_header *ovh = msg;
2981 struct nicira_header *nh;
2983 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
2984 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2986 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
2987 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2989 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
2990 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2994 switch (ntohl(nh->subtype)) {
2995 case NXT_STATUS_REQUEST:
2996 return switch_status_handle_request(p->switch_status, ofconn->rconn,
2999 case NXT_ACT_SET_CONFIG:
3000 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
3002 case NXT_ACT_GET_CONFIG:
3003 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
3006 return handle_ofmp(p, ofconn, msg);
3009 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3013 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3014 struct ofpbuf *ofp_msg)
3016 struct ofp_header *oh = ofp_msg->data;
3019 COVERAGE_INC(ofproto_recv_openflow);
3021 case OFPT_ECHO_REQUEST:
3022 error = handle_echo_request(ofconn, oh);
3025 case OFPT_ECHO_REPLY:
3029 case OFPT_FEATURES_REQUEST:
3030 error = handle_features_request(p, ofconn, oh);
3033 case OFPT_GET_CONFIG_REQUEST:
3034 error = handle_get_config_request(p, ofconn, oh);
3037 case OFPT_SET_CONFIG:
3038 error = handle_set_config(p, ofconn, ofp_msg->data);
3041 case OFPT_PACKET_OUT:
3042 error = handle_packet_out(p, ofconn, ofp_msg->data);
3046 error = handle_port_mod(p, oh);
3050 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3053 case OFPT_STATS_REQUEST:
3054 error = handle_stats_request(p, ofconn, oh);
3058 error = handle_vendor(p, ofconn, ofp_msg->data);
3062 if (VLOG_IS_WARN_ENABLED()) {
3063 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3064 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3067 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3072 send_error_oh(ofconn, ofp_msg->data, error);
3077 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3079 struct odp_msg *msg = packet->data;
3080 uint16_t in_port = odp_port_to_ofp_port(msg->port);
3082 struct ofpbuf payload;
3085 /* Handle controller actions. */
3086 if (msg->type == _ODPL_ACTION_NR) {
3087 COVERAGE_INC(ofproto_ctlr_action);
3088 pinsched_send(p->action_sched, in_port, packet,
3089 send_packet_in_action, p);
3093 payload.data = msg + 1;
3094 payload.size = msg->length - sizeof *msg;
3095 flow_extract(&payload, msg->port, &flow);
3097 /* Check with in-band control to see if this packet should be sent
3098 * to the local port regardless of the flow table. */
3099 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3100 union odp_action action;
3102 memset(&action, 0, sizeof(action));
3103 action.output.type = ODPAT_OUTPUT;
3104 action.output.port = ODPP_LOCAL;
3105 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3108 rule = lookup_valid_rule(p, &flow);
3110 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3111 struct ofport *port = port_array_get(&p->ports, msg->port);
3113 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3114 COVERAGE_INC(ofproto_no_packet_in);
3115 /* XXX install 'drop' flow entry */
3116 ofpbuf_delete(packet);
3120 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3123 COVERAGE_INC(ofproto_packet_in);
3124 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3128 if (rule->cr.wc.wildcards) {
3129 rule = rule_create_subrule(p, rule, &flow);
3130 rule_make_actions(p, rule, packet);
3132 if (!rule->may_install) {
3133 /* The rule is not installable, that is, we need to process every
3134 * packet, so process the current packet and set its actions into
3136 rule_make_actions(p, rule, packet);
3138 /* XXX revalidate rule if it needs it */
3142 rule_execute(p, rule, &payload, &flow);
3143 rule_reinstall(p, rule);
3145 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY
3146 && rconn_is_connected(p->controller->rconn)) {
3148 * Extra-special case for fail-open mode.
3150 * We are in fail-open mode and the packet matched the fail-open rule,
3151 * but we are connected to a controller too. We should send the packet
3152 * up to the controller in the hope that it will try to set up a flow
3153 * and thereby allow us to exit fail-open.
3155 * See the top-level comment in fail-open.c for more information.
3157 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3159 ofpbuf_delete(packet);
3164 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3166 struct rule *sub = rule_from_cls_rule(sub_);
3167 struct revalidate_cbdata *cbdata = cbdata_;
3169 if (cbdata->revalidate_all
3170 || (cbdata->revalidate_subrules && sub->super)
3171 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3172 revalidate_rule(cbdata->ofproto, sub);
3177 revalidate_rule(struct ofproto *p, struct rule *rule)
3179 const flow_t *flow = &rule->cr.flow;
3181 COVERAGE_INC(ofproto_revalidate_rule);
3184 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3186 rule_remove(p, rule);
3188 } else if (super != rule->super) {
3189 COVERAGE_INC(ofproto_revalidate_moved);
3190 list_remove(&rule->list);
3191 list_push_back(&super->list, &rule->list);
3192 rule->super = super;
3193 rule->hard_timeout = super->hard_timeout;
3194 rule->idle_timeout = super->idle_timeout;
3195 rule->created = super->created;
3200 rule_update_actions(p, rule);
3204 static struct ofpbuf *
3205 compose_flow_exp(const struct rule *rule, long long int now, uint8_t reason)
3207 struct ofp_flow_expired *ofe;
3210 ofe = make_openflow(sizeof *ofe, OFPT_FLOW_EXPIRED, &buf);
3211 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofe->match);
3212 ofe->priority = htons(rule->cr.priority);
3213 ofe->reason = reason;
3214 ofe->duration = htonl((now - rule->created) / 1000);
3215 ofe->packet_count = htonll(rule->packet_count);
3216 ofe->byte_count = htonll(rule->byte_count);
3222 send_flow_exp(struct ofproto *p, struct rule *rule,
3223 long long int now, uint8_t reason)
3225 struct ofconn *ofconn;
3226 struct ofconn *prev;
3227 struct ofpbuf *buf = NULL;
3229 /* We limit the maximum number of queued flow expirations it by accounting
3230 * them under the counter for replies. That works because preventing
3231 * OpenFlow requests from being processed also prevents new flows from
3232 * being added (and expiring). (It also prevents processing OpenFlow
3233 * requests that would not add new flows, so it is imperfect.) */
3236 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3237 if (ofconn->send_flow_exp && rconn_is_connected(ofconn->rconn)) {
3239 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3241 buf = compose_flow_exp(rule, now, reason);
3247 queue_tx(buf, prev, prev->reply_counter);
3252 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3254 assert(rule->installed);
3255 assert(!rule->cr.wc.wildcards);
3258 rule_remove(ofproto, rule);
3260 rule_uninstall(ofproto, rule);
3265 expire_rule(struct cls_rule *cls_rule, void *p_)
3267 struct ofproto *p = p_;
3268 struct rule *rule = rule_from_cls_rule(cls_rule);
3269 long long int hard_expire, idle_expire, expire, now;
3271 hard_expire = (rule->hard_timeout
3272 ? rule->created + rule->hard_timeout * 1000
3274 idle_expire = (rule->idle_timeout
3275 && (rule->super || list_is_empty(&rule->list))
3276 ? rule->used + rule->idle_timeout * 1000
3278 expire = MIN(hard_expire, idle_expire);
3282 if (rule->installed && now >= rule->used + 5000) {
3283 uninstall_idle_flow(p, rule);
3284 } else if (!rule->cr.wc.wildcards) {
3285 active_timeout(p, rule);
3291 COVERAGE_INC(ofproto_expired);
3293 /* Update stats. This code will be a no-op if the rule expired
3294 * due to an idle timeout. */
3295 if (rule->cr.wc.wildcards) {
3296 struct rule *subrule, *next;
3297 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3298 rule_remove(p, subrule);
3301 rule_uninstall(p, rule);
3304 if (!rule_is_hidden(rule)) {
3305 send_flow_exp(p, rule, now,
3307 ? OFPER_HARD_TIMEOUT : OFPER_IDLE_TIMEOUT));
3309 rule_remove(p, rule);
3313 active_timeout(struct ofproto *ofproto, struct rule *rule)
3315 if (ofproto->netflow && !is_controller_rule(rule) &&
3316 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
3317 struct ofexpired expired;
3318 struct odp_flow odp_flow;
3320 /* Get updated flow stats. */
3321 memset(&odp_flow, 0, sizeof odp_flow);
3322 if (rule->installed) {
3323 odp_flow.key = rule->cr.flow;
3324 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
3325 dpif_flow_get(ofproto->dpif, &odp_flow);
3327 if (odp_flow.stats.n_packets) {
3328 update_time(ofproto, rule, &odp_flow.stats);
3329 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
3330 odp_flow.stats.tcp_flags);
3334 expired.flow = rule->cr.flow;
3335 expired.packet_count = rule->packet_count +
3336 odp_flow.stats.n_packets;
3337 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
3338 expired.used = rule->used;
3340 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
3342 /* Schedule us to send the accumulated records once we have
3343 * collected all of them. */
3344 poll_immediate_wake();
3349 update_used(struct ofproto *p)
3351 struct odp_flow *flows;
3356 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
3361 for (i = 0; i < n_flows; i++) {
3362 struct odp_flow *f = &flows[i];
3365 rule = rule_from_cls_rule(
3366 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3367 if (!rule || !rule->installed) {
3368 COVERAGE_INC(ofproto_unexpected_rule);
3369 dpif_flow_del(p->dpif, f);
3373 update_time(p, rule, &f->stats);
3374 rule_account(p, rule, f->stats.n_bytes);
3380 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3381 const struct ofpbuf *packet, int send_len)
3383 struct odp_msg *msg = packet->data;
3384 struct ofpbuf payload;
3388 /* Extract packet payload from 'msg'. */
3389 payload.data = msg + 1;
3390 payload.size = msg->length - sizeof *msg;
3392 /* Construct ofp_packet_in message. */
3393 reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3394 opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port), reason,
3395 &payload, send_len);
3398 rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100);
3402 send_packet_in_action(struct ofpbuf *packet, void *p_)
3404 struct ofproto *p = p_;
3405 struct ofconn *ofconn;
3406 struct odp_msg *msg;
3409 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3410 if (ofconn == p->controller || ofconn->miss_send_len) {
3411 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3414 ofpbuf_delete(packet);
3418 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3420 struct ofproto *p = p_;
3421 bool in_fail_open = p->fail_open && fail_open_is_active(p->fail_open);
3422 struct ofconn *ofconn;
3423 struct ofpbuf payload;
3424 struct odp_msg *msg;
3427 payload.data = msg + 1;
3428 payload.size = msg->length - sizeof *msg;
3429 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3430 if (ofconn->miss_send_len) {
3431 struct pktbuf *pb = ofconn->pktbuf;
3432 uint32_t buffer_id = (in_fail_open
3434 : pktbuf_save(pb, &payload, msg->port));
3435 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3437 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3440 ofpbuf_delete(packet);
3444 pick_datapath_id(const struct ofproto *ofproto)
3446 const struct ofport *port;
3448 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
3450 uint8_t ea[ETH_ADDR_LEN];
3453 error = netdev_get_etheraddr(port->netdev, ea);
3455 return eth_addr_to_uint64(ea);
3457 VLOG_WARN("could not get MAC address for %s (%s)",
3458 netdev_get_name(port->netdev), strerror(error));
3460 return ofproto->fallback_dpid;
3464 pick_fallback_dpid(void)
3466 uint8_t ea[ETH_ADDR_LEN];
3467 eth_addr_random(ea);
3468 ea[0] = 0x00; /* Set Nicira OUI. */
3471 return eth_addr_to_uint64(ea);
3475 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3476 struct odp_actions *actions, tag_type *tags,
3477 uint16_t *nf_output_iface, void *ofproto_)
3479 struct ofproto *ofproto = ofproto_;
3482 /* Drop frames for reserved multicast addresses. */
3483 if (eth_addr_is_reserved(flow->dl_dst)) {
3487 /* Learn source MAC (but don't try to learn from revalidation). */
3488 if (packet != NULL) {
3489 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3492 /* The log messages here could actually be useful in debugging,
3493 * so keep the rate limit relatively high. */
3494 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3495 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3496 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3497 ofproto_revalidate(ofproto, rev_tag);
3501 /* Determine output port. */
3502 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3504 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
3505 } else if (out_port != flow->in_port) {
3506 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3507 *nf_output_iface = out_port;
3515 static const struct ofhooks default_ofhooks = {
3517 default_normal_ofhook_cb,