2 * Copyright (c) 2009, 2010 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
22 #include <netinet/in.h>
25 #include "classifier.h"
27 #include "discovery.h"
29 #include "dynamic-string.h"
30 #include "fail-open.h"
32 #include "mac-learning.h"
36 #include "ofp-print.h"
38 #include "openflow/nicira-ext.h"
39 #include "openflow/openflow.h"
40 #include "openflow/openflow-mgmt.h"
41 #include "openvswitch/datapath-protocol.h"
45 #include "poll-loop.h"
46 #include "port-array.h"
51 #include "stream-ssl.h"
59 #define THIS_MODULE VLM_ofproto
69 TABLEID_CLASSIFIER = 1
73 struct netdev *netdev;
74 struct ofp_phy_port opp; /* In host byte order. */
77 static void ofport_free(struct ofport *);
78 static void hton_ofp_phy_port(struct ofp_phy_port *);
80 static int xlate_actions(const union ofp_action *in, size_t n_in,
81 const flow_t *flow, struct ofproto *ofproto,
82 const struct ofpbuf *packet,
83 struct odp_actions *out, tag_type *tags,
84 bool *may_set_up_flow, uint16_t *nf_output_iface);
89 uint16_t idle_timeout; /* In seconds from time of last use. */
90 uint16_t hard_timeout; /* In seconds from time of creation. */
91 long long int used; /* Last-used time (0 if never used). */
92 long long int created; /* Creation time. */
93 uint64_t packet_count; /* Number of packets received. */
94 uint64_t byte_count; /* Number of bytes received. */
95 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
96 tag_type tags; /* Tags (set only by hooks). */
97 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
99 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
100 * exact-match rule (having cr.wc.wildcards of 0) generated from the
101 * wildcard rule 'super'. In this case, 'list' is an element of the
104 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
105 * a list of subrules. A super-rule with no wildcards (where
106 * cr.wc.wildcards is 0) will never have any subrules. */
112 * A subrule has no actions (it uses the super-rule's actions). */
114 union ofp_action *actions;
118 * A super-rule with wildcard fields never has ODP actions (since the
119 * datapath only supports exact-match flows). */
120 bool installed; /* Installed in datapath? */
121 bool may_install; /* True ordinarily; false if actions must
122 * be reassessed for every packet. */
124 union odp_action *odp_actions;
128 rule_is_hidden(const struct rule *rule)
130 /* Subrules are merely an implementation detail, so hide them from the
132 if (rule->super != NULL) {
136 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
137 * (e.g. by in-band control) and are intentionally hidden from the
139 if (rule->cr.priority > UINT16_MAX) {
146 static struct rule *rule_create(struct ofproto *, struct rule *super,
147 const union ofp_action *, size_t n_actions,
148 uint16_t idle_timeout, uint16_t hard_timeout);
149 static void rule_free(struct rule *);
150 static void rule_destroy(struct ofproto *, struct rule *);
151 static struct rule *rule_from_cls_rule(const struct cls_rule *);
152 static void rule_insert(struct ofproto *, struct rule *,
153 struct ofpbuf *packet, uint16_t in_port);
154 static void rule_remove(struct ofproto *, struct rule *);
155 static bool rule_make_actions(struct ofproto *, struct rule *,
156 const struct ofpbuf *packet);
157 static void rule_install(struct ofproto *, struct rule *,
158 struct rule *displaced_rule);
159 static void rule_uninstall(struct ofproto *, struct rule *);
160 static void rule_post_uninstall(struct ofproto *, struct rule *);
165 struct pktbuf *pktbuf;
169 struct rconn_packet_counter *packet_in_counter;
171 /* Number of OpenFlow messages queued as replies to OpenFlow requests, and
172 * the maximum number before we stop reading OpenFlow requests. */
173 #define OFCONN_REPLY_MAX 100
174 struct rconn_packet_counter *reply_counter;
177 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *);
178 static void ofconn_destroy(struct ofconn *);
179 static void ofconn_run(struct ofconn *, struct ofproto *);
180 static void ofconn_wait(struct ofconn *);
181 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
182 struct rconn_packet_counter *counter);
186 uint64_t datapath_id; /* Datapath ID. */
187 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
188 uint64_t mgmt_id; /* Management channel identifier. */
189 char *manufacturer; /* Manufacturer. */
190 char *hardware; /* Hardware. */
191 char *software; /* Software version. */
192 char *serial; /* Serial number. */
196 struct netdev_monitor *netdev_monitor;
197 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
199 struct shash port_by_name;
203 struct switch_status *switch_status;
204 struct status_category *ss_cat;
205 struct in_band *in_band;
206 struct discovery *discovery;
207 struct fail_open *fail_open;
208 struct pinsched *miss_sched, *action_sched;
209 struct netflow *netflow;
212 struct classifier cls;
213 bool need_revalidate;
214 long long int next_expiration;
215 struct tag_set revalidate_set;
217 /* OpenFlow connections. */
218 struct list all_conns;
219 struct ofconn *controller;
220 struct pvconn **listeners;
222 struct pvconn **snoops;
225 /* Hooks for ovs-vswitchd. */
226 const struct ofhooks *ofhooks;
229 /* Used by default ofhooks. */
230 struct mac_learning *ml;
233 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
235 static const struct ofhooks default_ofhooks;
237 static uint64_t pick_datapath_id(const struct ofproto *);
238 static uint64_t pick_fallback_dpid(void);
239 static void send_packet_in_miss(struct ofpbuf *, void *ofproto);
240 static void send_packet_in_action(struct ofpbuf *, void *ofproto);
241 static void update_used(struct ofproto *);
242 static void update_stats(struct ofproto *, struct rule *,
243 const struct odp_flow_stats *);
244 static void expire_rule(struct cls_rule *, void *ofproto);
245 static void active_timeout(struct ofproto *ofproto, struct rule *rule);
246 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
247 static void revalidate_cb(struct cls_rule *rule_, void *p_);
249 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
251 static void handle_openflow(struct ofconn *, struct ofproto *,
254 static void refresh_port_group(struct ofproto *, unsigned int group);
255 static void update_port(struct ofproto *, const char *devname);
256 static int init_ports(struct ofproto *);
257 static void reinit_ports(struct ofproto *);
260 ofproto_create(const char *datapath, const struct ofhooks *ofhooks, void *aux,
261 struct ofproto **ofprotop)
263 struct odp_stats stats;
270 /* Connect to datapath and start listening for messages. */
271 error = dpif_open(datapath, &dpif);
273 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
276 error = dpif_get_dp_stats(dpif, &stats);
278 VLOG_ERR("failed to obtain stats for datapath %s: %s",
279 datapath, strerror(error));
283 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION);
285 VLOG_ERR("failed to listen on datapath %s: %s",
286 datapath, strerror(error));
290 dpif_flow_flush(dpif);
291 dpif_recv_purge(dpif);
293 /* Initialize settings. */
294 p = xzalloc(sizeof *p);
295 p->fallback_dpid = pick_fallback_dpid();
296 p->datapath_id = p->fallback_dpid;
297 p->manufacturer = xstrdup("Nicira Networks, Inc.");
298 p->hardware = xstrdup("Reference Implementation");
299 p->software = xstrdup(VERSION BUILDNR);
300 p->serial = xstrdup("None");
302 /* Initialize datapath. */
304 p->netdev_monitor = netdev_monitor_create();
305 port_array_init(&p->ports);
306 shash_init(&p->port_by_name);
307 p->max_ports = stats.max_ports;
309 /* Initialize submodules. */
310 p->switch_status = switch_status_create(p);
314 p->miss_sched = p->action_sched = NULL;
317 /* Initialize flow table. */
318 classifier_init(&p->cls);
319 p->need_revalidate = false;
320 p->next_expiration = time_msec() + 1000;
321 tag_set_init(&p->revalidate_set);
323 /* Initialize OpenFlow connections. */
324 list_init(&p->all_conns);
325 p->controller = ofconn_create(p, rconn_create(5, 8));
326 p->controller->pktbuf = pktbuf_create();
327 p->controller->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
333 /* Initialize hooks. */
335 p->ofhooks = ofhooks;
339 p->ofhooks = &default_ofhooks;
341 p->ml = mac_learning_create();
344 /* Register switch status category. */
345 p->ss_cat = switch_status_register(p->switch_status, "remote",
346 rconn_status_cb, p->controller->rconn);
349 error = init_ports(p);
355 /* Pick final datapath ID. */
356 p->datapath_id = pick_datapath_id(p);
357 VLOG_INFO("using datapath ID %012"PRIx64, p->datapath_id);
364 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
366 uint64_t old_dpid = p->datapath_id;
367 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
368 if (p->datapath_id != old_dpid) {
369 VLOG_INFO("datapath ID changed to %012"PRIx64, p->datapath_id);
370 rconn_reconnect(p->controller->rconn);
375 ofproto_set_mgmt_id(struct ofproto *p, uint64_t mgmt_id)
377 p->mgmt_id = mgmt_id;
381 ofproto_set_probe_interval(struct ofproto *p, int probe_interval)
383 probe_interval = probe_interval ? MAX(probe_interval, 5) : 0;
384 rconn_set_probe_interval(p->controller->rconn, probe_interval);
386 int trigger_duration = probe_interval ? probe_interval * 3 : 15;
387 fail_open_set_trigger_duration(p->fail_open, trigger_duration);
392 ofproto_set_max_backoff(struct ofproto *p, int max_backoff)
394 rconn_set_max_backoff(p->controller->rconn, max_backoff);
398 ofproto_set_desc(struct ofproto *p,
399 const char *manufacturer, const char *hardware,
400 const char *software, const char *serial)
403 free(p->manufacturer);
404 p->manufacturer = xstrdup(manufacturer);
408 p->hardware = xstrdup(hardware);
412 p->software = xstrdup(software);
416 p->serial = xstrdup(serial);
421 ofproto_set_in_band(struct ofproto *p, bool in_band)
423 if (in_band != (p->in_band != NULL)) {
425 return in_band_create(p, p->dpif, p->switch_status,
426 p->controller->rconn, &p->in_band);
428 ofproto_set_discovery(p, false, NULL, true);
429 in_band_destroy(p->in_band);
432 rconn_reconnect(p->controller->rconn);
438 ofproto_set_discovery(struct ofproto *p, bool discovery,
439 const char *re, bool update_resolv_conf)
441 if (discovery != (p->discovery != NULL)) {
443 int error = ofproto_set_in_band(p, true);
447 error = discovery_create(re, update_resolv_conf,
448 p->dpif, p->switch_status,
454 discovery_destroy(p->discovery);
457 rconn_disconnect(p->controller->rconn);
458 } else if (discovery) {
459 discovery_set_update_resolv_conf(p->discovery, update_resolv_conf);
460 return discovery_set_accept_controller_re(p->discovery, re);
466 ofproto_set_controller(struct ofproto *ofproto, const char *controller)
468 if (ofproto->discovery) {
470 } else if (controller) {
471 if (strcmp(rconn_get_name(ofproto->controller->rconn), controller)) {
472 return rconn_connect(ofproto->controller->rconn, controller);
477 rconn_disconnect(ofproto->controller->rconn);
483 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
484 const struct svec *svec)
486 struct pvconn **pvconns = *pvconnsp;
487 size_t n_pvconns = *n_pvconnsp;
491 for (i = 0; i < n_pvconns; i++) {
492 pvconn_close(pvconns[i]);
496 pvconns = xmalloc(svec->n * sizeof *pvconns);
498 for (i = 0; i < svec->n; i++) {
499 const char *name = svec->names[i];
500 struct pvconn *pvconn;
503 error = pvconn_open(name, &pvconn);
505 pvconns[n_pvconns++] = pvconn;
507 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
515 *n_pvconnsp = n_pvconns;
521 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
523 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
527 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
529 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
533 ofproto_set_netflow(struct ofproto *ofproto,
534 const struct netflow_options *nf_options)
536 if (nf_options && nf_options->collectors.n) {
537 if (!ofproto->netflow) {
538 ofproto->netflow = netflow_create();
540 return netflow_set_options(ofproto->netflow, nf_options);
542 netflow_destroy(ofproto->netflow);
543 ofproto->netflow = NULL;
549 ofproto_set_failure(struct ofproto *ofproto, bool fail_open)
552 struct rconn *rconn = ofproto->controller->rconn;
553 int trigger_duration = rconn_get_probe_interval(rconn) * 3;
554 if (!ofproto->fail_open) {
555 ofproto->fail_open = fail_open_create(ofproto, trigger_duration,
556 ofproto->switch_status,
559 fail_open_set_trigger_duration(ofproto->fail_open,
563 fail_open_destroy(ofproto->fail_open);
564 ofproto->fail_open = NULL;
569 ofproto_set_rate_limit(struct ofproto *ofproto,
570 int rate_limit, int burst_limit)
572 if (rate_limit > 0) {
573 if (!ofproto->miss_sched) {
574 ofproto->miss_sched = pinsched_create(rate_limit, burst_limit,
575 ofproto->switch_status);
576 ofproto->action_sched = pinsched_create(rate_limit, burst_limit,
579 pinsched_set_limits(ofproto->miss_sched, rate_limit, burst_limit);
580 pinsched_set_limits(ofproto->action_sched,
581 rate_limit, burst_limit);
584 pinsched_destroy(ofproto->miss_sched);
585 ofproto->miss_sched = NULL;
586 pinsched_destroy(ofproto->action_sched);
587 ofproto->action_sched = NULL;
592 ofproto_set_stp(struct ofproto *ofproto UNUSED, bool enable_stp)
596 VLOG_WARN("STP is not yet implemented");
604 ofproto_get_datapath_id(const struct ofproto *ofproto)
606 return ofproto->datapath_id;
610 ofproto_get_mgmt_id(const struct ofproto *ofproto)
612 return ofproto->mgmt_id;
616 ofproto_get_probe_interval(const struct ofproto *ofproto)
618 return rconn_get_probe_interval(ofproto->controller->rconn);
622 ofproto_get_max_backoff(const struct ofproto *ofproto)
624 return rconn_get_max_backoff(ofproto->controller->rconn);
628 ofproto_get_in_band(const struct ofproto *ofproto)
630 return ofproto->in_band != NULL;
634 ofproto_get_discovery(const struct ofproto *ofproto)
636 return ofproto->discovery != NULL;
640 ofproto_get_controller(const struct ofproto *ofproto)
642 return rconn_get_name(ofproto->controller->rconn);
646 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
650 for (i = 0; i < ofproto->n_listeners; i++) {
651 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
656 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
660 for (i = 0; i < ofproto->n_snoops; i++) {
661 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
666 ofproto_destroy(struct ofproto *p)
668 struct ofconn *ofconn, *next_ofconn;
669 struct ofport *ofport;
670 unsigned int port_no;
677 ofproto_flush_flows(p);
678 classifier_destroy(&p->cls);
680 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
682 ofconn_destroy(ofconn);
686 netdev_monitor_destroy(p->netdev_monitor);
687 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
690 shash_destroy(&p->port_by_name);
692 switch_status_destroy(p->switch_status);
693 in_band_destroy(p->in_band);
694 discovery_destroy(p->discovery);
695 fail_open_destroy(p->fail_open);
696 pinsched_destroy(p->miss_sched);
697 pinsched_destroy(p->action_sched);
698 netflow_destroy(p->netflow);
700 switch_status_unregister(p->ss_cat);
702 for (i = 0; i < p->n_listeners; i++) {
703 pvconn_close(p->listeners[i]);
707 for (i = 0; i < p->n_snoops; i++) {
708 pvconn_close(p->snoops[i]);
712 mac_learning_destroy(p->ml);
718 ofproto_run(struct ofproto *p)
720 int error = ofproto_run1(p);
722 error = ofproto_run2(p, false);
728 process_port_change(struct ofproto *ofproto, int error, char *devname)
730 if (error == ENOBUFS) {
731 reinit_ports(ofproto);
733 update_port(ofproto, devname);
739 ofproto_run1(struct ofproto *p)
741 struct ofconn *ofconn, *next_ofconn;
746 for (i = 0; i < 50; i++) {
750 error = dpif_recv(p->dpif, &buf);
752 if (error == ENODEV) {
753 /* Someone destroyed the datapath behind our back. The caller
754 * better destroy us and give up, because we're just going to
755 * spin from here on out. */
756 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
757 VLOG_ERR_RL(&rl, "%s: datapath was destroyed externally",
764 handle_odp_msg(p, buf);
767 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
768 process_port_change(p, error, devname);
770 while ((error = netdev_monitor_poll(p->netdev_monitor,
771 &devname)) != EAGAIN) {
772 process_port_change(p, error, devname);
776 in_band_run(p->in_band);
779 char *controller_name;
780 if (rconn_is_connectivity_questionable(p->controller->rconn)) {
781 discovery_question_connectivity(p->discovery);
783 if (discovery_run(p->discovery, &controller_name)) {
784 if (controller_name) {
785 rconn_connect(p->controller->rconn, controller_name);
787 rconn_disconnect(p->controller->rconn);
791 pinsched_run(p->miss_sched, send_packet_in_miss, p);
792 pinsched_run(p->action_sched, send_packet_in_action, p);
794 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
796 ofconn_run(ofconn, p);
799 /* Fail-open maintenance. Do this after processing the ofconns since
800 * fail-open checks the status of the controller rconn. */
802 fail_open_run(p->fail_open);
805 for (i = 0; i < p->n_listeners; i++) {
809 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
811 ofconn_create(p, rconn_new_from_vconn("passive", vconn));
812 } else if (retval != EAGAIN) {
813 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
817 for (i = 0; i < p->n_snoops; i++) {
821 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
823 rconn_add_monitor(p->controller->rconn, vconn);
824 } else if (retval != EAGAIN) {
825 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
829 if (time_msec() >= p->next_expiration) {
830 COVERAGE_INC(ofproto_expiration);
831 p->next_expiration = time_msec() + 1000;
834 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
836 /* Let the hook know that we're at a stable point: all outstanding data
837 * in existing flows has been accounted to the account_cb. Thus, the
838 * hook can now reasonably do operations that depend on having accurate
839 * flow volume accounting (currently, that's just bond rebalancing). */
840 if (p->ofhooks->account_checkpoint_cb) {
841 p->ofhooks->account_checkpoint_cb(p->aux);
846 netflow_run(p->netflow);
852 struct revalidate_cbdata {
853 struct ofproto *ofproto;
854 bool revalidate_all; /* Revalidate all exact-match rules? */
855 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
856 struct tag_set revalidate_set; /* Set of tags to revalidate. */
860 ofproto_run2(struct ofproto *p, bool revalidate_all)
862 if (p->need_revalidate || revalidate_all
863 || !tag_set_is_empty(&p->revalidate_set)) {
864 struct revalidate_cbdata cbdata;
866 cbdata.revalidate_all = revalidate_all;
867 cbdata.revalidate_subrules = p->need_revalidate;
868 cbdata.revalidate_set = p->revalidate_set;
869 tag_set_init(&p->revalidate_set);
870 COVERAGE_INC(ofproto_revalidate);
871 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
872 p->need_revalidate = false;
879 ofproto_wait(struct ofproto *p)
881 struct ofconn *ofconn;
884 dpif_recv_wait(p->dpif);
885 dpif_port_poll_wait(p->dpif);
886 netdev_monitor_poll_wait(p->netdev_monitor);
887 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
891 in_band_wait(p->in_band);
894 discovery_wait(p->discovery);
897 fail_open_wait(p->fail_open);
899 pinsched_wait(p->miss_sched);
900 pinsched_wait(p->action_sched);
901 if (!tag_set_is_empty(&p->revalidate_set)) {
902 poll_immediate_wake();
904 if (p->need_revalidate) {
905 /* Shouldn't happen, but if it does just go around again. */
906 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
907 poll_immediate_wake();
908 } else if (p->next_expiration != LLONG_MAX) {
909 poll_timer_wait(p->next_expiration - time_msec());
911 for (i = 0; i < p->n_listeners; i++) {
912 pvconn_wait(p->listeners[i]);
914 for (i = 0; i < p->n_snoops; i++) {
915 pvconn_wait(p->snoops[i]);
920 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
922 tag_set_add(&ofproto->revalidate_set, tag);
926 ofproto_get_revalidate_set(struct ofproto *ofproto)
928 return &ofproto->revalidate_set;
932 ofproto_is_alive(const struct ofproto *p)
934 return p->discovery || rconn_is_alive(p->controller->rconn);
938 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
939 const union ofp_action *actions, size_t n_actions,
940 const struct ofpbuf *packet)
942 struct odp_actions odp_actions;
945 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
951 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
953 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
954 odp_actions.n_actions, packet);
959 ofproto_add_flow(struct ofproto *p,
960 const flow_t *flow, uint32_t wildcards, unsigned int priority,
961 const union ofp_action *actions, size_t n_actions,
965 rule = rule_create(p, NULL, actions, n_actions,
966 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */, 0);
967 cls_rule_from_flow(&rule->cr, flow, wildcards, priority);
968 rule_insert(p, rule, NULL, 0);
972 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
973 uint32_t wildcards, unsigned int priority)
977 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
981 rule_remove(ofproto, rule);
986 destroy_rule(struct cls_rule *rule_, void *ofproto_)
988 struct rule *rule = rule_from_cls_rule(rule_);
989 struct ofproto *ofproto = ofproto_;
991 /* Mark the flow as not installed, even though it might really be
992 * installed, so that rule_remove() doesn't bother trying to uninstall it.
993 * There is no point in uninstalling it individually since we are about to
994 * blow away all the flows with dpif_flow_flush(). */
995 rule->installed = false;
997 rule_remove(ofproto, rule);
1001 ofproto_flush_flows(struct ofproto *ofproto)
1003 COVERAGE_INC(ofproto_flush);
1004 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1005 dpif_flow_flush(ofproto->dpif);
1006 if (ofproto->in_band) {
1007 in_band_flushed(ofproto->in_band);
1009 if (ofproto->fail_open) {
1010 fail_open_flushed(ofproto->fail_open);
1015 reinit_ports(struct ofproto *p)
1017 struct svec devnames;
1018 struct ofport *ofport;
1019 unsigned int port_no;
1020 struct odp_port *odp_ports;
1024 svec_init(&devnames);
1025 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1026 svec_add (&devnames, (char *) ofport->opp.name);
1028 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1029 for (i = 0; i < n_odp_ports; i++) {
1030 svec_add (&devnames, odp_ports[i].devname);
1034 svec_sort_unique(&devnames);
1035 for (i = 0; i < devnames.n; i++) {
1036 update_port(p, devnames.names[i]);
1038 svec_destroy(&devnames);
1042 refresh_port_group(struct ofproto *p, unsigned int group)
1046 struct ofport *port;
1047 unsigned int port_no;
1049 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1051 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1053 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1054 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1055 ports[n_ports++] = port_no;
1058 dpif_port_group_set(p->dpif, group, ports, n_ports);
1063 refresh_port_groups(struct ofproto *p)
1065 refresh_port_group(p, DP_GROUP_FLOOD);
1066 refresh_port_group(p, DP_GROUP_ALL);
1069 static struct ofport *
1070 make_ofport(const struct odp_port *odp_port)
1072 enum netdev_flags flags;
1073 struct ofport *ofport;
1074 struct netdev *netdev;
1078 error = netdev_open(odp_port->devname, NETDEV_ETH_TYPE_NONE, &netdev);
1080 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1081 "cannot be opened (%s)",
1082 odp_port->devname, odp_port->port,
1083 odp_port->devname, strerror(error));
1087 ofport = xmalloc(sizeof *ofport);
1088 ofport->netdev = netdev;
1089 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1090 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1091 memcpy(ofport->opp.name, odp_port->devname,
1092 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1093 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1095 netdev_get_flags(netdev, &flags);
1096 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1098 netdev_get_carrier(netdev, &carrier);
1099 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1101 netdev_get_features(netdev,
1102 &ofport->opp.curr, &ofport->opp.advertised,
1103 &ofport->opp.supported, &ofport->opp.peer);
1108 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1110 if (port_array_get(&p->ports, odp_port->port)) {
1111 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1114 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1115 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1124 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1126 const struct ofp_phy_port *a = &a_->opp;
1127 const struct ofp_phy_port *b = &b_->opp;
1129 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1130 return (a->port_no == b->port_no
1131 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1132 && !strcmp((char *) a->name, (char *) b->name)
1133 && a->state == b->state
1134 && a->config == b->config
1135 && a->curr == b->curr
1136 && a->advertised == b->advertised
1137 && a->supported == b->supported
1138 && a->peer == b->peer);
1142 send_port_status(struct ofproto *p, const struct ofport *ofport,
1145 /* XXX Should limit the number of queued port status change messages. */
1146 struct ofconn *ofconn;
1147 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1148 struct ofp_port_status *ops;
1151 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1152 ops->reason = reason;
1153 ops->desc = ofport->opp;
1154 hton_ofp_phy_port(&ops->desc);
1155 queue_tx(b, ofconn, NULL);
1157 if (p->ofhooks->port_changed_cb) {
1158 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1163 ofport_install(struct ofproto *p, struct ofport *ofport)
1165 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1166 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no),
1168 shash_add(&p->port_by_name, (char *) ofport->opp.name, ofport);
1172 ofport_remove(struct ofproto *p, struct ofport *ofport)
1174 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1175 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no), NULL);
1176 shash_delete(&p->port_by_name,
1177 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1181 ofport_free(struct ofport *ofport)
1184 netdev_close(ofport->netdev);
1190 update_port(struct ofproto *p, const char *devname)
1192 struct odp_port odp_port;
1193 struct ofport *old_ofport;
1194 struct ofport *new_ofport;
1197 COVERAGE_INC(ofproto_update_port);
1199 /* Query the datapath for port information. */
1200 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1202 /* Find the old ofport. */
1203 old_ofport = shash_find_data(&p->port_by_name, devname);
1206 /* There's no port named 'devname' but there might be a port with
1207 * the same port number. This could happen if a port is deleted
1208 * and then a new one added in its place very quickly, or if a port
1209 * is renamed. In the former case we want to send an OFPPR_DELETE
1210 * and an OFPPR_ADD, and in the latter case we want to send a
1211 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1212 * the old port's ifindex against the new port, or perhaps less
1213 * reliably but more portably by comparing the old port's MAC
1214 * against the new port's MAC. However, this code isn't that smart
1215 * and always sends an OFPPR_MODIFY (XXX). */
1216 old_ofport = port_array_get(&p->ports, odp_port.port);
1218 } else if (error != ENOENT && error != ENODEV) {
1219 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1220 "%s", strerror(error));
1224 /* Create a new ofport. */
1225 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1227 /* Eliminate a few pathological cases. */
1228 if (!old_ofport && !new_ofport) {
1230 } else if (old_ofport && new_ofport) {
1231 /* Most of the 'config' bits are OpenFlow soft state, but
1232 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1233 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1234 * leaves the other bits 0.) */
1235 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1237 if (ofport_equal(old_ofport, new_ofport)) {
1238 /* False alarm--no change. */
1239 ofport_free(new_ofport);
1244 /* Now deal with the normal cases. */
1246 ofport_remove(p, old_ofport);
1249 ofport_install(p, new_ofport);
1251 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1252 (!old_ofport ? OFPPR_ADD
1253 : !new_ofport ? OFPPR_DELETE
1255 ofport_free(old_ofport);
1257 /* Update port groups. */
1258 refresh_port_groups(p);
1262 init_ports(struct ofproto *p)
1264 struct odp_port *ports;
1269 error = dpif_port_list(p->dpif, &ports, &n_ports);
1274 for (i = 0; i < n_ports; i++) {
1275 const struct odp_port *odp_port = &ports[i];
1276 if (!ofport_conflicts(p, odp_port)) {
1277 struct ofport *ofport = make_ofport(odp_port);
1279 ofport_install(p, ofport);
1284 refresh_port_groups(p);
1288 static struct ofconn *
1289 ofconn_create(struct ofproto *p, struct rconn *rconn)
1291 struct ofconn *ofconn = xmalloc(sizeof *ofconn);
1292 list_push_back(&p->all_conns, &ofconn->node);
1293 ofconn->rconn = rconn;
1294 ofconn->pktbuf = NULL;
1295 ofconn->send_flow_exp = false;
1296 ofconn->miss_send_len = 0;
1297 ofconn->packet_in_counter = rconn_packet_counter_create ();
1298 ofconn->reply_counter = rconn_packet_counter_create ();
1303 ofconn_destroy(struct ofconn *ofconn)
1305 list_remove(&ofconn->node);
1306 rconn_destroy(ofconn->rconn);
1307 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1308 rconn_packet_counter_destroy(ofconn->reply_counter);
1309 pktbuf_destroy(ofconn->pktbuf);
1314 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1318 rconn_run(ofconn->rconn);
1320 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1321 /* Limit the number of iterations to prevent other tasks from
1323 for (iteration = 0; iteration < 50; iteration++) {
1324 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1329 fail_open_maybe_recover(p->fail_open);
1331 handle_openflow(ofconn, p, of_msg);
1332 ofpbuf_delete(of_msg);
1336 if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
1337 ofconn_destroy(ofconn);
1342 ofconn_wait(struct ofconn *ofconn)
1344 rconn_run_wait(ofconn->rconn);
1345 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1346 rconn_recv_wait(ofconn->rconn);
1348 COVERAGE_INC(ofproto_ofconn_stuck);
1352 /* Caller is responsible for initializing the 'cr' member of the returned
1354 static struct rule *
1355 rule_create(struct ofproto *ofproto, struct rule *super,
1356 const union ofp_action *actions, size_t n_actions,
1357 uint16_t idle_timeout, uint16_t hard_timeout)
1359 struct rule *rule = xzalloc(sizeof *rule);
1360 rule->idle_timeout = idle_timeout;
1361 rule->hard_timeout = hard_timeout;
1362 rule->used = rule->created = time_msec();
1363 rule->super = super;
1365 list_push_back(&super->list, &rule->list);
1367 list_init(&rule->list);
1369 rule->n_actions = n_actions;
1370 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1371 netflow_flow_clear(&rule->nf_flow);
1372 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1377 static struct rule *
1378 rule_from_cls_rule(const struct cls_rule *cls_rule)
1380 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1384 rule_free(struct rule *rule)
1386 free(rule->actions);
1387 free(rule->odp_actions);
1391 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1392 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1393 * through all of its subrules and revalidates them, destroying any that no
1394 * longer has a super-rule (which is probably all of them).
1396 * Before calling this function, the caller must make have removed 'rule' from
1397 * the classifier. If 'rule' is an exact-match rule, the caller is also
1398 * responsible for ensuring that it has been uninstalled from the datapath. */
1400 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1403 struct rule *subrule, *next;
1404 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1405 revalidate_rule(ofproto, subrule);
1408 list_remove(&rule->list);
1414 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1416 const union ofp_action *oa;
1417 struct actions_iterator i;
1419 if (out_port == htons(OFPP_NONE)) {
1422 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1423 oa = actions_next(&i)) {
1424 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1431 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1432 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1434 * The flow that 'packet' actually contains does not need to actually match
1435 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1436 * the packet and byte counters for 'rule' will be credited for the packet sent
1437 * out whether or not the packet actually matches 'rule'.
1439 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1440 * the caller must already have accurately composed ODP actions for it given
1441 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1442 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1443 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1444 * actions and apply them to 'packet'. */
1446 rule_execute(struct ofproto *ofproto, struct rule *rule,
1447 struct ofpbuf *packet, const flow_t *flow)
1449 const union odp_action *actions;
1451 struct odp_actions a;
1453 /* Grab or compose the ODP actions.
1455 * The special case for an exact-match 'rule' where 'flow' is not the
1456 * rule's flow is important to avoid, e.g., sending a packet out its input
1457 * port simply because the ODP actions were composed for the wrong
1459 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1460 struct rule *super = rule->super ? rule->super : rule;
1461 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1462 packet, &a, NULL, 0, NULL)) {
1465 actions = a.actions;
1466 n_actions = a.n_actions;
1468 actions = rule->odp_actions;
1469 n_actions = rule->n_odp_actions;
1472 /* Execute the ODP actions. */
1473 if (!dpif_execute(ofproto->dpif, flow->in_port,
1474 actions, n_actions, packet)) {
1475 struct odp_flow_stats stats;
1476 flow_extract_stats(flow, packet, &stats);
1477 update_stats(ofproto, rule, &stats);
1478 rule->used = time_msec();
1479 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1484 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1487 struct rule *displaced_rule;
1489 /* Insert the rule in the classifier. */
1490 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1491 if (!rule->cr.wc.wildcards) {
1492 rule_make_actions(p, rule, packet);
1495 /* Send the packet and credit it to the rule. */
1498 flow_extract(packet, in_port, &flow);
1499 rule_execute(p, rule, packet, &flow);
1502 /* Install the rule in the datapath only after sending the packet, to
1503 * avoid packet reordering. */
1504 if (rule->cr.wc.wildcards) {
1505 COVERAGE_INC(ofproto_add_wc_flow);
1506 p->need_revalidate = true;
1508 rule_install(p, rule, displaced_rule);
1511 /* Free the rule that was displaced, if any. */
1512 if (displaced_rule) {
1513 rule_destroy(p, displaced_rule);
1517 static struct rule *
1518 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1521 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1522 rule->idle_timeout, rule->hard_timeout);
1523 COVERAGE_INC(ofproto_subrule_create);
1524 cls_rule_from_flow(&subrule->cr, flow, 0,
1525 (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1526 : rule->cr.priority));
1527 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1533 rule_remove(struct ofproto *ofproto, struct rule *rule)
1535 if (rule->cr.wc.wildcards) {
1536 COVERAGE_INC(ofproto_del_wc_flow);
1537 ofproto->need_revalidate = true;
1539 rule_uninstall(ofproto, rule);
1541 classifier_remove(&ofproto->cls, &rule->cr);
1542 rule_destroy(ofproto, rule);
1545 /* Returns true if the actions changed, false otherwise. */
1547 rule_make_actions(struct ofproto *p, struct rule *rule,
1548 const struct ofpbuf *packet)
1550 const struct rule *super;
1551 struct odp_actions a;
1554 assert(!rule->cr.wc.wildcards);
1556 super = rule->super ? rule->super : rule;
1558 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1559 packet, &a, &rule->tags, &rule->may_install,
1560 &rule->nf_flow.output_iface);
1562 actions_len = a.n_actions * sizeof *a.actions;
1563 if (rule->n_odp_actions != a.n_actions
1564 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1565 COVERAGE_INC(ofproto_odp_unchanged);
1566 free(rule->odp_actions);
1567 rule->n_odp_actions = a.n_actions;
1568 rule->odp_actions = xmemdup(a.actions, actions_len);
1576 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1577 struct odp_flow_put *put)
1579 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1580 put->flow.key = rule->cr.flow;
1581 put->flow.actions = rule->odp_actions;
1582 put->flow.n_actions = rule->n_odp_actions;
1584 return dpif_flow_put(ofproto->dpif, put);
1588 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1590 assert(!rule->cr.wc.wildcards);
1592 if (rule->may_install) {
1593 struct odp_flow_put put;
1594 if (!do_put_flow(p, rule,
1595 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1597 rule->installed = true;
1598 if (displaced_rule) {
1599 update_stats(p, displaced_rule, &put.flow.stats);
1600 rule_post_uninstall(p, displaced_rule);
1603 } else if (displaced_rule) {
1604 rule_uninstall(p, displaced_rule);
1609 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1611 if (rule->installed) {
1612 struct odp_flow_put put;
1613 COVERAGE_INC(ofproto_dp_missed);
1614 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1616 rule_install(ofproto, rule, NULL);
1621 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1623 bool actions_changed;
1624 uint16_t new_out_iface, old_out_iface;
1626 old_out_iface = rule->nf_flow.output_iface;
1627 actions_changed = rule_make_actions(ofproto, rule, NULL);
1629 if (rule->may_install) {
1630 if (rule->installed) {
1631 if (actions_changed) {
1632 struct odp_flow_put put;
1633 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
1634 | ODPPF_ZERO_STATS, &put);
1635 update_stats(ofproto, rule, &put.flow.stats);
1637 /* Temporarily set the old output iface so that NetFlow
1638 * messages have the correct output interface for the old
1640 new_out_iface = rule->nf_flow.output_iface;
1641 rule->nf_flow.output_iface = old_out_iface;
1642 rule_post_uninstall(ofproto, rule);
1643 rule->nf_flow.output_iface = new_out_iface;
1646 rule_install(ofproto, rule, NULL);
1649 rule_uninstall(ofproto, rule);
1654 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1656 uint64_t total_bytes = rule->byte_count + extra_bytes;
1658 if (ofproto->ofhooks->account_flow_cb
1659 && total_bytes > rule->accounted_bytes)
1661 ofproto->ofhooks->account_flow_cb(
1662 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1663 total_bytes - rule->accounted_bytes, ofproto->aux);
1664 rule->accounted_bytes = total_bytes;
1669 rule_uninstall(struct ofproto *p, struct rule *rule)
1671 assert(!rule->cr.wc.wildcards);
1672 if (rule->installed) {
1673 struct odp_flow odp_flow;
1675 odp_flow.key = rule->cr.flow;
1676 odp_flow.actions = NULL;
1677 odp_flow.n_actions = 0;
1678 if (!dpif_flow_del(p->dpif, &odp_flow)) {
1679 update_stats(p, rule, &odp_flow.stats);
1681 rule->installed = false;
1683 rule_post_uninstall(p, rule);
1688 is_controller_rule(struct rule *rule)
1690 /* If the only action is send to the controller then don't report
1691 * NetFlow expiration messages since it is just part of the control
1692 * logic for the network and not real traffic. */
1694 if (rule && rule->super) {
1695 struct rule *super = rule->super;
1697 return super->n_actions == 1 &&
1698 super->actions[0].type == htons(OFPAT_OUTPUT) &&
1699 super->actions[0].output.port == htons(OFPP_CONTROLLER);
1706 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1708 struct rule *super = rule->super;
1710 rule_account(ofproto, rule, 0);
1712 if (ofproto->netflow && !is_controller_rule(rule)) {
1713 struct ofexpired expired;
1714 expired.flow = rule->cr.flow;
1715 expired.packet_count = rule->packet_count;
1716 expired.byte_count = rule->byte_count;
1717 expired.used = rule->used;
1718 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
1721 super->packet_count += rule->packet_count;
1722 super->byte_count += rule->byte_count;
1724 /* Reset counters to prevent double counting if the rule ever gets
1726 rule->packet_count = 0;
1727 rule->byte_count = 0;
1728 rule->accounted_bytes = 0;
1730 netflow_flow_clear(&rule->nf_flow);
1735 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1736 struct rconn_packet_counter *counter)
1738 update_openflow_length(msg);
1739 if (rconn_send(ofconn->rconn, msg, counter)) {
1745 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1746 int error, const void *data, size_t len)
1749 struct ofp_error_msg *oem;
1751 if (!(error >> 16)) {
1752 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1757 COVERAGE_INC(ofproto_error);
1758 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1759 oh ? oh->xid : 0, &buf);
1760 oem->type = htons((unsigned int) error >> 16);
1761 oem->code = htons(error & 0xffff);
1762 memcpy(oem->data, data, len);
1763 queue_tx(buf, ofconn, ofconn->reply_counter);
1767 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1770 size_t oh_length = ntohs(oh->length);
1771 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1775 hton_ofp_phy_port(struct ofp_phy_port *opp)
1777 opp->port_no = htons(opp->port_no);
1778 opp->config = htonl(opp->config);
1779 opp->state = htonl(opp->state);
1780 opp->curr = htonl(opp->curr);
1781 opp->advertised = htonl(opp->advertised);
1782 opp->supported = htonl(opp->supported);
1783 opp->peer = htonl(opp->peer);
1787 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
1789 struct ofp_header *rq = oh;
1790 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
1795 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
1796 struct ofp_header *oh)
1798 struct ofp_switch_features *osf;
1800 unsigned int port_no;
1801 struct ofport *port;
1803 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1804 osf->datapath_id = htonll(p->datapath_id);
1805 osf->n_buffers = htonl(pktbuf_capacity());
1807 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1808 OFPC_PORT_STATS | OFPC_MULTI_PHY_TX);
1809 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1810 (1u << OFPAT_SET_VLAN_VID) |
1811 (1u << OFPAT_SET_VLAN_PCP) |
1812 (1u << OFPAT_STRIP_VLAN) |
1813 (1u << OFPAT_SET_DL_SRC) |
1814 (1u << OFPAT_SET_DL_DST) |
1815 (1u << OFPAT_SET_NW_SRC) |
1816 (1u << OFPAT_SET_NW_DST) |
1817 (1u << OFPAT_SET_TP_SRC) |
1818 (1u << OFPAT_SET_TP_DST));
1820 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1821 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1824 queue_tx(buf, ofconn, ofconn->reply_counter);
1829 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1830 struct ofp_header *oh)
1833 struct ofp_switch_config *osc;
1837 /* Figure out flags. */
1838 dpif_get_drop_frags(p->dpif, &drop_frags);
1839 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1840 if (ofconn->send_flow_exp) {
1841 flags |= OFPC_SEND_FLOW_EXP;
1845 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1846 osc->flags = htons(flags);
1847 osc->miss_send_len = htons(ofconn->miss_send_len);
1848 queue_tx(buf, ofconn, ofconn->reply_counter);
1854 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1855 struct ofp_switch_config *osc)
1860 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1864 flags = ntohs(osc->flags);
1866 ofconn->send_flow_exp = (flags & OFPC_SEND_FLOW_EXP) != 0;
1868 if (ofconn == p->controller) {
1869 switch (flags & OFPC_FRAG_MASK) {
1870 case OFPC_FRAG_NORMAL:
1871 dpif_set_drop_frags(p->dpif, false);
1873 case OFPC_FRAG_DROP:
1874 dpif_set_drop_frags(p->dpif, true);
1877 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1883 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1884 if (ntohs(osc->miss_send_len) != 0) {
1885 ofconn->pktbuf = pktbuf_create();
1887 pktbuf_destroy(ofconn->pktbuf);
1891 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1897 add_output_group_action(struct odp_actions *actions, uint16_t group,
1898 uint16_t *nf_output_iface)
1900 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1902 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
1903 *nf_output_iface = NF_OUT_FLOOD;
1908 add_controller_action(struct odp_actions *actions,
1909 const struct ofp_action_output *oao)
1911 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
1912 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
1915 struct action_xlate_ctx {
1917 const flow_t *flow; /* Flow to which these actions correspond. */
1918 int recurse; /* Recursion level, via xlate_table_action. */
1919 struct ofproto *ofproto;
1920 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
1921 * null pointer if we are revalidating
1922 * without a packet to refer to. */
1925 struct odp_actions *out; /* Datapath actions. */
1926 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
1927 bool may_set_up_flow; /* True ordinarily; false if the actions must
1928 * be reassessed for every packet. */
1929 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
1932 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
1933 struct action_xlate_ctx *ctx);
1936 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
1938 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
1941 if (ofport->opp.config & OFPPC_NO_FWD) {
1942 /* Forwarding disabled on port. */
1947 * We don't have an ofport record for this port, but it doesn't hurt to
1948 * allow forwarding to it anyhow. Maybe such a port will appear later
1949 * and we're pre-populating the flow table.
1953 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
1954 ctx->nf_output_iface = port;
1957 static struct rule *
1958 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
1961 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
1963 /* The rule we found might not be valid, since we could be in need of
1964 * revalidation. If it is not valid, don't return it. */
1967 && ofproto->need_revalidate
1968 && !revalidate_rule(ofproto, rule)) {
1969 COVERAGE_INC(ofproto_invalidated);
1977 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
1979 if (!ctx->recurse) {
1984 flow.in_port = in_port;
1986 rule = lookup_valid_rule(ctx->ofproto, &flow);
1993 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2000 xlate_output_action(struct action_xlate_ctx *ctx,
2001 const struct ofp_action_output *oao)
2004 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2006 ctx->nf_output_iface = NF_OUT_DROP;
2008 switch (ntohs(oao->port)) {
2010 add_output_action(ctx, ctx->flow->in_port);
2013 xlate_table_action(ctx, ctx->flow->in_port);
2016 if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet,
2017 ctx->out, ctx->tags,
2018 &ctx->nf_output_iface,
2019 ctx->ofproto->aux)) {
2020 COVERAGE_INC(ofproto_uninstallable);
2021 ctx->may_set_up_flow = false;
2025 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2026 &ctx->nf_output_iface);
2029 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2031 case OFPP_CONTROLLER:
2032 add_controller_action(ctx->out, oao);
2035 add_output_action(ctx, ODPP_LOCAL);
2038 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2039 if (odp_port != ctx->flow->in_port) {
2040 add_output_action(ctx, odp_port);
2045 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2046 ctx->nf_output_iface = NF_OUT_FLOOD;
2047 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2048 ctx->nf_output_iface = prev_nf_output_iface;
2049 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2050 ctx->nf_output_iface != NF_OUT_FLOOD) {
2051 ctx->nf_output_iface = NF_OUT_MULTI;
2056 xlate_nicira_action(struct action_xlate_ctx *ctx,
2057 const struct nx_action_header *nah)
2059 const struct nx_action_resubmit *nar;
2060 int subtype = ntohs(nah->subtype);
2062 assert(nah->vendor == htonl(NX_VENDOR_ID));
2064 case NXAST_RESUBMIT:
2065 nar = (const struct nx_action_resubmit *) nah;
2066 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2070 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2076 do_xlate_actions(const union ofp_action *in, size_t n_in,
2077 struct action_xlate_ctx *ctx)
2079 struct actions_iterator iter;
2080 const union ofp_action *ia;
2081 const struct ofport *port;
2083 port = port_array_get(&ctx->ofproto->ports, ctx->flow->in_port);
2084 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2085 port->opp.config & (eth_addr_equals(ctx->flow->dl_dst, stp_eth_addr)
2086 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2087 /* Drop this flow. */
2091 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2092 uint16_t type = ntohs(ia->type);
2093 union odp_action *oa;
2097 xlate_output_action(ctx, &ia->output);
2100 case OFPAT_SET_VLAN_VID:
2101 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2102 oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2105 case OFPAT_SET_VLAN_PCP:
2106 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2107 oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2110 case OFPAT_STRIP_VLAN:
2111 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2114 case OFPAT_SET_DL_SRC:
2115 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2116 memcpy(oa->dl_addr.dl_addr,
2117 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2120 case OFPAT_SET_DL_DST:
2121 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2122 memcpy(oa->dl_addr.dl_addr,
2123 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2126 case OFPAT_SET_NW_SRC:
2127 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2128 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2131 case OFPAT_SET_NW_DST:
2132 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2133 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2136 case OFPAT_SET_TP_SRC:
2137 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2138 oa->tp_port.tp_port = ia->tp_port.tp_port;
2141 case OFPAT_SET_TP_DST:
2142 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2143 oa->tp_port.tp_port = ia->tp_port.tp_port;
2147 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2151 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2158 xlate_actions(const union ofp_action *in, size_t n_in,
2159 const flow_t *flow, struct ofproto *ofproto,
2160 const struct ofpbuf *packet,
2161 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2162 uint16_t *nf_output_iface)
2164 tag_type no_tags = 0;
2165 struct action_xlate_ctx ctx;
2166 COVERAGE_INC(ofproto_ofp2odp);
2167 odp_actions_init(out);
2170 ctx.ofproto = ofproto;
2171 ctx.packet = packet;
2173 ctx.tags = tags ? tags : &no_tags;
2174 ctx.may_set_up_flow = true;
2175 ctx.nf_output_iface = NF_OUT_DROP;
2176 do_xlate_actions(in, n_in, &ctx);
2178 /* Check with in-band control to see if we're allowed to set up this
2180 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2181 ctx.may_set_up_flow = false;
2184 if (may_set_up_flow) {
2185 *may_set_up_flow = ctx.may_set_up_flow;
2187 if (nf_output_iface) {
2188 *nf_output_iface = ctx.nf_output_iface;
2190 if (odp_actions_overflow(out)) {
2191 odp_actions_init(out);
2192 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2198 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2199 struct ofp_header *oh)
2201 struct ofp_packet_out *opo;
2202 struct ofpbuf payload, *buffer;
2203 struct odp_actions actions;
2209 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2213 opo = (struct ofp_packet_out *) oh;
2215 COVERAGE_INC(ofproto_packet_out);
2216 if (opo->buffer_id != htonl(UINT32_MAX)) {
2217 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2219 if (error || !buffer) {
2227 flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2228 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2229 &flow, p, &payload, &actions, NULL, NULL, NULL);
2234 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2236 ofpbuf_delete(buffer);
2242 update_port_config(struct ofproto *p, struct ofport *port,
2243 uint32_t config, uint32_t mask)
2245 mask &= config ^ port->opp.config;
2246 if (mask & OFPPC_PORT_DOWN) {
2247 if (config & OFPPC_PORT_DOWN) {
2248 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2250 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2253 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2254 if (mask & REVALIDATE_BITS) {
2255 COVERAGE_INC(ofproto_costly_flags);
2256 port->opp.config ^= mask & REVALIDATE_BITS;
2257 p->need_revalidate = true;
2259 #undef REVALIDATE_BITS
2260 if (mask & OFPPC_NO_FLOOD) {
2261 port->opp.config ^= OFPPC_NO_FLOOD;
2262 refresh_port_group(p, DP_GROUP_FLOOD);
2264 if (mask & OFPPC_NO_PACKET_IN) {
2265 port->opp.config ^= OFPPC_NO_PACKET_IN;
2270 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2272 const struct ofp_port_mod *opm;
2273 struct ofport *port;
2276 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2280 opm = (struct ofp_port_mod *) oh;
2282 port = port_array_get(&p->ports,
2283 ofp_port_to_odp_port(ntohs(opm->port_no)));
2285 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2286 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2287 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2289 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2290 if (opm->advertise) {
2291 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2297 static struct ofpbuf *
2298 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2300 struct ofp_stats_reply *osr;
2303 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2304 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2306 osr->flags = htons(0);
2310 static struct ofpbuf *
2311 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2313 return make_stats_reply(request->header.xid, request->type, body_len);
2317 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2319 struct ofpbuf *msg = *msgp;
2320 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2321 if (nbytes + msg->size > UINT16_MAX) {
2322 struct ofp_stats_reply *reply = msg->data;
2323 reply->flags = htons(OFPSF_REPLY_MORE);
2324 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2325 queue_tx(msg, ofconn, ofconn->reply_counter);
2327 return ofpbuf_put_uninit(*msgp, nbytes);
2331 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2332 struct ofp_stats_request *request)
2334 struct ofp_desc_stats *ods;
2337 msg = start_stats_reply(request, sizeof *ods);
2338 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2339 strncpy(ods->mfr_desc, p->manufacturer, sizeof ods->mfr_desc);
2340 strncpy(ods->hw_desc, p->hardware, sizeof ods->hw_desc);
2341 strncpy(ods->sw_desc, p->software, sizeof ods->sw_desc);
2342 strncpy(ods->serial_num, p->serial, sizeof ods->serial_num);
2343 queue_tx(msg, ofconn, ofconn->reply_counter);
2349 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2351 struct rule *rule = rule_from_cls_rule(cls_rule);
2352 int *n_subrules = n_subrules_;
2360 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2361 struct ofp_stats_request *request)
2363 struct ofp_table_stats *ots;
2365 struct odp_stats dpstats;
2366 int n_exact, n_subrules, n_wild;
2368 msg = start_stats_reply(request, sizeof *ots * 2);
2370 /* Count rules of various kinds. */
2372 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2373 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2374 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2377 dpif_get_dp_stats(p->dpif, &dpstats);
2378 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2379 memset(ots, 0, sizeof *ots);
2380 ots->table_id = TABLEID_HASH;
2381 strcpy(ots->name, "hash");
2382 ots->wildcards = htonl(0);
2383 ots->max_entries = htonl(dpstats.max_capacity);
2384 ots->active_count = htonl(n_exact);
2385 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2387 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2389 /* Classifier table. */
2390 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2391 memset(ots, 0, sizeof *ots);
2392 ots->table_id = TABLEID_CLASSIFIER;
2393 strcpy(ots->name, "classifier");
2394 ots->wildcards = htonl(OFPFW_ALL);
2395 ots->max_entries = htonl(65536);
2396 ots->active_count = htonl(n_wild);
2397 ots->lookup_count = htonll(0); /* XXX */
2398 ots->matched_count = htonll(0); /* XXX */
2400 queue_tx(msg, ofconn, ofconn->reply_counter);
2405 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2406 struct ofp_stats_request *request)
2408 struct ofp_port_stats *ops;
2410 struct ofport *port;
2411 unsigned int port_no;
2413 msg = start_stats_reply(request, sizeof *ops * 16);
2414 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2415 struct netdev_stats stats;
2417 /* Intentionally ignore return value, since errors will set 'stats' to
2418 * all-1s, which is correct for OpenFlow, and netdev_get_stats() will
2420 netdev_get_stats(port->netdev, &stats);
2422 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2423 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2424 memset(ops->pad, 0, sizeof ops->pad);
2425 ops->rx_packets = htonll(stats.rx_packets);
2426 ops->tx_packets = htonll(stats.tx_packets);
2427 ops->rx_bytes = htonll(stats.rx_bytes);
2428 ops->tx_bytes = htonll(stats.tx_bytes);
2429 ops->rx_dropped = htonll(stats.rx_dropped);
2430 ops->tx_dropped = htonll(stats.tx_dropped);
2431 ops->rx_errors = htonll(stats.rx_errors);
2432 ops->tx_errors = htonll(stats.tx_errors);
2433 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2434 ops->rx_over_err = htonll(stats.rx_over_errors);
2435 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2436 ops->collisions = htonll(stats.collisions);
2439 queue_tx(msg, ofconn, ofconn->reply_counter);
2443 struct flow_stats_cbdata {
2444 struct ofproto *ofproto;
2445 struct ofconn *ofconn;
2451 query_stats(struct ofproto *p, struct rule *rule,
2452 uint64_t *packet_countp, uint64_t *byte_countp)
2454 uint64_t packet_count, byte_count;
2455 struct rule *subrule;
2456 struct odp_flow *odp_flows;
2459 packet_count = rule->packet_count;
2460 byte_count = rule->byte_count;
2462 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2463 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
2464 if (rule->cr.wc.wildcards) {
2466 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2467 odp_flows[i++].key = subrule->cr.flow;
2468 packet_count += subrule->packet_count;
2469 byte_count += subrule->byte_count;
2472 odp_flows[0].key = rule->cr.flow;
2475 packet_count = rule->packet_count;
2476 byte_count = rule->byte_count;
2477 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2479 for (i = 0; i < n_odp_flows; i++) {
2480 struct odp_flow *odp_flow = &odp_flows[i];
2481 packet_count += odp_flow->stats.n_packets;
2482 byte_count += odp_flow->stats.n_bytes;
2487 *packet_countp = packet_count;
2488 *byte_countp = byte_count;
2492 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2494 struct rule *rule = rule_from_cls_rule(rule_);
2495 struct flow_stats_cbdata *cbdata = cbdata_;
2496 struct ofp_flow_stats *ofs;
2497 uint64_t packet_count, byte_count;
2498 size_t act_len, len;
2500 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2504 act_len = sizeof *rule->actions * rule->n_actions;
2505 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2507 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2509 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2510 ofs->length = htons(len);
2511 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2513 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match);
2514 ofs->duration = htonl((time_msec() - rule->created) / 1000);
2515 ofs->priority = htons(rule->cr.priority);
2516 ofs->idle_timeout = htons(rule->idle_timeout);
2517 ofs->hard_timeout = htons(rule->hard_timeout);
2518 memset(ofs->pad2, 0, sizeof ofs->pad2);
2519 ofs->packet_count = htonll(packet_count);
2520 ofs->byte_count = htonll(byte_count);
2521 memcpy(ofs->actions, rule->actions, act_len);
2525 table_id_to_include(uint8_t table_id)
2527 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2528 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2529 : table_id == 0xff ? CLS_INC_ALL
2534 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2535 const struct ofp_stats_request *osr,
2538 struct ofp_flow_stats_request *fsr;
2539 struct flow_stats_cbdata cbdata;
2540 struct cls_rule target;
2542 if (arg_size != sizeof *fsr) {
2543 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2545 fsr = (struct ofp_flow_stats_request *) osr->body;
2547 COVERAGE_INC(ofproto_flows_req);
2549 cbdata.ofconn = ofconn;
2550 cbdata.out_port = fsr->out_port;
2551 cbdata.msg = start_stats_reply(osr, 1024);
2552 cls_rule_from_match(&target, &fsr->match, 0);
2553 classifier_for_each_match(&p->cls, &target,
2554 table_id_to_include(fsr->table_id),
2555 flow_stats_cb, &cbdata);
2556 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2560 struct flow_stats_ds_cbdata {
2561 struct ofproto *ofproto;
2566 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
2568 struct rule *rule = rule_from_cls_rule(rule_);
2569 struct flow_stats_ds_cbdata *cbdata = cbdata_;
2570 struct ds *results = cbdata->results;
2571 struct ofp_match match;
2572 uint64_t packet_count, byte_count;
2573 size_t act_len = sizeof *rule->actions * rule->n_actions;
2575 /* Don't report on subrules. */
2576 if (rule->super != NULL) {
2580 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2581 flow_to_ovs_match(&rule->cr.flow, rule->cr.wc.wildcards, &match);
2583 ds_put_format(results, "duration=%llds, ",
2584 (time_msec() - rule->created) / 1000);
2585 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2586 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2587 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2588 ofp_print_match(results, &match, true);
2589 ofp_print_actions(results, &rule->actions->header, act_len);
2590 ds_put_cstr(results, "\n");
2593 /* Adds a pretty-printed description of all flows to 'results', including
2594 * those marked hidden by secchan (e.g., by in-band control). */
2596 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2598 struct ofp_match match;
2599 struct cls_rule target;
2600 struct flow_stats_ds_cbdata cbdata;
2602 memset(&match, 0, sizeof match);
2603 match.wildcards = htonl(OFPFW_ALL);
2606 cbdata.results = results;
2608 cls_rule_from_match(&target, &match, 0);
2609 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2610 flow_stats_ds_cb, &cbdata);
2613 struct aggregate_stats_cbdata {
2614 struct ofproto *ofproto;
2616 uint64_t packet_count;
2617 uint64_t byte_count;
2622 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2624 struct rule *rule = rule_from_cls_rule(rule_);
2625 struct aggregate_stats_cbdata *cbdata = cbdata_;
2626 uint64_t packet_count, byte_count;
2628 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2632 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2634 cbdata->packet_count += packet_count;
2635 cbdata->byte_count += byte_count;
2640 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2641 const struct ofp_stats_request *osr,
2644 struct ofp_aggregate_stats_request *asr;
2645 struct ofp_aggregate_stats_reply *reply;
2646 struct aggregate_stats_cbdata cbdata;
2647 struct cls_rule target;
2650 if (arg_size != sizeof *asr) {
2651 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2653 asr = (struct ofp_aggregate_stats_request *) osr->body;
2655 COVERAGE_INC(ofproto_agg_request);
2657 cbdata.out_port = asr->out_port;
2658 cbdata.packet_count = 0;
2659 cbdata.byte_count = 0;
2661 cls_rule_from_match(&target, &asr->match, 0);
2662 classifier_for_each_match(&p->cls, &target,
2663 table_id_to_include(asr->table_id),
2664 aggregate_stats_cb, &cbdata);
2666 msg = start_stats_reply(osr, sizeof *reply);
2667 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2668 reply->flow_count = htonl(cbdata.n_flows);
2669 reply->packet_count = htonll(cbdata.packet_count);
2670 reply->byte_count = htonll(cbdata.byte_count);
2671 queue_tx(msg, ofconn, ofconn->reply_counter);
2676 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2677 struct ofp_header *oh)
2679 struct ofp_stats_request *osr;
2683 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2688 osr = (struct ofp_stats_request *) oh;
2690 switch (ntohs(osr->type)) {
2692 return handle_desc_stats_request(p, ofconn, osr);
2695 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2697 case OFPST_AGGREGATE:
2698 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2701 return handle_table_stats_request(p, ofconn, osr);
2704 return handle_port_stats_request(p, ofconn, osr);
2707 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2710 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2714 static long long int
2715 msec_from_nsec(uint64_t sec, uint32_t nsec)
2717 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2721 update_time(struct ofproto *ofproto, struct rule *rule,
2722 const struct odp_flow_stats *stats)
2724 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2725 if (used > rule->used) {
2727 if (rule->super && used > rule->super->used) {
2728 rule->super->used = used;
2730 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
2735 update_stats(struct ofproto *ofproto, struct rule *rule,
2736 const struct odp_flow_stats *stats)
2738 if (stats->n_packets) {
2739 update_time(ofproto, rule, stats);
2740 rule->packet_count += stats->n_packets;
2741 rule->byte_count += stats->n_bytes;
2742 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
2748 add_flow(struct ofproto *p, struct ofconn *ofconn,
2749 struct ofp_flow_mod *ofm, size_t n_actions)
2751 struct ofpbuf *packet;
2756 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
2757 n_actions, ntohs(ofm->idle_timeout),
2758 ntohs(ofm->hard_timeout));
2759 cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority));
2763 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2764 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2768 rule_insert(p, rule, packet, in_port);
2769 ofpbuf_delete(packet);
2774 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
2775 size_t n_actions, uint16_t command, struct rule *rule)
2777 if (rule_is_hidden(rule)) {
2781 if (command == OFPFC_DELETE) {
2782 rule_remove(p, rule);
2784 size_t actions_len = n_actions * sizeof *rule->actions;
2786 if (n_actions == rule->n_actions
2787 && !memcmp(ofm->actions, rule->actions, actions_len))
2792 free(rule->actions);
2793 rule->actions = xmemdup(ofm->actions, actions_len);
2794 rule->n_actions = n_actions;
2796 if (rule->cr.wc.wildcards) {
2797 COVERAGE_INC(ofproto_mod_wc_flow);
2798 p->need_revalidate = true;
2800 rule_update_actions(p, rule);
2808 modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm,
2809 size_t n_actions, uint16_t command)
2815 flow_from_match(&flow, &wildcards, &ofm->match);
2816 rule = rule_from_cls_rule(classifier_find_rule_exactly(
2817 &p->cls, &flow, wildcards,
2818 ntohs(ofm->priority)));
2821 if (command == OFPFC_DELETE
2822 && ofm->out_port != htons(OFPP_NONE)
2823 && !rule_has_out_port(rule, ofm->out_port)) {
2827 modify_flow(p, ofm, n_actions, command, rule);
2832 struct modify_flows_cbdata {
2833 struct ofproto *ofproto;
2834 const struct ofp_flow_mod *ofm;
2841 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
2843 struct rule *rule = rule_from_cls_rule(rule_);
2844 struct modify_flows_cbdata *cbdata = cbdata_;
2846 if (cbdata->out_port != htons(OFPP_NONE)
2847 && !rule_has_out_port(rule, cbdata->out_port)) {
2851 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions,
2852 cbdata->command, rule);
2856 modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm,
2857 size_t n_actions, uint16_t command)
2859 struct modify_flows_cbdata cbdata;
2860 struct cls_rule target;
2864 cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port
2865 : htons(OFPP_NONE));
2866 cbdata.n_actions = n_actions;
2867 cbdata.command = command;
2869 cls_rule_from_match(&target, &ofm->match, 0);
2871 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2872 modify_flows_cb, &cbdata);
2877 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
2878 struct ofp_flow_mod *ofm)
2883 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
2884 sizeof *ofm->actions, &n_actions);
2889 normalize_match(&ofm->match);
2890 if (!ofm->match.wildcards) {
2891 ofm->priority = htons(UINT16_MAX);
2894 error = validate_actions((const union ofp_action *) ofm->actions,
2895 n_actions, p->max_ports);
2900 switch (ntohs(ofm->command)) {
2902 return add_flow(p, ofconn, ofm, n_actions);
2905 return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY);
2907 case OFPFC_MODIFY_STRICT:
2908 return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY);
2911 return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE);
2913 case OFPFC_DELETE_STRICT:
2914 return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE);
2917 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
2922 send_capability_reply(struct ofproto *p, struct ofconn *ofconn, uint32_t xid)
2924 struct ofmp_capability_reply *ocr;
2926 char capabilities[] = "com.nicira.mgmt.manager=false\n";
2928 ocr = make_openflow_xid(sizeof(*ocr), OFPT_VENDOR, xid, &b);
2929 ocr->header.header.vendor = htonl(NX_VENDOR_ID);
2930 ocr->header.header.subtype = htonl(NXT_MGMT);
2931 ocr->header.type = htons(OFMPT_CAPABILITY_REPLY);
2933 ocr->format = htonl(OFMPCOF_SIMPLE);
2934 ocr->mgmt_id = htonll(p->mgmt_id);
2936 ofpbuf_put(b, capabilities, strlen(capabilities));
2938 queue_tx(b, ofconn, ofconn->reply_counter);
2942 handle_ofmp(struct ofproto *p, struct ofconn *ofconn,
2943 struct ofmp_header *ofmph)
2945 size_t msg_len = ntohs(ofmph->header.header.length);
2946 if (msg_len < sizeof(*ofmph)) {
2947 VLOG_WARN_RL(&rl, "dropping short managment message: %zu\n", msg_len);
2948 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2951 if (ofmph->type == htons(OFMPT_CAPABILITY_REQUEST)) {
2952 struct ofmp_capability_request *ofmpcr;
2954 if (msg_len < sizeof(struct ofmp_capability_request)) {
2955 VLOG_WARN_RL(&rl, "dropping short capability request: %zu\n",
2957 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2960 ofmpcr = (struct ofmp_capability_request *)ofmph;
2961 if (ofmpcr->format != htonl(OFMPCAF_SIMPLE)) {
2962 /* xxx Find a better type than bad subtype */
2963 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2966 send_capability_reply(p, ofconn, ofmph->header.header.xid);
2969 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2974 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
2976 struct ofp_vendor_header *ovh = msg;
2977 struct nicira_header *nh;
2979 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
2980 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2982 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
2983 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2985 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
2986 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2990 switch (ntohl(nh->subtype)) {
2991 case NXT_STATUS_REQUEST:
2992 return switch_status_handle_request(p->switch_status, ofconn->rconn,
2995 case NXT_ACT_SET_CONFIG:
2996 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
2998 case NXT_ACT_GET_CONFIG:
2999 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
3002 return handle_ofmp(p, ofconn, msg);
3005 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3009 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3010 struct ofpbuf *ofp_msg)
3012 struct ofp_header *oh = ofp_msg->data;
3015 COVERAGE_INC(ofproto_recv_openflow);
3017 case OFPT_ECHO_REQUEST:
3018 error = handle_echo_request(ofconn, oh);
3021 case OFPT_ECHO_REPLY:
3025 case OFPT_FEATURES_REQUEST:
3026 error = handle_features_request(p, ofconn, oh);
3029 case OFPT_GET_CONFIG_REQUEST:
3030 error = handle_get_config_request(p, ofconn, oh);
3033 case OFPT_SET_CONFIG:
3034 error = handle_set_config(p, ofconn, ofp_msg->data);
3037 case OFPT_PACKET_OUT:
3038 error = handle_packet_out(p, ofconn, ofp_msg->data);
3042 error = handle_port_mod(p, oh);
3046 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3049 case OFPT_STATS_REQUEST:
3050 error = handle_stats_request(p, ofconn, oh);
3054 error = handle_vendor(p, ofconn, ofp_msg->data);
3058 if (VLOG_IS_WARN_ENABLED()) {
3059 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3060 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3063 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3068 send_error_oh(ofconn, ofp_msg->data, error);
3073 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3075 struct odp_msg *msg = packet->data;
3076 uint16_t in_port = odp_port_to_ofp_port(msg->port);
3078 struct ofpbuf payload;
3081 /* Handle controller actions. */
3082 if (msg->type == _ODPL_ACTION_NR) {
3083 COVERAGE_INC(ofproto_ctlr_action);
3084 pinsched_send(p->action_sched, in_port, packet,
3085 send_packet_in_action, p);
3089 payload.data = msg + 1;
3090 payload.size = msg->length - sizeof *msg;
3091 flow_extract(&payload, msg->port, &flow);
3093 /* Check with in-band control to see if this packet should be sent
3094 * to the local port regardless of the flow table. */
3095 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3096 union odp_action action;
3098 memset(&action, 0, sizeof(action));
3099 action.output.type = ODPAT_OUTPUT;
3100 action.output.port = ODPP_LOCAL;
3101 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3104 rule = lookup_valid_rule(p, &flow);
3106 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3107 struct ofport *port = port_array_get(&p->ports, msg->port);
3109 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3110 COVERAGE_INC(ofproto_no_packet_in);
3111 /* XXX install 'drop' flow entry */
3112 ofpbuf_delete(packet);
3116 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3119 COVERAGE_INC(ofproto_packet_in);
3120 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3124 if (rule->cr.wc.wildcards) {
3125 rule = rule_create_subrule(p, rule, &flow);
3126 rule_make_actions(p, rule, packet);
3128 if (!rule->may_install) {
3129 /* The rule is not installable, that is, we need to process every
3130 * packet, so process the current packet and set its actions into
3132 rule_make_actions(p, rule, packet);
3134 /* XXX revalidate rule if it needs it */
3138 rule_execute(p, rule, &payload, &flow);
3139 rule_reinstall(p, rule);
3141 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY
3142 && rconn_is_connected(p->controller->rconn)) {
3144 * Extra-special case for fail-open mode.
3146 * We are in fail-open mode and the packet matched the fail-open rule,
3147 * but we are connected to a controller too. We should send the packet
3148 * up to the controller in the hope that it will try to set up a flow
3149 * and thereby allow us to exit fail-open.
3151 * See the top-level comment in fail-open.c for more information.
3153 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3155 ofpbuf_delete(packet);
3160 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3162 struct rule *sub = rule_from_cls_rule(sub_);
3163 struct revalidate_cbdata *cbdata = cbdata_;
3165 if (cbdata->revalidate_all
3166 || (cbdata->revalidate_subrules && sub->super)
3167 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3168 revalidate_rule(cbdata->ofproto, sub);
3173 revalidate_rule(struct ofproto *p, struct rule *rule)
3175 const flow_t *flow = &rule->cr.flow;
3177 COVERAGE_INC(ofproto_revalidate_rule);
3180 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3182 rule_remove(p, rule);
3184 } else if (super != rule->super) {
3185 COVERAGE_INC(ofproto_revalidate_moved);
3186 list_remove(&rule->list);
3187 list_push_back(&super->list, &rule->list);
3188 rule->super = super;
3189 rule->hard_timeout = super->hard_timeout;
3190 rule->idle_timeout = super->idle_timeout;
3191 rule->created = super->created;
3196 rule_update_actions(p, rule);
3200 static struct ofpbuf *
3201 compose_flow_exp(const struct rule *rule, long long int now, uint8_t reason)
3203 struct ofp_flow_expired *ofe;
3206 ofe = make_openflow(sizeof *ofe, OFPT_FLOW_EXPIRED, &buf);
3207 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofe->match);
3208 ofe->priority = htons(rule->cr.priority);
3209 ofe->reason = reason;
3210 ofe->duration = htonl((now - rule->created) / 1000);
3211 ofe->packet_count = htonll(rule->packet_count);
3212 ofe->byte_count = htonll(rule->byte_count);
3218 send_flow_exp(struct ofproto *p, struct rule *rule,
3219 long long int now, uint8_t reason)
3221 struct ofconn *ofconn;
3222 struct ofconn *prev;
3223 struct ofpbuf *buf = NULL;
3225 /* We limit the maximum number of queued flow expirations it by accounting
3226 * them under the counter for replies. That works because preventing
3227 * OpenFlow requests from being processed also prevents new flows from
3228 * being added (and expiring). (It also prevents processing OpenFlow
3229 * requests that would not add new flows, so it is imperfect.) */
3232 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3233 if (ofconn->send_flow_exp && rconn_is_connected(ofconn->rconn)) {
3235 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3237 buf = compose_flow_exp(rule, now, reason);
3243 queue_tx(buf, prev, prev->reply_counter);
3248 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3250 assert(rule->installed);
3251 assert(!rule->cr.wc.wildcards);
3254 rule_remove(ofproto, rule);
3256 rule_uninstall(ofproto, rule);
3261 expire_rule(struct cls_rule *cls_rule, void *p_)
3263 struct ofproto *p = p_;
3264 struct rule *rule = rule_from_cls_rule(cls_rule);
3265 long long int hard_expire, idle_expire, expire, now;
3267 hard_expire = (rule->hard_timeout
3268 ? rule->created + rule->hard_timeout * 1000
3270 idle_expire = (rule->idle_timeout
3271 && (rule->super || list_is_empty(&rule->list))
3272 ? rule->used + rule->idle_timeout * 1000
3274 expire = MIN(hard_expire, idle_expire);
3278 if (rule->installed && now >= rule->used + 5000) {
3279 uninstall_idle_flow(p, rule);
3280 } else if (!rule->cr.wc.wildcards) {
3281 active_timeout(p, rule);
3287 COVERAGE_INC(ofproto_expired);
3289 /* Update stats. This code will be a no-op if the rule expired
3290 * due to an idle timeout. */
3291 if (rule->cr.wc.wildcards) {
3292 struct rule *subrule, *next;
3293 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3294 rule_remove(p, subrule);
3297 rule_uninstall(p, rule);
3300 if (!rule_is_hidden(rule)) {
3301 send_flow_exp(p, rule, now,
3303 ? OFPER_HARD_TIMEOUT : OFPER_IDLE_TIMEOUT));
3305 rule_remove(p, rule);
3309 active_timeout(struct ofproto *ofproto, struct rule *rule)
3311 if (ofproto->netflow && !is_controller_rule(rule) &&
3312 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
3313 struct ofexpired expired;
3314 struct odp_flow odp_flow;
3316 /* Get updated flow stats. */
3317 memset(&odp_flow, 0, sizeof odp_flow);
3318 if (rule->installed) {
3319 odp_flow.key = rule->cr.flow;
3320 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
3321 dpif_flow_get(ofproto->dpif, &odp_flow);
3323 if (odp_flow.stats.n_packets) {
3324 update_time(ofproto, rule, &odp_flow.stats);
3325 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
3326 odp_flow.stats.tcp_flags);
3330 expired.flow = rule->cr.flow;
3331 expired.packet_count = rule->packet_count +
3332 odp_flow.stats.n_packets;
3333 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
3334 expired.used = rule->used;
3336 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
3338 /* Schedule us to send the accumulated records once we have
3339 * collected all of them. */
3340 poll_immediate_wake();
3345 update_used(struct ofproto *p)
3347 struct odp_flow *flows;
3352 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
3357 for (i = 0; i < n_flows; i++) {
3358 struct odp_flow *f = &flows[i];
3361 rule = rule_from_cls_rule(
3362 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3363 if (!rule || !rule->installed) {
3364 COVERAGE_INC(ofproto_unexpected_rule);
3365 dpif_flow_del(p->dpif, f);
3369 update_time(p, rule, &f->stats);
3370 rule_account(p, rule, f->stats.n_bytes);
3376 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3377 const struct ofpbuf *packet, int send_len)
3379 struct odp_msg *msg = packet->data;
3380 struct ofpbuf payload;
3384 /* Extract packet payload from 'msg'. */
3385 payload.data = msg + 1;
3386 payload.size = msg->length - sizeof *msg;
3388 /* Construct ofp_packet_in message. */
3389 reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3390 opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port), reason,
3391 &payload, send_len);
3394 rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100);
3398 send_packet_in_action(struct ofpbuf *packet, void *p_)
3400 struct ofproto *p = p_;
3401 struct ofconn *ofconn;
3402 struct odp_msg *msg;
3405 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3406 if (ofconn == p->controller || ofconn->miss_send_len) {
3407 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3410 ofpbuf_delete(packet);
3414 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3416 struct ofproto *p = p_;
3417 bool in_fail_open = p->fail_open && fail_open_is_active(p->fail_open);
3418 struct ofconn *ofconn;
3419 struct ofpbuf payload;
3420 struct odp_msg *msg;
3423 payload.data = msg + 1;
3424 payload.size = msg->length - sizeof *msg;
3425 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3426 if (ofconn->miss_send_len) {
3427 struct pktbuf *pb = ofconn->pktbuf;
3428 uint32_t buffer_id = (in_fail_open
3430 : pktbuf_save(pb, &payload, msg->port));
3431 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3433 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3436 ofpbuf_delete(packet);
3440 pick_datapath_id(const struct ofproto *ofproto)
3442 const struct ofport *port;
3444 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
3446 uint8_t ea[ETH_ADDR_LEN];
3449 error = netdev_get_etheraddr(port->netdev, ea);
3451 return eth_addr_to_uint64(ea);
3453 VLOG_WARN("could not get MAC address for %s (%s)",
3454 netdev_get_name(port->netdev), strerror(error));
3456 return ofproto->fallback_dpid;
3460 pick_fallback_dpid(void)
3462 uint8_t ea[ETH_ADDR_LEN];
3463 eth_addr_random(ea);
3464 ea[0] = 0x00; /* Set Nicira OUI. */
3467 return eth_addr_to_uint64(ea);
3471 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3472 struct odp_actions *actions, tag_type *tags,
3473 uint16_t *nf_output_iface, void *ofproto_)
3475 struct ofproto *ofproto = ofproto_;
3478 /* Drop frames for reserved multicast addresses. */
3479 if (eth_addr_is_reserved(flow->dl_dst)) {
3483 /* Learn source MAC (but don't try to learn from revalidation). */
3484 if (packet != NULL) {
3485 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3488 /* The log messages here could actually be useful in debugging,
3489 * so keep the rate limit relatively high. */
3490 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3491 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3492 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3493 ofproto_revalidate(ofproto, rev_tag);
3497 /* Determine output port. */
3498 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3500 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
3501 } else if (out_port != flow->in_port) {
3502 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3503 *nf_output_iface = out_port;
3511 static const struct ofhooks default_ofhooks = {
3513 default_normal_ofhook_cb,