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_probe_interval(struct ofproto *p, int probe_interval)
381 probe_interval = probe_interval ? MAX(probe_interval, 5) : 0;
382 rconn_set_probe_interval(p->controller->rconn, probe_interval);
384 int trigger_duration = probe_interval ? probe_interval * 3 : 15;
385 fail_open_set_trigger_duration(p->fail_open, trigger_duration);
390 ofproto_set_max_backoff(struct ofproto *p, int max_backoff)
392 rconn_set_max_backoff(p->controller->rconn, max_backoff);
396 ofproto_set_desc(struct ofproto *p,
397 const char *mfr_desc, const char *hw_desc,
398 const char *sw_desc, const char *serial_desc,
401 struct ofp_desc_stats *ods;
404 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
405 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
406 sizeof ods->mfr_desc);
409 p->mfr_desc = xstrdup(mfr_desc);
412 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
413 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
414 sizeof ods->hw_desc);
417 p->hw_desc = xstrdup(hw_desc);
420 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
421 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
422 sizeof ods->sw_desc);
425 p->sw_desc = xstrdup(sw_desc);
428 if (strlen(serial_desc) >= sizeof ods->serial_num) {
429 VLOG_WARN("truncating serial_desc, must be less than %zu "
431 sizeof ods->serial_num);
433 free(p->serial_desc);
434 p->serial_desc = xstrdup(serial_desc);
437 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
438 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
439 sizeof ods->dp_desc);
442 p->dp_desc = xstrdup(dp_desc);
447 ofproto_set_in_band(struct ofproto *p, bool in_band)
449 if (in_band != (p->in_band != NULL)) {
453 error = in_band_create(p, p->dpif, p->switch_status, &p->in_band);
457 in_band_set_remotes(p->in_band, &p->controller->rconn, 1);
459 ofproto_set_discovery(p, false, NULL, true);
460 in_band_destroy(p->in_band);
463 rconn_reconnect(p->controller->rconn);
469 ofproto_set_discovery(struct ofproto *p, bool discovery,
470 const char *re, bool update_resolv_conf)
472 if (discovery != (p->discovery != NULL)) {
474 int error = ofproto_set_in_band(p, true);
478 error = discovery_create(re, update_resolv_conf,
479 p->dpif, p->switch_status,
485 discovery_destroy(p->discovery);
488 rconn_disconnect(p->controller->rconn);
489 } else if (discovery) {
490 discovery_set_update_resolv_conf(p->discovery, update_resolv_conf);
491 return discovery_set_accept_controller_re(p->discovery, re);
497 ofproto_set_controller(struct ofproto *ofproto, const char *controller)
499 if (ofproto->discovery) {
501 } else if (controller) {
502 if (strcmp(rconn_get_name(ofproto->controller->rconn), controller)) {
503 return rconn_connect(ofproto->controller->rconn, controller);
508 rconn_disconnect(ofproto->controller->rconn);
514 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
515 const struct svec *svec)
517 struct pvconn **pvconns = *pvconnsp;
518 size_t n_pvconns = *n_pvconnsp;
522 for (i = 0; i < n_pvconns; i++) {
523 pvconn_close(pvconns[i]);
527 pvconns = xmalloc(svec->n * sizeof *pvconns);
529 for (i = 0; i < svec->n; i++) {
530 const char *name = svec->names[i];
531 struct pvconn *pvconn;
534 error = pvconn_open(name, &pvconn);
536 pvconns[n_pvconns++] = pvconn;
538 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
546 *n_pvconnsp = n_pvconns;
552 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
554 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
558 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
560 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
564 ofproto_set_netflow(struct ofproto *ofproto,
565 const struct netflow_options *nf_options)
567 if (nf_options && nf_options->collectors.n) {
568 if (!ofproto->netflow) {
569 ofproto->netflow = netflow_create();
571 return netflow_set_options(ofproto->netflow, nf_options);
573 netflow_destroy(ofproto->netflow);
574 ofproto->netflow = NULL;
580 ofproto_set_sflow(struct ofproto *ofproto,
581 const struct ofproto_sflow_options *oso)
583 struct ofproto_sflow *os = ofproto->sflow;
586 struct ofport *ofport;
587 unsigned int odp_port;
589 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
590 refresh_port_groups(ofproto);
591 PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) {
592 ofproto_sflow_add_port(os, odp_port,
593 netdev_get_name(ofport->netdev));
596 ofproto_sflow_set_options(os, oso);
598 ofproto_sflow_destroy(os);
599 ofproto->sflow = NULL;
604 ofproto_set_failure(struct ofproto *ofproto, bool fail_open)
607 struct rconn *rconn = ofproto->controller->rconn;
608 int trigger_duration = rconn_get_probe_interval(rconn) * 3;
609 if (!ofproto->fail_open) {
610 ofproto->fail_open = fail_open_create(ofproto, trigger_duration,
611 ofproto->switch_status,
614 fail_open_set_trigger_duration(ofproto->fail_open,
618 fail_open_destroy(ofproto->fail_open);
619 ofproto->fail_open = NULL;
624 ofproto_set_rate_limit(struct ofproto *ofproto,
625 int rate_limit, int burst_limit)
627 if (rate_limit > 0) {
628 if (!ofproto->miss_sched) {
629 ofproto->miss_sched = pinsched_create(rate_limit, burst_limit,
630 ofproto->switch_status);
631 ofproto->action_sched = pinsched_create(rate_limit, burst_limit,
634 pinsched_set_limits(ofproto->miss_sched, rate_limit, burst_limit);
635 pinsched_set_limits(ofproto->action_sched,
636 rate_limit, burst_limit);
639 pinsched_destroy(ofproto->miss_sched);
640 ofproto->miss_sched = NULL;
641 pinsched_destroy(ofproto->action_sched);
642 ofproto->action_sched = NULL;
647 ofproto_set_stp(struct ofproto *ofproto OVS_UNUSED, bool enable_stp)
651 VLOG_WARN("STP is not yet implemented");
659 ofproto_get_datapath_id(const struct ofproto *ofproto)
661 return ofproto->datapath_id;
665 ofproto_get_probe_interval(const struct ofproto *ofproto)
667 return rconn_get_probe_interval(ofproto->controller->rconn);
671 ofproto_get_max_backoff(const struct ofproto *ofproto)
673 return rconn_get_max_backoff(ofproto->controller->rconn);
677 ofproto_get_in_band(const struct ofproto *ofproto)
679 return ofproto->in_band != NULL;
683 ofproto_get_discovery(const struct ofproto *ofproto)
685 return ofproto->discovery != NULL;
689 ofproto_get_controller(const struct ofproto *ofproto)
691 return rconn_get_name(ofproto->controller->rconn);
695 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
699 for (i = 0; i < ofproto->n_listeners; i++) {
700 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
705 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
709 for (i = 0; i < ofproto->n_snoops; i++) {
710 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
715 ofproto_destroy(struct ofproto *p)
717 struct ofconn *ofconn, *next_ofconn;
718 struct ofport *ofport;
719 unsigned int port_no;
726 /* Destroy fail-open and in-band early, since they touch the classifier. */
727 ofproto_set_failure(p, false);
728 ofproto_set_in_band(p, false);
730 ofproto_flush_flows(p);
731 classifier_destroy(&p->cls);
733 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
735 ofconn_destroy(ofconn);
739 netdev_monitor_destroy(p->netdev_monitor);
740 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
743 shash_destroy(&p->port_by_name);
745 switch_status_destroy(p->switch_status);
746 discovery_destroy(p->discovery);
747 pinsched_destroy(p->miss_sched);
748 pinsched_destroy(p->action_sched);
749 netflow_destroy(p->netflow);
750 ofproto_sflow_destroy(p->sflow);
752 switch_status_unregister(p->ss_cat);
754 for (i = 0; i < p->n_listeners; i++) {
755 pvconn_close(p->listeners[i]);
759 for (i = 0; i < p->n_snoops; i++) {
760 pvconn_close(p->snoops[i]);
764 mac_learning_destroy(p->ml);
769 free(p->serial_desc);
772 port_array_destroy(&p->ports);
778 ofproto_run(struct ofproto *p)
780 int error = ofproto_run1(p);
782 error = ofproto_run2(p, false);
788 process_port_change(struct ofproto *ofproto, int error, char *devname)
790 if (error == ENOBUFS) {
791 reinit_ports(ofproto);
793 update_port(ofproto, devname);
799 ofproto_run1(struct ofproto *p)
801 struct ofconn *ofconn, *next_ofconn;
806 if (shash_is_empty(&p->port_by_name)) {
810 for (i = 0; i < 50; i++) {
814 error = dpif_recv(p->dpif, &buf);
816 if (error == ENODEV) {
817 /* Someone destroyed the datapath behind our back. The caller
818 * better destroy us and give up, because we're just going to
819 * spin from here on out. */
820 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
821 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
828 handle_odp_msg(p, buf);
831 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
832 process_port_change(p, error, devname);
834 while ((error = netdev_monitor_poll(p->netdev_monitor,
835 &devname)) != EAGAIN) {
836 process_port_change(p, error, devname);
840 in_band_run(p->in_band);
843 char *controller_name;
844 if (rconn_is_connectivity_questionable(p->controller->rconn)) {
845 discovery_question_connectivity(p->discovery);
847 if (discovery_run(p->discovery, &controller_name)) {
848 if (controller_name) {
849 rconn_connect(p->controller->rconn, controller_name);
851 rconn_disconnect(p->controller->rconn);
855 pinsched_run(p->miss_sched, send_packet_in_miss, p);
856 pinsched_run(p->action_sched, send_packet_in_action, p);
858 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
860 ofconn_run(ofconn, p);
863 /* Fail-open maintenance. Do this after processing the ofconns since
864 * fail-open checks the status of the controller rconn. */
866 fail_open_run(p->fail_open);
869 for (i = 0; i < p->n_listeners; i++) {
873 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
875 ofconn_create(p, rconn_new_from_vconn("passive", vconn));
876 } else if (retval != EAGAIN) {
877 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
881 for (i = 0; i < p->n_snoops; i++) {
885 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
887 rconn_add_monitor(p->controller->rconn, vconn);
888 } else if (retval != EAGAIN) {
889 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
893 if (time_msec() >= p->next_expiration) {
894 COVERAGE_INC(ofproto_expiration);
895 p->next_expiration = time_msec() + 1000;
898 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
900 /* Let the hook know that we're at a stable point: all outstanding data
901 * in existing flows has been accounted to the account_cb. Thus, the
902 * hook can now reasonably do operations that depend on having accurate
903 * flow volume accounting (currently, that's just bond rebalancing). */
904 if (p->ofhooks->account_checkpoint_cb) {
905 p->ofhooks->account_checkpoint_cb(p->aux);
910 netflow_run(p->netflow);
913 ofproto_sflow_run(p->sflow);
919 struct revalidate_cbdata {
920 struct ofproto *ofproto;
921 bool revalidate_all; /* Revalidate all exact-match rules? */
922 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
923 struct tag_set revalidate_set; /* Set of tags to revalidate. */
927 ofproto_run2(struct ofproto *p, bool revalidate_all)
929 if (p->need_revalidate || revalidate_all
930 || !tag_set_is_empty(&p->revalidate_set)) {
931 struct revalidate_cbdata cbdata;
933 cbdata.revalidate_all = revalidate_all;
934 cbdata.revalidate_subrules = p->need_revalidate;
935 cbdata.revalidate_set = p->revalidate_set;
936 tag_set_init(&p->revalidate_set);
937 COVERAGE_INC(ofproto_revalidate);
938 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
939 p->need_revalidate = false;
946 ofproto_wait(struct ofproto *p)
948 struct ofconn *ofconn;
951 dpif_recv_wait(p->dpif);
952 dpif_port_poll_wait(p->dpif);
953 netdev_monitor_poll_wait(p->netdev_monitor);
954 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
958 in_band_wait(p->in_band);
961 discovery_wait(p->discovery);
964 fail_open_wait(p->fail_open);
966 pinsched_wait(p->miss_sched);
967 pinsched_wait(p->action_sched);
969 ofproto_sflow_wait(p->sflow);
971 if (!tag_set_is_empty(&p->revalidate_set)) {
972 poll_immediate_wake();
974 if (p->need_revalidate) {
975 /* Shouldn't happen, but if it does just go around again. */
976 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
977 poll_immediate_wake();
978 } else if (p->next_expiration != LLONG_MAX) {
979 poll_timer_wait(p->next_expiration - time_msec());
981 for (i = 0; i < p->n_listeners; i++) {
982 pvconn_wait(p->listeners[i]);
984 for (i = 0; i < p->n_snoops; i++) {
985 pvconn_wait(p->snoops[i]);
990 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
992 tag_set_add(&ofproto->revalidate_set, tag);
996 ofproto_get_revalidate_set(struct ofproto *ofproto)
998 return &ofproto->revalidate_set;
1002 ofproto_is_alive(const struct ofproto *p)
1004 return p->discovery || rconn_is_alive(p->controller->rconn);
1008 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
1009 const union ofp_action *actions, size_t n_actions,
1010 const struct ofpbuf *packet)
1012 struct odp_actions odp_actions;
1015 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1021 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1023 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
1024 odp_actions.n_actions, packet);
1029 ofproto_add_flow(struct ofproto *p,
1030 const flow_t *flow, uint32_t wildcards, unsigned int priority,
1031 const union ofp_action *actions, size_t n_actions,
1035 rule = rule_create(p, NULL, actions, n_actions,
1036 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1038 cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
1039 rule_insert(p, rule, NULL, 0);
1043 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1044 uint32_t wildcards, unsigned int priority)
1048 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1052 rule_remove(ofproto, rule);
1057 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1059 struct rule *rule = rule_from_cls_rule(rule_);
1060 struct ofproto *ofproto = ofproto_;
1062 /* Mark the flow as not installed, even though it might really be
1063 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1064 * There is no point in uninstalling it individually since we are about to
1065 * blow away all the flows with dpif_flow_flush(). */
1066 rule->installed = false;
1068 rule_remove(ofproto, rule);
1072 ofproto_flush_flows(struct ofproto *ofproto)
1074 COVERAGE_INC(ofproto_flush);
1075 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1076 dpif_flow_flush(ofproto->dpif);
1077 if (ofproto->in_band) {
1078 in_band_flushed(ofproto->in_band);
1080 if (ofproto->fail_open) {
1081 fail_open_flushed(ofproto->fail_open);
1086 reinit_ports(struct ofproto *p)
1088 struct svec devnames;
1089 struct ofport *ofport;
1090 unsigned int port_no;
1091 struct odp_port *odp_ports;
1095 svec_init(&devnames);
1096 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1097 svec_add (&devnames, (char *) ofport->opp.name);
1099 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1100 for (i = 0; i < n_odp_ports; i++) {
1101 svec_add (&devnames, odp_ports[i].devname);
1105 svec_sort_unique(&devnames);
1106 for (i = 0; i < devnames.n; i++) {
1107 update_port(p, devnames.names[i]);
1109 svec_destroy(&devnames);
1113 refresh_port_group(struct ofproto *p, unsigned int group)
1117 struct ofport *port;
1118 unsigned int port_no;
1120 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1122 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1124 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1125 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1126 ports[n_ports++] = port_no;
1129 dpif_port_group_set(p->dpif, group, ports, n_ports);
1136 refresh_port_groups(struct ofproto *p)
1138 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1139 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1141 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1145 static struct ofport *
1146 make_ofport(const struct odp_port *odp_port)
1148 struct netdev_options netdev_options;
1149 enum netdev_flags flags;
1150 struct ofport *ofport;
1151 struct netdev *netdev;
1155 memset(&netdev_options, 0, sizeof netdev_options);
1156 netdev_options.name = odp_port->devname;
1157 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1158 netdev_options.may_open = true;
1160 error = netdev_open(&netdev_options, &netdev);
1162 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1163 "cannot be opened (%s)",
1164 odp_port->devname, odp_port->port,
1165 odp_port->devname, strerror(error));
1169 ofport = xmalloc(sizeof *ofport);
1170 ofport->netdev = netdev;
1171 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1172 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1173 memcpy(ofport->opp.name, odp_port->devname,
1174 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1175 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1177 netdev_get_flags(netdev, &flags);
1178 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1180 netdev_get_carrier(netdev, &carrier);
1181 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1183 netdev_get_features(netdev,
1184 &ofport->opp.curr, &ofport->opp.advertised,
1185 &ofport->opp.supported, &ofport->opp.peer);
1190 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1192 if (port_array_get(&p->ports, odp_port->port)) {
1193 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1196 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1197 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1206 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1208 const struct ofp_phy_port *a = &a_->opp;
1209 const struct ofp_phy_port *b = &b_->opp;
1211 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1212 return (a->port_no == b->port_no
1213 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1214 && !strcmp((char *) a->name, (char *) b->name)
1215 && a->state == b->state
1216 && a->config == b->config
1217 && a->curr == b->curr
1218 && a->advertised == b->advertised
1219 && a->supported == b->supported
1220 && a->peer == b->peer);
1224 send_port_status(struct ofproto *p, const struct ofport *ofport,
1227 /* XXX Should limit the number of queued port status change messages. */
1228 struct ofconn *ofconn;
1229 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1230 struct ofp_port_status *ops;
1233 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1234 ops->reason = reason;
1235 ops->desc = ofport->opp;
1236 hton_ofp_phy_port(&ops->desc);
1237 queue_tx(b, ofconn, NULL);
1239 if (p->ofhooks->port_changed_cb) {
1240 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1245 ofport_install(struct ofproto *p, struct ofport *ofport)
1247 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1248 const char *netdev_name = (const char *) ofport->opp.name;
1250 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1251 port_array_set(&p->ports, odp_port, ofport);
1252 shash_add(&p->port_by_name, netdev_name, ofport);
1254 ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
1259 ofport_remove(struct ofproto *p, struct ofport *ofport)
1261 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1263 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1264 port_array_set(&p->ports, odp_port, NULL);
1265 shash_delete(&p->port_by_name,
1266 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1268 ofproto_sflow_del_port(p->sflow, odp_port);
1273 ofport_free(struct ofport *ofport)
1276 netdev_close(ofport->netdev);
1282 update_port(struct ofproto *p, const char *devname)
1284 struct odp_port odp_port;
1285 struct ofport *old_ofport;
1286 struct ofport *new_ofport;
1289 COVERAGE_INC(ofproto_update_port);
1291 /* Query the datapath for port information. */
1292 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1294 /* Find the old ofport. */
1295 old_ofport = shash_find_data(&p->port_by_name, devname);
1298 /* There's no port named 'devname' but there might be a port with
1299 * the same port number. This could happen if a port is deleted
1300 * and then a new one added in its place very quickly, or if a port
1301 * is renamed. In the former case we want to send an OFPPR_DELETE
1302 * and an OFPPR_ADD, and in the latter case we want to send a
1303 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1304 * the old port's ifindex against the new port, or perhaps less
1305 * reliably but more portably by comparing the old port's MAC
1306 * against the new port's MAC. However, this code isn't that smart
1307 * and always sends an OFPPR_MODIFY (XXX). */
1308 old_ofport = port_array_get(&p->ports, odp_port.port);
1310 } else if (error != ENOENT && error != ENODEV) {
1311 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1312 "%s", strerror(error));
1316 /* Create a new ofport. */
1317 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1319 /* Eliminate a few pathological cases. */
1320 if (!old_ofport && !new_ofport) {
1322 } else if (old_ofport && new_ofport) {
1323 /* Most of the 'config' bits are OpenFlow soft state, but
1324 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1325 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1326 * leaves the other bits 0.) */
1327 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1329 if (ofport_equal(old_ofport, new_ofport)) {
1330 /* False alarm--no change. */
1331 ofport_free(new_ofport);
1336 /* Now deal with the normal cases. */
1338 ofport_remove(p, old_ofport);
1341 ofport_install(p, new_ofport);
1343 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1344 (!old_ofport ? OFPPR_ADD
1345 : !new_ofport ? OFPPR_DELETE
1347 ofport_free(old_ofport);
1349 /* Update port groups. */
1350 refresh_port_groups(p);
1354 init_ports(struct ofproto *p)
1356 struct odp_port *ports;
1361 error = dpif_port_list(p->dpif, &ports, &n_ports);
1366 for (i = 0; i < n_ports; i++) {
1367 const struct odp_port *odp_port = &ports[i];
1368 if (!ofport_conflicts(p, odp_port)) {
1369 struct ofport *ofport = make_ofport(odp_port);
1371 ofport_install(p, ofport);
1376 refresh_port_groups(p);
1380 static struct ofconn *
1381 ofconn_create(struct ofproto *p, struct rconn *rconn)
1383 struct ofconn *ofconn = xmalloc(sizeof *ofconn);
1384 list_push_back(&p->all_conns, &ofconn->node);
1385 ofconn->rconn = rconn;
1386 ofconn->pktbuf = NULL;
1387 ofconn->miss_send_len = 0;
1388 ofconn->packet_in_counter = rconn_packet_counter_create ();
1389 ofconn->reply_counter = rconn_packet_counter_create ();
1394 ofconn_destroy(struct ofconn *ofconn)
1396 list_remove(&ofconn->node);
1397 rconn_destroy(ofconn->rconn);
1398 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1399 rconn_packet_counter_destroy(ofconn->reply_counter);
1400 pktbuf_destroy(ofconn->pktbuf);
1405 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1409 rconn_run(ofconn->rconn);
1411 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1412 /* Limit the number of iterations to prevent other tasks from
1414 for (iteration = 0; iteration < 50; iteration++) {
1415 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1420 fail_open_maybe_recover(p->fail_open);
1422 handle_openflow(ofconn, p, of_msg);
1423 ofpbuf_delete(of_msg);
1427 if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
1428 ofconn_destroy(ofconn);
1433 ofconn_wait(struct ofconn *ofconn)
1435 rconn_run_wait(ofconn->rconn);
1436 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1437 rconn_recv_wait(ofconn->rconn);
1439 COVERAGE_INC(ofproto_ofconn_stuck);
1443 /* Caller is responsible for initializing the 'cr' member of the returned
1445 static struct rule *
1446 rule_create(struct ofproto *ofproto, struct rule *super,
1447 const union ofp_action *actions, size_t n_actions,
1448 uint16_t idle_timeout, uint16_t hard_timeout,
1449 uint64_t flow_cookie, bool send_flow_removed)
1451 struct rule *rule = xzalloc(sizeof *rule);
1452 rule->idle_timeout = idle_timeout;
1453 rule->hard_timeout = hard_timeout;
1454 rule->flow_cookie = flow_cookie;
1455 rule->used = rule->created = time_msec();
1456 rule->send_flow_removed = send_flow_removed;
1457 rule->super = super;
1459 list_push_back(&super->list, &rule->list);
1461 list_init(&rule->list);
1463 rule->n_actions = n_actions;
1464 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1465 netflow_flow_clear(&rule->nf_flow);
1466 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1471 static struct rule *
1472 rule_from_cls_rule(const struct cls_rule *cls_rule)
1474 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1478 rule_free(struct rule *rule)
1480 free(rule->actions);
1481 free(rule->odp_actions);
1485 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1486 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1487 * through all of its subrules and revalidates them, destroying any that no
1488 * longer has a super-rule (which is probably all of them).
1490 * Before calling this function, the caller must make have removed 'rule' from
1491 * the classifier. If 'rule' is an exact-match rule, the caller is also
1492 * responsible for ensuring that it has been uninstalled from the datapath. */
1494 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1497 struct rule *subrule, *next;
1498 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1499 revalidate_rule(ofproto, subrule);
1502 list_remove(&rule->list);
1508 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1510 const union ofp_action *oa;
1511 struct actions_iterator i;
1513 if (out_port == htons(OFPP_NONE)) {
1516 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1517 oa = actions_next(&i)) {
1518 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1525 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1526 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1528 * The flow that 'packet' actually contains does not need to actually match
1529 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1530 * the packet and byte counters for 'rule' will be credited for the packet sent
1531 * out whether or not the packet actually matches 'rule'.
1533 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1534 * the caller must already have accurately composed ODP actions for it given
1535 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1536 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1537 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1538 * actions and apply them to 'packet'. */
1540 rule_execute(struct ofproto *ofproto, struct rule *rule,
1541 struct ofpbuf *packet, const flow_t *flow)
1543 const union odp_action *actions;
1545 struct odp_actions a;
1547 /* Grab or compose the ODP actions.
1549 * The special case for an exact-match 'rule' where 'flow' is not the
1550 * rule's flow is important to avoid, e.g., sending a packet out its input
1551 * port simply because the ODP actions were composed for the wrong
1553 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1554 struct rule *super = rule->super ? rule->super : rule;
1555 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1556 packet, &a, NULL, 0, NULL)) {
1559 actions = a.actions;
1560 n_actions = a.n_actions;
1562 actions = rule->odp_actions;
1563 n_actions = rule->n_odp_actions;
1566 /* Execute the ODP actions. */
1567 if (!dpif_execute(ofproto->dpif, flow->in_port,
1568 actions, n_actions, packet)) {
1569 struct odp_flow_stats stats;
1570 flow_extract_stats(flow, packet, &stats);
1571 update_stats(ofproto, rule, &stats);
1572 rule->used = time_msec();
1573 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1578 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1581 struct rule *displaced_rule;
1583 /* Insert the rule in the classifier. */
1584 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1585 if (!rule->cr.wc.wildcards) {
1586 rule_make_actions(p, rule, packet);
1589 /* Send the packet and credit it to the rule. */
1592 flow_extract(packet, 0, in_port, &flow);
1593 rule_execute(p, rule, packet, &flow);
1596 /* Install the rule in the datapath only after sending the packet, to
1597 * avoid packet reordering. */
1598 if (rule->cr.wc.wildcards) {
1599 COVERAGE_INC(ofproto_add_wc_flow);
1600 p->need_revalidate = true;
1602 rule_install(p, rule, displaced_rule);
1605 /* Free the rule that was displaced, if any. */
1606 if (displaced_rule) {
1607 rule_destroy(p, displaced_rule);
1611 static struct rule *
1612 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1615 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1616 rule->idle_timeout, rule->hard_timeout,
1618 COVERAGE_INC(ofproto_subrule_create);
1619 cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1620 : rule->cr.priority), &subrule->cr);
1621 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1627 rule_remove(struct ofproto *ofproto, struct rule *rule)
1629 if (rule->cr.wc.wildcards) {
1630 COVERAGE_INC(ofproto_del_wc_flow);
1631 ofproto->need_revalidate = true;
1633 rule_uninstall(ofproto, rule);
1635 classifier_remove(&ofproto->cls, &rule->cr);
1636 rule_destroy(ofproto, rule);
1639 /* Returns true if the actions changed, false otherwise. */
1641 rule_make_actions(struct ofproto *p, struct rule *rule,
1642 const struct ofpbuf *packet)
1644 const struct rule *super;
1645 struct odp_actions a;
1648 assert(!rule->cr.wc.wildcards);
1650 super = rule->super ? rule->super : rule;
1652 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1653 packet, &a, &rule->tags, &rule->may_install,
1654 &rule->nf_flow.output_iface);
1656 actions_len = a.n_actions * sizeof *a.actions;
1657 if (rule->n_odp_actions != a.n_actions
1658 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1659 COVERAGE_INC(ofproto_odp_unchanged);
1660 free(rule->odp_actions);
1661 rule->n_odp_actions = a.n_actions;
1662 rule->odp_actions = xmemdup(a.actions, actions_len);
1670 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1671 struct odp_flow_put *put)
1673 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1674 put->flow.key = rule->cr.flow;
1675 put->flow.actions = rule->odp_actions;
1676 put->flow.n_actions = rule->n_odp_actions;
1677 put->flow.flags = 0;
1679 return dpif_flow_put(ofproto->dpif, put);
1683 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1685 assert(!rule->cr.wc.wildcards);
1687 if (rule->may_install) {
1688 struct odp_flow_put put;
1689 if (!do_put_flow(p, rule,
1690 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1692 rule->installed = true;
1693 if (displaced_rule) {
1694 update_stats(p, displaced_rule, &put.flow.stats);
1695 rule_post_uninstall(p, displaced_rule);
1698 } else if (displaced_rule) {
1699 rule_uninstall(p, displaced_rule);
1704 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1706 if (rule->installed) {
1707 struct odp_flow_put put;
1708 COVERAGE_INC(ofproto_dp_missed);
1709 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1711 rule_install(ofproto, rule, NULL);
1716 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1718 bool actions_changed;
1719 uint16_t new_out_iface, old_out_iface;
1721 old_out_iface = rule->nf_flow.output_iface;
1722 actions_changed = rule_make_actions(ofproto, rule, NULL);
1724 if (rule->may_install) {
1725 if (rule->installed) {
1726 if (actions_changed) {
1727 struct odp_flow_put put;
1728 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
1729 | ODPPF_ZERO_STATS, &put);
1730 update_stats(ofproto, rule, &put.flow.stats);
1732 /* Temporarily set the old output iface so that NetFlow
1733 * messages have the correct output interface for the old
1735 new_out_iface = rule->nf_flow.output_iface;
1736 rule->nf_flow.output_iface = old_out_iface;
1737 rule_post_uninstall(ofproto, rule);
1738 rule->nf_flow.output_iface = new_out_iface;
1741 rule_install(ofproto, rule, NULL);
1744 rule_uninstall(ofproto, rule);
1749 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1751 uint64_t total_bytes = rule->byte_count + extra_bytes;
1753 if (ofproto->ofhooks->account_flow_cb
1754 && total_bytes > rule->accounted_bytes)
1756 ofproto->ofhooks->account_flow_cb(
1757 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1758 total_bytes - rule->accounted_bytes, ofproto->aux);
1759 rule->accounted_bytes = total_bytes;
1764 rule_uninstall(struct ofproto *p, struct rule *rule)
1766 assert(!rule->cr.wc.wildcards);
1767 if (rule->installed) {
1768 struct odp_flow odp_flow;
1770 odp_flow.key = rule->cr.flow;
1771 odp_flow.actions = NULL;
1772 odp_flow.n_actions = 0;
1774 if (!dpif_flow_del(p->dpif, &odp_flow)) {
1775 update_stats(p, rule, &odp_flow.stats);
1777 rule->installed = false;
1779 rule_post_uninstall(p, rule);
1784 is_controller_rule(struct rule *rule)
1786 /* If the only action is send to the controller then don't report
1787 * NetFlow expiration messages since it is just part of the control
1788 * logic for the network and not real traffic. */
1790 if (rule && rule->super) {
1791 struct rule *super = rule->super;
1793 return super->n_actions == 1 &&
1794 super->actions[0].type == htons(OFPAT_OUTPUT) &&
1795 super->actions[0].output.port == htons(OFPP_CONTROLLER);
1802 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1804 struct rule *super = rule->super;
1806 rule_account(ofproto, rule, 0);
1808 if (ofproto->netflow && !is_controller_rule(rule)) {
1809 struct ofexpired expired;
1810 expired.flow = rule->cr.flow;
1811 expired.packet_count = rule->packet_count;
1812 expired.byte_count = rule->byte_count;
1813 expired.used = rule->used;
1814 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
1817 super->packet_count += rule->packet_count;
1818 super->byte_count += rule->byte_count;
1820 /* Reset counters to prevent double counting if the rule ever gets
1822 rule->packet_count = 0;
1823 rule->byte_count = 0;
1824 rule->accounted_bytes = 0;
1826 netflow_flow_clear(&rule->nf_flow);
1831 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1832 struct rconn_packet_counter *counter)
1834 update_openflow_length(msg);
1835 if (rconn_send(ofconn->rconn, msg, counter)) {
1841 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1842 int error, const void *data, size_t len)
1845 struct ofp_error_msg *oem;
1847 if (!(error >> 16)) {
1848 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1853 COVERAGE_INC(ofproto_error);
1854 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1855 oh ? oh->xid : 0, &buf);
1856 oem->type = htons((unsigned int) error >> 16);
1857 oem->code = htons(error & 0xffff);
1858 memcpy(oem->data, data, len);
1859 queue_tx(buf, ofconn, ofconn->reply_counter);
1863 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1866 size_t oh_length = ntohs(oh->length);
1867 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1871 hton_ofp_phy_port(struct ofp_phy_port *opp)
1873 opp->port_no = htons(opp->port_no);
1874 opp->config = htonl(opp->config);
1875 opp->state = htonl(opp->state);
1876 opp->curr = htonl(opp->curr);
1877 opp->advertised = htonl(opp->advertised);
1878 opp->supported = htonl(opp->supported);
1879 opp->peer = htonl(opp->peer);
1883 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
1885 struct ofp_header *rq = oh;
1886 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
1891 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
1892 struct ofp_header *oh)
1894 struct ofp_switch_features *osf;
1896 unsigned int port_no;
1897 struct ofport *port;
1899 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1900 osf->datapath_id = htonll(p->datapath_id);
1901 osf->n_buffers = htonl(pktbuf_capacity());
1903 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1904 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
1905 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1906 (1u << OFPAT_SET_VLAN_VID) |
1907 (1u << OFPAT_SET_VLAN_PCP) |
1908 (1u << OFPAT_STRIP_VLAN) |
1909 (1u << OFPAT_SET_DL_SRC) |
1910 (1u << OFPAT_SET_DL_DST) |
1911 (1u << OFPAT_SET_NW_SRC) |
1912 (1u << OFPAT_SET_NW_DST) |
1913 (1u << OFPAT_SET_NW_TOS) |
1914 (1u << OFPAT_SET_TP_SRC) |
1915 (1u << OFPAT_SET_TP_DST));
1917 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1918 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1921 queue_tx(buf, ofconn, ofconn->reply_counter);
1926 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1927 struct ofp_header *oh)
1930 struct ofp_switch_config *osc;
1934 /* Figure out flags. */
1935 dpif_get_drop_frags(p->dpif, &drop_frags);
1936 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1939 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1940 osc->flags = htons(flags);
1941 osc->miss_send_len = htons(ofconn->miss_send_len);
1942 queue_tx(buf, ofconn, ofconn->reply_counter);
1948 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1949 struct ofp_switch_config *osc)
1954 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1958 flags = ntohs(osc->flags);
1960 if (ofconn == p->controller) {
1961 switch (flags & OFPC_FRAG_MASK) {
1962 case OFPC_FRAG_NORMAL:
1963 dpif_set_drop_frags(p->dpif, false);
1965 case OFPC_FRAG_DROP:
1966 dpif_set_drop_frags(p->dpif, true);
1969 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1975 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1976 if (ntohs(osc->miss_send_len) != 0) {
1977 ofconn->pktbuf = pktbuf_create();
1979 pktbuf_destroy(ofconn->pktbuf);
1983 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1989 add_output_group_action(struct odp_actions *actions, uint16_t group,
1990 uint16_t *nf_output_iface)
1992 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1994 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
1995 *nf_output_iface = NF_OUT_FLOOD;
2000 add_controller_action(struct odp_actions *actions,
2001 const struct ofp_action_output *oao)
2003 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2004 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
2007 struct action_xlate_ctx {
2009 flow_t flow; /* Flow to which these actions correspond. */
2010 int recurse; /* Recursion level, via xlate_table_action. */
2011 struct ofproto *ofproto;
2012 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2013 * null pointer if we are revalidating
2014 * without a packet to refer to. */
2017 struct odp_actions *out; /* Datapath actions. */
2018 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2019 bool may_set_up_flow; /* True ordinarily; false if the actions must
2020 * be reassessed for every packet. */
2021 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2024 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2025 struct action_xlate_ctx *ctx);
2028 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2030 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
2033 if (ofport->opp.config & OFPPC_NO_FWD) {
2034 /* Forwarding disabled on port. */
2039 * We don't have an ofport record for this port, but it doesn't hurt to
2040 * allow forwarding to it anyhow. Maybe such a port will appear later
2041 * and we're pre-populating the flow table.
2045 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2046 ctx->nf_output_iface = port;
2049 static struct rule *
2050 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2053 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2055 /* The rule we found might not be valid, since we could be in need of
2056 * revalidation. If it is not valid, don't return it. */
2059 && ofproto->need_revalidate
2060 && !revalidate_rule(ofproto, rule)) {
2061 COVERAGE_INC(ofproto_invalidated);
2069 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2071 if (!ctx->recurse) {
2072 uint16_t old_in_port;
2075 /* Look up a flow with 'in_port' as the input port. Then restore the
2076 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2077 * have surprising behavior). */
2078 old_in_port = ctx->flow.in_port;
2079 ctx->flow.in_port = in_port;
2080 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2081 ctx->flow.in_port = old_in_port;
2089 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2096 xlate_output_action(struct action_xlate_ctx *ctx,
2097 const struct ofp_action_output *oao)
2100 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2102 ctx->nf_output_iface = NF_OUT_DROP;
2104 switch (ntohs(oao->port)) {
2106 add_output_action(ctx, ctx->flow.in_port);
2109 xlate_table_action(ctx, ctx->flow.in_port);
2112 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2113 ctx->out, ctx->tags,
2114 &ctx->nf_output_iface,
2115 ctx->ofproto->aux)) {
2116 COVERAGE_INC(ofproto_uninstallable);
2117 ctx->may_set_up_flow = false;
2121 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2122 &ctx->nf_output_iface);
2125 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2127 case OFPP_CONTROLLER:
2128 add_controller_action(ctx->out, oao);
2131 add_output_action(ctx, ODPP_LOCAL);
2134 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2135 if (odp_port != ctx->flow.in_port) {
2136 add_output_action(ctx, odp_port);
2141 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2142 ctx->nf_output_iface = NF_OUT_FLOOD;
2143 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2144 ctx->nf_output_iface = prev_nf_output_iface;
2145 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2146 ctx->nf_output_iface != NF_OUT_FLOOD) {
2147 ctx->nf_output_iface = NF_OUT_MULTI;
2152 xlate_nicira_action(struct action_xlate_ctx *ctx,
2153 const struct nx_action_header *nah)
2155 const struct nx_action_resubmit *nar;
2156 const struct nx_action_set_tunnel *nast;
2157 union odp_action *oa;
2158 int subtype = ntohs(nah->subtype);
2160 assert(nah->vendor == htonl(NX_VENDOR_ID));
2162 case NXAST_RESUBMIT:
2163 nar = (const struct nx_action_resubmit *) nah;
2164 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2167 case NXAST_SET_TUNNEL:
2168 nast = (const struct nx_action_set_tunnel *) nah;
2169 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2170 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2173 /* If you add a new action here that modifies flow data, don't forget to
2174 * update the flow key in ctx->flow in the same key. */
2177 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2183 do_xlate_actions(const union ofp_action *in, size_t n_in,
2184 struct action_xlate_ctx *ctx)
2186 struct actions_iterator iter;
2187 const union ofp_action *ia;
2188 const struct ofport *port;
2190 port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port);
2191 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2192 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, stp_eth_addr)
2193 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2194 /* Drop this flow. */
2198 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2199 uint16_t type = ntohs(ia->type);
2200 union odp_action *oa;
2204 xlate_output_action(ctx, &ia->output);
2207 case OFPAT_SET_VLAN_VID:
2208 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2209 ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2212 case OFPAT_SET_VLAN_PCP:
2213 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2214 ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2217 case OFPAT_STRIP_VLAN:
2218 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2219 ctx->flow.dl_vlan = OFP_VLAN_NONE;
2220 ctx->flow.dl_vlan_pcp = 0;
2223 case OFPAT_SET_DL_SRC:
2224 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2225 memcpy(oa->dl_addr.dl_addr,
2226 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2227 memcpy(ctx->flow.dl_src,
2228 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2231 case OFPAT_SET_DL_DST:
2232 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2233 memcpy(oa->dl_addr.dl_addr,
2234 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2235 memcpy(ctx->flow.dl_dst,
2236 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2239 case OFPAT_SET_NW_SRC:
2240 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2241 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2244 case OFPAT_SET_NW_DST:
2245 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2246 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2249 case OFPAT_SET_NW_TOS:
2250 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2251 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2254 case OFPAT_SET_TP_SRC:
2255 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2256 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2259 case OFPAT_SET_TP_DST:
2260 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2261 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2265 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2269 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2276 xlate_actions(const union ofp_action *in, size_t n_in,
2277 const flow_t *flow, struct ofproto *ofproto,
2278 const struct ofpbuf *packet,
2279 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2280 uint16_t *nf_output_iface)
2282 tag_type no_tags = 0;
2283 struct action_xlate_ctx ctx;
2284 COVERAGE_INC(ofproto_ofp2odp);
2285 odp_actions_init(out);
2288 ctx.ofproto = ofproto;
2289 ctx.packet = packet;
2291 ctx.tags = tags ? tags : &no_tags;
2292 ctx.may_set_up_flow = true;
2293 ctx.nf_output_iface = NF_OUT_DROP;
2294 do_xlate_actions(in, n_in, &ctx);
2296 /* Check with in-band control to see if we're allowed to set up this
2298 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2299 ctx.may_set_up_flow = false;
2302 if (may_set_up_flow) {
2303 *may_set_up_flow = ctx.may_set_up_flow;
2305 if (nf_output_iface) {
2306 *nf_output_iface = ctx.nf_output_iface;
2308 if (odp_actions_overflow(out)) {
2309 odp_actions_init(out);
2310 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2316 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2317 struct ofp_header *oh)
2319 struct ofp_packet_out *opo;
2320 struct ofpbuf payload, *buffer;
2321 struct odp_actions actions;
2327 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2331 opo = (struct ofp_packet_out *) oh;
2333 COVERAGE_INC(ofproto_packet_out);
2334 if (opo->buffer_id != htonl(UINT32_MAX)) {
2335 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2337 if (error || !buffer) {
2345 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2346 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2347 &flow, p, &payload, &actions, NULL, NULL, NULL);
2352 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2354 ofpbuf_delete(buffer);
2360 update_port_config(struct ofproto *p, struct ofport *port,
2361 uint32_t config, uint32_t mask)
2363 mask &= config ^ port->opp.config;
2364 if (mask & OFPPC_PORT_DOWN) {
2365 if (config & OFPPC_PORT_DOWN) {
2366 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2368 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2371 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2372 if (mask & REVALIDATE_BITS) {
2373 COVERAGE_INC(ofproto_costly_flags);
2374 port->opp.config ^= mask & REVALIDATE_BITS;
2375 p->need_revalidate = true;
2377 #undef REVALIDATE_BITS
2378 if (mask & OFPPC_NO_FLOOD) {
2379 port->opp.config ^= OFPPC_NO_FLOOD;
2380 refresh_port_groups(p);
2382 if (mask & OFPPC_NO_PACKET_IN) {
2383 port->opp.config ^= OFPPC_NO_PACKET_IN;
2388 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2390 const struct ofp_port_mod *opm;
2391 struct ofport *port;
2394 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2398 opm = (struct ofp_port_mod *) oh;
2400 port = port_array_get(&p->ports,
2401 ofp_port_to_odp_port(ntohs(opm->port_no)));
2403 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2404 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2405 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2407 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2408 if (opm->advertise) {
2409 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2415 static struct ofpbuf *
2416 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2418 struct ofp_stats_reply *osr;
2421 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2422 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2424 osr->flags = htons(0);
2428 static struct ofpbuf *
2429 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2431 return make_stats_reply(request->header.xid, request->type, body_len);
2435 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2437 struct ofpbuf *msg = *msgp;
2438 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2439 if (nbytes + msg->size > UINT16_MAX) {
2440 struct ofp_stats_reply *reply = msg->data;
2441 reply->flags = htons(OFPSF_REPLY_MORE);
2442 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2443 queue_tx(msg, ofconn, ofconn->reply_counter);
2445 return ofpbuf_put_uninit(*msgp, nbytes);
2449 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2450 struct ofp_stats_request *request)
2452 struct ofp_desc_stats *ods;
2455 msg = start_stats_reply(request, sizeof *ods);
2456 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2457 memset(ods, 0, sizeof *ods);
2458 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2459 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2460 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2461 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2462 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2463 queue_tx(msg, ofconn, ofconn->reply_counter);
2469 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2471 struct rule *rule = rule_from_cls_rule(cls_rule);
2472 int *n_subrules = n_subrules_;
2480 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2481 struct ofp_stats_request *request)
2483 struct ofp_table_stats *ots;
2485 struct odp_stats dpstats;
2486 int n_exact, n_subrules, n_wild;
2488 msg = start_stats_reply(request, sizeof *ots * 2);
2490 /* Count rules of various kinds. */
2492 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2493 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2494 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2497 dpif_get_dp_stats(p->dpif, &dpstats);
2498 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2499 memset(ots, 0, sizeof *ots);
2500 ots->table_id = TABLEID_HASH;
2501 strcpy(ots->name, "hash");
2502 ots->wildcards = htonl(0);
2503 ots->max_entries = htonl(dpstats.max_capacity);
2504 ots->active_count = htonl(n_exact);
2505 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2507 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2509 /* Classifier table. */
2510 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2511 memset(ots, 0, sizeof *ots);
2512 ots->table_id = TABLEID_CLASSIFIER;
2513 strcpy(ots->name, "classifier");
2514 ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
2516 ots->max_entries = htonl(65536);
2517 ots->active_count = htonl(n_wild);
2518 ots->lookup_count = htonll(0); /* XXX */
2519 ots->matched_count = htonll(0); /* XXX */
2521 queue_tx(msg, ofconn, ofconn->reply_counter);
2526 append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
2529 struct netdev_stats stats;
2530 struct ofp_port_stats *ops;
2532 /* Intentionally ignore return value, since errors will set
2533 * 'stats' to all-1s, which is correct for OpenFlow, and
2534 * netdev_get_stats() will log errors. */
2535 netdev_get_stats(port->netdev, &stats);
2537 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2538 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2539 memset(ops->pad, 0, sizeof ops->pad);
2540 ops->rx_packets = htonll(stats.rx_packets);
2541 ops->tx_packets = htonll(stats.tx_packets);
2542 ops->rx_bytes = htonll(stats.rx_bytes);
2543 ops->tx_bytes = htonll(stats.tx_bytes);
2544 ops->rx_dropped = htonll(stats.rx_dropped);
2545 ops->tx_dropped = htonll(stats.tx_dropped);
2546 ops->rx_errors = htonll(stats.rx_errors);
2547 ops->tx_errors = htonll(stats.tx_errors);
2548 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2549 ops->rx_over_err = htonll(stats.rx_over_errors);
2550 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2551 ops->collisions = htonll(stats.collisions);
2555 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2556 struct ofp_stats_request *osr,
2559 struct ofp_port_stats_request *psr;
2560 struct ofp_port_stats *ops;
2562 struct ofport *port;
2563 unsigned int port_no;
2565 if (arg_size != sizeof *psr) {
2566 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2568 psr = (struct ofp_port_stats_request *) osr->body;
2570 msg = start_stats_reply(osr, sizeof *ops * 16);
2571 if (psr->port_no != htons(OFPP_NONE)) {
2572 port = port_array_get(&p->ports,
2573 ofp_port_to_odp_port(ntohs(psr->port_no)));
2575 append_port_stat(port, ntohs(psr->port_no), ofconn, msg);
2578 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2579 append_port_stat(port, port_no, ofconn, msg);
2583 queue_tx(msg, ofconn, ofconn->reply_counter);
2587 struct flow_stats_cbdata {
2588 struct ofproto *ofproto;
2589 struct ofconn *ofconn;
2594 /* Obtains statistic counters for 'rule' within 'p' and stores them into
2595 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
2596 * returned statistic include statistics for all of 'rule''s subrules. */
2598 query_stats(struct ofproto *p, struct rule *rule,
2599 uint64_t *packet_countp, uint64_t *byte_countp)
2601 uint64_t packet_count, byte_count;
2602 struct rule *subrule;
2603 struct odp_flow *odp_flows;
2606 /* Start from historical data for 'rule' itself that are no longer tracked
2607 * by the datapath. This counts, for example, subrules that have
2609 packet_count = rule->packet_count;
2610 byte_count = rule->byte_count;
2612 /* Prepare to ask the datapath for statistics on 'rule', or if it is
2613 * wildcarded then on all of its subrules.
2615 * Also, add any statistics that are not tracked by the datapath for each
2616 * subrule. This includes, for example, statistics for packets that were
2617 * executed "by hand" by ofproto via dpif_execute() but must be accounted
2619 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2620 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
2621 if (rule->cr.wc.wildcards) {
2623 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2624 odp_flows[i++].key = subrule->cr.flow;
2625 packet_count += subrule->packet_count;
2626 byte_count += subrule->byte_count;
2629 odp_flows[0].key = rule->cr.flow;
2632 /* Fetch up-to-date statistics from the datapath and add them in. */
2633 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2635 for (i = 0; i < n_odp_flows; i++) {
2636 struct odp_flow *odp_flow = &odp_flows[i];
2637 packet_count += odp_flow->stats.n_packets;
2638 byte_count += odp_flow->stats.n_bytes;
2643 /* Return the stats to the caller. */
2644 *packet_countp = packet_count;
2645 *byte_countp = byte_count;
2649 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2651 struct rule *rule = rule_from_cls_rule(rule_);
2652 struct flow_stats_cbdata *cbdata = cbdata_;
2653 struct ofp_flow_stats *ofs;
2654 uint64_t packet_count, byte_count;
2655 size_t act_len, len;
2656 long long int tdiff = time_msec() - rule->created;
2657 uint32_t sec = tdiff / 1000;
2658 uint32_t msec = tdiff - (sec * 1000);
2660 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2664 act_len = sizeof *rule->actions * rule->n_actions;
2665 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2667 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2669 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2670 ofs->length = htons(len);
2671 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2673 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
2674 cbdata->ofproto->tun_id_from_cookie, &ofs->match);
2675 ofs->duration_sec = htonl(sec);
2676 ofs->duration_nsec = htonl(msec * 1000000);
2677 ofs->cookie = rule->flow_cookie;
2678 ofs->priority = htons(rule->cr.priority);
2679 ofs->idle_timeout = htons(rule->idle_timeout);
2680 ofs->hard_timeout = htons(rule->hard_timeout);
2681 memset(ofs->pad2, 0, sizeof ofs->pad2);
2682 ofs->packet_count = htonll(packet_count);
2683 ofs->byte_count = htonll(byte_count);
2684 memcpy(ofs->actions, rule->actions, act_len);
2688 table_id_to_include(uint8_t table_id)
2690 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2691 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2692 : table_id == 0xff ? CLS_INC_ALL
2697 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2698 const struct ofp_stats_request *osr,
2701 struct ofp_flow_stats_request *fsr;
2702 struct flow_stats_cbdata cbdata;
2703 struct cls_rule target;
2705 if (arg_size != sizeof *fsr) {
2706 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2708 fsr = (struct ofp_flow_stats_request *) osr->body;
2710 COVERAGE_INC(ofproto_flows_req);
2712 cbdata.ofconn = ofconn;
2713 cbdata.out_port = fsr->out_port;
2714 cbdata.msg = start_stats_reply(osr, 1024);
2715 cls_rule_from_match(&fsr->match, 0, false, 0, &target);
2716 classifier_for_each_match(&p->cls, &target,
2717 table_id_to_include(fsr->table_id),
2718 flow_stats_cb, &cbdata);
2719 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2723 struct flow_stats_ds_cbdata {
2724 struct ofproto *ofproto;
2729 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
2731 struct rule *rule = rule_from_cls_rule(rule_);
2732 struct flow_stats_ds_cbdata *cbdata = cbdata_;
2733 struct ds *results = cbdata->results;
2734 struct ofp_match match;
2735 uint64_t packet_count, byte_count;
2736 size_t act_len = sizeof *rule->actions * rule->n_actions;
2738 /* Don't report on subrules. */
2739 if (rule->super != NULL) {
2743 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2744 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
2745 cbdata->ofproto->tun_id_from_cookie, &match);
2747 ds_put_format(results, "duration=%llds, ",
2748 (time_msec() - rule->created) / 1000);
2749 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2750 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2751 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2752 ofp_print_match(results, &match, true);
2753 ofp_print_actions(results, &rule->actions->header, act_len);
2754 ds_put_cstr(results, "\n");
2757 /* Adds a pretty-printed description of all flows to 'results', including
2758 * those marked hidden by secchan (e.g., by in-band control). */
2760 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2762 struct ofp_match match;
2763 struct cls_rule target;
2764 struct flow_stats_ds_cbdata cbdata;
2766 memset(&match, 0, sizeof match);
2767 match.wildcards = htonl(OVSFW_ALL);
2770 cbdata.results = results;
2772 cls_rule_from_match(&match, 0, false, 0, &target);
2773 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2774 flow_stats_ds_cb, &cbdata);
2777 struct aggregate_stats_cbdata {
2778 struct ofproto *ofproto;
2780 uint64_t packet_count;
2781 uint64_t byte_count;
2786 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2788 struct rule *rule = rule_from_cls_rule(rule_);
2789 struct aggregate_stats_cbdata *cbdata = cbdata_;
2790 uint64_t packet_count, byte_count;
2792 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2796 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2798 cbdata->packet_count += packet_count;
2799 cbdata->byte_count += byte_count;
2804 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2805 const struct ofp_stats_request *osr,
2808 struct ofp_aggregate_stats_request *asr;
2809 struct ofp_aggregate_stats_reply *reply;
2810 struct aggregate_stats_cbdata cbdata;
2811 struct cls_rule target;
2814 if (arg_size != sizeof *asr) {
2815 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2817 asr = (struct ofp_aggregate_stats_request *) osr->body;
2819 COVERAGE_INC(ofproto_agg_request);
2821 cbdata.out_port = asr->out_port;
2822 cbdata.packet_count = 0;
2823 cbdata.byte_count = 0;
2825 cls_rule_from_match(&asr->match, 0, false, 0, &target);
2826 classifier_for_each_match(&p->cls, &target,
2827 table_id_to_include(asr->table_id),
2828 aggregate_stats_cb, &cbdata);
2830 msg = start_stats_reply(osr, sizeof *reply);
2831 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2832 reply->flow_count = htonl(cbdata.n_flows);
2833 reply->packet_count = htonll(cbdata.packet_count);
2834 reply->byte_count = htonll(cbdata.byte_count);
2835 queue_tx(msg, ofconn, ofconn->reply_counter);
2840 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2841 struct ofp_header *oh)
2843 struct ofp_stats_request *osr;
2847 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2852 osr = (struct ofp_stats_request *) oh;
2854 switch (ntohs(osr->type)) {
2856 return handle_desc_stats_request(p, ofconn, osr);
2859 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2861 case OFPST_AGGREGATE:
2862 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2865 return handle_table_stats_request(p, ofconn, osr);
2868 return handle_port_stats_request(p, ofconn, osr, arg_size);
2871 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2874 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2878 static long long int
2879 msec_from_nsec(uint64_t sec, uint32_t nsec)
2881 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2885 update_time(struct ofproto *ofproto, struct rule *rule,
2886 const struct odp_flow_stats *stats)
2888 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2889 if (used > rule->used) {
2891 if (rule->super && used > rule->super->used) {
2892 rule->super->used = used;
2894 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
2899 update_stats(struct ofproto *ofproto, struct rule *rule,
2900 const struct odp_flow_stats *stats)
2902 if (stats->n_packets) {
2903 update_time(ofproto, rule, stats);
2904 rule->packet_count += stats->n_packets;
2905 rule->byte_count += stats->n_bytes;
2906 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
2911 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
2912 * in which no matching flow already exists in the flow table.
2914 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
2915 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
2916 * code as encoded by ofp_mkerr() on failure.
2918 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2921 add_flow(struct ofproto *p, struct ofconn *ofconn,
2922 const struct ofp_flow_mod *ofm, size_t n_actions)
2924 struct ofpbuf *packet;
2929 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
2933 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
2935 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
2936 ntohs(ofm->priority))) {
2937 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
2941 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
2942 n_actions, ntohs(ofm->idle_timeout),
2943 ntohs(ofm->hard_timeout), ofm->cookie,
2944 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
2945 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
2946 p->tun_id_from_cookie, ofm->cookie, &rule->cr);
2949 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2950 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2954 in_port = UINT16_MAX;
2957 rule_insert(p, rule, packet, in_port);
2958 ofpbuf_delete(packet);
2962 static struct rule *
2963 find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
2968 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
2970 return rule_from_cls_rule(classifier_find_rule_exactly(
2971 &p->cls, &flow, wildcards,
2972 ntohs(ofm->priority)));
2976 send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
2977 struct rule *rule, const struct ofp_flow_mod *ofm)
2979 struct ofpbuf *packet;
2984 if (ofm->buffer_id == htonl(UINT32_MAX)) {
2988 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2994 flow_extract(packet, 0, in_port, &flow);
2995 rule_execute(ofproto, rule, packet, &flow);
2996 ofpbuf_delete(packet);
3001 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3003 struct modify_flows_cbdata {
3004 struct ofproto *ofproto;
3005 const struct ofp_flow_mod *ofm;
3010 static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
3011 size_t n_actions, struct rule *);
3012 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
3014 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3015 * encoded by ofp_mkerr() on failure.
3017 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3020 modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
3021 const struct ofp_flow_mod *ofm, size_t n_actions)
3023 struct modify_flows_cbdata cbdata;
3024 struct cls_rule target;
3028 cbdata.n_actions = n_actions;
3029 cbdata.match = NULL;
3031 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3034 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3035 modify_flows_cb, &cbdata);
3037 /* This credits the packet to whichever flow happened to happened to
3038 * match last. That's weird. Maybe we should do a lookup for the
3039 * flow that actually matches the packet? Who knows. */
3040 send_buffered_packet(p, ofconn, cbdata.match, ofm);
3043 return add_flow(p, ofconn, ofm, n_actions);
3047 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3048 * code as encoded by ofp_mkerr() on failure.
3050 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3053 modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
3054 struct ofp_flow_mod *ofm, size_t n_actions)
3056 struct rule *rule = find_flow_strict(p, ofm);
3057 if (rule && !rule_is_hidden(rule)) {
3058 modify_flow(p, ofm, n_actions, rule);
3059 return send_buffered_packet(p, ofconn, rule, ofm);
3061 return add_flow(p, ofconn, ofm, n_actions);
3065 /* Callback for modify_flows_loose(). */
3067 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3069 struct rule *rule = rule_from_cls_rule(rule_);
3070 struct modify_flows_cbdata *cbdata = cbdata_;
3072 if (!rule_is_hidden(rule)) {
3073 cbdata->match = rule;
3074 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
3078 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3079 * been identified as a flow in 'p''s flow table to be modified, by changing
3080 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3081 * ofp_action[] structures). */
3083 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
3084 size_t n_actions, struct rule *rule)
3086 size_t actions_len = n_actions * sizeof *rule->actions;
3088 rule->flow_cookie = ofm->cookie;
3090 /* If the actions are the same, do nothing. */
3091 if (n_actions == rule->n_actions
3092 && !memcmp(ofm->actions, rule->actions, actions_len))
3097 /* Replace actions. */
3098 free(rule->actions);
3099 rule->actions = xmemdup(ofm->actions, actions_len);
3100 rule->n_actions = n_actions;
3102 /* Make sure that the datapath gets updated properly. */
3103 if (rule->cr.wc.wildcards) {
3104 COVERAGE_INC(ofproto_mod_wc_flow);
3105 p->need_revalidate = true;
3107 rule_update_actions(p, rule);
3113 /* OFPFC_DELETE implementation. */
3115 struct delete_flows_cbdata {
3116 struct ofproto *ofproto;
3120 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3121 static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port);
3123 /* Implements OFPFC_DELETE. */
3125 delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
3127 struct delete_flows_cbdata cbdata;
3128 struct cls_rule target;
3131 cbdata.out_port = ofm->out_port;
3133 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3136 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3137 delete_flows_cb, &cbdata);
3140 /* Implements OFPFC_DELETE_STRICT. */
3142 delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
3144 struct rule *rule = find_flow_strict(p, ofm);
3146 delete_flow(p, rule, ofm->out_port);
3150 /* Callback for delete_flows_loose(). */
3152 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3154 struct rule *rule = rule_from_cls_rule(rule_);
3155 struct delete_flows_cbdata *cbdata = cbdata_;
3157 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
3160 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3161 * been identified as a flow to delete from 'p''s flow table, by deleting the
3162 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3165 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3166 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3167 * specified 'out_port'. */
3169 delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port)
3171 if (rule_is_hidden(rule)) {
3175 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3179 send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
3180 rule_remove(p, rule);
3184 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3185 struct ofp_flow_mod *ofm)
3190 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3191 sizeof *ofm->actions, &n_actions);
3196 /* We do not support the emergency flow cache. It will hopefully
3197 * get dropped from OpenFlow in the near future. */
3198 if (ofm->flags & htons(OFPFF_EMERG)) {
3199 /* There isn't a good fit for an error code, so just state that the
3200 * flow table is full. */
3201 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3204 normalize_match(&ofm->match);
3205 if (!ofm->match.wildcards) {
3206 ofm->priority = htons(UINT16_MAX);
3209 error = validate_actions((const union ofp_action *) ofm->actions,
3210 n_actions, p->max_ports);
3215 switch (ntohs(ofm->command)) {
3217 return add_flow(p, ofconn, ofm, n_actions);
3220 return modify_flows_loose(p, ofconn, ofm, n_actions);
3222 case OFPFC_MODIFY_STRICT:
3223 return modify_flow_strict(p, ofconn, ofm, n_actions);
3226 delete_flows_loose(p, ofm);
3229 case OFPFC_DELETE_STRICT:
3230 delete_flow_strict(p, ofm);
3234 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3239 handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg)
3243 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
3248 p->tun_id_from_cookie = !!msg->set;
3253 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3255 struct ofp_vendor_header *ovh = msg;
3256 struct nicira_header *nh;
3258 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3259 VLOG_WARN_RL(&rl, "received vendor message of length %zu "
3260 "(expected at least %zu)",
3261 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
3262 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3264 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3265 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3267 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3268 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %zu "
3269 "(expected at least %zu)",
3270 ntohs(ovh->header.length), sizeof(struct nicira_header));
3271 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3275 switch (ntohl(nh->subtype)) {
3276 case NXT_STATUS_REQUEST:
3277 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3280 case NXT_TUN_ID_FROM_COOKIE:
3281 return handle_tun_id_from_cookie(p, msg);
3284 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3288 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
3290 struct ofp_header *ob;
3293 /* Currently, everything executes synchronously, so we can just
3294 * immediately send the barrier reply. */
3295 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3296 queue_tx(buf, ofconn, ofconn->reply_counter);
3301 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3302 struct ofpbuf *ofp_msg)
3304 struct ofp_header *oh = ofp_msg->data;
3307 COVERAGE_INC(ofproto_recv_openflow);
3309 case OFPT_ECHO_REQUEST:
3310 error = handle_echo_request(ofconn, oh);
3313 case OFPT_ECHO_REPLY:
3317 case OFPT_FEATURES_REQUEST:
3318 error = handle_features_request(p, ofconn, oh);
3321 case OFPT_GET_CONFIG_REQUEST:
3322 error = handle_get_config_request(p, ofconn, oh);
3325 case OFPT_SET_CONFIG:
3326 error = handle_set_config(p, ofconn, ofp_msg->data);
3329 case OFPT_PACKET_OUT:
3330 error = handle_packet_out(p, ofconn, ofp_msg->data);
3334 error = handle_port_mod(p, oh);
3338 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3341 case OFPT_STATS_REQUEST:
3342 error = handle_stats_request(p, ofconn, oh);
3346 error = handle_vendor(p, ofconn, ofp_msg->data);
3349 case OFPT_BARRIER_REQUEST:
3350 error = handle_barrier_request(ofconn, oh);
3354 if (VLOG_IS_WARN_ENABLED()) {
3355 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3356 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3359 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3364 send_error_oh(ofconn, ofp_msg->data, error);
3369 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
3371 struct odp_msg *msg = packet->data;
3372 uint16_t in_port = odp_port_to_ofp_port(msg->port);
3374 struct ofpbuf payload;
3377 payload.data = msg + 1;
3378 payload.size = msg->length - sizeof *msg;
3379 flow_extract(&payload, msg->arg, msg->port, &flow);
3381 /* Check with in-band control to see if this packet should be sent
3382 * to the local port regardless of the flow table. */
3383 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3384 union odp_action action;
3386 memset(&action, 0, sizeof(action));
3387 action.output.type = ODPAT_OUTPUT;
3388 action.output.port = ODPP_LOCAL;
3389 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3392 rule = lookup_valid_rule(p, &flow);
3394 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3395 struct ofport *port = port_array_get(&p->ports, msg->port);
3397 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3398 COVERAGE_INC(ofproto_no_packet_in);
3399 /* XXX install 'drop' flow entry */
3400 ofpbuf_delete(packet);
3404 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3407 COVERAGE_INC(ofproto_packet_in);
3408 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3412 if (rule->cr.wc.wildcards) {
3413 rule = rule_create_subrule(p, rule, &flow);
3414 rule_make_actions(p, rule, packet);
3416 if (!rule->may_install) {
3417 /* The rule is not installable, that is, we need to process every
3418 * packet, so process the current packet and set its actions into
3420 rule_make_actions(p, rule, packet);
3422 /* XXX revalidate rule if it needs it */
3426 rule_execute(p, rule, &payload, &flow);
3427 rule_reinstall(p, rule);
3429 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY
3430 && rconn_is_connected(p->controller->rconn)) {
3432 * Extra-special case for fail-open mode.
3434 * We are in fail-open mode and the packet matched the fail-open rule,
3435 * but we are connected to a controller too. We should send the packet
3436 * up to the controller in the hope that it will try to set up a flow
3437 * and thereby allow us to exit fail-open.
3439 * See the top-level comment in fail-open.c for more information.
3441 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3443 ofpbuf_delete(packet);
3448 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3450 struct odp_msg *msg = packet->data;
3452 switch (msg->type) {
3453 case _ODPL_ACTION_NR:
3454 COVERAGE_INC(ofproto_ctlr_action);
3455 pinsched_send(p->action_sched, odp_port_to_ofp_port(msg->port), packet,
3456 send_packet_in_action, p);
3459 case _ODPL_SFLOW_NR:
3461 ofproto_sflow_received(p->sflow, msg);
3463 ofpbuf_delete(packet);
3467 handle_odp_miss_msg(p, packet);
3471 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
3478 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3480 struct rule *sub = rule_from_cls_rule(sub_);
3481 struct revalidate_cbdata *cbdata = cbdata_;
3483 if (cbdata->revalidate_all
3484 || (cbdata->revalidate_subrules && sub->super)
3485 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3486 revalidate_rule(cbdata->ofproto, sub);
3491 revalidate_rule(struct ofproto *p, struct rule *rule)
3493 const flow_t *flow = &rule->cr.flow;
3495 COVERAGE_INC(ofproto_revalidate_rule);
3498 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3500 rule_remove(p, rule);
3502 } else if (super != rule->super) {
3503 COVERAGE_INC(ofproto_revalidate_moved);
3504 list_remove(&rule->list);
3505 list_push_back(&super->list, &rule->list);
3506 rule->super = super;
3507 rule->hard_timeout = super->hard_timeout;
3508 rule->idle_timeout = super->idle_timeout;
3509 rule->created = super->created;
3514 rule_update_actions(p, rule);
3518 static struct ofpbuf *
3519 compose_flow_removed(struct ofproto *p, const struct rule *rule,
3520 long long int now, uint8_t reason)
3522 struct ofp_flow_removed *ofr;
3524 long long int tdiff = now - rule->created;
3525 uint32_t sec = tdiff / 1000;
3526 uint32_t msec = tdiff - (sec * 1000);
3528 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
3529 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
3531 ofr->cookie = rule->flow_cookie;
3532 ofr->priority = htons(rule->cr.priority);
3533 ofr->reason = reason;
3534 ofr->duration_sec = htonl(sec);
3535 ofr->duration_nsec = htonl(msec * 1000000);
3536 ofr->idle_timeout = htons(rule->idle_timeout);
3537 ofr->packet_count = htonll(rule->packet_count);
3538 ofr->byte_count = htonll(rule->byte_count);
3544 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3546 assert(rule->installed);
3547 assert(!rule->cr.wc.wildcards);
3550 rule_remove(ofproto, rule);
3552 rule_uninstall(ofproto, rule);
3556 send_flow_removed(struct ofproto *p, struct rule *rule,
3557 long long int now, uint8_t reason)
3559 struct ofconn *ofconn;
3560 struct ofconn *prev;
3561 struct ofpbuf *buf = NULL;
3563 /* We limit the maximum number of queued flow expirations it by accounting
3564 * them under the counter for replies. That works because preventing
3565 * OpenFlow requests from being processed also prevents new flows from
3566 * being added (and expiring). (It also prevents processing OpenFlow
3567 * requests that would not add new flows, so it is imperfect.) */
3570 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3571 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)) {
3573 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3575 buf = compose_flow_removed(p, rule, now, reason);
3581 queue_tx(buf, prev, prev->reply_counter);
3587 expire_rule(struct cls_rule *cls_rule, void *p_)
3589 struct ofproto *p = p_;
3590 struct rule *rule = rule_from_cls_rule(cls_rule);
3591 long long int hard_expire, idle_expire, expire, now;
3593 hard_expire = (rule->hard_timeout
3594 ? rule->created + rule->hard_timeout * 1000
3596 idle_expire = (rule->idle_timeout
3597 && (rule->super || list_is_empty(&rule->list))
3598 ? rule->used + rule->idle_timeout * 1000
3600 expire = MIN(hard_expire, idle_expire);
3604 if (rule->installed && now >= rule->used + 5000) {
3605 uninstall_idle_flow(p, rule);
3606 } else if (!rule->cr.wc.wildcards) {
3607 active_timeout(p, rule);
3613 COVERAGE_INC(ofproto_expired);
3615 /* Update stats. This code will be a no-op if the rule expired
3616 * due to an idle timeout. */
3617 if (rule->cr.wc.wildcards) {
3618 struct rule *subrule, *next;
3619 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3620 rule_remove(p, subrule);
3623 rule_uninstall(p, rule);
3626 if (!rule_is_hidden(rule)) {
3627 send_flow_removed(p, rule, now,
3629 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
3631 rule_remove(p, rule);
3635 active_timeout(struct ofproto *ofproto, struct rule *rule)
3637 if (ofproto->netflow && !is_controller_rule(rule) &&
3638 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
3639 struct ofexpired expired;
3640 struct odp_flow odp_flow;
3642 /* Get updated flow stats. */
3643 memset(&odp_flow, 0, sizeof odp_flow);
3644 if (rule->installed) {
3645 odp_flow.key = rule->cr.flow;
3646 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
3647 dpif_flow_get(ofproto->dpif, &odp_flow);
3649 if (odp_flow.stats.n_packets) {
3650 update_time(ofproto, rule, &odp_flow.stats);
3651 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
3652 odp_flow.stats.tcp_flags);
3656 expired.flow = rule->cr.flow;
3657 expired.packet_count = rule->packet_count +
3658 odp_flow.stats.n_packets;
3659 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
3660 expired.used = rule->used;
3662 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
3664 /* Schedule us to send the accumulated records once we have
3665 * collected all of them. */
3666 poll_immediate_wake();
3671 update_used(struct ofproto *p)
3673 struct odp_flow *flows;
3678 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
3683 for (i = 0; i < n_flows; i++) {
3684 struct odp_flow *f = &flows[i];
3687 rule = rule_from_cls_rule(
3688 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3689 if (!rule || !rule->installed) {
3690 COVERAGE_INC(ofproto_unexpected_rule);
3691 dpif_flow_del(p->dpif, f);
3695 update_time(p, rule, &f->stats);
3696 rule_account(p, rule, f->stats.n_bytes);
3702 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3703 const struct ofpbuf *packet, int send_len)
3705 struct odp_msg *msg = packet->data;
3706 struct ofpbuf payload;
3710 /* Extract packet payload from 'msg'. */
3711 payload.data = msg + 1;
3712 payload.size = msg->length - sizeof *msg;
3714 /* Construct ofp_packet_in message. */
3715 reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3716 opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port), reason,
3717 &payload, send_len);
3720 rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100);
3724 send_packet_in_action(struct ofpbuf *packet, void *p_)
3726 struct ofproto *p = p_;
3727 struct ofconn *ofconn;
3728 struct odp_msg *msg;
3731 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3732 if (ofconn == p->controller || ofconn->miss_send_len) {
3733 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3736 ofpbuf_delete(packet);
3740 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3742 struct ofproto *p = p_;
3743 bool in_fail_open = p->fail_open && fail_open_is_active(p->fail_open);
3744 struct ofconn *ofconn;
3745 struct ofpbuf payload;
3746 struct odp_msg *msg;
3749 payload.data = msg + 1;
3750 payload.size = msg->length - sizeof *msg;
3751 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3752 if (ofconn->miss_send_len) {
3753 struct pktbuf *pb = ofconn->pktbuf;
3754 uint32_t buffer_id = (in_fail_open
3756 : pktbuf_save(pb, &payload, msg->port));
3757 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3759 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3762 ofpbuf_delete(packet);
3766 pick_datapath_id(const struct ofproto *ofproto)
3768 const struct ofport *port;
3770 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
3772 uint8_t ea[ETH_ADDR_LEN];
3775 error = netdev_get_etheraddr(port->netdev, ea);
3777 return eth_addr_to_uint64(ea);
3779 VLOG_WARN("could not get MAC address for %s (%s)",
3780 netdev_get_name(port->netdev), strerror(error));
3782 return ofproto->fallback_dpid;
3786 pick_fallback_dpid(void)
3788 uint8_t ea[ETH_ADDR_LEN];
3789 eth_addr_nicira_random(ea);
3790 return eth_addr_to_uint64(ea);
3794 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3795 struct odp_actions *actions, tag_type *tags,
3796 uint16_t *nf_output_iface, void *ofproto_)
3798 struct ofproto *ofproto = ofproto_;
3801 /* Drop frames for reserved multicast addresses. */
3802 if (eth_addr_is_reserved(flow->dl_dst)) {
3806 /* Learn source MAC (but don't try to learn from revalidation). */
3807 if (packet != NULL) {
3808 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3811 /* The log messages here could actually be useful in debugging,
3812 * so keep the rate limit relatively high. */
3813 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3814 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3815 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3816 ofproto_revalidate(ofproto, rev_tag);
3820 /* Determine output port. */
3821 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3823 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
3824 } else if (out_port != flow->in_port) {
3825 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3826 *nf_output_iface = out_port;
3834 static const struct ofhooks default_ofhooks = {
3836 default_normal_ofhook_cb,