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"
29 #include "dynamic-string.h"
31 #include "fail-open.h"
33 #include "mac-learning.h"
37 #include "ofp-print.h"
39 #include "openflow/nicira-ext.h"
40 #include "openflow/openflow.h"
41 #include "openflow/openflow-mgmt.h"
42 #include "openvswitch/datapath-protocol.h"
46 #include "poll-loop.h"
47 #include "port-array.h"
57 #include "vconn-ssl.h"
60 #define THIS_MODULE VLM_ofproto
70 TABLEID_CLASSIFIER
= 1
74 struct netdev
*netdev
;
75 struct ofp_phy_port opp
; /* In host byte order. */
78 static void ofport_free(struct ofport
*);
79 static void hton_ofp_phy_port(struct ofp_phy_port
*);
81 static int xlate_actions(const union ofp_action
*in
, size_t n_in
,
82 const flow_t
*flow
, struct ofproto
*ofproto
,
83 const struct ofpbuf
*packet
,
84 struct odp_actions
*out
, tag_type
*tags
,
85 bool *may_setup_flow
);
90 uint16_t idle_timeout
; /* In seconds from time of last use. */
91 uint16_t hard_timeout
; /* In seconds from time of creation. */
92 long long int used
; /* Last-used time (0 if never used). */
93 long long int created
; /* Creation time. */
94 uint64_t packet_count
; /* Number of packets received. */
95 uint64_t byte_count
; /* Number of bytes received. */
96 uint64_t accounted_bytes
; /* Number of bytes passed to account_cb. */
97 uint8_t tcp_flags
; /* Bitwise-OR of all TCP flags seen. */
98 uint8_t ip_tos
; /* Last-seen IP type-of-service. */
99 tag_type tags
; /* Tags (set only by hooks). */
101 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
102 * exact-match rule (having cr.wc.wildcards of 0) generated from the
103 * wildcard rule 'super'. In this case, 'list' is an element of the
106 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
107 * a list of subrules. A super-rule with no wildcards (where
108 * cr.wc.wildcards is 0) will never have any subrules. */
114 * A subrule has no actions (it uses the super-rule's actions). */
116 union ofp_action
*actions
;
120 * A super-rule with wildcard fields never has ODP actions (since the
121 * datapath only supports exact-match flows). */
122 bool installed
; /* Installed in datapath? */
123 bool may_install
; /* True ordinarily; false if actions must
124 * be reassessed for every packet. */
126 union odp_action
*odp_actions
;
130 rule_is_hidden(const struct rule
*rule
)
132 /* Subrules are merely an implementation detail, so hide them from the
134 if (rule
->super
!= NULL
) {
138 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
139 * (e.g. by in-band control) and are intentionally hidden from the
141 if (rule
->cr
.priority
> UINT16_MAX
) {
148 static struct rule
*rule_create(struct rule
*super
, const union ofp_action
*,
149 size_t n_actions
, uint16_t idle_timeout
,
150 uint16_t hard_timeout
);
151 static void rule_free(struct rule
*);
152 static void rule_destroy(struct ofproto
*, struct rule
*);
153 static struct rule
*rule_from_cls_rule(const struct cls_rule
*);
154 static void rule_insert(struct ofproto
*, struct rule
*,
155 struct ofpbuf
*packet
, uint16_t in_port
);
156 static void rule_remove(struct ofproto
*, struct rule
*);
157 static bool rule_make_actions(struct ofproto
*, struct rule
*,
158 const struct ofpbuf
*packet
);
159 static void rule_install(struct ofproto
*, struct rule
*,
160 struct rule
*displaced_rule
);
161 static void rule_uninstall(struct ofproto
*, struct rule
*);
162 static void rule_post_uninstall(struct ofproto
*, struct rule
*);
167 struct pktbuf
*pktbuf
;
171 struct rconn_packet_counter
*packet_in_counter
;
173 /* Number of OpenFlow messages queued as replies to OpenFlow requests, and
174 * the maximum number before we stop reading OpenFlow requests. */
175 #define OFCONN_REPLY_MAX 100
176 struct rconn_packet_counter
*reply_counter
;
179 static struct ofconn
*ofconn_create(struct ofproto
*, struct rconn
*);
180 static void ofconn_destroy(struct ofconn
*, struct ofproto
*);
181 static void ofconn_run(struct ofconn
*, struct ofproto
*);
182 static void ofconn_wait(struct ofconn
*);
183 static void queue_tx(struct ofpbuf
*msg
, const struct ofconn
*ofconn
,
184 struct rconn_packet_counter
*counter
);
188 uint64_t datapath_id
; /* Datapath ID. */
189 uint64_t fallback_dpid
; /* Datapath ID if no better choice found. */
190 uint64_t mgmt_id
; /* Management channel identifier. */
191 char *manufacturer
; /* Manufacturer. */
192 char *hardware
; /* Hardware. */
193 char *software
; /* Software version. */
194 char *serial
; /* Serial number. */
198 struct netdev_monitor
*netdev_monitor
;
199 struct port_array ports
; /* Index is ODP port nr; ofport->opp.port_no is
201 struct shash port_by_name
;
205 struct switch_status
*switch_status
;
206 struct status_category
*ss_cat
;
207 struct in_band
*in_band
;
208 struct discovery
*discovery
;
209 struct fail_open
*fail_open
;
210 struct pinsched
*miss_sched
, *action_sched
;
211 struct executer
*executer
;
212 struct netflow
*netflow
;
215 struct classifier cls
;
216 bool need_revalidate
;
217 long long int next_expiration
;
218 struct tag_set revalidate_set
;
220 /* OpenFlow connections. */
221 struct list all_conns
;
222 struct ofconn
*controller
;
223 struct pvconn
**listeners
;
225 struct pvconn
**snoops
;
228 /* Hooks for ovs-vswitchd. */
229 const struct ofhooks
*ofhooks
;
232 /* Used by default ofhooks. */
233 struct mac_learning
*ml
;
236 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
238 static const struct ofhooks default_ofhooks
;
240 static uint64_t pick_datapath_id(const struct ofproto
*);
241 static uint64_t pick_fallback_dpid(void);
242 static void send_packet_in_miss(struct ofpbuf
*, void *ofproto
);
243 static void send_packet_in_action(struct ofpbuf
*, void *ofproto
);
244 static void update_used(struct ofproto
*);
245 static void update_stats(struct rule
*, const struct odp_flow_stats
*);
246 static void expire_rule(struct cls_rule
*, void *ofproto
);
247 static bool revalidate_rule(struct ofproto
*p
, struct rule
*rule
);
248 static void revalidate_cb(struct cls_rule
*rule_
, void *p_
);
250 static void handle_odp_msg(struct ofproto
*, struct ofpbuf
*);
252 static void handle_openflow(struct ofconn
*, struct ofproto
*,
255 static void refresh_port_group(struct ofproto
*, unsigned int group
);
256 static void update_port(struct ofproto
*, const char *devname
);
257 static int init_ports(struct ofproto
*);
258 static void reinit_ports(struct ofproto
*);
261 ofproto_create(const char *datapath
, const struct ofhooks
*ofhooks
, void *aux
,
262 struct ofproto
**ofprotop
)
264 struct odp_stats stats
;
271 /* Connect to datapath and start listening for messages. */
272 error
= dpif_open(datapath
, &dpif
);
274 VLOG_ERR("failed to open datapath %s: %s", datapath
, strerror(error
));
277 error
= dpif_get_dp_stats(dpif
, &stats
);
279 VLOG_ERR("failed to obtain stats for datapath %s: %s",
280 datapath
, strerror(error
));
284 error
= dpif_recv_set_mask(dpif
, ODPL_MISS
| ODPL_ACTION
);
286 VLOG_ERR("failed to listen on datapath %s: %s",
287 datapath
, strerror(error
));
291 dpif_flow_flush(dpif
);
292 dpif_recv_purge(dpif
);
294 /* Initialize settings. */
295 p
= xcalloc(1, sizeof *p
);
296 p
->fallback_dpid
= pick_fallback_dpid();
297 p
->datapath_id
= p
->fallback_dpid
;
298 p
->manufacturer
= xstrdup("Nicira Networks, Inc.");
299 p
->hardware
= xstrdup("Reference Implementation");
300 p
->software
= xstrdup(VERSION BUILDNR
);
301 p
->serial
= xstrdup("None");
303 /* Initialize datapath. */
305 p
->netdev_monitor
= netdev_monitor_create();
306 port_array_init(&p
->ports
);
307 shash_init(&p
->port_by_name
);
308 p
->max_ports
= stats
.max_ports
;
310 /* Initialize submodules. */
311 p
->switch_status
= switch_status_create(p
);
315 p
->miss_sched
= p
->action_sched
= NULL
;
319 /* Initialize flow table. */
320 classifier_init(&p
->cls
);
321 p
->need_revalidate
= false;
322 p
->next_expiration
= time_msec() + 1000;
323 tag_set_init(&p
->revalidate_set
);
325 /* Initialize OpenFlow connections. */
326 list_init(&p
->all_conns
);
327 p
->controller
= ofconn_create(p
, rconn_create(5, 8));
328 p
->controller
->pktbuf
= pktbuf_create();
329 p
->controller
->miss_send_len
= OFP_DEFAULT_MISS_SEND_LEN
;
335 /* Initialize hooks. */
337 p
->ofhooks
= ofhooks
;
341 p
->ofhooks
= &default_ofhooks
;
343 p
->ml
= mac_learning_create();
346 /* Register switch status category. */
347 p
->ss_cat
= switch_status_register(p
->switch_status
, "remote",
348 rconn_status_cb
, p
->controller
->rconn
);
351 error
= init_ports(p
);
357 /* Pick final datapath ID. */
358 p
->datapath_id
= pick_datapath_id(p
);
359 VLOG_INFO("using datapath ID %012"PRIx64
, p
->datapath_id
);
366 ofproto_set_datapath_id(struct ofproto
*p
, uint64_t datapath_id
)
368 uint64_t old_dpid
= p
->datapath_id
;
369 p
->datapath_id
= datapath_id
? datapath_id
: pick_datapath_id(p
);
370 if (p
->datapath_id
!= old_dpid
) {
371 VLOG_INFO("datapath ID changed to %012"PRIx64
, p
->datapath_id
);
372 rconn_reconnect(p
->controller
->rconn
);
377 ofproto_set_mgmt_id(struct ofproto
*p
, uint64_t mgmt_id
)
379 p
->mgmt_id
= mgmt_id
;
383 ofproto_set_probe_interval(struct ofproto
*p
, int probe_interval
)
385 probe_interval
= probe_interval
? MAX(probe_interval
, 5) : 0;
386 rconn_set_probe_interval(p
->controller
->rconn
, probe_interval
);
388 int trigger_duration
= probe_interval
? probe_interval
* 3 : 15;
389 fail_open_set_trigger_duration(p
->fail_open
, trigger_duration
);
394 ofproto_set_max_backoff(struct ofproto
*p
, int max_backoff
)
396 rconn_set_max_backoff(p
->controller
->rconn
, max_backoff
);
400 ofproto_set_desc(struct ofproto
*p
,
401 const char *manufacturer
, const char *hardware
,
402 const char *software
, const char *serial
)
405 free(p
->manufacturer
);
406 p
->manufacturer
= xstrdup(manufacturer
);
410 p
->hardware
= xstrdup(hardware
);
414 p
->software
= xstrdup(software
);
418 p
->serial
= xstrdup(serial
);
423 ofproto_set_in_band(struct ofproto
*p
, bool in_band
)
425 if (in_band
!= (p
->in_band
!= NULL
)) {
427 return in_band_create(p
, p
->dpif
, p
->switch_status
,
428 p
->controller
->rconn
, &p
->in_band
);
430 ofproto_set_discovery(p
, false, NULL
, true);
431 in_band_destroy(p
->in_band
);
434 rconn_reconnect(p
->controller
->rconn
);
440 ofproto_set_discovery(struct ofproto
*p
, bool discovery
,
441 const char *re
, bool update_resolv_conf
)
443 if (discovery
!= (p
->discovery
!= NULL
)) {
445 int error
= ofproto_set_in_band(p
, true);
449 error
= discovery_create(re
, update_resolv_conf
,
450 p
->dpif
, p
->switch_status
,
456 discovery_destroy(p
->discovery
);
459 rconn_disconnect(p
->controller
->rconn
);
460 } else if (discovery
) {
461 discovery_set_update_resolv_conf(p
->discovery
, update_resolv_conf
);
462 return discovery_set_accept_controller_re(p
->discovery
, re
);
468 ofproto_set_controller(struct ofproto
*ofproto
, const char *controller
)
470 if (ofproto
->discovery
) {
472 } else if (controller
) {
473 if (strcmp(rconn_get_name(ofproto
->controller
->rconn
), controller
)) {
474 return rconn_connect(ofproto
->controller
->rconn
, controller
);
479 rconn_disconnect(ofproto
->controller
->rconn
);
485 set_pvconns(struct pvconn
***pvconnsp
, size_t *n_pvconnsp
,
486 const struct svec
*svec
)
488 struct pvconn
**pvconns
= *pvconnsp
;
489 size_t n_pvconns
= *n_pvconnsp
;
493 for (i
= 0; i
< n_pvconns
; i
++) {
494 pvconn_close(pvconns
[i
]);
498 pvconns
= xmalloc(svec
->n
* sizeof *pvconns
);
500 for (i
= 0; i
< svec
->n
; i
++) {
501 const char *name
= svec
->names
[i
];
502 struct pvconn
*pvconn
;
505 error
= pvconn_open(name
, &pvconn
);
507 pvconns
[n_pvconns
++] = pvconn
;
509 VLOG_ERR("failed to listen on %s: %s", name
, strerror(error
));
517 *n_pvconnsp
= n_pvconns
;
523 ofproto_set_listeners(struct ofproto
*ofproto
, const struct svec
*listeners
)
525 return set_pvconns(&ofproto
->listeners
, &ofproto
->n_listeners
, listeners
);
529 ofproto_set_snoops(struct ofproto
*ofproto
, const struct svec
*snoops
)
531 return set_pvconns(&ofproto
->snoops
, &ofproto
->n_snoops
, snoops
);
535 ofproto_set_netflow(struct ofproto
*ofproto
, const struct svec
*collectors
,
536 uint8_t engine_type
, uint8_t engine_id
, bool add_id_to_iface
)
538 if (collectors
&& collectors
->n
) {
539 if (!ofproto
->netflow
) {
540 ofproto
->netflow
= netflow_create();
542 netflow_set_engine(ofproto
->netflow
, engine_type
, engine_id
,
544 return netflow_set_collectors(ofproto
->netflow
, collectors
);
546 netflow_destroy(ofproto
->netflow
);
547 ofproto
->netflow
= NULL
;
553 ofproto_set_failure(struct ofproto
*ofproto
, bool fail_open
)
556 struct rconn
*rconn
= ofproto
->controller
->rconn
;
557 int trigger_duration
= rconn_get_probe_interval(rconn
) * 3;
558 if (!ofproto
->fail_open
) {
559 ofproto
->fail_open
= fail_open_create(ofproto
, trigger_duration
,
560 ofproto
->switch_status
,
563 fail_open_set_trigger_duration(ofproto
->fail_open
,
567 fail_open_destroy(ofproto
->fail_open
);
568 ofproto
->fail_open
= NULL
;
573 ofproto_set_rate_limit(struct ofproto
*ofproto
,
574 int rate_limit
, int burst_limit
)
576 if (rate_limit
> 0) {
577 if (!ofproto
->miss_sched
) {
578 ofproto
->miss_sched
= pinsched_create(rate_limit
, burst_limit
,
579 ofproto
->switch_status
);
580 ofproto
->action_sched
= pinsched_create(rate_limit
, burst_limit
,
583 pinsched_set_limits(ofproto
->miss_sched
, rate_limit
, burst_limit
);
584 pinsched_set_limits(ofproto
->action_sched
,
585 rate_limit
, burst_limit
);
588 pinsched_destroy(ofproto
->miss_sched
);
589 ofproto
->miss_sched
= NULL
;
590 pinsched_destroy(ofproto
->action_sched
);
591 ofproto
->action_sched
= NULL
;
596 ofproto_set_stp(struct ofproto
*ofproto UNUSED
, bool enable_stp
)
600 VLOG_WARN("STP is not yet implemented");
608 ofproto_set_remote_execution(struct ofproto
*ofproto
, const char *command_acl
,
609 const char *command_dir
)
612 if (!ofproto
->executer
) {
613 return executer_create(command_acl
, command_dir
,
616 executer_set_acl(ofproto
->executer
, command_acl
, command_dir
);
619 executer_destroy(ofproto
->executer
);
620 ofproto
->executer
= NULL
;
626 ofproto_get_datapath_id(const struct ofproto
*ofproto
)
628 return ofproto
->datapath_id
;
632 ofproto_get_mgmt_id(const struct ofproto
*ofproto
)
634 return ofproto
->mgmt_id
;
638 ofproto_get_probe_interval(const struct ofproto
*ofproto
)
640 return rconn_get_probe_interval(ofproto
->controller
->rconn
);
644 ofproto_get_max_backoff(const struct ofproto
*ofproto
)
646 return rconn_get_max_backoff(ofproto
->controller
->rconn
);
650 ofproto_get_in_band(const struct ofproto
*ofproto
)
652 return ofproto
->in_band
!= NULL
;
656 ofproto_get_discovery(const struct ofproto
*ofproto
)
658 return ofproto
->discovery
!= NULL
;
662 ofproto_get_controller(const struct ofproto
*ofproto
)
664 return rconn_get_name(ofproto
->controller
->rconn
);
668 ofproto_get_listeners(const struct ofproto
*ofproto
, struct svec
*listeners
)
672 for (i
= 0; i
< ofproto
->n_listeners
; i
++) {
673 svec_add(listeners
, pvconn_get_name(ofproto
->listeners
[i
]));
678 ofproto_get_snoops(const struct ofproto
*ofproto
, struct svec
*snoops
)
682 for (i
= 0; i
< ofproto
->n_snoops
; i
++) {
683 svec_add(snoops
, pvconn_get_name(ofproto
->snoops
[i
]));
688 ofproto_destroy(struct ofproto
*p
)
690 struct ofconn
*ofconn
, *next_ofconn
;
691 struct ofport
*ofport
;
692 unsigned int port_no
;
699 ofproto_flush_flows(p
);
700 classifier_destroy(&p
->cls
);
702 LIST_FOR_EACH_SAFE (ofconn
, next_ofconn
, struct ofconn
, node
,
704 ofconn_destroy(ofconn
, p
);
708 netdev_monitor_destroy(p
->netdev_monitor
);
709 PORT_ARRAY_FOR_EACH (ofport
, &p
->ports
, port_no
) {
712 shash_destroy(&p
->port_by_name
);
714 switch_status_destroy(p
->switch_status
);
715 in_band_destroy(p
->in_band
);
716 discovery_destroy(p
->discovery
);
717 fail_open_destroy(p
->fail_open
);
718 pinsched_destroy(p
->miss_sched
);
719 pinsched_destroy(p
->action_sched
);
720 executer_destroy(p
->executer
);
721 netflow_destroy(p
->netflow
);
723 switch_status_unregister(p
->ss_cat
);
725 for (i
= 0; i
< p
->n_listeners
; i
++) {
726 pvconn_close(p
->listeners
[i
]);
730 for (i
= 0; i
< p
->n_snoops
; i
++) {
731 pvconn_close(p
->snoops
[i
]);
735 mac_learning_destroy(p
->ml
);
741 ofproto_run(struct ofproto
*p
)
743 int error
= ofproto_run1(p
);
745 error
= ofproto_run2(p
, false);
751 process_port_change(struct ofproto
*ofproto
, int error
, char *devname
)
753 if (error
== ENOBUFS
) {
754 reinit_ports(ofproto
);
756 update_port(ofproto
, devname
);
762 ofproto_run1(struct ofproto
*p
)
764 struct ofconn
*ofconn
, *next_ofconn
;
769 for (i
= 0; i
< 50; i
++) {
773 error
= dpif_recv(p
->dpif
, &buf
);
775 if (error
== ENODEV
) {
776 /* Someone destroyed the datapath behind our back. The caller
777 * better destroy us and give up, because we're just going to
778 * spin from here on out. */
779 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
780 VLOG_ERR_RL(&rl
, "%s: datapath was destroyed externally",
787 handle_odp_msg(p
, buf
);
790 while ((error
= dpif_port_poll(p
->dpif
, &devname
)) != EAGAIN
) {
791 process_port_change(p
, error
, devname
);
793 while ((error
= netdev_monitor_poll(p
->netdev_monitor
,
794 &devname
)) != EAGAIN
) {
795 process_port_change(p
, error
, devname
);
799 in_band_run(p
->in_band
);
802 char *controller_name
;
803 if (rconn_is_connectivity_questionable(p
->controller
->rconn
)) {
804 discovery_question_connectivity(p
->discovery
);
806 if (discovery_run(p
->discovery
, &controller_name
)) {
807 if (controller_name
) {
808 rconn_connect(p
->controller
->rconn
, controller_name
);
810 rconn_disconnect(p
->controller
->rconn
);
815 fail_open_run(p
->fail_open
);
817 pinsched_run(p
->miss_sched
, send_packet_in_miss
, p
);
818 pinsched_run(p
->action_sched
, send_packet_in_action
, p
);
820 executer_run(p
->executer
);
823 LIST_FOR_EACH_SAFE (ofconn
, next_ofconn
, struct ofconn
, node
,
825 ofconn_run(ofconn
, p
);
828 for (i
= 0; i
< p
->n_listeners
; i
++) {
832 retval
= pvconn_accept(p
->listeners
[i
], OFP_VERSION
, &vconn
);
834 ofconn_create(p
, rconn_new_from_vconn("passive", vconn
));
835 } else if (retval
!= EAGAIN
) {
836 VLOG_WARN_RL(&rl
, "accept failed (%s)", strerror(retval
));
840 for (i
= 0; i
< p
->n_snoops
; i
++) {
844 retval
= pvconn_accept(p
->snoops
[i
], OFP_VERSION
, &vconn
);
846 rconn_add_monitor(p
->controller
->rconn
, vconn
);
847 } else if (retval
!= EAGAIN
) {
848 VLOG_WARN_RL(&rl
, "accept failed (%s)", strerror(retval
));
852 if (time_msec() >= p
->next_expiration
) {
853 COVERAGE_INC(ofproto_expiration
);
854 p
->next_expiration
= time_msec() + 1000;
857 classifier_for_each(&p
->cls
, CLS_INC_ALL
, expire_rule
, p
);
859 /* Let the hook know that we're at a stable point: all outstanding data
860 * in existing flows has been accounted to the account_cb. Thus, the
861 * hook can now reasonably do operations that depend on having accurate
862 * flow volume accounting (currently, that's just bond rebalancing). */
863 if (p
->ofhooks
->account_checkpoint_cb
) {
864 p
->ofhooks
->account_checkpoint_cb(p
->aux
);
869 netflow_run(p
->netflow
);
875 struct revalidate_cbdata
{
876 struct ofproto
*ofproto
;
877 bool revalidate_all
; /* Revalidate all exact-match rules? */
878 bool revalidate_subrules
; /* Revalidate all exact-match subrules? */
879 struct tag_set revalidate_set
; /* Set of tags to revalidate. */
883 ofproto_run2(struct ofproto
*p
, bool revalidate_all
)
885 if (p
->need_revalidate
|| revalidate_all
886 || !tag_set_is_empty(&p
->revalidate_set
)) {
887 struct revalidate_cbdata cbdata
;
889 cbdata
.revalidate_all
= revalidate_all
;
890 cbdata
.revalidate_subrules
= p
->need_revalidate
;
891 cbdata
.revalidate_set
= p
->revalidate_set
;
892 tag_set_init(&p
->revalidate_set
);
893 COVERAGE_INC(ofproto_revalidate
);
894 classifier_for_each(&p
->cls
, CLS_INC_EXACT
, revalidate_cb
, &cbdata
);
895 p
->need_revalidate
= false;
902 ofproto_wait(struct ofproto
*p
)
904 struct ofconn
*ofconn
;
907 dpif_recv_wait(p
->dpif
);
908 dpif_port_poll_wait(p
->dpif
);
909 netdev_monitor_poll_wait(p
->netdev_monitor
);
910 LIST_FOR_EACH (ofconn
, struct ofconn
, node
, &p
->all_conns
) {
914 in_band_wait(p
->in_band
);
917 discovery_wait(p
->discovery
);
920 fail_open_wait(p
->fail_open
);
922 pinsched_wait(p
->miss_sched
);
923 pinsched_wait(p
->action_sched
);
925 executer_wait(p
->executer
);
927 if (!tag_set_is_empty(&p
->revalidate_set
)) {
928 poll_immediate_wake();
930 if (p
->need_revalidate
) {
931 /* Shouldn't happen, but if it does just go around again. */
932 VLOG_DBG_RL(&rl
, "need revalidate in ofproto_wait_cb()");
933 poll_immediate_wake();
934 } else if (p
->next_expiration
!= LLONG_MAX
) {
935 poll_timer_wait(p
->next_expiration
- time_msec());
937 for (i
= 0; i
< p
->n_listeners
; i
++) {
938 pvconn_wait(p
->listeners
[i
]);
940 for (i
= 0; i
< p
->n_snoops
; i
++) {
941 pvconn_wait(p
->snoops
[i
]);
946 ofproto_revalidate(struct ofproto
*ofproto
, tag_type tag
)
948 tag_set_add(&ofproto
->revalidate_set
, tag
);
952 ofproto_get_revalidate_set(struct ofproto
*ofproto
)
954 return &ofproto
->revalidate_set
;
958 ofproto_is_alive(const struct ofproto
*p
)
960 return p
->discovery
|| rconn_is_alive(p
->controller
->rconn
);
964 ofproto_send_packet(struct ofproto
*p
, const flow_t
*flow
,
965 const union ofp_action
*actions
, size_t n_actions
,
966 const struct ofpbuf
*packet
)
968 struct odp_actions odp_actions
;
971 error
= xlate_actions(actions
, n_actions
, flow
, p
, packet
, &odp_actions
,
977 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
979 dpif_execute(p
->dpif
, flow
->in_port
, odp_actions
.actions
,
980 odp_actions
.n_actions
, packet
);
985 ofproto_add_flow(struct ofproto
*p
,
986 const flow_t
*flow
, uint32_t wildcards
, unsigned int priority
,
987 const union ofp_action
*actions
, size_t n_actions
,
991 rule
= rule_create(NULL
, actions
, n_actions
,
992 idle_timeout
>= 0 ? idle_timeout
: 5 /* XXX */, 0);
993 cls_rule_from_flow(&rule
->cr
, flow
, wildcards
, priority
);
994 rule_insert(p
, rule
, NULL
, 0);
998 ofproto_delete_flow(struct ofproto
*ofproto
, const flow_t
*flow
,
999 uint32_t wildcards
, unsigned int priority
)
1003 rule
= rule_from_cls_rule(classifier_find_rule_exactly(&ofproto
->cls
,
1007 rule_remove(ofproto
, rule
);
1012 destroy_rule(struct cls_rule
*rule_
, void *ofproto_
)
1014 struct rule
*rule
= rule_from_cls_rule(rule_
);
1015 struct ofproto
*ofproto
= ofproto_
;
1017 /* Mark the flow as not installed, even though it might really be
1018 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1019 * There is no point in uninstalling it individually since we are about to
1020 * blow away all the flows with dpif_flow_flush(). */
1021 rule
->installed
= false;
1023 rule_remove(ofproto
, rule
);
1027 ofproto_flush_flows(struct ofproto
*ofproto
)
1029 COVERAGE_INC(ofproto_flush
);
1030 classifier_for_each(&ofproto
->cls
, CLS_INC_ALL
, destroy_rule
, ofproto
);
1031 dpif_flow_flush(ofproto
->dpif
);
1032 if (ofproto
->in_band
) {
1033 in_band_flushed(ofproto
->in_band
);
1035 if (ofproto
->fail_open
) {
1036 fail_open_flushed(ofproto
->fail_open
);
1041 reinit_ports(struct ofproto
*p
)
1043 struct svec devnames
;
1044 struct ofport
*ofport
;
1045 unsigned int port_no
;
1046 struct odp_port
*odp_ports
;
1050 svec_init(&devnames
);
1051 PORT_ARRAY_FOR_EACH (ofport
, &p
->ports
, port_no
) {
1052 svec_add (&devnames
, (char *) ofport
->opp
.name
);
1054 dpif_port_list(p
->dpif
, &odp_ports
, &n_odp_ports
);
1055 for (i
= 0; i
< n_odp_ports
; i
++) {
1056 svec_add (&devnames
, odp_ports
[i
].devname
);
1060 svec_sort_unique(&devnames
);
1061 for (i
= 0; i
< devnames
.n
; i
++) {
1062 update_port(p
, devnames
.names
[i
]);
1064 svec_destroy(&devnames
);
1068 refresh_port_group(struct ofproto
*p
, unsigned int group
)
1072 struct ofport
*port
;
1073 unsigned int port_no
;
1075 assert(group
== DP_GROUP_ALL
|| group
== DP_GROUP_FLOOD
);
1077 ports
= xmalloc(port_array_count(&p
->ports
) * sizeof *ports
);
1079 PORT_ARRAY_FOR_EACH (port
, &p
->ports
, port_no
) {
1080 if (group
== DP_GROUP_ALL
|| !(port
->opp
.config
& OFPPC_NO_FLOOD
)) {
1081 ports
[n_ports
++] = port_no
;
1084 dpif_port_group_set(p
->dpif
, group
, ports
, n_ports
);
1089 refresh_port_groups(struct ofproto
*p
)
1091 refresh_port_group(p
, DP_GROUP_FLOOD
);
1092 refresh_port_group(p
, DP_GROUP_ALL
);
1095 static struct ofport
*
1096 make_ofport(const struct odp_port
*odp_port
)
1098 enum netdev_flags flags
;
1099 struct ofport
*ofport
;
1100 struct netdev
*netdev
;
1104 error
= netdev_open(odp_port
->devname
, NETDEV_ETH_TYPE_NONE
, &netdev
);
1106 VLOG_WARN_RL(&rl
, "ignoring port %s (%"PRIu16
") because netdev %s "
1107 "cannot be opened (%s)",
1108 odp_port
->devname
, odp_port
->port
,
1109 odp_port
->devname
, strerror(error
));
1113 ofport
= xmalloc(sizeof *ofport
);
1114 ofport
->netdev
= netdev
;
1115 ofport
->opp
.port_no
= odp_port_to_ofp_port(odp_port
->port
);
1116 netdev_get_etheraddr(netdev
, ofport
->opp
.hw_addr
);
1117 memcpy(ofport
->opp
.name
, odp_port
->devname
,
1118 MIN(sizeof ofport
->opp
.name
, sizeof odp_port
->devname
));
1119 ofport
->opp
.name
[sizeof ofport
->opp
.name
- 1] = '\0';
1121 netdev_get_flags(netdev
, &flags
);
1122 ofport
->opp
.config
= flags
& NETDEV_UP
? 0 : OFPPC_PORT_DOWN
;
1124 netdev_get_carrier(netdev
, &carrier
);
1125 ofport
->opp
.state
= carrier
? 0 : OFPPS_LINK_DOWN
;
1127 netdev_get_features(netdev
,
1128 &ofport
->opp
.curr
, &ofport
->opp
.advertised
,
1129 &ofport
->opp
.supported
, &ofport
->opp
.peer
);
1134 ofport_conflicts(const struct ofproto
*p
, const struct odp_port
*odp_port
)
1136 if (port_array_get(&p
->ports
, odp_port
->port
)) {
1137 VLOG_WARN_RL(&rl
, "ignoring duplicate port %"PRIu16
" in datapath",
1140 } else if (shash_find(&p
->port_by_name
, odp_port
->devname
)) {
1141 VLOG_WARN_RL(&rl
, "ignoring duplicate device %s in datapath",
1150 ofport_equal(const struct ofport
*a_
, const struct ofport
*b_
)
1152 const struct ofp_phy_port
*a
= &a_
->opp
;
1153 const struct ofp_phy_port
*b
= &b_
->opp
;
1155 BUILD_ASSERT_DECL(sizeof *a
== 48); /* Detect ofp_phy_port changes. */
1156 return (a
->port_no
== b
->port_no
1157 && !memcmp(a
->hw_addr
, b
->hw_addr
, sizeof a
->hw_addr
)
1158 && !strcmp((char *) a
->name
, (char *) b
->name
)
1159 && a
->state
== b
->state
1160 && a
->config
== b
->config
1161 && a
->curr
== b
->curr
1162 && a
->advertised
== b
->advertised
1163 && a
->supported
== b
->supported
1164 && a
->peer
== b
->peer
);
1168 send_port_status(struct ofproto
*p
, const struct ofport
*ofport
,
1171 /* XXX Should limit the number of queued port status change messages. */
1172 struct ofconn
*ofconn
;
1173 LIST_FOR_EACH (ofconn
, struct ofconn
, node
, &p
->all_conns
) {
1174 struct ofp_port_status
*ops
;
1177 ops
= make_openflow_xid(sizeof *ops
, OFPT_PORT_STATUS
, 0, &b
);
1178 ops
->reason
= reason
;
1179 ops
->desc
= ofport
->opp
;
1180 hton_ofp_phy_port(&ops
->desc
);
1181 queue_tx(b
, ofconn
, NULL
);
1183 if (p
->ofhooks
->port_changed_cb
) {
1184 p
->ofhooks
->port_changed_cb(reason
, &ofport
->opp
, p
->aux
);
1189 ofport_install(struct ofproto
*p
, struct ofport
*ofport
)
1191 netdev_monitor_add(p
->netdev_monitor
, ofport
->netdev
);
1192 port_array_set(&p
->ports
, ofp_port_to_odp_port(ofport
->opp
.port_no
),
1194 shash_add(&p
->port_by_name
, (char *) ofport
->opp
.name
, ofport
);
1198 ofport_remove(struct ofproto
*p
, struct ofport
*ofport
)
1200 netdev_monitor_remove(p
->netdev_monitor
, ofport
->netdev
);
1201 port_array_set(&p
->ports
, ofp_port_to_odp_port(ofport
->opp
.port_no
), NULL
);
1202 shash_delete(&p
->port_by_name
,
1203 shash_find(&p
->port_by_name
, (char *) ofport
->opp
.name
));
1207 ofport_free(struct ofport
*ofport
)
1210 netdev_close(ofport
->netdev
);
1216 update_port(struct ofproto
*p
, const char *devname
)
1218 struct odp_port odp_port
;
1219 struct ofport
*old_ofport
;
1220 struct ofport
*new_ofport
;
1223 COVERAGE_INC(ofproto_update_port
);
1225 /* Query the datapath for port information. */
1226 error
= dpif_port_query_by_name(p
->dpif
, devname
, &odp_port
);
1228 /* Find the old ofport. */
1229 old_ofport
= shash_find_data(&p
->port_by_name
, devname
);
1232 /* There's no port named 'devname' but there might be a port with
1233 * the same port number. This could happen if a port is deleted
1234 * and then a new one added in its place very quickly, or if a port
1235 * is renamed. In the former case we want to send an OFPPR_DELETE
1236 * and an OFPPR_ADD, and in the latter case we want to send a
1237 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1238 * the old port's ifindex against the new port, or perhaps less
1239 * reliably but more portably by comparing the old port's MAC
1240 * against the new port's MAC. However, this code isn't that smart
1241 * and always sends an OFPPR_MODIFY (XXX). */
1242 old_ofport
= port_array_get(&p
->ports
, odp_port
.port
);
1244 } else if (error
!= ENOENT
&& error
!= ENODEV
) {
1245 VLOG_WARN_RL(&rl
, "dpif_port_query_by_name returned unexpected error "
1246 "%s", strerror(error
));
1250 /* Create a new ofport. */
1251 new_ofport
= !error
? make_ofport(&odp_port
) : NULL
;
1253 /* Eliminate a few pathological cases. */
1254 if (!old_ofport
&& !new_ofport
) {
1256 } else if (old_ofport
&& new_ofport
) {
1257 /* Most of the 'config' bits are OpenFlow soft state, but
1258 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1259 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1260 * leaves the other bits 0.) */
1261 new_ofport
->opp
.config
|= old_ofport
->opp
.config
& ~OFPPC_PORT_DOWN
;
1263 if (ofport_equal(old_ofport
, new_ofport
)) {
1264 /* False alarm--no change. */
1265 ofport_free(new_ofport
);
1270 /* Now deal with the normal cases. */
1272 ofport_remove(p
, old_ofport
);
1275 ofport_install(p
, new_ofport
);
1277 send_port_status(p
, new_ofport
? new_ofport
: old_ofport
,
1278 (!old_ofport
? OFPPR_ADD
1279 : !new_ofport
? OFPPR_DELETE
1281 ofport_free(old_ofport
);
1283 /* Update port groups. */
1284 refresh_port_groups(p
);
1288 init_ports(struct ofproto
*p
)
1290 struct odp_port
*ports
;
1295 error
= dpif_port_list(p
->dpif
, &ports
, &n_ports
);
1300 for (i
= 0; i
< n_ports
; i
++) {
1301 const struct odp_port
*odp_port
= &ports
[i
];
1302 if (!ofport_conflicts(p
, odp_port
)) {
1303 struct ofport
*ofport
= make_ofport(odp_port
);
1305 ofport_install(p
, ofport
);
1310 refresh_port_groups(p
);
1314 static struct ofconn
*
1315 ofconn_create(struct ofproto
*p
, struct rconn
*rconn
)
1317 struct ofconn
*ofconn
= xmalloc(sizeof *ofconn
);
1318 list_push_back(&p
->all_conns
, &ofconn
->node
);
1319 ofconn
->rconn
= rconn
;
1320 ofconn
->pktbuf
= NULL
;
1321 ofconn
->send_flow_exp
= false;
1322 ofconn
->miss_send_len
= 0;
1323 ofconn
->packet_in_counter
= rconn_packet_counter_create ();
1324 ofconn
->reply_counter
= rconn_packet_counter_create ();
1329 ofconn_destroy(struct ofconn
*ofconn
, struct ofproto
*p
)
1332 executer_rconn_closing(p
->executer
, ofconn
->rconn
);
1335 list_remove(&ofconn
->node
);
1336 rconn_destroy(ofconn
->rconn
);
1337 rconn_packet_counter_destroy(ofconn
->packet_in_counter
);
1338 rconn_packet_counter_destroy(ofconn
->reply_counter
);
1339 pktbuf_destroy(ofconn
->pktbuf
);
1344 ofconn_run(struct ofconn
*ofconn
, struct ofproto
*p
)
1348 rconn_run(ofconn
->rconn
);
1350 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1351 /* Limit the number of iterations to prevent other tasks from
1353 for (iteration
= 0; iteration
< 50; iteration
++) {
1354 struct ofpbuf
*of_msg
= rconn_recv(ofconn
->rconn
);
1358 handle_openflow(ofconn
, p
, of_msg
);
1359 ofpbuf_delete(of_msg
);
1363 if (ofconn
!= p
->controller
&& !rconn_is_alive(ofconn
->rconn
)) {
1364 ofconn_destroy(ofconn
, p
);
1369 ofconn_wait(struct ofconn
*ofconn
)
1371 rconn_run_wait(ofconn
->rconn
);
1372 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1373 rconn_recv_wait(ofconn
->rconn
);
1375 COVERAGE_INC(ofproto_ofconn_stuck
);
1379 /* Caller is responsible for initializing the 'cr' member of the returned
1381 static struct rule
*
1382 rule_create(struct rule
*super
,
1383 const union ofp_action
*actions
, size_t n_actions
,
1384 uint16_t idle_timeout
, uint16_t hard_timeout
)
1386 struct rule
*rule
= xcalloc(1, sizeof *rule
);
1387 rule
->idle_timeout
= idle_timeout
;
1388 rule
->hard_timeout
= hard_timeout
;
1389 rule
->used
= rule
->created
= time_msec();
1390 rule
->super
= super
;
1392 list_push_back(&super
->list
, &rule
->list
);
1394 list_init(&rule
->list
);
1396 rule
->n_actions
= n_actions
;
1397 rule
->actions
= xmemdup(actions
, n_actions
* sizeof *actions
);
1401 static struct rule
*
1402 rule_from_cls_rule(const struct cls_rule
*cls_rule
)
1404 return cls_rule
? CONTAINER_OF(cls_rule
, struct rule
, cr
) : NULL
;
1408 rule_free(struct rule
*rule
)
1410 free(rule
->actions
);
1411 free(rule
->odp_actions
);
1415 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1416 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1417 * through all of its subrules and revalidates them, destroying any that no
1418 * longer has a super-rule (which is probably all of them).
1420 * Before calling this function, the caller must make have removed 'rule' from
1421 * the classifier. If 'rule' is an exact-match rule, the caller is also
1422 * responsible for ensuring that it has been uninstalled from the datapath. */
1424 rule_destroy(struct ofproto
*ofproto
, struct rule
*rule
)
1427 struct rule
*subrule
, *next
;
1428 LIST_FOR_EACH_SAFE (subrule
, next
, struct rule
, list
, &rule
->list
) {
1429 revalidate_rule(ofproto
, subrule
);
1432 list_remove(&rule
->list
);
1438 rule_has_out_port(const struct rule
*rule
, uint16_t out_port
)
1440 const union ofp_action
*oa
;
1441 struct actions_iterator i
;
1443 if (out_port
== htons(OFPP_NONE
)) {
1446 for (oa
= actions_first(&i
, rule
->actions
, rule
->n_actions
); oa
;
1447 oa
= actions_next(&i
)) {
1448 if (oa
->type
== htons(OFPAT_OUTPUT
) && oa
->output
.port
== out_port
) {
1455 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1456 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1458 * The flow that 'packet' actually contains does not need to actually match
1459 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1460 * the packet and byte counters for 'rule' will be credited for the packet sent
1461 * out whether or not the packet actually matches 'rule'.
1463 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1464 * the caller must already have accurately composed ODP actions for it given
1465 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1466 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1467 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1468 * actions and apply them to 'packet'. */
1470 rule_execute(struct ofproto
*ofproto
, struct rule
*rule
,
1471 struct ofpbuf
*packet
, const flow_t
*flow
)
1473 const union odp_action
*actions
;
1475 struct odp_actions a
;
1477 /* Grab or compose the ODP actions.
1479 * The special case for an exact-match 'rule' where 'flow' is not the
1480 * rule's flow is important to avoid, e.g., sending a packet out its input
1481 * port simply because the ODP actions were composed for the wrong
1483 if (rule
->cr
.wc
.wildcards
|| !flow_equal(flow
, &rule
->cr
.flow
)) {
1484 struct rule
*super
= rule
->super
? rule
->super
: rule
;
1485 if (xlate_actions(super
->actions
, super
->n_actions
, flow
, ofproto
,
1486 packet
, &a
, NULL
, 0)) {
1489 actions
= a
.actions
;
1490 n_actions
= a
.n_actions
;
1492 actions
= rule
->odp_actions
;
1493 n_actions
= rule
->n_odp_actions
;
1496 /* Execute the ODP actions. */
1497 if (!dpif_execute(ofproto
->dpif
, flow
->in_port
,
1498 actions
, n_actions
, packet
)) {
1499 struct odp_flow_stats stats
;
1500 flow_extract_stats(flow
, packet
, &stats
);
1501 update_stats(rule
, &stats
);
1502 rule
->used
= time_msec();
1507 rule_insert(struct ofproto
*p
, struct rule
*rule
, struct ofpbuf
*packet
,
1510 struct rule
*displaced_rule
;
1512 /* Insert the rule in the classifier. */
1513 displaced_rule
= rule_from_cls_rule(classifier_insert(&p
->cls
, &rule
->cr
));
1514 if (!rule
->cr
.wc
.wildcards
) {
1515 rule_make_actions(p
, rule
, packet
);
1518 /* Send the packet and credit it to the rule. */
1521 flow_extract(packet
, in_port
, &flow
);
1522 rule_execute(p
, rule
, packet
, &flow
);
1525 /* Install the rule in the datapath only after sending the packet, to
1526 * avoid packet reordering. */
1527 if (rule
->cr
.wc
.wildcards
) {
1528 COVERAGE_INC(ofproto_add_wc_flow
);
1529 p
->need_revalidate
= true;
1531 rule_install(p
, rule
, displaced_rule
);
1534 /* Free the rule that was displaced, if any. */
1535 if (displaced_rule
) {
1536 rule_destroy(p
, displaced_rule
);
1540 static struct rule
*
1541 rule_create_subrule(struct ofproto
*ofproto
, struct rule
*rule
,
1544 struct rule
*subrule
= rule_create(rule
, NULL
, 0,
1545 rule
->idle_timeout
, rule
->hard_timeout
);
1546 COVERAGE_INC(ofproto_subrule_create
);
1547 cls_rule_from_flow(&subrule
->cr
, flow
, 0,
1548 (rule
->cr
.priority
<= UINT16_MAX
? UINT16_MAX
1549 : rule
->cr
.priority
));
1550 classifier_insert_exact(&ofproto
->cls
, &subrule
->cr
);
1556 rule_remove(struct ofproto
*ofproto
, struct rule
*rule
)
1558 if (rule
->cr
.wc
.wildcards
) {
1559 COVERAGE_INC(ofproto_del_wc_flow
);
1560 ofproto
->need_revalidate
= true;
1562 rule_uninstall(ofproto
, rule
);
1564 classifier_remove(&ofproto
->cls
, &rule
->cr
);
1565 rule_destroy(ofproto
, rule
);
1568 /* Returns true if the actions changed, false otherwise. */
1570 rule_make_actions(struct ofproto
*p
, struct rule
*rule
,
1571 const struct ofpbuf
*packet
)
1573 const struct rule
*super
;
1574 struct odp_actions a
;
1577 assert(!rule
->cr
.wc
.wildcards
);
1579 super
= rule
->super
? rule
->super
: rule
;
1581 xlate_actions(super
->actions
, super
->n_actions
, &rule
->cr
.flow
, p
,
1582 packet
, &a
, &rule
->tags
, &rule
->may_install
);
1584 actions_len
= a
.n_actions
* sizeof *a
.actions
;
1585 if (rule
->n_odp_actions
!= a
.n_actions
1586 || memcmp(rule
->odp_actions
, a
.actions
, actions_len
)) {
1587 COVERAGE_INC(ofproto_odp_unchanged
);
1588 free(rule
->odp_actions
);
1589 rule
->n_odp_actions
= a
.n_actions
;
1590 rule
->odp_actions
= xmemdup(a
.actions
, actions_len
);
1598 do_put_flow(struct ofproto
*ofproto
, struct rule
*rule
, int flags
,
1599 struct odp_flow_put
*put
)
1601 memset(&put
->flow
.stats
, 0, sizeof put
->flow
.stats
);
1602 put
->flow
.key
= rule
->cr
.flow
;
1603 put
->flow
.actions
= rule
->odp_actions
;
1604 put
->flow
.n_actions
= rule
->n_odp_actions
;
1606 return dpif_flow_put(ofproto
->dpif
, put
);
1610 rule_install(struct ofproto
*p
, struct rule
*rule
, struct rule
*displaced_rule
)
1612 assert(!rule
->cr
.wc
.wildcards
);
1614 if (rule
->may_install
) {
1615 struct odp_flow_put put
;
1616 if (!do_put_flow(p
, rule
,
1617 ODPPF_CREATE
| ODPPF_MODIFY
| ODPPF_ZERO_STATS
,
1619 rule
->installed
= true;
1620 if (displaced_rule
) {
1621 update_stats(rule
, &put
.flow
.stats
);
1622 rule_post_uninstall(p
, displaced_rule
);
1625 } else if (displaced_rule
) {
1626 rule_uninstall(p
, displaced_rule
);
1631 rule_reinstall(struct ofproto
*ofproto
, struct rule
*rule
)
1633 if (rule
->installed
) {
1634 struct odp_flow_put put
;
1635 COVERAGE_INC(ofproto_dp_missed
);
1636 do_put_flow(ofproto
, rule
, ODPPF_CREATE
| ODPPF_MODIFY
, &put
);
1638 rule_install(ofproto
, rule
, NULL
);
1643 rule_update_actions(struct ofproto
*ofproto
, struct rule
*rule
)
1645 bool actions_changed
= rule_make_actions(ofproto
, rule
, NULL
);
1646 if (rule
->may_install
) {
1647 if (rule
->installed
) {
1648 if (actions_changed
) {
1649 /* XXX should really do rule_post_uninstall() for the *old* set
1650 * of actions, and distinguish the old stats from the new. */
1651 struct odp_flow_put put
;
1652 do_put_flow(ofproto
, rule
, ODPPF_CREATE
| ODPPF_MODIFY
, &put
);
1655 rule_install(ofproto
, rule
, NULL
);
1658 rule_uninstall(ofproto
, rule
);
1663 rule_account(struct ofproto
*ofproto
, struct rule
*rule
, uint64_t extra_bytes
)
1665 uint64_t total_bytes
= rule
->byte_count
+ extra_bytes
;
1667 if (ofproto
->ofhooks
->account_flow_cb
1668 && total_bytes
> rule
->accounted_bytes
)
1670 ofproto
->ofhooks
->account_flow_cb(
1671 &rule
->cr
.flow
, rule
->odp_actions
, rule
->n_odp_actions
,
1672 total_bytes
- rule
->accounted_bytes
, ofproto
->aux
);
1673 rule
->accounted_bytes
= total_bytes
;
1678 rule_uninstall(struct ofproto
*p
, struct rule
*rule
)
1680 assert(!rule
->cr
.wc
.wildcards
);
1681 if (rule
->installed
) {
1682 struct odp_flow odp_flow
;
1684 odp_flow
.key
= rule
->cr
.flow
;
1685 odp_flow
.actions
= NULL
;
1686 odp_flow
.n_actions
= 0;
1687 if (!dpif_flow_del(p
->dpif
, &odp_flow
)) {
1688 update_stats(rule
, &odp_flow
.stats
);
1690 rule
->installed
= false;
1692 rule_post_uninstall(p
, rule
);
1697 rule_post_uninstall(struct ofproto
*ofproto
, struct rule
*rule
)
1699 struct rule
*super
= rule
->super
;
1701 rule_account(ofproto
, rule
, 0);
1702 if (ofproto
->netflow
&& rule
->byte_count
) {
1703 struct ofexpired expired
;
1704 expired
.flow
= rule
->cr
.flow
;
1705 expired
.packet_count
= rule
->packet_count
;
1706 expired
.byte_count
= rule
->byte_count
;
1707 expired
.used
= rule
->used
;
1708 expired
.created
= rule
->created
;
1709 expired
.tcp_flags
= rule
->tcp_flags
;
1710 expired
.ip_tos
= rule
->ip_tos
;
1711 netflow_expire(ofproto
->netflow
, &expired
);
1714 super
->packet_count
+= rule
->packet_count
;
1715 super
->byte_count
+= rule
->byte_count
;
1716 super
->tcp_flags
|= rule
->tcp_flags
;
1717 if (rule
->packet_count
) {
1718 super
->ip_tos
= rule
->ip_tos
;
1722 /* Reset counters to prevent double counting if the rule ever gets
1724 rule
->packet_count
= 0;
1725 rule
->byte_count
= 0;
1726 rule
->accounted_bytes
= 0;
1727 rule
->tcp_flags
= 0;
1732 queue_tx(struct ofpbuf
*msg
, const struct ofconn
*ofconn
,
1733 struct rconn_packet_counter
*counter
)
1735 update_openflow_length(msg
);
1736 if (rconn_send(ofconn
->rconn
, msg
, counter
)) {
1742 send_error(const struct ofconn
*ofconn
, const struct ofp_header
*oh
,
1743 int error
, const void *data
, size_t len
)
1746 struct ofp_error_msg
*oem
;
1748 if (!(error
>> 16)) {
1749 VLOG_WARN_RL(&rl
, "not sending bad error code %d to controller",
1754 COVERAGE_INC(ofproto_error
);
1755 oem
= make_openflow_xid(len
+ sizeof *oem
, OFPT_ERROR
,
1756 oh
? oh
->xid
: 0, &buf
);
1757 oem
->type
= htons((unsigned int) error
>> 16);
1758 oem
->code
= htons(error
& 0xffff);
1759 memcpy(oem
->data
, data
, len
);
1760 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
1764 send_error_oh(const struct ofconn
*ofconn
, const struct ofp_header
*oh
,
1767 size_t oh_length
= ntohs(oh
->length
);
1768 send_error(ofconn
, oh
, error
, oh
, MIN(oh_length
, 64));
1772 hton_ofp_phy_port(struct ofp_phy_port
*opp
)
1774 opp
->port_no
= htons(opp
->port_no
);
1775 opp
->config
= htonl(opp
->config
);
1776 opp
->state
= htonl(opp
->state
);
1777 opp
->curr
= htonl(opp
->curr
);
1778 opp
->advertised
= htonl(opp
->advertised
);
1779 opp
->supported
= htonl(opp
->supported
);
1780 opp
->peer
= htonl(opp
->peer
);
1784 handle_echo_request(struct ofconn
*ofconn
, struct ofp_header
*oh
)
1786 struct ofp_header
*rq
= oh
;
1787 queue_tx(make_echo_reply(rq
), ofconn
, ofconn
->reply_counter
);
1792 handle_features_request(struct ofproto
*p
, struct ofconn
*ofconn
,
1793 struct ofp_header
*oh
)
1795 struct ofp_switch_features
*osf
;
1797 unsigned int port_no
;
1798 struct ofport
*port
;
1800 osf
= make_openflow_xid(sizeof *osf
, OFPT_FEATURES_REPLY
, oh
->xid
, &buf
);
1801 osf
->datapath_id
= htonll(p
->datapath_id
);
1802 osf
->n_buffers
= htonl(pktbuf_capacity());
1804 osf
->capabilities
= htonl(OFPC_FLOW_STATS
| OFPC_TABLE_STATS
|
1805 OFPC_PORT_STATS
| OFPC_MULTI_PHY_TX
);
1806 osf
->actions
= htonl((1u << OFPAT_OUTPUT
) |
1807 (1u << OFPAT_SET_VLAN_VID
) |
1808 (1u << OFPAT_SET_VLAN_PCP
) |
1809 (1u << OFPAT_STRIP_VLAN
) |
1810 (1u << OFPAT_SET_DL_SRC
) |
1811 (1u << OFPAT_SET_DL_DST
) |
1812 (1u << OFPAT_SET_NW_SRC
) |
1813 (1u << OFPAT_SET_NW_DST
) |
1814 (1u << OFPAT_SET_TP_SRC
) |
1815 (1u << OFPAT_SET_TP_DST
));
1817 PORT_ARRAY_FOR_EACH (port
, &p
->ports
, port_no
) {
1818 hton_ofp_phy_port(ofpbuf_put(buf
, &port
->opp
, sizeof port
->opp
));
1821 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
1826 handle_get_config_request(struct ofproto
*p
, struct ofconn
*ofconn
,
1827 struct ofp_header
*oh
)
1830 struct ofp_switch_config
*osc
;
1834 /* Figure out flags. */
1835 dpif_get_drop_frags(p
->dpif
, &drop_frags
);
1836 flags
= drop_frags
? OFPC_FRAG_DROP
: OFPC_FRAG_NORMAL
;
1837 if (ofconn
->send_flow_exp
) {
1838 flags
|= OFPC_SEND_FLOW_EXP
;
1842 osc
= make_openflow_xid(sizeof *osc
, OFPT_GET_CONFIG_REPLY
, oh
->xid
, &buf
);
1843 osc
->flags
= htons(flags
);
1844 osc
->miss_send_len
= htons(ofconn
->miss_send_len
);
1845 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
1851 handle_set_config(struct ofproto
*p
, struct ofconn
*ofconn
,
1852 struct ofp_switch_config
*osc
)
1857 error
= check_ofp_message(&osc
->header
, OFPT_SET_CONFIG
, sizeof *osc
);
1861 flags
= ntohs(osc
->flags
);
1863 ofconn
->send_flow_exp
= (flags
& OFPC_SEND_FLOW_EXP
) != 0;
1865 if (ofconn
== p
->controller
) {
1866 switch (flags
& OFPC_FRAG_MASK
) {
1867 case OFPC_FRAG_NORMAL
:
1868 dpif_set_drop_frags(p
->dpif
, false);
1870 case OFPC_FRAG_DROP
:
1871 dpif_set_drop_frags(p
->dpif
, true);
1874 VLOG_WARN_RL(&rl
, "requested bad fragment mode (flags=%"PRIx16
")",
1880 if ((ntohs(osc
->miss_send_len
) != 0) != (ofconn
->miss_send_len
!= 0)) {
1881 if (ntohs(osc
->miss_send_len
) != 0) {
1882 ofconn
->pktbuf
= pktbuf_create();
1884 pktbuf_destroy(ofconn
->pktbuf
);
1888 ofconn
->miss_send_len
= ntohs(osc
->miss_send_len
);
1894 add_output_group_action(struct odp_actions
*actions
, uint16_t group
)
1896 odp_actions_add(actions
, ODPAT_OUTPUT_GROUP
)->output_group
.group
= group
;
1900 add_controller_action(struct odp_actions
*actions
,
1901 const struct ofp_action_output
*oao
)
1903 union odp_action
*a
= odp_actions_add(actions
, ODPAT_CONTROLLER
);
1904 a
->controller
.arg
= oao
->max_len
? ntohs(oao
->max_len
) : UINT32_MAX
;
1907 struct action_xlate_ctx
{
1909 const flow_t
*flow
; /* Flow to which these actions correspond. */
1910 int recurse
; /* Recursion level, via xlate_table_action. */
1911 struct ofproto
*ofproto
;
1912 const struct ofpbuf
*packet
; /* The packet corresponding to 'flow', or a
1913 * null pointer if we are revalidating
1914 * without a packet to refer to. */
1917 struct odp_actions
*out
; /* Datapath actions. */
1918 tag_type
*tags
; /* Tags associated with OFPP_NORMAL actions. */
1919 bool may_setup_flow
; /* True ordinarily; false if the actions must
1920 * be reassessed for every packet. */
1923 static void do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
1924 struct action_xlate_ctx
*ctx
);
1927 add_output_action(struct action_xlate_ctx
*ctx
, uint16_t port
)
1929 const struct ofport
*ofport
= port_array_get(&ctx
->ofproto
->ports
, port
);
1930 if (!ofport
|| !(ofport
->opp
.config
& OFPPC_NO_FWD
)) {
1931 odp_actions_add(ctx
->out
, ODPAT_OUTPUT
)->output
.port
= port
;
1935 static struct rule
*
1936 lookup_valid_rule(struct ofproto
*ofproto
, const flow_t
*flow
)
1939 rule
= rule_from_cls_rule(classifier_lookup(&ofproto
->cls
, flow
));
1941 /* The rule we found might not be valid, since we could be in need of
1942 * revalidation. If it is not valid, don't return it. */
1945 && ofproto
->need_revalidate
1946 && !revalidate_rule(ofproto
, rule
)) {
1947 COVERAGE_INC(ofproto_invalidated
);
1955 xlate_table_action(struct action_xlate_ctx
*ctx
, uint16_t in_port
)
1957 if (!ctx
->recurse
) {
1962 flow
.in_port
= in_port
;
1964 rule
= lookup_valid_rule(ctx
->ofproto
, &flow
);
1971 do_xlate_actions(rule
->actions
, rule
->n_actions
, ctx
);
1978 xlate_output_action(struct action_xlate_ctx
*ctx
,
1979 const struct ofp_action_output
*oao
)
1983 switch (ntohs(oao
->port
)) {
1985 add_output_action(ctx
, ctx
->flow
->in_port
);
1988 xlate_table_action(ctx
, ctx
->flow
->in_port
);
1991 if (!ctx
->ofproto
->ofhooks
->normal_cb(ctx
->flow
, ctx
->packet
,
1992 ctx
->out
, ctx
->tags
,
1993 ctx
->ofproto
->aux
)) {
1994 COVERAGE_INC(ofproto_uninstallable
);
1995 ctx
->may_setup_flow
= false;
1999 add_output_group_action(ctx
->out
, DP_GROUP_FLOOD
);
2002 add_output_group_action(ctx
->out
, DP_GROUP_ALL
);
2004 case OFPP_CONTROLLER
:
2005 add_controller_action(ctx
->out
, oao
);
2008 add_output_action(ctx
, ODPP_LOCAL
);
2011 odp_port
= ofp_port_to_odp_port(ntohs(oao
->port
));
2012 if (odp_port
!= ctx
->flow
->in_port
) {
2013 add_output_action(ctx
, odp_port
);
2020 xlate_nicira_action(struct action_xlate_ctx
*ctx
,
2021 const struct nx_action_header
*nah
)
2023 const struct nx_action_resubmit
*nar
;
2024 int subtype
= ntohs(nah
->subtype
);
2026 assert(nah
->vendor
== htonl(NX_VENDOR_ID
));
2028 case NXAST_RESUBMIT
:
2029 nar
= (const struct nx_action_resubmit
*) nah
;
2030 xlate_table_action(ctx
, ofp_port_to_odp_port(ntohs(nar
->in_port
)));
2034 VLOG_DBG_RL(&rl
, "unknown Nicira action type %"PRIu16
, subtype
);
2040 do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
2041 struct action_xlate_ctx
*ctx
)
2043 struct actions_iterator iter
;
2044 const union ofp_action
*ia
;
2045 const struct ofport
*port
;
2047 port
= port_array_get(&ctx
->ofproto
->ports
, ctx
->flow
->in_port
);
2048 if (port
&& port
->opp
.config
& (OFPPC_NO_RECV
| OFPPC_NO_RECV_STP
) &&
2049 port
->opp
.config
& (eth_addr_equals(ctx
->flow
->dl_dst
, stp_eth_addr
)
2050 ? OFPPC_NO_RECV_STP
: OFPPC_NO_RECV
)) {
2051 /* Drop this flow. */
2055 for (ia
= actions_first(&iter
, in
, n_in
); ia
; ia
= actions_next(&iter
)) {
2056 uint16_t type
= ntohs(ia
->type
);
2057 union odp_action
*oa
;
2061 xlate_output_action(ctx
, &ia
->output
);
2064 case OFPAT_SET_VLAN_VID
:
2065 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_VLAN_VID
);
2066 oa
->vlan_vid
.vlan_vid
= ia
->vlan_vid
.vlan_vid
;
2069 case OFPAT_SET_VLAN_PCP
:
2070 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_VLAN_PCP
);
2071 oa
->vlan_pcp
.vlan_pcp
= ia
->vlan_pcp
.vlan_pcp
;
2074 case OFPAT_STRIP_VLAN
:
2075 odp_actions_add(ctx
->out
, ODPAT_STRIP_VLAN
);
2078 case OFPAT_SET_DL_SRC
:
2079 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_DL_SRC
);
2080 memcpy(oa
->dl_addr
.dl_addr
,
2081 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2084 case OFPAT_SET_DL_DST
:
2085 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_DL_DST
);
2086 memcpy(oa
->dl_addr
.dl_addr
,
2087 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2090 case OFPAT_SET_NW_SRC
:
2091 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_NW_SRC
);
2092 oa
->nw_addr
.nw_addr
= ia
->nw_addr
.nw_addr
;
2095 case OFPAT_SET_TP_SRC
:
2096 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_TP_SRC
);
2097 oa
->tp_port
.tp_port
= ia
->tp_port
.tp_port
;
2101 xlate_nicira_action(ctx
, (const struct nx_action_header
*) ia
);
2105 VLOG_DBG_RL(&rl
, "unknown action type %"PRIu16
, type
);
2112 xlate_actions(const union ofp_action
*in
, size_t n_in
,
2113 const flow_t
*flow
, struct ofproto
*ofproto
,
2114 const struct ofpbuf
*packet
,
2115 struct odp_actions
*out
, tag_type
*tags
, bool *may_setup_flow
)
2117 tag_type no_tags
= 0;
2118 struct action_xlate_ctx ctx
;
2119 COVERAGE_INC(ofproto_ofp2odp
);
2120 odp_actions_init(out
);
2123 ctx
.ofproto
= ofproto
;
2124 ctx
.packet
= packet
;
2126 ctx
.tags
= tags
? tags
: &no_tags
;
2127 ctx
.may_setup_flow
= true;
2128 do_xlate_actions(in
, n_in
, &ctx
);
2130 /* Check with in-band control to see if we're allowed to setup this
2132 if (!in_band_rule_check(ofproto
->in_band
, flow
, out
)) {
2133 ctx
.may_setup_flow
= false;
2136 if (may_setup_flow
) {
2137 *may_setup_flow
= ctx
.may_setup_flow
;
2139 if (odp_actions_overflow(out
)) {
2140 odp_actions_init(out
);
2141 return ofp_mkerr(OFPET_BAD_ACTION
, OFPBAC_TOO_MANY
);
2147 handle_packet_out(struct ofproto
*p
, struct ofconn
*ofconn
,
2148 struct ofp_header
*oh
)
2150 struct ofp_packet_out
*opo
;
2151 struct ofpbuf payload
, *buffer
;
2152 struct odp_actions actions
;
2158 error
= check_ofp_packet_out(oh
, &payload
, &n_actions
, p
->max_ports
);
2162 opo
= (struct ofp_packet_out
*) oh
;
2164 COVERAGE_INC(ofproto_packet_out
);
2165 if (opo
->buffer_id
!= htonl(UINT32_MAX
)) {
2166 error
= pktbuf_retrieve(ofconn
->pktbuf
, ntohl(opo
->buffer_id
),
2176 flow_extract(&payload
, ofp_port_to_odp_port(ntohs(opo
->in_port
)), &flow
);
2177 error
= xlate_actions((const union ofp_action
*) opo
->actions
, n_actions
,
2178 &flow
, p
, &payload
, &actions
, NULL
, NULL
);
2183 dpif_execute(p
->dpif
, flow
.in_port
, actions
.actions
, actions
.n_actions
,
2185 ofpbuf_delete(buffer
);
2191 update_port_config(struct ofproto
*p
, struct ofport
*port
,
2192 uint32_t config
, uint32_t mask
)
2194 mask
&= config
^ port
->opp
.config
;
2195 if (mask
& OFPPC_PORT_DOWN
) {
2196 if (config
& OFPPC_PORT_DOWN
) {
2197 netdev_turn_flags_off(port
->netdev
, NETDEV_UP
, true);
2199 netdev_turn_flags_on(port
->netdev
, NETDEV_UP
, true);
2202 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2203 if (mask
& REVALIDATE_BITS
) {
2204 COVERAGE_INC(ofproto_costly_flags
);
2205 port
->opp
.config
^= mask
& REVALIDATE_BITS
;
2206 p
->need_revalidate
= true;
2208 #undef REVALIDATE_BITS
2209 if (mask
& OFPPC_NO_FLOOD
) {
2210 port
->opp
.config
^= OFPPC_NO_FLOOD
;
2211 refresh_port_group(p
, DP_GROUP_FLOOD
);
2213 if (mask
& OFPPC_NO_PACKET_IN
) {
2214 port
->opp
.config
^= OFPPC_NO_PACKET_IN
;
2219 handle_port_mod(struct ofproto
*p
, struct ofp_header
*oh
)
2221 const struct ofp_port_mod
*opm
;
2222 struct ofport
*port
;
2225 error
= check_ofp_message(oh
, OFPT_PORT_MOD
, sizeof *opm
);
2229 opm
= (struct ofp_port_mod
*) oh
;
2231 port
= port_array_get(&p
->ports
,
2232 ofp_port_to_odp_port(ntohs(opm
->port_no
)));
2234 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_PORT
);
2235 } else if (memcmp(port
->opp
.hw_addr
, opm
->hw_addr
, OFP_ETH_ALEN
)) {
2236 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_HW_ADDR
);
2238 update_port_config(p
, port
, ntohl(opm
->config
), ntohl(opm
->mask
));
2239 if (opm
->advertise
) {
2240 netdev_set_advertisements(port
->netdev
, ntohl(opm
->advertise
));
2246 static struct ofpbuf
*
2247 make_stats_reply(uint32_t xid
, uint16_t type
, size_t body_len
)
2249 struct ofp_stats_reply
*osr
;
2252 msg
= ofpbuf_new(MIN(sizeof *osr
+ body_len
, UINT16_MAX
));
2253 osr
= put_openflow_xid(sizeof *osr
, OFPT_STATS_REPLY
, xid
, msg
);
2255 osr
->flags
= htons(0);
2259 static struct ofpbuf
*
2260 start_stats_reply(const struct ofp_stats_request
*request
, size_t body_len
)
2262 return make_stats_reply(request
->header
.xid
, request
->type
, body_len
);
2266 append_stats_reply(size_t nbytes
, struct ofconn
*ofconn
, struct ofpbuf
**msgp
)
2268 struct ofpbuf
*msg
= *msgp
;
2269 assert(nbytes
<= UINT16_MAX
- sizeof(struct ofp_stats_reply
));
2270 if (nbytes
+ msg
->size
> UINT16_MAX
) {
2271 struct ofp_stats_reply
*reply
= msg
->data
;
2272 reply
->flags
= htons(OFPSF_REPLY_MORE
);
2273 *msgp
= make_stats_reply(reply
->header
.xid
, reply
->type
, nbytes
);
2274 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
2276 return ofpbuf_put_uninit(*msgp
, nbytes
);
2280 handle_desc_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
2281 struct ofp_stats_request
*request
)
2283 struct ofp_desc_stats
*ods
;
2286 msg
= start_stats_reply(request
, sizeof *ods
);
2287 ods
= append_stats_reply(sizeof *ods
, ofconn
, &msg
);
2288 strncpy(ods
->mfr_desc
, p
->manufacturer
, sizeof ods
->mfr_desc
);
2289 strncpy(ods
->hw_desc
, p
->hardware
, sizeof ods
->hw_desc
);
2290 strncpy(ods
->sw_desc
, p
->software
, sizeof ods
->sw_desc
);
2291 strncpy(ods
->serial_num
, p
->serial
, sizeof ods
->serial_num
);
2292 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
2298 count_subrules(struct cls_rule
*cls_rule
, void *n_subrules_
)
2300 struct rule
*rule
= rule_from_cls_rule(cls_rule
);
2301 int *n_subrules
= n_subrules_
;
2309 handle_table_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
2310 struct ofp_stats_request
*request
)
2312 struct ofp_table_stats
*ots
;
2314 struct odp_stats dpstats
;
2315 int n_exact
, n_subrules
, n_wild
;
2317 msg
= start_stats_reply(request
, sizeof *ots
* 2);
2319 /* Count rules of various kinds. */
2321 classifier_for_each(&p
->cls
, CLS_INC_EXACT
, count_subrules
, &n_subrules
);
2322 n_exact
= classifier_count_exact(&p
->cls
) - n_subrules
;
2323 n_wild
= classifier_count(&p
->cls
) - classifier_count_exact(&p
->cls
);
2326 dpif_get_dp_stats(p
->dpif
, &dpstats
);
2327 ots
= append_stats_reply(sizeof *ots
, ofconn
, &msg
);
2328 memset(ots
, 0, sizeof *ots
);
2329 ots
->table_id
= TABLEID_HASH
;
2330 strcpy(ots
->name
, "hash");
2331 ots
->wildcards
= htonl(0);
2332 ots
->max_entries
= htonl(dpstats
.max_capacity
);
2333 ots
->active_count
= htonl(n_exact
);
2334 ots
->lookup_count
= htonll(dpstats
.n_frags
+ dpstats
.n_hit
+
2336 ots
->matched_count
= htonll(dpstats
.n_hit
); /* XXX */
2338 /* Classifier table. */
2339 ots
= append_stats_reply(sizeof *ots
, ofconn
, &msg
);
2340 memset(ots
, 0, sizeof *ots
);
2341 ots
->table_id
= TABLEID_CLASSIFIER
;
2342 strcpy(ots
->name
, "classifier");
2343 ots
->wildcards
= htonl(OFPFW_ALL
);
2344 ots
->max_entries
= htonl(65536);
2345 ots
->active_count
= htonl(n_wild
);
2346 ots
->lookup_count
= htonll(0); /* XXX */
2347 ots
->matched_count
= htonll(0); /* XXX */
2349 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
2354 handle_port_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
2355 struct ofp_stats_request
*request
)
2357 struct ofp_port_stats
*ops
;
2359 struct ofport
*port
;
2360 unsigned int port_no
;
2362 msg
= start_stats_reply(request
, sizeof *ops
* 16);
2363 PORT_ARRAY_FOR_EACH (port
, &p
->ports
, port_no
) {
2364 struct netdev_stats stats
;
2366 /* Intentionally ignore return value, since errors will set 'stats' to
2367 * all-1s, which is correct for OpenFlow, and netdev_get_stats() will
2369 netdev_get_stats(port
->netdev
, &stats
);
2371 ops
= append_stats_reply(sizeof *ops
, ofconn
, &msg
);
2372 ops
->port_no
= htons(odp_port_to_ofp_port(port_no
));
2373 memset(ops
->pad
, 0, sizeof ops
->pad
);
2374 ops
->rx_packets
= htonll(stats
.rx_packets
);
2375 ops
->tx_packets
= htonll(stats
.tx_packets
);
2376 ops
->rx_bytes
= htonll(stats
.rx_bytes
);
2377 ops
->tx_bytes
= htonll(stats
.tx_bytes
);
2378 ops
->rx_dropped
= htonll(stats
.rx_dropped
);
2379 ops
->tx_dropped
= htonll(stats
.tx_dropped
);
2380 ops
->rx_errors
= htonll(stats
.rx_errors
);
2381 ops
->tx_errors
= htonll(stats
.tx_errors
);
2382 ops
->rx_frame_err
= htonll(stats
.rx_frame_errors
);
2383 ops
->rx_over_err
= htonll(stats
.rx_over_errors
);
2384 ops
->rx_crc_err
= htonll(stats
.rx_crc_errors
);
2385 ops
->collisions
= htonll(stats
.collisions
);
2388 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
2392 struct flow_stats_cbdata
{
2393 struct ofproto
*ofproto
;
2394 struct ofconn
*ofconn
;
2400 query_stats(struct ofproto
*p
, struct rule
*rule
,
2401 uint64_t *packet_countp
, uint64_t *byte_countp
)
2403 uint64_t packet_count
, byte_count
;
2404 struct rule
*subrule
;
2405 struct odp_flow
*odp_flows
;
2408 n_odp_flows
= rule
->cr
.wc
.wildcards
? list_size(&rule
->list
) : 1;
2409 odp_flows
= xcalloc(1, n_odp_flows
* sizeof *odp_flows
);
2410 if (rule
->cr
.wc
.wildcards
) {
2412 LIST_FOR_EACH (subrule
, struct rule
, list
, &rule
->list
) {
2413 odp_flows
[i
++].key
= subrule
->cr
.flow
;
2416 odp_flows
[0].key
= rule
->cr
.flow
;
2419 packet_count
= rule
->packet_count
;
2420 byte_count
= rule
->byte_count
;
2421 if (!dpif_flow_get_multiple(p
->dpif
, odp_flows
, n_odp_flows
)) {
2423 for (i
= 0; i
< n_odp_flows
; i
++) {
2424 struct odp_flow
*odp_flow
= &odp_flows
[i
];
2425 packet_count
+= odp_flow
->stats
.n_packets
;
2426 byte_count
+= odp_flow
->stats
.n_bytes
;
2431 *packet_countp
= packet_count
;
2432 *byte_countp
= byte_count
;
2436 flow_stats_cb(struct cls_rule
*rule_
, void *cbdata_
)
2438 struct rule
*rule
= rule_from_cls_rule(rule_
);
2439 struct flow_stats_cbdata
*cbdata
= cbdata_
;
2440 struct ofp_flow_stats
*ofs
;
2441 uint64_t packet_count
, byte_count
;
2442 size_t act_len
, len
;
2444 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, cbdata
->out_port
)) {
2448 act_len
= sizeof *rule
->actions
* rule
->n_actions
;
2449 len
= offsetof(struct ofp_flow_stats
, actions
) + act_len
;
2451 query_stats(cbdata
->ofproto
, rule
, &packet_count
, &byte_count
);
2453 ofs
= append_stats_reply(len
, cbdata
->ofconn
, &cbdata
->msg
);
2454 ofs
->length
= htons(len
);
2455 ofs
->table_id
= rule
->cr
.wc
.wildcards
? TABLEID_CLASSIFIER
: TABLEID_HASH
;
2457 flow_to_match(&rule
->cr
.flow
, rule
->cr
.wc
.wildcards
, &ofs
->match
);
2458 ofs
->duration
= htonl((time_msec() - rule
->created
) / 1000);
2459 ofs
->priority
= htons(rule
->cr
.priority
);
2460 ofs
->idle_timeout
= htons(rule
->idle_timeout
);
2461 ofs
->hard_timeout
= htons(rule
->hard_timeout
);
2462 memset(ofs
->pad2
, 0, sizeof ofs
->pad2
);
2463 ofs
->packet_count
= htonll(packet_count
);
2464 ofs
->byte_count
= htonll(byte_count
);
2465 memcpy(ofs
->actions
, rule
->actions
, act_len
);
2469 table_id_to_include(uint8_t table_id
)
2471 return (table_id
== TABLEID_HASH
? CLS_INC_EXACT
2472 : table_id
== TABLEID_CLASSIFIER
? CLS_INC_WILD
2473 : table_id
== 0xff ? CLS_INC_ALL
2478 handle_flow_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
2479 const struct ofp_stats_request
*osr
,
2482 struct ofp_flow_stats_request
*fsr
;
2483 struct flow_stats_cbdata cbdata
;
2484 struct cls_rule target
;
2486 if (arg_size
!= sizeof *fsr
) {
2487 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LENGTH
);
2489 fsr
= (struct ofp_flow_stats_request
*) osr
->body
;
2491 COVERAGE_INC(ofproto_flows_req
);
2493 cbdata
.ofconn
= ofconn
;
2494 cbdata
.out_port
= fsr
->out_port
;
2495 cbdata
.msg
= start_stats_reply(osr
, 1024);
2496 cls_rule_from_match(&target
, &fsr
->match
, 0);
2497 classifier_for_each_match(&p
->cls
, &target
,
2498 table_id_to_include(fsr
->table_id
),
2499 flow_stats_cb
, &cbdata
);
2500 queue_tx(cbdata
.msg
, ofconn
, ofconn
->reply_counter
);
2504 struct flow_stats_ds_cbdata
{
2505 struct ofproto
*ofproto
;
2510 flow_stats_ds_cb(struct cls_rule
*rule_
, void *cbdata_
)
2512 struct rule
*rule
= rule_from_cls_rule(rule_
);
2513 struct flow_stats_ds_cbdata
*cbdata
= cbdata_
;
2514 struct ds
*results
= cbdata
->results
;
2515 struct ofp_match match
;
2516 uint64_t packet_count
, byte_count
;
2517 size_t act_len
= sizeof *rule
->actions
* rule
->n_actions
;
2519 /* Don't report on subrules. */
2520 if (rule
->super
!= NULL
) {
2524 query_stats(cbdata
->ofproto
, rule
, &packet_count
, &byte_count
);
2525 flow_to_ovs_match(&rule
->cr
.flow
, rule
->cr
.wc
.wildcards
, &match
);
2527 ds_put_format(results
, "duration=%llds, ",
2528 (time_msec() - rule
->created
) / 1000);
2529 ds_put_format(results
, "priority=%u, ", rule
->cr
.priority
);
2530 ds_put_format(results
, "n_packets=%"PRIu64
", ", packet_count
);
2531 ds_put_format(results
, "n_bytes=%"PRIu64
", ", byte_count
);
2532 ofp_print_match(results
, &match
, true);
2533 ofp_print_actions(results
, &rule
->actions
->header
, act_len
);
2534 ds_put_cstr(results
, "\n");
2537 /* Adds a pretty-printed description of all flows to 'results', including
2538 * those marked hidden by secchan (e.g., by in-band control). */
2540 ofproto_get_all_flows(struct ofproto
*p
, struct ds
*results
)
2542 struct ofp_match match
;
2543 struct cls_rule target
;
2544 struct flow_stats_ds_cbdata cbdata
;
2546 memset(&match
, 0, sizeof match
);
2547 match
.wildcards
= htonl(OFPFW_ALL
);
2550 cbdata
.results
= results
;
2552 cls_rule_from_match(&target
, &match
, 0);
2553 classifier_for_each_match(&p
->cls
, &target
, CLS_INC_ALL
,
2554 flow_stats_ds_cb
, &cbdata
);
2557 struct aggregate_stats_cbdata
{
2558 struct ofproto
*ofproto
;
2560 uint64_t packet_count
;
2561 uint64_t byte_count
;
2566 aggregate_stats_cb(struct cls_rule
*rule_
, void *cbdata_
)
2568 struct rule
*rule
= rule_from_cls_rule(rule_
);
2569 struct aggregate_stats_cbdata
*cbdata
= cbdata_
;
2570 uint64_t packet_count
, byte_count
;
2572 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, cbdata
->out_port
)) {
2576 query_stats(cbdata
->ofproto
, rule
, &packet_count
, &byte_count
);
2578 cbdata
->packet_count
+= packet_count
;
2579 cbdata
->byte_count
+= byte_count
;
2584 handle_aggregate_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
2585 const struct ofp_stats_request
*osr
,
2588 struct ofp_aggregate_stats_request
*asr
;
2589 struct ofp_aggregate_stats_reply
*reply
;
2590 struct aggregate_stats_cbdata cbdata
;
2591 struct cls_rule target
;
2594 if (arg_size
!= sizeof *asr
) {
2595 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LENGTH
);
2597 asr
= (struct ofp_aggregate_stats_request
*) osr
->body
;
2599 COVERAGE_INC(ofproto_agg_request
);
2601 cbdata
.out_port
= asr
->out_port
;
2602 cbdata
.packet_count
= 0;
2603 cbdata
.byte_count
= 0;
2605 cls_rule_from_match(&target
, &asr
->match
, 0);
2606 classifier_for_each_match(&p
->cls
, &target
,
2607 table_id_to_include(asr
->table_id
),
2608 aggregate_stats_cb
, &cbdata
);
2610 msg
= start_stats_reply(osr
, sizeof *reply
);
2611 reply
= append_stats_reply(sizeof *reply
, ofconn
, &msg
);
2612 reply
->flow_count
= htonl(cbdata
.n_flows
);
2613 reply
->packet_count
= htonll(cbdata
.packet_count
);
2614 reply
->byte_count
= htonll(cbdata
.byte_count
);
2615 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
2620 handle_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
2621 struct ofp_header
*oh
)
2623 struct ofp_stats_request
*osr
;
2627 error
= check_ofp_message_array(oh
, OFPT_STATS_REQUEST
, sizeof *osr
,
2632 osr
= (struct ofp_stats_request
*) oh
;
2634 switch (ntohs(osr
->type
)) {
2636 return handle_desc_stats_request(p
, ofconn
, osr
);
2639 return handle_flow_stats_request(p
, ofconn
, osr
, arg_size
);
2641 case OFPST_AGGREGATE
:
2642 return handle_aggregate_stats_request(p
, ofconn
, osr
, arg_size
);
2645 return handle_table_stats_request(p
, ofconn
, osr
);
2648 return handle_port_stats_request(p
, ofconn
, osr
);
2651 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_VENDOR
);
2654 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_STAT
);
2658 static long long int
2659 msec_from_nsec(uint64_t sec
, uint32_t nsec
)
2661 return !sec
? 0 : sec
* 1000 + nsec
/ 1000000;
2665 update_time(struct rule
*rule
, const struct odp_flow_stats
*stats
)
2667 long long int used
= msec_from_nsec(stats
->used_sec
, stats
->used_nsec
);
2668 if (used
> rule
->used
) {
2674 update_stats(struct rule
*rule
, const struct odp_flow_stats
*stats
)
2676 update_time(rule
, stats
);
2677 rule
->packet_count
+= stats
->n_packets
;
2678 rule
->byte_count
+= stats
->n_bytes
;
2679 rule
->tcp_flags
|= stats
->tcp_flags
;
2680 if (stats
->n_packets
) {
2681 rule
->ip_tos
= stats
->ip_tos
;
2686 add_flow(struct ofproto
*p
, struct ofconn
*ofconn
,
2687 struct ofp_flow_mod
*ofm
, size_t n_actions
)
2689 struct ofpbuf
*packet
;
2694 rule
= rule_create(NULL
, (const union ofp_action
*) ofm
->actions
,
2695 n_actions
, ntohs(ofm
->idle_timeout
),
2696 ntohs(ofm
->hard_timeout
));
2697 cls_rule_from_match(&rule
->cr
, &ofm
->match
, ntohs(ofm
->priority
));
2701 if (ofm
->buffer_id
!= htonl(UINT32_MAX
)) {
2702 error
= pktbuf_retrieve(ofconn
->pktbuf
, ntohl(ofm
->buffer_id
),
2706 rule_insert(p
, rule
, packet
, in_port
);
2707 ofpbuf_delete(packet
);
2712 modify_flow(struct ofproto
*p
, const struct ofp_flow_mod
*ofm
,
2713 size_t n_actions
, uint16_t command
, struct rule
*rule
)
2715 if (rule_is_hidden(rule
)) {
2719 if (command
== OFPFC_DELETE
) {
2720 rule_remove(p
, rule
);
2722 size_t actions_len
= n_actions
* sizeof *rule
->actions
;
2724 if (n_actions
== rule
->n_actions
2725 && !memcmp(ofm
->actions
, rule
->actions
, actions_len
))
2730 free(rule
->actions
);
2731 rule
->actions
= xmemdup(ofm
->actions
, actions_len
);
2732 rule
->n_actions
= n_actions
;
2734 if (rule
->cr
.wc
.wildcards
) {
2735 COVERAGE_INC(ofproto_mod_wc_flow
);
2736 p
->need_revalidate
= true;
2738 rule_update_actions(p
, rule
);
2746 modify_flows_strict(struct ofproto
*p
, const struct ofp_flow_mod
*ofm
,
2747 size_t n_actions
, uint16_t command
)
2753 flow_from_match(&flow
, &wildcards
, &ofm
->match
);
2754 rule
= rule_from_cls_rule(classifier_find_rule_exactly(
2755 &p
->cls
, &flow
, wildcards
,
2756 ntohs(ofm
->priority
)));
2759 if (command
== OFPFC_DELETE
2760 && ofm
->out_port
!= htons(OFPP_NONE
)
2761 && !rule_has_out_port(rule
, ofm
->out_port
)) {
2765 modify_flow(p
, ofm
, n_actions
, command
, rule
);
2770 struct modify_flows_cbdata
{
2771 struct ofproto
*ofproto
;
2772 const struct ofp_flow_mod
*ofm
;
2779 modify_flows_cb(struct cls_rule
*rule_
, void *cbdata_
)
2781 struct rule
*rule
= rule_from_cls_rule(rule_
);
2782 struct modify_flows_cbdata
*cbdata
= cbdata_
;
2784 if (cbdata
->out_port
!= htons(OFPP_NONE
)
2785 && !rule_has_out_port(rule
, cbdata
->out_port
)) {
2789 modify_flow(cbdata
->ofproto
, cbdata
->ofm
, cbdata
->n_actions
,
2790 cbdata
->command
, rule
);
2794 modify_flows_loose(struct ofproto
*p
, const struct ofp_flow_mod
*ofm
,
2795 size_t n_actions
, uint16_t command
)
2797 struct modify_flows_cbdata cbdata
;
2798 struct cls_rule target
;
2802 cbdata
.out_port
= (command
== OFPFC_DELETE
? ofm
->out_port
2803 : htons(OFPP_NONE
));
2804 cbdata
.n_actions
= n_actions
;
2805 cbdata
.command
= command
;
2807 cls_rule_from_match(&target
, &ofm
->match
, 0);
2809 classifier_for_each_match(&p
->cls
, &target
, CLS_INC_ALL
,
2810 modify_flows_cb
, &cbdata
);
2815 handle_flow_mod(struct ofproto
*p
, struct ofconn
*ofconn
,
2816 struct ofp_flow_mod
*ofm
)
2821 error
= check_ofp_message_array(&ofm
->header
, OFPT_FLOW_MOD
, sizeof *ofm
,
2822 sizeof *ofm
->actions
, &n_actions
);
2827 normalize_match(&ofm
->match
);
2828 if (!ofm
->match
.wildcards
) {
2829 ofm
->priority
= htons(UINT16_MAX
);
2832 error
= validate_actions((const union ofp_action
*) ofm
->actions
,
2833 n_actions
, p
->max_ports
);
2838 switch (ntohs(ofm
->command
)) {
2840 return add_flow(p
, ofconn
, ofm
, n_actions
);
2843 return modify_flows_loose(p
, ofm
, n_actions
, OFPFC_MODIFY
);
2845 case OFPFC_MODIFY_STRICT
:
2846 return modify_flows_strict(p
, ofm
, n_actions
, OFPFC_MODIFY
);
2849 return modify_flows_loose(p
, ofm
, n_actions
, OFPFC_DELETE
);
2851 case OFPFC_DELETE_STRICT
:
2852 return modify_flows_strict(p
, ofm
, n_actions
, OFPFC_DELETE
);
2855 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_BAD_COMMAND
);
2860 send_capability_reply(struct ofproto
*p
, struct ofconn
*ofconn
, uint32_t xid
)
2862 struct ofmp_capability_reply
*ocr
;
2864 char capabilities
[] = "com.nicira.mgmt.manager=false\n";
2866 ocr
= make_openflow_xid(sizeof(*ocr
), OFPT_VENDOR
, xid
, &b
);
2867 ocr
->header
.header
.vendor
= htonl(NX_VENDOR_ID
);
2868 ocr
->header
.header
.subtype
= htonl(NXT_MGMT
);
2869 ocr
->header
.type
= htons(OFMPT_CAPABILITY_REPLY
);
2871 ocr
->format
= htonl(OFMPCOF_SIMPLE
);
2872 ocr
->mgmt_id
= htonll(p
->mgmt_id
);
2874 ofpbuf_put(b
, capabilities
, strlen(capabilities
));
2876 queue_tx(b
, ofconn
, ofconn
->reply_counter
);
2880 handle_ofmp(struct ofproto
*p
, struct ofconn
*ofconn
,
2881 struct ofmp_header
*ofmph
)
2883 size_t msg_len
= ntohs(ofmph
->header
.header
.length
);
2884 if (msg_len
< sizeof(*ofmph
)) {
2885 VLOG_WARN_RL(&rl
, "dropping short managment message: %d\n", msg_len
);
2886 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LENGTH
);
2889 if (ofmph
->type
== htons(OFMPT_CAPABILITY_REQUEST
)) {
2890 struct ofmp_capability_request
*ofmpcr
;
2892 if (msg_len
< sizeof(struct ofmp_capability_request
)) {
2893 VLOG_WARN_RL(&rl
, "dropping short capability request: %d\n",
2895 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LENGTH
);
2898 ofmpcr
= (struct ofmp_capability_request
*)ofmph
;
2899 if (ofmpcr
->format
!= htonl(OFMPCAF_SIMPLE
)) {
2900 /* xxx Find a better type than bad subtype */
2901 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_SUBTYPE
);
2904 send_capability_reply(p
, ofconn
, ofmph
->header
.header
.xid
);
2907 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_SUBTYPE
);
2912 handle_vendor(struct ofproto
*p
, struct ofconn
*ofconn
, void *msg
)
2914 struct ofp_vendor_header
*ovh
= msg
;
2915 struct nicira_header
*nh
;
2917 if (ntohs(ovh
->header
.length
) < sizeof(struct ofp_vendor_header
)) {
2918 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LENGTH
);
2920 if (ovh
->vendor
!= htonl(NX_VENDOR_ID
)) {
2921 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_VENDOR
);
2923 if (ntohs(ovh
->header
.length
) < sizeof(struct nicira_header
)) {
2924 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LENGTH
);
2928 switch (ntohl(nh
->subtype
)) {
2929 case NXT_STATUS_REQUEST
:
2930 return switch_status_handle_request(p
->switch_status
, ofconn
->rconn
,
2933 case NXT_ACT_SET_CONFIG
:
2934 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_SUBTYPE
); /* XXX */
2936 case NXT_ACT_GET_CONFIG
:
2937 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_SUBTYPE
); /* XXX */
2939 case NXT_COMMAND_REQUEST
:
2941 return executer_handle_request(p
->executer
, ofconn
->rconn
, msg
);
2946 return handle_ofmp(p
, ofconn
, msg
);
2949 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_SUBTYPE
);
2953 handle_openflow(struct ofconn
*ofconn
, struct ofproto
*p
,
2954 struct ofpbuf
*ofp_msg
)
2956 struct ofp_header
*oh
= ofp_msg
->data
;
2959 COVERAGE_INC(ofproto_recv_openflow
);
2961 case OFPT_ECHO_REQUEST
:
2962 error
= handle_echo_request(ofconn
, oh
);
2965 case OFPT_ECHO_REPLY
:
2969 case OFPT_FEATURES_REQUEST
:
2970 error
= handle_features_request(p
, ofconn
, oh
);
2973 case OFPT_GET_CONFIG_REQUEST
:
2974 error
= handle_get_config_request(p
, ofconn
, oh
);
2977 case OFPT_SET_CONFIG
:
2978 error
= handle_set_config(p
, ofconn
, ofp_msg
->data
);
2981 case OFPT_PACKET_OUT
:
2982 error
= handle_packet_out(p
, ofconn
, ofp_msg
->data
);
2986 error
= handle_port_mod(p
, oh
);
2990 error
= handle_flow_mod(p
, ofconn
, ofp_msg
->data
);
2993 case OFPT_STATS_REQUEST
:
2994 error
= handle_stats_request(p
, ofconn
, oh
);
2998 error
= handle_vendor(p
, ofconn
, ofp_msg
->data
);
3002 if (VLOG_IS_WARN_ENABLED()) {
3003 char *s
= ofp_to_string(oh
, ntohs(oh
->length
), 2);
3004 VLOG_DBG_RL(&rl
, "OpenFlow message ignored: %s", s
);
3007 error
= ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_TYPE
);
3012 send_error_oh(ofconn
, ofp_msg
->data
, error
);
3017 handle_odp_msg(struct ofproto
*p
, struct ofpbuf
*packet
)
3019 struct odp_msg
*msg
= packet
->data
;
3020 uint16_t in_port
= odp_port_to_ofp_port(msg
->port
);
3022 struct ofpbuf payload
;
3025 /* Handle controller actions. */
3026 if (msg
->type
== _ODPL_ACTION_NR
) {
3027 COVERAGE_INC(ofproto_ctlr_action
);
3028 pinsched_send(p
->action_sched
, in_port
, packet
,
3029 send_packet_in_action
, p
);
3033 payload
.data
= msg
+ 1;
3034 payload
.size
= msg
->length
- sizeof *msg
;
3035 flow_extract(&payload
, msg
->port
, &flow
);
3037 /* Check with in-band control to see if this packet should be sent
3038 * to the local port regardless of the flow table. */
3039 if (in_band_msg_in_hook(p
->in_band
, &flow
, &payload
)) {
3040 union odp_action action
;
3042 memset(&action
, 0, sizeof(action
));
3043 action
.output
.type
= ODPAT_OUTPUT
;
3044 action
.output
.port
= ODPP_LOCAL
;
3045 dpif_execute(p
->dpif
, flow
.in_port
, &action
, 1, &payload
);
3048 rule
= lookup_valid_rule(p
, &flow
);
3050 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3051 struct ofport
*port
= port_array_get(&p
->ports
, msg
->port
);
3053 if (port
->opp
.config
& OFPPC_NO_PACKET_IN
) {
3054 COVERAGE_INC(ofproto_no_packet_in
);
3055 /* XXX install 'drop' flow entry */
3056 ofpbuf_delete(packet
);
3060 VLOG_WARN_RL(&rl
, "packet-in on unknown port %"PRIu16
, msg
->port
);
3063 COVERAGE_INC(ofproto_packet_in
);
3064 pinsched_send(p
->miss_sched
, in_port
, packet
, send_packet_in_miss
, p
);
3068 if (rule
->cr
.wc
.wildcards
) {
3069 rule
= rule_create_subrule(p
, rule
, &flow
);
3070 rule_make_actions(p
, rule
, packet
);
3072 if (!rule
->may_install
) {
3073 /* The rule is not installable, that is, we need to process every
3074 * packet, so process the current packet and set its actions into
3076 rule_make_actions(p
, rule
, packet
);
3078 /* XXX revalidate rule if it needs it */
3082 rule_execute(p
, rule
, &payload
, &flow
);
3083 rule_reinstall(p
, rule
);
3084 ofpbuf_delete(packet
);
3088 revalidate_cb(struct cls_rule
*sub_
, void *cbdata_
)
3090 struct rule
*sub
= rule_from_cls_rule(sub_
);
3091 struct revalidate_cbdata
*cbdata
= cbdata_
;
3093 if (cbdata
->revalidate_all
3094 || (cbdata
->revalidate_subrules
&& sub
->super
)
3095 || (tag_set_intersects(&cbdata
->revalidate_set
, sub
->tags
))) {
3096 revalidate_rule(cbdata
->ofproto
, sub
);
3101 revalidate_rule(struct ofproto
*p
, struct rule
*rule
)
3103 const flow_t
*flow
= &rule
->cr
.flow
;
3105 COVERAGE_INC(ofproto_revalidate_rule
);
3108 super
= rule_from_cls_rule(classifier_lookup_wild(&p
->cls
, flow
));
3110 rule_remove(p
, rule
);
3112 } else if (super
!= rule
->super
) {
3113 COVERAGE_INC(ofproto_revalidate_moved
);
3114 list_remove(&rule
->list
);
3115 list_push_back(&super
->list
, &rule
->list
);
3116 rule
->super
= super
;
3117 rule
->hard_timeout
= super
->hard_timeout
;
3118 rule
->idle_timeout
= super
->idle_timeout
;
3119 rule
->created
= super
->created
;
3124 rule_update_actions(p
, rule
);
3128 static struct ofpbuf
*
3129 compose_flow_exp(const struct rule
*rule
, long long int now
, uint8_t reason
)
3131 struct ofp_flow_expired
*ofe
;
3134 ofe
= make_openflow(sizeof *ofe
, OFPT_FLOW_EXPIRED
, &buf
);
3135 flow_to_match(&rule
->cr
.flow
, rule
->cr
.wc
.wildcards
, &ofe
->match
);
3136 ofe
->priority
= htons(rule
->cr
.priority
);
3137 ofe
->reason
= reason
;
3138 ofe
->duration
= (now
- rule
->created
) / 1000;
3139 ofe
->packet_count
= rule
->packet_count
;
3140 ofe
->byte_count
= rule
->byte_count
;
3146 send_flow_exp(struct ofproto
*p
, struct rule
*rule
,
3147 long long int now
, uint8_t reason
)
3149 struct ofconn
*ofconn
;
3150 struct ofconn
*prev
;
3151 struct ofpbuf
*buf
= NULL
;
3153 /* We limit the maximum number of queued flow expirations it by accounting
3154 * them under the counter for replies. That works because preventing
3155 * OpenFlow requests from being processed also prevents new flows from
3156 * being added (and expiring). (It also prevents processing OpenFlow
3157 * requests that would not add new flows, so it is imperfect.) */
3160 LIST_FOR_EACH (ofconn
, struct ofconn
, node
, &p
->all_conns
) {
3161 if (ofconn
->send_flow_exp
&& rconn_is_connected(ofconn
->rconn
)) {
3163 queue_tx(ofpbuf_clone(buf
), prev
, ofconn
->reply_counter
);
3165 buf
= compose_flow_exp(rule
, now
, reason
);
3171 queue_tx(buf
, prev
, ofconn
->reply_counter
);
3176 uninstall_idle_flow(struct ofproto
*ofproto
, struct rule
*rule
)
3178 assert(rule
->installed
);
3179 assert(!rule
->cr
.wc
.wildcards
);
3182 rule_remove(ofproto
, rule
);
3184 rule_uninstall(ofproto
, rule
);
3189 expire_rule(struct cls_rule
*cls_rule
, void *p_
)
3191 struct ofproto
*p
= p_
;
3192 struct rule
*rule
= rule_from_cls_rule(cls_rule
);
3193 long long int hard_expire
, idle_expire
, expire
, now
;
3195 hard_expire
= (rule
->hard_timeout
3196 ? rule
->created
+ rule
->hard_timeout
* 1000
3198 idle_expire
= (rule
->idle_timeout
3199 && (rule
->super
|| list_is_empty(&rule
->list
))
3200 ? rule
->used
+ rule
->idle_timeout
* 1000
3202 expire
= MIN(hard_expire
, idle_expire
);
3203 if (expire
== LLONG_MAX
) {
3204 if (rule
->installed
&& time_msec() >= rule
->used
+ 5000) {
3205 uninstall_idle_flow(p
, rule
);
3212 if (rule
->installed
&& now
>= rule
->used
+ 5000) {
3213 uninstall_idle_flow(p
, rule
);
3218 COVERAGE_INC(ofproto_expired
);
3219 if (rule
->cr
.wc
.wildcards
) {
3220 /* Update stats. (This code will be a no-op if the rule expired
3221 * due to an idle timeout, because in that case the rule has no
3222 * subrules left.) */
3223 struct rule
*subrule
, *next
;
3224 LIST_FOR_EACH_SAFE (subrule
, next
, struct rule
, list
, &rule
->list
) {
3225 rule_remove(p
, subrule
);
3229 send_flow_exp(p
, rule
, now
,
3231 ? OFPER_HARD_TIMEOUT
: OFPER_IDLE_TIMEOUT
));
3232 rule_remove(p
, rule
);
3236 update_used(struct ofproto
*p
)
3238 struct odp_flow
*flows
;
3243 error
= dpif_flow_list_all(p
->dpif
, &flows
, &n_flows
);
3248 for (i
= 0; i
< n_flows
; i
++) {
3249 struct odp_flow
*f
= &flows
[i
];
3252 rule
= rule_from_cls_rule(
3253 classifier_find_rule_exactly(&p
->cls
, &f
->key
, 0, UINT16_MAX
));
3254 if (!rule
|| !rule
->installed
) {
3255 COVERAGE_INC(ofproto_unexpected_rule
);
3256 dpif_flow_del(p
->dpif
, f
);
3260 update_time(rule
, &f
->stats
);
3261 rule_account(p
, rule
, f
->stats
.n_bytes
);
3267 do_send_packet_in(struct ofconn
*ofconn
, uint32_t buffer_id
,
3268 const struct ofpbuf
*packet
, int send_len
)
3270 struct ofp_packet_in
*opi
;
3271 struct ofpbuf payload
, *buf
;
3272 struct odp_msg
*msg
;
3275 payload
.data
= msg
+ 1;
3276 payload
.size
= msg
->length
- sizeof *msg
;
3278 send_len
= MIN(send_len
, payload
.size
);
3279 buf
= ofpbuf_new(sizeof *opi
+ send_len
);
3280 opi
= put_openflow_xid(offsetof(struct ofp_packet_in
, data
),
3281 OFPT_PACKET_IN
, 0, buf
);
3282 opi
->buffer_id
= htonl(buffer_id
);
3283 opi
->total_len
= htons(payload
.size
);
3284 opi
->in_port
= htons(odp_port_to_ofp_port(msg
->port
));
3285 opi
->reason
= msg
->type
== _ODPL_ACTION_NR
? OFPR_ACTION
: OFPR_NO_MATCH
;
3286 ofpbuf_put(buf
, payload
.data
, MIN(send_len
, payload
.size
));
3287 update_openflow_length(buf
);
3288 rconn_send_with_limit(ofconn
->rconn
, buf
, ofconn
->packet_in_counter
, 100);
3292 send_packet_in_action(struct ofpbuf
*packet
, void *p_
)
3294 struct ofproto
*p
= p_
;
3295 struct ofconn
*ofconn
;
3296 struct odp_msg
*msg
;
3299 LIST_FOR_EACH (ofconn
, struct ofconn
, node
, &p
->all_conns
) {
3300 if (ofconn
== p
->controller
|| ofconn
->miss_send_len
) {
3301 do_send_packet_in(ofconn
, UINT32_MAX
, packet
, msg
->arg
);
3304 ofpbuf_delete(packet
);
3308 send_packet_in_miss(struct ofpbuf
*packet
, void *p_
)
3310 struct ofproto
*p
= p_
;
3311 struct ofconn
*ofconn
;
3312 struct ofpbuf payload
;
3313 struct odp_msg
*msg
;
3316 payload
.data
= msg
+ 1;
3317 payload
.size
= msg
->length
- sizeof *msg
;
3318 LIST_FOR_EACH (ofconn
, struct ofconn
, node
, &p
->all_conns
) {
3319 if (ofconn
->miss_send_len
) {
3320 uint32_t buffer_id
= pktbuf_save(ofconn
->pktbuf
, &payload
,
3322 int send_len
= (buffer_id
!= UINT32_MAX
? ofconn
->miss_send_len
3324 do_send_packet_in(ofconn
, buffer_id
, packet
, send_len
);
3327 ofpbuf_delete(packet
);
3331 pick_datapath_id(const struct ofproto
*ofproto
)
3333 const struct ofport
*port
;
3335 port
= port_array_get(&ofproto
->ports
, ODPP_LOCAL
);
3337 uint8_t ea
[ETH_ADDR_LEN
];
3340 error
= netdev_get_etheraddr(port
->netdev
, ea
);
3342 return eth_addr_to_uint64(ea
);
3344 VLOG_WARN("could not get MAC address for %s (%s)",
3345 netdev_get_name(port
->netdev
), strerror(error
));
3347 return ofproto
->fallback_dpid
;
3351 pick_fallback_dpid(void)
3353 uint8_t ea
[ETH_ADDR_LEN
];
3354 eth_addr_random(ea
);
3355 ea
[0] = 0x00; /* Set Nicira OUI. */
3358 return eth_addr_to_uint64(ea
);
3362 default_normal_ofhook_cb(const flow_t
*flow
, const struct ofpbuf
*packet
,
3363 struct odp_actions
*actions
, tag_type
*tags
,
3366 struct ofproto
*ofproto
= ofproto_
;
3369 /* Drop frames for reserved multicast addresses. */
3370 if (eth_addr_is_reserved(flow
->dl_dst
)) {
3374 /* Learn source MAC (but don't try to learn from revalidation). */
3375 if (packet
!= NULL
) {
3376 tag_type rev_tag
= mac_learning_learn(ofproto
->ml
, flow
->dl_src
,
3379 /* The log messages here could actually be useful in debugging,
3380 * so keep the rate limit relatively high. */
3381 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(30, 300);
3382 VLOG_DBG_RL(&rl
, "learned that "ETH_ADDR_FMT
" is on port %"PRIu16
,
3383 ETH_ADDR_ARGS(flow
->dl_src
), flow
->in_port
);
3384 ofproto_revalidate(ofproto
, rev_tag
);
3388 /* Determine output port. */
3389 out_port
= mac_learning_lookup_tag(ofproto
->ml
, flow
->dl_dst
, 0, tags
);
3391 add_output_group_action(actions
, DP_GROUP_FLOOD
);
3392 } else if (out_port
!= flow
->in_port
) {
3393 odp_actions_add(actions
, ODPAT_OUTPUT
)->output
.port
= out_port
;
3401 static const struct ofhooks default_ofhooks
= {
3403 default_normal_ofhook_cb
,