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 * 'n_actions' is the number of elements in the 'actions' array. A single
113 * action may take up more more than one element's worth of space.
115 * A subrule has no actions (it uses the super-rule's actions). */
117 union ofp_action *actions;
121 * A super-rule with wildcard fields never has ODP actions (since the
122 * datapath only supports exact-match flows). */
123 bool installed; /* Installed in datapath? */
124 bool may_install; /* True ordinarily; false if actions must
125 * be reassessed for every packet. */
127 union odp_action *odp_actions;
131 rule_is_hidden(const struct rule *rule)
133 /* Subrules are merely an implementation detail, so hide them from the
135 if (rule->super != NULL) {
139 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
140 * (e.g. by in-band control) and are intentionally hidden from the
142 if (rule->cr.priority > UINT16_MAX) {
149 static struct rule *rule_create(struct ofproto *, struct rule *super,
150 const union ofp_action *, size_t n_actions,
151 uint16_t idle_timeout, uint16_t hard_timeout,
152 uint64_t flow_cookie, bool send_flow_removed);
153 static void rule_free(struct rule *);
154 static void rule_destroy(struct ofproto *, struct rule *);
155 static struct rule *rule_from_cls_rule(const struct cls_rule *);
156 static void rule_insert(struct ofproto *, struct rule *,
157 struct ofpbuf *packet, uint16_t in_port);
158 static void rule_remove(struct ofproto *, struct rule *);
159 static bool rule_make_actions(struct ofproto *, struct rule *,
160 const struct ofpbuf *packet);
161 static void rule_install(struct ofproto *, struct rule *,
162 struct rule *displaced_rule);
163 static void rule_uninstall(struct ofproto *, struct rule *);
164 static void rule_post_uninstall(struct ofproto *, struct rule *);
165 static void send_flow_removed(struct ofproto *p, struct rule *rule,
166 long long int now, uint8_t reason);
171 struct pktbuf *pktbuf;
174 struct rconn_packet_counter *packet_in_counter;
176 /* Number of OpenFlow messages queued as replies to OpenFlow requests, and
177 * the maximum number before we stop reading OpenFlow requests. */
178 #define OFCONN_REPLY_MAX 100
179 struct rconn_packet_counter *reply_counter;
182 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *);
183 static void ofconn_destroy(struct ofconn *);
184 static void ofconn_run(struct ofconn *, struct ofproto *);
185 static void ofconn_wait(struct ofconn *);
186 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
187 struct rconn_packet_counter *counter);
191 uint64_t datapath_id; /* Datapath ID. */
192 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
193 char *mfr_desc; /* Manufacturer. */
194 char *hw_desc; /* Hardware. */
195 char *sw_desc; /* Software version. */
196 char *serial_desc; /* Serial number. */
197 char *dp_desc; /* Datapath description. */
201 struct netdev_monitor *netdev_monitor;
202 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
204 struct shash port_by_name;
208 struct switch_status *switch_status;
209 struct status_category *ss_cat;
210 struct in_band *in_band;
211 struct discovery *discovery;
212 struct fail_open *fail_open;
213 struct pinsched *miss_sched, *action_sched;
214 struct netflow *netflow;
215 struct ofproto_sflow *sflow;
218 struct classifier cls;
219 bool need_revalidate;
220 long long int next_expiration;
221 struct tag_set revalidate_set;
222 bool tun_id_from_cookie;
224 /* OpenFlow connections. */
225 struct list all_conns;
226 struct ofconn *controller;
227 struct pvconn **listeners;
229 struct pvconn **snoops;
232 /* Hooks for ovs-vswitchd. */
233 const struct ofhooks *ofhooks;
236 /* Used by default ofhooks. */
237 struct mac_learning *ml;
240 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
242 static const struct ofhooks default_ofhooks;
244 static uint64_t pick_datapath_id(const struct ofproto *);
245 static uint64_t pick_fallback_dpid(void);
246 static void send_packet_in_miss(struct ofpbuf *, void *ofproto);
247 static void send_packet_in_action(struct ofpbuf *, void *ofproto);
248 static void update_used(struct ofproto *);
249 static void update_stats(struct ofproto *, struct rule *,
250 const struct odp_flow_stats *);
251 static void expire_rule(struct cls_rule *, void *ofproto);
252 static void active_timeout(struct ofproto *ofproto, struct rule *rule);
253 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
254 static void revalidate_cb(struct cls_rule *rule_, void *p_);
256 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
258 static void handle_openflow(struct ofconn *, struct ofproto *,
261 static void refresh_port_groups(struct ofproto *);
263 static void update_port(struct ofproto *, const char *devname);
264 static int init_ports(struct ofproto *);
265 static void reinit_ports(struct ofproto *);
268 ofproto_create(const char *datapath, const char *datapath_type,
269 const struct ofhooks *ofhooks, void *aux,
270 struct ofproto **ofprotop)
272 struct odp_stats stats;
279 /* Connect to datapath and start listening for messages. */
280 error = dpif_open(datapath, datapath_type, &dpif);
282 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
285 error = dpif_get_dp_stats(dpif, &stats);
287 VLOG_ERR("failed to obtain stats for datapath %s: %s",
288 datapath, strerror(error));
292 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
294 VLOG_ERR("failed to listen on datapath %s: %s",
295 datapath, strerror(error));
299 dpif_flow_flush(dpif);
300 dpif_recv_purge(dpif);
302 /* Initialize settings. */
303 p = xzalloc(sizeof *p);
304 p->fallback_dpid = pick_fallback_dpid();
305 p->datapath_id = p->fallback_dpid;
306 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
307 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
308 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
309 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
310 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
312 /* Initialize datapath. */
314 p->netdev_monitor = netdev_monitor_create();
315 port_array_init(&p->ports);
316 shash_init(&p->port_by_name);
317 p->max_ports = stats.max_ports;
319 /* Initialize submodules. */
320 p->switch_status = switch_status_create(p);
324 p->miss_sched = p->action_sched = NULL;
328 /* Initialize flow table. */
329 classifier_init(&p->cls);
330 p->need_revalidate = false;
331 p->next_expiration = time_msec() + 1000;
332 tag_set_init(&p->revalidate_set);
334 /* Initialize OpenFlow connections. */
335 list_init(&p->all_conns);
336 p->controller = ofconn_create(p, rconn_create(5, 8));
337 p->controller->pktbuf = pktbuf_create();
338 p->controller->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
344 /* Initialize hooks. */
346 p->ofhooks = ofhooks;
350 p->ofhooks = &default_ofhooks;
352 p->ml = mac_learning_create();
355 /* Register switch status category. */
356 p->ss_cat = switch_status_register(p->switch_status, "remote",
357 rconn_status_cb, p->controller->rconn);
359 /* Pick final datapath ID. */
360 p->datapath_id = pick_datapath_id(p);
361 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
368 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
370 uint64_t old_dpid = p->datapath_id;
371 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
372 if (p->datapath_id != old_dpid) {
373 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
374 rconn_reconnect(p->controller->rconn);
379 ofproto_set_controller(struct ofproto *p, const struct ofproto_controller *c)
381 int rate_limit, burst_limit;
388 discovery = !strcmp(c->target, "discover");
389 in_band = discovery || c->band == OFPROTO_IN_BAND;
391 rconn_set_max_backoff(p->controller->rconn, c->max_backoff);
393 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
394 rconn_set_probe_interval(p->controller->rconn, probe_interval);
396 if (discovery != (p->discovery != NULL)) {
397 rconn_disconnect(p->controller->rconn);
399 if (discovery_create(c->accept_re, c->update_resolv_conf,
400 p->dpif, p->switch_status,
405 discovery_destroy(p->discovery);
411 discovery_set_update_resolv_conf(p->discovery,
412 c->update_resolv_conf);
413 discovery_set_accept_controller_re(p->discovery, c->accept_re);
415 if (strcmp(rconn_get_name(p->controller->rconn), c->target)) {
416 rconn_connect(p->controller->rconn, c->target);
420 rconn_disconnect(p->controller->rconn);
424 if (in_band != (p->in_band != NULL)) {
428 error = in_band_create(p, p->dpif, p->switch_status, &p->in_band);
430 in_band_set_remotes(p->in_band, &p->controller->rconn, 1);
433 in_band_destroy(p->in_band);
436 rconn_reconnect(p->controller->rconn);
439 if (c && c->fail == OFPROTO_FAIL_STANDALONE) {
440 struct rconn *rconn = p->controller->rconn;
441 int trigger_duration = rconn_get_probe_interval(rconn) * 3;
443 p->fail_open = fail_open_create(p, trigger_duration,
444 p->switch_status, rconn);
446 fail_open_set_trigger_duration(p->fail_open, trigger_duration);
449 fail_open_destroy(p->fail_open);
453 rate_limit = c ? c->rate_limit : 0;
454 burst_limit = c ? c->burst_limit : 0;
455 if (rate_limit > 0) {
456 if (!p->miss_sched) {
457 p->miss_sched = pinsched_create(rate_limit, burst_limit,
459 p->action_sched = pinsched_create(rate_limit, burst_limit,
462 pinsched_set_limits(p->miss_sched, rate_limit, burst_limit);
463 pinsched_set_limits(p->action_sched, rate_limit, burst_limit);
466 pinsched_destroy(p->miss_sched);
467 p->miss_sched = NULL;
468 pinsched_destroy(p->action_sched);
469 p->action_sched = NULL;
474 ofproto_set_desc(struct ofproto *p,
475 const char *mfr_desc, const char *hw_desc,
476 const char *sw_desc, const char *serial_desc,
479 struct ofp_desc_stats *ods;
482 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
483 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
484 sizeof ods->mfr_desc);
487 p->mfr_desc = xstrdup(mfr_desc);
490 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
491 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
492 sizeof ods->hw_desc);
495 p->hw_desc = xstrdup(hw_desc);
498 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
499 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
500 sizeof ods->sw_desc);
503 p->sw_desc = xstrdup(sw_desc);
506 if (strlen(serial_desc) >= sizeof ods->serial_num) {
507 VLOG_WARN("truncating serial_desc, must be less than %zu "
509 sizeof ods->serial_num);
511 free(p->serial_desc);
512 p->serial_desc = xstrdup(serial_desc);
515 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
516 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
517 sizeof ods->dp_desc);
520 p->dp_desc = xstrdup(dp_desc);
525 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
526 const struct svec *svec)
528 struct pvconn **pvconns = *pvconnsp;
529 size_t n_pvconns = *n_pvconnsp;
533 for (i = 0; i < n_pvconns; i++) {
534 pvconn_close(pvconns[i]);
538 pvconns = xmalloc(svec->n * sizeof *pvconns);
540 for (i = 0; i < svec->n; i++) {
541 const char *name = svec->names[i];
542 struct pvconn *pvconn;
545 error = pvconn_open(name, &pvconn);
547 pvconns[n_pvconns++] = pvconn;
549 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
557 *n_pvconnsp = n_pvconns;
563 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
565 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
569 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
571 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
575 ofproto_set_netflow(struct ofproto *ofproto,
576 const struct netflow_options *nf_options)
578 if (nf_options && nf_options->collectors.n) {
579 if (!ofproto->netflow) {
580 ofproto->netflow = netflow_create();
582 return netflow_set_options(ofproto->netflow, nf_options);
584 netflow_destroy(ofproto->netflow);
585 ofproto->netflow = NULL;
591 ofproto_set_sflow(struct ofproto *ofproto,
592 const struct ofproto_sflow_options *oso)
594 struct ofproto_sflow *os = ofproto->sflow;
597 struct ofport *ofport;
598 unsigned int odp_port;
600 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
601 refresh_port_groups(ofproto);
602 PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) {
603 ofproto_sflow_add_port(os, odp_port,
604 netdev_get_name(ofport->netdev));
607 ofproto_sflow_set_options(os, oso);
609 ofproto_sflow_destroy(os);
610 ofproto->sflow = NULL;
615 ofproto_set_stp(struct ofproto *ofproto OVS_UNUSED, bool enable_stp)
619 VLOG_WARN("STP is not yet implemented");
627 ofproto_get_datapath_id(const struct ofproto *ofproto)
629 return ofproto->datapath_id;
633 ofproto_get_controller(const struct ofproto *p, struct ofproto_controller *c)
635 memset(c, 0, sizeof *c);
637 struct discovery *d = p->discovery;
639 c->target = "discover";
640 c->accept_re = (char *) discovery_get_accept_controller_re(d);
641 c->update_resolv_conf = discovery_get_update_resolv_conf(d);
642 } else if (p->controller) {
643 c->target = (char *) rconn_get_name(p->controller->rconn);
648 c->max_backoff = rconn_get_max_backoff(p->controller->rconn);
649 c->probe_interval = rconn_get_probe_interval(p->controller->rconn);
650 c->fail = p->fail_open ? OFPROTO_FAIL_STANDALONE : OFPROTO_FAIL_SECURE;
651 c->band = p->in_band ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND;
652 pinsched_get_limits(p->miss_sched, &c->rate_limit, &c->burst_limit);
656 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
660 for (i = 0; i < ofproto->n_listeners; i++) {
661 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
666 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
670 for (i = 0; i < ofproto->n_snoops; i++) {
671 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
676 ofproto_destroy(struct ofproto *p)
678 struct ofconn *ofconn, *next_ofconn;
679 struct ofport *ofport;
680 unsigned int port_no;
687 /* Destroy fail-open and in-band early, since they touch the classifier. */
688 fail_open_destroy(p->fail_open);
691 in_band_destroy(p->in_band);
694 ofproto_flush_flows(p);
695 classifier_destroy(&p->cls);
697 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
699 ofconn_destroy(ofconn);
703 netdev_monitor_destroy(p->netdev_monitor);
704 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
707 shash_destroy(&p->port_by_name);
709 switch_status_destroy(p->switch_status);
710 discovery_destroy(p->discovery);
711 pinsched_destroy(p->miss_sched);
712 pinsched_destroy(p->action_sched);
713 netflow_destroy(p->netflow);
714 ofproto_sflow_destroy(p->sflow);
716 switch_status_unregister(p->ss_cat);
718 for (i = 0; i < p->n_listeners; i++) {
719 pvconn_close(p->listeners[i]);
723 for (i = 0; i < p->n_snoops; i++) {
724 pvconn_close(p->snoops[i]);
728 mac_learning_destroy(p->ml);
733 free(p->serial_desc);
736 port_array_destroy(&p->ports);
742 ofproto_run(struct ofproto *p)
744 int error = ofproto_run1(p);
746 error = ofproto_run2(p, false);
752 process_port_change(struct ofproto *ofproto, int error, char *devname)
754 if (error == ENOBUFS) {
755 reinit_ports(ofproto);
757 update_port(ofproto, devname);
763 ofproto_run1(struct ofproto *p)
765 struct ofconn *ofconn, *next_ofconn;
770 if (shash_is_empty(&p->port_by_name)) {
774 for (i = 0; i < 50; i++) {
778 error = dpif_recv(p->dpif, &buf);
780 if (error == ENODEV) {
781 /* Someone destroyed the datapath behind our back. The caller
782 * better destroy us and give up, because we're just going to
783 * spin from here on out. */
784 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
785 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
792 handle_odp_msg(p, buf);
795 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
796 process_port_change(p, error, devname);
798 while ((error = netdev_monitor_poll(p->netdev_monitor,
799 &devname)) != EAGAIN) {
800 process_port_change(p, error, devname);
804 in_band_run(p->in_band);
807 char *controller_name;
808 if (rconn_is_connectivity_questionable(p->controller->rconn)) {
809 discovery_question_connectivity(p->discovery);
811 if (discovery_run(p->discovery, &controller_name)) {
812 if (controller_name) {
813 rconn_connect(p->controller->rconn, controller_name);
815 rconn_disconnect(p->controller->rconn);
819 pinsched_run(p->miss_sched, send_packet_in_miss, p);
820 pinsched_run(p->action_sched, send_packet_in_action, p);
822 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
824 ofconn_run(ofconn, p);
827 /* Fail-open maintenance. Do this after processing the ofconns since
828 * fail-open checks the status of the controller rconn. */
830 fail_open_run(p->fail_open);
833 for (i = 0; i < p->n_listeners; i++) {
837 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
839 ofconn_create(p, rconn_new_from_vconn("passive", vconn));
840 } else if (retval != EAGAIN) {
841 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
845 for (i = 0; i < p->n_snoops; i++) {
849 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
851 rconn_add_monitor(p->controller->rconn, vconn);
852 } else if (retval != EAGAIN) {
853 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
857 if (time_msec() >= p->next_expiration) {
858 COVERAGE_INC(ofproto_expiration);
859 p->next_expiration = time_msec() + 1000;
862 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
864 /* Let the hook know that we're at a stable point: all outstanding data
865 * in existing flows has been accounted to the account_cb. Thus, the
866 * hook can now reasonably do operations that depend on having accurate
867 * flow volume accounting (currently, that's just bond rebalancing). */
868 if (p->ofhooks->account_checkpoint_cb) {
869 p->ofhooks->account_checkpoint_cb(p->aux);
874 netflow_run(p->netflow);
877 ofproto_sflow_run(p->sflow);
883 struct revalidate_cbdata {
884 struct ofproto *ofproto;
885 bool revalidate_all; /* Revalidate all exact-match rules? */
886 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
887 struct tag_set revalidate_set; /* Set of tags to revalidate. */
891 ofproto_run2(struct ofproto *p, bool revalidate_all)
893 if (p->need_revalidate || revalidate_all
894 || !tag_set_is_empty(&p->revalidate_set)) {
895 struct revalidate_cbdata cbdata;
897 cbdata.revalidate_all = revalidate_all;
898 cbdata.revalidate_subrules = p->need_revalidate;
899 cbdata.revalidate_set = p->revalidate_set;
900 tag_set_init(&p->revalidate_set);
901 COVERAGE_INC(ofproto_revalidate);
902 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
903 p->need_revalidate = false;
910 ofproto_wait(struct ofproto *p)
912 struct ofconn *ofconn;
915 dpif_recv_wait(p->dpif);
916 dpif_port_poll_wait(p->dpif);
917 netdev_monitor_poll_wait(p->netdev_monitor);
918 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
922 in_band_wait(p->in_band);
925 discovery_wait(p->discovery);
928 fail_open_wait(p->fail_open);
930 pinsched_wait(p->miss_sched);
931 pinsched_wait(p->action_sched);
933 ofproto_sflow_wait(p->sflow);
935 if (!tag_set_is_empty(&p->revalidate_set)) {
936 poll_immediate_wake();
938 if (p->need_revalidate) {
939 /* Shouldn't happen, but if it does just go around again. */
940 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
941 poll_immediate_wake();
942 } else if (p->next_expiration != LLONG_MAX) {
943 poll_timer_wait(p->next_expiration - time_msec());
945 for (i = 0; i < p->n_listeners; i++) {
946 pvconn_wait(p->listeners[i]);
948 for (i = 0; i < p->n_snoops; i++) {
949 pvconn_wait(p->snoops[i]);
954 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
956 tag_set_add(&ofproto->revalidate_set, tag);
960 ofproto_get_revalidate_set(struct ofproto *ofproto)
962 return &ofproto->revalidate_set;
966 ofproto_is_alive(const struct ofproto *p)
968 return p->discovery || rconn_is_alive(p->controller->rconn);
972 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
973 const union ofp_action *actions, size_t n_actions,
974 const struct ofpbuf *packet)
976 struct odp_actions odp_actions;
979 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
985 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
987 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
988 odp_actions.n_actions, packet);
993 ofproto_add_flow(struct ofproto *p,
994 const flow_t *flow, uint32_t wildcards, unsigned int priority,
995 const union ofp_action *actions, size_t n_actions,
999 rule = rule_create(p, NULL, actions, n_actions,
1000 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1002 cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
1003 rule_insert(p, rule, NULL, 0);
1007 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1008 uint32_t wildcards, unsigned int priority)
1012 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1016 rule_remove(ofproto, rule);
1021 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1023 struct rule *rule = rule_from_cls_rule(rule_);
1024 struct ofproto *ofproto = ofproto_;
1026 /* Mark the flow as not installed, even though it might really be
1027 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1028 * There is no point in uninstalling it individually since we are about to
1029 * blow away all the flows with dpif_flow_flush(). */
1030 rule->installed = false;
1032 rule_remove(ofproto, rule);
1036 ofproto_flush_flows(struct ofproto *ofproto)
1038 COVERAGE_INC(ofproto_flush);
1039 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1040 dpif_flow_flush(ofproto->dpif);
1041 if (ofproto->in_band) {
1042 in_band_flushed(ofproto->in_band);
1044 if (ofproto->fail_open) {
1045 fail_open_flushed(ofproto->fail_open);
1050 reinit_ports(struct ofproto *p)
1052 struct svec devnames;
1053 struct ofport *ofport;
1054 unsigned int port_no;
1055 struct odp_port *odp_ports;
1059 svec_init(&devnames);
1060 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1061 svec_add (&devnames, (char *) ofport->opp.name);
1063 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1064 for (i = 0; i < n_odp_ports; i++) {
1065 svec_add (&devnames, odp_ports[i].devname);
1069 svec_sort_unique(&devnames);
1070 for (i = 0; i < devnames.n; i++) {
1071 update_port(p, devnames.names[i]);
1073 svec_destroy(&devnames);
1077 refresh_port_group(struct ofproto *p, unsigned int group)
1081 struct ofport *port;
1082 unsigned int port_no;
1084 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1086 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1088 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1089 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1090 ports[n_ports++] = port_no;
1093 dpif_port_group_set(p->dpif, group, ports, n_ports);
1100 refresh_port_groups(struct ofproto *p)
1102 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1103 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1105 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1109 static struct ofport *
1110 make_ofport(const struct odp_port *odp_port)
1112 struct netdev_options netdev_options;
1113 enum netdev_flags flags;
1114 struct ofport *ofport;
1115 struct netdev *netdev;
1119 memset(&netdev_options, 0, sizeof netdev_options);
1120 netdev_options.name = odp_port->devname;
1121 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1122 netdev_options.may_open = true;
1124 error = netdev_open(&netdev_options, &netdev);
1126 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1127 "cannot be opened (%s)",
1128 odp_port->devname, odp_port->port,
1129 odp_port->devname, strerror(error));
1133 ofport = xmalloc(sizeof *ofport);
1134 ofport->netdev = netdev;
1135 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1136 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1137 memcpy(ofport->opp.name, odp_port->devname,
1138 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1139 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1141 netdev_get_flags(netdev, &flags);
1142 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1144 netdev_get_carrier(netdev, &carrier);
1145 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1147 netdev_get_features(netdev,
1148 &ofport->opp.curr, &ofport->opp.advertised,
1149 &ofport->opp.supported, &ofport->opp.peer);
1154 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1156 if (port_array_get(&p->ports, odp_port->port)) {
1157 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1160 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1161 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1170 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1172 const struct ofp_phy_port *a = &a_->opp;
1173 const struct ofp_phy_port *b = &b_->opp;
1175 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1176 return (a->port_no == b->port_no
1177 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1178 && !strcmp((char *) a->name, (char *) b->name)
1179 && a->state == b->state
1180 && a->config == b->config
1181 && a->curr == b->curr
1182 && a->advertised == b->advertised
1183 && a->supported == b->supported
1184 && a->peer == b->peer);
1188 send_port_status(struct ofproto *p, const struct ofport *ofport,
1191 /* XXX Should limit the number of queued port status change messages. */
1192 struct ofconn *ofconn;
1193 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1194 struct ofp_port_status *ops;
1197 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1198 ops->reason = reason;
1199 ops->desc = ofport->opp;
1200 hton_ofp_phy_port(&ops->desc);
1201 queue_tx(b, ofconn, NULL);
1203 if (p->ofhooks->port_changed_cb) {
1204 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1209 ofport_install(struct ofproto *p, struct ofport *ofport)
1211 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1212 const char *netdev_name = (const char *) ofport->opp.name;
1214 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1215 port_array_set(&p->ports, odp_port, ofport);
1216 shash_add(&p->port_by_name, netdev_name, ofport);
1218 ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
1223 ofport_remove(struct ofproto *p, struct ofport *ofport)
1225 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1227 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1228 port_array_set(&p->ports, odp_port, NULL);
1229 shash_delete(&p->port_by_name,
1230 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1232 ofproto_sflow_del_port(p->sflow, odp_port);
1237 ofport_free(struct ofport *ofport)
1240 netdev_close(ofport->netdev);
1246 update_port(struct ofproto *p, const char *devname)
1248 struct odp_port odp_port;
1249 struct ofport *old_ofport;
1250 struct ofport *new_ofport;
1253 COVERAGE_INC(ofproto_update_port);
1255 /* Query the datapath for port information. */
1256 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1258 /* Find the old ofport. */
1259 old_ofport = shash_find_data(&p->port_by_name, devname);
1262 /* There's no port named 'devname' but there might be a port with
1263 * the same port number. This could happen if a port is deleted
1264 * and then a new one added in its place very quickly, or if a port
1265 * is renamed. In the former case we want to send an OFPPR_DELETE
1266 * and an OFPPR_ADD, and in the latter case we want to send a
1267 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1268 * the old port's ifindex against the new port, or perhaps less
1269 * reliably but more portably by comparing the old port's MAC
1270 * against the new port's MAC. However, this code isn't that smart
1271 * and always sends an OFPPR_MODIFY (XXX). */
1272 old_ofport = port_array_get(&p->ports, odp_port.port);
1274 } else if (error != ENOENT && error != ENODEV) {
1275 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1276 "%s", strerror(error));
1280 /* Create a new ofport. */
1281 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1283 /* Eliminate a few pathological cases. */
1284 if (!old_ofport && !new_ofport) {
1286 } else if (old_ofport && new_ofport) {
1287 /* Most of the 'config' bits are OpenFlow soft state, but
1288 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1289 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1290 * leaves the other bits 0.) */
1291 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1293 if (ofport_equal(old_ofport, new_ofport)) {
1294 /* False alarm--no change. */
1295 ofport_free(new_ofport);
1300 /* Now deal with the normal cases. */
1302 ofport_remove(p, old_ofport);
1305 ofport_install(p, new_ofport);
1307 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1308 (!old_ofport ? OFPPR_ADD
1309 : !new_ofport ? OFPPR_DELETE
1311 ofport_free(old_ofport);
1313 /* Update port groups. */
1314 refresh_port_groups(p);
1318 init_ports(struct ofproto *p)
1320 struct odp_port *ports;
1325 error = dpif_port_list(p->dpif, &ports, &n_ports);
1330 for (i = 0; i < n_ports; i++) {
1331 const struct odp_port *odp_port = &ports[i];
1332 if (!ofport_conflicts(p, odp_port)) {
1333 struct ofport *ofport = make_ofport(odp_port);
1335 ofport_install(p, ofport);
1340 refresh_port_groups(p);
1344 static struct ofconn *
1345 ofconn_create(struct ofproto *p, struct rconn *rconn)
1347 struct ofconn *ofconn = xmalloc(sizeof *ofconn);
1348 list_push_back(&p->all_conns, &ofconn->node);
1349 ofconn->rconn = rconn;
1350 ofconn->pktbuf = NULL;
1351 ofconn->miss_send_len = 0;
1352 ofconn->packet_in_counter = rconn_packet_counter_create ();
1353 ofconn->reply_counter = rconn_packet_counter_create ();
1358 ofconn_destroy(struct ofconn *ofconn)
1360 list_remove(&ofconn->node);
1361 rconn_destroy(ofconn->rconn);
1362 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1363 rconn_packet_counter_destroy(ofconn->reply_counter);
1364 pktbuf_destroy(ofconn->pktbuf);
1369 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1373 rconn_run(ofconn->rconn);
1375 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1376 /* Limit the number of iterations to prevent other tasks from
1378 for (iteration = 0; iteration < 50; iteration++) {
1379 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1384 fail_open_maybe_recover(p->fail_open);
1386 handle_openflow(ofconn, p, of_msg);
1387 ofpbuf_delete(of_msg);
1391 if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
1392 ofconn_destroy(ofconn);
1397 ofconn_wait(struct ofconn *ofconn)
1399 rconn_run_wait(ofconn->rconn);
1400 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1401 rconn_recv_wait(ofconn->rconn);
1403 COVERAGE_INC(ofproto_ofconn_stuck);
1407 /* Caller is responsible for initializing the 'cr' member of the returned
1409 static struct rule *
1410 rule_create(struct ofproto *ofproto, struct rule *super,
1411 const union ofp_action *actions, size_t n_actions,
1412 uint16_t idle_timeout, uint16_t hard_timeout,
1413 uint64_t flow_cookie, bool send_flow_removed)
1415 struct rule *rule = xzalloc(sizeof *rule);
1416 rule->idle_timeout = idle_timeout;
1417 rule->hard_timeout = hard_timeout;
1418 rule->flow_cookie = flow_cookie;
1419 rule->used = rule->created = time_msec();
1420 rule->send_flow_removed = send_flow_removed;
1421 rule->super = super;
1423 list_push_back(&super->list, &rule->list);
1425 list_init(&rule->list);
1427 rule->n_actions = n_actions;
1428 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1429 netflow_flow_clear(&rule->nf_flow);
1430 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1435 static struct rule *
1436 rule_from_cls_rule(const struct cls_rule *cls_rule)
1438 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1442 rule_free(struct rule *rule)
1444 free(rule->actions);
1445 free(rule->odp_actions);
1449 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1450 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1451 * through all of its subrules and revalidates them, destroying any that no
1452 * longer has a super-rule (which is probably all of them).
1454 * Before calling this function, the caller must make have removed 'rule' from
1455 * the classifier. If 'rule' is an exact-match rule, the caller is also
1456 * responsible for ensuring that it has been uninstalled from the datapath. */
1458 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1461 struct rule *subrule, *next;
1462 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1463 revalidate_rule(ofproto, subrule);
1466 list_remove(&rule->list);
1472 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1474 const union ofp_action *oa;
1475 struct actions_iterator i;
1477 if (out_port == htons(OFPP_NONE)) {
1480 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1481 oa = actions_next(&i)) {
1482 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1489 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1490 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1492 * The flow that 'packet' actually contains does not need to actually match
1493 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1494 * the packet and byte counters for 'rule' will be credited for the packet sent
1495 * out whether or not the packet actually matches 'rule'.
1497 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1498 * the caller must already have accurately composed ODP actions for it given
1499 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1500 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1501 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1502 * actions and apply them to 'packet'. */
1504 rule_execute(struct ofproto *ofproto, struct rule *rule,
1505 struct ofpbuf *packet, const flow_t *flow)
1507 const union odp_action *actions;
1509 struct odp_actions a;
1511 /* Grab or compose the ODP actions.
1513 * The special case for an exact-match 'rule' where 'flow' is not the
1514 * rule's flow is important to avoid, e.g., sending a packet out its input
1515 * port simply because the ODP actions were composed for the wrong
1517 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1518 struct rule *super = rule->super ? rule->super : rule;
1519 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1520 packet, &a, NULL, 0, NULL)) {
1523 actions = a.actions;
1524 n_actions = a.n_actions;
1526 actions = rule->odp_actions;
1527 n_actions = rule->n_odp_actions;
1530 /* Execute the ODP actions. */
1531 if (!dpif_execute(ofproto->dpif, flow->in_port,
1532 actions, n_actions, packet)) {
1533 struct odp_flow_stats stats;
1534 flow_extract_stats(flow, packet, &stats);
1535 update_stats(ofproto, rule, &stats);
1536 rule->used = time_msec();
1537 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1542 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1545 struct rule *displaced_rule;
1547 /* Insert the rule in the classifier. */
1548 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1549 if (!rule->cr.wc.wildcards) {
1550 rule_make_actions(p, rule, packet);
1553 /* Send the packet and credit it to the rule. */
1556 flow_extract(packet, 0, in_port, &flow);
1557 rule_execute(p, rule, packet, &flow);
1560 /* Install the rule in the datapath only after sending the packet, to
1561 * avoid packet reordering. */
1562 if (rule->cr.wc.wildcards) {
1563 COVERAGE_INC(ofproto_add_wc_flow);
1564 p->need_revalidate = true;
1566 rule_install(p, rule, displaced_rule);
1569 /* Free the rule that was displaced, if any. */
1570 if (displaced_rule) {
1571 rule_destroy(p, displaced_rule);
1575 static struct rule *
1576 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1579 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1580 rule->idle_timeout, rule->hard_timeout,
1582 COVERAGE_INC(ofproto_subrule_create);
1583 cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1584 : rule->cr.priority), &subrule->cr);
1585 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1591 rule_remove(struct ofproto *ofproto, struct rule *rule)
1593 if (rule->cr.wc.wildcards) {
1594 COVERAGE_INC(ofproto_del_wc_flow);
1595 ofproto->need_revalidate = true;
1597 rule_uninstall(ofproto, rule);
1599 classifier_remove(&ofproto->cls, &rule->cr);
1600 rule_destroy(ofproto, rule);
1603 /* Returns true if the actions changed, false otherwise. */
1605 rule_make_actions(struct ofproto *p, struct rule *rule,
1606 const struct ofpbuf *packet)
1608 const struct rule *super;
1609 struct odp_actions a;
1612 assert(!rule->cr.wc.wildcards);
1614 super = rule->super ? rule->super : rule;
1616 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1617 packet, &a, &rule->tags, &rule->may_install,
1618 &rule->nf_flow.output_iface);
1620 actions_len = a.n_actions * sizeof *a.actions;
1621 if (rule->n_odp_actions != a.n_actions
1622 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1623 COVERAGE_INC(ofproto_odp_unchanged);
1624 free(rule->odp_actions);
1625 rule->n_odp_actions = a.n_actions;
1626 rule->odp_actions = xmemdup(a.actions, actions_len);
1634 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1635 struct odp_flow_put *put)
1637 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1638 put->flow.key = rule->cr.flow;
1639 put->flow.actions = rule->odp_actions;
1640 put->flow.n_actions = rule->n_odp_actions;
1641 put->flow.flags = 0;
1643 return dpif_flow_put(ofproto->dpif, put);
1647 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1649 assert(!rule->cr.wc.wildcards);
1651 if (rule->may_install) {
1652 struct odp_flow_put put;
1653 if (!do_put_flow(p, rule,
1654 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1656 rule->installed = true;
1657 if (displaced_rule) {
1658 update_stats(p, displaced_rule, &put.flow.stats);
1659 rule_post_uninstall(p, displaced_rule);
1662 } else if (displaced_rule) {
1663 rule_uninstall(p, displaced_rule);
1668 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1670 if (rule->installed) {
1671 struct odp_flow_put put;
1672 COVERAGE_INC(ofproto_dp_missed);
1673 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1675 rule_install(ofproto, rule, NULL);
1680 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1682 bool actions_changed;
1683 uint16_t new_out_iface, old_out_iface;
1685 old_out_iface = rule->nf_flow.output_iface;
1686 actions_changed = rule_make_actions(ofproto, rule, NULL);
1688 if (rule->may_install) {
1689 if (rule->installed) {
1690 if (actions_changed) {
1691 struct odp_flow_put put;
1692 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
1693 | ODPPF_ZERO_STATS, &put);
1694 update_stats(ofproto, rule, &put.flow.stats);
1696 /* Temporarily set the old output iface so that NetFlow
1697 * messages have the correct output interface for the old
1699 new_out_iface = rule->nf_flow.output_iface;
1700 rule->nf_flow.output_iface = old_out_iface;
1701 rule_post_uninstall(ofproto, rule);
1702 rule->nf_flow.output_iface = new_out_iface;
1705 rule_install(ofproto, rule, NULL);
1708 rule_uninstall(ofproto, rule);
1713 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1715 uint64_t total_bytes = rule->byte_count + extra_bytes;
1717 if (ofproto->ofhooks->account_flow_cb
1718 && total_bytes > rule->accounted_bytes)
1720 ofproto->ofhooks->account_flow_cb(
1721 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1722 total_bytes - rule->accounted_bytes, ofproto->aux);
1723 rule->accounted_bytes = total_bytes;
1728 rule_uninstall(struct ofproto *p, struct rule *rule)
1730 assert(!rule->cr.wc.wildcards);
1731 if (rule->installed) {
1732 struct odp_flow odp_flow;
1734 odp_flow.key = rule->cr.flow;
1735 odp_flow.actions = NULL;
1736 odp_flow.n_actions = 0;
1738 if (!dpif_flow_del(p->dpif, &odp_flow)) {
1739 update_stats(p, rule, &odp_flow.stats);
1741 rule->installed = false;
1743 rule_post_uninstall(p, rule);
1748 is_controller_rule(struct rule *rule)
1750 /* If the only action is send to the controller then don't report
1751 * NetFlow expiration messages since it is just part of the control
1752 * logic for the network and not real traffic. */
1754 if (rule && rule->super) {
1755 struct rule *super = rule->super;
1757 return super->n_actions == 1 &&
1758 super->actions[0].type == htons(OFPAT_OUTPUT) &&
1759 super->actions[0].output.port == htons(OFPP_CONTROLLER);
1766 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1768 struct rule *super = rule->super;
1770 rule_account(ofproto, rule, 0);
1772 if (ofproto->netflow && !is_controller_rule(rule)) {
1773 struct ofexpired expired;
1774 expired.flow = rule->cr.flow;
1775 expired.packet_count = rule->packet_count;
1776 expired.byte_count = rule->byte_count;
1777 expired.used = rule->used;
1778 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
1781 super->packet_count += rule->packet_count;
1782 super->byte_count += rule->byte_count;
1784 /* Reset counters to prevent double counting if the rule ever gets
1786 rule->packet_count = 0;
1787 rule->byte_count = 0;
1788 rule->accounted_bytes = 0;
1790 netflow_flow_clear(&rule->nf_flow);
1795 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1796 struct rconn_packet_counter *counter)
1798 update_openflow_length(msg);
1799 if (rconn_send(ofconn->rconn, msg, counter)) {
1805 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1806 int error, const void *data, size_t len)
1809 struct ofp_error_msg *oem;
1811 if (!(error >> 16)) {
1812 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1817 COVERAGE_INC(ofproto_error);
1818 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1819 oh ? oh->xid : 0, &buf);
1820 oem->type = htons((unsigned int) error >> 16);
1821 oem->code = htons(error & 0xffff);
1822 memcpy(oem->data, data, len);
1823 queue_tx(buf, ofconn, ofconn->reply_counter);
1827 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1830 size_t oh_length = ntohs(oh->length);
1831 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1835 hton_ofp_phy_port(struct ofp_phy_port *opp)
1837 opp->port_no = htons(opp->port_no);
1838 opp->config = htonl(opp->config);
1839 opp->state = htonl(opp->state);
1840 opp->curr = htonl(opp->curr);
1841 opp->advertised = htonl(opp->advertised);
1842 opp->supported = htonl(opp->supported);
1843 opp->peer = htonl(opp->peer);
1847 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
1849 struct ofp_header *rq = oh;
1850 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
1855 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
1856 struct ofp_header *oh)
1858 struct ofp_switch_features *osf;
1860 unsigned int port_no;
1861 struct ofport *port;
1863 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1864 osf->datapath_id = htonll(p->datapath_id);
1865 osf->n_buffers = htonl(pktbuf_capacity());
1867 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1868 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
1869 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1870 (1u << OFPAT_SET_VLAN_VID) |
1871 (1u << OFPAT_SET_VLAN_PCP) |
1872 (1u << OFPAT_STRIP_VLAN) |
1873 (1u << OFPAT_SET_DL_SRC) |
1874 (1u << OFPAT_SET_DL_DST) |
1875 (1u << OFPAT_SET_NW_SRC) |
1876 (1u << OFPAT_SET_NW_DST) |
1877 (1u << OFPAT_SET_NW_TOS) |
1878 (1u << OFPAT_SET_TP_SRC) |
1879 (1u << OFPAT_SET_TP_DST));
1881 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1882 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1885 queue_tx(buf, ofconn, ofconn->reply_counter);
1890 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1891 struct ofp_header *oh)
1894 struct ofp_switch_config *osc;
1898 /* Figure out flags. */
1899 dpif_get_drop_frags(p->dpif, &drop_frags);
1900 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1903 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1904 osc->flags = htons(flags);
1905 osc->miss_send_len = htons(ofconn->miss_send_len);
1906 queue_tx(buf, ofconn, ofconn->reply_counter);
1912 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1913 struct ofp_switch_config *osc)
1918 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1922 flags = ntohs(osc->flags);
1924 if (ofconn == p->controller) {
1925 switch (flags & OFPC_FRAG_MASK) {
1926 case OFPC_FRAG_NORMAL:
1927 dpif_set_drop_frags(p->dpif, false);
1929 case OFPC_FRAG_DROP:
1930 dpif_set_drop_frags(p->dpif, true);
1933 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1939 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1940 if (ntohs(osc->miss_send_len) != 0) {
1941 ofconn->pktbuf = pktbuf_create();
1943 pktbuf_destroy(ofconn->pktbuf);
1947 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1953 add_output_group_action(struct odp_actions *actions, uint16_t group,
1954 uint16_t *nf_output_iface)
1956 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1958 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
1959 *nf_output_iface = NF_OUT_FLOOD;
1964 add_controller_action(struct odp_actions *actions,
1965 const struct ofp_action_output *oao)
1967 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
1968 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
1971 struct action_xlate_ctx {
1973 flow_t flow; /* Flow to which these actions correspond. */
1974 int recurse; /* Recursion level, via xlate_table_action. */
1975 struct ofproto *ofproto;
1976 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
1977 * null pointer if we are revalidating
1978 * without a packet to refer to. */
1981 struct odp_actions *out; /* Datapath actions. */
1982 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
1983 bool may_set_up_flow; /* True ordinarily; false if the actions must
1984 * be reassessed for every packet. */
1985 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
1988 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
1989 struct action_xlate_ctx *ctx);
1992 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
1994 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
1997 if (ofport->opp.config & OFPPC_NO_FWD) {
1998 /* Forwarding disabled on port. */
2003 * We don't have an ofport record for this port, but it doesn't hurt to
2004 * allow forwarding to it anyhow. Maybe such a port will appear later
2005 * and we're pre-populating the flow table.
2009 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2010 ctx->nf_output_iface = port;
2013 static struct rule *
2014 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2017 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2019 /* The rule we found might not be valid, since we could be in need of
2020 * revalidation. If it is not valid, don't return it. */
2023 && ofproto->need_revalidate
2024 && !revalidate_rule(ofproto, rule)) {
2025 COVERAGE_INC(ofproto_invalidated);
2033 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2035 if (!ctx->recurse) {
2036 uint16_t old_in_port;
2039 /* Look up a flow with 'in_port' as the input port. Then restore the
2040 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2041 * have surprising behavior). */
2042 old_in_port = ctx->flow.in_port;
2043 ctx->flow.in_port = in_port;
2044 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2045 ctx->flow.in_port = old_in_port;
2053 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2060 xlate_output_action(struct action_xlate_ctx *ctx,
2061 const struct ofp_action_output *oao)
2064 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2066 ctx->nf_output_iface = NF_OUT_DROP;
2068 switch (ntohs(oao->port)) {
2070 add_output_action(ctx, ctx->flow.in_port);
2073 xlate_table_action(ctx, ctx->flow.in_port);
2076 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2077 ctx->out, ctx->tags,
2078 &ctx->nf_output_iface,
2079 ctx->ofproto->aux)) {
2080 COVERAGE_INC(ofproto_uninstallable);
2081 ctx->may_set_up_flow = false;
2085 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2086 &ctx->nf_output_iface);
2089 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2091 case OFPP_CONTROLLER:
2092 add_controller_action(ctx->out, oao);
2095 add_output_action(ctx, ODPP_LOCAL);
2098 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2099 if (odp_port != ctx->flow.in_port) {
2100 add_output_action(ctx, odp_port);
2105 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2106 ctx->nf_output_iface = NF_OUT_FLOOD;
2107 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2108 ctx->nf_output_iface = prev_nf_output_iface;
2109 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2110 ctx->nf_output_iface != NF_OUT_FLOOD) {
2111 ctx->nf_output_iface = NF_OUT_MULTI;
2116 xlate_nicira_action(struct action_xlate_ctx *ctx,
2117 const struct nx_action_header *nah)
2119 const struct nx_action_resubmit *nar;
2120 const struct nx_action_set_tunnel *nast;
2121 union odp_action *oa;
2122 int subtype = ntohs(nah->subtype);
2124 assert(nah->vendor == htonl(NX_VENDOR_ID));
2126 case NXAST_RESUBMIT:
2127 nar = (const struct nx_action_resubmit *) nah;
2128 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2131 case NXAST_SET_TUNNEL:
2132 nast = (const struct nx_action_set_tunnel *) nah;
2133 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2134 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2137 /* If you add a new action here that modifies flow data, don't forget to
2138 * update the flow key in ctx->flow in the same key. */
2141 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2147 do_xlate_actions(const union ofp_action *in, size_t n_in,
2148 struct action_xlate_ctx *ctx)
2150 struct actions_iterator iter;
2151 const union ofp_action *ia;
2152 const struct ofport *port;
2154 port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port);
2155 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2156 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, stp_eth_addr)
2157 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2158 /* Drop this flow. */
2162 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2163 uint16_t type = ntohs(ia->type);
2164 union odp_action *oa;
2168 xlate_output_action(ctx, &ia->output);
2171 case OFPAT_SET_VLAN_VID:
2172 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2173 ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2176 case OFPAT_SET_VLAN_PCP:
2177 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2178 ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2181 case OFPAT_STRIP_VLAN:
2182 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2183 ctx->flow.dl_vlan = OFP_VLAN_NONE;
2184 ctx->flow.dl_vlan_pcp = 0;
2187 case OFPAT_SET_DL_SRC:
2188 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2189 memcpy(oa->dl_addr.dl_addr,
2190 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2191 memcpy(ctx->flow.dl_src,
2192 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2195 case OFPAT_SET_DL_DST:
2196 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2197 memcpy(oa->dl_addr.dl_addr,
2198 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2199 memcpy(ctx->flow.dl_dst,
2200 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2203 case OFPAT_SET_NW_SRC:
2204 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2205 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2208 case OFPAT_SET_NW_DST:
2209 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2210 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2213 case OFPAT_SET_NW_TOS:
2214 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2215 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2218 case OFPAT_SET_TP_SRC:
2219 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2220 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2223 case OFPAT_SET_TP_DST:
2224 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2225 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2229 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2233 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2240 xlate_actions(const union ofp_action *in, size_t n_in,
2241 const flow_t *flow, struct ofproto *ofproto,
2242 const struct ofpbuf *packet,
2243 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2244 uint16_t *nf_output_iface)
2246 tag_type no_tags = 0;
2247 struct action_xlate_ctx ctx;
2248 COVERAGE_INC(ofproto_ofp2odp);
2249 odp_actions_init(out);
2252 ctx.ofproto = ofproto;
2253 ctx.packet = packet;
2255 ctx.tags = tags ? tags : &no_tags;
2256 ctx.may_set_up_flow = true;
2257 ctx.nf_output_iface = NF_OUT_DROP;
2258 do_xlate_actions(in, n_in, &ctx);
2260 /* Check with in-band control to see if we're allowed to set up this
2262 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2263 ctx.may_set_up_flow = false;
2266 if (may_set_up_flow) {
2267 *may_set_up_flow = ctx.may_set_up_flow;
2269 if (nf_output_iface) {
2270 *nf_output_iface = ctx.nf_output_iface;
2272 if (odp_actions_overflow(out)) {
2273 odp_actions_init(out);
2274 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2280 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2281 struct ofp_header *oh)
2283 struct ofp_packet_out *opo;
2284 struct ofpbuf payload, *buffer;
2285 struct odp_actions actions;
2291 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2295 opo = (struct ofp_packet_out *) oh;
2297 COVERAGE_INC(ofproto_packet_out);
2298 if (opo->buffer_id != htonl(UINT32_MAX)) {
2299 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2301 if (error || !buffer) {
2309 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2310 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2311 &flow, p, &payload, &actions, NULL, NULL, NULL);
2316 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2318 ofpbuf_delete(buffer);
2324 update_port_config(struct ofproto *p, struct ofport *port,
2325 uint32_t config, uint32_t mask)
2327 mask &= config ^ port->opp.config;
2328 if (mask & OFPPC_PORT_DOWN) {
2329 if (config & OFPPC_PORT_DOWN) {
2330 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2332 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2335 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2336 if (mask & REVALIDATE_BITS) {
2337 COVERAGE_INC(ofproto_costly_flags);
2338 port->opp.config ^= mask & REVALIDATE_BITS;
2339 p->need_revalidate = true;
2341 #undef REVALIDATE_BITS
2342 if (mask & OFPPC_NO_FLOOD) {
2343 port->opp.config ^= OFPPC_NO_FLOOD;
2344 refresh_port_groups(p);
2346 if (mask & OFPPC_NO_PACKET_IN) {
2347 port->opp.config ^= OFPPC_NO_PACKET_IN;
2352 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2354 const struct ofp_port_mod *opm;
2355 struct ofport *port;
2358 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2362 opm = (struct ofp_port_mod *) oh;
2364 port = port_array_get(&p->ports,
2365 ofp_port_to_odp_port(ntohs(opm->port_no)));
2367 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2368 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2369 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2371 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2372 if (opm->advertise) {
2373 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2379 static struct ofpbuf *
2380 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2382 struct ofp_stats_reply *osr;
2385 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2386 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2388 osr->flags = htons(0);
2392 static struct ofpbuf *
2393 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2395 return make_stats_reply(request->header.xid, request->type, body_len);
2399 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2401 struct ofpbuf *msg = *msgp;
2402 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2403 if (nbytes + msg->size > UINT16_MAX) {
2404 struct ofp_stats_reply *reply = msg->data;
2405 reply->flags = htons(OFPSF_REPLY_MORE);
2406 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2407 queue_tx(msg, ofconn, ofconn->reply_counter);
2409 return ofpbuf_put_uninit(*msgp, nbytes);
2413 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2414 struct ofp_stats_request *request)
2416 struct ofp_desc_stats *ods;
2419 msg = start_stats_reply(request, sizeof *ods);
2420 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2421 memset(ods, 0, sizeof *ods);
2422 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2423 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2424 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2425 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2426 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2427 queue_tx(msg, ofconn, ofconn->reply_counter);
2433 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2435 struct rule *rule = rule_from_cls_rule(cls_rule);
2436 int *n_subrules = n_subrules_;
2444 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2445 struct ofp_stats_request *request)
2447 struct ofp_table_stats *ots;
2449 struct odp_stats dpstats;
2450 int n_exact, n_subrules, n_wild;
2452 msg = start_stats_reply(request, sizeof *ots * 2);
2454 /* Count rules of various kinds. */
2456 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2457 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2458 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2461 dpif_get_dp_stats(p->dpif, &dpstats);
2462 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2463 memset(ots, 0, sizeof *ots);
2464 ots->table_id = TABLEID_HASH;
2465 strcpy(ots->name, "hash");
2466 ots->wildcards = htonl(0);
2467 ots->max_entries = htonl(dpstats.max_capacity);
2468 ots->active_count = htonl(n_exact);
2469 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2471 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2473 /* Classifier table. */
2474 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2475 memset(ots, 0, sizeof *ots);
2476 ots->table_id = TABLEID_CLASSIFIER;
2477 strcpy(ots->name, "classifier");
2478 ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
2480 ots->max_entries = htonl(65536);
2481 ots->active_count = htonl(n_wild);
2482 ots->lookup_count = htonll(0); /* XXX */
2483 ots->matched_count = htonll(0); /* XXX */
2485 queue_tx(msg, ofconn, ofconn->reply_counter);
2490 append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
2493 struct netdev_stats stats;
2494 struct ofp_port_stats *ops;
2496 /* Intentionally ignore return value, since errors will set
2497 * 'stats' to all-1s, which is correct for OpenFlow, and
2498 * netdev_get_stats() will log errors. */
2499 netdev_get_stats(port->netdev, &stats);
2501 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2502 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2503 memset(ops->pad, 0, sizeof ops->pad);
2504 ops->rx_packets = htonll(stats.rx_packets);
2505 ops->tx_packets = htonll(stats.tx_packets);
2506 ops->rx_bytes = htonll(stats.rx_bytes);
2507 ops->tx_bytes = htonll(stats.tx_bytes);
2508 ops->rx_dropped = htonll(stats.rx_dropped);
2509 ops->tx_dropped = htonll(stats.tx_dropped);
2510 ops->rx_errors = htonll(stats.rx_errors);
2511 ops->tx_errors = htonll(stats.tx_errors);
2512 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2513 ops->rx_over_err = htonll(stats.rx_over_errors);
2514 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2515 ops->collisions = htonll(stats.collisions);
2519 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2520 struct ofp_stats_request *osr,
2523 struct ofp_port_stats_request *psr;
2524 struct ofp_port_stats *ops;
2526 struct ofport *port;
2527 unsigned int port_no;
2529 if (arg_size != sizeof *psr) {
2530 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2532 psr = (struct ofp_port_stats_request *) osr->body;
2534 msg = start_stats_reply(osr, sizeof *ops * 16);
2535 if (psr->port_no != htons(OFPP_NONE)) {
2536 port = port_array_get(&p->ports,
2537 ofp_port_to_odp_port(ntohs(psr->port_no)));
2539 append_port_stat(port, ntohs(psr->port_no), ofconn, msg);
2542 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2543 append_port_stat(port, port_no, ofconn, msg);
2547 queue_tx(msg, ofconn, ofconn->reply_counter);
2551 struct flow_stats_cbdata {
2552 struct ofproto *ofproto;
2553 struct ofconn *ofconn;
2558 /* Obtains statistic counters for 'rule' within 'p' and stores them into
2559 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
2560 * returned statistic include statistics for all of 'rule''s subrules. */
2562 query_stats(struct ofproto *p, struct rule *rule,
2563 uint64_t *packet_countp, uint64_t *byte_countp)
2565 uint64_t packet_count, byte_count;
2566 struct rule *subrule;
2567 struct odp_flow *odp_flows;
2570 /* Start from historical data for 'rule' itself that are no longer tracked
2571 * by the datapath. This counts, for example, subrules that have
2573 packet_count = rule->packet_count;
2574 byte_count = rule->byte_count;
2576 /* Prepare to ask the datapath for statistics on 'rule', or if it is
2577 * wildcarded then on all of its subrules.
2579 * Also, add any statistics that are not tracked by the datapath for each
2580 * subrule. This includes, for example, statistics for packets that were
2581 * executed "by hand" by ofproto via dpif_execute() but must be accounted
2583 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2584 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
2585 if (rule->cr.wc.wildcards) {
2587 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2588 odp_flows[i++].key = subrule->cr.flow;
2589 packet_count += subrule->packet_count;
2590 byte_count += subrule->byte_count;
2593 odp_flows[0].key = rule->cr.flow;
2596 /* Fetch up-to-date statistics from the datapath and add them in. */
2597 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2599 for (i = 0; i < n_odp_flows; i++) {
2600 struct odp_flow *odp_flow = &odp_flows[i];
2601 packet_count += odp_flow->stats.n_packets;
2602 byte_count += odp_flow->stats.n_bytes;
2607 /* Return the stats to the caller. */
2608 *packet_countp = packet_count;
2609 *byte_countp = byte_count;
2613 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2615 struct rule *rule = rule_from_cls_rule(rule_);
2616 struct flow_stats_cbdata *cbdata = cbdata_;
2617 struct ofp_flow_stats *ofs;
2618 uint64_t packet_count, byte_count;
2619 size_t act_len, len;
2620 long long int tdiff = time_msec() - rule->created;
2621 uint32_t sec = tdiff / 1000;
2622 uint32_t msec = tdiff - (sec * 1000);
2624 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2628 act_len = sizeof *rule->actions * rule->n_actions;
2629 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2631 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2633 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2634 ofs->length = htons(len);
2635 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2637 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
2638 cbdata->ofproto->tun_id_from_cookie, &ofs->match);
2639 ofs->duration_sec = htonl(sec);
2640 ofs->duration_nsec = htonl(msec * 1000000);
2641 ofs->cookie = rule->flow_cookie;
2642 ofs->priority = htons(rule->cr.priority);
2643 ofs->idle_timeout = htons(rule->idle_timeout);
2644 ofs->hard_timeout = htons(rule->hard_timeout);
2645 memset(ofs->pad2, 0, sizeof ofs->pad2);
2646 ofs->packet_count = htonll(packet_count);
2647 ofs->byte_count = htonll(byte_count);
2648 memcpy(ofs->actions, rule->actions, act_len);
2652 table_id_to_include(uint8_t table_id)
2654 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2655 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2656 : table_id == 0xff ? CLS_INC_ALL
2661 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2662 const struct ofp_stats_request *osr,
2665 struct ofp_flow_stats_request *fsr;
2666 struct flow_stats_cbdata cbdata;
2667 struct cls_rule target;
2669 if (arg_size != sizeof *fsr) {
2670 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2672 fsr = (struct ofp_flow_stats_request *) osr->body;
2674 COVERAGE_INC(ofproto_flows_req);
2676 cbdata.ofconn = ofconn;
2677 cbdata.out_port = fsr->out_port;
2678 cbdata.msg = start_stats_reply(osr, 1024);
2679 cls_rule_from_match(&fsr->match, 0, false, 0, &target);
2680 classifier_for_each_match(&p->cls, &target,
2681 table_id_to_include(fsr->table_id),
2682 flow_stats_cb, &cbdata);
2683 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2687 struct flow_stats_ds_cbdata {
2688 struct ofproto *ofproto;
2693 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
2695 struct rule *rule = rule_from_cls_rule(rule_);
2696 struct flow_stats_ds_cbdata *cbdata = cbdata_;
2697 struct ds *results = cbdata->results;
2698 struct ofp_match match;
2699 uint64_t packet_count, byte_count;
2700 size_t act_len = sizeof *rule->actions * rule->n_actions;
2702 /* Don't report on subrules. */
2703 if (rule->super != NULL) {
2707 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2708 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
2709 cbdata->ofproto->tun_id_from_cookie, &match);
2711 ds_put_format(results, "duration=%llds, ",
2712 (time_msec() - rule->created) / 1000);
2713 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2714 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2715 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2716 ofp_print_match(results, &match, true);
2717 ofp_print_actions(results, &rule->actions->header, act_len);
2718 ds_put_cstr(results, "\n");
2721 /* Adds a pretty-printed description of all flows to 'results', including
2722 * those marked hidden by secchan (e.g., by in-band control). */
2724 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2726 struct ofp_match match;
2727 struct cls_rule target;
2728 struct flow_stats_ds_cbdata cbdata;
2730 memset(&match, 0, sizeof match);
2731 match.wildcards = htonl(OVSFW_ALL);
2734 cbdata.results = results;
2736 cls_rule_from_match(&match, 0, false, 0, &target);
2737 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2738 flow_stats_ds_cb, &cbdata);
2741 struct aggregate_stats_cbdata {
2742 struct ofproto *ofproto;
2744 uint64_t packet_count;
2745 uint64_t byte_count;
2750 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2752 struct rule *rule = rule_from_cls_rule(rule_);
2753 struct aggregate_stats_cbdata *cbdata = cbdata_;
2754 uint64_t packet_count, byte_count;
2756 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2760 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2762 cbdata->packet_count += packet_count;
2763 cbdata->byte_count += byte_count;
2768 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2769 const struct ofp_stats_request *osr,
2772 struct ofp_aggregate_stats_request *asr;
2773 struct ofp_aggregate_stats_reply *reply;
2774 struct aggregate_stats_cbdata cbdata;
2775 struct cls_rule target;
2778 if (arg_size != sizeof *asr) {
2779 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2781 asr = (struct ofp_aggregate_stats_request *) osr->body;
2783 COVERAGE_INC(ofproto_agg_request);
2785 cbdata.out_port = asr->out_port;
2786 cbdata.packet_count = 0;
2787 cbdata.byte_count = 0;
2789 cls_rule_from_match(&asr->match, 0, false, 0, &target);
2790 classifier_for_each_match(&p->cls, &target,
2791 table_id_to_include(asr->table_id),
2792 aggregate_stats_cb, &cbdata);
2794 msg = start_stats_reply(osr, sizeof *reply);
2795 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2796 reply->flow_count = htonl(cbdata.n_flows);
2797 reply->packet_count = htonll(cbdata.packet_count);
2798 reply->byte_count = htonll(cbdata.byte_count);
2799 queue_tx(msg, ofconn, ofconn->reply_counter);
2804 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2805 struct ofp_header *oh)
2807 struct ofp_stats_request *osr;
2811 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2816 osr = (struct ofp_stats_request *) oh;
2818 switch (ntohs(osr->type)) {
2820 return handle_desc_stats_request(p, ofconn, osr);
2823 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2825 case OFPST_AGGREGATE:
2826 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2829 return handle_table_stats_request(p, ofconn, osr);
2832 return handle_port_stats_request(p, ofconn, osr, arg_size);
2835 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2838 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2842 static long long int
2843 msec_from_nsec(uint64_t sec, uint32_t nsec)
2845 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2849 update_time(struct ofproto *ofproto, struct rule *rule,
2850 const struct odp_flow_stats *stats)
2852 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2853 if (used > rule->used) {
2855 if (rule->super && used > rule->super->used) {
2856 rule->super->used = used;
2858 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
2863 update_stats(struct ofproto *ofproto, struct rule *rule,
2864 const struct odp_flow_stats *stats)
2866 if (stats->n_packets) {
2867 update_time(ofproto, rule, stats);
2868 rule->packet_count += stats->n_packets;
2869 rule->byte_count += stats->n_bytes;
2870 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
2875 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
2876 * in which no matching flow already exists in the flow table.
2878 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
2879 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
2880 * code as encoded by ofp_mkerr() on failure.
2882 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2885 add_flow(struct ofproto *p, struct ofconn *ofconn,
2886 const struct ofp_flow_mod *ofm, size_t n_actions)
2888 struct ofpbuf *packet;
2893 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
2897 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
2899 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
2900 ntohs(ofm->priority))) {
2901 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
2905 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
2906 n_actions, ntohs(ofm->idle_timeout),
2907 ntohs(ofm->hard_timeout), ofm->cookie,
2908 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
2909 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
2910 p->tun_id_from_cookie, ofm->cookie, &rule->cr);
2913 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2914 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2918 in_port = UINT16_MAX;
2921 rule_insert(p, rule, packet, in_port);
2922 ofpbuf_delete(packet);
2926 static struct rule *
2927 find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
2932 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
2934 return rule_from_cls_rule(classifier_find_rule_exactly(
2935 &p->cls, &flow, wildcards,
2936 ntohs(ofm->priority)));
2940 send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
2941 struct rule *rule, const struct ofp_flow_mod *ofm)
2943 struct ofpbuf *packet;
2948 if (ofm->buffer_id == htonl(UINT32_MAX)) {
2952 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2958 flow_extract(packet, 0, in_port, &flow);
2959 rule_execute(ofproto, rule, packet, &flow);
2960 ofpbuf_delete(packet);
2965 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
2967 struct modify_flows_cbdata {
2968 struct ofproto *ofproto;
2969 const struct ofp_flow_mod *ofm;
2974 static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
2975 size_t n_actions, struct rule *);
2976 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
2978 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
2979 * encoded by ofp_mkerr() on failure.
2981 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2984 modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
2985 const struct ofp_flow_mod *ofm, size_t n_actions)
2987 struct modify_flows_cbdata cbdata;
2988 struct cls_rule target;
2992 cbdata.n_actions = n_actions;
2993 cbdata.match = NULL;
2995 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
2998 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2999 modify_flows_cb, &cbdata);
3001 /* This credits the packet to whichever flow happened to happened to
3002 * match last. That's weird. Maybe we should do a lookup for the
3003 * flow that actually matches the packet? Who knows. */
3004 send_buffered_packet(p, ofconn, cbdata.match, ofm);
3007 return add_flow(p, ofconn, ofm, n_actions);
3011 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3012 * code as encoded by ofp_mkerr() on failure.
3014 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3017 modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
3018 struct ofp_flow_mod *ofm, size_t n_actions)
3020 struct rule *rule = find_flow_strict(p, ofm);
3021 if (rule && !rule_is_hidden(rule)) {
3022 modify_flow(p, ofm, n_actions, rule);
3023 return send_buffered_packet(p, ofconn, rule, ofm);
3025 return add_flow(p, ofconn, ofm, n_actions);
3029 /* Callback for modify_flows_loose(). */
3031 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3033 struct rule *rule = rule_from_cls_rule(rule_);
3034 struct modify_flows_cbdata *cbdata = cbdata_;
3036 if (!rule_is_hidden(rule)) {
3037 cbdata->match = rule;
3038 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
3042 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3043 * been identified as a flow in 'p''s flow table to be modified, by changing
3044 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3045 * ofp_action[] structures). */
3047 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
3048 size_t n_actions, struct rule *rule)
3050 size_t actions_len = n_actions * sizeof *rule->actions;
3052 rule->flow_cookie = ofm->cookie;
3054 /* If the actions are the same, do nothing. */
3055 if (n_actions == rule->n_actions
3056 && !memcmp(ofm->actions, rule->actions, actions_len))
3061 /* Replace actions. */
3062 free(rule->actions);
3063 rule->actions = xmemdup(ofm->actions, actions_len);
3064 rule->n_actions = n_actions;
3066 /* Make sure that the datapath gets updated properly. */
3067 if (rule->cr.wc.wildcards) {
3068 COVERAGE_INC(ofproto_mod_wc_flow);
3069 p->need_revalidate = true;
3071 rule_update_actions(p, rule);
3077 /* OFPFC_DELETE implementation. */
3079 struct delete_flows_cbdata {
3080 struct ofproto *ofproto;
3084 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3085 static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port);
3087 /* Implements OFPFC_DELETE. */
3089 delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
3091 struct delete_flows_cbdata cbdata;
3092 struct cls_rule target;
3095 cbdata.out_port = ofm->out_port;
3097 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3100 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3101 delete_flows_cb, &cbdata);
3104 /* Implements OFPFC_DELETE_STRICT. */
3106 delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
3108 struct rule *rule = find_flow_strict(p, ofm);
3110 delete_flow(p, rule, ofm->out_port);
3114 /* Callback for delete_flows_loose(). */
3116 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3118 struct rule *rule = rule_from_cls_rule(rule_);
3119 struct delete_flows_cbdata *cbdata = cbdata_;
3121 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
3124 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3125 * been identified as a flow to delete from 'p''s flow table, by deleting the
3126 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3129 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3130 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3131 * specified 'out_port'. */
3133 delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port)
3135 if (rule_is_hidden(rule)) {
3139 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3143 send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
3144 rule_remove(p, rule);
3148 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3149 struct ofp_flow_mod *ofm)
3154 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3155 sizeof *ofm->actions, &n_actions);
3160 /* We do not support the emergency flow cache. It will hopefully
3161 * get dropped from OpenFlow in the near future. */
3162 if (ofm->flags & htons(OFPFF_EMERG)) {
3163 /* There isn't a good fit for an error code, so just state that the
3164 * flow table is full. */
3165 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3168 normalize_match(&ofm->match);
3169 if (!ofm->match.wildcards) {
3170 ofm->priority = htons(UINT16_MAX);
3173 error = validate_actions((const union ofp_action *) ofm->actions,
3174 n_actions, p->max_ports);
3179 switch (ntohs(ofm->command)) {
3181 return add_flow(p, ofconn, ofm, n_actions);
3184 return modify_flows_loose(p, ofconn, ofm, n_actions);
3186 case OFPFC_MODIFY_STRICT:
3187 return modify_flow_strict(p, ofconn, ofm, n_actions);
3190 delete_flows_loose(p, ofm);
3193 case OFPFC_DELETE_STRICT:
3194 delete_flow_strict(p, ofm);
3198 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3203 handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg)
3207 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
3212 p->tun_id_from_cookie = !!msg->set;
3217 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3219 struct ofp_vendor_header *ovh = msg;
3220 struct nicira_header *nh;
3222 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3223 VLOG_WARN_RL(&rl, "received vendor message of length %zu "
3224 "(expected at least %zu)",
3225 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
3226 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3228 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3229 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3231 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3232 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %zu "
3233 "(expected at least %zu)",
3234 ntohs(ovh->header.length), sizeof(struct nicira_header));
3235 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3239 switch (ntohl(nh->subtype)) {
3240 case NXT_STATUS_REQUEST:
3241 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3244 case NXT_TUN_ID_FROM_COOKIE:
3245 return handle_tun_id_from_cookie(p, msg);
3248 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3252 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
3254 struct ofp_header *ob;
3257 /* Currently, everything executes synchronously, so we can just
3258 * immediately send the barrier reply. */
3259 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3260 queue_tx(buf, ofconn, ofconn->reply_counter);
3265 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3266 struct ofpbuf *ofp_msg)
3268 struct ofp_header *oh = ofp_msg->data;
3271 COVERAGE_INC(ofproto_recv_openflow);
3273 case OFPT_ECHO_REQUEST:
3274 error = handle_echo_request(ofconn, oh);
3277 case OFPT_ECHO_REPLY:
3281 case OFPT_FEATURES_REQUEST:
3282 error = handle_features_request(p, ofconn, oh);
3285 case OFPT_GET_CONFIG_REQUEST:
3286 error = handle_get_config_request(p, ofconn, oh);
3289 case OFPT_SET_CONFIG:
3290 error = handle_set_config(p, ofconn, ofp_msg->data);
3293 case OFPT_PACKET_OUT:
3294 error = handle_packet_out(p, ofconn, ofp_msg->data);
3298 error = handle_port_mod(p, oh);
3302 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3305 case OFPT_STATS_REQUEST:
3306 error = handle_stats_request(p, ofconn, oh);
3310 error = handle_vendor(p, ofconn, ofp_msg->data);
3313 case OFPT_BARRIER_REQUEST:
3314 error = handle_barrier_request(ofconn, oh);
3318 if (VLOG_IS_WARN_ENABLED()) {
3319 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3320 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3323 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3328 send_error_oh(ofconn, ofp_msg->data, error);
3333 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
3335 struct odp_msg *msg = packet->data;
3336 uint16_t in_port = odp_port_to_ofp_port(msg->port);
3338 struct ofpbuf payload;
3341 payload.data = msg + 1;
3342 payload.size = msg->length - sizeof *msg;
3343 flow_extract(&payload, msg->arg, msg->port, &flow);
3345 /* Check with in-band control to see if this packet should be sent
3346 * to the local port regardless of the flow table. */
3347 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3348 union odp_action action;
3350 memset(&action, 0, sizeof(action));
3351 action.output.type = ODPAT_OUTPUT;
3352 action.output.port = ODPP_LOCAL;
3353 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3356 rule = lookup_valid_rule(p, &flow);
3358 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3359 struct ofport *port = port_array_get(&p->ports, msg->port);
3361 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3362 COVERAGE_INC(ofproto_no_packet_in);
3363 /* XXX install 'drop' flow entry */
3364 ofpbuf_delete(packet);
3368 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3371 COVERAGE_INC(ofproto_packet_in);
3372 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3376 if (rule->cr.wc.wildcards) {
3377 rule = rule_create_subrule(p, rule, &flow);
3378 rule_make_actions(p, rule, packet);
3380 if (!rule->may_install) {
3381 /* The rule is not installable, that is, we need to process every
3382 * packet, so process the current packet and set its actions into
3384 rule_make_actions(p, rule, packet);
3386 /* XXX revalidate rule if it needs it */
3390 rule_execute(p, rule, &payload, &flow);
3391 rule_reinstall(p, rule);
3393 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY
3394 && rconn_is_connected(p->controller->rconn)) {
3396 * Extra-special case for fail-open mode.
3398 * We are in fail-open mode and the packet matched the fail-open rule,
3399 * but we are connected to a controller too. We should send the packet
3400 * up to the controller in the hope that it will try to set up a flow
3401 * and thereby allow us to exit fail-open.
3403 * See the top-level comment in fail-open.c for more information.
3405 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3407 ofpbuf_delete(packet);
3412 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3414 struct odp_msg *msg = packet->data;
3416 switch (msg->type) {
3417 case _ODPL_ACTION_NR:
3418 COVERAGE_INC(ofproto_ctlr_action);
3419 pinsched_send(p->action_sched, odp_port_to_ofp_port(msg->port), packet,
3420 send_packet_in_action, p);
3423 case _ODPL_SFLOW_NR:
3425 ofproto_sflow_received(p->sflow, msg);
3427 ofpbuf_delete(packet);
3431 handle_odp_miss_msg(p, packet);
3435 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
3442 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3444 struct rule *sub = rule_from_cls_rule(sub_);
3445 struct revalidate_cbdata *cbdata = cbdata_;
3447 if (cbdata->revalidate_all
3448 || (cbdata->revalidate_subrules && sub->super)
3449 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3450 revalidate_rule(cbdata->ofproto, sub);
3455 revalidate_rule(struct ofproto *p, struct rule *rule)
3457 const flow_t *flow = &rule->cr.flow;
3459 COVERAGE_INC(ofproto_revalidate_rule);
3462 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3464 rule_remove(p, rule);
3466 } else if (super != rule->super) {
3467 COVERAGE_INC(ofproto_revalidate_moved);
3468 list_remove(&rule->list);
3469 list_push_back(&super->list, &rule->list);
3470 rule->super = super;
3471 rule->hard_timeout = super->hard_timeout;
3472 rule->idle_timeout = super->idle_timeout;
3473 rule->created = super->created;
3478 rule_update_actions(p, rule);
3482 static struct ofpbuf *
3483 compose_flow_removed(struct ofproto *p, const struct rule *rule,
3484 long long int now, uint8_t reason)
3486 struct ofp_flow_removed *ofr;
3488 long long int tdiff = now - rule->created;
3489 uint32_t sec = tdiff / 1000;
3490 uint32_t msec = tdiff - (sec * 1000);
3492 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
3493 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
3495 ofr->cookie = rule->flow_cookie;
3496 ofr->priority = htons(rule->cr.priority);
3497 ofr->reason = reason;
3498 ofr->duration_sec = htonl(sec);
3499 ofr->duration_nsec = htonl(msec * 1000000);
3500 ofr->idle_timeout = htons(rule->idle_timeout);
3501 ofr->packet_count = htonll(rule->packet_count);
3502 ofr->byte_count = htonll(rule->byte_count);
3508 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3510 assert(rule->installed);
3511 assert(!rule->cr.wc.wildcards);
3514 rule_remove(ofproto, rule);
3516 rule_uninstall(ofproto, rule);
3520 send_flow_removed(struct ofproto *p, struct rule *rule,
3521 long long int now, uint8_t reason)
3523 struct ofconn *ofconn;
3524 struct ofconn *prev;
3525 struct ofpbuf *buf = NULL;
3527 /* We limit the maximum number of queued flow expirations it by accounting
3528 * them under the counter for replies. That works because preventing
3529 * OpenFlow requests from being processed also prevents new flows from
3530 * being added (and expiring). (It also prevents processing OpenFlow
3531 * requests that would not add new flows, so it is imperfect.) */
3534 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3535 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)) {
3537 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3539 buf = compose_flow_removed(p, rule, now, reason);
3545 queue_tx(buf, prev, prev->reply_counter);
3551 expire_rule(struct cls_rule *cls_rule, void *p_)
3553 struct ofproto *p = p_;
3554 struct rule *rule = rule_from_cls_rule(cls_rule);
3555 long long int hard_expire, idle_expire, expire, now;
3557 hard_expire = (rule->hard_timeout
3558 ? rule->created + rule->hard_timeout * 1000
3560 idle_expire = (rule->idle_timeout
3561 && (rule->super || list_is_empty(&rule->list))
3562 ? rule->used + rule->idle_timeout * 1000
3564 expire = MIN(hard_expire, idle_expire);
3568 if (rule->installed && now >= rule->used + 5000) {
3569 uninstall_idle_flow(p, rule);
3570 } else if (!rule->cr.wc.wildcards) {
3571 active_timeout(p, rule);
3577 COVERAGE_INC(ofproto_expired);
3579 /* Update stats. This code will be a no-op if the rule expired
3580 * due to an idle timeout. */
3581 if (rule->cr.wc.wildcards) {
3582 struct rule *subrule, *next;
3583 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3584 rule_remove(p, subrule);
3587 rule_uninstall(p, rule);
3590 if (!rule_is_hidden(rule)) {
3591 send_flow_removed(p, rule, now,
3593 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
3595 rule_remove(p, rule);
3599 active_timeout(struct ofproto *ofproto, struct rule *rule)
3601 if (ofproto->netflow && !is_controller_rule(rule) &&
3602 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
3603 struct ofexpired expired;
3604 struct odp_flow odp_flow;
3606 /* Get updated flow stats. */
3607 memset(&odp_flow, 0, sizeof odp_flow);
3608 if (rule->installed) {
3609 odp_flow.key = rule->cr.flow;
3610 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
3611 dpif_flow_get(ofproto->dpif, &odp_flow);
3613 if (odp_flow.stats.n_packets) {
3614 update_time(ofproto, rule, &odp_flow.stats);
3615 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
3616 odp_flow.stats.tcp_flags);
3620 expired.flow = rule->cr.flow;
3621 expired.packet_count = rule->packet_count +
3622 odp_flow.stats.n_packets;
3623 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
3624 expired.used = rule->used;
3626 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
3628 /* Schedule us to send the accumulated records once we have
3629 * collected all of them. */
3630 poll_immediate_wake();
3635 update_used(struct ofproto *p)
3637 struct odp_flow *flows;
3642 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
3647 for (i = 0; i < n_flows; i++) {
3648 struct odp_flow *f = &flows[i];
3651 rule = rule_from_cls_rule(
3652 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3653 if (!rule || !rule->installed) {
3654 COVERAGE_INC(ofproto_unexpected_rule);
3655 dpif_flow_del(p->dpif, f);
3659 update_time(p, rule, &f->stats);
3660 rule_account(p, rule, f->stats.n_bytes);
3666 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3667 const struct ofpbuf *packet, int send_len)
3669 struct odp_msg *msg = packet->data;
3670 struct ofpbuf payload;
3674 /* Extract packet payload from 'msg'. */
3675 payload.data = msg + 1;
3676 payload.size = msg->length - sizeof *msg;
3678 /* Construct ofp_packet_in message. */
3679 reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3680 opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port), reason,
3681 &payload, send_len);
3684 rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100);
3688 send_packet_in_action(struct ofpbuf *packet, void *p_)
3690 struct ofproto *p = p_;
3691 struct ofconn *ofconn;
3692 struct odp_msg *msg;
3695 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3696 if (ofconn == p->controller || ofconn->miss_send_len) {
3697 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3700 ofpbuf_delete(packet);
3704 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3706 struct ofproto *p = p_;
3707 bool in_fail_open = p->fail_open && fail_open_is_active(p->fail_open);
3708 struct ofconn *ofconn;
3709 struct ofpbuf payload;
3710 struct odp_msg *msg;
3713 payload.data = msg + 1;
3714 payload.size = msg->length - sizeof *msg;
3715 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3716 if (ofconn->miss_send_len) {
3717 struct pktbuf *pb = ofconn->pktbuf;
3718 uint32_t buffer_id = (in_fail_open
3720 : pktbuf_save(pb, &payload, msg->port));
3721 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3723 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3726 ofpbuf_delete(packet);
3730 pick_datapath_id(const struct ofproto *ofproto)
3732 const struct ofport *port;
3734 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
3736 uint8_t ea[ETH_ADDR_LEN];
3739 error = netdev_get_etheraddr(port->netdev, ea);
3741 return eth_addr_to_uint64(ea);
3743 VLOG_WARN("could not get MAC address for %s (%s)",
3744 netdev_get_name(port->netdev), strerror(error));
3746 return ofproto->fallback_dpid;
3750 pick_fallback_dpid(void)
3752 uint8_t ea[ETH_ADDR_LEN];
3753 eth_addr_nicira_random(ea);
3754 return eth_addr_to_uint64(ea);
3758 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3759 struct odp_actions *actions, tag_type *tags,
3760 uint16_t *nf_output_iface, void *ofproto_)
3762 struct ofproto *ofproto = ofproto_;
3765 /* Drop frames for reserved multicast addresses. */
3766 if (eth_addr_is_reserved(flow->dl_dst)) {
3770 /* Learn source MAC (but don't try to learn from revalidation). */
3771 if (packet != NULL) {
3772 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3775 /* The log messages here could actually be useful in debugging,
3776 * so keep the rate limit relatively high. */
3777 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3778 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3779 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3780 ofproto_revalidate(ofproto, rev_tag);
3784 /* Determine output port. */
3785 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3787 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
3788 } else if (out_port != flow->in_port) {
3789 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3790 *nf_output_iface = out_port;
3798 static const struct ofhooks default_ofhooks = {
3800 default_normal_ofhook_cb,