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);
763 port_array_destroy(&p->ports);
769 ofproto_run(struct ofproto *p)
771 int error = ofproto_run1(p);
773 error = ofproto_run2(p, false);
779 process_port_change(struct ofproto *ofproto, int error, char *devname)
781 if (error == ENOBUFS) {
782 reinit_ports(ofproto);
784 update_port(ofproto, devname);
790 ofproto_run1(struct ofproto *p)
792 struct ofconn *ofconn, *next_ofconn;
797 if (shash_is_empty(&p->port_by_name)) {
801 for (i = 0; i < 50; i++) {
805 error = dpif_recv(p->dpif, &buf);
807 if (error == ENODEV) {
808 /* Someone destroyed the datapath behind our back. The caller
809 * better destroy us and give up, because we're just going to
810 * spin from here on out. */
811 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
812 VLOG_ERR_RL(&rl, "%s: datapath was destroyed externally",
819 handle_odp_msg(p, buf);
822 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
823 process_port_change(p, error, devname);
825 while ((error = netdev_monitor_poll(p->netdev_monitor,
826 &devname)) != EAGAIN) {
827 process_port_change(p, error, devname);
831 in_band_run(p->in_band);
834 char *controller_name;
835 if (rconn_is_connectivity_questionable(p->controller->rconn)) {
836 discovery_question_connectivity(p->discovery);
838 if (discovery_run(p->discovery, &controller_name)) {
839 if (controller_name) {
840 rconn_connect(p->controller->rconn, controller_name);
842 rconn_disconnect(p->controller->rconn);
846 pinsched_run(p->miss_sched, send_packet_in_miss, p);
847 pinsched_run(p->action_sched, send_packet_in_action, p);
849 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
851 ofconn_run(ofconn, p);
854 /* Fail-open maintenance. Do this after processing the ofconns since
855 * fail-open checks the status of the controller rconn. */
857 fail_open_run(p->fail_open);
860 for (i = 0; i < p->n_listeners; i++) {
864 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
866 ofconn_create(p, rconn_new_from_vconn("passive", vconn));
867 } else if (retval != EAGAIN) {
868 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
872 for (i = 0; i < p->n_snoops; i++) {
876 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
878 rconn_add_monitor(p->controller->rconn, vconn);
879 } else if (retval != EAGAIN) {
880 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
884 if (time_msec() >= p->next_expiration) {
885 COVERAGE_INC(ofproto_expiration);
886 p->next_expiration = time_msec() + 1000;
889 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
891 /* Let the hook know that we're at a stable point: all outstanding data
892 * in existing flows has been accounted to the account_cb. Thus, the
893 * hook can now reasonably do operations that depend on having accurate
894 * flow volume accounting (currently, that's just bond rebalancing). */
895 if (p->ofhooks->account_checkpoint_cb) {
896 p->ofhooks->account_checkpoint_cb(p->aux);
901 netflow_run(p->netflow);
904 ofproto_sflow_run(p->sflow);
910 struct revalidate_cbdata {
911 struct ofproto *ofproto;
912 bool revalidate_all; /* Revalidate all exact-match rules? */
913 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
914 struct tag_set revalidate_set; /* Set of tags to revalidate. */
918 ofproto_run2(struct ofproto *p, bool revalidate_all)
920 if (p->need_revalidate || revalidate_all
921 || !tag_set_is_empty(&p->revalidate_set)) {
922 struct revalidate_cbdata cbdata;
924 cbdata.revalidate_all = revalidate_all;
925 cbdata.revalidate_subrules = p->need_revalidate;
926 cbdata.revalidate_set = p->revalidate_set;
927 tag_set_init(&p->revalidate_set);
928 COVERAGE_INC(ofproto_revalidate);
929 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
930 p->need_revalidate = false;
937 ofproto_wait(struct ofproto *p)
939 struct ofconn *ofconn;
942 dpif_recv_wait(p->dpif);
943 dpif_port_poll_wait(p->dpif);
944 netdev_monitor_poll_wait(p->netdev_monitor);
945 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
949 in_band_wait(p->in_band);
952 discovery_wait(p->discovery);
955 fail_open_wait(p->fail_open);
957 pinsched_wait(p->miss_sched);
958 pinsched_wait(p->action_sched);
960 ofproto_sflow_wait(p->sflow);
962 if (!tag_set_is_empty(&p->revalidate_set)) {
963 poll_immediate_wake();
965 if (p->need_revalidate) {
966 /* Shouldn't happen, but if it does just go around again. */
967 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
968 poll_immediate_wake();
969 } else if (p->next_expiration != LLONG_MAX) {
970 poll_timer_wait(p->next_expiration - time_msec());
972 for (i = 0; i < p->n_listeners; i++) {
973 pvconn_wait(p->listeners[i]);
975 for (i = 0; i < p->n_snoops; i++) {
976 pvconn_wait(p->snoops[i]);
981 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
983 tag_set_add(&ofproto->revalidate_set, tag);
987 ofproto_get_revalidate_set(struct ofproto *ofproto)
989 return &ofproto->revalidate_set;
993 ofproto_is_alive(const struct ofproto *p)
995 return p->discovery || rconn_is_alive(p->controller->rconn);
999 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
1000 const union ofp_action *actions, size_t n_actions,
1001 const struct ofpbuf *packet)
1003 struct odp_actions odp_actions;
1006 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1012 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1014 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
1015 odp_actions.n_actions, packet);
1020 ofproto_add_flow(struct ofproto *p,
1021 const flow_t *flow, uint32_t wildcards, unsigned int priority,
1022 const union ofp_action *actions, size_t n_actions,
1026 rule = rule_create(p, NULL, actions, n_actions,
1027 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1029 cls_rule_from_flow(&rule->cr, flow, wildcards, priority);
1030 rule_insert(p, rule, NULL, 0);
1034 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1035 uint32_t wildcards, unsigned int priority)
1039 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1043 rule_remove(ofproto, rule);
1048 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1050 struct rule *rule = rule_from_cls_rule(rule_);
1051 struct ofproto *ofproto = ofproto_;
1053 /* Mark the flow as not installed, even though it might really be
1054 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1055 * There is no point in uninstalling it individually since we are about to
1056 * blow away all the flows with dpif_flow_flush(). */
1057 rule->installed = false;
1059 rule_remove(ofproto, rule);
1063 ofproto_flush_flows(struct ofproto *ofproto)
1065 COVERAGE_INC(ofproto_flush);
1066 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1067 dpif_flow_flush(ofproto->dpif);
1068 if (ofproto->in_band) {
1069 in_band_flushed(ofproto->in_band);
1071 if (ofproto->fail_open) {
1072 fail_open_flushed(ofproto->fail_open);
1077 reinit_ports(struct ofproto *p)
1079 struct svec devnames;
1080 struct ofport *ofport;
1081 unsigned int port_no;
1082 struct odp_port *odp_ports;
1086 svec_init(&devnames);
1087 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1088 svec_add (&devnames, (char *) ofport->opp.name);
1090 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1091 for (i = 0; i < n_odp_ports; i++) {
1092 svec_add (&devnames, odp_ports[i].devname);
1096 svec_sort_unique(&devnames);
1097 for (i = 0; i < devnames.n; i++) {
1098 update_port(p, devnames.names[i]);
1100 svec_destroy(&devnames);
1104 refresh_port_group(struct ofproto *p, unsigned int group)
1108 struct ofport *port;
1109 unsigned int port_no;
1111 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1113 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1115 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1116 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1117 ports[n_ports++] = port_no;
1120 dpif_port_group_set(p->dpif, group, ports, n_ports);
1127 refresh_port_groups(struct ofproto *p)
1129 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1130 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1132 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1136 static struct ofport *
1137 make_ofport(const struct odp_port *odp_port)
1139 struct netdev_options netdev_options;
1140 enum netdev_flags flags;
1141 struct ofport *ofport;
1142 struct netdev *netdev;
1146 memset(&netdev_options, 0, sizeof netdev_options);
1147 netdev_options.name = odp_port->devname;
1148 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1149 netdev_options.may_open = true;
1151 error = netdev_open(&netdev_options, &netdev);
1153 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1154 "cannot be opened (%s)",
1155 odp_port->devname, odp_port->port,
1156 odp_port->devname, strerror(error));
1160 ofport = xmalloc(sizeof *ofport);
1161 ofport->netdev = netdev;
1162 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1163 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1164 memcpy(ofport->opp.name, odp_port->devname,
1165 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1166 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1168 netdev_get_flags(netdev, &flags);
1169 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1171 netdev_get_carrier(netdev, &carrier);
1172 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1174 netdev_get_features(netdev,
1175 &ofport->opp.curr, &ofport->opp.advertised,
1176 &ofport->opp.supported, &ofport->opp.peer);
1181 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1183 if (port_array_get(&p->ports, odp_port->port)) {
1184 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1187 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1188 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1197 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1199 const struct ofp_phy_port *a = &a_->opp;
1200 const struct ofp_phy_port *b = &b_->opp;
1202 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1203 return (a->port_no == b->port_no
1204 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1205 && !strcmp((char *) a->name, (char *) b->name)
1206 && a->state == b->state
1207 && a->config == b->config
1208 && a->curr == b->curr
1209 && a->advertised == b->advertised
1210 && a->supported == b->supported
1211 && a->peer == b->peer);
1215 send_port_status(struct ofproto *p, const struct ofport *ofport,
1218 /* XXX Should limit the number of queued port status change messages. */
1219 struct ofconn *ofconn;
1220 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1221 struct ofp_port_status *ops;
1224 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1225 ops->reason = reason;
1226 ops->desc = ofport->opp;
1227 hton_ofp_phy_port(&ops->desc);
1228 queue_tx(b, ofconn, NULL);
1230 if (p->ofhooks->port_changed_cb) {
1231 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1236 ofport_install(struct ofproto *p, struct ofport *ofport)
1238 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1239 const char *netdev_name = (const char *) ofport->opp.name;
1241 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1242 port_array_set(&p->ports, odp_port, ofport);
1243 shash_add(&p->port_by_name, netdev_name, ofport);
1245 ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
1250 ofport_remove(struct ofproto *p, struct ofport *ofport)
1252 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1254 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1255 port_array_set(&p->ports, odp_port, NULL);
1256 shash_delete(&p->port_by_name,
1257 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1259 ofproto_sflow_del_port(p->sflow, odp_port);
1264 ofport_free(struct ofport *ofport)
1267 netdev_close(ofport->netdev);
1273 update_port(struct ofproto *p, const char *devname)
1275 struct odp_port odp_port;
1276 struct ofport *old_ofport;
1277 struct ofport *new_ofport;
1280 COVERAGE_INC(ofproto_update_port);
1282 /* Query the datapath for port information. */
1283 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1285 /* Find the old ofport. */
1286 old_ofport = shash_find_data(&p->port_by_name, devname);
1289 /* There's no port named 'devname' but there might be a port with
1290 * the same port number. This could happen if a port is deleted
1291 * and then a new one added in its place very quickly, or if a port
1292 * is renamed. In the former case we want to send an OFPPR_DELETE
1293 * and an OFPPR_ADD, and in the latter case we want to send a
1294 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1295 * the old port's ifindex against the new port, or perhaps less
1296 * reliably but more portably by comparing the old port's MAC
1297 * against the new port's MAC. However, this code isn't that smart
1298 * and always sends an OFPPR_MODIFY (XXX). */
1299 old_ofport = port_array_get(&p->ports, odp_port.port);
1301 } else if (error != ENOENT && error != ENODEV) {
1302 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1303 "%s", strerror(error));
1307 /* Create a new ofport. */
1308 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1310 /* Eliminate a few pathological cases. */
1311 if (!old_ofport && !new_ofport) {
1313 } else if (old_ofport && new_ofport) {
1314 /* Most of the 'config' bits are OpenFlow soft state, but
1315 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1316 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1317 * leaves the other bits 0.) */
1318 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1320 if (ofport_equal(old_ofport, new_ofport)) {
1321 /* False alarm--no change. */
1322 ofport_free(new_ofport);
1327 /* Now deal with the normal cases. */
1329 ofport_remove(p, old_ofport);
1332 ofport_install(p, new_ofport);
1334 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1335 (!old_ofport ? OFPPR_ADD
1336 : !new_ofport ? OFPPR_DELETE
1338 ofport_free(old_ofport);
1340 /* Update port groups. */
1341 refresh_port_groups(p);
1345 init_ports(struct ofproto *p)
1347 struct odp_port *ports;
1352 error = dpif_port_list(p->dpif, &ports, &n_ports);
1357 for (i = 0; i < n_ports; i++) {
1358 const struct odp_port *odp_port = &ports[i];
1359 if (!ofport_conflicts(p, odp_port)) {
1360 struct ofport *ofport = make_ofport(odp_port);
1362 ofport_install(p, ofport);
1367 refresh_port_groups(p);
1371 static struct ofconn *
1372 ofconn_create(struct ofproto *p, struct rconn *rconn)
1374 struct ofconn *ofconn = xmalloc(sizeof *ofconn);
1375 list_push_back(&p->all_conns, &ofconn->node);
1376 ofconn->rconn = rconn;
1377 ofconn->pktbuf = NULL;
1378 ofconn->miss_send_len = 0;
1379 ofconn->packet_in_counter = rconn_packet_counter_create ();
1380 ofconn->reply_counter = rconn_packet_counter_create ();
1385 ofconn_destroy(struct ofconn *ofconn)
1387 list_remove(&ofconn->node);
1388 rconn_destroy(ofconn->rconn);
1389 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1390 rconn_packet_counter_destroy(ofconn->reply_counter);
1391 pktbuf_destroy(ofconn->pktbuf);
1396 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1400 rconn_run(ofconn->rconn);
1402 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1403 /* Limit the number of iterations to prevent other tasks from
1405 for (iteration = 0; iteration < 50; iteration++) {
1406 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1411 fail_open_maybe_recover(p->fail_open);
1413 handle_openflow(ofconn, p, of_msg);
1414 ofpbuf_delete(of_msg);
1418 if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
1419 ofconn_destroy(ofconn);
1424 ofconn_wait(struct ofconn *ofconn)
1426 rconn_run_wait(ofconn->rconn);
1427 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1428 rconn_recv_wait(ofconn->rconn);
1430 COVERAGE_INC(ofproto_ofconn_stuck);
1434 /* Caller is responsible for initializing the 'cr' member of the returned
1436 static struct rule *
1437 rule_create(struct ofproto *ofproto, struct rule *super,
1438 const union ofp_action *actions, size_t n_actions,
1439 uint16_t idle_timeout, uint16_t hard_timeout,
1440 uint64_t flow_cookie, bool send_flow_removed)
1442 struct rule *rule = xzalloc(sizeof *rule);
1443 rule->idle_timeout = idle_timeout;
1444 rule->hard_timeout = hard_timeout;
1445 rule->flow_cookie = flow_cookie;
1446 rule->used = rule->created = time_msec();
1447 rule->send_flow_removed = send_flow_removed;
1448 rule->super = super;
1450 list_push_back(&super->list, &rule->list);
1452 list_init(&rule->list);
1454 rule->n_actions = n_actions;
1455 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1456 netflow_flow_clear(&rule->nf_flow);
1457 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1462 static struct rule *
1463 rule_from_cls_rule(const struct cls_rule *cls_rule)
1465 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1469 rule_free(struct rule *rule)
1471 free(rule->actions);
1472 free(rule->odp_actions);
1476 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1477 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1478 * through all of its subrules and revalidates them, destroying any that no
1479 * longer has a super-rule (which is probably all of them).
1481 * Before calling this function, the caller must make have removed 'rule' from
1482 * the classifier. If 'rule' is an exact-match rule, the caller is also
1483 * responsible for ensuring that it has been uninstalled from the datapath. */
1485 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1488 struct rule *subrule, *next;
1489 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1490 revalidate_rule(ofproto, subrule);
1493 list_remove(&rule->list);
1499 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1501 const union ofp_action *oa;
1502 struct actions_iterator i;
1504 if (out_port == htons(OFPP_NONE)) {
1507 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1508 oa = actions_next(&i)) {
1509 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1516 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1517 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1519 * The flow that 'packet' actually contains does not need to actually match
1520 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1521 * the packet and byte counters for 'rule' will be credited for the packet sent
1522 * out whether or not the packet actually matches 'rule'.
1524 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1525 * the caller must already have accurately composed ODP actions for it given
1526 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1527 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1528 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1529 * actions and apply them to 'packet'. */
1531 rule_execute(struct ofproto *ofproto, struct rule *rule,
1532 struct ofpbuf *packet, const flow_t *flow)
1534 const union odp_action *actions;
1536 struct odp_actions a;
1538 /* Grab or compose the ODP actions.
1540 * The special case for an exact-match 'rule' where 'flow' is not the
1541 * rule's flow is important to avoid, e.g., sending a packet out its input
1542 * port simply because the ODP actions were composed for the wrong
1544 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1545 struct rule *super = rule->super ? rule->super : rule;
1546 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1547 packet, &a, NULL, 0, NULL)) {
1550 actions = a.actions;
1551 n_actions = a.n_actions;
1553 actions = rule->odp_actions;
1554 n_actions = rule->n_odp_actions;
1557 /* Execute the ODP actions. */
1558 if (!dpif_execute(ofproto->dpif, flow->in_port,
1559 actions, n_actions, packet)) {
1560 struct odp_flow_stats stats;
1561 flow_extract_stats(flow, packet, &stats);
1562 update_stats(ofproto, rule, &stats);
1563 rule->used = time_msec();
1564 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1569 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1572 struct rule *displaced_rule;
1574 /* Insert the rule in the classifier. */
1575 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1576 if (!rule->cr.wc.wildcards) {
1577 rule_make_actions(p, rule, packet);
1580 /* Send the packet and credit it to the rule. */
1583 flow_extract(packet, in_port, &flow);
1584 rule_execute(p, rule, packet, &flow);
1587 /* Install the rule in the datapath only after sending the packet, to
1588 * avoid packet reordering. */
1589 if (rule->cr.wc.wildcards) {
1590 COVERAGE_INC(ofproto_add_wc_flow);
1591 p->need_revalidate = true;
1593 rule_install(p, rule, displaced_rule);
1596 /* Free the rule that was displaced, if any. */
1597 if (displaced_rule) {
1598 rule_destroy(p, displaced_rule);
1602 static struct rule *
1603 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1606 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1607 rule->idle_timeout, rule->hard_timeout,
1609 COVERAGE_INC(ofproto_subrule_create);
1610 cls_rule_from_flow(&subrule->cr, flow, 0,
1611 (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1612 : rule->cr.priority));
1613 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1619 rule_remove(struct ofproto *ofproto, struct rule *rule)
1621 if (rule->cr.wc.wildcards) {
1622 COVERAGE_INC(ofproto_del_wc_flow);
1623 ofproto->need_revalidate = true;
1625 rule_uninstall(ofproto, rule);
1627 classifier_remove(&ofproto->cls, &rule->cr);
1628 rule_destroy(ofproto, rule);
1631 /* Returns true if the actions changed, false otherwise. */
1633 rule_make_actions(struct ofproto *p, struct rule *rule,
1634 const struct ofpbuf *packet)
1636 const struct rule *super;
1637 struct odp_actions a;
1640 assert(!rule->cr.wc.wildcards);
1642 super = rule->super ? rule->super : rule;
1644 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1645 packet, &a, &rule->tags, &rule->may_install,
1646 &rule->nf_flow.output_iface);
1648 actions_len = a.n_actions * sizeof *a.actions;
1649 if (rule->n_odp_actions != a.n_actions
1650 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1651 COVERAGE_INC(ofproto_odp_unchanged);
1652 free(rule->odp_actions);
1653 rule->n_odp_actions = a.n_actions;
1654 rule->odp_actions = xmemdup(a.actions, actions_len);
1662 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1663 struct odp_flow_put *put)
1665 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1666 put->flow.key = rule->cr.flow;
1667 put->flow.actions = rule->odp_actions;
1668 put->flow.n_actions = rule->n_odp_actions;
1670 return dpif_flow_put(ofproto->dpif, put);
1674 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1676 assert(!rule->cr.wc.wildcards);
1678 if (rule->may_install) {
1679 struct odp_flow_put put;
1680 if (!do_put_flow(p, rule,
1681 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1683 rule->installed = true;
1684 if (displaced_rule) {
1685 update_stats(p, displaced_rule, &put.flow.stats);
1686 rule_post_uninstall(p, displaced_rule);
1689 } else if (displaced_rule) {
1690 rule_uninstall(p, displaced_rule);
1695 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1697 if (rule->installed) {
1698 struct odp_flow_put put;
1699 COVERAGE_INC(ofproto_dp_missed);
1700 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1702 rule_install(ofproto, rule, NULL);
1707 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1709 bool actions_changed;
1710 uint16_t new_out_iface, old_out_iface;
1712 old_out_iface = rule->nf_flow.output_iface;
1713 actions_changed = rule_make_actions(ofproto, rule, NULL);
1715 if (rule->may_install) {
1716 if (rule->installed) {
1717 if (actions_changed) {
1718 struct odp_flow_put put;
1719 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
1720 | ODPPF_ZERO_STATS, &put);
1721 update_stats(ofproto, rule, &put.flow.stats);
1723 /* Temporarily set the old output iface so that NetFlow
1724 * messages have the correct output interface for the old
1726 new_out_iface = rule->nf_flow.output_iface;
1727 rule->nf_flow.output_iface = old_out_iface;
1728 rule_post_uninstall(ofproto, rule);
1729 rule->nf_flow.output_iface = new_out_iface;
1732 rule_install(ofproto, rule, NULL);
1735 rule_uninstall(ofproto, rule);
1740 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1742 uint64_t total_bytes = rule->byte_count + extra_bytes;
1744 if (ofproto->ofhooks->account_flow_cb
1745 && total_bytes > rule->accounted_bytes)
1747 ofproto->ofhooks->account_flow_cb(
1748 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1749 total_bytes - rule->accounted_bytes, ofproto->aux);
1750 rule->accounted_bytes = total_bytes;
1755 rule_uninstall(struct ofproto *p, struct rule *rule)
1757 assert(!rule->cr.wc.wildcards);
1758 if (rule->installed) {
1759 struct odp_flow odp_flow;
1761 odp_flow.key = rule->cr.flow;
1762 odp_flow.actions = NULL;
1763 odp_flow.n_actions = 0;
1764 if (!dpif_flow_del(p->dpif, &odp_flow)) {
1765 update_stats(p, rule, &odp_flow.stats);
1767 rule->installed = false;
1769 rule_post_uninstall(p, rule);
1774 is_controller_rule(struct rule *rule)
1776 /* If the only action is send to the controller then don't report
1777 * NetFlow expiration messages since it is just part of the control
1778 * logic for the network and not real traffic. */
1780 if (rule && rule->super) {
1781 struct rule *super = rule->super;
1783 return super->n_actions == 1 &&
1784 super->actions[0].type == htons(OFPAT_OUTPUT) &&
1785 super->actions[0].output.port == htons(OFPP_CONTROLLER);
1792 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1794 struct rule *super = rule->super;
1796 rule_account(ofproto, rule, 0);
1798 if (ofproto->netflow && !is_controller_rule(rule)) {
1799 struct ofexpired expired;
1800 expired.flow = rule->cr.flow;
1801 expired.packet_count = rule->packet_count;
1802 expired.byte_count = rule->byte_count;
1803 expired.used = rule->used;
1804 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
1807 super->packet_count += rule->packet_count;
1808 super->byte_count += rule->byte_count;
1810 /* Reset counters to prevent double counting if the rule ever gets
1812 rule->packet_count = 0;
1813 rule->byte_count = 0;
1814 rule->accounted_bytes = 0;
1816 netflow_flow_clear(&rule->nf_flow);
1821 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1822 struct rconn_packet_counter *counter)
1824 update_openflow_length(msg);
1825 if (rconn_send(ofconn->rconn, msg, counter)) {
1831 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1832 int error, const void *data, size_t len)
1835 struct ofp_error_msg *oem;
1837 if (!(error >> 16)) {
1838 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1843 COVERAGE_INC(ofproto_error);
1844 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1845 oh ? oh->xid : 0, &buf);
1846 oem->type = htons((unsigned int) error >> 16);
1847 oem->code = htons(error & 0xffff);
1848 memcpy(oem->data, data, len);
1849 queue_tx(buf, ofconn, ofconn->reply_counter);
1853 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1856 size_t oh_length = ntohs(oh->length);
1857 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1861 hton_ofp_phy_port(struct ofp_phy_port *opp)
1863 opp->port_no = htons(opp->port_no);
1864 opp->config = htonl(opp->config);
1865 opp->state = htonl(opp->state);
1866 opp->curr = htonl(opp->curr);
1867 opp->advertised = htonl(opp->advertised);
1868 opp->supported = htonl(opp->supported);
1869 opp->peer = htonl(opp->peer);
1873 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
1875 struct ofp_header *rq = oh;
1876 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
1881 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
1882 struct ofp_header *oh)
1884 struct ofp_switch_features *osf;
1886 unsigned int port_no;
1887 struct ofport *port;
1889 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1890 osf->datapath_id = htonll(p->datapath_id);
1891 osf->n_buffers = htonl(pktbuf_capacity());
1893 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1894 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
1895 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1896 (1u << OFPAT_SET_VLAN_VID) |
1897 (1u << OFPAT_SET_VLAN_PCP) |
1898 (1u << OFPAT_STRIP_VLAN) |
1899 (1u << OFPAT_SET_DL_SRC) |
1900 (1u << OFPAT_SET_DL_DST) |
1901 (1u << OFPAT_SET_NW_SRC) |
1902 (1u << OFPAT_SET_NW_DST) |
1903 (1u << OFPAT_SET_NW_TOS) |
1904 (1u << OFPAT_SET_TP_SRC) |
1905 (1u << OFPAT_SET_TP_DST));
1907 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1908 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1911 queue_tx(buf, ofconn, ofconn->reply_counter);
1916 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1917 struct ofp_header *oh)
1920 struct ofp_switch_config *osc;
1924 /* Figure out flags. */
1925 dpif_get_drop_frags(p->dpif, &drop_frags);
1926 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1929 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1930 osc->flags = htons(flags);
1931 osc->miss_send_len = htons(ofconn->miss_send_len);
1932 queue_tx(buf, ofconn, ofconn->reply_counter);
1938 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1939 struct ofp_switch_config *osc)
1944 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1948 flags = ntohs(osc->flags);
1950 if (ofconn == p->controller) {
1951 switch (flags & OFPC_FRAG_MASK) {
1952 case OFPC_FRAG_NORMAL:
1953 dpif_set_drop_frags(p->dpif, false);
1955 case OFPC_FRAG_DROP:
1956 dpif_set_drop_frags(p->dpif, true);
1959 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1965 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1966 if (ntohs(osc->miss_send_len) != 0) {
1967 ofconn->pktbuf = pktbuf_create();
1969 pktbuf_destroy(ofconn->pktbuf);
1973 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1979 add_output_group_action(struct odp_actions *actions, uint16_t group,
1980 uint16_t *nf_output_iface)
1982 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1984 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
1985 *nf_output_iface = NF_OUT_FLOOD;
1990 add_controller_action(struct odp_actions *actions,
1991 const struct ofp_action_output *oao)
1993 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
1994 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
1997 struct action_xlate_ctx {
1999 const flow_t *flow; /* Flow to which these actions correspond. */
2000 int recurse; /* Recursion level, via xlate_table_action. */
2001 struct ofproto *ofproto;
2002 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2003 * null pointer if we are revalidating
2004 * without a packet to refer to. */
2007 struct odp_actions *out; /* Datapath actions. */
2008 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2009 bool may_set_up_flow; /* True ordinarily; false if the actions must
2010 * be reassessed for every packet. */
2011 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2014 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2015 struct action_xlate_ctx *ctx);
2018 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2020 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
2023 if (ofport->opp.config & OFPPC_NO_FWD) {
2024 /* Forwarding disabled on port. */
2029 * We don't have an ofport record for this port, but it doesn't hurt to
2030 * allow forwarding to it anyhow. Maybe such a port will appear later
2031 * and we're pre-populating the flow table.
2035 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2036 ctx->nf_output_iface = port;
2039 static struct rule *
2040 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2043 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2045 /* The rule we found might not be valid, since we could be in need of
2046 * revalidation. If it is not valid, don't return it. */
2049 && ofproto->need_revalidate
2050 && !revalidate_rule(ofproto, rule)) {
2051 COVERAGE_INC(ofproto_invalidated);
2059 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2061 if (!ctx->recurse) {
2066 flow.in_port = in_port;
2068 rule = lookup_valid_rule(ctx->ofproto, &flow);
2075 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2082 xlate_output_action(struct action_xlate_ctx *ctx,
2083 const struct ofp_action_output *oao)
2086 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2088 ctx->nf_output_iface = NF_OUT_DROP;
2090 switch (ntohs(oao->port)) {
2092 add_output_action(ctx, ctx->flow->in_port);
2095 xlate_table_action(ctx, ctx->flow->in_port);
2098 if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet,
2099 ctx->out, ctx->tags,
2100 &ctx->nf_output_iface,
2101 ctx->ofproto->aux)) {
2102 COVERAGE_INC(ofproto_uninstallable);
2103 ctx->may_set_up_flow = false;
2107 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2108 &ctx->nf_output_iface);
2111 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2113 case OFPP_CONTROLLER:
2114 add_controller_action(ctx->out, oao);
2117 add_output_action(ctx, ODPP_LOCAL);
2120 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2121 if (odp_port != ctx->flow->in_port) {
2122 add_output_action(ctx, odp_port);
2127 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2128 ctx->nf_output_iface = NF_OUT_FLOOD;
2129 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2130 ctx->nf_output_iface = prev_nf_output_iface;
2131 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2132 ctx->nf_output_iface != NF_OUT_FLOOD) {
2133 ctx->nf_output_iface = NF_OUT_MULTI;
2138 xlate_nicira_action(struct action_xlate_ctx *ctx,
2139 const struct nx_action_header *nah)
2141 const struct nx_action_resubmit *nar;
2142 int subtype = ntohs(nah->subtype);
2144 assert(nah->vendor == htonl(NX_VENDOR_ID));
2146 case NXAST_RESUBMIT:
2147 nar = (const struct nx_action_resubmit *) nah;
2148 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2152 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2158 do_xlate_actions(const union ofp_action *in, size_t n_in,
2159 struct action_xlate_ctx *ctx)
2161 struct actions_iterator iter;
2162 const union ofp_action *ia;
2163 const struct ofport *port;
2165 port = port_array_get(&ctx->ofproto->ports, ctx->flow->in_port);
2166 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2167 port->opp.config & (eth_addr_equals(ctx->flow->dl_dst, stp_eth_addr)
2168 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2169 /* Drop this flow. */
2173 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2174 uint16_t type = ntohs(ia->type);
2175 union odp_action *oa;
2179 xlate_output_action(ctx, &ia->output);
2182 case OFPAT_SET_VLAN_VID:
2183 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2184 oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2187 case OFPAT_SET_VLAN_PCP:
2188 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2189 oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2192 case OFPAT_STRIP_VLAN:
2193 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2196 case OFPAT_SET_DL_SRC:
2197 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2198 memcpy(oa->dl_addr.dl_addr,
2199 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2202 case OFPAT_SET_DL_DST:
2203 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2204 memcpy(oa->dl_addr.dl_addr,
2205 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2208 case OFPAT_SET_NW_SRC:
2209 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2210 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2213 case OFPAT_SET_NW_DST:
2214 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2215 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2218 case OFPAT_SET_NW_TOS:
2219 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2220 oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2223 case OFPAT_SET_TP_SRC:
2224 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2225 oa->tp_port.tp_port = ia->tp_port.tp_port;
2228 case OFPAT_SET_TP_DST:
2229 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2230 oa->tp_port.tp_port = ia->tp_port.tp_port;
2234 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2238 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2245 xlate_actions(const union ofp_action *in, size_t n_in,
2246 const flow_t *flow, struct ofproto *ofproto,
2247 const struct ofpbuf *packet,
2248 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2249 uint16_t *nf_output_iface)
2251 tag_type no_tags = 0;
2252 struct action_xlate_ctx ctx;
2253 COVERAGE_INC(ofproto_ofp2odp);
2254 odp_actions_init(out);
2257 ctx.ofproto = ofproto;
2258 ctx.packet = packet;
2260 ctx.tags = tags ? tags : &no_tags;
2261 ctx.may_set_up_flow = true;
2262 ctx.nf_output_iface = NF_OUT_DROP;
2263 do_xlate_actions(in, n_in, &ctx);
2265 /* Check with in-band control to see if we're allowed to set up this
2267 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2268 ctx.may_set_up_flow = false;
2271 if (may_set_up_flow) {
2272 *may_set_up_flow = ctx.may_set_up_flow;
2274 if (nf_output_iface) {
2275 *nf_output_iface = ctx.nf_output_iface;
2277 if (odp_actions_overflow(out)) {
2278 odp_actions_init(out);
2279 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2285 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2286 struct ofp_header *oh)
2288 struct ofp_packet_out *opo;
2289 struct ofpbuf payload, *buffer;
2290 struct odp_actions actions;
2296 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2300 opo = (struct ofp_packet_out *) oh;
2302 COVERAGE_INC(ofproto_packet_out);
2303 if (opo->buffer_id != htonl(UINT32_MAX)) {
2304 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2306 if (error || !buffer) {
2314 flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2315 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2316 &flow, p, &payload, &actions, NULL, NULL, NULL);
2321 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2323 ofpbuf_delete(buffer);
2329 update_port_config(struct ofproto *p, struct ofport *port,
2330 uint32_t config, uint32_t mask)
2332 mask &= config ^ port->opp.config;
2333 if (mask & OFPPC_PORT_DOWN) {
2334 if (config & OFPPC_PORT_DOWN) {
2335 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2337 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2340 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2341 if (mask & REVALIDATE_BITS) {
2342 COVERAGE_INC(ofproto_costly_flags);
2343 port->opp.config ^= mask & REVALIDATE_BITS;
2344 p->need_revalidate = true;
2346 #undef REVALIDATE_BITS
2347 if (mask & OFPPC_NO_FLOOD) {
2348 port->opp.config ^= OFPPC_NO_FLOOD;
2349 refresh_port_groups(p);
2351 if (mask & OFPPC_NO_PACKET_IN) {
2352 port->opp.config ^= OFPPC_NO_PACKET_IN;
2357 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2359 const struct ofp_port_mod *opm;
2360 struct ofport *port;
2363 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2367 opm = (struct ofp_port_mod *) oh;
2369 port = port_array_get(&p->ports,
2370 ofp_port_to_odp_port(ntohs(opm->port_no)));
2372 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2373 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2374 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2376 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2377 if (opm->advertise) {
2378 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2384 static struct ofpbuf *
2385 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2387 struct ofp_stats_reply *osr;
2390 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2391 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2393 osr->flags = htons(0);
2397 static struct ofpbuf *
2398 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2400 return make_stats_reply(request->header.xid, request->type, body_len);
2404 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2406 struct ofpbuf *msg = *msgp;
2407 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2408 if (nbytes + msg->size > UINT16_MAX) {
2409 struct ofp_stats_reply *reply = msg->data;
2410 reply->flags = htons(OFPSF_REPLY_MORE);
2411 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2412 queue_tx(msg, ofconn, ofconn->reply_counter);
2414 return ofpbuf_put_uninit(*msgp, nbytes);
2418 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2419 struct ofp_stats_request *request)
2421 struct ofp_desc_stats *ods;
2424 msg = start_stats_reply(request, sizeof *ods);
2425 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2426 memset(ods, 0, sizeof *ods);
2427 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2428 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2429 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2430 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2431 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2432 queue_tx(msg, ofconn, ofconn->reply_counter);
2438 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2440 struct rule *rule = rule_from_cls_rule(cls_rule);
2441 int *n_subrules = n_subrules_;
2449 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2450 struct ofp_stats_request *request)
2452 struct ofp_table_stats *ots;
2454 struct odp_stats dpstats;
2455 int n_exact, n_subrules, n_wild;
2457 msg = start_stats_reply(request, sizeof *ots * 2);
2459 /* Count rules of various kinds. */
2461 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2462 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2463 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2466 dpif_get_dp_stats(p->dpif, &dpstats);
2467 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2468 memset(ots, 0, sizeof *ots);
2469 ots->table_id = TABLEID_HASH;
2470 strcpy(ots->name, "hash");
2471 ots->wildcards = htonl(0);
2472 ots->max_entries = htonl(dpstats.max_capacity);
2473 ots->active_count = htonl(n_exact);
2474 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2476 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2478 /* Classifier table. */
2479 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2480 memset(ots, 0, sizeof *ots);
2481 ots->table_id = TABLEID_CLASSIFIER;
2482 strcpy(ots->name, "classifier");
2483 ots->wildcards = htonl(OFPFW_ALL);
2484 ots->max_entries = htonl(65536);
2485 ots->active_count = htonl(n_wild);
2486 ots->lookup_count = htonll(0); /* XXX */
2487 ots->matched_count = htonll(0); /* XXX */
2489 queue_tx(msg, ofconn, ofconn->reply_counter);
2494 append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
2497 struct netdev_stats stats;
2498 struct ofp_port_stats *ops;
2500 /* Intentionally ignore return value, since errors will set
2501 * 'stats' to all-1s, which is correct for OpenFlow, and
2502 * netdev_get_stats() will log errors. */
2503 netdev_get_stats(port->netdev, &stats);
2505 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2506 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2507 memset(ops->pad, 0, sizeof ops->pad);
2508 ops->rx_packets = htonll(stats.rx_packets);
2509 ops->tx_packets = htonll(stats.tx_packets);
2510 ops->rx_bytes = htonll(stats.rx_bytes);
2511 ops->tx_bytes = htonll(stats.tx_bytes);
2512 ops->rx_dropped = htonll(stats.rx_dropped);
2513 ops->tx_dropped = htonll(stats.tx_dropped);
2514 ops->rx_errors = htonll(stats.rx_errors);
2515 ops->tx_errors = htonll(stats.tx_errors);
2516 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2517 ops->rx_over_err = htonll(stats.rx_over_errors);
2518 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2519 ops->collisions = htonll(stats.collisions);
2523 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2524 struct ofp_stats_request *osr,
2527 struct ofp_port_stats_request *psr;
2528 struct ofp_port_stats *ops;
2530 struct ofport *port;
2531 unsigned int port_no;
2533 if (arg_size != sizeof *psr) {
2534 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2536 psr = (struct ofp_port_stats_request *) osr->body;
2538 msg = start_stats_reply(osr, sizeof *ops * 16);
2539 if (psr->port_no != htons(OFPP_NONE)) {
2540 port = port_array_get(&p->ports,
2541 ofp_port_to_odp_port(ntohs(psr->port_no)));
2543 append_port_stat(port, ntohs(psr->port_no), ofconn, msg);
2546 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2547 append_port_stat(port, port_no, ofconn, msg);
2551 queue_tx(msg, ofconn, ofconn->reply_counter);
2555 struct flow_stats_cbdata {
2556 struct ofproto *ofproto;
2557 struct ofconn *ofconn;
2563 query_stats(struct ofproto *p, struct rule *rule,
2564 uint64_t *packet_countp, uint64_t *byte_countp)
2566 uint64_t packet_count, byte_count;
2567 struct rule *subrule;
2568 struct odp_flow *odp_flows;
2571 packet_count = rule->packet_count;
2572 byte_count = rule->byte_count;
2574 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2575 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
2576 if (rule->cr.wc.wildcards) {
2578 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2579 odp_flows[i++].key = subrule->cr.flow;
2580 packet_count += subrule->packet_count;
2581 byte_count += subrule->byte_count;
2584 odp_flows[0].key = rule->cr.flow;
2587 packet_count = rule->packet_count;
2588 byte_count = rule->byte_count;
2589 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2591 for (i = 0; i < n_odp_flows; i++) {
2592 struct odp_flow *odp_flow = &odp_flows[i];
2593 packet_count += odp_flow->stats.n_packets;
2594 byte_count += odp_flow->stats.n_bytes;
2599 *packet_countp = packet_count;
2600 *byte_countp = byte_count;
2604 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2606 struct rule *rule = rule_from_cls_rule(rule_);
2607 struct flow_stats_cbdata *cbdata = cbdata_;
2608 struct ofp_flow_stats *ofs;
2609 uint64_t packet_count, byte_count;
2610 size_t act_len, len;
2611 long long int tdiff = time_msec() - rule->created;
2612 uint32_t sec = tdiff / 1000;
2613 uint32_t msec = tdiff - (sec * 1000);
2615 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2619 act_len = sizeof *rule->actions * rule->n_actions;
2620 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2622 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2624 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2625 ofs->length = htons(len);
2626 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2628 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match);
2629 ofs->duration_sec = htonl(sec);
2630 ofs->duration_nsec = htonl(msec * 1000000);
2631 ofs->cookie = rule->flow_cookie;
2632 ofs->priority = htons(rule->cr.priority);
2633 ofs->idle_timeout = htons(rule->idle_timeout);
2634 ofs->hard_timeout = htons(rule->hard_timeout);
2635 memset(ofs->pad2, 0, sizeof ofs->pad2);
2636 ofs->packet_count = htonll(packet_count);
2637 ofs->byte_count = htonll(byte_count);
2638 memcpy(ofs->actions, rule->actions, act_len);
2642 table_id_to_include(uint8_t table_id)
2644 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2645 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2646 : table_id == 0xff ? CLS_INC_ALL
2651 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2652 const struct ofp_stats_request *osr,
2655 struct ofp_flow_stats_request *fsr;
2656 struct flow_stats_cbdata cbdata;
2657 struct cls_rule target;
2659 if (arg_size != sizeof *fsr) {
2660 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2662 fsr = (struct ofp_flow_stats_request *) osr->body;
2664 COVERAGE_INC(ofproto_flows_req);
2666 cbdata.ofconn = ofconn;
2667 cbdata.out_port = fsr->out_port;
2668 cbdata.msg = start_stats_reply(osr, 1024);
2669 cls_rule_from_match(&target, &fsr->match, 0);
2670 classifier_for_each_match(&p->cls, &target,
2671 table_id_to_include(fsr->table_id),
2672 flow_stats_cb, &cbdata);
2673 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2677 struct flow_stats_ds_cbdata {
2678 struct ofproto *ofproto;
2683 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
2685 struct rule *rule = rule_from_cls_rule(rule_);
2686 struct flow_stats_ds_cbdata *cbdata = cbdata_;
2687 struct ds *results = cbdata->results;
2688 struct ofp_match match;
2689 uint64_t packet_count, byte_count;
2690 size_t act_len = sizeof *rule->actions * rule->n_actions;
2692 /* Don't report on subrules. */
2693 if (rule->super != NULL) {
2697 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2698 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &match);
2700 ds_put_format(results, "duration=%llds, ",
2701 (time_msec() - rule->created) / 1000);
2702 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2703 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2704 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2705 ofp_print_match(results, &match, true);
2706 ofp_print_actions(results, &rule->actions->header, act_len);
2707 ds_put_cstr(results, "\n");
2710 /* Adds a pretty-printed description of all flows to 'results', including
2711 * those marked hidden by secchan (e.g., by in-band control). */
2713 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2715 struct ofp_match match;
2716 struct cls_rule target;
2717 struct flow_stats_ds_cbdata cbdata;
2719 memset(&match, 0, sizeof match);
2720 match.wildcards = htonl(OFPFW_ALL);
2723 cbdata.results = results;
2725 cls_rule_from_match(&target, &match, 0);
2726 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2727 flow_stats_ds_cb, &cbdata);
2730 struct aggregate_stats_cbdata {
2731 struct ofproto *ofproto;
2733 uint64_t packet_count;
2734 uint64_t byte_count;
2739 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2741 struct rule *rule = rule_from_cls_rule(rule_);
2742 struct aggregate_stats_cbdata *cbdata = cbdata_;
2743 uint64_t packet_count, byte_count;
2745 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2749 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2751 cbdata->packet_count += packet_count;
2752 cbdata->byte_count += byte_count;
2757 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2758 const struct ofp_stats_request *osr,
2761 struct ofp_aggregate_stats_request *asr;
2762 struct ofp_aggregate_stats_reply *reply;
2763 struct aggregate_stats_cbdata cbdata;
2764 struct cls_rule target;
2767 if (arg_size != sizeof *asr) {
2768 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2770 asr = (struct ofp_aggregate_stats_request *) osr->body;
2772 COVERAGE_INC(ofproto_agg_request);
2774 cbdata.out_port = asr->out_port;
2775 cbdata.packet_count = 0;
2776 cbdata.byte_count = 0;
2778 cls_rule_from_match(&target, &asr->match, 0);
2779 classifier_for_each_match(&p->cls, &target,
2780 table_id_to_include(asr->table_id),
2781 aggregate_stats_cb, &cbdata);
2783 msg = start_stats_reply(osr, sizeof *reply);
2784 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2785 reply->flow_count = htonl(cbdata.n_flows);
2786 reply->packet_count = htonll(cbdata.packet_count);
2787 reply->byte_count = htonll(cbdata.byte_count);
2788 queue_tx(msg, ofconn, ofconn->reply_counter);
2793 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2794 struct ofp_header *oh)
2796 struct ofp_stats_request *osr;
2800 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2805 osr = (struct ofp_stats_request *) oh;
2807 switch (ntohs(osr->type)) {
2809 return handle_desc_stats_request(p, ofconn, osr);
2812 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2814 case OFPST_AGGREGATE:
2815 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2818 return handle_table_stats_request(p, ofconn, osr);
2821 return handle_port_stats_request(p, ofconn, osr, arg_size);
2824 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2827 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2831 static long long int
2832 msec_from_nsec(uint64_t sec, uint32_t nsec)
2834 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2838 update_time(struct ofproto *ofproto, struct rule *rule,
2839 const struct odp_flow_stats *stats)
2841 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2842 if (used > rule->used) {
2844 if (rule->super && used > rule->super->used) {
2845 rule->super->used = used;
2847 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
2852 update_stats(struct ofproto *ofproto, struct rule *rule,
2853 const struct odp_flow_stats *stats)
2855 if (stats->n_packets) {
2856 update_time(ofproto, rule, stats);
2857 rule->packet_count += stats->n_packets;
2858 rule->byte_count += stats->n_bytes;
2859 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
2865 add_flow(struct ofproto *p, struct ofconn *ofconn,
2866 struct ofp_flow_mod *ofm, size_t n_actions)
2868 struct ofpbuf *packet;
2873 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
2877 flow_from_match(&flow, &wildcards, &ofm->match);
2878 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
2879 ntohs(ofm->priority))) {
2880 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
2884 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
2885 n_actions, ntohs(ofm->idle_timeout),
2886 ntohs(ofm->hard_timeout), ofm->cookie,
2887 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
2888 cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority));
2891 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2892 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2896 in_port = UINT16_MAX;
2899 rule_insert(p, rule, packet, in_port);
2900 ofpbuf_delete(packet);
2905 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
2906 size_t n_actions, uint16_t command, struct rule *rule)
2908 if (rule_is_hidden(rule)) {
2912 if (command == OFPFC_DELETE) {
2913 long long int now = time_msec();
2914 send_flow_removed(p, rule, now, OFPRR_DELETE);
2915 rule_remove(p, rule);
2917 size_t actions_len = n_actions * sizeof *rule->actions;
2919 if (n_actions == rule->n_actions
2920 && !memcmp(ofm->actions, rule->actions, actions_len))
2925 free(rule->actions);
2926 rule->actions = xmemdup(ofm->actions, actions_len);
2927 rule->n_actions = n_actions;
2928 rule->flow_cookie = ofm->cookie;
2930 if (rule->cr.wc.wildcards) {
2931 COVERAGE_INC(ofproto_mod_wc_flow);
2932 p->need_revalidate = true;
2934 rule_update_actions(p, rule);
2942 modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm,
2943 size_t n_actions, uint16_t command)
2949 flow_from_match(&flow, &wildcards, &ofm->match);
2950 rule = rule_from_cls_rule(classifier_find_rule_exactly(
2951 &p->cls, &flow, wildcards,
2952 ntohs(ofm->priority)));
2955 if (command == OFPFC_DELETE
2956 && ofm->out_port != htons(OFPP_NONE)
2957 && !rule_has_out_port(rule, ofm->out_port)) {
2961 modify_flow(p, ofm, n_actions, command, rule);
2966 struct modify_flows_cbdata {
2967 struct ofproto *ofproto;
2968 const struct ofp_flow_mod *ofm;
2975 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
2977 struct rule *rule = rule_from_cls_rule(rule_);
2978 struct modify_flows_cbdata *cbdata = cbdata_;
2980 if (cbdata->out_port != htons(OFPP_NONE)
2981 && !rule_has_out_port(rule, cbdata->out_port)) {
2985 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions,
2986 cbdata->command, rule);
2990 modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm,
2991 size_t n_actions, uint16_t command)
2993 struct modify_flows_cbdata cbdata;
2994 struct cls_rule target;
2998 cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port
2999 : htons(OFPP_NONE));
3000 cbdata.n_actions = n_actions;
3001 cbdata.command = command;
3003 cls_rule_from_match(&target, &ofm->match, 0);
3005 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3006 modify_flows_cb, &cbdata);
3011 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3012 struct ofp_flow_mod *ofm)
3017 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3018 sizeof *ofm->actions, &n_actions);
3023 /* We do not support the emergency flow cache. It will hopefully
3024 * get dropped from OpenFlow in the near future. */
3025 if (ofm->flags & htons(OFPFF_EMERG)) {
3026 /* There isn't a good fit for an error code, so just state that the
3027 * flow table is full. */
3028 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3031 normalize_match(&ofm->match);
3032 if (!ofm->match.wildcards) {
3033 ofm->priority = htons(UINT16_MAX);
3036 error = validate_actions((const union ofp_action *) ofm->actions,
3037 n_actions, p->max_ports);
3042 switch (ntohs(ofm->command)) {
3044 return add_flow(p, ofconn, ofm, n_actions);
3047 return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY);
3049 case OFPFC_MODIFY_STRICT:
3050 return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY);
3053 return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE);
3055 case OFPFC_DELETE_STRICT:
3056 return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE);
3059 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3064 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3066 struct ofp_vendor_header *ovh = msg;
3067 struct nicira_header *nh;
3069 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3070 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3072 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3073 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3075 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3076 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3080 switch (ntohl(nh->subtype)) {
3081 case NXT_STATUS_REQUEST:
3082 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3086 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3090 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
3092 struct ofp_header *ob;
3095 /* Currently, everything executes synchronously, so we can just
3096 * immediately send the barrier reply. */
3097 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3098 queue_tx(buf, ofconn, ofconn->reply_counter);
3103 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3104 struct ofpbuf *ofp_msg)
3106 struct ofp_header *oh = ofp_msg->data;
3109 COVERAGE_INC(ofproto_recv_openflow);
3111 case OFPT_ECHO_REQUEST:
3112 error = handle_echo_request(ofconn, oh);
3115 case OFPT_ECHO_REPLY:
3119 case OFPT_FEATURES_REQUEST:
3120 error = handle_features_request(p, ofconn, oh);
3123 case OFPT_GET_CONFIG_REQUEST:
3124 error = handle_get_config_request(p, ofconn, oh);
3127 case OFPT_SET_CONFIG:
3128 error = handle_set_config(p, ofconn, ofp_msg->data);
3131 case OFPT_PACKET_OUT:
3132 error = handle_packet_out(p, ofconn, ofp_msg->data);
3136 error = handle_port_mod(p, oh);
3140 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3143 case OFPT_STATS_REQUEST:
3144 error = handle_stats_request(p, ofconn, oh);
3148 error = handle_vendor(p, ofconn, ofp_msg->data);
3151 case OFPT_BARRIER_REQUEST:
3152 error = handle_barrier_request(ofconn, oh);
3156 if (VLOG_IS_WARN_ENABLED()) {
3157 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3158 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3161 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3166 send_error_oh(ofconn, ofp_msg->data, error);
3171 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
3173 struct odp_msg *msg = packet->data;
3174 uint16_t in_port = odp_port_to_ofp_port(msg->port);
3176 struct ofpbuf payload;
3179 payload.data = msg + 1;
3180 payload.size = msg->length - sizeof *msg;
3181 flow_extract(&payload, msg->port, &flow);
3183 /* Check with in-band control to see if this packet should be sent
3184 * to the local port regardless of the flow table. */
3185 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3186 union odp_action action;
3188 memset(&action, 0, sizeof(action));
3189 action.output.type = ODPAT_OUTPUT;
3190 action.output.port = ODPP_LOCAL;
3191 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3194 rule = lookup_valid_rule(p, &flow);
3196 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3197 struct ofport *port = port_array_get(&p->ports, msg->port);
3199 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3200 COVERAGE_INC(ofproto_no_packet_in);
3201 /* XXX install 'drop' flow entry */
3202 ofpbuf_delete(packet);
3206 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3209 COVERAGE_INC(ofproto_packet_in);
3210 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3214 if (rule->cr.wc.wildcards) {
3215 rule = rule_create_subrule(p, rule, &flow);
3216 rule_make_actions(p, rule, packet);
3218 if (!rule->may_install) {
3219 /* The rule is not installable, that is, we need to process every
3220 * packet, so process the current packet and set its actions into
3222 rule_make_actions(p, rule, packet);
3224 /* XXX revalidate rule if it needs it */
3228 rule_execute(p, rule, &payload, &flow);
3229 rule_reinstall(p, rule);
3231 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY
3232 && rconn_is_connected(p->controller->rconn)) {
3234 * Extra-special case for fail-open mode.
3236 * We are in fail-open mode and the packet matched the fail-open rule,
3237 * but we are connected to a controller too. We should send the packet
3238 * up to the controller in the hope that it will try to set up a flow
3239 * and thereby allow us to exit fail-open.
3241 * See the top-level comment in fail-open.c for more information.
3243 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3245 ofpbuf_delete(packet);
3250 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3252 struct odp_msg *msg = packet->data;
3254 switch (msg->type) {
3255 case _ODPL_ACTION_NR:
3256 COVERAGE_INC(ofproto_ctlr_action);
3257 pinsched_send(p->action_sched, odp_port_to_ofp_port(msg->port), packet,
3258 send_packet_in_action, p);
3261 case _ODPL_SFLOW_NR:
3263 ofproto_sflow_received(p->sflow, msg);
3265 ofpbuf_delete(packet);
3269 handle_odp_miss_msg(p, packet);
3273 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
3280 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3282 struct rule *sub = rule_from_cls_rule(sub_);
3283 struct revalidate_cbdata *cbdata = cbdata_;
3285 if (cbdata->revalidate_all
3286 || (cbdata->revalidate_subrules && sub->super)
3287 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3288 revalidate_rule(cbdata->ofproto, sub);
3293 revalidate_rule(struct ofproto *p, struct rule *rule)
3295 const flow_t *flow = &rule->cr.flow;
3297 COVERAGE_INC(ofproto_revalidate_rule);
3300 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3302 rule_remove(p, rule);
3304 } else if (super != rule->super) {
3305 COVERAGE_INC(ofproto_revalidate_moved);
3306 list_remove(&rule->list);
3307 list_push_back(&super->list, &rule->list);
3308 rule->super = super;
3309 rule->hard_timeout = super->hard_timeout;
3310 rule->idle_timeout = super->idle_timeout;
3311 rule->created = super->created;
3316 rule_update_actions(p, rule);
3320 static struct ofpbuf *
3321 compose_flow_removed(const struct rule *rule, long long int now, uint8_t reason)
3323 struct ofp_flow_removed *ofr;
3325 long long int tdiff = time_msec() - rule->created;
3326 uint32_t sec = tdiff / 1000;
3327 uint32_t msec = tdiff - (sec * 1000);
3329 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
3330 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofr->match);
3331 ofr->cookie = rule->flow_cookie;
3332 ofr->priority = htons(rule->cr.priority);
3333 ofr->reason = reason;
3334 ofr->duration_sec = htonl(sec);
3335 ofr->duration_nsec = htonl(msec * 1000000);
3336 ofr->idle_timeout = htons(rule->idle_timeout);
3337 ofr->packet_count = htonll(rule->packet_count);
3338 ofr->byte_count = htonll(rule->byte_count);
3344 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3346 assert(rule->installed);
3347 assert(!rule->cr.wc.wildcards);
3350 rule_remove(ofproto, rule);
3352 rule_uninstall(ofproto, rule);
3356 send_flow_removed(struct ofproto *p, struct rule *rule,
3357 long long int now, uint8_t reason)
3359 struct ofconn *ofconn;
3360 struct ofconn *prev;
3361 struct ofpbuf *buf = NULL;
3363 /* We limit the maximum number of queued flow expirations it by accounting
3364 * them under the counter for replies. That works because preventing
3365 * OpenFlow requests from being processed also prevents new flows from
3366 * being added (and expiring). (It also prevents processing OpenFlow
3367 * requests that would not add new flows, so it is imperfect.) */
3370 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3371 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)) {
3373 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3375 buf = compose_flow_removed(rule, now, reason);
3381 queue_tx(buf, prev, prev->reply_counter);
3387 expire_rule(struct cls_rule *cls_rule, void *p_)
3389 struct ofproto *p = p_;
3390 struct rule *rule = rule_from_cls_rule(cls_rule);
3391 long long int hard_expire, idle_expire, expire, now;
3393 hard_expire = (rule->hard_timeout
3394 ? rule->created + rule->hard_timeout * 1000
3396 idle_expire = (rule->idle_timeout
3397 && (rule->super || list_is_empty(&rule->list))
3398 ? rule->used + rule->idle_timeout * 1000
3400 expire = MIN(hard_expire, idle_expire);
3404 if (rule->installed && now >= rule->used + 5000) {
3405 uninstall_idle_flow(p, rule);
3406 } else if (!rule->cr.wc.wildcards) {
3407 active_timeout(p, rule);
3413 COVERAGE_INC(ofproto_expired);
3415 /* Update stats. This code will be a no-op if the rule expired
3416 * due to an idle timeout. */
3417 if (rule->cr.wc.wildcards) {
3418 struct rule *subrule, *next;
3419 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3420 rule_remove(p, subrule);
3423 rule_uninstall(p, rule);
3426 if (!rule_is_hidden(rule)) {
3427 send_flow_removed(p, rule, now,
3429 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
3431 rule_remove(p, rule);
3435 active_timeout(struct ofproto *ofproto, struct rule *rule)
3437 if (ofproto->netflow && !is_controller_rule(rule) &&
3438 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
3439 struct ofexpired expired;
3440 struct odp_flow odp_flow;
3442 /* Get updated flow stats. */
3443 memset(&odp_flow, 0, sizeof odp_flow);
3444 if (rule->installed) {
3445 odp_flow.key = rule->cr.flow;
3446 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
3447 dpif_flow_get(ofproto->dpif, &odp_flow);
3449 if (odp_flow.stats.n_packets) {
3450 update_time(ofproto, rule, &odp_flow.stats);
3451 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
3452 odp_flow.stats.tcp_flags);
3456 expired.flow = rule->cr.flow;
3457 expired.packet_count = rule->packet_count +
3458 odp_flow.stats.n_packets;
3459 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
3460 expired.used = rule->used;
3462 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
3464 /* Schedule us to send the accumulated records once we have
3465 * collected all of them. */
3466 poll_immediate_wake();
3471 update_used(struct ofproto *p)
3473 struct odp_flow *flows;
3478 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
3483 for (i = 0; i < n_flows; i++) {
3484 struct odp_flow *f = &flows[i];
3487 rule = rule_from_cls_rule(
3488 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3489 if (!rule || !rule->installed) {
3490 COVERAGE_INC(ofproto_unexpected_rule);
3491 dpif_flow_del(p->dpif, f);
3495 update_time(p, rule, &f->stats);
3496 rule_account(p, rule, f->stats.n_bytes);
3502 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3503 const struct ofpbuf *packet, int send_len)
3505 struct odp_msg *msg = packet->data;
3506 struct ofpbuf payload;
3510 /* Extract packet payload from 'msg'. */
3511 payload.data = msg + 1;
3512 payload.size = msg->length - sizeof *msg;
3514 /* Construct ofp_packet_in message. */
3515 reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3516 opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port), reason,
3517 &payload, send_len);
3520 rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100);
3524 send_packet_in_action(struct ofpbuf *packet, void *p_)
3526 struct ofproto *p = p_;
3527 struct ofconn *ofconn;
3528 struct odp_msg *msg;
3531 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3532 if (ofconn == p->controller || ofconn->miss_send_len) {
3533 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3536 ofpbuf_delete(packet);
3540 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3542 struct ofproto *p = p_;
3543 bool in_fail_open = p->fail_open && fail_open_is_active(p->fail_open);
3544 struct ofconn *ofconn;
3545 struct ofpbuf payload;
3546 struct odp_msg *msg;
3549 payload.data = msg + 1;
3550 payload.size = msg->length - sizeof *msg;
3551 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3552 if (ofconn->miss_send_len) {
3553 struct pktbuf *pb = ofconn->pktbuf;
3554 uint32_t buffer_id = (in_fail_open
3556 : pktbuf_save(pb, &payload, msg->port));
3557 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3559 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3562 ofpbuf_delete(packet);
3566 pick_datapath_id(const struct ofproto *ofproto)
3568 const struct ofport *port;
3570 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
3572 uint8_t ea[ETH_ADDR_LEN];
3575 error = netdev_get_etheraddr(port->netdev, ea);
3577 return eth_addr_to_uint64(ea);
3579 VLOG_WARN("could not get MAC address for %s (%s)",
3580 netdev_get_name(port->netdev), strerror(error));
3582 return ofproto->fallback_dpid;
3586 pick_fallback_dpid(void)
3588 uint8_t ea[ETH_ADDR_LEN];
3589 eth_addr_nicira_random(ea);
3590 return eth_addr_to_uint64(ea);
3594 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3595 struct odp_actions *actions, tag_type *tags,
3596 uint16_t *nf_output_iface, void *ofproto_)
3598 struct ofproto *ofproto = ofproto_;
3601 /* Drop frames for reserved multicast addresses. */
3602 if (eth_addr_is_reserved(flow->dl_dst)) {
3606 /* Learn source MAC (but don't try to learn from revalidation). */
3607 if (packet != NULL) {
3608 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3611 /* The log messages here could actually be useful in debugging,
3612 * so keep the rate limit relatively high. */
3613 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3614 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3615 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3616 ofproto_revalidate(ofproto, rev_tag);
3620 /* Determine output port. */
3621 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3623 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
3624 } else if (out_port != flow->in_port) {
3625 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3626 *nf_output_iface = out_port;
3634 static const struct ofhooks default_ofhooks = {
3636 default_normal_ofhook_cb,