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"
37 #include "ofproto-sflow.h"
39 #include "openflow/nicira-ext.h"
40 #include "openflow/openflow.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
62 #include "sflow_api.h"
66 TABLEID_CLASSIFIER = 1
70 struct netdev *netdev;
71 struct ofp_phy_port opp; /* In host byte order. */
74 static void ofport_free(struct ofport *);
75 static void hton_ofp_phy_port(struct ofp_phy_port *);
77 static int xlate_actions(const union ofp_action *in, size_t n_in,
78 const flow_t *flow, struct ofproto *ofproto,
79 const struct ofpbuf *packet,
80 struct odp_actions *out, tag_type *tags,
81 bool *may_set_up_flow, uint16_t *nf_output_iface);
86 uint64_t flow_cookie; /* Controller-issued identifier.
87 (Kept in network-byte order.) */
88 uint16_t idle_timeout; /* In seconds from time of last use. */
89 uint16_t hard_timeout; /* In seconds from time of creation. */
90 bool send_flow_removed; /* Send a flow removed message? */
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 uint64_t flow_cookie, bool send_flow_removed);
150 static void rule_free(struct rule *);
151 static void rule_destroy(struct ofproto *, struct rule *);
152 static struct rule *rule_from_cls_rule(const struct cls_rule *);
153 static void rule_insert(struct ofproto *, struct rule *,
154 struct ofpbuf *packet, uint16_t in_port);
155 static void rule_remove(struct ofproto *, struct rule *);
156 static bool rule_make_actions(struct ofproto *, struct rule *,
157 const struct ofpbuf *packet);
158 static void rule_install(struct ofproto *, struct rule *,
159 struct rule *displaced_rule);
160 static void rule_uninstall(struct ofproto *, struct rule *);
161 static void rule_post_uninstall(struct ofproto *, struct rule *);
162 static void send_flow_removed(struct ofproto *p, struct rule *rule,
163 long long int now, uint8_t reason);
168 struct pktbuf *pktbuf;
171 struct rconn_packet_counter *packet_in_counter;
173 /* Number of OpenFlow messages queued as replies to OpenFlow requests, and
174 * the maximum number before we stop reading OpenFlow requests. */
175 #define OFCONN_REPLY_MAX 100
176 struct rconn_packet_counter *reply_counter;
179 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *);
180 static void ofconn_destroy(struct ofconn *);
181 static void ofconn_run(struct ofconn *, struct ofproto *);
182 static void ofconn_wait(struct ofconn *);
183 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
184 struct rconn_packet_counter *counter);
188 uint64_t datapath_id; /* Datapath ID. */
189 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
190 char *manufacturer; /* Manufacturer. */
191 char *hardware; /* Hardware. */
192 char *software; /* Software version. */
193 char *serial; /* Serial number. */
194 char *dp_desc; /* Datapath description. */
198 struct netdev_monitor *netdev_monitor;
199 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
201 struct shash port_by_name;
205 struct switch_status *switch_status;
206 struct status_category *ss_cat;
207 struct in_band *in_band;
208 struct discovery *discovery;
209 struct fail_open *fail_open;
210 struct pinsched *miss_sched, *action_sched;
211 struct netflow *netflow;
212 struct ofproto_sflow *sflow;
215 struct classifier cls;
216 bool need_revalidate;
217 long long int next_expiration;
218 struct tag_set revalidate_set;
220 /* OpenFlow connections. */
221 struct list all_conns;
222 struct ofconn *controller;
223 struct pvconn **listeners;
225 struct pvconn **snoops;
228 /* Hooks for ovs-vswitchd. */
229 const struct ofhooks *ofhooks;
232 /* Used by default ofhooks. */
233 struct mac_learning *ml;
236 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
238 static const struct ofhooks default_ofhooks;
240 static uint64_t pick_datapath_id(const struct ofproto *);
241 static uint64_t pick_fallback_dpid(void);
242 static void send_packet_in_miss(struct ofpbuf *, void *ofproto);
243 static void send_packet_in_action(struct ofpbuf *, void *ofproto);
244 static void update_used(struct ofproto *);
245 static void update_stats(struct ofproto *, struct rule *,
246 const struct odp_flow_stats *);
247 static void expire_rule(struct cls_rule *, void *ofproto);
248 static void active_timeout(struct ofproto *ofproto, struct rule *rule);
249 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
250 static void revalidate_cb(struct cls_rule *rule_, void *p_);
252 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
254 static void handle_openflow(struct ofconn *, struct ofproto *,
257 static void refresh_port_groups(struct ofproto *);
259 static void update_port(struct ofproto *, const char *devname);
260 static int init_ports(struct ofproto *);
261 static void reinit_ports(struct ofproto *);
264 ofproto_create(const char *datapath, const char *datapath_type,
265 const struct ofhooks *ofhooks, void *aux,
266 struct ofproto **ofprotop)
268 struct odp_stats stats;
275 /* Connect to datapath and start listening for messages. */
276 error = dpif_open(datapath, datapath_type, &dpif);
278 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
281 error = dpif_get_dp_stats(dpif, &stats);
283 VLOG_ERR("failed to obtain stats for datapath %s: %s",
284 datapath, strerror(error));
288 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
290 VLOG_ERR("failed to listen on datapath %s: %s",
291 datapath, strerror(error));
295 dpif_flow_flush(dpif);
296 dpif_recv_purge(dpif);
298 /* Initialize settings. */
299 p = xzalloc(sizeof *p);
300 p->fallback_dpid = pick_fallback_dpid();
301 p->datapath_id = p->fallback_dpid;
302 p->manufacturer = xstrdup("Nicira Networks, Inc.");
303 p->hardware = xstrdup("Reference Implementation");
304 p->software = xstrdup(VERSION BUILDNR);
305 p->serial = xstrdup("None");
306 p->dp_desc = xstrdup("None");
308 /* Initialize datapath. */
310 p->netdev_monitor = netdev_monitor_create();
311 port_array_init(&p->ports);
312 shash_init(&p->port_by_name);
313 p->max_ports = stats.max_ports;
315 /* Initialize submodules. */
316 p->switch_status = switch_status_create(p);
320 p->miss_sched = p->action_sched = NULL;
324 /* Initialize flow table. */
325 classifier_init(&p->cls);
326 p->need_revalidate = false;
327 p->next_expiration = time_msec() + 1000;
328 tag_set_init(&p->revalidate_set);
330 /* Initialize OpenFlow connections. */
331 list_init(&p->all_conns);
332 p->controller = ofconn_create(p, rconn_create(5, 8));
333 p->controller->pktbuf = pktbuf_create();
334 p->controller->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
340 /* Initialize hooks. */
342 p->ofhooks = ofhooks;
346 p->ofhooks = &default_ofhooks;
348 p->ml = mac_learning_create();
351 /* Register switch status category. */
352 p->ss_cat = switch_status_register(p->switch_status, "remote",
353 rconn_status_cb, p->controller->rconn);
355 /* Pick final datapath ID. */
356 p->datapath_id = pick_datapath_id(p);
357 VLOG_INFO("using datapath ID %016"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 %016"PRIx64, p->datapath_id);
370 rconn_reconnect(p->controller->rconn);
375 ofproto_set_probe_interval(struct ofproto *p, int probe_interval)
377 probe_interval = probe_interval ? MAX(probe_interval, 5) : 0;
378 rconn_set_probe_interval(p->controller->rconn, probe_interval);
380 int trigger_duration = probe_interval ? probe_interval * 3 : 15;
381 fail_open_set_trigger_duration(p->fail_open, trigger_duration);
386 ofproto_set_max_backoff(struct ofproto *p, int max_backoff)
388 rconn_set_max_backoff(p->controller->rconn, max_backoff);
392 ofproto_set_desc(struct ofproto *p,
393 const char *manufacturer, const char *hardware,
394 const char *software, const char *serial,
398 free(p->manufacturer);
399 p->manufacturer = xstrdup(manufacturer);
403 p->hardware = xstrdup(hardware);
407 p->software = xstrdup(software);
411 p->serial = xstrdup(serial);
415 p->dp_desc = xstrdup(dp_desc);
420 ofproto_set_in_band(struct ofproto *p, bool in_band)
422 if (in_band != (p->in_band != NULL)) {
424 return in_band_create(p, p->dpif, p->switch_status,
425 p->controller->rconn, &p->in_band);
427 ofproto_set_discovery(p, false, NULL, true);
428 in_band_destroy(p->in_band);
431 rconn_reconnect(p->controller->rconn);
437 ofproto_set_discovery(struct ofproto *p, bool discovery,
438 const char *re, bool update_resolv_conf)
440 if (discovery != (p->discovery != NULL)) {
442 int error = ofproto_set_in_band(p, true);
446 error = discovery_create(re, update_resolv_conf,
447 p->dpif, p->switch_status,
453 discovery_destroy(p->discovery);
456 rconn_disconnect(p->controller->rconn);
457 } else if (discovery) {
458 discovery_set_update_resolv_conf(p->discovery, update_resolv_conf);
459 return discovery_set_accept_controller_re(p->discovery, re);
465 ofproto_set_controller(struct ofproto *ofproto, const char *controller)
467 if (ofproto->discovery) {
469 } else if (controller) {
470 if (strcmp(rconn_get_name(ofproto->controller->rconn), controller)) {
471 return rconn_connect(ofproto->controller->rconn, controller);
476 rconn_disconnect(ofproto->controller->rconn);
482 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
483 const struct svec *svec)
485 struct pvconn **pvconns = *pvconnsp;
486 size_t n_pvconns = *n_pvconnsp;
490 for (i = 0; i < n_pvconns; i++) {
491 pvconn_close(pvconns[i]);
495 pvconns = xmalloc(svec->n * sizeof *pvconns);
497 for (i = 0; i < svec->n; i++) {
498 const char *name = svec->names[i];
499 struct pvconn *pvconn;
502 error = pvconn_open(name, &pvconn);
504 pvconns[n_pvconns++] = pvconn;
506 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
514 *n_pvconnsp = n_pvconns;
520 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
522 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
526 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
528 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
532 ofproto_set_netflow(struct ofproto *ofproto,
533 const struct netflow_options *nf_options)
535 if (nf_options && nf_options->collectors.n) {
536 if (!ofproto->netflow) {
537 ofproto->netflow = netflow_create();
539 return netflow_set_options(ofproto->netflow, nf_options);
541 netflow_destroy(ofproto->netflow);
542 ofproto->netflow = NULL;
548 ofproto_set_sflow(struct ofproto *ofproto,
549 const struct ofproto_sflow_options *oso)
551 struct ofproto_sflow *os = ofproto->sflow;
554 struct ofport *ofport;
555 unsigned int odp_port;
557 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
558 refresh_port_groups(ofproto);
559 PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) {
560 ofproto_sflow_add_port(os, odp_port,
561 netdev_get_name(ofport->netdev));
564 ofproto_sflow_set_options(os, oso);
566 ofproto_sflow_destroy(os);
567 ofproto->sflow = NULL;
572 ofproto_set_failure(struct ofproto *ofproto, bool fail_open)
575 struct rconn *rconn = ofproto->controller->rconn;
576 int trigger_duration = rconn_get_probe_interval(rconn) * 3;
577 if (!ofproto->fail_open) {
578 ofproto->fail_open = fail_open_create(ofproto, trigger_duration,
579 ofproto->switch_status,
582 fail_open_set_trigger_duration(ofproto->fail_open,
586 fail_open_destroy(ofproto->fail_open);
587 ofproto->fail_open = NULL;
592 ofproto_set_rate_limit(struct ofproto *ofproto,
593 int rate_limit, int burst_limit)
595 if (rate_limit > 0) {
596 if (!ofproto->miss_sched) {
597 ofproto->miss_sched = pinsched_create(rate_limit, burst_limit,
598 ofproto->switch_status);
599 ofproto->action_sched = pinsched_create(rate_limit, burst_limit,
602 pinsched_set_limits(ofproto->miss_sched, rate_limit, burst_limit);
603 pinsched_set_limits(ofproto->action_sched,
604 rate_limit, burst_limit);
607 pinsched_destroy(ofproto->miss_sched);
608 ofproto->miss_sched = NULL;
609 pinsched_destroy(ofproto->action_sched);
610 ofproto->action_sched = NULL;
615 ofproto_set_stp(struct ofproto *ofproto OVS_UNUSED, bool enable_stp)
619 VLOG_WARN("STP is not yet implemented");
627 ofproto_get_datapath_id(const struct ofproto *ofproto)
629 return ofproto->datapath_id;
633 ofproto_get_probe_interval(const struct ofproto *ofproto)
635 return rconn_get_probe_interval(ofproto->controller->rconn);
639 ofproto_get_max_backoff(const struct ofproto *ofproto)
641 return rconn_get_max_backoff(ofproto->controller->rconn);
645 ofproto_get_in_band(const struct ofproto *ofproto)
647 return ofproto->in_band != NULL;
651 ofproto_get_discovery(const struct ofproto *ofproto)
653 return ofproto->discovery != NULL;
657 ofproto_get_controller(const struct ofproto *ofproto)
659 return rconn_get_name(ofproto->controller->rconn);
663 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
667 for (i = 0; i < ofproto->n_listeners; i++) {
668 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
673 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
677 for (i = 0; i < ofproto->n_snoops; i++) {
678 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
683 ofproto_destroy(struct ofproto *p)
685 struct ofconn *ofconn, *next_ofconn;
686 struct ofport *ofport;
687 unsigned int port_no;
694 /* Destroy fail-open early, because it touches the classifier. */
695 ofproto_set_failure(p, false);
697 ofproto_flush_flows(p);
698 classifier_destroy(&p->cls);
700 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
702 ofconn_destroy(ofconn);
706 netdev_monitor_destroy(p->netdev_monitor);
707 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
710 shash_destroy(&p->port_by_name);
712 switch_status_destroy(p->switch_status);
713 in_band_destroy(p->in_band);
714 discovery_destroy(p->discovery);
715 pinsched_destroy(p->miss_sched);
716 pinsched_destroy(p->action_sched);
717 netflow_destroy(p->netflow);
718 ofproto_sflow_destroy(p->sflow);
720 switch_status_unregister(p->ss_cat);
722 for (i = 0; i < p->n_listeners; i++) {
723 pvconn_close(p->listeners[i]);
727 for (i = 0; i < p->n_snoops; i++) {
728 pvconn_close(p->snoops[i]);
732 mac_learning_destroy(p->ml);
734 free(p->manufacturer);
744 ofproto_run(struct ofproto *p)
746 int error = ofproto_run1(p);
748 error = ofproto_run2(p, false);
754 process_port_change(struct ofproto *ofproto, int error, char *devname)
756 if (error == ENOBUFS) {
757 reinit_ports(ofproto);
759 update_port(ofproto, devname);
765 ofproto_run1(struct ofproto *p)
767 struct ofconn *ofconn, *next_ofconn;
772 if (shash_is_empty(&p->port_by_name)) {
776 for (i = 0; i < 50; i++) {
780 error = dpif_recv(p->dpif, &buf);
782 if (error == ENODEV) {
783 /* Someone destroyed the datapath behind our back. The caller
784 * better destroy us and give up, because we're just going to
785 * spin from here on out. */
786 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
787 VLOG_ERR_RL(&rl, "%s: datapath was destroyed externally",
794 handle_odp_msg(p, buf);
797 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
798 process_port_change(p, error, devname);
800 while ((error = netdev_monitor_poll(p->netdev_monitor,
801 &devname)) != EAGAIN) {
802 process_port_change(p, error, devname);
806 in_band_run(p->in_band);
809 char *controller_name;
810 if (rconn_is_connectivity_questionable(p->controller->rconn)) {
811 discovery_question_connectivity(p->discovery);
813 if (discovery_run(p->discovery, &controller_name)) {
814 if (controller_name) {
815 rconn_connect(p->controller->rconn, controller_name);
817 rconn_disconnect(p->controller->rconn);
821 pinsched_run(p->miss_sched, send_packet_in_miss, p);
822 pinsched_run(p->action_sched, send_packet_in_action, p);
824 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
826 ofconn_run(ofconn, p);
829 /* Fail-open maintenance. Do this after processing the ofconns since
830 * fail-open checks the status of the controller rconn. */
832 fail_open_run(p->fail_open);
835 for (i = 0; i < p->n_listeners; i++) {
839 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
841 ofconn_create(p, rconn_new_from_vconn("passive", vconn));
842 } else if (retval != EAGAIN) {
843 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
847 for (i = 0; i < p->n_snoops; i++) {
851 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
853 rconn_add_monitor(p->controller->rconn, vconn);
854 } else if (retval != EAGAIN) {
855 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
859 if (time_msec() >= p->next_expiration) {
860 COVERAGE_INC(ofproto_expiration);
861 p->next_expiration = time_msec() + 1000;
864 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
866 /* Let the hook know that we're at a stable point: all outstanding data
867 * in existing flows has been accounted to the account_cb. Thus, the
868 * hook can now reasonably do operations that depend on having accurate
869 * flow volume accounting (currently, that's just bond rebalancing). */
870 if (p->ofhooks->account_checkpoint_cb) {
871 p->ofhooks->account_checkpoint_cb(p->aux);
876 netflow_run(p->netflow);
879 ofproto_sflow_run(p->sflow);
885 struct revalidate_cbdata {
886 struct ofproto *ofproto;
887 bool revalidate_all; /* Revalidate all exact-match rules? */
888 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
889 struct tag_set revalidate_set; /* Set of tags to revalidate. */
893 ofproto_run2(struct ofproto *p, bool revalidate_all)
895 if (p->need_revalidate || revalidate_all
896 || !tag_set_is_empty(&p->revalidate_set)) {
897 struct revalidate_cbdata cbdata;
899 cbdata.revalidate_all = revalidate_all;
900 cbdata.revalidate_subrules = p->need_revalidate;
901 cbdata.revalidate_set = p->revalidate_set;
902 tag_set_init(&p->revalidate_set);
903 COVERAGE_INC(ofproto_revalidate);
904 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
905 p->need_revalidate = false;
912 ofproto_wait(struct ofproto *p)
914 struct ofconn *ofconn;
917 dpif_recv_wait(p->dpif);
918 dpif_port_poll_wait(p->dpif);
919 netdev_monitor_poll_wait(p->netdev_monitor);
920 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
924 in_band_wait(p->in_band);
927 discovery_wait(p->discovery);
930 fail_open_wait(p->fail_open);
932 pinsched_wait(p->miss_sched);
933 pinsched_wait(p->action_sched);
935 ofproto_sflow_wait(p->sflow);
937 if (!tag_set_is_empty(&p->revalidate_set)) {
938 poll_immediate_wake();
940 if (p->need_revalidate) {
941 /* Shouldn't happen, but if it does just go around again. */
942 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
943 poll_immediate_wake();
944 } else if (p->next_expiration != LLONG_MAX) {
945 poll_timer_wait(p->next_expiration - time_msec());
947 for (i = 0; i < p->n_listeners; i++) {
948 pvconn_wait(p->listeners[i]);
950 for (i = 0; i < p->n_snoops; i++) {
951 pvconn_wait(p->snoops[i]);
956 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
958 tag_set_add(&ofproto->revalidate_set, tag);
962 ofproto_get_revalidate_set(struct ofproto *ofproto)
964 return &ofproto->revalidate_set;
968 ofproto_is_alive(const struct ofproto *p)
970 return p->discovery || rconn_is_alive(p->controller->rconn);
974 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
975 const union ofp_action *actions, size_t n_actions,
976 const struct ofpbuf *packet)
978 struct odp_actions odp_actions;
981 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
987 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
989 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
990 odp_actions.n_actions, packet);
995 ofproto_add_flow(struct ofproto *p,
996 const flow_t *flow, uint32_t wildcards, unsigned int priority,
997 const union ofp_action *actions, size_t n_actions,
1001 rule = rule_create(p, NULL, actions, n_actions,
1002 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1004 cls_rule_from_flow(&rule->cr, flow, wildcards, priority);
1005 rule_insert(p, rule, NULL, 0);
1009 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1010 uint32_t wildcards, unsigned int priority)
1014 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1018 rule_remove(ofproto, rule);
1023 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1025 struct rule *rule = rule_from_cls_rule(rule_);
1026 struct ofproto *ofproto = ofproto_;
1028 /* Mark the flow as not installed, even though it might really be
1029 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1030 * There is no point in uninstalling it individually since we are about to
1031 * blow away all the flows with dpif_flow_flush(). */
1032 rule->installed = false;
1034 rule_remove(ofproto, rule);
1038 ofproto_flush_flows(struct ofproto *ofproto)
1040 COVERAGE_INC(ofproto_flush);
1041 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1042 dpif_flow_flush(ofproto->dpif);
1043 if (ofproto->in_band) {
1044 in_band_flushed(ofproto->in_band);
1046 if (ofproto->fail_open) {
1047 fail_open_flushed(ofproto->fail_open);
1052 reinit_ports(struct ofproto *p)
1054 struct svec devnames;
1055 struct ofport *ofport;
1056 unsigned int port_no;
1057 struct odp_port *odp_ports;
1061 svec_init(&devnames);
1062 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1063 svec_add (&devnames, (char *) ofport->opp.name);
1065 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1066 for (i = 0; i < n_odp_ports; i++) {
1067 svec_add (&devnames, odp_ports[i].devname);
1071 svec_sort_unique(&devnames);
1072 for (i = 0; i < devnames.n; i++) {
1073 update_port(p, devnames.names[i]);
1075 svec_destroy(&devnames);
1079 refresh_port_group(struct ofproto *p, unsigned int group)
1083 struct ofport *port;
1084 unsigned int port_no;
1086 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1088 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1090 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1091 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1092 ports[n_ports++] = port_no;
1095 dpif_port_group_set(p->dpif, group, ports, n_ports);
1102 refresh_port_groups(struct ofproto *p)
1104 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1105 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1107 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1111 static struct ofport *
1112 make_ofport(const struct odp_port *odp_port)
1114 struct netdev_options netdev_options;
1115 enum netdev_flags flags;
1116 struct ofport *ofport;
1117 struct netdev *netdev;
1121 memset(&netdev_options, 0, sizeof netdev_options);
1122 netdev_options.name = odp_port->devname;
1123 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1124 netdev_options.may_open = true;
1126 error = netdev_open(&netdev_options, &netdev);
1128 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1129 "cannot be opened (%s)",
1130 odp_port->devname, odp_port->port,
1131 odp_port->devname, strerror(error));
1135 ofport = xmalloc(sizeof *ofport);
1136 ofport->netdev = netdev;
1137 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1138 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1139 memcpy(ofport->opp.name, odp_port->devname,
1140 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1141 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1143 netdev_get_flags(netdev, &flags);
1144 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1146 netdev_get_carrier(netdev, &carrier);
1147 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1149 netdev_get_features(netdev,
1150 &ofport->opp.curr, &ofport->opp.advertised,
1151 &ofport->opp.supported, &ofport->opp.peer);
1156 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1158 if (port_array_get(&p->ports, odp_port->port)) {
1159 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1162 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1163 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1172 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1174 const struct ofp_phy_port *a = &a_->opp;
1175 const struct ofp_phy_port *b = &b_->opp;
1177 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1178 return (a->port_no == b->port_no
1179 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1180 && !strcmp((char *) a->name, (char *) b->name)
1181 && a->state == b->state
1182 && a->config == b->config
1183 && a->curr == b->curr
1184 && a->advertised == b->advertised
1185 && a->supported == b->supported
1186 && a->peer == b->peer);
1190 send_port_status(struct ofproto *p, const struct ofport *ofport,
1193 /* XXX Should limit the number of queued port status change messages. */
1194 struct ofconn *ofconn;
1195 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1196 struct ofp_port_status *ops;
1199 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1200 ops->reason = reason;
1201 ops->desc = ofport->opp;
1202 hton_ofp_phy_port(&ops->desc);
1203 queue_tx(b, ofconn, NULL);
1205 if (p->ofhooks->port_changed_cb) {
1206 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1211 ofport_install(struct ofproto *p, struct ofport *ofport)
1213 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1214 const char *netdev_name = (const char *) ofport->opp.name;
1216 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1217 port_array_set(&p->ports, odp_port, ofport);
1218 shash_add(&p->port_by_name, netdev_name, ofport);
1220 ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
1225 ofport_remove(struct ofproto *p, struct ofport *ofport)
1227 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1229 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1230 port_array_set(&p->ports, odp_port, NULL);
1231 shash_delete(&p->port_by_name,
1232 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1234 ofproto_sflow_del_port(p->sflow, odp_port);
1239 ofport_free(struct ofport *ofport)
1242 netdev_close(ofport->netdev);
1248 update_port(struct ofproto *p, const char *devname)
1250 struct odp_port odp_port;
1251 struct ofport *old_ofport;
1252 struct ofport *new_ofport;
1255 COVERAGE_INC(ofproto_update_port);
1257 /* Query the datapath for port information. */
1258 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1260 /* Find the old ofport. */
1261 old_ofport = shash_find_data(&p->port_by_name, devname);
1264 /* There's no port named 'devname' but there might be a port with
1265 * the same port number. This could happen if a port is deleted
1266 * and then a new one added in its place very quickly, or if a port
1267 * is renamed. In the former case we want to send an OFPPR_DELETE
1268 * and an OFPPR_ADD, and in the latter case we want to send a
1269 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1270 * the old port's ifindex against the new port, or perhaps less
1271 * reliably but more portably by comparing the old port's MAC
1272 * against the new port's MAC. However, this code isn't that smart
1273 * and always sends an OFPPR_MODIFY (XXX). */
1274 old_ofport = port_array_get(&p->ports, odp_port.port);
1276 } else if (error != ENOENT && error != ENODEV) {
1277 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1278 "%s", strerror(error));
1282 /* Create a new ofport. */
1283 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1285 /* Eliminate a few pathological cases. */
1286 if (!old_ofport && !new_ofport) {
1288 } else if (old_ofport && new_ofport) {
1289 /* Most of the 'config' bits are OpenFlow soft state, but
1290 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1291 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1292 * leaves the other bits 0.) */
1293 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1295 if (ofport_equal(old_ofport, new_ofport)) {
1296 /* False alarm--no change. */
1297 ofport_free(new_ofport);
1302 /* Now deal with the normal cases. */
1304 ofport_remove(p, old_ofport);
1307 ofport_install(p, new_ofport);
1309 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1310 (!old_ofport ? OFPPR_ADD
1311 : !new_ofport ? OFPPR_DELETE
1313 ofport_free(old_ofport);
1315 /* Update port groups. */
1316 refresh_port_groups(p);
1320 init_ports(struct ofproto *p)
1322 struct odp_port *ports;
1327 error = dpif_port_list(p->dpif, &ports, &n_ports);
1332 for (i = 0; i < n_ports; i++) {
1333 const struct odp_port *odp_port = &ports[i];
1334 if (!ofport_conflicts(p, odp_port)) {
1335 struct ofport *ofport = make_ofport(odp_port);
1337 ofport_install(p, ofport);
1342 refresh_port_groups(p);
1346 static struct ofconn *
1347 ofconn_create(struct ofproto *p, struct rconn *rconn)
1349 struct ofconn *ofconn = xmalloc(sizeof *ofconn);
1350 list_push_back(&p->all_conns, &ofconn->node);
1351 ofconn->rconn = rconn;
1352 ofconn->pktbuf = NULL;
1353 ofconn->miss_send_len = 0;
1354 ofconn->packet_in_counter = rconn_packet_counter_create ();
1355 ofconn->reply_counter = rconn_packet_counter_create ();
1360 ofconn_destroy(struct ofconn *ofconn)
1362 list_remove(&ofconn->node);
1363 rconn_destroy(ofconn->rconn);
1364 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1365 rconn_packet_counter_destroy(ofconn->reply_counter);
1366 pktbuf_destroy(ofconn->pktbuf);
1371 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1375 rconn_run(ofconn->rconn);
1377 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1378 /* Limit the number of iterations to prevent other tasks from
1380 for (iteration = 0; iteration < 50; iteration++) {
1381 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1386 fail_open_maybe_recover(p->fail_open);
1388 handle_openflow(ofconn, p, of_msg);
1389 ofpbuf_delete(of_msg);
1393 if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
1394 ofconn_destroy(ofconn);
1399 ofconn_wait(struct ofconn *ofconn)
1401 rconn_run_wait(ofconn->rconn);
1402 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1403 rconn_recv_wait(ofconn->rconn);
1405 COVERAGE_INC(ofproto_ofconn_stuck);
1409 /* Caller is responsible for initializing the 'cr' member of the returned
1411 static struct rule *
1412 rule_create(struct ofproto *ofproto, struct rule *super,
1413 const union ofp_action *actions, size_t n_actions,
1414 uint16_t idle_timeout, uint16_t hard_timeout,
1415 uint64_t flow_cookie, bool send_flow_removed)
1417 struct rule *rule = xzalloc(sizeof *rule);
1418 rule->idle_timeout = idle_timeout;
1419 rule->hard_timeout = hard_timeout;
1420 rule->flow_cookie = flow_cookie;
1421 rule->used = rule->created = time_msec();
1422 rule->send_flow_removed = send_flow_removed;
1423 rule->super = super;
1425 list_push_back(&super->list, &rule->list);
1427 list_init(&rule->list);
1429 rule->n_actions = n_actions;
1430 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1431 netflow_flow_clear(&rule->nf_flow);
1432 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1437 static struct rule *
1438 rule_from_cls_rule(const struct cls_rule *cls_rule)
1440 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1444 rule_free(struct rule *rule)
1446 free(rule->actions);
1447 free(rule->odp_actions);
1451 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1452 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1453 * through all of its subrules and revalidates them, destroying any that no
1454 * longer has a super-rule (which is probably all of them).
1456 * Before calling this function, the caller must make have removed 'rule' from
1457 * the classifier. If 'rule' is an exact-match rule, the caller is also
1458 * responsible for ensuring that it has been uninstalled from the datapath. */
1460 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1463 struct rule *subrule, *next;
1464 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1465 revalidate_rule(ofproto, subrule);
1468 list_remove(&rule->list);
1474 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1476 const union ofp_action *oa;
1477 struct actions_iterator i;
1479 if (out_port == htons(OFPP_NONE)) {
1482 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1483 oa = actions_next(&i)) {
1484 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1491 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1492 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1494 * The flow that 'packet' actually contains does not need to actually match
1495 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1496 * the packet and byte counters for 'rule' will be credited for the packet sent
1497 * out whether or not the packet actually matches 'rule'.
1499 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1500 * the caller must already have accurately composed ODP actions for it given
1501 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1502 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1503 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1504 * actions and apply them to 'packet'. */
1506 rule_execute(struct ofproto *ofproto, struct rule *rule,
1507 struct ofpbuf *packet, const flow_t *flow)
1509 const union odp_action *actions;
1511 struct odp_actions a;
1513 /* Grab or compose the ODP actions.
1515 * The special case for an exact-match 'rule' where 'flow' is not the
1516 * rule's flow is important to avoid, e.g., sending a packet out its input
1517 * port simply because the ODP actions were composed for the wrong
1519 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1520 struct rule *super = rule->super ? rule->super : rule;
1521 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1522 packet, &a, NULL, 0, NULL)) {
1525 actions = a.actions;
1526 n_actions = a.n_actions;
1528 actions = rule->odp_actions;
1529 n_actions = rule->n_odp_actions;
1532 /* Execute the ODP actions. */
1533 if (!dpif_execute(ofproto->dpif, flow->in_port,
1534 actions, n_actions, packet)) {
1535 struct odp_flow_stats stats;
1536 flow_extract_stats(flow, packet, &stats);
1537 update_stats(ofproto, rule, &stats);
1538 rule->used = time_msec();
1539 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1544 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1547 struct rule *displaced_rule;
1549 /* Insert the rule in the classifier. */
1550 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1551 if (!rule->cr.wc.wildcards) {
1552 rule_make_actions(p, rule, packet);
1555 /* Send the packet and credit it to the rule. */
1558 flow_extract(packet, in_port, &flow);
1559 rule_execute(p, rule, packet, &flow);
1562 /* Install the rule in the datapath only after sending the packet, to
1563 * avoid packet reordering. */
1564 if (rule->cr.wc.wildcards) {
1565 COVERAGE_INC(ofproto_add_wc_flow);
1566 p->need_revalidate = true;
1568 rule_install(p, rule, displaced_rule);
1571 /* Free the rule that was displaced, if any. */
1572 if (displaced_rule) {
1573 rule_destroy(p, displaced_rule);
1577 static struct rule *
1578 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1581 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1582 rule->idle_timeout, rule->hard_timeout,
1584 COVERAGE_INC(ofproto_subrule_create);
1585 cls_rule_from_flow(&subrule->cr, flow, 0,
1586 (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1587 : rule->cr.priority));
1588 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1594 rule_remove(struct ofproto *ofproto, struct rule *rule)
1596 if (rule->cr.wc.wildcards) {
1597 COVERAGE_INC(ofproto_del_wc_flow);
1598 ofproto->need_revalidate = true;
1600 rule_uninstall(ofproto, rule);
1602 classifier_remove(&ofproto->cls, &rule->cr);
1603 rule_destroy(ofproto, rule);
1606 /* Returns true if the actions changed, false otherwise. */
1608 rule_make_actions(struct ofproto *p, struct rule *rule,
1609 const struct ofpbuf *packet)
1611 const struct rule *super;
1612 struct odp_actions a;
1615 assert(!rule->cr.wc.wildcards);
1617 super = rule->super ? rule->super : rule;
1619 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1620 packet, &a, &rule->tags, &rule->may_install,
1621 &rule->nf_flow.output_iface);
1623 actions_len = a.n_actions * sizeof *a.actions;
1624 if (rule->n_odp_actions != a.n_actions
1625 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1626 COVERAGE_INC(ofproto_odp_unchanged);
1627 free(rule->odp_actions);
1628 rule->n_odp_actions = a.n_actions;
1629 rule->odp_actions = xmemdup(a.actions, actions_len);
1637 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1638 struct odp_flow_put *put)
1640 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1641 put->flow.key = rule->cr.flow;
1642 put->flow.actions = rule->odp_actions;
1643 put->flow.n_actions = rule->n_odp_actions;
1645 return dpif_flow_put(ofproto->dpif, put);
1649 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1651 assert(!rule->cr.wc.wildcards);
1653 if (rule->may_install) {
1654 struct odp_flow_put put;
1655 if (!do_put_flow(p, rule,
1656 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1658 rule->installed = true;
1659 if (displaced_rule) {
1660 update_stats(p, displaced_rule, &put.flow.stats);
1661 rule_post_uninstall(p, displaced_rule);
1664 } else if (displaced_rule) {
1665 rule_uninstall(p, displaced_rule);
1670 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1672 if (rule->installed) {
1673 struct odp_flow_put put;
1674 COVERAGE_INC(ofproto_dp_missed);
1675 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1677 rule_install(ofproto, rule, NULL);
1682 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1684 bool actions_changed;
1685 uint16_t new_out_iface, old_out_iface;
1687 old_out_iface = rule->nf_flow.output_iface;
1688 actions_changed = rule_make_actions(ofproto, rule, NULL);
1690 if (rule->may_install) {
1691 if (rule->installed) {
1692 if (actions_changed) {
1693 struct odp_flow_put put;
1694 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
1695 | ODPPF_ZERO_STATS, &put);
1696 update_stats(ofproto, rule, &put.flow.stats);
1698 /* Temporarily set the old output iface so that NetFlow
1699 * messages have the correct output interface for the old
1701 new_out_iface = rule->nf_flow.output_iface;
1702 rule->nf_flow.output_iface = old_out_iface;
1703 rule_post_uninstall(ofproto, rule);
1704 rule->nf_flow.output_iface = new_out_iface;
1707 rule_install(ofproto, rule, NULL);
1710 rule_uninstall(ofproto, rule);
1715 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1717 uint64_t total_bytes = rule->byte_count + extra_bytes;
1719 if (ofproto->ofhooks->account_flow_cb
1720 && total_bytes > rule->accounted_bytes)
1722 ofproto->ofhooks->account_flow_cb(
1723 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1724 total_bytes - rule->accounted_bytes, ofproto->aux);
1725 rule->accounted_bytes = total_bytes;
1730 rule_uninstall(struct ofproto *p, struct rule *rule)
1732 assert(!rule->cr.wc.wildcards);
1733 if (rule->installed) {
1734 struct odp_flow odp_flow;
1736 odp_flow.key = rule->cr.flow;
1737 odp_flow.actions = NULL;
1738 odp_flow.n_actions = 0;
1739 if (!dpif_flow_del(p->dpif, &odp_flow)) {
1740 update_stats(p, rule, &odp_flow.stats);
1742 rule->installed = false;
1744 rule_post_uninstall(p, rule);
1749 is_controller_rule(struct rule *rule)
1751 /* If the only action is send to the controller then don't report
1752 * NetFlow expiration messages since it is just part of the control
1753 * logic for the network and not real traffic. */
1755 if (rule && rule->super) {
1756 struct rule *super = rule->super;
1758 return super->n_actions == 1 &&
1759 super->actions[0].type == htons(OFPAT_OUTPUT) &&
1760 super->actions[0].output.port == htons(OFPP_CONTROLLER);
1767 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1769 struct rule *super = rule->super;
1771 rule_account(ofproto, rule, 0);
1773 if (ofproto->netflow && !is_controller_rule(rule)) {
1774 struct ofexpired expired;
1775 expired.flow = rule->cr.flow;
1776 expired.packet_count = rule->packet_count;
1777 expired.byte_count = rule->byte_count;
1778 expired.used = rule->used;
1779 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
1782 super->packet_count += rule->packet_count;
1783 super->byte_count += rule->byte_count;
1785 /* Reset counters to prevent double counting if the rule ever gets
1787 rule->packet_count = 0;
1788 rule->byte_count = 0;
1789 rule->accounted_bytes = 0;
1791 netflow_flow_clear(&rule->nf_flow);
1796 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1797 struct rconn_packet_counter *counter)
1799 update_openflow_length(msg);
1800 if (rconn_send(ofconn->rconn, msg, counter)) {
1806 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1807 int error, const void *data, size_t len)
1810 struct ofp_error_msg *oem;
1812 if (!(error >> 16)) {
1813 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1818 COVERAGE_INC(ofproto_error);
1819 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1820 oh ? oh->xid : 0, &buf);
1821 oem->type = htons((unsigned int) error >> 16);
1822 oem->code = htons(error & 0xffff);
1823 memcpy(oem->data, data, len);
1824 queue_tx(buf, ofconn, ofconn->reply_counter);
1828 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1831 size_t oh_length = ntohs(oh->length);
1832 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1836 hton_ofp_phy_port(struct ofp_phy_port *opp)
1838 opp->port_no = htons(opp->port_no);
1839 opp->config = htonl(opp->config);
1840 opp->state = htonl(opp->state);
1841 opp->curr = htonl(opp->curr);
1842 opp->advertised = htonl(opp->advertised);
1843 opp->supported = htonl(opp->supported);
1844 opp->peer = htonl(opp->peer);
1848 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
1850 struct ofp_header *rq = oh;
1851 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
1856 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
1857 struct ofp_header *oh)
1859 struct ofp_switch_features *osf;
1861 unsigned int port_no;
1862 struct ofport *port;
1864 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1865 osf->datapath_id = htonll(p->datapath_id);
1866 osf->n_buffers = htonl(pktbuf_capacity());
1868 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1869 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
1870 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1871 (1u << OFPAT_SET_VLAN_VID) |
1872 (1u << OFPAT_SET_VLAN_PCP) |
1873 (1u << OFPAT_STRIP_VLAN) |
1874 (1u << OFPAT_SET_DL_SRC) |
1875 (1u << OFPAT_SET_DL_DST) |
1876 (1u << OFPAT_SET_NW_SRC) |
1877 (1u << OFPAT_SET_NW_DST) |
1878 (1u << OFPAT_SET_NW_TOS) |
1879 (1u << OFPAT_SET_TP_SRC) |
1880 (1u << OFPAT_SET_TP_DST));
1882 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1883 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1886 queue_tx(buf, ofconn, ofconn->reply_counter);
1891 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1892 struct ofp_header *oh)
1895 struct ofp_switch_config *osc;
1899 /* Figure out flags. */
1900 dpif_get_drop_frags(p->dpif, &drop_frags);
1901 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1904 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1905 osc->flags = htons(flags);
1906 osc->miss_send_len = htons(ofconn->miss_send_len);
1907 queue_tx(buf, ofconn, ofconn->reply_counter);
1913 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1914 struct ofp_switch_config *osc)
1919 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1923 flags = ntohs(osc->flags);
1925 if (ofconn == p->controller) {
1926 switch (flags & OFPC_FRAG_MASK) {
1927 case OFPC_FRAG_NORMAL:
1928 dpif_set_drop_frags(p->dpif, false);
1930 case OFPC_FRAG_DROP:
1931 dpif_set_drop_frags(p->dpif, true);
1934 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1940 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1941 if (ntohs(osc->miss_send_len) != 0) {
1942 ofconn->pktbuf = pktbuf_create();
1944 pktbuf_destroy(ofconn->pktbuf);
1948 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1954 add_output_group_action(struct odp_actions *actions, uint16_t group,
1955 uint16_t *nf_output_iface)
1957 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1959 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
1960 *nf_output_iface = NF_OUT_FLOOD;
1965 add_controller_action(struct odp_actions *actions,
1966 const struct ofp_action_output *oao)
1968 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
1969 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
1972 struct action_xlate_ctx {
1974 const flow_t *flow; /* Flow to which these actions correspond. */
1975 int recurse; /* Recursion level, via xlate_table_action. */
1976 struct ofproto *ofproto;
1977 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
1978 * null pointer if we are revalidating
1979 * without a packet to refer to. */
1982 struct odp_actions *out; /* Datapath actions. */
1983 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
1984 bool may_set_up_flow; /* True ordinarily; false if the actions must
1985 * be reassessed for every packet. */
1986 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
1989 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
1990 struct action_xlate_ctx *ctx);
1993 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
1995 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
1998 if (ofport->opp.config & OFPPC_NO_FWD) {
1999 /* Forwarding disabled on port. */
2004 * We don't have an ofport record for this port, but it doesn't hurt to
2005 * allow forwarding to it anyhow. Maybe such a port will appear later
2006 * and we're pre-populating the flow table.
2010 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2011 ctx->nf_output_iface = port;
2014 static struct rule *
2015 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2018 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2020 /* The rule we found might not be valid, since we could be in need of
2021 * revalidation. If it is not valid, don't return it. */
2024 && ofproto->need_revalidate
2025 && !revalidate_rule(ofproto, rule)) {
2026 COVERAGE_INC(ofproto_invalidated);
2034 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2036 if (!ctx->recurse) {
2041 flow.in_port = in_port;
2043 rule = lookup_valid_rule(ctx->ofproto, &flow);
2050 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2057 xlate_output_action(struct action_xlate_ctx *ctx,
2058 const struct ofp_action_output *oao)
2061 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2063 ctx->nf_output_iface = NF_OUT_DROP;
2065 switch (ntohs(oao->port)) {
2067 add_output_action(ctx, ctx->flow->in_port);
2070 xlate_table_action(ctx, ctx->flow->in_port);
2073 if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet,
2074 ctx->out, ctx->tags,
2075 &ctx->nf_output_iface,
2076 ctx->ofproto->aux)) {
2077 COVERAGE_INC(ofproto_uninstallable);
2078 ctx->may_set_up_flow = false;
2082 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2083 &ctx->nf_output_iface);
2086 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2088 case OFPP_CONTROLLER:
2089 add_controller_action(ctx->out, oao);
2092 add_output_action(ctx, ODPP_LOCAL);
2095 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2096 if (odp_port != ctx->flow->in_port) {
2097 add_output_action(ctx, odp_port);
2102 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2103 ctx->nf_output_iface = NF_OUT_FLOOD;
2104 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2105 ctx->nf_output_iface = prev_nf_output_iface;
2106 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2107 ctx->nf_output_iface != NF_OUT_FLOOD) {
2108 ctx->nf_output_iface = NF_OUT_MULTI;
2113 xlate_nicira_action(struct action_xlate_ctx *ctx,
2114 const struct nx_action_header *nah)
2116 const struct nx_action_resubmit *nar;
2117 int subtype = ntohs(nah->subtype);
2119 assert(nah->vendor == htonl(NX_VENDOR_ID));
2121 case NXAST_RESUBMIT:
2122 nar = (const struct nx_action_resubmit *) nah;
2123 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2127 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2133 do_xlate_actions(const union ofp_action *in, size_t n_in,
2134 struct action_xlate_ctx *ctx)
2136 struct actions_iterator iter;
2137 const union ofp_action *ia;
2138 const struct ofport *port;
2140 port = port_array_get(&ctx->ofproto->ports, ctx->flow->in_port);
2141 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2142 port->opp.config & (eth_addr_equals(ctx->flow->dl_dst, stp_eth_addr)
2143 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2144 /* Drop this flow. */
2148 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2149 uint16_t type = ntohs(ia->type);
2150 union odp_action *oa;
2154 xlate_output_action(ctx, &ia->output);
2157 case OFPAT_SET_VLAN_VID:
2158 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2159 oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2162 case OFPAT_SET_VLAN_PCP:
2163 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2164 oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2167 case OFPAT_STRIP_VLAN:
2168 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2171 case OFPAT_SET_DL_SRC:
2172 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2173 memcpy(oa->dl_addr.dl_addr,
2174 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2177 case OFPAT_SET_DL_DST:
2178 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2179 memcpy(oa->dl_addr.dl_addr,
2180 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2183 case OFPAT_SET_NW_SRC:
2184 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2185 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2188 case OFPAT_SET_NW_DST:
2189 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2190 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2192 case OFPAT_SET_NW_TOS:
2193 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2194 oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2197 case OFPAT_SET_TP_SRC:
2198 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2199 oa->tp_port.tp_port = ia->tp_port.tp_port;
2202 case OFPAT_SET_TP_DST:
2203 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2204 oa->tp_port.tp_port = ia->tp_port.tp_port;
2208 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2212 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2219 xlate_actions(const union ofp_action *in, size_t n_in,
2220 const flow_t *flow, struct ofproto *ofproto,
2221 const struct ofpbuf *packet,
2222 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2223 uint16_t *nf_output_iface)
2225 tag_type no_tags = 0;
2226 struct action_xlate_ctx ctx;
2227 COVERAGE_INC(ofproto_ofp2odp);
2228 odp_actions_init(out);
2231 ctx.ofproto = ofproto;
2232 ctx.packet = packet;
2234 ctx.tags = tags ? tags : &no_tags;
2235 ctx.may_set_up_flow = true;
2236 ctx.nf_output_iface = NF_OUT_DROP;
2237 do_xlate_actions(in, n_in, &ctx);
2239 /* Check with in-band control to see if we're allowed to set up this
2241 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2242 ctx.may_set_up_flow = false;
2245 if (may_set_up_flow) {
2246 *may_set_up_flow = ctx.may_set_up_flow;
2248 if (nf_output_iface) {
2249 *nf_output_iface = ctx.nf_output_iface;
2251 if (odp_actions_overflow(out)) {
2252 odp_actions_init(out);
2253 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2259 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2260 struct ofp_header *oh)
2262 struct ofp_packet_out *opo;
2263 struct ofpbuf payload, *buffer;
2264 struct odp_actions actions;
2270 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2274 opo = (struct ofp_packet_out *) oh;
2276 COVERAGE_INC(ofproto_packet_out);
2277 if (opo->buffer_id != htonl(UINT32_MAX)) {
2278 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2280 if (error || !buffer) {
2288 flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2289 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2290 &flow, p, &payload, &actions, NULL, NULL, NULL);
2295 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2297 ofpbuf_delete(buffer);
2303 update_port_config(struct ofproto *p, struct ofport *port,
2304 uint32_t config, uint32_t mask)
2306 mask &= config ^ port->opp.config;
2307 if (mask & OFPPC_PORT_DOWN) {
2308 if (config & OFPPC_PORT_DOWN) {
2309 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2311 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2314 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2315 if (mask & REVALIDATE_BITS) {
2316 COVERAGE_INC(ofproto_costly_flags);
2317 port->opp.config ^= mask & REVALIDATE_BITS;
2318 p->need_revalidate = true;
2320 #undef REVALIDATE_BITS
2321 if (mask & OFPPC_NO_FLOOD) {
2322 port->opp.config ^= OFPPC_NO_FLOOD;
2323 refresh_port_groups(p);
2325 if (mask & OFPPC_NO_PACKET_IN) {
2326 port->opp.config ^= OFPPC_NO_PACKET_IN;
2331 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2333 const struct ofp_port_mod *opm;
2334 struct ofport *port;
2337 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2341 opm = (struct ofp_port_mod *) oh;
2343 port = port_array_get(&p->ports,
2344 ofp_port_to_odp_port(ntohs(opm->port_no)));
2346 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2347 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2348 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2350 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2351 if (opm->advertise) {
2352 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2358 static struct ofpbuf *
2359 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2361 struct ofp_stats_reply *osr;
2364 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2365 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2367 osr->flags = htons(0);
2371 static struct ofpbuf *
2372 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2374 return make_stats_reply(request->header.xid, request->type, body_len);
2378 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2380 struct ofpbuf *msg = *msgp;
2381 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2382 if (nbytes + msg->size > UINT16_MAX) {
2383 struct ofp_stats_reply *reply = msg->data;
2384 reply->flags = htons(OFPSF_REPLY_MORE);
2385 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2386 queue_tx(msg, ofconn, ofconn->reply_counter);
2388 return ofpbuf_put_uninit(*msgp, nbytes);
2392 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2393 struct ofp_stats_request *request)
2395 struct ofp_desc_stats *ods;
2398 msg = start_stats_reply(request, sizeof *ods);
2399 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2400 strncpy(ods->mfr_desc, p->manufacturer, sizeof ods->mfr_desc);
2401 strncpy(ods->hw_desc, p->hardware, sizeof ods->hw_desc);
2402 strncpy(ods->sw_desc, p->software, sizeof ods->sw_desc);
2403 strncpy(ods->serial_num, p->serial, sizeof ods->serial_num);
2404 strncpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2405 queue_tx(msg, ofconn, ofconn->reply_counter);
2411 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2413 struct rule *rule = rule_from_cls_rule(cls_rule);
2414 int *n_subrules = n_subrules_;
2422 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2423 struct ofp_stats_request *request)
2425 struct ofp_table_stats *ots;
2427 struct odp_stats dpstats;
2428 int n_exact, n_subrules, n_wild;
2430 msg = start_stats_reply(request, sizeof *ots * 2);
2432 /* Count rules of various kinds. */
2434 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2435 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2436 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2439 dpif_get_dp_stats(p->dpif, &dpstats);
2440 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2441 memset(ots, 0, sizeof *ots);
2442 ots->table_id = TABLEID_HASH;
2443 strcpy(ots->name, "hash");
2444 ots->wildcards = htonl(0);
2445 ots->max_entries = htonl(dpstats.max_capacity);
2446 ots->active_count = htonl(n_exact);
2447 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2449 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2451 /* Classifier table. */
2452 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2453 memset(ots, 0, sizeof *ots);
2454 ots->table_id = TABLEID_CLASSIFIER;
2455 strcpy(ots->name, "classifier");
2456 ots->wildcards = htonl(OFPFW_ALL);
2457 ots->max_entries = htonl(65536);
2458 ots->active_count = htonl(n_wild);
2459 ots->lookup_count = htonll(0); /* XXX */
2460 ots->matched_count = htonll(0); /* XXX */
2462 queue_tx(msg, ofconn, ofconn->reply_counter);
2467 append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
2470 struct netdev_stats stats;
2471 struct ofp_port_stats *ops;
2473 /* Intentionally ignore return value, since errors will set
2474 * 'stats' to all-1s, which is correct for OpenFlow, and
2475 * netdev_get_stats() will log errors. */
2476 netdev_get_stats(port->netdev, &stats);
2478 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2479 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2480 memset(ops->pad, 0, sizeof ops->pad);
2481 ops->rx_packets = htonll(stats.rx_packets);
2482 ops->tx_packets = htonll(stats.tx_packets);
2483 ops->rx_bytes = htonll(stats.rx_bytes);
2484 ops->tx_bytes = htonll(stats.tx_bytes);
2485 ops->rx_dropped = htonll(stats.rx_dropped);
2486 ops->tx_dropped = htonll(stats.tx_dropped);
2487 ops->rx_errors = htonll(stats.rx_errors);
2488 ops->tx_errors = htonll(stats.tx_errors);
2489 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2490 ops->rx_over_err = htonll(stats.rx_over_errors);
2491 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2492 ops->collisions = htonll(stats.collisions);
2496 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2497 struct ofp_stats_request *osr,
2500 struct ofp_port_stats_request *psr;
2501 struct ofp_port_stats *ops;
2503 struct ofport *port;
2504 unsigned int port_no;
2506 if (arg_size != sizeof *psr) {
2507 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2509 psr = (struct ofp_port_stats_request *) osr->body;
2511 msg = start_stats_reply(osr, sizeof *ops * 16);
2512 if (psr->port_no != htons(OFPP_NONE)) {
2513 port = port_array_get(&p->ports,
2514 ofp_port_to_odp_port(ntohs(psr->port_no)));
2516 append_port_stat(port, ntohs(psr->port_no), ofconn, msg);
2519 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2520 append_port_stat(port, port_no, ofconn, msg);
2524 queue_tx(msg, ofconn, ofconn->reply_counter);
2528 struct flow_stats_cbdata {
2529 struct ofproto *ofproto;
2530 struct ofconn *ofconn;
2536 query_stats(struct ofproto *p, struct rule *rule,
2537 uint64_t *packet_countp, uint64_t *byte_countp)
2539 uint64_t packet_count, byte_count;
2540 struct rule *subrule;
2541 struct odp_flow *odp_flows;
2544 packet_count = rule->packet_count;
2545 byte_count = rule->byte_count;
2547 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2548 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
2549 if (rule->cr.wc.wildcards) {
2551 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2552 odp_flows[i++].key = subrule->cr.flow;
2553 packet_count += subrule->packet_count;
2554 byte_count += subrule->byte_count;
2557 odp_flows[0].key = rule->cr.flow;
2560 packet_count = rule->packet_count;
2561 byte_count = rule->byte_count;
2562 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2564 for (i = 0; i < n_odp_flows; i++) {
2565 struct odp_flow *odp_flow = &odp_flows[i];
2566 packet_count += odp_flow->stats.n_packets;
2567 byte_count += odp_flow->stats.n_bytes;
2572 *packet_countp = packet_count;
2573 *byte_countp = byte_count;
2577 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2579 struct rule *rule = rule_from_cls_rule(rule_);
2580 struct flow_stats_cbdata *cbdata = cbdata_;
2581 struct ofp_flow_stats *ofs;
2582 uint64_t packet_count, byte_count;
2583 size_t act_len, len;
2584 long long int tdiff = time_msec() - rule->created;
2585 uint32_t sec = tdiff / 1000;
2586 uint32_t msec = tdiff - (sec * 1000);
2588 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2592 act_len = sizeof *rule->actions * rule->n_actions;
2593 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2595 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2597 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2598 ofs->length = htons(len);
2599 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2601 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match);
2602 ofs->duration_sec = htonl(sec);
2603 ofs->duration_nsec = htonl(msec * 1000000);
2604 ofs->cookie = rule->flow_cookie;
2605 ofs->priority = htons(rule->cr.priority);
2606 ofs->idle_timeout = htons(rule->idle_timeout);
2607 ofs->hard_timeout = htons(rule->hard_timeout);
2608 memset(ofs->pad2, 0, sizeof ofs->pad2);
2609 ofs->packet_count = htonll(packet_count);
2610 ofs->byte_count = htonll(byte_count);
2611 memcpy(ofs->actions, rule->actions, act_len);
2615 table_id_to_include(uint8_t table_id)
2617 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2618 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2619 : table_id == 0xff ? CLS_INC_ALL
2624 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2625 const struct ofp_stats_request *osr,
2628 struct ofp_flow_stats_request *fsr;
2629 struct flow_stats_cbdata cbdata;
2630 struct cls_rule target;
2632 if (arg_size != sizeof *fsr) {
2633 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2635 fsr = (struct ofp_flow_stats_request *) osr->body;
2637 COVERAGE_INC(ofproto_flows_req);
2639 cbdata.ofconn = ofconn;
2640 cbdata.out_port = fsr->out_port;
2641 cbdata.msg = start_stats_reply(osr, 1024);
2642 cls_rule_from_match(&target, &fsr->match, 0);
2643 classifier_for_each_match(&p->cls, &target,
2644 table_id_to_include(fsr->table_id),
2645 flow_stats_cb, &cbdata);
2646 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2650 struct flow_stats_ds_cbdata {
2651 struct ofproto *ofproto;
2656 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
2658 struct rule *rule = rule_from_cls_rule(rule_);
2659 struct flow_stats_ds_cbdata *cbdata = cbdata_;
2660 struct ds *results = cbdata->results;
2661 struct ofp_match match;
2662 uint64_t packet_count, byte_count;
2663 size_t act_len = sizeof *rule->actions * rule->n_actions;
2665 /* Don't report on subrules. */
2666 if (rule->super != NULL) {
2670 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2671 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &match);
2673 ds_put_format(results, "duration=%llds, ",
2674 (time_msec() - rule->created) / 1000);
2675 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2676 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2677 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2678 ofp_print_match(results, &match, true);
2679 ofp_print_actions(results, &rule->actions->header, act_len);
2680 ds_put_cstr(results, "\n");
2683 /* Adds a pretty-printed description of all flows to 'results', including
2684 * those marked hidden by secchan (e.g., by in-band control). */
2686 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2688 struct ofp_match match;
2689 struct cls_rule target;
2690 struct flow_stats_ds_cbdata cbdata;
2692 memset(&match, 0, sizeof match);
2693 match.wildcards = htonl(OFPFW_ALL);
2696 cbdata.results = results;
2698 cls_rule_from_match(&target, &match, 0);
2699 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2700 flow_stats_ds_cb, &cbdata);
2703 struct aggregate_stats_cbdata {
2704 struct ofproto *ofproto;
2706 uint64_t packet_count;
2707 uint64_t byte_count;
2712 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2714 struct rule *rule = rule_from_cls_rule(rule_);
2715 struct aggregate_stats_cbdata *cbdata = cbdata_;
2716 uint64_t packet_count, byte_count;
2718 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2722 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2724 cbdata->packet_count += packet_count;
2725 cbdata->byte_count += byte_count;
2730 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2731 const struct ofp_stats_request *osr,
2734 struct ofp_aggregate_stats_request *asr;
2735 struct ofp_aggregate_stats_reply *reply;
2736 struct aggregate_stats_cbdata cbdata;
2737 struct cls_rule target;
2740 if (arg_size != sizeof *asr) {
2741 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2743 asr = (struct ofp_aggregate_stats_request *) osr->body;
2745 COVERAGE_INC(ofproto_agg_request);
2747 cbdata.out_port = asr->out_port;
2748 cbdata.packet_count = 0;
2749 cbdata.byte_count = 0;
2751 cls_rule_from_match(&target, &asr->match, 0);
2752 classifier_for_each_match(&p->cls, &target,
2753 table_id_to_include(asr->table_id),
2754 aggregate_stats_cb, &cbdata);
2756 msg = start_stats_reply(osr, sizeof *reply);
2757 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2758 reply->flow_count = htonl(cbdata.n_flows);
2759 reply->packet_count = htonll(cbdata.packet_count);
2760 reply->byte_count = htonll(cbdata.byte_count);
2761 queue_tx(msg, ofconn, ofconn->reply_counter);
2766 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2767 struct ofp_header *oh)
2769 struct ofp_stats_request *osr;
2773 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2778 osr = (struct ofp_stats_request *) oh;
2780 switch (ntohs(osr->type)) {
2782 return handle_desc_stats_request(p, ofconn, osr);
2785 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2787 case OFPST_AGGREGATE:
2788 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2791 return handle_table_stats_request(p, ofconn, osr);
2794 return handle_port_stats_request(p, ofconn, osr, arg_size);
2797 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2800 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2804 static long long int
2805 msec_from_nsec(uint64_t sec, uint32_t nsec)
2807 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2811 update_time(struct ofproto *ofproto, struct rule *rule,
2812 const struct odp_flow_stats *stats)
2814 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2815 if (used > rule->used) {
2817 if (rule->super && used > rule->super->used) {
2818 rule->super->used = used;
2820 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
2825 update_stats(struct ofproto *ofproto, struct rule *rule,
2826 const struct odp_flow_stats *stats)
2828 if (stats->n_packets) {
2829 update_time(ofproto, rule, stats);
2830 rule->packet_count += stats->n_packets;
2831 rule->byte_count += stats->n_bytes;
2832 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
2838 add_flow(struct ofproto *p, struct ofconn *ofconn,
2839 struct ofp_flow_mod *ofm, size_t n_actions)
2841 struct ofpbuf *packet;
2846 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
2850 flow_from_match(&flow, &wildcards, &ofm->match);
2851 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
2852 ntohs(ofm->priority))) {
2853 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
2857 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
2858 n_actions, ntohs(ofm->idle_timeout),
2859 ntohs(ofm->hard_timeout), ofm->cookie,
2860 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
2861 cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority));
2864 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2865 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2869 in_port = UINT16_MAX;
2872 rule_insert(p, rule, packet, in_port);
2873 ofpbuf_delete(packet);
2878 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
2879 size_t n_actions, uint16_t command, struct rule *rule)
2881 if (rule_is_hidden(rule)) {
2885 if (command == OFPFC_DELETE) {
2886 long long int now = time_msec();
2887 send_flow_removed(p, rule, now, OFPRR_DELETE);
2888 rule_remove(p, rule);
2890 size_t actions_len = n_actions * sizeof *rule->actions;
2892 if (n_actions == rule->n_actions
2893 && !memcmp(ofm->actions, rule->actions, actions_len))
2898 free(rule->actions);
2899 rule->actions = xmemdup(ofm->actions, actions_len);
2900 rule->n_actions = n_actions;
2901 rule->flow_cookie = ofm->cookie;
2903 if (rule->cr.wc.wildcards) {
2904 COVERAGE_INC(ofproto_mod_wc_flow);
2905 p->need_revalidate = true;
2907 rule_update_actions(p, rule);
2915 modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm,
2916 size_t n_actions, uint16_t command)
2922 flow_from_match(&flow, &wildcards, &ofm->match);
2923 rule = rule_from_cls_rule(classifier_find_rule_exactly(
2924 &p->cls, &flow, wildcards,
2925 ntohs(ofm->priority)));
2928 if (command == OFPFC_DELETE
2929 && ofm->out_port != htons(OFPP_NONE)
2930 && !rule_has_out_port(rule, ofm->out_port)) {
2934 modify_flow(p, ofm, n_actions, command, rule);
2939 struct modify_flows_cbdata {
2940 struct ofproto *ofproto;
2941 const struct ofp_flow_mod *ofm;
2948 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
2950 struct rule *rule = rule_from_cls_rule(rule_);
2951 struct modify_flows_cbdata *cbdata = cbdata_;
2953 if (cbdata->out_port != htons(OFPP_NONE)
2954 && !rule_has_out_port(rule, cbdata->out_port)) {
2958 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions,
2959 cbdata->command, rule);
2963 modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm,
2964 size_t n_actions, uint16_t command)
2966 struct modify_flows_cbdata cbdata;
2967 struct cls_rule target;
2971 cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port
2972 : htons(OFPP_NONE));
2973 cbdata.n_actions = n_actions;
2974 cbdata.command = command;
2976 cls_rule_from_match(&target, &ofm->match, 0);
2978 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2979 modify_flows_cb, &cbdata);
2984 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
2985 struct ofp_flow_mod *ofm)
2990 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
2991 sizeof *ofm->actions, &n_actions);
2996 /* We do not support the emergency flow cache. It will hopefully
2997 * get dropped from OpenFlow in the near future. */
2998 if (ofm->flags & htons(OFPFF_EMERG)) {
2999 /* There isn't a good fit for an error code, so just state that the
3000 * flow table is full. */
3001 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3004 normalize_match(&ofm->match);
3005 if (!ofm->match.wildcards) {
3006 ofm->priority = htons(UINT16_MAX);
3009 error = validate_actions((const union ofp_action *) ofm->actions,
3010 n_actions, p->max_ports);
3015 switch (ntohs(ofm->command)) {
3017 return add_flow(p, ofconn, ofm, n_actions);
3020 return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY);
3022 case OFPFC_MODIFY_STRICT:
3023 return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY);
3026 return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE);
3028 case OFPFC_DELETE_STRICT:
3029 return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE);
3032 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3037 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3039 struct ofp_vendor_header *ovh = msg;
3040 struct nicira_header *nh;
3042 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3043 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3045 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3046 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3048 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3049 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3053 switch (ntohl(nh->subtype)) {
3054 case NXT_STATUS_REQUEST:
3055 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3059 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3063 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
3065 struct ofp_header *ob;
3068 /* Currently, everything executes synchronously, so we can just
3069 * immediately send the barrier reply. */
3070 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3071 queue_tx(buf, ofconn, ofconn->reply_counter);
3076 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3077 struct ofpbuf *ofp_msg)
3079 struct ofp_header *oh = ofp_msg->data;
3082 COVERAGE_INC(ofproto_recv_openflow);
3084 case OFPT_ECHO_REQUEST:
3085 error = handle_echo_request(ofconn, oh);
3088 case OFPT_ECHO_REPLY:
3092 case OFPT_FEATURES_REQUEST:
3093 error = handle_features_request(p, ofconn, oh);
3096 case OFPT_GET_CONFIG_REQUEST:
3097 error = handle_get_config_request(p, ofconn, oh);
3100 case OFPT_SET_CONFIG:
3101 error = handle_set_config(p, ofconn, ofp_msg->data);
3104 case OFPT_PACKET_OUT:
3105 error = handle_packet_out(p, ofconn, ofp_msg->data);
3109 error = handle_port_mod(p, oh);
3113 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3116 case OFPT_STATS_REQUEST:
3117 error = handle_stats_request(p, ofconn, oh);
3121 error = handle_vendor(p, ofconn, ofp_msg->data);
3124 case OFPT_BARRIER_REQUEST:
3125 error = handle_barrier_request(ofconn, oh);
3129 if (VLOG_IS_WARN_ENABLED()) {
3130 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3131 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3134 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3139 send_error_oh(ofconn, ofp_msg->data, error);
3144 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
3146 struct odp_msg *msg = packet->data;
3147 uint16_t in_port = odp_port_to_ofp_port(msg->port);
3149 struct ofpbuf payload;
3152 payload.data = msg + 1;
3153 payload.size = msg->length - sizeof *msg;
3154 flow_extract(&payload, msg->port, &flow);
3156 /* Check with in-band control to see if this packet should be sent
3157 * to the local port regardless of the flow table. */
3158 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3159 union odp_action action;
3161 memset(&action, 0, sizeof(action));
3162 action.output.type = ODPAT_OUTPUT;
3163 action.output.port = ODPP_LOCAL;
3164 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3167 rule = lookup_valid_rule(p, &flow);
3169 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3170 struct ofport *port = port_array_get(&p->ports, msg->port);
3172 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3173 COVERAGE_INC(ofproto_no_packet_in);
3174 /* XXX install 'drop' flow entry */
3175 ofpbuf_delete(packet);
3179 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3182 COVERAGE_INC(ofproto_packet_in);
3183 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3187 if (rule->cr.wc.wildcards) {
3188 rule = rule_create_subrule(p, rule, &flow);
3189 rule_make_actions(p, rule, packet);
3191 if (!rule->may_install) {
3192 /* The rule is not installable, that is, we need to process every
3193 * packet, so process the current packet and set its actions into
3195 rule_make_actions(p, rule, packet);
3197 /* XXX revalidate rule if it needs it */
3201 rule_execute(p, rule, &payload, &flow);
3202 rule_reinstall(p, rule);
3204 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY
3205 && rconn_is_connected(p->controller->rconn)) {
3207 * Extra-special case for fail-open mode.
3209 * We are in fail-open mode and the packet matched the fail-open rule,
3210 * but we are connected to a controller too. We should send the packet
3211 * up to the controller in the hope that it will try to set up a flow
3212 * and thereby allow us to exit fail-open.
3214 * See the top-level comment in fail-open.c for more information.
3216 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3218 ofpbuf_delete(packet);
3223 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3225 struct odp_msg *msg = packet->data;
3227 switch (msg->type) {
3228 case _ODPL_ACTION_NR:
3229 COVERAGE_INC(ofproto_ctlr_action);
3230 pinsched_send(p->action_sched, odp_port_to_ofp_port(msg->port), packet,
3231 send_packet_in_action, p);
3234 case _ODPL_SFLOW_NR:
3236 ofproto_sflow_received(p->sflow, msg);
3238 ofpbuf_delete(packet);
3242 handle_odp_miss_msg(p, packet);
3246 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
3253 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3255 struct rule *sub = rule_from_cls_rule(sub_);
3256 struct revalidate_cbdata *cbdata = cbdata_;
3258 if (cbdata->revalidate_all
3259 || (cbdata->revalidate_subrules && sub->super)
3260 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3261 revalidate_rule(cbdata->ofproto, sub);
3266 revalidate_rule(struct ofproto *p, struct rule *rule)
3268 const flow_t *flow = &rule->cr.flow;
3270 COVERAGE_INC(ofproto_revalidate_rule);
3273 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3275 rule_remove(p, rule);
3277 } else if (super != rule->super) {
3278 COVERAGE_INC(ofproto_revalidate_moved);
3279 list_remove(&rule->list);
3280 list_push_back(&super->list, &rule->list);
3281 rule->super = super;
3282 rule->hard_timeout = super->hard_timeout;
3283 rule->idle_timeout = super->idle_timeout;
3284 rule->created = super->created;
3289 rule_update_actions(p, rule);
3293 static struct ofpbuf *
3294 compose_flow_removed(const struct rule *rule, long long int now, uint8_t reason)
3296 struct ofp_flow_removed *ofr;
3298 long long int tdiff = time_msec() - rule->created;
3299 uint32_t sec = tdiff / 1000;
3300 uint32_t msec = tdiff - (sec * 1000);
3302 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
3303 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofr->match);
3304 ofr->cookie = rule->flow_cookie;
3305 ofr->priority = htons(rule->cr.priority);
3306 ofr->reason = reason;
3307 ofr->duration_sec = htonl(sec);
3308 ofr->duration_nsec = htonl(msec * 1000000);
3309 ofr->idle_timeout = htons(rule->idle_timeout);
3310 ofr->packet_count = htonll(rule->packet_count);
3311 ofr->byte_count = htonll(rule->byte_count);
3317 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3319 assert(rule->installed);
3320 assert(!rule->cr.wc.wildcards);
3323 rule_remove(ofproto, rule);
3325 rule_uninstall(ofproto, rule);
3329 send_flow_removed(struct ofproto *p, struct rule *rule,
3330 long long int now, uint8_t reason)
3332 struct ofconn *ofconn;
3333 struct ofconn *prev;
3334 struct ofpbuf *buf = NULL;
3336 /* We limit the maximum number of queued flow expirations it by accounting
3337 * them under the counter for replies. That works because preventing
3338 * OpenFlow requests from being processed also prevents new flows from
3339 * being added (and expiring). (It also prevents processing OpenFlow
3340 * requests that would not add new flows, so it is imperfect.) */
3343 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3344 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)) {
3346 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3348 buf = compose_flow_removed(rule, now, reason);
3354 queue_tx(buf, prev, prev->reply_counter);
3360 expire_rule(struct cls_rule *cls_rule, void *p_)
3362 struct ofproto *p = p_;
3363 struct rule *rule = rule_from_cls_rule(cls_rule);
3364 long long int hard_expire, idle_expire, expire, now;
3366 hard_expire = (rule->hard_timeout
3367 ? rule->created + rule->hard_timeout * 1000
3369 idle_expire = (rule->idle_timeout
3370 && (rule->super || list_is_empty(&rule->list))
3371 ? rule->used + rule->idle_timeout * 1000
3373 expire = MIN(hard_expire, idle_expire);
3377 if (rule->installed && now >= rule->used + 5000) {
3378 uninstall_idle_flow(p, rule);
3379 } else if (!rule->cr.wc.wildcards) {
3380 active_timeout(p, rule);
3386 COVERAGE_INC(ofproto_expired);
3388 /* Update stats. This code will be a no-op if the rule expired
3389 * due to an idle timeout. */
3390 if (rule->cr.wc.wildcards) {
3391 struct rule *subrule, *next;
3392 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3393 rule_remove(p, subrule);
3396 rule_uninstall(p, rule);
3399 if (!rule_is_hidden(rule)) {
3400 send_flow_removed(p, rule, now,
3402 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
3404 rule_remove(p, rule);
3408 active_timeout(struct ofproto *ofproto, struct rule *rule)
3410 if (ofproto->netflow && !is_controller_rule(rule) &&
3411 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
3412 struct ofexpired expired;
3413 struct odp_flow odp_flow;
3415 /* Get updated flow stats. */
3416 memset(&odp_flow, 0, sizeof odp_flow);
3417 if (rule->installed) {
3418 odp_flow.key = rule->cr.flow;
3419 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
3420 dpif_flow_get(ofproto->dpif, &odp_flow);
3422 if (odp_flow.stats.n_packets) {
3423 update_time(ofproto, rule, &odp_flow.stats);
3424 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
3425 odp_flow.stats.tcp_flags);
3429 expired.flow = rule->cr.flow;
3430 expired.packet_count = rule->packet_count +
3431 odp_flow.stats.n_packets;
3432 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
3433 expired.used = rule->used;
3435 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
3437 /* Schedule us to send the accumulated records once we have
3438 * collected all of them. */
3439 poll_immediate_wake();
3444 update_used(struct ofproto *p)
3446 struct odp_flow *flows;
3451 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
3456 for (i = 0; i < n_flows; i++) {
3457 struct odp_flow *f = &flows[i];
3460 rule = rule_from_cls_rule(
3461 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3462 if (!rule || !rule->installed) {
3463 COVERAGE_INC(ofproto_unexpected_rule);
3464 dpif_flow_del(p->dpif, f);
3468 update_time(p, rule, &f->stats);
3469 rule_account(p, rule, f->stats.n_bytes);
3475 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3476 const struct ofpbuf *packet, int send_len)
3478 struct odp_msg *msg = packet->data;
3479 struct ofpbuf payload;
3483 /* Extract packet payload from 'msg'. */
3484 payload.data = msg + 1;
3485 payload.size = msg->length - sizeof *msg;
3487 /* Construct ofp_packet_in message. */
3488 reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3489 opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port), reason,
3490 &payload, send_len);
3493 rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100);
3497 send_packet_in_action(struct ofpbuf *packet, void *p_)
3499 struct ofproto *p = p_;
3500 struct ofconn *ofconn;
3501 struct odp_msg *msg;
3504 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3505 if (ofconn == p->controller || ofconn->miss_send_len) {
3506 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3509 ofpbuf_delete(packet);
3513 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3515 struct ofproto *p = p_;
3516 bool in_fail_open = p->fail_open && fail_open_is_active(p->fail_open);
3517 struct ofconn *ofconn;
3518 struct ofpbuf payload;
3519 struct odp_msg *msg;
3522 payload.data = msg + 1;
3523 payload.size = msg->length - sizeof *msg;
3524 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3525 if (ofconn->miss_send_len) {
3526 struct pktbuf *pb = ofconn->pktbuf;
3527 uint32_t buffer_id = (in_fail_open
3529 : pktbuf_save(pb, &payload, msg->port));
3530 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3532 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3535 ofpbuf_delete(packet);
3539 pick_datapath_id(const struct ofproto *ofproto)
3541 const struct ofport *port;
3543 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
3545 uint8_t ea[ETH_ADDR_LEN];
3548 error = netdev_get_etheraddr(port->netdev, ea);
3550 return eth_addr_to_uint64(ea);
3552 VLOG_WARN("could not get MAC address for %s (%s)",
3553 netdev_get_name(port->netdev), strerror(error));
3555 return ofproto->fallback_dpid;
3559 pick_fallback_dpid(void)
3561 uint8_t ea[ETH_ADDR_LEN];
3562 eth_addr_nicira_random(ea);
3563 return eth_addr_to_uint64(ea);
3567 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3568 struct odp_actions *actions, tag_type *tags,
3569 uint16_t *nf_output_iface, void *ofproto_)
3571 struct ofproto *ofproto = ofproto_;
3574 /* Drop frames for reserved multicast addresses. */
3575 if (eth_addr_is_reserved(flow->dl_dst)) {
3579 /* Learn source MAC (but don't try to learn from revalidation). */
3580 if (packet != NULL) {
3581 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3584 /* The log messages here could actually be useful in debugging,
3585 * so keep the rate limit relatively high. */
3586 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3587 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3588 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3589 ofproto_revalidate(ofproto, rev_tag);
3593 /* Determine output port. */
3594 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3596 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
3597 } else if (out_port != flow->in_port) {
3598 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3599 *nf_output_iface = out_port;
3607 static const struct ofhooks default_ofhooks = {
3609 default_normal_ofhook_cb,