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"
37 #include "ofproto-sflow.h"
39 #include "openflow/nicira-ext.h"
40 #include "openflow/openflow.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
62 #include "sflow_api.h"
66 TABLEID_CLASSIFIER = 1
70 struct netdev *netdev;
71 struct ofp_phy_port opp; /* In host byte order. */
74 static void ofport_free(struct ofport *);
75 static void hton_ofp_phy_port(struct ofp_phy_port *);
77 static int xlate_actions(const union ofp_action *in, size_t n_in,
78 const flow_t *flow, struct ofproto *ofproto,
79 const struct ofpbuf *packet,
80 struct odp_actions *out, tag_type *tags,
81 bool *may_set_up_flow, uint16_t *nf_output_iface);
86 uint64_t flow_cookie; /* Controller-issued identifier.
87 (Kept in network-byte order.) */
88 uint16_t idle_timeout; /* In seconds from time of last use. */
89 uint16_t hard_timeout; /* In seconds from time of creation. */
90 bool send_flow_removed; /* Send a flow removed message? */
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 uint64_t flow_cookie, bool send_flow_removed);
150 static void rule_free(struct rule *);
151 static void rule_destroy(struct ofproto *, struct rule *);
152 static struct rule *rule_from_cls_rule(const struct cls_rule *);
153 static void rule_insert(struct ofproto *, struct rule *,
154 struct ofpbuf *packet, uint16_t in_port);
155 static void rule_remove(struct ofproto *, struct rule *);
156 static bool rule_make_actions(struct ofproto *, struct rule *,
157 const struct ofpbuf *packet);
158 static void rule_install(struct ofproto *, struct rule *,
159 struct rule *displaced_rule);
160 static void rule_uninstall(struct ofproto *, struct rule *);
161 static void rule_post_uninstall(struct ofproto *, struct rule *);
162 static void send_flow_removed(struct ofproto *p, struct rule *rule,
163 long long int now, uint8_t reason);
168 struct pktbuf *pktbuf;
171 struct rconn_packet_counter *packet_in_counter;
173 /* Number of OpenFlow messages queued as replies to OpenFlow requests, and
174 * the maximum number before we stop reading OpenFlow requests. */
175 #define OFCONN_REPLY_MAX 100
176 struct rconn_packet_counter *reply_counter;
179 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *);
180 static void ofconn_destroy(struct ofconn *);
181 static void ofconn_run(struct ofconn *, struct ofproto *);
182 static void ofconn_wait(struct ofconn *);
183 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
184 struct rconn_packet_counter *counter);
188 uint64_t datapath_id; /* Datapath ID. */
189 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
190 char *manufacturer; /* Manufacturer. */
191 char *hardware; /* Hardware. */
192 char *software; /* Software version. */
193 char *serial; /* Serial number. */
194 char *dp_desc; /* Datapath description. */
198 struct netdev_monitor *netdev_monitor;
199 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
201 struct shash port_by_name;
205 struct switch_status *switch_status;
206 struct status_category *ss_cat;
207 struct in_band *in_band;
208 struct discovery *discovery;
209 struct fail_open *fail_open;
210 struct pinsched *miss_sched, *action_sched;
211 struct netflow *netflow;
212 struct ofproto_sflow *sflow;
215 struct classifier cls;
216 bool need_revalidate;
217 long long int next_expiration;
218 struct tag_set revalidate_set;
220 /* OpenFlow connections. */
221 struct list all_conns;
222 struct ofconn *controller;
223 struct pvconn **listeners;
225 struct pvconn **snoops;
228 /* Hooks for ovs-vswitchd. */
229 const struct ofhooks *ofhooks;
232 /* Used by default ofhooks. */
233 struct mac_learning *ml;
236 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
238 static const struct ofhooks default_ofhooks;
240 static uint64_t pick_datapath_id(const struct ofproto *);
241 static uint64_t pick_fallback_dpid(void);
242 static void send_packet_in_miss(struct ofpbuf *, void *ofproto);
243 static void send_packet_in_action(struct ofpbuf *, void *ofproto);
244 static void update_used(struct ofproto *);
245 static void update_stats(struct ofproto *, struct rule *,
246 const struct odp_flow_stats *);
247 static void expire_rule(struct cls_rule *, void *ofproto);
248 static void active_timeout(struct ofproto *ofproto, struct rule *rule);
249 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
250 static void revalidate_cb(struct cls_rule *rule_, void *p_);
252 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
254 static void handle_openflow(struct ofconn *, struct ofproto *,
257 static void refresh_port_groups(struct ofproto *);
259 static void update_port(struct ofproto *, const char *devname);
260 static int init_ports(struct ofproto *);
261 static void reinit_ports(struct ofproto *);
264 ofproto_create(const char *datapath, const char *datapath_type,
265 const struct ofhooks *ofhooks, void *aux,
266 struct ofproto **ofprotop)
268 struct odp_stats stats;
275 /* Connect to datapath and start listening for messages. */
276 error = dpif_open(datapath, datapath_type, &dpif);
278 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
281 error = dpif_get_dp_stats(dpif, &stats);
283 VLOG_ERR("failed to obtain stats for datapath %s: %s",
284 datapath, strerror(error));
288 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
290 VLOG_ERR("failed to listen on datapath %s: %s",
291 datapath, strerror(error));
295 dpif_flow_flush(dpif);
296 dpif_recv_purge(dpif);
298 /* Initialize settings. */
299 p = xzalloc(sizeof *p);
300 p->fallback_dpid = pick_fallback_dpid();
301 p->datapath_id = p->fallback_dpid;
302 p->manufacturer = xstrdup("Nicira Networks, Inc.");
303 p->hardware = xstrdup("Reference Implementation");
304 p->software = xstrdup(VERSION BUILDNR);
305 p->serial = xstrdup("None");
306 p->dp_desc = xstrdup("None");
308 /* Initialize datapath. */
310 p->netdev_monitor = netdev_monitor_create();
311 port_array_init(&p->ports);
312 shash_init(&p->port_by_name);
313 p->max_ports = stats.max_ports;
315 /* Initialize submodules. */
316 p->switch_status = switch_status_create(p);
320 p->miss_sched = p->action_sched = NULL;
324 /* Initialize flow table. */
325 classifier_init(&p->cls);
326 p->need_revalidate = false;
327 p->next_expiration = time_msec() + 1000;
328 tag_set_init(&p->revalidate_set);
330 /* Initialize OpenFlow connections. */
331 list_init(&p->all_conns);
332 p->controller = ofconn_create(p, rconn_create(5, 8));
333 p->controller->pktbuf = pktbuf_create();
334 p->controller->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
340 /* Initialize hooks. */
342 p->ofhooks = ofhooks;
346 p->ofhooks = &default_ofhooks;
348 p->ml = mac_learning_create();
351 /* Register switch status category. */
352 p->ss_cat = switch_status_register(p->switch_status, "remote",
353 rconn_status_cb, p->controller->rconn);
355 /* Pick final datapath ID. */
356 p->datapath_id = pick_datapath_id(p);
357 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
364 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
366 uint64_t old_dpid = p->datapath_id;
367 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
368 if (p->datapath_id != old_dpid) {
369 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
370 rconn_reconnect(p->controller->rconn);
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,
398 free(p->manufacturer);
399 p->manufacturer = xstrdup(manufacturer);
403 p->hardware = xstrdup(hardware);
407 p->software = xstrdup(software);
411 p->serial = xstrdup(serial);
415 p->dp_desc = xstrdup(dp_desc);
420 ofproto_set_in_band(struct ofproto *p, bool in_band)
422 if (in_band != (p->in_band != NULL)) {
424 return in_band_create(p, p->dpif, p->switch_status,
425 p->controller->rconn, &p->in_band);
427 ofproto_set_discovery(p, false, NULL, true);
428 in_band_destroy(p->in_band);
431 rconn_reconnect(p->controller->rconn);
437 ofproto_set_discovery(struct ofproto *p, bool discovery,
438 const char *re, bool update_resolv_conf)
440 if (discovery != (p->discovery != NULL)) {
442 int error = ofproto_set_in_band(p, true);
446 error = discovery_create(re, update_resolv_conf,
447 p->dpif, p->switch_status,
453 discovery_destroy(p->discovery);
456 rconn_disconnect(p->controller->rconn);
457 } else if (discovery) {
458 discovery_set_update_resolv_conf(p->discovery, update_resolv_conf);
459 return discovery_set_accept_controller_re(p->discovery, re);
465 ofproto_set_controller(struct ofproto *ofproto, const char *controller)
467 if (ofproto->discovery) {
469 } else if (controller) {
470 if (strcmp(rconn_get_name(ofproto->controller->rconn), controller)) {
471 return rconn_connect(ofproto->controller->rconn, controller);
476 rconn_disconnect(ofproto->controller->rconn);
482 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
483 const struct svec *svec)
485 struct pvconn **pvconns = *pvconnsp;
486 size_t n_pvconns = *n_pvconnsp;
490 for (i = 0; i < n_pvconns; i++) {
491 pvconn_close(pvconns[i]);
495 pvconns = xmalloc(svec->n * sizeof *pvconns);
497 for (i = 0; i < svec->n; i++) {
498 const char *name = svec->names[i];
499 struct pvconn *pvconn;
502 error = pvconn_open(name, &pvconn);
504 pvconns[n_pvconns++] = pvconn;
506 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
514 *n_pvconnsp = n_pvconns;
520 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
522 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
526 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
528 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
532 ofproto_set_netflow(struct ofproto *ofproto,
533 const struct netflow_options *nf_options)
535 if (nf_options && nf_options->collectors.n) {
536 if (!ofproto->netflow) {
537 ofproto->netflow = netflow_create();
539 return netflow_set_options(ofproto->netflow, nf_options);
541 netflow_destroy(ofproto->netflow);
542 ofproto->netflow = NULL;
548 ofproto_set_sflow(struct ofproto *ofproto,
549 const struct ofproto_sflow_options *oso)
551 struct ofproto_sflow *os = ofproto->sflow;
554 struct ofport *ofport;
555 unsigned int odp_port;
557 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
558 refresh_port_groups(ofproto);
559 PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) {
560 ofproto_sflow_add_port(os, odp_port,
561 netdev_get_name(ofport->netdev));
564 ofproto_sflow_set_options(os, oso);
566 ofproto_sflow_destroy(os);
567 ofproto->sflow = NULL;
572 ofproto_set_failure(struct ofproto *ofproto, bool fail_open)
575 struct rconn *rconn = ofproto->controller->rconn;
576 int trigger_duration = rconn_get_probe_interval(rconn) * 3;
577 if (!ofproto->fail_open) {
578 ofproto->fail_open = fail_open_create(ofproto, trigger_duration,
579 ofproto->switch_status,
582 fail_open_set_trigger_duration(ofproto->fail_open,
586 fail_open_destroy(ofproto->fail_open);
587 ofproto->fail_open = NULL;
592 ofproto_set_rate_limit(struct ofproto *ofproto,
593 int rate_limit, int burst_limit)
595 if (rate_limit > 0) {
596 if (!ofproto->miss_sched) {
597 ofproto->miss_sched = pinsched_create(rate_limit, burst_limit,
598 ofproto->switch_status);
599 ofproto->action_sched = pinsched_create(rate_limit, burst_limit,
602 pinsched_set_limits(ofproto->miss_sched, rate_limit, burst_limit);
603 pinsched_set_limits(ofproto->action_sched,
604 rate_limit, burst_limit);
607 pinsched_destroy(ofproto->miss_sched);
608 ofproto->miss_sched = NULL;
609 pinsched_destroy(ofproto->action_sched);
610 ofproto->action_sched = NULL;
615 ofproto_set_stp(struct ofproto *ofproto OVS_UNUSED, bool enable_stp)
619 VLOG_WARN("STP is not yet implemented");
627 ofproto_get_datapath_id(const struct ofproto *ofproto)
629 return ofproto->datapath_id;
633 ofproto_get_probe_interval(const struct ofproto *ofproto)
635 return rconn_get_probe_interval(ofproto->controller->rconn);
639 ofproto_get_max_backoff(const struct ofproto *ofproto)
641 return rconn_get_max_backoff(ofproto->controller->rconn);
645 ofproto_get_in_band(const struct ofproto *ofproto)
647 return ofproto->in_band != NULL;
651 ofproto_get_discovery(const struct ofproto *ofproto)
653 return ofproto->discovery != NULL;
657 ofproto_get_controller(const struct ofproto *ofproto)
659 return rconn_get_name(ofproto->controller->rconn);
663 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
667 for (i = 0; i < ofproto->n_listeners; i++) {
668 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
673 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
677 for (i = 0; i < ofproto->n_snoops; i++) {
678 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
683 ofproto_destroy(struct ofproto *p)
685 struct ofconn *ofconn, *next_ofconn;
686 struct ofport *ofport;
687 unsigned int port_no;
694 /* Destroy fail-open early, because it touches the classifier. */
695 ofproto_set_failure(p, false);
697 ofproto_flush_flows(p);
698 classifier_destroy(&p->cls);
700 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
702 ofconn_destroy(ofconn);
706 netdev_monitor_destroy(p->netdev_monitor);
707 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
710 shash_destroy(&p->port_by_name);
712 switch_status_destroy(p->switch_status);
713 in_band_destroy(p->in_band);
714 discovery_destroy(p->discovery);
715 pinsched_destroy(p->miss_sched);
716 pinsched_destroy(p->action_sched);
717 netflow_destroy(p->netflow);
718 ofproto_sflow_destroy(p->sflow);
720 switch_status_unregister(p->ss_cat);
722 for (i = 0; i < p->n_listeners; i++) {
723 pvconn_close(p->listeners[i]);
727 for (i = 0; i < p->n_snoops; i++) {
728 pvconn_close(p->snoops[i]);
732 mac_learning_destroy(p->ml);
738 ofproto_run(struct ofproto *p)
740 int error = ofproto_run1(p);
742 error = ofproto_run2(p, false);
748 process_port_change(struct ofproto *ofproto, int error, char *devname)
750 if (error == ENOBUFS) {
751 reinit_ports(ofproto);
753 update_port(ofproto, devname);
759 ofproto_run1(struct ofproto *p)
761 struct ofconn *ofconn, *next_ofconn;
766 if (shash_is_empty(&p->port_by_name)) {
770 for (i = 0; i < 50; i++) {
774 error = dpif_recv(p->dpif, &buf);
776 if (error == ENODEV) {
777 /* Someone destroyed the datapath behind our back. The caller
778 * better destroy us and give up, because we're just going to
779 * spin from here on out. */
780 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
781 VLOG_ERR_RL(&rl, "%s: datapath was destroyed externally",
788 handle_odp_msg(p, buf);
791 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
792 process_port_change(p, error, devname);
794 while ((error = netdev_monitor_poll(p->netdev_monitor,
795 &devname)) != EAGAIN) {
796 process_port_change(p, error, devname);
800 in_band_run(p->in_band);
803 char *controller_name;
804 if (rconn_is_connectivity_questionable(p->controller->rconn)) {
805 discovery_question_connectivity(p->discovery);
807 if (discovery_run(p->discovery, &controller_name)) {
808 if (controller_name) {
809 rconn_connect(p->controller->rconn, controller_name);
811 rconn_disconnect(p->controller->rconn);
815 pinsched_run(p->miss_sched, send_packet_in_miss, p);
816 pinsched_run(p->action_sched, send_packet_in_action, p);
818 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
820 ofconn_run(ofconn, p);
823 /* Fail-open maintenance. Do this after processing the ofconns since
824 * fail-open checks the status of the controller rconn. */
826 fail_open_run(p->fail_open);
829 for (i = 0; i < p->n_listeners; i++) {
833 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
835 ofconn_create(p, rconn_new_from_vconn("passive", vconn));
836 } else if (retval != EAGAIN) {
837 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
841 for (i = 0; i < p->n_snoops; i++) {
845 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
847 rconn_add_monitor(p->controller->rconn, vconn);
848 } else if (retval != EAGAIN) {
849 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
853 if (time_msec() >= p->next_expiration) {
854 COVERAGE_INC(ofproto_expiration);
855 p->next_expiration = time_msec() + 1000;
858 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
860 /* Let the hook know that we're at a stable point: all outstanding data
861 * in existing flows has been accounted to the account_cb. Thus, the
862 * hook can now reasonably do operations that depend on having accurate
863 * flow volume accounting (currently, that's just bond rebalancing). */
864 if (p->ofhooks->account_checkpoint_cb) {
865 p->ofhooks->account_checkpoint_cb(p->aux);
870 netflow_run(p->netflow);
873 ofproto_sflow_run(p->sflow);
879 struct revalidate_cbdata {
880 struct ofproto *ofproto;
881 bool revalidate_all; /* Revalidate all exact-match rules? */
882 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
883 struct tag_set revalidate_set; /* Set of tags to revalidate. */
887 ofproto_run2(struct ofproto *p, bool revalidate_all)
889 if (p->need_revalidate || revalidate_all
890 || !tag_set_is_empty(&p->revalidate_set)) {
891 struct revalidate_cbdata cbdata;
893 cbdata.revalidate_all = revalidate_all;
894 cbdata.revalidate_subrules = p->need_revalidate;
895 cbdata.revalidate_set = p->revalidate_set;
896 tag_set_init(&p->revalidate_set);
897 COVERAGE_INC(ofproto_revalidate);
898 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
899 p->need_revalidate = false;
906 ofproto_wait(struct ofproto *p)
908 struct ofconn *ofconn;
911 dpif_recv_wait(p->dpif);
912 dpif_port_poll_wait(p->dpif);
913 netdev_monitor_poll_wait(p->netdev_monitor);
914 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
918 in_band_wait(p->in_band);
921 discovery_wait(p->discovery);
924 fail_open_wait(p->fail_open);
926 pinsched_wait(p->miss_sched);
927 pinsched_wait(p->action_sched);
929 ofproto_sflow_wait(p->sflow);
931 if (!tag_set_is_empty(&p->revalidate_set)) {
932 poll_immediate_wake();
934 if (p->need_revalidate) {
935 /* Shouldn't happen, but if it does just go around again. */
936 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
937 poll_immediate_wake();
938 } else if (p->next_expiration != LLONG_MAX) {
939 poll_timer_wait(p->next_expiration - time_msec());
941 for (i = 0; i < p->n_listeners; i++) {
942 pvconn_wait(p->listeners[i]);
944 for (i = 0; i < p->n_snoops; i++) {
945 pvconn_wait(p->snoops[i]);
950 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
952 tag_set_add(&ofproto->revalidate_set, tag);
956 ofproto_get_revalidate_set(struct ofproto *ofproto)
958 return &ofproto->revalidate_set;
962 ofproto_is_alive(const struct ofproto *p)
964 return p->discovery || rconn_is_alive(p->controller->rconn);
968 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
969 const union ofp_action *actions, size_t n_actions,
970 const struct ofpbuf *packet)
972 struct odp_actions odp_actions;
975 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
981 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
983 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
984 odp_actions.n_actions, packet);
989 ofproto_add_flow(struct ofproto *p,
990 const flow_t *flow, uint32_t wildcards, unsigned int priority,
991 const union ofp_action *actions, size_t n_actions,
995 rule = rule_create(p, NULL, actions, n_actions,
996 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
998 cls_rule_from_flow(&rule->cr, flow, wildcards, priority);
999 rule_insert(p, rule, NULL, 0);
1003 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1004 uint32_t wildcards, unsigned int priority)
1008 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1012 rule_remove(ofproto, rule);
1017 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1019 struct rule *rule = rule_from_cls_rule(rule_);
1020 struct ofproto *ofproto = ofproto_;
1022 /* Mark the flow as not installed, even though it might really be
1023 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1024 * There is no point in uninstalling it individually since we are about to
1025 * blow away all the flows with dpif_flow_flush(). */
1026 rule->installed = false;
1028 rule_remove(ofproto, rule);
1032 ofproto_flush_flows(struct ofproto *ofproto)
1034 COVERAGE_INC(ofproto_flush);
1035 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1036 dpif_flow_flush(ofproto->dpif);
1037 if (ofproto->in_band) {
1038 in_band_flushed(ofproto->in_band);
1040 if (ofproto->fail_open) {
1041 fail_open_flushed(ofproto->fail_open);
1046 reinit_ports(struct ofproto *p)
1048 struct svec devnames;
1049 struct ofport *ofport;
1050 unsigned int port_no;
1051 struct odp_port *odp_ports;
1055 svec_init(&devnames);
1056 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1057 svec_add (&devnames, (char *) ofport->opp.name);
1059 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1060 for (i = 0; i < n_odp_ports; i++) {
1061 svec_add (&devnames, odp_ports[i].devname);
1065 svec_sort_unique(&devnames);
1066 for (i = 0; i < devnames.n; i++) {
1067 update_port(p, devnames.names[i]);
1069 svec_destroy(&devnames);
1073 refresh_port_group(struct ofproto *p, unsigned int group)
1077 struct ofport *port;
1078 unsigned int port_no;
1080 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1082 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1084 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1085 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1086 ports[n_ports++] = port_no;
1089 dpif_port_group_set(p->dpif, group, ports, n_ports);
1096 refresh_port_groups(struct ofproto *p)
1098 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1099 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1101 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1105 static struct ofport *
1106 make_ofport(const struct odp_port *odp_port)
1108 struct netdev_options netdev_options;
1109 enum netdev_flags flags;
1110 struct ofport *ofport;
1111 struct netdev *netdev;
1115 memset(&netdev_options, 0, sizeof netdev_options);
1116 netdev_options.name = odp_port->devname;
1117 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1118 netdev_options.may_open = true;
1120 error = netdev_open(&netdev_options, &netdev);
1122 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1123 "cannot be opened (%s)",
1124 odp_port->devname, odp_port->port,
1125 odp_port->devname, strerror(error));
1129 ofport = xmalloc(sizeof *ofport);
1130 ofport->netdev = netdev;
1131 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1132 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1133 memcpy(ofport->opp.name, odp_port->devname,
1134 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1135 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1137 netdev_get_flags(netdev, &flags);
1138 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1140 netdev_get_carrier(netdev, &carrier);
1141 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1143 netdev_get_features(netdev,
1144 &ofport->opp.curr, &ofport->opp.advertised,
1145 &ofport->opp.supported, &ofport->opp.peer);
1150 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1152 if (port_array_get(&p->ports, odp_port->port)) {
1153 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1156 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1157 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1166 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1168 const struct ofp_phy_port *a = &a_->opp;
1169 const struct ofp_phy_port *b = &b_->opp;
1171 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1172 return (a->port_no == b->port_no
1173 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1174 && !strcmp((char *) a->name, (char *) b->name)
1175 && a->state == b->state
1176 && a->config == b->config
1177 && a->curr == b->curr
1178 && a->advertised == b->advertised
1179 && a->supported == b->supported
1180 && a->peer == b->peer);
1184 send_port_status(struct ofproto *p, const struct ofport *ofport,
1187 /* XXX Should limit the number of queued port status change messages. */
1188 struct ofconn *ofconn;
1189 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1190 struct ofp_port_status *ops;
1193 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1194 ops->reason = reason;
1195 ops->desc = ofport->opp;
1196 hton_ofp_phy_port(&ops->desc);
1197 queue_tx(b, ofconn, NULL);
1199 if (p->ofhooks->port_changed_cb) {
1200 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1205 ofport_install(struct ofproto *p, struct ofport *ofport)
1207 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1208 const char *netdev_name = (const char *) ofport->opp.name;
1210 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1211 port_array_set(&p->ports, odp_port, ofport);
1212 shash_add(&p->port_by_name, netdev_name, ofport);
1214 ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
1219 ofport_remove(struct ofproto *p, struct ofport *ofport)
1221 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1223 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1224 port_array_set(&p->ports, odp_port, NULL);
1225 shash_delete(&p->port_by_name,
1226 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1228 ofproto_sflow_del_port(p->sflow, odp_port);
1233 ofport_free(struct ofport *ofport)
1236 netdev_close(ofport->netdev);
1242 update_port(struct ofproto *p, const char *devname)
1244 struct odp_port odp_port;
1245 struct ofport *old_ofport;
1246 struct ofport *new_ofport;
1249 COVERAGE_INC(ofproto_update_port);
1251 /* Query the datapath for port information. */
1252 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1254 /* Find the old ofport. */
1255 old_ofport = shash_find_data(&p->port_by_name, devname);
1258 /* There's no port named 'devname' but there might be a port with
1259 * the same port number. This could happen if a port is deleted
1260 * and then a new one added in its place very quickly, or if a port
1261 * is renamed. In the former case we want to send an OFPPR_DELETE
1262 * and an OFPPR_ADD, and in the latter case we want to send a
1263 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1264 * the old port's ifindex against the new port, or perhaps less
1265 * reliably but more portably by comparing the old port's MAC
1266 * against the new port's MAC. However, this code isn't that smart
1267 * and always sends an OFPPR_MODIFY (XXX). */
1268 old_ofport = port_array_get(&p->ports, odp_port.port);
1270 } else if (error != ENOENT && error != ENODEV) {
1271 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1272 "%s", strerror(error));
1276 /* Create a new ofport. */
1277 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1279 /* Eliminate a few pathological cases. */
1280 if (!old_ofport && !new_ofport) {
1282 } else if (old_ofport && new_ofport) {
1283 /* Most of the 'config' bits are OpenFlow soft state, but
1284 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1285 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1286 * leaves the other bits 0.) */
1287 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1289 if (ofport_equal(old_ofport, new_ofport)) {
1290 /* False alarm--no change. */
1291 ofport_free(new_ofport);
1296 /* Now deal with the normal cases. */
1298 ofport_remove(p, old_ofport);
1301 ofport_install(p, new_ofport);
1303 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1304 (!old_ofport ? OFPPR_ADD
1305 : !new_ofport ? OFPPR_DELETE
1307 ofport_free(old_ofport);
1309 /* Update port groups. */
1310 refresh_port_groups(p);
1314 init_ports(struct ofproto *p)
1316 struct odp_port *ports;
1321 error = dpif_port_list(p->dpif, &ports, &n_ports);
1326 for (i = 0; i < n_ports; i++) {
1327 const struct odp_port *odp_port = &ports[i];
1328 if (!ofport_conflicts(p, odp_port)) {
1329 struct ofport *ofport = make_ofport(odp_port);
1331 ofport_install(p, ofport);
1336 refresh_port_groups(p);
1340 static struct ofconn *
1341 ofconn_create(struct ofproto *p, struct rconn *rconn)
1343 struct ofconn *ofconn = xmalloc(sizeof *ofconn);
1344 list_push_back(&p->all_conns, &ofconn->node);
1345 ofconn->rconn = rconn;
1346 ofconn->pktbuf = NULL;
1347 ofconn->miss_send_len = 0;
1348 ofconn->packet_in_counter = rconn_packet_counter_create ();
1349 ofconn->reply_counter = rconn_packet_counter_create ();
1354 ofconn_destroy(struct ofconn *ofconn)
1356 list_remove(&ofconn->node);
1357 rconn_destroy(ofconn->rconn);
1358 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1359 rconn_packet_counter_destroy(ofconn->reply_counter);
1360 pktbuf_destroy(ofconn->pktbuf);
1365 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1369 rconn_run(ofconn->rconn);
1371 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1372 /* Limit the number of iterations to prevent other tasks from
1374 for (iteration = 0; iteration < 50; iteration++) {
1375 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1380 fail_open_maybe_recover(p->fail_open);
1382 handle_openflow(ofconn, p, of_msg);
1383 ofpbuf_delete(of_msg);
1387 if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
1388 ofconn_destroy(ofconn);
1393 ofconn_wait(struct ofconn *ofconn)
1395 rconn_run_wait(ofconn->rconn);
1396 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1397 rconn_recv_wait(ofconn->rconn);
1399 COVERAGE_INC(ofproto_ofconn_stuck);
1403 /* Caller is responsible for initializing the 'cr' member of the returned
1405 static struct rule *
1406 rule_create(struct ofproto *ofproto, struct rule *super,
1407 const union ofp_action *actions, size_t n_actions,
1408 uint16_t idle_timeout, uint16_t hard_timeout,
1409 uint64_t flow_cookie, bool send_flow_removed)
1411 struct rule *rule = xzalloc(sizeof *rule);
1412 rule->idle_timeout = idle_timeout;
1413 rule->hard_timeout = hard_timeout;
1414 rule->flow_cookie = flow_cookie;
1415 rule->used = rule->created = time_msec();
1416 rule->send_flow_removed = send_flow_removed;
1417 rule->super = super;
1419 list_push_back(&super->list, &rule->list);
1421 list_init(&rule->list);
1423 rule->n_actions = n_actions;
1424 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1425 netflow_flow_clear(&rule->nf_flow);
1426 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1431 static struct rule *
1432 rule_from_cls_rule(const struct cls_rule *cls_rule)
1434 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1438 rule_free(struct rule *rule)
1440 free(rule->actions);
1441 free(rule->odp_actions);
1445 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1446 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1447 * through all of its subrules and revalidates them, destroying any that no
1448 * longer has a super-rule (which is probably all of them).
1450 * Before calling this function, the caller must make have removed 'rule' from
1451 * the classifier. If 'rule' is an exact-match rule, the caller is also
1452 * responsible for ensuring that it has been uninstalled from the datapath. */
1454 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1457 struct rule *subrule, *next;
1458 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1459 revalidate_rule(ofproto, subrule);
1462 list_remove(&rule->list);
1468 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1470 const union ofp_action *oa;
1471 struct actions_iterator i;
1473 if (out_port == htons(OFPP_NONE)) {
1476 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1477 oa = actions_next(&i)) {
1478 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1485 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1486 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1488 * The flow that 'packet' actually contains does not need to actually match
1489 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1490 * the packet and byte counters for 'rule' will be credited for the packet sent
1491 * out whether or not the packet actually matches 'rule'.
1493 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1494 * the caller must already have accurately composed ODP actions for it given
1495 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1496 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1497 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1498 * actions and apply them to 'packet'. */
1500 rule_execute(struct ofproto *ofproto, struct rule *rule,
1501 struct ofpbuf *packet, const flow_t *flow)
1503 const union odp_action *actions;
1505 struct odp_actions a;
1507 /* Grab or compose the ODP actions.
1509 * The special case for an exact-match 'rule' where 'flow' is not the
1510 * rule's flow is important to avoid, e.g., sending a packet out its input
1511 * port simply because the ODP actions were composed for the wrong
1513 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1514 struct rule *super = rule->super ? rule->super : rule;
1515 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1516 packet, &a, NULL, 0, NULL)) {
1519 actions = a.actions;
1520 n_actions = a.n_actions;
1522 actions = rule->odp_actions;
1523 n_actions = rule->n_odp_actions;
1526 /* Execute the ODP actions. */
1527 if (!dpif_execute(ofproto->dpif, flow->in_port,
1528 actions, n_actions, packet)) {
1529 struct odp_flow_stats stats;
1530 flow_extract_stats(flow, packet, &stats);
1531 update_stats(ofproto, rule, &stats);
1532 rule->used = time_msec();
1533 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1538 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1541 struct rule *displaced_rule;
1543 /* Insert the rule in the classifier. */
1544 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1545 if (!rule->cr.wc.wildcards) {
1546 rule_make_actions(p, rule, packet);
1549 /* Send the packet and credit it to the rule. */
1552 flow_extract(packet, in_port, &flow);
1553 rule_execute(p, rule, packet, &flow);
1556 /* Install the rule in the datapath only after sending the packet, to
1557 * avoid packet reordering. */
1558 if (rule->cr.wc.wildcards) {
1559 COVERAGE_INC(ofproto_add_wc_flow);
1560 p->need_revalidate = true;
1562 rule_install(p, rule, displaced_rule);
1565 /* Free the rule that was displaced, if any. */
1566 if (displaced_rule) {
1567 rule_destroy(p, displaced_rule);
1571 static struct rule *
1572 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1575 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1576 rule->idle_timeout, rule->hard_timeout,
1578 COVERAGE_INC(ofproto_subrule_create);
1579 cls_rule_from_flow(&subrule->cr, flow, 0,
1580 (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1581 : rule->cr.priority));
1582 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1588 rule_remove(struct ofproto *ofproto, struct rule *rule)
1590 if (rule->cr.wc.wildcards) {
1591 COVERAGE_INC(ofproto_del_wc_flow);
1592 ofproto->need_revalidate = true;
1594 rule_uninstall(ofproto, rule);
1596 classifier_remove(&ofproto->cls, &rule->cr);
1597 rule_destroy(ofproto, rule);
1600 /* Returns true if the actions changed, false otherwise. */
1602 rule_make_actions(struct ofproto *p, struct rule *rule,
1603 const struct ofpbuf *packet)
1605 const struct rule *super;
1606 struct odp_actions a;
1609 assert(!rule->cr.wc.wildcards);
1611 super = rule->super ? rule->super : rule;
1613 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1614 packet, &a, &rule->tags, &rule->may_install,
1615 &rule->nf_flow.output_iface);
1617 actions_len = a.n_actions * sizeof *a.actions;
1618 if (rule->n_odp_actions != a.n_actions
1619 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1620 COVERAGE_INC(ofproto_odp_unchanged);
1621 free(rule->odp_actions);
1622 rule->n_odp_actions = a.n_actions;
1623 rule->odp_actions = xmemdup(a.actions, actions_len);
1631 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1632 struct odp_flow_put *put)
1634 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1635 put->flow.key = rule->cr.flow;
1636 put->flow.actions = rule->odp_actions;
1637 put->flow.n_actions = rule->n_odp_actions;
1639 return dpif_flow_put(ofproto->dpif, put);
1643 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1645 assert(!rule->cr.wc.wildcards);
1647 if (rule->may_install) {
1648 struct odp_flow_put put;
1649 if (!do_put_flow(p, rule,
1650 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1652 rule->installed = true;
1653 if (displaced_rule) {
1654 update_stats(p, displaced_rule, &put.flow.stats);
1655 rule_post_uninstall(p, displaced_rule);
1658 } else if (displaced_rule) {
1659 rule_uninstall(p, displaced_rule);
1664 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1666 if (rule->installed) {
1667 struct odp_flow_put put;
1668 COVERAGE_INC(ofproto_dp_missed);
1669 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1671 rule_install(ofproto, rule, NULL);
1676 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1678 bool actions_changed;
1679 uint16_t new_out_iface, old_out_iface;
1681 old_out_iface = rule->nf_flow.output_iface;
1682 actions_changed = rule_make_actions(ofproto, rule, NULL);
1684 if (rule->may_install) {
1685 if (rule->installed) {
1686 if (actions_changed) {
1687 struct odp_flow_put put;
1688 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
1689 | ODPPF_ZERO_STATS, &put);
1690 update_stats(ofproto, rule, &put.flow.stats);
1692 /* Temporarily set the old output iface so that NetFlow
1693 * messages have the correct output interface for the old
1695 new_out_iface = rule->nf_flow.output_iface;
1696 rule->nf_flow.output_iface = old_out_iface;
1697 rule_post_uninstall(ofproto, rule);
1698 rule->nf_flow.output_iface = new_out_iface;
1701 rule_install(ofproto, rule, NULL);
1704 rule_uninstall(ofproto, rule);
1709 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1711 uint64_t total_bytes = rule->byte_count + extra_bytes;
1713 if (ofproto->ofhooks->account_flow_cb
1714 && total_bytes > rule->accounted_bytes)
1716 ofproto->ofhooks->account_flow_cb(
1717 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1718 total_bytes - rule->accounted_bytes, ofproto->aux);
1719 rule->accounted_bytes = total_bytes;
1724 rule_uninstall(struct ofproto *p, struct rule *rule)
1726 assert(!rule->cr.wc.wildcards);
1727 if (rule->installed) {
1728 struct odp_flow odp_flow;
1730 odp_flow.key = rule->cr.flow;
1731 odp_flow.actions = NULL;
1732 odp_flow.n_actions = 0;
1733 if (!dpif_flow_del(p->dpif, &odp_flow)) {
1734 update_stats(p, rule, &odp_flow.stats);
1736 rule->installed = false;
1738 rule_post_uninstall(p, rule);
1743 is_controller_rule(struct rule *rule)
1745 /* If the only action is send to the controller then don't report
1746 * NetFlow expiration messages since it is just part of the control
1747 * logic for the network and not real traffic. */
1749 if (rule && rule->super) {
1750 struct rule *super = rule->super;
1752 return super->n_actions == 1 &&
1753 super->actions[0].type == htons(OFPAT_OUTPUT) &&
1754 super->actions[0].output.port == htons(OFPP_CONTROLLER);
1761 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1763 struct rule *super = rule->super;
1765 rule_account(ofproto, rule, 0);
1767 if (ofproto->netflow && !is_controller_rule(rule)) {
1768 struct ofexpired expired;
1769 expired.flow = rule->cr.flow;
1770 expired.packet_count = rule->packet_count;
1771 expired.byte_count = rule->byte_count;
1772 expired.used = rule->used;
1773 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
1776 super->packet_count += rule->packet_count;
1777 super->byte_count += rule->byte_count;
1779 /* Reset counters to prevent double counting if the rule ever gets
1781 rule->packet_count = 0;
1782 rule->byte_count = 0;
1783 rule->accounted_bytes = 0;
1785 netflow_flow_clear(&rule->nf_flow);
1790 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1791 struct rconn_packet_counter *counter)
1793 update_openflow_length(msg);
1794 if (rconn_send(ofconn->rconn, msg, counter)) {
1800 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1801 int error, const void *data, size_t len)
1804 struct ofp_error_msg *oem;
1806 if (!(error >> 16)) {
1807 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1812 COVERAGE_INC(ofproto_error);
1813 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1814 oh ? oh->xid : 0, &buf);
1815 oem->type = htons((unsigned int) error >> 16);
1816 oem->code = htons(error & 0xffff);
1817 memcpy(oem->data, data, len);
1818 queue_tx(buf, ofconn, ofconn->reply_counter);
1822 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1825 size_t oh_length = ntohs(oh->length);
1826 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1830 hton_ofp_phy_port(struct ofp_phy_port *opp)
1832 opp->port_no = htons(opp->port_no);
1833 opp->config = htonl(opp->config);
1834 opp->state = htonl(opp->state);
1835 opp->curr = htonl(opp->curr);
1836 opp->advertised = htonl(opp->advertised);
1837 opp->supported = htonl(opp->supported);
1838 opp->peer = htonl(opp->peer);
1842 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
1844 struct ofp_header *rq = oh;
1845 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
1850 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
1851 struct ofp_header *oh)
1853 struct ofp_switch_features *osf;
1855 unsigned int port_no;
1856 struct ofport *port;
1858 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1859 osf->datapath_id = htonll(p->datapath_id);
1860 osf->n_buffers = htonl(pktbuf_capacity());
1862 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1863 OFPC_PORT_STATS | OFPC_MULTI_PHY_TX |
1865 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1866 (1u << OFPAT_SET_VLAN_VID) |
1867 (1u << OFPAT_SET_VLAN_PCP) |
1868 (1u << OFPAT_STRIP_VLAN) |
1869 (1u << OFPAT_SET_DL_SRC) |
1870 (1u << OFPAT_SET_DL_DST) |
1871 (1u << OFPAT_SET_NW_SRC) |
1872 (1u << OFPAT_SET_NW_DST) |
1873 (1u << OFPAT_SET_NW_TOS) |
1874 (1u << OFPAT_SET_TP_SRC) |
1875 (1u << OFPAT_SET_TP_DST));
1877 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1878 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1881 queue_tx(buf, ofconn, ofconn->reply_counter);
1886 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1887 struct ofp_header *oh)
1890 struct ofp_switch_config *osc;
1894 /* Figure out flags. */
1895 dpif_get_drop_frags(p->dpif, &drop_frags);
1896 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1899 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1900 osc->flags = htons(flags);
1901 osc->miss_send_len = htons(ofconn->miss_send_len);
1902 queue_tx(buf, ofconn, ofconn->reply_counter);
1908 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1909 struct ofp_switch_config *osc)
1914 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1918 flags = ntohs(osc->flags);
1920 if (ofconn == p->controller) {
1921 switch (flags & OFPC_FRAG_MASK) {
1922 case OFPC_FRAG_NORMAL:
1923 dpif_set_drop_frags(p->dpif, false);
1925 case OFPC_FRAG_DROP:
1926 dpif_set_drop_frags(p->dpif, true);
1929 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1935 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1936 if (ntohs(osc->miss_send_len) != 0) {
1937 ofconn->pktbuf = pktbuf_create();
1939 pktbuf_destroy(ofconn->pktbuf);
1943 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1949 add_output_group_action(struct odp_actions *actions, uint16_t group,
1950 uint16_t *nf_output_iface)
1952 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1954 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
1955 *nf_output_iface = NF_OUT_FLOOD;
1960 add_controller_action(struct odp_actions *actions,
1961 const struct ofp_action_output *oao)
1963 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
1964 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
1967 struct action_xlate_ctx {
1969 const flow_t *flow; /* Flow to which these actions correspond. */
1970 int recurse; /* Recursion level, via xlate_table_action. */
1971 struct ofproto *ofproto;
1972 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
1973 * null pointer if we are revalidating
1974 * without a packet to refer to. */
1977 struct odp_actions *out; /* Datapath actions. */
1978 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
1979 bool may_set_up_flow; /* True ordinarily; false if the actions must
1980 * be reassessed for every packet. */
1981 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
1984 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
1985 struct action_xlate_ctx *ctx);
1988 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
1990 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
1993 if (ofport->opp.config & OFPPC_NO_FWD) {
1994 /* Forwarding disabled on port. */
1999 * We don't have an ofport record for this port, but it doesn't hurt to
2000 * allow forwarding to it anyhow. Maybe such a port will appear later
2001 * and we're pre-populating the flow table.
2005 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2006 ctx->nf_output_iface = port;
2009 static struct rule *
2010 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2013 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2015 /* The rule we found might not be valid, since we could be in need of
2016 * revalidation. If it is not valid, don't return it. */
2019 && ofproto->need_revalidate
2020 && !revalidate_rule(ofproto, rule)) {
2021 COVERAGE_INC(ofproto_invalidated);
2029 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2031 if (!ctx->recurse) {
2036 flow.in_port = in_port;
2038 rule = lookup_valid_rule(ctx->ofproto, &flow);
2045 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2052 xlate_output_action(struct action_xlate_ctx *ctx,
2053 const struct ofp_action_output *oao)
2056 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2058 ctx->nf_output_iface = NF_OUT_DROP;
2060 switch (ntohs(oao->port)) {
2062 add_output_action(ctx, ctx->flow->in_port);
2065 xlate_table_action(ctx, ctx->flow->in_port);
2068 if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet,
2069 ctx->out, ctx->tags,
2070 &ctx->nf_output_iface,
2071 ctx->ofproto->aux)) {
2072 COVERAGE_INC(ofproto_uninstallable);
2073 ctx->may_set_up_flow = false;
2077 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2078 &ctx->nf_output_iface);
2081 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2083 case OFPP_CONTROLLER:
2084 add_controller_action(ctx->out, oao);
2087 add_output_action(ctx, ODPP_LOCAL);
2090 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2091 if (odp_port != ctx->flow->in_port) {
2092 add_output_action(ctx, odp_port);
2097 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2098 ctx->nf_output_iface = NF_OUT_FLOOD;
2099 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2100 ctx->nf_output_iface = prev_nf_output_iface;
2101 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2102 ctx->nf_output_iface != NF_OUT_FLOOD) {
2103 ctx->nf_output_iface = NF_OUT_MULTI;
2108 xlate_nicira_action(struct action_xlate_ctx *ctx,
2109 const struct nx_action_header *nah)
2111 const struct nx_action_resubmit *nar;
2112 int subtype = ntohs(nah->subtype);
2114 assert(nah->vendor == htonl(NX_VENDOR_ID));
2116 case NXAST_RESUBMIT:
2117 nar = (const struct nx_action_resubmit *) nah;
2118 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2122 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2128 do_xlate_actions(const union ofp_action *in, size_t n_in,
2129 struct action_xlate_ctx *ctx)
2131 struct actions_iterator iter;
2132 const union ofp_action *ia;
2133 const struct ofport *port;
2135 port = port_array_get(&ctx->ofproto->ports, ctx->flow->in_port);
2136 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2137 port->opp.config & (eth_addr_equals(ctx->flow->dl_dst, stp_eth_addr)
2138 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2139 /* Drop this flow. */
2143 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2144 uint16_t type = ntohs(ia->type);
2145 union odp_action *oa;
2149 xlate_output_action(ctx, &ia->output);
2152 case OFPAT_SET_VLAN_VID:
2153 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2154 oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2157 case OFPAT_SET_VLAN_PCP:
2158 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2159 oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2162 case OFPAT_STRIP_VLAN:
2163 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2166 case OFPAT_SET_DL_SRC:
2167 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2168 memcpy(oa->dl_addr.dl_addr,
2169 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2172 case OFPAT_SET_DL_DST:
2173 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2174 memcpy(oa->dl_addr.dl_addr,
2175 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2178 case OFPAT_SET_NW_SRC:
2179 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2180 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2183 case OFPAT_SET_NW_DST:
2184 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2185 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2187 case OFPAT_SET_NW_TOS:
2188 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2189 oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2192 case OFPAT_SET_TP_SRC:
2193 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2194 oa->tp_port.tp_port = ia->tp_port.tp_port;
2197 case OFPAT_SET_TP_DST:
2198 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2199 oa->tp_port.tp_port = ia->tp_port.tp_port;
2203 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2207 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2214 xlate_actions(const union ofp_action *in, size_t n_in,
2215 const flow_t *flow, struct ofproto *ofproto,
2216 const struct ofpbuf *packet,
2217 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2218 uint16_t *nf_output_iface)
2220 tag_type no_tags = 0;
2221 struct action_xlate_ctx ctx;
2222 COVERAGE_INC(ofproto_ofp2odp);
2223 odp_actions_init(out);
2226 ctx.ofproto = ofproto;
2227 ctx.packet = packet;
2229 ctx.tags = tags ? tags : &no_tags;
2230 ctx.may_set_up_flow = true;
2231 ctx.nf_output_iface = NF_OUT_DROP;
2232 do_xlate_actions(in, n_in, &ctx);
2234 /* Check with in-band control to see if we're allowed to set up this
2236 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2237 ctx.may_set_up_flow = false;
2240 if (may_set_up_flow) {
2241 *may_set_up_flow = ctx.may_set_up_flow;
2243 if (nf_output_iface) {
2244 *nf_output_iface = ctx.nf_output_iface;
2246 if (odp_actions_overflow(out)) {
2247 odp_actions_init(out);
2248 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2254 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2255 struct ofp_header *oh)
2257 struct ofp_packet_out *opo;
2258 struct ofpbuf payload, *buffer;
2259 struct odp_actions actions;
2265 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2269 opo = (struct ofp_packet_out *) oh;
2271 COVERAGE_INC(ofproto_packet_out);
2272 if (opo->buffer_id != htonl(UINT32_MAX)) {
2273 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2275 if (error || !buffer) {
2283 flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2284 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2285 &flow, p, &payload, &actions, NULL, NULL, NULL);
2290 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2292 ofpbuf_delete(buffer);
2298 update_port_config(struct ofproto *p, struct ofport *port,
2299 uint32_t config, uint32_t mask)
2301 mask &= config ^ port->opp.config;
2302 if (mask & OFPPC_PORT_DOWN) {
2303 if (config & OFPPC_PORT_DOWN) {
2304 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2306 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2309 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2310 if (mask & REVALIDATE_BITS) {
2311 COVERAGE_INC(ofproto_costly_flags);
2312 port->opp.config ^= mask & REVALIDATE_BITS;
2313 p->need_revalidate = true;
2315 #undef REVALIDATE_BITS
2316 if (mask & OFPPC_NO_FLOOD) {
2317 port->opp.config ^= OFPPC_NO_FLOOD;
2318 refresh_port_groups(p);
2320 if (mask & OFPPC_NO_PACKET_IN) {
2321 port->opp.config ^= OFPPC_NO_PACKET_IN;
2326 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2328 const struct ofp_port_mod *opm;
2329 struct ofport *port;
2332 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2336 opm = (struct ofp_port_mod *) oh;
2338 port = port_array_get(&p->ports,
2339 ofp_port_to_odp_port(ntohs(opm->port_no)));
2341 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2342 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2343 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2345 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2346 if (opm->advertise) {
2347 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2353 static struct ofpbuf *
2354 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2356 struct ofp_stats_reply *osr;
2359 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2360 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2362 osr->flags = htons(0);
2366 static struct ofpbuf *
2367 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2369 return make_stats_reply(request->header.xid, request->type, body_len);
2373 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2375 struct ofpbuf *msg = *msgp;
2376 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2377 if (nbytes + msg->size > UINT16_MAX) {
2378 struct ofp_stats_reply *reply = msg->data;
2379 reply->flags = htons(OFPSF_REPLY_MORE);
2380 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2381 queue_tx(msg, ofconn, ofconn->reply_counter);
2383 return ofpbuf_put_uninit(*msgp, nbytes);
2387 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2388 struct ofp_stats_request *request)
2390 struct ofp_desc_stats *ods;
2393 msg = start_stats_reply(request, sizeof *ods);
2394 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2395 strncpy(ods->mfr_desc, p->manufacturer, sizeof ods->mfr_desc);
2396 strncpy(ods->hw_desc, p->hardware, sizeof ods->hw_desc);
2397 strncpy(ods->sw_desc, p->software, sizeof ods->sw_desc);
2398 strncpy(ods->serial_num, p->serial, sizeof ods->serial_num);
2399 strncpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2400 queue_tx(msg, ofconn, ofconn->reply_counter);
2406 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2408 struct rule *rule = rule_from_cls_rule(cls_rule);
2409 int *n_subrules = n_subrules_;
2417 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2418 struct ofp_stats_request *request)
2420 struct ofp_table_stats *ots;
2422 struct odp_stats dpstats;
2423 int n_exact, n_subrules, n_wild;
2425 msg = start_stats_reply(request, sizeof *ots * 2);
2427 /* Count rules of various kinds. */
2429 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2430 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2431 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2434 dpif_get_dp_stats(p->dpif, &dpstats);
2435 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2436 memset(ots, 0, sizeof *ots);
2437 ots->table_id = TABLEID_HASH;
2438 strcpy(ots->name, "hash");
2439 ots->wildcards = htonl(0);
2440 ots->max_entries = htonl(dpstats.max_capacity);
2441 ots->active_count = htonl(n_exact);
2442 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2444 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2446 /* Classifier table. */
2447 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2448 memset(ots, 0, sizeof *ots);
2449 ots->table_id = TABLEID_CLASSIFIER;
2450 strcpy(ots->name, "classifier");
2451 ots->wildcards = htonl(OFPFW_ALL);
2452 ots->max_entries = htonl(65536);
2453 ots->active_count = htonl(n_wild);
2454 ots->lookup_count = htonll(0); /* XXX */
2455 ots->matched_count = htonll(0); /* XXX */
2457 queue_tx(msg, ofconn, ofconn->reply_counter);
2462 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2463 struct ofp_stats_request *request)
2465 struct ofp_port_stats *ops;
2467 struct ofport *port;
2468 unsigned int port_no;
2470 msg = start_stats_reply(request, sizeof *ops * 16);
2471 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2472 struct netdev_stats stats;
2474 /* Intentionally ignore return value, since errors will set 'stats' to
2475 * all-1s, which is correct for OpenFlow, and netdev_get_stats() will
2477 netdev_get_stats(port->netdev, &stats);
2479 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2480 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2481 memset(ops->pad, 0, sizeof ops->pad);
2482 ops->rx_packets = htonll(stats.rx_packets);
2483 ops->tx_packets = htonll(stats.tx_packets);
2484 ops->rx_bytes = htonll(stats.rx_bytes);
2485 ops->tx_bytes = htonll(stats.tx_bytes);
2486 ops->rx_dropped = htonll(stats.rx_dropped);
2487 ops->tx_dropped = htonll(stats.tx_dropped);
2488 ops->rx_errors = htonll(stats.rx_errors);
2489 ops->tx_errors = htonll(stats.tx_errors);
2490 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2491 ops->rx_over_err = htonll(stats.rx_over_errors);
2492 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2493 ops->collisions = htonll(stats.collisions);
2496 queue_tx(msg, ofconn, ofconn->reply_counter);
2500 struct flow_stats_cbdata {
2501 struct ofproto *ofproto;
2502 struct ofconn *ofconn;
2508 query_stats(struct ofproto *p, struct rule *rule,
2509 uint64_t *packet_countp, uint64_t *byte_countp)
2511 uint64_t packet_count, byte_count;
2512 struct rule *subrule;
2513 struct odp_flow *odp_flows;
2516 packet_count = rule->packet_count;
2517 byte_count = rule->byte_count;
2519 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2520 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
2521 if (rule->cr.wc.wildcards) {
2523 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2524 odp_flows[i++].key = subrule->cr.flow;
2525 packet_count += subrule->packet_count;
2526 byte_count += subrule->byte_count;
2529 odp_flows[0].key = rule->cr.flow;
2532 packet_count = rule->packet_count;
2533 byte_count = rule->byte_count;
2534 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2536 for (i = 0; i < n_odp_flows; i++) {
2537 struct odp_flow *odp_flow = &odp_flows[i];
2538 packet_count += odp_flow->stats.n_packets;
2539 byte_count += odp_flow->stats.n_bytes;
2544 *packet_countp = packet_count;
2545 *byte_countp = byte_count;
2549 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2551 struct rule *rule = rule_from_cls_rule(rule_);
2552 struct flow_stats_cbdata *cbdata = cbdata_;
2553 struct ofp_flow_stats *ofs;
2554 uint64_t packet_count, byte_count;
2555 size_t act_len, len;
2557 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2561 act_len = sizeof *rule->actions * rule->n_actions;
2562 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2564 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2566 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2567 ofs->length = htons(len);
2568 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2570 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match);
2571 ofs->duration = htonl((time_msec() - rule->created) / 1000);
2572 ofs->cookie = rule->flow_cookie;
2573 ofs->priority = htons(rule->cr.priority);
2574 ofs->idle_timeout = htons(rule->idle_timeout);
2575 ofs->hard_timeout = htons(rule->hard_timeout);
2576 memset(ofs->pad2, 0, sizeof ofs->pad2);
2577 ofs->packet_count = htonll(packet_count);
2578 ofs->byte_count = htonll(byte_count);
2579 memcpy(ofs->actions, rule->actions, act_len);
2583 table_id_to_include(uint8_t table_id)
2585 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2586 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2587 : table_id == 0xff ? CLS_INC_ALL
2592 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2593 const struct ofp_stats_request *osr,
2596 struct ofp_flow_stats_request *fsr;
2597 struct flow_stats_cbdata cbdata;
2598 struct cls_rule target;
2600 if (arg_size != sizeof *fsr) {
2601 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2603 fsr = (struct ofp_flow_stats_request *) osr->body;
2605 COVERAGE_INC(ofproto_flows_req);
2607 cbdata.ofconn = ofconn;
2608 cbdata.out_port = fsr->out_port;
2609 cbdata.msg = start_stats_reply(osr, 1024);
2610 cls_rule_from_match(&target, &fsr->match, 0);
2611 classifier_for_each_match(&p->cls, &target,
2612 table_id_to_include(fsr->table_id),
2613 flow_stats_cb, &cbdata);
2614 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2618 struct flow_stats_ds_cbdata {
2619 struct ofproto *ofproto;
2624 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
2626 struct rule *rule = rule_from_cls_rule(rule_);
2627 struct flow_stats_ds_cbdata *cbdata = cbdata_;
2628 struct ds *results = cbdata->results;
2629 struct ofp_match match;
2630 uint64_t packet_count, byte_count;
2631 size_t act_len = sizeof *rule->actions * rule->n_actions;
2633 /* Don't report on subrules. */
2634 if (rule->super != NULL) {
2638 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2639 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &match);
2641 ds_put_format(results, "duration=%llds, ",
2642 (time_msec() - rule->created) / 1000);
2643 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2644 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2645 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2646 ofp_print_match(results, &match, true);
2647 ofp_print_actions(results, &rule->actions->header, act_len);
2648 ds_put_cstr(results, "\n");
2651 /* Adds a pretty-printed description of all flows to 'results', including
2652 * those marked hidden by secchan (e.g., by in-band control). */
2654 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2656 struct ofp_match match;
2657 struct cls_rule target;
2658 struct flow_stats_ds_cbdata cbdata;
2660 memset(&match, 0, sizeof match);
2661 match.wildcards = htonl(OFPFW_ALL);
2664 cbdata.results = results;
2666 cls_rule_from_match(&target, &match, 0);
2667 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2668 flow_stats_ds_cb, &cbdata);
2671 struct aggregate_stats_cbdata {
2672 struct ofproto *ofproto;
2674 uint64_t packet_count;
2675 uint64_t byte_count;
2680 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2682 struct rule *rule = rule_from_cls_rule(rule_);
2683 struct aggregate_stats_cbdata *cbdata = cbdata_;
2684 uint64_t packet_count, byte_count;
2686 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2690 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2692 cbdata->packet_count += packet_count;
2693 cbdata->byte_count += byte_count;
2698 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2699 const struct ofp_stats_request *osr,
2702 struct ofp_aggregate_stats_request *asr;
2703 struct ofp_aggregate_stats_reply *reply;
2704 struct aggregate_stats_cbdata cbdata;
2705 struct cls_rule target;
2708 if (arg_size != sizeof *asr) {
2709 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2711 asr = (struct ofp_aggregate_stats_request *) osr->body;
2713 COVERAGE_INC(ofproto_agg_request);
2715 cbdata.out_port = asr->out_port;
2716 cbdata.packet_count = 0;
2717 cbdata.byte_count = 0;
2719 cls_rule_from_match(&target, &asr->match, 0);
2720 classifier_for_each_match(&p->cls, &target,
2721 table_id_to_include(asr->table_id),
2722 aggregate_stats_cb, &cbdata);
2724 msg = start_stats_reply(osr, sizeof *reply);
2725 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2726 reply->flow_count = htonl(cbdata.n_flows);
2727 reply->packet_count = htonll(cbdata.packet_count);
2728 reply->byte_count = htonll(cbdata.byte_count);
2729 queue_tx(msg, ofconn, ofconn->reply_counter);
2734 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2735 struct ofp_header *oh)
2737 struct ofp_stats_request *osr;
2741 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2746 osr = (struct ofp_stats_request *) oh;
2748 switch (ntohs(osr->type)) {
2750 return handle_desc_stats_request(p, ofconn, osr);
2753 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2755 case OFPST_AGGREGATE:
2756 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2759 return handle_table_stats_request(p, ofconn, osr);
2762 return handle_port_stats_request(p, ofconn, osr);
2765 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2768 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2772 static long long int
2773 msec_from_nsec(uint64_t sec, uint32_t nsec)
2775 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2779 update_time(struct ofproto *ofproto, struct rule *rule,
2780 const struct odp_flow_stats *stats)
2782 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2783 if (used > rule->used) {
2785 if (rule->super && used > rule->super->used) {
2786 rule->super->used = used;
2788 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
2793 update_stats(struct ofproto *ofproto, struct rule *rule,
2794 const struct odp_flow_stats *stats)
2796 if (stats->n_packets) {
2797 update_time(ofproto, rule, stats);
2798 rule->packet_count += stats->n_packets;
2799 rule->byte_count += stats->n_bytes;
2800 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
2806 add_flow(struct ofproto *p, struct ofconn *ofconn,
2807 struct ofp_flow_mod *ofm, size_t n_actions)
2809 struct ofpbuf *packet;
2814 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
2818 flow_from_match(&flow, &wildcards, &ofm->match);
2819 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
2820 ntohs(ofm->priority))) {
2821 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
2825 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
2826 n_actions, ntohs(ofm->idle_timeout),
2827 ntohs(ofm->hard_timeout), ofm->cookie,
2828 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
2829 cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority));
2832 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2833 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2837 in_port = UINT16_MAX;
2840 rule_insert(p, rule, packet, in_port);
2841 ofpbuf_delete(packet);
2846 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
2847 size_t n_actions, uint16_t command, struct rule *rule)
2849 if (rule_is_hidden(rule)) {
2853 if (command == OFPFC_DELETE) {
2854 long long int now = time_msec();
2855 send_flow_removed(p, rule, now, OFPRR_DELETE);
2856 rule_remove(p, rule);
2858 size_t actions_len = n_actions * sizeof *rule->actions;
2860 if (n_actions == rule->n_actions
2861 && !memcmp(ofm->actions, rule->actions, actions_len))
2866 free(rule->actions);
2867 rule->actions = xmemdup(ofm->actions, actions_len);
2868 rule->n_actions = n_actions;
2869 rule->flow_cookie = ofm->cookie;
2871 if (rule->cr.wc.wildcards) {
2872 COVERAGE_INC(ofproto_mod_wc_flow);
2873 p->need_revalidate = true;
2875 rule_update_actions(p, rule);
2883 modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm,
2884 size_t n_actions, uint16_t command)
2890 flow_from_match(&flow, &wildcards, &ofm->match);
2891 rule = rule_from_cls_rule(classifier_find_rule_exactly(
2892 &p->cls, &flow, wildcards,
2893 ntohs(ofm->priority)));
2896 if (command == OFPFC_DELETE
2897 && ofm->out_port != htons(OFPP_NONE)
2898 && !rule_has_out_port(rule, ofm->out_port)) {
2902 modify_flow(p, ofm, n_actions, command, rule);
2907 struct modify_flows_cbdata {
2908 struct ofproto *ofproto;
2909 const struct ofp_flow_mod *ofm;
2916 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
2918 struct rule *rule = rule_from_cls_rule(rule_);
2919 struct modify_flows_cbdata *cbdata = cbdata_;
2921 if (cbdata->out_port != htons(OFPP_NONE)
2922 && !rule_has_out_port(rule, cbdata->out_port)) {
2926 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions,
2927 cbdata->command, rule);
2931 modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm,
2932 size_t n_actions, uint16_t command)
2934 struct modify_flows_cbdata cbdata;
2935 struct cls_rule target;
2939 cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port
2940 : htons(OFPP_NONE));
2941 cbdata.n_actions = n_actions;
2942 cbdata.command = command;
2944 cls_rule_from_match(&target, &ofm->match, 0);
2946 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2947 modify_flows_cb, &cbdata);
2952 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
2953 struct ofp_flow_mod *ofm)
2958 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
2959 sizeof *ofm->actions, &n_actions);
2964 /* We do not support the emergency flow cache. It will hopefully
2965 * get dropped from OpenFlow in the near future. */
2966 if (ofm->flags & htons(OFPFF_EMERG)) {
2967 /* There isn't a good fit for an error code, so just state that the
2968 * flow table is full. */
2969 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
2972 normalize_match(&ofm->match);
2973 if (!ofm->match.wildcards) {
2974 ofm->priority = htons(UINT16_MAX);
2977 error = validate_actions((const union ofp_action *) ofm->actions,
2978 n_actions, p->max_ports);
2983 switch (ntohs(ofm->command)) {
2985 return add_flow(p, ofconn, ofm, n_actions);
2988 return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY);
2990 case OFPFC_MODIFY_STRICT:
2991 return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY);
2994 return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE);
2996 case OFPFC_DELETE_STRICT:
2997 return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE);
3000 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3005 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3007 struct ofp_vendor_header *ovh = msg;
3008 struct nicira_header *nh;
3010 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3011 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3013 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3014 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3016 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3017 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3021 switch (ntohl(nh->subtype)) {
3022 case NXT_STATUS_REQUEST:
3023 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3027 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3031 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
3033 struct ofp_header *ob;
3036 /* Currently, everything executes synchronously, so we can just
3037 * immediately send the barrier reply. */
3038 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3039 queue_tx(buf, ofconn, ofconn->reply_counter);
3044 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3045 struct ofpbuf *ofp_msg)
3047 struct ofp_header *oh = ofp_msg->data;
3050 COVERAGE_INC(ofproto_recv_openflow);
3052 case OFPT_ECHO_REQUEST:
3053 error = handle_echo_request(ofconn, oh);
3056 case OFPT_ECHO_REPLY:
3060 case OFPT_FEATURES_REQUEST:
3061 error = handle_features_request(p, ofconn, oh);
3064 case OFPT_GET_CONFIG_REQUEST:
3065 error = handle_get_config_request(p, ofconn, oh);
3068 case OFPT_SET_CONFIG:
3069 error = handle_set_config(p, ofconn, ofp_msg->data);
3072 case OFPT_PACKET_OUT:
3073 error = handle_packet_out(p, ofconn, ofp_msg->data);
3077 error = handle_port_mod(p, oh);
3081 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3084 case OFPT_STATS_REQUEST:
3085 error = handle_stats_request(p, ofconn, oh);
3089 error = handle_vendor(p, ofconn, ofp_msg->data);
3092 case OFPT_BARRIER_REQUEST:
3093 error = handle_barrier_request(ofconn, oh);
3097 if (VLOG_IS_WARN_ENABLED()) {
3098 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3099 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3102 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3107 send_error_oh(ofconn, ofp_msg->data, error);
3112 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
3114 struct odp_msg *msg = packet->data;
3115 uint16_t in_port = odp_port_to_ofp_port(msg->port);
3117 struct ofpbuf payload;
3120 payload.data = msg + 1;
3121 payload.size = msg->length - sizeof *msg;
3122 flow_extract(&payload, msg->port, &flow);
3124 /* Check with in-band control to see if this packet should be sent
3125 * to the local port regardless of the flow table. */
3126 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3127 union odp_action action;
3129 memset(&action, 0, sizeof(action));
3130 action.output.type = ODPAT_OUTPUT;
3131 action.output.port = ODPP_LOCAL;
3132 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3135 rule = lookup_valid_rule(p, &flow);
3137 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3138 struct ofport *port = port_array_get(&p->ports, msg->port);
3140 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3141 COVERAGE_INC(ofproto_no_packet_in);
3142 /* XXX install 'drop' flow entry */
3143 ofpbuf_delete(packet);
3147 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3150 COVERAGE_INC(ofproto_packet_in);
3151 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3155 if (rule->cr.wc.wildcards) {
3156 rule = rule_create_subrule(p, rule, &flow);
3157 rule_make_actions(p, rule, packet);
3159 if (!rule->may_install) {
3160 /* The rule is not installable, that is, we need to process every
3161 * packet, so process the current packet and set its actions into
3163 rule_make_actions(p, rule, packet);
3165 /* XXX revalidate rule if it needs it */
3169 rule_execute(p, rule, &payload, &flow);
3170 rule_reinstall(p, rule);
3172 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY
3173 && rconn_is_connected(p->controller->rconn)) {
3175 * Extra-special case for fail-open mode.
3177 * We are in fail-open mode and the packet matched the fail-open rule,
3178 * but we are connected to a controller too. We should send the packet
3179 * up to the controller in the hope that it will try to set up a flow
3180 * and thereby allow us to exit fail-open.
3182 * See the top-level comment in fail-open.c for more information.
3184 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3186 ofpbuf_delete(packet);
3191 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3193 struct odp_msg *msg = packet->data;
3195 switch (msg->type) {
3196 case _ODPL_ACTION_NR:
3197 COVERAGE_INC(ofproto_ctlr_action);
3198 pinsched_send(p->action_sched, odp_port_to_ofp_port(msg->port), packet,
3199 send_packet_in_action, p);
3202 case _ODPL_SFLOW_NR:
3204 ofproto_sflow_received(p->sflow, msg);
3206 ofpbuf_delete(packet);
3210 handle_odp_miss_msg(p, packet);
3214 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
3221 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3223 struct rule *sub = rule_from_cls_rule(sub_);
3224 struct revalidate_cbdata *cbdata = cbdata_;
3226 if (cbdata->revalidate_all
3227 || (cbdata->revalidate_subrules && sub->super)
3228 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3229 revalidate_rule(cbdata->ofproto, sub);
3234 revalidate_rule(struct ofproto *p, struct rule *rule)
3236 const flow_t *flow = &rule->cr.flow;
3238 COVERAGE_INC(ofproto_revalidate_rule);
3241 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3243 rule_remove(p, rule);
3245 } else if (super != rule->super) {
3246 COVERAGE_INC(ofproto_revalidate_moved);
3247 list_remove(&rule->list);
3248 list_push_back(&super->list, &rule->list);
3249 rule->super = super;
3250 rule->hard_timeout = super->hard_timeout;
3251 rule->idle_timeout = super->idle_timeout;
3252 rule->created = super->created;
3257 rule_update_actions(p, rule);
3261 static struct ofpbuf *
3262 compose_flow_removed(const struct rule *rule, long long int now, uint8_t reason)
3264 struct ofp_flow_removed *ofr;
3267 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
3268 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofr->match);
3269 ofr->cookie = rule->flow_cookie;
3270 ofr->priority = htons(rule->cr.priority);
3271 ofr->reason = reason;
3272 ofr->duration = htonl((now - rule->created) / 1000);
3273 ofr->idle_timeout = htons(rule->idle_timeout);
3274 ofr->packet_count = htonll(rule->packet_count);
3275 ofr->byte_count = htonll(rule->byte_count);
3281 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3283 assert(rule->installed);
3284 assert(!rule->cr.wc.wildcards);
3287 rule_remove(ofproto, rule);
3289 rule_uninstall(ofproto, rule);
3293 send_flow_removed(struct ofproto *p, struct rule *rule,
3294 long long int now, uint8_t reason)
3296 struct ofconn *ofconn;
3297 struct ofconn *prev;
3298 struct ofpbuf *buf = NULL;
3300 /* We limit the maximum number of queued flow expirations it by accounting
3301 * them under the counter for replies. That works because preventing
3302 * OpenFlow requests from being processed also prevents new flows from
3303 * being added (and expiring). (It also prevents processing OpenFlow
3304 * requests that would not add new flows, so it is imperfect.) */
3307 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3308 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)) {
3310 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3312 buf = compose_flow_removed(rule, now, reason);
3318 queue_tx(buf, prev, prev->reply_counter);
3324 expire_rule(struct cls_rule *cls_rule, void *p_)
3326 struct ofproto *p = p_;
3327 struct rule *rule = rule_from_cls_rule(cls_rule);
3328 long long int hard_expire, idle_expire, expire, now;
3330 hard_expire = (rule->hard_timeout
3331 ? rule->created + rule->hard_timeout * 1000
3333 idle_expire = (rule->idle_timeout
3334 && (rule->super || list_is_empty(&rule->list))
3335 ? rule->used + rule->idle_timeout * 1000
3337 expire = MIN(hard_expire, idle_expire);
3341 if (rule->installed && now >= rule->used + 5000) {
3342 uninstall_idle_flow(p, rule);
3343 } else if (!rule->cr.wc.wildcards) {
3344 active_timeout(p, rule);
3350 COVERAGE_INC(ofproto_expired);
3352 /* Update stats. This code will be a no-op if the rule expired
3353 * due to an idle timeout. */
3354 if (rule->cr.wc.wildcards) {
3355 struct rule *subrule, *next;
3356 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3357 rule_remove(p, subrule);
3360 rule_uninstall(p, rule);
3363 if (!rule_is_hidden(rule)) {
3364 send_flow_removed(p, rule, now,
3366 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
3368 rule_remove(p, rule);
3372 active_timeout(struct ofproto *ofproto, struct rule *rule)
3374 if (ofproto->netflow && !is_controller_rule(rule) &&
3375 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
3376 struct ofexpired expired;
3377 struct odp_flow odp_flow;
3379 /* Get updated flow stats. */
3380 memset(&odp_flow, 0, sizeof odp_flow);
3381 if (rule->installed) {
3382 odp_flow.key = rule->cr.flow;
3383 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
3384 dpif_flow_get(ofproto->dpif, &odp_flow);
3386 if (odp_flow.stats.n_packets) {
3387 update_time(ofproto, rule, &odp_flow.stats);
3388 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
3389 odp_flow.stats.tcp_flags);
3393 expired.flow = rule->cr.flow;
3394 expired.packet_count = rule->packet_count +
3395 odp_flow.stats.n_packets;
3396 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
3397 expired.used = rule->used;
3399 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
3401 /* Schedule us to send the accumulated records once we have
3402 * collected all of them. */
3403 poll_immediate_wake();
3408 update_used(struct ofproto *p)
3410 struct odp_flow *flows;
3415 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
3420 for (i = 0; i < n_flows; i++) {
3421 struct odp_flow *f = &flows[i];
3424 rule = rule_from_cls_rule(
3425 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3426 if (!rule || !rule->installed) {
3427 COVERAGE_INC(ofproto_unexpected_rule);
3428 dpif_flow_del(p->dpif, f);
3432 update_time(p, rule, &f->stats);
3433 rule_account(p, rule, f->stats.n_bytes);
3439 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3440 const struct ofpbuf *packet, int send_len)
3442 struct odp_msg *msg = packet->data;
3443 struct ofpbuf payload;
3447 /* Extract packet payload from 'msg'. */
3448 payload.data = msg + 1;
3449 payload.size = msg->length - sizeof *msg;
3451 /* Construct ofp_packet_in message. */
3452 reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3453 opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port), reason,
3454 &payload, send_len);
3457 rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100);
3461 send_packet_in_action(struct ofpbuf *packet, void *p_)
3463 struct ofproto *p = p_;
3464 struct ofconn *ofconn;
3465 struct odp_msg *msg;
3468 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3469 if (ofconn == p->controller || ofconn->miss_send_len) {
3470 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3473 ofpbuf_delete(packet);
3477 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3479 struct ofproto *p = p_;
3480 bool in_fail_open = p->fail_open && fail_open_is_active(p->fail_open);
3481 struct ofconn *ofconn;
3482 struct ofpbuf payload;
3483 struct odp_msg *msg;
3486 payload.data = msg + 1;
3487 payload.size = msg->length - sizeof *msg;
3488 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3489 if (ofconn->miss_send_len) {
3490 struct pktbuf *pb = ofconn->pktbuf;
3491 uint32_t buffer_id = (in_fail_open
3493 : pktbuf_save(pb, &payload, msg->port));
3494 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3496 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3499 ofpbuf_delete(packet);
3503 pick_datapath_id(const struct ofproto *ofproto)
3505 const struct ofport *port;
3507 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
3509 uint8_t ea[ETH_ADDR_LEN];
3512 error = netdev_get_etheraddr(port->netdev, ea);
3514 return eth_addr_to_uint64(ea);
3516 VLOG_WARN("could not get MAC address for %s (%s)",
3517 netdev_get_name(port->netdev), strerror(error));
3519 return ofproto->fallback_dpid;
3523 pick_fallback_dpid(void)
3525 uint8_t ea[ETH_ADDR_LEN];
3526 eth_addr_nicira_random(ea);
3527 return eth_addr_to_uint64(ea);
3531 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3532 struct odp_actions *actions, tag_type *tags,
3533 uint16_t *nf_output_iface, void *ofproto_)
3535 struct ofproto *ofproto = ofproto_;
3538 /* Drop frames for reserved multicast addresses. */
3539 if (eth_addr_is_reserved(flow->dl_dst)) {
3543 /* Learn source MAC (but don't try to learn from revalidation). */
3544 if (packet != NULL) {
3545 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3548 /* The log messages here could actually be useful in debugging,
3549 * so keep the rate limit relatively high. */
3550 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3551 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3552 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3553 ofproto_revalidate(ofproto, rev_tag);
3557 /* Determine output port. */
3558 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3560 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
3561 } else if (out_port != flow->in_port) {
3562 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3563 *nf_output_iface = out_port;
3571 static const struct ofhooks default_ofhooks = {
3573 default_normal_ofhook_cb,