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;
223 /* OpenFlow connections. */
224 struct list all_conns;
225 struct ofconn *controller;
226 struct pvconn **listeners;
228 struct pvconn **snoops;
231 /* Hooks for ovs-vswitchd. */
232 const struct ofhooks *ofhooks;
235 /* Used by default ofhooks. */
236 struct mac_learning *ml;
239 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
241 static const struct ofhooks default_ofhooks;
243 static uint64_t pick_datapath_id(const struct ofproto *);
244 static uint64_t pick_fallback_dpid(void);
245 static void send_packet_in_miss(struct ofpbuf *, void *ofproto);
246 static void send_packet_in_action(struct ofpbuf *, void *ofproto);
247 static void update_used(struct ofproto *);
248 static void update_stats(struct ofproto *, struct rule *,
249 const struct odp_flow_stats *);
250 static void expire_rule(struct cls_rule *, void *ofproto);
251 static void active_timeout(struct ofproto *ofproto, struct rule *rule);
252 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
253 static void revalidate_cb(struct cls_rule *rule_, void *p_);
255 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
257 static void handle_openflow(struct ofconn *, struct ofproto *,
260 static void refresh_port_groups(struct ofproto *);
262 static void update_port(struct ofproto *, const char *devname);
263 static int init_ports(struct ofproto *);
264 static void reinit_ports(struct ofproto *);
267 ofproto_create(const char *datapath, const char *datapath_type,
268 const struct ofhooks *ofhooks, void *aux,
269 struct ofproto **ofprotop)
271 struct odp_stats stats;
278 /* Connect to datapath and start listening for messages. */
279 error = dpif_open(datapath, datapath_type, &dpif);
281 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
284 error = dpif_get_dp_stats(dpif, &stats);
286 VLOG_ERR("failed to obtain stats for datapath %s: %s",
287 datapath, strerror(error));
291 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
293 VLOG_ERR("failed to listen on datapath %s: %s",
294 datapath, strerror(error));
298 dpif_flow_flush(dpif);
299 dpif_recv_purge(dpif);
301 /* Initialize settings. */
302 p = xzalloc(sizeof *p);
303 p->fallback_dpid = pick_fallback_dpid();
304 p->datapath_id = p->fallback_dpid;
305 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
306 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
307 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
308 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
309 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
311 /* Initialize datapath. */
313 p->netdev_monitor = netdev_monitor_create();
314 port_array_init(&p->ports);
315 shash_init(&p->port_by_name);
316 p->max_ports = stats.max_ports;
318 /* Initialize submodules. */
319 p->switch_status = switch_status_create(p);
323 p->miss_sched = p->action_sched = NULL;
327 /* Initialize flow table. */
328 classifier_init(&p->cls);
329 p->need_revalidate = false;
330 p->next_expiration = time_msec() + 1000;
331 tag_set_init(&p->revalidate_set);
333 /* Initialize OpenFlow connections. */
334 list_init(&p->all_conns);
335 p->controller = ofconn_create(p, rconn_create(5, 8));
336 p->controller->pktbuf = pktbuf_create();
337 p->controller->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
343 /* Initialize hooks. */
345 p->ofhooks = ofhooks;
349 p->ofhooks = &default_ofhooks;
351 p->ml = mac_learning_create();
354 /* Register switch status category. */
355 p->ss_cat = switch_status_register(p->switch_status, "remote",
356 rconn_status_cb, p->controller->rconn);
358 /* Pick final datapath ID. */
359 p->datapath_id = pick_datapath_id(p);
360 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
367 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
369 uint64_t old_dpid = p->datapath_id;
370 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
371 if (p->datapath_id != old_dpid) {
372 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
373 rconn_reconnect(p->controller->rconn);
378 ofproto_set_probe_interval(struct ofproto *p, int probe_interval)
380 probe_interval = probe_interval ? MAX(probe_interval, 5) : 0;
381 rconn_set_probe_interval(p->controller->rconn, probe_interval);
383 int trigger_duration = probe_interval ? probe_interval * 3 : 15;
384 fail_open_set_trigger_duration(p->fail_open, trigger_duration);
389 ofproto_set_max_backoff(struct ofproto *p, int max_backoff)
391 rconn_set_max_backoff(p->controller->rconn, max_backoff);
395 ofproto_set_desc(struct ofproto *p,
396 const char *mfr_desc, const char *hw_desc,
397 const char *sw_desc, const char *serial_desc,
400 struct ofp_desc_stats *ods;
403 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
404 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
405 sizeof ods->mfr_desc);
408 p->mfr_desc = xstrdup(mfr_desc);
411 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
412 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
413 sizeof ods->hw_desc);
416 p->hw_desc = xstrdup(hw_desc);
419 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
420 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
421 sizeof ods->sw_desc);
424 p->sw_desc = xstrdup(sw_desc);
427 if (strlen(serial_desc) >= sizeof ods->serial_num) {
428 VLOG_WARN("truncating serial_desc, must be less than %zu "
430 sizeof ods->serial_num);
432 free(p->serial_desc);
433 p->serial_desc = xstrdup(serial_desc);
436 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
437 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
438 sizeof ods->dp_desc);
441 p->dp_desc = xstrdup(dp_desc);
446 ofproto_set_in_band(struct ofproto *p, bool in_band)
448 if (in_band != (p->in_band != NULL)) {
450 return in_band_create(p, p->dpif, p->switch_status,
451 p->controller->rconn, &p->in_band);
453 ofproto_set_discovery(p, false, NULL, true);
454 in_band_destroy(p->in_band);
457 rconn_reconnect(p->controller->rconn);
463 ofproto_set_discovery(struct ofproto *p, bool discovery,
464 const char *re, bool update_resolv_conf)
466 if (discovery != (p->discovery != NULL)) {
468 int error = ofproto_set_in_band(p, true);
472 error = discovery_create(re, update_resolv_conf,
473 p->dpif, p->switch_status,
479 discovery_destroy(p->discovery);
482 rconn_disconnect(p->controller->rconn);
483 } else if (discovery) {
484 discovery_set_update_resolv_conf(p->discovery, update_resolv_conf);
485 return discovery_set_accept_controller_re(p->discovery, re);
491 ofproto_set_controller(struct ofproto *ofproto, const char *controller)
493 if (ofproto->discovery) {
495 } else if (controller) {
496 if (strcmp(rconn_get_name(ofproto->controller->rconn), controller)) {
497 return rconn_connect(ofproto->controller->rconn, controller);
502 rconn_disconnect(ofproto->controller->rconn);
508 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
509 const struct svec *svec)
511 struct pvconn **pvconns = *pvconnsp;
512 size_t n_pvconns = *n_pvconnsp;
516 for (i = 0; i < n_pvconns; i++) {
517 pvconn_close(pvconns[i]);
521 pvconns = xmalloc(svec->n * sizeof *pvconns);
523 for (i = 0; i < svec->n; i++) {
524 const char *name = svec->names[i];
525 struct pvconn *pvconn;
528 error = pvconn_open(name, &pvconn);
530 pvconns[n_pvconns++] = pvconn;
532 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
540 *n_pvconnsp = n_pvconns;
546 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
548 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
552 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
554 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
558 ofproto_set_netflow(struct ofproto *ofproto,
559 const struct netflow_options *nf_options)
561 if (nf_options && nf_options->collectors.n) {
562 if (!ofproto->netflow) {
563 ofproto->netflow = netflow_create();
565 return netflow_set_options(ofproto->netflow, nf_options);
567 netflow_destroy(ofproto->netflow);
568 ofproto->netflow = NULL;
574 ofproto_set_sflow(struct ofproto *ofproto,
575 const struct ofproto_sflow_options *oso)
577 struct ofproto_sflow *os = ofproto->sflow;
580 struct ofport *ofport;
581 unsigned int odp_port;
583 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
584 refresh_port_groups(ofproto);
585 PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) {
586 ofproto_sflow_add_port(os, odp_port,
587 netdev_get_name(ofport->netdev));
590 ofproto_sflow_set_options(os, oso);
592 ofproto_sflow_destroy(os);
593 ofproto->sflow = NULL;
598 ofproto_set_failure(struct ofproto *ofproto, bool fail_open)
601 struct rconn *rconn = ofproto->controller->rconn;
602 int trigger_duration = rconn_get_probe_interval(rconn) * 3;
603 if (!ofproto->fail_open) {
604 ofproto->fail_open = fail_open_create(ofproto, trigger_duration,
605 ofproto->switch_status,
608 fail_open_set_trigger_duration(ofproto->fail_open,
612 fail_open_destroy(ofproto->fail_open);
613 ofproto->fail_open = NULL;
618 ofproto_set_rate_limit(struct ofproto *ofproto,
619 int rate_limit, int burst_limit)
621 if (rate_limit > 0) {
622 if (!ofproto->miss_sched) {
623 ofproto->miss_sched = pinsched_create(rate_limit, burst_limit,
624 ofproto->switch_status);
625 ofproto->action_sched = pinsched_create(rate_limit, burst_limit,
628 pinsched_set_limits(ofproto->miss_sched, rate_limit, burst_limit);
629 pinsched_set_limits(ofproto->action_sched,
630 rate_limit, burst_limit);
633 pinsched_destroy(ofproto->miss_sched);
634 ofproto->miss_sched = NULL;
635 pinsched_destroy(ofproto->action_sched);
636 ofproto->action_sched = NULL;
641 ofproto_set_stp(struct ofproto *ofproto OVS_UNUSED, bool enable_stp)
645 VLOG_WARN("STP is not yet implemented");
653 ofproto_get_datapath_id(const struct ofproto *ofproto)
655 return ofproto->datapath_id;
659 ofproto_get_probe_interval(const struct ofproto *ofproto)
661 return rconn_get_probe_interval(ofproto->controller->rconn);
665 ofproto_get_max_backoff(const struct ofproto *ofproto)
667 return rconn_get_max_backoff(ofproto->controller->rconn);
671 ofproto_get_in_band(const struct ofproto *ofproto)
673 return ofproto->in_band != NULL;
677 ofproto_get_discovery(const struct ofproto *ofproto)
679 return ofproto->discovery != NULL;
683 ofproto_get_controller(const struct ofproto *ofproto)
685 return rconn_get_name(ofproto->controller->rconn);
689 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
693 for (i = 0; i < ofproto->n_listeners; i++) {
694 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
699 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
703 for (i = 0; i < ofproto->n_snoops; i++) {
704 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
709 ofproto_destroy(struct ofproto *p)
711 struct ofconn *ofconn, *next_ofconn;
712 struct ofport *ofport;
713 unsigned int port_no;
720 /* Destroy fail-open early, because it touches the classifier. */
721 ofproto_set_failure(p, false);
723 ofproto_flush_flows(p);
724 classifier_destroy(&p->cls);
726 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
728 ofconn_destroy(ofconn);
732 netdev_monitor_destroy(p->netdev_monitor);
733 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
736 shash_destroy(&p->port_by_name);
738 switch_status_destroy(p->switch_status);
739 in_band_destroy(p->in_band);
740 discovery_destroy(p->discovery);
741 pinsched_destroy(p->miss_sched);
742 pinsched_destroy(p->action_sched);
743 netflow_destroy(p->netflow);
744 ofproto_sflow_destroy(p->sflow);
746 switch_status_unregister(p->ss_cat);
748 for (i = 0; i < p->n_listeners; i++) {
749 pvconn_close(p->listeners[i]);
753 for (i = 0; i < p->n_snoops; i++) {
754 pvconn_close(p->snoops[i]);
758 mac_learning_destroy(p->ml);
763 free(p->serial_desc);
766 port_array_destroy(&p->ports);
772 ofproto_run(struct ofproto *p)
774 int error = ofproto_run1(p);
776 error = ofproto_run2(p, false);
782 process_port_change(struct ofproto *ofproto, int error, char *devname)
784 if (error == ENOBUFS) {
785 reinit_ports(ofproto);
787 update_port(ofproto, devname);
793 ofproto_run1(struct ofproto *p)
795 struct ofconn *ofconn, *next_ofconn;
800 if (shash_is_empty(&p->port_by_name)) {
804 for (i = 0; i < 50; i++) {
808 error = dpif_recv(p->dpif, &buf);
810 if (error == ENODEV) {
811 /* Someone destroyed the datapath behind our back. The caller
812 * better destroy us and give up, because we're just going to
813 * spin from here on out. */
814 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
815 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
822 handle_odp_msg(p, buf);
825 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
826 process_port_change(p, error, devname);
828 while ((error = netdev_monitor_poll(p->netdev_monitor,
829 &devname)) != EAGAIN) {
830 process_port_change(p, error, devname);
834 in_band_run(p->in_band);
837 char *controller_name;
838 if (rconn_is_connectivity_questionable(p->controller->rconn)) {
839 discovery_question_connectivity(p->discovery);
841 if (discovery_run(p->discovery, &controller_name)) {
842 if (controller_name) {
843 rconn_connect(p->controller->rconn, controller_name);
845 rconn_disconnect(p->controller->rconn);
849 pinsched_run(p->miss_sched, send_packet_in_miss, p);
850 pinsched_run(p->action_sched, send_packet_in_action, p);
852 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
854 ofconn_run(ofconn, p);
857 /* Fail-open maintenance. Do this after processing the ofconns since
858 * fail-open checks the status of the controller rconn. */
860 fail_open_run(p->fail_open);
863 for (i = 0; i < p->n_listeners; i++) {
867 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
869 ofconn_create(p, rconn_new_from_vconn("passive", vconn));
870 } else if (retval != EAGAIN) {
871 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
875 for (i = 0; i < p->n_snoops; i++) {
879 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
881 rconn_add_monitor(p->controller->rconn, vconn);
882 } else if (retval != EAGAIN) {
883 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
887 if (time_msec() >= p->next_expiration) {
888 COVERAGE_INC(ofproto_expiration);
889 p->next_expiration = time_msec() + 1000;
892 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
894 /* Let the hook know that we're at a stable point: all outstanding data
895 * in existing flows has been accounted to the account_cb. Thus, the
896 * hook can now reasonably do operations that depend on having accurate
897 * flow volume accounting (currently, that's just bond rebalancing). */
898 if (p->ofhooks->account_checkpoint_cb) {
899 p->ofhooks->account_checkpoint_cb(p->aux);
904 netflow_run(p->netflow);
907 ofproto_sflow_run(p->sflow);
913 struct revalidate_cbdata {
914 struct ofproto *ofproto;
915 bool revalidate_all; /* Revalidate all exact-match rules? */
916 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
917 struct tag_set revalidate_set; /* Set of tags to revalidate. */
921 ofproto_run2(struct ofproto *p, bool revalidate_all)
923 if (p->need_revalidate || revalidate_all
924 || !tag_set_is_empty(&p->revalidate_set)) {
925 struct revalidate_cbdata cbdata;
927 cbdata.revalidate_all = revalidate_all;
928 cbdata.revalidate_subrules = p->need_revalidate;
929 cbdata.revalidate_set = p->revalidate_set;
930 tag_set_init(&p->revalidate_set);
931 COVERAGE_INC(ofproto_revalidate);
932 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
933 p->need_revalidate = false;
940 ofproto_wait(struct ofproto *p)
942 struct ofconn *ofconn;
945 dpif_recv_wait(p->dpif);
946 dpif_port_poll_wait(p->dpif);
947 netdev_monitor_poll_wait(p->netdev_monitor);
948 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
952 in_band_wait(p->in_band);
955 discovery_wait(p->discovery);
958 fail_open_wait(p->fail_open);
960 pinsched_wait(p->miss_sched);
961 pinsched_wait(p->action_sched);
963 ofproto_sflow_wait(p->sflow);
965 if (!tag_set_is_empty(&p->revalidate_set)) {
966 poll_immediate_wake();
968 if (p->need_revalidate) {
969 /* Shouldn't happen, but if it does just go around again. */
970 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
971 poll_immediate_wake();
972 } else if (p->next_expiration != LLONG_MAX) {
973 poll_timer_wait(p->next_expiration - time_msec());
975 for (i = 0; i < p->n_listeners; i++) {
976 pvconn_wait(p->listeners[i]);
978 for (i = 0; i < p->n_snoops; i++) {
979 pvconn_wait(p->snoops[i]);
984 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
986 tag_set_add(&ofproto->revalidate_set, tag);
990 ofproto_get_revalidate_set(struct ofproto *ofproto)
992 return &ofproto->revalidate_set;
996 ofproto_is_alive(const struct ofproto *p)
998 return p->discovery || rconn_is_alive(p->controller->rconn);
1002 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
1003 const union ofp_action *actions, size_t n_actions,
1004 const struct ofpbuf *packet)
1006 struct odp_actions odp_actions;
1009 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1015 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1017 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
1018 odp_actions.n_actions, packet);
1023 ofproto_add_flow(struct ofproto *p,
1024 const flow_t *flow, uint32_t wildcards, unsigned int priority,
1025 const union ofp_action *actions, size_t n_actions,
1029 rule = rule_create(p, NULL, actions, n_actions,
1030 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1032 cls_rule_from_flow(&rule->cr, flow, wildcards, priority);
1033 rule_insert(p, rule, NULL, 0);
1037 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1038 uint32_t wildcards, unsigned int priority)
1042 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1046 rule_remove(ofproto, rule);
1051 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1053 struct rule *rule = rule_from_cls_rule(rule_);
1054 struct ofproto *ofproto = ofproto_;
1056 /* Mark the flow as not installed, even though it might really be
1057 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1058 * There is no point in uninstalling it individually since we are about to
1059 * blow away all the flows with dpif_flow_flush(). */
1060 rule->installed = false;
1062 rule_remove(ofproto, rule);
1066 ofproto_flush_flows(struct ofproto *ofproto)
1068 COVERAGE_INC(ofproto_flush);
1069 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1070 dpif_flow_flush(ofproto->dpif);
1071 if (ofproto->in_band) {
1072 in_band_flushed(ofproto->in_band);
1074 if (ofproto->fail_open) {
1075 fail_open_flushed(ofproto->fail_open);
1080 reinit_ports(struct ofproto *p)
1082 struct svec devnames;
1083 struct ofport *ofport;
1084 unsigned int port_no;
1085 struct odp_port *odp_ports;
1089 svec_init(&devnames);
1090 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1091 svec_add (&devnames, (char *) ofport->opp.name);
1093 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1094 for (i = 0; i < n_odp_ports; i++) {
1095 svec_add (&devnames, odp_ports[i].devname);
1099 svec_sort_unique(&devnames);
1100 for (i = 0; i < devnames.n; i++) {
1101 update_port(p, devnames.names[i]);
1103 svec_destroy(&devnames);
1107 refresh_port_group(struct ofproto *p, unsigned int group)
1111 struct ofport *port;
1112 unsigned int port_no;
1114 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1116 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1118 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1119 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1120 ports[n_ports++] = port_no;
1123 dpif_port_group_set(p->dpif, group, ports, n_ports);
1130 refresh_port_groups(struct ofproto *p)
1132 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1133 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1135 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1139 static struct ofport *
1140 make_ofport(const struct odp_port *odp_port)
1142 struct netdev_options netdev_options;
1143 enum netdev_flags flags;
1144 struct ofport *ofport;
1145 struct netdev *netdev;
1149 memset(&netdev_options, 0, sizeof netdev_options);
1150 netdev_options.name = odp_port->devname;
1151 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1152 netdev_options.may_open = true;
1154 error = netdev_open(&netdev_options, &netdev);
1156 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1157 "cannot be opened (%s)",
1158 odp_port->devname, odp_port->port,
1159 odp_port->devname, strerror(error));
1163 ofport = xmalloc(sizeof *ofport);
1164 ofport->netdev = netdev;
1165 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1166 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1167 memcpy(ofport->opp.name, odp_port->devname,
1168 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1169 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1171 netdev_get_flags(netdev, &flags);
1172 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1174 netdev_get_carrier(netdev, &carrier);
1175 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1177 netdev_get_features(netdev,
1178 &ofport->opp.curr, &ofport->opp.advertised,
1179 &ofport->opp.supported, &ofport->opp.peer);
1184 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1186 if (port_array_get(&p->ports, odp_port->port)) {
1187 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1190 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1191 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1200 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1202 const struct ofp_phy_port *a = &a_->opp;
1203 const struct ofp_phy_port *b = &b_->opp;
1205 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1206 return (a->port_no == b->port_no
1207 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1208 && !strcmp((char *) a->name, (char *) b->name)
1209 && a->state == b->state
1210 && a->config == b->config
1211 && a->curr == b->curr
1212 && a->advertised == b->advertised
1213 && a->supported == b->supported
1214 && a->peer == b->peer);
1218 send_port_status(struct ofproto *p, const struct ofport *ofport,
1221 /* XXX Should limit the number of queued port status change messages. */
1222 struct ofconn *ofconn;
1223 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1224 struct ofp_port_status *ops;
1227 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1228 ops->reason = reason;
1229 ops->desc = ofport->opp;
1230 hton_ofp_phy_port(&ops->desc);
1231 queue_tx(b, ofconn, NULL);
1233 if (p->ofhooks->port_changed_cb) {
1234 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1239 ofport_install(struct ofproto *p, struct ofport *ofport)
1241 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1242 const char *netdev_name = (const char *) ofport->opp.name;
1244 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1245 port_array_set(&p->ports, odp_port, ofport);
1246 shash_add(&p->port_by_name, netdev_name, ofport);
1248 ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
1253 ofport_remove(struct ofproto *p, struct ofport *ofport)
1255 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1257 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1258 port_array_set(&p->ports, odp_port, NULL);
1259 shash_delete(&p->port_by_name,
1260 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1262 ofproto_sflow_del_port(p->sflow, odp_port);
1267 ofport_free(struct ofport *ofport)
1270 netdev_close(ofport->netdev);
1276 update_port(struct ofproto *p, const char *devname)
1278 struct odp_port odp_port;
1279 struct ofport *old_ofport;
1280 struct ofport *new_ofport;
1283 COVERAGE_INC(ofproto_update_port);
1285 /* Query the datapath for port information. */
1286 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1288 /* Find the old ofport. */
1289 old_ofport = shash_find_data(&p->port_by_name, devname);
1292 /* There's no port named 'devname' but there might be a port with
1293 * the same port number. This could happen if a port is deleted
1294 * and then a new one added in its place very quickly, or if a port
1295 * is renamed. In the former case we want to send an OFPPR_DELETE
1296 * and an OFPPR_ADD, and in the latter case we want to send a
1297 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1298 * the old port's ifindex against the new port, or perhaps less
1299 * reliably but more portably by comparing the old port's MAC
1300 * against the new port's MAC. However, this code isn't that smart
1301 * and always sends an OFPPR_MODIFY (XXX). */
1302 old_ofport = port_array_get(&p->ports, odp_port.port);
1304 } else if (error != ENOENT && error != ENODEV) {
1305 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1306 "%s", strerror(error));
1310 /* Create a new ofport. */
1311 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1313 /* Eliminate a few pathological cases. */
1314 if (!old_ofport && !new_ofport) {
1316 } else if (old_ofport && new_ofport) {
1317 /* Most of the 'config' bits are OpenFlow soft state, but
1318 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1319 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1320 * leaves the other bits 0.) */
1321 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1323 if (ofport_equal(old_ofport, new_ofport)) {
1324 /* False alarm--no change. */
1325 ofport_free(new_ofport);
1330 /* Now deal with the normal cases. */
1332 ofport_remove(p, old_ofport);
1335 ofport_install(p, new_ofport);
1337 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1338 (!old_ofport ? OFPPR_ADD
1339 : !new_ofport ? OFPPR_DELETE
1341 ofport_free(old_ofport);
1343 /* Update port groups. */
1344 refresh_port_groups(p);
1348 init_ports(struct ofproto *p)
1350 struct odp_port *ports;
1355 error = dpif_port_list(p->dpif, &ports, &n_ports);
1360 for (i = 0; i < n_ports; i++) {
1361 const struct odp_port *odp_port = &ports[i];
1362 if (!ofport_conflicts(p, odp_port)) {
1363 struct ofport *ofport = make_ofport(odp_port);
1365 ofport_install(p, ofport);
1370 refresh_port_groups(p);
1374 static struct ofconn *
1375 ofconn_create(struct ofproto *p, struct rconn *rconn)
1377 struct ofconn *ofconn = xmalloc(sizeof *ofconn);
1378 list_push_back(&p->all_conns, &ofconn->node);
1379 ofconn->rconn = rconn;
1380 ofconn->pktbuf = NULL;
1381 ofconn->miss_send_len = 0;
1382 ofconn->packet_in_counter = rconn_packet_counter_create ();
1383 ofconn->reply_counter = rconn_packet_counter_create ();
1388 ofconn_destroy(struct ofconn *ofconn)
1390 list_remove(&ofconn->node);
1391 rconn_destroy(ofconn->rconn);
1392 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1393 rconn_packet_counter_destroy(ofconn->reply_counter);
1394 pktbuf_destroy(ofconn->pktbuf);
1399 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1403 rconn_run(ofconn->rconn);
1405 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1406 /* Limit the number of iterations to prevent other tasks from
1408 for (iteration = 0; iteration < 50; iteration++) {
1409 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1414 fail_open_maybe_recover(p->fail_open);
1416 handle_openflow(ofconn, p, of_msg);
1417 ofpbuf_delete(of_msg);
1421 if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
1422 ofconn_destroy(ofconn);
1427 ofconn_wait(struct ofconn *ofconn)
1429 rconn_run_wait(ofconn->rconn);
1430 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1431 rconn_recv_wait(ofconn->rconn);
1433 COVERAGE_INC(ofproto_ofconn_stuck);
1437 /* Caller is responsible for initializing the 'cr' member of the returned
1439 static struct rule *
1440 rule_create(struct ofproto *ofproto, struct rule *super,
1441 const union ofp_action *actions, size_t n_actions,
1442 uint16_t idle_timeout, uint16_t hard_timeout,
1443 uint64_t flow_cookie, bool send_flow_removed)
1445 struct rule *rule = xzalloc(sizeof *rule);
1446 rule->idle_timeout = idle_timeout;
1447 rule->hard_timeout = hard_timeout;
1448 rule->flow_cookie = flow_cookie;
1449 rule->used = rule->created = time_msec();
1450 rule->send_flow_removed = send_flow_removed;
1451 rule->super = super;
1453 list_push_back(&super->list, &rule->list);
1455 list_init(&rule->list);
1457 rule->n_actions = n_actions;
1458 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1459 netflow_flow_clear(&rule->nf_flow);
1460 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1465 static struct rule *
1466 rule_from_cls_rule(const struct cls_rule *cls_rule)
1468 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1472 rule_free(struct rule *rule)
1474 free(rule->actions);
1475 free(rule->odp_actions);
1479 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1480 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1481 * through all of its subrules and revalidates them, destroying any that no
1482 * longer has a super-rule (which is probably all of them).
1484 * Before calling this function, the caller must make have removed 'rule' from
1485 * the classifier. If 'rule' is an exact-match rule, the caller is also
1486 * responsible for ensuring that it has been uninstalled from the datapath. */
1488 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1491 struct rule *subrule, *next;
1492 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1493 revalidate_rule(ofproto, subrule);
1496 list_remove(&rule->list);
1502 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1504 const union ofp_action *oa;
1505 struct actions_iterator i;
1507 if (out_port == htons(OFPP_NONE)) {
1510 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1511 oa = actions_next(&i)) {
1512 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1519 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1520 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1522 * The flow that 'packet' actually contains does not need to actually match
1523 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1524 * the packet and byte counters for 'rule' will be credited for the packet sent
1525 * out whether or not the packet actually matches 'rule'.
1527 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1528 * the caller must already have accurately composed ODP actions for it given
1529 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1530 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1531 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1532 * actions and apply them to 'packet'. */
1534 rule_execute(struct ofproto *ofproto, struct rule *rule,
1535 struct ofpbuf *packet, const flow_t *flow)
1537 const union odp_action *actions;
1539 struct odp_actions a;
1541 /* Grab or compose the ODP actions.
1543 * The special case for an exact-match 'rule' where 'flow' is not the
1544 * rule's flow is important to avoid, e.g., sending a packet out its input
1545 * port simply because the ODP actions were composed for the wrong
1547 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1548 struct rule *super = rule->super ? rule->super : rule;
1549 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1550 packet, &a, NULL, 0, NULL)) {
1553 actions = a.actions;
1554 n_actions = a.n_actions;
1556 actions = rule->odp_actions;
1557 n_actions = rule->n_odp_actions;
1560 /* Execute the ODP actions. */
1561 if (!dpif_execute(ofproto->dpif, flow->in_port,
1562 actions, n_actions, packet)) {
1563 struct odp_flow_stats stats;
1564 flow_extract_stats(flow, packet, &stats);
1565 update_stats(ofproto, rule, &stats);
1566 rule->used = time_msec();
1567 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1572 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1575 struct rule *displaced_rule;
1577 /* Insert the rule in the classifier. */
1578 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1579 if (!rule->cr.wc.wildcards) {
1580 rule_make_actions(p, rule, packet);
1583 /* Send the packet and credit it to the rule. */
1586 flow_extract(packet, in_port, &flow);
1587 rule_execute(p, rule, packet, &flow);
1590 /* Install the rule in the datapath only after sending the packet, to
1591 * avoid packet reordering. */
1592 if (rule->cr.wc.wildcards) {
1593 COVERAGE_INC(ofproto_add_wc_flow);
1594 p->need_revalidate = true;
1596 rule_install(p, rule, displaced_rule);
1599 /* Free the rule that was displaced, if any. */
1600 if (displaced_rule) {
1601 rule_destroy(p, displaced_rule);
1605 static struct rule *
1606 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1609 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1610 rule->idle_timeout, rule->hard_timeout,
1612 COVERAGE_INC(ofproto_subrule_create);
1613 cls_rule_from_flow(&subrule->cr, flow, 0,
1614 (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1615 : rule->cr.priority));
1616 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1622 rule_remove(struct ofproto *ofproto, struct rule *rule)
1624 if (rule->cr.wc.wildcards) {
1625 COVERAGE_INC(ofproto_del_wc_flow);
1626 ofproto->need_revalidate = true;
1628 rule_uninstall(ofproto, rule);
1630 classifier_remove(&ofproto->cls, &rule->cr);
1631 rule_destroy(ofproto, rule);
1634 /* Returns true if the actions changed, false otherwise. */
1636 rule_make_actions(struct ofproto *p, struct rule *rule,
1637 const struct ofpbuf *packet)
1639 const struct rule *super;
1640 struct odp_actions a;
1643 assert(!rule->cr.wc.wildcards);
1645 super = rule->super ? rule->super : rule;
1647 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1648 packet, &a, &rule->tags, &rule->may_install,
1649 &rule->nf_flow.output_iface);
1651 actions_len = a.n_actions * sizeof *a.actions;
1652 if (rule->n_odp_actions != a.n_actions
1653 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1654 COVERAGE_INC(ofproto_odp_unchanged);
1655 free(rule->odp_actions);
1656 rule->n_odp_actions = a.n_actions;
1657 rule->odp_actions = xmemdup(a.actions, actions_len);
1665 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1666 struct odp_flow_put *put)
1668 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1669 put->flow.key = rule->cr.flow;
1670 put->flow.actions = rule->odp_actions;
1671 put->flow.n_actions = rule->n_odp_actions;
1673 return dpif_flow_put(ofproto->dpif, put);
1677 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1679 assert(!rule->cr.wc.wildcards);
1681 if (rule->may_install) {
1682 struct odp_flow_put put;
1683 if (!do_put_flow(p, rule,
1684 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1686 rule->installed = true;
1687 if (displaced_rule) {
1688 update_stats(p, displaced_rule, &put.flow.stats);
1689 rule_post_uninstall(p, displaced_rule);
1692 } else if (displaced_rule) {
1693 rule_uninstall(p, displaced_rule);
1698 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1700 if (rule->installed) {
1701 struct odp_flow_put put;
1702 COVERAGE_INC(ofproto_dp_missed);
1703 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1705 rule_install(ofproto, rule, NULL);
1710 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1712 bool actions_changed;
1713 uint16_t new_out_iface, old_out_iface;
1715 old_out_iface = rule->nf_flow.output_iface;
1716 actions_changed = rule_make_actions(ofproto, rule, NULL);
1718 if (rule->may_install) {
1719 if (rule->installed) {
1720 if (actions_changed) {
1721 struct odp_flow_put put;
1722 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
1723 | ODPPF_ZERO_STATS, &put);
1724 update_stats(ofproto, rule, &put.flow.stats);
1726 /* Temporarily set the old output iface so that NetFlow
1727 * messages have the correct output interface for the old
1729 new_out_iface = rule->nf_flow.output_iface;
1730 rule->nf_flow.output_iface = old_out_iface;
1731 rule_post_uninstall(ofproto, rule);
1732 rule->nf_flow.output_iface = new_out_iface;
1735 rule_install(ofproto, rule, NULL);
1738 rule_uninstall(ofproto, rule);
1743 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1745 uint64_t total_bytes = rule->byte_count + extra_bytes;
1747 if (ofproto->ofhooks->account_flow_cb
1748 && total_bytes > rule->accounted_bytes)
1750 ofproto->ofhooks->account_flow_cb(
1751 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1752 total_bytes - rule->accounted_bytes, ofproto->aux);
1753 rule->accounted_bytes = total_bytes;
1758 rule_uninstall(struct ofproto *p, struct rule *rule)
1760 assert(!rule->cr.wc.wildcards);
1761 if (rule->installed) {
1762 struct odp_flow odp_flow;
1764 odp_flow.key = rule->cr.flow;
1765 odp_flow.actions = NULL;
1766 odp_flow.n_actions = 0;
1767 if (!dpif_flow_del(p->dpif, &odp_flow)) {
1768 update_stats(p, rule, &odp_flow.stats);
1770 rule->installed = false;
1772 rule_post_uninstall(p, rule);
1777 is_controller_rule(struct rule *rule)
1779 /* If the only action is send to the controller then don't report
1780 * NetFlow expiration messages since it is just part of the control
1781 * logic for the network and not real traffic. */
1783 if (rule && rule->super) {
1784 struct rule *super = rule->super;
1786 return super->n_actions == 1 &&
1787 super->actions[0].type == htons(OFPAT_OUTPUT) &&
1788 super->actions[0].output.port == htons(OFPP_CONTROLLER);
1795 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1797 struct rule *super = rule->super;
1799 rule_account(ofproto, rule, 0);
1801 if (ofproto->netflow && !is_controller_rule(rule)) {
1802 struct ofexpired expired;
1803 expired.flow = rule->cr.flow;
1804 expired.packet_count = rule->packet_count;
1805 expired.byte_count = rule->byte_count;
1806 expired.used = rule->used;
1807 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
1810 super->packet_count += rule->packet_count;
1811 super->byte_count += rule->byte_count;
1813 /* Reset counters to prevent double counting if the rule ever gets
1815 rule->packet_count = 0;
1816 rule->byte_count = 0;
1817 rule->accounted_bytes = 0;
1819 netflow_flow_clear(&rule->nf_flow);
1824 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1825 struct rconn_packet_counter *counter)
1827 update_openflow_length(msg);
1828 if (rconn_send(ofconn->rconn, msg, counter)) {
1834 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1835 int error, const void *data, size_t len)
1838 struct ofp_error_msg *oem;
1840 if (!(error >> 16)) {
1841 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1846 COVERAGE_INC(ofproto_error);
1847 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1848 oh ? oh->xid : 0, &buf);
1849 oem->type = htons((unsigned int) error >> 16);
1850 oem->code = htons(error & 0xffff);
1851 memcpy(oem->data, data, len);
1852 queue_tx(buf, ofconn, ofconn->reply_counter);
1856 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1859 size_t oh_length = ntohs(oh->length);
1860 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1864 hton_ofp_phy_port(struct ofp_phy_port *opp)
1866 opp->port_no = htons(opp->port_no);
1867 opp->config = htonl(opp->config);
1868 opp->state = htonl(opp->state);
1869 opp->curr = htonl(opp->curr);
1870 opp->advertised = htonl(opp->advertised);
1871 opp->supported = htonl(opp->supported);
1872 opp->peer = htonl(opp->peer);
1876 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
1878 struct ofp_header *rq = oh;
1879 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
1884 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
1885 struct ofp_header *oh)
1887 struct ofp_switch_features *osf;
1889 unsigned int port_no;
1890 struct ofport *port;
1892 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1893 osf->datapath_id = htonll(p->datapath_id);
1894 osf->n_buffers = htonl(pktbuf_capacity());
1896 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1897 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
1898 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1899 (1u << OFPAT_SET_VLAN_VID) |
1900 (1u << OFPAT_SET_VLAN_PCP) |
1901 (1u << OFPAT_STRIP_VLAN) |
1902 (1u << OFPAT_SET_DL_SRC) |
1903 (1u << OFPAT_SET_DL_DST) |
1904 (1u << OFPAT_SET_NW_SRC) |
1905 (1u << OFPAT_SET_NW_DST) |
1906 (1u << OFPAT_SET_NW_TOS) |
1907 (1u << OFPAT_SET_TP_SRC) |
1908 (1u << OFPAT_SET_TP_DST));
1910 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1911 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1914 queue_tx(buf, ofconn, ofconn->reply_counter);
1919 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1920 struct ofp_header *oh)
1923 struct ofp_switch_config *osc;
1927 /* Figure out flags. */
1928 dpif_get_drop_frags(p->dpif, &drop_frags);
1929 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1932 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1933 osc->flags = htons(flags);
1934 osc->miss_send_len = htons(ofconn->miss_send_len);
1935 queue_tx(buf, ofconn, ofconn->reply_counter);
1941 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1942 struct ofp_switch_config *osc)
1947 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1951 flags = ntohs(osc->flags);
1953 if (ofconn == p->controller) {
1954 switch (flags & OFPC_FRAG_MASK) {
1955 case OFPC_FRAG_NORMAL:
1956 dpif_set_drop_frags(p->dpif, false);
1958 case OFPC_FRAG_DROP:
1959 dpif_set_drop_frags(p->dpif, true);
1962 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1968 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1969 if (ntohs(osc->miss_send_len) != 0) {
1970 ofconn->pktbuf = pktbuf_create();
1972 pktbuf_destroy(ofconn->pktbuf);
1976 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1982 add_output_group_action(struct odp_actions *actions, uint16_t group,
1983 uint16_t *nf_output_iface)
1985 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1987 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
1988 *nf_output_iface = NF_OUT_FLOOD;
1993 add_controller_action(struct odp_actions *actions,
1994 const struct ofp_action_output *oao)
1996 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
1997 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
2000 struct action_xlate_ctx {
2002 flow_t flow; /* Flow to which these actions correspond. */
2003 int recurse; /* Recursion level, via xlate_table_action. */
2004 struct ofproto *ofproto;
2005 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2006 * null pointer if we are revalidating
2007 * without a packet to refer to. */
2010 struct odp_actions *out; /* Datapath actions. */
2011 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2012 bool may_set_up_flow; /* True ordinarily; false if the actions must
2013 * be reassessed for every packet. */
2014 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2017 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2018 struct action_xlate_ctx *ctx);
2021 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2023 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
2026 if (ofport->opp.config & OFPPC_NO_FWD) {
2027 /* Forwarding disabled on port. */
2032 * We don't have an ofport record for this port, but it doesn't hurt to
2033 * allow forwarding to it anyhow. Maybe such a port will appear later
2034 * and we're pre-populating the flow table.
2038 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2039 ctx->nf_output_iface = port;
2042 static struct rule *
2043 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2046 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2048 /* The rule we found might not be valid, since we could be in need of
2049 * revalidation. If it is not valid, don't return it. */
2052 && ofproto->need_revalidate
2053 && !revalidate_rule(ofproto, rule)) {
2054 COVERAGE_INC(ofproto_invalidated);
2062 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2064 if (!ctx->recurse) {
2065 uint16_t old_in_port;
2068 /* Look up a flow with 'in_port' as the input port. Then restore the
2069 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2070 * have surprising behavior). */
2071 old_in_port = ctx->flow.in_port;
2072 ctx->flow.in_port = in_port;
2073 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2074 ctx->flow.in_port = old_in_port;
2082 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2089 xlate_output_action(struct action_xlate_ctx *ctx,
2090 const struct ofp_action_output *oao)
2093 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2095 ctx->nf_output_iface = NF_OUT_DROP;
2097 switch (ntohs(oao->port)) {
2099 add_output_action(ctx, ctx->flow.in_port);
2102 xlate_table_action(ctx, ctx->flow.in_port);
2105 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2106 ctx->out, ctx->tags,
2107 &ctx->nf_output_iface,
2108 ctx->ofproto->aux)) {
2109 COVERAGE_INC(ofproto_uninstallable);
2110 ctx->may_set_up_flow = false;
2114 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2115 &ctx->nf_output_iface);
2118 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2120 case OFPP_CONTROLLER:
2121 add_controller_action(ctx->out, oao);
2124 add_output_action(ctx, ODPP_LOCAL);
2127 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2128 if (odp_port != ctx->flow.in_port) {
2129 add_output_action(ctx, odp_port);
2134 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2135 ctx->nf_output_iface = NF_OUT_FLOOD;
2136 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2137 ctx->nf_output_iface = prev_nf_output_iface;
2138 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2139 ctx->nf_output_iface != NF_OUT_FLOOD) {
2140 ctx->nf_output_iface = NF_OUT_MULTI;
2145 xlate_nicira_action(struct action_xlate_ctx *ctx,
2146 const struct nx_action_header *nah)
2148 const struct nx_action_resubmit *nar;
2149 int subtype = ntohs(nah->subtype);
2151 assert(nah->vendor == htonl(NX_VENDOR_ID));
2153 case NXAST_RESUBMIT:
2154 nar = (const struct nx_action_resubmit *) nah;
2155 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2158 /* If you add a new action here that modifies flow data, don't forget to
2159 * update the flow key in ctx->flow in the same key. */
2162 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2168 do_xlate_actions(const union ofp_action *in, size_t n_in,
2169 struct action_xlate_ctx *ctx)
2171 struct actions_iterator iter;
2172 const union ofp_action *ia;
2173 const struct ofport *port;
2175 port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port);
2176 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2177 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, stp_eth_addr)
2178 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2179 /* Drop this flow. */
2183 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2184 uint16_t type = ntohs(ia->type);
2185 union odp_action *oa;
2189 xlate_output_action(ctx, &ia->output);
2192 case OFPAT_SET_VLAN_VID:
2193 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2194 ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2197 case OFPAT_SET_VLAN_PCP:
2198 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2199 ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2202 case OFPAT_STRIP_VLAN:
2203 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2204 ctx->flow.dl_vlan = OFP_VLAN_NONE;
2205 ctx->flow.dl_vlan_pcp = 0;
2208 case OFPAT_SET_DL_SRC:
2209 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2210 memcpy(oa->dl_addr.dl_addr,
2211 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2212 memcpy(ctx->flow.dl_src,
2213 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2216 case OFPAT_SET_DL_DST:
2217 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2218 memcpy(oa->dl_addr.dl_addr,
2219 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2220 memcpy(ctx->flow.dl_dst,
2221 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2224 case OFPAT_SET_NW_SRC:
2225 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2226 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2229 case OFPAT_SET_NW_DST:
2230 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2231 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2234 case OFPAT_SET_NW_TOS:
2235 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2236 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2239 case OFPAT_SET_TP_SRC:
2240 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2241 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2244 case OFPAT_SET_TP_DST:
2245 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2246 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2250 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2254 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2261 xlate_actions(const union ofp_action *in, size_t n_in,
2262 const flow_t *flow, struct ofproto *ofproto,
2263 const struct ofpbuf *packet,
2264 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2265 uint16_t *nf_output_iface)
2267 tag_type no_tags = 0;
2268 struct action_xlate_ctx ctx;
2269 COVERAGE_INC(ofproto_ofp2odp);
2270 odp_actions_init(out);
2273 ctx.ofproto = ofproto;
2274 ctx.packet = packet;
2276 ctx.tags = tags ? tags : &no_tags;
2277 ctx.may_set_up_flow = true;
2278 ctx.nf_output_iface = NF_OUT_DROP;
2279 do_xlate_actions(in, n_in, &ctx);
2281 /* Check with in-band control to see if we're allowed to set up this
2283 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2284 ctx.may_set_up_flow = false;
2287 if (may_set_up_flow) {
2288 *may_set_up_flow = ctx.may_set_up_flow;
2290 if (nf_output_iface) {
2291 *nf_output_iface = ctx.nf_output_iface;
2293 if (odp_actions_overflow(out)) {
2294 odp_actions_init(out);
2295 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2301 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2302 struct ofp_header *oh)
2304 struct ofp_packet_out *opo;
2305 struct ofpbuf payload, *buffer;
2306 struct odp_actions actions;
2312 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2316 opo = (struct ofp_packet_out *) oh;
2318 COVERAGE_INC(ofproto_packet_out);
2319 if (opo->buffer_id != htonl(UINT32_MAX)) {
2320 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2322 if (error || !buffer) {
2330 flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2331 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2332 &flow, p, &payload, &actions, NULL, NULL, NULL);
2337 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2339 ofpbuf_delete(buffer);
2345 update_port_config(struct ofproto *p, struct ofport *port,
2346 uint32_t config, uint32_t mask)
2348 mask &= config ^ port->opp.config;
2349 if (mask & OFPPC_PORT_DOWN) {
2350 if (config & OFPPC_PORT_DOWN) {
2351 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2353 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2356 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2357 if (mask & REVALIDATE_BITS) {
2358 COVERAGE_INC(ofproto_costly_flags);
2359 port->opp.config ^= mask & REVALIDATE_BITS;
2360 p->need_revalidate = true;
2362 #undef REVALIDATE_BITS
2363 if (mask & OFPPC_NO_FLOOD) {
2364 port->opp.config ^= OFPPC_NO_FLOOD;
2365 refresh_port_groups(p);
2367 if (mask & OFPPC_NO_PACKET_IN) {
2368 port->opp.config ^= OFPPC_NO_PACKET_IN;
2373 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2375 const struct ofp_port_mod *opm;
2376 struct ofport *port;
2379 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2383 opm = (struct ofp_port_mod *) oh;
2385 port = port_array_get(&p->ports,
2386 ofp_port_to_odp_port(ntohs(opm->port_no)));
2388 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2389 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2390 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2392 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2393 if (opm->advertise) {
2394 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2400 static struct ofpbuf *
2401 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2403 struct ofp_stats_reply *osr;
2406 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2407 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2409 osr->flags = htons(0);
2413 static struct ofpbuf *
2414 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2416 return make_stats_reply(request->header.xid, request->type, body_len);
2420 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2422 struct ofpbuf *msg = *msgp;
2423 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2424 if (nbytes + msg->size > UINT16_MAX) {
2425 struct ofp_stats_reply *reply = msg->data;
2426 reply->flags = htons(OFPSF_REPLY_MORE);
2427 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2428 queue_tx(msg, ofconn, ofconn->reply_counter);
2430 return ofpbuf_put_uninit(*msgp, nbytes);
2434 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2435 struct ofp_stats_request *request)
2437 struct ofp_desc_stats *ods;
2440 msg = start_stats_reply(request, sizeof *ods);
2441 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2442 memset(ods, 0, sizeof *ods);
2443 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2444 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2445 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2446 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2447 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2448 queue_tx(msg, ofconn, ofconn->reply_counter);
2454 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2456 struct rule *rule = rule_from_cls_rule(cls_rule);
2457 int *n_subrules = n_subrules_;
2465 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2466 struct ofp_stats_request *request)
2468 struct ofp_table_stats *ots;
2470 struct odp_stats dpstats;
2471 int n_exact, n_subrules, n_wild;
2473 msg = start_stats_reply(request, sizeof *ots * 2);
2475 /* Count rules of various kinds. */
2477 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2478 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2479 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2482 dpif_get_dp_stats(p->dpif, &dpstats);
2483 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2484 memset(ots, 0, sizeof *ots);
2485 ots->table_id = TABLEID_HASH;
2486 strcpy(ots->name, "hash");
2487 ots->wildcards = htonl(0);
2488 ots->max_entries = htonl(dpstats.max_capacity);
2489 ots->active_count = htonl(n_exact);
2490 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2492 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2494 /* Classifier table. */
2495 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2496 memset(ots, 0, sizeof *ots);
2497 ots->table_id = TABLEID_CLASSIFIER;
2498 strcpy(ots->name, "classifier");
2499 ots->wildcards = htonl(OFPFW_ALL);
2500 ots->max_entries = htonl(65536);
2501 ots->active_count = htonl(n_wild);
2502 ots->lookup_count = htonll(0); /* XXX */
2503 ots->matched_count = htonll(0); /* XXX */
2505 queue_tx(msg, ofconn, ofconn->reply_counter);
2510 append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
2513 struct netdev_stats stats;
2514 struct ofp_port_stats *ops;
2516 /* Intentionally ignore return value, since errors will set
2517 * 'stats' to all-1s, which is correct for OpenFlow, and
2518 * netdev_get_stats() will log errors. */
2519 netdev_get_stats(port->netdev, &stats);
2521 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2522 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2523 memset(ops->pad, 0, sizeof ops->pad);
2524 ops->rx_packets = htonll(stats.rx_packets);
2525 ops->tx_packets = htonll(stats.tx_packets);
2526 ops->rx_bytes = htonll(stats.rx_bytes);
2527 ops->tx_bytes = htonll(stats.tx_bytes);
2528 ops->rx_dropped = htonll(stats.rx_dropped);
2529 ops->tx_dropped = htonll(stats.tx_dropped);
2530 ops->rx_errors = htonll(stats.rx_errors);
2531 ops->tx_errors = htonll(stats.tx_errors);
2532 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2533 ops->rx_over_err = htonll(stats.rx_over_errors);
2534 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2535 ops->collisions = htonll(stats.collisions);
2539 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2540 struct ofp_stats_request *osr,
2543 struct ofp_port_stats_request *psr;
2544 struct ofp_port_stats *ops;
2546 struct ofport *port;
2547 unsigned int port_no;
2549 if (arg_size != sizeof *psr) {
2550 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2552 psr = (struct ofp_port_stats_request *) osr->body;
2554 msg = start_stats_reply(osr, sizeof *ops * 16);
2555 if (psr->port_no != htons(OFPP_NONE)) {
2556 port = port_array_get(&p->ports,
2557 ofp_port_to_odp_port(ntohs(psr->port_no)));
2559 append_port_stat(port, ntohs(psr->port_no), ofconn, msg);
2562 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2563 append_port_stat(port, port_no, ofconn, msg);
2567 queue_tx(msg, ofconn, ofconn->reply_counter);
2571 struct flow_stats_cbdata {
2572 struct ofproto *ofproto;
2573 struct ofconn *ofconn;
2578 /* Obtains statistic counters for 'rule' within 'p' and stores them into
2579 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
2580 * returned statistic include statistics for all of 'rule''s subrules. */
2582 query_stats(struct ofproto *p, struct rule *rule,
2583 uint64_t *packet_countp, uint64_t *byte_countp)
2585 uint64_t packet_count, byte_count;
2586 struct rule *subrule;
2587 struct odp_flow *odp_flows;
2590 /* Start from historical data for 'rule' itself that are no longer tracked
2591 * by the datapath. This counts, for example, subrules that have
2593 packet_count = rule->packet_count;
2594 byte_count = rule->byte_count;
2596 /* Prepare to ask the datapath for statistics on 'rule', or if it is
2597 * wildcarded then on all of its subrules.
2599 * Also, add any statistics that are not tracked by the datapath for each
2600 * subrule. This includes, for example, statistics for packets that were
2601 * executed "by hand" by ofproto via dpif_execute() but must be accounted
2603 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2604 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
2605 if (rule->cr.wc.wildcards) {
2607 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2608 odp_flows[i++].key = subrule->cr.flow;
2609 packet_count += subrule->packet_count;
2610 byte_count += subrule->byte_count;
2613 odp_flows[0].key = rule->cr.flow;
2616 /* Fetch up-to-date statistics from the datapath and add them in. */
2617 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2619 for (i = 0; i < n_odp_flows; i++) {
2620 struct odp_flow *odp_flow = &odp_flows[i];
2621 packet_count += odp_flow->stats.n_packets;
2622 byte_count += odp_flow->stats.n_bytes;
2627 /* Return the stats to the caller. */
2628 *packet_countp = packet_count;
2629 *byte_countp = byte_count;
2633 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2635 struct rule *rule = rule_from_cls_rule(rule_);
2636 struct flow_stats_cbdata *cbdata = cbdata_;
2637 struct ofp_flow_stats *ofs;
2638 uint64_t packet_count, byte_count;
2639 size_t act_len, len;
2640 long long int tdiff = time_msec() - rule->created;
2641 uint32_t sec = tdiff / 1000;
2642 uint32_t msec = tdiff - (sec * 1000);
2644 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2648 act_len = sizeof *rule->actions * rule->n_actions;
2649 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2651 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2653 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2654 ofs->length = htons(len);
2655 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2657 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match);
2658 ofs->duration_sec = htonl(sec);
2659 ofs->duration_nsec = htonl(msec * 1000000);
2660 ofs->cookie = rule->flow_cookie;
2661 ofs->priority = htons(rule->cr.priority);
2662 ofs->idle_timeout = htons(rule->idle_timeout);
2663 ofs->hard_timeout = htons(rule->hard_timeout);
2664 memset(ofs->pad2, 0, sizeof ofs->pad2);
2665 ofs->packet_count = htonll(packet_count);
2666 ofs->byte_count = htonll(byte_count);
2667 memcpy(ofs->actions, rule->actions, act_len);
2671 table_id_to_include(uint8_t table_id)
2673 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2674 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2675 : table_id == 0xff ? CLS_INC_ALL
2680 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2681 const struct ofp_stats_request *osr,
2684 struct ofp_flow_stats_request *fsr;
2685 struct flow_stats_cbdata cbdata;
2686 struct cls_rule target;
2688 if (arg_size != sizeof *fsr) {
2689 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2691 fsr = (struct ofp_flow_stats_request *) osr->body;
2693 COVERAGE_INC(ofproto_flows_req);
2695 cbdata.ofconn = ofconn;
2696 cbdata.out_port = fsr->out_port;
2697 cbdata.msg = start_stats_reply(osr, 1024);
2698 cls_rule_from_match(&target, &fsr->match, 0);
2699 classifier_for_each_match(&p->cls, &target,
2700 table_id_to_include(fsr->table_id),
2701 flow_stats_cb, &cbdata);
2702 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2706 struct flow_stats_ds_cbdata {
2707 struct ofproto *ofproto;
2712 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
2714 struct rule *rule = rule_from_cls_rule(rule_);
2715 struct flow_stats_ds_cbdata *cbdata = cbdata_;
2716 struct ds *results = cbdata->results;
2717 struct ofp_match match;
2718 uint64_t packet_count, byte_count;
2719 size_t act_len = sizeof *rule->actions * rule->n_actions;
2721 /* Don't report on subrules. */
2722 if (rule->super != NULL) {
2726 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2727 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &match);
2729 ds_put_format(results, "duration=%llds, ",
2730 (time_msec() - rule->created) / 1000);
2731 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2732 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2733 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2734 ofp_print_match(results, &match, true);
2735 ofp_print_actions(results, &rule->actions->header, act_len);
2736 ds_put_cstr(results, "\n");
2739 /* Adds a pretty-printed description of all flows to 'results', including
2740 * those marked hidden by secchan (e.g., by in-band control). */
2742 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2744 struct ofp_match match;
2745 struct cls_rule target;
2746 struct flow_stats_ds_cbdata cbdata;
2748 memset(&match, 0, sizeof match);
2749 match.wildcards = htonl(OFPFW_ALL);
2752 cbdata.results = results;
2754 cls_rule_from_match(&target, &match, 0);
2755 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2756 flow_stats_ds_cb, &cbdata);
2759 struct aggregate_stats_cbdata {
2760 struct ofproto *ofproto;
2762 uint64_t packet_count;
2763 uint64_t byte_count;
2768 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2770 struct rule *rule = rule_from_cls_rule(rule_);
2771 struct aggregate_stats_cbdata *cbdata = cbdata_;
2772 uint64_t packet_count, byte_count;
2774 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2778 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2780 cbdata->packet_count += packet_count;
2781 cbdata->byte_count += byte_count;
2786 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2787 const struct ofp_stats_request *osr,
2790 struct ofp_aggregate_stats_request *asr;
2791 struct ofp_aggregate_stats_reply *reply;
2792 struct aggregate_stats_cbdata cbdata;
2793 struct cls_rule target;
2796 if (arg_size != sizeof *asr) {
2797 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2799 asr = (struct ofp_aggregate_stats_request *) osr->body;
2801 COVERAGE_INC(ofproto_agg_request);
2803 cbdata.out_port = asr->out_port;
2804 cbdata.packet_count = 0;
2805 cbdata.byte_count = 0;
2807 cls_rule_from_match(&target, &asr->match, 0);
2808 classifier_for_each_match(&p->cls, &target,
2809 table_id_to_include(asr->table_id),
2810 aggregate_stats_cb, &cbdata);
2812 msg = start_stats_reply(osr, sizeof *reply);
2813 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2814 reply->flow_count = htonl(cbdata.n_flows);
2815 reply->packet_count = htonll(cbdata.packet_count);
2816 reply->byte_count = htonll(cbdata.byte_count);
2817 queue_tx(msg, ofconn, ofconn->reply_counter);
2822 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2823 struct ofp_header *oh)
2825 struct ofp_stats_request *osr;
2829 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2834 osr = (struct ofp_stats_request *) oh;
2836 switch (ntohs(osr->type)) {
2838 return handle_desc_stats_request(p, ofconn, osr);
2841 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2843 case OFPST_AGGREGATE:
2844 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2847 return handle_table_stats_request(p, ofconn, osr);
2850 return handle_port_stats_request(p, ofconn, osr, arg_size);
2853 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2856 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2860 static long long int
2861 msec_from_nsec(uint64_t sec, uint32_t nsec)
2863 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2867 update_time(struct ofproto *ofproto, struct rule *rule,
2868 const struct odp_flow_stats *stats)
2870 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2871 if (used > rule->used) {
2873 if (rule->super && used > rule->super->used) {
2874 rule->super->used = used;
2876 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
2881 update_stats(struct ofproto *ofproto, struct rule *rule,
2882 const struct odp_flow_stats *stats)
2884 if (stats->n_packets) {
2885 update_time(ofproto, rule, stats);
2886 rule->packet_count += stats->n_packets;
2887 rule->byte_count += stats->n_bytes;
2888 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
2893 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
2894 * in which no matching flow already exists in the flow table.
2896 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
2897 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
2898 * code as encoded by ofp_mkerr() on failure.
2900 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2903 add_flow(struct ofproto *p, struct ofconn *ofconn,
2904 const struct ofp_flow_mod *ofm, size_t n_actions)
2906 struct ofpbuf *packet;
2911 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
2915 flow_from_match(&flow, &wildcards, &ofm->match);
2916 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
2917 ntohs(ofm->priority))) {
2918 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
2922 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
2923 n_actions, ntohs(ofm->idle_timeout),
2924 ntohs(ofm->hard_timeout), ofm->cookie,
2925 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
2926 cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority));
2929 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2930 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2934 in_port = UINT16_MAX;
2937 rule_insert(p, rule, packet, in_port);
2938 ofpbuf_delete(packet);
2942 static struct rule *
2943 find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
2948 flow_from_match(&flow, &wildcards, &ofm->match);
2949 return rule_from_cls_rule(classifier_find_rule_exactly(
2950 &p->cls, &flow, wildcards,
2951 ntohs(ofm->priority)));
2955 send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
2956 struct rule *rule, const struct ofp_flow_mod *ofm)
2958 struct ofpbuf *packet;
2963 if (ofm->buffer_id == htonl(UINT32_MAX)) {
2967 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2973 flow_extract(packet, in_port, &flow);
2974 rule_execute(ofproto, rule, packet, &flow);
2975 ofpbuf_delete(packet);
2980 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
2982 struct modify_flows_cbdata {
2983 struct ofproto *ofproto;
2984 const struct ofp_flow_mod *ofm;
2989 static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
2990 size_t n_actions, struct rule *);
2991 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
2993 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
2994 * encoded by ofp_mkerr() on failure.
2996 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2999 modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
3000 const struct ofp_flow_mod *ofm, size_t n_actions)
3002 struct modify_flows_cbdata cbdata;
3003 struct cls_rule target;
3007 cbdata.n_actions = n_actions;
3008 cbdata.match = NULL;
3010 cls_rule_from_match(&target, &ofm->match, 0);
3012 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3013 modify_flows_cb, &cbdata);
3015 /* This credits the packet to whichever flow happened to happened to
3016 * match last. That's weird. Maybe we should do a lookup for the
3017 * flow that actually matches the packet? Who knows. */
3018 send_buffered_packet(p, ofconn, cbdata.match, ofm);
3021 return add_flow(p, ofconn, ofm, n_actions);
3025 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3026 * code as encoded by ofp_mkerr() on failure.
3028 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3031 modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
3032 struct ofp_flow_mod *ofm, size_t n_actions)
3034 struct rule *rule = find_flow_strict(p, ofm);
3035 if (rule && !rule_is_hidden(rule)) {
3036 modify_flow(p, ofm, n_actions, rule);
3037 return send_buffered_packet(p, ofconn, rule, ofm);
3039 return add_flow(p, ofconn, ofm, n_actions);
3043 /* Callback for modify_flows_loose(). */
3045 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3047 struct rule *rule = rule_from_cls_rule(rule_);
3048 struct modify_flows_cbdata *cbdata = cbdata_;
3050 if (!rule_is_hidden(rule)) {
3051 cbdata->match = rule;
3052 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
3056 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3057 * been identified as a flow in 'p''s flow table to be modified, by changing
3058 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3059 * ofp_action[] structures). */
3061 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
3062 size_t n_actions, struct rule *rule)
3064 size_t actions_len = n_actions * sizeof *rule->actions;
3066 rule->flow_cookie = ofm->cookie;
3068 /* If the actions are the same, do nothing. */
3069 if (n_actions == rule->n_actions
3070 && !memcmp(ofm->actions, rule->actions, actions_len))
3075 /* Replace actions. */
3076 free(rule->actions);
3077 rule->actions = xmemdup(ofm->actions, actions_len);
3078 rule->n_actions = n_actions;
3080 /* Make sure that the datapath gets updated properly. */
3081 if (rule->cr.wc.wildcards) {
3082 COVERAGE_INC(ofproto_mod_wc_flow);
3083 p->need_revalidate = true;
3085 rule_update_actions(p, rule);
3091 /* OFPFC_DELETE implementation. */
3093 struct delete_flows_cbdata {
3094 struct ofproto *ofproto;
3098 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3099 static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port);
3101 /* Implements OFPFC_DELETE. */
3103 delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
3105 struct delete_flows_cbdata cbdata;
3106 struct cls_rule target;
3109 cbdata.out_port = ofm->out_port;
3111 cls_rule_from_match(&target, &ofm->match, 0);
3113 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3114 delete_flows_cb, &cbdata);
3117 /* Implements OFPFC_DELETE_STRICT. */
3119 delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
3121 struct rule *rule = find_flow_strict(p, ofm);
3123 delete_flow(p, rule, ofm->out_port);
3127 /* Callback for delete_flows_loose(). */
3129 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3131 struct rule *rule = rule_from_cls_rule(rule_);
3132 struct delete_flows_cbdata *cbdata = cbdata_;
3134 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
3137 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3138 * been identified as a flow to delete from 'p''s flow table, by deleting the
3139 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3142 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3143 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3144 * specified 'out_port'. */
3146 delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port)
3148 if (rule_is_hidden(rule)) {
3152 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3156 send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
3157 rule_remove(p, rule);
3161 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3162 struct ofp_flow_mod *ofm)
3167 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3168 sizeof *ofm->actions, &n_actions);
3173 /* We do not support the emergency flow cache. It will hopefully
3174 * get dropped from OpenFlow in the near future. */
3175 if (ofm->flags & htons(OFPFF_EMERG)) {
3176 /* There isn't a good fit for an error code, so just state that the
3177 * flow table is full. */
3178 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3181 normalize_match(&ofm->match);
3182 if (!ofm->match.wildcards) {
3183 ofm->priority = htons(UINT16_MAX);
3186 error = validate_actions((const union ofp_action *) ofm->actions,
3187 n_actions, p->max_ports);
3192 switch (ntohs(ofm->command)) {
3194 return add_flow(p, ofconn, ofm, n_actions);
3197 return modify_flows_loose(p, ofconn, ofm, n_actions);
3199 case OFPFC_MODIFY_STRICT:
3200 return modify_flow_strict(p, ofconn, ofm, n_actions);
3203 delete_flows_loose(p, ofm);
3206 case OFPFC_DELETE_STRICT:
3207 delete_flow_strict(p, ofm);
3211 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3216 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3218 struct ofp_vendor_header *ovh = msg;
3219 struct nicira_header *nh;
3221 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3222 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3224 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3225 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3227 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3228 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3232 switch (ntohl(nh->subtype)) {
3233 case NXT_STATUS_REQUEST:
3234 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3238 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3242 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
3244 struct ofp_header *ob;
3247 /* Currently, everything executes synchronously, so we can just
3248 * immediately send the barrier reply. */
3249 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3250 queue_tx(buf, ofconn, ofconn->reply_counter);
3255 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3256 struct ofpbuf *ofp_msg)
3258 struct ofp_header *oh = ofp_msg->data;
3261 COVERAGE_INC(ofproto_recv_openflow);
3263 case OFPT_ECHO_REQUEST:
3264 error = handle_echo_request(ofconn, oh);
3267 case OFPT_ECHO_REPLY:
3271 case OFPT_FEATURES_REQUEST:
3272 error = handle_features_request(p, ofconn, oh);
3275 case OFPT_GET_CONFIG_REQUEST:
3276 error = handle_get_config_request(p, ofconn, oh);
3279 case OFPT_SET_CONFIG:
3280 error = handle_set_config(p, ofconn, ofp_msg->data);
3283 case OFPT_PACKET_OUT:
3284 error = handle_packet_out(p, ofconn, ofp_msg->data);
3288 error = handle_port_mod(p, oh);
3292 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3295 case OFPT_STATS_REQUEST:
3296 error = handle_stats_request(p, ofconn, oh);
3300 error = handle_vendor(p, ofconn, ofp_msg->data);
3303 case OFPT_BARRIER_REQUEST:
3304 error = handle_barrier_request(ofconn, oh);
3308 if (VLOG_IS_WARN_ENABLED()) {
3309 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3310 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3313 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3318 send_error_oh(ofconn, ofp_msg->data, error);
3323 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
3325 struct odp_msg *msg = packet->data;
3326 uint16_t in_port = odp_port_to_ofp_port(msg->port);
3328 struct ofpbuf payload;
3331 payload.data = msg + 1;
3332 payload.size = msg->length - sizeof *msg;
3333 flow_extract(&payload, msg->port, &flow);
3335 /* Check with in-band control to see if this packet should be sent
3336 * to the local port regardless of the flow table. */
3337 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3338 union odp_action action;
3340 memset(&action, 0, sizeof(action));
3341 action.output.type = ODPAT_OUTPUT;
3342 action.output.port = ODPP_LOCAL;
3343 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3346 rule = lookup_valid_rule(p, &flow);
3348 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3349 struct ofport *port = port_array_get(&p->ports, msg->port);
3351 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3352 COVERAGE_INC(ofproto_no_packet_in);
3353 /* XXX install 'drop' flow entry */
3354 ofpbuf_delete(packet);
3358 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3361 COVERAGE_INC(ofproto_packet_in);
3362 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3366 if (rule->cr.wc.wildcards) {
3367 rule = rule_create_subrule(p, rule, &flow);
3368 rule_make_actions(p, rule, packet);
3370 if (!rule->may_install) {
3371 /* The rule is not installable, that is, we need to process every
3372 * packet, so process the current packet and set its actions into
3374 rule_make_actions(p, rule, packet);
3376 /* XXX revalidate rule if it needs it */
3380 rule_execute(p, rule, &payload, &flow);
3381 rule_reinstall(p, rule);
3383 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY
3384 && rconn_is_connected(p->controller->rconn)) {
3386 * Extra-special case for fail-open mode.
3388 * We are in fail-open mode and the packet matched the fail-open rule,
3389 * but we are connected to a controller too. We should send the packet
3390 * up to the controller in the hope that it will try to set up a flow
3391 * and thereby allow us to exit fail-open.
3393 * See the top-level comment in fail-open.c for more information.
3395 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
3397 ofpbuf_delete(packet);
3402 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3404 struct odp_msg *msg = packet->data;
3406 switch (msg->type) {
3407 case _ODPL_ACTION_NR:
3408 COVERAGE_INC(ofproto_ctlr_action);
3409 pinsched_send(p->action_sched, odp_port_to_ofp_port(msg->port), packet,
3410 send_packet_in_action, p);
3413 case _ODPL_SFLOW_NR:
3415 ofproto_sflow_received(p->sflow, msg);
3417 ofpbuf_delete(packet);
3421 handle_odp_miss_msg(p, packet);
3425 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
3432 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3434 struct rule *sub = rule_from_cls_rule(sub_);
3435 struct revalidate_cbdata *cbdata = cbdata_;
3437 if (cbdata->revalidate_all
3438 || (cbdata->revalidate_subrules && sub->super)
3439 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3440 revalidate_rule(cbdata->ofproto, sub);
3445 revalidate_rule(struct ofproto *p, struct rule *rule)
3447 const flow_t *flow = &rule->cr.flow;
3449 COVERAGE_INC(ofproto_revalidate_rule);
3452 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3454 rule_remove(p, rule);
3456 } else if (super != rule->super) {
3457 COVERAGE_INC(ofproto_revalidate_moved);
3458 list_remove(&rule->list);
3459 list_push_back(&super->list, &rule->list);
3460 rule->super = super;
3461 rule->hard_timeout = super->hard_timeout;
3462 rule->idle_timeout = super->idle_timeout;
3463 rule->created = super->created;
3468 rule_update_actions(p, rule);
3472 static struct ofpbuf *
3473 compose_flow_removed(const struct rule *rule, long long int now, uint8_t reason)
3475 struct ofp_flow_removed *ofr;
3477 long long int tdiff = now - rule->created;
3478 uint32_t sec = tdiff / 1000;
3479 uint32_t msec = tdiff - (sec * 1000);
3481 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
3482 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofr->match);
3483 ofr->cookie = rule->flow_cookie;
3484 ofr->priority = htons(rule->cr.priority);
3485 ofr->reason = reason;
3486 ofr->duration_sec = htonl(sec);
3487 ofr->duration_nsec = htonl(msec * 1000000);
3488 ofr->idle_timeout = htons(rule->idle_timeout);
3489 ofr->packet_count = htonll(rule->packet_count);
3490 ofr->byte_count = htonll(rule->byte_count);
3496 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3498 assert(rule->installed);
3499 assert(!rule->cr.wc.wildcards);
3502 rule_remove(ofproto, rule);
3504 rule_uninstall(ofproto, rule);
3508 send_flow_removed(struct ofproto *p, struct rule *rule,
3509 long long int now, uint8_t reason)
3511 struct ofconn *ofconn;
3512 struct ofconn *prev;
3513 struct ofpbuf *buf = NULL;
3515 /* We limit the maximum number of queued flow expirations it by accounting
3516 * them under the counter for replies. That works because preventing
3517 * OpenFlow requests from being processed also prevents new flows from
3518 * being added (and expiring). (It also prevents processing OpenFlow
3519 * requests that would not add new flows, so it is imperfect.) */
3522 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3523 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)) {
3525 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3527 buf = compose_flow_removed(rule, now, reason);
3533 queue_tx(buf, prev, prev->reply_counter);
3539 expire_rule(struct cls_rule *cls_rule, void *p_)
3541 struct ofproto *p = p_;
3542 struct rule *rule = rule_from_cls_rule(cls_rule);
3543 long long int hard_expire, idle_expire, expire, now;
3545 hard_expire = (rule->hard_timeout
3546 ? rule->created + rule->hard_timeout * 1000
3548 idle_expire = (rule->idle_timeout
3549 && (rule->super || list_is_empty(&rule->list))
3550 ? rule->used + rule->idle_timeout * 1000
3552 expire = MIN(hard_expire, idle_expire);
3556 if (rule->installed && now >= rule->used + 5000) {
3557 uninstall_idle_flow(p, rule);
3558 } else if (!rule->cr.wc.wildcards) {
3559 active_timeout(p, rule);
3565 COVERAGE_INC(ofproto_expired);
3567 /* Update stats. This code will be a no-op if the rule expired
3568 * due to an idle timeout. */
3569 if (rule->cr.wc.wildcards) {
3570 struct rule *subrule, *next;
3571 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3572 rule_remove(p, subrule);
3575 rule_uninstall(p, rule);
3578 if (!rule_is_hidden(rule)) {
3579 send_flow_removed(p, rule, now,
3581 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
3583 rule_remove(p, rule);
3587 active_timeout(struct ofproto *ofproto, struct rule *rule)
3589 if (ofproto->netflow && !is_controller_rule(rule) &&
3590 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
3591 struct ofexpired expired;
3592 struct odp_flow odp_flow;
3594 /* Get updated flow stats. */
3595 memset(&odp_flow, 0, sizeof odp_flow);
3596 if (rule->installed) {
3597 odp_flow.key = rule->cr.flow;
3598 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
3599 dpif_flow_get(ofproto->dpif, &odp_flow);
3601 if (odp_flow.stats.n_packets) {
3602 update_time(ofproto, rule, &odp_flow.stats);
3603 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
3604 odp_flow.stats.tcp_flags);
3608 expired.flow = rule->cr.flow;
3609 expired.packet_count = rule->packet_count +
3610 odp_flow.stats.n_packets;
3611 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
3612 expired.used = rule->used;
3614 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
3616 /* Schedule us to send the accumulated records once we have
3617 * collected all of them. */
3618 poll_immediate_wake();
3623 update_used(struct ofproto *p)
3625 struct odp_flow *flows;
3630 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
3635 for (i = 0; i < n_flows; i++) {
3636 struct odp_flow *f = &flows[i];
3639 rule = rule_from_cls_rule(
3640 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3641 if (!rule || !rule->installed) {
3642 COVERAGE_INC(ofproto_unexpected_rule);
3643 dpif_flow_del(p->dpif, f);
3647 update_time(p, rule, &f->stats);
3648 rule_account(p, rule, f->stats.n_bytes);
3654 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3655 const struct ofpbuf *packet, int send_len)
3657 struct odp_msg *msg = packet->data;
3658 struct ofpbuf payload;
3662 /* Extract packet payload from 'msg'. */
3663 payload.data = msg + 1;
3664 payload.size = msg->length - sizeof *msg;
3666 /* Construct ofp_packet_in message. */
3667 reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3668 opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port), reason,
3669 &payload, send_len);
3672 rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100);
3676 send_packet_in_action(struct ofpbuf *packet, void *p_)
3678 struct ofproto *p = p_;
3679 struct ofconn *ofconn;
3680 struct odp_msg *msg;
3683 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3684 if (ofconn == p->controller || ofconn->miss_send_len) {
3685 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3688 ofpbuf_delete(packet);
3692 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3694 struct ofproto *p = p_;
3695 bool in_fail_open = p->fail_open && fail_open_is_active(p->fail_open);
3696 struct ofconn *ofconn;
3697 struct ofpbuf payload;
3698 struct odp_msg *msg;
3701 payload.data = msg + 1;
3702 payload.size = msg->length - sizeof *msg;
3703 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3704 if (ofconn->miss_send_len) {
3705 struct pktbuf *pb = ofconn->pktbuf;
3706 uint32_t buffer_id = (in_fail_open
3708 : pktbuf_save(pb, &payload, msg->port));
3709 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3711 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3714 ofpbuf_delete(packet);
3718 pick_datapath_id(const struct ofproto *ofproto)
3720 const struct ofport *port;
3722 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
3724 uint8_t ea[ETH_ADDR_LEN];
3727 error = netdev_get_etheraddr(port->netdev, ea);
3729 return eth_addr_to_uint64(ea);
3731 VLOG_WARN("could not get MAC address for %s (%s)",
3732 netdev_get_name(port->netdev), strerror(error));
3734 return ofproto->fallback_dpid;
3738 pick_fallback_dpid(void)
3740 uint8_t ea[ETH_ADDR_LEN];
3741 eth_addr_nicira_random(ea);
3742 return eth_addr_to_uint64(ea);
3746 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3747 struct odp_actions *actions, tag_type *tags,
3748 uint16_t *nf_output_iface, void *ofproto_)
3750 struct ofproto *ofproto = ofproto_;
3753 /* Drop frames for reserved multicast addresses. */
3754 if (eth_addr_is_reserved(flow->dl_dst)) {
3758 /* Learn source MAC (but don't try to learn from revalidation). */
3759 if (packet != NULL) {
3760 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3763 /* The log messages here could actually be useful in debugging,
3764 * so keep the rate limit relatively high. */
3765 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3766 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3767 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3768 ofproto_revalidate(ofproto, rev_tag);
3772 /* Determine output port. */
3773 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3775 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
3776 } else if (out_port != flow->in_port) {
3777 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3778 *nf_output_iface = out_port;
3786 static const struct ofhooks default_ofhooks = {
3788 default_normal_ofhook_cb,