2 * Copyright (c) 2009 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"
30 #include "fail-open.h"
32 #include "mac-learning.h"
36 #include "ofp-print.h"
38 #include "openflow/nicira-ext.h"
39 #include "openflow/openflow.h"
40 #include "openflow/openflow-mgmt.h"
41 #include "openvswitch/datapath-protocol.h"
45 #include "poll-loop.h"
46 #include "port-array.h"
55 #include "vconn-ssl.h"
58 #define THIS_MODULE VLM_ofproto
68 TABLEID_CLASSIFIER = 1
72 struct netdev *netdev;
73 struct ofp_phy_port opp; /* In host byte order. */
76 static void ofport_free(struct ofport *);
77 static void hton_ofp_phy_port(struct ofp_phy_port *);
79 static int xlate_actions(const union ofp_action *in, size_t n_in,
80 const flow_t *flow, struct ofproto *ofproto,
81 const struct ofpbuf *packet,
82 struct odp_actions *out, tag_type *tags,
83 bool *may_setup_flow);
88 uint16_t idle_timeout; /* In seconds from time of last use. */
89 uint16_t hard_timeout; /* In seconds from time of creation. */
90 long long int used; /* Last-used time (0 if never used). */
91 long long int created; /* Creation time. */
92 uint64_t packet_count; /* Number of packets received. */
93 uint64_t byte_count; /* Number of bytes received. */
94 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
95 uint8_t tcp_flags; /* Bitwise-OR of all TCP flags seen. */
96 uint8_t ip_tos; /* Last-seen IP type-of-service. */
97 tag_type tags; /* Tags (set only by hooks). */
99 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
100 * exact-match rule (having cr.wc.wildcards of 0) generated from the
101 * wildcard rule 'super'. In this case, 'list' is an element of the
104 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
105 * a list of subrules. A super-rule with no wildcards (where
106 * cr.wc.wildcards is 0) will never have any subrules. */
112 * A subrule has no actions (it uses the super-rule's actions). */
114 union ofp_action *actions;
118 * A super-rule with wildcard fields never has ODP actions (since the
119 * datapath only supports exact-match flows). */
120 bool installed; /* Installed in datapath? */
121 bool may_install; /* True ordinarily; false if actions must
122 * be reassessed for every packet. */
124 union odp_action *odp_actions;
128 rule_is_hidden(const struct rule *rule)
130 /* Subrules are merely an implementation detail, so hide them from the
132 if (rule->super != NULL) {
136 /* Rules with priority higher than UINT16_MAX are set up by secchan itself
137 * (e.g. by in-band control) and are intentionally hidden from the
139 if (rule->cr.priority > UINT16_MAX) {
146 static struct rule *rule_create(struct rule *super, const union ofp_action *,
147 size_t n_actions, uint16_t idle_timeout,
148 uint16_t hard_timeout);
149 static void rule_free(struct rule *);
150 static void rule_destroy(struct ofproto *, struct rule *);
151 static struct rule *rule_from_cls_rule(const struct cls_rule *);
152 static void rule_insert(struct ofproto *, struct rule *,
153 struct ofpbuf *packet, uint16_t in_port);
154 static void rule_remove(struct ofproto *, struct rule *);
155 static bool rule_make_actions(struct ofproto *, struct rule *,
156 const struct ofpbuf *packet);
157 static void rule_install(struct ofproto *, struct rule *,
158 struct rule *displaced_rule);
159 static void rule_uninstall(struct ofproto *, struct rule *);
160 static void rule_post_uninstall(struct ofproto *, struct rule *);
165 struct pktbuf *pktbuf;
169 struct rconn_packet_counter *packet_in_counter;
171 /* Number of OpenFlow messages queued as replies to OpenFlow requests, and
172 * the maximum number before we stop reading OpenFlow requests. */
173 #define OFCONN_REPLY_MAX 100
174 struct rconn_packet_counter *reply_counter;
177 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *);
178 static void ofconn_destroy(struct ofconn *, struct ofproto *);
179 static void ofconn_run(struct ofconn *, struct ofproto *);
180 static void ofconn_wait(struct ofconn *);
181 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
182 struct rconn_packet_counter *counter);
186 uint64_t datapath_id; /* Datapath ID. */
187 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
188 uint64_t mgmt_id; /* Management channel identifier. */
189 char *manufacturer; /* Manufacturer. */
190 char *hardware; /* Hardware. */
191 char *software; /* Software version. */
192 char *serial; /* Serial number. */
196 struct dpifmon *dpifmon;
197 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
199 struct shash port_by_name;
203 struct switch_status *switch_status;
204 struct status_category *ss_cat;
205 struct in_band *in_band;
206 struct discovery *discovery;
207 struct fail_open *fail_open;
208 struct pinsched *miss_sched, *action_sched;
209 struct executer *executer;
210 struct netflow *netflow;
213 struct classifier cls;
214 bool need_revalidate;
215 long long int next_expiration;
216 struct tag_set revalidate_set;
218 /* OpenFlow connections. */
219 struct list all_conns;
220 struct ofconn *controller;
221 struct pvconn **listeners;
223 struct pvconn **snoops;
226 /* Hooks for ovs-vswitchd. */
227 const struct ofhooks *ofhooks;
230 /* Used by default ofhooks. */
231 struct mac_learning *ml;
234 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
236 static const struct ofhooks default_ofhooks;
238 static uint64_t pick_datapath_id(struct dpif *, uint64_t fallback_dpid);
239 static uint64_t pick_fallback_dpid(void);
240 static void send_packet_in_miss(struct ofpbuf *, void *ofproto);
241 static void send_packet_in_action(struct ofpbuf *, void *ofproto);
242 static void update_used(struct ofproto *);
243 static void update_stats(struct rule *, const struct odp_flow_stats *);
244 static void expire_rule(struct cls_rule *, void *ofproto);
245 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
246 static void revalidate_cb(struct cls_rule *rule_, void *p_);
248 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
250 static void handle_openflow(struct ofconn *, struct ofproto *,
253 static void refresh_port_group(struct ofproto *, unsigned int group);
254 static void update_port(struct ofproto *, const char *devname);
255 static int init_ports(struct ofproto *);
256 static void reinit_ports(struct ofproto *);
259 ofproto_create(const char *datapath, const struct ofhooks *ofhooks, void *aux,
260 struct ofproto **ofprotop)
262 struct dpifmon *dpifmon;
263 struct odp_stats stats;
270 /* Connect to datapath and start listening for messages. */
271 error = dpif_open(datapath, &dpif);
273 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
276 error = dpif_get_dp_stats(&dpif, &stats);
278 VLOG_ERR("failed to obtain stats for datapath %s: %s",
279 datapath, strerror(error));
283 error = dpif_set_listen_mask(&dpif, ODPL_MISS | ODPL_ACTION);
285 VLOG_ERR("failed to listen on datapath %s: %s",
286 datapath, strerror(error));
290 dpif_flow_flush(&dpif);
293 /* Start monitoring datapath ports for status changes. */
294 error = dpifmon_create(datapath, &dpifmon);
296 VLOG_ERR("failed to starting monitoring datapath %s: %s",
297 datapath, strerror(error));
302 /* Initialize settings. */
303 p = xcalloc(1, sizeof *p);
304 p->fallback_dpid = pick_fallback_dpid();
305 p->datapath_id = pick_datapath_id(&dpif, p->fallback_dpid);
306 VLOG_INFO("using datapath ID %012"PRIx64, p->datapath_id);
307 p->manufacturer = xstrdup("Nicira Networks, Inc.");
308 p->hardware = xstrdup("Reference Implementation");
309 p->software = xstrdup(VERSION BUILDNR);
310 p->serial = xstrdup("None");
312 /* Initialize datapath. */
314 p->dpifmon = dpifmon;
315 port_array_init(&p->ports);
316 shash_init(&p->port_by_name);
317 p->max_ports = stats.max_ports;
319 /* Initialize submodules. */
320 p->switch_status = switch_status_create(p);
324 p->miss_sched = p->action_sched = NULL;
328 /* Initialize flow table. */
329 classifier_init(&p->cls);
330 p->need_revalidate = false;
331 p->next_expiration = time_msec() + 1000;
332 tag_set_init(&p->revalidate_set);
334 /* Initialize OpenFlow connections. */
335 list_init(&p->all_conns);
336 p->controller = ofconn_create(p, rconn_create(15, 15));
337 p->controller->pktbuf = pktbuf_create();
338 p->controller->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
344 /* Initialize hooks. */
346 p->ofhooks = ofhooks;
350 p->ofhooks = &default_ofhooks;
352 p->ml = mac_learning_create();
355 /* Register switch status category. */
356 p->ss_cat = switch_status_register(p->switch_status, "remote",
357 rconn_status_cb, p->controller->rconn);
360 error = init_ports(p);
371 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
373 uint64_t old_dpid = p->datapath_id;
374 p->datapath_id = (datapath_id
376 : pick_datapath_id(&p->dpif, p->fallback_dpid));
377 if (p->datapath_id != old_dpid) {
378 VLOG_INFO("datapath ID changed to %012"PRIx64, p->datapath_id);
379 rconn_reconnect(p->controller->rconn);
384 ofproto_set_mgmt_id(struct ofproto *p, uint64_t mgmt_id)
386 p->mgmt_id = mgmt_id;
390 ofproto_set_probe_interval(struct ofproto *p, int probe_interval)
392 probe_interval = probe_interval ? MAX(probe_interval, 5) : 0;
393 rconn_set_probe_interval(p->controller->rconn, probe_interval);
395 int trigger_duration = probe_interval ? probe_interval * 3 : 15;
396 fail_open_set_trigger_duration(p->fail_open, trigger_duration);
401 ofproto_set_max_backoff(struct ofproto *p, int max_backoff)
403 rconn_set_max_backoff(p->controller->rconn, max_backoff);
407 ofproto_set_desc(struct ofproto *p,
408 const char *manufacturer, const char *hardware,
409 const char *software, const char *serial)
412 free(p->manufacturer);
413 p->manufacturer = xstrdup(manufacturer);
417 p->hardware = xstrdup(hardware);
421 p->software = xstrdup(software);
425 p->serial = xstrdup(serial);
430 ofproto_set_in_band(struct ofproto *p, bool in_band)
432 if (in_band != (p->in_band != NULL)) {
434 in_band_create(p, p->switch_status, p->controller->rconn,
438 ofproto_set_discovery(p, false, NULL, true);
439 in_band_destroy(p->in_band);
442 rconn_reconnect(p->controller->rconn);
448 ofproto_set_discovery(struct ofproto *p, bool discovery,
449 const char *re, bool update_resolv_conf)
451 if (discovery != (p->discovery != NULL)) {
453 int error = ofproto_set_in_band(p, true);
457 error = discovery_create(re, update_resolv_conf,
458 &p->dpif, p->switch_status,
464 discovery_destroy(p->discovery);
467 rconn_disconnect(p->controller->rconn);
468 } else if (discovery) {
469 discovery_set_update_resolv_conf(p->discovery, update_resolv_conf);
470 return discovery_set_accept_controller_re(p->discovery, re);
476 ofproto_set_controller(struct ofproto *ofproto, const char *controller)
478 if (ofproto->discovery) {
480 } else if (controller) {
481 if (strcmp(rconn_get_name(ofproto->controller->rconn), controller)) {
482 return rconn_connect(ofproto->controller->rconn, controller);
487 rconn_disconnect(ofproto->controller->rconn);
493 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
494 const struct svec *svec)
496 struct pvconn **pvconns = *pvconnsp;
497 size_t n_pvconns = *n_pvconnsp;
501 for (i = 0; i < n_pvconns; i++) {
502 pvconn_close(pvconns[i]);
506 pvconns = xmalloc(svec->n * sizeof *pvconns);
508 for (i = 0; i < svec->n; i++) {
509 const char *name = svec->names[i];
510 struct pvconn *pvconn;
513 error = pvconn_open(name, &pvconn);
515 pvconns[n_pvconns++] = pvconn;
517 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
525 *n_pvconnsp = n_pvconns;
531 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
533 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
537 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
539 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
543 ofproto_set_netflow(struct ofproto *ofproto, const struct svec *collectors,
544 uint8_t engine_type, uint8_t engine_id, bool add_id_to_iface)
546 if (collectors && collectors->n) {
547 if (!ofproto->netflow) {
548 ofproto->netflow = netflow_create();
550 netflow_set_engine(ofproto->netflow, engine_type, engine_id,
552 return netflow_set_collectors(ofproto->netflow, collectors);
554 netflow_destroy(ofproto->netflow);
555 ofproto->netflow = NULL;
561 ofproto_set_failure(struct ofproto *ofproto, bool fail_open)
564 struct rconn *rconn = ofproto->controller->rconn;
565 int trigger_duration = rconn_get_probe_interval(rconn) * 3;
566 if (!ofproto->fail_open) {
567 ofproto->fail_open = fail_open_create(ofproto, trigger_duration,
568 ofproto->switch_status,
571 fail_open_set_trigger_duration(ofproto->fail_open,
575 fail_open_destroy(ofproto->fail_open);
576 ofproto->fail_open = NULL;
581 ofproto_set_rate_limit(struct ofproto *ofproto,
582 int rate_limit, int burst_limit)
584 if (rate_limit > 0) {
585 if (!ofproto->miss_sched) {
586 ofproto->miss_sched = pinsched_create(rate_limit, burst_limit,
587 ofproto->switch_status);
588 ofproto->action_sched = pinsched_create(rate_limit, burst_limit,
591 pinsched_set_limits(ofproto->miss_sched, rate_limit, burst_limit);
592 pinsched_set_limits(ofproto->action_sched,
593 rate_limit, burst_limit);
596 pinsched_destroy(ofproto->miss_sched);
597 ofproto->miss_sched = NULL;
598 pinsched_destroy(ofproto->action_sched);
599 ofproto->action_sched = NULL;
604 ofproto_set_stp(struct ofproto *ofproto UNUSED, bool enable_stp)
608 VLOG_WARN("STP is not yet implemented");
616 ofproto_set_remote_execution(struct ofproto *ofproto, const char *command_acl,
617 const char *command_dir)
620 if (!ofproto->executer) {
621 return executer_create(command_acl, command_dir,
624 executer_set_acl(ofproto->executer, command_acl, command_dir);
627 executer_destroy(ofproto->executer);
628 ofproto->executer = NULL;
634 ofproto_get_datapath_id(const struct ofproto *ofproto)
636 return ofproto->datapath_id;
640 ofproto_get_mgmt_id(const struct ofproto *ofproto)
642 return ofproto->mgmt_id;
646 ofproto_get_probe_interval(const struct ofproto *ofproto)
648 return rconn_get_probe_interval(ofproto->controller->rconn);
652 ofproto_get_max_backoff(const struct ofproto *ofproto)
654 return rconn_get_max_backoff(ofproto->controller->rconn);
658 ofproto_get_in_band(const struct ofproto *ofproto)
660 return ofproto->in_band != NULL;
664 ofproto_get_discovery(const struct ofproto *ofproto)
666 return ofproto->discovery != NULL;
670 ofproto_get_controller(const struct ofproto *ofproto)
672 return rconn_get_name(ofproto->controller->rconn);
676 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
680 for (i = 0; i < ofproto->n_listeners; i++) {
681 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
686 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
690 for (i = 0; i < ofproto->n_snoops; i++) {
691 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
696 ofproto_destroy(struct ofproto *p)
698 struct ofconn *ofconn, *next_ofconn;
699 struct ofport *ofport;
700 unsigned int port_no;
707 ofproto_flush_flows(p);
708 classifier_destroy(&p->cls);
710 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
712 ofconn_destroy(ofconn, p);
715 dpif_close(&p->dpif);
716 dpifmon_destroy(p->dpifmon);
717 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
720 shash_destroy(&p->port_by_name);
722 switch_status_destroy(p->switch_status);
723 in_band_destroy(p->in_band);
724 discovery_destroy(p->discovery);
725 fail_open_destroy(p->fail_open);
726 pinsched_destroy(p->miss_sched);
727 pinsched_destroy(p->action_sched);
728 executer_destroy(p->executer);
729 netflow_destroy(p->netflow);
731 switch_status_unregister(p->ss_cat);
733 for (i = 0; i < p->n_listeners; i++) {
734 pvconn_close(p->listeners[i]);
738 for (i = 0; i < p->n_snoops; i++) {
739 pvconn_close(p->snoops[i]);
743 mac_learning_destroy(p->ml);
749 ofproto_run(struct ofproto *p)
751 int error = ofproto_run1(p);
753 error = ofproto_run2(p, false);
759 ofproto_run1(struct ofproto *p)
761 struct ofconn *ofconn, *next_ofconn;
766 for (i = 0; i < 50; i++) {
770 error = dpif_recv(&p->dpif, &buf);
772 if (error == ENODEV) {
773 /* Someone destroyed the datapath behind our back. The caller
774 * better destroy us and give up, because we're just going to
775 * spin from here on out. */
776 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
777 VLOG_ERR_RL(&rl, "dp%u: datapath was destroyed externally",
784 handle_odp_msg(p, buf);
787 while ((error = dpifmon_poll(p->dpifmon, &devname)) != EAGAIN) {
788 if (error == ENOBUFS) {
791 update_port(p, devname);
797 in_band_run(p->in_band);
800 char *controller_name;
801 if (rconn_is_connectivity_questionable(p->controller->rconn)) {
802 discovery_question_connectivity(p->discovery);
804 if (discovery_run(p->discovery, &controller_name)) {
805 if (controller_name) {
806 rconn_connect(p->controller->rconn, controller_name);
808 rconn_disconnect(p->controller->rconn);
813 fail_open_run(p->fail_open);
815 pinsched_run(p->miss_sched, send_packet_in_miss, p);
816 pinsched_run(p->action_sched, send_packet_in_action, p);
818 executer_run(p->executer);
821 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
823 ofconn_run(ofconn, p);
826 for (i = 0; i < p->n_listeners; i++) {
830 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
832 ofconn_create(p, rconn_new_from_vconn("passive", vconn));
833 } else if (retval != EAGAIN) {
834 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
838 for (i = 0; i < p->n_snoops; i++) {
842 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
844 rconn_add_monitor(p->controller->rconn, vconn);
845 } else if (retval != EAGAIN) {
846 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
850 if (time_msec() >= p->next_expiration) {
851 COVERAGE_INC(ofproto_expiration);
852 p->next_expiration = time_msec() + 1000;
855 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
857 /* Let the hook know that we're at a stable point: all outstanding data
858 * in existing flows has been accounted to the account_cb. Thus, the
859 * hook can now reasonably do operations that depend on having accurate
860 * flow volume accounting (currently, that's just bond rebalancing). */
861 if (p->ofhooks->account_checkpoint_cb) {
862 p->ofhooks->account_checkpoint_cb(p->aux);
867 netflow_run(p->netflow);
873 struct revalidate_cbdata {
874 struct ofproto *ofproto;
875 bool revalidate_all; /* Revalidate all exact-match rules? */
876 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
877 struct tag_set revalidate_set; /* Set of tags to revalidate. */
881 ofproto_run2(struct ofproto *p, bool revalidate_all)
883 if (p->need_revalidate || revalidate_all
884 || !tag_set_is_empty(&p->revalidate_set)) {
885 struct revalidate_cbdata cbdata;
887 cbdata.revalidate_all = revalidate_all;
888 cbdata.revalidate_subrules = p->need_revalidate;
889 cbdata.revalidate_set = p->revalidate_set;
890 tag_set_init(&p->revalidate_set);
891 COVERAGE_INC(ofproto_revalidate);
892 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
893 p->need_revalidate = false;
900 ofproto_wait(struct ofproto *p)
902 struct ofconn *ofconn;
905 dpif_recv_wait(&p->dpif);
906 dpifmon_wait(p->dpifmon);
907 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
911 in_band_wait(p->in_band);
914 discovery_wait(p->discovery);
917 fail_open_wait(p->fail_open);
919 pinsched_wait(p->miss_sched);
920 pinsched_wait(p->action_sched);
922 executer_wait(p->executer);
924 if (!tag_set_is_empty(&p->revalidate_set)) {
925 poll_immediate_wake();
927 if (p->need_revalidate) {
928 /* Shouldn't happen, but if it does just go around again. */
929 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
930 poll_immediate_wake();
931 } else if (p->next_expiration != LLONG_MAX) {
932 poll_timer_wait(p->next_expiration - time_msec());
934 for (i = 0; i < p->n_listeners; i++) {
935 pvconn_wait(p->listeners[i]);
937 for (i = 0; i < p->n_snoops; i++) {
938 pvconn_wait(p->snoops[i]);
943 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
945 tag_set_add(&ofproto->revalidate_set, tag);
949 ofproto_get_revalidate_set(struct ofproto *ofproto)
951 return &ofproto->revalidate_set;
955 ofproto_is_alive(const struct ofproto *p)
957 return p->discovery || rconn_is_alive(p->controller->rconn);
961 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
962 const union ofp_action *actions, size_t n_actions,
963 const struct ofpbuf *packet)
965 struct odp_actions odp_actions;
968 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
974 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
976 dpif_execute(&p->dpif, flow->in_port, odp_actions.actions,
977 odp_actions.n_actions, packet);
982 ofproto_add_flow(struct ofproto *p,
983 const flow_t *flow, uint32_t wildcards, unsigned int priority,
984 const union ofp_action *actions, size_t n_actions,
988 rule = rule_create(NULL, actions, n_actions,
989 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */, 0);
990 cls_rule_from_flow(&rule->cr, flow, wildcards, priority);
991 rule_insert(p, rule, NULL, 0);
995 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
996 uint32_t wildcards, unsigned int priority)
1000 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1004 rule_remove(ofproto, rule);
1009 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1011 struct rule *rule = rule_from_cls_rule(rule_);
1012 struct ofproto *ofproto = ofproto_;
1014 /* Mark the flow as not installed, even though it might really be
1015 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1016 * There is no point in uninstalling it individually since we are about to
1017 * blow away all the flows with dpif_flow_flush(). */
1018 rule->installed = false;
1020 rule_remove(ofproto, rule);
1024 ofproto_flush_flows(struct ofproto *ofproto)
1026 COVERAGE_INC(ofproto_flush);
1027 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1028 dpif_flow_flush(&ofproto->dpif);
1029 if (ofproto->in_band) {
1030 in_band_flushed(ofproto->in_band);
1032 if (ofproto->fail_open) {
1033 fail_open_flushed(ofproto->fail_open);
1038 reinit_ports(struct ofproto *p)
1040 struct svec devnames;
1041 struct ofport *ofport;
1042 unsigned int port_no;
1043 struct odp_port *odp_ports;
1047 svec_init(&devnames);
1048 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1049 svec_add (&devnames, (char *) ofport->opp.name);
1051 dpif_port_list(&p->dpif, &odp_ports, &n_odp_ports);
1052 for (i = 0; i < n_odp_ports; i++) {
1053 svec_add (&devnames, odp_ports[i].devname);
1057 svec_sort_unique(&devnames);
1058 for (i = 0; i < devnames.n; i++) {
1059 update_port(p, devnames.names[i]);
1061 svec_destroy(&devnames);
1065 refresh_port_group(struct ofproto *p, unsigned int group)
1069 struct ofport *port;
1070 unsigned int port_no;
1072 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1074 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1076 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1077 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1078 ports[n_ports++] = port_no;
1081 dpif_port_group_set(&p->dpif, group, ports, n_ports);
1086 refresh_port_groups(struct ofproto *p)
1088 refresh_port_group(p, DP_GROUP_FLOOD);
1089 refresh_port_group(p, DP_GROUP_ALL);
1092 static struct ofport *
1093 make_ofport(const struct odp_port *odp_port)
1095 enum netdev_flags flags;
1096 struct ofport *ofport;
1097 struct netdev *netdev;
1101 error = netdev_open(odp_port->devname, NETDEV_ETH_TYPE_NONE, &netdev);
1103 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1104 "cannot be opened (%s)",
1105 odp_port->devname, odp_port->port,
1106 odp_port->devname, strerror(error));
1110 ofport = xmalloc(sizeof *ofport);
1111 ofport->netdev = netdev;
1112 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1113 memcpy(ofport->opp.hw_addr, netdev_get_etheraddr(netdev), ETH_ALEN);
1114 memcpy(ofport->opp.name, odp_port->devname,
1115 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1116 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1118 netdev_get_flags(netdev, &flags);
1119 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1121 netdev_get_carrier(netdev, &carrier);
1122 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1124 netdev_get_features(netdev,
1125 &ofport->opp.curr, &ofport->opp.advertised,
1126 &ofport->opp.supported, &ofport->opp.peer);
1131 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1133 if (port_array_get(&p->ports, odp_port->port)) {
1134 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1137 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1138 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1147 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1149 const struct ofp_phy_port *a = &a_->opp;
1150 const struct ofp_phy_port *b = &b_->opp;
1152 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1153 return (a->port_no == b->port_no
1154 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1155 && !strcmp((char *) a->name, (char *) b->name)
1156 && a->state == b->state
1157 && a->config == b->config
1158 && a->curr == b->curr
1159 && a->advertised == b->advertised
1160 && a->supported == b->supported
1161 && a->peer == b->peer);
1165 send_port_status(struct ofproto *p, const struct ofport *ofport,
1168 /* XXX Should limit the number of queued port status change messages. */
1169 struct ofconn *ofconn;
1170 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1171 struct ofp_port_status *ops;
1174 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1175 ops->reason = reason;
1176 ops->desc = ofport->opp;
1177 hton_ofp_phy_port(&ops->desc);
1178 queue_tx(b, ofconn, NULL);
1180 if (p->ofhooks->port_changed_cb) {
1181 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1186 ofport_install(struct ofproto *p, struct ofport *ofport)
1188 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no),
1190 shash_add(&p->port_by_name, (char *) ofport->opp.name, ofport);
1194 ofport_remove(struct ofproto *p, struct ofport *ofport)
1196 port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no), NULL);
1197 shash_delete(&p->port_by_name,
1198 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1202 ofport_free(struct ofport *ofport)
1205 netdev_close(ofport->netdev);
1211 update_port(struct ofproto *p, const char *devname)
1213 struct odp_port odp_port;
1214 struct ofport *ofport;
1217 COVERAGE_INC(ofproto_update_port);
1218 ofport = shash_find_data(&p->port_by_name, devname);
1219 error = dpif_port_query_by_name(&p->dpif, devname, &odp_port);
1223 if (!ofport_conflicts(p, &odp_port)) {
1224 ofport = make_ofport(&odp_port);
1226 ofport_install(p, ofport);
1227 send_port_status(p, ofport, OFPPR_ADD);
1231 /* Modified port. */
1232 struct ofport *new_ofport = make_ofport(&odp_port);
1237 new_ofport->opp.config &= OFPPC_PORT_DOWN;
1238 new_ofport->opp.config |= ofport->opp.config & ~OFPPC_PORT_DOWN;
1239 if (ofport_equal(ofport, new_ofport)) {
1240 /* False alarm--no change. */
1241 ofport_free(new_ofport);
1243 ofport_remove(p, ofport);
1244 ofport_install(p, new_ofport);
1245 ofport_free(ofport);
1246 send_port_status(p, new_ofport, OFPPR_MODIFY);
1249 } else if (error == ENOENT || error == ENODEV) {
1252 send_port_status(p, ofport, OFPPR_DELETE);
1253 ofport_remove(p, ofport);
1254 ofport_free(ofport);
1257 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1258 "%s", strerror(error));
1261 refresh_port_groups(p);
1265 init_ports(struct ofproto *p)
1267 struct odp_port *ports;
1272 error = dpif_port_list(&p->dpif, &ports, &n_ports);
1277 for (i = 0; i < n_ports; i++) {
1278 const struct odp_port *odp_port = &ports[i];
1279 if (!ofport_conflicts(p, odp_port)) {
1280 struct ofport *ofport = make_ofport(odp_port);
1282 ofport_install(p, ofport);
1287 refresh_port_groups(p);
1291 static struct ofconn *
1292 ofconn_create(struct ofproto *p, struct rconn *rconn)
1294 struct ofconn *ofconn = xmalloc(sizeof *ofconn);
1295 list_push_back(&p->all_conns, &ofconn->node);
1296 ofconn->rconn = rconn;
1297 ofconn->pktbuf = NULL;
1298 ofconn->send_flow_exp = false;
1299 ofconn->miss_send_len = 0;
1300 ofconn->packet_in_counter = rconn_packet_counter_create ();
1301 ofconn->reply_counter = rconn_packet_counter_create ();
1306 ofconn_destroy(struct ofconn *ofconn, struct ofproto *p)
1309 executer_rconn_closing(p->executer, ofconn->rconn);
1312 list_remove(&ofconn->node);
1313 rconn_destroy(ofconn->rconn);
1314 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1315 rconn_packet_counter_destroy(ofconn->reply_counter);
1316 pktbuf_destroy(ofconn->pktbuf);
1321 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1325 rconn_run(ofconn->rconn);
1327 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1328 /* Limit the number of iterations to prevent other tasks from
1330 for (iteration = 0; iteration < 50; iteration++) {
1331 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1335 handle_openflow(ofconn, p, of_msg);
1336 ofpbuf_delete(of_msg);
1340 if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) {
1341 ofconn_destroy(ofconn, p);
1346 ofconn_wait(struct ofconn *ofconn)
1348 rconn_run_wait(ofconn->rconn);
1349 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1350 rconn_recv_wait(ofconn->rconn);
1352 COVERAGE_INC(ofproto_ofconn_stuck);
1356 /* Caller is responsible for initializing the 'cr' member of the returned
1358 static struct rule *
1359 rule_create(struct rule *super,
1360 const union ofp_action *actions, size_t n_actions,
1361 uint16_t idle_timeout, uint16_t hard_timeout)
1363 struct rule *rule = xcalloc(1, sizeof *rule);
1364 rule->idle_timeout = idle_timeout;
1365 rule->hard_timeout = hard_timeout;
1366 rule->used = rule->created = time_msec();
1367 rule->super = super;
1369 list_push_back(&super->list, &rule->list);
1371 list_init(&rule->list);
1373 rule->n_actions = n_actions;
1374 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1378 static struct rule *
1379 rule_from_cls_rule(const struct cls_rule *cls_rule)
1381 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1385 rule_free(struct rule *rule)
1387 free(rule->actions);
1388 free(rule->odp_actions);
1392 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1393 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1394 * through all of its subrules and revalidates them, destroying any that no
1395 * longer has a super-rule (which is probably all of them).
1397 * Before calling this function, the caller must make have removed 'rule' from
1398 * the classifier. If 'rule' is an exact-match rule, the caller is also
1399 * responsible for ensuring that it has been uninstalled from the datapath. */
1401 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1404 struct rule *subrule, *next;
1405 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1406 revalidate_rule(ofproto, subrule);
1409 list_remove(&rule->list);
1415 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1417 const union ofp_action *oa;
1418 struct actions_iterator i;
1420 if (out_port == htons(OFPP_NONE)) {
1423 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1424 oa = actions_next(&i)) {
1425 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1432 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1433 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1435 * The flow that 'packet' actually contains does not need to actually match
1436 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1437 * the packet and byte counters for 'rule' will be credited for the packet sent
1438 * out whether or not the packet actually matches 'rule'.
1440 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1441 * the caller must already have accurately composed ODP actions for it given
1442 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1443 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1444 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1445 * actions and apply them to 'packet'. */
1447 rule_execute(struct ofproto *ofproto, struct rule *rule,
1448 struct ofpbuf *packet, const flow_t *flow)
1450 const union odp_action *actions;
1452 struct odp_actions a;
1454 /* Grab or compose the ODP actions.
1456 * The special case for an exact-match 'rule' where 'flow' is not the
1457 * rule's flow is important to avoid, e.g., sending a packet out its input
1458 * port simply because the ODP actions were composed for the wrong
1460 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1461 struct rule *super = rule->super ? rule->super : rule;
1462 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1463 packet, &a, NULL, 0)) {
1466 actions = a.actions;
1467 n_actions = a.n_actions;
1469 actions = rule->odp_actions;
1470 n_actions = rule->n_odp_actions;
1473 /* Execute the ODP actions. */
1474 if (!dpif_execute(&ofproto->dpif, flow->in_port,
1475 actions, n_actions, packet)) {
1476 struct odp_flow_stats stats;
1477 flow_extract_stats(flow, packet, &stats);
1478 update_stats(rule, &stats);
1479 rule->used = time_msec();
1484 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1487 struct rule *displaced_rule;
1489 /* Insert the rule in the classifier. */
1490 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1491 if (!rule->cr.wc.wildcards) {
1492 rule_make_actions(p, rule, packet);
1495 /* Send the packet and credit it to the rule. */
1498 flow_extract(packet, in_port, &flow);
1499 rule_execute(p, rule, packet, &flow);
1502 /* Install the rule in the datapath only after sending the packet, to
1503 * avoid packet reordering. */
1504 if (rule->cr.wc.wildcards) {
1505 COVERAGE_INC(ofproto_add_wc_flow);
1506 p->need_revalidate = true;
1508 rule_install(p, rule, displaced_rule);
1511 /* Free the rule that was displaced, if any. */
1512 if (displaced_rule) {
1513 rule_destroy(p, displaced_rule);
1517 static struct rule *
1518 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1521 struct rule *subrule = rule_create(rule, NULL, 0,
1522 rule->idle_timeout, rule->hard_timeout);
1523 COVERAGE_INC(ofproto_subrule_create);
1524 cls_rule_from_flow(&subrule->cr, flow, 0,
1525 (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1526 : rule->cr.priority));
1527 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1533 rule_remove(struct ofproto *ofproto, struct rule *rule)
1535 if (rule->cr.wc.wildcards) {
1536 COVERAGE_INC(ofproto_del_wc_flow);
1537 ofproto->need_revalidate = true;
1539 rule_uninstall(ofproto, rule);
1541 classifier_remove(&ofproto->cls, &rule->cr);
1542 rule_destroy(ofproto, rule);
1545 /* Returns true if the actions changed, false otherwise. */
1547 rule_make_actions(struct ofproto *p, struct rule *rule,
1548 const struct ofpbuf *packet)
1550 const struct rule *super;
1551 struct odp_actions a;
1554 assert(!rule->cr.wc.wildcards);
1556 super = rule->super ? rule->super : rule;
1558 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1559 packet, &a, &rule->tags, &rule->may_install);
1561 actions_len = a.n_actions * sizeof *a.actions;
1562 if (rule->n_odp_actions != a.n_actions
1563 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1564 COVERAGE_INC(ofproto_odp_unchanged);
1565 free(rule->odp_actions);
1566 rule->n_odp_actions = a.n_actions;
1567 rule->odp_actions = xmemdup(a.actions, actions_len);
1575 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1576 struct odp_flow_put *put)
1578 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1579 put->flow.key = rule->cr.flow;
1580 put->flow.actions = rule->odp_actions;
1581 put->flow.n_actions = rule->n_odp_actions;
1583 return dpif_flow_put(&ofproto->dpif, put);
1587 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1589 assert(!rule->cr.wc.wildcards);
1591 if (rule->may_install) {
1592 struct odp_flow_put put;
1593 if (!do_put_flow(p, rule,
1594 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1596 rule->installed = true;
1597 if (displaced_rule) {
1598 update_stats(rule, &put.flow.stats);
1599 rule_post_uninstall(p, displaced_rule);
1602 } else if (displaced_rule) {
1603 rule_uninstall(p, displaced_rule);
1608 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1610 if (rule->installed) {
1611 struct odp_flow_put put;
1612 COVERAGE_INC(ofproto_dp_missed);
1613 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1615 rule_install(ofproto, rule, NULL);
1620 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1622 bool actions_changed = rule_make_actions(ofproto, rule, NULL);
1623 if (rule->may_install) {
1624 if (rule->installed) {
1625 if (actions_changed) {
1626 /* XXX should really do rule_post_uninstall() for the *old* set
1627 * of actions, and distinguish the old stats from the new. */
1628 struct odp_flow_put put;
1629 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1632 rule_install(ofproto, rule, NULL);
1635 rule_uninstall(ofproto, rule);
1640 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1642 uint64_t total_bytes = rule->byte_count + extra_bytes;
1644 if (ofproto->ofhooks->account_flow_cb
1645 && total_bytes > rule->accounted_bytes)
1647 ofproto->ofhooks->account_flow_cb(
1648 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1649 total_bytes - rule->accounted_bytes, ofproto->aux);
1650 rule->accounted_bytes = total_bytes;
1655 rule_uninstall(struct ofproto *p, struct rule *rule)
1657 assert(!rule->cr.wc.wildcards);
1658 if (rule->installed) {
1659 struct odp_flow odp_flow;
1661 odp_flow.key = rule->cr.flow;
1662 odp_flow.actions = NULL;
1663 odp_flow.n_actions = 0;
1664 if (!dpif_flow_del(&p->dpif, &odp_flow)) {
1665 update_stats(rule, &odp_flow.stats);
1667 rule->installed = false;
1669 rule_post_uninstall(p, rule);
1674 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1676 struct rule *super = rule->super;
1678 rule_account(ofproto, rule, 0);
1679 if (ofproto->netflow) {
1680 struct ofexpired expired;
1681 expired.flow = rule->cr.flow;
1682 expired.packet_count = rule->packet_count;
1683 expired.byte_count = rule->byte_count;
1684 expired.used = rule->used;
1685 expired.created = rule->created;
1686 expired.tcp_flags = rule->tcp_flags;
1687 expired.ip_tos = rule->ip_tos;
1688 netflow_expire(ofproto->netflow, &expired);
1691 super->packet_count += rule->packet_count;
1692 super->byte_count += rule->byte_count;
1693 super->tcp_flags |= rule->tcp_flags;
1694 if (rule->packet_count) {
1695 super->ip_tos = rule->ip_tos;
1699 /* Reset counters to prevent double counting if the rule ever gets
1701 rule->packet_count = 0;
1702 rule->byte_count = 0;
1703 rule->accounted_bytes = 0;
1704 rule->tcp_flags = 0;
1709 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1710 struct rconn_packet_counter *counter)
1712 update_openflow_length(msg);
1713 if (rconn_send(ofconn->rconn, msg, counter)) {
1719 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1720 int error, const void *data, size_t len)
1723 struct ofp_error_msg *oem;
1725 if (!(error >> 16)) {
1726 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1731 COVERAGE_INC(ofproto_error);
1732 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1733 oh ? oh->xid : 0, &buf);
1734 oem->type = htons((unsigned int) error >> 16);
1735 oem->code = htons(error & 0xffff);
1736 memcpy(oem->data, data, len);
1737 queue_tx(buf, ofconn, ofconn->reply_counter);
1741 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1744 size_t oh_length = ntohs(oh->length);
1745 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1749 hton_ofp_phy_port(struct ofp_phy_port *opp)
1751 opp->port_no = htons(opp->port_no);
1752 opp->config = htonl(opp->config);
1753 opp->state = htonl(opp->state);
1754 opp->curr = htonl(opp->curr);
1755 opp->advertised = htonl(opp->advertised);
1756 opp->supported = htonl(opp->supported);
1757 opp->peer = htonl(opp->peer);
1761 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
1763 struct ofp_header *rq = oh;
1764 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
1769 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
1770 struct ofp_header *oh)
1772 struct ofp_switch_features *osf;
1774 unsigned int port_no;
1775 struct ofport *port;
1777 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1778 osf->datapath_id = htonll(p->datapath_id);
1779 osf->n_buffers = htonl(pktbuf_capacity());
1781 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1782 OFPC_PORT_STATS | OFPC_MULTI_PHY_TX);
1783 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1784 (1u << OFPAT_SET_VLAN_VID) |
1785 (1u << OFPAT_SET_VLAN_PCP) |
1786 (1u << OFPAT_STRIP_VLAN) |
1787 (1u << OFPAT_SET_DL_SRC) |
1788 (1u << OFPAT_SET_DL_DST) |
1789 (1u << OFPAT_SET_NW_SRC) |
1790 (1u << OFPAT_SET_NW_DST) |
1791 (1u << OFPAT_SET_TP_SRC) |
1792 (1u << OFPAT_SET_TP_DST));
1794 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1795 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1798 queue_tx(buf, ofconn, ofconn->reply_counter);
1803 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
1804 struct ofp_header *oh)
1807 struct ofp_switch_config *osc;
1811 /* Figure out flags. */
1812 dpif_get_drop_frags(&p->dpif, &drop_frags);
1813 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1814 if (ofconn->send_flow_exp) {
1815 flags |= OFPC_SEND_FLOW_EXP;
1819 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1820 osc->flags = htons(flags);
1821 osc->miss_send_len = htons(ofconn->miss_send_len);
1822 queue_tx(buf, ofconn, ofconn->reply_counter);
1828 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
1829 struct ofp_switch_config *osc)
1834 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
1838 flags = ntohs(osc->flags);
1840 ofconn->send_flow_exp = (flags & OFPC_SEND_FLOW_EXP) != 0;
1842 if (ofconn == p->controller) {
1843 switch (flags & OFPC_FRAG_MASK) {
1844 case OFPC_FRAG_NORMAL:
1845 dpif_set_drop_frags(&p->dpif, false);
1847 case OFPC_FRAG_DROP:
1848 dpif_set_drop_frags(&p->dpif, true);
1851 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1857 if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) {
1858 if (ntohs(osc->miss_send_len) != 0) {
1859 ofconn->pktbuf = pktbuf_create();
1861 pktbuf_destroy(ofconn->pktbuf);
1865 ofconn->miss_send_len = ntohs(osc->miss_send_len);
1871 add_output_group_action(struct odp_actions *actions, uint16_t group)
1873 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
1877 add_controller_action(struct odp_actions *actions,
1878 const struct ofp_action_output *oao)
1880 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
1881 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
1884 struct action_xlate_ctx {
1886 const flow_t *flow; /* Flow to which these actions correspond. */
1887 int recurse; /* Recursion level, via xlate_table_action. */
1888 struct ofproto *ofproto;
1889 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
1890 * null pointer if we are revalidating
1891 * without a packet to refer to. */
1894 struct odp_actions *out; /* Datapath actions. */
1895 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
1896 bool may_setup_flow; /* True ordinarily; false if the actions must
1897 * be reassessed for every packet. */
1900 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
1901 struct action_xlate_ctx *ctx);
1904 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
1906 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
1907 if (!ofport || !(ofport->opp.config & OFPPC_NO_FWD)) {
1908 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
1912 static struct rule *
1913 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
1916 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
1918 /* The rule we found might not be valid, since we could be in need of
1919 * revalidation. If it is not valid, don't return it. */
1922 && ofproto->need_revalidate
1923 && !revalidate_rule(ofproto, rule)) {
1924 COVERAGE_INC(ofproto_invalidated);
1932 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
1934 if (!ctx->recurse) {
1939 flow.in_port = in_port;
1941 rule = lookup_valid_rule(ctx->ofproto, &flow);
1948 do_xlate_actions(rule->actions, rule->n_actions, ctx);
1955 xlate_output_action(struct action_xlate_ctx *ctx,
1956 const struct ofp_action_output *oao)
1960 switch (ntohs(oao->port)) {
1962 add_output_action(ctx, ctx->flow->in_port);
1965 xlate_table_action(ctx, ctx->flow->in_port);
1968 if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet,
1969 ctx->out, ctx->tags,
1970 ctx->ofproto->aux)) {
1971 COVERAGE_INC(ofproto_uninstallable);
1972 ctx->may_setup_flow = false;
1976 add_output_group_action(ctx->out, DP_GROUP_FLOOD);
1979 add_output_group_action(ctx->out, DP_GROUP_ALL);
1981 case OFPP_CONTROLLER:
1982 add_controller_action(ctx->out, oao);
1985 add_output_action(ctx, ODPP_LOCAL);
1988 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
1989 if (odp_port != ctx->flow->in_port) {
1990 add_output_action(ctx, odp_port);
1997 xlate_nicira_action(struct action_xlate_ctx *ctx,
1998 const struct nx_action_header *nah)
2000 const struct nx_action_resubmit *nar;
2001 int subtype = ntohs(nah->subtype);
2003 assert(nah->vendor == htonl(NX_VENDOR_ID));
2005 case NXAST_RESUBMIT:
2006 nar = (const struct nx_action_resubmit *) nah;
2007 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2011 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2017 do_xlate_actions(const union ofp_action *in, size_t n_in,
2018 struct action_xlate_ctx *ctx)
2020 struct actions_iterator iter;
2021 const union ofp_action *ia;
2022 const struct ofport *port;
2024 port = port_array_get(&ctx->ofproto->ports, ctx->flow->in_port);
2025 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2026 port->opp.config & (eth_addr_equals(ctx->flow->dl_dst, stp_eth_addr)
2027 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2028 /* Drop this flow. */
2032 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2033 uint16_t type = ntohs(ia->type);
2034 union odp_action *oa;
2038 xlate_output_action(ctx, &ia->output);
2041 case OFPAT_SET_VLAN_VID:
2042 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2043 oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2046 case OFPAT_SET_VLAN_PCP:
2047 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2048 oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2051 case OFPAT_STRIP_VLAN:
2052 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2055 case OFPAT_SET_DL_SRC:
2056 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2057 memcpy(oa->dl_addr.dl_addr,
2058 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2061 case OFPAT_SET_DL_DST:
2062 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2063 memcpy(oa->dl_addr.dl_addr,
2064 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2067 case OFPAT_SET_NW_SRC:
2068 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2069 oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2072 case OFPAT_SET_TP_SRC:
2073 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2074 oa->tp_port.tp_port = ia->tp_port.tp_port;
2078 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2082 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2089 xlate_actions(const union ofp_action *in, size_t n_in,
2090 const flow_t *flow, struct ofproto *ofproto,
2091 const struct ofpbuf *packet,
2092 struct odp_actions *out, tag_type *tags, bool *may_setup_flow)
2094 tag_type no_tags = 0;
2095 struct action_xlate_ctx ctx;
2096 COVERAGE_INC(ofproto_ofp2odp);
2097 odp_actions_init(out);
2100 ctx.ofproto = ofproto;
2101 ctx.packet = packet;
2103 ctx.tags = tags ? tags : &no_tags;
2104 ctx.may_setup_flow = true;
2105 do_xlate_actions(in, n_in, &ctx);
2106 if (may_setup_flow) {
2107 *may_setup_flow = ctx.may_setup_flow;
2109 if (odp_actions_overflow(out)) {
2110 odp_actions_init(out);
2111 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2117 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2118 struct ofp_header *oh)
2120 struct ofp_packet_out *opo;
2121 struct ofpbuf payload, *buffer;
2122 struct odp_actions actions;
2128 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2132 opo = (struct ofp_packet_out *) oh;
2134 COVERAGE_INC(ofproto_packet_out);
2135 if (opo->buffer_id != htonl(UINT32_MAX)) {
2136 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2146 flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2147 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2148 &flow, p, &payload, &actions, NULL, NULL);
2153 dpif_execute(&p->dpif, flow.in_port, actions.actions, actions.n_actions,
2155 ofpbuf_delete(buffer);
2161 update_port_config(struct ofproto *p, struct ofport *port,
2162 uint32_t config, uint32_t mask)
2164 mask &= config ^ port->opp.config;
2165 if (mask & OFPPC_PORT_DOWN) {
2166 if (config & OFPPC_PORT_DOWN) {
2167 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2169 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2172 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2173 if (mask & REVALIDATE_BITS) {
2174 COVERAGE_INC(ofproto_costly_flags);
2175 port->opp.config ^= mask & REVALIDATE_BITS;
2176 p->need_revalidate = true;
2178 #undef REVALIDATE_BITS
2179 if (mask & OFPPC_NO_FLOOD) {
2180 port->opp.config ^= OFPPC_NO_FLOOD;
2181 refresh_port_group(p, DP_GROUP_FLOOD);
2183 if (mask & OFPPC_NO_PACKET_IN) {
2184 port->opp.config ^= OFPPC_NO_PACKET_IN;
2189 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2191 const struct ofp_port_mod *opm;
2192 struct ofport *port;
2195 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2199 opm = (struct ofp_port_mod *) oh;
2201 port = port_array_get(&p->ports,
2202 ofp_port_to_odp_port(ntohs(opm->port_no)));
2204 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2205 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2206 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2208 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2209 if (opm->advertise) {
2210 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2216 static struct ofpbuf *
2217 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2219 struct ofp_stats_reply *osr;
2222 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2223 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2225 osr->flags = htons(0);
2229 static struct ofpbuf *
2230 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2232 return make_stats_reply(request->header.xid, request->type, body_len);
2236 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2238 struct ofpbuf *msg = *msgp;
2239 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2240 if (nbytes + msg->size > UINT16_MAX) {
2241 struct ofp_stats_reply *reply = msg->data;
2242 reply->flags = htons(OFPSF_REPLY_MORE);
2243 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2244 queue_tx(msg, ofconn, ofconn->reply_counter);
2246 return ofpbuf_put_uninit(*msgp, nbytes);
2250 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2251 struct ofp_stats_request *request)
2253 struct ofp_desc_stats *ods;
2256 msg = start_stats_reply(request, sizeof *ods);
2257 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2258 strncpy(ods->mfr_desc, p->manufacturer, sizeof ods->mfr_desc);
2259 strncpy(ods->hw_desc, p->hardware, sizeof ods->hw_desc);
2260 strncpy(ods->sw_desc, p->software, sizeof ods->sw_desc);
2261 strncpy(ods->serial_num, p->serial, sizeof ods->serial_num);
2262 queue_tx(msg, ofconn, ofconn->reply_counter);
2268 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2270 struct rule *rule = rule_from_cls_rule(cls_rule);
2271 int *n_subrules = n_subrules_;
2279 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2280 struct ofp_stats_request *request)
2282 struct ofp_table_stats *ots;
2284 struct odp_stats dpstats;
2285 int n_exact, n_subrules, n_wild;
2287 msg = start_stats_reply(request, sizeof *ots * 2);
2289 /* Count rules of various kinds. */
2291 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2292 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2293 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2296 dpif_get_dp_stats(&p->dpif, &dpstats);
2297 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2298 memset(ots, 0, sizeof *ots);
2299 ots->table_id = TABLEID_HASH;
2300 strcpy(ots->name, "hash");
2301 ots->wildcards = htonl(0);
2302 ots->max_entries = htonl(dpstats.max_capacity);
2303 ots->active_count = htonl(n_exact);
2304 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2306 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2308 /* Classifier table. */
2309 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2310 memset(ots, 0, sizeof *ots);
2311 ots->table_id = TABLEID_CLASSIFIER;
2312 strcpy(ots->name, "classifier");
2313 ots->wildcards = htonl(OFPFW_ALL);
2314 ots->max_entries = htonl(65536);
2315 ots->active_count = htonl(n_wild);
2316 ots->lookup_count = htonll(0); /* XXX */
2317 ots->matched_count = htonll(0); /* XXX */
2319 queue_tx(msg, ofconn, ofconn->reply_counter);
2324 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2325 struct ofp_stats_request *request)
2327 struct ofp_port_stats *ops;
2329 struct ofport *port;
2330 unsigned int port_no;
2332 msg = start_stats_reply(request, sizeof *ops * 16);
2333 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2334 struct netdev_stats stats;
2336 /* Intentionally ignore return value, since errors will set 'stats' to
2337 * all-1s, which is correct for OpenFlow, and netdev_get_stats() will
2339 netdev_get_stats(port->netdev, &stats);
2341 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2342 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2343 memset(ops->pad, 0, sizeof ops->pad);
2344 ops->rx_packets = htonll(stats.rx_packets);
2345 ops->tx_packets = htonll(stats.tx_packets);
2346 ops->rx_bytes = htonll(stats.rx_bytes);
2347 ops->tx_bytes = htonll(stats.tx_bytes);
2348 ops->rx_dropped = htonll(stats.rx_dropped);
2349 ops->tx_dropped = htonll(stats.tx_dropped);
2350 ops->rx_errors = htonll(stats.rx_errors);
2351 ops->tx_errors = htonll(stats.tx_errors);
2352 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2353 ops->rx_over_err = htonll(stats.rx_over_errors);
2354 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2355 ops->collisions = htonll(stats.collisions);
2358 queue_tx(msg, ofconn, ofconn->reply_counter);
2362 struct flow_stats_cbdata {
2363 struct ofproto *ofproto;
2364 struct ofconn *ofconn;
2370 query_stats(struct ofproto *p, struct rule *rule,
2371 uint64_t *packet_countp, uint64_t *byte_countp)
2373 uint64_t packet_count, byte_count;
2374 struct rule *subrule;
2375 struct odp_flow *odp_flows;
2378 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2379 odp_flows = xcalloc(1, n_odp_flows * sizeof *odp_flows);
2380 if (rule->cr.wc.wildcards) {
2382 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2383 odp_flows[i++].key = subrule->cr.flow;
2386 odp_flows[0].key = rule->cr.flow;
2389 packet_count = rule->packet_count;
2390 byte_count = rule->byte_count;
2391 if (!dpif_flow_get_multiple(&p->dpif, odp_flows, n_odp_flows)) {
2393 for (i = 0; i < n_odp_flows; i++) {
2394 struct odp_flow *odp_flow = &odp_flows[i];
2395 packet_count += odp_flow->stats.n_packets;
2396 byte_count += odp_flow->stats.n_bytes;
2401 *packet_countp = packet_count;
2402 *byte_countp = byte_count;
2406 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2408 struct rule *rule = rule_from_cls_rule(rule_);
2409 struct flow_stats_cbdata *cbdata = cbdata_;
2410 struct ofp_flow_stats *ofs;
2411 uint64_t packet_count, byte_count;
2412 size_t act_len, len;
2414 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2418 act_len = sizeof *rule->actions * rule->n_actions;
2419 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2421 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2423 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2424 ofs->length = htons(len);
2425 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2427 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match);
2428 ofs->duration = htonl((time_msec() - rule->created) / 1000);
2429 ofs->priority = htons(rule->cr.priority);
2430 ofs->idle_timeout = htons(rule->idle_timeout);
2431 ofs->hard_timeout = htons(rule->hard_timeout);
2432 memset(ofs->pad2, 0, sizeof ofs->pad2);
2433 ofs->packet_count = htonll(packet_count);
2434 ofs->byte_count = htonll(byte_count);
2435 memcpy(ofs->actions, rule->actions, act_len);
2439 table_id_to_include(uint8_t table_id)
2441 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2442 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2443 : table_id == 0xff ? CLS_INC_ALL
2448 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2449 const struct ofp_stats_request *osr,
2452 struct ofp_flow_stats_request *fsr;
2453 struct flow_stats_cbdata cbdata;
2454 struct cls_rule target;
2456 if (arg_size != sizeof *fsr) {
2457 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2459 fsr = (struct ofp_flow_stats_request *) osr->body;
2461 COVERAGE_INC(ofproto_flows_req);
2463 cbdata.ofconn = ofconn;
2464 cbdata.out_port = fsr->out_port;
2465 cbdata.msg = start_stats_reply(osr, 1024);
2466 cls_rule_from_match(&target, &fsr->match, 0);
2467 classifier_for_each_match(&p->cls, &target,
2468 table_id_to_include(fsr->table_id),
2469 flow_stats_cb, &cbdata);
2470 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2474 struct aggregate_stats_cbdata {
2475 struct ofproto *ofproto;
2477 uint64_t packet_count;
2478 uint64_t byte_count;
2483 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2485 struct rule *rule = rule_from_cls_rule(rule_);
2486 struct aggregate_stats_cbdata *cbdata = cbdata_;
2487 uint64_t packet_count, byte_count;
2489 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2493 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2495 cbdata->packet_count += packet_count;
2496 cbdata->byte_count += byte_count;
2501 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2502 const struct ofp_stats_request *osr,
2505 struct ofp_aggregate_stats_request *asr;
2506 struct ofp_aggregate_stats_reply *reply;
2507 struct aggregate_stats_cbdata cbdata;
2508 struct cls_rule target;
2511 if (arg_size != sizeof *asr) {
2512 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2514 asr = (struct ofp_aggregate_stats_request *) osr->body;
2516 COVERAGE_INC(ofproto_agg_request);
2518 cbdata.out_port = asr->out_port;
2519 cbdata.packet_count = 0;
2520 cbdata.byte_count = 0;
2522 cls_rule_from_match(&target, &asr->match, 0);
2523 classifier_for_each_match(&p->cls, &target,
2524 table_id_to_include(asr->table_id),
2525 aggregate_stats_cb, &cbdata);
2527 msg = start_stats_reply(osr, sizeof *reply);
2528 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2529 reply->flow_count = htonl(cbdata.n_flows);
2530 reply->packet_count = htonll(cbdata.packet_count);
2531 reply->byte_count = htonll(cbdata.byte_count);
2532 queue_tx(msg, ofconn, ofconn->reply_counter);
2537 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2538 struct ofp_header *oh)
2540 struct ofp_stats_request *osr;
2544 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2549 osr = (struct ofp_stats_request *) oh;
2551 switch (ntohs(osr->type)) {
2553 return handle_desc_stats_request(p, ofconn, osr);
2556 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2558 case OFPST_AGGREGATE:
2559 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2562 return handle_table_stats_request(p, ofconn, osr);
2565 return handle_port_stats_request(p, ofconn, osr);
2568 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2571 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2575 static long long int
2576 msec_from_nsec(uint64_t sec, uint32_t nsec)
2578 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2582 update_time(struct rule *rule, const struct odp_flow_stats *stats)
2584 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2585 if (used > rule->used) {
2591 update_stats(struct rule *rule, const struct odp_flow_stats *stats)
2593 update_time(rule, stats);
2594 rule->packet_count += stats->n_packets;
2595 rule->byte_count += stats->n_bytes;
2596 rule->tcp_flags |= stats->tcp_flags;
2597 if (stats->n_packets) {
2598 rule->ip_tos = stats->ip_tos;
2603 add_flow(struct ofproto *p, struct ofconn *ofconn,
2604 struct ofp_flow_mod *ofm, size_t n_actions)
2606 struct ofpbuf *packet;
2611 rule = rule_create(NULL, (const union ofp_action *) ofm->actions,
2612 n_actions, ntohs(ofm->idle_timeout),
2613 ntohs(ofm->hard_timeout));
2614 cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority));
2618 if (ofm->buffer_id != htonl(UINT32_MAX)) {
2619 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
2623 rule_insert(p, rule, packet, in_port);
2624 ofpbuf_delete(packet);
2629 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
2630 size_t n_actions, uint16_t command, struct rule *rule)
2632 if (rule_is_hidden(rule)) {
2636 if (command == OFPFC_DELETE) {
2637 rule_remove(p, rule);
2639 size_t actions_len = n_actions * sizeof *rule->actions;
2641 if (n_actions == rule->n_actions
2642 && !memcmp(ofm->actions, rule->actions, actions_len))
2647 free(rule->actions);
2648 rule->actions = xmemdup(ofm->actions, actions_len);
2649 rule->n_actions = n_actions;
2651 if (rule->cr.wc.wildcards) {
2652 COVERAGE_INC(ofproto_mod_wc_flow);
2653 p->need_revalidate = true;
2655 rule_update_actions(p, rule);
2663 modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm,
2664 size_t n_actions, uint16_t command)
2670 flow_from_match(&flow, &wildcards, &ofm->match);
2671 rule = rule_from_cls_rule(classifier_find_rule_exactly(
2672 &p->cls, &flow, wildcards,
2673 ntohs(ofm->priority)));
2676 if (command == OFPFC_DELETE
2677 && ofm->out_port != htons(OFPP_NONE)
2678 && !rule_has_out_port(rule, ofm->out_port)) {
2682 modify_flow(p, ofm, n_actions, command, rule);
2687 struct modify_flows_cbdata {
2688 struct ofproto *ofproto;
2689 const struct ofp_flow_mod *ofm;
2696 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
2698 struct rule *rule = rule_from_cls_rule(rule_);
2699 struct modify_flows_cbdata *cbdata = cbdata_;
2701 if (cbdata->out_port != htons(OFPP_NONE)
2702 && !rule_has_out_port(rule, cbdata->out_port)) {
2706 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions,
2707 cbdata->command, rule);
2711 modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm,
2712 size_t n_actions, uint16_t command)
2714 struct modify_flows_cbdata cbdata;
2715 struct cls_rule target;
2719 cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port
2720 : htons(OFPP_NONE));
2721 cbdata.n_actions = n_actions;
2722 cbdata.command = command;
2724 cls_rule_from_match(&target, &ofm->match, 0);
2726 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2727 modify_flows_cb, &cbdata);
2732 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
2733 struct ofp_flow_mod *ofm)
2738 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
2739 sizeof *ofm->actions, &n_actions);
2744 normalize_match(&ofm->match);
2745 if (!ofm->match.wildcards) {
2746 ofm->priority = htons(UINT16_MAX);
2749 error = validate_actions((const union ofp_action *) ofm->actions,
2750 n_actions, p->max_ports);
2755 switch (ntohs(ofm->command)) {
2757 return add_flow(p, ofconn, ofm, n_actions);
2760 return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY);
2762 case OFPFC_MODIFY_STRICT:
2763 return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY);
2766 return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE);
2768 case OFPFC_DELETE_STRICT:
2769 return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE);
2772 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
2777 send_capability_reply(struct ofproto *p, struct ofconn *ofconn, uint32_t xid)
2779 struct ofmp_capability_reply *ocr;
2781 char capabilities[] = "com.nicira.mgmt.manager=false\n";
2783 ocr = make_openflow_xid(sizeof(*ocr), OFPT_VENDOR, xid, &b);
2784 ocr->header.header.vendor = htonl(NX_VENDOR_ID);
2785 ocr->header.header.subtype = htonl(NXT_MGMT);
2786 ocr->header.type = htons(OFMPT_CAPABILITY_REPLY);
2788 ocr->format = htonl(OFMPCOF_SIMPLE);
2789 ocr->mgmt_id = htonll(p->mgmt_id);
2791 ofpbuf_put(b, capabilities, strlen(capabilities));
2793 queue_tx(b, ofconn, ofconn->reply_counter);
2797 handle_ofmp(struct ofproto *p, struct ofconn *ofconn,
2798 struct ofmp_header *ofmph)
2800 size_t msg_len = ntohs(ofmph->header.header.length);
2801 if (msg_len < sizeof(*ofmph)) {
2802 VLOG_WARN_RL(&rl, "dropping short managment message: %d\n", msg_len);
2803 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2806 if (ofmph->type == htons(OFMPT_CAPABILITY_REQUEST)) {
2807 struct ofmp_capability_request *ofmpcr;
2809 if (msg_len < sizeof(struct ofmp_capability_request)) {
2810 VLOG_WARN_RL(&rl, "dropping short capability request: %d\n",
2812 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2815 ofmpcr = (struct ofmp_capability_request *)ofmph;
2816 if (ofmpcr->format != htonl(OFMPCAF_SIMPLE)) {
2817 /* xxx Find a better type than bad subtype */
2818 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2821 send_capability_reply(p, ofconn, ofmph->header.header.xid);
2824 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2829 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
2831 struct ofp_vendor_header *ovh = msg;
2832 struct nicira_header *nh;
2834 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
2835 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2837 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
2838 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2840 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
2841 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH);
2845 switch (ntohl(nh->subtype)) {
2846 case NXT_STATUS_REQUEST:
2847 return switch_status_handle_request(p->switch_status, ofconn->rconn,
2850 case NXT_ACT_SET_CONFIG:
2851 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
2853 case NXT_ACT_GET_CONFIG:
2854 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */
2856 case NXT_COMMAND_REQUEST:
2858 return executer_handle_request(p->executer, ofconn->rconn, msg);
2863 return handle_ofmp(p, ofconn, msg);
2866 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
2870 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
2871 struct ofpbuf *ofp_msg)
2873 struct ofp_header *oh = ofp_msg->data;
2876 COVERAGE_INC(ofproto_recv_openflow);
2878 case OFPT_ECHO_REQUEST:
2879 error = handle_echo_request(ofconn, oh);
2882 case OFPT_ECHO_REPLY:
2886 case OFPT_FEATURES_REQUEST:
2887 error = handle_features_request(p, ofconn, oh);
2890 case OFPT_GET_CONFIG_REQUEST:
2891 error = handle_get_config_request(p, ofconn, oh);
2894 case OFPT_SET_CONFIG:
2895 error = handle_set_config(p, ofconn, ofp_msg->data);
2898 case OFPT_PACKET_OUT:
2899 error = handle_packet_out(p, ofconn, ofp_msg->data);
2903 error = handle_port_mod(p, oh);
2907 error = handle_flow_mod(p, ofconn, ofp_msg->data);
2910 case OFPT_STATS_REQUEST:
2911 error = handle_stats_request(p, ofconn, oh);
2915 error = handle_vendor(p, ofconn, ofp_msg->data);
2919 if (VLOG_IS_WARN_ENABLED()) {
2920 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
2921 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
2924 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
2929 send_error_oh(ofconn, ofp_msg->data, error);
2934 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
2936 struct odp_msg *msg = packet->data;
2937 uint16_t in_port = odp_port_to_ofp_port(msg->port);
2939 struct ofpbuf payload;
2942 /* Handle controller actions. */
2943 if (msg->type == _ODPL_ACTION_NR) {
2944 COVERAGE_INC(ofproto_ctlr_action);
2945 pinsched_send(p->action_sched, in_port, packet,
2946 send_packet_in_action, p);
2950 payload.data = msg + 1;
2951 payload.size = msg->length - sizeof *msg;
2952 flow_extract(&payload, msg->port, &flow);
2954 rule = lookup_valid_rule(p, &flow);
2956 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2957 struct ofport *port = port_array_get(&p->ports, msg->port);
2959 if (port->opp.config & OFPPC_NO_PACKET_IN) {
2960 COVERAGE_INC(ofproto_no_packet_in);
2961 /* XXX install 'drop' flow entry */
2962 ofpbuf_delete(packet);
2966 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
2969 COVERAGE_INC(ofproto_packet_in);
2970 pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p);
2974 if (rule->cr.wc.wildcards) {
2975 rule = rule_create_subrule(p, rule, &flow);
2976 rule_make_actions(p, rule, packet);
2978 if (!rule->may_install) {
2979 /* The rule is not installable, that is, we need to process every
2980 * packet, so process the current packet and set its actions into
2982 rule_make_actions(p, rule, packet);
2984 /* XXX revalidate rule if it needs it */
2988 rule_execute(p, rule, &payload, &flow);
2989 rule_reinstall(p, rule);
2990 ofpbuf_delete(packet);
2994 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
2996 struct rule *sub = rule_from_cls_rule(sub_);
2997 struct revalidate_cbdata *cbdata = cbdata_;
2999 if (cbdata->revalidate_all
3000 || (cbdata->revalidate_subrules && sub->super)
3001 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3002 revalidate_rule(cbdata->ofproto, sub);
3007 revalidate_rule(struct ofproto *p, struct rule *rule)
3009 const flow_t *flow = &rule->cr.flow;
3011 COVERAGE_INC(ofproto_revalidate_rule);
3014 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3016 rule_remove(p, rule);
3018 } else if (super != rule->super) {
3019 COVERAGE_INC(ofproto_revalidate_moved);
3020 list_remove(&rule->list);
3021 list_push_back(&super->list, &rule->list);
3022 rule->super = super;
3023 rule->hard_timeout = super->hard_timeout;
3024 rule->idle_timeout = super->idle_timeout;
3025 rule->created = super->created;
3030 rule_update_actions(p, rule);
3034 static struct ofpbuf *
3035 compose_flow_exp(const struct rule *rule, long long int now, uint8_t reason)
3037 struct ofp_flow_expired *ofe;
3040 ofe = make_openflow(sizeof *ofe, OFPT_FLOW_EXPIRED, &buf);
3041 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofe->match);
3042 ofe->priority = htons(rule->cr.priority);
3043 ofe->reason = reason;
3044 ofe->duration = (now - rule->created) / 1000;
3045 ofe->packet_count = rule->packet_count;
3046 ofe->byte_count = rule->byte_count;
3052 send_flow_exp(struct ofproto *p, struct rule *rule,
3053 long long int now, uint8_t reason)
3055 struct ofconn *ofconn;
3056 struct ofconn *prev;
3059 /* We limit the maximum number of queued flow expirations it by accounting
3060 * them under the counter for replies. That works because preventing
3061 * OpenFlow requests from being processed also prevents new flows from
3062 * being added (and expiring). (It also prevents processing OpenFlow
3063 * requests that would not add new flows, so it is imperfect.) */
3066 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3067 if (ofconn->send_flow_exp && rconn_is_connected(ofconn->rconn)) {
3069 queue_tx(ofpbuf_clone(buf), prev, ofconn->reply_counter);
3071 buf = compose_flow_exp(rule, now, reason);
3077 queue_tx(buf, prev, ofconn->reply_counter);
3082 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3084 assert(rule->installed);
3085 assert(!rule->cr.wc.wildcards);
3088 rule_remove(ofproto, rule);
3090 rule_uninstall(ofproto, rule);
3095 expire_rule(struct cls_rule *cls_rule, void *p_)
3097 struct ofproto *p = p_;
3098 struct rule *rule = rule_from_cls_rule(cls_rule);
3099 long long int hard_expire, idle_expire, expire, now;
3101 hard_expire = (rule->hard_timeout
3102 ? rule->created + rule->hard_timeout * 1000
3104 idle_expire = (rule->idle_timeout
3105 && (rule->super || list_is_empty(&rule->list))
3106 ? rule->used + rule->idle_timeout * 1000
3108 expire = MIN(hard_expire, idle_expire);
3109 if (expire == LLONG_MAX) {
3110 if (rule->installed && time_msec() >= rule->used + 5000) {
3111 uninstall_idle_flow(p, rule);
3118 if (rule->installed && now >= rule->used + 5000) {
3119 uninstall_idle_flow(p, rule);
3124 COVERAGE_INC(ofproto_expired);
3125 if (rule->cr.wc.wildcards) {
3126 /* Update stats. (This code will be a no-op if the rule expired
3127 * due to an idle timeout, because in that case the rule has no
3128 * subrules left.) */
3129 struct rule *subrule, *next;
3130 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3131 rule_remove(p, subrule);
3135 send_flow_exp(p, rule, now,
3137 ? OFPER_HARD_TIMEOUT : OFPER_IDLE_TIMEOUT));
3138 rule_remove(p, rule);
3142 update_used(struct ofproto *p)
3144 struct odp_flow *flows;
3149 error = dpif_flow_list_all(&p->dpif, &flows, &n_flows);
3154 for (i = 0; i < n_flows; i++) {
3155 struct odp_flow *f = &flows[i];
3158 rule = rule_from_cls_rule(
3159 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3160 if (!rule || !rule->installed) {
3161 COVERAGE_INC(ofproto_unexpected_rule);
3162 dpif_flow_del(&p->dpif, f);
3166 update_time(rule, &f->stats);
3167 rule_account(p, rule, f->stats.n_bytes);
3173 do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id,
3174 const struct ofpbuf *packet, int send_len)
3176 struct ofp_packet_in *opi;
3177 struct ofpbuf payload, *buf;
3178 struct odp_msg *msg;
3181 payload.data = msg + 1;
3182 payload.size = msg->length - sizeof *msg;
3184 send_len = MIN(send_len, payload.size);
3185 buf = ofpbuf_new(sizeof *opi + send_len);
3186 opi = put_openflow_xid(offsetof(struct ofp_packet_in, data),
3187 OFPT_PACKET_IN, 0, buf);
3188 opi->buffer_id = htonl(buffer_id);
3189 opi->total_len = htons(payload.size);
3190 opi->in_port = htons(odp_port_to_ofp_port(msg->port));
3191 opi->reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH;
3192 ofpbuf_put(buf, payload.data, MIN(send_len, payload.size));
3193 update_openflow_length(buf);
3194 rconn_send_with_limit(ofconn->rconn, buf, ofconn->packet_in_counter, 100);
3198 send_packet_in_action(struct ofpbuf *packet, void *p_)
3200 struct ofproto *p = p_;
3201 struct ofconn *ofconn;
3202 struct odp_msg *msg;
3205 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3206 if (ofconn == p->controller || ofconn->miss_send_len) {
3207 do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg);
3210 ofpbuf_delete(packet);
3214 send_packet_in_miss(struct ofpbuf *packet, void *p_)
3216 struct ofproto *p = p_;
3217 struct ofconn *ofconn;
3218 struct ofpbuf payload;
3219 struct odp_msg *msg;
3222 payload.data = msg + 1;
3223 payload.size = msg->length - sizeof *msg;
3224 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3225 if (ofconn->miss_send_len) {
3226 uint32_t buffer_id = pktbuf_save(ofconn->pktbuf, &payload,
3228 int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len
3230 do_send_packet_in(ofconn, buffer_id, packet, send_len);
3233 ofpbuf_delete(packet);
3237 pick_datapath_id(struct dpif *dpif, uint64_t fallback_dpid)
3239 char local_name[IF_NAMESIZE];
3240 uint8_t ea[ETH_ADDR_LEN];
3243 error = dpif_get_name(dpif, local_name, sizeof local_name);
3245 error = netdev_nodev_get_etheraddr(local_name, ea);
3247 return eth_addr_to_uint64(ea);
3249 VLOG_WARN("could not get MAC address for %s (%s)",
3250 local_name, strerror(error));
3253 return fallback_dpid;
3257 pick_fallback_dpid(void)
3259 uint8_t ea[ETH_ADDR_LEN];
3260 eth_addr_random(ea);
3261 ea[0] = 0x00; /* Set Nicira OUI. */
3264 return eth_addr_to_uint64(ea);
3268 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3269 struct odp_actions *actions, tag_type *tags,
3272 struct ofproto *ofproto = ofproto_;
3275 /* Drop frames for reserved multicast addresses. */
3276 if (eth_addr_is_reserved(flow->dl_dst)) {
3280 /* Learn source MAC (but don't try to learn from revalidation). */
3281 if (packet != NULL) {
3282 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3285 /* The log messages here could actually be useful in debugging,
3286 * so keep the rate limit relatively high. */
3287 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3288 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3289 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3290 ofproto_revalidate(ofproto, rev_tag);
3294 /* Determine output port. */
3295 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3297 add_output_group_action(actions, DP_GROUP_FLOOD);
3298 } else if (out_port != flow->in_port) {
3299 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3307 static const struct ofhooks default_ofhooks = {
3309 default_normal_ofhook_cb,