2 * Copyright (c) 2009, 2010 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
22 #include <netinet/in.h>
25 #include "classifier.h"
27 #include "discovery.h"
29 #include "dynamic-string.h"
31 #include "fail-open.h"
33 #include "mac-learning.h"
37 #include "ofp-print.h"
38 #include "ofproto-sflow.h"
40 #include "openflow/nicira-ext.h"
41 #include "openflow/openflow.h"
42 #include "openflow/openflow-mgmt.h"
43 #include "openvswitch/datapath-protocol.h"
47 #include "poll-loop.h"
48 #include "port-array.h"
58 #include "vconn-ssl.h"
61 #define THIS_MODULE VLM_ofproto
64 #include "sflow_api.h"
68 TABLEID_CLASSIFIER = 1
72 struct netdev *netdev;
73 struct ofp_phy_port opp; /* In host byte order. */
76 static void ofport_free(struct ofport *);
77 static void hton_ofp_phy_port(struct ofp_phy_port *);
79 static int xlate_actions(const union ofp_action *in, size_t n_in,
80 const flow_t *flow, struct ofproto *ofproto,
81 const struct ofpbuf *packet,
82 struct odp_actions *out, tag_type *tags,
83 bool *may_set_up_flow, uint16_t *nf_output_iface);
88 uint16_t idle_timeout; /* In seconds from time of last use. */
89 uint16_t hard_timeout; /* In seconds from time of creation. */
90 long long int used; /* Last-used time (0 if never used). */
91 long long int created; /* Creation time. */
92 uint64_t packet_count; /* Number of packets received. */
93 uint64_t byte_count; /* Number of bytes received. */
94 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
95 tag_type tags; /* Tags (set only by hooks). */
96 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
98 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
99 * exact-match rule (having cr.wc.wildcards of 0) generated from the
100 * wildcard rule 'super'. In this case, 'list' is an element of the
103 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
104 * a list of subrules. A super-rule with no wildcards (where
105 * cr.wc.wildcards is 0) will never have any subrules. */
111 * A subrule has no actions (it uses the super-rule's actions). */
113 union ofp_action *actions;
117 * A super-rule with wildcard fields never has ODP actions (since the
118 * datapath only supports exact-match flows). */
119 bool installed; /* Installed in datapath? */
120 bool may_install; /* True ordinarily; false if actions must
121 * be reassessed for every packet. */
123 union odp_action *odp_actions;
127 rule_is_hidden(const struct rule *rule)
129 /* Subrules are merely an implementation detail, so hide them from the
131 if (rule->super != NULL) {
135 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
136 * (e.g. by in-band control) and are intentionally hidden from the
138 if (rule->cr.priority > UINT16_MAX) {
145 static struct rule *rule_create(struct ofproto *, struct rule *super,
146 const union ofp_action *, size_t n_actions,
147 uint16_t idle_timeout, uint16_t hard_timeout);
148 static void rule_free(struct rule *);
149 static void rule_destroy(struct ofproto *, struct rule *);
150 static struct rule *rule_from_cls_rule(const struct cls_rule *);
151 static void rule_insert(struct ofproto *, struct rule *,
152 struct ofpbuf *packet, uint16_t in_port);
153 static void rule_remove(struct ofproto *, struct rule *);
154 static bool rule_make_actions(struct ofproto *, struct rule *,
155 const struct ofpbuf *packet);
156 static void rule_install(struct ofproto *, struct rule *,
157 struct rule *displaced_rule);
158 static void rule_uninstall(struct ofproto *, struct rule *);
159 static void rule_post_uninstall(struct ofproto *, struct rule *);
164 struct pktbuf *pktbuf;
168 struct rconn_packet_counter *packet_in_counter;
170 /* Number of OpenFlow messages queued as replies to OpenFlow requests, and
171 * the maximum number before we stop reading OpenFlow requests. */
172 #define OFCONN_REPLY_MAX 100
173 struct rconn_packet_counter *reply_counter;
176 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *);
177 static void ofconn_destroy(struct ofconn *, struct ofproto *);
178 static void ofconn_run(struct ofconn *, struct ofproto *);
179 static void ofconn_wait(struct ofconn *);
180 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
181 struct rconn_packet_counter *counter);
185 uint64_t datapath_id; /* Datapath ID. */
186 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
187 uint64_t mgmt_id; /* Management channel identifier. */
188 char *manufacturer; /* Manufacturer. */
189 char *hardware; /* Hardware. */
190 char *software; /* Software version. */
191 char *serial; /* Serial number. */
195 struct netdev_monitor *netdev_monitor;
196 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
198 struct shash port_by_name;
202 struct switch_status *switch_status;
203 struct status_category *ss_cat;
204 struct in_band *in_band;
205 struct discovery *discovery;
206 struct fail_open *fail_open;
207 struct pinsched *miss_sched, *action_sched;
208 struct executer *executer;
209 struct netflow *netflow;
210 struct ofproto_sflow *sflow;
213 struct classifier cls;
214 bool need_revalidate;
215 long long int next_expiration;
216 struct tag_set revalidate_set;
218 /* OpenFlow connections. */
219 struct list all_conns;
220 struct ofconn *controller;
221 struct pvconn **listeners;
223 struct pvconn **snoops;
226 /* Hooks for ovs-vswitchd. */
227 const struct ofhooks *ofhooks;
230 /* Used by default ofhooks. */
231 struct mac_learning *ml;
234 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
236 static const struct ofhooks default_ofhooks;
238 static uint64_t pick_datapath_id(const struct ofproto *);
239 static uint64_t pick_fallback_dpid(void);
240 static void send_packet_in_miss(struct ofpbuf *, void *ofproto);
241 static void send_packet_in_action(struct ofpbuf *, void *ofproto);
242 static void update_used(struct ofproto *);
243 static void update_stats(struct ofproto *, struct rule *,
244 const struct odp_flow_stats *);
245 static void expire_rule(struct cls_rule *, void *ofproto);
246 static void active_timeout(struct ofproto *ofproto, struct rule *rule);
247 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
248 static void revalidate_cb(struct cls_rule *rule_, void *p_);
250 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
252 static void handle_openflow(struct ofconn *, struct ofproto *,
255 static void refresh_port_groups(struct ofproto *);
257 static void update_port(struct ofproto *, const char *devname);
258 static int init_ports(struct ofproto *);
259 static void reinit_ports(struct ofproto *);
262 ofproto_create(const char *datapath, const struct ofhooks *ofhooks, void *aux,
263 struct ofproto **ofprotop)
265 struct odp_stats stats;
272 /* Connect to datapath and start listening for messages. */
273 error = dpif_open(datapath, &dpif);
275 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
278 error = dpif_get_dp_stats(dpif, &stats);
280 VLOG_ERR("failed to obtain stats for datapath %s: %s",
281 datapath, strerror(error));
285 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
287 VLOG_ERR("failed to listen on datapath %s: %s",
288 datapath, strerror(error));
292 dpif_flow_flush(dpif);
293 dpif_recv_purge(dpif);
295 /* Initialize settings. */
296 p = xcalloc(1, sizeof *p);
297 p->fallback_dpid = pick_fallback_dpid();
298 p->datapath_id = p->fallback_dpid;
299 p->manufacturer = xstrdup("Nicira Networks, Inc.");
300 p->hardware = xstrdup("Reference Implementation");
301 p->software = xstrdup(VERSION BUILDNR);
302 p->serial = xstrdup("None");
304 /* Initialize datapath. */
306 p->netdev_monitor = netdev_monitor_create();
307 port_array_init(&p->ports);
308 shash_init(&p->port_by_name);
309 p->max_ports = stats.max_ports;
311 /* Initialize submodules. */
312 p->switch_status = switch_status_create(p);
316 p->miss_sched = p->action_sched = NULL;
321 /* Initialize flow table. */
322 classifier_init(&p->cls);
323 p->need_revalidate = false;
324 p->next_expiration = time_msec() + 1000;
325 tag_set_init(&p->revalidate_set);
327 /* Initialize OpenFlow connections. */
328 list_init(&p->all_conns);
329 p->controller = ofconn_create(p, rconn_create(5, 8));
330 p->controller->pktbuf = pktbuf_create();
331 p->controller->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
337 /* Initialize hooks. */
339 p->ofhooks = ofhooks;
343 p->ofhooks = &default_ofhooks;
345 p->ml = mac_learning_create();
348 /* Register switch status category. */
349 p->ss_cat = switch_status_register(p->switch_status, "remote",
350 rconn_status_cb, p->controller->rconn);
353 error = init_ports(p);
359 /* Pick final datapath ID. */
360 p->datapath_id = pick_datapath_id(p);
361 VLOG_INFO("using datapath ID %012"PRIx64, p->datapath_id);
368 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
370 uint64_t old_dpid = p->datapath_id;
371 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
372 if (p->datapath_id != old_dpid) {
373 VLOG_INFO("datapath ID changed to %012"PRIx64, p->datapath_id);
374 rconn_reconnect(p->controller->rconn);
379 ofproto_set_mgmt_id(struct ofproto *p, uint64_t mgmt_id)
381 p->mgmt_id = mgmt_id;
385 ofproto_set_probe_interval(struct ofproto *p, int probe_interval)
387 probe_interval = probe_interval ? MAX(probe_interval, 5) : 0;
388 rconn_set_probe_interval(p->controller->rconn, probe_interval);
390 int trigger_duration = probe_interval ? probe_interval * 3 : 15;
391 fail_open_set_trigger_duration(p->fail_open, trigger_duration);
396 ofproto_set_max_backoff(struct ofproto *p, int max_backoff)
398 rconn_set_max_backoff(p->controller->rconn, max_backoff);
402 ofproto_set_desc(struct ofproto *p,
403 const char *manufacturer, const char *hardware,
404 const char *software, const char *serial)
407 free(p->manufacturer);
408 p->manufacturer = xstrdup(manufacturer);
412 p->hardware = xstrdup(hardware);
416 p->software = xstrdup(software);
420 p->serial = xstrdup(serial);
425 ofproto_set_in_band(struct ofproto *p, bool in_band)
427 if (in_band != (p->in_band != NULL)) {
429 return in_band_create(p, p->dpif, p->switch_status,
430 p->controller->rconn, &p->in_band);
432 ofproto_set_discovery(p, false, NULL, true);
433 in_band_destroy(p->in_band);
436 rconn_reconnect(p->controller->rconn);
442 ofproto_set_discovery(struct ofproto *p, bool discovery,
443 const char *re, bool update_resolv_conf)
445 if (discovery != (p->discovery != NULL)) {
447 int error = ofproto_set_in_band(p, true);
451 error = discovery_create(re, update_resolv_conf,
452 p->dpif, p->switch_status,
458 discovery_destroy(p->discovery);
461 rconn_disconnect(p->controller->rconn);
462 } else if (discovery) {
463 discovery_set_update_resolv_conf(p->discovery, update_resolv_conf);
464 return discovery_set_accept_controller_re(p->discovery, re);
470 ofproto_set_controller(struct ofproto *ofproto, const char *controller)
472 if (ofproto->discovery) {
474 } else if (controller) {
475 if (strcmp(rconn_get_name(ofproto->controller->rconn), controller)) {
476 return rconn_connect(ofproto->controller->rconn, controller);
481 rconn_disconnect(ofproto->controller->rconn);
487 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
488 const struct svec *svec)
490 struct pvconn **pvconns = *pvconnsp;
491 size_t n_pvconns = *n_pvconnsp;
495 for (i = 0; i < n_pvconns; i++) {
496 pvconn_close(pvconns[i]);
500 pvconns = xmalloc(svec->n * sizeof *pvconns);
502 for (i = 0; i < svec->n; i++) {
503 const char *name = svec->names[i];
504 struct pvconn *pvconn;
507 error = pvconn_open(name, &pvconn);
509 pvconns[n_pvconns++] = pvconn;
511 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
519 *n_pvconnsp = n_pvconns;
525 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
527 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
531 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
533 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
537 ofproto_set_netflow(struct ofproto *ofproto,
538 const struct netflow_options *nf_options)
540 if (nf_options->collectors.n) {
541 if (!ofproto->netflow) {
542 ofproto->netflow = netflow_create();
544 return netflow_set_options(ofproto->netflow, nf_options);
546 netflow_destroy(ofproto->netflow);
547 ofproto->netflow = NULL;
553 ofproto_set_sflow(struct ofproto *ofproto,
554 const struct ofproto_sflow_options *oso)
556 struct ofproto_sflow *os = ofproto->sflow;
559 struct ofport *ofport;
560 unsigned int odp_port;
562 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
563 refresh_port_groups(ofproto);
564 PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) {
565 ofproto_sflow_add_port(os, odp_port,
566 netdev_get_name(ofport->netdev));
569 ofproto_sflow_set_options(os, oso);
571 ofproto_sflow_destroy(os);
572 ofproto->sflow = NULL;
577 ofproto_set_failure(struct ofproto *ofproto, bool fail_open)
580 struct rconn *rconn = ofproto->controller->rconn;
581 int trigger_duration = rconn_get_probe_interval(rconn) * 3;
582 if (!ofproto->fail_open) {
583 ofproto->fail_open = fail_open_create(ofproto, trigger_duration,
584 ofproto->switch_status,
587 fail_open_set_trigger_duration(ofproto->fail_open,
591 fail_open_destroy(ofproto->fail_open);
592 ofproto->fail_open = NULL;
597 ofproto_set_rate_limit(struct ofproto *ofproto,
598 int rate_limit, int burst_limit)
600 if (rate_limit > 0) {
601 if (!ofproto->miss_sched) {
602 ofproto->miss_sched = pinsched_create(rate_limit, burst_limit,
603 ofproto->switch_status);
604 ofproto->action_sched = pinsched_create(rate_limit, burst_limit,
607 pinsched_set_limits(ofproto->miss_sched, rate_limit, burst_limit);
608 pinsched_set_limits(ofproto->action_sched,
609 rate_limit, burst_limit);
612 pinsched_destroy(ofproto->miss_sched);
613 ofproto->miss_sched = NULL;
614 pinsched_destroy(ofproto->action_sched);
615 ofproto->action_sched = NULL;
620 ofproto_set_stp(struct ofproto *ofproto OVS_UNUSED, bool enable_stp)
624 VLOG_WARN("STP is not yet implemented");
632 ofproto_set_remote_execution(struct ofproto *ofproto, const char *command_acl,
633 const char *command_dir)
636 if (!ofproto->executer) {
637 return executer_create(command_acl, command_dir,
640 executer_set_acl(ofproto->executer, command_acl, command_dir);
643 executer_destroy(ofproto->executer);
644 ofproto->executer = NULL;
650 ofproto_get_datapath_id(const struct ofproto *ofproto)
652 return ofproto->datapath_id;
656 ofproto_get_mgmt_id(const struct ofproto *ofproto)
658 return ofproto->mgmt_id;
662 ofproto_get_probe_interval(const struct ofproto *ofproto)
664 return rconn_get_probe_interval(ofproto->controller->rconn);
668 ofproto_get_max_backoff(const struct ofproto *ofproto)
670 return rconn_get_max_backoff(ofproto->controller->rconn);
674 ofproto_get_in_band(const struct ofproto *ofproto)
676 return ofproto->in_band != NULL;
680 ofproto_get_discovery(const struct ofproto *ofproto)
682 return ofproto->discovery != NULL;
686 ofproto_get_controller(const struct ofproto *ofproto)
688 return rconn_get_name(ofproto->controller->rconn);
692 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
696 for (i = 0; i < ofproto->n_listeners; i++) {
697 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
702 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
706 for (i = 0; i < ofproto->n_snoops; i++) {
707 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
712 ofproto_destroy(struct ofproto *p)
714 struct ofconn *ofconn, *next_ofconn;
715 struct ofport *ofport;
716 unsigned int port_no;
723 ofproto_flush_flows(p);
724 classifier_destroy(&p->cls);
726 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
728 ofconn_destroy(ofconn, p);
732 netdev_monitor_destroy(p->netdev_monitor);
733 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
736 shash_destroy(&p->port_by_name);
738 switch_status_destroy(p->switch_status);
739 in_band_destroy(p->in_band);
740 discovery_destroy(p->discovery);
741 fail_open_destroy(p->fail_open);
742 pinsched_destroy(p->miss_sched);
743 pinsched_destroy(p->action_sched);
744 executer_destroy(p->executer);
745 netflow_destroy(p->netflow);
746 ofproto_sflow_destroy(p->sflow);
748 switch_status_unregister(p->ss_cat);
750 for (i = 0; i < p->n_listeners; i++) {
751 pvconn_close(p->listeners[i]);
755 for (i = 0; i < p->n_snoops; i++) {
756 pvconn_close(p->snoops[i]);
760 mac_learning_destroy(p->ml);
762 port_array_destroy(&p->ports);
768 ofproto_run(struct ofproto *p)
770 int error = ofproto_run1(p);
772 error = ofproto_run2(p, false);
778 process_port_change(struct ofproto *ofproto, int error, char *devname)
780 if (error == ENOBUFS) {
781 reinit_ports(ofproto);
783 update_port(ofproto, devname);
789 ofproto_run1(struct ofproto *p)
791 struct ofconn *ofconn, *next_ofconn;
796 for (i = 0; i < 50; i++) {
800 error = dpif_recv(p->dpif, &buf);
802 if (error == ENODEV) {
803 /* Someone destroyed the datapath behind our back. The caller
804 * better destroy us and give up, because we're just going to
805 * spin from here on out. */
806 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
807 VLOG_ERR_RL(&rl, "%s: datapath was destroyed externally",
814 handle_odp_msg(p, buf);
817 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
818 process_port_change(p, error, devname);
820 while ((error = netdev_monitor_poll(p->netdev_monitor,
821 &devname)) != EAGAIN) {
822 process_port_change(p, error, devname);
826 in_band_run(p->in_band);
829 char *controller_name;
830 if (rconn_is_connectivity_questionable(p->controller->rconn)) {
831 discovery_question_connectivity(p->discovery);
833 if (discovery_run(p->discovery, &controller_name)) {
834 if (controller_name) {
835 rconn_connect(p->controller->rconn, controller_name);
837 rconn_disconnect(p->controller->rconn);
841 pinsched_run(p->miss_sched, send_packet_in_miss, p);
842 pinsched_run(p->action_sched, send_packet_in_action, p);
844 executer_run(p->executer);
847 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
849 ofconn_run(ofconn, p);
852 /* Fail-open maintenance. Do this after processing the ofconns since
853 * fail-open checks the status of the controller rconn. */
855 fail_open_run(p->fail_open);
858 for (i = 0; i < p->n_listeners; i++) {
862 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
864 ofconn_create(p, rconn_new_from_vconn("passive", vconn));
865 } else if (retval != EAGAIN) {
866 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
870 for (i = 0; i < p->n_snoops; i++) {
874 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
876 rconn_add_monitor(p->controller->rconn, vconn);
877 } else if (retval != EAGAIN) {
878 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
882 if (time_msec() >= p->next_expiration) {
883 COVERAGE_INC(ofproto_expiration);
884 p->next_expiration = time_msec() + 1000;
887 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
889 /* Let the hook know that we're at a stable point: all outstanding data
890 * in existing flows has been accounted to the account_cb. Thus, the
891 * hook can now reasonably do operations that depend on having accurate
892 * flow volume accounting (currently, that's just bond rebalancing). */
893 if (p->ofhooks->account_checkpoint_cb) {
894 p->ofhooks->account_checkpoint_cb(p->aux);
899 netflow_run(p->netflow);
902 ofproto_sflow_run(p->sflow);
908 struct revalidate_cbdata {
909 struct ofproto *ofproto;
910 bool revalidate_all; /* Revalidate all exact-match rules? */
911 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
912 struct tag_set revalidate_set; /* Set of tags to revalidate. */
916 ofproto_run2(struct ofproto *p, bool revalidate_all)
918 if (p->need_revalidate || revalidate_all
919 || !tag_set_is_empty(&p->revalidate_set)) {
920 struct revalidate_cbdata cbdata;
922 cbdata.revalidate_all = revalidate_all;
923 cbdata.revalidate_subrules = p->need_revalidate;
924 cbdata.revalidate_set = p->revalidate_set;
925 tag_set_init(&p->revalidate_set);
926 COVERAGE_INC(ofproto_revalidate);
927 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
928 p->need_revalidate = false;
935 ofproto_wait(struct ofproto *p)
937 struct ofconn *ofconn;
940 dpif_recv_wait(p->dpif);
941 dpif_port_poll_wait(p->dpif);
942 netdev_monitor_poll_wait(p->netdev_monitor);
943 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
947 in_band_wait(p->in_band);
950 discovery_wait(p->discovery);
953 fail_open_wait(p->fail_open);
955 pinsched_wait(p->miss_sched);
956 pinsched_wait(p->action_sched);
958 executer_wait(p->executer);
961 ofproto_sflow_wait(p->sflow);
963 if (!tag_set_is_empty(&p->revalidate_set)) {
964 poll_immediate_wake();
966 if (p->need_revalidate) {
967 /* Shouldn't happen, but if it does just go around again. */
968 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
969 poll_immediate_wake();
970 } else if (p->next_expiration != LLONG_MAX) {
971 poll_timer_wait(p->next_expiration - time_msec());
973 for (i = 0; i < p->n_listeners; i++) {
974 pvconn_wait(p->listeners[i]);
976 for (i = 0; i < p->n_snoops; i++) {
977 pvconn_wait(p->snoops[i]);
982 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
984 tag_set_add(&ofproto->revalidate_set, tag);
988 ofproto_get_revalidate_set(struct ofproto *ofproto)
990 return &ofproto->revalidate_set;
994 ofproto_is_alive(const struct ofproto *p)
996 return p->discovery || rconn_is_alive(p->controller->rconn);
1000 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
1001 const union ofp_action *actions, size_t n_actions,
1002 const struct ofpbuf *packet)
1004 struct odp_actions odp_actions;
1007 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1013 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1015 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
1016 odp_actions.n_actions, packet);
1021 ofproto_add_flow(struct ofproto *p,
1022 const flow_t *flow, uint32_t wildcards, unsigned int priority,
1023 const union ofp_action *actions, size_t n_actions,
1027 rule = rule_create(p, NULL, actions, n_actions,
1028 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */, 0);
1029 cls_rule_from_flow(&rule->cr, flow, wildcards, priority);
1030 rule_insert(p, rule, NULL, 0);
1034 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1035 uint32_t wildcards, unsigned int priority)
1039 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1043 rule_remove(ofproto, rule);
1048 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1050 struct rule *rule = rule_from_cls_rule(rule_);
1051 struct ofproto *ofproto = ofproto_;
1053 /* Mark the flow as not installed, even though it might really be
1054 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1055 * There is no point in uninstalling it individually since we are about to
1056 * blow away all the flows with dpif_flow_flush(). */
1057 rule->installed = false;
1059 rule_remove(ofproto, rule);
1063 ofproto_flush_flows(struct ofproto *ofproto)
1065 COVERAGE_INC(ofproto_flush);
1066 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1067 dpif_flow_flush(ofproto->dpif);
1068 if (ofproto->in_band) {
1069 in_band_flushed(ofproto->in_band);
1071 if (ofproto->fail_open) {
1072 fail_open_flushed(ofproto->fail_open);
1077 reinit_ports(struct ofproto *p)
1079 struct svec devnames;
1080 struct ofport *ofport;
1081 unsigned int port_no;
1082 struct odp_port *odp_ports;
1086 svec_init(&devnames);
1087 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1088 svec_add (&devnames, (char *) ofport->opp.name);
1090 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1091 for (i = 0; i < n_odp_ports; i++) {
1092 svec_add (&devnames, odp_ports[i].devname);
1096 svec_sort_unique(&devnames);
1097 for (i = 0; i < devnames.n; i++) {
1098 update_port(p, devnames.names[i]);
1100 svec_destroy(&devnames);
1104 refresh_port_group(struct ofproto *p, unsigned int group)
1108 struct ofport *port;
1109 unsigned int port_no;
1111 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1113 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1115 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1116 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1117 ports[n_ports++] = port_no;
1120 dpif_port_group_set(p->dpif, group, ports, n_ports);
1127 refresh_port_groups(struct ofproto *p)
1129 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1130 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1132 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1136 static struct ofport *
1137 make_ofport(const struct odp_port *odp_port)
1139 enum netdev_flags flags;
1140 struct ofport *ofport;
1141 struct netdev *netdev;
1145 error = netdev_open(odp_port->devname, NETDEV_ETH_TYPE_NONE, &netdev);
1147 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1148 "cannot be opened (%s)",
1149 odp_port->devname, odp_port->port,
1150 odp_port->devname, strerror(error));
1154 ofport = xmalloc(sizeof *ofport);
1155 ofport->netdev = netdev;
1156 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1157 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1158 memcpy(ofport->opp.name, odp_port->devname,
1159 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1160 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1162 netdev_get_flags(netdev, &flags);
1163 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1165 netdev_get_carrier(netdev, &carrier);
1166 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1168 netdev_get_features(netdev,
1169 &ofport->opp.curr, &ofport->opp.advertised,
1170 &ofport->opp.supported, &ofport->opp.peer);
1175 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1177 if (port_array_get(&p->ports, odp_port->port)) {
1178 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1181 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1182 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1191 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1193 const struct ofp_phy_port *a = &a_->opp;
1194 const struct ofp_phy_port *b = &b_->opp;
1196 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1197 return (a->port_no == b->port_no
1198 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1199 && !strcmp((char *) a->name, (char *) b->name)
1200 && a->state == b->state
1201 && a->config == b->config
1202 && a->curr == b->curr
1203 && a->advertised == b->advertised
1204 && a->supported == b->supported
1205 && a->peer == b->peer);
1209 send_port_status(struct ofproto *p, const struct ofport *ofport,
1212 /* XXX Should limit the number of queued port status change messages. */
1213 struct ofconn *ofconn;
1214 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1215 struct ofp_port_status *ops;
1218 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1219 ops->reason = reason;
1220 ops->desc = ofport->opp;
1221 hton_ofp_phy_port(&ops->desc);
1222 queue_tx(b, ofconn, NULL);
1224 if (p->ofhooks->port_changed_cb) {
1225 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1230 ofport_install(struct ofproto *p, struct ofport *ofport)
1232 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1233 const char *netdev_name = (const char *) ofport->opp.name;
1235 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1236 port_array_set(&p->ports, odp_port, ofport);
1237 shash_add(&p->port_by_name, netdev_name, ofport);
1239 ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
1244 ofport_remove(struct ofproto *p, struct ofport *ofport)
1246 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1248 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1249 port_array_set(&p->ports, odp_port, NULL);
1250 shash_delete(&p->port_by_name,
1251 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1253 ofproto_sflow_del_port(p->sflow, odp_port);
1258 ofport_free(struct ofport *ofport)
1261 netdev_close(ofport->netdev);
1267 update_port(struct ofproto *p, const char *devname)
1269 struct odp_port odp_port;
1270 struct ofport *old_ofport;
1271 struct ofport *new_ofport;
1274 COVERAGE_INC(ofproto_update_port);
1276 /* Query the datapath for port information. */
1277 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1279 /* Find the old ofport. */
1280 old_ofport = shash_find_data(&p->port_by_name, devname);
1283 /* There's no port named 'devname' but there might be a port with
1284 * the same port number. This could happen if a port is deleted
1285 * and then a new one added in its place very quickly, or if a port
1286 * is renamed. In the former case we want to send an OFPPR_DELETE
1287 * and an OFPPR_ADD, and in the latter case we want to send a
1288 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1289 * the old port's ifindex against the new port, or perhaps less
1290 * reliably but more portably by comparing the old port's MAC
1291 * against the new port's MAC. However, this code isn't that smart
1292 * and always sends an OFPPR_MODIFY (XXX). */
1293 old_ofport = port_array_get(&p->ports, odp_port.port);
1295 } else if (error != ENOENT && error != ENODEV) {
1296 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1297 "%s", strerror(error));
1301 /* Create a new ofport. */
1302 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1304 /* Eliminate a few pathological cases. */
1305 if (!old_ofport && !new_ofport) {
1307 } else if (old_ofport && new_ofport) {
1308 /* Most of the 'config' bits are OpenFlow soft state, but
1309 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1310 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1311 * leaves the other bits 0.) */
1312 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1314 if (ofport_equal(old_ofport, new_ofport)) {
1315 /* False alarm--no change. */
1316 ofport_free(new_ofport);
1321 /* Now deal with the normal cases. */
1323 ofport_remove(p, old_ofport);
1326 ofport_install(p, new_ofport);
1328 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1329 (!old_ofport ? OFPPR_ADD
1330 : !new_ofport ? OFPPR_DELETE
1332 ofport_free(old_ofport);
1334 /* Update port groups. */
1335 refresh_port_groups(p);
1339 init_ports(struct ofproto *p)
1341 struct odp_port *ports;
1346 error = dpif_port_list(p->dpif, &ports, &n_ports);
1351 for (i = 0; i < n_ports; i++) {
1352 const struct odp_port *odp_port = &ports[i];
1353 if (!ofport_conflicts(p, odp_port)) {
1354 struct ofport *ofport = make_ofport(odp_port);
1356 ofport_install(p, ofport);
1361 refresh_port_groups(p);
1365 static struct ofconn *
1366 ofconn_create(struct ofproto *p, struct rconn *rconn)
1368 struct ofconn *ofconn = xmalloc(sizeof *ofconn);
1369 list_push_back(&p->all_conns, &ofconn->node);
1370 ofconn->rconn = rconn;
1371 ofconn->pktbuf = NULL;
1372 ofconn->send_flow_exp = false;
1373 ofconn->miss_send_len = 0;
1374 ofconn->packet_in_counter = rconn_packet_counter_create ();
1375 ofconn->reply_counter = rconn_packet_counter_create ();
1380 ofconn_destroy(struct ofconn *ofconn, struct ofproto *p)
1383 executer_rconn_closing(p->executer, ofconn->rconn);
1386 list_remove(&ofconn->node);
1387 rconn_destroy(ofconn->rconn);
1388 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1389 rconn_packet_counter_destroy(ofconn->reply_counter);
1390 pktbuf_destroy(ofconn->pktbuf);
1395 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1399 rconn_run(ofconn->rconn);
1401 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1402 /* Limit the number of iterations to prevent other tasks from
1404 for (iteration = 0; iteration < 50; iteration++) {
1405 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1410 fail_open_maybe_recover(p->fail_open);
1412 handle_openflow(ofconn, p, of_msg);
1413 ofpbuf_delete(of_msg);
1417 if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
1418 ofconn_destroy(ofconn, p);
1423 ofconn_wait(struct ofconn *ofconn)
1425 rconn_run_wait(ofconn->rconn);
1426 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1427 rconn_recv_wait(ofconn->rconn);
1429 COVERAGE_INC(ofproto_ofconn_stuck);
1433 /* Caller is responsible for initializing the 'cr' member of the returned
1435 static struct rule *
1436 rule_create(struct ofproto *ofproto, struct rule *super,
1437 const union ofp_action *actions, size_t n_actions,
1438 uint16_t idle_timeout, uint16_t hard_timeout)
1440 struct rule *rule = xcalloc(1, sizeof *rule);
1441 rule->idle_timeout = idle_timeout;
1442 rule->hard_timeout = hard_timeout;
1443 rule->used = rule->created = time_msec();
1444 rule->super = super;
1446 list_push_back(&super->list, &rule->list);
1448 list_init(&rule->list);
1450 rule->n_actions = n_actions;
1451 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1452 netflow_flow_clear(&rule->nf_flow);
1453 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1458 static struct rule *
1459 rule_from_cls_rule(const struct cls_rule *cls_rule)
1461 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1465 rule_free(struct rule *rule)
1467 free(rule->actions);
1468 free(rule->odp_actions);
1472 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1473 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1474 * through all of its subrules and revalidates them, destroying any that no
1475 * longer has a super-rule (which is probably all of them).
1477 * Before calling this function, the caller must make have removed 'rule' from
1478 * the classifier. If 'rule' is an exact-match rule, the caller is also
1479 * responsible for ensuring that it has been uninstalled from the datapath. */
1481 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1484 struct rule *subrule, *next;
1485 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1486 revalidate_rule(ofproto, subrule);
1489 list_remove(&rule->list);
1495 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1497 const union ofp_action *oa;
1498 struct actions_iterator i;
1500 if (out_port == htons(OFPP_NONE)) {
1503 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1504 oa = actions_next(&i)) {
1505 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1512 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1513 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1515 * The flow that 'packet' actually contains does not need to actually match
1516 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1517 * the packet and byte counters for 'rule' will be credited for the packet sent
1518 * out whether or not the packet actually matches 'rule'.
1520 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1521 * the caller must already have accurately composed ODP actions for it given
1522 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1523 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1524 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1525 * actions and apply them to 'packet'. */
1527 rule_execute(struct ofproto *ofproto, struct rule *rule,
1528 struct ofpbuf *packet, const flow_t *flow)
1530 const union odp_action *actions;
1532 struct odp_actions a;
1534 /* Grab or compose the ODP actions.
1536 * The special case for an exact-match 'rule' where 'flow' is not the
1537 * rule's flow is important to avoid, e.g., sending a packet out its input
1538 * port simply because the ODP actions were composed for the wrong
1540 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1541 struct rule *super = rule->super ? rule->super : rule;
1542 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1543 packet, &a, NULL, 0, NULL)) {
1546 actions = a.actions;
1547 n_actions = a.n_actions;
1549 actions = rule->odp_actions;
1550 n_actions = rule->n_odp_actions;
1553 /* Execute the ODP actions. */
1554 if (!dpif_execute(ofproto->dpif, flow->in_port,
1555 actions, n_actions, packet)) {
1556 struct odp_flow_stats stats;
1557 flow_extract_stats(flow, packet, &stats);
1558 update_stats(ofproto, rule, &stats);
1559 rule->used = time_msec();
1560 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1565 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1568 struct rule *displaced_rule;
1570 /* Insert the rule in the classifier. */
1571 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1572 if (!rule->cr.wc.wildcards) {
1573 rule_make_actions(p, rule, packet);
1576 /* Send the packet and credit it to the rule. */
1579 flow_extract(packet, in_port, &flow);
1580 rule_execute(p, rule, packet, &flow);
1583 /* Install the rule in the datapath only after sending the packet, to
1584 * avoid packet reordering. */
1585 if (rule->cr.wc.wildcards) {
1586 COVERAGE_INC(ofproto_add_wc_flow);
1587 p->need_revalidate = true;
1589 rule_install(p, rule, displaced_rule);
1592 /* Free the rule that was displaced, if any. */
1593 if (displaced_rule) {
1594 rule_destroy(p, displaced_rule);
1598 static struct rule *
1599 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1602 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1603 rule->idle_timeout, rule->hard_timeout);
1604 COVERAGE_INC(ofproto_subrule_create);
1605 cls_rule_from_flow(&subrule->cr, flow, 0,
1606 (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1607 : rule->cr.priority));
1608 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1614 rule_remove(struct ofproto *ofproto, struct rule *rule)
1616 if (rule->cr.wc.wildcards) {
1617 COVERAGE_INC(ofproto_del_wc_flow);
1618 ofproto->need_revalidate = true;
1620 rule_uninstall(ofproto, rule);
1622 classifier_remove(&ofproto->cls, &rule->cr);
1623 rule_destroy(ofproto, rule);
1626 /* Returns true if the actions changed, false otherwise. */
1628 rule_make_actions(struct ofproto *p, struct rule *rule,
1629 const struct ofpbuf *packet)
1631 const struct rule *super;
1632 struct odp_actions a;
1635 assert(!rule->cr.wc.wildcards);
1637 super = rule->super ? rule->super : rule;
1639 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1640 packet, &a, &rule->tags, &rule->may_install,
1641 &rule->nf_flow.output_iface);
1643 actions_len = a.n_actions * sizeof *a.actions;
1644 if (rule->n_odp_actions != a.n_actions
1645 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1646 COVERAGE_INC(ofproto_odp_unchanged);
1647 free(rule->odp_actions);
1648 rule->n_odp_actions = a.n_actions;
1649 rule->odp_actions = xmemdup(a.actions, actions_len);
1657 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1658 struct odp_flow_put *put)
1660 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1661 put->flow.key = rule->cr.flow;
1662 put->flow.actions = rule->odp_actions;
1663 put->flow.n_actions = rule->n_odp_actions;
1665 return dpif_flow_put(ofproto->dpif, put);
1669 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1671 assert(!rule->cr.wc.wildcards);
1673 if (rule->may_install) {
1674 struct odp_flow_put put;
1675 if (!do_put_flow(p, rule,
1676 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1678 rule->installed = true;
1679 if (displaced_rule) {
1680 update_stats(p, displaced_rule, &put.flow.stats);
1681 rule_post_uninstall(p, displaced_rule);
1684 } else if (displaced_rule) {
1685 rule_uninstall(p, displaced_rule);
1690 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1692 if (rule->installed) {
1693 struct odp_flow_put put;
1694 COVERAGE_INC(ofproto_dp_missed);
1695 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1697 rule_install(ofproto, rule, NULL);
1702 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1704 bool actions_changed;
1705 uint16_t new_out_iface, old_out_iface;
1707 old_out_iface = rule->nf_flow.output_iface;
1708 actions_changed = rule_make_actions(ofproto, rule, NULL);
1710 if (rule->may_install) {
1711 if (rule->installed) {
1712 if (actions_changed) {
1713 struct odp_flow_put put;
1714 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
1715 | ODPPF_ZERO_STATS, &put);
1716 update_stats(ofproto, rule, &put.flow.stats);
1718 /* Temporarily set the old output iface so that NetFlow
1719 * messages have the correct output interface for the old
1721 new_out_iface = rule->nf_flow.output_iface;
1722 rule->nf_flow.output_iface = old_out_iface;
1723 rule_post_uninstall(ofproto, rule);
1724 rule->nf_flow.output_iface = new_out_iface;
1727 rule_install(ofproto, rule, NULL);
1730 rule_uninstall(ofproto, rule);
1735 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1737 uint64_t total_bytes = rule->byte_count + extra_bytes;
1739 if (ofproto->ofhooks->account_flow_cb
1740 && total_bytes > rule->accounted_bytes)
1742 ofproto->ofhooks->account_flow_cb(
1743 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1744 total_bytes - rule->accounted_bytes, ofproto->aux);
1745 rule->accounted_bytes = total_bytes;
1750 rule_uninstall(struct ofproto *p, struct rule *rule)
1752 assert(!rule->cr.wc.wildcards);
1753 if (rule->installed) {
1754 struct odp_flow odp_flow;
1756 odp_flow.key = rule->cr.flow;
1757 odp_flow.actions = NULL;
1758 odp_flow.n_actions = 0;
1759 if (!dpif_flow_del(p->dpif, &odp_flow)) {
1760 update_stats(p, rule, &odp_flow.stats);
1762 rule->installed = false;
1764 rule_post_uninstall(p, rule);
1769 is_controller_rule(struct rule *rule)
1771 /* If the only action is send to the controller then don't report
1772 * NetFlow expiration messages since it is just part of the control
1773 * logic for the network and not real traffic. */
1775 if (rule && rule->super) {
1776 struct rule *super = rule->super;
1778 return super->n_actions == 1 &&
1779 super->actions[0].type == htons(OFPAT_OUTPUT) &&
1780 super->actions[0].output.port == htons(OFPP_CONTROLLER);
1787 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1789 struct rule *super = rule->super;
1791 rule_account(ofproto, rule, 0);
1793 if (ofproto->netflow && !is_controller_rule(rule)) {
1794 struct ofexpired expired;
1795 expired.flow = rule->cr.flow;
1796 expired.packet_count = rule->packet_count;
1797 expired.byte_count = rule->byte_count;
1798 expired.used = rule->used;
1799 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
1802 super->packet_count += rule->packet_count;
1803 super->byte_count += rule->byte_count;
1805 /* Reset counters to prevent double counting if the rule ever gets
1807 rule->packet_count = 0;
1808 rule->byte_count = 0;
1809 rule->accounted_bytes = 0;
1811 netflow_flow_clear(&rule->nf_flow);
1816 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1817 struct rconn_packet_counter *counter)
1819 update_openflow_length(msg);
1820 if (rconn_send(ofconn->rconn, msg, counter)) {
1826 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1827 int error, const void *data, size_t len)
1830 struct ofp_error_msg *oem;
1832 if (!(error >> 16)) {
1833 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1838 COVERAGE_INC(ofproto_error);
1839 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1840 oh ? oh->xid : 0, &buf);
1841 oem->type = htons((unsigned int) error >> 16);
1842 oem->code = htons(error & 0xffff);
1843 memcpy(oem->data, data, len);
1844 queue_tx(buf, ofconn, ofconn->reply_counter);
1848 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1851 size_t oh_length = ntohs(oh->length);
1852 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1856 hton_ofp_phy_port(struct ofp_phy_port *opp)
1858 opp->port_no = htons(opp->port_no);
1859 opp->config = htonl(opp->config);
1860 opp->state = htonl(opp->state);
1861 opp->curr = htonl(opp->curr);
1862 opp->advertised = htonl(opp->advertised);
1863 opp->supported = htonl(opp->supported);
1864 opp->peer = htonl(opp->peer);
1868 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
1870 struct ofp_header *rq = oh;
1871 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
1876 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
1877 struct ofp_header *oh)
1879 struct ofp_switch_features *osf;
1881 unsigned int port_no;
1882 struct ofport *port;
1884 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1885 osf->datapath_id = htonll(p->datapath_id);
1886 osf->n_buffers = htonl(pktbuf_capacity());
1888 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1889 OFPC_PORT_STATS | OFPC_MULTI_PHY_TX);
1890 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1891 (1u << OFPAT_SET_VLAN_VID) |
1892 (1u << OFPAT_SET_VLAN_PCP) |
1893 (1u << OFPAT_STRIP_VLAN) |
1894 (1u << OFPAT_SET_DL_SRC) |
1895 (1u << OFPAT_SET_DL_DST) |
1896 (1u << OFPAT_SET_NW_SRC) |
1897 (1u << OFPAT_SET_NW_DST) |
1898 (1u << OFPAT_SET_TP_SRC) |
1899 (1u << OFPAT_SET_TP_DST));
1901 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1902 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1905 queue_tx(buf, ofconn, ofconn->reply_counter);
1910 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1911 struct ofp_header *oh)
1914 struct ofp_switch_config *osc;
1918 /* Figure out flags. */
1919 dpif_get_drop_frags(p->dpif, &drop_frags);
1920 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1921 if (ofconn->send_flow_exp) {
1922 flags |= OFPC_SEND_FLOW_EXP;
1926 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1927 osc->flags = htons(flags);
1928 osc->miss_send_len = htons(ofconn->miss_send_len);
1929 queue_tx(buf, ofconn, ofconn->reply_counter);
1935 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1936 struct ofp_switch_config *osc)
1941 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1945 flags = ntohs(osc->flags);
1947 ofconn->send_flow_exp = (flags & OFPC_SEND_FLOW_EXP) != 0;
1949 if (ofconn == p->controller) {
1950 switch (flags & OFPC_FRAG_MASK) {
1951 case OFPC_FRAG_NORMAL:
1952 dpif_set_drop_frags(p->dpif, false);
1954 case OFPC_FRAG_DROP:
1955 dpif_set_drop_frags(p->dpif, true);
1958 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1964 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1965 if (ntohs(osc->miss_send_len) != 0) {
1966 ofconn->pktbuf = pktbuf_create();
1968 pktbuf_destroy(ofconn->pktbuf);
1972 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1978 add_output_group_action(struct odp_actions *actions, uint16_t group,
1979 uint16_t *nf_output_iface)
1981 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1983 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
1984 *nf_output_iface = NF_OUT_FLOOD;
1989 add_controller_action(struct odp_actions *actions,
1990 const struct ofp_action_output *oao)
1992 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
1993 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
1996 struct action_xlate_ctx {
1998 const flow_t *flow; /* Flow to which these actions correspond. */
1999 int recurse; /* Recursion level, via xlate_table_action. */
2000 struct ofproto *ofproto;
2001 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2002 * null pointer if we are revalidating
2003 * without a packet to refer to. */
2006 struct odp_actions *out; /* Datapath actions. */
2007 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2008 bool may_set_up_flow; /* True ordinarily; false if the actions must
2009 * be reassessed for every packet. */
2010 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2013 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2014 struct action_xlate_ctx *ctx);
2017 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2019 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
2022 if (ofport->opp.config & OFPPC_NO_FWD) {
2023 /* Forwarding disabled on port. */
2028 * We don't have an ofport record for this port, but it doesn't hurt to
2029 * allow forwarding to it anyhow. Maybe such a port will appear later
2030 * and we're pre-populating the flow table.
2034 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2035 ctx->nf_output_iface = port;
2038 static struct rule *
2039 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2042 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2044 /* The rule we found might not be valid, since we could be in need of
2045 * revalidation. If it is not valid, don't return it. */
2048 && ofproto->need_revalidate
2049 && !revalidate_rule(ofproto, rule)) {
2050 COVERAGE_INC(ofproto_invalidated);
2058 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2060 if (!ctx->recurse) {
2065 flow.in_port = in_port;
2067 rule = lookup_valid_rule(ctx->ofproto, &flow);
2074 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2081 xlate_output_action(struct action_xlate_ctx *ctx,
2082 const struct ofp_action_output *oao)
2085 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2087 ctx->nf_output_iface = NF_OUT_DROP;
2089 switch (ntohs(oao->port)) {
2091 add_output_action(ctx, ctx->flow->in_port);
2094 xlate_table_action(ctx, ctx->flow->in_port);
2097 if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet,
2098 ctx->out, ctx->tags,
2099 &ctx->nf_output_iface,
2100 ctx->ofproto->aux)) {
2101 COVERAGE_INC(ofproto_uninstallable);
2102 ctx->may_set_up_flow = false;
2106 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2107 &ctx->nf_output_iface);
2110 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2112 case OFPP_CONTROLLER:
2113 add_controller_action(ctx->out, oao);
2116 add_output_action(ctx, ODPP_LOCAL);
2119 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2120 if (odp_port != ctx->flow->in_port) {
2121 add_output_action(ctx, odp_port);
2126 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2127 ctx->nf_output_iface = NF_OUT_FLOOD;
2128 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2129 ctx->nf_output_iface = prev_nf_output_iface;
2130 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2131 ctx->nf_output_iface != NF_OUT_FLOOD) {
2132 ctx->nf_output_iface = NF_OUT_MULTI;
2137 xlate_nicira_action(struct action_xlate_ctx *ctx,
2138 const struct nx_action_header *nah)
2140 const struct nx_action_resubmit *nar;
2141 int subtype = ntohs(nah->subtype);
2143 assert(nah->vendor == htonl(NX_VENDOR_ID));
2145 case NXAST_RESUBMIT:
2146 nar = (const struct nx_action_resubmit *) nah;
2147 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2151 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2157 do_xlate_actions(const union ofp_action *in, size_t n_in,
2158 struct action_xlate_ctx *ctx)
2160 struct actions_iterator iter;
2161 const union ofp_action *ia;
2162 const struct ofport *port;
2164 port = port_array_get(&ctx->ofproto->ports, ctx->flow->in_port);
2165 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2166 port->opp.config & (eth_addr_equals(ctx->flow->dl_dst, stp_eth_addr)
2167 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2168 /* Drop this flow. */
2172 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2173 uint16_t type = ntohs(ia->type);
2174 union odp_action *oa;
2178 xlate_output_action(ctx, &ia->output);
2181 case OFPAT_SET_VLAN_VID:
2182 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2183 oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2186 case OFPAT_SET_VLAN_PCP:
2187 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2188 oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2191 case OFPAT_STRIP_VLAN:
2192 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2195 case OFPAT_SET_DL_SRC:
2196 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2197 memcpy(oa->dl_addr.dl_addr,
2198 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2201 case OFPAT_SET_DL_DST:
2202 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2203 memcpy(oa->dl_addr.dl_addr,
2204 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2207 case OFPAT_SET_NW_SRC:
2208 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2209 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2212 case OFPAT_SET_NW_DST:
2213 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2214 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2217 case OFPAT_SET_TP_SRC:
2218 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2219 oa->tp_port.tp_port = ia->tp_port.tp_port;
2222 case OFPAT_SET_TP_DST:
2223 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2224 oa->tp_port.tp_port = ia->tp_port.tp_port;
2228 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2232 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2239 xlate_actions(const union ofp_action *in, size_t n_in,
2240 const flow_t *flow, struct ofproto *ofproto,
2241 const struct ofpbuf *packet,
2242 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2243 uint16_t *nf_output_iface)
2245 tag_type no_tags = 0;
2246 struct action_xlate_ctx ctx;
2247 COVERAGE_INC(ofproto_ofp2odp);
2248 odp_actions_init(out);
2251 ctx.ofproto = ofproto;
2252 ctx.packet = packet;
2254 ctx.tags = tags ? tags : &no_tags;
2255 ctx.may_set_up_flow = true;
2256 ctx.nf_output_iface = NF_OUT_DROP;
2257 do_xlate_actions(in, n_in, &ctx);
2259 /* Check with in-band control to see if we're allowed to set up this
2261 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2262 ctx.may_set_up_flow = false;
2265 if (may_set_up_flow) {
2266 *may_set_up_flow = ctx.may_set_up_flow;
2268 if (nf_output_iface) {
2269 *nf_output_iface = ctx.nf_output_iface;
2271 if (odp_actions_overflow(out)) {
2272 odp_actions_init(out);
2273 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2279 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2280 struct ofp_header *oh)
2282 struct ofp_packet_out *opo;
2283 struct ofpbuf payload, *buffer;
2284 struct odp_actions actions;
2290 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2294 opo = (struct ofp_packet_out *) oh;
2296 COVERAGE_INC(ofproto_packet_out);
2297 if (opo->buffer_id != htonl(UINT32_MAX)) {
2298 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2300 if (error || !buffer) {
2308 flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2309 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2310 &flow, p, &payload, &actions, NULL, NULL, NULL);
2315 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2317 ofpbuf_delete(buffer);
2323 update_port_config(struct ofproto *p, struct ofport *port,
2324 uint32_t config, uint32_t mask)
2326 mask &= config ^ port->opp.config;
2327 if (mask & OFPPC_PORT_DOWN) {
2328 if (config & OFPPC_PORT_DOWN) {
2329 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2331 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2334 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2335 if (mask & REVALIDATE_BITS) {
2336 COVERAGE_INC(ofproto_costly_flags);
2337 port->opp.config ^= mask & REVALIDATE_BITS;
2338 p->need_revalidate = true;
2340 #undef REVALIDATE_BITS
2341 if (mask & OFPPC_NO_FLOOD) {
2342 port->opp.config ^= OFPPC_NO_FLOOD;
2343 refresh_port_groups(p);
2345 if (mask & OFPPC_NO_PACKET_IN) {
2346 port->opp.config ^= OFPPC_NO_PACKET_IN;
2351 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2353 const struct ofp_port_mod *opm;
2354 struct ofport *port;
2357 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2361 opm = (struct ofp_port_mod *) oh;
2363 port = port_array_get(&p->ports,
2364 ofp_port_to_odp_port(ntohs(opm->port_no)));
2366 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2367 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2368 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2370 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2371 if (opm->advertise) {
2372 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2378 static struct ofpbuf *
2379 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2381 struct ofp_stats_reply *osr;
2384 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2385 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2387 osr->flags = htons(0);
2391 static struct ofpbuf *
2392 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2394 return make_stats_reply(request->header.xid, request->type, body_len);
2398 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2400 struct ofpbuf *msg = *msgp;
2401 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2402 if (nbytes + msg->size > UINT16_MAX) {
2403 struct ofp_stats_reply *reply = msg->data;
2404 reply->flags = htons(OFPSF_REPLY_MORE);
2405 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2406 queue_tx(msg, ofconn, ofconn->reply_counter);
2408 return ofpbuf_put_uninit(*msgp, nbytes);
2412 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2413 struct ofp_stats_request *request)
2415 struct ofp_desc_stats *ods;
2418 msg = start_stats_reply(request, sizeof *ods);
2419 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2420 strncpy(ods->mfr_desc, p->manufacturer, sizeof ods->mfr_desc);
2421 strncpy(ods->hw_desc, p->hardware, sizeof ods->hw_desc);
2422 strncpy(ods->sw_desc, p->software, sizeof ods->sw_desc);
2423 strncpy(ods->serial_num, p->serial, sizeof ods->serial_num);
2424 queue_tx(msg, ofconn, ofconn->reply_counter);
2430 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2432 struct rule *rule = rule_from_cls_rule(cls_rule);
2433 int *n_subrules = n_subrules_;
2441 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2442 struct ofp_stats_request *request)
2444 struct ofp_table_stats *ots;
2446 struct odp_stats dpstats;
2447 int n_exact, n_subrules, n_wild;
2449 msg = start_stats_reply(request, sizeof *ots * 2);
2451 /* Count rules of various kinds. */
2453 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2454 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2455 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2458 dpif_get_dp_stats(p->dpif, &dpstats);
2459 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2460 memset(ots, 0, sizeof *ots);
2461 ots->table_id = TABLEID_HASH;
2462 strcpy(ots->name, "hash");
2463 ots->wildcards = htonl(0);
2464 ots->max_entries = htonl(dpstats.max_capacity);
2465 ots->active_count = htonl(n_exact);
2466 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2468 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2470 /* Classifier table. */
2471 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2472 memset(ots, 0, sizeof *ots);
2473 ots->table_id = TABLEID_CLASSIFIER;
2474 strcpy(ots->name, "classifier");
2475 ots->wildcards = htonl(OFPFW_ALL);
2476 ots->max_entries = htonl(65536);
2477 ots->active_count = htonl(n_wild);
2478 ots->lookup_count = htonll(0); /* XXX */
2479 ots->matched_count = htonll(0); /* XXX */
2481 queue_tx(msg, ofconn, ofconn->reply_counter);
2486 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2487 struct ofp_stats_request *request)
2489 struct ofp_port_stats *ops;
2491 struct ofport *port;
2492 unsigned int port_no;
2494 msg = start_stats_reply(request, sizeof *ops * 16);
2495 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2496 struct netdev_stats stats;
2498 /* Intentionally ignore return value, since errors will set 'stats' to
2499 * all-1s, which is correct for OpenFlow, and netdev_get_stats() will
2501 netdev_get_stats(port->netdev, &stats);
2503 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2504 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2505 memset(ops->pad, 0, sizeof ops->pad);
2506 ops->rx_packets = htonll(stats.rx_packets);
2507 ops->tx_packets = htonll(stats.tx_packets);
2508 ops->rx_bytes = htonll(stats.rx_bytes);
2509 ops->tx_bytes = htonll(stats.tx_bytes);
2510 ops->rx_dropped = htonll(stats.rx_dropped);
2511 ops->tx_dropped = htonll(stats.tx_dropped);
2512 ops->rx_errors = htonll(stats.rx_errors);
2513 ops->tx_errors = htonll(stats.tx_errors);
2514 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2515 ops->rx_over_err = htonll(stats.rx_over_errors);
2516 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2517 ops->collisions = htonll(stats.collisions);
2520 queue_tx(msg, ofconn, ofconn->reply_counter);
2524 struct flow_stats_cbdata {
2525 struct ofproto *ofproto;
2526 struct ofconn *ofconn;
2532 query_stats(struct ofproto *p, struct rule *rule,
2533 uint64_t *packet_countp, uint64_t *byte_countp)
2535 uint64_t packet_count, byte_count;
2536 struct rule *subrule;
2537 struct odp_flow *odp_flows;
2540 packet_count = rule->packet_count;
2541 byte_count = rule->byte_count;
2543 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2544 odp_flows = xcalloc(1, n_odp_flows * sizeof *odp_flows);
2545 if (rule->cr.wc.wildcards) {
2547 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2548 odp_flows[i++].key = subrule->cr.flow;
2549 packet_count += subrule->packet_count;
2550 byte_count += subrule->byte_count;
2553 odp_flows[0].key = rule->cr.flow;
2556 packet_count = rule->packet_count;
2557 byte_count = rule->byte_count;
2558 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2560 for (i = 0; i < n_odp_flows; i++) {
2561 struct odp_flow *odp_flow = &odp_flows[i];
2562 packet_count += odp_flow->stats.n_packets;
2563 byte_count += odp_flow->stats.n_bytes;
2568 *packet_countp = packet_count;
2569 *byte_countp = byte_count;
2573 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2575 struct rule *rule = rule_from_cls_rule(rule_);
2576 struct flow_stats_cbdata *cbdata = cbdata_;
2577 struct ofp_flow_stats *ofs;
2578 uint64_t packet_count, byte_count;
2579 size_t act_len, len;
2581 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2585 act_len = sizeof *rule->actions * rule->n_actions;
2586 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2588 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2590 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2591 ofs->length = htons(len);
2592 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2594 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match);
2595 ofs->duration = htonl((time_msec() - rule->created) / 1000);
2596 ofs->priority = htons(rule->cr.priority);
2597 ofs->idle_timeout = htons(rule->idle_timeout);
2598 ofs->hard_timeout = htons(rule->hard_timeout);
2599 memset(ofs->pad2, 0, sizeof ofs->pad2);
2600 ofs->packet_count = htonll(packet_count);
2601 ofs->byte_count = htonll(byte_count);
2602 memcpy(ofs->actions, rule->actions, act_len);
2606 table_id_to_include(uint8_t table_id)
2608 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2609 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2610 : table_id == 0xff ? CLS_INC_ALL
2615 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2616 const struct ofp_stats_request *osr,
2619 struct ofp_flow_stats_request *fsr;
2620 struct flow_stats_cbdata cbdata;
2621 struct cls_rule target;
2623 if (arg_size != sizeof *fsr) {
2624 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2626 fsr = (struct ofp_flow_stats_request *) osr->body;
2628 COVERAGE_INC(ofproto_flows_req);
2630 cbdata.ofconn = ofconn;
2631 cbdata.out_port = fsr->out_port;
2632 cbdata.msg = start_stats_reply(osr, 1024);
2633 cls_rule_from_match(&target, &fsr->match, 0);
2634 classifier_for_each_match(&p->cls, &target,
2635 table_id_to_include(fsr->table_id),
2636 flow_stats_cb, &cbdata);
2637 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2641 struct flow_stats_ds_cbdata {
2642 struct ofproto *ofproto;
2647 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
2649 struct rule *rule = rule_from_cls_rule(rule_);
2650 struct flow_stats_ds_cbdata *cbdata = cbdata_;
2651 struct ds *results = cbdata->results;
2652 struct ofp_match match;
2653 uint64_t packet_count, byte_count;
2654 size_t act_len = sizeof *rule->actions * rule->n_actions;
2656 /* Don't report on subrules. */
2657 if (rule->super != NULL) {
2661 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2662 flow_to_ovs_match(&rule->cr.flow, rule->cr.wc.wildcards, &match);
2664 ds_put_format(results, "duration=%llds, ",
2665 (time_msec() - rule->created) / 1000);
2666 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2667 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2668 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2669 ofp_print_match(results, &match, true);
2670 ofp_print_actions(results, &rule->actions->header, act_len);
2671 ds_put_cstr(results, "\n");
2674 /* Adds a pretty-printed description of all flows to 'results', including
2675 * those marked hidden by secchan (e.g., by in-band control). */
2677 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2679 struct ofp_match match;
2680 struct cls_rule target;
2681 struct flow_stats_ds_cbdata cbdata;
2683 memset(&match, 0, sizeof match);
2684 match.wildcards = htonl(OFPFW_ALL);
2687 cbdata.results = results;
2689 cls_rule_from_match(&target, &match, 0);
2690 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2691 flow_stats_ds_cb, &cbdata);
2694 struct aggregate_stats_cbdata {
2695 struct ofproto *ofproto;
2697 uint64_t packet_count;
2698 uint64_t byte_count;
2703 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2705 struct rule *rule = rule_from_cls_rule(rule_);
2706 struct aggregate_stats_cbdata *cbdata = cbdata_;
2707 uint64_t packet_count, byte_count;
2709 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2713 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2715 cbdata->packet_count += packet_count;
2716 cbdata->byte_count += byte_count;
2721 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2722 const struct ofp_stats_request *osr,
2725 struct ofp_aggregate_stats_request *asr;
2726 struct ofp_aggregate_stats_reply *reply;
2727 struct aggregate_stats_cbdata cbdata;
2728 struct cls_rule target;
2731 if (arg_size != sizeof *asr) {
2732 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2734 asr = (struct ofp_aggregate_stats_request *) osr->body;
2736 COVERAGE_INC(ofproto_agg_request);
2738 cbdata.out_port = asr->out_port;
2739 cbdata.packet_count = 0;
2740 cbdata.byte_count = 0;
2742 cls_rule_from_match(&target, &asr->match, 0);
2743 classifier_for_each_match(&p->cls, &target,
2744 table_id_to_include(asr->table_id),
2745 aggregate_stats_cb, &cbdata);
2747 msg = start_stats_reply(osr, sizeof *reply);
2748 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2749 reply->flow_count = htonl(cbdata.n_flows);
2750 reply->packet_count = htonll(cbdata.packet_count);
2751 reply->byte_count = htonll(cbdata.byte_count);
2752 queue_tx(msg, ofconn, ofconn->reply_counter);
2757 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2758 struct ofp_header *oh)
2760 struct ofp_stats_request *osr;
2764 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2769 osr = (struct ofp_stats_request *) oh;
2771 switch (ntohs(osr->type)) {
2773 return handle_desc_stats_request(p, ofconn, osr);
2776 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2778 case OFPST_AGGREGATE:
2779 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2782 return handle_table_stats_request(p, ofconn, osr);
2785 return handle_port_stats_request(p, ofconn, osr);
2788 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2791 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2795 static long long int
2796 msec_from_nsec(uint64_t sec, uint32_t nsec)
2798 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2802 update_time(struct ofproto *ofproto, struct rule *rule,
2803 const struct odp_flow_stats *stats)
2805 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2806 if (used > rule->used) {
2808 if (rule->super && used > rule->super->used) {
2809 rule->super->used = used;
2811 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
2816 update_stats(struct ofproto *ofproto, struct rule *rule,
2817 const struct odp_flow_stats *stats)
2819 if (stats->n_packets) {
2820 update_time(ofproto, rule, stats);
2821 rule->packet_count += stats->n_packets;
2822 rule->byte_count += stats->n_bytes;
2823 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
2829 add_flow(struct ofproto *p, struct ofconn *ofconn,
2830 struct ofp_flow_mod *ofm, size_t n_actions)
2832 struct ofpbuf *packet;
2837 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
2838 n_actions, ntohs(ofm->idle_timeout),
2839 ntohs(ofm->hard_timeout));
2840 cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority));
2844 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2845 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2849 rule_insert(p, rule, packet, in_port);
2850 ofpbuf_delete(packet);
2855 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
2856 size_t n_actions, uint16_t command, struct rule *rule)
2858 if (rule_is_hidden(rule)) {
2862 if (command == OFPFC_DELETE) {
2863 rule_remove(p, rule);
2865 size_t actions_len = n_actions * sizeof *rule->actions;
2867 if (n_actions == rule->n_actions
2868 && !memcmp(ofm->actions, rule->actions, actions_len))
2873 free(rule->actions);
2874 rule->actions = xmemdup(ofm->actions, actions_len);
2875 rule->n_actions = n_actions;
2877 if (rule->cr.wc.wildcards) {
2878 COVERAGE_INC(ofproto_mod_wc_flow);
2879 p->need_revalidate = true;
2881 rule_update_actions(p, rule);
2889 modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm,
2890 size_t n_actions, uint16_t command)
2896 flow_from_match(&flow, &wildcards, &ofm->match);
2897 rule = rule_from_cls_rule(classifier_find_rule_exactly(
2898 &p->cls, &flow, wildcards,
2899 ntohs(ofm->priority)));
2902 if (command == OFPFC_DELETE
2903 && ofm->out_port != htons(OFPP_NONE)
2904 && !rule_has_out_port(rule, ofm->out_port)) {
2908 modify_flow(p, ofm, n_actions, command, rule);
2913 struct modify_flows_cbdata {
2914 struct ofproto *ofproto;
2915 const struct ofp_flow_mod *ofm;
2922 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
2924 struct rule *rule = rule_from_cls_rule(rule_);
2925 struct modify_flows_cbdata *cbdata = cbdata_;
2927 if (cbdata->out_port != htons(OFPP_NONE)
2928 && !rule_has_out_port(rule, cbdata->out_port)) {
2932 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions,
2933 cbdata->command, rule);
2937 modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm,
2938 size_t n_actions, uint16_t command)
2940 struct modify_flows_cbdata cbdata;
2941 struct cls_rule target;
2945 cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port
2946 : htons(OFPP_NONE));
2947 cbdata.n_actions = n_actions;
2948 cbdata.command = command;
2950 cls_rule_from_match(&target, &ofm->match, 0);
2952 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2953 modify_flows_cb, &cbdata);
2958 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
2959 struct ofp_flow_mod *ofm)
2964 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
2965 sizeof *ofm->actions, &n_actions);
2970 normalize_match(&ofm->match);
2971 if (!ofm->match.wildcards) {
2972 ofm->priority = htons(UINT16_MAX);
2975 error = validate_actions((const union ofp_action *) ofm->actions,
2976 n_actions, p->max_ports);
2981 switch (ntohs(ofm->command)) {
2983 return add_flow(p, ofconn, ofm, n_actions);
2986 return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY);
2988 case OFPFC_MODIFY_STRICT:
2989 return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY);
2992 return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE);
2994 case OFPFC_DELETE_STRICT:
2995 return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE);
2998 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3003 send_capability_reply(struct ofproto *p, struct ofconn *ofconn, uint32_t xid)
3005 struct ofmp_capability_reply *ocr;
3007 char capabilities[] = "com.nicira.mgmt.manager=false\n";
3009 ocr = make_openflow_xid(sizeof(*ocr), OFPT_VENDOR, xid, &b);
3010 ocr->header.header.vendor = htonl(NX_VENDOR_ID);
3011 ocr->header.header.subtype = htonl(NXT_MGMT);
3012 ocr->header.type = htons(OFMPT_CAPABILITY_REPLY);
3014 ocr->format = htonl(OFMPCOF_SIMPLE);
3015 ocr->mgmt_id = htonll(p->mgmt_id);
3017 ofpbuf_put(b, capabilities, strlen(capabilities));
3019 queue_tx(b, ofconn, ofconn->reply_counter);
3023 handle_ofmp(struct ofproto *p, struct ofconn *ofconn,
3024 struct ofmp_header *ofmph)
3026 size_t msg_len = ntohs(ofmph->header.header.length);
3027 if (msg_len < sizeof(*ofmph)) {
3028 VLOG_WARN_RL(&rl, "dropping short managment message: %zu\n", msg_len);
3029 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
3032 if (ofmph->type == htons(OFMPT_CAPABILITY_REQUEST)) {
3033 struct ofmp_capability_request *ofmpcr;
3035 if (msg_len < sizeof(struct ofmp_capability_request)) {
3036 VLOG_WARN_RL(&rl, "dropping short capability request: %zu\n",
3038 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
3041 ofmpcr = (struct ofmp_capability_request *)ofmph;
3042 if (ofmpcr->format != htonl(OFMPCAF_SIMPLE)) {
3043 /* xxx Find a better type than bad subtype */
3044 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3047 send_capability_reply(p, ofconn, ofmph->header.header.xid);
3050 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3055 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3057 struct ofp_vendor_header *ovh = msg;
3058 struct nicira_header *nh;
3060 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3061 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
3063 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3064 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3066 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3067 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
3071 switch (ntohl(nh->subtype)) {
3072 case NXT_STATUS_REQUEST:
3073 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3076 case NXT_ACT_SET_CONFIG:
3077 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
3079 case NXT_ACT_GET_CONFIG:
3080 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
3082 case NXT_COMMAND_REQUEST:
3084 return executer_handle_request(p->executer, ofconn->rconn, msg);
3089 return handle_ofmp(p, ofconn, msg);
3092 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3096 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3097 struct ofpbuf *ofp_msg)
3099 struct ofp_header *oh = ofp_msg->data;
3102 COVERAGE_INC(ofproto_recv_openflow);
3104 case OFPT_ECHO_REQUEST:
3105 error = handle_echo_request(ofconn, oh);
3108 case OFPT_ECHO_REPLY:
3112 case OFPT_FEATURES_REQUEST:
3113 error = handle_features_request(p, ofconn, oh);
3116 case OFPT_GET_CONFIG_REQUEST:
3117 error = handle_get_config_request(p, ofconn, oh);
3120 case OFPT_SET_CONFIG:
3121 error = handle_set_config(p, ofconn, ofp_msg->data);
3124 case OFPT_PACKET_OUT:
3125 error = handle_packet_out(p, ofconn, ofp_msg->data);
3129 error = handle_port_mod(p, oh);
3133 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3136 case OFPT_STATS_REQUEST:
3137 error = handle_stats_request(p, ofconn, oh);
3141 error = handle_vendor(p, ofconn, ofp_msg->data);
3145 if (VLOG_IS_WARN_ENABLED()) {
3146 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3147 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3150 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3155 send_error_oh(ofconn, ofp_msg->data, error);
3160 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
3162 struct odp_msg *msg = packet->data;
3163 uint16_t in_port = odp_port_to_ofp_port(msg->port);
3165 struct ofpbuf payload;
3168 payload.data = msg + 1;
3169 payload.size = msg->length - sizeof *msg;
3170 flow_extract(&payload, msg->port, &flow);
3172 /* Check with in-band control to see if this packet should be sent
3173 * to the local port regardless of the flow table. */
3174 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3175 union odp_action action;
3177 memset(&action, 0, sizeof(action));
3178 action.output.type = ODPAT_OUTPUT;
3179 action.output.port = ODPP_LOCAL;
3180 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3183 rule = lookup_valid_rule(p, &flow);
3185 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3186 struct ofport *port = port_array_get(&p->ports, msg->port);
3188 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3189 COVERAGE_INC(ofproto_no_packet_in);
3190 /* XXX install 'drop' flow entry */
3191 ofpbuf_delete(packet);
3195 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3198 COVERAGE_INC(ofproto_packet_in);
3199 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3203 if (rule->cr.wc.wildcards) {
3204 rule = rule_create_subrule(p, rule, &flow);
3205 rule_make_actions(p, rule, packet);
3207 if (!rule->may_install) {
3208 /* The rule is not installable, that is, we need to process every
3209 * packet, so process the current packet and set its actions into
3211 rule_make_actions(p, rule, packet);
3213 /* XXX revalidate rule if it needs it */
3217 rule_execute(p, rule, &payload, &flow);
3218 rule_reinstall(p, rule);
3220 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY
3221 && rconn_is_connected(p->controller->rconn)) {
3223 * Extra-special case for fail-open mode.
3225 * We are in fail-open mode and the packet matched the fail-open rule,
3226 * but we are connected to a controller too. We should send the packet
3227 * up to the controller in the hope that it will try to set up a flow
3228 * and thereby allow us to exit fail-open.
3230 * See the top-level comment in fail-open.c for more information.
3232 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3234 ofpbuf_delete(packet);
3239 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3241 struct odp_msg *msg = packet->data;
3243 switch (msg->type) {
3244 case _ODPL_ACTION_NR:
3245 COVERAGE_INC(ofproto_ctlr_action);
3246 pinsched_send(p->action_sched, odp_port_to_ofp_port(msg->port), packet,
3247 send_packet_in_action, p);
3250 case _ODPL_SFLOW_NR:
3252 ofproto_sflow_received(p->sflow, msg);
3254 ofpbuf_delete(packet);
3258 handle_odp_miss_msg(p, packet);
3262 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
3269 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3271 struct rule *sub = rule_from_cls_rule(sub_);
3272 struct revalidate_cbdata *cbdata = cbdata_;
3274 if (cbdata->revalidate_all
3275 || (cbdata->revalidate_subrules && sub->super)
3276 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3277 revalidate_rule(cbdata->ofproto, sub);
3282 revalidate_rule(struct ofproto *p, struct rule *rule)
3284 const flow_t *flow = &rule->cr.flow;
3286 COVERAGE_INC(ofproto_revalidate_rule);
3289 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3291 rule_remove(p, rule);
3293 } else if (super != rule->super) {
3294 COVERAGE_INC(ofproto_revalidate_moved);
3295 list_remove(&rule->list);
3296 list_push_back(&super->list, &rule->list);
3297 rule->super = super;
3298 rule->hard_timeout = super->hard_timeout;
3299 rule->idle_timeout = super->idle_timeout;
3300 rule->created = super->created;
3305 rule_update_actions(p, rule);
3309 static struct ofpbuf *
3310 compose_flow_exp(const struct rule *rule, long long int now, uint8_t reason)
3312 struct ofp_flow_expired *ofe;
3315 ofe = make_openflow(sizeof *ofe, OFPT_FLOW_EXPIRED, &buf);
3316 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofe->match);
3317 ofe->priority = htons(rule->cr.priority);
3318 ofe->reason = reason;
3319 ofe->duration = htonl((now - rule->created) / 1000);
3320 ofe->packet_count = htonll(rule->packet_count);
3321 ofe->byte_count = htonll(rule->byte_count);
3327 send_flow_exp(struct ofproto *p, struct rule *rule,
3328 long long int now, uint8_t reason)
3330 struct ofconn *ofconn;
3331 struct ofconn *prev;
3332 struct ofpbuf *buf = NULL;
3334 /* We limit the maximum number of queued flow expirations it by accounting
3335 * them under the counter for replies. That works because preventing
3336 * OpenFlow requests from being processed also prevents new flows from
3337 * being added (and expiring). (It also prevents processing OpenFlow
3338 * requests that would not add new flows, so it is imperfect.) */
3341 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3342 if (ofconn->send_flow_exp && rconn_is_connected(ofconn->rconn)) {
3344 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3346 buf = compose_flow_exp(rule, now, reason);
3352 queue_tx(buf, prev, prev->reply_counter);
3357 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3359 assert(rule->installed);
3360 assert(!rule->cr.wc.wildcards);
3363 rule_remove(ofproto, rule);
3365 rule_uninstall(ofproto, rule);
3370 expire_rule(struct cls_rule *cls_rule, void *p_)
3372 struct ofproto *p = p_;
3373 struct rule *rule = rule_from_cls_rule(cls_rule);
3374 long long int hard_expire, idle_expire, expire, now;
3376 hard_expire = (rule->hard_timeout
3377 ? rule->created + rule->hard_timeout * 1000
3379 idle_expire = (rule->idle_timeout
3380 && (rule->super || list_is_empty(&rule->list))
3381 ? rule->used + rule->idle_timeout * 1000
3383 expire = MIN(hard_expire, idle_expire);
3387 if (rule->installed && now >= rule->used + 5000) {
3388 uninstall_idle_flow(p, rule);
3389 } else if (!rule->cr.wc.wildcards) {
3390 active_timeout(p, rule);
3396 COVERAGE_INC(ofproto_expired);
3398 /* Update stats. This code will be a no-op if the rule expired
3399 * due to an idle timeout. */
3400 if (rule->cr.wc.wildcards) {
3401 struct rule *subrule, *next;
3402 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3403 rule_remove(p, subrule);
3406 rule_uninstall(p, rule);
3409 if (!rule_is_hidden(rule)) {
3410 send_flow_exp(p, rule, now,
3412 ? OFPER_HARD_TIMEOUT : OFPER_IDLE_TIMEOUT));
3414 rule_remove(p, rule);
3418 active_timeout(struct ofproto *ofproto, struct rule *rule)
3420 if (ofproto->netflow && !is_controller_rule(rule) &&
3421 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
3422 struct ofexpired expired;
3423 struct odp_flow odp_flow;
3425 /* Get updated flow stats. */
3426 memset(&odp_flow, 0, sizeof odp_flow);
3427 if (rule->installed) {
3428 odp_flow.key = rule->cr.flow;
3429 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
3430 dpif_flow_get(ofproto->dpif, &odp_flow);
3432 if (odp_flow.stats.n_packets) {
3433 update_time(ofproto, rule, &odp_flow.stats);
3434 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
3435 odp_flow.stats.tcp_flags);
3439 expired.flow = rule->cr.flow;
3440 expired.packet_count = rule->packet_count +
3441 odp_flow.stats.n_packets;
3442 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
3443 expired.used = rule->used;
3445 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
3447 /* Schedule us to send the accumulated records once we have
3448 * collected all of them. */
3449 poll_immediate_wake();
3454 update_used(struct ofproto *p)
3456 struct odp_flow *flows;
3461 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
3466 for (i = 0; i < n_flows; i++) {
3467 struct odp_flow *f = &flows[i];
3470 rule = rule_from_cls_rule(
3471 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3472 if (!rule || !rule->installed) {
3473 COVERAGE_INC(ofproto_unexpected_rule);
3474 dpif_flow_del(p->dpif, f);
3478 update_time(p, rule, &f->stats);
3479 rule_account(p, rule, f->stats.n_bytes);
3485 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3486 const struct ofpbuf *packet, int send_len)
3488 struct odp_msg *msg = packet->data;
3489 struct ofpbuf payload;
3493 /* Extract packet payload from 'msg'. */
3494 payload.data = msg + 1;
3495 payload.size = msg->length - sizeof *msg;
3497 /* Construct ofp_packet_in message. */
3498 reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3499 opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port), reason,
3500 &payload, send_len);
3503 rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100);
3507 send_packet_in_action(struct ofpbuf *packet, void *p_)
3509 struct ofproto *p = p_;
3510 struct ofconn *ofconn;
3511 struct odp_msg *msg;
3514 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3515 if (ofconn == p->controller || ofconn->miss_send_len) {
3516 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3519 ofpbuf_delete(packet);
3523 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3525 struct ofproto *p = p_;
3526 bool in_fail_open = p->fail_open && fail_open_is_active(p->fail_open);
3527 struct ofconn *ofconn;
3528 struct ofpbuf payload;
3529 struct odp_msg *msg;
3532 payload.data = msg + 1;
3533 payload.size = msg->length - sizeof *msg;
3534 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3535 if (ofconn->miss_send_len) {
3536 struct pktbuf *pb = ofconn->pktbuf;
3537 uint32_t buffer_id = (in_fail_open
3539 : pktbuf_save(pb, &payload, msg->port));
3540 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3542 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3545 ofpbuf_delete(packet);
3549 pick_datapath_id(const struct ofproto *ofproto)
3551 const struct ofport *port;
3553 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
3555 uint8_t ea[ETH_ADDR_LEN];
3558 error = netdev_get_etheraddr(port->netdev, ea);
3560 return eth_addr_to_uint64(ea);
3562 VLOG_WARN("could not get MAC address for %s (%s)",
3563 netdev_get_name(port->netdev), strerror(error));
3565 return ofproto->fallback_dpid;
3569 pick_fallback_dpid(void)
3571 uint8_t ea[ETH_ADDR_LEN];
3572 eth_addr_random(ea);
3573 ea[0] = 0x00; /* Set Nicira OUI. */
3576 return eth_addr_to_uint64(ea);
3580 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3581 struct odp_actions *actions, tag_type *tags,
3582 uint16_t *nf_output_iface, void *ofproto_)
3584 struct ofproto *ofproto = ofproto_;
3587 /* Drop frames for reserved multicast addresses. */
3588 if (eth_addr_is_reserved(flow->dl_dst)) {
3592 /* Learn source MAC (but don't try to learn from revalidation). */
3593 if (packet != NULL) {
3594 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3597 /* The log messages here could actually be useful in debugging,
3598 * so keep the rate limit relatively high. */
3599 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3600 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3601 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3602 ofproto_revalidate(ofproto, rev_tag);
3606 /* Determine output port. */
3607 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3609 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
3610 } else if (out_port != flow->in_port) {
3611 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3612 *nf_output_iface = out_port;
3620 static const struct ofhooks default_ofhooks = {
3622 default_normal_ofhook_cb,