2 * Copyright (c) 2009 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
22 #include <netinet/in.h>
25 #include "classifier.h"
27 #include "discovery.h"
29 #include "dynamic-string.h"
31 #include "fail-open.h"
33 #include "mac-learning.h"
37 #include "ofp-print.h"
39 #include "openflow/nicira-ext.h"
40 #include "openflow/openflow.h"
41 #include "openflow/openflow-mgmt.h"
42 #include "openvswitch/datapath-protocol.h"
46 #include "poll-loop.h"
47 #include "port-array.h"
57 #include "vconn-ssl.h"
60 #define THIS_MODULE VLM_ofproto
70 TABLEID_CLASSIFIER = 1
74 struct netdev *netdev;
75 struct ofp_phy_port opp; /* In host byte order. */
78 static void ofport_free(struct ofport *);
79 static void hton_ofp_phy_port(struct ofp_phy_port *);
81 static int xlate_actions(const union ofp_action *in, size_t n_in,
82 const flow_t *flow, struct ofproto *ofproto,
83 const struct ofpbuf *packet,
84 struct odp_actions *out, tag_type *tags,
85 bool *may_set_up_flow, uint16_t *nf_output_iface);
90 uint16_t idle_timeout; /* In seconds from time of last use. */
91 uint16_t hard_timeout; /* In seconds from time of creation. */
92 bool send_flow_removed; /* Send a flow removed message? */
93 long long int used; /* Last-used time (0 if never used). */
94 long long int created; /* Creation time. */
95 uint64_t packet_count; /* Number of packets received. */
96 uint64_t byte_count; /* Number of bytes received. */
97 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
98 tag_type tags; /* Tags (set only by hooks). */
99 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
101 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
102 * exact-match rule (having cr.wc.wildcards of 0) generated from the
103 * wildcard rule 'super'. In this case, 'list' is an element of the
106 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
107 * a list of subrules. A super-rule with no wildcards (where
108 * cr.wc.wildcards is 0) will never have any subrules. */
114 * A subrule has no actions (it uses the super-rule's actions). */
116 union ofp_action *actions;
120 * A super-rule with wildcard fields never has ODP actions (since the
121 * datapath only supports exact-match flows). */
122 bool installed; /* Installed in datapath? */
123 bool may_install; /* True ordinarily; false if actions must
124 * be reassessed for every packet. */
126 union odp_action *odp_actions;
130 rule_is_hidden(const struct rule *rule)
132 /* Subrules are merely an implementation detail, so hide them from the
134 if (rule->super != NULL) {
138 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
139 * (e.g. by in-band control) and are intentionally hidden from the
141 if (rule->cr.priority > UINT16_MAX) {
148 static struct rule *rule_create(struct ofproto *, struct rule *super,
149 const union ofp_action *, size_t n_actions,
150 uint16_t idle_timeout, uint16_t hard_timeout,
151 bool send_flow_removed);
152 static void rule_free(struct rule *);
153 static void rule_destroy(struct ofproto *, struct rule *);
154 static struct rule *rule_from_cls_rule(const struct cls_rule *);
155 static void rule_insert(struct ofproto *, struct rule *,
156 struct ofpbuf *packet, uint16_t in_port);
157 static void rule_remove(struct ofproto *, struct rule *);
158 static bool rule_make_actions(struct ofproto *, struct rule *,
159 const struct ofpbuf *packet);
160 static void rule_install(struct ofproto *, struct rule *,
161 struct rule *displaced_rule);
162 static void rule_uninstall(struct ofproto *, struct rule *);
163 static void rule_post_uninstall(struct ofproto *, struct rule *);
164 static void send_flow_removed(struct ofproto *p, struct rule *rule,
165 long long int now, uint8_t reason);
170 struct pktbuf *pktbuf;
173 struct rconn_packet_counter *packet_in_counter;
175 /* Number of OpenFlow messages queued as replies to OpenFlow requests, and
176 * the maximum number before we stop reading OpenFlow requests. */
177 #define OFCONN_REPLY_MAX 100
178 struct rconn_packet_counter *reply_counter;
181 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *);
182 static void ofconn_destroy(struct ofconn *, struct ofproto *);
183 static void ofconn_run(struct ofconn *, struct ofproto *);
184 static void ofconn_wait(struct ofconn *);
185 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
186 struct rconn_packet_counter *counter);
190 uint64_t datapath_id; /* Datapath ID. */
191 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
192 uint64_t mgmt_id; /* Management channel identifier. */
193 char *manufacturer; /* Manufacturer. */
194 char *hardware; /* Hardware. */
195 char *software; /* Software version. */
196 char *serial; /* Serial number. */
200 struct netdev_monitor *netdev_monitor;
201 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
203 struct shash port_by_name;
207 struct switch_status *switch_status;
208 struct status_category *ss_cat;
209 struct in_band *in_band;
210 struct discovery *discovery;
211 struct fail_open *fail_open;
212 struct pinsched *miss_sched, *action_sched;
213 struct executer *executer;
214 struct netflow *netflow;
217 struct classifier cls;
218 bool need_revalidate;
219 long long int next_expiration;
220 struct tag_set revalidate_set;
222 /* OpenFlow connections. */
223 struct list all_conns;
224 struct ofconn *controller;
225 struct pvconn **listeners;
227 struct pvconn **snoops;
230 /* Hooks for ovs-vswitchd. */
231 const struct ofhooks *ofhooks;
234 /* Used by default ofhooks. */
235 struct mac_learning *ml;
238 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
240 static const struct ofhooks default_ofhooks;
242 static uint64_t pick_datapath_id(const struct ofproto *);
243 static uint64_t pick_fallback_dpid(void);
244 static void send_packet_in_miss(struct ofpbuf *, void *ofproto);
245 static void send_packet_in_action(struct ofpbuf *, void *ofproto);
246 static void update_used(struct ofproto *);
247 static void update_stats(struct ofproto *, struct rule *,
248 const struct odp_flow_stats *);
249 static void expire_rule(struct cls_rule *, void *ofproto);
250 static void active_timeout(struct ofproto *ofproto, struct rule *rule);
251 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
252 static void revalidate_cb(struct cls_rule *rule_, void *p_);
254 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
256 static void handle_openflow(struct ofconn *, struct ofproto *,
259 static void refresh_port_group(struct ofproto *, unsigned int group);
260 static void update_port(struct ofproto *, const char *devname);
261 static int init_ports(struct ofproto *);
262 static void reinit_ports(struct ofproto *);
265 ofproto_create(const char *datapath, 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, &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);
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 = xcalloc(1, 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");
307 /* Initialize datapath. */
309 p->netdev_monitor = netdev_monitor_create();
310 port_array_init(&p->ports);
311 shash_init(&p->port_by_name);
312 p->max_ports = stats.max_ports;
314 /* Initialize submodules. */
315 p->switch_status = switch_status_create(p);
319 p->miss_sched = p->action_sched = NULL;
323 /* Initialize flow table. */
324 classifier_init(&p->cls);
325 p->need_revalidate = false;
326 p->next_expiration = time_msec() + 1000;
327 tag_set_init(&p->revalidate_set);
329 /* Initialize OpenFlow connections. */
330 list_init(&p->all_conns);
331 p->controller = ofconn_create(p, rconn_create(5, 8));
332 p->controller->pktbuf = pktbuf_create();
333 p->controller->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
339 /* Initialize hooks. */
341 p->ofhooks = ofhooks;
345 p->ofhooks = &default_ofhooks;
347 p->ml = mac_learning_create();
350 /* Register switch status category. */
351 p->ss_cat = switch_status_register(p->switch_status, "remote",
352 rconn_status_cb, p->controller->rconn);
355 error = init_ports(p);
361 /* Pick final datapath ID. */
362 p->datapath_id = pick_datapath_id(p);
363 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
370 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
372 uint64_t old_dpid = p->datapath_id;
373 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
374 if (p->datapath_id != old_dpid) {
375 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
376 rconn_reconnect(p->controller->rconn);
381 ofproto_set_mgmt_id(struct ofproto *p, uint64_t mgmt_id)
383 p->mgmt_id = mgmt_id;
387 ofproto_set_probe_interval(struct ofproto *p, int probe_interval)
389 probe_interval = probe_interval ? MAX(probe_interval, 5) : 0;
390 rconn_set_probe_interval(p->controller->rconn, probe_interval);
392 int trigger_duration = probe_interval ? probe_interval * 3 : 15;
393 fail_open_set_trigger_duration(p->fail_open, trigger_duration);
398 ofproto_set_max_backoff(struct ofproto *p, int max_backoff)
400 rconn_set_max_backoff(p->controller->rconn, max_backoff);
404 ofproto_set_desc(struct ofproto *p,
405 const char *manufacturer, const char *hardware,
406 const char *software, const char *serial)
409 free(p->manufacturer);
410 p->manufacturer = xstrdup(manufacturer);
414 p->hardware = xstrdup(hardware);
418 p->software = xstrdup(software);
422 p->serial = xstrdup(serial);
427 ofproto_set_in_band(struct ofproto *p, bool in_band)
429 if (in_band != (p->in_band != NULL)) {
431 return in_band_create(p, p->dpif, p->switch_status,
432 p->controller->rconn, &p->in_band);
434 ofproto_set_discovery(p, false, NULL, true);
435 in_band_destroy(p->in_band);
438 rconn_reconnect(p->controller->rconn);
444 ofproto_set_discovery(struct ofproto *p, bool discovery,
445 const char *re, bool update_resolv_conf)
447 if (discovery != (p->discovery != NULL)) {
449 int error = ofproto_set_in_band(p, true);
453 error = discovery_create(re, update_resolv_conf,
454 p->dpif, p->switch_status,
460 discovery_destroy(p->discovery);
463 rconn_disconnect(p->controller->rconn);
464 } else if (discovery) {
465 discovery_set_update_resolv_conf(p->discovery, update_resolv_conf);
466 return discovery_set_accept_controller_re(p->discovery, re);
472 ofproto_set_controller(struct ofproto *ofproto, const char *controller)
474 if (ofproto->discovery) {
476 } else if (controller) {
477 if (strcmp(rconn_get_name(ofproto->controller->rconn), controller)) {
478 return rconn_connect(ofproto->controller->rconn, controller);
483 rconn_disconnect(ofproto->controller->rconn);
489 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
490 const struct svec *svec)
492 struct pvconn **pvconns = *pvconnsp;
493 size_t n_pvconns = *n_pvconnsp;
497 for (i = 0; i < n_pvconns; i++) {
498 pvconn_close(pvconns[i]);
502 pvconns = xmalloc(svec->n * sizeof *pvconns);
504 for (i = 0; i < svec->n; i++) {
505 const char *name = svec->names[i];
506 struct pvconn *pvconn;
509 error = pvconn_open(name, &pvconn);
511 pvconns[n_pvconns++] = pvconn;
513 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
521 *n_pvconnsp = n_pvconns;
527 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
529 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
533 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
535 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
539 ofproto_set_netflow(struct ofproto *ofproto,
540 const struct netflow_options *nf_options)
542 if (nf_options->collectors.n) {
543 if (!ofproto->netflow) {
544 ofproto->netflow = netflow_create();
546 return netflow_set_options(ofproto->netflow, nf_options);
548 netflow_destroy(ofproto->netflow);
549 ofproto->netflow = NULL;
555 ofproto_set_failure(struct ofproto *ofproto, bool fail_open)
558 struct rconn *rconn = ofproto->controller->rconn;
559 int trigger_duration = rconn_get_probe_interval(rconn) * 3;
560 if (!ofproto->fail_open) {
561 ofproto->fail_open = fail_open_create(ofproto, trigger_duration,
562 ofproto->switch_status,
565 fail_open_set_trigger_duration(ofproto->fail_open,
569 fail_open_destroy(ofproto->fail_open);
570 ofproto->fail_open = NULL;
575 ofproto_set_rate_limit(struct ofproto *ofproto,
576 int rate_limit, int burst_limit)
578 if (rate_limit > 0) {
579 if (!ofproto->miss_sched) {
580 ofproto->miss_sched = pinsched_create(rate_limit, burst_limit,
581 ofproto->switch_status);
582 ofproto->action_sched = pinsched_create(rate_limit, burst_limit,
585 pinsched_set_limits(ofproto->miss_sched, rate_limit, burst_limit);
586 pinsched_set_limits(ofproto->action_sched,
587 rate_limit, burst_limit);
590 pinsched_destroy(ofproto->miss_sched);
591 ofproto->miss_sched = NULL;
592 pinsched_destroy(ofproto->action_sched);
593 ofproto->action_sched = NULL;
598 ofproto_set_stp(struct ofproto *ofproto UNUSED, bool enable_stp)
602 VLOG_WARN("STP is not yet implemented");
610 ofproto_set_remote_execution(struct ofproto *ofproto, const char *command_acl,
611 const char *command_dir)
614 if (!ofproto->executer) {
615 return executer_create(command_acl, command_dir,
618 executer_set_acl(ofproto->executer, command_acl, command_dir);
621 executer_destroy(ofproto->executer);
622 ofproto->executer = NULL;
628 ofproto_get_datapath_id(const struct ofproto *ofproto)
630 return ofproto->datapath_id;
634 ofproto_get_mgmt_id(const struct ofproto *ofproto)
636 return ofproto->mgmt_id;
640 ofproto_get_probe_interval(const struct ofproto *ofproto)
642 return rconn_get_probe_interval(ofproto->controller->rconn);
646 ofproto_get_max_backoff(const struct ofproto *ofproto)
648 return rconn_get_max_backoff(ofproto->controller->rconn);
652 ofproto_get_in_band(const struct ofproto *ofproto)
654 return ofproto->in_band != NULL;
658 ofproto_get_discovery(const struct ofproto *ofproto)
660 return ofproto->discovery != NULL;
664 ofproto_get_controller(const struct ofproto *ofproto)
666 return rconn_get_name(ofproto->controller->rconn);
670 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
674 for (i = 0; i < ofproto->n_listeners; i++) {
675 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
680 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
684 for (i = 0; i < ofproto->n_snoops; i++) {
685 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
690 ofproto_destroy(struct ofproto *p)
692 struct ofconn *ofconn, *next_ofconn;
693 struct ofport *ofport;
694 unsigned int port_no;
701 ofproto_flush_flows(p);
702 classifier_destroy(&p->cls);
704 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
706 ofconn_destroy(ofconn, p);
710 netdev_monitor_destroy(p->netdev_monitor);
711 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
714 shash_destroy(&p->port_by_name);
716 switch_status_destroy(p->switch_status);
717 in_band_destroy(p->in_band);
718 discovery_destroy(p->discovery);
719 fail_open_destroy(p->fail_open);
720 pinsched_destroy(p->miss_sched);
721 pinsched_destroy(p->action_sched);
722 executer_destroy(p->executer);
723 netflow_destroy(p->netflow);
725 switch_status_unregister(p->ss_cat);
727 for (i = 0; i < p->n_listeners; i++) {
728 pvconn_close(p->listeners[i]);
732 for (i = 0; i < p->n_snoops; i++) {
733 pvconn_close(p->snoops[i]);
737 mac_learning_destroy(p->ml);
743 ofproto_run(struct ofproto *p)
745 int error = ofproto_run1(p);
747 error = ofproto_run2(p, false);
753 process_port_change(struct ofproto *ofproto, int error, char *devname)
755 if (error == ENOBUFS) {
756 reinit_ports(ofproto);
758 update_port(ofproto, devname);
764 ofproto_run1(struct ofproto *p)
766 struct ofconn *ofconn, *next_ofconn;
771 for (i = 0; i < 50; i++) {
775 error = dpif_recv(p->dpif, &buf);
777 if (error == ENODEV) {
778 /* Someone destroyed the datapath behind our back. The caller
779 * better destroy us and give up, because we're just going to
780 * spin from here on out. */
781 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
782 VLOG_ERR_RL(&rl, "%s: datapath was destroyed externally",
789 handle_odp_msg(p, buf);
792 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
793 process_port_change(p, error, devname);
795 while ((error = netdev_monitor_poll(p->netdev_monitor,
796 &devname)) != EAGAIN) {
797 process_port_change(p, error, devname);
801 in_band_run(p->in_band);
804 char *controller_name;
805 if (rconn_is_connectivity_questionable(p->controller->rconn)) {
806 discovery_question_connectivity(p->discovery);
808 if (discovery_run(p->discovery, &controller_name)) {
809 if (controller_name) {
810 rconn_connect(p->controller->rconn, controller_name);
812 rconn_disconnect(p->controller->rconn);
816 pinsched_run(p->miss_sched, send_packet_in_miss, p);
817 pinsched_run(p->action_sched, send_packet_in_action, p);
819 executer_run(p->executer);
822 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
824 ofconn_run(ofconn, p);
827 /* Fail-open maintenance. Do this after processing the ofconns since
828 * fail-open checks the status of the controller rconn. */
830 fail_open_run(p->fail_open);
833 for (i = 0; i < p->n_listeners; i++) {
837 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
839 ofconn_create(p, rconn_new_from_vconn("passive", vconn));
840 } else if (retval != EAGAIN) {
841 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
845 for (i = 0; i < p->n_snoops; i++) {
849 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
851 rconn_add_monitor(p->controller->rconn, vconn);
852 } else if (retval != EAGAIN) {
853 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
857 if (time_msec() >= p->next_expiration) {
858 COVERAGE_INC(ofproto_expiration);
859 p->next_expiration = time_msec() + 1000;
862 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
864 /* Let the hook know that we're at a stable point: all outstanding data
865 * in existing flows has been accounted to the account_cb. Thus, the
866 * hook can now reasonably do operations that depend on having accurate
867 * flow volume accounting (currently, that's just bond rebalancing). */
868 if (p->ofhooks->account_checkpoint_cb) {
869 p->ofhooks->account_checkpoint_cb(p->aux);
874 netflow_run(p->netflow);
880 struct revalidate_cbdata {
881 struct ofproto *ofproto;
882 bool revalidate_all; /* Revalidate all exact-match rules? */
883 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
884 struct tag_set revalidate_set; /* Set of tags to revalidate. */
888 ofproto_run2(struct ofproto *p, bool revalidate_all)
890 if (p->need_revalidate || revalidate_all
891 || !tag_set_is_empty(&p->revalidate_set)) {
892 struct revalidate_cbdata cbdata;
894 cbdata.revalidate_all = revalidate_all;
895 cbdata.revalidate_subrules = p->need_revalidate;
896 cbdata.revalidate_set = p->revalidate_set;
897 tag_set_init(&p->revalidate_set);
898 COVERAGE_INC(ofproto_revalidate);
899 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
900 p->need_revalidate = false;
907 ofproto_wait(struct ofproto *p)
909 struct ofconn *ofconn;
912 dpif_recv_wait(p->dpif);
913 dpif_port_poll_wait(p->dpif);
914 netdev_monitor_poll_wait(p->netdev_monitor);
915 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
919 in_band_wait(p->in_band);
922 discovery_wait(p->discovery);
925 fail_open_wait(p->fail_open);
927 pinsched_wait(p->miss_sched);
928 pinsched_wait(p->action_sched);
930 executer_wait(p->executer);
932 if (!tag_set_is_empty(&p->revalidate_set)) {
933 poll_immediate_wake();
935 if (p->need_revalidate) {
936 /* Shouldn't happen, but if it does just go around again. */
937 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
938 poll_immediate_wake();
939 } else if (p->next_expiration != LLONG_MAX) {
940 poll_timer_wait(p->next_expiration - time_msec());
942 for (i = 0; i < p->n_listeners; i++) {
943 pvconn_wait(p->listeners[i]);
945 for (i = 0; i < p->n_snoops; i++) {
946 pvconn_wait(p->snoops[i]);
951 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
953 tag_set_add(&ofproto->revalidate_set, tag);
957 ofproto_get_revalidate_set(struct ofproto *ofproto)
959 return &ofproto->revalidate_set;
963 ofproto_is_alive(const struct ofproto *p)
965 return p->discovery || rconn_is_alive(p->controller->rconn);
969 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
970 const union ofp_action *actions, size_t n_actions,
971 const struct ofpbuf *packet)
973 struct odp_actions odp_actions;
976 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
982 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
984 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
985 odp_actions.n_actions, packet);
990 ofproto_add_flow(struct ofproto *p,
991 const flow_t *flow, uint32_t wildcards, unsigned int priority,
992 const union ofp_action *actions, size_t n_actions,
996 rule = rule_create(p, NULL, actions, n_actions,
997 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
999 cls_rule_from_flow(&rule->cr, flow, wildcards, priority);
1000 rule_insert(p, rule, NULL, 0);
1004 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1005 uint32_t wildcards, unsigned int priority)
1009 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1013 rule_remove(ofproto, rule);
1018 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1020 struct rule *rule = rule_from_cls_rule(rule_);
1021 struct ofproto *ofproto = ofproto_;
1023 /* Mark the flow as not installed, even though it might really be
1024 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1025 * There is no point in uninstalling it individually since we are about to
1026 * blow away all the flows with dpif_flow_flush(). */
1027 rule->installed = false;
1029 rule_remove(ofproto, rule);
1033 ofproto_flush_flows(struct ofproto *ofproto)
1035 COVERAGE_INC(ofproto_flush);
1036 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1037 dpif_flow_flush(ofproto->dpif);
1038 if (ofproto->in_band) {
1039 in_band_flushed(ofproto->in_band);
1041 if (ofproto->fail_open) {
1042 fail_open_flushed(ofproto->fail_open);
1047 reinit_ports(struct ofproto *p)
1049 struct svec devnames;
1050 struct ofport *ofport;
1051 unsigned int port_no;
1052 struct odp_port *odp_ports;
1056 svec_init(&devnames);
1057 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1058 svec_add (&devnames, (char *) ofport->opp.name);
1060 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1061 for (i = 0; i < n_odp_ports; i++) {
1062 svec_add (&devnames, odp_ports[i].devname);
1066 svec_sort_unique(&devnames);
1067 for (i = 0; i < devnames.n; i++) {
1068 update_port(p, devnames.names[i]);
1070 svec_destroy(&devnames);
1074 refresh_port_group(struct ofproto *p, unsigned int group)
1078 struct ofport *port;
1079 unsigned int port_no;
1081 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1083 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1085 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1086 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1087 ports[n_ports++] = port_no;
1090 dpif_port_group_set(p->dpif, group, ports, n_ports);
1095 refresh_port_groups(struct ofproto *p)
1097 refresh_port_group(p, DP_GROUP_FLOOD);
1098 refresh_port_group(p, DP_GROUP_ALL);
1101 static struct ofport *
1102 make_ofport(const struct odp_port *odp_port)
1104 enum netdev_flags flags;
1105 struct ofport *ofport;
1106 struct netdev *netdev;
1110 error = netdev_open(odp_port->devname, NETDEV_ETH_TYPE_NONE, &netdev);
1112 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1113 "cannot be opened (%s)",
1114 odp_port->devname, odp_port->port,
1115 odp_port->devname, strerror(error));
1119 ofport = xmalloc(sizeof *ofport);
1120 ofport->netdev = netdev;
1121 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1122 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1123 memcpy(ofport->opp.name, odp_port->devname,
1124 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1125 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1127 netdev_get_flags(netdev, &flags);
1128 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1130 netdev_get_carrier(netdev, &carrier);
1131 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1133 netdev_get_features(netdev,
1134 &ofport->opp.curr, &ofport->opp.advertised,
1135 &ofport->opp.supported, &ofport->opp.peer);
1140 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1142 if (port_array_get(&p->ports, odp_port->port)) {
1143 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1146 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1147 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1156 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1158 const struct ofp_phy_port *a = &a_->opp;
1159 const struct ofp_phy_port *b = &b_->opp;
1161 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1162 return (a->port_no == b->port_no
1163 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1164 && !strcmp((char *) a->name, (char *) b->name)
1165 && a->state == b->state
1166 && a->config == b->config
1167 && a->curr == b->curr
1168 && a->advertised == b->advertised
1169 && a->supported == b->supported
1170 && a->peer == b->peer);
1174 send_port_status(struct ofproto *p, const struct ofport *ofport,
1177 /* XXX Should limit the number of queued port status change messages. */
1178 struct ofconn *ofconn;
1179 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1180 struct ofp_port_status *ops;
1183 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1184 ops->reason = reason;
1185 ops->desc = ofport->opp;
1186 hton_ofp_phy_port(&ops->desc);
1187 queue_tx(b, ofconn, NULL);
1189 if (p->ofhooks->port_changed_cb) {
1190 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1195 ofport_install(struct ofproto *p, struct ofport *ofport)
1197 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1198 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no),
1200 shash_add(&p->port_by_name, (char *) ofport->opp.name, ofport);
1204 ofport_remove(struct ofproto *p, struct ofport *ofport)
1206 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1207 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no), NULL);
1208 shash_delete(&p->port_by_name,
1209 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1213 ofport_free(struct ofport *ofport)
1216 netdev_close(ofport->netdev);
1222 update_port(struct ofproto *p, const char *devname)
1224 struct odp_port odp_port;
1225 struct ofport *old_ofport;
1226 struct ofport *new_ofport;
1229 COVERAGE_INC(ofproto_update_port);
1231 /* Query the datapath for port information. */
1232 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1234 /* Find the old ofport. */
1235 old_ofport = shash_find_data(&p->port_by_name, devname);
1238 /* There's no port named 'devname' but there might be a port with
1239 * the same port number. This could happen if a port is deleted
1240 * and then a new one added in its place very quickly, or if a port
1241 * is renamed. In the former case we want to send an OFPPR_DELETE
1242 * and an OFPPR_ADD, and in the latter case we want to send a
1243 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1244 * the old port's ifindex against the new port, or perhaps less
1245 * reliably but more portably by comparing the old port's MAC
1246 * against the new port's MAC. However, this code isn't that smart
1247 * and always sends an OFPPR_MODIFY (XXX). */
1248 old_ofport = port_array_get(&p->ports, odp_port.port);
1250 } else if (error != ENOENT && error != ENODEV) {
1251 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1252 "%s", strerror(error));
1256 /* Create a new ofport. */
1257 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1259 /* Eliminate a few pathological cases. */
1260 if (!old_ofport && !new_ofport) {
1262 } else if (old_ofport && new_ofport) {
1263 /* Most of the 'config' bits are OpenFlow soft state, but
1264 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1265 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1266 * leaves the other bits 0.) */
1267 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1269 if (ofport_equal(old_ofport, new_ofport)) {
1270 /* False alarm--no change. */
1271 ofport_free(new_ofport);
1276 /* Now deal with the normal cases. */
1278 ofport_remove(p, old_ofport);
1281 ofport_install(p, new_ofport);
1283 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1284 (!old_ofport ? OFPPR_ADD
1285 : !new_ofport ? OFPPR_DELETE
1287 ofport_free(old_ofport);
1289 /* Update port groups. */
1290 refresh_port_groups(p);
1294 init_ports(struct ofproto *p)
1296 struct odp_port *ports;
1301 error = dpif_port_list(p->dpif, &ports, &n_ports);
1306 for (i = 0; i < n_ports; i++) {
1307 const struct odp_port *odp_port = &ports[i];
1308 if (!ofport_conflicts(p, odp_port)) {
1309 struct ofport *ofport = make_ofport(odp_port);
1311 ofport_install(p, ofport);
1316 refresh_port_groups(p);
1320 static struct ofconn *
1321 ofconn_create(struct ofproto *p, struct rconn *rconn)
1323 struct ofconn *ofconn = xmalloc(sizeof *ofconn);
1324 list_push_back(&p->all_conns, &ofconn->node);
1325 ofconn->rconn = rconn;
1326 ofconn->pktbuf = NULL;
1327 ofconn->miss_send_len = 0;
1328 ofconn->packet_in_counter = rconn_packet_counter_create ();
1329 ofconn->reply_counter = rconn_packet_counter_create ();
1334 ofconn_destroy(struct ofconn *ofconn, struct ofproto *p)
1337 executer_rconn_closing(p->executer, ofconn->rconn);
1340 list_remove(&ofconn->node);
1341 rconn_destroy(ofconn->rconn);
1342 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1343 rconn_packet_counter_destroy(ofconn->reply_counter);
1344 pktbuf_destroy(ofconn->pktbuf);
1349 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1353 rconn_run(ofconn->rconn);
1355 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1356 /* Limit the number of iterations to prevent other tasks from
1358 for (iteration = 0; iteration < 50; iteration++) {
1359 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1364 fail_open_maybe_recover(p->fail_open);
1366 handle_openflow(ofconn, p, of_msg);
1367 ofpbuf_delete(of_msg);
1371 if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
1372 ofconn_destroy(ofconn, p);
1377 ofconn_wait(struct ofconn *ofconn)
1379 rconn_run_wait(ofconn->rconn);
1380 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1381 rconn_recv_wait(ofconn->rconn);
1383 COVERAGE_INC(ofproto_ofconn_stuck);
1387 /* Caller is responsible for initializing the 'cr' member of the returned
1389 static struct rule *
1390 rule_create(struct ofproto *ofproto, struct rule *super,
1391 const union ofp_action *actions, size_t n_actions,
1392 uint16_t idle_timeout, uint16_t hard_timeout,
1393 bool send_flow_removed)
1395 struct rule *rule = xcalloc(1, sizeof *rule);
1396 rule->idle_timeout = idle_timeout;
1397 rule->hard_timeout = hard_timeout;
1398 rule->used = rule->created = time_msec();
1399 rule->send_flow_removed = send_flow_removed;
1400 rule->super = super;
1402 list_push_back(&super->list, &rule->list);
1404 list_init(&rule->list);
1406 rule->n_actions = n_actions;
1407 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1408 netflow_flow_clear(&rule->nf_flow);
1409 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1414 static struct rule *
1415 rule_from_cls_rule(const struct cls_rule *cls_rule)
1417 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1421 rule_free(struct rule *rule)
1423 free(rule->actions);
1424 free(rule->odp_actions);
1428 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1429 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1430 * through all of its subrules and revalidates them, destroying any that no
1431 * longer has a super-rule (which is probably all of them).
1433 * Before calling this function, the caller must make have removed 'rule' from
1434 * the classifier. If 'rule' is an exact-match rule, the caller is also
1435 * responsible for ensuring that it has been uninstalled from the datapath. */
1437 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1440 struct rule *subrule, *next;
1441 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1442 revalidate_rule(ofproto, subrule);
1445 list_remove(&rule->list);
1451 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1453 const union ofp_action *oa;
1454 struct actions_iterator i;
1456 if (out_port == htons(OFPP_NONE)) {
1459 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1460 oa = actions_next(&i)) {
1461 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1468 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1469 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1471 * The flow that 'packet' actually contains does not need to actually match
1472 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1473 * the packet and byte counters for 'rule' will be credited for the packet sent
1474 * out whether or not the packet actually matches 'rule'.
1476 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1477 * the caller must already have accurately composed ODP actions for it given
1478 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1479 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1480 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1481 * actions and apply them to 'packet'. */
1483 rule_execute(struct ofproto *ofproto, struct rule *rule,
1484 struct ofpbuf *packet, const flow_t *flow)
1486 const union odp_action *actions;
1488 struct odp_actions a;
1490 /* Grab or compose the ODP actions.
1492 * The special case for an exact-match 'rule' where 'flow' is not the
1493 * rule's flow is important to avoid, e.g., sending a packet out its input
1494 * port simply because the ODP actions were composed for the wrong
1496 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1497 struct rule *super = rule->super ? rule->super : rule;
1498 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1499 packet, &a, NULL, 0, NULL)) {
1502 actions = a.actions;
1503 n_actions = a.n_actions;
1505 actions = rule->odp_actions;
1506 n_actions = rule->n_odp_actions;
1509 /* Execute the ODP actions. */
1510 if (!dpif_execute(ofproto->dpif, flow->in_port,
1511 actions, n_actions, packet)) {
1512 struct odp_flow_stats stats;
1513 flow_extract_stats(flow, packet, &stats);
1514 update_stats(ofproto, rule, &stats);
1515 rule->used = time_msec();
1516 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1521 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1524 struct rule *displaced_rule;
1526 /* Insert the rule in the classifier. */
1527 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1528 if (!rule->cr.wc.wildcards) {
1529 rule_make_actions(p, rule, packet);
1532 /* Send the packet and credit it to the rule. */
1535 flow_extract(packet, in_port, &flow);
1536 rule_execute(p, rule, packet, &flow);
1539 /* Install the rule in the datapath only after sending the packet, to
1540 * avoid packet reordering. */
1541 if (rule->cr.wc.wildcards) {
1542 COVERAGE_INC(ofproto_add_wc_flow);
1543 p->need_revalidate = true;
1545 rule_install(p, rule, displaced_rule);
1548 /* Free the rule that was displaced, if any. */
1549 if (displaced_rule) {
1550 rule_destroy(p, displaced_rule);
1554 static struct rule *
1555 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1558 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1559 rule->idle_timeout, rule->hard_timeout,
1561 COVERAGE_INC(ofproto_subrule_create);
1562 cls_rule_from_flow(&subrule->cr, flow, 0,
1563 (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1564 : rule->cr.priority));
1565 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1571 rule_remove(struct ofproto *ofproto, struct rule *rule)
1573 if (rule->cr.wc.wildcards) {
1574 COVERAGE_INC(ofproto_del_wc_flow);
1575 ofproto->need_revalidate = true;
1577 rule_uninstall(ofproto, rule);
1579 classifier_remove(&ofproto->cls, &rule->cr);
1580 rule_destroy(ofproto, rule);
1583 /* Returns true if the actions changed, false otherwise. */
1585 rule_make_actions(struct ofproto *p, struct rule *rule,
1586 const struct ofpbuf *packet)
1588 const struct rule *super;
1589 struct odp_actions a;
1592 assert(!rule->cr.wc.wildcards);
1594 super = rule->super ? rule->super : rule;
1596 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1597 packet, &a, &rule->tags, &rule->may_install,
1598 &rule->nf_flow.output_iface);
1600 actions_len = a.n_actions * sizeof *a.actions;
1601 if (rule->n_odp_actions != a.n_actions
1602 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1603 COVERAGE_INC(ofproto_odp_unchanged);
1604 free(rule->odp_actions);
1605 rule->n_odp_actions = a.n_actions;
1606 rule->odp_actions = xmemdup(a.actions, actions_len);
1614 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1615 struct odp_flow_put *put)
1617 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1618 put->flow.key = rule->cr.flow;
1619 put->flow.actions = rule->odp_actions;
1620 put->flow.n_actions = rule->n_odp_actions;
1622 return dpif_flow_put(ofproto->dpif, put);
1626 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1628 assert(!rule->cr.wc.wildcards);
1630 if (rule->may_install) {
1631 struct odp_flow_put put;
1632 if (!do_put_flow(p, rule,
1633 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1635 rule->installed = true;
1636 if (displaced_rule) {
1637 update_stats(p, displaced_rule, &put.flow.stats);
1638 rule_post_uninstall(p, displaced_rule);
1641 } else if (displaced_rule) {
1642 rule_uninstall(p, displaced_rule);
1647 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1649 if (rule->installed) {
1650 struct odp_flow_put put;
1651 COVERAGE_INC(ofproto_dp_missed);
1652 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1654 rule_install(ofproto, rule, NULL);
1659 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1661 bool actions_changed;
1662 uint16_t new_out_iface, old_out_iface;
1664 old_out_iface = rule->nf_flow.output_iface;
1665 actions_changed = rule_make_actions(ofproto, rule, NULL);
1667 if (rule->may_install) {
1668 if (rule->installed) {
1669 if (actions_changed) {
1670 struct odp_flow_put put;
1671 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
1672 | ODPPF_ZERO_STATS, &put);
1673 update_stats(ofproto, rule, &put.flow.stats);
1675 /* Temporarily set the old output iface so that NetFlow
1676 * messages have the correct output interface for the old
1678 new_out_iface = rule->nf_flow.output_iface;
1679 rule->nf_flow.output_iface = old_out_iface;
1680 rule_post_uninstall(ofproto, rule);
1681 rule->nf_flow.output_iface = new_out_iface;
1684 rule_install(ofproto, rule, NULL);
1687 rule_uninstall(ofproto, rule);
1692 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1694 uint64_t total_bytes = rule->byte_count + extra_bytes;
1696 if (ofproto->ofhooks->account_flow_cb
1697 && total_bytes > rule->accounted_bytes)
1699 ofproto->ofhooks->account_flow_cb(
1700 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1701 total_bytes - rule->accounted_bytes, ofproto->aux);
1702 rule->accounted_bytes = total_bytes;
1707 rule_uninstall(struct ofproto *p, struct rule *rule)
1709 assert(!rule->cr.wc.wildcards);
1710 if (rule->installed) {
1711 struct odp_flow odp_flow;
1713 odp_flow.key = rule->cr.flow;
1714 odp_flow.actions = NULL;
1715 odp_flow.n_actions = 0;
1716 if (!dpif_flow_del(p->dpif, &odp_flow)) {
1717 update_stats(p, rule, &odp_flow.stats);
1719 rule->installed = false;
1721 rule_post_uninstall(p, rule);
1726 is_controller_rule(struct rule *rule)
1728 /* If the only action is send to the controller then don't report
1729 * NetFlow expiration messages since it is just part of the control
1730 * logic for the network and not real traffic. */
1732 if (rule && rule->super) {
1733 struct rule *super = rule->super;
1735 return super->n_actions == 1 &&
1736 super->actions[0].type == htons(OFPAT_OUTPUT) &&
1737 super->actions[0].output.port == htons(OFPP_CONTROLLER);
1744 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1746 struct rule *super = rule->super;
1748 rule_account(ofproto, rule, 0);
1750 if (ofproto->netflow && !is_controller_rule(rule)) {
1751 struct ofexpired expired;
1752 expired.flow = rule->cr.flow;
1753 expired.packet_count = rule->packet_count;
1754 expired.byte_count = rule->byte_count;
1755 expired.used = rule->used;
1756 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
1759 super->packet_count += rule->packet_count;
1760 super->byte_count += rule->byte_count;
1762 /* Reset counters to prevent double counting if the rule ever gets
1764 rule->packet_count = 0;
1765 rule->byte_count = 0;
1766 rule->accounted_bytes = 0;
1768 netflow_flow_clear(&rule->nf_flow);
1773 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1774 struct rconn_packet_counter *counter)
1776 update_openflow_length(msg);
1777 if (rconn_send(ofconn->rconn, msg, counter)) {
1783 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1784 int error, const void *data, size_t len)
1787 struct ofp_error_msg *oem;
1789 if (!(error >> 16)) {
1790 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1795 COVERAGE_INC(ofproto_error);
1796 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1797 oh ? oh->xid : 0, &buf);
1798 oem->type = htons((unsigned int) error >> 16);
1799 oem->code = htons(error & 0xffff);
1800 memcpy(oem->data, data, len);
1801 queue_tx(buf, ofconn, ofconn->reply_counter);
1805 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1808 size_t oh_length = ntohs(oh->length);
1809 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1813 hton_ofp_phy_port(struct ofp_phy_port *opp)
1815 opp->port_no = htons(opp->port_no);
1816 opp->config = htonl(opp->config);
1817 opp->state = htonl(opp->state);
1818 opp->curr = htonl(opp->curr);
1819 opp->advertised = htonl(opp->advertised);
1820 opp->supported = htonl(opp->supported);
1821 opp->peer = htonl(opp->peer);
1825 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
1827 struct ofp_header *rq = oh;
1828 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
1833 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
1834 struct ofp_header *oh)
1836 struct ofp_switch_features *osf;
1838 unsigned int port_no;
1839 struct ofport *port;
1841 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1842 osf->datapath_id = htonll(p->datapath_id);
1843 osf->n_buffers = htonl(pktbuf_capacity());
1845 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1846 OFPC_PORT_STATS | OFPC_MULTI_PHY_TX);
1847 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1848 (1u << OFPAT_SET_VLAN_VID) |
1849 (1u << OFPAT_SET_VLAN_PCP) |
1850 (1u << OFPAT_STRIP_VLAN) |
1851 (1u << OFPAT_SET_DL_SRC) |
1852 (1u << OFPAT_SET_DL_DST) |
1853 (1u << OFPAT_SET_NW_SRC) |
1854 (1u << OFPAT_SET_NW_DST) |
1855 (1u << OFPAT_SET_NW_TOS) |
1856 (1u << OFPAT_SET_TP_SRC) |
1857 (1u << OFPAT_SET_TP_DST));
1859 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1860 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1863 queue_tx(buf, ofconn, ofconn->reply_counter);
1868 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1869 struct ofp_header *oh)
1872 struct ofp_switch_config *osc;
1876 /* Figure out flags. */
1877 dpif_get_drop_frags(p->dpif, &drop_frags);
1878 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1881 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1882 osc->flags = htons(flags);
1883 osc->miss_send_len = htons(ofconn->miss_send_len);
1884 queue_tx(buf, ofconn, ofconn->reply_counter);
1890 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1891 struct ofp_switch_config *osc)
1896 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1900 flags = ntohs(osc->flags);
1902 if (ofconn == p->controller) {
1903 switch (flags & OFPC_FRAG_MASK) {
1904 case OFPC_FRAG_NORMAL:
1905 dpif_set_drop_frags(p->dpif, false);
1907 case OFPC_FRAG_DROP:
1908 dpif_set_drop_frags(p->dpif, true);
1911 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1917 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1918 if (ntohs(osc->miss_send_len) != 0) {
1919 ofconn->pktbuf = pktbuf_create();
1921 pktbuf_destroy(ofconn->pktbuf);
1925 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1931 add_output_group_action(struct odp_actions *actions, uint16_t group,
1932 uint16_t *nf_output_iface)
1934 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1936 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
1937 *nf_output_iface = NF_OUT_FLOOD;
1942 add_controller_action(struct odp_actions *actions,
1943 const struct ofp_action_output *oao)
1945 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
1946 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
1949 struct action_xlate_ctx {
1951 const flow_t *flow; /* Flow to which these actions correspond. */
1952 int recurse; /* Recursion level, via xlate_table_action. */
1953 struct ofproto *ofproto;
1954 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
1955 * null pointer if we are revalidating
1956 * without a packet to refer to. */
1959 struct odp_actions *out; /* Datapath actions. */
1960 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
1961 bool may_set_up_flow; /* True ordinarily; false if the actions must
1962 * be reassessed for every packet. */
1963 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
1966 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
1967 struct action_xlate_ctx *ctx);
1970 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
1972 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
1975 if (ofport->opp.config & OFPPC_NO_FWD) {
1976 /* Forwarding disabled on port. */
1981 * We don't have an ofport record for this port, but it doesn't hurt to
1982 * allow forwarding to it anyhow. Maybe such a port will appear later
1983 * and we're pre-populating the flow table.
1987 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
1988 ctx->nf_output_iface = port;
1991 static struct rule *
1992 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
1995 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
1997 /* The rule we found might not be valid, since we could be in need of
1998 * revalidation. If it is not valid, don't return it. */
2001 && ofproto->need_revalidate
2002 && !revalidate_rule(ofproto, rule)) {
2003 COVERAGE_INC(ofproto_invalidated);
2011 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2013 if (!ctx->recurse) {
2018 flow.in_port = in_port;
2020 rule = lookup_valid_rule(ctx->ofproto, &flow);
2027 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2034 xlate_output_action(struct action_xlate_ctx *ctx,
2035 const struct ofp_action_output *oao)
2038 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2040 ctx->nf_output_iface = NF_OUT_DROP;
2042 switch (ntohs(oao->port)) {
2044 add_output_action(ctx, ctx->flow->in_port);
2047 xlate_table_action(ctx, ctx->flow->in_port);
2050 if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet,
2051 ctx->out, ctx->tags,
2052 &ctx->nf_output_iface,
2053 ctx->ofproto->aux)) {
2054 COVERAGE_INC(ofproto_uninstallable);
2055 ctx->may_set_up_flow = false;
2059 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2060 &ctx->nf_output_iface);
2063 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2065 case OFPP_CONTROLLER:
2066 add_controller_action(ctx->out, oao);
2069 add_output_action(ctx, ODPP_LOCAL);
2072 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2073 if (odp_port != ctx->flow->in_port) {
2074 add_output_action(ctx, odp_port);
2079 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2080 ctx->nf_output_iface = NF_OUT_FLOOD;
2081 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2082 ctx->nf_output_iface = prev_nf_output_iface;
2083 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2084 ctx->nf_output_iface != NF_OUT_FLOOD) {
2085 ctx->nf_output_iface = NF_OUT_MULTI;
2090 xlate_nicira_action(struct action_xlate_ctx *ctx,
2091 const struct nx_action_header *nah)
2093 const struct nx_action_resubmit *nar;
2094 int subtype = ntohs(nah->subtype);
2096 assert(nah->vendor == htonl(NX_VENDOR_ID));
2098 case NXAST_RESUBMIT:
2099 nar = (const struct nx_action_resubmit *) nah;
2100 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2104 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2110 do_xlate_actions(const union ofp_action *in, size_t n_in,
2111 struct action_xlate_ctx *ctx)
2113 struct actions_iterator iter;
2114 const union ofp_action *ia;
2115 const struct ofport *port;
2117 port = port_array_get(&ctx->ofproto->ports, ctx->flow->in_port);
2118 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2119 port->opp.config & (eth_addr_equals(ctx->flow->dl_dst, stp_eth_addr)
2120 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2121 /* Drop this flow. */
2125 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2126 uint16_t type = ntohs(ia->type);
2127 union odp_action *oa;
2131 xlate_output_action(ctx, &ia->output);
2134 case OFPAT_SET_VLAN_VID:
2135 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2136 oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2139 case OFPAT_SET_VLAN_PCP:
2140 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2141 oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2144 case OFPAT_STRIP_VLAN:
2145 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2148 case OFPAT_SET_DL_SRC:
2149 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2150 memcpy(oa->dl_addr.dl_addr,
2151 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2154 case OFPAT_SET_DL_DST:
2155 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2156 memcpy(oa->dl_addr.dl_addr,
2157 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2160 case OFPAT_SET_NW_SRC:
2161 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2162 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2165 case OFPAT_SET_NW_DST:
2166 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2167 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2169 case OFPAT_SET_NW_TOS:
2170 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2171 oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2174 case OFPAT_SET_TP_SRC:
2175 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2176 oa->tp_port.tp_port = ia->tp_port.tp_port;
2179 case OFPAT_SET_TP_DST:
2180 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2181 oa->tp_port.tp_port = ia->tp_port.tp_port;
2185 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2189 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2196 xlate_actions(const union ofp_action *in, size_t n_in,
2197 const flow_t *flow, struct ofproto *ofproto,
2198 const struct ofpbuf *packet,
2199 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2200 uint16_t *nf_output_iface)
2202 tag_type no_tags = 0;
2203 struct action_xlate_ctx ctx;
2204 COVERAGE_INC(ofproto_ofp2odp);
2205 odp_actions_init(out);
2208 ctx.ofproto = ofproto;
2209 ctx.packet = packet;
2211 ctx.tags = tags ? tags : &no_tags;
2212 ctx.may_set_up_flow = true;
2213 ctx.nf_output_iface = NF_OUT_DROP;
2214 do_xlate_actions(in, n_in, &ctx);
2216 /* Check with in-band control to see if we're allowed to set up this
2218 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2219 ctx.may_set_up_flow = false;
2222 if (may_set_up_flow) {
2223 *may_set_up_flow = ctx.may_set_up_flow;
2225 if (nf_output_iface) {
2226 *nf_output_iface = ctx.nf_output_iface;
2228 if (odp_actions_overflow(out)) {
2229 odp_actions_init(out);
2230 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2236 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2237 struct ofp_header *oh)
2239 struct ofp_packet_out *opo;
2240 struct ofpbuf payload, *buffer;
2241 struct odp_actions actions;
2247 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2251 opo = (struct ofp_packet_out *) oh;
2253 COVERAGE_INC(ofproto_packet_out);
2254 if (opo->buffer_id != htonl(UINT32_MAX)) {
2255 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2257 if (error || !buffer) {
2265 flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2266 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2267 &flow, p, &payload, &actions, NULL, NULL, NULL);
2272 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2274 ofpbuf_delete(buffer);
2280 update_port_config(struct ofproto *p, struct ofport *port,
2281 uint32_t config, uint32_t mask)
2283 mask &= config ^ port->opp.config;
2284 if (mask & OFPPC_PORT_DOWN) {
2285 if (config & OFPPC_PORT_DOWN) {
2286 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2288 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2291 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2292 if (mask & REVALIDATE_BITS) {
2293 COVERAGE_INC(ofproto_costly_flags);
2294 port->opp.config ^= mask & REVALIDATE_BITS;
2295 p->need_revalidate = true;
2297 #undef REVALIDATE_BITS
2298 if (mask & OFPPC_NO_FLOOD) {
2299 port->opp.config ^= OFPPC_NO_FLOOD;
2300 refresh_port_group(p, DP_GROUP_FLOOD);
2302 if (mask & OFPPC_NO_PACKET_IN) {
2303 port->opp.config ^= OFPPC_NO_PACKET_IN;
2308 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2310 const struct ofp_port_mod *opm;
2311 struct ofport *port;
2314 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2318 opm = (struct ofp_port_mod *) oh;
2320 port = port_array_get(&p->ports,
2321 ofp_port_to_odp_port(ntohs(opm->port_no)));
2323 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2324 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2325 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2327 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2328 if (opm->advertise) {
2329 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2335 static struct ofpbuf *
2336 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2338 struct ofp_stats_reply *osr;
2341 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2342 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2344 osr->flags = htons(0);
2348 static struct ofpbuf *
2349 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2351 return make_stats_reply(request->header.xid, request->type, body_len);
2355 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2357 struct ofpbuf *msg = *msgp;
2358 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2359 if (nbytes + msg->size > UINT16_MAX) {
2360 struct ofp_stats_reply *reply = msg->data;
2361 reply->flags = htons(OFPSF_REPLY_MORE);
2362 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2363 queue_tx(msg, ofconn, ofconn->reply_counter);
2365 return ofpbuf_put_uninit(*msgp, nbytes);
2369 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2370 struct ofp_stats_request *request)
2372 struct ofp_desc_stats *ods;
2375 msg = start_stats_reply(request, sizeof *ods);
2376 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2377 strncpy(ods->mfr_desc, p->manufacturer, sizeof ods->mfr_desc);
2378 strncpy(ods->hw_desc, p->hardware, sizeof ods->hw_desc);
2379 strncpy(ods->sw_desc, p->software, sizeof ods->sw_desc);
2380 strncpy(ods->serial_num, p->serial, sizeof ods->serial_num);
2381 queue_tx(msg, ofconn, ofconn->reply_counter);
2387 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2389 struct rule *rule = rule_from_cls_rule(cls_rule);
2390 int *n_subrules = n_subrules_;
2398 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2399 struct ofp_stats_request *request)
2401 struct ofp_table_stats *ots;
2403 struct odp_stats dpstats;
2404 int n_exact, n_subrules, n_wild;
2406 msg = start_stats_reply(request, sizeof *ots * 2);
2408 /* Count rules of various kinds. */
2410 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2411 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2412 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2415 dpif_get_dp_stats(p->dpif, &dpstats);
2416 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2417 memset(ots, 0, sizeof *ots);
2418 ots->table_id = TABLEID_HASH;
2419 strcpy(ots->name, "hash");
2420 ots->wildcards = htonl(0);
2421 ots->max_entries = htonl(dpstats.max_capacity);
2422 ots->active_count = htonl(n_exact);
2423 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2425 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2427 /* Classifier table. */
2428 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2429 memset(ots, 0, sizeof *ots);
2430 ots->table_id = TABLEID_CLASSIFIER;
2431 strcpy(ots->name, "classifier");
2432 ots->wildcards = htonl(OFPFW_ALL);
2433 ots->max_entries = htonl(65536);
2434 ots->active_count = htonl(n_wild);
2435 ots->lookup_count = htonll(0); /* XXX */
2436 ots->matched_count = htonll(0); /* XXX */
2438 queue_tx(msg, ofconn, ofconn->reply_counter);
2443 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2444 struct ofp_stats_request *request)
2446 struct ofp_port_stats *ops;
2448 struct ofport *port;
2449 unsigned int port_no;
2451 msg = start_stats_reply(request, sizeof *ops * 16);
2452 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2453 struct netdev_stats stats;
2455 /* Intentionally ignore return value, since errors will set 'stats' to
2456 * all-1s, which is correct for OpenFlow, and netdev_get_stats() will
2458 netdev_get_stats(port->netdev, &stats);
2460 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2461 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2462 memset(ops->pad, 0, sizeof ops->pad);
2463 ops->rx_packets = htonll(stats.rx_packets);
2464 ops->tx_packets = htonll(stats.tx_packets);
2465 ops->rx_bytes = htonll(stats.rx_bytes);
2466 ops->tx_bytes = htonll(stats.tx_bytes);
2467 ops->rx_dropped = htonll(stats.rx_dropped);
2468 ops->tx_dropped = htonll(stats.tx_dropped);
2469 ops->rx_errors = htonll(stats.rx_errors);
2470 ops->tx_errors = htonll(stats.tx_errors);
2471 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2472 ops->rx_over_err = htonll(stats.rx_over_errors);
2473 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2474 ops->collisions = htonll(stats.collisions);
2477 queue_tx(msg, ofconn, ofconn->reply_counter);
2481 struct flow_stats_cbdata {
2482 struct ofproto *ofproto;
2483 struct ofconn *ofconn;
2489 query_stats(struct ofproto *p, struct rule *rule,
2490 uint64_t *packet_countp, uint64_t *byte_countp)
2492 uint64_t packet_count, byte_count;
2493 struct rule *subrule;
2494 struct odp_flow *odp_flows;
2497 packet_count = rule->packet_count;
2498 byte_count = rule->byte_count;
2500 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2501 odp_flows = xcalloc(1, n_odp_flows * sizeof *odp_flows);
2502 if (rule->cr.wc.wildcards) {
2504 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2505 odp_flows[i++].key = subrule->cr.flow;
2506 packet_count += subrule->packet_count;
2507 byte_count += subrule->byte_count;
2510 odp_flows[0].key = rule->cr.flow;
2513 packet_count = rule->packet_count;
2514 byte_count = rule->byte_count;
2515 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2517 for (i = 0; i < n_odp_flows; i++) {
2518 struct odp_flow *odp_flow = &odp_flows[i];
2519 packet_count += odp_flow->stats.n_packets;
2520 byte_count += odp_flow->stats.n_bytes;
2525 *packet_countp = packet_count;
2526 *byte_countp = byte_count;
2530 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2532 struct rule *rule = rule_from_cls_rule(rule_);
2533 struct flow_stats_cbdata *cbdata = cbdata_;
2534 struct ofp_flow_stats *ofs;
2535 uint64_t packet_count, byte_count;
2536 size_t act_len, len;
2538 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2542 act_len = sizeof *rule->actions * rule->n_actions;
2543 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2545 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2547 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2548 ofs->length = htons(len);
2549 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2551 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match);
2552 ofs->duration = htonl((time_msec() - rule->created) / 1000);
2553 ofs->priority = htons(rule->cr.priority);
2554 ofs->idle_timeout = htons(rule->idle_timeout);
2555 ofs->hard_timeout = htons(rule->hard_timeout);
2557 ofs->packet_count = htonll(packet_count);
2558 ofs->byte_count = htonll(byte_count);
2559 memcpy(ofs->actions, rule->actions, act_len);
2563 table_id_to_include(uint8_t table_id)
2565 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2566 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2567 : table_id == 0xff ? CLS_INC_ALL
2572 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2573 const struct ofp_stats_request *osr,
2576 struct ofp_flow_stats_request *fsr;
2577 struct flow_stats_cbdata cbdata;
2578 struct cls_rule target;
2580 if (arg_size != sizeof *fsr) {
2581 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2583 fsr = (struct ofp_flow_stats_request *) osr->body;
2585 COVERAGE_INC(ofproto_flows_req);
2587 cbdata.ofconn = ofconn;
2588 cbdata.out_port = fsr->out_port;
2589 cbdata.msg = start_stats_reply(osr, 1024);
2590 cls_rule_from_match(&target, &fsr->match, 0);
2591 classifier_for_each_match(&p->cls, &target,
2592 table_id_to_include(fsr->table_id),
2593 flow_stats_cb, &cbdata);
2594 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2598 struct flow_stats_ds_cbdata {
2599 struct ofproto *ofproto;
2604 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
2606 struct rule *rule = rule_from_cls_rule(rule_);
2607 struct flow_stats_ds_cbdata *cbdata = cbdata_;
2608 struct ds *results = cbdata->results;
2609 struct ofp_match match;
2610 uint64_t packet_count, byte_count;
2611 size_t act_len = sizeof *rule->actions * rule->n_actions;
2613 /* Don't report on subrules. */
2614 if (rule->super != NULL) {
2618 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2619 flow_to_ovs_match(&rule->cr.flow, rule->cr.wc.wildcards, &match);
2621 ds_put_format(results, "duration=%llds, ",
2622 (time_msec() - rule->created) / 1000);
2623 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2624 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2625 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2626 ofp_print_match(results, &match, true);
2627 ofp_print_actions(results, &rule->actions->header, act_len);
2628 ds_put_cstr(results, "\n");
2631 /* Adds a pretty-printed description of all flows to 'results', including
2632 * those marked hidden by secchan (e.g., by in-band control). */
2634 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2636 struct ofp_match match;
2637 struct cls_rule target;
2638 struct flow_stats_ds_cbdata cbdata;
2640 memset(&match, 0, sizeof match);
2641 match.wildcards = htonl(OFPFW_ALL);
2644 cbdata.results = results;
2646 cls_rule_from_match(&target, &match, 0);
2647 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2648 flow_stats_ds_cb, &cbdata);
2651 struct aggregate_stats_cbdata {
2652 struct ofproto *ofproto;
2654 uint64_t packet_count;
2655 uint64_t byte_count;
2660 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2662 struct rule *rule = rule_from_cls_rule(rule_);
2663 struct aggregate_stats_cbdata *cbdata = cbdata_;
2664 uint64_t packet_count, byte_count;
2666 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2670 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2672 cbdata->packet_count += packet_count;
2673 cbdata->byte_count += byte_count;
2678 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2679 const struct ofp_stats_request *osr,
2682 struct ofp_aggregate_stats_request *asr;
2683 struct ofp_aggregate_stats_reply *reply;
2684 struct aggregate_stats_cbdata cbdata;
2685 struct cls_rule target;
2688 if (arg_size != sizeof *asr) {
2689 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2691 asr = (struct ofp_aggregate_stats_request *) osr->body;
2693 COVERAGE_INC(ofproto_agg_request);
2695 cbdata.out_port = asr->out_port;
2696 cbdata.packet_count = 0;
2697 cbdata.byte_count = 0;
2699 cls_rule_from_match(&target, &asr->match, 0);
2700 classifier_for_each_match(&p->cls, &target,
2701 table_id_to_include(asr->table_id),
2702 aggregate_stats_cb, &cbdata);
2704 msg = start_stats_reply(osr, sizeof *reply);
2705 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2706 reply->flow_count = htonl(cbdata.n_flows);
2707 reply->packet_count = htonll(cbdata.packet_count);
2708 reply->byte_count = htonll(cbdata.byte_count);
2709 queue_tx(msg, ofconn, ofconn->reply_counter);
2714 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2715 struct ofp_header *oh)
2717 struct ofp_stats_request *osr;
2721 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2726 osr = (struct ofp_stats_request *) oh;
2728 switch (ntohs(osr->type)) {
2730 return handle_desc_stats_request(p, ofconn, osr);
2733 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2735 case OFPST_AGGREGATE:
2736 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2739 return handle_table_stats_request(p, ofconn, osr);
2742 return handle_port_stats_request(p, ofconn, osr);
2745 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2748 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2752 static long long int
2753 msec_from_nsec(uint64_t sec, uint32_t nsec)
2755 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2759 update_time(struct ofproto *ofproto, struct rule *rule,
2760 const struct odp_flow_stats *stats)
2762 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2763 if (used > rule->used) {
2765 if (rule->super && used > rule->super->used) {
2766 rule->super->used = used;
2768 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
2773 update_stats(struct ofproto *ofproto, struct rule *rule,
2774 const struct odp_flow_stats *stats)
2776 if (stats->n_packets) {
2777 update_time(ofproto, rule, stats);
2778 rule->packet_count += stats->n_packets;
2779 rule->byte_count += stats->n_bytes;
2780 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
2786 add_flow(struct ofproto *p, struct ofconn *ofconn,
2787 struct ofp_flow_mod *ofm, size_t n_actions)
2789 struct ofpbuf *packet;
2794 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
2798 flow_from_match(&flow, &wildcards, &ofm->match);
2799 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
2800 ntohs(ofm->priority))) {
2801 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
2805 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
2806 n_actions, ntohs(ofm->idle_timeout),
2807 ntohs(ofm->hard_timeout),
2808 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
2809 cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority));
2813 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2814 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2818 rule_insert(p, rule, packet, in_port);
2819 ofpbuf_delete(packet);
2824 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
2825 size_t n_actions, uint16_t command, struct rule *rule)
2827 if (rule_is_hidden(rule)) {
2831 if (command == OFPFC_DELETE) {
2832 long long int now = time_msec();
2833 send_flow_removed(p, rule, now, OFPRR_DELETE);
2834 rule_remove(p, rule);
2836 size_t actions_len = n_actions * sizeof *rule->actions;
2838 if (n_actions == rule->n_actions
2839 && !memcmp(ofm->actions, rule->actions, actions_len))
2844 free(rule->actions);
2845 rule->actions = xmemdup(ofm->actions, actions_len);
2846 rule->n_actions = n_actions;
2848 if (rule->cr.wc.wildcards) {
2849 COVERAGE_INC(ofproto_mod_wc_flow);
2850 p->need_revalidate = true;
2852 rule_update_actions(p, rule);
2860 modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm,
2861 size_t n_actions, uint16_t command)
2867 flow_from_match(&flow, &wildcards, &ofm->match);
2868 rule = rule_from_cls_rule(classifier_find_rule_exactly(
2869 &p->cls, &flow, wildcards,
2870 ntohs(ofm->priority)));
2873 if (command == OFPFC_DELETE
2874 && ofm->out_port != htons(OFPP_NONE)
2875 && !rule_has_out_port(rule, ofm->out_port)) {
2879 modify_flow(p, ofm, n_actions, command, rule);
2884 struct modify_flows_cbdata {
2885 struct ofproto *ofproto;
2886 const struct ofp_flow_mod *ofm;
2893 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
2895 struct rule *rule = rule_from_cls_rule(rule_);
2896 struct modify_flows_cbdata *cbdata = cbdata_;
2898 if (cbdata->out_port != htons(OFPP_NONE)
2899 && !rule_has_out_port(rule, cbdata->out_port)) {
2903 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions,
2904 cbdata->command, rule);
2908 modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm,
2909 size_t n_actions, uint16_t command)
2911 struct modify_flows_cbdata cbdata;
2912 struct cls_rule target;
2916 cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port
2917 : htons(OFPP_NONE));
2918 cbdata.n_actions = n_actions;
2919 cbdata.command = command;
2921 cls_rule_from_match(&target, &ofm->match, 0);
2923 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2924 modify_flows_cb, &cbdata);
2929 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
2930 struct ofp_flow_mod *ofm)
2935 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
2936 sizeof *ofm->actions, &n_actions);
2941 /* We do not support the emergency flow cache. It will hopefully
2942 * get dropped from OpenFlow in the near future. */
2943 if (ofm->flags & htons(OFPFF_EMERG)) {
2944 /* There isn't a good fit for an error code, so just state that the
2945 * flow table is full. */
2946 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
2949 normalize_match(&ofm->match);
2950 if (!ofm->match.wildcards) {
2951 ofm->priority = htons(UINT16_MAX);
2954 error = validate_actions((const union ofp_action *) ofm->actions,
2955 n_actions, p->max_ports);
2960 switch (ntohs(ofm->command)) {
2962 return add_flow(p, ofconn, ofm, n_actions);
2965 return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY);
2967 case OFPFC_MODIFY_STRICT:
2968 return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY);
2971 return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE);
2973 case OFPFC_DELETE_STRICT:
2974 return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE);
2977 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
2982 send_capability_reply(struct ofproto *p, struct ofconn *ofconn, uint32_t xid)
2984 struct ofmp_capability_reply *ocr;
2986 char capabilities[] = "com.nicira.mgmt.manager=false\n";
2988 ocr = make_openflow_xid(sizeof(*ocr), OFPT_VENDOR, xid, &b);
2989 ocr->header.header.vendor = htonl(NX_VENDOR_ID);
2990 ocr->header.header.subtype = htonl(NXT_MGMT);
2991 ocr->header.type = htons(OFMPT_CAPABILITY_REPLY);
2993 ocr->format = htonl(OFMPCOF_SIMPLE);
2994 ocr->mgmt_id = htonll(p->mgmt_id);
2996 ofpbuf_put(b, capabilities, strlen(capabilities));
2998 queue_tx(b, ofconn, ofconn->reply_counter);
3002 handle_ofmp(struct ofproto *p, struct ofconn *ofconn,
3003 struct ofmp_header *ofmph)
3005 size_t msg_len = ntohs(ofmph->header.header.length);
3006 if (msg_len < sizeof(*ofmph)) {
3007 VLOG_WARN_RL(&rl, "dropping short managment message: %zu\n", msg_len);
3008 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3011 if (ofmph->type == htons(OFMPT_CAPABILITY_REQUEST)) {
3012 struct ofmp_capability_request *ofmpcr;
3014 if (msg_len < sizeof(struct ofmp_capability_request)) {
3015 VLOG_WARN_RL(&rl, "dropping short capability request: %zu\n",
3017 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3020 ofmpcr = (struct ofmp_capability_request *)ofmph;
3021 if (ofmpcr->format != htonl(OFMPCAF_SIMPLE)) {
3022 /* xxx Find a better type than bad subtype */
3023 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3026 send_capability_reply(p, ofconn, ofmph->header.header.xid);
3029 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3034 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3036 struct ofp_vendor_header *ovh = msg;
3037 struct nicira_header *nh;
3039 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3040 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3042 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3043 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3045 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3046 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3050 switch (ntohl(nh->subtype)) {
3051 case NXT_STATUS_REQUEST:
3052 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3055 case NXT_ACT_SET_CONFIG:
3056 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
3058 case NXT_ACT_GET_CONFIG:
3059 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
3061 case NXT_COMMAND_REQUEST:
3063 return executer_handle_request(p->executer, ofconn->rconn, msg);
3068 return handle_ofmp(p, ofconn, msg);
3071 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3075 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
3077 struct ofp_header *ob;
3080 /* Currently, everything executes synchronously, so we can just
3081 * immediately send the barrier reply. */
3082 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3083 queue_tx(buf, ofconn, ofconn->reply_counter);
3088 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3089 struct ofpbuf *ofp_msg)
3091 struct ofp_header *oh = ofp_msg->data;
3094 COVERAGE_INC(ofproto_recv_openflow);
3096 case OFPT_ECHO_REQUEST:
3097 error = handle_echo_request(ofconn, oh);
3100 case OFPT_ECHO_REPLY:
3104 case OFPT_FEATURES_REQUEST:
3105 error = handle_features_request(p, ofconn, oh);
3108 case OFPT_GET_CONFIG_REQUEST:
3109 error = handle_get_config_request(p, ofconn, oh);
3112 case OFPT_SET_CONFIG:
3113 error = handle_set_config(p, ofconn, ofp_msg->data);
3116 case OFPT_PACKET_OUT:
3117 error = handle_packet_out(p, ofconn, ofp_msg->data);
3121 error = handle_port_mod(p, oh);
3125 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3128 case OFPT_STATS_REQUEST:
3129 error = handle_stats_request(p, ofconn, oh);
3133 error = handle_vendor(p, ofconn, ofp_msg->data);
3136 case OFPT_BARRIER_REQUEST:
3137 error = handle_barrier_request(ofconn, oh);
3141 if (VLOG_IS_WARN_ENABLED()) {
3142 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3143 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3146 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3151 send_error_oh(ofconn, ofp_msg->data, error);
3156 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3158 struct odp_msg *msg = packet->data;
3159 uint16_t in_port = odp_port_to_ofp_port(msg->port);
3161 struct ofpbuf payload;
3164 /* Handle controller actions. */
3165 if (msg->type == _ODPL_ACTION_NR) {
3166 COVERAGE_INC(ofproto_ctlr_action);
3167 pinsched_send(p->action_sched, in_port, packet,
3168 send_packet_in_action, p);
3172 payload.data = msg + 1;
3173 payload.size = msg->length - sizeof *msg;
3174 flow_extract(&payload, msg->port, &flow);
3176 /* Check with in-band control to see if this packet should be sent
3177 * to the local port regardless of the flow table. */
3178 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3179 union odp_action action;
3181 memset(&action, 0, sizeof(action));
3182 action.output.type = ODPAT_OUTPUT;
3183 action.output.port = ODPP_LOCAL;
3184 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3187 rule = lookup_valid_rule(p, &flow);
3189 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3190 struct ofport *port = port_array_get(&p->ports, msg->port);
3192 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3193 COVERAGE_INC(ofproto_no_packet_in);
3194 /* XXX install 'drop' flow entry */
3195 ofpbuf_delete(packet);
3199 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3202 COVERAGE_INC(ofproto_packet_in);
3203 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3207 if (rule->cr.wc.wildcards) {
3208 rule = rule_create_subrule(p, rule, &flow);
3209 rule_make_actions(p, rule, packet);
3211 if (!rule->may_install) {
3212 /* The rule is not installable, that is, we need to process every
3213 * packet, so process the current packet and set its actions into
3215 rule_make_actions(p, rule, packet);
3217 /* XXX revalidate rule if it needs it */
3221 rule_execute(p, rule, &payload, &flow);
3222 rule_reinstall(p, rule);
3224 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY
3225 && rconn_is_connected(p->controller->rconn)) {
3227 * Extra-special case for fail-open mode.
3229 * We are in fail-open mode and the packet matched the fail-open rule,
3230 * but we are connected to a controller too. We should send the packet
3231 * up to the controller in the hope that it will try to set up a flow
3232 * and thereby allow us to exit fail-open.
3234 * See the top-level comment in fail-open.c for more information.
3236 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3238 ofpbuf_delete(packet);
3243 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3245 struct rule *sub = rule_from_cls_rule(sub_);
3246 struct revalidate_cbdata *cbdata = cbdata_;
3248 if (cbdata->revalidate_all
3249 || (cbdata->revalidate_subrules && sub->super)
3250 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3251 revalidate_rule(cbdata->ofproto, sub);
3256 revalidate_rule(struct ofproto *p, struct rule *rule)
3258 const flow_t *flow = &rule->cr.flow;
3260 COVERAGE_INC(ofproto_revalidate_rule);
3263 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3265 rule_remove(p, rule);
3267 } else if (super != rule->super) {
3268 COVERAGE_INC(ofproto_revalidate_moved);
3269 list_remove(&rule->list);
3270 list_push_back(&super->list, &rule->list);
3271 rule->super = super;
3272 rule->hard_timeout = super->hard_timeout;
3273 rule->idle_timeout = super->idle_timeout;
3274 rule->created = super->created;
3279 rule_update_actions(p, rule);
3283 static struct ofpbuf *
3284 compose_flow_removed(const struct rule *rule, long long int now, uint8_t reason)
3286 struct ofp_flow_removed *ofr;
3288 long long int last_used = rule->used ? now - rule->used : 0;
3290 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
3291 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofr->match);
3292 ofr->priority = htons(rule->cr.priority);
3293 ofr->reason = reason;
3294 ofr->duration = htonl((now - rule->created - last_used) / 1000);
3295 ofr->idle_timeout = htons(rule->idle_timeout);
3296 ofr->packet_count = htonll(rule->packet_count);
3297 ofr->byte_count = htonll(rule->byte_count);
3303 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3305 assert(rule->installed);
3306 assert(!rule->cr.wc.wildcards);
3309 rule_remove(ofproto, rule);
3311 rule_uninstall(ofproto, rule);
3315 send_flow_removed(struct ofproto *p, struct rule *rule,
3316 long long int now, uint8_t reason)
3318 struct ofconn *ofconn;
3319 struct ofconn *prev;
3320 struct ofpbuf *buf = NULL;
3322 /* We limit the maximum number of queued flow expirations it by accounting
3323 * them under the counter for replies. That works because preventing
3324 * OpenFlow requests from being processed also prevents new flows from
3325 * being added (and expiring). (It also prevents processing OpenFlow
3326 * requests that would not add new flows, so it is imperfect.) */
3329 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3330 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)) {
3332 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3334 buf = compose_flow_removed(rule, now, reason);
3340 queue_tx(buf, prev, prev->reply_counter);
3346 expire_rule(struct cls_rule *cls_rule, void *p_)
3348 struct ofproto *p = p_;
3349 struct rule *rule = rule_from_cls_rule(cls_rule);
3350 long long int hard_expire, idle_expire, expire, now;
3352 hard_expire = (rule->hard_timeout
3353 ? rule->created + rule->hard_timeout * 1000
3355 idle_expire = (rule->idle_timeout
3356 && (rule->super || list_is_empty(&rule->list))
3357 ? rule->used + rule->idle_timeout * 1000
3359 expire = MIN(hard_expire, idle_expire);
3363 if (rule->installed && now >= rule->used + 5000) {
3364 uninstall_idle_flow(p, rule);
3365 } else if (!rule->cr.wc.wildcards) {
3366 active_timeout(p, rule);
3372 COVERAGE_INC(ofproto_expired);
3374 /* Update stats. This code will be a no-op if the rule expired
3375 * due to an idle timeout. */
3376 if (rule->cr.wc.wildcards) {
3377 struct rule *subrule, *next;
3378 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3379 rule_remove(p, subrule);
3382 rule_uninstall(p, rule);
3385 if (!rule_is_hidden(rule)) {
3386 send_flow_removed(p, rule, now,
3388 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
3390 rule_remove(p, rule);
3394 active_timeout(struct ofproto *ofproto, struct rule *rule)
3396 if (ofproto->netflow && !is_controller_rule(rule) &&
3397 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
3398 struct ofexpired expired;
3399 struct odp_flow odp_flow;
3401 /* Get updated flow stats. */
3402 memset(&odp_flow, 0, sizeof odp_flow);
3403 if (rule->installed) {
3404 odp_flow.key = rule->cr.flow;
3405 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
3406 dpif_flow_get(ofproto->dpif, &odp_flow);
3408 if (odp_flow.stats.n_packets) {
3409 update_time(ofproto, rule, &odp_flow.stats);
3410 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
3411 odp_flow.stats.tcp_flags);
3415 expired.flow = rule->cr.flow;
3416 expired.packet_count = rule->packet_count +
3417 odp_flow.stats.n_packets;
3418 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
3419 expired.used = rule->used;
3421 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
3423 /* Schedule us to send the accumulated records once we have
3424 * collected all of them. */
3425 poll_immediate_wake();
3430 update_used(struct ofproto *p)
3432 struct odp_flow *flows;
3437 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
3442 for (i = 0; i < n_flows; i++) {
3443 struct odp_flow *f = &flows[i];
3446 rule = rule_from_cls_rule(
3447 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3448 if (!rule || !rule->installed) {
3449 COVERAGE_INC(ofproto_unexpected_rule);
3450 dpif_flow_del(p->dpif, f);
3454 update_time(p, rule, &f->stats);
3455 rule_account(p, rule, f->stats.n_bytes);
3461 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3462 const struct ofpbuf *packet, int send_len)
3464 struct odp_msg *msg = packet->data;
3465 struct ofpbuf payload;
3469 /* Extract packet payload from 'msg'. */
3470 payload.data = msg + 1;
3471 payload.size = msg->length - sizeof *msg;
3473 /* Construct ofp_packet_in message. */
3474 reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3475 opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port), reason,
3476 &payload, send_len);
3479 rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100);
3483 send_packet_in_action(struct ofpbuf *packet, void *p_)
3485 struct ofproto *p = p_;
3486 struct ofconn *ofconn;
3487 struct odp_msg *msg;
3490 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3491 if (ofconn == p->controller || ofconn->miss_send_len) {
3492 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3495 ofpbuf_delete(packet);
3499 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3501 struct ofproto *p = p_;
3502 bool in_fail_open = p->fail_open && fail_open_is_active(p->fail_open);
3503 struct ofconn *ofconn;
3504 struct ofpbuf payload;
3505 struct odp_msg *msg;
3508 payload.data = msg + 1;
3509 payload.size = msg->length - sizeof *msg;
3510 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3511 if (ofconn->miss_send_len) {
3512 struct pktbuf *pb = ofconn->pktbuf;
3513 uint32_t buffer_id = (in_fail_open
3515 : pktbuf_save(pb, &payload, msg->port));
3516 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3518 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3521 ofpbuf_delete(packet);
3525 pick_datapath_id(const struct ofproto *ofproto)
3527 const struct ofport *port;
3529 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
3531 uint8_t ea[ETH_ADDR_LEN];
3534 error = netdev_get_etheraddr(port->netdev, ea);
3536 return eth_addr_to_uint64(ea);
3538 VLOG_WARN("could not get MAC address for %s (%s)",
3539 netdev_get_name(port->netdev), strerror(error));
3541 return ofproto->fallback_dpid;
3545 pick_fallback_dpid(void)
3547 uint8_t ea[ETH_ADDR_LEN];
3548 eth_addr_random(ea);
3549 ea[0] = 0x00; /* Set Nicira OUI. */
3552 return eth_addr_to_uint64(ea);
3556 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3557 struct odp_actions *actions, tag_type *tags,
3558 uint16_t *nf_output_iface, void *ofproto_)
3560 struct ofproto *ofproto = ofproto_;
3563 /* Drop frames for reserved multicast addresses. */
3564 if (eth_addr_is_reserved(flow->dl_dst)) {
3568 /* Learn source MAC (but don't try to learn from revalidation). */
3569 if (packet != NULL) {
3570 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3573 /* The log messages here could actually be useful in debugging,
3574 * so keep the rate limit relatively high. */
3575 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3576 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3577 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3578 ofproto_revalidate(ofproto, rev_tag);
3582 /* Determine output port. */
3583 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3585 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
3586 } else if (out_port != flow->in_port) {
3587 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3588 *nf_output_iface = out_port;
3596 static const struct ofhooks default_ofhooks = {
3598 default_normal_ofhook_cb,