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 *mfr_desc; /* Manufacturer. */
191 char *hw_desc; /* Hardware. */
192 char *sw_desc; /* Software version. */
193 char *serial_desc; /* 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->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
303 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
304 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
305 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
306 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
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 *mfr_desc, const char *hw_desc,
394 const char *sw_desc, const char *serial_desc,
397 struct ofp_desc_stats *ods;
400 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
401 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
402 sizeof ods->mfr_desc);
405 p->mfr_desc = xstrdup(mfr_desc);
408 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
409 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
410 sizeof ods->hw_desc);
413 p->hw_desc = xstrdup(hw_desc);
416 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
417 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
418 sizeof ods->sw_desc);
421 p->sw_desc = xstrdup(sw_desc);
424 if (strlen(serial_desc) >= sizeof ods->serial_num) {
425 VLOG_WARN("truncating serial_desc, must be less than %zu "
427 sizeof ods->serial_num);
429 free(p->serial_desc);
430 p->serial_desc = xstrdup(serial_desc);
433 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
434 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
435 sizeof ods->dp_desc);
438 p->dp_desc = xstrdup(dp_desc);
443 ofproto_set_in_band(struct ofproto *p, bool in_band)
445 if (in_band != (p->in_band != NULL)) {
447 return in_band_create(p, p->dpif, p->switch_status,
448 p->controller->rconn, &p->in_band);
450 ofproto_set_discovery(p, false, NULL, true);
451 in_band_destroy(p->in_band);
454 rconn_reconnect(p->controller->rconn);
460 ofproto_set_discovery(struct ofproto *p, bool discovery,
461 const char *re, bool update_resolv_conf)
463 if (discovery != (p->discovery != NULL)) {
465 int error = ofproto_set_in_band(p, true);
469 error = discovery_create(re, update_resolv_conf,
470 p->dpif, p->switch_status,
476 discovery_destroy(p->discovery);
479 rconn_disconnect(p->controller->rconn);
480 } else if (discovery) {
481 discovery_set_update_resolv_conf(p->discovery, update_resolv_conf);
482 return discovery_set_accept_controller_re(p->discovery, re);
488 ofproto_set_controller(struct ofproto *ofproto, const char *controller)
490 if (ofproto->discovery) {
492 } else if (controller) {
493 if (strcmp(rconn_get_name(ofproto->controller->rconn), controller)) {
494 return rconn_connect(ofproto->controller->rconn, controller);
499 rconn_disconnect(ofproto->controller->rconn);
505 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
506 const struct svec *svec)
508 struct pvconn **pvconns = *pvconnsp;
509 size_t n_pvconns = *n_pvconnsp;
513 for (i = 0; i < n_pvconns; i++) {
514 pvconn_close(pvconns[i]);
518 pvconns = xmalloc(svec->n * sizeof *pvconns);
520 for (i = 0; i < svec->n; i++) {
521 const char *name = svec->names[i];
522 struct pvconn *pvconn;
525 error = pvconn_open(name, &pvconn);
527 pvconns[n_pvconns++] = pvconn;
529 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
537 *n_pvconnsp = n_pvconns;
543 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
545 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
549 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
551 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
555 ofproto_set_netflow(struct ofproto *ofproto,
556 const struct netflow_options *nf_options)
558 if (nf_options && nf_options->collectors.n) {
559 if (!ofproto->netflow) {
560 ofproto->netflow = netflow_create();
562 return netflow_set_options(ofproto->netflow, nf_options);
564 netflow_destroy(ofproto->netflow);
565 ofproto->netflow = NULL;
571 ofproto_set_sflow(struct ofproto *ofproto,
572 const struct ofproto_sflow_options *oso)
574 struct ofproto_sflow *os = ofproto->sflow;
577 struct ofport *ofport;
578 unsigned int odp_port;
580 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
581 refresh_port_groups(ofproto);
582 PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) {
583 ofproto_sflow_add_port(os, odp_port,
584 netdev_get_name(ofport->netdev));
587 ofproto_sflow_set_options(os, oso);
589 ofproto_sflow_destroy(os);
590 ofproto->sflow = NULL;
595 ofproto_set_failure(struct ofproto *ofproto, bool fail_open)
598 struct rconn *rconn = ofproto->controller->rconn;
599 int trigger_duration = rconn_get_probe_interval(rconn) * 3;
600 if (!ofproto->fail_open) {
601 ofproto->fail_open = fail_open_create(ofproto, trigger_duration,
602 ofproto->switch_status,
605 fail_open_set_trigger_duration(ofproto->fail_open,
609 fail_open_destroy(ofproto->fail_open);
610 ofproto->fail_open = NULL;
615 ofproto_set_rate_limit(struct ofproto *ofproto,
616 int rate_limit, int burst_limit)
618 if (rate_limit > 0) {
619 if (!ofproto->miss_sched) {
620 ofproto->miss_sched = pinsched_create(rate_limit, burst_limit,
621 ofproto->switch_status);
622 ofproto->action_sched = pinsched_create(rate_limit, burst_limit,
625 pinsched_set_limits(ofproto->miss_sched, rate_limit, burst_limit);
626 pinsched_set_limits(ofproto->action_sched,
627 rate_limit, burst_limit);
630 pinsched_destroy(ofproto->miss_sched);
631 ofproto->miss_sched = NULL;
632 pinsched_destroy(ofproto->action_sched);
633 ofproto->action_sched = NULL;
638 ofproto_set_stp(struct ofproto *ofproto OVS_UNUSED, bool enable_stp)
642 VLOG_WARN("STP is not yet implemented");
650 ofproto_get_datapath_id(const struct ofproto *ofproto)
652 return ofproto->datapath_id;
656 ofproto_get_probe_interval(const struct ofproto *ofproto)
658 return rconn_get_probe_interval(ofproto->controller->rconn);
662 ofproto_get_max_backoff(const struct ofproto *ofproto)
664 return rconn_get_max_backoff(ofproto->controller->rconn);
668 ofproto_get_in_band(const struct ofproto *ofproto)
670 return ofproto->in_band != NULL;
674 ofproto_get_discovery(const struct ofproto *ofproto)
676 return ofproto->discovery != NULL;
680 ofproto_get_controller(const struct ofproto *ofproto)
682 return rconn_get_name(ofproto->controller->rconn);
686 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
690 for (i = 0; i < ofproto->n_listeners; i++) {
691 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
696 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
700 for (i = 0; i < ofproto->n_snoops; i++) {
701 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
706 ofproto_destroy(struct ofproto *p)
708 struct ofconn *ofconn, *next_ofconn;
709 struct ofport *ofport;
710 unsigned int port_no;
717 /* Destroy fail-open early, because it touches the classifier. */
718 ofproto_set_failure(p, false);
720 ofproto_flush_flows(p);
721 classifier_destroy(&p->cls);
723 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
725 ofconn_destroy(ofconn);
729 netdev_monitor_destroy(p->netdev_monitor);
730 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
733 shash_destroy(&p->port_by_name);
735 switch_status_destroy(p->switch_status);
736 in_band_destroy(p->in_band);
737 discovery_destroy(p->discovery);
738 pinsched_destroy(p->miss_sched);
739 pinsched_destroy(p->action_sched);
740 netflow_destroy(p->netflow);
741 ofproto_sflow_destroy(p->sflow);
743 switch_status_unregister(p->ss_cat);
745 for (i = 0; i < p->n_listeners; i++) {
746 pvconn_close(p->listeners[i]);
750 for (i = 0; i < p->n_snoops; i++) {
751 pvconn_close(p->snoops[i]);
755 mac_learning_destroy(p->ml);
760 free(p->serial_desc);
767 ofproto_run(struct ofproto *p)
769 int error = ofproto_run1(p);
771 error = ofproto_run2(p, false);
777 process_port_change(struct ofproto *ofproto, int error, char *devname)
779 if (error == ENOBUFS) {
780 reinit_ports(ofproto);
782 update_port(ofproto, devname);
788 ofproto_run1(struct ofproto *p)
790 struct ofconn *ofconn, *next_ofconn;
795 if (shash_is_empty(&p->port_by_name)) {
799 for (i = 0; i < 50; i++) {
803 error = dpif_recv(p->dpif, &buf);
805 if (error == ENODEV) {
806 /* Someone destroyed the datapath behind our back. The caller
807 * better destroy us and give up, because we're just going to
808 * spin from here on out. */
809 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
810 VLOG_ERR_RL(&rl, "%s: datapath was destroyed externally",
817 handle_odp_msg(p, buf);
820 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
821 process_port_change(p, error, devname);
823 while ((error = netdev_monitor_poll(p->netdev_monitor,
824 &devname)) != EAGAIN) {
825 process_port_change(p, error, devname);
829 in_band_run(p->in_band);
832 char *controller_name;
833 if (rconn_is_connectivity_questionable(p->controller->rconn)) {
834 discovery_question_connectivity(p->discovery);
836 if (discovery_run(p->discovery, &controller_name)) {
837 if (controller_name) {
838 rconn_connect(p->controller->rconn, controller_name);
840 rconn_disconnect(p->controller->rconn);
844 pinsched_run(p->miss_sched, send_packet_in_miss, p);
845 pinsched_run(p->action_sched, send_packet_in_action, p);
847 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
849 ofconn_run(ofconn, p);
852 /* Fail-open maintenance. Do this after processing the ofconns since
853 * fail-open checks the status of the controller rconn. */
855 fail_open_run(p->fail_open);
858 for (i = 0; i < p->n_listeners; i++) {
862 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
864 ofconn_create(p, rconn_new_from_vconn("passive", vconn));
865 } else if (retval != EAGAIN) {
866 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
870 for (i = 0; i < p->n_snoops; i++) {
874 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
876 rconn_add_monitor(p->controller->rconn, vconn);
877 } else if (retval != EAGAIN) {
878 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
882 if (time_msec() >= p->next_expiration) {
883 COVERAGE_INC(ofproto_expiration);
884 p->next_expiration = time_msec() + 1000;
887 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
889 /* Let the hook know that we're at a stable point: all outstanding data
890 * in existing flows has been accounted to the account_cb. Thus, the
891 * hook can now reasonably do operations that depend on having accurate
892 * flow volume accounting (currently, that's just bond rebalancing). */
893 if (p->ofhooks->account_checkpoint_cb) {
894 p->ofhooks->account_checkpoint_cb(p->aux);
899 netflow_run(p->netflow);
902 ofproto_sflow_run(p->sflow);
908 struct revalidate_cbdata {
909 struct ofproto *ofproto;
910 bool revalidate_all; /* Revalidate all exact-match rules? */
911 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
912 struct tag_set revalidate_set; /* Set of tags to revalidate. */
916 ofproto_run2(struct ofproto *p, bool revalidate_all)
918 if (p->need_revalidate || revalidate_all
919 || !tag_set_is_empty(&p->revalidate_set)) {
920 struct revalidate_cbdata cbdata;
922 cbdata.revalidate_all = revalidate_all;
923 cbdata.revalidate_subrules = p->need_revalidate;
924 cbdata.revalidate_set = p->revalidate_set;
925 tag_set_init(&p->revalidate_set);
926 COVERAGE_INC(ofproto_revalidate);
927 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
928 p->need_revalidate = false;
935 ofproto_wait(struct ofproto *p)
937 struct ofconn *ofconn;
940 dpif_recv_wait(p->dpif);
941 dpif_port_poll_wait(p->dpif);
942 netdev_monitor_poll_wait(p->netdev_monitor);
943 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
947 in_band_wait(p->in_band);
950 discovery_wait(p->discovery);
953 fail_open_wait(p->fail_open);
955 pinsched_wait(p->miss_sched);
956 pinsched_wait(p->action_sched);
958 ofproto_sflow_wait(p->sflow);
960 if (!tag_set_is_empty(&p->revalidate_set)) {
961 poll_immediate_wake();
963 if (p->need_revalidate) {
964 /* Shouldn't happen, but if it does just go around again. */
965 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
966 poll_immediate_wake();
967 } else if (p->next_expiration != LLONG_MAX) {
968 poll_timer_wait(p->next_expiration - time_msec());
970 for (i = 0; i < p->n_listeners; i++) {
971 pvconn_wait(p->listeners[i]);
973 for (i = 0; i < p->n_snoops; i++) {
974 pvconn_wait(p->snoops[i]);
979 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
981 tag_set_add(&ofproto->revalidate_set, tag);
985 ofproto_get_revalidate_set(struct ofproto *ofproto)
987 return &ofproto->revalidate_set;
991 ofproto_is_alive(const struct ofproto *p)
993 return p->discovery || rconn_is_alive(p->controller->rconn);
997 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
998 const union ofp_action *actions, size_t n_actions,
999 const struct ofpbuf *packet)
1001 struct odp_actions odp_actions;
1004 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1010 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1012 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
1013 odp_actions.n_actions, packet);
1018 ofproto_add_flow(struct ofproto *p,
1019 const flow_t *flow, uint32_t wildcards, unsigned int priority,
1020 const union ofp_action *actions, size_t n_actions,
1024 rule = rule_create(p, NULL, actions, n_actions,
1025 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1027 cls_rule_from_flow(&rule->cr, flow, wildcards, priority);
1028 rule_insert(p, rule, NULL, 0);
1032 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1033 uint32_t wildcards, unsigned int priority)
1037 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1041 rule_remove(ofproto, rule);
1046 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1048 struct rule *rule = rule_from_cls_rule(rule_);
1049 struct ofproto *ofproto = ofproto_;
1051 /* Mark the flow as not installed, even though it might really be
1052 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1053 * There is no point in uninstalling it individually since we are about to
1054 * blow away all the flows with dpif_flow_flush(). */
1055 rule->installed = false;
1057 rule_remove(ofproto, rule);
1061 ofproto_flush_flows(struct ofproto *ofproto)
1063 COVERAGE_INC(ofproto_flush);
1064 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1065 dpif_flow_flush(ofproto->dpif);
1066 if (ofproto->in_band) {
1067 in_band_flushed(ofproto->in_band);
1069 if (ofproto->fail_open) {
1070 fail_open_flushed(ofproto->fail_open);
1075 reinit_ports(struct ofproto *p)
1077 struct svec devnames;
1078 struct ofport *ofport;
1079 unsigned int port_no;
1080 struct odp_port *odp_ports;
1084 svec_init(&devnames);
1085 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1086 svec_add (&devnames, (char *) ofport->opp.name);
1088 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1089 for (i = 0; i < n_odp_ports; i++) {
1090 svec_add (&devnames, odp_ports[i].devname);
1094 svec_sort_unique(&devnames);
1095 for (i = 0; i < devnames.n; i++) {
1096 update_port(p, devnames.names[i]);
1098 svec_destroy(&devnames);
1102 refresh_port_group(struct ofproto *p, unsigned int group)
1106 struct ofport *port;
1107 unsigned int port_no;
1109 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1111 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1113 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1114 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1115 ports[n_ports++] = port_no;
1118 dpif_port_group_set(p->dpif, group, ports, n_ports);
1125 refresh_port_groups(struct ofproto *p)
1127 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1128 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1130 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1134 static struct ofport *
1135 make_ofport(const struct odp_port *odp_port)
1137 struct netdev_options netdev_options;
1138 enum netdev_flags flags;
1139 struct ofport *ofport;
1140 struct netdev *netdev;
1144 memset(&netdev_options, 0, sizeof netdev_options);
1145 netdev_options.name = odp_port->devname;
1146 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1147 netdev_options.may_open = true;
1149 error = netdev_open(&netdev_options, &netdev);
1151 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1152 "cannot be opened (%s)",
1153 odp_port->devname, odp_port->port,
1154 odp_port->devname, strerror(error));
1158 ofport = xmalloc(sizeof *ofport);
1159 ofport->netdev = netdev;
1160 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1161 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1162 memcpy(ofport->opp.name, odp_port->devname,
1163 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1164 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1166 netdev_get_flags(netdev, &flags);
1167 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1169 netdev_get_carrier(netdev, &carrier);
1170 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1172 netdev_get_features(netdev,
1173 &ofport->opp.curr, &ofport->opp.advertised,
1174 &ofport->opp.supported, &ofport->opp.peer);
1179 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1181 if (port_array_get(&p->ports, odp_port->port)) {
1182 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1185 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1186 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1195 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1197 const struct ofp_phy_port *a = &a_->opp;
1198 const struct ofp_phy_port *b = &b_->opp;
1200 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1201 return (a->port_no == b->port_no
1202 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1203 && !strcmp((char *) a->name, (char *) b->name)
1204 && a->state == b->state
1205 && a->config == b->config
1206 && a->curr == b->curr
1207 && a->advertised == b->advertised
1208 && a->supported == b->supported
1209 && a->peer == b->peer);
1213 send_port_status(struct ofproto *p, const struct ofport *ofport,
1216 /* XXX Should limit the number of queued port status change messages. */
1217 struct ofconn *ofconn;
1218 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1219 struct ofp_port_status *ops;
1222 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1223 ops->reason = reason;
1224 ops->desc = ofport->opp;
1225 hton_ofp_phy_port(&ops->desc);
1226 queue_tx(b, ofconn, NULL);
1228 if (p->ofhooks->port_changed_cb) {
1229 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1234 ofport_install(struct ofproto *p, struct ofport *ofport)
1236 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1237 const char *netdev_name = (const char *) ofport->opp.name;
1239 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1240 port_array_set(&p->ports, odp_port, ofport);
1241 shash_add(&p->port_by_name, netdev_name, ofport);
1243 ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
1248 ofport_remove(struct ofproto *p, struct ofport *ofport)
1250 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1252 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1253 port_array_set(&p->ports, odp_port, NULL);
1254 shash_delete(&p->port_by_name,
1255 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1257 ofproto_sflow_del_port(p->sflow, odp_port);
1262 ofport_free(struct ofport *ofport)
1265 netdev_close(ofport->netdev);
1271 update_port(struct ofproto *p, const char *devname)
1273 struct odp_port odp_port;
1274 struct ofport *old_ofport;
1275 struct ofport *new_ofport;
1278 COVERAGE_INC(ofproto_update_port);
1280 /* Query the datapath for port information. */
1281 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1283 /* Find the old ofport. */
1284 old_ofport = shash_find_data(&p->port_by_name, devname);
1287 /* There's no port named 'devname' but there might be a port with
1288 * the same port number. This could happen if a port is deleted
1289 * and then a new one added in its place very quickly, or if a port
1290 * is renamed. In the former case we want to send an OFPPR_DELETE
1291 * and an OFPPR_ADD, and in the latter case we want to send a
1292 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1293 * the old port's ifindex against the new port, or perhaps less
1294 * reliably but more portably by comparing the old port's MAC
1295 * against the new port's MAC. However, this code isn't that smart
1296 * and always sends an OFPPR_MODIFY (XXX). */
1297 old_ofport = port_array_get(&p->ports, odp_port.port);
1299 } else if (error != ENOENT && error != ENODEV) {
1300 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1301 "%s", strerror(error));
1305 /* Create a new ofport. */
1306 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1308 /* Eliminate a few pathological cases. */
1309 if (!old_ofport && !new_ofport) {
1311 } else if (old_ofport && new_ofport) {
1312 /* Most of the 'config' bits are OpenFlow soft state, but
1313 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1314 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1315 * leaves the other bits 0.) */
1316 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1318 if (ofport_equal(old_ofport, new_ofport)) {
1319 /* False alarm--no change. */
1320 ofport_free(new_ofport);
1325 /* Now deal with the normal cases. */
1327 ofport_remove(p, old_ofport);
1330 ofport_install(p, new_ofport);
1332 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1333 (!old_ofport ? OFPPR_ADD
1334 : !new_ofport ? OFPPR_DELETE
1336 ofport_free(old_ofport);
1338 /* Update port groups. */
1339 refresh_port_groups(p);
1343 init_ports(struct ofproto *p)
1345 struct odp_port *ports;
1350 error = dpif_port_list(p->dpif, &ports, &n_ports);
1355 for (i = 0; i < n_ports; i++) {
1356 const struct odp_port *odp_port = &ports[i];
1357 if (!ofport_conflicts(p, odp_port)) {
1358 struct ofport *ofport = make_ofport(odp_port);
1360 ofport_install(p, ofport);
1365 refresh_port_groups(p);
1369 static struct ofconn *
1370 ofconn_create(struct ofproto *p, struct rconn *rconn)
1372 struct ofconn *ofconn = xmalloc(sizeof *ofconn);
1373 list_push_back(&p->all_conns, &ofconn->node);
1374 ofconn->rconn = rconn;
1375 ofconn->pktbuf = NULL;
1376 ofconn->miss_send_len = 0;
1377 ofconn->packet_in_counter = rconn_packet_counter_create ();
1378 ofconn->reply_counter = rconn_packet_counter_create ();
1383 ofconn_destroy(struct ofconn *ofconn)
1385 list_remove(&ofconn->node);
1386 rconn_destroy(ofconn->rconn);
1387 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1388 rconn_packet_counter_destroy(ofconn->reply_counter);
1389 pktbuf_destroy(ofconn->pktbuf);
1394 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1398 rconn_run(ofconn->rconn);
1400 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1401 /* Limit the number of iterations to prevent other tasks from
1403 for (iteration = 0; iteration < 50; iteration++) {
1404 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1409 fail_open_maybe_recover(p->fail_open);
1411 handle_openflow(ofconn, p, of_msg);
1412 ofpbuf_delete(of_msg);
1416 if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
1417 ofconn_destroy(ofconn);
1422 ofconn_wait(struct ofconn *ofconn)
1424 rconn_run_wait(ofconn->rconn);
1425 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1426 rconn_recv_wait(ofconn->rconn);
1428 COVERAGE_INC(ofproto_ofconn_stuck);
1432 /* Caller is responsible for initializing the 'cr' member of the returned
1434 static struct rule *
1435 rule_create(struct ofproto *ofproto, struct rule *super,
1436 const union ofp_action *actions, size_t n_actions,
1437 uint16_t idle_timeout, uint16_t hard_timeout,
1438 uint64_t flow_cookie, bool send_flow_removed)
1440 struct rule *rule = xzalloc(sizeof *rule);
1441 rule->idle_timeout = idle_timeout;
1442 rule->hard_timeout = hard_timeout;
1443 rule->flow_cookie = flow_cookie;
1444 rule->used = rule->created = time_msec();
1445 rule->send_flow_removed = send_flow_removed;
1446 rule->super = super;
1448 list_push_back(&super->list, &rule->list);
1450 list_init(&rule->list);
1452 rule->n_actions = n_actions;
1453 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1454 netflow_flow_clear(&rule->nf_flow);
1455 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1460 static struct rule *
1461 rule_from_cls_rule(const struct cls_rule *cls_rule)
1463 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1467 rule_free(struct rule *rule)
1469 free(rule->actions);
1470 free(rule->odp_actions);
1474 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1475 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1476 * through all of its subrules and revalidates them, destroying any that no
1477 * longer has a super-rule (which is probably all of them).
1479 * Before calling this function, the caller must make have removed 'rule' from
1480 * the classifier. If 'rule' is an exact-match rule, the caller is also
1481 * responsible for ensuring that it has been uninstalled from the datapath. */
1483 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1486 struct rule *subrule, *next;
1487 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1488 revalidate_rule(ofproto, subrule);
1491 list_remove(&rule->list);
1497 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1499 const union ofp_action *oa;
1500 struct actions_iterator i;
1502 if (out_port == htons(OFPP_NONE)) {
1505 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1506 oa = actions_next(&i)) {
1507 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1514 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1515 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1517 * The flow that 'packet' actually contains does not need to actually match
1518 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1519 * the packet and byte counters for 'rule' will be credited for the packet sent
1520 * out whether or not the packet actually matches 'rule'.
1522 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1523 * the caller must already have accurately composed ODP actions for it given
1524 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1525 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1526 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1527 * actions and apply them to 'packet'. */
1529 rule_execute(struct ofproto *ofproto, struct rule *rule,
1530 struct ofpbuf *packet, const flow_t *flow)
1532 const union odp_action *actions;
1534 struct odp_actions a;
1536 /* Grab or compose the ODP actions.
1538 * The special case for an exact-match 'rule' where 'flow' is not the
1539 * rule's flow is important to avoid, e.g., sending a packet out its input
1540 * port simply because the ODP actions were composed for the wrong
1542 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1543 struct rule *super = rule->super ? rule->super : rule;
1544 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1545 packet, &a, NULL, 0, NULL)) {
1548 actions = a.actions;
1549 n_actions = a.n_actions;
1551 actions = rule->odp_actions;
1552 n_actions = rule->n_odp_actions;
1555 /* Execute the ODP actions. */
1556 if (!dpif_execute(ofproto->dpif, flow->in_port,
1557 actions, n_actions, packet)) {
1558 struct odp_flow_stats stats;
1559 flow_extract_stats(flow, packet, &stats);
1560 update_stats(ofproto, rule, &stats);
1561 rule->used = time_msec();
1562 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1567 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1570 struct rule *displaced_rule;
1572 /* Insert the rule in the classifier. */
1573 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1574 if (!rule->cr.wc.wildcards) {
1575 rule_make_actions(p, rule, packet);
1578 /* Send the packet and credit it to the rule. */
1581 flow_extract(packet, in_port, &flow);
1582 rule_execute(p, rule, packet, &flow);
1585 /* Install the rule in the datapath only after sending the packet, to
1586 * avoid packet reordering. */
1587 if (rule->cr.wc.wildcards) {
1588 COVERAGE_INC(ofproto_add_wc_flow);
1589 p->need_revalidate = true;
1591 rule_install(p, rule, displaced_rule);
1594 /* Free the rule that was displaced, if any. */
1595 if (displaced_rule) {
1596 rule_destroy(p, displaced_rule);
1600 static struct rule *
1601 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1604 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1605 rule->idle_timeout, rule->hard_timeout,
1607 COVERAGE_INC(ofproto_subrule_create);
1608 cls_rule_from_flow(&subrule->cr, flow, 0,
1609 (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1610 : rule->cr.priority));
1611 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1617 rule_remove(struct ofproto *ofproto, struct rule *rule)
1619 if (rule->cr.wc.wildcards) {
1620 COVERAGE_INC(ofproto_del_wc_flow);
1621 ofproto->need_revalidate = true;
1623 rule_uninstall(ofproto, rule);
1625 classifier_remove(&ofproto->cls, &rule->cr);
1626 rule_destroy(ofproto, rule);
1629 /* Returns true if the actions changed, false otherwise. */
1631 rule_make_actions(struct ofproto *p, struct rule *rule,
1632 const struct ofpbuf *packet)
1634 const struct rule *super;
1635 struct odp_actions a;
1638 assert(!rule->cr.wc.wildcards);
1640 super = rule->super ? rule->super : rule;
1642 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1643 packet, &a, &rule->tags, &rule->may_install,
1644 &rule->nf_flow.output_iface);
1646 actions_len = a.n_actions * sizeof *a.actions;
1647 if (rule->n_odp_actions != a.n_actions
1648 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1649 COVERAGE_INC(ofproto_odp_unchanged);
1650 free(rule->odp_actions);
1651 rule->n_odp_actions = a.n_actions;
1652 rule->odp_actions = xmemdup(a.actions, actions_len);
1660 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1661 struct odp_flow_put *put)
1663 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1664 put->flow.key = rule->cr.flow;
1665 put->flow.actions = rule->odp_actions;
1666 put->flow.n_actions = rule->n_odp_actions;
1668 return dpif_flow_put(ofproto->dpif, put);
1672 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1674 assert(!rule->cr.wc.wildcards);
1676 if (rule->may_install) {
1677 struct odp_flow_put put;
1678 if (!do_put_flow(p, rule,
1679 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1681 rule->installed = true;
1682 if (displaced_rule) {
1683 update_stats(p, displaced_rule, &put.flow.stats);
1684 rule_post_uninstall(p, displaced_rule);
1687 } else if (displaced_rule) {
1688 rule_uninstall(p, displaced_rule);
1693 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1695 if (rule->installed) {
1696 struct odp_flow_put put;
1697 COVERAGE_INC(ofproto_dp_missed);
1698 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1700 rule_install(ofproto, rule, NULL);
1705 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1707 bool actions_changed;
1708 uint16_t new_out_iface, old_out_iface;
1710 old_out_iface = rule->nf_flow.output_iface;
1711 actions_changed = rule_make_actions(ofproto, rule, NULL);
1713 if (rule->may_install) {
1714 if (rule->installed) {
1715 if (actions_changed) {
1716 struct odp_flow_put put;
1717 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
1718 | ODPPF_ZERO_STATS, &put);
1719 update_stats(ofproto, rule, &put.flow.stats);
1721 /* Temporarily set the old output iface so that NetFlow
1722 * messages have the correct output interface for the old
1724 new_out_iface = rule->nf_flow.output_iface;
1725 rule->nf_flow.output_iface = old_out_iface;
1726 rule_post_uninstall(ofproto, rule);
1727 rule->nf_flow.output_iface = new_out_iface;
1730 rule_install(ofproto, rule, NULL);
1733 rule_uninstall(ofproto, rule);
1738 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1740 uint64_t total_bytes = rule->byte_count + extra_bytes;
1742 if (ofproto->ofhooks->account_flow_cb
1743 && total_bytes > rule->accounted_bytes)
1745 ofproto->ofhooks->account_flow_cb(
1746 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1747 total_bytes - rule->accounted_bytes, ofproto->aux);
1748 rule->accounted_bytes = total_bytes;
1753 rule_uninstall(struct ofproto *p, struct rule *rule)
1755 assert(!rule->cr.wc.wildcards);
1756 if (rule->installed) {
1757 struct odp_flow odp_flow;
1759 odp_flow.key = rule->cr.flow;
1760 odp_flow.actions = NULL;
1761 odp_flow.n_actions = 0;
1762 if (!dpif_flow_del(p->dpif, &odp_flow)) {
1763 update_stats(p, rule, &odp_flow.stats);
1765 rule->installed = false;
1767 rule_post_uninstall(p, rule);
1772 is_controller_rule(struct rule *rule)
1774 /* If the only action is send to the controller then don't report
1775 * NetFlow expiration messages since it is just part of the control
1776 * logic for the network and not real traffic. */
1778 if (rule && rule->super) {
1779 struct rule *super = rule->super;
1781 return super->n_actions == 1 &&
1782 super->actions[0].type == htons(OFPAT_OUTPUT) &&
1783 super->actions[0].output.port == htons(OFPP_CONTROLLER);
1790 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1792 struct rule *super = rule->super;
1794 rule_account(ofproto, rule, 0);
1796 if (ofproto->netflow && !is_controller_rule(rule)) {
1797 struct ofexpired expired;
1798 expired.flow = rule->cr.flow;
1799 expired.packet_count = rule->packet_count;
1800 expired.byte_count = rule->byte_count;
1801 expired.used = rule->used;
1802 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
1805 super->packet_count += rule->packet_count;
1806 super->byte_count += rule->byte_count;
1808 /* Reset counters to prevent double counting if the rule ever gets
1810 rule->packet_count = 0;
1811 rule->byte_count = 0;
1812 rule->accounted_bytes = 0;
1814 netflow_flow_clear(&rule->nf_flow);
1819 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1820 struct rconn_packet_counter *counter)
1822 update_openflow_length(msg);
1823 if (rconn_send(ofconn->rconn, msg, counter)) {
1829 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1830 int error, const void *data, size_t len)
1833 struct ofp_error_msg *oem;
1835 if (!(error >> 16)) {
1836 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1841 COVERAGE_INC(ofproto_error);
1842 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1843 oh ? oh->xid : 0, &buf);
1844 oem->type = htons((unsigned int) error >> 16);
1845 oem->code = htons(error & 0xffff);
1846 memcpy(oem->data, data, len);
1847 queue_tx(buf, ofconn, ofconn->reply_counter);
1851 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1854 size_t oh_length = ntohs(oh->length);
1855 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1859 hton_ofp_phy_port(struct ofp_phy_port *opp)
1861 opp->port_no = htons(opp->port_no);
1862 opp->config = htonl(opp->config);
1863 opp->state = htonl(opp->state);
1864 opp->curr = htonl(opp->curr);
1865 opp->advertised = htonl(opp->advertised);
1866 opp->supported = htonl(opp->supported);
1867 opp->peer = htonl(opp->peer);
1871 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
1873 struct ofp_header *rq = oh;
1874 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
1879 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
1880 struct ofp_header *oh)
1882 struct ofp_switch_features *osf;
1884 unsigned int port_no;
1885 struct ofport *port;
1887 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1888 osf->datapath_id = htonll(p->datapath_id);
1889 osf->n_buffers = htonl(pktbuf_capacity());
1891 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1892 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
1893 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1894 (1u << OFPAT_SET_VLAN_VID) |
1895 (1u << OFPAT_SET_VLAN_PCP) |
1896 (1u << OFPAT_STRIP_VLAN) |
1897 (1u << OFPAT_SET_DL_SRC) |
1898 (1u << OFPAT_SET_DL_DST) |
1899 (1u << OFPAT_SET_NW_SRC) |
1900 (1u << OFPAT_SET_NW_DST) |
1901 (1u << OFPAT_SET_NW_TOS) |
1902 (1u << OFPAT_SET_TP_SRC) |
1903 (1u << OFPAT_SET_TP_DST));
1905 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1906 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1909 queue_tx(buf, ofconn, ofconn->reply_counter);
1914 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1915 struct ofp_header *oh)
1918 struct ofp_switch_config *osc;
1922 /* Figure out flags. */
1923 dpif_get_drop_frags(p->dpif, &drop_frags);
1924 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1927 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1928 osc->flags = htons(flags);
1929 osc->miss_send_len = htons(ofconn->miss_send_len);
1930 queue_tx(buf, ofconn, ofconn->reply_counter);
1936 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1937 struct ofp_switch_config *osc)
1942 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1946 flags = ntohs(osc->flags);
1948 if (ofconn == p->controller) {
1949 switch (flags & OFPC_FRAG_MASK) {
1950 case OFPC_FRAG_NORMAL:
1951 dpif_set_drop_frags(p->dpif, false);
1953 case OFPC_FRAG_DROP:
1954 dpif_set_drop_frags(p->dpif, true);
1957 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1963 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1964 if (ntohs(osc->miss_send_len) != 0) {
1965 ofconn->pktbuf = pktbuf_create();
1967 pktbuf_destroy(ofconn->pktbuf);
1971 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1977 add_output_group_action(struct odp_actions *actions, uint16_t group,
1978 uint16_t *nf_output_iface)
1980 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1982 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
1983 *nf_output_iface = NF_OUT_FLOOD;
1988 add_controller_action(struct odp_actions *actions,
1989 const struct ofp_action_output *oao)
1991 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
1992 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
1995 struct action_xlate_ctx {
1997 const flow_t *flow; /* Flow to which these actions correspond. */
1998 int recurse; /* Recursion level, via xlate_table_action. */
1999 struct ofproto *ofproto;
2000 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2001 * null pointer if we are revalidating
2002 * without a packet to refer to. */
2005 struct odp_actions *out; /* Datapath actions. */
2006 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2007 bool may_set_up_flow; /* True ordinarily; false if the actions must
2008 * be reassessed for every packet. */
2009 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2012 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2013 struct action_xlate_ctx *ctx);
2016 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2018 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
2021 if (ofport->opp.config & OFPPC_NO_FWD) {
2022 /* Forwarding disabled on port. */
2027 * We don't have an ofport record for this port, but it doesn't hurt to
2028 * allow forwarding to it anyhow. Maybe such a port will appear later
2029 * and we're pre-populating the flow table.
2033 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2034 ctx->nf_output_iface = port;
2037 static struct rule *
2038 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2041 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2043 /* The rule we found might not be valid, since we could be in need of
2044 * revalidation. If it is not valid, don't return it. */
2047 && ofproto->need_revalidate
2048 && !revalidate_rule(ofproto, rule)) {
2049 COVERAGE_INC(ofproto_invalidated);
2057 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2059 if (!ctx->recurse) {
2064 flow.in_port = in_port;
2066 rule = lookup_valid_rule(ctx->ofproto, &flow);
2073 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2080 xlate_output_action(struct action_xlate_ctx *ctx,
2081 const struct ofp_action_output *oao)
2084 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2086 ctx->nf_output_iface = NF_OUT_DROP;
2088 switch (ntohs(oao->port)) {
2090 add_output_action(ctx, ctx->flow->in_port);
2093 xlate_table_action(ctx, ctx->flow->in_port);
2096 if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet,
2097 ctx->out, ctx->tags,
2098 &ctx->nf_output_iface,
2099 ctx->ofproto->aux)) {
2100 COVERAGE_INC(ofproto_uninstallable);
2101 ctx->may_set_up_flow = false;
2105 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2106 &ctx->nf_output_iface);
2109 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2111 case OFPP_CONTROLLER:
2112 add_controller_action(ctx->out, oao);
2115 add_output_action(ctx, ODPP_LOCAL);
2118 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2119 if (odp_port != ctx->flow->in_port) {
2120 add_output_action(ctx, odp_port);
2125 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2126 ctx->nf_output_iface = NF_OUT_FLOOD;
2127 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2128 ctx->nf_output_iface = prev_nf_output_iface;
2129 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2130 ctx->nf_output_iface != NF_OUT_FLOOD) {
2131 ctx->nf_output_iface = NF_OUT_MULTI;
2136 xlate_nicira_action(struct action_xlate_ctx *ctx,
2137 const struct nx_action_header *nah)
2139 const struct nx_action_resubmit *nar;
2140 int subtype = ntohs(nah->subtype);
2142 assert(nah->vendor == htonl(NX_VENDOR_ID));
2144 case NXAST_RESUBMIT:
2145 nar = (const struct nx_action_resubmit *) nah;
2146 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2150 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2156 do_xlate_actions(const union ofp_action *in, size_t n_in,
2157 struct action_xlate_ctx *ctx)
2159 struct actions_iterator iter;
2160 const union ofp_action *ia;
2161 const struct ofport *port;
2163 port = port_array_get(&ctx->ofproto->ports, ctx->flow->in_port);
2164 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2165 port->opp.config & (eth_addr_equals(ctx->flow->dl_dst, stp_eth_addr)
2166 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2167 /* Drop this flow. */
2171 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2172 uint16_t type = ntohs(ia->type);
2173 union odp_action *oa;
2177 xlate_output_action(ctx, &ia->output);
2180 case OFPAT_SET_VLAN_VID:
2181 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2182 oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2185 case OFPAT_SET_VLAN_PCP:
2186 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2187 oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2190 case OFPAT_STRIP_VLAN:
2191 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2194 case OFPAT_SET_DL_SRC:
2195 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2196 memcpy(oa->dl_addr.dl_addr,
2197 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2200 case OFPAT_SET_DL_DST:
2201 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2202 memcpy(oa->dl_addr.dl_addr,
2203 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2206 case OFPAT_SET_NW_SRC:
2207 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2208 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2211 case OFPAT_SET_NW_DST:
2212 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2213 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2216 case OFPAT_SET_NW_TOS:
2217 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2218 oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2221 case OFPAT_SET_TP_SRC:
2222 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2223 oa->tp_port.tp_port = ia->tp_port.tp_port;
2226 case OFPAT_SET_TP_DST:
2227 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2228 oa->tp_port.tp_port = ia->tp_port.tp_port;
2232 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2236 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2243 xlate_actions(const union ofp_action *in, size_t n_in,
2244 const flow_t *flow, struct ofproto *ofproto,
2245 const struct ofpbuf *packet,
2246 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2247 uint16_t *nf_output_iface)
2249 tag_type no_tags = 0;
2250 struct action_xlate_ctx ctx;
2251 COVERAGE_INC(ofproto_ofp2odp);
2252 odp_actions_init(out);
2255 ctx.ofproto = ofproto;
2256 ctx.packet = packet;
2258 ctx.tags = tags ? tags : &no_tags;
2259 ctx.may_set_up_flow = true;
2260 ctx.nf_output_iface = NF_OUT_DROP;
2261 do_xlate_actions(in, n_in, &ctx);
2263 /* Check with in-band control to see if we're allowed to set up this
2265 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2266 ctx.may_set_up_flow = false;
2269 if (may_set_up_flow) {
2270 *may_set_up_flow = ctx.may_set_up_flow;
2272 if (nf_output_iface) {
2273 *nf_output_iface = ctx.nf_output_iface;
2275 if (odp_actions_overflow(out)) {
2276 odp_actions_init(out);
2277 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2283 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2284 struct ofp_header *oh)
2286 struct ofp_packet_out *opo;
2287 struct ofpbuf payload, *buffer;
2288 struct odp_actions actions;
2294 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2298 opo = (struct ofp_packet_out *) oh;
2300 COVERAGE_INC(ofproto_packet_out);
2301 if (opo->buffer_id != htonl(UINT32_MAX)) {
2302 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2304 if (error || !buffer) {
2312 flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2313 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2314 &flow, p, &payload, &actions, NULL, NULL, NULL);
2319 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2321 ofpbuf_delete(buffer);
2327 update_port_config(struct ofproto *p, struct ofport *port,
2328 uint32_t config, uint32_t mask)
2330 mask &= config ^ port->opp.config;
2331 if (mask & OFPPC_PORT_DOWN) {
2332 if (config & OFPPC_PORT_DOWN) {
2333 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2335 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2338 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2339 if (mask & REVALIDATE_BITS) {
2340 COVERAGE_INC(ofproto_costly_flags);
2341 port->opp.config ^= mask & REVALIDATE_BITS;
2342 p->need_revalidate = true;
2344 #undef REVALIDATE_BITS
2345 if (mask & OFPPC_NO_FLOOD) {
2346 port->opp.config ^= OFPPC_NO_FLOOD;
2347 refresh_port_groups(p);
2349 if (mask & OFPPC_NO_PACKET_IN) {
2350 port->opp.config ^= OFPPC_NO_PACKET_IN;
2355 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2357 const struct ofp_port_mod *opm;
2358 struct ofport *port;
2361 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2365 opm = (struct ofp_port_mod *) oh;
2367 port = port_array_get(&p->ports,
2368 ofp_port_to_odp_port(ntohs(opm->port_no)));
2370 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2371 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2372 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2374 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2375 if (opm->advertise) {
2376 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2382 static struct ofpbuf *
2383 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2385 struct ofp_stats_reply *osr;
2388 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2389 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2391 osr->flags = htons(0);
2395 static struct ofpbuf *
2396 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2398 return make_stats_reply(request->header.xid, request->type, body_len);
2402 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2404 struct ofpbuf *msg = *msgp;
2405 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2406 if (nbytes + msg->size > UINT16_MAX) {
2407 struct ofp_stats_reply *reply = msg->data;
2408 reply->flags = htons(OFPSF_REPLY_MORE);
2409 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2410 queue_tx(msg, ofconn, ofconn->reply_counter);
2412 return ofpbuf_put_uninit(*msgp, nbytes);
2416 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2417 struct ofp_stats_request *request)
2419 struct ofp_desc_stats *ods;
2422 msg = start_stats_reply(request, sizeof *ods);
2423 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2424 memset(ods, 0, sizeof *ods);
2425 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2426 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2427 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2428 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2429 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2430 queue_tx(msg, ofconn, ofconn->reply_counter);
2436 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2438 struct rule *rule = rule_from_cls_rule(cls_rule);
2439 int *n_subrules = n_subrules_;
2447 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2448 struct ofp_stats_request *request)
2450 struct ofp_table_stats *ots;
2452 struct odp_stats dpstats;
2453 int n_exact, n_subrules, n_wild;
2455 msg = start_stats_reply(request, sizeof *ots * 2);
2457 /* Count rules of various kinds. */
2459 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2460 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2461 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2464 dpif_get_dp_stats(p->dpif, &dpstats);
2465 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2466 memset(ots, 0, sizeof *ots);
2467 ots->table_id = TABLEID_HASH;
2468 strcpy(ots->name, "hash");
2469 ots->wildcards = htonl(0);
2470 ots->max_entries = htonl(dpstats.max_capacity);
2471 ots->active_count = htonl(n_exact);
2472 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2474 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2476 /* Classifier table. */
2477 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2478 memset(ots, 0, sizeof *ots);
2479 ots->table_id = TABLEID_CLASSIFIER;
2480 strcpy(ots->name, "classifier");
2481 ots->wildcards = htonl(OFPFW_ALL);
2482 ots->max_entries = htonl(65536);
2483 ots->active_count = htonl(n_wild);
2484 ots->lookup_count = htonll(0); /* XXX */
2485 ots->matched_count = htonll(0); /* XXX */
2487 queue_tx(msg, ofconn, ofconn->reply_counter);
2492 append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
2495 struct netdev_stats stats;
2496 struct ofp_port_stats *ops;
2498 /* Intentionally ignore return value, since errors will set
2499 * 'stats' to all-1s, which is correct for OpenFlow, and
2500 * netdev_get_stats() will log errors. */
2501 netdev_get_stats(port->netdev, &stats);
2503 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2504 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2505 memset(ops->pad, 0, sizeof ops->pad);
2506 ops->rx_packets = htonll(stats.rx_packets);
2507 ops->tx_packets = htonll(stats.tx_packets);
2508 ops->rx_bytes = htonll(stats.rx_bytes);
2509 ops->tx_bytes = htonll(stats.tx_bytes);
2510 ops->rx_dropped = htonll(stats.rx_dropped);
2511 ops->tx_dropped = htonll(stats.tx_dropped);
2512 ops->rx_errors = htonll(stats.rx_errors);
2513 ops->tx_errors = htonll(stats.tx_errors);
2514 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2515 ops->rx_over_err = htonll(stats.rx_over_errors);
2516 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2517 ops->collisions = htonll(stats.collisions);
2521 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2522 struct ofp_stats_request *osr,
2525 struct ofp_port_stats_request *psr;
2526 struct ofp_port_stats *ops;
2528 struct ofport *port;
2529 unsigned int port_no;
2531 if (arg_size != sizeof *psr) {
2532 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2534 psr = (struct ofp_port_stats_request *) osr->body;
2536 msg = start_stats_reply(osr, sizeof *ops * 16);
2537 if (psr->port_no != htons(OFPP_NONE)) {
2538 port = port_array_get(&p->ports,
2539 ofp_port_to_odp_port(ntohs(psr->port_no)));
2541 append_port_stat(port, ntohs(psr->port_no), ofconn, msg);
2544 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2545 append_port_stat(port, port_no, ofconn, msg);
2549 queue_tx(msg, ofconn, ofconn->reply_counter);
2553 struct flow_stats_cbdata {
2554 struct ofproto *ofproto;
2555 struct ofconn *ofconn;
2561 query_stats(struct ofproto *p, struct rule *rule,
2562 uint64_t *packet_countp, uint64_t *byte_countp)
2564 uint64_t packet_count, byte_count;
2565 struct rule *subrule;
2566 struct odp_flow *odp_flows;
2569 packet_count = rule->packet_count;
2570 byte_count = rule->byte_count;
2572 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2573 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
2574 if (rule->cr.wc.wildcards) {
2576 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2577 odp_flows[i++].key = subrule->cr.flow;
2578 packet_count += subrule->packet_count;
2579 byte_count += subrule->byte_count;
2582 odp_flows[0].key = rule->cr.flow;
2585 packet_count = rule->packet_count;
2586 byte_count = rule->byte_count;
2587 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2589 for (i = 0; i < n_odp_flows; i++) {
2590 struct odp_flow *odp_flow = &odp_flows[i];
2591 packet_count += odp_flow->stats.n_packets;
2592 byte_count += odp_flow->stats.n_bytes;
2597 *packet_countp = packet_count;
2598 *byte_countp = byte_count;
2602 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2604 struct rule *rule = rule_from_cls_rule(rule_);
2605 struct flow_stats_cbdata *cbdata = cbdata_;
2606 struct ofp_flow_stats *ofs;
2607 uint64_t packet_count, byte_count;
2608 size_t act_len, len;
2609 long long int tdiff = time_msec() - rule->created;
2610 uint32_t sec = tdiff / 1000;
2611 uint32_t msec = tdiff - (sec * 1000);
2613 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2617 act_len = sizeof *rule->actions * rule->n_actions;
2618 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2620 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2622 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2623 ofs->length = htons(len);
2624 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2626 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match);
2627 ofs->duration_sec = htonl(sec);
2628 ofs->duration_nsec = htonl(msec * 1000000);
2629 ofs->cookie = rule->flow_cookie;
2630 ofs->priority = htons(rule->cr.priority);
2631 ofs->idle_timeout = htons(rule->idle_timeout);
2632 ofs->hard_timeout = htons(rule->hard_timeout);
2633 memset(ofs->pad2, 0, sizeof ofs->pad2);
2634 ofs->packet_count = htonll(packet_count);
2635 ofs->byte_count = htonll(byte_count);
2636 memcpy(ofs->actions, rule->actions, act_len);
2640 table_id_to_include(uint8_t table_id)
2642 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2643 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2644 : table_id == 0xff ? CLS_INC_ALL
2649 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2650 const struct ofp_stats_request *osr,
2653 struct ofp_flow_stats_request *fsr;
2654 struct flow_stats_cbdata cbdata;
2655 struct cls_rule target;
2657 if (arg_size != sizeof *fsr) {
2658 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2660 fsr = (struct ofp_flow_stats_request *) osr->body;
2662 COVERAGE_INC(ofproto_flows_req);
2664 cbdata.ofconn = ofconn;
2665 cbdata.out_port = fsr->out_port;
2666 cbdata.msg = start_stats_reply(osr, 1024);
2667 cls_rule_from_match(&target, &fsr->match, 0);
2668 classifier_for_each_match(&p->cls, &target,
2669 table_id_to_include(fsr->table_id),
2670 flow_stats_cb, &cbdata);
2671 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2675 struct flow_stats_ds_cbdata {
2676 struct ofproto *ofproto;
2681 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
2683 struct rule *rule = rule_from_cls_rule(rule_);
2684 struct flow_stats_ds_cbdata *cbdata = cbdata_;
2685 struct ds *results = cbdata->results;
2686 struct ofp_match match;
2687 uint64_t packet_count, byte_count;
2688 size_t act_len = sizeof *rule->actions * rule->n_actions;
2690 /* Don't report on subrules. */
2691 if (rule->super != NULL) {
2695 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2696 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &match);
2698 ds_put_format(results, "duration=%llds, ",
2699 (time_msec() - rule->created) / 1000);
2700 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2701 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2702 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2703 ofp_print_match(results, &match, true);
2704 ofp_print_actions(results, &rule->actions->header, act_len);
2705 ds_put_cstr(results, "\n");
2708 /* Adds a pretty-printed description of all flows to 'results', including
2709 * those marked hidden by secchan (e.g., by in-band control). */
2711 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2713 struct ofp_match match;
2714 struct cls_rule target;
2715 struct flow_stats_ds_cbdata cbdata;
2717 memset(&match, 0, sizeof match);
2718 match.wildcards = htonl(OFPFW_ALL);
2721 cbdata.results = results;
2723 cls_rule_from_match(&target, &match, 0);
2724 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2725 flow_stats_ds_cb, &cbdata);
2728 struct aggregate_stats_cbdata {
2729 struct ofproto *ofproto;
2731 uint64_t packet_count;
2732 uint64_t byte_count;
2737 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2739 struct rule *rule = rule_from_cls_rule(rule_);
2740 struct aggregate_stats_cbdata *cbdata = cbdata_;
2741 uint64_t packet_count, byte_count;
2743 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2747 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2749 cbdata->packet_count += packet_count;
2750 cbdata->byte_count += byte_count;
2755 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2756 const struct ofp_stats_request *osr,
2759 struct ofp_aggregate_stats_request *asr;
2760 struct ofp_aggregate_stats_reply *reply;
2761 struct aggregate_stats_cbdata cbdata;
2762 struct cls_rule target;
2765 if (arg_size != sizeof *asr) {
2766 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2768 asr = (struct ofp_aggregate_stats_request *) osr->body;
2770 COVERAGE_INC(ofproto_agg_request);
2772 cbdata.out_port = asr->out_port;
2773 cbdata.packet_count = 0;
2774 cbdata.byte_count = 0;
2776 cls_rule_from_match(&target, &asr->match, 0);
2777 classifier_for_each_match(&p->cls, &target,
2778 table_id_to_include(asr->table_id),
2779 aggregate_stats_cb, &cbdata);
2781 msg = start_stats_reply(osr, sizeof *reply);
2782 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2783 reply->flow_count = htonl(cbdata.n_flows);
2784 reply->packet_count = htonll(cbdata.packet_count);
2785 reply->byte_count = htonll(cbdata.byte_count);
2786 queue_tx(msg, ofconn, ofconn->reply_counter);
2791 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2792 struct ofp_header *oh)
2794 struct ofp_stats_request *osr;
2798 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2803 osr = (struct ofp_stats_request *) oh;
2805 switch (ntohs(osr->type)) {
2807 return handle_desc_stats_request(p, ofconn, osr);
2810 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2812 case OFPST_AGGREGATE:
2813 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2816 return handle_table_stats_request(p, ofconn, osr);
2819 return handle_port_stats_request(p, ofconn, osr, arg_size);
2822 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2825 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2829 static long long int
2830 msec_from_nsec(uint64_t sec, uint32_t nsec)
2832 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2836 update_time(struct ofproto *ofproto, struct rule *rule,
2837 const struct odp_flow_stats *stats)
2839 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2840 if (used > rule->used) {
2842 if (rule->super && used > rule->super->used) {
2843 rule->super->used = used;
2845 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
2850 update_stats(struct ofproto *ofproto, struct rule *rule,
2851 const struct odp_flow_stats *stats)
2853 if (stats->n_packets) {
2854 update_time(ofproto, rule, stats);
2855 rule->packet_count += stats->n_packets;
2856 rule->byte_count += stats->n_bytes;
2857 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
2863 add_flow(struct ofproto *p, struct ofconn *ofconn,
2864 struct ofp_flow_mod *ofm, size_t n_actions)
2866 struct ofpbuf *packet;
2871 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
2875 flow_from_match(&flow, &wildcards, &ofm->match);
2876 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
2877 ntohs(ofm->priority))) {
2878 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
2882 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
2883 n_actions, ntohs(ofm->idle_timeout),
2884 ntohs(ofm->hard_timeout), ofm->cookie,
2885 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
2886 cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority));
2889 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2890 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2894 in_port = UINT16_MAX;
2897 rule_insert(p, rule, packet, in_port);
2898 ofpbuf_delete(packet);
2903 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
2904 size_t n_actions, uint16_t command, struct rule *rule)
2906 if (rule_is_hidden(rule)) {
2910 if (command == OFPFC_DELETE) {
2911 long long int now = time_msec();
2912 send_flow_removed(p, rule, now, OFPRR_DELETE);
2913 rule_remove(p, rule);
2915 size_t actions_len = n_actions * sizeof *rule->actions;
2917 if (n_actions == rule->n_actions
2918 && !memcmp(ofm->actions, rule->actions, actions_len))
2923 free(rule->actions);
2924 rule->actions = xmemdup(ofm->actions, actions_len);
2925 rule->n_actions = n_actions;
2926 rule->flow_cookie = ofm->cookie;
2928 if (rule->cr.wc.wildcards) {
2929 COVERAGE_INC(ofproto_mod_wc_flow);
2930 p->need_revalidate = true;
2932 rule_update_actions(p, rule);
2940 modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm,
2941 size_t n_actions, uint16_t command)
2947 flow_from_match(&flow, &wildcards, &ofm->match);
2948 rule = rule_from_cls_rule(classifier_find_rule_exactly(
2949 &p->cls, &flow, wildcards,
2950 ntohs(ofm->priority)));
2953 if (command == OFPFC_DELETE
2954 && ofm->out_port != htons(OFPP_NONE)
2955 && !rule_has_out_port(rule, ofm->out_port)) {
2959 modify_flow(p, ofm, n_actions, command, rule);
2964 struct modify_flows_cbdata {
2965 struct ofproto *ofproto;
2966 const struct ofp_flow_mod *ofm;
2973 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
2975 struct rule *rule = rule_from_cls_rule(rule_);
2976 struct modify_flows_cbdata *cbdata = cbdata_;
2978 if (cbdata->out_port != htons(OFPP_NONE)
2979 && !rule_has_out_port(rule, cbdata->out_port)) {
2983 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions,
2984 cbdata->command, rule);
2988 modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm,
2989 size_t n_actions, uint16_t command)
2991 struct modify_flows_cbdata cbdata;
2992 struct cls_rule target;
2996 cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port
2997 : htons(OFPP_NONE));
2998 cbdata.n_actions = n_actions;
2999 cbdata.command = command;
3001 cls_rule_from_match(&target, &ofm->match, 0);
3003 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3004 modify_flows_cb, &cbdata);
3009 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3010 struct ofp_flow_mod *ofm)
3015 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3016 sizeof *ofm->actions, &n_actions);
3021 /* We do not support the emergency flow cache. It will hopefully
3022 * get dropped from OpenFlow in the near future. */
3023 if (ofm->flags & htons(OFPFF_EMERG)) {
3024 /* There isn't a good fit for an error code, so just state that the
3025 * flow table is full. */
3026 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3029 normalize_match(&ofm->match);
3030 if (!ofm->match.wildcards) {
3031 ofm->priority = htons(UINT16_MAX);
3034 error = validate_actions((const union ofp_action *) ofm->actions,
3035 n_actions, p->max_ports);
3040 switch (ntohs(ofm->command)) {
3042 return add_flow(p, ofconn, ofm, n_actions);
3045 return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY);
3047 case OFPFC_MODIFY_STRICT:
3048 return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY);
3051 return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE);
3053 case OFPFC_DELETE_STRICT:
3054 return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE);
3057 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3062 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3064 struct ofp_vendor_header *ovh = msg;
3065 struct nicira_header *nh;
3067 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3068 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3070 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3071 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3073 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3074 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3078 switch (ntohl(nh->subtype)) {
3079 case NXT_STATUS_REQUEST:
3080 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3084 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3088 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
3090 struct ofp_header *ob;
3093 /* Currently, everything executes synchronously, so we can just
3094 * immediately send the barrier reply. */
3095 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3096 queue_tx(buf, ofconn, ofconn->reply_counter);
3101 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3102 struct ofpbuf *ofp_msg)
3104 struct ofp_header *oh = ofp_msg->data;
3107 COVERAGE_INC(ofproto_recv_openflow);
3109 case OFPT_ECHO_REQUEST:
3110 error = handle_echo_request(ofconn, oh);
3113 case OFPT_ECHO_REPLY:
3117 case OFPT_FEATURES_REQUEST:
3118 error = handle_features_request(p, ofconn, oh);
3121 case OFPT_GET_CONFIG_REQUEST:
3122 error = handle_get_config_request(p, ofconn, oh);
3125 case OFPT_SET_CONFIG:
3126 error = handle_set_config(p, ofconn, ofp_msg->data);
3129 case OFPT_PACKET_OUT:
3130 error = handle_packet_out(p, ofconn, ofp_msg->data);
3134 error = handle_port_mod(p, oh);
3138 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3141 case OFPT_STATS_REQUEST:
3142 error = handle_stats_request(p, ofconn, oh);
3146 error = handle_vendor(p, ofconn, ofp_msg->data);
3149 case OFPT_BARRIER_REQUEST:
3150 error = handle_barrier_request(ofconn, oh);
3154 if (VLOG_IS_WARN_ENABLED()) {
3155 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3156 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3159 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3164 send_error_oh(ofconn, ofp_msg->data, error);
3169 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
3171 struct odp_msg *msg = packet->data;
3172 uint16_t in_port = odp_port_to_ofp_port(msg->port);
3174 struct ofpbuf payload;
3177 payload.data = msg + 1;
3178 payload.size = msg->length - sizeof *msg;
3179 flow_extract(&payload, msg->port, &flow);
3181 /* Check with in-band control to see if this packet should be sent
3182 * to the local port regardless of the flow table. */
3183 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3184 union odp_action action;
3186 memset(&action, 0, sizeof(action));
3187 action.output.type = ODPAT_OUTPUT;
3188 action.output.port = ODPP_LOCAL;
3189 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3192 rule = lookup_valid_rule(p, &flow);
3194 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3195 struct ofport *port = port_array_get(&p->ports, msg->port);
3197 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3198 COVERAGE_INC(ofproto_no_packet_in);
3199 /* XXX install 'drop' flow entry */
3200 ofpbuf_delete(packet);
3204 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3207 COVERAGE_INC(ofproto_packet_in);
3208 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3212 if (rule->cr.wc.wildcards) {
3213 rule = rule_create_subrule(p, rule, &flow);
3214 rule_make_actions(p, rule, packet);
3216 if (!rule->may_install) {
3217 /* The rule is not installable, that is, we need to process every
3218 * packet, so process the current packet and set its actions into
3220 rule_make_actions(p, rule, packet);
3222 /* XXX revalidate rule if it needs it */
3226 rule_execute(p, rule, &payload, &flow);
3227 rule_reinstall(p, rule);
3229 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY
3230 && rconn_is_connected(p->controller->rconn)) {
3232 * Extra-special case for fail-open mode.
3234 * We are in fail-open mode and the packet matched the fail-open rule,
3235 * but we are connected to a controller too. We should send the packet
3236 * up to the controller in the hope that it will try to set up a flow
3237 * and thereby allow us to exit fail-open.
3239 * See the top-level comment in fail-open.c for more information.
3241 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3243 ofpbuf_delete(packet);
3248 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3250 struct odp_msg *msg = packet->data;
3252 switch (msg->type) {
3253 case _ODPL_ACTION_NR:
3254 COVERAGE_INC(ofproto_ctlr_action);
3255 pinsched_send(p->action_sched, odp_port_to_ofp_port(msg->port), packet,
3256 send_packet_in_action, p);
3259 case _ODPL_SFLOW_NR:
3261 ofproto_sflow_received(p->sflow, msg);
3263 ofpbuf_delete(packet);
3267 handle_odp_miss_msg(p, packet);
3271 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
3278 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3280 struct rule *sub = rule_from_cls_rule(sub_);
3281 struct revalidate_cbdata *cbdata = cbdata_;
3283 if (cbdata->revalidate_all
3284 || (cbdata->revalidate_subrules && sub->super)
3285 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3286 revalidate_rule(cbdata->ofproto, sub);
3291 revalidate_rule(struct ofproto *p, struct rule *rule)
3293 const flow_t *flow = &rule->cr.flow;
3295 COVERAGE_INC(ofproto_revalidate_rule);
3298 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3300 rule_remove(p, rule);
3302 } else if (super != rule->super) {
3303 COVERAGE_INC(ofproto_revalidate_moved);
3304 list_remove(&rule->list);
3305 list_push_back(&super->list, &rule->list);
3306 rule->super = super;
3307 rule->hard_timeout = super->hard_timeout;
3308 rule->idle_timeout = super->idle_timeout;
3309 rule->created = super->created;
3314 rule_update_actions(p, rule);
3318 static struct ofpbuf *
3319 compose_flow_removed(const struct rule *rule, long long int now, uint8_t reason)
3321 struct ofp_flow_removed *ofr;
3323 long long int tdiff = time_msec() - rule->created;
3324 uint32_t sec = tdiff / 1000;
3325 uint32_t msec = tdiff - (sec * 1000);
3327 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
3328 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofr->match);
3329 ofr->cookie = rule->flow_cookie;
3330 ofr->priority = htons(rule->cr.priority);
3331 ofr->reason = reason;
3332 ofr->duration_sec = htonl(sec);
3333 ofr->duration_nsec = htonl(msec * 1000000);
3334 ofr->idle_timeout = htons(rule->idle_timeout);
3335 ofr->packet_count = htonll(rule->packet_count);
3336 ofr->byte_count = htonll(rule->byte_count);
3342 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3344 assert(rule->installed);
3345 assert(!rule->cr.wc.wildcards);
3348 rule_remove(ofproto, rule);
3350 rule_uninstall(ofproto, rule);
3354 send_flow_removed(struct ofproto *p, struct rule *rule,
3355 long long int now, uint8_t reason)
3357 struct ofconn *ofconn;
3358 struct ofconn *prev;
3359 struct ofpbuf *buf = NULL;
3361 /* We limit the maximum number of queued flow expirations it by accounting
3362 * them under the counter for replies. That works because preventing
3363 * OpenFlow requests from being processed also prevents new flows from
3364 * being added (and expiring). (It also prevents processing OpenFlow
3365 * requests that would not add new flows, so it is imperfect.) */
3368 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3369 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)) {
3371 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3373 buf = compose_flow_removed(rule, now, reason);
3379 queue_tx(buf, prev, prev->reply_counter);
3385 expire_rule(struct cls_rule *cls_rule, void *p_)
3387 struct ofproto *p = p_;
3388 struct rule *rule = rule_from_cls_rule(cls_rule);
3389 long long int hard_expire, idle_expire, expire, now;
3391 hard_expire = (rule->hard_timeout
3392 ? rule->created + rule->hard_timeout * 1000
3394 idle_expire = (rule->idle_timeout
3395 && (rule->super || list_is_empty(&rule->list))
3396 ? rule->used + rule->idle_timeout * 1000
3398 expire = MIN(hard_expire, idle_expire);
3402 if (rule->installed && now >= rule->used + 5000) {
3403 uninstall_idle_flow(p, rule);
3404 } else if (!rule->cr.wc.wildcards) {
3405 active_timeout(p, rule);
3411 COVERAGE_INC(ofproto_expired);
3413 /* Update stats. This code will be a no-op if the rule expired
3414 * due to an idle timeout. */
3415 if (rule->cr.wc.wildcards) {
3416 struct rule *subrule, *next;
3417 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3418 rule_remove(p, subrule);
3421 rule_uninstall(p, rule);
3424 if (!rule_is_hidden(rule)) {
3425 send_flow_removed(p, rule, now,
3427 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
3429 rule_remove(p, rule);
3433 active_timeout(struct ofproto *ofproto, struct rule *rule)
3435 if (ofproto->netflow && !is_controller_rule(rule) &&
3436 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
3437 struct ofexpired expired;
3438 struct odp_flow odp_flow;
3440 /* Get updated flow stats. */
3441 memset(&odp_flow, 0, sizeof odp_flow);
3442 if (rule->installed) {
3443 odp_flow.key = rule->cr.flow;
3444 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
3445 dpif_flow_get(ofproto->dpif, &odp_flow);
3447 if (odp_flow.stats.n_packets) {
3448 update_time(ofproto, rule, &odp_flow.stats);
3449 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
3450 odp_flow.stats.tcp_flags);
3454 expired.flow = rule->cr.flow;
3455 expired.packet_count = rule->packet_count +
3456 odp_flow.stats.n_packets;
3457 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
3458 expired.used = rule->used;
3460 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
3462 /* Schedule us to send the accumulated records once we have
3463 * collected all of them. */
3464 poll_immediate_wake();
3469 update_used(struct ofproto *p)
3471 struct odp_flow *flows;
3476 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
3481 for (i = 0; i < n_flows; i++) {
3482 struct odp_flow *f = &flows[i];
3485 rule = rule_from_cls_rule(
3486 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3487 if (!rule || !rule->installed) {
3488 COVERAGE_INC(ofproto_unexpected_rule);
3489 dpif_flow_del(p->dpif, f);
3493 update_time(p, rule, &f->stats);
3494 rule_account(p, rule, f->stats.n_bytes);
3500 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3501 const struct ofpbuf *packet, int send_len)
3503 struct odp_msg *msg = packet->data;
3504 struct ofpbuf payload;
3508 /* Extract packet payload from 'msg'. */
3509 payload.data = msg + 1;
3510 payload.size = msg->length - sizeof *msg;
3512 /* Construct ofp_packet_in message. */
3513 reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3514 opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port), reason,
3515 &payload, send_len);
3518 rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100);
3522 send_packet_in_action(struct ofpbuf *packet, void *p_)
3524 struct ofproto *p = p_;
3525 struct ofconn *ofconn;
3526 struct odp_msg *msg;
3529 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3530 if (ofconn == p->controller || ofconn->miss_send_len) {
3531 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3534 ofpbuf_delete(packet);
3538 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3540 struct ofproto *p = p_;
3541 bool in_fail_open = p->fail_open && fail_open_is_active(p->fail_open);
3542 struct ofconn *ofconn;
3543 struct ofpbuf payload;
3544 struct odp_msg *msg;
3547 payload.data = msg + 1;
3548 payload.size = msg->length - sizeof *msg;
3549 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3550 if (ofconn->miss_send_len) {
3551 struct pktbuf *pb = ofconn->pktbuf;
3552 uint32_t buffer_id = (in_fail_open
3554 : pktbuf_save(pb, &payload, msg->port));
3555 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3557 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3560 ofpbuf_delete(packet);
3564 pick_datapath_id(const struct ofproto *ofproto)
3566 const struct ofport *port;
3568 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
3570 uint8_t ea[ETH_ADDR_LEN];
3573 error = netdev_get_etheraddr(port->netdev, ea);
3575 return eth_addr_to_uint64(ea);
3577 VLOG_WARN("could not get MAC address for %s (%s)",
3578 netdev_get_name(port->netdev), strerror(error));
3580 return ofproto->fallback_dpid;
3584 pick_fallback_dpid(void)
3586 uint8_t ea[ETH_ADDR_LEN];
3587 eth_addr_nicira_random(ea);
3588 return eth_addr_to_uint64(ea);
3592 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3593 struct odp_actions *actions, tag_type *tags,
3594 uint16_t *nf_output_iface, void *ofproto_)
3596 struct ofproto *ofproto = ofproto_;
3599 /* Drop frames for reserved multicast addresses. */
3600 if (eth_addr_is_reserved(flow->dl_dst)) {
3604 /* Learn source MAC (but don't try to learn from revalidation). */
3605 if (packet != NULL) {
3606 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3609 /* The log messages here could actually be useful in debugging,
3610 * so keep the rate limit relatively high. */
3611 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3612 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3613 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3614 ofproto_revalidate(ofproto, rev_tag);
3618 /* Determine output port. */
3619 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3621 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
3622 } else if (out_port != flow->in_port) {
3623 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3624 *nf_output_iface = out_port;
3632 static const struct ofhooks default_ofhooks = {
3634 default_normal_ofhook_cb,