2 * Copyright (c) 2009, 2010 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
22 #include <netinet/in.h>
25 #include "classifier.h"
27 #include "discovery.h"
29 #include "dynamic-string.h"
31 #include "fail-open.h"
33 #include "mac-learning.h"
37 #include "ofp-print.h"
39 #include "openflow/nicira-ext.h"
40 #include "openflow/openflow.h"
41 #include "openflow/openflow-mgmt.h"
42 #include "openvswitch/datapath-protocol.h"
46 #include "poll-loop.h"
47 #include "port-array.h"
57 #include "vconn-ssl.h"
60 #define THIS_MODULE VLM_ofproto
70 TABLEID_CLASSIFIER = 1
74 struct netdev *netdev;
75 struct ofp_phy_port opp; /* In host byte order. */
78 static void ofport_free(struct ofport *);
79 static void hton_ofp_phy_port(struct ofp_phy_port *);
81 static int xlate_actions(const union ofp_action *in, size_t n_in,
82 const flow_t *flow, struct ofproto *ofproto,
83 const struct ofpbuf *packet,
84 struct odp_actions *out, tag_type *tags,
85 bool *may_set_up_flow, uint16_t *nf_output_iface);
90 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. */
197 char *dp_desc; /* Datapath description. */
201 struct netdev_monitor *netdev_monitor;
202 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
204 struct shash port_by_name;
208 struct switch_status *switch_status;
209 struct status_category *ss_cat;
210 struct in_band *in_band;
211 struct discovery *discovery;
212 struct fail_open *fail_open;
213 struct pinsched *miss_sched, *action_sched;
214 struct executer *executer;
215 struct netflow *netflow;
218 struct classifier cls;
219 bool need_revalidate;
220 long long int next_expiration;
221 struct tag_set revalidate_set;
223 /* OpenFlow connections. */
224 struct list all_conns;
225 struct ofconn *controller;
226 struct pvconn **listeners;
228 struct pvconn **snoops;
231 /* Hooks for ovs-vswitchd. */
232 const struct ofhooks *ofhooks;
235 /* Used by default ofhooks. */
236 struct mac_learning *ml;
239 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
241 static const struct ofhooks default_ofhooks;
243 static uint64_t pick_datapath_id(const struct ofproto *);
244 static uint64_t pick_fallback_dpid(void);
245 static void send_packet_in_miss(struct ofpbuf *, void *ofproto);
246 static void send_packet_in_action(struct ofpbuf *, void *ofproto);
247 static void update_used(struct ofproto *);
248 static void update_stats(struct ofproto *, struct rule *,
249 const struct odp_flow_stats *);
250 static void expire_rule(struct cls_rule *, void *ofproto);
251 static void active_timeout(struct ofproto *ofproto, struct rule *rule);
252 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
253 static void revalidate_cb(struct cls_rule *rule_, void *p_);
255 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
257 static void handle_openflow(struct ofconn *, struct ofproto *,
260 static void refresh_port_group(struct ofproto *, unsigned int group);
261 static void update_port(struct ofproto *, const char *devname);
262 static int init_ports(struct ofproto *);
263 static void reinit_ports(struct ofproto *);
266 ofproto_create(const char *datapath, const struct ofhooks *ofhooks, void *aux,
267 struct ofproto **ofprotop)
269 struct odp_stats stats;
276 /* Connect to datapath and start listening for messages. */
277 error = dpif_open(datapath, &dpif);
279 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
282 error = dpif_get_dp_stats(dpif, &stats);
284 VLOG_ERR("failed to obtain stats for datapath %s: %s",
285 datapath, strerror(error));
289 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION);
291 VLOG_ERR("failed to listen on datapath %s: %s",
292 datapath, strerror(error));
296 dpif_flow_flush(dpif);
297 dpif_recv_purge(dpif);
299 /* Initialize settings. */
300 p = xcalloc(1, sizeof *p);
301 p->fallback_dpid = pick_fallback_dpid();
302 p->datapath_id = p->fallback_dpid;
303 p->manufacturer = xstrdup("Nicira Networks, Inc.");
304 p->hardware = xstrdup("Reference Implementation");
305 p->software = xstrdup(VERSION BUILDNR);
306 p->serial = xstrdup("None");
307 p->dp_desc = xstrdup("None");
309 /* Initialize datapath. */
311 p->netdev_monitor = netdev_monitor_create();
312 port_array_init(&p->ports);
313 shash_init(&p->port_by_name);
314 p->max_ports = stats.max_ports;
316 /* Initialize submodules. */
317 p->switch_status = switch_status_create(p);
321 p->miss_sched = p->action_sched = NULL;
325 /* Initialize flow table. */
326 classifier_init(&p->cls);
327 p->need_revalidate = false;
328 p->next_expiration = time_msec() + 1000;
329 tag_set_init(&p->revalidate_set);
331 /* Initialize OpenFlow connections. */
332 list_init(&p->all_conns);
333 p->controller = ofconn_create(p, rconn_create(5, 8));
334 p->controller->pktbuf = pktbuf_create();
335 p->controller->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
341 /* Initialize hooks. */
343 p->ofhooks = ofhooks;
347 p->ofhooks = &default_ofhooks;
349 p->ml = mac_learning_create();
352 /* Register switch status category. */
353 p->ss_cat = switch_status_register(p->switch_status, "remote",
354 rconn_status_cb, p->controller->rconn);
357 error = init_ports(p);
363 /* Pick final datapath ID. */
364 p->datapath_id = pick_datapath_id(p);
365 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
372 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
374 uint64_t old_dpid = p->datapath_id;
375 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
376 if (p->datapath_id != old_dpid) {
377 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
378 rconn_reconnect(p->controller->rconn);
383 ofproto_set_mgmt_id(struct ofproto *p, uint64_t mgmt_id)
385 p->mgmt_id = mgmt_id;
389 ofproto_set_probe_interval(struct ofproto *p, int probe_interval)
391 probe_interval = probe_interval ? MAX(probe_interval, 5) : 0;
392 rconn_set_probe_interval(p->controller->rconn, probe_interval);
394 int trigger_duration = probe_interval ? probe_interval * 3 : 15;
395 fail_open_set_trigger_duration(p->fail_open, trigger_duration);
400 ofproto_set_max_backoff(struct ofproto *p, int max_backoff)
402 rconn_set_max_backoff(p->controller->rconn, max_backoff);
406 ofproto_set_desc(struct ofproto *p,
407 const char *manufacturer, const char *hardware,
408 const char *software, const char *serial,
412 free(p->manufacturer);
413 p->manufacturer = xstrdup(manufacturer);
417 p->hardware = xstrdup(hardware);
421 p->software = xstrdup(software);
425 p->serial = xstrdup(serial);
429 p->dp_desc = xstrdup(dp_desc);
434 ofproto_set_in_band(struct ofproto *p, bool in_band)
436 if (in_band != (p->in_band != NULL)) {
438 return in_band_create(p, p->dpif, p->switch_status,
439 p->controller->rconn, &p->in_band);
441 ofproto_set_discovery(p, false, NULL, true);
442 in_band_destroy(p->in_band);
445 rconn_reconnect(p->controller->rconn);
451 ofproto_set_discovery(struct ofproto *p, bool discovery,
452 const char *re, bool update_resolv_conf)
454 if (discovery != (p->discovery != NULL)) {
456 int error = ofproto_set_in_band(p, true);
460 error = discovery_create(re, update_resolv_conf,
461 p->dpif, p->switch_status,
467 discovery_destroy(p->discovery);
470 rconn_disconnect(p->controller->rconn);
471 } else if (discovery) {
472 discovery_set_update_resolv_conf(p->discovery, update_resolv_conf);
473 return discovery_set_accept_controller_re(p->discovery, re);
479 ofproto_set_controller(struct ofproto *ofproto, const char *controller)
481 if (ofproto->discovery) {
483 } else if (controller) {
484 if (strcmp(rconn_get_name(ofproto->controller->rconn), controller)) {
485 return rconn_connect(ofproto->controller->rconn, controller);
490 rconn_disconnect(ofproto->controller->rconn);
496 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
497 const struct svec *svec)
499 struct pvconn **pvconns = *pvconnsp;
500 size_t n_pvconns = *n_pvconnsp;
504 for (i = 0; i < n_pvconns; i++) {
505 pvconn_close(pvconns[i]);
509 pvconns = xmalloc(svec->n * sizeof *pvconns);
511 for (i = 0; i < svec->n; i++) {
512 const char *name = svec->names[i];
513 struct pvconn *pvconn;
516 error = pvconn_open(name, &pvconn);
518 pvconns[n_pvconns++] = pvconn;
520 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
528 *n_pvconnsp = n_pvconns;
534 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
536 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
540 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
542 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
546 ofproto_set_netflow(struct ofproto *ofproto,
547 const struct netflow_options *nf_options)
549 if (nf_options->collectors.n) {
550 if (!ofproto->netflow) {
551 ofproto->netflow = netflow_create();
553 return netflow_set_options(ofproto->netflow, nf_options);
555 netflow_destroy(ofproto->netflow);
556 ofproto->netflow = NULL;
562 ofproto_set_failure(struct ofproto *ofproto, bool fail_open)
565 struct rconn *rconn = ofproto->controller->rconn;
566 int trigger_duration = rconn_get_probe_interval(rconn) * 3;
567 if (!ofproto->fail_open) {
568 ofproto->fail_open = fail_open_create(ofproto, trigger_duration,
569 ofproto->switch_status,
572 fail_open_set_trigger_duration(ofproto->fail_open,
576 fail_open_destroy(ofproto->fail_open);
577 ofproto->fail_open = NULL;
582 ofproto_set_rate_limit(struct ofproto *ofproto,
583 int rate_limit, int burst_limit)
585 if (rate_limit > 0) {
586 if (!ofproto->miss_sched) {
587 ofproto->miss_sched = pinsched_create(rate_limit, burst_limit,
588 ofproto->switch_status);
589 ofproto->action_sched = pinsched_create(rate_limit, burst_limit,
592 pinsched_set_limits(ofproto->miss_sched, rate_limit, burst_limit);
593 pinsched_set_limits(ofproto->action_sched,
594 rate_limit, burst_limit);
597 pinsched_destroy(ofproto->miss_sched);
598 ofproto->miss_sched = NULL;
599 pinsched_destroy(ofproto->action_sched);
600 ofproto->action_sched = NULL;
605 ofproto_set_stp(struct ofproto *ofproto UNUSED, bool enable_stp)
609 VLOG_WARN("STP is not yet implemented");
617 ofproto_set_remote_execution(struct ofproto *ofproto, const char *command_acl,
618 const char *command_dir)
621 if (!ofproto->executer) {
622 return executer_create(command_acl, command_dir,
625 executer_set_acl(ofproto->executer, command_acl, command_dir);
628 executer_destroy(ofproto->executer);
629 ofproto->executer = NULL;
635 ofproto_get_datapath_id(const struct ofproto *ofproto)
637 return ofproto->datapath_id;
641 ofproto_get_mgmt_id(const struct ofproto *ofproto)
643 return ofproto->mgmt_id;
647 ofproto_get_probe_interval(const struct ofproto *ofproto)
649 return rconn_get_probe_interval(ofproto->controller->rconn);
653 ofproto_get_max_backoff(const struct ofproto *ofproto)
655 return rconn_get_max_backoff(ofproto->controller->rconn);
659 ofproto_get_in_band(const struct ofproto *ofproto)
661 return ofproto->in_band != NULL;
665 ofproto_get_discovery(const struct ofproto *ofproto)
667 return ofproto->discovery != NULL;
671 ofproto_get_controller(const struct ofproto *ofproto)
673 return rconn_get_name(ofproto->controller->rconn);
677 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
681 for (i = 0; i < ofproto->n_listeners; i++) {
682 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
687 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
691 for (i = 0; i < ofproto->n_snoops; i++) {
692 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
697 ofproto_destroy(struct ofproto *p)
699 struct ofconn *ofconn, *next_ofconn;
700 struct ofport *ofport;
701 unsigned int port_no;
708 ofproto_flush_flows(p);
709 classifier_destroy(&p->cls);
711 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
713 ofconn_destroy(ofconn, p);
717 netdev_monitor_destroy(p->netdev_monitor);
718 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
721 shash_destroy(&p->port_by_name);
723 switch_status_destroy(p->switch_status);
724 in_band_destroy(p->in_band);
725 discovery_destroy(p->discovery);
726 fail_open_destroy(p->fail_open);
727 pinsched_destroy(p->miss_sched);
728 pinsched_destroy(p->action_sched);
729 executer_destroy(p->executer);
730 netflow_destroy(p->netflow);
732 switch_status_unregister(p->ss_cat);
734 for (i = 0; i < p->n_listeners; i++) {
735 pvconn_close(p->listeners[i]);
739 for (i = 0; i < p->n_snoops; i++) {
740 pvconn_close(p->snoops[i]);
744 mac_learning_destroy(p->ml);
750 ofproto_run(struct ofproto *p)
752 int error = ofproto_run1(p);
754 error = ofproto_run2(p, false);
760 process_port_change(struct ofproto *ofproto, int error, char *devname)
762 if (error == ENOBUFS) {
763 reinit_ports(ofproto);
765 update_port(ofproto, devname);
771 ofproto_run1(struct ofproto *p)
773 struct ofconn *ofconn, *next_ofconn;
778 for (i = 0; i < 50; i++) {
782 error = dpif_recv(p->dpif, &buf);
784 if (error == ENODEV) {
785 /* Someone destroyed the datapath behind our back. The caller
786 * better destroy us and give up, because we're just going to
787 * spin from here on out. */
788 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
789 VLOG_ERR_RL(&rl, "%s: datapath was destroyed externally",
796 handle_odp_msg(p, buf);
799 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
800 process_port_change(p, error, devname);
802 while ((error = netdev_monitor_poll(p->netdev_monitor,
803 &devname)) != EAGAIN) {
804 process_port_change(p, error, devname);
808 in_band_run(p->in_band);
811 char *controller_name;
812 if (rconn_is_connectivity_questionable(p->controller->rconn)) {
813 discovery_question_connectivity(p->discovery);
815 if (discovery_run(p->discovery, &controller_name)) {
816 if (controller_name) {
817 rconn_connect(p->controller->rconn, controller_name);
819 rconn_disconnect(p->controller->rconn);
823 pinsched_run(p->miss_sched, send_packet_in_miss, p);
824 pinsched_run(p->action_sched, send_packet_in_action, p);
826 executer_run(p->executer);
829 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
831 ofconn_run(ofconn, p);
834 /* Fail-open maintenance. Do this after processing the ofconns since
835 * fail-open checks the status of the controller rconn. */
837 fail_open_run(p->fail_open);
840 for (i = 0; i < p->n_listeners; i++) {
844 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
846 ofconn_create(p, rconn_new_from_vconn("passive", vconn));
847 } else if (retval != EAGAIN) {
848 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
852 for (i = 0; i < p->n_snoops; i++) {
856 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
858 rconn_add_monitor(p->controller->rconn, vconn);
859 } else if (retval != EAGAIN) {
860 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
864 if (time_msec() >= p->next_expiration) {
865 COVERAGE_INC(ofproto_expiration);
866 p->next_expiration = time_msec() + 1000;
869 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
871 /* Let the hook know that we're at a stable point: all outstanding data
872 * in existing flows has been accounted to the account_cb. Thus, the
873 * hook can now reasonably do operations that depend on having accurate
874 * flow volume accounting (currently, that's just bond rebalancing). */
875 if (p->ofhooks->account_checkpoint_cb) {
876 p->ofhooks->account_checkpoint_cb(p->aux);
881 netflow_run(p->netflow);
887 struct revalidate_cbdata {
888 struct ofproto *ofproto;
889 bool revalidate_all; /* Revalidate all exact-match rules? */
890 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
891 struct tag_set revalidate_set; /* Set of tags to revalidate. */
895 ofproto_run2(struct ofproto *p, bool revalidate_all)
897 if (p->need_revalidate || revalidate_all
898 || !tag_set_is_empty(&p->revalidate_set)) {
899 struct revalidate_cbdata cbdata;
901 cbdata.revalidate_all = revalidate_all;
902 cbdata.revalidate_subrules = p->need_revalidate;
903 cbdata.revalidate_set = p->revalidate_set;
904 tag_set_init(&p->revalidate_set);
905 COVERAGE_INC(ofproto_revalidate);
906 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
907 p->need_revalidate = false;
914 ofproto_wait(struct ofproto *p)
916 struct ofconn *ofconn;
919 dpif_recv_wait(p->dpif);
920 dpif_port_poll_wait(p->dpif);
921 netdev_monitor_poll_wait(p->netdev_monitor);
922 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
926 in_band_wait(p->in_band);
929 discovery_wait(p->discovery);
932 fail_open_wait(p->fail_open);
934 pinsched_wait(p->miss_sched);
935 pinsched_wait(p->action_sched);
937 executer_wait(p->executer);
939 if (!tag_set_is_empty(&p->revalidate_set)) {
940 poll_immediate_wake();
942 if (p->need_revalidate) {
943 /* Shouldn't happen, but if it does just go around again. */
944 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
945 poll_immediate_wake();
946 } else if (p->next_expiration != LLONG_MAX) {
947 poll_timer_wait(p->next_expiration - time_msec());
949 for (i = 0; i < p->n_listeners; i++) {
950 pvconn_wait(p->listeners[i]);
952 for (i = 0; i < p->n_snoops; i++) {
953 pvconn_wait(p->snoops[i]);
958 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
960 tag_set_add(&ofproto->revalidate_set, tag);
964 ofproto_get_revalidate_set(struct ofproto *ofproto)
966 return &ofproto->revalidate_set;
970 ofproto_is_alive(const struct ofproto *p)
972 return p->discovery || rconn_is_alive(p->controller->rconn);
976 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
977 const union ofp_action *actions, size_t n_actions,
978 const struct ofpbuf *packet)
980 struct odp_actions odp_actions;
983 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
989 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
991 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
992 odp_actions.n_actions, packet);
997 ofproto_add_flow(struct ofproto *p,
998 const flow_t *flow, uint32_t wildcards, unsigned int priority,
999 const union ofp_action *actions, size_t n_actions,
1003 rule = rule_create(p, NULL, actions, n_actions,
1004 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1006 cls_rule_from_flow(&rule->cr, flow, wildcards, priority);
1007 rule_insert(p, rule, NULL, 0);
1011 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1012 uint32_t wildcards, unsigned int priority)
1016 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1020 rule_remove(ofproto, rule);
1025 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1027 struct rule *rule = rule_from_cls_rule(rule_);
1028 struct ofproto *ofproto = ofproto_;
1030 /* Mark the flow as not installed, even though it might really be
1031 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1032 * There is no point in uninstalling it individually since we are about to
1033 * blow away all the flows with dpif_flow_flush(). */
1034 rule->installed = false;
1036 rule_remove(ofproto, rule);
1040 ofproto_flush_flows(struct ofproto *ofproto)
1042 COVERAGE_INC(ofproto_flush);
1043 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1044 dpif_flow_flush(ofproto->dpif);
1045 if (ofproto->in_band) {
1046 in_band_flushed(ofproto->in_band);
1048 if (ofproto->fail_open) {
1049 fail_open_flushed(ofproto->fail_open);
1054 reinit_ports(struct ofproto *p)
1056 struct svec devnames;
1057 struct ofport *ofport;
1058 unsigned int port_no;
1059 struct odp_port *odp_ports;
1063 svec_init(&devnames);
1064 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1065 svec_add (&devnames, (char *) ofport->opp.name);
1067 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1068 for (i = 0; i < n_odp_ports; i++) {
1069 svec_add (&devnames, odp_ports[i].devname);
1073 svec_sort_unique(&devnames);
1074 for (i = 0; i < devnames.n; i++) {
1075 update_port(p, devnames.names[i]);
1077 svec_destroy(&devnames);
1081 refresh_port_group(struct ofproto *p, unsigned int group)
1085 struct ofport *port;
1086 unsigned int port_no;
1088 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1090 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1092 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1093 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1094 ports[n_ports++] = port_no;
1097 dpif_port_group_set(p->dpif, group, ports, n_ports);
1102 refresh_port_groups(struct ofproto *p)
1104 refresh_port_group(p, DP_GROUP_FLOOD);
1105 refresh_port_group(p, DP_GROUP_ALL);
1108 static struct ofport *
1109 make_ofport(const struct odp_port *odp_port)
1111 enum netdev_flags flags;
1112 struct ofport *ofport;
1113 struct netdev *netdev;
1117 error = netdev_open(odp_port->devname, NETDEV_ETH_TYPE_NONE, &netdev);
1119 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1120 "cannot be opened (%s)",
1121 odp_port->devname, odp_port->port,
1122 odp_port->devname, strerror(error));
1126 ofport = xmalloc(sizeof *ofport);
1127 ofport->netdev = netdev;
1128 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1129 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1130 memcpy(ofport->opp.name, odp_port->devname,
1131 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1132 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1134 netdev_get_flags(netdev, &flags);
1135 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1137 netdev_get_carrier(netdev, &carrier);
1138 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1140 netdev_get_features(netdev,
1141 &ofport->opp.curr, &ofport->opp.advertised,
1142 &ofport->opp.supported, &ofport->opp.peer);
1147 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1149 if (port_array_get(&p->ports, odp_port->port)) {
1150 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1153 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1154 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1163 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1165 const struct ofp_phy_port *a = &a_->opp;
1166 const struct ofp_phy_port *b = &b_->opp;
1168 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1169 return (a->port_no == b->port_no
1170 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1171 && !strcmp((char *) a->name, (char *) b->name)
1172 && a->state == b->state
1173 && a->config == b->config
1174 && a->curr == b->curr
1175 && a->advertised == b->advertised
1176 && a->supported == b->supported
1177 && a->peer == b->peer);
1181 send_port_status(struct ofproto *p, const struct ofport *ofport,
1184 /* XXX Should limit the number of queued port status change messages. */
1185 struct ofconn *ofconn;
1186 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1187 struct ofp_port_status *ops;
1190 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1191 ops->reason = reason;
1192 ops->desc = ofport->opp;
1193 hton_ofp_phy_port(&ops->desc);
1194 queue_tx(b, ofconn, NULL);
1196 if (p->ofhooks->port_changed_cb) {
1197 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1202 ofport_install(struct ofproto *p, struct ofport *ofport)
1204 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1205 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no),
1207 shash_add(&p->port_by_name, (char *) ofport->opp.name, ofport);
1211 ofport_remove(struct ofproto *p, struct ofport *ofport)
1213 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1214 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no), NULL);
1215 shash_delete(&p->port_by_name,
1216 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1220 ofport_free(struct ofport *ofport)
1223 netdev_close(ofport->netdev);
1229 update_port(struct ofproto *p, const char *devname)
1231 struct odp_port odp_port;
1232 struct ofport *old_ofport;
1233 struct ofport *new_ofport;
1236 COVERAGE_INC(ofproto_update_port);
1238 /* Query the datapath for port information. */
1239 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1241 /* Find the old ofport. */
1242 old_ofport = shash_find_data(&p->port_by_name, devname);
1245 /* There's no port named 'devname' but there might be a port with
1246 * the same port number. This could happen if a port is deleted
1247 * and then a new one added in its place very quickly, or if a port
1248 * is renamed. In the former case we want to send an OFPPR_DELETE
1249 * and an OFPPR_ADD, and in the latter case we want to send a
1250 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1251 * the old port's ifindex against the new port, or perhaps less
1252 * reliably but more portably by comparing the old port's MAC
1253 * against the new port's MAC. However, this code isn't that smart
1254 * and always sends an OFPPR_MODIFY (XXX). */
1255 old_ofport = port_array_get(&p->ports, odp_port.port);
1257 } else if (error != ENOENT && error != ENODEV) {
1258 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1259 "%s", strerror(error));
1263 /* Create a new ofport. */
1264 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1266 /* Eliminate a few pathological cases. */
1267 if (!old_ofport && !new_ofport) {
1269 } else if (old_ofport && new_ofport) {
1270 /* Most of the 'config' bits are OpenFlow soft state, but
1271 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1272 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1273 * leaves the other bits 0.) */
1274 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1276 if (ofport_equal(old_ofport, new_ofport)) {
1277 /* False alarm--no change. */
1278 ofport_free(new_ofport);
1283 /* Now deal with the normal cases. */
1285 ofport_remove(p, old_ofport);
1288 ofport_install(p, new_ofport);
1290 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1291 (!old_ofport ? OFPPR_ADD
1292 : !new_ofport ? OFPPR_DELETE
1294 ofport_free(old_ofport);
1296 /* Update port groups. */
1297 refresh_port_groups(p);
1301 init_ports(struct ofproto *p)
1303 struct odp_port *ports;
1308 error = dpif_port_list(p->dpif, &ports, &n_ports);
1313 for (i = 0; i < n_ports; i++) {
1314 const struct odp_port *odp_port = &ports[i];
1315 if (!ofport_conflicts(p, odp_port)) {
1316 struct ofport *ofport = make_ofport(odp_port);
1318 ofport_install(p, ofport);
1323 refresh_port_groups(p);
1327 static struct ofconn *
1328 ofconn_create(struct ofproto *p, struct rconn *rconn)
1330 struct ofconn *ofconn = xmalloc(sizeof *ofconn);
1331 list_push_back(&p->all_conns, &ofconn->node);
1332 ofconn->rconn = rconn;
1333 ofconn->pktbuf = NULL;
1334 ofconn->miss_send_len = 0;
1335 ofconn->packet_in_counter = rconn_packet_counter_create ();
1336 ofconn->reply_counter = rconn_packet_counter_create ();
1341 ofconn_destroy(struct ofconn *ofconn, struct ofproto *p)
1344 executer_rconn_closing(p->executer, ofconn->rconn);
1347 list_remove(&ofconn->node);
1348 rconn_destroy(ofconn->rconn);
1349 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1350 rconn_packet_counter_destroy(ofconn->reply_counter);
1351 pktbuf_destroy(ofconn->pktbuf);
1356 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1360 rconn_run(ofconn->rconn);
1362 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1363 /* Limit the number of iterations to prevent other tasks from
1365 for (iteration = 0; iteration < 50; iteration++) {
1366 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1371 fail_open_maybe_recover(p->fail_open);
1373 handle_openflow(ofconn, p, of_msg);
1374 ofpbuf_delete(of_msg);
1378 if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
1379 ofconn_destroy(ofconn, p);
1384 ofconn_wait(struct ofconn *ofconn)
1386 rconn_run_wait(ofconn->rconn);
1387 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1388 rconn_recv_wait(ofconn->rconn);
1390 COVERAGE_INC(ofproto_ofconn_stuck);
1394 /* Caller is responsible for initializing the 'cr' member of the returned
1396 static struct rule *
1397 rule_create(struct ofproto *ofproto, struct rule *super,
1398 const union ofp_action *actions, size_t n_actions,
1399 uint16_t idle_timeout, uint16_t hard_timeout,
1400 bool send_flow_removed)
1402 struct rule *rule = xcalloc(1, sizeof *rule);
1403 rule->idle_timeout = idle_timeout;
1404 rule->hard_timeout = hard_timeout;
1405 rule->used = rule->created = time_msec();
1406 rule->send_flow_removed = send_flow_removed;
1407 rule->super = super;
1409 list_push_back(&super->list, &rule->list);
1411 list_init(&rule->list);
1413 rule->n_actions = n_actions;
1414 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1415 netflow_flow_clear(&rule->nf_flow);
1416 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1421 static struct rule *
1422 rule_from_cls_rule(const struct cls_rule *cls_rule)
1424 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1428 rule_free(struct rule *rule)
1430 free(rule->actions);
1431 free(rule->odp_actions);
1435 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1436 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1437 * through all of its subrules and revalidates them, destroying any that no
1438 * longer has a super-rule (which is probably all of them).
1440 * Before calling this function, the caller must make have removed 'rule' from
1441 * the classifier. If 'rule' is an exact-match rule, the caller is also
1442 * responsible for ensuring that it has been uninstalled from the datapath. */
1444 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1447 struct rule *subrule, *next;
1448 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1449 revalidate_rule(ofproto, subrule);
1452 list_remove(&rule->list);
1458 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1460 const union ofp_action *oa;
1461 struct actions_iterator i;
1463 if (out_port == htons(OFPP_NONE)) {
1466 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1467 oa = actions_next(&i)) {
1468 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1475 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1476 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1478 * The flow that 'packet' actually contains does not need to actually match
1479 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1480 * the packet and byte counters for 'rule' will be credited for the packet sent
1481 * out whether or not the packet actually matches 'rule'.
1483 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1484 * the caller must already have accurately composed ODP actions for it given
1485 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1486 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1487 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1488 * actions and apply them to 'packet'. */
1490 rule_execute(struct ofproto *ofproto, struct rule *rule,
1491 struct ofpbuf *packet, const flow_t *flow)
1493 const union odp_action *actions;
1495 struct odp_actions a;
1497 /* Grab or compose the ODP actions.
1499 * The special case for an exact-match 'rule' where 'flow' is not the
1500 * rule's flow is important to avoid, e.g., sending a packet out its input
1501 * port simply because the ODP actions were composed for the wrong
1503 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1504 struct rule *super = rule->super ? rule->super : rule;
1505 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1506 packet, &a, NULL, 0, NULL)) {
1509 actions = a.actions;
1510 n_actions = a.n_actions;
1512 actions = rule->odp_actions;
1513 n_actions = rule->n_odp_actions;
1516 /* Execute the ODP actions. */
1517 if (!dpif_execute(ofproto->dpif, flow->in_port,
1518 actions, n_actions, packet)) {
1519 struct odp_flow_stats stats;
1520 flow_extract_stats(flow, packet, &stats);
1521 update_stats(ofproto, rule, &stats);
1522 rule->used = time_msec();
1523 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1528 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1531 struct rule *displaced_rule;
1533 /* Insert the rule in the classifier. */
1534 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1535 if (!rule->cr.wc.wildcards) {
1536 rule_make_actions(p, rule, packet);
1539 /* Send the packet and credit it to the rule. */
1542 flow_extract(packet, in_port, &flow);
1543 rule_execute(p, rule, packet, &flow);
1546 /* Install the rule in the datapath only after sending the packet, to
1547 * avoid packet reordering. */
1548 if (rule->cr.wc.wildcards) {
1549 COVERAGE_INC(ofproto_add_wc_flow);
1550 p->need_revalidate = true;
1552 rule_install(p, rule, displaced_rule);
1555 /* Free the rule that was displaced, if any. */
1556 if (displaced_rule) {
1557 rule_destroy(p, displaced_rule);
1561 static struct rule *
1562 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1565 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1566 rule->idle_timeout, rule->hard_timeout,
1568 COVERAGE_INC(ofproto_subrule_create);
1569 cls_rule_from_flow(&subrule->cr, flow, 0,
1570 (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1571 : rule->cr.priority));
1572 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1578 rule_remove(struct ofproto *ofproto, struct rule *rule)
1580 if (rule->cr.wc.wildcards) {
1581 COVERAGE_INC(ofproto_del_wc_flow);
1582 ofproto->need_revalidate = true;
1584 rule_uninstall(ofproto, rule);
1586 classifier_remove(&ofproto->cls, &rule->cr);
1587 rule_destroy(ofproto, rule);
1590 /* Returns true if the actions changed, false otherwise. */
1592 rule_make_actions(struct ofproto *p, struct rule *rule,
1593 const struct ofpbuf *packet)
1595 const struct rule *super;
1596 struct odp_actions a;
1599 assert(!rule->cr.wc.wildcards);
1601 super = rule->super ? rule->super : rule;
1603 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1604 packet, &a, &rule->tags, &rule->may_install,
1605 &rule->nf_flow.output_iface);
1607 actions_len = a.n_actions * sizeof *a.actions;
1608 if (rule->n_odp_actions != a.n_actions
1609 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1610 COVERAGE_INC(ofproto_odp_unchanged);
1611 free(rule->odp_actions);
1612 rule->n_odp_actions = a.n_actions;
1613 rule->odp_actions = xmemdup(a.actions, actions_len);
1621 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1622 struct odp_flow_put *put)
1624 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1625 put->flow.key = rule->cr.flow;
1626 put->flow.actions = rule->odp_actions;
1627 put->flow.n_actions = rule->n_odp_actions;
1629 return dpif_flow_put(ofproto->dpif, put);
1633 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1635 assert(!rule->cr.wc.wildcards);
1637 if (rule->may_install) {
1638 struct odp_flow_put put;
1639 if (!do_put_flow(p, rule,
1640 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1642 rule->installed = true;
1643 if (displaced_rule) {
1644 update_stats(p, displaced_rule, &put.flow.stats);
1645 rule_post_uninstall(p, displaced_rule);
1648 } else if (displaced_rule) {
1649 rule_uninstall(p, displaced_rule);
1654 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1656 if (rule->installed) {
1657 struct odp_flow_put put;
1658 COVERAGE_INC(ofproto_dp_missed);
1659 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1661 rule_install(ofproto, rule, NULL);
1666 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1668 bool actions_changed;
1669 uint16_t new_out_iface, old_out_iface;
1671 old_out_iface = rule->nf_flow.output_iface;
1672 actions_changed = rule_make_actions(ofproto, rule, NULL);
1674 if (rule->may_install) {
1675 if (rule->installed) {
1676 if (actions_changed) {
1677 struct odp_flow_put put;
1678 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
1679 | ODPPF_ZERO_STATS, &put);
1680 update_stats(ofproto, rule, &put.flow.stats);
1682 /* Temporarily set the old output iface so that NetFlow
1683 * messages have the correct output interface for the old
1685 new_out_iface = rule->nf_flow.output_iface;
1686 rule->nf_flow.output_iface = old_out_iface;
1687 rule_post_uninstall(ofproto, rule);
1688 rule->nf_flow.output_iface = new_out_iface;
1691 rule_install(ofproto, rule, NULL);
1694 rule_uninstall(ofproto, rule);
1699 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1701 uint64_t total_bytes = rule->byte_count + extra_bytes;
1703 if (ofproto->ofhooks->account_flow_cb
1704 && total_bytes > rule->accounted_bytes)
1706 ofproto->ofhooks->account_flow_cb(
1707 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1708 total_bytes - rule->accounted_bytes, ofproto->aux);
1709 rule->accounted_bytes = total_bytes;
1714 rule_uninstall(struct ofproto *p, struct rule *rule)
1716 assert(!rule->cr.wc.wildcards);
1717 if (rule->installed) {
1718 struct odp_flow odp_flow;
1720 odp_flow.key = rule->cr.flow;
1721 odp_flow.actions = NULL;
1722 odp_flow.n_actions = 0;
1723 if (!dpif_flow_del(p->dpif, &odp_flow)) {
1724 update_stats(p, rule, &odp_flow.stats);
1726 rule->installed = false;
1728 rule_post_uninstall(p, rule);
1733 is_controller_rule(struct rule *rule)
1735 /* If the only action is send to the controller then don't report
1736 * NetFlow expiration messages since it is just part of the control
1737 * logic for the network and not real traffic. */
1739 if (rule && rule->super) {
1740 struct rule *super = rule->super;
1742 return super->n_actions == 1 &&
1743 super->actions[0].type == htons(OFPAT_OUTPUT) &&
1744 super->actions[0].output.port == htons(OFPP_CONTROLLER);
1751 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1753 struct rule *super = rule->super;
1755 rule_account(ofproto, rule, 0);
1757 if (ofproto->netflow && !is_controller_rule(rule)) {
1758 struct ofexpired expired;
1759 expired.flow = rule->cr.flow;
1760 expired.packet_count = rule->packet_count;
1761 expired.byte_count = rule->byte_count;
1762 expired.used = rule->used;
1763 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
1766 super->packet_count += rule->packet_count;
1767 super->byte_count += rule->byte_count;
1769 /* Reset counters to prevent double counting if the rule ever gets
1771 rule->packet_count = 0;
1772 rule->byte_count = 0;
1773 rule->accounted_bytes = 0;
1775 netflow_flow_clear(&rule->nf_flow);
1780 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1781 struct rconn_packet_counter *counter)
1783 update_openflow_length(msg);
1784 if (rconn_send(ofconn->rconn, msg, counter)) {
1790 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1791 int error, const void *data, size_t len)
1794 struct ofp_error_msg *oem;
1796 if (!(error >> 16)) {
1797 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1802 COVERAGE_INC(ofproto_error);
1803 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1804 oh ? oh->xid : 0, &buf);
1805 oem->type = htons((unsigned int) error >> 16);
1806 oem->code = htons(error & 0xffff);
1807 memcpy(oem->data, data, len);
1808 queue_tx(buf, ofconn, ofconn->reply_counter);
1812 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1815 size_t oh_length = ntohs(oh->length);
1816 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1820 hton_ofp_phy_port(struct ofp_phy_port *opp)
1822 opp->port_no = htons(opp->port_no);
1823 opp->config = htonl(opp->config);
1824 opp->state = htonl(opp->state);
1825 opp->curr = htonl(opp->curr);
1826 opp->advertised = htonl(opp->advertised);
1827 opp->supported = htonl(opp->supported);
1828 opp->peer = htonl(opp->peer);
1832 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
1834 struct ofp_header *rq = oh;
1835 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
1840 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
1841 struct ofp_header *oh)
1843 struct ofp_switch_features *osf;
1845 unsigned int port_no;
1846 struct ofport *port;
1848 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1849 osf->datapath_id = htonll(p->datapath_id);
1850 osf->n_buffers = htonl(pktbuf_capacity());
1852 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1853 OFPC_PORT_STATS | OFPC_MULTI_PHY_TX);
1854 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1855 (1u << OFPAT_SET_VLAN_VID) |
1856 (1u << OFPAT_SET_VLAN_PCP) |
1857 (1u << OFPAT_STRIP_VLAN) |
1858 (1u << OFPAT_SET_DL_SRC) |
1859 (1u << OFPAT_SET_DL_DST) |
1860 (1u << OFPAT_SET_NW_SRC) |
1861 (1u << OFPAT_SET_NW_DST) |
1862 (1u << OFPAT_SET_NW_TOS) |
1863 (1u << OFPAT_SET_TP_SRC) |
1864 (1u << OFPAT_SET_TP_DST));
1866 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1867 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1870 queue_tx(buf, ofconn, ofconn->reply_counter);
1875 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1876 struct ofp_header *oh)
1879 struct ofp_switch_config *osc;
1883 /* Figure out flags. */
1884 dpif_get_drop_frags(p->dpif, &drop_frags);
1885 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1888 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1889 osc->flags = htons(flags);
1890 osc->miss_send_len = htons(ofconn->miss_send_len);
1891 queue_tx(buf, ofconn, ofconn->reply_counter);
1897 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1898 struct ofp_switch_config *osc)
1903 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1907 flags = ntohs(osc->flags);
1909 if (ofconn == p->controller) {
1910 switch (flags & OFPC_FRAG_MASK) {
1911 case OFPC_FRAG_NORMAL:
1912 dpif_set_drop_frags(p->dpif, false);
1914 case OFPC_FRAG_DROP:
1915 dpif_set_drop_frags(p->dpif, true);
1918 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1924 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1925 if (ntohs(osc->miss_send_len) != 0) {
1926 ofconn->pktbuf = pktbuf_create();
1928 pktbuf_destroy(ofconn->pktbuf);
1932 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1938 add_output_group_action(struct odp_actions *actions, uint16_t group,
1939 uint16_t *nf_output_iface)
1941 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1943 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
1944 *nf_output_iface = NF_OUT_FLOOD;
1949 add_controller_action(struct odp_actions *actions,
1950 const struct ofp_action_output *oao)
1952 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
1953 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
1956 struct action_xlate_ctx {
1958 const flow_t *flow; /* Flow to which these actions correspond. */
1959 int recurse; /* Recursion level, via xlate_table_action. */
1960 struct ofproto *ofproto;
1961 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
1962 * null pointer if we are revalidating
1963 * without a packet to refer to. */
1966 struct odp_actions *out; /* Datapath actions. */
1967 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
1968 bool may_set_up_flow; /* True ordinarily; false if the actions must
1969 * be reassessed for every packet. */
1970 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
1973 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
1974 struct action_xlate_ctx *ctx);
1977 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
1979 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
1982 if (ofport->opp.config & OFPPC_NO_FWD) {
1983 /* Forwarding disabled on port. */
1988 * We don't have an ofport record for this port, but it doesn't hurt to
1989 * allow forwarding to it anyhow. Maybe such a port will appear later
1990 * and we're pre-populating the flow table.
1994 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
1995 ctx->nf_output_iface = port;
1998 static struct rule *
1999 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2002 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2004 /* The rule we found might not be valid, since we could be in need of
2005 * revalidation. If it is not valid, don't return it. */
2008 && ofproto->need_revalidate
2009 && !revalidate_rule(ofproto, rule)) {
2010 COVERAGE_INC(ofproto_invalidated);
2018 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2020 if (!ctx->recurse) {
2025 flow.in_port = in_port;
2027 rule = lookup_valid_rule(ctx->ofproto, &flow);
2034 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2041 xlate_output_action(struct action_xlate_ctx *ctx,
2042 const struct ofp_action_output *oao)
2045 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2047 ctx->nf_output_iface = NF_OUT_DROP;
2049 switch (ntohs(oao->port)) {
2051 add_output_action(ctx, ctx->flow->in_port);
2054 xlate_table_action(ctx, ctx->flow->in_port);
2057 if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet,
2058 ctx->out, ctx->tags,
2059 &ctx->nf_output_iface,
2060 ctx->ofproto->aux)) {
2061 COVERAGE_INC(ofproto_uninstallable);
2062 ctx->may_set_up_flow = false;
2066 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2067 &ctx->nf_output_iface);
2070 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2072 case OFPP_CONTROLLER:
2073 add_controller_action(ctx->out, oao);
2076 add_output_action(ctx, ODPP_LOCAL);
2079 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2080 if (odp_port != ctx->flow->in_port) {
2081 add_output_action(ctx, odp_port);
2086 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2087 ctx->nf_output_iface = NF_OUT_FLOOD;
2088 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2089 ctx->nf_output_iface = prev_nf_output_iface;
2090 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2091 ctx->nf_output_iface != NF_OUT_FLOOD) {
2092 ctx->nf_output_iface = NF_OUT_MULTI;
2097 xlate_nicira_action(struct action_xlate_ctx *ctx,
2098 const struct nx_action_header *nah)
2100 const struct nx_action_resubmit *nar;
2101 int subtype = ntohs(nah->subtype);
2103 assert(nah->vendor == htonl(NX_VENDOR_ID));
2105 case NXAST_RESUBMIT:
2106 nar = (const struct nx_action_resubmit *) nah;
2107 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2111 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2117 do_xlate_actions(const union ofp_action *in, size_t n_in,
2118 struct action_xlate_ctx *ctx)
2120 struct actions_iterator iter;
2121 const union ofp_action *ia;
2122 const struct ofport *port;
2124 port = port_array_get(&ctx->ofproto->ports, ctx->flow->in_port);
2125 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2126 port->opp.config & (eth_addr_equals(ctx->flow->dl_dst, stp_eth_addr)
2127 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2128 /* Drop this flow. */
2132 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2133 uint16_t type = ntohs(ia->type);
2134 union odp_action *oa;
2138 xlate_output_action(ctx, &ia->output);
2141 case OFPAT_SET_VLAN_VID:
2142 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2143 oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2146 case OFPAT_SET_VLAN_PCP:
2147 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2148 oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2151 case OFPAT_STRIP_VLAN:
2152 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2155 case OFPAT_SET_DL_SRC:
2156 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2157 memcpy(oa->dl_addr.dl_addr,
2158 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2161 case OFPAT_SET_DL_DST:
2162 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2163 memcpy(oa->dl_addr.dl_addr,
2164 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2167 case OFPAT_SET_NW_SRC:
2168 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2169 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2172 case OFPAT_SET_NW_DST:
2173 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2174 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2176 case OFPAT_SET_NW_TOS:
2177 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2178 oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2181 case OFPAT_SET_TP_SRC:
2182 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2183 oa->tp_port.tp_port = ia->tp_port.tp_port;
2186 case OFPAT_SET_TP_DST:
2187 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2188 oa->tp_port.tp_port = ia->tp_port.tp_port;
2192 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2196 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2203 xlate_actions(const union ofp_action *in, size_t n_in,
2204 const flow_t *flow, struct ofproto *ofproto,
2205 const struct ofpbuf *packet,
2206 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2207 uint16_t *nf_output_iface)
2209 tag_type no_tags = 0;
2210 struct action_xlate_ctx ctx;
2211 COVERAGE_INC(ofproto_ofp2odp);
2212 odp_actions_init(out);
2215 ctx.ofproto = ofproto;
2216 ctx.packet = packet;
2218 ctx.tags = tags ? tags : &no_tags;
2219 ctx.may_set_up_flow = true;
2220 ctx.nf_output_iface = NF_OUT_DROP;
2221 do_xlate_actions(in, n_in, &ctx);
2223 /* Check with in-band control to see if we're allowed to set up this
2225 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2226 ctx.may_set_up_flow = false;
2229 if (may_set_up_flow) {
2230 *may_set_up_flow = ctx.may_set_up_flow;
2232 if (nf_output_iface) {
2233 *nf_output_iface = ctx.nf_output_iface;
2235 if (odp_actions_overflow(out)) {
2236 odp_actions_init(out);
2237 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2243 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2244 struct ofp_header *oh)
2246 struct ofp_packet_out *opo;
2247 struct ofpbuf payload, *buffer;
2248 struct odp_actions actions;
2254 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2258 opo = (struct ofp_packet_out *) oh;
2260 COVERAGE_INC(ofproto_packet_out);
2261 if (opo->buffer_id != htonl(UINT32_MAX)) {
2262 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2264 if (error || !buffer) {
2272 flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2273 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2274 &flow, p, &payload, &actions, NULL, NULL, NULL);
2279 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2281 ofpbuf_delete(buffer);
2287 update_port_config(struct ofproto *p, struct ofport *port,
2288 uint32_t config, uint32_t mask)
2290 mask &= config ^ port->opp.config;
2291 if (mask & OFPPC_PORT_DOWN) {
2292 if (config & OFPPC_PORT_DOWN) {
2293 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2295 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2298 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2299 if (mask & REVALIDATE_BITS) {
2300 COVERAGE_INC(ofproto_costly_flags);
2301 port->opp.config ^= mask & REVALIDATE_BITS;
2302 p->need_revalidate = true;
2304 #undef REVALIDATE_BITS
2305 if (mask & OFPPC_NO_FLOOD) {
2306 port->opp.config ^= OFPPC_NO_FLOOD;
2307 refresh_port_group(p, DP_GROUP_FLOOD);
2309 if (mask & OFPPC_NO_PACKET_IN) {
2310 port->opp.config ^= OFPPC_NO_PACKET_IN;
2315 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2317 const struct ofp_port_mod *opm;
2318 struct ofport *port;
2321 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2325 opm = (struct ofp_port_mod *) oh;
2327 port = port_array_get(&p->ports,
2328 ofp_port_to_odp_port(ntohs(opm->port_no)));
2330 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2331 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2332 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2334 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2335 if (opm->advertise) {
2336 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2342 static struct ofpbuf *
2343 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2345 struct ofp_stats_reply *osr;
2348 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2349 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2351 osr->flags = htons(0);
2355 static struct ofpbuf *
2356 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2358 return make_stats_reply(request->header.xid, request->type, body_len);
2362 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2364 struct ofpbuf *msg = *msgp;
2365 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2366 if (nbytes + msg->size > UINT16_MAX) {
2367 struct ofp_stats_reply *reply = msg->data;
2368 reply->flags = htons(OFPSF_REPLY_MORE);
2369 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2370 queue_tx(msg, ofconn, ofconn->reply_counter);
2372 return ofpbuf_put_uninit(*msgp, nbytes);
2376 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2377 struct ofp_stats_request *request)
2379 struct ofp_desc_stats *ods;
2382 msg = start_stats_reply(request, sizeof *ods);
2383 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2384 strncpy(ods->mfr_desc, p->manufacturer, sizeof ods->mfr_desc);
2385 strncpy(ods->hw_desc, p->hardware, sizeof ods->hw_desc);
2386 strncpy(ods->sw_desc, p->software, sizeof ods->sw_desc);
2387 strncpy(ods->serial_num, p->serial, sizeof ods->serial_num);
2388 strncpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2389 queue_tx(msg, ofconn, ofconn->reply_counter);
2395 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2397 struct rule *rule = rule_from_cls_rule(cls_rule);
2398 int *n_subrules = n_subrules_;
2406 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2407 struct ofp_stats_request *request)
2409 struct ofp_table_stats *ots;
2411 struct odp_stats dpstats;
2412 int n_exact, n_subrules, n_wild;
2414 msg = start_stats_reply(request, sizeof *ots * 2);
2416 /* Count rules of various kinds. */
2418 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2419 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2420 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2423 dpif_get_dp_stats(p->dpif, &dpstats);
2424 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2425 memset(ots, 0, sizeof *ots);
2426 ots->table_id = TABLEID_HASH;
2427 strcpy(ots->name, "hash");
2428 ots->wildcards = htonl(0);
2429 ots->max_entries = htonl(dpstats.max_capacity);
2430 ots->active_count = htonl(n_exact);
2431 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2433 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2435 /* Classifier table. */
2436 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2437 memset(ots, 0, sizeof *ots);
2438 ots->table_id = TABLEID_CLASSIFIER;
2439 strcpy(ots->name, "classifier");
2440 ots->wildcards = htonl(OFPFW_ALL);
2441 ots->max_entries = htonl(65536);
2442 ots->active_count = htonl(n_wild);
2443 ots->lookup_count = htonll(0); /* XXX */
2444 ots->matched_count = htonll(0); /* XXX */
2446 queue_tx(msg, ofconn, ofconn->reply_counter);
2451 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2452 struct ofp_stats_request *request)
2454 struct ofp_port_stats *ops;
2456 struct ofport *port;
2457 unsigned int port_no;
2459 msg = start_stats_reply(request, sizeof *ops * 16);
2460 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2461 struct netdev_stats stats;
2463 /* Intentionally ignore return value, since errors will set 'stats' to
2464 * all-1s, which is correct for OpenFlow, and netdev_get_stats() will
2466 netdev_get_stats(port->netdev, &stats);
2468 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2469 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2470 memset(ops->pad, 0, sizeof ops->pad);
2471 ops->rx_packets = htonll(stats.rx_packets);
2472 ops->tx_packets = htonll(stats.tx_packets);
2473 ops->rx_bytes = htonll(stats.rx_bytes);
2474 ops->tx_bytes = htonll(stats.tx_bytes);
2475 ops->rx_dropped = htonll(stats.rx_dropped);
2476 ops->tx_dropped = htonll(stats.tx_dropped);
2477 ops->rx_errors = htonll(stats.rx_errors);
2478 ops->tx_errors = htonll(stats.tx_errors);
2479 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2480 ops->rx_over_err = htonll(stats.rx_over_errors);
2481 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2482 ops->collisions = htonll(stats.collisions);
2485 queue_tx(msg, ofconn, ofconn->reply_counter);
2489 struct flow_stats_cbdata {
2490 struct ofproto *ofproto;
2491 struct ofconn *ofconn;
2497 query_stats(struct ofproto *p, struct rule *rule,
2498 uint64_t *packet_countp, uint64_t *byte_countp)
2500 uint64_t packet_count, byte_count;
2501 struct rule *subrule;
2502 struct odp_flow *odp_flows;
2505 packet_count = rule->packet_count;
2506 byte_count = rule->byte_count;
2508 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2509 odp_flows = xcalloc(1, n_odp_flows * sizeof *odp_flows);
2510 if (rule->cr.wc.wildcards) {
2512 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2513 odp_flows[i++].key = subrule->cr.flow;
2514 packet_count += subrule->packet_count;
2515 byte_count += subrule->byte_count;
2518 odp_flows[0].key = rule->cr.flow;
2521 packet_count = rule->packet_count;
2522 byte_count = rule->byte_count;
2523 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2525 for (i = 0; i < n_odp_flows; i++) {
2526 struct odp_flow *odp_flow = &odp_flows[i];
2527 packet_count += odp_flow->stats.n_packets;
2528 byte_count += odp_flow->stats.n_bytes;
2533 *packet_countp = packet_count;
2534 *byte_countp = byte_count;
2538 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2540 struct rule *rule = rule_from_cls_rule(rule_);
2541 struct flow_stats_cbdata *cbdata = cbdata_;
2542 struct ofp_flow_stats *ofs;
2543 uint64_t packet_count, byte_count;
2544 size_t act_len, len;
2546 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2550 act_len = sizeof *rule->actions * rule->n_actions;
2551 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2553 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2555 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2556 ofs->length = htons(len);
2557 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2559 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match);
2560 ofs->duration = htonl((time_msec() - rule->created) / 1000);
2561 ofs->priority = htons(rule->cr.priority);
2562 ofs->idle_timeout = htons(rule->idle_timeout);
2563 ofs->hard_timeout = htons(rule->hard_timeout);
2565 ofs->packet_count = htonll(packet_count);
2566 ofs->byte_count = htonll(byte_count);
2567 memcpy(ofs->actions, rule->actions, act_len);
2571 table_id_to_include(uint8_t table_id)
2573 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2574 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2575 : table_id == 0xff ? CLS_INC_ALL
2580 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2581 const struct ofp_stats_request *osr,
2584 struct ofp_flow_stats_request *fsr;
2585 struct flow_stats_cbdata cbdata;
2586 struct cls_rule target;
2588 if (arg_size != sizeof *fsr) {
2589 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2591 fsr = (struct ofp_flow_stats_request *) osr->body;
2593 COVERAGE_INC(ofproto_flows_req);
2595 cbdata.ofconn = ofconn;
2596 cbdata.out_port = fsr->out_port;
2597 cbdata.msg = start_stats_reply(osr, 1024);
2598 cls_rule_from_match(&target, &fsr->match, 0);
2599 classifier_for_each_match(&p->cls, &target,
2600 table_id_to_include(fsr->table_id),
2601 flow_stats_cb, &cbdata);
2602 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2606 struct flow_stats_ds_cbdata {
2607 struct ofproto *ofproto;
2612 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
2614 struct rule *rule = rule_from_cls_rule(rule_);
2615 struct flow_stats_ds_cbdata *cbdata = cbdata_;
2616 struct ds *results = cbdata->results;
2617 struct ofp_match match;
2618 uint64_t packet_count, byte_count;
2619 size_t act_len = sizeof *rule->actions * rule->n_actions;
2621 /* Don't report on subrules. */
2622 if (rule->super != NULL) {
2626 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2627 flow_to_ovs_match(&rule->cr.flow, rule->cr.wc.wildcards, &match);
2629 ds_put_format(results, "duration=%llds, ",
2630 (time_msec() - rule->created) / 1000);
2631 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2632 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2633 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2634 ofp_print_match(results, &match, true);
2635 ofp_print_actions(results, &rule->actions->header, act_len);
2636 ds_put_cstr(results, "\n");
2639 /* Adds a pretty-printed description of all flows to 'results', including
2640 * those marked hidden by secchan (e.g., by in-band control). */
2642 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2644 struct ofp_match match;
2645 struct cls_rule target;
2646 struct flow_stats_ds_cbdata cbdata;
2648 memset(&match, 0, sizeof match);
2649 match.wildcards = htonl(OFPFW_ALL);
2652 cbdata.results = results;
2654 cls_rule_from_match(&target, &match, 0);
2655 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2656 flow_stats_ds_cb, &cbdata);
2659 struct aggregate_stats_cbdata {
2660 struct ofproto *ofproto;
2662 uint64_t packet_count;
2663 uint64_t byte_count;
2668 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2670 struct rule *rule = rule_from_cls_rule(rule_);
2671 struct aggregate_stats_cbdata *cbdata = cbdata_;
2672 uint64_t packet_count, byte_count;
2674 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2678 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2680 cbdata->packet_count += packet_count;
2681 cbdata->byte_count += byte_count;
2686 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2687 const struct ofp_stats_request *osr,
2690 struct ofp_aggregate_stats_request *asr;
2691 struct ofp_aggregate_stats_reply *reply;
2692 struct aggregate_stats_cbdata cbdata;
2693 struct cls_rule target;
2696 if (arg_size != sizeof *asr) {
2697 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2699 asr = (struct ofp_aggregate_stats_request *) osr->body;
2701 COVERAGE_INC(ofproto_agg_request);
2703 cbdata.out_port = asr->out_port;
2704 cbdata.packet_count = 0;
2705 cbdata.byte_count = 0;
2707 cls_rule_from_match(&target, &asr->match, 0);
2708 classifier_for_each_match(&p->cls, &target,
2709 table_id_to_include(asr->table_id),
2710 aggregate_stats_cb, &cbdata);
2712 msg = start_stats_reply(osr, sizeof *reply);
2713 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2714 reply->flow_count = htonl(cbdata.n_flows);
2715 reply->packet_count = htonll(cbdata.packet_count);
2716 reply->byte_count = htonll(cbdata.byte_count);
2717 queue_tx(msg, ofconn, ofconn->reply_counter);
2722 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2723 struct ofp_header *oh)
2725 struct ofp_stats_request *osr;
2729 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2734 osr = (struct ofp_stats_request *) oh;
2736 switch (ntohs(osr->type)) {
2738 return handle_desc_stats_request(p, ofconn, osr);
2741 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2743 case OFPST_AGGREGATE:
2744 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2747 return handle_table_stats_request(p, ofconn, osr);
2750 return handle_port_stats_request(p, ofconn, osr);
2753 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2756 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2760 static long long int
2761 msec_from_nsec(uint64_t sec, uint32_t nsec)
2763 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2767 update_time(struct ofproto *ofproto, struct rule *rule,
2768 const struct odp_flow_stats *stats)
2770 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2771 if (used > rule->used) {
2773 if (rule->super && used > rule->super->used) {
2774 rule->super->used = used;
2776 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
2781 update_stats(struct ofproto *ofproto, struct rule *rule,
2782 const struct odp_flow_stats *stats)
2784 if (stats->n_packets) {
2785 update_time(ofproto, rule, stats);
2786 rule->packet_count += stats->n_packets;
2787 rule->byte_count += stats->n_bytes;
2788 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
2794 add_flow(struct ofproto *p, struct ofconn *ofconn,
2795 struct ofp_flow_mod *ofm, size_t n_actions)
2797 struct ofpbuf *packet;
2802 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
2806 flow_from_match(&flow, &wildcards, &ofm->match);
2807 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
2808 ntohs(ofm->priority))) {
2809 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
2813 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
2814 n_actions, ntohs(ofm->idle_timeout),
2815 ntohs(ofm->hard_timeout),
2816 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
2817 cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority));
2821 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2822 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2826 rule_insert(p, rule, packet, in_port);
2827 ofpbuf_delete(packet);
2832 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
2833 size_t n_actions, uint16_t command, struct rule *rule)
2835 if (rule_is_hidden(rule)) {
2839 if (command == OFPFC_DELETE) {
2840 long long int now = time_msec();
2841 send_flow_removed(p, rule, now, OFPRR_DELETE);
2842 rule_remove(p, rule);
2844 size_t actions_len = n_actions * sizeof *rule->actions;
2846 if (n_actions == rule->n_actions
2847 && !memcmp(ofm->actions, rule->actions, actions_len))
2852 free(rule->actions);
2853 rule->actions = xmemdup(ofm->actions, actions_len);
2854 rule->n_actions = n_actions;
2856 if (rule->cr.wc.wildcards) {
2857 COVERAGE_INC(ofproto_mod_wc_flow);
2858 p->need_revalidate = true;
2860 rule_update_actions(p, rule);
2868 modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm,
2869 size_t n_actions, uint16_t command)
2875 flow_from_match(&flow, &wildcards, &ofm->match);
2876 rule = rule_from_cls_rule(classifier_find_rule_exactly(
2877 &p->cls, &flow, wildcards,
2878 ntohs(ofm->priority)));
2881 if (command == OFPFC_DELETE
2882 && ofm->out_port != htons(OFPP_NONE)
2883 && !rule_has_out_port(rule, ofm->out_port)) {
2887 modify_flow(p, ofm, n_actions, command, rule);
2892 struct modify_flows_cbdata {
2893 struct ofproto *ofproto;
2894 const struct ofp_flow_mod *ofm;
2901 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
2903 struct rule *rule = rule_from_cls_rule(rule_);
2904 struct modify_flows_cbdata *cbdata = cbdata_;
2906 if (cbdata->out_port != htons(OFPP_NONE)
2907 && !rule_has_out_port(rule, cbdata->out_port)) {
2911 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions,
2912 cbdata->command, rule);
2916 modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm,
2917 size_t n_actions, uint16_t command)
2919 struct modify_flows_cbdata cbdata;
2920 struct cls_rule target;
2924 cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port
2925 : htons(OFPP_NONE));
2926 cbdata.n_actions = n_actions;
2927 cbdata.command = command;
2929 cls_rule_from_match(&target, &ofm->match, 0);
2931 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2932 modify_flows_cb, &cbdata);
2937 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
2938 struct ofp_flow_mod *ofm)
2943 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
2944 sizeof *ofm->actions, &n_actions);
2949 /* We do not support the emergency flow cache. It will hopefully
2950 * get dropped from OpenFlow in the near future. */
2951 if (ofm->flags & htons(OFPFF_EMERG)) {
2952 /* There isn't a good fit for an error code, so just state that the
2953 * flow table is full. */
2954 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
2957 normalize_match(&ofm->match);
2958 if (!ofm->match.wildcards) {
2959 ofm->priority = htons(UINT16_MAX);
2962 error = validate_actions((const union ofp_action *) ofm->actions,
2963 n_actions, p->max_ports);
2968 switch (ntohs(ofm->command)) {
2970 return add_flow(p, ofconn, ofm, n_actions);
2973 return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY);
2975 case OFPFC_MODIFY_STRICT:
2976 return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY);
2979 return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE);
2981 case OFPFC_DELETE_STRICT:
2982 return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE);
2985 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
2990 send_capability_reply(struct ofproto *p, struct ofconn *ofconn, uint32_t xid)
2992 struct ofmp_capability_reply *ocr;
2994 char capabilities[] = "com.nicira.mgmt.manager=false\n";
2996 ocr = make_openflow_xid(sizeof(*ocr), OFPT_VENDOR, xid, &b);
2997 ocr->header.header.vendor = htonl(NX_VENDOR_ID);
2998 ocr->header.header.subtype = htonl(NXT_MGMT);
2999 ocr->header.type = htons(OFMPT_CAPABILITY_REPLY);
3001 ocr->format = htonl(OFMPCOF_SIMPLE);
3002 ocr->mgmt_id = htonll(p->mgmt_id);
3004 ofpbuf_put(b, capabilities, strlen(capabilities));
3006 queue_tx(b, ofconn, ofconn->reply_counter);
3010 handle_ofmp(struct ofproto *p, struct ofconn *ofconn,
3011 struct ofmp_header *ofmph)
3013 size_t msg_len = ntohs(ofmph->header.header.length);
3014 if (msg_len < sizeof(*ofmph)) {
3015 VLOG_WARN_RL(&rl, "dropping short managment message: %zu\n", msg_len);
3016 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3019 if (ofmph->type == htons(OFMPT_CAPABILITY_REQUEST)) {
3020 struct ofmp_capability_request *ofmpcr;
3022 if (msg_len < sizeof(struct ofmp_capability_request)) {
3023 VLOG_WARN_RL(&rl, "dropping short capability request: %zu\n",
3025 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3028 ofmpcr = (struct ofmp_capability_request *)ofmph;
3029 if (ofmpcr->format != htonl(OFMPCAF_SIMPLE)) {
3030 /* xxx Find a better type than bad subtype */
3031 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3034 send_capability_reply(p, ofconn, ofmph->header.header.xid);
3037 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3042 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3044 struct ofp_vendor_header *ovh = msg;
3045 struct nicira_header *nh;
3047 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3048 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3050 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3051 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3053 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3054 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3058 switch (ntohl(nh->subtype)) {
3059 case NXT_STATUS_REQUEST:
3060 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3063 case NXT_ACT_SET_CONFIG:
3064 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
3066 case NXT_ACT_GET_CONFIG:
3067 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
3069 case NXT_COMMAND_REQUEST:
3071 return executer_handle_request(p->executer, ofconn->rconn, msg);
3076 return handle_ofmp(p, ofconn, msg);
3079 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3083 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
3085 struct ofp_header *ob;
3088 /* Currently, everything executes synchronously, so we can just
3089 * immediately send the barrier reply. */
3090 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3091 queue_tx(buf, ofconn, ofconn->reply_counter);
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);
3144 case OFPT_BARRIER_REQUEST:
3145 error = handle_barrier_request(ofconn, oh);
3149 if (VLOG_IS_WARN_ENABLED()) {
3150 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3151 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3154 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3159 send_error_oh(ofconn, ofp_msg->data, error);
3164 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3166 struct odp_msg *msg = packet->data;
3167 uint16_t in_port = odp_port_to_ofp_port(msg->port);
3169 struct ofpbuf payload;
3172 /* Handle controller actions. */
3173 if (msg->type == _ODPL_ACTION_NR) {
3174 COVERAGE_INC(ofproto_ctlr_action);
3175 pinsched_send(p->action_sched, in_port, packet,
3176 send_packet_in_action, p);
3180 payload.data = msg + 1;
3181 payload.size = msg->length - sizeof *msg;
3182 flow_extract(&payload, msg->port, &flow);
3184 /* Check with in-band control to see if this packet should be sent
3185 * to the local port regardless of the flow table. */
3186 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3187 union odp_action action;
3189 memset(&action, 0, sizeof(action));
3190 action.output.type = ODPAT_OUTPUT;
3191 action.output.port = ODPP_LOCAL;
3192 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3195 rule = lookup_valid_rule(p, &flow);
3197 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3198 struct ofport *port = port_array_get(&p->ports, msg->port);
3200 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3201 COVERAGE_INC(ofproto_no_packet_in);
3202 /* XXX install 'drop' flow entry */
3203 ofpbuf_delete(packet);
3207 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3210 COVERAGE_INC(ofproto_packet_in);
3211 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3215 if (rule->cr.wc.wildcards) {
3216 rule = rule_create_subrule(p, rule, &flow);
3217 rule_make_actions(p, rule, packet);
3219 if (!rule->may_install) {
3220 /* The rule is not installable, that is, we need to process every
3221 * packet, so process the current packet and set its actions into
3223 rule_make_actions(p, rule, packet);
3225 /* XXX revalidate rule if it needs it */
3229 rule_execute(p, rule, &payload, &flow);
3230 rule_reinstall(p, rule);
3232 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY
3233 && rconn_is_connected(p->controller->rconn)) {
3235 * Extra-special case for fail-open mode.
3237 * We are in fail-open mode and the packet matched the fail-open rule,
3238 * but we are connected to a controller too. We should send the packet
3239 * up to the controller in the hope that it will try to set up a flow
3240 * and thereby allow us to exit fail-open.
3242 * See the top-level comment in fail-open.c for more information.
3244 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3246 ofpbuf_delete(packet);
3251 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3253 struct rule *sub = rule_from_cls_rule(sub_);
3254 struct revalidate_cbdata *cbdata = cbdata_;
3256 if (cbdata->revalidate_all
3257 || (cbdata->revalidate_subrules && sub->super)
3258 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3259 revalidate_rule(cbdata->ofproto, sub);
3264 revalidate_rule(struct ofproto *p, struct rule *rule)
3266 const flow_t *flow = &rule->cr.flow;
3268 COVERAGE_INC(ofproto_revalidate_rule);
3271 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3273 rule_remove(p, rule);
3275 } else if (super != rule->super) {
3276 COVERAGE_INC(ofproto_revalidate_moved);
3277 list_remove(&rule->list);
3278 list_push_back(&super->list, &rule->list);
3279 rule->super = super;
3280 rule->hard_timeout = super->hard_timeout;
3281 rule->idle_timeout = super->idle_timeout;
3282 rule->created = super->created;
3287 rule_update_actions(p, rule);
3291 static struct ofpbuf *
3292 compose_flow_removed(const struct rule *rule, long long int now, uint8_t reason)
3294 struct ofp_flow_removed *ofr;
3296 long long int last_used = rule->used ? now - rule->used : 0;
3298 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
3299 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofr->match);
3300 ofr->priority = htons(rule->cr.priority);
3301 ofr->reason = reason;
3302 ofr->duration = htonl((now - rule->created - last_used) / 1000);
3303 ofr->idle_timeout = htons(rule->idle_timeout);
3304 ofr->packet_count = htonll(rule->packet_count);
3305 ofr->byte_count = htonll(rule->byte_count);
3311 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3313 assert(rule->installed);
3314 assert(!rule->cr.wc.wildcards);
3317 rule_remove(ofproto, rule);
3319 rule_uninstall(ofproto, rule);
3323 send_flow_removed(struct ofproto *p, struct rule *rule,
3324 long long int now, uint8_t reason)
3326 struct ofconn *ofconn;
3327 struct ofconn *prev;
3328 struct ofpbuf *buf = NULL;
3330 /* We limit the maximum number of queued flow expirations it by accounting
3331 * them under the counter for replies. That works because preventing
3332 * OpenFlow requests from being processed also prevents new flows from
3333 * being added (and expiring). (It also prevents processing OpenFlow
3334 * requests that would not add new flows, so it is imperfect.) */
3337 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3338 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)) {
3340 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3342 buf = compose_flow_removed(rule, now, reason);
3348 queue_tx(buf, prev, prev->reply_counter);
3354 expire_rule(struct cls_rule *cls_rule, void *p_)
3356 struct ofproto *p = p_;
3357 struct rule *rule = rule_from_cls_rule(cls_rule);
3358 long long int hard_expire, idle_expire, expire, now;
3360 hard_expire = (rule->hard_timeout
3361 ? rule->created + rule->hard_timeout * 1000
3363 idle_expire = (rule->idle_timeout
3364 && (rule->super || list_is_empty(&rule->list))
3365 ? rule->used + rule->idle_timeout * 1000
3367 expire = MIN(hard_expire, idle_expire);
3371 if (rule->installed && now >= rule->used + 5000) {
3372 uninstall_idle_flow(p, rule);
3373 } else if (!rule->cr.wc.wildcards) {
3374 active_timeout(p, rule);
3380 COVERAGE_INC(ofproto_expired);
3382 /* Update stats. This code will be a no-op if the rule expired
3383 * due to an idle timeout. */
3384 if (rule->cr.wc.wildcards) {
3385 struct rule *subrule, *next;
3386 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3387 rule_remove(p, subrule);
3390 rule_uninstall(p, rule);
3393 if (!rule_is_hidden(rule)) {
3394 send_flow_removed(p, rule, now,
3396 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
3398 rule_remove(p, rule);
3402 active_timeout(struct ofproto *ofproto, struct rule *rule)
3404 if (ofproto->netflow && !is_controller_rule(rule) &&
3405 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
3406 struct ofexpired expired;
3407 struct odp_flow odp_flow;
3409 /* Get updated flow stats. */
3410 memset(&odp_flow, 0, sizeof odp_flow);
3411 if (rule->installed) {
3412 odp_flow.key = rule->cr.flow;
3413 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
3414 dpif_flow_get(ofproto->dpif, &odp_flow);
3416 if (odp_flow.stats.n_packets) {
3417 update_time(ofproto, rule, &odp_flow.stats);
3418 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
3419 odp_flow.stats.tcp_flags);
3423 expired.flow = rule->cr.flow;
3424 expired.packet_count = rule->packet_count +
3425 odp_flow.stats.n_packets;
3426 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
3427 expired.used = rule->used;
3429 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
3431 /* Schedule us to send the accumulated records once we have
3432 * collected all of them. */
3433 poll_immediate_wake();
3438 update_used(struct ofproto *p)
3440 struct odp_flow *flows;
3445 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
3450 for (i = 0; i < n_flows; i++) {
3451 struct odp_flow *f = &flows[i];
3454 rule = rule_from_cls_rule(
3455 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3456 if (!rule || !rule->installed) {
3457 COVERAGE_INC(ofproto_unexpected_rule);
3458 dpif_flow_del(p->dpif, f);
3462 update_time(p, rule, &f->stats);
3463 rule_account(p, rule, f->stats.n_bytes);
3469 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3470 const struct ofpbuf *packet, int send_len)
3472 struct odp_msg *msg = packet->data;
3473 struct ofpbuf payload;
3477 /* Extract packet payload from 'msg'. */
3478 payload.data = msg + 1;
3479 payload.size = msg->length - sizeof *msg;
3481 /* Construct ofp_packet_in message. */
3482 reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3483 opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port), reason,
3484 &payload, send_len);
3487 rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100);
3491 send_packet_in_action(struct ofpbuf *packet, void *p_)
3493 struct ofproto *p = p_;
3494 struct ofconn *ofconn;
3495 struct odp_msg *msg;
3498 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3499 if (ofconn == p->controller || ofconn->miss_send_len) {
3500 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3503 ofpbuf_delete(packet);
3507 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3509 struct ofproto *p = p_;
3510 bool in_fail_open = p->fail_open && fail_open_is_active(p->fail_open);
3511 struct ofconn *ofconn;
3512 struct ofpbuf payload;
3513 struct odp_msg *msg;
3516 payload.data = msg + 1;
3517 payload.size = msg->length - sizeof *msg;
3518 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3519 if (ofconn->miss_send_len) {
3520 struct pktbuf *pb = ofconn->pktbuf;
3521 uint32_t buffer_id = (in_fail_open
3523 : pktbuf_save(pb, &payload, msg->port));
3524 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3526 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3529 ofpbuf_delete(packet);
3533 pick_datapath_id(const struct ofproto *ofproto)
3535 const struct ofport *port;
3537 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
3539 uint8_t ea[ETH_ADDR_LEN];
3542 error = netdev_get_etheraddr(port->netdev, ea);
3544 return eth_addr_to_uint64(ea);
3546 VLOG_WARN("could not get MAC address for %s (%s)",
3547 netdev_get_name(port->netdev), strerror(error));
3549 return ofproto->fallback_dpid;
3553 pick_fallback_dpid(void)
3555 uint8_t ea[ETH_ADDR_LEN];
3556 eth_addr_random(ea);
3557 ea[0] = 0x00; /* Set Nicira OUI. */
3560 return eth_addr_to_uint64(ea);
3564 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3565 struct odp_actions *actions, tag_type *tags,
3566 uint16_t *nf_output_iface, void *ofproto_)
3568 struct ofproto *ofproto = ofproto_;
3571 /* Drop frames for reserved multicast addresses. */
3572 if (eth_addr_is_reserved(flow->dl_dst)) {
3576 /* Learn source MAC (but don't try to learn from revalidation). */
3577 if (packet != NULL) {
3578 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3581 /* The log messages here could actually be useful in debugging,
3582 * so keep the rate limit relatively high. */
3583 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3584 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3585 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3586 ofproto_revalidate(ofproto, rev_tag);
3590 /* Determine output port. */
3591 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3593 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
3594 } else if (out_port != flow->in_port) {
3595 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3596 *nf_output_iface = out_port;
3604 static const struct ofhooks default_ofhooks = {
3606 default_normal_ofhook_cb,