3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
22 #include "byte-order.h"
23 #include "connectivity.h"
28 #include "fail-open.h"
29 #include "guarded-list.h"
33 #include "mac-learning.h"
35 #include "mcast-snooping.h"
36 #include "multipath.h"
37 #include "netdev-vport.h"
42 #include "odp-execute.h"
43 #include "ofproto/ofproto-dpif.h"
44 #include "ofproto/ofproto-provider.h"
45 #include "ofproto-dpif-ipfix.h"
46 #include "ofproto-dpif-mirror.h"
47 #include "ofproto-dpif-monitor.h"
48 #include "ofproto-dpif-rid.h"
49 #include "ofproto-dpif-sflow.h"
50 #include "ofproto-dpif-trace.h"
51 #include "ofproto-dpif-upcall.h"
52 #include "ofproto-dpif-xlate.h"
53 #include "ofproto-dpif-xlate-cache.h"
54 #include "openvswitch/ofp-actions.h"
55 #include "openvswitch/dynamic-string.h"
56 #include "openvswitch/meta-flow.h"
57 #include "openvswitch/ofp-print.h"
58 #include "openvswitch/ofpbuf.h"
59 #include "openvswitch/uuid.h"
60 #include "openvswitch/vlog.h"
63 #include "ovs-router.h"
64 #include "openvswitch/poll-loop.h"
70 #include "unaligned.h"
74 #include "vlan-bitmap.h"
76 VLOG_DEFINE_THIS_MODULE(ofproto_dpif
);
78 COVERAGE_DEFINE(ofproto_dpif_expired
);
79 COVERAGE_DEFINE(packet_in_overflow
);
83 static void rule_get_stats(struct rule
*, struct pkt_stats
*stats
,
85 static struct rule_dpif
*rule_dpif_cast(const struct rule
*);
86 static void rule_expire(struct rule_dpif
*, long long now
);
89 struct hmap_node hmap_node
; /* In struct ofproto's "bundles" hmap. */
90 struct ofproto_dpif
*ofproto
; /* Owning ofproto. */
91 void *aux
; /* Key supplied by ofproto's client. */
92 char *name
; /* Identifier for log messages. */
95 struct ovs_list ports
; /* Contains "struct ofport_dpif"s. */
96 enum port_vlan_mode vlan_mode
; /* VLAN mode */
97 uint16_t qinq_ethtype
;
98 int vlan
; /* -1=trunk port, else a 12-bit VLAN ID. */
99 unsigned long *trunks
; /* Bitmap of trunked VLANs, if 'vlan' == -1.
100 * NULL if all VLANs are trunked. */
101 unsigned long *cvlans
;
102 struct lacp
*lacp
; /* LACP if LACP is enabled, otherwise NULL. */
103 struct bond
*bond
; /* Nonnull iff more than one port. */
104 enum port_priority_tags_mode use_priority_tags
;
105 /* Use 802.1p tag for frames in VLAN 0? */
107 bool protected; /* Protected port mode */
110 bool floodable
; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
113 static void bundle_remove(struct ofport
*);
114 static void bundle_update(struct ofbundle
*);
115 static void bundle_destroy(struct ofbundle
*);
116 static void bundle_del_port(struct ofport_dpif
*);
117 static void bundle_run(struct ofbundle
*);
118 static void bundle_wait(struct ofbundle
*);
119 static void bundle_flush_macs(struct ofbundle
*, bool);
120 static void bundle_move(struct ofbundle
*, struct ofbundle
*);
122 static void stp_run(struct ofproto_dpif
*ofproto
);
123 static void stp_wait(struct ofproto_dpif
*ofproto
);
124 static int set_stp_port(struct ofport
*,
125 const struct ofproto_port_stp_settings
*);
127 static void rstp_run(struct ofproto_dpif
*ofproto
);
128 static void set_rstp_port(struct ofport
*,
129 const struct ofproto_port_rstp_settings
*);
132 struct hmap_node odp_port_node
; /* In dpif_backer's "odp_to_ofport_map". */
136 struct ofbundle
*bundle
; /* Bundle that contains this port, if any. */
137 struct ovs_list bundle_node
;/* In struct ofbundle's "ports" list. */
138 struct cfm
*cfm
; /* Connectivity Fault Management, if any. */
139 struct bfd
*bfd
; /* BFD, if any. */
140 struct lldp
*lldp
; /* lldp, if any. */
141 bool is_tunnel
; /* This port is a tunnel. */
142 long long int carrier_seq
; /* Carrier status changes. */
143 struct ofport_dpif
*peer
; /* Peer if patch port. */
146 struct stp_port
*stp_port
; /* Spanning Tree Protocol, if any. */
147 enum stp_state stp_state
; /* Always STP_DISABLED if STP not in use. */
148 long long int stp_state_entered
;
150 /* Rapid Spanning Tree. */
151 struct rstp_port
*rstp_port
; /* Rapid Spanning Tree Protocol, if any. */
152 enum rstp_state rstp_state
; /* Always RSTP_DISABLED if RSTP not in use. */
154 /* Queue to DSCP mapping. */
155 struct ofproto_port_queue
*qdscp
;
159 struct ct_timeout_policy
{
160 int ref_count
; /* The number of ct zones that use this
162 uint32_t tp_id
; /* Timeout policy id in the datapath. */
163 struct simap tp
; /* A map from timeout policy attribute to
165 struct hmap_node node
; /* Element in struct dpif_backer's "ct_tps"
167 struct ovs_list list_node
; /* Element in struct dpif_backer's
168 * "ct_tp_kill_list" list. */
171 /* Periodically try to purge deleted timeout policies from the datapath. Retry
172 * may be necessary if the kernel datapath has a non-zero datapath flow
173 * reference count for the timeout policy. */
174 #define TIMEOUT_POLICY_CLEANUP_INTERVAL (20000) /* 20 seconds. */
175 static long long int timeout_policy_cleanup_timer
= LLONG_MIN
;
179 struct ct_timeout_policy
*ct_tp
;
180 struct cmap_node node
; /* Element in struct dpif_backer's
181 * "ct_zones" cmap. */
184 static odp_port_t
ofp_port_to_odp_port(const struct ofproto_dpif
*,
187 static ofp_port_t
odp_port_to_ofp_port(const struct ofproto_dpif
*,
190 static struct ofport_dpif
*
191 ofport_dpif_cast(const struct ofport
*ofport
)
193 return ofport
? CONTAINER_OF(ofport
, struct ofport_dpif
, up
) : NULL
;
196 static void port_run(struct ofport_dpif
*);
197 static int set_bfd(struct ofport
*, const struct smap
*);
198 static int set_cfm(struct ofport
*, const struct cfm_settings
*);
199 static int set_lldp(struct ofport
*ofport_
, const struct smap
*cfg
);
200 static void ofport_update_peer(struct ofport_dpif
*);
202 COVERAGE_DEFINE(rev_reconfigure
);
203 COVERAGE_DEFINE(rev_stp
);
204 COVERAGE_DEFINE(rev_rstp
);
205 COVERAGE_DEFINE(rev_bond
);
206 COVERAGE_DEFINE(rev_port_toggled
);
207 COVERAGE_DEFINE(rev_flow_table
);
208 COVERAGE_DEFINE(rev_mac_learning
);
209 COVERAGE_DEFINE(rev_mcast_snooping
);
211 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
212 struct shash all_dpif_backers
= SHASH_INITIALIZER(&all_dpif_backers
);
214 /* All existing ofproto_dpif instances, indexed by ->up.name. */
215 static struct hmap all_ofproto_dpifs_by_name
=
216 HMAP_INITIALIZER(&all_ofproto_dpifs_by_name
);
218 /* All existing ofproto_dpif instances, indexed by ->uuid. */
219 static struct hmap all_ofproto_dpifs_by_uuid
=
220 HMAP_INITIALIZER(&all_ofproto_dpifs_by_uuid
);
222 static bool ofproto_use_tnl_push_pop
= true;
223 static void ofproto_unixctl_init(void);
224 static void ct_zone_config_init(struct dpif_backer
*backer
);
225 static void ct_zone_config_uninit(struct dpif_backer
*backer
);
226 static void ct_zone_timeout_policy_sweep(struct dpif_backer
*backer
);
228 static inline struct ofproto_dpif
*
229 ofproto_dpif_cast(const struct ofproto
*ofproto
)
231 ovs_assert(ofproto
->ofproto_class
== &ofproto_dpif_class
);
232 return CONTAINER_OF(ofproto
, struct ofproto_dpif
, up
);
235 /* Global variables. */
236 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
238 /* Initial mappings of port to bridge mappings. */
239 static struct shash init_ofp_ports
= SHASH_INITIALIZER(&init_ofp_ports
);
241 /* Initialize 'ofm' for a learn action. If the rule already existed, reference
242 * to that rule is taken, otherwise a new rule is created. 'ofm' keeps the
243 * rule reference in both cases. */
245 ofproto_dpif_flow_mod_init_for_learn(struct ofproto_dpif
*ofproto
,
246 const struct ofputil_flow_mod
*fm
,
247 struct ofproto_flow_mod
*ofm
)
249 /* This will not take the global 'ofproto_mutex'. */
250 return ofproto_flow_mod_init_for_learn(&ofproto
->up
, fm
, ofm
);
253 /* Appends 'am' to the queue of asynchronous messages to be sent to the
254 * controller. Takes ownership of 'am' and any data it points to. */
256 ofproto_dpif_send_async_msg(struct ofproto_dpif
*ofproto
,
257 struct ofproto_async_msg
*am
)
259 if (!guarded_list_push_back(&ofproto
->ams
, &am
->list_node
, 1024)) {
260 COVERAGE_INC(packet_in_overflow
);
261 ofproto_async_msg_free(am
);
264 /* Wakes up main thread for packet-in I/O. */
265 seq_change(ofproto
->ams_seq
);
268 /* Factory functions. */
271 init(const struct shash
*iface_hints
)
273 struct shash_node
*node
;
275 /* Make a local copy, since we don't own 'iface_hints' elements. */
276 SHASH_FOR_EACH(node
, iface_hints
) {
277 const struct iface_hint
*orig_hint
= node
->data
;
278 struct iface_hint
*new_hint
= xmalloc(sizeof *new_hint
);
280 new_hint
->br_name
= xstrdup(orig_hint
->br_name
);
281 new_hint
->br_type
= xstrdup(orig_hint
->br_type
);
282 new_hint
->ofp_port
= orig_hint
->ofp_port
;
284 shash_add(&init_ofp_ports
, node
->name
, new_hint
);
287 ofproto_unixctl_init();
288 ofproto_dpif_trace_init();
293 enumerate_types(struct sset
*types
)
295 dp_enumerate_types(types
);
299 enumerate_names(const char *type
, struct sset
*names
)
301 struct ofproto_dpif
*ofproto
;
304 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_by_name_node
,
305 &all_ofproto_dpifs_by_name
) {
306 if (strcmp(type
, ofproto
->up
.type
)) {
309 sset_add(names
, ofproto
->up
.name
);
316 del(const char *type
, const char *name
)
321 error
= dpif_open(name
, type
, &dpif
);
323 error
= dpif_delete(dpif
);
330 port_open_type(const char *datapath_type
, const char *port_type
)
332 return dpif_port_open_type(datapath_type
, port_type
);
335 /* Type functions. */
337 static void process_dpif_port_changes(struct dpif_backer
*);
338 static void process_dpif_all_ports_changed(struct dpif_backer
*);
339 static void process_dpif_port_change(struct dpif_backer
*,
340 const char *devname
);
341 static void process_dpif_port_error(struct dpif_backer
*, int error
);
343 static struct ofproto_dpif
*
344 lookup_ofproto_dpif_by_port_name(const char *name
)
346 struct ofproto_dpif
*ofproto
;
348 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_by_name_node
,
349 &all_ofproto_dpifs_by_name
) {
350 if (sset_contains(&ofproto
->ports
, name
)) {
359 type_run(const char *type
)
361 struct dpif_backer
*backer
;
363 backer
= shash_find_data(&all_dpif_backers
, type
);
365 /* This is not necessarily a problem, since backers are only
366 * created on demand. */
370 if (dpif_run(backer
->dpif
)) {
371 backer
->need_revalidate
= REV_RECONFIGURE
;
374 udpif_run(backer
->udpif
);
376 /* If vswitchd started with other_config:flow_restore_wait set as "true",
377 * and the configuration has now changed to "false", enable receiving
378 * packets from the datapath. */
379 if (!backer
->recv_set_enable
&& !ofproto_get_flow_restore_wait()) {
382 backer
->recv_set_enable
= true;
384 error
= dpif_recv_set(backer
->dpif
, backer
->recv_set_enable
);
386 VLOG_ERR("Failed to enable receiving packets in dpif.");
389 dpif_flow_flush(backer
->dpif
);
390 backer
->need_revalidate
= REV_RECONFIGURE
;
393 if (backer
->recv_set_enable
) {
394 udpif_set_threads(backer
->udpif
, n_handlers
, n_revalidators
);
397 if (backer
->need_revalidate
) {
398 struct ofproto_dpif
*ofproto
;
399 struct simap_node
*node
;
400 struct simap tmp_backers
;
402 /* Handle tunnel garbage collection. */
403 simap_init(&tmp_backers
);
404 simap_swap(&backer
->tnl_backers
, &tmp_backers
);
406 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_by_name_node
,
407 &all_ofproto_dpifs_by_name
) {
408 struct ofport_dpif
*iter
;
410 if (backer
!= ofproto
->backer
) {
414 HMAP_FOR_EACH (iter
, up
.hmap_node
, &ofproto
->up
.ports
) {
415 char namebuf
[NETDEV_VPORT_NAME_BUFSIZE
];
417 odp_port_t old_odp_port
;
419 if (!iter
->is_tunnel
) {
423 dp_port
= netdev_vport_get_dpif_port(iter
->up
.netdev
,
424 namebuf
, sizeof namebuf
);
425 old_odp_port
= iter
->odp_port
;
426 node
= simap_find(&tmp_backers
, dp_port
);
428 simap_put(&backer
->tnl_backers
, dp_port
, node
->data
);
429 simap_delete(&tmp_backers
, node
);
430 node
= simap_find(&backer
->tnl_backers
, dp_port
);
432 node
= simap_find(&backer
->tnl_backers
, dp_port
);
434 odp_port_t odp_port
= ODPP_NONE
;
436 if (!dpif_port_add(backer
->dpif
, iter
->up
.netdev
,
438 simap_put(&backer
->tnl_backers
, dp_port
,
439 odp_to_u32(odp_port
));
440 node
= simap_find(&backer
->tnl_backers
, dp_port
);
445 iter
->odp_port
= node
? u32_to_odp(node
->data
) : ODPP_NONE
;
446 if (tnl_port_reconfigure(iter
, iter
->up
.netdev
,
447 iter
->odp_port
, old_odp_port
,
448 ovs_native_tunneling_is_on(ofproto
), dp_port
)) {
449 backer
->need_revalidate
= REV_RECONFIGURE
;
454 SIMAP_FOR_EACH (node
, &tmp_backers
) {
455 dpif_port_del(backer
->dpif
, u32_to_odp(node
->data
), false);
457 simap_destroy(&tmp_backers
);
459 switch (backer
->need_revalidate
) {
460 case REV_RECONFIGURE
: COVERAGE_INC(rev_reconfigure
); break;
461 case REV_STP
: COVERAGE_INC(rev_stp
); break;
462 case REV_RSTP
: COVERAGE_INC(rev_rstp
); break;
463 case REV_BOND
: COVERAGE_INC(rev_bond
); break;
464 case REV_PORT_TOGGLED
: COVERAGE_INC(rev_port_toggled
); break;
465 case REV_FLOW_TABLE
: COVERAGE_INC(rev_flow_table
); break;
466 case REV_MAC_LEARNING
: COVERAGE_INC(rev_mac_learning
); break;
467 case REV_MCAST_SNOOPING
: COVERAGE_INC(rev_mcast_snooping
); break;
469 backer
->need_revalidate
= 0;
472 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_by_name_node
,
473 &all_ofproto_dpifs_by_name
) {
474 struct ofport_dpif
*ofport
;
475 struct ofbundle
*bundle
;
477 if (ofproto
->backer
!= backer
) {
481 xlate_ofproto_set(ofproto
, ofproto
->up
.name
,
482 ofproto
->backer
->dpif
, ofproto
->ml
,
483 ofproto
->stp
, ofproto
->rstp
, ofproto
->ms
,
484 ofproto
->mbridge
, ofproto
->sflow
, ofproto
->ipfix
,
486 ofproto
->up
.forward_bpdu
,
487 connmgr_has_in_band(ofproto
->up
.connmgr
),
488 &ofproto
->backer
->rt_support
);
490 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
491 xlate_bundle_set(ofproto
, bundle
, bundle
->name
,
492 bundle
->vlan_mode
, bundle
->qinq_ethtype
,
493 bundle
->vlan
, bundle
->trunks
, bundle
->cvlans
,
494 bundle
->use_priority_tags
,
495 bundle
->bond
, bundle
->lacp
,
496 bundle
->floodable
, bundle
->protected);
499 HMAP_FOR_EACH (ofport
, up
.hmap_node
, &ofproto
->up
.ports
) {
500 int stp_port
= ofport
->stp_port
501 ? stp_port_no(ofport
->stp_port
)
503 xlate_ofport_set(ofproto
, ofport
->bundle
, ofport
,
504 ofport
->up
.ofp_port
, ofport
->odp_port
,
505 ofport
->up
.netdev
, ofport
->cfm
, ofport
->bfd
,
506 ofport
->lldp
, ofport
->peer
, stp_port
,
507 ofport
->rstp_port
, ofport
->qdscp
,
508 ofport
->n_qdscp
, ofport
->up
.pp
.config
,
509 ofport
->up
.pp
.state
, ofport
->is_tunnel
,
510 ofport
->up
.may_enable
);
515 udpif_revalidate(backer
->udpif
);
518 process_dpif_port_changes(backer
);
519 ct_zone_timeout_policy_sweep(backer
);
524 /* Check for and handle port changes in 'backer''s dpif. */
526 process_dpif_port_changes(struct dpif_backer
*backer
)
532 error
= dpif_port_poll(backer
->dpif
, &devname
);
538 process_dpif_all_ports_changed(backer
);
542 process_dpif_port_change(backer
, devname
);
547 process_dpif_port_error(backer
, error
);
554 process_dpif_all_ports_changed(struct dpif_backer
*backer
)
556 struct ofproto_dpif
*ofproto
;
557 struct dpif_port dpif_port
;
558 struct dpif_port_dump dump
;
559 struct sset devnames
;
562 sset_init(&devnames
);
563 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_by_name_node
,
564 &all_ofproto_dpifs_by_name
) {
565 if (ofproto
->backer
== backer
) {
566 struct ofport
*ofport
;
568 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->up
.ports
) {
569 sset_add(&devnames
, netdev_get_name(ofport
->netdev
));
573 DPIF_PORT_FOR_EACH (&dpif_port
, &dump
, backer
->dpif
) {
574 sset_add(&devnames
, dpif_port
.name
);
577 SSET_FOR_EACH (devname
, &devnames
) {
578 process_dpif_port_change(backer
, devname
);
580 sset_destroy(&devnames
);
584 process_dpif_port_change(struct dpif_backer
*backer
, const char *devname
)
586 struct ofproto_dpif
*ofproto
;
587 struct dpif_port port
;
589 /* Don't report on the datapath's device. */
590 if (!strcmp(devname
, dpif_base_name(backer
->dpif
))) {
594 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_by_name_node
,
595 &all_ofproto_dpifs_by_name
) {
596 if (simap_contains(&ofproto
->backer
->tnl_backers
, devname
)) {
601 ofproto
= lookup_ofproto_dpif_by_port_name(devname
);
602 if (dpif_port_query_by_name(backer
->dpif
, devname
, &port
)) {
603 /* The port was removed. If we know the datapath,
604 * report it through poll_set(). If we don't, it may be
605 * notifying us of a removal we initiated, so ignore it.
606 * If there's a pending ENOBUFS, let it stand, since
607 * everything will be reevaluated. */
608 if (ofproto
&& ofproto
->port_poll_errno
!= ENOBUFS
) {
609 sset_add(&ofproto
->port_poll_set
, devname
);
610 ofproto
->port_poll_errno
= 0;
612 } else if (!ofproto
) {
613 /* The port was added, but we don't know with which
614 * ofproto we should associate it. Delete it. */
615 dpif_port_del(backer
->dpif
, port
.port_no
, false);
617 struct ofport_dpif
*ofport
;
619 ofport
= ofport_dpif_cast(shash_find_data(
620 &ofproto
->up
.port_by_name
, devname
));
622 && ofport
->odp_port
!= port
.port_no
623 && !odp_port_to_ofport(backer
, port
.port_no
))
625 /* 'ofport''s datapath port number has changed from
626 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
627 * structures to match. */
628 ovs_rwlock_wrlock(&backer
->odp_to_ofport_lock
);
629 hmap_remove(&backer
->odp_to_ofport_map
, &ofport
->odp_port_node
);
630 ofport
->odp_port
= port
.port_no
;
631 hmap_insert(&backer
->odp_to_ofport_map
, &ofport
->odp_port_node
,
632 hash_odp_port(port
.port_no
));
633 ovs_rwlock_unlock(&backer
->odp_to_ofport_lock
);
634 backer
->need_revalidate
= REV_RECONFIGURE
;
637 dpif_port_destroy(&port
);
640 /* Propagate 'error' to all ofprotos based on 'backer'. */
642 process_dpif_port_error(struct dpif_backer
*backer
, int error
)
644 struct ofproto_dpif
*ofproto
;
646 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_by_name_node
,
647 &all_ofproto_dpifs_by_name
) {
648 if (ofproto
->backer
== backer
) {
649 sset_clear(&ofproto
->port_poll_set
);
650 ofproto
->port_poll_errno
= error
;
656 type_wait(const char *type
)
658 struct dpif_backer
*backer
;
660 backer
= shash_find_data(&all_dpif_backers
, type
);
662 /* This is not necessarily a problem, since backers are only
663 * created on demand. */
667 dpif_wait(backer
->dpif
);
670 /* Basic life-cycle. */
672 static int add_internal_flows(struct ofproto_dpif
*);
674 static struct ofproto
*
677 struct ofproto_dpif
*ofproto
= xzalloc(sizeof *ofproto
);
682 dealloc(struct ofproto
*ofproto_
)
684 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
689 close_dpif_backer(struct dpif_backer
*backer
, bool del
)
691 struct simap_node
*node
;
693 ovs_assert(backer
->refcount
> 0);
695 if (--backer
->refcount
) {
699 udpif_destroy(backer
->udpif
);
702 SIMAP_FOR_EACH (node
, &backer
->tnl_backers
) {
703 dpif_port_del(backer
->dpif
, u32_to_odp(node
->data
), false);
706 simap_destroy(&backer
->tnl_backers
);
707 ovs_rwlock_destroy(&backer
->odp_to_ofport_lock
);
708 hmap_destroy(&backer
->odp_to_ofport_map
);
709 shash_find_and_delete(&all_dpif_backers
, backer
->type
);
711 free(backer
->dp_version_string
);
713 dpif_delete(backer
->dpif
);
715 dpif_close(backer
->dpif
);
716 id_pool_destroy(backer
->meter_ids
);
717 ct_zone_config_uninit(backer
);
721 /* Datapath port slated for removal from datapath. */
723 struct ovs_list list_node
;
727 static void check_support(struct dpif_backer
*backer
);
730 open_dpif_backer(const char *type
, struct dpif_backer
**backerp
)
732 struct dpif_backer
*backer
;
733 struct dpif_port_dump port_dump
;
734 struct dpif_port port
;
735 struct shash_node
*node
;
736 struct ovs_list garbage_list
;
737 struct odp_garbage
*garbage
;
744 backer
= shash_find_data(&all_dpif_backers
, type
);
751 backer_name
= xasprintf("ovs-%s", type
);
753 /* Remove any existing datapaths, since we assume we're the only
754 * userspace controlling the datapath. */
756 dp_enumerate_names(type
, &names
);
757 SSET_FOR_EACH(name
, &names
) {
758 struct dpif
*old_dpif
;
760 /* Don't remove our backer if it exists. */
761 if (!strcmp(name
, backer_name
)) {
765 if (dpif_open(name
, type
, &old_dpif
)) {
766 VLOG_WARN("couldn't open old datapath %s to remove it", name
);
768 dpif_delete(old_dpif
);
769 dpif_close(old_dpif
);
772 sset_destroy(&names
);
774 backer
= xmalloc(sizeof *backer
);
776 error
= dpif_create_and_open(backer_name
, type
, &backer
->dpif
);
779 VLOG_ERR("failed to open datapath of type %s: %s", type
,
780 ovs_strerror(error
));
784 backer
->udpif
= udpif_create(backer
, backer
->dpif
);
786 backer
->type
= xstrdup(type
);
787 backer
->refcount
= 1;
788 hmap_init(&backer
->odp_to_ofport_map
);
789 ovs_rwlock_init(&backer
->odp_to_ofport_lock
);
790 backer
->need_revalidate
= 0;
791 simap_init(&backer
->tnl_backers
);
792 backer
->recv_set_enable
= !ofproto_get_flow_restore_wait();
795 if (backer
->recv_set_enable
) {
796 dpif_flow_flush(backer
->dpif
);
799 /* Loop through the ports already on the datapath and remove any
800 * that we don't need anymore. */
801 ovs_list_init(&garbage_list
);
802 dpif_port_dump_start(&port_dump
, backer
->dpif
);
803 while (dpif_port_dump_next(&port_dump
, &port
)) {
804 node
= shash_find(&init_ofp_ports
, port
.name
);
805 if (!node
&& strcmp(port
.name
, dpif_base_name(backer
->dpif
))) {
806 garbage
= xmalloc(sizeof *garbage
);
807 garbage
->odp_port
= port
.port_no
;
808 ovs_list_push_front(&garbage_list
, &garbage
->list_node
);
811 dpif_port_dump_done(&port_dump
);
813 LIST_FOR_EACH_POP (garbage
, list_node
, &garbage_list
) {
814 dpif_port_del(backer
->dpif
, garbage
->odp_port
, false);
818 shash_add(&all_dpif_backers
, type
, backer
);
820 check_support(backer
);
821 atomic_count_init(&backer
->tnl_count
, 0);
823 error
= dpif_recv_set(backer
->dpif
, backer
->recv_set_enable
);
825 VLOG_ERR("failed to listen on datapath of type %s: %s",
826 type
, ovs_strerror(error
));
827 close_dpif_backer(backer
, false);
831 if (backer
->recv_set_enable
) {
832 udpif_set_threads(backer
->udpif
, n_handlers
, n_revalidators
);
835 backer
->dp_version_string
= dpif_get_dp_version(backer
->dpif
);
837 /* Manage Datapath meter IDs if supported. */
838 struct ofputil_meter_features features
;
839 dpif_meter_get_features(backer
->dpif
, &features
);
840 if (features
.max_meters
) {
841 backer
->meter_ids
= id_pool_create(0, features
.max_meters
);
843 backer
->meter_ids
= NULL
;
846 ct_zone_config_init(backer
);
848 /* Make a pristine snapshot of 'support' into 'boottime_support'.
849 * 'boottime_support' can be checked to prevent 'support' to be changed
850 * beyond the datapath capabilities. In case 'support' is changed by
851 * the user, 'boottime_support' can be used to restore it. */
852 backer
->bt_support
= backer
->rt_support
;
858 ovs_native_tunneling_is_on(struct ofproto_dpif
*ofproto
)
860 return ofproto_use_tnl_push_pop
861 && ofproto
->backer
->rt_support
.tnl_push_pop
862 && atomic_count_get(&ofproto
->backer
->tnl_count
);
866 ovs_explicit_drop_action_supported(struct ofproto_dpif
*ofproto
)
868 return ofproto
->backer
->rt_support
.explicit_drop_action
;
872 ovs_lb_output_action_supported(struct ofproto_dpif
*ofproto
)
874 return ofproto
->backer
->rt_support
.lb_output_action
;
877 /* Tests whether 'backer''s datapath supports recirculation. Only newer
878 * datapaths support OVS_KEY_ATTR_RECIRC_ID in keys. We need to disable some
879 * features on older datapaths that don't support this feature.
881 * Returns false if 'backer' definitely does not support recirculation, true if
882 * it seems to support recirculation or if at least the error we get is
885 check_recirc(struct dpif_backer
*backer
)
888 struct odputil_keybuf keybuf
;
891 struct odp_flow_key_parms odp_parms
= {
898 memset(&flow
, 0, sizeof flow
);
902 ofpbuf_use_stack(&key
, &keybuf
, sizeof keybuf
);
903 odp_flow_key_from_flow(&odp_parms
, &key
);
904 enable_recirc
= dpif_probe_feature(backer
->dpif
, "recirculation", &key
,
908 VLOG_INFO("%s: Datapath supports recirculation",
909 dpif_name(backer
->dpif
));
911 VLOG_INFO("%s: Datapath does not support recirculation",
912 dpif_name(backer
->dpif
));
915 return enable_recirc
;
918 /* Tests whether 'dpif' supports unique flow ids. We can skip serializing
919 * some flow attributes for datapaths that support this feature.
921 * Returns true if 'dpif' supports UFID for flow operations.
922 * Returns false if 'dpif' does not support UFID. */
924 check_ufid(struct dpif_backer
*backer
)
927 struct odputil_keybuf keybuf
;
931 struct odp_flow_key_parms odp_parms
= {
935 memset(&flow
, 0, sizeof flow
);
936 flow
.dl_type
= htons(0x1234);
938 ofpbuf_use_stack(&key
, &keybuf
, sizeof keybuf
);
939 odp_flow_key_from_flow(&odp_parms
, &key
);
940 odp_flow_key_hash(key
.data
, key
.size
, &ufid
);
942 enable_ufid
= dpif_probe_feature(backer
->dpif
, "UFID", &key
, NULL
, &ufid
);
945 VLOG_INFO("%s: Datapath supports unique flow ids",
946 dpif_name(backer
->dpif
));
948 VLOG_INFO("%s: Datapath does not support unique flow ids",
949 dpif_name(backer
->dpif
));
954 /* Tests number of 802.1q VLAN headers supported by 'backer''s datapath.
956 * Returns the number of elements in a struct flow's vlan
957 * if the datapath supports at least that many VLAN headers. */
959 check_max_vlan_headers(struct dpif_backer
*backer
)
962 struct odp_flow_key_parms odp_parms
= {
968 memset(&flow
, 0, sizeof flow
);
969 flow
.dl_type
= htons(ETH_TYPE_IP
);
970 for (n
= 0; n
< FLOW_MAX_VLAN_HEADERS
; n
++) {
971 struct odputil_keybuf keybuf
;
974 flow_push_vlan_uninit(&flow
, NULL
);
975 flow
.vlans
[0].tpid
= htons(ETH_TYPE_VLAN
);
976 flow
.vlans
[0].tci
= htons(1) | htons(VLAN_CFI
);
978 ofpbuf_use_stack(&key
, &keybuf
, sizeof keybuf
);
979 odp_flow_key_from_flow(&odp_parms
, &key
);
980 if (!dpif_probe_feature(backer
->dpif
, "VLAN", &key
, NULL
, NULL
)) {
985 VLOG_INFO("%s: VLAN header stack length probed as %d",
986 dpif_name(backer
->dpif
), n
);
989 /* Tests the MPLS label stack depth supported by 'backer''s datapath.
991 * Returns the number of elements in a struct flow's mpls_lse field
992 * if the datapath supports at least that many entries in an
994 * Otherwise returns the number of MPLS push actions supported by
997 check_max_mpls_depth(struct dpif_backer
*backer
)
1002 for (n
= 0; n
< FLOW_MAX_MPLS_LABELS
; n
++) {
1003 struct odputil_keybuf keybuf
;
1005 struct odp_flow_key_parms odp_parms
= {
1009 memset(&flow
, 0, sizeof flow
);
1010 flow
.dl_type
= htons(ETH_TYPE_MPLS
);
1011 flow_set_mpls_bos(&flow
, n
, 1);
1013 ofpbuf_use_stack(&key
, &keybuf
, sizeof keybuf
);
1014 odp_flow_key_from_flow(&odp_parms
, &key
);
1015 if (!dpif_probe_feature(backer
->dpif
, "MPLS", &key
, NULL
, NULL
)) {
1020 VLOG_INFO("%s: MPLS label stack length probed as %d",
1021 dpif_name(backer
->dpif
), n
);
1026 add_sample_actions(struct ofpbuf
*actions
, int nesting
)
1029 nl_msg_put_odp_port(actions
, OVS_ACTION_ATTR_OUTPUT
, u32_to_odp(1));
1033 size_t start
, actions_start
;
1035 start
= nl_msg_start_nested(actions
, OVS_ACTION_ATTR_SAMPLE
);
1036 actions_start
= nl_msg_start_nested(actions
, OVS_SAMPLE_ATTR_ACTIONS
);
1037 add_sample_actions(actions
, nesting
- 1);
1038 nl_msg_end_nested(actions
, actions_start
);
1039 nl_msg_put_u32(actions
, OVS_SAMPLE_ATTR_PROBABILITY
, UINT32_MAX
);
1040 nl_msg_end_nested(actions
, start
);
1043 /* Tests the nested sample actions levels supported by 'backer''s datapath.
1045 * Returns the number of nested sample actions accepted by the datapath. */
1047 check_max_sample_nesting(struct dpif_backer
*backer
)
1049 struct odputil_keybuf keybuf
;
1054 struct odp_flow_key_parms odp_parms
= {
1058 memset(&flow
, 0, sizeof flow
);
1059 ofpbuf_use_stack(&key
, &keybuf
, sizeof keybuf
);
1060 odp_flow_key_from_flow(&odp_parms
, &key
);
1062 /* OVS datapath has always supported at least 3 nested levels. */
1063 for (n
= 3; n
< FLOW_MAX_SAMPLE_NESTING
; n
++) {
1064 struct ofpbuf actions
;
1067 ofpbuf_init(&actions
, 300);
1068 add_sample_actions(&actions
, n
);
1069 ok
= dpif_probe_feature(backer
->dpif
, "Sample action nesting", &key
,
1071 ofpbuf_uninit(&actions
);
1077 VLOG_INFO("%s: Max sample nesting level probed as %d",
1078 dpif_name(backer
->dpif
), n
);
1082 /* Tests whether 'backer''s datapath supports masked data in
1083 * OVS_ACTION_ATTR_SET actions. We need to disable some features on older
1084 * datapaths that don't support this feature. */
1086 check_masked_set_action(struct dpif_backer
*backer
)
1088 struct eth_header
*eth
;
1089 struct ofpbuf actions
;
1090 struct dp_packet packet
;
1093 struct ovs_key_ethernet key
, mask
;
1095 /* Compose a set action that will cause an EINVAL error on older
1096 * datapaths that don't support masked set actions.
1097 * Avoid using a full mask, as it could be translated to a non-masked
1098 * set action instead. */
1099 ofpbuf_init(&actions
, 64);
1100 memset(&key
, 0x53, sizeof key
);
1101 memset(&mask
, 0x7f, sizeof mask
);
1102 commit_masked_set_action(&actions
, OVS_KEY_ATTR_ETHERNET
, &key
, &mask
,
1105 /* Compose a dummy ethernet packet. */
1106 dp_packet_init(&packet
, ETH_HEADER_LEN
);
1107 eth
= dp_packet_put_zeros(&packet
, ETH_HEADER_LEN
);
1108 eth
->eth_type
= htons(0x1234);
1110 flow_extract(&packet
, &flow
);
1112 /* Execute the actions. On older datapaths this fails with EINVAL, on
1113 * newer datapaths it succeeds. */
1114 struct dpif_execute execute
= {
1115 .actions
= actions
.data
,
1116 .actions_len
= actions
.size
,
1121 error
= dpif_execute(backer
->dpif
, &execute
);
1123 dp_packet_uninit(&packet
);
1124 ofpbuf_uninit(&actions
);
1127 /* Masked set action is not supported. */
1128 VLOG_INFO("%s: datapath does not support masked set action feature.",
1129 dpif_name(backer
->dpif
));
1134 /* Tests whether 'backer''s datapath supports truncation of a packet in
1135 * OVS_ACTION_ATTR_TRUNC. We need to disable some features on older
1136 * datapaths that don't support this feature. */
1138 check_trunc_action(struct dpif_backer
*backer
)
1140 struct eth_header
*eth
;
1141 struct ofpbuf actions
;
1142 struct dp_packet packet
;
1143 struct ovs_action_trunc
*trunc
;
1147 /* Compose an action with output(port:1,
1148 * max_len:OVS_ACTION_OUTPUT_MIN + 1).
1149 * This translates to one truncate action and one output action. */
1150 ofpbuf_init(&actions
, 64);
1151 trunc
= nl_msg_put_unspec_uninit(&actions
,
1152 OVS_ACTION_ATTR_TRUNC
, sizeof *trunc
);
1154 trunc
->max_len
= ETH_HEADER_LEN
+ 1;
1155 nl_msg_put_odp_port(&actions
, OVS_ACTION_ATTR_OUTPUT
, u32_to_odp(1));
1157 /* Compose a dummy Ethernet packet. */
1158 dp_packet_init(&packet
, ETH_HEADER_LEN
);
1159 eth
= dp_packet_put_zeros(&packet
, ETH_HEADER_LEN
);
1160 eth
->eth_type
= htons(0x1234);
1162 flow_extract(&packet
, &flow
);
1164 /* Execute the actions. On older datapaths this fails with EINVAL, on
1165 * newer datapaths it succeeds. */
1166 struct dpif_execute execute
= {
1167 .actions
= actions
.data
,
1168 .actions_len
= actions
.size
,
1173 error
= dpif_execute(backer
->dpif
, &execute
);
1175 dp_packet_uninit(&packet
);
1176 ofpbuf_uninit(&actions
);
1179 VLOG_INFO("%s: Datapath does not support truncate action",
1180 dpif_name(backer
->dpif
));
1182 VLOG_INFO("%s: Datapath supports truncate action",
1183 dpif_name(backer
->dpif
));
1189 /* Tests whether 'backer''s datapath supports the clone action
1190 * OVS_ACTION_ATTR_CLONE. */
1192 check_clone(struct dpif_backer
*backer
)
1194 struct eth_header
*eth
;
1196 struct dp_packet packet
;
1197 struct ofpbuf actions
;
1201 /* Compose clone with an empty action list.
1202 * and check if datapath can decode the message. */
1203 ofpbuf_init(&actions
, 64);
1204 clone_start
= nl_msg_start_nested(&actions
, OVS_ACTION_ATTR_CLONE
);
1205 nl_msg_end_nested(&actions
, clone_start
);
1207 /* Compose a dummy Ethernet packet. */
1208 dp_packet_init(&packet
, ETH_HEADER_LEN
);
1209 eth
= dp_packet_put_zeros(&packet
, ETH_HEADER_LEN
);
1210 eth
->eth_type
= htons(0x1234);
1212 flow_extract(&packet
, &flow
);
1214 /* Execute the actions. On older datapaths this fails with EINVAL, on
1215 * newer datapaths it succeeds. */
1216 struct dpif_execute execute
= {
1217 .actions
= actions
.data
,
1218 .actions_len
= actions
.size
,
1223 error
= dpif_execute(backer
->dpif
, &execute
);
1225 dp_packet_uninit(&packet
);
1226 ofpbuf_uninit(&actions
);
1229 VLOG_INFO("%s: Datapath does not support clone action",
1230 dpif_name(backer
->dpif
));
1232 VLOG_INFO("%s: Datapath supports clone action",
1233 dpif_name(backer
->dpif
));
1239 /* Tests whether 'backer''s datapath supports the OVS_CT_ATTR_EVENTMASK
1240 * attribute in OVS_ACTION_ATTR_CT. */
1242 check_ct_eventmask(struct dpif_backer
*backer
)
1244 struct dp_packet packet
;
1245 struct ofpbuf actions
;
1246 struct flow flow
= {
1247 .dl_type
= CONSTANT_HTONS(ETH_TYPE_IP
),
1248 .nw_proto
= IPPROTO_UDP
,
1250 /* Use the broadcast address on the loopback address range 127/8 to
1251 * avoid hitting any real conntrack entries. We leave the UDP ports to
1252 * zeroes for the same purpose. */
1253 .nw_src
= CONSTANT_HTONL(0x7fffffff),
1254 .nw_dst
= CONSTANT_HTONL(0x7fffffff),
1259 /* Compose CT action with eventmask attribute and check if datapath can
1260 * decode the message. */
1261 ofpbuf_init(&actions
, 64);
1262 ct_start
= nl_msg_start_nested(&actions
, OVS_ACTION_ATTR_CT
);
1263 /* Eventmask has no effect without the commit flag, but currently the
1264 * datapath will accept an eventmask even without commit. This is useful
1265 * as we do not want to persist the probe connection in the conntrack
1267 nl_msg_put_u32(&actions
, OVS_CT_ATTR_EVENTMASK
, ~0);
1268 nl_msg_end_nested(&actions
, ct_start
);
1270 /* Compose a dummy UDP packet. */
1271 dp_packet_init(&packet
, 0);
1272 flow_compose(&packet
, &flow
, NULL
, 64);
1274 /* Execute the actions. On older datapaths this fails with EINVAL, on
1275 * newer datapaths it succeeds. */
1276 struct dpif_execute execute
= {
1277 .actions
= actions
.data
,
1278 .actions_len
= actions
.size
,
1283 error
= dpif_execute(backer
->dpif
, &execute
);
1285 dp_packet_uninit(&packet
);
1286 ofpbuf_uninit(&actions
);
1289 VLOG_INFO("%s: Datapath does not support eventmask in conntrack action",
1290 dpif_name(backer
->dpif
));
1292 VLOG_INFO("%s: Datapath supports eventmask in conntrack action",
1293 dpif_name(backer
->dpif
));
1299 /* Tests whether 'backer''s datapath supports the OVS_ACTION_ATTR_CT_CLEAR
1302 check_ct_clear(struct dpif_backer
*backer
)
1304 struct odputil_keybuf keybuf
;
1305 uint8_t actbuf
[NL_A_FLAG_SIZE
];
1306 struct ofpbuf actions
;
1311 struct odp_flow_key_parms odp_parms
= {
1316 memset(&flow
, 0, sizeof flow
);
1317 ofpbuf_use_stack(&key
, &keybuf
, sizeof keybuf
);
1318 odp_flow_key_from_flow(&odp_parms
, &key
);
1320 ofpbuf_use_stack(&actions
, &actbuf
, sizeof actbuf
);
1321 nl_msg_put_flag(&actions
, OVS_ACTION_ATTR_CT_CLEAR
);
1323 supported
= dpif_probe_feature(backer
->dpif
, "ct_clear", &key
,
1326 VLOG_INFO("%s: Datapath %s ct_clear action",
1327 dpif_name(backer
->dpif
), (supported
) ? "supports"
1328 : "does not support");
1332 /* Tests whether 'backer''s datapath supports the OVS_CT_ATTR_TIMEOUT
1333 * attribute in OVS_ACTION_ATTR_CT. */
1335 check_ct_timeout_policy(struct dpif_backer
*backer
)
1337 struct dp_packet packet
;
1338 struct ofpbuf actions
;
1339 struct flow flow
= {
1340 .dl_type
= CONSTANT_HTONS(ETH_TYPE_IP
),
1341 .nw_proto
= IPPROTO_UDP
,
1343 /* Use the broadcast address on the loopback address range 127/8 to
1344 * avoid hitting any real conntrack entries. We leave the UDP ports to
1345 * zeroes for the same purpose. */
1346 .nw_src
= CONSTANT_HTONL(0x7fffffff),
1347 .nw_dst
= CONSTANT_HTONL(0x7fffffff),
1352 /* Compose CT action with timeout policy attribute and check if datapath
1353 * can decode the message. */
1354 ofpbuf_init(&actions
, 64);
1355 ct_start
= nl_msg_start_nested(&actions
, OVS_ACTION_ATTR_CT
);
1356 /* Timeout policy has no effect without the commit flag, but currently the
1357 * datapath will accept a timeout policy even without commit. This is
1358 * useful as we do not want to persist the probe connection in the
1359 * conntrack table. */
1360 nl_msg_put_string(&actions
, OVS_CT_ATTR_TIMEOUT
, "ovs_test_tp");
1361 nl_msg_end_nested(&actions
, ct_start
);
1363 /* Compose a dummy UDP packet. */
1364 dp_packet_init(&packet
, 0);
1365 flow_compose(&packet
, &flow
, NULL
, 64);
1367 /* Execute the actions. On older datapaths this fails with EINVAL, on
1368 * newer datapaths it succeeds. */
1369 struct dpif_execute execute
= {
1370 .actions
= actions
.data
,
1371 .actions_len
= actions
.size
,
1376 error
= dpif_execute(backer
->dpif
, &execute
);
1378 dp_packet_uninit(&packet
);
1379 ofpbuf_uninit(&actions
);
1382 VLOG_INFO("%s: Datapath does not support timeout policy in conntrack "
1383 "action", dpif_name(backer
->dpif
));
1385 VLOG_INFO("%s: Datapath supports timeout policy in conntrack action",
1386 dpif_name(backer
->dpif
));
1392 /* Tests whether 'backer''s datapath supports the
1393 * OVS_ACTION_ATTR_CHECK_PKT_LEN action. */
1395 check_check_pkt_len(struct dpif_backer
*backer
)
1397 struct odputil_keybuf keybuf
;
1398 struct ofpbuf actions
;
1403 struct odp_flow_key_parms odp_parms
= {
1408 memset(&flow
, 0, sizeof flow
);
1409 ofpbuf_use_stack(&key
, &keybuf
, sizeof keybuf
);
1410 odp_flow_key_from_flow(&odp_parms
, &key
);
1411 ofpbuf_init(&actions
, 64);
1414 cpl_start
= nl_msg_start_nested(&actions
, OVS_ACTION_ATTR_CHECK_PKT_LEN
);
1415 nl_msg_put_u16(&actions
, OVS_CHECK_PKT_LEN_ATTR_PKT_LEN
, 100);
1417 /* Putting these actions without any data is good enough to check
1418 * if check_pkt_len is supported or not. */
1419 nl_msg_put_flag(&actions
, OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER
);
1420 nl_msg_put_flag(&actions
, OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL
);
1422 nl_msg_end_nested(&actions
, cpl_start
);
1424 supported
= dpif_probe_feature(backer
->dpif
, "check_pkt_len", &key
,
1426 ofpbuf_uninit(&actions
);
1427 VLOG_INFO("%s: Datapath %s check_pkt_len action",
1428 dpif_name(backer
->dpif
), supported
? "supports"
1429 : "does not support");
1433 /* Probe the highest dp_hash algorithm supported by the datapath. */
1435 check_max_dp_hash_alg(struct dpif_backer
*backer
)
1437 struct odputil_keybuf keybuf
;
1440 struct ovs_action_hash
*hash
;
1443 struct odp_flow_key_parms odp_parms
= {
1448 memset(&flow
, 0, sizeof flow
);
1449 ofpbuf_use_stack(&key
, &keybuf
, sizeof keybuf
);
1450 odp_flow_key_from_flow(&odp_parms
, &key
);
1452 /* All datapaths support algortithm 0 (OVS_HASH_ALG_L4). */
1453 for (int alg
= 1; alg
< __OVS_HASH_MAX
; alg
++) {
1454 struct ofpbuf actions
;
1457 ofpbuf_init(&actions
, 300);
1458 hash
= nl_msg_put_unspec_uninit(&actions
,
1459 OVS_ACTION_ATTR_HASH
, sizeof *hash
);
1460 hash
->hash_basis
= 0;
1461 hash
->hash_alg
= alg
;
1462 ok
= dpif_probe_feature(backer
->dpif
, "Max dp_hash algorithm", &key
,
1464 ofpbuf_uninit(&actions
);
1472 VLOG_INFO("%s: Max dp_hash algorithm probed to be %d",
1473 dpif_name(backer
->dpif
), max_alg
);
1477 /* Tests whether 'backer''s datapath supports IPv6 ND extensions.
1478 * Only userspace datapath support OVS_KEY_ATTR_ND_EXTENSIONS in keys.
1480 * Returns false if 'backer' definitely does not support matching and
1481 * setting reserved and options type, true if it seems to support. */
1483 check_nd_extensions(struct dpif_backer
*backer
)
1485 struct eth_header
*eth
;
1486 struct ofpbuf actions
;
1487 struct dp_packet packet
;
1490 struct ovs_key_nd_extensions key
, mask
;
1492 ofpbuf_init(&actions
, 64);
1493 memset(&key
, 0x53, sizeof key
);
1494 memset(&mask
, 0x7f, sizeof mask
);
1495 commit_masked_set_action(&actions
, OVS_KEY_ATTR_ND_EXTENSIONS
, &key
, &mask
,
1498 /* Compose a dummy ethernet packet. */
1499 dp_packet_init(&packet
, ETH_HEADER_LEN
);
1500 eth
= dp_packet_put_zeros(&packet
, ETH_HEADER_LEN
);
1501 eth
->eth_type
= htons(0x1234);
1503 flow_extract(&packet
, &flow
);
1505 /* Execute the actions. On datapaths without support fails with EINVAL. */
1506 struct dpif_execute execute
= {
1507 .actions
= actions
.data
,
1508 .actions_len
= actions
.size
,
1513 error
= dpif_execute(backer
->dpif
, &execute
);
1515 dp_packet_uninit(&packet
);
1516 ofpbuf_uninit(&actions
);
1518 VLOG_INFO("%s: Datapath %s IPv6 ND Extensions", dpif_name(backer
->dpif
),
1519 error
? "does not support" : "supports");
1524 #define CHECK_FEATURE__(NAME, SUPPORT, FIELD, VALUE, ETHTYPE) \
1526 check_##NAME(struct dpif_backer *backer) \
1529 struct odputil_keybuf keybuf; \
1530 struct ofpbuf key; \
1532 struct odp_flow_key_parms odp_parms = { \
1539 memset(&flow, 0, sizeof flow); \
1540 flow.FIELD = VALUE; \
1541 flow.dl_type = htons(ETHTYPE); \
1543 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); \
1544 odp_flow_key_from_flow(&odp_parms, &key); \
1545 enable = dpif_probe_feature(backer->dpif, #NAME, &key, NULL, NULL); \
1548 VLOG_INFO("%s: Datapath supports "#NAME, dpif_name(backer->dpif)); \
1550 VLOG_INFO("%s: Datapath does not support "#NAME, \
1551 dpif_name(backer->dpif)); \
1556 #define CHECK_FEATURE(FIELD) CHECK_FEATURE__(FIELD, FIELD, FIELD, 1, \
1559 CHECK_FEATURE(ct_state
)
1560 CHECK_FEATURE(ct_zone
)
1561 CHECK_FEATURE(ct_mark
)
1562 CHECK_FEATURE__(ct_label
, ct_label
, ct_label
.u64
.lo
, 1, ETH_TYPE_IP
)
1563 CHECK_FEATURE__(ct_state_nat
, ct_state
, ct_state
, \
1564 CS_TRACKED
|CS_SRC_NAT
, ETH_TYPE_IP
)
1565 CHECK_FEATURE__(ct_orig_tuple
, ct_orig_tuple
, ct_nw_proto
, 1, ETH_TYPE_IP
)
1566 CHECK_FEATURE__(ct_orig_tuple6
, ct_orig_tuple6
, ct_nw_proto
, 1, ETH_TYPE_IPV6
)
1568 #undef CHECK_FEATURE
1569 #undef CHECK_FEATURE__
1572 check_support(struct dpif_backer
*backer
)
1575 backer
->rt_support
.odp
.recirc
= check_recirc(backer
);
1576 backer
->rt_support
.odp
.max_vlan_headers
= check_max_vlan_headers(backer
);
1577 backer
->rt_support
.odp
.max_mpls_depth
= check_max_mpls_depth(backer
);
1578 backer
->rt_support
.masked_set_action
= check_masked_set_action(backer
);
1579 backer
->rt_support
.trunc
= check_trunc_action(backer
);
1580 backer
->rt_support
.ufid
= check_ufid(backer
);
1581 backer
->rt_support
.tnl_push_pop
= dpif_supports_tnl_push_pop(backer
->dpif
);
1582 backer
->rt_support
.clone
= check_clone(backer
);
1583 backer
->rt_support
.sample_nesting
= check_max_sample_nesting(backer
);
1584 backer
->rt_support
.ct_eventmask
= check_ct_eventmask(backer
);
1585 backer
->rt_support
.ct_clear
= check_ct_clear(backer
);
1586 backer
->rt_support
.max_hash_alg
= check_max_dp_hash_alg(backer
);
1587 backer
->rt_support
.check_pkt_len
= check_check_pkt_len(backer
);
1588 backer
->rt_support
.ct_timeout
= check_ct_timeout_policy(backer
);
1589 backer
->rt_support
.explicit_drop_action
=
1590 dpif_supports_explicit_drop_action(backer
->dpif
);
1591 backer
->rt_support
.lb_output_action
=
1592 dpif_supports_lb_output_action(backer
->dpif
);
1595 backer
->rt_support
.odp
.ct_state
= check_ct_state(backer
);
1596 backer
->rt_support
.odp
.ct_zone
= check_ct_zone(backer
);
1597 backer
->rt_support
.odp
.ct_mark
= check_ct_mark(backer
);
1598 backer
->rt_support
.odp
.ct_label
= check_ct_label(backer
);
1599 backer
->rt_support
.odp
.ct_state_nat
= check_ct_state_nat(backer
);
1600 backer
->rt_support
.odp
.ct_orig_tuple
= check_ct_orig_tuple(backer
);
1601 backer
->rt_support
.odp
.ct_orig_tuple6
= check_ct_orig_tuple6(backer
);
1602 backer
->rt_support
.odp
.nd_ext
= check_nd_extensions(backer
);
1606 construct(struct ofproto
*ofproto_
)
1608 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1609 struct shash_node
*node
, *next
;
1612 /* Tunnel module can get used right after the udpif threads are running. */
1613 ofproto_tunnel_init();
1615 error
= open_dpif_backer(ofproto
->up
.type
, &ofproto
->backer
);
1620 uuid_generate(&ofproto
->uuid
);
1621 atomic_init(&ofproto
->tables_version
, OVS_VERSION_MIN
);
1622 ofproto
->netflow
= NULL
;
1623 ofproto
->sflow
= NULL
;
1624 ofproto
->ipfix
= NULL
;
1625 ofproto
->stp
= NULL
;
1626 ofproto
->rstp
= NULL
;
1627 ofproto
->dump_seq
= 0;
1628 hmap_init(&ofproto
->bundles
);
1629 ofproto
->ml
= mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME
);
1631 ofproto
->mbridge
= mbridge_create();
1632 ofproto
->has_bonded_bundles
= false;
1633 ofproto
->lacp_enabled
= false;
1634 ovs_mutex_init_adaptive(&ofproto
->stats_mutex
);
1636 guarded_list_init(&ofproto
->ams
);
1638 sset_init(&ofproto
->ports
);
1639 sset_init(&ofproto
->ghost_ports
);
1640 sset_init(&ofproto
->port_poll_set
);
1641 ofproto
->port_poll_errno
= 0;
1642 ofproto
->change_seq
= 0;
1643 ofproto
->ams_seq
= seq_create();
1644 ofproto
->ams_seqno
= seq_read(ofproto
->ams_seq
);
1647 SHASH_FOR_EACH_SAFE (node
, next
, &init_ofp_ports
) {
1648 struct iface_hint
*iface_hint
= node
->data
;
1650 if (!strcmp(iface_hint
->br_name
, ofproto
->up
.name
)) {
1651 /* Check if the datapath already has this port. */
1652 if (dpif_port_exists(ofproto
->backer
->dpif
, node
->name
)) {
1653 sset_add(&ofproto
->ports
, node
->name
);
1656 free(iface_hint
->br_name
);
1657 free(iface_hint
->br_type
);
1659 shash_delete(&init_ofp_ports
, node
);
1663 hmap_insert(&all_ofproto_dpifs_by_name
,
1664 &ofproto
->all_ofproto_dpifs_by_name_node
,
1665 hash_string(ofproto
->up
.name
, 0));
1666 hmap_insert(&all_ofproto_dpifs_by_uuid
,
1667 &ofproto
->all_ofproto_dpifs_by_uuid_node
,
1668 uuid_hash(&ofproto
->uuid
));
1669 memset(&ofproto
->stats
, 0, sizeof ofproto
->stats
);
1671 ofproto_init_tables(ofproto_
, N_TABLES
);
1672 error
= add_internal_flows(ofproto
);
1674 ofproto
->up
.tables
[TBL_INTERNAL
].flags
= OFTABLE_HIDDEN
| OFTABLE_READONLY
;
1680 add_internal_miss_flow(struct ofproto_dpif
*ofproto
, int id
,
1681 const struct ofpbuf
*ofpacts
, struct rule_dpif
**rulep
)
1687 match_init_catchall(&match
);
1688 match_set_reg(&match
, 0, id
);
1690 error
= ofproto_dpif_add_internal_flow(ofproto
, &match
, 0, 0, ofpacts
,
1692 *rulep
= error
? NULL
: rule_dpif_cast(rule
);
1698 add_internal_flows(struct ofproto_dpif
*ofproto
)
1700 struct ofpact_controller
*controller
;
1701 uint64_t ofpacts_stub
[128 / 8];
1702 struct ofpbuf ofpacts
;
1703 struct rule
*unused_rulep OVS_UNUSED
;
1708 ofpbuf_use_stack(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
1711 controller
= ofpact_put_CONTROLLER(&ofpacts
);
1712 controller
->max_len
= UINT16_MAX
;
1713 controller
->controller_id
= 0;
1714 controller
->reason
= OFPR_IMPLICIT_MISS
;
1715 controller
->meter_id
= NX_CTLR_NO_METER
;
1716 ofpact_finish_CONTROLLER(&ofpacts
, &controller
);
1718 error
= add_internal_miss_flow(ofproto
, id
++, &ofpacts
,
1719 &ofproto
->miss_rule
);
1724 ofpbuf_clear(&ofpacts
);
1725 error
= add_internal_miss_flow(ofproto
, id
++, &ofpacts
,
1726 &ofproto
->no_packet_in_rule
);
1731 error
= add_internal_miss_flow(ofproto
, id
++, &ofpacts
,
1732 &ofproto
->drop_frags_rule
);
1737 /* Drop any run away non-recirc rule lookups. Recirc_id has to be
1738 * zero when reaching this rule.
1740 * (priority=2), recirc_id=0, actions=drop
1742 ofpbuf_clear(&ofpacts
);
1743 match_init_catchall(&match
);
1744 match_set_recirc_id(&match
, 0);
1745 error
= ofproto_dpif_add_internal_flow(ofproto
, &match
, 2, 0, &ofpacts
,
1751 destruct(struct ofproto
*ofproto_
, bool del
)
1753 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1754 struct ofproto_async_msg
*am
;
1755 struct rule_dpif
*rule
;
1756 struct oftable
*table
;
1757 struct ovs_list ams
;
1759 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1761 xlate_remove_ofproto(ofproto
);
1764 hmap_remove(&all_ofproto_dpifs_by_name
,
1765 &ofproto
->all_ofproto_dpifs_by_name_node
);
1766 hmap_remove(&all_ofproto_dpifs_by_uuid
,
1767 &ofproto
->all_ofproto_dpifs_by_uuid_node
);
1769 OFPROTO_FOR_EACH_TABLE (table
, &ofproto
->up
) {
1770 CLS_FOR_EACH (rule
, up
.cr
, &table
->cls
) {
1771 ofproto_rule_delete(&ofproto
->up
, &rule
->up
);
1774 ofproto_group_delete_all(&ofproto
->up
);
1776 guarded_list_pop_all(&ofproto
->ams
, &ams
);
1777 LIST_FOR_EACH_POP (am
, list_node
, &ams
) {
1778 ofproto_async_msg_free(am
);
1780 guarded_list_destroy(&ofproto
->ams
);
1782 recirc_free_ofproto(ofproto
, ofproto
->up
.name
);
1784 mbridge_unref(ofproto
->mbridge
);
1786 netflow_unref(ofproto
->netflow
);
1787 dpif_sflow_unref(ofproto
->sflow
);
1788 dpif_ipfix_unref(ofproto
->ipfix
);
1789 hmap_destroy(&ofproto
->bundles
);
1790 mac_learning_unref(ofproto
->ml
);
1791 mcast_snooping_unref(ofproto
->ms
);
1792 stp_unref(ofproto
->stp
);
1793 rstp_unref(ofproto
->rstp
);
1795 sset_destroy(&ofproto
->ports
);
1796 sset_destroy(&ofproto
->ghost_ports
);
1797 sset_destroy(&ofproto
->port_poll_set
);
1799 ovs_mutex_destroy(&ofproto
->stats_mutex
);
1801 seq_destroy(ofproto
->ams_seq
);
1803 close_dpif_backer(ofproto
->backer
, del
);
1807 run(struct ofproto
*ofproto_
)
1809 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1810 uint64_t new_seq
, new_dump_seq
;
1813 if (mbridge_need_revalidate(ofproto
->mbridge
)) {
1814 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1815 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
1816 mac_learning_flush(ofproto
->ml
);
1817 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
1818 mcast_snooping_mdb_flush(ofproto
->ms
);
1821 /* Always updates the ofproto->ams_seqno to avoid frequent wakeup during
1822 * flow restore. Even though nothing is processed during flow restore,
1823 * all queued 'ams' will be handled immediately when flow restore
1825 ofproto
->ams_seqno
= seq_read(ofproto
->ams_seq
);
1827 /* Do not perform any periodic activity required by 'ofproto' while
1828 * waiting for flow restore to complete. */
1829 if (!ofproto_get_flow_restore_wait()) {
1830 struct ofproto_async_msg
*am
;
1831 struct ovs_list ams
;
1833 guarded_list_pop_all(&ofproto
->ams
, &ams
);
1834 LIST_FOR_EACH_POP (am
, list_node
, &ams
) {
1835 connmgr_send_async_msg(ofproto
->up
.connmgr
, am
);
1836 ofproto_async_msg_free(am
);
1840 if (ofproto
->netflow
) {
1841 netflow_run(ofproto
->netflow
);
1843 if (ofproto
->sflow
) {
1844 dpif_sflow_run(ofproto
->sflow
);
1846 if (ofproto
->ipfix
) {
1847 dpif_ipfix_run(ofproto
->ipfix
);
1850 new_seq
= seq_read(connectivity_seq_get());
1851 if (ofproto
->change_seq
!= new_seq
) {
1852 struct ofport_dpif
*ofport
;
1854 HMAP_FOR_EACH (ofport
, up
.hmap_node
, &ofproto
->up
.ports
) {
1858 ofproto
->change_seq
= new_seq
;
1860 if (ofproto
->lacp_enabled
|| ofproto
->has_bonded_bundles
) {
1861 struct ofbundle
*bundle
;
1863 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
1870 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
1871 if (mac_learning_run(ofproto
->ml
)) {
1872 ofproto
->backer
->need_revalidate
= REV_MAC_LEARNING
;
1874 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
1876 if (mcast_snooping_run(ofproto
->ms
)) {
1877 ofproto
->backer
->need_revalidate
= REV_MCAST_SNOOPING
;
1880 /* Check if controller connection is toggled. */
1881 is_connected
= ofproto_is_alive(&ofproto
->up
);
1882 if (ofproto
->is_controller_connected
!= is_connected
) {
1883 ofproto
->is_controller_connected
= is_connected
;
1884 /* Trigger revalidation as fast failover group monitoring
1885 * controller port may need to check liveness again. */
1886 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1889 new_dump_seq
= seq_read(udpif_dump_seq(ofproto
->backer
->udpif
));
1890 if (ofproto
->dump_seq
!= new_dump_seq
) {
1891 struct rule
*rule
, *next_rule
;
1892 long long now
= time_msec();
1894 /* We know stats are relatively fresh, so now is a good time to do some
1896 ofproto
->dump_seq
= new_dump_seq
;
1898 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1900 ovs_mutex_lock(&ofproto_mutex
);
1901 LIST_FOR_EACH_SAFE (rule
, next_rule
, expirable
,
1902 &ofproto
->up
.expirable
) {
1903 rule_expire(rule_dpif_cast(rule
), now
);
1905 ovs_mutex_unlock(&ofproto_mutex
);
1907 /* All outstanding data in existing flows has been accounted, so it's a
1908 * good time to do bond rebalancing. */
1909 if (ofproto
->has_bonded_bundles
) {
1910 struct ofbundle
*bundle
;
1912 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
1914 bond_rebalance(bundle
->bond
);
1923 ofproto_dpif_wait(struct ofproto
*ofproto_
)
1925 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1927 if (ofproto_get_flow_restore_wait()) {
1931 if (ofproto
->sflow
) {
1932 dpif_sflow_wait(ofproto
->sflow
);
1934 if (ofproto
->ipfix
) {
1935 dpif_ipfix_wait(ofproto
->ipfix
);
1937 if (ofproto
->lacp_enabled
|| ofproto
->has_bonded_bundles
) {
1938 struct ofbundle
*bundle
;
1940 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
1941 bundle_wait(bundle
);
1944 if (ofproto
->netflow
) {
1945 netflow_wait(ofproto
->netflow
);
1947 ovs_rwlock_rdlock(&ofproto
->ml
->rwlock
);
1948 mac_learning_wait(ofproto
->ml
);
1949 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
1950 mcast_snooping_wait(ofproto
->ms
);
1952 if (ofproto
->backer
->need_revalidate
) {
1953 poll_immediate_wake();
1956 seq_wait(udpif_dump_seq(ofproto
->backer
->udpif
), ofproto
->dump_seq
);
1957 seq_wait(ofproto
->ams_seq
, ofproto
->ams_seqno
);
1961 type_get_memory_usage(const char *type
, struct simap
*usage
)
1963 struct dpif_backer
*backer
;
1965 backer
= shash_find_data(&all_dpif_backers
, type
);
1967 udpif_get_memory_usage(backer
->udpif
, usage
);
1972 flush(struct ofproto
*ofproto_
)
1974 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1975 struct dpif_backer
*backer
= ofproto
->backer
;
1978 udpif_flush(backer
->udpif
);
1983 query_tables(struct ofproto
*ofproto
,
1984 struct ofputil_table_features
*features OVS_UNUSED
,
1985 struct ofputil_table_stats
*stats
)
1990 for (i
= 0; i
< ofproto
->n_tables
; i
++) {
1991 unsigned long missed
, matched
;
1993 atomic_read_relaxed(&ofproto
->tables
[i
].n_matched
, &matched
);
1994 atomic_read_relaxed(&ofproto
->tables
[i
].n_missed
, &missed
);
1996 stats
[i
].matched_count
= matched
;
1997 stats
[i
].lookup_count
= matched
+ missed
;
2003 set_tables_version(struct ofproto
*ofproto_
, ovs_version_t version
)
2005 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2007 /* Use memory_order_release to signify that any prior memory accesses can
2008 * not be reordered to happen after this atomic store. This makes sure the
2009 * new version is properly set up when the readers can read this 'version'
2011 atomic_store_explicit(&ofproto
->tables_version
, version
,
2012 memory_order_release
);
2013 /* 'need_revalidate' can be reordered to happen before the atomic_store
2014 * above, but it does not matter as this variable is not accessed by other
2016 ofproto
->backer
->need_revalidate
= REV_FLOW_TABLE
;
2019 static struct ofport
*
2022 struct ofport_dpif
*port
= xzalloc(sizeof *port
);
2027 port_dealloc(struct ofport
*port_
)
2029 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
2034 port_construct(struct ofport
*port_
)
2036 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
2037 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
2038 const struct netdev
*netdev
= port
->up
.netdev
;
2039 char namebuf
[NETDEV_VPORT_NAME_BUFSIZE
];
2040 const char *dp_port_name
;
2041 struct dpif_port dpif_port
;
2044 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2045 port
->bundle
= NULL
;
2049 port
->stp_port
= NULL
;
2050 port
->stp_state
= STP_DISABLED
;
2051 port
->rstp_port
= NULL
;
2052 port
->rstp_state
= RSTP_DISABLED
;
2053 port
->is_tunnel
= false;
2057 port
->carrier_seq
= netdev_get_carrier_resets(netdev
);
2059 if (netdev_vport_is_patch(netdev
)) {
2060 /* By bailing out here, we don't submit the port to the sFlow module
2061 * to be considered for counter polling export. This is correct
2062 * because the patch port represents an interface that sFlow considers
2063 * to be "internal" to the switch as a whole, and therefore not a
2064 * candidate for counter polling. */
2065 port
->odp_port
= ODPP_NONE
;
2066 ofport_update_peer(port
);
2070 dp_port_name
= netdev_vport_get_dpif_port(netdev
, namebuf
, sizeof namebuf
);
2071 error
= dpif_port_query_by_name(ofproto
->backer
->dpif
, dp_port_name
,
2077 port
->odp_port
= dpif_port
.port_no
;
2079 if (netdev_get_tunnel_config(netdev
)) {
2080 atomic_count_inc(&ofproto
->backer
->tnl_count
);
2081 error
= tnl_port_add(port
, port
->up
.netdev
, port
->odp_port
,
2082 ovs_native_tunneling_is_on(ofproto
), dp_port_name
);
2084 atomic_count_dec(&ofproto
->backer
->tnl_count
);
2085 dpif_port_destroy(&dpif_port
);
2089 port
->is_tunnel
= true;
2091 /* Sanity-check that a mapping doesn't already exist. This
2092 * shouldn't happen for non-tunnel ports. */
2093 if (odp_port_to_ofp_port(ofproto
, port
->odp_port
) != OFPP_NONE
) {
2094 VLOG_ERR("port %s already has an OpenFlow port number",
2096 dpif_port_destroy(&dpif_port
);
2100 ovs_rwlock_wrlock(&ofproto
->backer
->odp_to_ofport_lock
);
2101 hmap_insert(&ofproto
->backer
->odp_to_ofport_map
, &port
->odp_port_node
,
2102 hash_odp_port(port
->odp_port
));
2103 ovs_rwlock_unlock(&ofproto
->backer
->odp_to_ofport_lock
);
2105 dpif_port_destroy(&dpif_port
);
2107 if (ofproto
->sflow
) {
2108 dpif_sflow_add_port(ofproto
->sflow
, port_
, port
->odp_port
);
2110 if (ofproto
->ipfix
) {
2111 dpif_ipfix_add_port(ofproto
->ipfix
, port_
, port
->odp_port
);
2118 port_destruct(struct ofport
*port_
, bool del
)
2120 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
2121 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
2122 const char *devname
= netdev_get_name(port
->up
.netdev
);
2123 const char *netdev_type
= netdev_get_type(port
->up
.netdev
);
2124 char namebuf
[NETDEV_VPORT_NAME_BUFSIZE
];
2125 const char *dp_port_name
;
2127 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2129 xlate_ofport_remove(port
);
2132 if (!del
&& strcmp(netdev_type
,
2133 ofproto_port_open_type(port
->up
.ofproto
, "internal"))) {
2134 /* Check if datapath requires removal of attached ports. Avoid
2135 * removal of 'internal' ports to preserve user ip/route settings. */
2136 del
= dpif_cleanup_required(ofproto
->backer
->dpif
);
2139 dp_port_name
= netdev_vport_get_dpif_port(port
->up
.netdev
, namebuf
,
2141 if (del
&& dpif_port_exists(ofproto
->backer
->dpif
, dp_port_name
)) {
2142 /* The underlying device is still there, so delete it. This
2143 * happens when the ofproto is being destroyed, since the caller
2144 * assumes that removal of attached ports will happen as part of
2146 if (!port
->is_tunnel
) {
2147 dpif_port_del(ofproto
->backer
->dpif
, port
->odp_port
, false);
2150 /* The underlying device is already deleted (e.g. tunctl -d).
2151 * Calling dpif_port_remove to do local cleanup for the netdev */
2152 if (!port
->is_tunnel
) {
2153 dpif_port_del(ofproto
->backer
->dpif
, port
->odp_port
, true);
2158 port
->peer
->peer
= NULL
;
2162 if (port
->odp_port
!= ODPP_NONE
&& !port
->is_tunnel
) {
2163 ovs_rwlock_wrlock(&ofproto
->backer
->odp_to_ofport_lock
);
2164 hmap_remove(&ofproto
->backer
->odp_to_ofport_map
, &port
->odp_port_node
);
2165 ovs_rwlock_unlock(&ofproto
->backer
->odp_to_ofport_lock
);
2168 if (port
->is_tunnel
) {
2169 atomic_count_dec(&ofproto
->backer
->tnl_count
);
2172 tnl_port_del(port
, port
->odp_port
);
2173 sset_find_and_delete(&ofproto
->ports
, devname
);
2174 sset_find_and_delete(&ofproto
->ghost_ports
, devname
);
2175 bundle_remove(port_
);
2176 set_cfm(port_
, NULL
);
2177 set_bfd(port_
, NULL
);
2178 set_lldp(port_
, NULL
);
2179 if (port
->stp_port
) {
2180 stp_port_disable(port
->stp_port
);
2182 set_rstp_port(port_
, NULL
);
2183 if (ofproto
->sflow
) {
2184 dpif_sflow_del_port(ofproto
->sflow
, port
->odp_port
);
2186 if (ofproto
->ipfix
) {
2187 dpif_ipfix_del_port(ofproto
->ipfix
, port
->odp_port
);
2194 port_modified(struct ofport
*port_
)
2196 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
2197 char namebuf
[NETDEV_VPORT_NAME_BUFSIZE
];
2198 const char *dp_port_name
;
2199 struct netdev
*netdev
= port
->up
.netdev
;
2201 if (port
->bundle
&& port
->bundle
->bond
) {
2202 bond_member_set_netdev(port
->bundle
->bond
, port
, netdev
);
2206 cfm_set_netdev(port
->cfm
, netdev
);
2210 bfd_set_netdev(port
->bfd
, netdev
);
2213 ofproto_dpif_monitor_port_update(port
, port
->bfd
, port
->cfm
,
2214 port
->lldp
, &port
->up
.pp
.hw_addr
);
2216 dp_port_name
= netdev_vport_get_dpif_port(netdev
, namebuf
, sizeof namebuf
);
2218 if (port
->is_tunnel
) {
2219 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
2221 if (tnl_port_reconfigure(port
, netdev
, port
->odp_port
, port
->odp_port
,
2222 ovs_native_tunneling_is_on(ofproto
),
2224 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2228 ofport_update_peer(port
);
2232 port_reconfigured(struct ofport
*port_
, enum ofputil_port_config old_config
)
2234 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
2235 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
2236 enum ofputil_port_config changed
= old_config
^ port
->up
.pp
.config
;
2238 if (changed
& (OFPUTIL_PC_NO_RECV
| OFPUTIL_PC_NO_RECV_STP
|
2239 OFPUTIL_PC_NO_FWD
| OFPUTIL_PC_NO_FLOOD
|
2240 OFPUTIL_PC_NO_PACKET_IN
)) {
2241 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2243 if (changed
& OFPUTIL_PC_NO_FLOOD
&& port
->bundle
) {
2244 bundle_update(port
->bundle
);
2251 set_sflow(struct ofproto
*ofproto_
,
2252 const struct ofproto_sflow_options
*sflow_options
)
2254 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2255 struct dpif_sflow
*ds
= ofproto
->sflow
;
2257 if (sflow_options
) {
2258 uint32_t old_probability
= ds
? dpif_sflow_get_probability(ds
) : 0;
2260 struct ofport_dpif
*ofport
;
2262 ds
= ofproto
->sflow
= dpif_sflow_create();
2263 HMAP_FOR_EACH (ofport
, up
.hmap_node
, &ofproto
->up
.ports
) {
2264 dpif_sflow_add_port(ds
, &ofport
->up
, ofport
->odp_port
);
2267 dpif_sflow_set_options(ds
, sflow_options
);
2268 if (dpif_sflow_get_probability(ds
) != old_probability
) {
2269 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2273 dpif_sflow_unref(ds
);
2274 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2275 ofproto
->sflow
= NULL
;
2283 struct ofproto
*ofproto_
,
2284 const struct ofproto_ipfix_bridge_exporter_options
*bridge_exporter_options
,
2285 const struct ofproto_ipfix_flow_exporter_options
*flow_exporters_options
,
2286 size_t n_flow_exporters_options
)
2288 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2289 struct dpif_ipfix
*di
= ofproto
->ipfix
;
2290 bool has_options
= bridge_exporter_options
|| flow_exporters_options
;
2291 bool new_di
= false;
2293 if (has_options
&& !di
) {
2294 di
= ofproto
->ipfix
= dpif_ipfix_create();
2299 /* Call set_options in any case to cleanly flush the flow
2300 * caches in the last exporters that are to be destroyed. */
2301 dpif_ipfix_set_options(
2302 di
, bridge_exporter_options
, flow_exporters_options
,
2303 n_flow_exporters_options
);
2305 /* Add ports only when a new ipfix created */
2306 if (new_di
== true) {
2307 struct ofport_dpif
*ofport
;
2308 HMAP_FOR_EACH (ofport
, up
.hmap_node
, &ofproto
->up
.ports
) {
2309 dpif_ipfix_add_port(di
, &ofport
->up
, ofport
->odp_port
);
2314 dpif_ipfix_unref(di
);
2315 ofproto
->ipfix
= NULL
;
2323 get_ipfix_stats(const struct ofproto
*ofproto_
,
2325 struct ovs_list
*replies
)
2327 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2328 struct dpif_ipfix
*di
= ofproto
->ipfix
;
2331 return OFPERR_NXST_NOT_CONFIGURED
;
2334 return dpif_ipfix_get_stats(di
, bridge_ipfix
, replies
);
2338 set_cfm(struct ofport
*ofport_
, const struct cfm_settings
*s
)
2340 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2341 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2342 struct cfm
*old
= ofport
->cfm
;
2347 ofport
->cfm
= cfm_create(ofport
->up
.netdev
);
2350 if (cfm_configure(ofport
->cfm
, s
)) {
2357 cfm_unref(ofport
->cfm
);
2360 if (ofport
->cfm
!= old
) {
2361 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2363 ofproto_dpif_monitor_port_update(ofport
, ofport
->bfd
, ofport
->cfm
,
2364 ofport
->lldp
, &ofport
->up
.pp
.hw_addr
);
2369 cfm_status_changed(struct ofport
*ofport_
)
2371 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2373 return ofport
->cfm
? cfm_check_status_change(ofport
->cfm
) : true;
2377 get_cfm_status(const struct ofport
*ofport_
,
2378 struct cfm_status
*status
)
2380 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2384 cfm_get_status(ofport
->cfm
, status
);
2393 set_bfd(struct ofport
*ofport_
, const struct smap
*cfg
)
2395 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport_
->ofproto
);
2396 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2400 ofport
->bfd
= bfd_configure(old
, netdev_get_name(ofport
->up
.netdev
),
2401 cfg
, ofport
->up
.netdev
);
2402 if (ofport
->bfd
!= old
) {
2403 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2405 ofproto_dpif_monitor_port_update(ofport
, ofport
->bfd
, ofport
->cfm
,
2406 ofport
->lldp
, &ofport
->up
.pp
.hw_addr
);
2411 bfd_status_changed(struct ofport
*ofport_
)
2413 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2415 return ofport
->bfd
? bfd_check_status_change(ofport
->bfd
) : true;
2419 get_bfd_status(struct ofport
*ofport_
, struct smap
*smap
)
2421 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2425 bfd_get_status(ofport
->bfd
, smap
);
2434 set_lldp(struct ofport
*ofport_
,
2435 const struct smap
*cfg
)
2437 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2438 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2442 if (!ofport
->lldp
) {
2443 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2444 ofport
->lldp
= lldp_create(ofport
->up
.netdev
, ofport_
->mtu
, cfg
);
2447 if (!lldp_configure(ofport
->lldp
, cfg
)) {
2448 lldp_unref(ofport
->lldp
);
2449 ofport
->lldp
= NULL
;
2452 } else if (ofport
->lldp
) {
2453 lldp_unref(ofport
->lldp
);
2454 ofport
->lldp
= NULL
;
2455 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2458 ofproto_dpif_monitor_port_update(ofport
,
2462 &ofport
->up
.pp
.hw_addr
);
2467 get_lldp_status(const struct ofport
*ofport_
,
2468 struct lldp_status
*status OVS_UNUSED
)
2470 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2472 return ofport
->lldp
? true : false;
2476 set_aa(struct ofproto
*ofproto OVS_UNUSED
,
2477 const struct aa_settings
*s
)
2479 return aa_configure(s
);
2483 aa_mapping_set(struct ofproto
*ofproto_ OVS_UNUSED
, void *aux
,
2484 const struct aa_mapping_settings
*s
)
2486 return aa_mapping_register(aux
, s
);
2490 aa_mapping_unset(struct ofproto
*ofproto OVS_UNUSED
, void *aux
)
2492 return aa_mapping_unregister(aux
);
2496 aa_vlan_get_queued(struct ofproto
*ofproto OVS_UNUSED
, struct ovs_list
*list
)
2498 return aa_get_vlan_queued(list
);
2502 aa_vlan_get_queue_size(struct ofproto
*ofproto OVS_UNUSED
)
2504 return aa_get_vlan_queue_size();
2508 /* Spanning Tree. */
2510 /* Called while rstp_mutex is held. */
2512 rstp_send_bpdu_cb(struct dp_packet
*pkt
, void *ofport_
, void *ofproto_
)
2514 struct ofproto_dpif
*ofproto
= ofproto_
;
2515 struct ofport_dpif
*ofport
= ofport_
;
2516 struct eth_header
*eth
= dp_packet_eth(pkt
);
2518 netdev_get_etheraddr(ofport
->up
.netdev
, ð
->eth_src
);
2519 if (eth_addr_is_zero(eth
->eth_src
)) {
2520 VLOG_WARN_RL(&rl
, "%s port %d: cannot send RSTP BPDU on a port which "
2521 "does not have a configured source MAC address.",
2522 ofproto
->up
.name
, ofp_to_u16(ofport
->up
.ofp_port
));
2524 ofproto_dpif_send_packet(ofport
, false, pkt
);
2526 dp_packet_delete(pkt
);
2530 send_bpdu_cb(struct dp_packet
*pkt
, int port_num
, void *ofproto_
)
2532 struct ofproto_dpif
*ofproto
= ofproto_
;
2533 struct stp_port
*sp
= stp_get_port(ofproto
->stp
, port_num
);
2534 struct ofport_dpif
*ofport
;
2536 ofport
= stp_port_get_aux(sp
);
2538 VLOG_WARN_RL(&rl
, "%s: cannot send BPDU on unknown port %d",
2539 ofproto
->up
.name
, port_num
);
2541 struct eth_header
*eth
= dp_packet_eth(pkt
);
2543 netdev_get_etheraddr(ofport
->up
.netdev
, ð
->eth_src
);
2544 if (eth_addr_is_zero(eth
->eth_src
)) {
2545 VLOG_WARN_RL(&rl
, "%s: cannot send BPDU on port %d "
2546 "with unknown MAC", ofproto
->up
.name
, port_num
);
2548 ofproto_dpif_send_packet(ofport
, false, pkt
);
2551 dp_packet_delete(pkt
);
2554 /* Configure RSTP on 'ofproto_' using the settings defined in 's'. */
2556 set_rstp(struct ofproto
*ofproto_
, const struct ofproto_rstp_settings
*s
)
2558 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2560 /* Only revalidate flows if the configuration changed. */
2561 if (!s
!= !ofproto
->rstp
) {
2562 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2566 if (!ofproto
->rstp
) {
2567 ofproto
->rstp
= rstp_create(ofproto_
->name
, s
->address
,
2568 rstp_send_bpdu_cb
, ofproto
);
2569 ofproto
->rstp_last_tick
= time_msec();
2571 rstp_set_bridge_address(ofproto
->rstp
, s
->address
);
2572 rstp_set_bridge_priority(ofproto
->rstp
, s
->priority
);
2573 rstp_set_bridge_ageing_time(ofproto
->rstp
, s
->ageing_time
);
2574 rstp_set_bridge_force_protocol_version(ofproto
->rstp
,
2575 s
->force_protocol_version
);
2576 rstp_set_bridge_max_age(ofproto
->rstp
, s
->bridge_max_age
);
2577 rstp_set_bridge_forward_delay(ofproto
->rstp
, s
->bridge_forward_delay
);
2578 rstp_set_bridge_transmit_hold_count(ofproto
->rstp
,
2579 s
->transmit_hold_count
);
2581 struct ofport
*ofport
;
2582 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->up
.ports
) {
2583 set_rstp_port(ofport
, NULL
);
2585 rstp_unref(ofproto
->rstp
);
2586 ofproto
->rstp
= NULL
;
2591 get_rstp_status(struct ofproto
*ofproto_
, struct ofproto_rstp_status
*s
)
2593 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2595 if (ofproto
->rstp
) {
2597 s
->root_id
= rstp_get_root_id(ofproto
->rstp
);
2598 s
->bridge_id
= rstp_get_bridge_id(ofproto
->rstp
);
2599 s
->designated_id
= rstp_get_designated_id(ofproto
->rstp
);
2600 s
->root_path_cost
= rstp_get_root_path_cost(ofproto
->rstp
);
2601 s
->designated_port_id
= rstp_get_designated_port_id(ofproto
->rstp
);
2602 s
->bridge_port_id
= rstp_get_bridge_port_id(ofproto
->rstp
);
2609 update_rstp_port_state(struct ofport_dpif
*ofport
)
2611 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2612 enum rstp_state state
;
2614 /* Figure out new state. */
2615 state
= ofport
->rstp_port
? rstp_port_get_state(ofport
->rstp_port
)
2619 if (ofport
->rstp_state
!= state
) {
2620 enum ofputil_port_state of_state
;
2623 VLOG_DBG("port %s: RSTP state changed from %s to %s",
2624 netdev_get_name(ofport
->up
.netdev
),
2625 rstp_state_name(ofport
->rstp_state
),
2626 rstp_state_name(state
));
2628 if (rstp_learn_in_state(ofport
->rstp_state
)
2629 != rstp_learn_in_state(state
)) {
2630 /* XXX: Learning action flows should also be flushed. */
2631 if (ofport
->bundle
) {
2632 if (!rstp_shift_root_learned_address(ofproto
->rstp
)
2633 || rstp_get_old_root_aux(ofproto
->rstp
) != ofport
) {
2634 bundle_flush_macs(ofport
->bundle
, false);
2638 fwd_change
= rstp_forward_in_state(ofport
->rstp_state
)
2639 != rstp_forward_in_state(state
);
2641 ofproto
->backer
->need_revalidate
= REV_RSTP
;
2642 ofport
->rstp_state
= state
;
2644 if (fwd_change
&& ofport
->bundle
) {
2645 bundle_update(ofport
->bundle
);
2648 /* Update the RSTP state bits in the OpenFlow port description. */
2649 of_state
= ofport
->up
.pp
.state
& ~OFPUTIL_PS_STP_MASK
;
2650 of_state
|= (state
== RSTP_LEARNING
? OFPUTIL_PS_STP_LEARN
2651 : state
== RSTP_FORWARDING
? OFPUTIL_PS_STP_FORWARD
2652 : state
== RSTP_DISCARDING
? OFPUTIL_PS_STP_LISTEN
2654 ofproto_port_set_state(&ofport
->up
, of_state
);
2659 rstp_run(struct ofproto_dpif
*ofproto
)
2661 if (ofproto
->rstp
) {
2662 long long int now
= time_msec();
2663 long long int elapsed
= now
- ofproto
->rstp_last_tick
;
2664 struct rstp_port
*rp
;
2665 struct ofport_dpif
*ofport
;
2667 /* Every second, decrease the values of the timers. */
2668 if (elapsed
>= 1000) {
2669 rstp_tick_timers(ofproto
->rstp
);
2670 ofproto
->rstp_last_tick
= now
;
2673 while ((ofport
= rstp_get_next_changed_port_aux(ofproto
->rstp
, &rp
))) {
2674 update_rstp_port_state(ofport
);
2678 /* FIXME: This check should be done on-event (i.e., when setting
2679 * p->fdb_flush) and not periodically.
2681 while ((ofport
= rstp_check_and_reset_fdb_flush(ofproto
->rstp
, &rp
))) {
2682 if (!rstp_shift_root_learned_address(ofproto
->rstp
)
2683 || rstp_get_old_root_aux(ofproto
->rstp
) != ofport
) {
2684 bundle_flush_macs(ofport
->bundle
, false);
2688 if (rstp_shift_root_learned_address(ofproto
->rstp
)) {
2689 struct ofport_dpif
*old_root_aux
=
2690 (struct ofport_dpif
*)rstp_get_old_root_aux(ofproto
->rstp
);
2691 struct ofport_dpif
*new_root_aux
=
2692 (struct ofport_dpif
*)rstp_get_new_root_aux(ofproto
->rstp
);
2693 if (old_root_aux
!= NULL
&& new_root_aux
!= NULL
) {
2694 bundle_move(old_root_aux
->bundle
, new_root_aux
->bundle
);
2695 rstp_reset_root_changed(ofproto
->rstp
);
2701 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2703 set_stp(struct ofproto
*ofproto_
, const struct ofproto_stp_settings
*s
)
2705 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2707 /* Only revalidate flows if the configuration changed. */
2708 if (!s
!= !ofproto
->stp
) {
2709 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2713 if (!ofproto
->stp
) {
2714 ofproto
->stp
= stp_create(ofproto_
->name
, s
->system_id
,
2715 send_bpdu_cb
, ofproto
);
2716 ofproto
->stp_last_tick
= time_msec();
2719 stp_set_bridge_id(ofproto
->stp
, s
->system_id
);
2720 stp_set_bridge_priority(ofproto
->stp
, s
->priority
);
2721 stp_set_hello_time(ofproto
->stp
, s
->hello_time
);
2722 stp_set_max_age(ofproto
->stp
, s
->max_age
);
2723 stp_set_forward_delay(ofproto
->stp
, s
->fwd_delay
);
2725 struct ofport
*ofport
;
2727 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->up
.ports
) {
2728 set_stp_port(ofport
, NULL
);
2731 stp_unref(ofproto
->stp
);
2732 ofproto
->stp
= NULL
;
2739 get_stp_status(struct ofproto
*ofproto_
, struct ofproto_stp_status
*s
)
2741 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2745 s
->bridge_id
= stp_get_bridge_id(ofproto
->stp
);
2746 s
->designated_root
= stp_get_designated_root(ofproto
->stp
);
2747 s
->root_path_cost
= stp_get_root_path_cost(ofproto
->stp
);
2756 update_stp_port_state(struct ofport_dpif
*ofport
)
2758 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2759 enum stp_state state
;
2761 /* Figure out new state. */
2762 state
= ofport
->stp_port
? stp_port_get_state(ofport
->stp_port
)
2766 if (ofport
->stp_state
!= state
) {
2767 enum ofputil_port_state of_state
;
2770 VLOG_DBG("port %s: STP state changed from %s to %s",
2771 netdev_get_name(ofport
->up
.netdev
),
2772 stp_state_name(ofport
->stp_state
),
2773 stp_state_name(state
));
2774 if (stp_learn_in_state(ofport
->stp_state
)
2775 != stp_learn_in_state(state
)) {
2776 /* xxx Learning action flows should also be flushed. */
2777 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
2778 mac_learning_flush(ofproto
->ml
);
2779 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
2780 mcast_snooping_mdb_flush(ofproto
->ms
);
2782 fwd_change
= stp_forward_in_state(ofport
->stp_state
)
2783 != stp_forward_in_state(state
);
2785 ofproto
->backer
->need_revalidate
= REV_STP
;
2786 ofport
->stp_state
= state
;
2787 ofport
->stp_state_entered
= time_msec();
2789 if (fwd_change
&& ofport
->bundle
) {
2790 bundle_update(ofport
->bundle
);
2793 /* Update the STP state bits in the OpenFlow port description. */
2794 of_state
= ofport
->up
.pp
.state
& ~OFPUTIL_PS_STP_MASK
;
2795 of_state
|= (state
== STP_LISTENING
? OFPUTIL_PS_STP_LISTEN
2796 : state
== STP_LEARNING
? OFPUTIL_PS_STP_LEARN
2797 : state
== STP_FORWARDING
? OFPUTIL_PS_STP_FORWARD
2798 : state
== STP_BLOCKING
? OFPUTIL_PS_STP_BLOCK
2800 ofproto_port_set_state(&ofport
->up
, of_state
);
2805 stp_check_and_update_link_state(struct ofproto_dpif
*ofproto
)
2807 struct ofport_dpif
*ofport
;
2809 HMAP_FOR_EACH (ofport
, up
.hmap_node
, &ofproto
->up
.ports
) {
2810 bool up
= netdev_get_carrier(ofport
->up
.netdev
);
2812 if (ofport
->stp_port
&&
2813 up
!= (stp_port_get_state(ofport
->stp_port
) != STP_DISABLED
)) {
2815 VLOG_DBG("bridge %s, port %s is %s, %s it.",
2816 ofproto
->up
.name
, netdev_get_name(ofport
->up
.netdev
),
2818 up
? "enabling" : "disabling");
2821 stp_port_enable(ofport
->stp_port
);
2822 stp_port_set_aux(ofport
->stp_port
, ofport
);
2824 stp_port_disable(ofport
->stp_port
);
2827 update_stp_port_state(ofport
);
2832 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2833 * caller is responsible for assigning STP port numbers and ensuring
2834 * there are no duplicates. */
2836 set_stp_port(struct ofport
*ofport_
,
2837 const struct ofproto_port_stp_settings
*s
)
2839 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2840 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2841 struct stp_port
*sp
= ofport
->stp_port
;
2843 if (!s
|| !s
->enable
) {
2845 ofport
->stp_port
= NULL
;
2846 stp_port_disable(sp
);
2847 update_stp_port_state(ofport
);
2850 } else if (sp
&& stp_port_no(sp
) != s
->port_num
2851 && ofport
== stp_port_get_aux(sp
)) {
2852 /* The port-id changed, so disable the old one if it's not
2853 * already in use by another port. */
2854 stp_port_disable(sp
);
2857 sp
= ofport
->stp_port
= stp_get_port(ofproto
->stp
, s
->port_num
);
2859 /* Set name before enabling the port so that debugging messages can print
2861 stp_port_set_name(sp
, netdev_get_name(ofport
->up
.netdev
));
2863 if (netdev_get_carrier(ofport_
->netdev
)) {
2864 stp_port_enable(sp
);
2866 stp_port_disable(sp
);
2869 stp_port_set_aux(sp
, ofport
);
2870 stp_port_set_priority(sp
, s
->priority
);
2871 stp_port_set_path_cost(sp
, s
->path_cost
);
2873 update_stp_port_state(ofport
);
2879 get_stp_port_status(struct ofport
*ofport_
,
2880 struct ofproto_port_stp_status
*s
)
2882 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2883 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2884 struct stp_port
*sp
= ofport
->stp_port
;
2886 if (!ofproto
->stp
|| !sp
) {
2892 stp_port_get_status(sp
, &s
->port_id
, &s
->state
, &s
->role
);
2893 s
->sec_in_state
= (time_msec() - ofport
->stp_state_entered
) / 1000;
2899 get_stp_port_stats(struct ofport
*ofport_
,
2900 struct ofproto_port_stp_stats
*s
)
2902 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2903 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2904 struct stp_port
*sp
= ofport
->stp_port
;
2906 if (!ofproto
->stp
|| !sp
) {
2912 stp_port_get_counts(sp
, &s
->tx_count
, &s
->rx_count
, &s
->error_count
);
2918 stp_run(struct ofproto_dpif
*ofproto
)
2921 long long int now
= time_msec();
2922 long long int elapsed
= now
- ofproto
->stp_last_tick
;
2923 struct stp_port
*sp
;
2926 stp_tick(ofproto
->stp
, MIN(INT_MAX
, elapsed
));
2927 ofproto
->stp_last_tick
= now
;
2930 stp_check_and_update_link_state(ofproto
);
2932 while (stp_get_changed_port(ofproto
->stp
, &sp
)) {
2933 struct ofport_dpif
*ofport
= stp_port_get_aux(sp
);
2936 update_stp_port_state(ofport
);
2940 if (stp_check_and_reset_fdb_flush(ofproto
->stp
)) {
2941 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
2942 mac_learning_flush(ofproto
->ml
);
2943 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
2944 mcast_snooping_mdb_flush(ofproto
->ms
);
2950 stp_wait(struct ofproto_dpif
*ofproto
)
2953 poll_timer_wait(1000);
2957 /* Configures RSTP on 'ofport_' using the settings defined in 's'. The
2958 * caller is responsible for assigning RSTP port numbers and ensuring
2959 * there are no duplicates. */
2961 set_rstp_port(struct ofport
*ofport_
,
2962 const struct ofproto_port_rstp_settings
*s
)
2964 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2965 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2966 struct rstp_port
*rp
= ofport
->rstp_port
;
2968 if (!s
|| !s
->enable
) {
2970 rstp_port_set_aux(rp
, NULL
);
2971 rstp_port_set_state(rp
, RSTP_DISABLED
);
2972 rstp_port_set_mac_operational(rp
, false);
2973 ofport
->rstp_port
= NULL
;
2974 rstp_port_unref(rp
);
2975 update_rstp_port_state(ofport
);
2980 /* Check if need to add a new port. */
2982 rp
= ofport
->rstp_port
= rstp_add_port(ofproto
->rstp
);
2985 rstp_port_set(rp
, s
->port_num
, s
->priority
, s
->path_cost
,
2986 s
->admin_edge_port
, s
->auto_edge
,
2987 s
->admin_p2p_mac_state
, s
->admin_port_state
, s
->mcheck
,
2988 ofport
, netdev_get_name(ofport
->up
.netdev
));
2989 update_rstp_port_state(ofport
);
2990 /* Synchronize operational status. */
2991 rstp_port_set_mac_operational(rp
, ofport
->up
.may_enable
);
2995 get_rstp_port_status(struct ofport
*ofport_
,
2996 struct ofproto_port_rstp_status
*s
)
2998 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2999 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
3000 struct rstp_port
*rp
= ofport
->rstp_port
;
3002 if (!ofproto
->rstp
|| !rp
) {
3008 rstp_port_get_status(rp
, &s
->port_id
, &s
->state
, &s
->role
,
3009 &s
->designated_bridge_id
, &s
->designated_port_id
,
3010 &s
->designated_path_cost
, &s
->tx_count
,
3011 &s
->rx_count
, &s
->error_count
, &s
->uptime
);
3016 set_queues(struct ofport
*ofport_
, const struct ofproto_port_queue
*qdscp
,
3019 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
3020 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
3022 if (ofport
->n_qdscp
!= n_qdscp
3023 || (n_qdscp
&& memcmp(ofport
->qdscp
, qdscp
,
3024 n_qdscp
* sizeof *qdscp
))) {
3025 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3026 free(ofport
->qdscp
);
3027 ofport
->qdscp
= n_qdscp
3028 ? xmemdup(qdscp
, n_qdscp
* sizeof *qdscp
)
3030 ofport
->n_qdscp
= n_qdscp
;
3038 /* Expires all MAC learning entries associated with 'bundle' and forces its
3039 * ofproto to revalidate every flow.
3041 * Normally MAC learning entries are removed only from the ofproto associated
3042 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
3043 * are removed from every ofproto. When patch ports and SLB bonds are in use
3044 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
3045 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
3046 * with the host from which it migrated. */
3048 bundle_flush_macs(struct ofbundle
*bundle
, bool all_ofprotos
)
3050 struct ofproto_dpif
*ofproto
= bundle
->ofproto
;
3051 struct mac_learning
*ml
= ofproto
->ml
;
3052 struct mac_entry
*mac
, *next_mac
;
3054 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3055 ovs_rwlock_wrlock(&ml
->rwlock
);
3056 LIST_FOR_EACH_SAFE (mac
, next_mac
, lru_node
, &ml
->lrus
) {
3057 if (mac_entry_get_port(ml
, mac
) == bundle
) {
3059 struct ofproto_dpif
*o
;
3061 HMAP_FOR_EACH (o
, all_ofproto_dpifs_by_name_node
,
3062 &all_ofproto_dpifs_by_name
) {
3064 struct mac_entry
*e
;
3066 ovs_rwlock_wrlock(&o
->ml
->rwlock
);
3067 e
= mac_learning_lookup(o
->ml
, mac
->mac
, mac
->vlan
);
3069 mac_learning_expire(o
->ml
, e
);
3071 ovs_rwlock_unlock(&o
->ml
->rwlock
);
3076 mac_learning_expire(ml
, mac
);
3079 ovs_rwlock_unlock(&ml
->rwlock
);
3083 bundle_move(struct ofbundle
*old
, struct ofbundle
*new)
3085 struct ofproto_dpif
*ofproto
= old
->ofproto
;
3086 struct mac_learning
*ml
= ofproto
->ml
;
3087 struct mac_entry
*mac
, *next_mac
;
3089 ovs_assert(new->ofproto
== old
->ofproto
);
3091 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3092 ovs_rwlock_wrlock(&ml
->rwlock
);
3093 LIST_FOR_EACH_SAFE (mac
, next_mac
, lru_node
, &ml
->lrus
) {
3094 if (mac_entry_get_port(ml
, mac
) == old
) {
3095 mac_entry_set_port(ml
, mac
, new);
3098 ovs_rwlock_unlock(&ml
->rwlock
);
3101 static struct ofbundle
*
3102 bundle_lookup(const struct ofproto_dpif
*ofproto
, void *aux
)
3104 struct ofbundle
*bundle
;
3106 HMAP_FOR_EACH_IN_BUCKET (bundle
, hmap_node
, hash_pointer(aux
, 0),
3107 &ofproto
->bundles
) {
3108 if (bundle
->aux
== aux
) {
3116 bundle_update(struct ofbundle
*bundle
)
3118 struct ofport_dpif
*port
;
3120 bundle
->floodable
= true;
3121 LIST_FOR_EACH (port
, bundle_node
, &bundle
->ports
) {
3122 if (port
->up
.pp
.config
& OFPUTIL_PC_NO_FLOOD
3123 || netdev_get_pt_mode(port
->up
.netdev
) == NETDEV_PT_LEGACY_L3
3124 || (bundle
->ofproto
->stp
&& !stp_forward_in_state(port
->stp_state
))
3125 || (bundle
->ofproto
->rstp
&& !rstp_forward_in_state(port
->rstp_state
))) {
3126 bundle
->floodable
= false;
3133 bundle_del_port(struct ofport_dpif
*port
)
3135 struct ofbundle
*bundle
= port
->bundle
;
3137 bundle
->ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3139 ovs_list_remove(&port
->bundle_node
);
3140 port
->bundle
= NULL
;
3143 lacp_member_unregister(bundle
->lacp
, port
);
3146 bond_member_unregister(bundle
->bond
, port
);
3149 bundle_update(bundle
);
3153 bundle_add_port(struct ofbundle
*bundle
, ofp_port_t ofp_port
,
3154 struct lacp_member_settings
*lacp
)
3156 struct ofport_dpif
*port
;
3158 port
= ofp_port_to_ofport(bundle
->ofproto
, ofp_port
);
3163 if (port
->bundle
!= bundle
) {
3164 bundle
->ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3166 bundle_remove(&port
->up
);
3169 port
->bundle
= bundle
;
3170 ovs_list_push_back(&bundle
->ports
, &port
->bundle_node
);
3171 if (port
->up
.pp
.config
& OFPUTIL_PC_NO_FLOOD
3172 || netdev_get_pt_mode(port
->up
.netdev
) == NETDEV_PT_LEGACY_L3
3173 || (bundle
->ofproto
->stp
&& !stp_forward_in_state(port
->stp_state
))
3174 || (bundle
->ofproto
->rstp
&& !rstp_forward_in_state(port
->rstp_state
))) {
3175 bundle
->floodable
= false;
3179 bundle
->ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3180 lacp_member_register(bundle
->lacp
, port
, lacp
);
3187 bundle_destroy(struct ofbundle
*bundle
)
3189 struct ofproto_dpif
*ofproto
;
3190 struct ofport_dpif
*port
, *next_port
;
3196 ofproto
= bundle
->ofproto
;
3197 mbridge_unregister_bundle(ofproto
->mbridge
, bundle
);
3200 xlate_bundle_remove(bundle
);
3203 LIST_FOR_EACH_SAFE (port
, next_port
, bundle_node
, &bundle
->ports
) {
3204 bundle_del_port(port
);
3207 bundle_flush_macs(bundle
, true);
3208 mcast_snooping_flush_bundle(ofproto
->ms
, bundle
);
3209 hmap_remove(&ofproto
->bundles
, &bundle
->hmap_node
);
3211 free(bundle
->trunks
);
3212 free(bundle
->cvlans
);
3213 lacp_unref(bundle
->lacp
);
3214 bond_unref(bundle
->bond
);
3219 bundle_set(struct ofproto
*ofproto_
, void *aux
,
3220 const struct ofproto_bundle_settings
*s
)
3222 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3223 bool need_flush
= false;
3224 struct ofport_dpif
*port
;
3225 struct ofbundle
*bundle
;
3226 unsigned long *trunks
= NULL
;
3227 unsigned long *cvlans
= NULL
;
3232 bundle
= bundle_lookup(ofproto
, aux
);
3235 bundle_destroy(bundle
);
3239 ovs_assert(s
->n_members
== 1 || s
->bond
!= NULL
);
3240 ovs_assert((s
->lacp
!= NULL
) == (s
->lacp_members
!= NULL
));
3243 bundle
= xmalloc(sizeof *bundle
);
3245 bundle
->ofproto
= ofproto
;
3246 hmap_insert(&ofproto
->bundles
, &bundle
->hmap_node
,
3247 hash_pointer(aux
, 0));
3249 bundle
->name
= NULL
;
3251 ovs_list_init(&bundle
->ports
);
3252 bundle
->vlan_mode
= PORT_VLAN_TRUNK
;
3253 bundle
->qinq_ethtype
= ETH_TYPE_VLAN_8021AD
;
3255 bundle
->trunks
= NULL
;
3256 bundle
->cvlans
= NULL
;
3257 bundle
->use_priority_tags
= s
->use_priority_tags
;
3258 bundle
->lacp
= NULL
;
3259 bundle
->bond
= NULL
;
3261 bundle
->floodable
= true;
3262 bundle
->protected = false;
3263 mbridge_register_bundle(ofproto
->mbridge
, bundle
);
3266 if (!bundle
->name
|| strcmp(s
->name
, bundle
->name
)) {
3268 bundle
->name
= xstrdup(s
->name
);
3273 ofproto
->lacp_enabled
= true;
3274 if (!bundle
->lacp
) {
3275 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3276 bundle
->lacp
= lacp_create();
3278 lacp_configure(bundle
->lacp
, s
->lacp
);
3280 lacp_unref(bundle
->lacp
);
3281 bundle
->lacp
= NULL
;
3284 /* Update set of ports. */
3286 for (i
= 0; i
< s
->n_members
; i
++) {
3287 if (!bundle_add_port(bundle
, s
->members
[i
],
3288 s
->lacp
? &s
->lacp_members
[i
] : NULL
)) {
3292 if (!ok
|| ovs_list_size(&bundle
->ports
) != s
->n_members
) {
3293 struct ofport_dpif
*next_port
;
3295 LIST_FOR_EACH_SAFE (port
, next_port
, bundle_node
, &bundle
->ports
) {
3296 for (i
= 0; i
< s
->n_members
; i
++) {
3297 if (s
->members
[i
] == port
->up
.ofp_port
) {
3302 bundle_del_port(port
);
3306 ovs_assert(ovs_list_size(&bundle
->ports
) <= s
->n_members
);
3308 if (ovs_list_is_empty(&bundle
->ports
)) {
3309 bundle_destroy(bundle
);
3313 /* Set VLAN tagging mode */
3314 if (s
->vlan_mode
!= bundle
->vlan_mode
3315 || s
->use_priority_tags
!= bundle
->use_priority_tags
) {
3316 bundle
->vlan_mode
= s
->vlan_mode
;
3317 bundle
->use_priority_tags
= s
->use_priority_tags
;
3321 if (s
->qinq_ethtype
!= bundle
->qinq_ethtype
) {
3322 bundle
->qinq_ethtype
= s
->qinq_ethtype
;
3327 vlan
= (s
->vlan_mode
== PORT_VLAN_TRUNK
? -1
3328 : s
->vlan
>= 0 && s
->vlan
<= 4095 ? s
->vlan
3330 if (vlan
!= bundle
->vlan
) {
3331 bundle
->vlan
= vlan
;
3335 /* Get trunked VLANs. */
3336 switch (s
->vlan_mode
) {
3337 case PORT_VLAN_ACCESS
:
3341 case PORT_VLAN_TRUNK
:
3342 trunks
= CONST_CAST(unsigned long *, s
->trunks
);
3345 case PORT_VLAN_NATIVE_UNTAGGED
:
3346 case PORT_VLAN_NATIVE_TAGGED
:
3347 if (vlan
!= 0 && (!s
->trunks
3348 || !bitmap_is_set(s
->trunks
, vlan
)
3349 || bitmap_is_set(s
->trunks
, 0))) {
3350 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
3352 trunks
= bitmap_clone(s
->trunks
, 4096);
3354 trunks
= bitmap_allocate1(4096);
3356 bitmap_set1(trunks
, vlan
);
3357 bitmap_set0(trunks
, 0);
3359 trunks
= CONST_CAST(unsigned long *, s
->trunks
);
3363 case PORT_VLAN_DOT1Q_TUNNEL
:
3364 cvlans
= CONST_CAST(unsigned long *, s
->cvlans
);
3370 if (!vlan_bitmap_equal(trunks
, bundle
->trunks
)) {
3371 free(bundle
->trunks
);
3372 if (trunks
== s
->trunks
) {
3373 bundle
->trunks
= vlan_bitmap_clone(trunks
);
3375 bundle
->trunks
= trunks
;
3380 if (trunks
!= s
->trunks
) {
3384 if (!vlan_bitmap_equal(cvlans
, bundle
->cvlans
)) {
3385 free(bundle
->cvlans
);
3386 if (cvlans
== s
->cvlans
) {
3387 bundle
->cvlans
= vlan_bitmap_clone(cvlans
);
3389 bundle
->cvlans
= cvlans
;
3394 if (cvlans
!= s
->cvlans
) {
3399 if (!ovs_list_is_short(&bundle
->ports
)) {
3400 bundle
->ofproto
->has_bonded_bundles
= true;
3402 if (bond_reconfigure(bundle
->bond
, s
->bond
)) {
3403 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3406 bundle
->bond
= bond_create(s
->bond
, ofproto
);
3407 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3410 LIST_FOR_EACH (port
, bundle_node
, &bundle
->ports
) {
3411 bond_member_register(bundle
->bond
, port
,
3412 port
->up
.ofp_port
, port
->up
.netdev
);
3415 bond_unref(bundle
->bond
);
3416 bundle
->bond
= NULL
;
3419 /* Set proteced port mode */
3420 if (s
->protected != bundle
->protected) {
3421 bundle
->protected = s
->protected;
3425 /* If we changed something that would affect MAC learning, un-learn
3426 * everything on this port and force flow revalidation. */
3428 bundle_flush_macs(bundle
, false);
3429 mcast_snooping_flush_bundle(ofproto
->ms
, bundle
);
3436 bundle_remove(struct ofport
*port_
)
3438 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
3439 struct ofbundle
*bundle
= port
->bundle
;
3442 bundle_del_port(port
);
3443 if (ovs_list_is_empty(&bundle
->ports
)) {
3444 bundle_destroy(bundle
);
3445 } else if (ovs_list_is_short(&bundle
->ports
)) {
3446 bond_unref(bundle
->bond
);
3447 bundle
->bond
= NULL
;
3453 ofproto_dpif_add_lb_output_buckets(struct ofproto_dpif
*ofproto
,
3455 const ofp_port_t
*slave_map
)
3457 odp_port_t odp_map
[BOND_BUCKETS
];
3459 for (int bucket
= 0; bucket
< BOND_BUCKETS
; bucket
++) {
3460 /* Convert ofp_port to odp_port. */
3461 odp_map
[bucket
] = ofp_port_to_odp_port(ofproto
, slave_map
[bucket
]);
3463 return dpif_bond_add(ofproto
->backer
->dpif
, bond_id
, odp_map
);
3467 ofproto_dpif_delete_lb_output_buckets(struct ofproto_dpif
*ofproto
,
3470 return dpif_bond_del(ofproto
->backer
->dpif
, bond_id
);
3474 send_pdu_cb(void *port_
, const void *pdu
, size_t pdu_size
)
3476 struct ofport_dpif
*port
= port_
;
3480 error
= netdev_get_etheraddr(port
->up
.netdev
, &ea
);
3482 struct dp_packet packet
;
3485 dp_packet_init(&packet
, 0);
3486 packet_pdu
= eth_compose(&packet
, eth_addr_lacp
, ea
, ETH_TYPE_LACP
,
3488 memcpy(packet_pdu
, pdu
, pdu_size
);
3490 error
= ofproto_dpif_send_packet(port
, false, &packet
);
3492 VLOG_WARN_RL(&rl
, "port %s: cannot transmit LACP PDU (%s).",
3493 port
->bundle
->name
, ovs_strerror(error
));
3495 dp_packet_uninit(&packet
);
3497 static struct vlog_rate_limit rll
= VLOG_RATE_LIMIT_INIT(1, 10);
3498 VLOG_ERR_RL(&rll
, "port %s: cannot obtain Ethernet address of iface "
3499 "%s (%s)", port
->bundle
->name
,
3500 netdev_get_name(port
->up
.netdev
), ovs_strerror(error
));
3505 bundle_send_learning_packets(struct ofbundle
*bundle
)
3507 struct ofproto_dpif
*ofproto
= bundle
->ofproto
;
3508 int error
, n_packets
, n_errors
;
3509 struct mac_entry
*e
;
3511 struct ovs_list list_node
;
3512 struct ofport_dpif
*port
;
3513 struct dp_packet
*pkt
;
3515 struct ovs_list packets
;
3517 ovs_list_init(&packets
);
3518 ovs_rwlock_rdlock(&ofproto
->ml
->rwlock
);
3519 LIST_FOR_EACH (e
, lru_node
, &ofproto
->ml
->lrus
) {
3520 if (mac_entry_get_port(ofproto
->ml
, e
) != bundle
) {
3521 pkt_node
= xmalloc(sizeof *pkt_node
);
3522 pkt_node
->pkt
= bond_compose_learning_packet(bundle
->bond
,
3524 (void **)&pkt_node
->port
);
3525 ovs_list_push_back(&packets
, &pkt_node
->list_node
);
3528 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
3530 error
= n_packets
= n_errors
= 0;
3531 LIST_FOR_EACH_POP (pkt_node
, list_node
, &packets
) {
3534 ret
= ofproto_dpif_send_packet(pkt_node
->port
, false, pkt_node
->pkt
);
3535 dp_packet_delete(pkt_node
->pkt
);
3545 static struct vlog_rate_limit rll
= VLOG_RATE_LIMIT_INIT(1, 5);
3546 VLOG_WARN_RL(&rll
, "bond %s: %d errors sending %d gratuitous learning "
3547 "packets, last error was: %s",
3548 bundle
->name
, n_errors
, n_packets
, ovs_strerror(error
));
3550 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3551 bundle
->name
, n_packets
);
3556 bundle_run(struct ofbundle
*bundle
)
3559 lacp_run(bundle
->lacp
, send_pdu_cb
);
3562 struct ofport_dpif
*port
;
3564 LIST_FOR_EACH (port
, bundle_node
, &bundle
->ports
) {
3565 bond_member_set_may_enable(bundle
->bond
, port
, port
->up
.may_enable
);
3568 if (bond_run(bundle
->bond
, lacp_status(bundle
->lacp
))) {
3569 bundle
->ofproto
->backer
->need_revalidate
= REV_BOND
;
3572 if (bond_should_send_learning_packets(bundle
->bond
)) {
3573 bundle_send_learning_packets(bundle
);
3579 bundle_wait(struct ofbundle
*bundle
)
3582 lacp_wait(bundle
->lacp
);
3585 bond_wait(bundle
->bond
);
3592 mirror_set__(struct ofproto
*ofproto_
, void *aux
,
3593 const struct ofproto_mirror_settings
*s
)
3595 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3596 struct ofbundle
**srcs
, **dsts
;
3601 mirror_destroy(ofproto
->mbridge
, aux
);
3605 srcs
= xmalloc(s
->n_srcs
* sizeof *srcs
);
3606 dsts
= xmalloc(s
->n_dsts
* sizeof *dsts
);
3608 for (i
= 0; i
< s
->n_srcs
; i
++) {
3609 srcs
[i
] = bundle_lookup(ofproto
, s
->srcs
[i
]);
3612 for (i
= 0; i
< s
->n_dsts
; i
++) {
3613 dsts
[i
] = bundle_lookup(ofproto
, s
->dsts
[i
]);
3616 error
= mirror_set(ofproto
->mbridge
, aux
, s
->name
, srcs
, s
->n_srcs
, dsts
,
3617 s
->n_dsts
, s
->src_vlans
,
3618 bundle_lookup(ofproto
, s
->out_bundle
),
3619 s
->snaplen
, s
->out_vlan
);
3626 mirror_get_stats__(struct ofproto
*ofproto
, void *aux
,
3627 uint64_t *packets
, uint64_t *bytes
)
3629 return mirror_get_stats(ofproto_dpif_cast(ofproto
)->mbridge
, aux
, packets
,
3634 set_flood_vlans(struct ofproto
*ofproto_
, unsigned long *flood_vlans
)
3636 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3637 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
3638 if (mac_learning_set_flood_vlans(ofproto
->ml
, flood_vlans
)) {
3639 mac_learning_flush(ofproto
->ml
);
3641 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
3646 is_mirror_output_bundle(const struct ofproto
*ofproto_
, void *aux
)
3648 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3649 struct ofbundle
*bundle
= bundle_lookup(ofproto
, aux
);
3650 return bundle
&& mirror_bundle_out(ofproto
->mbridge
, bundle
) != 0;
3654 forward_bpdu_changed(struct ofproto
*ofproto_
)
3656 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3657 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3661 set_mac_table_config(struct ofproto
*ofproto_
, unsigned int idle_time
,
3664 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3665 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
3666 mac_learning_set_idle_time(ofproto
->ml
, idle_time
);
3667 mac_learning_set_max_entries(ofproto
->ml
, max_entries
);
3668 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
3671 /* Configures multicast snooping on 'ofport' using the settings
3672 * defined in 's'. */
3674 set_mcast_snooping(struct ofproto
*ofproto_
,
3675 const struct ofproto_mcast_snooping_settings
*s
)
3677 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3679 /* Only revalidate flows if the configuration changed. */
3680 if (!s
!= !ofproto
->ms
) {
3681 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3686 ofproto
->ms
= mcast_snooping_create();
3689 ovs_rwlock_wrlock(&ofproto
->ms
->rwlock
);
3690 mcast_snooping_set_idle_time(ofproto
->ms
, s
->idle_time
);
3691 mcast_snooping_set_max_entries(ofproto
->ms
, s
->max_entries
);
3692 if (mcast_snooping_set_flood_unreg(ofproto
->ms
, s
->flood_unreg
)) {
3693 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3695 ovs_rwlock_unlock(&ofproto
->ms
->rwlock
);
3697 mcast_snooping_unref(ofproto
->ms
);
3704 /* Configures multicast snooping port's flood settings on 'ofproto'. */
3706 set_mcast_snooping_port(struct ofproto
*ofproto_
, void *aux
,
3707 const struct ofproto_mcast_snooping_port_settings
*s
)
3709 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3710 struct ofbundle
*bundle
= bundle_lookup(ofproto
, aux
);
3712 if (ofproto
->ms
&& s
) {
3713 ovs_rwlock_wrlock(&ofproto
->ms
->rwlock
);
3714 mcast_snooping_set_port_flood(ofproto
->ms
, bundle
, s
->flood
);
3715 mcast_snooping_set_port_flood_reports(ofproto
->ms
, bundle
,
3717 ovs_rwlock_unlock(&ofproto
->ms
->rwlock
);
3725 struct ofport_dpif
*
3726 ofp_port_to_ofport(const struct ofproto_dpif
*ofproto
, ofp_port_t ofp_port
)
3728 struct ofport
*ofport
= ofproto_get_port(&ofproto
->up
, ofp_port
);
3729 return ofport
? ofport_dpif_cast(ofport
) : NULL
;
3733 ofproto_port_from_dpif_port(struct ofproto_dpif
*ofproto
,
3734 struct ofproto_port
*ofproto_port
,
3735 struct dpif_port
*dpif_port
)
3737 ofproto_port
->name
= dpif_port
->name
;
3738 ofproto_port
->type
= dpif_port
->type
;
3739 ofproto_port
->ofp_port
= odp_port_to_ofp_port(ofproto
, dpif_port
->port_no
);
3743 ofport_update_peer(struct ofport_dpif
*ofport
)
3745 const struct ofproto_dpif
*ofproto
;
3746 struct dpif_backer
*backer
;
3749 if (!netdev_vport_is_patch(ofport
->up
.netdev
)) {
3753 backer
= ofproto_dpif_cast(ofport
->up
.ofproto
)->backer
;
3754 backer
->need_revalidate
= REV_RECONFIGURE
;
3757 ofport
->peer
->peer
= NULL
;
3758 ofport
->peer
= NULL
;
3761 peer_name
= netdev_vport_patch_peer(ofport
->up
.netdev
);
3766 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_by_name_node
,
3767 &all_ofproto_dpifs_by_name
) {
3768 struct ofport
*peer_ofport
;
3769 struct ofport_dpif
*peer
;
3772 if (ofproto
->backer
!= backer
) {
3776 peer_ofport
= shash_find_data(&ofproto
->up
.port_by_name
, peer_name
);
3781 peer
= ofport_dpif_cast(peer_ofport
);
3782 peer_peer
= netdev_vport_patch_peer(peer
->up
.netdev
);
3783 if (peer_peer
&& !strcmp(netdev_get_name(ofport
->up
.netdev
),
3785 ofport
->peer
= peer
;
3786 ofport
->peer
->peer
= ofport
;
3796 may_enable_port(struct ofport_dpif
*ofport
)
3798 /* If CFM or BFD is enabled, then at least one of them must report that the
3800 if ((ofport
->bfd
|| ofport
->cfm
)
3802 && !cfm_get_fault(ofport
->cfm
)
3803 && cfm_get_opup(ofport
->cfm
) != 0)
3805 && bfd_forwarding(ofport
->bfd
))) {
3809 /* If LACP is enabled, it must report that the link is enabled. */
3811 && !lacp_member_may_enable(ofport
->bundle
->lacp
, ofport
)) {
3819 port_run(struct ofport_dpif
*ofport
)
3821 long long int carrier_seq
= netdev_get_carrier_resets(ofport
->up
.netdev
);
3822 bool carrier_changed
= carrier_seq
!= ofport
->carrier_seq
;
3823 bool enable
= netdev_get_carrier(ofport
->up
.netdev
);
3825 ofport
->carrier_seq
= carrier_seq
;
3826 if (carrier_changed
&& ofport
->bundle
) {
3827 lacp_member_carrier_changed(ofport
->bundle
->lacp
, ofport
, enable
);
3831 enable
= may_enable_port(ofport
);
3834 if (ofport
->up
.may_enable
!= enable
) {
3835 ofproto_port_set_enable(&ofport
->up
, enable
);
3837 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
3838 ofproto
->backer
->need_revalidate
= REV_PORT_TOGGLED
;
3840 if (ofport
->rstp_port
) {
3841 rstp_port_set_mac_operational(ofport
->rstp_port
, enable
);
3847 port_query_by_name(const struct ofproto
*ofproto_
, const char *devname
,
3848 struct ofproto_port
*ofproto_port
)
3850 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3851 struct dpif_port dpif_port
;
3854 if (sset_contains(&ofproto
->ghost_ports
, devname
)) {
3855 const char *type
= netdev_get_type_from_name(devname
);
3857 /* We may be called before ofproto->up.port_by_name is populated with
3858 * the appropriate ofport. For this reason, we must get the name and
3859 * type from the netdev layer directly. */
3861 const struct ofport
*ofport
;
3863 ofport
= shash_find_data(&ofproto
->up
.port_by_name
, devname
);
3864 ofproto_port
->ofp_port
= ofport
? ofport
->ofp_port
: OFPP_NONE
;
3865 ofproto_port
->name
= xstrdup(devname
);
3866 ofproto_port
->type
= xstrdup(type
);
3872 if (!sset_contains(&ofproto
->ports
, devname
)) {
3875 error
= dpif_port_query_by_name(ofproto
->backer
->dpif
,
3876 devname
, &dpif_port
);
3878 ofproto_port_from_dpif_port(ofproto
, ofproto_port
, &dpif_port
);
3884 port_add(struct ofproto
*ofproto_
, struct netdev
*netdev
)
3886 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3887 const char *devname
= netdev_get_name(netdev
);
3888 char namebuf
[NETDEV_VPORT_NAME_BUFSIZE
];
3889 const char *dp_port_name
;
3891 if (netdev_vport_is_patch(netdev
)) {
3892 sset_add(&ofproto
->ghost_ports
, netdev_get_name(netdev
));
3896 dp_port_name
= netdev_vport_get_dpif_port(netdev
, namebuf
, sizeof namebuf
);
3897 if (!dpif_port_exists(ofproto
->backer
->dpif
, dp_port_name
)) {
3898 odp_port_t port_no
= ODPP_NONE
;
3901 error
= dpif_port_add(ofproto
->backer
->dpif
, netdev
, &port_no
);
3905 if (netdev_get_tunnel_config(netdev
)) {
3906 simap_put(&ofproto
->backer
->tnl_backers
,
3907 dp_port_name
, odp_to_u32(port_no
));
3911 if (netdev_get_tunnel_config(netdev
)) {
3912 sset_add(&ofproto
->ghost_ports
, devname
);
3914 sset_add(&ofproto
->ports
, devname
);
3920 port_del(struct ofproto
*ofproto_
, ofp_port_t ofp_port
)
3922 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3923 struct ofport_dpif
*ofport
= ofp_port_to_ofport(ofproto
, ofp_port
);
3930 sset_find_and_delete(&ofproto
->ghost_ports
,
3931 netdev_get_name(ofport
->up
.netdev
));
3932 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3933 if (!ofport
->is_tunnel
&& !netdev_vport_is_patch(ofport
->up
.netdev
)) {
3934 error
= dpif_port_del(ofproto
->backer
->dpif
, ofport
->odp_port
, false);
3936 /* The caller is going to close ofport->up.netdev. If this is a
3937 * bonded port, then the bond is using that netdev, so remove it
3938 * from the bond. The client will need to reconfigure everything
3939 * after deleting ports, so then the member will get re-added. */
3940 bundle_remove(&ofport
->up
);
3947 port_set_config(const struct ofport
*ofport_
, const struct smap
*cfg
)
3949 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
3950 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
3952 if (sset_contains(&ofproto
->ghost_ports
,
3953 netdev_get_name(ofport
->up
.netdev
))) {
3957 return dpif_port_set_config(ofproto
->backer
->dpif
, ofport
->odp_port
, cfg
);
3961 port_get_stats(const struct ofport
*ofport_
, struct netdev_stats
*stats
)
3963 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
3966 error
= netdev_get_stats(ofport
->up
.netdev
, stats
);
3968 if (!error
&& ofport_
->ofp_port
== OFPP_LOCAL
) {
3969 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
3971 ovs_mutex_lock(&ofproto
->stats_mutex
);
3972 /* ofproto->stats.tx_packets represents packets that we created
3973 * internally and sent to some port (e.g. packets sent with
3974 * ofproto_dpif_send_packet()). Account for them as if they had
3975 * come from OFPP_LOCAL and got forwarded. */
3977 if (stats
->rx_packets
!= UINT64_MAX
) {
3978 stats
->rx_packets
+= ofproto
->stats
.tx_packets
;
3981 if (stats
->rx_bytes
!= UINT64_MAX
) {
3982 stats
->rx_bytes
+= ofproto
->stats
.tx_bytes
;
3985 /* ofproto->stats.rx_packets represents packets that were received on
3986 * some port and we processed internally and dropped (e.g. STP).
3987 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3989 if (stats
->tx_packets
!= UINT64_MAX
) {
3990 stats
->tx_packets
+= ofproto
->stats
.rx_packets
;
3993 if (stats
->tx_bytes
!= UINT64_MAX
) {
3994 stats
->tx_bytes
+= ofproto
->stats
.rx_bytes
;
3996 ovs_mutex_unlock(&ofproto
->stats_mutex
);
4003 vport_get_status(const struct ofport
*ofport_
, char **errp
)
4005 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
4008 if (!netdev_vport_is_patch(ofport
->up
.netdev
) || ofport
->peer
) {
4012 peer_name
= netdev_vport_patch_peer(ofport
->up
.netdev
);
4016 *errp
= xasprintf("No usable peer '%s' exists in '%s' datapath.",
4017 peer_name
, ofport
->up
.ofproto
->type
);
4023 port_get_lacp_stats(const struct ofport
*ofport_
,
4024 struct lacp_member_stats
*stats
)
4026 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
4027 if (ofport
->bundle
&& ofport
->bundle
->lacp
) {
4028 if (lacp_get_member_stats(ofport
->bundle
->lacp
, ofport
, stats
)) {
4035 struct port_dump_state
{
4036 struct sset_position pos
;
4039 struct ofproto_port port
;
4044 port_dump_start(const struct ofproto
*ofproto_ OVS_UNUSED
, void **statep
)
4046 *statep
= xzalloc(sizeof(struct port_dump_state
));
4051 port_dump_next(const struct ofproto
*ofproto_
, void *state_
,
4052 struct ofproto_port
*port
)
4054 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
4055 struct port_dump_state
*state
= state_
;
4056 const struct sset
*sset
;
4057 struct sset_node
*node
;
4059 if (state
->has_port
) {
4060 ofproto_port_destroy(&state
->port
);
4061 state
->has_port
= false;
4063 sset
= state
->ghost
? &ofproto
->ghost_ports
: &ofproto
->ports
;
4064 while ((node
= sset_at_position(sset
, &state
->pos
))) {
4067 error
= port_query_by_name(ofproto_
, node
->name
, &state
->port
);
4069 *port
= state
->port
;
4070 state
->has_port
= true;
4072 } else if (error
!= ENODEV
) {
4077 if (!state
->ghost
) {
4078 state
->ghost
= true;
4079 memset(&state
->pos
, 0, sizeof state
->pos
);
4080 return port_dump_next(ofproto_
, state_
, port
);
4087 port_dump_done(const struct ofproto
*ofproto_ OVS_UNUSED
, void *state_
)
4089 struct port_dump_state
*state
= state_
;
4091 if (state
->has_port
) {
4092 ofproto_port_destroy(&state
->port
);
4099 port_poll(const struct ofproto
*ofproto_
, char **devnamep
)
4101 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
4103 if (ofproto
->port_poll_errno
) {
4104 int error
= ofproto
->port_poll_errno
;
4105 ofproto
->port_poll_errno
= 0;
4109 if (sset_is_empty(&ofproto
->port_poll_set
)) {
4113 *devnamep
= sset_pop(&ofproto
->port_poll_set
);
4118 port_poll_wait(const struct ofproto
*ofproto_
)
4120 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
4121 dpif_port_poll_wait(ofproto
->backer
->dpif
);
4125 port_is_lacp_current(const struct ofport
*ofport_
)
4127 const struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
4128 return (ofport
->bundle
&& ofport
->bundle
->lacp
4129 ? lacp_member_is_current(ofport
->bundle
->lacp
, ofport
)
4133 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4134 * then delete it entirely. */
4136 rule_expire(struct rule_dpif
*rule
, long long now
)
4137 OVS_REQUIRES(ofproto_mutex
)
4139 uint16_t hard_timeout
, idle_timeout
;
4142 hard_timeout
= rule
->up
.hard_timeout
;
4143 idle_timeout
= rule
->up
.idle_timeout
;
4145 /* Has 'rule' expired? */
4147 long long int modified
;
4149 ovs_mutex_lock(&rule
->up
.mutex
);
4150 modified
= rule
->up
.modified
;
4151 ovs_mutex_unlock(&rule
->up
.mutex
);
4153 if (now
> modified
+ hard_timeout
* 1000) {
4154 reason
= OFPRR_HARD_TIMEOUT
;
4158 if (reason
< 0 && idle_timeout
) {
4161 ovs_mutex_lock(&rule
->stats_mutex
);
4162 used
= rule
->stats
.used
;
4163 ovs_mutex_unlock(&rule
->stats_mutex
);
4165 if (now
> used
+ idle_timeout
* 1000) {
4166 reason
= OFPRR_IDLE_TIMEOUT
;
4171 COVERAGE_INC(ofproto_dpif_expired
);
4172 ofproto_rule_expire(&rule
->up
, reason
);
4177 ofproto_dpif_set_packet_odp_port(const struct ofproto_dpif
*ofproto
,
4178 ofp_port_t in_port
, struct dp_packet
*packet
)
4180 if (in_port
== OFPP_NONE
) {
4181 in_port
= OFPP_LOCAL
;
4183 packet
->md
.in_port
.odp_port
= ofp_port_to_odp_port(ofproto
, in_port
);
4187 ofproto_dpif_execute_actions__(struct ofproto_dpif
*ofproto
,
4188 ovs_version_t version
, const struct flow
*flow
,
4189 struct rule_dpif
*rule
,
4190 const struct ofpact
*ofpacts
, size_t ofpacts_len
,
4191 int depth
, int resubmits
,
4192 struct dp_packet
*packet
)
4194 struct dpif_flow_stats stats
;
4195 struct xlate_out xout
;
4196 struct xlate_in xin
;
4199 ovs_assert((rule
!= NULL
) != (ofpacts
!= NULL
));
4201 dpif_flow_stats_extract(flow
, packet
, time_msec(), &stats
);
4204 rule_dpif_credit_stats(rule
, &stats
, false);
4207 uint64_t odp_actions_stub
[1024 / 8];
4208 struct ofpbuf odp_actions
= OFPBUF_STUB_INITIALIZER(odp_actions_stub
);
4209 xlate_in_init(&xin
, ofproto
, version
, flow
, flow
->in_port
.ofp_port
, rule
,
4210 stats
.tcp_flags
, packet
, NULL
, &odp_actions
);
4211 xin
.ofpacts
= ofpacts
;
4212 xin
.ofpacts_len
= ofpacts_len
;
4213 xin
.resubmit_stats
= &stats
;
4215 xin
.resubmits
= resubmits
;
4216 if (xlate_actions(&xin
, &xout
) != XLATE_OK
) {
4221 pkt_metadata_from_flow(&packet
->md
, flow
);
4223 struct dpif_execute execute
= {
4224 .actions
= odp_actions
.data
,
4225 .actions_len
= odp_actions
.size
,
4228 .needs_help
= (xout
.slow
& SLOW_ACTION
) != 0,
4231 /* Fix up in_port. */
4232 ofproto_dpif_set_packet_odp_port(ofproto
, flow
->in_port
.ofp_port
, packet
);
4234 error
= dpif_execute(ofproto
->backer
->dpif
, &execute
);
4236 xlate_out_uninit(&xout
);
4237 ofpbuf_uninit(&odp_actions
);
4242 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
4243 * 'flow' must reflect the data in 'packet'. */
4245 ofproto_dpif_execute_actions(struct ofproto_dpif
*ofproto
,
4246 ovs_version_t version
, const struct flow
*flow
,
4247 struct rule_dpif
*rule
,
4248 const struct ofpact
*ofpacts
, size_t ofpacts_len
,
4249 struct dp_packet
*packet
)
4251 return ofproto_dpif_execute_actions__(ofproto
, version
, flow
, rule
,
4252 ofpacts
, ofpacts_len
, 0, 0, packet
);
4256 rule_dpif_credit_stats__(struct rule_dpif
*rule
,
4257 const struct dpif_flow_stats
*stats
,
4258 bool credit_counts
, bool offloaded
)
4259 OVS_REQUIRES(rule
->stats_mutex
)
4261 if (credit_counts
) {
4263 rule
->stats
.n_offload_packets
+= stats
->n_packets
;
4264 rule
->stats
.n_offload_bytes
+= stats
->n_bytes
;
4266 rule
->stats
.n_packets
+= stats
->n_packets
;
4267 rule
->stats
.n_bytes
+= stats
->n_bytes
;
4269 rule
->stats
.used
= MAX(rule
->stats
.used
, stats
->used
);
4273 rule_dpif_credit_stats(struct rule_dpif
*rule
,
4274 const struct dpif_flow_stats
*stats
, bool offloaded
)
4276 ovs_mutex_lock(&rule
->stats_mutex
);
4277 if (OVS_UNLIKELY(rule
->new_rule
)) {
4278 ovs_mutex_lock(&rule
->new_rule
->stats_mutex
);
4279 rule_dpif_credit_stats__(rule
->new_rule
, stats
, rule
->forward_counts
,
4281 ovs_mutex_unlock(&rule
->new_rule
->stats_mutex
);
4283 rule_dpif_credit_stats__(rule
, stats
, true, offloaded
);
4285 ovs_mutex_unlock(&rule
->stats_mutex
);
4288 /* Sets 'rule''s recirculation id. */
4290 rule_dpif_set_recirc_id(struct rule_dpif
*rule
, uint32_t id
)
4291 OVS_REQUIRES(rule
->up
.mutex
)
4293 ovs_assert(!rule
->recirc_id
|| rule
->recirc_id
== id
);
4294 if (rule
->recirc_id
== id
) {
4295 /* Release the new reference to the same id. */
4298 rule
->recirc_id
= id
;
4302 /* Sets 'rule''s recirculation id. */
4304 rule_set_recirc_id(struct rule
*rule_
, uint32_t id
)
4306 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
4308 ovs_mutex_lock(&rule
->up
.mutex
);
4309 rule_dpif_set_recirc_id(rule
, id
);
4310 ovs_mutex_unlock(&rule
->up
.mutex
);
4314 ofproto_dpif_get_tables_version(struct ofproto_dpif
*ofproto
)
4316 ovs_version_t version
;
4318 /* Use memory_order_acquire to signify that any following memory accesses
4319 * can not be reordered to happen before this atomic read. This makes sure
4320 * all following reads relate to this or a newer version, but never to an
4322 atomic_read_explicit(&ofproto
->tables_version
, &version
,
4323 memory_order_acquire
);
4327 /* The returned rule (if any) is valid at least until the next RCU quiescent
4328 * period. If the rule needs to stay around longer, the caller should take
4331 * 'flow' is non-const to allow for temporary modifications during the lookup.
4332 * Any changes are restored before returning. */
4333 static struct rule_dpif
*
4334 rule_dpif_lookup_in_table(struct ofproto_dpif
*ofproto
, ovs_version_t version
,
4335 uint8_t table_id
, struct flow
*flow
,
4336 struct flow_wildcards
*wc
)
4338 struct classifier
*cls
= &ofproto
->up
.tables
[table_id
].cls
;
4339 return rule_dpif_cast(rule_from_cls_rule(classifier_lookup(cls
, version
,
4344 ofproto_dpif_credit_table_stats(struct ofproto_dpif
*ofproto
, uint8_t table_id
,
4345 uint64_t n_matches
, uint64_t n_misses
)
4347 struct oftable
*tbl
= &ofproto
->up
.tables
[table_id
];
4351 atomic_add_relaxed(&tbl
->n_matched
, n_matches
, &orig
);
4354 atomic_add_relaxed(&tbl
->n_missed
, n_misses
, &orig
);
4358 /* Look up 'flow' in 'ofproto''s classifier version 'version', starting from
4359 * table '*table_id'. Returns the rule that was found, which may be one of the
4360 * special rules according to packet miss hadling. If 'may_packet_in' is
4361 * false, returning of the miss_rule (which issues packet ins for the
4362 * controller) is avoided. Updates 'wc', if nonnull, to reflect the fields
4363 * that were used during the lookup.
4365 * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
4366 * if none is found then the table miss configuration for that table is
4367 * honored, which can result in additional lookups in other OpenFlow tables.
4368 * In this case the function updates '*table_id' to reflect the final OpenFlow
4369 * table that was searched.
4371 * If 'honor_table_miss' is false, then only one table lookup occurs, in
4374 * The rule is returned in '*rule', which is valid at least until the next
4375 * RCU quiescent period. If the '*rule' needs to stay around longer, the
4376 * caller must take a reference.
4378 * 'in_port' allows the lookup to take place as if the in port had the value
4379 * 'in_port'. This is needed for resubmit action support.
4381 * 'flow' is non-const to allow for temporary modifications during the lookup.
4382 * Any changes are restored before returning. */
4384 rule_dpif_lookup_from_table(struct ofproto_dpif
*ofproto
,
4385 ovs_version_t version
, struct flow
*flow
,
4386 struct flow_wildcards
*wc
,
4387 const struct dpif_flow_stats
*stats
,
4388 uint8_t *table_id
, ofp_port_t in_port
,
4389 bool may_packet_in
, bool honor_table_miss
,
4390 struct xlate_cache
*xcache
)
4392 ovs_be16 old_tp_src
= flow
->tp_src
, old_tp_dst
= flow
->tp_dst
;
4393 ofp_port_t old_in_port
= flow
->in_port
.ofp_port
;
4394 enum ofputil_table_miss miss_config
;
4395 struct rule_dpif
*rule
;
4398 /* We always unwildcard nw_frag (for IP), so they
4399 * need not be unwildcarded here. */
4400 if (flow
->nw_frag
& FLOW_NW_FRAG_ANY
4401 && ofproto
->up
.frag_handling
!= OFPUTIL_FRAG_NX_MATCH
) {
4402 if (ofproto
->up
.frag_handling
== OFPUTIL_FRAG_NORMAL
) {
4403 /* We must pretend that transport ports are unavailable. */
4404 flow
->tp_src
= htons(0);
4405 flow
->tp_dst
= htons(0);
4407 /* Must be OFPUTIL_FRAG_DROP (we don't have OFPUTIL_FRAG_REASM).
4408 * Use the drop_frags_rule (which cannot disappear). */
4409 rule
= ofproto
->drop_frags_rule
;
4411 struct oftable
*tbl
= &ofproto
->up
.tables
[*table_id
];
4414 atomic_add_relaxed(&tbl
->n_matched
, stats
->n_packets
, &orig
);
4417 struct xc_entry
*entry
;
4419 entry
= xlate_cache_add_entry(xcache
, XC_TABLE
);
4420 entry
->table
.ofproto
= ofproto
;
4421 entry
->table
.id
= *table_id
;
4422 entry
->table
.match
= true;
4428 /* Look up a flow with 'in_port' as the input port. Then restore the
4429 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
4430 * have surprising behavior). */
4431 flow
->in_port
.ofp_port
= in_port
;
4433 /* Our current implementation depends on n_tables == N_TABLES, and
4434 * TBL_INTERNAL being the last table. */
4435 BUILD_ASSERT_DECL(N_TABLES
== TBL_INTERNAL
+ 1);
4437 miss_config
= OFPUTIL_TABLE_MISS_CONTINUE
;
4439 for (next_id
= *table_id
;
4440 next_id
< ofproto
->up
.n_tables
;
4441 next_id
++, next_id
+= (next_id
== TBL_INTERNAL
))
4443 *table_id
= next_id
;
4444 rule
= rule_dpif_lookup_in_table(ofproto
, version
, next_id
, flow
, wc
);
4446 struct oftable
*tbl
= &ofproto
->up
.tables
[next_id
];
4449 atomic_add_relaxed(rule
? &tbl
->n_matched
: &tbl
->n_missed
,
4450 stats
->n_packets
, &orig
);
4453 struct xc_entry
*entry
;
4455 entry
= xlate_cache_add_entry(xcache
, XC_TABLE
);
4456 entry
->table
.ofproto
= ofproto
;
4457 entry
->table
.id
= next_id
;
4458 entry
->table
.match
= (rule
!= NULL
);
4461 goto out
; /* Match. */
4463 if (honor_table_miss
) {
4464 miss_config
= ofproto_table_get_miss_config(&ofproto
->up
,
4466 if (miss_config
== OFPUTIL_TABLE_MISS_CONTINUE
) {
4473 rule
= ofproto
->no_packet_in_rule
;
4474 if (may_packet_in
) {
4475 if (miss_config
== OFPUTIL_TABLE_MISS_CONTINUE
4476 || miss_config
== OFPUTIL_TABLE_MISS_CONTROLLER
) {
4477 struct ofport_dpif
*port
;
4479 port
= ofp_port_to_ofport(ofproto
, old_in_port
);
4481 VLOG_WARN_RL(&rl
, "packet-in on unknown OpenFlow port %"PRIu32
,
4483 } else if (!(port
->up
.pp
.config
& OFPUTIL_PC_NO_PACKET_IN
)) {
4484 rule
= ofproto
->miss_rule
;
4486 } else if (miss_config
== OFPUTIL_TABLE_MISS_DEFAULT
&&
4487 connmgr_wants_packet_in_on_miss(ofproto
->up
.connmgr
)) {
4488 rule
= ofproto
->miss_rule
;
4492 /* Restore port numbers, as they may have been modified above. */
4493 flow
->tp_src
= old_tp_src
;
4494 flow
->tp_dst
= old_tp_dst
;
4495 /* Restore the old in port. */
4496 flow
->in_port
.ofp_port
= old_in_port
;
4501 static struct rule_dpif
*rule_dpif_cast(const struct rule
*rule
)
4503 return rule
? CONTAINER_OF(rule
, struct rule_dpif
, up
) : NULL
;
4506 static struct rule
*
4509 struct rule_dpif
*rule
= xzalloc(sizeof *rule
);
4514 rule_dealloc(struct rule
*rule_
)
4516 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
4521 check_mask(struct ofproto_dpif
*ofproto
, const struct miniflow
*flow
)
4523 const struct odp_support
*support
;
4524 uint16_t ct_state
, ct_zone
;
4528 support
= &ofproto
->backer
->rt_support
.odp
;
4529 ct_state
= MINIFLOW_GET_U8(flow
, ct_state
);
4531 if (ct_state
& CS_UNSUPPORTED_MASK
) {
4532 return OFPERR_OFPBMC_BAD_MASK
;
4535 /* Do not bother dissecting the flow further if the datapath supports all
4536 * the features we know of. */
4537 if (support
->ct_state
&& support
->ct_zone
&& support
->ct_mark
4538 && support
->ct_label
&& support
->ct_state_nat
4539 && support
->ct_orig_tuple
&& support
->ct_orig_tuple6
) {
4543 ct_zone
= MINIFLOW_GET_U16(flow
, ct_zone
);
4544 ct_mark
= MINIFLOW_GET_U32(flow
, ct_mark
);
4545 ct_label
= MINIFLOW_GET_U128(flow
, ct_label
);
4547 if ((ct_state
&& !support
->ct_state
)
4548 || ((ct_state
& (CS_SRC_NAT
| CS_DST_NAT
)) && !support
->ct_state_nat
)
4549 || (ct_zone
&& !support
->ct_zone
)
4550 || (ct_mark
&& !support
->ct_mark
)
4551 || (!ovs_u128_is_zero(ct_label
) && !support
->ct_label
)) {
4552 return OFPERR_NXBMC_CT_DATAPATH_SUPPORT
;
4555 if (!support
->ct_orig_tuple
&& !support
->ct_orig_tuple6
4556 && (MINIFLOW_GET_U8(flow
, ct_nw_proto
)
4557 || MINIFLOW_GET_U16(flow
, ct_tp_src
)
4558 || MINIFLOW_GET_U16(flow
, ct_tp_dst
))) {
4559 return OFPERR_NXBMC_CT_DATAPATH_SUPPORT
;
4562 if (!support
->ct_orig_tuple
4563 && (MINIFLOW_GET_U32(flow
, ct_nw_src
)
4564 || MINIFLOW_GET_U32(flow
, ct_nw_dst
))) {
4565 return OFPERR_NXBMC_CT_DATAPATH_SUPPORT
;
4568 if (!support
->ct_orig_tuple6
4569 && (!ovs_u128_is_zero(MINIFLOW_GET_U128(flow
, ct_ipv6_src
))
4570 || !ovs_u128_is_zero(MINIFLOW_GET_U128(flow
, ct_ipv6_dst
)))) {
4571 return OFPERR_NXBMC_CT_DATAPATH_SUPPORT
;
4578 report_unsupported_act(const char *action
, const char *detail
)
4580 static struct vlog_rate_limit rll
= VLOG_RATE_LIMIT_INIT(1, 5);
4581 VLOG_WARN_RL(&rll
, "Rejecting %s action because datapath does not support"
4582 "%s%s (your kernel module may be out of date)",
4583 action
, detail
? " " : "", detail
? detail
: "");
4587 check_actions(const struct ofproto_dpif
*ofproto
,
4588 const struct rule_actions
*const actions
)
4590 const struct ofpact
*ofpact
;
4591 const struct odp_support
*support
= &ofproto
->backer
->rt_support
.odp
;
4593 OFPACT_FOR_EACH (ofpact
, actions
->ofpacts
, actions
->ofpacts_len
) {
4594 if (ofpact
->type
== OFPACT_CT
) {
4595 const struct ofpact_conntrack
*ct
;
4596 const struct ofpact
*a
;
4598 ct
= CONTAINER_OF(ofpact
, struct ofpact_conntrack
, ofpact
);
4600 if (!support
->ct_state
) {
4601 report_unsupported_act("ct", "ct action");
4602 return OFPERR_NXBAC_CT_DATAPATH_SUPPORT
;
4604 if ((ct
->zone_imm
|| ct
->zone_src
.field
) && !support
->ct_zone
) {
4605 report_unsupported_act("ct", "ct zones");
4606 return OFPERR_NXBAC_CT_DATAPATH_SUPPORT
;
4608 /* So far the force commit feature is implemented together with the
4609 * original direction tuple feature by all datapaths, so we use the
4610 * support flag for the 'ct_orig_tuple' to indicate support for the
4611 * force commit feature as well. */
4612 if ((ct
->flags
& NX_CT_F_FORCE
) && !support
->ct_orig_tuple
) {
4613 report_unsupported_act("ct", "force commit");
4614 return OFPERR_NXBAC_CT_DATAPATH_SUPPORT
;
4617 OFPACT_FOR_EACH(a
, ct
->actions
, ofpact_ct_get_action_len(ct
)) {
4618 const struct mf_field
*dst
= ofpact_get_mf_dst(a
);
4620 if (a
->type
== OFPACT_NAT
&& !support
->ct_state_nat
) {
4621 /* The backer doesn't seem to support the NAT bits in
4622 * 'ct_state': assume that it doesn't support the NAT
4624 report_unsupported_act("ct", "nat");
4625 return OFPERR_NXBAC_CT_DATAPATH_SUPPORT
;
4627 if (dst
&& ((dst
->id
== MFF_CT_MARK
&& !support
->ct_mark
) ||
4628 (dst
->id
== MFF_CT_LABEL
&& !support
->ct_label
))) {
4629 report_unsupported_act("ct", "setting mark and/or label");
4630 return OFPERR_NXBAC_CT_DATAPATH_SUPPORT
;
4633 } else if (ofpact
->type
== OFPACT_RESUBMIT
) {
4634 struct ofpact_resubmit
*resubmit
= ofpact_get_RESUBMIT(ofpact
);
4636 if (resubmit
->with_ct_orig
&& !support
->ct_orig_tuple
) {
4637 report_unsupported_act("resubmit",
4638 "ct original direction tuple");
4639 return OFPERR_NXBAC_CT_DATAPATH_SUPPORT
;
4641 } else if (!support
->nd_ext
&& ofpact
->type
== OFPACT_SET_FIELD
) {
4642 const struct mf_field
*dst
= ofpact_get_mf_dst(ofpact
);
4644 if (dst
->id
== MFF_ND_RESERVED
|| dst
->id
== MFF_ND_OPTIONS_TYPE
) {
4645 report_unsupported_act("set field",
4646 "setting IPv6 ND Extensions fields");
4647 return OFPERR_OFPBAC_BAD_SET_ARGUMENT
;
4656 rule_check(struct rule
*rule
)
4658 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->ofproto
);
4661 err
= check_mask(ofproto
, &rule
->cr
.match
.mask
->masks
);
4665 return check_actions(ofproto
, rule
->actions
);
4669 rule_construct(struct rule
*rule_
)
4670 OVS_NO_THREAD_SAFETY_ANALYSIS
4672 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
4675 error
= rule_check(rule_
);
4680 ovs_mutex_init_adaptive(&rule
->stats_mutex
);
4681 rule
->stats
.n_packets
= 0;
4682 rule
->stats
.n_bytes
= 0;
4683 rule
->stats
.used
= rule
->up
.modified
;
4684 rule
->recirc_id
= 0;
4685 rule
->new_rule
= NULL
;
4686 rule
->forward_counts
= false;
4692 rule_insert(struct rule
*rule_
, struct rule
*old_rule_
, bool forward_counts
)
4693 OVS_REQUIRES(ofproto_mutex
)
4695 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
4698 struct rule_dpif
*old_rule
= rule_dpif_cast(old_rule_
);
4700 ovs_assert(!old_rule
->new_rule
);
4702 /* Take a reference to the new rule, and refer all stats updates from
4703 * the old rule to the new rule. */
4704 ofproto_rule_ref(&rule
->up
);
4706 ovs_mutex_lock(&old_rule
->stats_mutex
);
4707 ovs_mutex_lock(&rule
->stats_mutex
);
4708 old_rule
->new_rule
= rule
; /* Forward future stats. */
4709 old_rule
->forward_counts
= forward_counts
;
4711 if (forward_counts
) {
4712 rule
->stats
= old_rule
->stats
; /* Transfer stats to the new
4715 /* Used timestamp must be forwarded whenever a rule is modified. */
4716 rule
->stats
.used
= old_rule
->stats
.used
;
4718 ovs_mutex_unlock(&rule
->stats_mutex
);
4719 ovs_mutex_unlock(&old_rule
->stats_mutex
);
4726 rule_destruct(struct rule
*rule_
)
4727 OVS_NO_THREAD_SAFETY_ANALYSIS
4729 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
4731 ovs_mutex_destroy(&rule
->stats_mutex
);
4732 /* Release reference to the new rule, if any. */
4733 if (rule
->new_rule
) {
4734 ofproto_rule_unref(&rule
->new_rule
->up
);
4736 if (rule
->recirc_id
) {
4737 recirc_free_id(rule
->recirc_id
);
4742 rule_get_stats(struct rule
*rule_
, struct pkt_stats
*stats
,
4743 long long int *used
)
4745 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
4747 ovs_mutex_lock(&rule
->stats_mutex
);
4748 if (OVS_UNLIKELY(rule
->new_rule
)) {
4749 rule_get_stats(&rule
->new_rule
->up
, stats
, used
);
4751 stats
->n_packets
= rule
->stats
.n_packets
;
4752 stats
->n_bytes
= rule
->stats
.n_bytes
;
4753 stats
->n_offload_packets
= rule
->stats
.n_offload_packets
;
4754 stats
->n_offload_bytes
= rule
->stats
.n_offload_bytes
;
4755 *used
= rule
->stats
.used
;
4757 ovs_mutex_unlock(&rule
->stats_mutex
);
4760 struct ofproto_dpif_packet_out
{
4761 struct xlate_cache xcache
;
4762 struct ofpbuf odp_actions
;
4763 struct recirc_refs rr
;
4768 static struct ofproto_dpif_packet_out
*
4769 ofproto_dpif_packet_out_new(void)
4771 struct ofproto_dpif_packet_out
*aux
= xmalloc(sizeof *aux
);
4772 xlate_cache_init(&aux
->xcache
);
4773 ofpbuf_init(&aux
->odp_actions
, 64);
4774 aux
->rr
= RECIRC_REFS_EMPTY_INITIALIZER
;
4775 aux
->needs_help
= false;
4781 ofproto_dpif_packet_out_delete(struct ofproto_dpif_packet_out
*aux
)
4784 xlate_cache_uninit(&aux
->xcache
);
4785 ofpbuf_uninit(&aux
->odp_actions
);
4786 recirc_refs_unref(&aux
->rr
);
4792 packet_xlate(struct ofproto
*ofproto_
, struct ofproto_packet_out
*opo
)
4793 OVS_REQUIRES(ofproto_mutex
)
4795 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
4796 struct xlate_out xout
;
4797 struct xlate_in xin
;
4798 enum ofperr error
= 0;
4800 struct ofproto_dpif_packet_out
*aux
= ofproto_dpif_packet_out_new();
4802 xlate_in_init(&xin
, ofproto
, opo
->version
, opo
->flow
,
4803 opo
->flow
->in_port
.ofp_port
, NULL
, 0, opo
->packet
, NULL
,
4805 xin
.ofpacts
= opo
->ofpacts
;
4806 xin
.ofpacts_len
= opo
->ofpacts_len
;
4807 /* No learning or stats, but collect side effects to xcache. */
4808 xin
.allow_side_effects
= false;
4809 xin
.resubmit_stats
= NULL
;
4810 xin
.xcache
= &aux
->xcache
;
4811 xin
.in_packet_out
= true;
4813 if (xlate_actions(&xin
, &xout
) != XLATE_OK
) {
4814 error
= OFPERR_OFPFMFC_UNKNOWN
; /* Error processing actions. */
4817 /* Prepare learn actions. */
4818 struct xc_entry
*entry
;
4819 struct ofpbuf entries
= aux
->xcache
.entries
;
4821 XC_ENTRY_FOR_EACH (entry
, &entries
) {
4822 if (entry
->type
== XC_LEARN
) {
4823 struct ofproto_flow_mod
*ofm
= entry
->learn
.ofm
;
4825 error
= ofproto_flow_mod_learn_refresh(ofm
);
4829 struct rule
*rule
= ofm
->temp_rule
;
4830 ofm
->learn_adds_rule
= (rule
->state
== RULE_INITIALIZED
);
4831 if (ofm
->learn_adds_rule
) {
4832 /* If learning on a different bridge, must use its next
4833 * version number. */
4834 ofm
->version
= (rule
->ofproto
== ofproto_
)
4835 ? opo
->version
: rule
->ofproto
->tables_version
+ 1;
4836 error
= ofproto_flow_mod_learn_start(ofm
);
4845 aux
->needs_help
= (xout
.slow
& SLOW_ACTION
) != 0;
4846 recirc_refs_swap(&aux
->rr
, &xout
.recircs
); /* Hold recirc refs. */
4848 xlate_out_uninit(&xout
);
4853 xlate_out_uninit(&xout
);
4854 ofproto_dpif_packet_out_delete(aux
);
4860 packet_xlate_revert(struct ofproto
*ofproto OVS_UNUSED
,
4861 struct ofproto_packet_out
*opo
)
4862 OVS_REQUIRES(ofproto_mutex
)
4864 struct ofproto_dpif_packet_out
*aux
= opo
->aux
;
4867 /* Revert the learned flows. */
4868 struct xc_entry
*entry
;
4869 struct ofpbuf entries
= aux
->xcache
.entries
;
4871 XC_ENTRY_FOR_EACH (entry
, &entries
) {
4872 if (entry
->type
== XC_LEARN
&& entry
->learn
.ofm
->learn_adds_rule
) {
4873 ofproto_flow_mod_learn_revert(entry
->learn
.ofm
);
4877 ofproto_dpif_packet_out_delete(aux
);
4881 /* Push stats and perform side effects of flow translation. */
4883 ofproto_dpif_xcache_execute(struct ofproto_dpif
*ofproto
,
4884 struct xlate_cache
*xcache
,
4885 struct dpif_flow_stats
*stats
)
4886 OVS_REQUIRES(ofproto_mutex
)
4888 struct xc_entry
*entry
;
4889 struct ofpbuf entries
= xcache
->entries
;
4891 XC_ENTRY_FOR_EACH (entry
, &entries
) {
4892 switch (entry
->type
) {
4894 /* Finish the learned flows. */
4895 if (entry
->learn
.ofm
->learn_adds_rule
) {
4896 ofproto_flow_mod_learn_finish(entry
->learn
.ofm
, &ofproto
->up
);
4899 case XC_FIN_TIMEOUT
:
4900 if (stats
->tcp_flags
& (TCP_FIN
| TCP_RST
)) {
4901 /* 'ofproto_mutex' already held */
4902 ofproto_rule_reduce_timeouts__(&entry
->fin
.rule
->up
,
4907 /* All the rest can be dealt with by the xlate layer. */
4917 case XC_TUNNEL_HEADER
:
4918 xlate_push_stats_entry(entry
, stats
, false);
4927 packet_execute_prepare(struct ofproto
*ofproto_
,
4928 struct ofproto_packet_out
*opo
)
4929 OVS_REQUIRES(ofproto_mutex
)
4931 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
4932 struct dpif_flow_stats stats
;
4933 struct dpif_execute
*execute
;
4935 struct ofproto_dpif_packet_out
*aux
= opo
->aux
;
4938 /* Run the side effects from the xcache. */
4939 dpif_flow_stats_extract(opo
->flow
, opo
->packet
, time_msec(), &stats
);
4940 ofproto_dpif_xcache_execute(ofproto
, &aux
->xcache
, &stats
);
4942 execute
= xzalloc(sizeof *execute
);
4943 execute
->actions
= xmemdup(aux
->odp_actions
.data
, aux
->odp_actions
.size
);
4944 execute
->actions_len
= aux
->odp_actions
.size
;
4946 pkt_metadata_from_flow(&opo
->packet
->md
, opo
->flow
);
4947 execute
->packet
= opo
->packet
;
4948 execute
->flow
= opo
->flow
;
4949 execute
->needs_help
= aux
->needs_help
;
4950 execute
->probe
= false;
4953 /* Fix up in_port. */
4954 ofproto_dpif_set_packet_odp_port(ofproto
, opo
->flow
->in_port
.ofp_port
,
4957 ofproto_dpif_packet_out_delete(aux
);
4962 packet_execute(struct ofproto
*ofproto_
, struct ofproto_packet_out
*opo
)
4963 OVS_EXCLUDED(ofproto_mutex
)
4965 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
4966 struct dpif_execute
*execute
= opo
->aux
;
4972 dpif_execute(ofproto
->backer
->dpif
, execute
);
4974 free(CONST_CAST(struct nlattr
*, execute
->actions
));
4979 static struct group_dpif
*group_dpif_cast(const struct ofgroup
*group
)
4981 return group
? CONTAINER_OF(group
, struct group_dpif
, up
) : NULL
;
4984 static struct ofgroup
*
4987 struct group_dpif
*group
= xzalloc(sizeof *group
);
4992 group_dealloc(struct ofgroup
*group_
)
4994 struct group_dpif
*group
= group_dpif_cast(group_
);
4999 group_construct_stats(struct group_dpif
*group
)
5000 OVS_REQUIRES(group
->stats_mutex
)
5002 group
->packet_count
= 0;
5003 group
->byte_count
= 0;
5005 struct ofputil_bucket
*bucket
;
5006 LIST_FOR_EACH (bucket
, list_node
, &group
->up
.buckets
) {
5007 bucket
->stats
.packet_count
= 0;
5008 bucket
->stats
.byte_count
= 0;
5013 group_dpif_credit_stats(struct group_dpif
*group
,
5014 struct ofputil_bucket
*bucket
,
5015 const struct dpif_flow_stats
*stats
)
5017 ovs_mutex_lock(&group
->stats_mutex
);
5018 group
->packet_count
+= stats
->n_packets
;
5019 group
->byte_count
+= stats
->n_bytes
;
5021 bucket
->stats
.packet_count
+= stats
->n_packets
;
5022 bucket
->stats
.byte_count
+= stats
->n_bytes
;
5023 } else { /* Credit to all buckets */
5024 LIST_FOR_EACH (bucket
, list_node
, &group
->up
.buckets
) {
5025 bucket
->stats
.packet_count
+= stats
->n_packets
;
5026 bucket
->stats
.byte_count
+= stats
->n_bytes
;
5029 ovs_mutex_unlock(&group
->stats_mutex
);
5032 /* Calculate the dp_hash mask needed to provide the least weighted bucket
5033 * with at least one hash value and construct a mapping table from masked
5034 * dp_hash value to group bucket using the Webster method.
5035 * If the caller specifies a non-zero max_hash value, abort and return false
5036 * if more hash values would be required. The absolute maximum number of
5037 * hash values supported is 256. */
5039 #define MAX_SELECT_GROUP_HASH_VALUES 256
5042 group_setup_dp_hash_table(struct group_dpif
*group
, size_t max_hash
)
5044 struct ofputil_bucket
*bucket
;
5045 uint32_t n_buckets
= group
->up
.n_buckets
;
5046 uint64_t total_weight
= 0;
5047 uint16_t min_weight
= UINT16_MAX
;
5049 struct ofputil_bucket
*bucket
;
5055 if (n_buckets
== 0) {
5056 VLOG_DBG(" Don't apply dp_hash method without buckets.");
5060 webster
= xcalloc(n_buckets
, sizeof(struct webster
));
5062 LIST_FOR_EACH (bucket
, list_node
, &group
->up
.buckets
) {
5063 if (bucket
->weight
> 0 && bucket
->weight
< min_weight
) {
5064 min_weight
= bucket
->weight
;
5066 total_weight
+= bucket
->weight
;
5067 webster
[i
].bucket
= bucket
;
5068 webster
[i
].divisor
= 1;
5069 webster
[i
].value
= bucket
->weight
;
5070 webster
[i
].hits
= 0;
5074 if (total_weight
== 0) {
5075 VLOG_DBG(" Total weight is zero. No active buckets.");
5079 VLOG_DBG(" Minimum weight: %d, total weight: %"PRIu64
,
5080 min_weight
, total_weight
);
5082 uint64_t min_slots
= DIV_ROUND_UP(total_weight
, min_weight
);
5083 uint64_t min_slots2
= ROUND_UP_POW2(min_slots
);
5084 uint64_t n_hash
= MAX(16, min_slots2
);
5085 if (n_hash
> MAX_SELECT_GROUP_HASH_VALUES
||
5086 (max_hash
!= 0 && n_hash
> max_hash
)) {
5087 VLOG_DBG(" Too many hash values required: %"PRIu64
, n_hash
);
5092 VLOG_DBG(" Using %"PRIu64
" hash values:", n_hash
);
5093 group
->hash_mask
= n_hash
- 1;
5094 if (group
->hash_map
) {
5095 free(group
->hash_map
);
5097 group
->hash_map
= xcalloc(n_hash
, sizeof(struct ofputil_bucket
*));
5099 /* Use Webster method to distribute hash values over buckets. */
5100 for (int hash
= 0; hash
< n_hash
; hash
++) {
5101 struct webster
*winner
= &webster
[0];
5102 for (i
= 1; i
< n_buckets
; i
++) {
5103 if (webster
[i
].value
> winner
->value
) {
5104 winner
= &webster
[i
];
5108 winner
->divisor
+= 2;
5109 winner
->value
= (double) winner
->bucket
->weight
/ winner
->divisor
;
5110 group
->hash_map
[hash
] = winner
->bucket
;
5114 LIST_FOR_EACH (bucket
, list_node
, &group
->up
.buckets
) {
5115 double target
= (n_hash
* bucket
->weight
) / (double) total_weight
;
5116 VLOG_DBG(" Bucket %d: weight=%d, target=%.2f hits=%d",
5117 bucket
->bucket_id
, bucket
->weight
,
5118 target
, webster
[i
].hits
);
5127 group_set_selection_method(struct group_dpif
*group
)
5129 const struct ofputil_group_props
*props
= &group
->up
.props
;
5130 const char *selection_method
= props
->selection_method
;
5132 VLOG_DBG("Constructing select group %"PRIu32
, group
->up
.group_id
);
5133 if (selection_method
[0] == '\0') {
5134 VLOG_DBG("No selection method specified. Trying dp_hash.");
5135 /* If the controller has not specified a selection method, check if
5136 * the dp_hash selection method with max 64 hash values is appropriate
5137 * for the given bucket configuration. */
5138 if (group_setup_dp_hash_table(group
, 64)) {
5139 /* Use dp_hash selection method with symmetric L4 hash. */
5140 group
->selection_method
= SEL_METHOD_DP_HASH
;
5141 group
->hash_alg
= OVS_HASH_ALG_SYM_L4
;
5142 group
->hash_basis
= 0;
5143 VLOG_DBG("Use dp_hash with %d hash values using algorithm %d.",
5144 group
->hash_mask
+ 1, group
->hash_alg
);
5146 /* Fall back to original default hashing in slow path. */
5147 VLOG_DBG("Falling back to default hash method.");
5148 group
->selection_method
= SEL_METHOD_DEFAULT
;
5150 } else if (!strcmp(selection_method
, "dp_hash")) {
5151 VLOG_DBG("Selection method specified: dp_hash.");
5152 /* Try to use dp_hash if possible at all. */
5153 if (group_setup_dp_hash_table(group
, 0)) {
5154 group
->selection_method
= SEL_METHOD_DP_HASH
;
5155 group
->hash_alg
= props
->selection_method_param
>> 32;
5156 if (group
->hash_alg
>= __OVS_HASH_MAX
) {
5157 VLOG_DBG("Invalid dp_hash algorithm %d. "
5158 "Defaulting to OVS_HASH_ALG_L4", group
->hash_alg
);
5159 group
->hash_alg
= OVS_HASH_ALG_L4
;
5161 group
->hash_basis
= (uint32_t) props
->selection_method_param
;
5162 VLOG_DBG("Use dp_hash with %d hash values using algorithm %d.",
5163 group
->hash_mask
+ 1, group
->hash_alg
);
5165 /* Fall back to original default hashing in slow path. */
5166 VLOG_DBG("Falling back to default hash method.");
5167 group
->selection_method
= SEL_METHOD_DEFAULT
;
5169 } else if (!strcmp(selection_method
, "hash")) {
5170 VLOG_DBG("Selection method specified: hash.");
5171 if (props
->fields
.values_size
> 0) {
5172 /* Controller has specified hash fields. */
5173 struct ds s
= DS_EMPTY_INITIALIZER
;
5174 oxm_format_field_array(&s
, &props
->fields
);
5175 VLOG_DBG("Hash fields: %s", ds_cstr(&s
));
5177 group
->selection_method
= SEL_METHOD_HASH
;
5179 /* No hash fields. Fall back to original default hashing. */
5180 VLOG_DBG("No hash fields. Falling back to default hash method.");
5181 group
->selection_method
= SEL_METHOD_DEFAULT
;
5184 /* Parsing of groups should ensure this never happens */
5190 group_construct(struct ofgroup
*group_
)
5192 struct group_dpif
*group
= group_dpif_cast(group_
);
5194 ovs_mutex_init_adaptive(&group
->stats_mutex
);
5195 ovs_mutex_lock(&group
->stats_mutex
);
5196 group_construct_stats(group
);
5197 group
->hash_map
= NULL
;
5198 if (group
->up
.type
== OFPGT11_SELECT
) {
5199 group_set_selection_method(group
);
5201 ovs_mutex_unlock(&group
->stats_mutex
);
5206 group_destruct(struct ofgroup
*group_
)
5208 struct group_dpif
*group
= group_dpif_cast(group_
);
5209 ovs_mutex_destroy(&group
->stats_mutex
);
5210 if (group
->hash_map
) {
5211 free(group
->hash_map
);
5212 group
->hash_map
= NULL
;
5217 group_get_stats(const struct ofgroup
*group_
, struct ofputil_group_stats
*ogs
)
5219 struct group_dpif
*group
= group_dpif_cast(group_
);
5221 ovs_mutex_lock(&group
->stats_mutex
);
5222 ogs
->packet_count
= group
->packet_count
;
5223 ogs
->byte_count
= group
->byte_count
;
5225 struct bucket_counter
*bucket_stats
= ogs
->bucket_stats
;
5226 struct ofputil_bucket
*bucket
;
5227 LIST_FOR_EACH (bucket
, list_node
, &group
->up
.buckets
) {
5228 bucket_stats
->packet_count
= bucket
->stats
.packet_count
;
5229 bucket_stats
->byte_count
= bucket
->stats
.byte_count
;
5232 ovs_mutex_unlock(&group
->stats_mutex
);
5237 /* If the group exists, this function increments the groups's reference count.
5239 * Make sure to call ofproto_group_unref() after no longer needing to maintain
5240 * a reference to the group. */
5242 group_dpif_lookup(struct ofproto_dpif
*ofproto
, uint32_t group_id
,
5243 ovs_version_t version
, bool take_ref
)
5245 struct ofgroup
*ofgroup
= ofproto_group_lookup(&ofproto
->up
, group_id
,
5247 return ofgroup
? group_dpif_cast(ofgroup
) : NULL
;
5250 /* Sends 'packet' out 'ofport'. If 'port' is a tunnel and that tunnel type
5251 * supports a notion of an OAM flag, sets it if 'oam' is true.
5252 * May modify 'packet'.
5253 * Returns 0 if successful, otherwise a positive errno value. */
5255 ofproto_dpif_send_packet(const struct ofport_dpif
*ofport
, bool oam
,
5256 struct dp_packet
*packet
)
5258 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
5261 error
= xlate_send_packet(ofport
, oam
, packet
);
5263 ovs_mutex_lock(&ofproto
->stats_mutex
);
5264 ofproto
->stats
.tx_packets
++;
5265 ofproto
->stats
.tx_bytes
+= dp_packet_size(packet
);
5266 ovs_mutex_unlock(&ofproto
->stats_mutex
);
5270 /* Return the version string of the datapath that backs up
5274 get_datapath_version(const struct ofproto
*ofproto_
)
5276 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
5278 return ofproto
->backer
->dp_version_string
;
5282 type_set_config(const char *type
, const struct smap
*other_config
)
5284 struct dpif_backer
*backer
;
5286 backer
= shash_find_data(&all_dpif_backers
, type
);
5288 /* This is not necessarily a problem, since backers are only
5289 * created on demand. */
5293 dpif_set_config(backer
->dpif
, other_config
);
5297 ct_flush(const struct ofproto
*ofproto_
, const uint16_t *zone
)
5299 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
5301 ct_dpif_flush(ofproto
->backer
->dpif
, zone
, NULL
);
5304 static struct ct_timeout_policy
*
5305 ct_timeout_policy_lookup(const struct hmap
*ct_tps
, struct simap
*tp
)
5307 struct ct_timeout_policy
*ct_tp
;
5309 HMAP_FOR_EACH_WITH_HASH (ct_tp
, node
, simap_hash(tp
), ct_tps
) {
5310 if (simap_equal(&ct_tp
->tp
, tp
)) {
5317 static struct ct_timeout_policy
*
5318 ct_timeout_policy_alloc__(void)
5320 struct ct_timeout_policy
*ct_tp
= xzalloc(sizeof *ct_tp
);
5321 simap_init(&ct_tp
->tp
);
5325 static struct ct_timeout_policy
*
5326 ct_timeout_policy_alloc(struct simap
*tp
, struct id_pool
*tp_ids
)
5328 struct simap_node
*node
;
5330 struct ct_timeout_policy
*ct_tp
= ct_timeout_policy_alloc__();
5331 SIMAP_FOR_EACH (node
, tp
) {
5332 simap_put(&ct_tp
->tp
, node
->name
, node
->data
);
5335 if (!id_pool_alloc_id(tp_ids
, &ct_tp
->tp_id
)) {
5336 VLOG_ERR_RL(&rl
, "failed to allocate timeout policy id.");
5337 simap_destroy(&ct_tp
->tp
);
5346 ct_timeout_policy_destroy__(struct ct_timeout_policy
*ct_tp
)
5348 simap_destroy(&ct_tp
->tp
);
5353 ct_timeout_policy_destroy(struct ct_timeout_policy
*ct_tp
,
5354 struct id_pool
*tp_ids
)
5356 id_pool_free_id(tp_ids
, ct_tp
->tp_id
);
5357 ovsrcu_postpone(ct_timeout_policy_destroy__
, ct_tp
);
5361 ct_timeout_policy_unref(struct dpif_backer
*backer
,
5362 struct ct_timeout_policy
*ct_tp
)
5367 if (!ct_tp
->ref_count
) {
5368 hmap_remove(&backer
->ct_tps
, &ct_tp
->node
);
5369 ovs_list_push_back(&backer
->ct_tp_kill_list
, &ct_tp
->list_node
);
5374 static struct ct_zone
*
5375 ct_zone_lookup(const struct cmap
*ct_zones
, uint16_t zone_id
)
5377 struct ct_zone
*ct_zone
;
5379 CMAP_FOR_EACH_WITH_HASH (ct_zone
, node
, hash_int(zone_id
, 0), ct_zones
) {
5380 if (ct_zone
->zone_id
== zone_id
) {
5387 static struct ct_zone
*
5388 ct_zone_alloc(uint16_t zone_id
)
5390 struct ct_zone
*ct_zone
= xzalloc(sizeof *ct_zone
);
5391 ct_zone
->zone_id
= zone_id
;
5396 ct_zone_destroy(struct ct_zone
*ct_zone
)
5398 ovsrcu_postpone(free
, ct_zone
);
5402 ct_zone_remove_and_destroy(struct dpif_backer
*backer
, struct ct_zone
*ct_zone
)
5404 cmap_remove(&backer
->ct_zones
, &ct_zone
->node
,
5405 hash_int(ct_zone
->zone_id
, 0));
5406 ct_zone_destroy(ct_zone
);
5410 ct_add_timeout_policy_to_dpif(struct dpif
*dpif
,
5411 struct ct_timeout_policy
*ct_tp
)
5413 struct ct_dpif_timeout_policy cdtp
;
5414 struct simap_node
*node
;
5416 cdtp
.id
= ct_tp
->tp_id
;
5417 SIMAP_FOR_EACH (node
, &ct_tp
->tp
) {
5418 ct_dpif_set_timeout_policy_attr_by_name(&cdtp
, node
->name
, node
->data
);
5421 int err
= ct_dpif_set_timeout_policy(dpif
, &cdtp
);
5423 VLOG_ERR_RL(&rl
, "failed to set timeout policy %"PRIu32
" (%s)",
5424 ct_tp
->tp_id
, ovs_strerror(err
));
5429 clear_existing_ct_timeout_policies(struct dpif_backer
*backer
)
5431 /* In kernel datapath, when OVS starts, there may be some pre-existing
5432 * timeout policies in the kernel. To avoid reassigning the same timeout
5433 * policy ids, we dump all the pre-existing timeout policies and keep
5434 * the ids in the pool. Since OVS will not use those timeout policies
5435 * for new datapath flow, we add them to the kill list and remove
5437 struct ct_dpif_timeout_policy cdtp
;
5440 if (ct_dpif_timeout_policy_dump_start(backer
->dpif
, &state
)) {
5444 while (!ct_dpif_timeout_policy_dump_next(backer
->dpif
, state
, &cdtp
)) {
5445 struct ct_timeout_policy
*ct_tp
= ct_timeout_policy_alloc__();
5446 ct_tp
->tp_id
= cdtp
.id
;
5447 id_pool_add(backer
->tp_ids
, cdtp
.id
);
5448 ovs_list_push_back(&backer
->ct_tp_kill_list
, &ct_tp
->list_node
);
5451 ct_dpif_timeout_policy_dump_done(backer
->dpif
, state
);
5454 #define MAX_TIMEOUT_POLICY_ID UINT32_MAX
5457 ct_zone_config_init(struct dpif_backer
*backer
)
5459 backer
->tp_ids
= id_pool_create(DEFAULT_TP_ID
+ 1,
5460 MAX_TIMEOUT_POLICY_ID
- 1);
5461 cmap_init(&backer
->ct_zones
);
5462 hmap_init(&backer
->ct_tps
);
5463 ovs_list_init(&backer
->ct_tp_kill_list
);
5464 clear_existing_ct_timeout_policies(backer
);
5468 ct_zone_config_uninit(struct dpif_backer
*backer
)
5470 struct ct_zone
*ct_zone
;
5471 CMAP_FOR_EACH (ct_zone
, node
, &backer
->ct_zones
) {
5472 ct_zone_remove_and_destroy(backer
, ct_zone
);
5475 struct ct_timeout_policy
*ct_tp
;
5476 HMAP_FOR_EACH_POP (ct_tp
, node
, &backer
->ct_tps
) {
5477 ct_timeout_policy_destroy(ct_tp
, backer
->tp_ids
);
5480 LIST_FOR_EACH_POP (ct_tp
, list_node
, &backer
->ct_tp_kill_list
) {
5481 ct_timeout_policy_destroy(ct_tp
, backer
->tp_ids
);
5484 id_pool_destroy(backer
->tp_ids
);
5485 cmap_destroy(&backer
->ct_zones
);
5486 hmap_destroy(&backer
->ct_tps
);
5490 ct_zone_timeout_policy_sweep(struct dpif_backer
*backer
)
5492 if (!ovs_list_is_empty(&backer
->ct_tp_kill_list
)
5493 && time_msec() >= timeout_policy_cleanup_timer
) {
5494 struct ct_timeout_policy
*ct_tp
, *next
;
5496 LIST_FOR_EACH_SAFE (ct_tp
, next
, list_node
, &backer
->ct_tp_kill_list
) {
5497 if (!ct_dpif_del_timeout_policy(backer
->dpif
, ct_tp
->tp_id
)) {
5498 ovs_list_remove(&ct_tp
->list_node
);
5499 ct_timeout_policy_destroy(ct_tp
, backer
->tp_ids
);
5501 /* INFO log raised by 'dpif' layer. */
5504 timeout_policy_cleanup_timer
= time_msec() +
5505 TIMEOUT_POLICY_CLEANUP_INTERVAL
;
5510 ct_set_zone_timeout_policy(const char *datapath_type
, uint16_t zone_id
,
5511 struct simap
*timeout_policy
)
5513 struct dpif_backer
*backer
= shash_find_data(&all_dpif_backers
,
5519 struct ct_timeout_policy
*ct_tp
= ct_timeout_policy_lookup(&backer
->ct_tps
,
5522 ct_tp
= ct_timeout_policy_alloc(timeout_policy
, backer
->tp_ids
);
5524 hmap_insert(&backer
->ct_tps
, &ct_tp
->node
, simap_hash(&ct_tp
->tp
));
5525 ct_add_timeout_policy_to_dpif(backer
->dpif
, ct_tp
);
5531 struct ct_zone
*ct_zone
= ct_zone_lookup(&backer
->ct_zones
, zone_id
);
5533 if (ct_zone
->ct_tp
!= ct_tp
) {
5534 /* Update the zone timeout policy. */
5535 ct_timeout_policy_unref(backer
, ct_zone
->ct_tp
);
5536 ct_zone
->ct_tp
= ct_tp
;
5540 struct ct_zone
*new_ct_zone
= ct_zone_alloc(zone_id
);
5541 new_ct_zone
->ct_tp
= ct_tp
;
5542 cmap_insert(&backer
->ct_zones
, &new_ct_zone
->node
,
5543 hash_int(zone_id
, 0));
5549 ct_del_zone_timeout_policy(const char *datapath_type
, uint16_t zone_id
)
5551 struct dpif_backer
*backer
= shash_find_data(&all_dpif_backers
,
5557 struct ct_zone
*ct_zone
= ct_zone_lookup(&backer
->ct_zones
, zone_id
);
5559 ct_timeout_policy_unref(backer
, ct_zone
->ct_tp
);
5560 ct_zone_remove_and_destroy(backer
, ct_zone
);
5565 get_datapath_cap(const char *datapath_type
, struct smap
*cap
)
5567 struct odp_support odp
;
5568 struct dpif_backer_support s
;
5569 struct dpif_backer
*backer
= shash_find_data(&all_dpif_backers
,
5574 s
= backer
->rt_support
;
5577 /* ODP_SUPPORT_FIELDS */
5578 smap_add_format(cap
, "max_vlan_headers", "%"PRIuSIZE
,
5579 odp
.max_vlan_headers
);
5580 smap_add_format(cap
, "max_mpls_depth", "%"PRIuSIZE
, odp
.max_mpls_depth
);
5581 smap_add(cap
, "recirc", odp
.recirc
? "true" : "false");
5582 smap_add(cap
, "ct_state", odp
.ct_state
? "true" : "false");
5583 smap_add(cap
, "ct_zone", odp
.ct_zone
? "true" : "false");
5584 smap_add(cap
, "ct_mark", odp
.ct_mark
? "true" : "false");
5585 smap_add(cap
, "ct_label", odp
.ct_label
? "true" : "false");
5586 smap_add(cap
, "ct_state_nat", odp
.ct_state_nat
? "true" : "false");
5587 smap_add(cap
, "ct_orig_tuple", odp
.ct_orig_tuple
? "true" : "false");
5588 smap_add(cap
, "ct_orig_tuple6", odp
.ct_orig_tuple6
? "true" : "false");
5589 smap_add(cap
, "nd_ext", odp
.nd_ext
? "true" : "false");
5591 /* DPIF_SUPPORT_FIELDS */
5592 smap_add(cap
, "masked_set_action", s
.masked_set_action
? "true" : "false");
5593 smap_add(cap
, "tnl_push_pop", s
.tnl_push_pop
? "true" : "false");
5594 smap_add(cap
, "ufid", s
.ufid
? "true" : "false");
5595 smap_add(cap
, "trunc", s
.trunc
? "true" : "false");
5596 smap_add(cap
, "clone", s
.clone
? "true" : "false");
5597 smap_add(cap
, "sample_nesting", s
.sample_nesting
? "true" : "false");
5598 smap_add(cap
, "ct_eventmask", s
.ct_eventmask
? "true" : "false");
5599 smap_add(cap
, "ct_clear", s
.ct_clear
? "true" : "false");
5600 smap_add_format(cap
, "max_hash_alg", "%"PRIuSIZE
, s
.max_hash_alg
);
5601 smap_add(cap
, "check_pkt_len", s
.check_pkt_len
? "true" : "false");
5602 smap_add(cap
, "ct_timeout", s
.ct_timeout
? "true" : "false");
5603 smap_add(cap
, "explicit_drop_action",
5604 s
.explicit_drop_action
? "true" :"false");
5605 smap_add(cap
, "lb_output_action", s
.lb_output_action
? "true" : "false");
5608 /* Gets timeout policy name in 'backer' based on 'zone', 'dl_type' and
5609 * 'nw_proto'. Returns true if the zone-based timeout policy is configured.
5610 * On success, stores the timeout policy name in 'tp_name', and sets
5611 * 'unwildcard' based on the dpif implementation. If 'unwildcard' is true,
5612 * the returned timeout policy is 'dl_type' and 'nw_proto' specific, and OVS
5613 * needs to unwildcard the datapath flow for this timeout policy in flow
5616 * The caller is responsible for freeing 'tp_name'. */
5618 ofproto_dpif_ct_zone_timeout_policy_get_name(
5619 const struct dpif_backer
*backer
, uint16_t zone
, uint16_t dl_type
,
5620 uint8_t nw_proto
, char **tp_name
, bool *unwildcard
)
5622 if (!ct_dpif_timeout_policy_support_ipproto(nw_proto
)) {
5626 struct ct_zone
*ct_zone
= ct_zone_lookup(&backer
->ct_zones
, zone
);
5632 if (ct_dpif_get_timeout_policy_name(backer
->dpif
,
5633 ct_zone
->ct_tp
->tp_id
, dl_type
,
5634 nw_proto
, tp_name
, &is_generic
)) {
5638 /* Unwildcard datapath flow if it is not a generic timeout policy. */
5639 *unwildcard
= !is_generic
;
5644 set_frag_handling(struct ofproto
*ofproto_
,
5645 enum ofputil_frag_handling frag_handling
)
5647 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
5648 if (frag_handling
!= OFPUTIL_FRAG_REASM
) {
5649 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
5657 nxt_resume(struct ofproto
*ofproto_
,
5658 const struct ofputil_packet_in_private
*pin
)
5660 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
5661 struct dpif_flow_stats stats
;
5662 struct xlate_cache xcache
;
5664 xlate_cache_init(&xcache
);
5666 /* Translate pin into datapath actions. */
5667 uint64_t odp_actions_stub
[1024 / 8];
5668 struct ofpbuf odp_actions
= OFPBUF_STUB_INITIALIZER(odp_actions_stub
);
5669 enum slow_path_reason slow
;
5670 enum ofperr error
= xlate_resume(ofproto
, pin
, &odp_actions
, &slow
,
5673 /* Steal 'pin->packet' and put it into a dp_packet. */
5674 struct dp_packet packet
;
5675 dp_packet_init(&packet
, pin
->base
.packet_len
);
5676 dp_packet_put(&packet
, pin
->base
.packet
, pin
->base
.packet_len
);
5678 /* Run the side effects from the xcache. */
5679 dpif_flow_stats_extract(&flow
, &packet
, time_msec(), &stats
);
5680 ovs_mutex_lock(&ofproto_mutex
);
5681 ofproto_dpif_xcache_execute(ofproto
, &xcache
, &stats
);
5682 ovs_mutex_unlock(&ofproto_mutex
);
5684 pkt_metadata_from_flow(&packet
.md
, &pin
->base
.flow_metadata
.flow
);
5686 /* Fix up in_port. */
5687 packet
.md
.in_port
.odp_port
= pin
->odp_port
;
5689 struct flow headers
;
5690 flow_extract(&packet
, &headers
);
5692 /* Execute the datapath actions on the packet. */
5693 struct dpif_execute execute
= {
5694 .actions
= odp_actions
.data
,
5695 .actions_len
= odp_actions
.size
,
5696 .needs_help
= (slow
& SLOW_ACTION
) != 0,
5700 dpif_execute(ofproto
->backer
->dpif
, &execute
);
5703 ofpbuf_uninit(&odp_actions
);
5704 dp_packet_uninit(&packet
);
5705 xlate_cache_uninit(&xcache
);
5713 set_netflow(struct ofproto
*ofproto_
,
5714 const struct netflow_options
*netflow_options
)
5716 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
5718 if (netflow_options
) {
5719 if (!ofproto
->netflow
) {
5720 ofproto
->netflow
= netflow_create();
5721 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
5723 return netflow_set_options(ofproto
->netflow
, netflow_options
);
5724 } else if (ofproto
->netflow
) {
5725 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
5726 netflow_unref(ofproto
->netflow
);
5727 ofproto
->netflow
= NULL
;
5734 get_netflow_ids(const struct ofproto
*ofproto_
,
5735 uint8_t *engine_type
, uint8_t *engine_id
)
5737 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
5739 dpif_get_netflow_ids(ofproto
->backer
->dpif
, engine_type
, engine_id
);
5742 struct ofproto_dpif
*
5743 ofproto_dpif_lookup_by_name(const char *name
)
5745 struct ofproto_dpif
*ofproto
;
5747 HMAP_FOR_EACH_WITH_HASH (ofproto
, all_ofproto_dpifs_by_name_node
,
5748 hash_string(name
, 0),
5749 &all_ofproto_dpifs_by_name
) {
5750 if (!strcmp(ofproto
->up
.name
, name
)) {
5757 struct ofproto_dpif
*
5758 ofproto_dpif_lookup_by_uuid(const struct uuid
*uuid
)
5760 struct ofproto_dpif
*ofproto
;
5762 HMAP_FOR_EACH_WITH_HASH (ofproto
, all_ofproto_dpifs_by_uuid_node
,
5763 uuid_hash(uuid
), &all_ofproto_dpifs_by_uuid
) {
5764 if (uuid_equals(&ofproto
->uuid
, uuid
)) {
5772 ofproto_unixctl_fdb_flush(struct unixctl_conn
*conn
, int argc
,
5773 const char *argv
[], void *aux OVS_UNUSED
)
5775 struct ofproto_dpif
*ofproto
;
5778 ofproto
= ofproto_dpif_lookup_by_name(argv
[1]);
5780 unixctl_command_reply_error(conn
, "no such bridge");
5783 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
5784 mac_learning_flush(ofproto
->ml
);
5785 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
5787 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_by_name_node
,
5788 &all_ofproto_dpifs_by_name
) {
5789 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
5790 mac_learning_flush(ofproto
->ml
);
5791 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
5795 unixctl_command_reply(conn
, "table successfully flushed");
5799 ofproto_unixctl_mcast_snooping_flush(struct unixctl_conn
*conn
, int argc
,
5800 const char *argv
[], void *aux OVS_UNUSED
)
5802 struct ofproto_dpif
*ofproto
;
5805 ofproto
= ofproto_dpif_lookup_by_name(argv
[1]);
5807 unixctl_command_reply_error(conn
, "no such bridge");
5811 if (!mcast_snooping_enabled(ofproto
->ms
)) {
5812 unixctl_command_reply_error(conn
, "multicast snooping is disabled");
5815 mcast_snooping_mdb_flush(ofproto
->ms
);
5817 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_by_name_node
,
5818 &all_ofproto_dpifs_by_name
) {
5819 if (!mcast_snooping_enabled(ofproto
->ms
)) {
5822 mcast_snooping_mdb_flush(ofproto
->ms
);
5826 unixctl_command_reply(conn
, "table successfully flushed");
5829 static struct ofport_dpif
*
5830 ofbundle_get_a_port(const struct ofbundle
*bundle
)
5832 return CONTAINER_OF(ovs_list_front(&bundle
->ports
), struct ofport_dpif
,
5837 ofproto_unixctl_fdb_show(struct unixctl_conn
*conn
, int argc OVS_UNUSED
,
5838 const char *argv
[], void *aux OVS_UNUSED
)
5840 struct ds ds
= DS_EMPTY_INITIALIZER
;
5841 const struct ofproto_dpif
*ofproto
;
5842 const struct mac_entry
*e
;
5844 ofproto
= ofproto_dpif_lookup_by_name(argv
[1]);
5846 unixctl_command_reply_error(conn
, "no such bridge");
5850 ds_put_cstr(&ds
, " port VLAN MAC Age\n");
5851 ovs_rwlock_rdlock(&ofproto
->ml
->rwlock
);
5852 LIST_FOR_EACH (e
, lru_node
, &ofproto
->ml
->lrus
) {
5853 struct ofbundle
*bundle
= mac_entry_get_port(ofproto
->ml
, e
);
5854 char name
[OFP_MAX_PORT_NAME_LEN
];
5856 ofputil_port_to_string(ofbundle_get_a_port(bundle
)->up
.ofp_port
,
5857 NULL
, name
, sizeof name
);
5858 ds_put_format(&ds
, "%5s %4d "ETH_ADDR_FMT
" %3d\n",
5859 name
, e
->vlan
, ETH_ADDR_ARGS(e
->mac
),
5860 mac_entry_age(ofproto
->ml
, e
));
5862 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
5863 unixctl_command_reply(conn
, ds_cstr(&ds
));
5868 ofproto_unixctl_fdb_stats_clear(struct unixctl_conn
*conn
, int argc
,
5869 const char *argv
[], void *aux OVS_UNUSED
)
5871 struct ofproto_dpif
*ofproto
;
5874 ofproto
= ofproto_dpif_lookup_by_name(argv
[1]);
5876 unixctl_command_reply_error(conn
, "no such bridge");
5879 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
5880 mac_learning_clear_statistics(ofproto
->ml
);
5881 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
5883 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_by_name_node
,
5884 &all_ofproto_dpifs_by_name
) {
5885 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
5886 mac_learning_clear_statistics(ofproto
->ml
);
5887 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
5891 unixctl_command_reply(conn
, "statistics successfully cleared");
5895 ofproto_unixctl_fdb_stats_show(struct unixctl_conn
*conn
, int argc OVS_UNUSED
,
5896 const char *argv
[], void *aux OVS_UNUSED
)
5898 struct ds ds
= DS_EMPTY_INITIALIZER
;
5899 const struct ofproto_dpif
*ofproto
;
5900 ofproto
= ofproto_dpif_lookup_by_name(argv
[1]);
5902 unixctl_command_reply_error(conn
, "no such bridge");
5906 ds_put_format(&ds
, "Statistics for bridge \"%s\":\n", argv
[1]);
5907 ovs_rwlock_rdlock(&ofproto
->ml
->rwlock
);
5909 ds_put_format(&ds
, " Current/maximum MAC entries in the table: %"
5910 PRIuSIZE
"/%"PRIuSIZE
"\n",
5911 hmap_count(&ofproto
->ml
->table
), ofproto
->ml
->max_entries
);
5913 " Total number of learned MAC entries : %"PRIu64
"\n",
5914 ofproto
->ml
->total_learned
);
5916 " Total number of expired MAC entries : %"PRIu64
"\n",
5917 ofproto
->ml
->total_expired
);
5919 " Total number of evicted MAC entries : %"PRIu64
"\n",
5920 ofproto
->ml
->total_evicted
);
5922 " Total number of port moved MAC entries : %"PRIu64
"\n",
5923 ofproto
->ml
->total_moved
);
5925 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
5926 unixctl_command_reply(conn
, ds_cstr(&ds
));
5931 ofproto_unixctl_mcast_snooping_show(struct unixctl_conn
*conn
,
5932 int argc OVS_UNUSED
,
5934 void *aux OVS_UNUSED
)
5936 struct ds ds
= DS_EMPTY_INITIALIZER
;
5937 const struct ofproto_dpif
*ofproto
;
5938 const struct ofbundle
*bundle
;
5939 const struct mcast_group
*grp
;
5940 struct mcast_group_bundle
*b
;
5941 struct mcast_mrouter_bundle
*mrouter
;
5943 ofproto
= ofproto_dpif_lookup_by_name(argv
[1]);
5945 unixctl_command_reply_error(conn
, "no such bridge");
5949 if (!mcast_snooping_enabled(ofproto
->ms
)) {
5950 unixctl_command_reply_error(conn
, "multicast snooping is disabled");
5954 ds_put_cstr(&ds
, " port VLAN GROUP Age\n");
5955 ovs_rwlock_rdlock(&ofproto
->ms
->rwlock
);
5956 LIST_FOR_EACH (grp
, group_node
, &ofproto
->ms
->group_lru
) {
5957 LIST_FOR_EACH(b
, bundle_node
, &grp
->bundle_lru
) {
5958 char name
[OFP_MAX_PORT_NAME_LEN
];
5961 ofputil_port_to_string(ofbundle_get_a_port(bundle
)->up
.ofp_port
,
5962 NULL
, name
, sizeof name
);
5963 ds_put_format(&ds
, "%5s %4d ", name
, grp
->vlan
);
5964 ipv6_format_mapped(&grp
->addr
, &ds
);
5965 ds_put_format(&ds
, " %3d\n",
5966 mcast_bundle_age(ofproto
->ms
, b
));
5970 /* ports connected to multicast routers */
5971 LIST_FOR_EACH(mrouter
, mrouter_node
, &ofproto
->ms
->mrouter_lru
) {
5972 char name
[OFP_MAX_PORT_NAME_LEN
];
5974 bundle
= mrouter
->port
;
5975 ofputil_port_to_string(ofbundle_get_a_port(bundle
)->up
.ofp_port
,
5976 NULL
, name
, sizeof name
);
5977 ds_put_format(&ds
, "%5s %4d querier %3d\n",
5978 name
, mrouter
->vlan
,
5979 mcast_mrouter_age(ofproto
->ms
, mrouter
));
5981 ovs_rwlock_unlock(&ofproto
->ms
->rwlock
);
5982 unixctl_command_reply(conn
, ds_cstr(&ds
));
5986 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5987 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5988 * to destroy 'ofproto_shash' and free the returned value. */
5989 static const struct shash_node
**
5990 get_ofprotos(struct shash
*ofproto_shash
)
5992 const struct ofproto_dpif
*ofproto
;
5994 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_by_name_node
,
5995 &all_ofproto_dpifs_by_name
) {
5996 char *name
= xasprintf("%s@%s", ofproto
->up
.type
, ofproto
->up
.name
);
5997 shash_add_nocopy(ofproto_shash
, name
, ofproto
);
6000 return shash_sort(ofproto_shash
);
6004 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn
*conn
, int argc OVS_UNUSED
,
6005 const char *argv
[] OVS_UNUSED
,
6006 void *aux OVS_UNUSED
)
6008 struct ds ds
= DS_EMPTY_INITIALIZER
;
6009 struct shash ofproto_shash
;
6010 const struct shash_node
**sorted_ofprotos
;
6013 shash_init(&ofproto_shash
);
6014 sorted_ofprotos
= get_ofprotos(&ofproto_shash
);
6015 for (i
= 0; i
< shash_count(&ofproto_shash
); i
++) {
6016 const struct shash_node
*node
= sorted_ofprotos
[i
];
6017 ds_put_format(&ds
, "%s\n", node
->name
);
6020 shash_destroy(&ofproto_shash
);
6021 free(sorted_ofprotos
);
6023 unixctl_command_reply(conn
, ds_cstr(&ds
));
6028 show_dp_feature_bool(struct ds
*ds
, const char *feature
, bool b
)
6030 ds_put_format(ds
, "%s: %s\n", feature
, b
? "Yes" : "No");
6034 show_dp_feature_size_t(struct ds
*ds
, const char *feature
, size_t s
)
6036 ds_put_format(ds
, "%s: %"PRIuSIZE
"\n", feature
, s
);
6039 enum dpif_support_field_type
{
6040 DPIF_SUPPORT_FIELD_bool
,
6041 DPIF_SUPPORT_FIELD_size_t
,
6044 struct dpif_support_field
{
6045 void *rt_ptr
; /* Points to the 'rt_support' field. */
6046 const void *bt_ptr
; /* Points to the 'bt_support' field. */
6048 enum dpif_support_field_type type
;
6051 #define DPIF_SUPPORT_FIELD_INTIALIZER(RT_PTR, BT_PTR, TITLE, TYPE) \
6052 (struct dpif_support_field) {RT_PTR, BT_PTR, TITLE, TYPE}
6055 dpif_show_support(const struct dpif_backer_support
*support
, struct ds
*ds
)
6057 #define DPIF_SUPPORT_FIELD(TYPE, NAME, TITLE) \
6058 show_dp_feature_##TYPE (ds, TITLE, support->NAME);
6060 #undef DPIF_SUPPORT_FIELD
6062 #define ODP_SUPPORT_FIELD(TYPE, NAME, TITLE) \
6063 show_dp_feature_##TYPE (ds, TITLE, support->odp.NAME );
6065 #undef ODP_SUPPORT_FIELD
6069 display_support_field(const char *name
,
6070 const struct dpif_support_field
*field
,
6073 switch (field
->type
) {
6074 case DPIF_SUPPORT_FIELD_bool
: {
6075 bool v
= *(bool *)field
->rt_ptr
;
6076 bool b
= *(bool *)field
->bt_ptr
;
6077 ds_put_format(ds
, "%s (%s) : [run time]:%s, [boot time]:%s\n", name
,
6078 field
->title
, v
? "true" : "false",
6079 b
? "true" : "false");
6082 case DPIF_SUPPORT_FIELD_size_t
:
6083 ds_put_format(ds
, "%s (%s) : [run time]:%"PRIuSIZE
6084 ", [boot time]:%"PRIuSIZE
"\n", name
,
6085 field
->title
, *(size_t *)field
->rt_ptr
,
6086 *(size_t *)field
->bt_ptr
);
6093 /* Set a field of 'rt_support' to a new value.
6095 * Returns 'true' if the value is actually set. */
6097 dpif_set_support(struct dpif_backer_support
*rt_support
,
6098 struct dpif_backer_support
*bt_support
,
6099 const char *name
, const char *value
, struct ds
*ds
)
6101 struct shash all_fields
= SHASH_INITIALIZER(&all_fields
);
6102 struct dpif_support_field
*field
;
6103 struct shash_node
*node
;
6104 bool changed
= false;
6106 #define DPIF_SUPPORT_FIELD(TYPE, NAME, TITLE) \
6108 struct dpif_support_field *f = xmalloc(sizeof *f); \
6109 *f = DPIF_SUPPORT_FIELD_INTIALIZER(&rt_support->NAME, \
6110 &bt_support->NAME, \
6112 DPIF_SUPPORT_FIELD_##TYPE);\
6113 shash_add_once(&all_fields, #NAME, f); \
6115 DPIF_SUPPORT_FIELDS
;
6116 #undef DPIF_SUPPORT_FIELD
6118 #define ODP_SUPPORT_FIELD(TYPE, NAME, TITLE) \
6120 struct dpif_support_field *f = xmalloc(sizeof *f); \
6121 *f = DPIF_SUPPORT_FIELD_INTIALIZER(&rt_support->odp.NAME, \
6122 &bt_support->odp.NAME, \
6124 DPIF_SUPPORT_FIELD_##TYPE);\
6125 shash_add_once(&all_fields, #NAME, f); \
6128 #undef ODP_SUPPORT_FIELD
6131 SHASH_FOR_EACH (node
, &all_fields
) {
6132 display_support_field(node
->name
, node
->data
, ds
);
6137 node
= shash_find(&all_fields
, name
);
6139 ds_put_cstr(ds
, "Unexpected support field");
6145 display_support_field(node
->name
, field
, ds
);
6149 if (field
->type
== DPIF_SUPPORT_FIELD_bool
) {
6150 if (!strcasecmp(value
, "true")) {
6151 if (*(bool *)field
->bt_ptr
) {
6152 *(bool *)field
->rt_ptr
= true;
6155 ds_put_cstr(ds
, "Can not enable features not supported by the datapth");
6157 } else if (!strcasecmp(value
, "false")) {
6158 *(bool *)field
->rt_ptr
= false;
6161 ds_put_cstr(ds
, "Boolean value expected");
6163 } else if (field
->type
== DPIF_SUPPORT_FIELD_size_t
) {
6165 if (str_to_int(value
, 10, &v
)) {
6167 if (v
<= *(size_t *)field
->bt_ptr
) {
6168 *(size_t *)field
->rt_ptr
= v
;
6171 ds_put_cstr(ds
, "Can not set value beyond the datapath capability");
6174 ds_put_format(ds
, "Negative number not expected");
6177 ds_put_cstr(ds
, "Integer number expected");
6182 shash_destroy_free_data(&all_fields
);
6187 dpif_show_backer(const struct dpif_backer
*backer
, struct ds
*ds
)
6189 const struct shash_node
**ofprotos
;
6190 struct dpif_dp_stats dp_stats
;
6191 struct shash ofproto_shash
;
6194 dpif_get_dp_stats(backer
->dpif
, &dp_stats
);
6195 ds_put_format(ds
, "%s: hit:%"PRIu64
" missed:%"PRIu64
"\n",
6196 dpif_name(backer
->dpif
), dp_stats
.n_hit
, dp_stats
.n_missed
);
6198 shash_init(&ofproto_shash
);
6199 ofprotos
= get_ofprotos(&ofproto_shash
);
6200 for (i
= 0; i
< shash_count(&ofproto_shash
); i
++) {
6201 struct ofproto_dpif
*ofproto
= ofprotos
[i
]->data
;
6202 const struct shash_node
**ports
;
6205 if (ofproto
->backer
!= backer
) {
6209 ds_put_format(ds
, " %s:\n", ofproto
->up
.name
);
6211 ports
= shash_sort(&ofproto
->up
.port_by_name
);
6212 for (j
= 0; j
< shash_count(&ofproto
->up
.port_by_name
); j
++) {
6213 const struct shash_node
*node
= ports
[j
];
6214 struct ofport
*ofport
= node
->data
;
6216 odp_port_t odp_port
;
6218 ds_put_format(ds
, " %s %u/", netdev_get_name(ofport
->netdev
),
6221 odp_port
= ofp_port_to_odp_port(ofproto
, ofport
->ofp_port
);
6222 if (odp_port
!= ODPP_NONE
) {
6223 ds_put_format(ds
, "%"PRIu32
":", odp_port
);
6225 ds_put_cstr(ds
, "none:");
6228 ds_put_format(ds
, " (%s", netdev_get_type(ofport
->netdev
));
6231 if (!netdev_get_config(ofport
->netdev
, &config
)) {
6232 const struct smap_node
**nodes
= smap_sort(&config
);
6233 for (size_t k
= 0; k
< smap_count(&config
); k
++) {
6234 ds_put_format(ds
, "%c %s=%s", k
? ',' : ':',
6235 nodes
[k
]->key
, nodes
[k
]->value
);
6239 smap_destroy(&config
);
6241 ds_put_char(ds
, ')');
6242 ds_put_char(ds
, '\n');
6246 shash_destroy(&ofproto_shash
);
6251 ofproto_unixctl_dpif_show(struct unixctl_conn
*conn
, int argc OVS_UNUSED
,
6252 const char *argv
[] OVS_UNUSED
, void *aux OVS_UNUSED
)
6254 struct ds ds
= DS_EMPTY_INITIALIZER
;
6255 const struct shash_node
**backers
;
6258 backers
= shash_sort(&all_dpif_backers
);
6259 for (i
= 0; i
< shash_count(&all_dpif_backers
); i
++) {
6260 dpif_show_backer(backers
[i
]->data
, &ds
);
6264 unixctl_command_reply(conn
, ds_cstr(&ds
));
6269 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn
*conn
,
6270 int argc OVS_UNUSED
, const char *argv
[],
6271 void *aux OVS_UNUSED
)
6273 const struct ofproto_dpif
*ofproto
;
6275 struct ds ds
= DS_EMPTY_INITIALIZER
;
6277 struct dpif_flow_dump
*flow_dump
;
6278 struct dpif_flow_dump_thread
*flow_dump_thread
;
6282 ofproto
= ofproto_dpif_lookup_by_name(argv
[argc
- 1]);
6284 unixctl_command_reply_error(conn
, "no such bridge");
6288 bool verbosity
= false;
6290 bool set_names
= false;
6291 for (int i
= 1; i
< argc
- 1; i
++) {
6292 if (!strcmp(argv
[i
], "-m")) {
6294 } else if (!strcmp(argv
[i
], "--names")) {
6297 } else if (!strcmp(argv
[i
], "--no-names")) {
6306 struct hmap
*portno_names
= NULL
;
6308 portno_names
= xmalloc(sizeof *portno_names
);
6309 hmap_init(portno_names
);
6311 struct dpif_port dpif_port
;
6312 struct dpif_port_dump port_dump
;
6313 DPIF_PORT_FOR_EACH (&dpif_port
, &port_dump
, ofproto
->backer
->dpif
) {
6314 odp_portno_names_set(portno_names
, dpif_port
.port_no
,
6320 flow_dump
= dpif_flow_dump_create(ofproto
->backer
->dpif
, false, NULL
);
6321 flow_dump_thread
= dpif_flow_dump_thread_create(flow_dump
);
6322 while (dpif_flow_dump_next(flow_dump_thread
, &f
, 1)) {
6325 if ((odp_flow_key_to_flow(f
.key
, f
.key_len
, &flow
, NULL
)
6327 || (xlate_lookup_ofproto(ofproto
->backer
, &flow
, NULL
, NULL
)
6333 odp_format_ufid(&f
.ufid
, &ds
);
6334 ds_put_cstr(&ds
, " ");
6336 odp_flow_format(f
.key
, f
.key_len
, f
.mask
, f
.mask_len
,
6337 portno_names
, &ds
, verbosity
);
6338 ds_put_cstr(&ds
, ", ");
6339 dpif_flow_stats_format(&f
.stats
, &ds
);
6340 ds_put_cstr(&ds
, ", actions:");
6341 format_odp_actions(&ds
, f
.actions
, f
.actions_len
, portno_names
);
6342 ds_put_char(&ds
, '\n');
6344 dpif_flow_dump_thread_destroy(flow_dump_thread
);
6345 error
= dpif_flow_dump_destroy(flow_dump
);
6349 ds_put_format(&ds
, "dpif/dump_flows failed: %s", ovs_strerror(errno
));
6350 unixctl_command_reply_error(conn
, ds_cstr(&ds
));
6352 unixctl_command_reply(conn
, ds_cstr(&ds
));
6355 odp_portno_names_destroy(portno_names
);
6356 hmap_destroy(portno_names
);
6363 ofproto_unixctl_dpif_show_dp_features(struct unixctl_conn
*conn
,
6364 int argc
, const char *argv
[],
6365 void *aux OVS_UNUSED
)
6367 struct ds ds
= DS_EMPTY_INITIALIZER
;
6368 const char *br
= argv
[argc
-1];
6369 struct ofproto_dpif
*ofproto
= ofproto_dpif_lookup_by_name(br
);
6372 unixctl_command_reply_error(conn
, "no such bridge");
6376 dpif_show_support(&ofproto
->backer
->bt_support
, &ds
);
6377 unixctl_command_reply(conn
, ds_cstr(&ds
));
6381 ofproto_unixctl_dpif_set_dp_features(struct unixctl_conn
*conn
,
6382 int argc
, const char *argv
[],
6383 void *aux OVS_UNUSED
)
6385 struct ds ds
= DS_EMPTY_INITIALIZER
;
6386 const char *br
= argv
[1];
6387 const char *name
, *value
;
6388 struct ofproto_dpif
*ofproto
= ofproto_dpif_lookup_by_name(br
);
6392 unixctl_command_reply_error(conn
, "no such bridge");
6396 name
= argc
> 2 ? argv
[2] : NULL
;
6397 value
= argc
> 3 ? argv
[3] : NULL
;
6398 changed
= dpif_set_support(&ofproto
->backer
->rt_support
,
6399 &ofproto
->backer
->bt_support
,
6402 xlate_set_support(ofproto
, &ofproto
->backer
->rt_support
);
6403 udpif_flush(ofproto
->backer
->udpif
);
6405 unixctl_command_reply(conn
, ds_cstr(&ds
));
6410 ofproto_unixctl_init(void)
6412 static bool registered
;
6418 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6419 ofproto_unixctl_fdb_flush
, NULL
);
6420 unixctl_command_register("fdb/show", "bridge", 1, 1,
6421 ofproto_unixctl_fdb_show
, NULL
);
6422 unixctl_command_register("fdb/stats-clear", "[bridge]", 0, 1,
6423 ofproto_unixctl_fdb_stats_clear
, NULL
);
6424 unixctl_command_register("fdb/stats-show", "bridge", 1, 1,
6425 ofproto_unixctl_fdb_stats_show
, NULL
);
6426 unixctl_command_register("mdb/flush", "[bridge]", 0, 1,
6427 ofproto_unixctl_mcast_snooping_flush
, NULL
);
6428 unixctl_command_register("mdb/show", "bridge", 1, 1,
6429 ofproto_unixctl_mcast_snooping_show
, NULL
);
6430 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6431 ofproto_unixctl_dpif_dump_dps
, NULL
);
6432 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show
,
6434 unixctl_command_register("dpif/show-dp-features", "bridge", 1, 1,
6435 ofproto_unixctl_dpif_show_dp_features
, NULL
);
6436 unixctl_command_register("dpif/dump-flows",
6437 "[-m] [--names | --no-names] bridge", 1, INT_MAX
,
6438 ofproto_unixctl_dpif_dump_flows
, NULL
);
6439 unixctl_command_register("dpif/set-dp-features", "bridge", 1, 3 ,
6440 ofproto_unixctl_dpif_set_dp_features
, NULL
);
6444 ofp_port_to_odp_port(const struct ofproto_dpif
*ofproto
, ofp_port_t ofp_port
)
6446 const struct ofport_dpif
*ofport
= ofp_port_to_ofport(ofproto
, ofp_port
);
6447 return ofport
? ofport
->odp_port
: ODPP_NONE
;
6450 struct ofport_dpif
*
6451 odp_port_to_ofport(const struct dpif_backer
*backer
, odp_port_t odp_port
)
6453 struct ofport_dpif
*port
;
6455 ovs_rwlock_rdlock(&backer
->odp_to_ofport_lock
);
6456 HMAP_FOR_EACH_IN_BUCKET (port
, odp_port_node
, hash_odp_port(odp_port
),
6457 &backer
->odp_to_ofport_map
) {
6458 if (port
->odp_port
== odp_port
) {
6459 ovs_rwlock_unlock(&backer
->odp_to_ofport_lock
);
6464 ovs_rwlock_unlock(&backer
->odp_to_ofport_lock
);
6469 odp_port_to_ofp_port(const struct ofproto_dpif
*ofproto
, odp_port_t odp_port
)
6471 struct ofport_dpif
*port
;
6473 port
= odp_port_to_ofport(ofproto
->backer
, odp_port
);
6474 if (port
&& &ofproto
->up
== port
->up
.ofproto
) {
6475 return port
->up
.ofp_port
;
6481 /* 'match' is non-const to allow for temporary modifications. Any changes are
6482 * restored before returning. */
6484 ofproto_dpif_add_internal_flow(struct ofproto_dpif
*ofproto
,
6485 struct match
*match
, int priority
,
6486 uint16_t idle_timeout
,
6487 const struct ofpbuf
*ofpacts
,
6488 struct rule
**rulep
)
6490 struct ofputil_flow_mod fm
;
6491 struct rule_dpif
*rule
;
6494 fm
= (struct ofputil_flow_mod
) {
6495 .buffer_id
= UINT32_MAX
,
6496 .priority
= priority
,
6497 .table_id
= TBL_INTERNAL
,
6498 .command
= OFPFC_ADD
,
6499 .idle_timeout
= idle_timeout
,
6500 .flags
= OFPUTIL_FF_HIDDEN_FIELDS
| OFPUTIL_FF_NO_READONLY
,
6501 .ofpacts
= ofpacts
->data
,
6502 .ofpacts_len
= ofpacts
->size
,
6504 minimatch_init(&fm
.match
, match
);
6505 error
= ofproto_flow_mod(&ofproto
->up
, &fm
);
6506 minimatch_destroy(&fm
.match
);
6509 VLOG_ERR_RL(&rl
, "failed to add internal flow (%s)",
6510 ofperr_to_string(error
));
6515 rule
= rule_dpif_lookup_in_table(ofproto
,
6516 ofproto_dpif_get_tables_version(ofproto
),
6517 TBL_INTERNAL
, &match
->flow
, &match
->wc
);
6527 ofproto_dpif_delete_internal_flow(struct ofproto_dpif
*ofproto
,
6528 struct match
*match
, int priority
)
6530 struct ofputil_flow_mod fm
;
6533 fm
= (struct ofputil_flow_mod
) {
6534 .buffer_id
= UINT32_MAX
,
6535 .priority
= priority
,
6536 .table_id
= TBL_INTERNAL
,
6537 .out_port
= OFPP_ANY
,
6538 .out_group
= OFPG_ANY
,
6539 .flags
= OFPUTIL_FF_HIDDEN_FIELDS
| OFPUTIL_FF_NO_READONLY
,
6540 .command
= OFPFC_DELETE_STRICT
,
6542 minimatch_init(&fm
.match
, match
);
6543 error
= ofproto_flow_mod(&ofproto
->up
, &fm
);
6544 minimatch_destroy(&fm
.match
);
6547 VLOG_ERR_RL(&rl
, "failed to delete internal flow (%s)",
6548 ofperr_to_string(error
));
6556 meter_get_features(const struct ofproto
*ofproto_
,
6557 struct ofputil_meter_features
*features
)
6559 const struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
6561 dpif_meter_get_features(ofproto
->backer
->dpif
, features
);
6565 meter_set(struct ofproto
*ofproto_
, ofproto_meter_id
*meter_id
,
6566 struct ofputil_meter_config
*config
)
6568 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
6570 /* Provider ID unknown. Use backer to allocate a new DP meter */
6571 if (meter_id
->uint32
== UINT32_MAX
) {
6572 if (!ofproto
->backer
->meter_ids
) {
6573 return OFPERR_OFPMMFC_OUT_OF_METERS
; /* Meters not supported. */
6576 if(!id_pool_alloc_id(ofproto
->backer
->meter_ids
, &meter_id
->uint32
)) {
6577 return OFPERR_OFPMMFC_OUT_OF_METERS
; /* Can't allocate meter. */
6581 switch (dpif_meter_set(ofproto
->backer
->dpif
, *meter_id
, config
)) {
6584 case EFBIG
: /* meter_id out of range */
6585 case ENOMEM
: /* Cannot allocate meter */
6586 return OFPERR_OFPMMFC_OUT_OF_METERS
;
6587 case EBADF
: /* Unsupported flags */
6588 return OFPERR_OFPMMFC_BAD_FLAGS
;
6589 case EINVAL
: /* Too many bands */
6590 return OFPERR_OFPMMFC_OUT_OF_BANDS
;
6591 case ENODEV
: /* Unsupported band type */
6592 return OFPERR_OFPMMFC_BAD_BAND
;
6593 case EDOM
: /* Rate must be non-zero */
6594 return OFPERR_OFPMMFC_BAD_RATE
;
6596 return OFPERR_OFPMMFC_UNKNOWN
;
6601 meter_get(const struct ofproto
*ofproto_
, ofproto_meter_id meter_id
,
6602 struct ofputil_meter_stats
*stats
, uint16_t n_bands
)
6604 const struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
6606 if (!dpif_meter_get(ofproto
->backer
->dpif
, meter_id
, stats
, n_bands
)) {
6609 return OFPERR_OFPMMFC_UNKNOWN_METER
;
6612 struct free_meter_id_args
{
6613 struct ofproto_dpif
*ofproto
;
6614 ofproto_meter_id meter_id
;
6618 free_meter_id(struct free_meter_id_args
*args
)
6620 struct ofproto_dpif
*ofproto
= args
->ofproto
;
6622 dpif_meter_del(ofproto
->backer
->dpif
, args
->meter_id
, NULL
, 0);
6623 id_pool_free_id(ofproto
->backer
->meter_ids
, args
->meter_id
.uint32
);
6628 meter_del(struct ofproto
*ofproto_
, ofproto_meter_id meter_id
)
6630 struct free_meter_id_args
*arg
= xmalloc(sizeof *arg
);
6632 /* Before a meter can be deleted, Openflow spec requires all rules
6633 * referring to the meter to be (automatically) removed before the
6634 * meter is deleted. However, since vswitchd is multi-threaded,
6635 * those rules and their actions remain accessible by other threads,
6636 * especially by the handler and revalidator threads.
6637 * Postpone meter deletion after RCU grace period, so that ongoing
6638 * upcall translation or flow revalidation can complete. */
6639 arg
->ofproto
= ofproto_dpif_cast(ofproto_
);
6640 arg
->meter_id
= meter_id
;
6641 ovsrcu_postpone(free_meter_id
, arg
);
6644 const struct ofproto_class ofproto_dpif_class
= {
6658 NULL
, /* get_memory_usage. */
6659 type_get_memory_usage
,
6662 NULL
, /* modify_tables */
6681 port_is_lacp_current
,
6682 port_get_lacp_stats
,
6683 NULL
, /* rule_choose_table */
6687 NULL
, /* rule_delete */
6692 packet_xlate_revert
,
6693 packet_execute_prepare
,
6711 aa_vlan_get_queue_size
,
6718 get_stp_port_status
,
6723 get_rstp_port_status
,
6730 is_mirror_output_bundle
,
6731 forward_bpdu_changed
,
6732 set_mac_table_config
,
6734 set_mcast_snooping_port
,
6739 group_alloc
, /* group_alloc */
6740 group_construct
, /* group_construct */
6741 group_destruct
, /* group_destruct */
6742 group_dealloc
, /* group_dealloc */
6743 NULL
, /* group_modify */
6744 group_get_stats
, /* group_get_stats */
6745 get_datapath_version
, /* get_datapath_version */
6748 ct_flush
, /* ct_flush */
6749 ct_set_zone_timeout_policy
,
6750 ct_del_zone_timeout_policy
,