2 * Copyright (c) 2009 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
22 #include <netinet/in.h>
25 #include "classifier.h"
27 #include "discovery.h"
29 #include "dynamic-string.h"
31 #include "fail-open.h"
33 #include "mac-learning.h"
37 #include "ofp-print.h"
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 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);
766 ofproto_run(struct ofproto *p)
768 int error = ofproto_run1(p);
770 error = ofproto_run2(p, false);
776 process_port_change(struct ofproto *ofproto, int error, char *devname)
778 if (error == ENOBUFS) {
779 reinit_ports(ofproto);
781 update_port(ofproto, devname);
787 ofproto_run1(struct ofproto *p)
789 struct ofconn *ofconn, *next_ofconn;
794 for (i = 0; i < 50; i++) {
798 error = dpif_recv(p->dpif, &buf);
800 if (error == ENODEV) {
801 /* Someone destroyed the datapath behind our back. The caller
802 * better destroy us and give up, because we're just going to
803 * spin from here on out. */
804 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
805 VLOG_ERR_RL(&rl, "%s: datapath was destroyed externally",
812 handle_odp_msg(p, buf);
815 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
816 process_port_change(p, error, devname);
818 while ((error = netdev_monitor_poll(p->netdev_monitor,
819 &devname)) != EAGAIN) {
820 process_port_change(p, error, devname);
824 in_band_run(p->in_band);
827 char *controller_name;
828 if (rconn_is_connectivity_questionable(p->controller->rconn)) {
829 discovery_question_connectivity(p->discovery);
831 if (discovery_run(p->discovery, &controller_name)) {
832 if (controller_name) {
833 rconn_connect(p->controller->rconn, controller_name);
835 rconn_disconnect(p->controller->rconn);
839 pinsched_run(p->miss_sched, send_packet_in_miss, p);
840 pinsched_run(p->action_sched, send_packet_in_action, p);
842 executer_run(p->executer);
845 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
847 ofconn_run(ofconn, p);
850 /* Fail-open maintenance. Do this after processing the ofconns since
851 * fail-open checks the status of the controller rconn. */
853 fail_open_run(p->fail_open);
856 for (i = 0; i < p->n_listeners; i++) {
860 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
862 ofconn_create(p, rconn_new_from_vconn("passive", vconn));
863 } else if (retval != EAGAIN) {
864 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
868 for (i = 0; i < p->n_snoops; i++) {
872 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
874 rconn_add_monitor(p->controller->rconn, vconn);
875 } else if (retval != EAGAIN) {
876 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
880 if (time_msec() >= p->next_expiration) {
881 COVERAGE_INC(ofproto_expiration);
882 p->next_expiration = time_msec() + 1000;
885 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
887 /* Let the hook know that we're at a stable point: all outstanding data
888 * in existing flows has been accounted to the account_cb. Thus, the
889 * hook can now reasonably do operations that depend on having accurate
890 * flow volume accounting (currently, that's just bond rebalancing). */
891 if (p->ofhooks->account_checkpoint_cb) {
892 p->ofhooks->account_checkpoint_cb(p->aux);
897 netflow_run(p->netflow);
900 ofproto_sflow_run(p->sflow);
906 struct revalidate_cbdata {
907 struct ofproto *ofproto;
908 bool revalidate_all; /* Revalidate all exact-match rules? */
909 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
910 struct tag_set revalidate_set; /* Set of tags to revalidate. */
914 ofproto_run2(struct ofproto *p, bool revalidate_all)
916 if (p->need_revalidate || revalidate_all
917 || !tag_set_is_empty(&p->revalidate_set)) {
918 struct revalidate_cbdata cbdata;
920 cbdata.revalidate_all = revalidate_all;
921 cbdata.revalidate_subrules = p->need_revalidate;
922 cbdata.revalidate_set = p->revalidate_set;
923 tag_set_init(&p->revalidate_set);
924 COVERAGE_INC(ofproto_revalidate);
925 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
926 p->need_revalidate = false;
933 ofproto_wait(struct ofproto *p)
935 struct ofconn *ofconn;
938 dpif_recv_wait(p->dpif);
939 dpif_port_poll_wait(p->dpif);
940 netdev_monitor_poll_wait(p->netdev_monitor);
941 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
945 in_band_wait(p->in_band);
948 discovery_wait(p->discovery);
951 fail_open_wait(p->fail_open);
953 pinsched_wait(p->miss_sched);
954 pinsched_wait(p->action_sched);
956 executer_wait(p->executer);
959 ofproto_sflow_wait(p->sflow);
961 if (!tag_set_is_empty(&p->revalidate_set)) {
962 poll_immediate_wake();
964 if (p->need_revalidate) {
965 /* Shouldn't happen, but if it does just go around again. */
966 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
967 poll_immediate_wake();
968 } else if (p->next_expiration != LLONG_MAX) {
969 poll_timer_wait(p->next_expiration - time_msec());
971 for (i = 0; i < p->n_listeners; i++) {
972 pvconn_wait(p->listeners[i]);
974 for (i = 0; i < p->n_snoops; i++) {
975 pvconn_wait(p->snoops[i]);
980 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
982 tag_set_add(&ofproto->revalidate_set, tag);
986 ofproto_get_revalidate_set(struct ofproto *ofproto)
988 return &ofproto->revalidate_set;
992 ofproto_is_alive(const struct ofproto *p)
994 return p->discovery || rconn_is_alive(p->controller->rconn);
998 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
999 const union ofp_action *actions, size_t n_actions,
1000 const struct ofpbuf *packet)
1002 struct odp_actions odp_actions;
1005 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1011 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1013 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
1014 odp_actions.n_actions, packet);
1019 ofproto_add_flow(struct ofproto *p,
1020 const flow_t *flow, uint32_t wildcards, unsigned int priority,
1021 const union ofp_action *actions, size_t n_actions,
1025 rule = rule_create(p, NULL, actions, n_actions,
1026 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */, 0);
1027 cls_rule_from_flow(&rule->cr, flow, wildcards, priority);
1028 rule_insert(p, rule, NULL, 0);
1032 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1033 uint32_t wildcards, unsigned int priority)
1037 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1041 rule_remove(ofproto, rule);
1046 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1048 struct rule *rule = rule_from_cls_rule(rule_);
1049 struct ofproto *ofproto = ofproto_;
1051 /* Mark the flow as not installed, even though it might really be
1052 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1053 * There is no point in uninstalling it individually since we are about to
1054 * blow away all the flows with dpif_flow_flush(). */
1055 rule->installed = false;
1057 rule_remove(ofproto, rule);
1061 ofproto_flush_flows(struct ofproto *ofproto)
1063 COVERAGE_INC(ofproto_flush);
1064 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1065 dpif_flow_flush(ofproto->dpif);
1066 if (ofproto->in_band) {
1067 in_band_flushed(ofproto->in_band);
1069 if (ofproto->fail_open) {
1070 fail_open_flushed(ofproto->fail_open);
1075 reinit_ports(struct ofproto *p)
1077 struct svec devnames;
1078 struct ofport *ofport;
1079 unsigned int port_no;
1080 struct odp_port *odp_ports;
1084 svec_init(&devnames);
1085 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1086 svec_add (&devnames, (char *) ofport->opp.name);
1088 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1089 for (i = 0; i < n_odp_ports; i++) {
1090 svec_add (&devnames, odp_ports[i].devname);
1094 svec_sort_unique(&devnames);
1095 for (i = 0; i < devnames.n; i++) {
1096 update_port(p, devnames.names[i]);
1098 svec_destroy(&devnames);
1102 refresh_port_group(struct ofproto *p, unsigned int group)
1106 struct ofport *port;
1107 unsigned int port_no;
1109 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1111 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1113 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1114 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1115 ports[n_ports++] = port_no;
1118 dpif_port_group_set(p->dpif, group, ports, n_ports);
1125 refresh_port_groups(struct ofproto *p)
1127 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1128 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1130 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1134 static struct ofport *
1135 make_ofport(const struct odp_port *odp_port)
1137 enum netdev_flags flags;
1138 struct ofport *ofport;
1139 struct netdev *netdev;
1143 error = netdev_open(odp_port->devname, NETDEV_ETH_TYPE_NONE, &netdev);
1145 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1146 "cannot be opened (%s)",
1147 odp_port->devname, odp_port->port,
1148 odp_port->devname, strerror(error));
1152 ofport = xmalloc(sizeof *ofport);
1153 ofport->netdev = netdev;
1154 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1155 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1156 memcpy(ofport->opp.name, odp_port->devname,
1157 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1158 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1160 netdev_get_flags(netdev, &flags);
1161 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1163 netdev_get_carrier(netdev, &carrier);
1164 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1166 netdev_get_features(netdev,
1167 &ofport->opp.curr, &ofport->opp.advertised,
1168 &ofport->opp.supported, &ofport->opp.peer);
1173 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1175 if (port_array_get(&p->ports, odp_port->port)) {
1176 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1179 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1180 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1189 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1191 const struct ofp_phy_port *a = &a_->opp;
1192 const struct ofp_phy_port *b = &b_->opp;
1194 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1195 return (a->port_no == b->port_no
1196 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1197 && !strcmp((char *) a->name, (char *) b->name)
1198 && a->state == b->state
1199 && a->config == b->config
1200 && a->curr == b->curr
1201 && a->advertised == b->advertised
1202 && a->supported == b->supported
1203 && a->peer == b->peer);
1207 send_port_status(struct ofproto *p, const struct ofport *ofport,
1210 /* XXX Should limit the number of queued port status change messages. */
1211 struct ofconn *ofconn;
1212 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1213 struct ofp_port_status *ops;
1216 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1217 ops->reason = reason;
1218 ops->desc = ofport->opp;
1219 hton_ofp_phy_port(&ops->desc);
1220 queue_tx(b, ofconn, NULL);
1222 if (p->ofhooks->port_changed_cb) {
1223 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1228 ofport_install(struct ofproto *p, struct ofport *ofport)
1230 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1231 const char *netdev_name = (const char *) ofport->opp.name;
1233 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1234 port_array_set(&p->ports, odp_port, ofport);
1235 shash_add(&p->port_by_name, netdev_name, ofport);
1237 ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
1242 ofport_remove(struct ofproto *p, struct ofport *ofport)
1244 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1246 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1247 port_array_set(&p->ports, odp_port, NULL);
1248 shash_delete(&p->port_by_name,
1249 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1251 ofproto_sflow_del_port(p->sflow, odp_port);
1256 ofport_free(struct ofport *ofport)
1259 netdev_close(ofport->netdev);
1265 update_port(struct ofproto *p, const char *devname)
1267 struct odp_port odp_port;
1268 struct ofport *old_ofport;
1269 struct ofport *new_ofport;
1272 COVERAGE_INC(ofproto_update_port);
1274 /* Query the datapath for port information. */
1275 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1277 /* Find the old ofport. */
1278 old_ofport = shash_find_data(&p->port_by_name, devname);
1281 /* There's no port named 'devname' but there might be a port with
1282 * the same port number. This could happen if a port is deleted
1283 * and then a new one added in its place very quickly, or if a port
1284 * is renamed. In the former case we want to send an OFPPR_DELETE
1285 * and an OFPPR_ADD, and in the latter case we want to send a
1286 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1287 * the old port's ifindex against the new port, or perhaps less
1288 * reliably but more portably by comparing the old port's MAC
1289 * against the new port's MAC. However, this code isn't that smart
1290 * and always sends an OFPPR_MODIFY (XXX). */
1291 old_ofport = port_array_get(&p->ports, odp_port.port);
1293 } else if (error != ENOENT && error != ENODEV) {
1294 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1295 "%s", strerror(error));
1299 /* Create a new ofport. */
1300 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1302 /* Eliminate a few pathological cases. */
1303 if (!old_ofport && !new_ofport) {
1305 } else if (old_ofport && new_ofport) {
1306 /* Most of the 'config' bits are OpenFlow soft state, but
1307 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1308 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1309 * leaves the other bits 0.) */
1310 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1312 if (ofport_equal(old_ofport, new_ofport)) {
1313 /* False alarm--no change. */
1314 ofport_free(new_ofport);
1319 /* Now deal with the normal cases. */
1321 ofport_remove(p, old_ofport);
1324 ofport_install(p, new_ofport);
1326 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1327 (!old_ofport ? OFPPR_ADD
1328 : !new_ofport ? OFPPR_DELETE
1330 ofport_free(old_ofport);
1332 /* Update port groups. */
1333 refresh_port_groups(p);
1337 init_ports(struct ofproto *p)
1339 struct odp_port *ports;
1344 error = dpif_port_list(p->dpif, &ports, &n_ports);
1349 for (i = 0; i < n_ports; i++) {
1350 const struct odp_port *odp_port = &ports[i];
1351 if (!ofport_conflicts(p, odp_port)) {
1352 struct ofport *ofport = make_ofport(odp_port);
1354 ofport_install(p, ofport);
1359 refresh_port_groups(p);
1363 static struct ofconn *
1364 ofconn_create(struct ofproto *p, struct rconn *rconn)
1366 struct ofconn *ofconn = xmalloc(sizeof *ofconn);
1367 list_push_back(&p->all_conns, &ofconn->node);
1368 ofconn->rconn = rconn;
1369 ofconn->pktbuf = NULL;
1370 ofconn->send_flow_exp = false;
1371 ofconn->miss_send_len = 0;
1372 ofconn->packet_in_counter = rconn_packet_counter_create ();
1373 ofconn->reply_counter = rconn_packet_counter_create ();
1378 ofconn_destroy(struct ofconn *ofconn, struct ofproto *p)
1381 executer_rconn_closing(p->executer, ofconn->rconn);
1384 list_remove(&ofconn->node);
1385 rconn_destroy(ofconn->rconn);
1386 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1387 rconn_packet_counter_destroy(ofconn->reply_counter);
1388 pktbuf_destroy(ofconn->pktbuf);
1393 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1397 rconn_run(ofconn->rconn);
1399 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1400 /* Limit the number of iterations to prevent other tasks from
1402 for (iteration = 0; iteration < 50; iteration++) {
1403 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1408 fail_open_maybe_recover(p->fail_open);
1410 handle_openflow(ofconn, p, of_msg);
1411 ofpbuf_delete(of_msg);
1415 if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
1416 ofconn_destroy(ofconn, p);
1421 ofconn_wait(struct ofconn *ofconn)
1423 rconn_run_wait(ofconn->rconn);
1424 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1425 rconn_recv_wait(ofconn->rconn);
1427 COVERAGE_INC(ofproto_ofconn_stuck);
1431 /* Caller is responsible for initializing the 'cr' member of the returned
1433 static struct rule *
1434 rule_create(struct ofproto *ofproto, struct rule *super,
1435 const union ofp_action *actions, size_t n_actions,
1436 uint16_t idle_timeout, uint16_t hard_timeout)
1438 struct rule *rule = xcalloc(1, sizeof *rule);
1439 rule->idle_timeout = idle_timeout;
1440 rule->hard_timeout = hard_timeout;
1441 rule->used = rule->created = time_msec();
1442 rule->super = super;
1444 list_push_back(&super->list, &rule->list);
1446 list_init(&rule->list);
1448 rule->n_actions = n_actions;
1449 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1450 netflow_flow_clear(&rule->nf_flow);
1451 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1456 static struct rule *
1457 rule_from_cls_rule(const struct cls_rule *cls_rule)
1459 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1463 rule_free(struct rule *rule)
1465 free(rule->actions);
1466 free(rule->odp_actions);
1470 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1471 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1472 * through all of its subrules and revalidates them, destroying any that no
1473 * longer has a super-rule (which is probably all of them).
1475 * Before calling this function, the caller must make have removed 'rule' from
1476 * the classifier. If 'rule' is an exact-match rule, the caller is also
1477 * responsible for ensuring that it has been uninstalled from the datapath. */
1479 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1482 struct rule *subrule, *next;
1483 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1484 revalidate_rule(ofproto, subrule);
1487 list_remove(&rule->list);
1493 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1495 const union ofp_action *oa;
1496 struct actions_iterator i;
1498 if (out_port == htons(OFPP_NONE)) {
1501 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1502 oa = actions_next(&i)) {
1503 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1510 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1511 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1513 * The flow that 'packet' actually contains does not need to actually match
1514 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1515 * the packet and byte counters for 'rule' will be credited for the packet sent
1516 * out whether or not the packet actually matches 'rule'.
1518 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1519 * the caller must already have accurately composed ODP actions for it given
1520 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1521 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1522 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1523 * actions and apply them to 'packet'. */
1525 rule_execute(struct ofproto *ofproto, struct rule *rule,
1526 struct ofpbuf *packet, const flow_t *flow)
1528 const union odp_action *actions;
1530 struct odp_actions a;
1532 /* Grab or compose the ODP actions.
1534 * The special case for an exact-match 'rule' where 'flow' is not the
1535 * rule's flow is important to avoid, e.g., sending a packet out its input
1536 * port simply because the ODP actions were composed for the wrong
1538 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1539 struct rule *super = rule->super ? rule->super : rule;
1540 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1541 packet, &a, NULL, 0, NULL)) {
1544 actions = a.actions;
1545 n_actions = a.n_actions;
1547 actions = rule->odp_actions;
1548 n_actions = rule->n_odp_actions;
1551 /* Execute the ODP actions. */
1552 if (!dpif_execute(ofproto->dpif, flow->in_port,
1553 actions, n_actions, packet)) {
1554 struct odp_flow_stats stats;
1555 flow_extract_stats(flow, packet, &stats);
1556 update_stats(ofproto, rule, &stats);
1557 rule->used = time_msec();
1558 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1563 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1566 struct rule *displaced_rule;
1568 /* Insert the rule in the classifier. */
1569 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1570 if (!rule->cr.wc.wildcards) {
1571 rule_make_actions(p, rule, packet);
1574 /* Send the packet and credit it to the rule. */
1577 flow_extract(packet, in_port, &flow);
1578 rule_execute(p, rule, packet, &flow);
1581 /* Install the rule in the datapath only after sending the packet, to
1582 * avoid packet reordering. */
1583 if (rule->cr.wc.wildcards) {
1584 COVERAGE_INC(ofproto_add_wc_flow);
1585 p->need_revalidate = true;
1587 rule_install(p, rule, displaced_rule);
1590 /* Free the rule that was displaced, if any. */
1591 if (displaced_rule) {
1592 rule_destroy(p, displaced_rule);
1596 static struct rule *
1597 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1600 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1601 rule->idle_timeout, rule->hard_timeout);
1602 COVERAGE_INC(ofproto_subrule_create);
1603 cls_rule_from_flow(&subrule->cr, flow, 0,
1604 (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1605 : rule->cr.priority));
1606 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1612 rule_remove(struct ofproto *ofproto, struct rule *rule)
1614 if (rule->cr.wc.wildcards) {
1615 COVERAGE_INC(ofproto_del_wc_flow);
1616 ofproto->need_revalidate = true;
1618 rule_uninstall(ofproto, rule);
1620 classifier_remove(&ofproto->cls, &rule->cr);
1621 rule_destroy(ofproto, rule);
1624 /* Returns true if the actions changed, false otherwise. */
1626 rule_make_actions(struct ofproto *p, struct rule *rule,
1627 const struct ofpbuf *packet)
1629 const struct rule *super;
1630 struct odp_actions a;
1633 assert(!rule->cr.wc.wildcards);
1635 super = rule->super ? rule->super : rule;
1637 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1638 packet, &a, &rule->tags, &rule->may_install,
1639 &rule->nf_flow.output_iface);
1641 actions_len = a.n_actions * sizeof *a.actions;
1642 if (rule->n_odp_actions != a.n_actions
1643 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1644 COVERAGE_INC(ofproto_odp_unchanged);
1645 free(rule->odp_actions);
1646 rule->n_odp_actions = a.n_actions;
1647 rule->odp_actions = xmemdup(a.actions, actions_len);
1655 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1656 struct odp_flow_put *put)
1658 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1659 put->flow.key = rule->cr.flow;
1660 put->flow.actions = rule->odp_actions;
1661 put->flow.n_actions = rule->n_odp_actions;
1663 return dpif_flow_put(ofproto->dpif, put);
1667 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1669 assert(!rule->cr.wc.wildcards);
1671 if (rule->may_install) {
1672 struct odp_flow_put put;
1673 if (!do_put_flow(p, rule,
1674 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1676 rule->installed = true;
1677 if (displaced_rule) {
1678 update_stats(p, displaced_rule, &put.flow.stats);
1679 rule_post_uninstall(p, displaced_rule);
1682 } else if (displaced_rule) {
1683 rule_uninstall(p, displaced_rule);
1688 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1690 if (rule->installed) {
1691 struct odp_flow_put put;
1692 COVERAGE_INC(ofproto_dp_missed);
1693 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1695 rule_install(ofproto, rule, NULL);
1700 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1702 bool actions_changed;
1703 uint16_t new_out_iface, old_out_iface;
1705 old_out_iface = rule->nf_flow.output_iface;
1706 actions_changed = rule_make_actions(ofproto, rule, NULL);
1708 if (rule->may_install) {
1709 if (rule->installed) {
1710 if (actions_changed) {
1711 struct odp_flow_put put;
1712 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
1713 | ODPPF_ZERO_STATS, &put);
1714 update_stats(ofproto, rule, &put.flow.stats);
1716 /* Temporarily set the old output iface so that NetFlow
1717 * messages have the correct output interface for the old
1719 new_out_iface = rule->nf_flow.output_iface;
1720 rule->nf_flow.output_iface = old_out_iface;
1721 rule_post_uninstall(ofproto, rule);
1722 rule->nf_flow.output_iface = new_out_iface;
1725 rule_install(ofproto, rule, NULL);
1728 rule_uninstall(ofproto, rule);
1733 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1735 uint64_t total_bytes = rule->byte_count + extra_bytes;
1737 if (ofproto->ofhooks->account_flow_cb
1738 && total_bytes > rule->accounted_bytes)
1740 ofproto->ofhooks->account_flow_cb(
1741 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1742 total_bytes - rule->accounted_bytes, ofproto->aux);
1743 rule->accounted_bytes = total_bytes;
1748 rule_uninstall(struct ofproto *p, struct rule *rule)
1750 assert(!rule->cr.wc.wildcards);
1751 if (rule->installed) {
1752 struct odp_flow odp_flow;
1754 odp_flow.key = rule->cr.flow;
1755 odp_flow.actions = NULL;
1756 odp_flow.n_actions = 0;
1757 if (!dpif_flow_del(p->dpif, &odp_flow)) {
1758 update_stats(p, rule, &odp_flow.stats);
1760 rule->installed = false;
1762 rule_post_uninstall(p, rule);
1767 is_controller_rule(struct rule *rule)
1769 /* If the only action is send to the controller then don't report
1770 * NetFlow expiration messages since it is just part of the control
1771 * logic for the network and not real traffic. */
1773 if (rule && rule->super) {
1774 struct rule *super = rule->super;
1776 return super->n_actions == 1 &&
1777 super->actions[0].type == htons(OFPAT_OUTPUT) &&
1778 super->actions[0].output.port == htons(OFPP_CONTROLLER);
1785 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1787 struct rule *super = rule->super;
1789 rule_account(ofproto, rule, 0);
1791 if (ofproto->netflow && !is_controller_rule(rule)) {
1792 struct ofexpired expired;
1793 expired.flow = rule->cr.flow;
1794 expired.packet_count = rule->packet_count;
1795 expired.byte_count = rule->byte_count;
1796 expired.used = rule->used;
1797 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
1800 super->packet_count += rule->packet_count;
1801 super->byte_count += rule->byte_count;
1803 /* Reset counters to prevent double counting if the rule ever gets
1805 rule->packet_count = 0;
1806 rule->byte_count = 0;
1807 rule->accounted_bytes = 0;
1809 netflow_flow_clear(&rule->nf_flow);
1814 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1815 struct rconn_packet_counter *counter)
1817 update_openflow_length(msg);
1818 if (rconn_send(ofconn->rconn, msg, counter)) {
1824 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1825 int error, const void *data, size_t len)
1828 struct ofp_error_msg *oem;
1830 if (!(error >> 16)) {
1831 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1836 COVERAGE_INC(ofproto_error);
1837 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1838 oh ? oh->xid : 0, &buf);
1839 oem->type = htons((unsigned int) error >> 16);
1840 oem->code = htons(error & 0xffff);
1841 memcpy(oem->data, data, len);
1842 queue_tx(buf, ofconn, ofconn->reply_counter);
1846 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1849 size_t oh_length = ntohs(oh->length);
1850 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1854 hton_ofp_phy_port(struct ofp_phy_port *opp)
1856 opp->port_no = htons(opp->port_no);
1857 opp->config = htonl(opp->config);
1858 opp->state = htonl(opp->state);
1859 opp->curr = htonl(opp->curr);
1860 opp->advertised = htonl(opp->advertised);
1861 opp->supported = htonl(opp->supported);
1862 opp->peer = htonl(opp->peer);
1866 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
1868 struct ofp_header *rq = oh;
1869 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
1874 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
1875 struct ofp_header *oh)
1877 struct ofp_switch_features *osf;
1879 unsigned int port_no;
1880 struct ofport *port;
1882 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1883 osf->datapath_id = htonll(p->datapath_id);
1884 osf->n_buffers = htonl(pktbuf_capacity());
1886 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1887 OFPC_PORT_STATS | OFPC_MULTI_PHY_TX);
1888 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1889 (1u << OFPAT_SET_VLAN_VID) |
1890 (1u << OFPAT_SET_VLAN_PCP) |
1891 (1u << OFPAT_STRIP_VLAN) |
1892 (1u << OFPAT_SET_DL_SRC) |
1893 (1u << OFPAT_SET_DL_DST) |
1894 (1u << OFPAT_SET_NW_SRC) |
1895 (1u << OFPAT_SET_NW_DST) |
1896 (1u << OFPAT_SET_TP_SRC) |
1897 (1u << OFPAT_SET_TP_DST));
1899 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1900 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1903 queue_tx(buf, ofconn, ofconn->reply_counter);
1908 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1909 struct ofp_header *oh)
1912 struct ofp_switch_config *osc;
1916 /* Figure out flags. */
1917 dpif_get_drop_frags(p->dpif, &drop_frags);
1918 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1919 if (ofconn->send_flow_exp) {
1920 flags |= OFPC_SEND_FLOW_EXP;
1924 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1925 osc->flags = htons(flags);
1926 osc->miss_send_len = htons(ofconn->miss_send_len);
1927 queue_tx(buf, ofconn, ofconn->reply_counter);
1933 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1934 struct ofp_switch_config *osc)
1939 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1943 flags = ntohs(osc->flags);
1945 ofconn->send_flow_exp = (flags & OFPC_SEND_FLOW_EXP) != 0;
1947 if (ofconn == p->controller) {
1948 switch (flags & OFPC_FRAG_MASK) {
1949 case OFPC_FRAG_NORMAL:
1950 dpif_set_drop_frags(p->dpif, false);
1952 case OFPC_FRAG_DROP:
1953 dpif_set_drop_frags(p->dpif, true);
1956 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1962 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1963 if (ntohs(osc->miss_send_len) != 0) {
1964 ofconn->pktbuf = pktbuf_create();
1966 pktbuf_destroy(ofconn->pktbuf);
1970 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1976 add_output_group_action(struct odp_actions *actions, uint16_t group,
1977 uint16_t *nf_output_iface)
1979 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1981 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
1982 *nf_output_iface = NF_OUT_FLOOD;
1987 add_controller_action(struct odp_actions *actions,
1988 const struct ofp_action_output *oao)
1990 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
1991 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
1994 struct action_xlate_ctx {
1996 const flow_t *flow; /* Flow to which these actions correspond. */
1997 int recurse; /* Recursion level, via xlate_table_action. */
1998 struct ofproto *ofproto;
1999 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2000 * null pointer if we are revalidating
2001 * without a packet to refer to. */
2004 struct odp_actions *out; /* Datapath actions. */
2005 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2006 bool may_set_up_flow; /* True ordinarily; false if the actions must
2007 * be reassessed for every packet. */
2008 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2011 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2012 struct action_xlate_ctx *ctx);
2015 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2017 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
2020 if (ofport->opp.config & OFPPC_NO_FWD) {
2021 /* Forwarding disabled on port. */
2026 * We don't have an ofport record for this port, but it doesn't hurt to
2027 * allow forwarding to it anyhow. Maybe such a port will appear later
2028 * and we're pre-populating the flow table.
2032 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2033 ctx->nf_output_iface = port;
2036 static struct rule *
2037 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2040 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2042 /* The rule we found might not be valid, since we could be in need of
2043 * revalidation. If it is not valid, don't return it. */
2046 && ofproto->need_revalidate
2047 && !revalidate_rule(ofproto, rule)) {
2048 COVERAGE_INC(ofproto_invalidated);
2056 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2058 if (!ctx->recurse) {
2063 flow.in_port = in_port;
2065 rule = lookup_valid_rule(ctx->ofproto, &flow);
2072 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2079 xlate_output_action(struct action_xlate_ctx *ctx,
2080 const struct ofp_action_output *oao)
2083 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2085 ctx->nf_output_iface = NF_OUT_DROP;
2087 switch (ntohs(oao->port)) {
2089 add_output_action(ctx, ctx->flow->in_port);
2092 xlate_table_action(ctx, ctx->flow->in_port);
2095 if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet,
2096 ctx->out, ctx->tags,
2097 &ctx->nf_output_iface,
2098 ctx->ofproto->aux)) {
2099 COVERAGE_INC(ofproto_uninstallable);
2100 ctx->may_set_up_flow = false;
2104 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2105 &ctx->nf_output_iface);
2108 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2110 case OFPP_CONTROLLER:
2111 add_controller_action(ctx->out, oao);
2114 add_output_action(ctx, ODPP_LOCAL);
2117 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2118 if (odp_port != ctx->flow->in_port) {
2119 add_output_action(ctx, odp_port);
2124 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2125 ctx->nf_output_iface = NF_OUT_FLOOD;
2126 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2127 ctx->nf_output_iface = prev_nf_output_iface;
2128 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2129 ctx->nf_output_iface != NF_OUT_FLOOD) {
2130 ctx->nf_output_iface = NF_OUT_MULTI;
2135 xlate_nicira_action(struct action_xlate_ctx *ctx,
2136 const struct nx_action_header *nah)
2138 const struct nx_action_resubmit *nar;
2139 int subtype = ntohs(nah->subtype);
2141 assert(nah->vendor == htonl(NX_VENDOR_ID));
2143 case NXAST_RESUBMIT:
2144 nar = (const struct nx_action_resubmit *) nah;
2145 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2149 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2155 do_xlate_actions(const union ofp_action *in, size_t n_in,
2156 struct action_xlate_ctx *ctx)
2158 struct actions_iterator iter;
2159 const union ofp_action *ia;
2160 const struct ofport *port;
2162 port = port_array_get(&ctx->ofproto->ports, ctx->flow->in_port);
2163 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2164 port->opp.config & (eth_addr_equals(ctx->flow->dl_dst, stp_eth_addr)
2165 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2166 /* Drop this flow. */
2170 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2171 uint16_t type = ntohs(ia->type);
2172 union odp_action *oa;
2176 xlate_output_action(ctx, &ia->output);
2179 case OFPAT_SET_VLAN_VID:
2180 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2181 oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2184 case OFPAT_SET_VLAN_PCP:
2185 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2186 oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2189 case OFPAT_STRIP_VLAN:
2190 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2193 case OFPAT_SET_DL_SRC:
2194 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2195 memcpy(oa->dl_addr.dl_addr,
2196 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2199 case OFPAT_SET_DL_DST:
2200 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2201 memcpy(oa->dl_addr.dl_addr,
2202 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2205 case OFPAT_SET_NW_SRC:
2206 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2207 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2210 case OFPAT_SET_NW_DST:
2211 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2212 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2215 case OFPAT_SET_TP_SRC:
2216 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2217 oa->tp_port.tp_port = ia->tp_port.tp_port;
2220 case OFPAT_SET_TP_DST:
2221 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2222 oa->tp_port.tp_port = ia->tp_port.tp_port;
2226 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2230 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2237 xlate_actions(const union ofp_action *in, size_t n_in,
2238 const flow_t *flow, struct ofproto *ofproto,
2239 const struct ofpbuf *packet,
2240 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2241 uint16_t *nf_output_iface)
2243 tag_type no_tags = 0;
2244 struct action_xlate_ctx ctx;
2245 COVERAGE_INC(ofproto_ofp2odp);
2246 odp_actions_init(out);
2249 ctx.ofproto = ofproto;
2250 ctx.packet = packet;
2252 ctx.tags = tags ? tags : &no_tags;
2253 ctx.may_set_up_flow = true;
2254 ctx.nf_output_iface = NF_OUT_DROP;
2255 do_xlate_actions(in, n_in, &ctx);
2257 /* Check with in-band control to see if we're allowed to set up this
2259 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2260 ctx.may_set_up_flow = false;
2263 if (may_set_up_flow) {
2264 *may_set_up_flow = ctx.may_set_up_flow;
2266 if (nf_output_iface) {
2267 *nf_output_iface = ctx.nf_output_iface;
2269 if (odp_actions_overflow(out)) {
2270 odp_actions_init(out);
2271 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2277 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2278 struct ofp_header *oh)
2280 struct ofp_packet_out *opo;
2281 struct ofpbuf payload, *buffer;
2282 struct odp_actions actions;
2288 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2292 opo = (struct ofp_packet_out *) oh;
2294 COVERAGE_INC(ofproto_packet_out);
2295 if (opo->buffer_id != htonl(UINT32_MAX)) {
2296 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2298 if (error || !buffer) {
2306 flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2307 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2308 &flow, p, &payload, &actions, NULL, NULL, NULL);
2313 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2315 ofpbuf_delete(buffer);
2321 update_port_config(struct ofproto *p, struct ofport *port,
2322 uint32_t config, uint32_t mask)
2324 mask &= config ^ port->opp.config;
2325 if (mask & OFPPC_PORT_DOWN) {
2326 if (config & OFPPC_PORT_DOWN) {
2327 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2329 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2332 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2333 if (mask & REVALIDATE_BITS) {
2334 COVERAGE_INC(ofproto_costly_flags);
2335 port->opp.config ^= mask & REVALIDATE_BITS;
2336 p->need_revalidate = true;
2338 #undef REVALIDATE_BITS
2339 if (mask & OFPPC_NO_FLOOD) {
2340 port->opp.config ^= OFPPC_NO_FLOOD;
2341 refresh_port_groups(p);
2343 if (mask & OFPPC_NO_PACKET_IN) {
2344 port->opp.config ^= OFPPC_NO_PACKET_IN;
2349 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2351 const struct ofp_port_mod *opm;
2352 struct ofport *port;
2355 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2359 opm = (struct ofp_port_mod *) oh;
2361 port = port_array_get(&p->ports,
2362 ofp_port_to_odp_port(ntohs(opm->port_no)));
2364 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2365 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2366 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2368 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2369 if (opm->advertise) {
2370 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2376 static struct ofpbuf *
2377 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2379 struct ofp_stats_reply *osr;
2382 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2383 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2385 osr->flags = htons(0);
2389 static struct ofpbuf *
2390 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2392 return make_stats_reply(request->header.xid, request->type, body_len);
2396 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2398 struct ofpbuf *msg = *msgp;
2399 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2400 if (nbytes + msg->size > UINT16_MAX) {
2401 struct ofp_stats_reply *reply = msg->data;
2402 reply->flags = htons(OFPSF_REPLY_MORE);
2403 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2404 queue_tx(msg, ofconn, ofconn->reply_counter);
2406 return ofpbuf_put_uninit(*msgp, nbytes);
2410 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2411 struct ofp_stats_request *request)
2413 struct ofp_desc_stats *ods;
2416 msg = start_stats_reply(request, sizeof *ods);
2417 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2418 strncpy(ods->mfr_desc, p->manufacturer, sizeof ods->mfr_desc);
2419 strncpy(ods->hw_desc, p->hardware, sizeof ods->hw_desc);
2420 strncpy(ods->sw_desc, p->software, sizeof ods->sw_desc);
2421 strncpy(ods->serial_num, p->serial, sizeof ods->serial_num);
2422 queue_tx(msg, ofconn, ofconn->reply_counter);
2428 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2430 struct rule *rule = rule_from_cls_rule(cls_rule);
2431 int *n_subrules = n_subrules_;
2439 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2440 struct ofp_stats_request *request)
2442 struct ofp_table_stats *ots;
2444 struct odp_stats dpstats;
2445 int n_exact, n_subrules, n_wild;
2447 msg = start_stats_reply(request, sizeof *ots * 2);
2449 /* Count rules of various kinds. */
2451 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2452 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2453 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2456 dpif_get_dp_stats(p->dpif, &dpstats);
2457 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2458 memset(ots, 0, sizeof *ots);
2459 ots->table_id = TABLEID_HASH;
2460 strcpy(ots->name, "hash");
2461 ots->wildcards = htonl(0);
2462 ots->max_entries = htonl(dpstats.max_capacity);
2463 ots->active_count = htonl(n_exact);
2464 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2466 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2468 /* Classifier table. */
2469 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2470 memset(ots, 0, sizeof *ots);
2471 ots->table_id = TABLEID_CLASSIFIER;
2472 strcpy(ots->name, "classifier");
2473 ots->wildcards = htonl(OFPFW_ALL);
2474 ots->max_entries = htonl(65536);
2475 ots->active_count = htonl(n_wild);
2476 ots->lookup_count = htonll(0); /* XXX */
2477 ots->matched_count = htonll(0); /* XXX */
2479 queue_tx(msg, ofconn, ofconn->reply_counter);
2484 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2485 struct ofp_stats_request *request)
2487 struct ofp_port_stats *ops;
2489 struct ofport *port;
2490 unsigned int port_no;
2492 msg = start_stats_reply(request, sizeof *ops * 16);
2493 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2494 struct netdev_stats stats;
2496 /* Intentionally ignore return value, since errors will set 'stats' to
2497 * all-1s, which is correct for OpenFlow, and netdev_get_stats() will
2499 netdev_get_stats(port->netdev, &stats);
2501 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2502 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2503 memset(ops->pad, 0, sizeof ops->pad);
2504 ops->rx_packets = htonll(stats.rx_packets);
2505 ops->tx_packets = htonll(stats.tx_packets);
2506 ops->rx_bytes = htonll(stats.rx_bytes);
2507 ops->tx_bytes = htonll(stats.tx_bytes);
2508 ops->rx_dropped = htonll(stats.rx_dropped);
2509 ops->tx_dropped = htonll(stats.tx_dropped);
2510 ops->rx_errors = htonll(stats.rx_errors);
2511 ops->tx_errors = htonll(stats.tx_errors);
2512 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2513 ops->rx_over_err = htonll(stats.rx_over_errors);
2514 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2515 ops->collisions = htonll(stats.collisions);
2518 queue_tx(msg, ofconn, ofconn->reply_counter);
2522 struct flow_stats_cbdata {
2523 struct ofproto *ofproto;
2524 struct ofconn *ofconn;
2530 query_stats(struct ofproto *p, struct rule *rule,
2531 uint64_t *packet_countp, uint64_t *byte_countp)
2533 uint64_t packet_count, byte_count;
2534 struct rule *subrule;
2535 struct odp_flow *odp_flows;
2538 packet_count = rule->packet_count;
2539 byte_count = rule->byte_count;
2541 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2542 odp_flows = xcalloc(1, n_odp_flows * sizeof *odp_flows);
2543 if (rule->cr.wc.wildcards) {
2545 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2546 odp_flows[i++].key = subrule->cr.flow;
2547 packet_count += subrule->packet_count;
2548 byte_count += subrule->byte_count;
2551 odp_flows[0].key = rule->cr.flow;
2554 packet_count = rule->packet_count;
2555 byte_count = rule->byte_count;
2556 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2558 for (i = 0; i < n_odp_flows; i++) {
2559 struct odp_flow *odp_flow = &odp_flows[i];
2560 packet_count += odp_flow->stats.n_packets;
2561 byte_count += odp_flow->stats.n_bytes;
2566 *packet_countp = packet_count;
2567 *byte_countp = byte_count;
2571 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2573 struct rule *rule = rule_from_cls_rule(rule_);
2574 struct flow_stats_cbdata *cbdata = cbdata_;
2575 struct ofp_flow_stats *ofs;
2576 uint64_t packet_count, byte_count;
2577 size_t act_len, len;
2579 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2583 act_len = sizeof *rule->actions * rule->n_actions;
2584 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2586 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2588 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2589 ofs->length = htons(len);
2590 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2592 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match);
2593 ofs->duration = htonl((time_msec() - rule->created) / 1000);
2594 ofs->priority = htons(rule->cr.priority);
2595 ofs->idle_timeout = htons(rule->idle_timeout);
2596 ofs->hard_timeout = htons(rule->hard_timeout);
2597 memset(ofs->pad2, 0, sizeof ofs->pad2);
2598 ofs->packet_count = htonll(packet_count);
2599 ofs->byte_count = htonll(byte_count);
2600 memcpy(ofs->actions, rule->actions, act_len);
2604 table_id_to_include(uint8_t table_id)
2606 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2607 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2608 : table_id == 0xff ? CLS_INC_ALL
2613 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2614 const struct ofp_stats_request *osr,
2617 struct ofp_flow_stats_request *fsr;
2618 struct flow_stats_cbdata cbdata;
2619 struct cls_rule target;
2621 if (arg_size != sizeof *fsr) {
2622 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2624 fsr = (struct ofp_flow_stats_request *) osr->body;
2626 COVERAGE_INC(ofproto_flows_req);
2628 cbdata.ofconn = ofconn;
2629 cbdata.out_port = fsr->out_port;
2630 cbdata.msg = start_stats_reply(osr, 1024);
2631 cls_rule_from_match(&target, &fsr->match, 0);
2632 classifier_for_each_match(&p->cls, &target,
2633 table_id_to_include(fsr->table_id),
2634 flow_stats_cb, &cbdata);
2635 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2639 struct flow_stats_ds_cbdata {
2640 struct ofproto *ofproto;
2645 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
2647 struct rule *rule = rule_from_cls_rule(rule_);
2648 struct flow_stats_ds_cbdata *cbdata = cbdata_;
2649 struct ds *results = cbdata->results;
2650 struct ofp_match match;
2651 uint64_t packet_count, byte_count;
2652 size_t act_len = sizeof *rule->actions * rule->n_actions;
2654 /* Don't report on subrules. */
2655 if (rule->super != NULL) {
2659 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2660 flow_to_ovs_match(&rule->cr.flow, rule->cr.wc.wildcards, &match);
2662 ds_put_format(results, "duration=%llds, ",
2663 (time_msec() - rule->created) / 1000);
2664 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2665 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2666 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2667 ofp_print_match(results, &match, true);
2668 ofp_print_actions(results, &rule->actions->header, act_len);
2669 ds_put_cstr(results, "\n");
2672 /* Adds a pretty-printed description of all flows to 'results', including
2673 * those marked hidden by secchan (e.g., by in-band control). */
2675 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2677 struct ofp_match match;
2678 struct cls_rule target;
2679 struct flow_stats_ds_cbdata cbdata;
2681 memset(&match, 0, sizeof match);
2682 match.wildcards = htonl(OFPFW_ALL);
2685 cbdata.results = results;
2687 cls_rule_from_match(&target, &match, 0);
2688 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2689 flow_stats_ds_cb, &cbdata);
2692 struct aggregate_stats_cbdata {
2693 struct ofproto *ofproto;
2695 uint64_t packet_count;
2696 uint64_t byte_count;
2701 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2703 struct rule *rule = rule_from_cls_rule(rule_);
2704 struct aggregate_stats_cbdata *cbdata = cbdata_;
2705 uint64_t packet_count, byte_count;
2707 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2711 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2713 cbdata->packet_count += packet_count;
2714 cbdata->byte_count += byte_count;
2719 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2720 const struct ofp_stats_request *osr,
2723 struct ofp_aggregate_stats_request *asr;
2724 struct ofp_aggregate_stats_reply *reply;
2725 struct aggregate_stats_cbdata cbdata;
2726 struct cls_rule target;
2729 if (arg_size != sizeof *asr) {
2730 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2732 asr = (struct ofp_aggregate_stats_request *) osr->body;
2734 COVERAGE_INC(ofproto_agg_request);
2736 cbdata.out_port = asr->out_port;
2737 cbdata.packet_count = 0;
2738 cbdata.byte_count = 0;
2740 cls_rule_from_match(&target, &asr->match, 0);
2741 classifier_for_each_match(&p->cls, &target,
2742 table_id_to_include(asr->table_id),
2743 aggregate_stats_cb, &cbdata);
2745 msg = start_stats_reply(osr, sizeof *reply);
2746 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2747 reply->flow_count = htonl(cbdata.n_flows);
2748 reply->packet_count = htonll(cbdata.packet_count);
2749 reply->byte_count = htonll(cbdata.byte_count);
2750 queue_tx(msg, ofconn, ofconn->reply_counter);
2755 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2756 struct ofp_header *oh)
2758 struct ofp_stats_request *osr;
2762 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2767 osr = (struct ofp_stats_request *) oh;
2769 switch (ntohs(osr->type)) {
2771 return handle_desc_stats_request(p, ofconn, osr);
2774 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2776 case OFPST_AGGREGATE:
2777 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2780 return handle_table_stats_request(p, ofconn, osr);
2783 return handle_port_stats_request(p, ofconn, osr);
2786 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2789 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2793 static long long int
2794 msec_from_nsec(uint64_t sec, uint32_t nsec)
2796 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2800 update_time(struct ofproto *ofproto, struct rule *rule,
2801 const struct odp_flow_stats *stats)
2803 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2804 if (used > rule->used) {
2806 if (rule->super && used > rule->super->used) {
2807 rule->super->used = used;
2809 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
2814 update_stats(struct ofproto *ofproto, struct rule *rule,
2815 const struct odp_flow_stats *stats)
2817 if (stats->n_packets) {
2818 update_time(ofproto, rule, stats);
2819 rule->packet_count += stats->n_packets;
2820 rule->byte_count += stats->n_bytes;
2821 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
2827 add_flow(struct ofproto *p, struct ofconn *ofconn,
2828 struct ofp_flow_mod *ofm, size_t n_actions)
2830 struct ofpbuf *packet;
2835 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
2836 n_actions, ntohs(ofm->idle_timeout),
2837 ntohs(ofm->hard_timeout));
2838 cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority));
2842 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2843 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2847 rule_insert(p, rule, packet, in_port);
2848 ofpbuf_delete(packet);
2853 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
2854 size_t n_actions, uint16_t command, struct rule *rule)
2856 if (rule_is_hidden(rule)) {
2860 if (command == OFPFC_DELETE) {
2861 rule_remove(p, rule);
2863 size_t actions_len = n_actions * sizeof *rule->actions;
2865 if (n_actions == rule->n_actions
2866 && !memcmp(ofm->actions, rule->actions, actions_len))
2871 free(rule->actions);
2872 rule->actions = xmemdup(ofm->actions, actions_len);
2873 rule->n_actions = n_actions;
2875 if (rule->cr.wc.wildcards) {
2876 COVERAGE_INC(ofproto_mod_wc_flow);
2877 p->need_revalidate = true;
2879 rule_update_actions(p, rule);
2887 modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm,
2888 size_t n_actions, uint16_t command)
2894 flow_from_match(&flow, &wildcards, &ofm->match);
2895 rule = rule_from_cls_rule(classifier_find_rule_exactly(
2896 &p->cls, &flow, wildcards,
2897 ntohs(ofm->priority)));
2900 if (command == OFPFC_DELETE
2901 && ofm->out_port != htons(OFPP_NONE)
2902 && !rule_has_out_port(rule, ofm->out_port)) {
2906 modify_flow(p, ofm, n_actions, command, rule);
2911 struct modify_flows_cbdata {
2912 struct ofproto *ofproto;
2913 const struct ofp_flow_mod *ofm;
2920 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
2922 struct rule *rule = rule_from_cls_rule(rule_);
2923 struct modify_flows_cbdata *cbdata = cbdata_;
2925 if (cbdata->out_port != htons(OFPP_NONE)
2926 && !rule_has_out_port(rule, cbdata->out_port)) {
2930 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions,
2931 cbdata->command, rule);
2935 modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm,
2936 size_t n_actions, uint16_t command)
2938 struct modify_flows_cbdata cbdata;
2939 struct cls_rule target;
2943 cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port
2944 : htons(OFPP_NONE));
2945 cbdata.n_actions = n_actions;
2946 cbdata.command = command;
2948 cls_rule_from_match(&target, &ofm->match, 0);
2950 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2951 modify_flows_cb, &cbdata);
2956 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
2957 struct ofp_flow_mod *ofm)
2962 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
2963 sizeof *ofm->actions, &n_actions);
2968 normalize_match(&ofm->match);
2969 if (!ofm->match.wildcards) {
2970 ofm->priority = htons(UINT16_MAX);
2973 error = validate_actions((const union ofp_action *) ofm->actions,
2974 n_actions, p->max_ports);
2979 switch (ntohs(ofm->command)) {
2981 return add_flow(p, ofconn, ofm, n_actions);
2984 return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY);
2986 case OFPFC_MODIFY_STRICT:
2987 return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY);
2990 return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE);
2992 case OFPFC_DELETE_STRICT:
2993 return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE);
2996 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3001 send_capability_reply(struct ofproto *p, struct ofconn *ofconn, uint32_t xid)
3003 struct ofmp_capability_reply *ocr;
3005 char capabilities[] = "com.nicira.mgmt.manager=false\n";
3007 ocr = make_openflow_xid(sizeof(*ocr), OFPT_VENDOR, xid, &b);
3008 ocr->header.header.vendor = htonl(NX_VENDOR_ID);
3009 ocr->header.header.subtype = htonl(NXT_MGMT);
3010 ocr->header.type = htons(OFMPT_CAPABILITY_REPLY);
3012 ocr->format = htonl(OFMPCOF_SIMPLE);
3013 ocr->mgmt_id = htonll(p->mgmt_id);
3015 ofpbuf_put(b, capabilities, strlen(capabilities));
3017 queue_tx(b, ofconn, ofconn->reply_counter);
3021 handle_ofmp(struct ofproto *p, struct ofconn *ofconn,
3022 struct ofmp_header *ofmph)
3024 size_t msg_len = ntohs(ofmph->header.header.length);
3025 if (msg_len < sizeof(*ofmph)) {
3026 VLOG_WARN_RL(&rl, "dropping short managment message: %zu\n", msg_len);
3027 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
3030 if (ofmph->type == htons(OFMPT_CAPABILITY_REQUEST)) {
3031 struct ofmp_capability_request *ofmpcr;
3033 if (msg_len < sizeof(struct ofmp_capability_request)) {
3034 VLOG_WARN_RL(&rl, "dropping short capability request: %zu\n",
3036 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
3039 ofmpcr = (struct ofmp_capability_request *)ofmph;
3040 if (ofmpcr->format != htonl(OFMPCAF_SIMPLE)) {
3041 /* xxx Find a better type than bad subtype */
3042 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3045 send_capability_reply(p, ofconn, ofmph->header.header.xid);
3048 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3053 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3055 struct ofp_vendor_header *ovh = msg;
3056 struct nicira_header *nh;
3058 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3059 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
3061 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3062 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3064 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3065 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
3069 switch (ntohl(nh->subtype)) {
3070 case NXT_STATUS_REQUEST:
3071 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3074 case NXT_ACT_SET_CONFIG:
3075 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
3077 case NXT_ACT_GET_CONFIG:
3078 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
3080 case NXT_COMMAND_REQUEST:
3082 return executer_handle_request(p->executer, ofconn->rconn, msg);
3087 return handle_ofmp(p, ofconn, msg);
3090 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3094 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3095 struct ofpbuf *ofp_msg)
3097 struct ofp_header *oh = ofp_msg->data;
3100 COVERAGE_INC(ofproto_recv_openflow);
3102 case OFPT_ECHO_REQUEST:
3103 error = handle_echo_request(ofconn, oh);
3106 case OFPT_ECHO_REPLY:
3110 case OFPT_FEATURES_REQUEST:
3111 error = handle_features_request(p, ofconn, oh);
3114 case OFPT_GET_CONFIG_REQUEST:
3115 error = handle_get_config_request(p, ofconn, oh);
3118 case OFPT_SET_CONFIG:
3119 error = handle_set_config(p, ofconn, ofp_msg->data);
3122 case OFPT_PACKET_OUT:
3123 error = handle_packet_out(p, ofconn, ofp_msg->data);
3127 error = handle_port_mod(p, oh);
3131 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3134 case OFPT_STATS_REQUEST:
3135 error = handle_stats_request(p, ofconn, oh);
3139 error = handle_vendor(p, ofconn, ofp_msg->data);
3143 if (VLOG_IS_WARN_ENABLED()) {
3144 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3145 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3148 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3153 send_error_oh(ofconn, ofp_msg->data, error);
3158 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
3160 struct odp_msg *msg = packet->data;
3161 uint16_t in_port = odp_port_to_ofp_port(msg->port);
3163 struct ofpbuf payload;
3166 payload.data = msg + 1;
3167 payload.size = msg->length - sizeof *msg;
3168 flow_extract(&payload, msg->port, &flow);
3170 /* Check with in-band control to see if this packet should be sent
3171 * to the local port regardless of the flow table. */
3172 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3173 union odp_action action;
3175 memset(&action, 0, sizeof(action));
3176 action.output.type = ODPAT_OUTPUT;
3177 action.output.port = ODPP_LOCAL;
3178 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3181 rule = lookup_valid_rule(p, &flow);
3183 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3184 struct ofport *port = port_array_get(&p->ports, msg->port);
3186 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3187 COVERAGE_INC(ofproto_no_packet_in);
3188 /* XXX install 'drop' flow entry */
3189 ofpbuf_delete(packet);
3193 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3196 COVERAGE_INC(ofproto_packet_in);
3197 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3201 if (rule->cr.wc.wildcards) {
3202 rule = rule_create_subrule(p, rule, &flow);
3203 rule_make_actions(p, rule, packet);
3205 if (!rule->may_install) {
3206 /* The rule is not installable, that is, we need to process every
3207 * packet, so process the current packet and set its actions into
3209 rule_make_actions(p, rule, packet);
3211 /* XXX revalidate rule if it needs it */
3215 rule_execute(p, rule, &payload, &flow);
3216 rule_reinstall(p, rule);
3218 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY
3219 && rconn_is_connected(p->controller->rconn)) {
3221 * Extra-special case for fail-open mode.
3223 * We are in fail-open mode and the packet matched the fail-open rule,
3224 * but we are connected to a controller too. We should send the packet
3225 * up to the controller in the hope that it will try to set up a flow
3226 * and thereby allow us to exit fail-open.
3228 * See the top-level comment in fail-open.c for more information.
3230 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3232 ofpbuf_delete(packet);
3237 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3239 struct odp_msg *msg = packet->data;
3241 switch (msg->type) {
3242 case _ODPL_ACTION_NR:
3243 COVERAGE_INC(ofproto_ctlr_action);
3244 pinsched_send(p->action_sched, odp_port_to_ofp_port(msg->port), packet,
3245 send_packet_in_action, p);
3248 case _ODPL_SFLOW_NR:
3250 ofproto_sflow_received(p->sflow, msg);
3252 ofpbuf_delete(packet);
3256 handle_odp_miss_msg(p, packet);
3260 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
3267 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3269 struct rule *sub = rule_from_cls_rule(sub_);
3270 struct revalidate_cbdata *cbdata = cbdata_;
3272 if (cbdata->revalidate_all
3273 || (cbdata->revalidate_subrules && sub->super)
3274 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3275 revalidate_rule(cbdata->ofproto, sub);
3280 revalidate_rule(struct ofproto *p, struct rule *rule)
3282 const flow_t *flow = &rule->cr.flow;
3284 COVERAGE_INC(ofproto_revalidate_rule);
3287 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3289 rule_remove(p, rule);
3291 } else if (super != rule->super) {
3292 COVERAGE_INC(ofproto_revalidate_moved);
3293 list_remove(&rule->list);
3294 list_push_back(&super->list, &rule->list);
3295 rule->super = super;
3296 rule->hard_timeout = super->hard_timeout;
3297 rule->idle_timeout = super->idle_timeout;
3298 rule->created = super->created;
3303 rule_update_actions(p, rule);
3307 static struct ofpbuf *
3308 compose_flow_exp(const struct rule *rule, long long int now, uint8_t reason)
3310 struct ofp_flow_expired *ofe;
3313 ofe = make_openflow(sizeof *ofe, OFPT_FLOW_EXPIRED, &buf);
3314 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofe->match);
3315 ofe->priority = htons(rule->cr.priority);
3316 ofe->reason = reason;
3317 ofe->duration = htonl((now - rule->created) / 1000);
3318 ofe->packet_count = htonll(rule->packet_count);
3319 ofe->byte_count = htonll(rule->byte_count);
3325 send_flow_exp(struct ofproto *p, struct rule *rule,
3326 long long int now, uint8_t reason)
3328 struct ofconn *ofconn;
3329 struct ofconn *prev;
3330 struct ofpbuf *buf = NULL;
3332 /* We limit the maximum number of queued flow expirations it by accounting
3333 * them under the counter for replies. That works because preventing
3334 * OpenFlow requests from being processed also prevents new flows from
3335 * being added (and expiring). (It also prevents processing OpenFlow
3336 * requests that would not add new flows, so it is imperfect.) */
3339 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3340 if (ofconn->send_flow_exp && rconn_is_connected(ofconn->rconn)) {
3342 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3344 buf = compose_flow_exp(rule, now, reason);
3350 queue_tx(buf, prev, prev->reply_counter);
3355 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3357 assert(rule->installed);
3358 assert(!rule->cr.wc.wildcards);
3361 rule_remove(ofproto, rule);
3363 rule_uninstall(ofproto, rule);
3368 expire_rule(struct cls_rule *cls_rule, void *p_)
3370 struct ofproto *p = p_;
3371 struct rule *rule = rule_from_cls_rule(cls_rule);
3372 long long int hard_expire, idle_expire, expire, now;
3374 hard_expire = (rule->hard_timeout
3375 ? rule->created + rule->hard_timeout * 1000
3377 idle_expire = (rule->idle_timeout
3378 && (rule->super || list_is_empty(&rule->list))
3379 ? rule->used + rule->idle_timeout * 1000
3381 expire = MIN(hard_expire, idle_expire);
3385 if (rule->installed && now >= rule->used + 5000) {
3386 uninstall_idle_flow(p, rule);
3387 } else if (!rule->cr.wc.wildcards) {
3388 active_timeout(p, rule);
3394 COVERAGE_INC(ofproto_expired);
3396 /* Update stats. This code will be a no-op if the rule expired
3397 * due to an idle timeout. */
3398 if (rule->cr.wc.wildcards) {
3399 struct rule *subrule, *next;
3400 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3401 rule_remove(p, subrule);
3404 rule_uninstall(p, rule);
3407 if (!rule_is_hidden(rule)) {
3408 send_flow_exp(p, rule, now,
3410 ? OFPER_HARD_TIMEOUT : OFPER_IDLE_TIMEOUT));
3412 rule_remove(p, rule);
3416 active_timeout(struct ofproto *ofproto, struct rule *rule)
3418 if (ofproto->netflow && !is_controller_rule(rule) &&
3419 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
3420 struct ofexpired expired;
3421 struct odp_flow odp_flow;
3423 /* Get updated flow stats. */
3424 memset(&odp_flow, 0, sizeof odp_flow);
3425 if (rule->installed) {
3426 odp_flow.key = rule->cr.flow;
3427 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
3428 dpif_flow_get(ofproto->dpif, &odp_flow);
3430 if (odp_flow.stats.n_packets) {
3431 update_time(ofproto, rule, &odp_flow.stats);
3432 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
3433 odp_flow.stats.tcp_flags);
3437 expired.flow = rule->cr.flow;
3438 expired.packet_count = rule->packet_count +
3439 odp_flow.stats.n_packets;
3440 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
3441 expired.used = rule->used;
3443 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
3445 /* Schedule us to send the accumulated records once we have
3446 * collected all of them. */
3447 poll_immediate_wake();
3452 update_used(struct ofproto *p)
3454 struct odp_flow *flows;
3459 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
3464 for (i = 0; i < n_flows; i++) {
3465 struct odp_flow *f = &flows[i];
3468 rule = rule_from_cls_rule(
3469 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3470 if (!rule || !rule->installed) {
3471 COVERAGE_INC(ofproto_unexpected_rule);
3472 dpif_flow_del(p->dpif, f);
3476 update_time(p, rule, &f->stats);
3477 rule_account(p, rule, f->stats.n_bytes);
3483 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3484 const struct ofpbuf *packet, int send_len)
3486 struct odp_msg *msg = packet->data;
3487 struct ofpbuf payload;
3491 /* Extract packet payload from 'msg'. */
3492 payload.data = msg + 1;
3493 payload.size = msg->length - sizeof *msg;
3495 /* Construct ofp_packet_in message. */
3496 reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3497 opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port), reason,
3498 &payload, send_len);
3501 rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100);
3505 send_packet_in_action(struct ofpbuf *packet, void *p_)
3507 struct ofproto *p = p_;
3508 struct ofconn *ofconn;
3509 struct odp_msg *msg;
3512 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3513 if (ofconn == p->controller || ofconn->miss_send_len) {
3514 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3517 ofpbuf_delete(packet);
3521 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3523 struct ofproto *p = p_;
3524 bool in_fail_open = p->fail_open && fail_open_is_active(p->fail_open);
3525 struct ofconn *ofconn;
3526 struct ofpbuf payload;
3527 struct odp_msg *msg;
3530 payload.data = msg + 1;
3531 payload.size = msg->length - sizeof *msg;
3532 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3533 if (ofconn->miss_send_len) {
3534 struct pktbuf *pb = ofconn->pktbuf;
3535 uint32_t buffer_id = (in_fail_open
3537 : pktbuf_save(pb, &payload, msg->port));
3538 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3540 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3543 ofpbuf_delete(packet);
3547 pick_datapath_id(const struct ofproto *ofproto)
3549 const struct ofport *port;
3551 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
3553 uint8_t ea[ETH_ADDR_LEN];
3556 error = netdev_get_etheraddr(port->netdev, ea);
3558 return eth_addr_to_uint64(ea);
3560 VLOG_WARN("could not get MAC address for %s (%s)",
3561 netdev_get_name(port->netdev), strerror(error));
3563 return ofproto->fallback_dpid;
3567 pick_fallback_dpid(void)
3569 uint8_t ea[ETH_ADDR_LEN];
3570 eth_addr_random(ea);
3571 ea[0] = 0x00; /* Set Nicira OUI. */
3574 return eth_addr_to_uint64(ea);
3578 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3579 struct odp_actions *actions, tag_type *tags,
3580 uint16_t *nf_output_iface, void *ofproto_)
3582 struct ofproto *ofproto = ofproto_;
3585 /* Drop frames for reserved multicast addresses. */
3586 if (eth_addr_is_reserved(flow->dl_dst)) {
3590 /* Learn source MAC (but don't try to learn from revalidation). */
3591 if (packet != NULL) {
3592 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3595 /* The log messages here could actually be useful in debugging,
3596 * so keep the rate limit relatively high. */
3597 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3598 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3599 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3600 ofproto_revalidate(ofproto, rev_tag);
3604 /* Determine output port. */
3605 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3607 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
3608 } else if (out_port != flow->in_port) {
3609 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3610 *nf_output_iface = out_port;
3618 static const struct ofhooks default_ofhooks = {
3620 default_normal_ofhook_cb,