2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-dpif.h"
20 #include "ofproto/ofproto-provider.h"
27 #include "byte-order.h"
28 #include "connectivity.h"
33 #include "dynamic-string.h"
34 #include "fail-open.h"
35 #include "guarded-list.h"
39 #include "mac-learning.h"
40 #include "meta-flow.h"
41 #include "multipath.h"
42 #include "netdev-vport.h"
47 #include "odp-execute.h"
50 #include "ofp-actions.h"
51 #include "ofp-parse.h"
52 #include "ofp-print.h"
53 #include "ofproto-dpif-ipfix.h"
54 #include "ofproto-dpif-mirror.h"
55 #include "ofproto-dpif-monitor.h"
56 #include "ofproto-dpif-rid.h"
57 #include "ofproto-dpif-sflow.h"
58 #include "ofproto-dpif-upcall.h"
59 #include "ofproto-dpif-xlate.h"
60 #include "poll-loop.h"
66 #include "unaligned.h"
68 #include "vlan-bitmap.h"
71 VLOG_DEFINE_THIS_MODULE(ofproto_dpif
);
73 COVERAGE_DEFINE(ofproto_dpif_expired
);
74 COVERAGE_DEFINE(packet_in_overflow
);
76 /* Number of implemented OpenFlow tables. */
77 enum { N_TABLES
= 255 };
78 enum { TBL_INTERNAL
= N_TABLES
- 1 }; /* Used for internal hidden rules. */
79 BUILD_ASSERT_DECL(N_TABLES
>= 2 && N_TABLES
<= 255);
88 * - Do include packets and bytes from datapath flows which have not
89 * recently been processed by a revalidator. */
90 struct ovs_mutex stats_mutex
;
91 struct dpif_flow_stats stats OVS_GUARDED
;
94 static void rule_get_stats(struct rule
*, uint64_t *packets
, uint64_t *bytes
,
96 static struct rule_dpif
*rule_dpif_cast(const struct rule
*);
97 static void rule_expire(struct rule_dpif
*);
104 * - Do include packets and bytes from datapath flows which have not
105 * recently been processed by a revalidator. */
106 struct ovs_mutex stats_mutex
;
107 uint64_t packet_count OVS_GUARDED
; /* Number of packets received. */
108 uint64_t byte_count OVS_GUARDED
; /* Number of bytes received. */
109 struct bucket_counter
*bucket_stats OVS_GUARDED
; /* Bucket statistics. */
113 struct hmap_node hmap_node
; /* In struct ofproto's "bundles" hmap. */
114 struct ofproto_dpif
*ofproto
; /* Owning ofproto. */
115 void *aux
; /* Key supplied by ofproto's client. */
116 char *name
; /* Identifier for log messages. */
119 struct list ports
; /* Contains "struct ofport"s. */
120 enum port_vlan_mode vlan_mode
; /* VLAN mode */
121 int vlan
; /* -1=trunk port, else a 12-bit VLAN ID. */
122 unsigned long *trunks
; /* Bitmap of trunked VLANs, if 'vlan' == -1.
123 * NULL if all VLANs are trunked. */
124 struct lacp
*lacp
; /* LACP if LACP is enabled, otherwise NULL. */
125 struct bond
*bond
; /* Nonnull iff more than one port. */
126 bool use_priority_tags
; /* Use 802.1p tag for frames in VLAN 0? */
129 bool floodable
; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
132 static void bundle_remove(struct ofport
*);
133 static void bundle_update(struct ofbundle
*);
134 static void bundle_destroy(struct ofbundle
*);
135 static void bundle_del_port(struct ofport_dpif
*);
136 static void bundle_run(struct ofbundle
*);
137 static void bundle_wait(struct ofbundle
*);
139 static void stp_run(struct ofproto_dpif
*ofproto
);
140 static void stp_wait(struct ofproto_dpif
*ofproto
);
141 static int set_stp_port(struct ofport
*,
142 const struct ofproto_port_stp_settings
*);
145 struct hmap_node odp_port_node
; /* In dpif_backer's "odp_to_ofport_map". */
149 struct ofbundle
*bundle
; /* Bundle that contains this port, if any. */
150 struct list bundle_node
; /* In struct ofbundle's "ports" list. */
151 struct cfm
*cfm
; /* Connectivity Fault Management, if any. */
152 struct bfd
*bfd
; /* BFD, if any. */
153 bool may_enable
; /* May be enabled in bonds. */
154 bool is_tunnel
; /* This port is a tunnel. */
155 bool is_layer3
; /* This is a layer 3 port. */
156 long long int carrier_seq
; /* Carrier status changes. */
157 struct ofport_dpif
*peer
; /* Peer if patch port. */
160 struct stp_port
*stp_port
; /* Spanning Tree Protocol, if any. */
161 enum stp_state stp_state
; /* Always STP_DISABLED if STP not in use. */
162 long long int stp_state_entered
;
164 /* Queue to DSCP mapping. */
165 struct ofproto_port_queue
*qdscp
;
168 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
170 * This is deprecated. It is only for compatibility with broken device
171 * drivers in old versions of Linux that do not properly support VLANs when
172 * VLAN devices are not used. When broken device drivers are no longer in
173 * widespread use, we will delete these interfaces. */
174 ofp_port_t realdev_ofp_port
;
178 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
180 * This is deprecated. It is only for compatibility with broken device drivers
181 * in old versions of Linux that do not properly support VLANs when VLAN
182 * devices are not used. When broken device drivers are no longer in
183 * widespread use, we will delete these interfaces. */
184 struct vlan_splinter
{
185 struct hmap_node realdev_vid_node
;
186 struct hmap_node vlandev_node
;
187 ofp_port_t realdev_ofp_port
;
188 ofp_port_t vlandev_ofp_port
;
192 static void vsp_remove(struct ofport_dpif
*);
193 static void vsp_add(struct ofport_dpif
*, ofp_port_t realdev_ofp_port
, int vid
);
195 static odp_port_t
ofp_port_to_odp_port(const struct ofproto_dpif
*,
198 static ofp_port_t
odp_port_to_ofp_port(const struct ofproto_dpif
*,
201 static struct ofport_dpif
*
202 ofport_dpif_cast(const struct ofport
*ofport
)
204 return ofport
? CONTAINER_OF(ofport
, struct ofport_dpif
, up
) : NULL
;
207 static void port_run(struct ofport_dpif
*);
208 static int set_bfd(struct ofport
*, const struct smap
*);
209 static int set_cfm(struct ofport
*, const struct cfm_settings
*);
210 static void ofport_update_peer(struct ofport_dpif
*);
212 struct dpif_completion
{
213 struct list list_node
;
214 struct ofoperation
*op
;
217 /* Reasons that we might need to revalidate every datapath flow, and
218 * corresponding coverage counters.
220 * A value of 0 means that there is no need to revalidate.
222 * It would be nice to have some cleaner way to integrate with coverage
223 * counters, but with only a few reasons I guess this is good enough for
225 enum revalidate_reason
{
226 REV_RECONFIGURE
= 1, /* Switch configuration changed. */
227 REV_STP
, /* Spanning tree protocol port status change. */
228 REV_BOND
, /* Bonding changed. */
229 REV_PORT_TOGGLED
, /* Port enabled or disabled by CFM, LACP, ...*/
230 REV_FLOW_TABLE
, /* Flow table changed. */
231 REV_MAC_LEARNING
, /* Mac learning changed. */
233 COVERAGE_DEFINE(rev_reconfigure
);
234 COVERAGE_DEFINE(rev_stp
);
235 COVERAGE_DEFINE(rev_bond
);
236 COVERAGE_DEFINE(rev_port_toggled
);
237 COVERAGE_DEFINE(rev_flow_table
);
238 COVERAGE_DEFINE(rev_mac_learning
);
240 /* All datapaths of a given type share a single dpif backer instance. */
247 struct ovs_rwlock odp_to_ofport_lock
;
248 struct hmap odp_to_ofport_map OVS_GUARDED
; /* Contains "struct ofport"s. */
250 struct simap tnl_backers
; /* Set of dpif ports backing tunnels. */
252 enum revalidate_reason need_revalidate
; /* Revalidate all flows. */
254 bool recv_set_enable
; /* Enables or disables receiving packets. */
257 struct recirc_id_pool
*rid_pool
; /* Recirculation ID pool. */
258 bool enable_recirc
; /* True if the datapath supports recirculation */
260 /* True if the datapath supports variable-length
261 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.
262 * False if the datapath supports only 8-byte (or shorter) userdata. */
263 bool variable_length_userdata
;
265 /* Maximum number of MPLS label stack entries that the datapath supports
267 size_t max_mpls_depth
;
270 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
271 static struct shash all_dpif_backers
= SHASH_INITIALIZER(&all_dpif_backers
);
273 struct ofproto_dpif
{
274 struct hmap_node all_ofproto_dpifs_node
; /* In 'all_ofproto_dpifs'. */
276 struct dpif_backer
*backer
;
278 uint64_t dump_seq
; /* Last read of udpif_dump_seq(). */
280 /* Special OpenFlow rules. */
281 struct rule_dpif
*miss_rule
; /* Sends flow table misses to controller. */
282 struct rule_dpif
*no_packet_in_rule
; /* Drops flow table misses. */
283 struct rule_dpif
*drop_frags_rule
; /* Used in OFPC_FRAG_DROP mode. */
286 struct netflow
*netflow
;
287 struct dpif_sflow
*sflow
;
288 struct dpif_ipfix
*ipfix
;
289 struct hmap bundles
; /* Contains "struct ofbundle"s. */
290 struct mac_learning
*ml
;
291 bool has_bonded_bundles
;
293 struct mbridge
*mbridge
;
295 struct ovs_mutex stats_mutex
;
296 struct netdev_stats stats OVS_GUARDED
; /* To account packets generated and
297 * consumed in userspace. */
301 long long int stp_last_tick
;
303 /* VLAN splinters. */
304 struct ovs_mutex vsp_mutex
;
305 struct hmap realdev_vid_map OVS_GUARDED
; /* (realdev,vid) -> vlandev. */
306 struct hmap vlandev_map OVS_GUARDED
; /* vlandev -> (realdev,vid). */
309 struct sset ports
; /* Set of standard port names. */
310 struct sset ghost_ports
; /* Ports with no datapath port. */
311 struct sset port_poll_set
; /* Queued names for port_poll() reply. */
312 int port_poll_errno
; /* Last errno for port_poll() reply. */
313 uint64_t change_seq
; /* Connectivity status changes. */
316 struct guarded_list pins
; /* Contains "struct ofputil_packet_in"s. */
319 /* All existing ofproto_dpif instances, indexed by ->up.name. */
320 static struct hmap all_ofproto_dpifs
= HMAP_INITIALIZER(&all_ofproto_dpifs
);
322 static void ofproto_dpif_unixctl_init(void);
324 static inline struct ofproto_dpif
*
325 ofproto_dpif_cast(const struct ofproto
*ofproto
)
327 ovs_assert(ofproto
->ofproto_class
== &ofproto_dpif_class
);
328 return CONTAINER_OF(ofproto
, struct ofproto_dpif
, up
);
332 ofproto_dpif_get_max_mpls_depth(const struct ofproto_dpif
*ofproto
)
334 return ofproto
->backer
->max_mpls_depth
;
338 ofproto_dpif_get_enable_recirc(const struct ofproto_dpif
*ofproto
)
340 return ofproto
->backer
->enable_recirc
;
343 static struct ofport_dpif
*get_ofp_port(const struct ofproto_dpif
*ofproto
,
344 ofp_port_t ofp_port
);
345 static void ofproto_trace(struct ofproto_dpif
*, struct flow
*,
346 const struct ofpbuf
*packet
,
347 const struct ofpact
[], size_t ofpacts_len
,
350 /* Global variables. */
351 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
353 /* Initial mappings of port to bridge mappings. */
354 static struct shash init_ofp_ports
= SHASH_INITIALIZER(&init_ofp_ports
);
356 /* Executes 'fm'. The caller retains ownership of 'fm' and everything in
359 ofproto_dpif_flow_mod(struct ofproto_dpif
*ofproto
,
360 struct ofputil_flow_mod
*fm
)
362 ofproto_flow_mod(&ofproto
->up
, fm
);
365 /* Resets the modified time for 'rule' or an equivalent rule. If 'rule' is not
366 * in the classifier, but an equivalent rule is, unref 'rule' and ref the new
367 * rule. Otherwise if 'rule' is no longer installed in the classifier,
370 * Returns the rule whose modified time has been reset. */
372 ofproto_dpif_refresh_rule(struct rule_dpif
*rule
)
374 return rule_dpif_cast(ofproto_refresh_rule(&rule
->up
));
377 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
378 * Takes ownership of 'pin' and pin->packet. */
380 ofproto_dpif_send_packet_in(struct ofproto_dpif
*ofproto
,
381 struct ofproto_packet_in
*pin
)
383 if (!guarded_list_push_back(&ofproto
->pins
, &pin
->list_node
, 1024)) {
384 COVERAGE_INC(packet_in_overflow
);
385 free(CONST_CAST(void *, pin
->up
.packet
));
390 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
391 * packet rather than to send the packet to the controller.
393 * This function returns false to indicate that a packet_in message
394 * for a "table-miss" should be sent to at least one controller.
395 * False otherwise. */
397 ofproto_dpif_wants_packet_in_on_miss(struct ofproto_dpif
*ofproto
)
399 return connmgr_wants_packet_in_on_miss(ofproto
->up
.connmgr
);
402 /* Factory functions. */
405 init(const struct shash
*iface_hints
)
407 struct shash_node
*node
;
409 /* Make a local copy, since we don't own 'iface_hints' elements. */
410 SHASH_FOR_EACH(node
, iface_hints
) {
411 const struct iface_hint
*orig_hint
= node
->data
;
412 struct iface_hint
*new_hint
= xmalloc(sizeof *new_hint
);
414 new_hint
->br_name
= xstrdup(orig_hint
->br_name
);
415 new_hint
->br_type
= xstrdup(orig_hint
->br_type
);
416 new_hint
->ofp_port
= orig_hint
->ofp_port
;
418 shash_add(&init_ofp_ports
, node
->name
, new_hint
);
423 enumerate_types(struct sset
*types
)
425 dp_enumerate_types(types
);
429 enumerate_names(const char *type
, struct sset
*names
)
431 struct ofproto_dpif
*ofproto
;
434 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
435 if (strcmp(type
, ofproto
->up
.type
)) {
438 sset_add(names
, ofproto
->up
.name
);
445 del(const char *type
, const char *name
)
450 error
= dpif_open(name
, type
, &dpif
);
452 error
= dpif_delete(dpif
);
459 port_open_type(const char *datapath_type
, const char *port_type
)
461 return dpif_port_open_type(datapath_type
, port_type
);
464 /* Type functions. */
466 static void process_dpif_port_changes(struct dpif_backer
*);
467 static void process_dpif_all_ports_changed(struct dpif_backer
*);
468 static void process_dpif_port_change(struct dpif_backer
*,
469 const char *devname
);
470 static void process_dpif_port_error(struct dpif_backer
*, int error
);
472 static struct ofproto_dpif
*
473 lookup_ofproto_dpif_by_port_name(const char *name
)
475 struct ofproto_dpif
*ofproto
;
477 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
478 if (sset_contains(&ofproto
->ports
, name
)) {
487 type_run(const char *type
)
489 struct dpif_backer
*backer
;
491 backer
= shash_find_data(&all_dpif_backers
, type
);
493 /* This is not necessarily a problem, since backers are only
494 * created on demand. */
498 dpif_run(backer
->dpif
);
500 /* If vswitchd started with other_config:flow_restore_wait set as "true",
501 * and the configuration has now changed to "false", enable receiving
502 * packets from the datapath. */
503 if (!backer
->recv_set_enable
&& !ofproto_get_flow_restore_wait()) {
506 backer
->recv_set_enable
= true;
508 error
= dpif_recv_set(backer
->dpif
, backer
->recv_set_enable
);
510 VLOG_ERR("Failed to enable receiving packets in dpif.");
513 dpif_flow_flush(backer
->dpif
);
514 backer
->need_revalidate
= REV_RECONFIGURE
;
517 if (backer
->recv_set_enable
) {
518 udpif_set_threads(backer
->udpif
, n_handlers
, n_revalidators
);
521 if (backer
->need_revalidate
) {
522 struct ofproto_dpif
*ofproto
;
523 struct simap_node
*node
;
524 struct simap tmp_backers
;
526 /* Handle tunnel garbage collection. */
527 simap_init(&tmp_backers
);
528 simap_swap(&backer
->tnl_backers
, &tmp_backers
);
530 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
531 struct ofport_dpif
*iter
;
533 if (backer
!= ofproto
->backer
) {
537 HMAP_FOR_EACH (iter
, up
.hmap_node
, &ofproto
->up
.ports
) {
538 char namebuf
[NETDEV_VPORT_NAME_BUFSIZE
];
541 if (!iter
->is_tunnel
) {
545 dp_port
= netdev_vport_get_dpif_port(iter
->up
.netdev
,
546 namebuf
, sizeof namebuf
);
547 node
= simap_find(&tmp_backers
, dp_port
);
549 simap_put(&backer
->tnl_backers
, dp_port
, node
->data
);
550 simap_delete(&tmp_backers
, node
);
551 node
= simap_find(&backer
->tnl_backers
, dp_port
);
553 node
= simap_find(&backer
->tnl_backers
, dp_port
);
555 odp_port_t odp_port
= ODPP_NONE
;
557 if (!dpif_port_add(backer
->dpif
, iter
->up
.netdev
,
559 simap_put(&backer
->tnl_backers
, dp_port
,
560 odp_to_u32(odp_port
));
561 node
= simap_find(&backer
->tnl_backers
, dp_port
);
566 iter
->odp_port
= node
? u32_to_odp(node
->data
) : ODPP_NONE
;
567 if (tnl_port_reconfigure(iter
, iter
->up
.netdev
,
569 backer
->need_revalidate
= REV_RECONFIGURE
;
574 SIMAP_FOR_EACH (node
, &tmp_backers
) {
575 dpif_port_del(backer
->dpif
, u32_to_odp(node
->data
));
577 simap_destroy(&tmp_backers
);
579 switch (backer
->need_revalidate
) {
580 case REV_RECONFIGURE
: COVERAGE_INC(rev_reconfigure
); break;
581 case REV_STP
: COVERAGE_INC(rev_stp
); break;
582 case REV_BOND
: COVERAGE_INC(rev_bond
); break;
583 case REV_PORT_TOGGLED
: COVERAGE_INC(rev_port_toggled
); break;
584 case REV_FLOW_TABLE
: COVERAGE_INC(rev_flow_table
); break;
585 case REV_MAC_LEARNING
: COVERAGE_INC(rev_mac_learning
); break;
587 backer
->need_revalidate
= 0;
589 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
590 struct ofport_dpif
*ofport
;
591 struct ofbundle
*bundle
;
593 if (ofproto
->backer
!= backer
) {
597 ovs_rwlock_wrlock(&xlate_rwlock
);
598 xlate_ofproto_set(ofproto
, ofproto
->up
.name
,
599 ofproto
->backer
->dpif
, ofproto
->miss_rule
,
600 ofproto
->no_packet_in_rule
, ofproto
->ml
,
601 ofproto
->stp
, ofproto
->mbridge
,
602 ofproto
->sflow
, ofproto
->ipfix
,
603 ofproto
->netflow
, ofproto
->up
.frag_handling
,
604 ofproto
->up
.forward_bpdu
,
605 connmgr_has_in_band(ofproto
->up
.connmgr
),
606 ofproto
->backer
->enable_recirc
,
607 ofproto
->backer
->variable_length_userdata
,
608 ofproto
->backer
->max_mpls_depth
);
610 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
611 xlate_bundle_set(ofproto
, bundle
, bundle
->name
,
612 bundle
->vlan_mode
, bundle
->vlan
,
613 bundle
->trunks
, bundle
->use_priority_tags
,
614 bundle
->bond
, bundle
->lacp
,
618 HMAP_FOR_EACH (ofport
, up
.hmap_node
, &ofproto
->up
.ports
) {
619 int stp_port
= ofport
->stp_port
620 ? stp_port_no(ofport
->stp_port
)
622 xlate_ofport_set(ofproto
, ofport
->bundle
, ofport
,
623 ofport
->up
.ofp_port
, ofport
->odp_port
,
624 ofport
->up
.netdev
, ofport
->cfm
,
625 ofport
->bfd
, ofport
->peer
, stp_port
,
626 ofport
->qdscp
, ofport
->n_qdscp
,
627 ofport
->up
.pp
.config
, ofport
->up
.pp
.state
,
628 ofport
->is_tunnel
, ofport
->may_enable
);
630 ovs_rwlock_unlock(&xlate_rwlock
);
633 udpif_revalidate(backer
->udpif
);
636 process_dpif_port_changes(backer
);
641 /* Check for and handle port changes in 'backer''s dpif. */
643 process_dpif_port_changes(struct dpif_backer
*backer
)
649 error
= dpif_port_poll(backer
->dpif
, &devname
);
655 process_dpif_all_ports_changed(backer
);
659 process_dpif_port_change(backer
, devname
);
664 process_dpif_port_error(backer
, error
);
671 process_dpif_all_ports_changed(struct dpif_backer
*backer
)
673 struct ofproto_dpif
*ofproto
;
674 struct dpif_port dpif_port
;
675 struct dpif_port_dump dump
;
676 struct sset devnames
;
679 sset_init(&devnames
);
680 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
681 if (ofproto
->backer
== backer
) {
682 struct ofport
*ofport
;
684 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->up
.ports
) {
685 sset_add(&devnames
, netdev_get_name(ofport
->netdev
));
689 DPIF_PORT_FOR_EACH (&dpif_port
, &dump
, backer
->dpif
) {
690 sset_add(&devnames
, dpif_port
.name
);
693 SSET_FOR_EACH (devname
, &devnames
) {
694 process_dpif_port_change(backer
, devname
);
696 sset_destroy(&devnames
);
700 process_dpif_port_change(struct dpif_backer
*backer
, const char *devname
)
702 struct ofproto_dpif
*ofproto
;
703 struct dpif_port port
;
705 /* Don't report on the datapath's device. */
706 if (!strcmp(devname
, dpif_base_name(backer
->dpif
))) {
710 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
,
711 &all_ofproto_dpifs
) {
712 if (simap_contains(&ofproto
->backer
->tnl_backers
, devname
)) {
717 ofproto
= lookup_ofproto_dpif_by_port_name(devname
);
718 if (dpif_port_query_by_name(backer
->dpif
, devname
, &port
)) {
719 /* The port was removed. If we know the datapath,
720 * report it through poll_set(). If we don't, it may be
721 * notifying us of a removal we initiated, so ignore it.
722 * If there's a pending ENOBUFS, let it stand, since
723 * everything will be reevaluated. */
724 if (ofproto
&& ofproto
->port_poll_errno
!= ENOBUFS
) {
725 sset_add(&ofproto
->port_poll_set
, devname
);
726 ofproto
->port_poll_errno
= 0;
728 } else if (!ofproto
) {
729 /* The port was added, but we don't know with which
730 * ofproto we should associate it. Delete it. */
731 dpif_port_del(backer
->dpif
, port
.port_no
);
733 struct ofport_dpif
*ofport
;
735 ofport
= ofport_dpif_cast(shash_find_data(
736 &ofproto
->up
.port_by_name
, devname
));
738 && ofport
->odp_port
!= port
.port_no
739 && !odp_port_to_ofport(backer
, port
.port_no
))
741 /* 'ofport''s datapath port number has changed from
742 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
743 * structures to match. */
744 ovs_rwlock_wrlock(&backer
->odp_to_ofport_lock
);
745 hmap_remove(&backer
->odp_to_ofport_map
, &ofport
->odp_port_node
);
746 ofport
->odp_port
= port
.port_no
;
747 hmap_insert(&backer
->odp_to_ofport_map
, &ofport
->odp_port_node
,
748 hash_odp_port(port
.port_no
));
749 ovs_rwlock_unlock(&backer
->odp_to_ofport_lock
);
750 backer
->need_revalidate
= REV_RECONFIGURE
;
753 dpif_port_destroy(&port
);
756 /* Propagate 'error' to all ofprotos based on 'backer'. */
758 process_dpif_port_error(struct dpif_backer
*backer
, int error
)
760 struct ofproto_dpif
*ofproto
;
762 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
763 if (ofproto
->backer
== backer
) {
764 sset_clear(&ofproto
->port_poll_set
);
765 ofproto
->port_poll_errno
= error
;
771 type_wait(const char *type
)
773 struct dpif_backer
*backer
;
775 backer
= shash_find_data(&all_dpif_backers
, type
);
777 /* This is not necessarily a problem, since backers are only
778 * created on demand. */
782 dpif_wait(backer
->dpif
);
785 /* Basic life-cycle. */
787 static int add_internal_flows(struct ofproto_dpif
*);
789 static struct ofproto
*
792 struct ofproto_dpif
*ofproto
= xmalloc(sizeof *ofproto
);
797 dealloc(struct ofproto
*ofproto_
)
799 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
804 close_dpif_backer(struct dpif_backer
*backer
)
806 ovs_assert(backer
->refcount
> 0);
808 if (--backer
->refcount
) {
812 udpif_destroy(backer
->udpif
);
814 simap_destroy(&backer
->tnl_backers
);
815 ovs_rwlock_destroy(&backer
->odp_to_ofport_lock
);
816 hmap_destroy(&backer
->odp_to_ofport_map
);
817 shash_find_and_delete(&all_dpif_backers
, backer
->type
);
818 recirc_id_pool_destroy(backer
->rid_pool
);
820 dpif_close(backer
->dpif
);
824 /* Datapath port slated for removal from datapath. */
826 struct list list_node
;
830 static bool check_variable_length_userdata(struct dpif_backer
*backer
);
831 static size_t check_max_mpls_depth(struct dpif_backer
*backer
);
832 static bool check_recirc(struct dpif_backer
*backer
);
835 open_dpif_backer(const char *type
, struct dpif_backer
**backerp
)
837 struct dpif_backer
*backer
;
838 struct dpif_port_dump port_dump
;
839 struct dpif_port port
;
840 struct shash_node
*node
;
841 struct list garbage_list
;
842 struct odp_garbage
*garbage
, *next
;
849 backer
= shash_find_data(&all_dpif_backers
, type
);
856 backer_name
= xasprintf("ovs-%s", type
);
858 /* Remove any existing datapaths, since we assume we're the only
859 * userspace controlling the datapath. */
861 dp_enumerate_names(type
, &names
);
862 SSET_FOR_EACH(name
, &names
) {
863 struct dpif
*old_dpif
;
865 /* Don't remove our backer if it exists. */
866 if (!strcmp(name
, backer_name
)) {
870 if (dpif_open(name
, type
, &old_dpif
)) {
871 VLOG_WARN("couldn't open old datapath %s to remove it", name
);
873 dpif_delete(old_dpif
);
874 dpif_close(old_dpif
);
877 sset_destroy(&names
);
879 backer
= xmalloc(sizeof *backer
);
881 error
= dpif_create_and_open(backer_name
, type
, &backer
->dpif
);
884 VLOG_ERR("failed to open datapath of type %s: %s", type
,
885 ovs_strerror(error
));
889 backer
->udpif
= udpif_create(backer
, backer
->dpif
);
891 backer
->type
= xstrdup(type
);
892 backer
->refcount
= 1;
893 hmap_init(&backer
->odp_to_ofport_map
);
894 ovs_rwlock_init(&backer
->odp_to_ofport_lock
);
895 backer
->need_revalidate
= 0;
896 simap_init(&backer
->tnl_backers
);
897 backer
->recv_set_enable
= !ofproto_get_flow_restore_wait();
900 if (backer
->recv_set_enable
) {
901 dpif_flow_flush(backer
->dpif
);
904 /* Loop through the ports already on the datapath and remove any
905 * that we don't need anymore. */
906 list_init(&garbage_list
);
907 dpif_port_dump_start(&port_dump
, backer
->dpif
);
908 while (dpif_port_dump_next(&port_dump
, &port
)) {
909 node
= shash_find(&init_ofp_ports
, port
.name
);
910 if (!node
&& strcmp(port
.name
, dpif_base_name(backer
->dpif
))) {
911 garbage
= xmalloc(sizeof *garbage
);
912 garbage
->odp_port
= port
.port_no
;
913 list_push_front(&garbage_list
, &garbage
->list_node
);
916 dpif_port_dump_done(&port_dump
);
918 LIST_FOR_EACH_SAFE (garbage
, next
, list_node
, &garbage_list
) {
919 dpif_port_del(backer
->dpif
, garbage
->odp_port
);
920 list_remove(&garbage
->list_node
);
924 shash_add(&all_dpif_backers
, type
, backer
);
926 error
= dpif_recv_set(backer
->dpif
, backer
->recv_set_enable
);
928 VLOG_ERR("failed to listen on datapath of type %s: %s",
929 type
, ovs_strerror(error
));
930 close_dpif_backer(backer
);
933 backer
->enable_recirc
= check_recirc(backer
);
934 backer
->variable_length_userdata
= check_variable_length_userdata(backer
);
935 backer
->max_mpls_depth
= check_max_mpls_depth(backer
);
936 backer
->rid_pool
= recirc_id_pool_create();
938 if (backer
->recv_set_enable
) {
939 udpif_set_threads(backer
->udpif
, n_handlers
, n_revalidators
);
945 /* Tests whether 'backer''s datapath supports recirculation Only newer datapath
946 * supports OVS_KEY_ATTR in OVS_ACTION_ATTR_USERSPACE actions. We need to
947 * disable some features on older datapaths that don't support this feature.
949 * Returns false if 'backer' definitely does not support recirculation, true if
950 * it seems to support recirculation or if at least the error we get is
953 check_recirc(struct dpif_backer
*backer
)
956 struct odputil_keybuf keybuf
;
959 bool enable_recirc
= false;
961 memset(&flow
, 0, sizeof flow
);
965 ofpbuf_use_stack(&key
, &keybuf
, sizeof keybuf
);
966 odp_flow_key_from_flow(&key
, &flow
, NULL
, 0);
968 error
= dpif_flow_put(backer
->dpif
, DPIF_FP_CREATE
| DPIF_FP_MODIFY
,
969 ofpbuf_data(&key
), ofpbuf_size(&key
), NULL
, 0, NULL
,
971 if (error
&& error
!= EEXIST
) {
972 if (error
!= EINVAL
) {
973 VLOG_WARN("%s: Reciculation flow probe failed (%s)",
974 dpif_name(backer
->dpif
), ovs_strerror(error
));
979 error
= dpif_flow_del(backer
->dpif
, ofpbuf_data(&key
), ofpbuf_size(&key
),
982 VLOG_WARN("%s: failed to delete recirculation feature probe flow",
983 dpif_name(backer
->dpif
));
986 enable_recirc
= true;
990 VLOG_INFO("%s: Datapath supports recirculation",
991 dpif_name(backer
->dpif
));
993 VLOG_INFO("%s: Datapath does not support recirculation",
994 dpif_name(backer
->dpif
));
997 return enable_recirc
;
1000 /* Tests whether 'backer''s datapath supports variable-length
1001 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions. We need
1002 * to disable some features on older datapaths that don't support this
1005 * Returns false if 'backer' definitely does not support variable-length
1006 * userdata, true if it seems to support them or if at least the error we get
1009 check_variable_length_userdata(struct dpif_backer
*backer
)
1011 struct eth_header
*eth
;
1012 struct ofpbuf actions
;
1013 struct dpif_execute execute
;
1014 struct ofpbuf packet
;
1018 /* Compose a userspace action that will cause an ERANGE error on older
1019 * datapaths that don't support variable-length userdata.
1021 * We really test for using userdata longer than 8 bytes, but older
1022 * datapaths accepted these, silently truncating the userdata to 8 bytes.
1023 * The same older datapaths rejected userdata shorter than 8 bytes, so we
1024 * test for that instead as a proxy for longer userdata support. */
1025 ofpbuf_init(&actions
, 64);
1026 start
= nl_msg_start_nested(&actions
, OVS_ACTION_ATTR_USERSPACE
);
1027 nl_msg_put_u32(&actions
, OVS_USERSPACE_ATTR_PID
,
1028 dpif_port_get_pid(backer
->dpif
, ODPP_NONE
, 0));
1029 nl_msg_put_unspec_zero(&actions
, OVS_USERSPACE_ATTR_USERDATA
, 4);
1030 nl_msg_end_nested(&actions
, start
);
1032 /* Compose a dummy ethernet packet. */
1033 ofpbuf_init(&packet
, ETH_HEADER_LEN
);
1034 eth
= ofpbuf_put_zeros(&packet
, ETH_HEADER_LEN
);
1035 eth
->eth_type
= htons(0x1234);
1037 /* Execute the actions. On older datapaths this fails with ERANGE, on
1038 * newer datapaths it succeeds. */
1039 execute
.actions
= ofpbuf_data(&actions
);
1040 execute
.actions_len
= ofpbuf_size(&actions
);
1041 execute
.packet
= &packet
;
1042 execute
.md
= PKT_METADATA_INITIALIZER(0);
1043 execute
.needs_help
= false;
1045 error
= dpif_execute(backer
->dpif
, &execute
);
1047 ofpbuf_uninit(&packet
);
1048 ofpbuf_uninit(&actions
);
1052 /* Variable-length userdata is supported.
1054 * Purge received packets to avoid processing the nonsense packet we
1055 * sent to userspace, then report success. */
1056 dpif_recv_purge(backer
->dpif
);
1060 /* Variable-length userdata is not supported. */
1061 VLOG_WARN("%s: datapath does not support variable-length userdata "
1062 "feature (needs Linux 3.10+ or kernel module from OVS "
1063 "1..11+). The NXAST_SAMPLE action will be ignored.",
1064 dpif_name(backer
->dpif
));
1068 /* Something odd happened. We're not sure whether variable-length
1069 * userdata is supported. Default to "yes". */
1070 VLOG_WARN("%s: variable-length userdata feature probe failed (%s)",
1071 dpif_name(backer
->dpif
), ovs_strerror(error
));
1076 /* Tests the MPLS label stack depth supported by 'backer''s datapath.
1078 * Returns the number of elements in a struct flow's mpls_lse field
1079 * if the datapath supports at least that many entries in an
1081 * Otherwise returns the number of MPLS push actions supported by
1084 check_max_mpls_depth(struct dpif_backer
*backer
)
1089 for (n
= 0; n
< FLOW_MAX_MPLS_LABELS
; n
++) {
1090 struct odputil_keybuf keybuf
;
1094 memset(&flow
, 0, sizeof flow
);
1095 flow
.dl_type
= htons(ETH_TYPE_MPLS
);
1096 flow_set_mpls_bos(&flow
, n
, 1);
1098 ofpbuf_use_stack(&key
, &keybuf
, sizeof keybuf
);
1099 odp_flow_key_from_flow(&key
, &flow
, NULL
, 0);
1101 error
= dpif_flow_put(backer
->dpif
, DPIF_FP_CREATE
| DPIF_FP_MODIFY
,
1102 ofpbuf_data(&key
), ofpbuf_size(&key
), NULL
, 0, NULL
, 0, NULL
);
1103 if (error
&& error
!= EEXIST
) {
1104 if (error
!= EINVAL
) {
1105 VLOG_WARN("%s: MPLS stack length feature probe failed (%s)",
1106 dpif_name(backer
->dpif
), ovs_strerror(error
));
1111 error
= dpif_flow_del(backer
->dpif
, ofpbuf_data(&key
), ofpbuf_size(&key
), NULL
);
1113 VLOG_WARN("%s: failed to delete MPLS feature probe flow",
1114 dpif_name(backer
->dpif
));
1118 VLOG_INFO("%s: MPLS label stack length probed as %d",
1119 dpif_name(backer
->dpif
), n
);
1124 construct(struct ofproto
*ofproto_
)
1126 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1127 struct shash_node
*node
, *next
;
1130 error
= open_dpif_backer(ofproto
->up
.type
, &ofproto
->backer
);
1135 ofproto
->netflow
= NULL
;
1136 ofproto
->sflow
= NULL
;
1137 ofproto
->ipfix
= NULL
;
1138 ofproto
->stp
= NULL
;
1139 ofproto
->dump_seq
= 0;
1140 hmap_init(&ofproto
->bundles
);
1141 ofproto
->ml
= mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME
);
1142 ofproto
->mbridge
= mbridge_create();
1143 ofproto
->has_bonded_bundles
= false;
1144 ofproto
->lacp_enabled
= false;
1145 ovs_mutex_init_adaptive(&ofproto
->stats_mutex
);
1146 ovs_mutex_init(&ofproto
->vsp_mutex
);
1148 guarded_list_init(&ofproto
->pins
);
1150 ofproto_dpif_unixctl_init();
1152 hmap_init(&ofproto
->vlandev_map
);
1153 hmap_init(&ofproto
->realdev_vid_map
);
1155 sset_init(&ofproto
->ports
);
1156 sset_init(&ofproto
->ghost_ports
);
1157 sset_init(&ofproto
->port_poll_set
);
1158 ofproto
->port_poll_errno
= 0;
1159 ofproto
->change_seq
= 0;
1161 SHASH_FOR_EACH_SAFE (node
, next
, &init_ofp_ports
) {
1162 struct iface_hint
*iface_hint
= node
->data
;
1164 if (!strcmp(iface_hint
->br_name
, ofproto
->up
.name
)) {
1165 /* Check if the datapath already has this port. */
1166 if (dpif_port_exists(ofproto
->backer
->dpif
, node
->name
)) {
1167 sset_add(&ofproto
->ports
, node
->name
);
1170 free(iface_hint
->br_name
);
1171 free(iface_hint
->br_type
);
1173 shash_delete(&init_ofp_ports
, node
);
1177 hmap_insert(&all_ofproto_dpifs
, &ofproto
->all_ofproto_dpifs_node
,
1178 hash_string(ofproto
->up
.name
, 0));
1179 memset(&ofproto
->stats
, 0, sizeof ofproto
->stats
);
1181 ofproto_init_tables(ofproto_
, N_TABLES
);
1182 error
= add_internal_flows(ofproto
);
1184 ofproto
->up
.tables
[TBL_INTERNAL
].flags
= OFTABLE_HIDDEN
| OFTABLE_READONLY
;
1190 add_internal_miss_flow(struct ofproto_dpif
*ofproto
, int id
,
1191 const struct ofpbuf
*ofpacts
, struct rule_dpif
**rulep
)
1197 match_init_catchall(&match
);
1198 match_set_reg(&match
, 0, id
);
1200 error
= ofproto_dpif_add_internal_flow(ofproto
, &match
, 0, ofpacts
, &rule
);
1201 *rulep
= error
? NULL
: rule_dpif_cast(rule
);
1207 add_internal_flows(struct ofproto_dpif
*ofproto
)
1209 struct ofpact_controller
*controller
;
1210 uint64_t ofpacts_stub
[128 / 8];
1211 struct ofpbuf ofpacts
;
1212 struct rule
*unused_rulep OVS_UNUSED
;
1213 struct ofpact_resubmit
*resubmit
;
1218 ofpbuf_use_stack(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
1221 controller
= ofpact_put_CONTROLLER(&ofpacts
);
1222 controller
->max_len
= UINT16_MAX
;
1223 controller
->controller_id
= 0;
1224 controller
->reason
= OFPR_NO_MATCH
;
1225 ofpact_pad(&ofpacts
);
1227 error
= add_internal_miss_flow(ofproto
, id
++, &ofpacts
,
1228 &ofproto
->miss_rule
);
1233 ofpbuf_clear(&ofpacts
);
1234 error
= add_internal_miss_flow(ofproto
, id
++, &ofpacts
,
1235 &ofproto
->no_packet_in_rule
);
1240 error
= add_internal_miss_flow(ofproto
, id
++, &ofpacts
,
1241 &ofproto
->drop_frags_rule
);
1246 /* Continue non-recirculation rule lookups from table 0.
1248 * (priority=2), recirc=0, actions=resubmit(, 0)
1250 resubmit
= ofpact_put_RESUBMIT(&ofpacts
);
1251 resubmit
->ofpact
.compat
= 0;
1252 resubmit
->in_port
= OFPP_IN_PORT
;
1253 resubmit
->table_id
= 0;
1255 match_init_catchall(&match
);
1256 match_set_recirc_id(&match
, 0);
1258 error
= ofproto_dpif_add_internal_flow(ofproto
, &match
, 2, &ofpacts
,
1264 /* Drop any run away recirc rule lookups. Recirc_id has to be
1265 * non-zero when reaching this rule.
1267 * (priority=1), *, actions=drop
1269 ofpbuf_clear(&ofpacts
);
1270 match_init_catchall(&match
);
1271 error
= ofproto_dpif_add_internal_flow(ofproto
, &match
, 1, &ofpacts
,
1278 destruct(struct ofproto
*ofproto_
)
1280 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1281 struct rule_dpif
*rule
, *next_rule
;
1282 struct ofproto_packet_in
*pin
, *next_pin
;
1283 struct oftable
*table
;
1286 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1287 ovs_rwlock_wrlock(&xlate_rwlock
);
1288 xlate_remove_ofproto(ofproto
);
1289 ovs_rwlock_unlock(&xlate_rwlock
);
1291 /* Ensure that the upcall processing threads have no remaining references
1292 * to the ofproto or anything in it. */
1293 udpif_synchronize(ofproto
->backer
->udpif
);
1295 hmap_remove(&all_ofproto_dpifs
, &ofproto
->all_ofproto_dpifs_node
);
1297 OFPROTO_FOR_EACH_TABLE (table
, &ofproto
->up
) {
1298 struct cls_cursor cursor
;
1300 fat_rwlock_rdlock(&table
->cls
.rwlock
);
1301 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
1302 fat_rwlock_unlock(&table
->cls
.rwlock
);
1303 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, up
.cr
, &cursor
) {
1304 ofproto_rule_delete(&ofproto
->up
, &rule
->up
);
1308 guarded_list_pop_all(&ofproto
->pins
, &pins
);
1309 LIST_FOR_EACH_SAFE (pin
, next_pin
, list_node
, &pins
) {
1310 list_remove(&pin
->list_node
);
1311 free(CONST_CAST(void *, pin
->up
.packet
));
1314 guarded_list_destroy(&ofproto
->pins
);
1316 mbridge_unref(ofproto
->mbridge
);
1318 netflow_unref(ofproto
->netflow
);
1319 dpif_sflow_unref(ofproto
->sflow
);
1320 hmap_destroy(&ofproto
->bundles
);
1321 mac_learning_unref(ofproto
->ml
);
1323 hmap_destroy(&ofproto
->vlandev_map
);
1324 hmap_destroy(&ofproto
->realdev_vid_map
);
1326 sset_destroy(&ofproto
->ports
);
1327 sset_destroy(&ofproto
->ghost_ports
);
1328 sset_destroy(&ofproto
->port_poll_set
);
1330 ovs_mutex_destroy(&ofproto
->stats_mutex
);
1331 ovs_mutex_destroy(&ofproto
->vsp_mutex
);
1333 close_dpif_backer(ofproto
->backer
);
1337 run(struct ofproto
*ofproto_
)
1339 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1340 uint64_t new_seq
, new_dump_seq
;
1341 const bool enable_recirc
= ofproto_dpif_get_enable_recirc(ofproto
);
1343 if (mbridge_need_revalidate(ofproto
->mbridge
)) {
1344 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1345 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
1346 mac_learning_flush(ofproto
->ml
);
1347 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
1350 /* Do not perform any periodic activity required by 'ofproto' while
1351 * waiting for flow restore to complete. */
1352 if (!ofproto_get_flow_restore_wait()) {
1353 struct ofproto_packet_in
*pin
, *next_pin
;
1356 guarded_list_pop_all(&ofproto
->pins
, &pins
);
1357 LIST_FOR_EACH_SAFE (pin
, next_pin
, list_node
, &pins
) {
1358 connmgr_send_packet_in(ofproto
->up
.connmgr
, pin
);
1359 list_remove(&pin
->list_node
);
1360 free(CONST_CAST(void *, pin
->up
.packet
));
1365 if (ofproto
->netflow
) {
1366 netflow_run(ofproto
->netflow
);
1368 if (ofproto
->sflow
) {
1369 dpif_sflow_run(ofproto
->sflow
);
1371 if (ofproto
->ipfix
) {
1372 dpif_ipfix_run(ofproto
->ipfix
);
1375 new_seq
= seq_read(connectivity_seq_get());
1376 if (ofproto
->change_seq
!= new_seq
) {
1377 struct ofport_dpif
*ofport
;
1379 HMAP_FOR_EACH (ofport
, up
.hmap_node
, &ofproto
->up
.ports
) {
1383 ofproto
->change_seq
= new_seq
;
1385 if (ofproto
->lacp_enabled
|| ofproto
->has_bonded_bundles
) {
1386 struct ofbundle
*bundle
;
1388 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
1394 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
1395 if (mac_learning_run(ofproto
->ml
)) {
1396 ofproto
->backer
->need_revalidate
= REV_MAC_LEARNING
;
1398 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
1400 new_dump_seq
= seq_read(udpif_dump_seq(ofproto
->backer
->udpif
));
1401 if (ofproto
->dump_seq
!= new_dump_seq
) {
1402 struct rule
*rule
, *next_rule
;
1404 /* We know stats are relatively fresh, so now is a good time to do some
1406 ofproto
->dump_seq
= new_dump_seq
;
1408 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1410 ovs_mutex_lock(&ofproto_mutex
);
1411 LIST_FOR_EACH_SAFE (rule
, next_rule
, expirable
,
1412 &ofproto
->up
.expirable
) {
1413 rule_expire(rule_dpif_cast(rule
));
1415 ovs_mutex_unlock(&ofproto_mutex
);
1417 /* All outstanding data in existing flows has been accounted, so it's a
1418 * good time to do bond rebalancing. */
1419 if (enable_recirc
&& ofproto
->has_bonded_bundles
) {
1420 struct ofbundle
*bundle
;
1422 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
1423 struct bond
*bond
= bundle
->bond
;
1425 if (bond
&& bond_may_recirc(bond
, NULL
, NULL
)) {
1426 bond_recirculation_account(bond
);
1427 if (bond_rebalance(bundle
->bond
)) {
1428 bond_update_post_recirc_rules(bond
, true);
1439 wait(struct ofproto
*ofproto_
)
1441 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1443 if (ofproto_get_flow_restore_wait()) {
1447 if (ofproto
->sflow
) {
1448 dpif_sflow_wait(ofproto
->sflow
);
1450 if (ofproto
->ipfix
) {
1451 dpif_ipfix_wait(ofproto
->ipfix
);
1453 if (ofproto
->lacp_enabled
|| ofproto
->has_bonded_bundles
) {
1454 struct ofbundle
*bundle
;
1456 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
1457 bundle_wait(bundle
);
1460 if (ofproto
->netflow
) {
1461 netflow_wait(ofproto
->netflow
);
1463 ovs_rwlock_rdlock(&ofproto
->ml
->rwlock
);
1464 mac_learning_wait(ofproto
->ml
);
1465 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
1467 if (ofproto
->backer
->need_revalidate
) {
1468 /* Shouldn't happen, but if it does just go around again. */
1469 VLOG_DBG_RL(&rl
, "need revalidate in ofproto_wait_cb()");
1470 poll_immediate_wake();
1473 seq_wait(udpif_dump_seq(ofproto
->backer
->udpif
), ofproto
->dump_seq
);
1477 type_get_memory_usage(const char *type
, struct simap
*usage
)
1479 struct dpif_backer
*backer
;
1481 backer
= shash_find_data(&all_dpif_backers
, type
);
1483 udpif_get_memory_usage(backer
->udpif
, usage
);
1488 flush(struct ofproto
*ofproto_
)
1490 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1491 struct dpif_backer
*backer
= ofproto
->backer
;
1494 udpif_flush(backer
->udpif
);
1499 get_features(struct ofproto
*ofproto_ OVS_UNUSED
,
1500 bool *arp_match_ip
, enum ofputil_action_bitmap
*actions
)
1502 *arp_match_ip
= true;
1503 *actions
= (OFPUTIL_A_OUTPUT
|
1504 OFPUTIL_A_SET_VLAN_VID
|
1505 OFPUTIL_A_SET_VLAN_PCP
|
1506 OFPUTIL_A_STRIP_VLAN
|
1507 OFPUTIL_A_SET_DL_SRC
|
1508 OFPUTIL_A_SET_DL_DST
|
1509 OFPUTIL_A_SET_NW_SRC
|
1510 OFPUTIL_A_SET_NW_DST
|
1511 OFPUTIL_A_SET_NW_TOS
|
1512 OFPUTIL_A_SET_TP_SRC
|
1513 OFPUTIL_A_SET_TP_DST
|
1518 get_tables(struct ofproto
*ofproto_
, struct ofp12_table_stats
*ots
)
1520 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1521 struct dpif_dp_stats s
;
1522 uint64_t n_miss
, n_no_pkt_in
, n_bytes
, n_dropped_frags
;
1526 strcpy(ots
->name
, "classifier");
1528 dpif_get_dp_stats(ofproto
->backer
->dpif
, &s
);
1529 rule_get_stats(&ofproto
->miss_rule
->up
, &n_miss
, &n_bytes
, &used
);
1530 rule_get_stats(&ofproto
->no_packet_in_rule
->up
, &n_no_pkt_in
, &n_bytes
,
1532 rule_get_stats(&ofproto
->drop_frags_rule
->up
, &n_dropped_frags
, &n_bytes
,
1534 n_lookup
= s
.n_hit
+ s
.n_missed
- n_dropped_frags
;
1535 ots
->lookup_count
= htonll(n_lookup
);
1536 ots
->matched_count
= htonll(n_lookup
- n_miss
- n_no_pkt_in
);
1539 static struct ofport
*
1542 struct ofport_dpif
*port
= xmalloc(sizeof *port
);
1547 port_dealloc(struct ofport
*port_
)
1549 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
1554 port_construct(struct ofport
*port_
)
1556 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
1557 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
1558 const struct netdev
*netdev
= port
->up
.netdev
;
1559 char namebuf
[NETDEV_VPORT_NAME_BUFSIZE
];
1560 struct dpif_port dpif_port
;
1563 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1564 port
->bundle
= NULL
;
1567 port
->may_enable
= true;
1568 port
->stp_port
= NULL
;
1569 port
->stp_state
= STP_DISABLED
;
1570 port
->is_tunnel
= false;
1574 port
->realdev_ofp_port
= 0;
1575 port
->vlandev_vid
= 0;
1576 port
->carrier_seq
= netdev_get_carrier_resets(netdev
);
1577 port
->is_layer3
= netdev_vport_is_layer3(netdev
);
1579 if (netdev_vport_is_patch(netdev
)) {
1580 /* By bailing out here, we don't submit the port to the sFlow module
1581 * to be considered for counter polling export. This is correct
1582 * because the patch port represents an interface that sFlow considers
1583 * to be "internal" to the switch as a whole, and therefore not an
1584 * candidate for counter polling. */
1585 port
->odp_port
= ODPP_NONE
;
1586 ofport_update_peer(port
);
1590 error
= dpif_port_query_by_name(ofproto
->backer
->dpif
,
1591 netdev_vport_get_dpif_port(netdev
, namebuf
,
1598 port
->odp_port
= dpif_port
.port_no
;
1600 if (netdev_get_tunnel_config(netdev
)) {
1601 tnl_port_add(port
, port
->up
.netdev
, port
->odp_port
);
1602 port
->is_tunnel
= true;
1604 /* Sanity-check that a mapping doesn't already exist. This
1605 * shouldn't happen for non-tunnel ports. */
1606 if (odp_port_to_ofp_port(ofproto
, port
->odp_port
) != OFPP_NONE
) {
1607 VLOG_ERR("port %s already has an OpenFlow port number",
1609 dpif_port_destroy(&dpif_port
);
1613 ovs_rwlock_wrlock(&ofproto
->backer
->odp_to_ofport_lock
);
1614 hmap_insert(&ofproto
->backer
->odp_to_ofport_map
, &port
->odp_port_node
,
1615 hash_odp_port(port
->odp_port
));
1616 ovs_rwlock_unlock(&ofproto
->backer
->odp_to_ofport_lock
);
1618 dpif_port_destroy(&dpif_port
);
1620 if (ofproto
->sflow
) {
1621 dpif_sflow_add_port(ofproto
->sflow
, port_
, port
->odp_port
);
1628 port_destruct(struct ofport
*port_
)
1630 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
1631 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
1632 const char *devname
= netdev_get_name(port
->up
.netdev
);
1633 char namebuf
[NETDEV_VPORT_NAME_BUFSIZE
];
1634 const char *dp_port_name
;
1636 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1637 ovs_rwlock_wrlock(&xlate_rwlock
);
1638 xlate_ofport_remove(port
);
1639 ovs_rwlock_unlock(&xlate_rwlock
);
1641 dp_port_name
= netdev_vport_get_dpif_port(port
->up
.netdev
, namebuf
,
1643 if (dpif_port_exists(ofproto
->backer
->dpif
, dp_port_name
)) {
1644 /* The underlying device is still there, so delete it. This
1645 * happens when the ofproto is being destroyed, since the caller
1646 * assumes that removal of attached ports will happen as part of
1648 if (!port
->is_tunnel
) {
1649 dpif_port_del(ofproto
->backer
->dpif
, port
->odp_port
);
1654 port
->peer
->peer
= NULL
;
1658 if (port
->odp_port
!= ODPP_NONE
&& !port
->is_tunnel
) {
1659 ovs_rwlock_wrlock(&ofproto
->backer
->odp_to_ofport_lock
);
1660 hmap_remove(&ofproto
->backer
->odp_to_ofport_map
, &port
->odp_port_node
);
1661 ovs_rwlock_unlock(&ofproto
->backer
->odp_to_ofport_lock
);
1665 sset_find_and_delete(&ofproto
->ports
, devname
);
1666 sset_find_and_delete(&ofproto
->ghost_ports
, devname
);
1667 bundle_remove(port_
);
1668 set_cfm(port_
, NULL
);
1669 set_bfd(port_
, NULL
);
1670 if (port
->stp_port
) {
1671 stp_port_disable(port
->stp_port
);
1673 if (ofproto
->sflow
) {
1674 dpif_sflow_del_port(ofproto
->sflow
, port
->odp_port
);
1681 port_modified(struct ofport
*port_
)
1683 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
1685 if (port
->bundle
&& port
->bundle
->bond
) {
1686 bond_slave_set_netdev(port
->bundle
->bond
, port
, port
->up
.netdev
);
1690 cfm_set_netdev(port
->cfm
, port
->up
.netdev
);
1694 bfd_set_netdev(port
->bfd
, port
->up
.netdev
);
1697 ofproto_dpif_monitor_port_update(port
, port
->bfd
, port
->cfm
,
1698 port
->up
.pp
.hw_addr
);
1700 if (port
->is_tunnel
&& tnl_port_reconfigure(port
, port
->up
.netdev
,
1702 ofproto_dpif_cast(port
->up
.ofproto
)->backer
->need_revalidate
=
1706 ofport_update_peer(port
);
1710 port_reconfigured(struct ofport
*port_
, enum ofputil_port_config old_config
)
1712 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
1713 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
1714 enum ofputil_port_config changed
= old_config
^ port
->up
.pp
.config
;
1716 if (changed
& (OFPUTIL_PC_NO_RECV
| OFPUTIL_PC_NO_RECV_STP
|
1717 OFPUTIL_PC_NO_FWD
| OFPUTIL_PC_NO_FLOOD
|
1718 OFPUTIL_PC_NO_PACKET_IN
)) {
1719 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1721 if (changed
& OFPUTIL_PC_NO_FLOOD
&& port
->bundle
) {
1722 bundle_update(port
->bundle
);
1728 set_sflow(struct ofproto
*ofproto_
,
1729 const struct ofproto_sflow_options
*sflow_options
)
1731 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1732 struct dpif_sflow
*ds
= ofproto
->sflow
;
1734 if (sflow_options
) {
1736 struct ofport_dpif
*ofport
;
1738 ds
= ofproto
->sflow
= dpif_sflow_create();
1739 HMAP_FOR_EACH (ofport
, up
.hmap_node
, &ofproto
->up
.ports
) {
1740 dpif_sflow_add_port(ds
, &ofport
->up
, ofport
->odp_port
);
1742 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1744 dpif_sflow_set_options(ds
, sflow_options
);
1747 dpif_sflow_unref(ds
);
1748 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1749 ofproto
->sflow
= NULL
;
1757 struct ofproto
*ofproto_
,
1758 const struct ofproto_ipfix_bridge_exporter_options
*bridge_exporter_options
,
1759 const struct ofproto_ipfix_flow_exporter_options
*flow_exporters_options
,
1760 size_t n_flow_exporters_options
)
1762 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1763 struct dpif_ipfix
*di
= ofproto
->ipfix
;
1764 bool has_options
= bridge_exporter_options
|| flow_exporters_options
;
1766 if (has_options
&& !di
) {
1767 di
= ofproto
->ipfix
= dpif_ipfix_create();
1771 /* Call set_options in any case to cleanly flush the flow
1772 * caches in the last exporters that are to be destroyed. */
1773 dpif_ipfix_set_options(
1774 di
, bridge_exporter_options
, flow_exporters_options
,
1775 n_flow_exporters_options
);
1778 dpif_ipfix_unref(di
);
1779 ofproto
->ipfix
= NULL
;
1787 set_cfm(struct ofport
*ofport_
, const struct cfm_settings
*s
)
1789 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
1794 struct ofproto_dpif
*ofproto
;
1796 ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
1797 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1798 ofport
->cfm
= cfm_create(ofport
->up
.netdev
);
1801 if (cfm_configure(ofport
->cfm
, s
)) {
1808 cfm_unref(ofport
->cfm
);
1811 ofproto_dpif_monitor_port_update(ofport
, ofport
->bfd
, ofport
->cfm
,
1812 ofport
->up
.pp
.hw_addr
);
1817 get_cfm_status(const struct ofport
*ofport_
,
1818 struct ofproto_cfm_status
*status
)
1820 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
1823 status
->faults
= cfm_get_fault(ofport
->cfm
);
1824 status
->flap_count
= cfm_get_flap_count(ofport
->cfm
);
1825 status
->remote_opstate
= cfm_get_opup(ofport
->cfm
);
1826 status
->health
= cfm_get_health(ofport
->cfm
);
1827 cfm_get_remote_mpids(ofport
->cfm
, &status
->rmps
, &status
->n_rmps
);
1835 set_bfd(struct ofport
*ofport_
, const struct smap
*cfg
)
1837 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport_
->ofproto
);
1838 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
1842 ofport
->bfd
= bfd_configure(old
, netdev_get_name(ofport
->up
.netdev
),
1843 cfg
, ofport
->up
.netdev
);
1844 if (ofport
->bfd
!= old
) {
1845 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1847 ofproto_dpif_monitor_port_update(ofport
, ofport
->bfd
, ofport
->cfm
,
1848 ofport
->up
.pp
.hw_addr
);
1853 get_bfd_status(struct ofport
*ofport_
, struct smap
*smap
)
1855 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
1858 bfd_get_status(ofport
->bfd
, smap
);
1865 /* Spanning Tree. */
1868 send_bpdu_cb(struct ofpbuf
*pkt
, int port_num
, void *ofproto_
)
1870 struct ofproto_dpif
*ofproto
= ofproto_
;
1871 struct stp_port
*sp
= stp_get_port(ofproto
->stp
, port_num
);
1872 struct ofport_dpif
*ofport
;
1874 ofport
= stp_port_get_aux(sp
);
1876 VLOG_WARN_RL(&rl
, "%s: cannot send BPDU on unknown port %d",
1877 ofproto
->up
.name
, port_num
);
1879 struct eth_header
*eth
= ofpbuf_l2(pkt
);
1881 netdev_get_etheraddr(ofport
->up
.netdev
, eth
->eth_src
);
1882 if (eth_addr_is_zero(eth
->eth_src
)) {
1883 VLOG_WARN_RL(&rl
, "%s: cannot send BPDU on port %d "
1884 "with unknown MAC", ofproto
->up
.name
, port_num
);
1886 ofproto_dpif_send_packet(ofport
, pkt
);
1892 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1894 set_stp(struct ofproto
*ofproto_
, const struct ofproto_stp_settings
*s
)
1896 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1898 /* Only revalidate flows if the configuration changed. */
1899 if (!s
!= !ofproto
->stp
) {
1900 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1904 if (!ofproto
->stp
) {
1905 ofproto
->stp
= stp_create(ofproto_
->name
, s
->system_id
,
1906 send_bpdu_cb
, ofproto
);
1907 ofproto
->stp_last_tick
= time_msec();
1910 stp_set_bridge_id(ofproto
->stp
, s
->system_id
);
1911 stp_set_bridge_priority(ofproto
->stp
, s
->priority
);
1912 stp_set_hello_time(ofproto
->stp
, s
->hello_time
);
1913 stp_set_max_age(ofproto
->stp
, s
->max_age
);
1914 stp_set_forward_delay(ofproto
->stp
, s
->fwd_delay
);
1916 struct ofport
*ofport
;
1918 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->up
.ports
) {
1919 set_stp_port(ofport
, NULL
);
1922 stp_unref(ofproto
->stp
);
1923 ofproto
->stp
= NULL
;
1930 get_stp_status(struct ofproto
*ofproto_
, struct ofproto_stp_status
*s
)
1932 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1936 s
->bridge_id
= stp_get_bridge_id(ofproto
->stp
);
1937 s
->designated_root
= stp_get_designated_root(ofproto
->stp
);
1938 s
->root_path_cost
= stp_get_root_path_cost(ofproto
->stp
);
1947 update_stp_port_state(struct ofport_dpif
*ofport
)
1949 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
1950 enum stp_state state
;
1952 /* Figure out new state. */
1953 state
= ofport
->stp_port
? stp_port_get_state(ofport
->stp_port
)
1957 if (ofport
->stp_state
!= state
) {
1958 enum ofputil_port_state of_state
;
1961 VLOG_DBG_RL(&rl
, "port %s: STP state changed from %s to %s",
1962 netdev_get_name(ofport
->up
.netdev
),
1963 stp_state_name(ofport
->stp_state
),
1964 stp_state_name(state
));
1965 if (stp_learn_in_state(ofport
->stp_state
)
1966 != stp_learn_in_state(state
)) {
1967 /* xxx Learning action flows should also be flushed. */
1968 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
1969 mac_learning_flush(ofproto
->ml
);
1970 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
1972 fwd_change
= stp_forward_in_state(ofport
->stp_state
)
1973 != stp_forward_in_state(state
);
1975 ofproto
->backer
->need_revalidate
= REV_STP
;
1976 ofport
->stp_state
= state
;
1977 ofport
->stp_state_entered
= time_msec();
1979 if (fwd_change
&& ofport
->bundle
) {
1980 bundle_update(ofport
->bundle
);
1983 /* Update the STP state bits in the OpenFlow port description. */
1984 of_state
= ofport
->up
.pp
.state
& ~OFPUTIL_PS_STP_MASK
;
1985 of_state
|= (state
== STP_LISTENING
? OFPUTIL_PS_STP_LISTEN
1986 : state
== STP_LEARNING
? OFPUTIL_PS_STP_LEARN
1987 : state
== STP_FORWARDING
? OFPUTIL_PS_STP_FORWARD
1988 : state
== STP_BLOCKING
? OFPUTIL_PS_STP_BLOCK
1990 ofproto_port_set_state(&ofport
->up
, of_state
);
1994 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1995 * caller is responsible for assigning STP port numbers and ensuring
1996 * there are no duplicates. */
1998 set_stp_port(struct ofport
*ofport_
,
1999 const struct ofproto_port_stp_settings
*s
)
2001 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2002 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2003 struct stp_port
*sp
= ofport
->stp_port
;
2005 if (!s
|| !s
->enable
) {
2007 ofport
->stp_port
= NULL
;
2008 stp_port_disable(sp
);
2009 update_stp_port_state(ofport
);
2012 } else if (sp
&& stp_port_no(sp
) != s
->port_num
2013 && ofport
== stp_port_get_aux(sp
)) {
2014 /* The port-id changed, so disable the old one if it's not
2015 * already in use by another port. */
2016 stp_port_disable(sp
);
2019 sp
= ofport
->stp_port
= stp_get_port(ofproto
->stp
, s
->port_num
);
2020 stp_port_enable(sp
);
2022 stp_port_set_aux(sp
, ofport
);
2023 stp_port_set_priority(sp
, s
->priority
);
2024 stp_port_set_path_cost(sp
, s
->path_cost
);
2026 update_stp_port_state(ofport
);
2032 get_stp_port_status(struct ofport
*ofport_
,
2033 struct ofproto_port_stp_status
*s
)
2035 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2036 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2037 struct stp_port
*sp
= ofport
->stp_port
;
2039 if (!ofproto
->stp
|| !sp
) {
2045 s
->port_id
= stp_port_get_id(sp
);
2046 s
->state
= stp_port_get_state(sp
);
2047 s
->sec_in_state
= (time_msec() - ofport
->stp_state_entered
) / 1000;
2048 s
->role
= stp_port_get_role(sp
);
2054 get_stp_port_stats(struct ofport
*ofport_
,
2055 struct ofproto_port_stp_stats
*s
)
2057 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2058 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2059 struct stp_port
*sp
= ofport
->stp_port
;
2061 if (!ofproto
->stp
|| !sp
) {
2067 stp_port_get_counts(sp
, &s
->tx_count
, &s
->rx_count
, &s
->error_count
);
2073 stp_run(struct ofproto_dpif
*ofproto
)
2076 long long int now
= time_msec();
2077 long long int elapsed
= now
- ofproto
->stp_last_tick
;
2078 struct stp_port
*sp
;
2081 stp_tick(ofproto
->stp
, MIN(INT_MAX
, elapsed
));
2082 ofproto
->stp_last_tick
= now
;
2084 while (stp_get_changed_port(ofproto
->stp
, &sp
)) {
2085 struct ofport_dpif
*ofport
= stp_port_get_aux(sp
);
2088 update_stp_port_state(ofport
);
2092 if (stp_check_and_reset_fdb_flush(ofproto
->stp
)) {
2093 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
2094 mac_learning_flush(ofproto
->ml
);
2095 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
2101 stp_wait(struct ofproto_dpif
*ofproto
)
2104 poll_timer_wait(1000);
2109 set_queues(struct ofport
*ofport_
, const struct ofproto_port_queue
*qdscp
,
2112 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2113 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2115 if (ofport
->n_qdscp
!= n_qdscp
2116 || (n_qdscp
&& memcmp(ofport
->qdscp
, qdscp
,
2117 n_qdscp
* sizeof *qdscp
))) {
2118 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2119 free(ofport
->qdscp
);
2120 ofport
->qdscp
= n_qdscp
2121 ? xmemdup(qdscp
, n_qdscp
* sizeof *qdscp
)
2123 ofport
->n_qdscp
= n_qdscp
;
2131 /* Expires all MAC learning entries associated with 'bundle' and forces its
2132 * ofproto to revalidate every flow.
2134 * Normally MAC learning entries are removed only from the ofproto associated
2135 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2136 * are removed from every ofproto. When patch ports and SLB bonds are in use
2137 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2138 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2139 * with the host from which it migrated. */
2141 bundle_flush_macs(struct ofbundle
*bundle
, bool all_ofprotos
)
2143 struct ofproto_dpif
*ofproto
= bundle
->ofproto
;
2144 struct mac_learning
*ml
= ofproto
->ml
;
2145 struct mac_entry
*mac
, *next_mac
;
2147 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2148 ovs_rwlock_wrlock(&ml
->rwlock
);
2149 LIST_FOR_EACH_SAFE (mac
, next_mac
, lru_node
, &ml
->lrus
) {
2150 if (mac
->port
.p
== bundle
) {
2152 struct ofproto_dpif
*o
;
2154 HMAP_FOR_EACH (o
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
2156 struct mac_entry
*e
;
2158 ovs_rwlock_wrlock(&o
->ml
->rwlock
);
2159 e
= mac_learning_lookup(o
->ml
, mac
->mac
, mac
->vlan
);
2161 mac_learning_expire(o
->ml
, e
);
2163 ovs_rwlock_unlock(&o
->ml
->rwlock
);
2168 mac_learning_expire(ml
, mac
);
2171 ovs_rwlock_unlock(&ml
->rwlock
);
2174 static struct ofbundle
*
2175 bundle_lookup(const struct ofproto_dpif
*ofproto
, void *aux
)
2177 struct ofbundle
*bundle
;
2179 HMAP_FOR_EACH_IN_BUCKET (bundle
, hmap_node
, hash_pointer(aux
, 0),
2180 &ofproto
->bundles
) {
2181 if (bundle
->aux
== aux
) {
2189 bundle_update(struct ofbundle
*bundle
)
2191 struct ofport_dpif
*port
;
2193 bundle
->floodable
= true;
2194 LIST_FOR_EACH (port
, bundle_node
, &bundle
->ports
) {
2195 if (port
->up
.pp
.config
& OFPUTIL_PC_NO_FLOOD
2197 || !stp_forward_in_state(port
->stp_state
)) {
2198 bundle
->floodable
= false;
2205 bundle_del_port(struct ofport_dpif
*port
)
2207 struct ofbundle
*bundle
= port
->bundle
;
2209 bundle
->ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2211 list_remove(&port
->bundle_node
);
2212 port
->bundle
= NULL
;
2215 lacp_slave_unregister(bundle
->lacp
, port
);
2218 bond_slave_unregister(bundle
->bond
, port
);
2221 bundle_update(bundle
);
2225 bundle_add_port(struct ofbundle
*bundle
, ofp_port_t ofp_port
,
2226 struct lacp_slave_settings
*lacp
)
2228 struct ofport_dpif
*port
;
2230 port
= get_ofp_port(bundle
->ofproto
, ofp_port
);
2235 if (port
->bundle
!= bundle
) {
2236 bundle
->ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2238 bundle_remove(&port
->up
);
2241 port
->bundle
= bundle
;
2242 list_push_back(&bundle
->ports
, &port
->bundle_node
);
2243 if (port
->up
.pp
.config
& OFPUTIL_PC_NO_FLOOD
2245 || !stp_forward_in_state(port
->stp_state
)) {
2246 bundle
->floodable
= false;
2250 bundle
->ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2251 lacp_slave_register(bundle
->lacp
, port
, lacp
);
2258 bundle_destroy(struct ofbundle
*bundle
)
2260 struct ofproto_dpif
*ofproto
;
2261 struct ofport_dpif
*port
, *next_port
;
2267 ofproto
= bundle
->ofproto
;
2268 mbridge_unregister_bundle(ofproto
->mbridge
, bundle
->aux
);
2270 ovs_rwlock_wrlock(&xlate_rwlock
);
2271 xlate_bundle_remove(bundle
);
2272 ovs_rwlock_unlock(&xlate_rwlock
);
2274 LIST_FOR_EACH_SAFE (port
, next_port
, bundle_node
, &bundle
->ports
) {
2275 bundle_del_port(port
);
2278 bundle_flush_macs(bundle
, true);
2279 hmap_remove(&ofproto
->bundles
, &bundle
->hmap_node
);
2281 free(bundle
->trunks
);
2282 lacp_unref(bundle
->lacp
);
2283 bond_unref(bundle
->bond
);
2288 bundle_set(struct ofproto
*ofproto_
, void *aux
,
2289 const struct ofproto_bundle_settings
*s
)
2291 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2292 bool need_flush
= false;
2293 struct ofport_dpif
*port
;
2294 struct ofbundle
*bundle
;
2295 unsigned long *trunks
;
2301 bundle_destroy(bundle_lookup(ofproto
, aux
));
2305 ovs_assert(s
->n_slaves
== 1 || s
->bond
!= NULL
);
2306 ovs_assert((s
->lacp
!= NULL
) == (s
->lacp_slaves
!= NULL
));
2308 bundle
= bundle_lookup(ofproto
, aux
);
2310 bundle
= xmalloc(sizeof *bundle
);
2312 bundle
->ofproto
= ofproto
;
2313 hmap_insert(&ofproto
->bundles
, &bundle
->hmap_node
,
2314 hash_pointer(aux
, 0));
2316 bundle
->name
= NULL
;
2318 list_init(&bundle
->ports
);
2319 bundle
->vlan_mode
= PORT_VLAN_TRUNK
;
2321 bundle
->trunks
= NULL
;
2322 bundle
->use_priority_tags
= s
->use_priority_tags
;
2323 bundle
->lacp
= NULL
;
2324 bundle
->bond
= NULL
;
2326 bundle
->floodable
= true;
2327 mbridge_register_bundle(ofproto
->mbridge
, bundle
);
2330 if (!bundle
->name
|| strcmp(s
->name
, bundle
->name
)) {
2332 bundle
->name
= xstrdup(s
->name
);
2337 ofproto
->lacp_enabled
= true;
2338 if (!bundle
->lacp
) {
2339 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2340 bundle
->lacp
= lacp_create();
2342 lacp_configure(bundle
->lacp
, s
->lacp
);
2344 lacp_unref(bundle
->lacp
);
2345 bundle
->lacp
= NULL
;
2348 /* Update set of ports. */
2350 for (i
= 0; i
< s
->n_slaves
; i
++) {
2351 if (!bundle_add_port(bundle
, s
->slaves
[i
],
2352 s
->lacp
? &s
->lacp_slaves
[i
] : NULL
)) {
2356 if (!ok
|| list_size(&bundle
->ports
) != s
->n_slaves
) {
2357 struct ofport_dpif
*next_port
;
2359 LIST_FOR_EACH_SAFE (port
, next_port
, bundle_node
, &bundle
->ports
) {
2360 for (i
= 0; i
< s
->n_slaves
; i
++) {
2361 if (s
->slaves
[i
] == port
->up
.ofp_port
) {
2366 bundle_del_port(port
);
2370 ovs_assert(list_size(&bundle
->ports
) <= s
->n_slaves
);
2372 if (list_is_empty(&bundle
->ports
)) {
2373 bundle_destroy(bundle
);
2377 /* Set VLAN tagging mode */
2378 if (s
->vlan_mode
!= bundle
->vlan_mode
2379 || s
->use_priority_tags
!= bundle
->use_priority_tags
) {
2380 bundle
->vlan_mode
= s
->vlan_mode
;
2381 bundle
->use_priority_tags
= s
->use_priority_tags
;
2386 vlan
= (s
->vlan_mode
== PORT_VLAN_TRUNK
? -1
2387 : s
->vlan
>= 0 && s
->vlan
<= 4095 ? s
->vlan
2389 if (vlan
!= bundle
->vlan
) {
2390 bundle
->vlan
= vlan
;
2394 /* Get trunked VLANs. */
2395 switch (s
->vlan_mode
) {
2396 case PORT_VLAN_ACCESS
:
2400 case PORT_VLAN_TRUNK
:
2401 trunks
= CONST_CAST(unsigned long *, s
->trunks
);
2404 case PORT_VLAN_NATIVE_UNTAGGED
:
2405 case PORT_VLAN_NATIVE_TAGGED
:
2406 if (vlan
!= 0 && (!s
->trunks
2407 || !bitmap_is_set(s
->trunks
, vlan
)
2408 || bitmap_is_set(s
->trunks
, 0))) {
2409 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2411 trunks
= bitmap_clone(s
->trunks
, 4096);
2413 trunks
= bitmap_allocate1(4096);
2415 bitmap_set1(trunks
, vlan
);
2416 bitmap_set0(trunks
, 0);
2418 trunks
= CONST_CAST(unsigned long *, s
->trunks
);
2425 if (!vlan_bitmap_equal(trunks
, bundle
->trunks
)) {
2426 free(bundle
->trunks
);
2427 if (trunks
== s
->trunks
) {
2428 bundle
->trunks
= vlan_bitmap_clone(trunks
);
2430 bundle
->trunks
= trunks
;
2435 if (trunks
!= s
->trunks
) {
2440 if (!list_is_short(&bundle
->ports
)) {
2441 bundle
->ofproto
->has_bonded_bundles
= true;
2443 if (bond_reconfigure(bundle
->bond
, s
->bond
)) {
2444 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2447 bundle
->bond
= bond_create(s
->bond
, ofproto
);
2448 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2451 LIST_FOR_EACH (port
, bundle_node
, &bundle
->ports
) {
2452 bond_slave_register(bundle
->bond
, port
,
2453 port
->up
.ofp_port
, port
->up
.netdev
);
2456 bond_unref(bundle
->bond
);
2457 bundle
->bond
= NULL
;
2460 /* If we changed something that would affect MAC learning, un-learn
2461 * everything on this port and force flow revalidation. */
2463 bundle_flush_macs(bundle
, false);
2470 bundle_remove(struct ofport
*port_
)
2472 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
2473 struct ofbundle
*bundle
= port
->bundle
;
2476 bundle_del_port(port
);
2477 if (list_is_empty(&bundle
->ports
)) {
2478 bundle_destroy(bundle
);
2479 } else if (list_is_short(&bundle
->ports
)) {
2480 bond_unref(bundle
->bond
);
2481 bundle
->bond
= NULL
;
2487 send_pdu_cb(void *port_
, const void *pdu
, size_t pdu_size
)
2489 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 10);
2490 struct ofport_dpif
*port
= port_
;
2491 uint8_t ea
[ETH_ADDR_LEN
];
2494 error
= netdev_get_etheraddr(port
->up
.netdev
, ea
);
2496 struct ofpbuf packet
;
2499 ofpbuf_init(&packet
, 0);
2500 packet_pdu
= eth_compose(&packet
, eth_addr_lacp
, ea
, ETH_TYPE_LACP
,
2502 memcpy(packet_pdu
, pdu
, pdu_size
);
2504 ofproto_dpif_send_packet(port
, &packet
);
2505 ofpbuf_uninit(&packet
);
2507 VLOG_ERR_RL(&rl
, "port %s: cannot obtain Ethernet address of iface "
2508 "%s (%s)", port
->bundle
->name
,
2509 netdev_get_name(port
->up
.netdev
), ovs_strerror(error
));
2514 bundle_send_learning_packets(struct ofbundle
*bundle
)
2516 struct ofproto_dpif
*ofproto
= bundle
->ofproto
;
2517 struct ofpbuf
*learning_packet
;
2518 int error
, n_packets
, n_errors
;
2519 struct mac_entry
*e
;
2520 struct list packets
;
2522 list_init(&packets
);
2523 ovs_rwlock_rdlock(&ofproto
->ml
->rwlock
);
2524 LIST_FOR_EACH (e
, lru_node
, &ofproto
->ml
->lrus
) {
2525 if (e
->port
.p
!= bundle
) {
2528 learning_packet
= bond_compose_learning_packet(bundle
->bond
,
2531 /* Temporarily use 'frame' as a private pointer (see below). */
2532 ovs_assert(learning_packet
->frame
== ofpbuf_data(learning_packet
));
2533 learning_packet
->frame
= port_void
;
2534 list_push_back(&packets
, &learning_packet
->list_node
);
2537 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
2539 error
= n_packets
= n_errors
= 0;
2540 LIST_FOR_EACH (learning_packet
, list_node
, &packets
) {
2542 void *port_void
= learning_packet
->frame
;
2544 /* Restore 'frame'. */
2545 learning_packet
->frame
= ofpbuf_data(learning_packet
);
2546 ret
= ofproto_dpif_send_packet(port_void
, learning_packet
);
2553 ofpbuf_list_delete(&packets
);
2556 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
2557 VLOG_WARN_RL(&rl
, "bond %s: %d errors sending %d gratuitous learning "
2558 "packets, last error was: %s",
2559 bundle
->name
, n_errors
, n_packets
, ovs_strerror(error
));
2561 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2562 bundle
->name
, n_packets
);
2567 bundle_run(struct ofbundle
*bundle
)
2570 lacp_run(bundle
->lacp
, send_pdu_cb
);
2573 struct ofport_dpif
*port
;
2575 LIST_FOR_EACH (port
, bundle_node
, &bundle
->ports
) {
2576 bond_slave_set_may_enable(bundle
->bond
, port
, port
->may_enable
);
2579 if (bond_run(bundle
->bond
, lacp_status(bundle
->lacp
))) {
2580 bundle
->ofproto
->backer
->need_revalidate
= REV_BOND
;
2583 if (bond_should_send_learning_packets(bundle
->bond
)) {
2584 bundle_send_learning_packets(bundle
);
2590 bundle_wait(struct ofbundle
*bundle
)
2593 lacp_wait(bundle
->lacp
);
2596 bond_wait(bundle
->bond
);
2603 mirror_set__(struct ofproto
*ofproto_
, void *aux
,
2604 const struct ofproto_mirror_settings
*s
)
2606 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2607 struct ofbundle
**srcs
, **dsts
;
2612 mirror_destroy(ofproto
->mbridge
, aux
);
2616 srcs
= xmalloc(s
->n_srcs
* sizeof *srcs
);
2617 dsts
= xmalloc(s
->n_dsts
* sizeof *dsts
);
2619 for (i
= 0; i
< s
->n_srcs
; i
++) {
2620 srcs
[i
] = bundle_lookup(ofproto
, s
->srcs
[i
]);
2623 for (i
= 0; i
< s
->n_dsts
; i
++) {
2624 dsts
[i
] = bundle_lookup(ofproto
, s
->dsts
[i
]);
2627 error
= mirror_set(ofproto
->mbridge
, aux
, s
->name
, srcs
, s
->n_srcs
, dsts
,
2628 s
->n_dsts
, s
->src_vlans
,
2629 bundle_lookup(ofproto
, s
->out_bundle
), s
->out_vlan
);
2636 mirror_get_stats__(struct ofproto
*ofproto
, void *aux
,
2637 uint64_t *packets
, uint64_t *bytes
)
2639 return mirror_get_stats(ofproto_dpif_cast(ofproto
)->mbridge
, aux
, packets
,
2644 set_flood_vlans(struct ofproto
*ofproto_
, unsigned long *flood_vlans
)
2646 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2647 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
2648 if (mac_learning_set_flood_vlans(ofproto
->ml
, flood_vlans
)) {
2649 mac_learning_flush(ofproto
->ml
);
2651 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
2656 is_mirror_output_bundle(const struct ofproto
*ofproto_
, void *aux
)
2658 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2659 struct ofbundle
*bundle
= bundle_lookup(ofproto
, aux
);
2660 return bundle
&& mirror_bundle_out(ofproto
->mbridge
, bundle
) != 0;
2664 forward_bpdu_changed(struct ofproto
*ofproto_
)
2666 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2667 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2671 set_mac_table_config(struct ofproto
*ofproto_
, unsigned int idle_time
,
2674 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2675 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
2676 mac_learning_set_idle_time(ofproto
->ml
, idle_time
);
2677 mac_learning_set_max_entries(ofproto
->ml
, max_entries
);
2678 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
2683 static struct ofport_dpif
*
2684 get_ofp_port(const struct ofproto_dpif
*ofproto
, ofp_port_t ofp_port
)
2686 struct ofport
*ofport
= ofproto_get_port(&ofproto
->up
, ofp_port
);
2687 return ofport
? ofport_dpif_cast(ofport
) : NULL
;
2691 ofproto_port_from_dpif_port(struct ofproto_dpif
*ofproto
,
2692 struct ofproto_port
*ofproto_port
,
2693 struct dpif_port
*dpif_port
)
2695 ofproto_port
->name
= dpif_port
->name
;
2696 ofproto_port
->type
= dpif_port
->type
;
2697 ofproto_port
->ofp_port
= odp_port_to_ofp_port(ofproto
, dpif_port
->port_no
);
2701 ofport_update_peer(struct ofport_dpif
*ofport
)
2703 const struct ofproto_dpif
*ofproto
;
2704 struct dpif_backer
*backer
;
2707 if (!netdev_vport_is_patch(ofport
->up
.netdev
)) {
2711 backer
= ofproto_dpif_cast(ofport
->up
.ofproto
)->backer
;
2712 backer
->need_revalidate
= REV_RECONFIGURE
;
2715 ofport
->peer
->peer
= NULL
;
2716 ofport
->peer
= NULL
;
2719 peer_name
= netdev_vport_patch_peer(ofport
->up
.netdev
);
2724 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
2725 struct ofport
*peer_ofport
;
2726 struct ofport_dpif
*peer
;
2729 if (ofproto
->backer
!= backer
) {
2733 peer_ofport
= shash_find_data(&ofproto
->up
.port_by_name
, peer_name
);
2738 peer
= ofport_dpif_cast(peer_ofport
);
2739 peer_peer
= netdev_vport_patch_peer(peer
->up
.netdev
);
2740 if (peer_peer
&& !strcmp(netdev_get_name(ofport
->up
.netdev
),
2742 ofport
->peer
= peer
;
2743 ofport
->peer
->peer
= ofport
;
2753 port_run(struct ofport_dpif
*ofport
)
2755 long long int carrier_seq
= netdev_get_carrier_resets(ofport
->up
.netdev
);
2756 bool carrier_changed
= carrier_seq
!= ofport
->carrier_seq
;
2757 bool enable
= netdev_get_carrier(ofport
->up
.netdev
);
2758 bool cfm_enable
= false;
2759 bool bfd_enable
= false;
2761 ofport
->carrier_seq
= carrier_seq
;
2764 int cfm_opup
= cfm_get_opup(ofport
->cfm
);
2766 cfm_enable
= !cfm_get_fault(ofport
->cfm
);
2768 if (cfm_opup
>= 0) {
2769 cfm_enable
= cfm_enable
&& cfm_opup
;
2774 bfd_enable
= bfd_forwarding(ofport
->bfd
);
2777 if (ofport
->bfd
|| ofport
->cfm
) {
2778 enable
= enable
&& (cfm_enable
|| bfd_enable
);
2781 if (ofport
->bundle
) {
2782 enable
= enable
&& lacp_slave_may_enable(ofport
->bundle
->lacp
, ofport
);
2783 if (carrier_changed
) {
2784 lacp_slave_carrier_changed(ofport
->bundle
->lacp
, ofport
);
2788 if (ofport
->may_enable
!= enable
) {
2789 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2790 ofproto
->backer
->need_revalidate
= REV_PORT_TOGGLED
;
2793 ofport
->may_enable
= enable
;
2797 port_query_by_name(const struct ofproto
*ofproto_
, const char *devname
,
2798 struct ofproto_port
*ofproto_port
)
2800 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2801 struct dpif_port dpif_port
;
2804 if (sset_contains(&ofproto
->ghost_ports
, devname
)) {
2805 const char *type
= netdev_get_type_from_name(devname
);
2807 /* We may be called before ofproto->up.port_by_name is populated with
2808 * the appropriate ofport. For this reason, we must get the name and
2809 * type from the netdev layer directly. */
2811 const struct ofport
*ofport
;
2813 ofport
= shash_find_data(&ofproto
->up
.port_by_name
, devname
);
2814 ofproto_port
->ofp_port
= ofport
? ofport
->ofp_port
: OFPP_NONE
;
2815 ofproto_port
->name
= xstrdup(devname
);
2816 ofproto_port
->type
= xstrdup(type
);
2822 if (!sset_contains(&ofproto
->ports
, devname
)) {
2825 error
= dpif_port_query_by_name(ofproto
->backer
->dpif
,
2826 devname
, &dpif_port
);
2828 ofproto_port_from_dpif_port(ofproto
, ofproto_port
, &dpif_port
);
2834 port_add(struct ofproto
*ofproto_
, struct netdev
*netdev
)
2836 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2837 const char *devname
= netdev_get_name(netdev
);
2838 char namebuf
[NETDEV_VPORT_NAME_BUFSIZE
];
2839 const char *dp_port_name
;
2841 if (netdev_vport_is_patch(netdev
)) {
2842 sset_add(&ofproto
->ghost_ports
, netdev_get_name(netdev
));
2846 dp_port_name
= netdev_vport_get_dpif_port(netdev
, namebuf
, sizeof namebuf
);
2847 if (!dpif_port_exists(ofproto
->backer
->dpif
, dp_port_name
)) {
2848 odp_port_t port_no
= ODPP_NONE
;
2851 error
= dpif_port_add(ofproto
->backer
->dpif
, netdev
, &port_no
);
2855 if (netdev_get_tunnel_config(netdev
)) {
2856 simap_put(&ofproto
->backer
->tnl_backers
,
2857 dp_port_name
, odp_to_u32(port_no
));
2861 if (netdev_get_tunnel_config(netdev
)) {
2862 sset_add(&ofproto
->ghost_ports
, devname
);
2864 sset_add(&ofproto
->ports
, devname
);
2870 port_del(struct ofproto
*ofproto_
, ofp_port_t ofp_port
)
2872 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2873 struct ofport_dpif
*ofport
= get_ofp_port(ofproto
, ofp_port
);
2880 sset_find_and_delete(&ofproto
->ghost_ports
,
2881 netdev_get_name(ofport
->up
.netdev
));
2882 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2883 if (!ofport
->is_tunnel
&& !netdev_vport_is_patch(ofport
->up
.netdev
)) {
2884 error
= dpif_port_del(ofproto
->backer
->dpif
, ofport
->odp_port
);
2886 /* The caller is going to close ofport->up.netdev. If this is a
2887 * bonded port, then the bond is using that netdev, so remove it
2888 * from the bond. The client will need to reconfigure everything
2889 * after deleting ports, so then the slave will get re-added. */
2890 bundle_remove(&ofport
->up
);
2897 port_get_stats(const struct ofport
*ofport_
, struct netdev_stats
*stats
)
2899 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2902 error
= netdev_get_stats(ofport
->up
.netdev
, stats
);
2904 if (!error
&& ofport_
->ofp_port
== OFPP_LOCAL
) {
2905 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2907 ovs_mutex_lock(&ofproto
->stats_mutex
);
2908 /* ofproto->stats.tx_packets represents packets that we created
2909 * internally and sent to some port (e.g. packets sent with
2910 * ofproto_dpif_send_packet()). Account for them as if they had
2911 * come from OFPP_LOCAL and got forwarded. */
2913 if (stats
->rx_packets
!= UINT64_MAX
) {
2914 stats
->rx_packets
+= ofproto
->stats
.tx_packets
;
2917 if (stats
->rx_bytes
!= UINT64_MAX
) {
2918 stats
->rx_bytes
+= ofproto
->stats
.tx_bytes
;
2921 /* ofproto->stats.rx_packets represents packets that were received on
2922 * some port and we processed internally and dropped (e.g. STP).
2923 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2925 if (stats
->tx_packets
!= UINT64_MAX
) {
2926 stats
->tx_packets
+= ofproto
->stats
.rx_packets
;
2929 if (stats
->tx_bytes
!= UINT64_MAX
) {
2930 stats
->tx_bytes
+= ofproto
->stats
.rx_bytes
;
2932 ovs_mutex_unlock(&ofproto
->stats_mutex
);
2938 struct port_dump_state
{
2943 struct ofproto_port port
;
2948 port_dump_start(const struct ofproto
*ofproto_ OVS_UNUSED
, void **statep
)
2950 *statep
= xzalloc(sizeof(struct port_dump_state
));
2955 port_dump_next(const struct ofproto
*ofproto_
, void *state_
,
2956 struct ofproto_port
*port
)
2958 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2959 struct port_dump_state
*state
= state_
;
2960 const struct sset
*sset
;
2961 struct sset_node
*node
;
2963 if (state
->has_port
) {
2964 ofproto_port_destroy(&state
->port
);
2965 state
->has_port
= false;
2967 sset
= state
->ghost
? &ofproto
->ghost_ports
: &ofproto
->ports
;
2968 while ((node
= sset_at_position(sset
, &state
->bucket
, &state
->offset
))) {
2971 error
= port_query_by_name(ofproto_
, node
->name
, &state
->port
);
2973 *port
= state
->port
;
2974 state
->has_port
= true;
2976 } else if (error
!= ENODEV
) {
2981 if (!state
->ghost
) {
2982 state
->ghost
= true;
2985 return port_dump_next(ofproto_
, state_
, port
);
2992 port_dump_done(const struct ofproto
*ofproto_ OVS_UNUSED
, void *state_
)
2994 struct port_dump_state
*state
= state_
;
2996 if (state
->has_port
) {
2997 ofproto_port_destroy(&state
->port
);
3004 port_poll(const struct ofproto
*ofproto_
, char **devnamep
)
3006 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3008 if (ofproto
->port_poll_errno
) {
3009 int error
= ofproto
->port_poll_errno
;
3010 ofproto
->port_poll_errno
= 0;
3014 if (sset_is_empty(&ofproto
->port_poll_set
)) {
3018 *devnamep
= sset_pop(&ofproto
->port_poll_set
);
3023 port_poll_wait(const struct ofproto
*ofproto_
)
3025 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3026 dpif_port_poll_wait(ofproto
->backer
->dpif
);
3030 port_is_lacp_current(const struct ofport
*ofport_
)
3032 const struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
3033 return (ofport
->bundle
&& ofport
->bundle
->lacp
3034 ? lacp_slave_is_current(ofport
->bundle
->lacp
, ofport
)
3038 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3039 * then delete it entirely. */
3041 rule_expire(struct rule_dpif
*rule
)
3042 OVS_REQUIRES(ofproto_mutex
)
3044 uint16_t hard_timeout
, idle_timeout
;
3045 long long int now
= time_msec();
3048 ovs_assert(!rule
->up
.pending
);
3050 hard_timeout
= rule
->up
.hard_timeout
;
3051 idle_timeout
= rule
->up
.idle_timeout
;
3053 /* Has 'rule' expired? */
3055 long long int modified
;
3057 ovs_mutex_lock(&rule
->up
.mutex
);
3058 modified
= rule
->up
.modified
;
3059 ovs_mutex_unlock(&rule
->up
.mutex
);
3061 if (now
> modified
+ hard_timeout
* 1000) {
3062 reason
= OFPRR_HARD_TIMEOUT
;
3066 if (reason
< 0 && idle_timeout
) {
3069 ovs_mutex_lock(&rule
->stats_mutex
);
3070 used
= rule
->stats
.used
;
3071 ovs_mutex_unlock(&rule
->stats_mutex
);
3073 if (now
> used
+ idle_timeout
* 1000) {
3074 reason
= OFPRR_IDLE_TIMEOUT
;
3079 COVERAGE_INC(ofproto_dpif_expired
);
3080 ofproto_rule_expire(&rule
->up
, reason
);
3084 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3085 * 'flow' must reflect the data in 'packet'. */
3087 ofproto_dpif_execute_actions(struct ofproto_dpif
*ofproto
,
3088 const struct flow
*flow
,
3089 struct rule_dpif
*rule
,
3090 const struct ofpact
*ofpacts
, size_t ofpacts_len
,
3091 struct ofpbuf
*packet
)
3093 struct dpif_flow_stats stats
;
3094 struct xlate_out xout
;
3095 struct xlate_in xin
;
3097 struct dpif_execute execute
;
3100 ovs_assert((rule
!= NULL
) != (ofpacts
!= NULL
));
3102 dpif_flow_stats_extract(flow
, packet
, time_msec(), &stats
);
3105 rule_dpif_credit_stats(rule
, &stats
);
3108 xlate_in_init(&xin
, ofproto
, flow
, rule
, stats
.tcp_flags
, packet
);
3109 xin
.ofpacts
= ofpacts
;
3110 xin
.ofpacts_len
= ofpacts_len
;
3111 xin
.resubmit_stats
= &stats
;
3112 xlate_actions(&xin
, &xout
);
3114 in_port
= flow
->in_port
.ofp_port
;
3115 if (in_port
== OFPP_NONE
) {
3116 in_port
= OFPP_LOCAL
;
3118 execute
.actions
= ofpbuf_data(&xout
.odp_actions
);
3119 execute
.actions_len
= ofpbuf_size(&xout
.odp_actions
);
3120 execute
.packet
= packet
;
3121 execute
.md
.tunnel
= flow
->tunnel
;
3122 execute
.md
.skb_priority
= flow
->skb_priority
;
3123 execute
.md
.pkt_mark
= flow
->pkt_mark
;
3124 execute
.md
.in_port
.odp_port
= ofp_port_to_odp_port(ofproto
, in_port
);
3125 execute
.needs_help
= (xout
.slow
& SLOW_ACTION
) != 0;
3127 error
= dpif_execute(ofproto
->backer
->dpif
, &execute
);
3129 xlate_out_uninit(&xout
);
3135 rule_dpif_credit_stats(struct rule_dpif
*rule
,
3136 const struct dpif_flow_stats
*stats
)
3138 ovs_mutex_lock(&rule
->stats_mutex
);
3139 rule
->stats
.n_packets
+= stats
->n_packets
;
3140 rule
->stats
.n_bytes
+= stats
->n_bytes
;
3141 rule
->stats
.used
= MAX(rule
->stats
.used
, stats
->used
);
3142 ovs_mutex_unlock(&rule
->stats_mutex
);
3146 rule_dpif_is_fail_open(const struct rule_dpif
*rule
)
3148 return is_fail_open_rule(&rule
->up
);
3152 rule_dpif_is_table_miss(const struct rule_dpif
*rule
)
3154 return rule_is_table_miss(&rule
->up
);
3158 rule_dpif_is_internal(const struct rule_dpif
*rule
)
3160 return rule_is_internal(&rule
->up
);
3164 rule_dpif_get_flow_cookie(const struct rule_dpif
*rule
)
3165 OVS_REQUIRES(rule
->up
.mutex
)
3167 return rule
->up
.flow_cookie
;
3171 rule_dpif_reduce_timeouts(struct rule_dpif
*rule
, uint16_t idle_timeout
,
3172 uint16_t hard_timeout
)
3174 ofproto_rule_reduce_timeouts(&rule
->up
, idle_timeout
, hard_timeout
);
3177 /* Returns 'rule''s actions. The caller owns a reference on the returned
3178 * actions and must eventually release it (with rule_actions_unref()) to avoid
3180 struct rule_actions
*
3181 rule_dpif_get_actions(const struct rule_dpif
*rule
)
3183 return rule_get_actions(&rule
->up
);
3186 /* Lookup 'flow' in table 0 of 'ofproto''s classifier.
3187 * If 'wc' is non-null, sets the fields that were relevant as part of
3188 * the lookup. Returns the table_id where a match or miss occurred.
3190 * The return value will be zero unless there was a miss and
3191 * OFPTC11_TABLE_MISS_CONTINUE is in effect for the sequence of tables
3192 * where misses occur.
3194 * The rule is returned in '*rule', which is valid at least until the next
3195 * RCU quiescent period. If the '*rule' needs to stay around longer,
3196 * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3197 * on it before this returns. */
3199 rule_dpif_lookup(struct ofproto_dpif
*ofproto
, struct flow
*flow
,
3200 struct flow_wildcards
*wc
, struct rule_dpif
**rule
,
3203 enum rule_dpif_lookup_verdict verdict
;
3204 enum ofputil_port_config config
= 0;
3207 if (ofproto_dpif_get_enable_recirc(ofproto
)) {
3208 /* Always exactly match recirc_id since datapath supports
3211 wc
->masks
.recirc_id
= UINT32_MAX
;
3214 /* Start looking up from internal table for post recirculation flows
3215 * or packets. We can also simply send all, including normal flows
3216 * or packets to the internal table. They will not match any post
3217 * recirculation rules except the 'catch all' rule that resubmit
3220 * As an optimization, we send normal flows and packets to table 0
3221 * directly, saving one table lookup. */
3222 table_id
= flow
->recirc_id
? TBL_INTERNAL
: 0;
3227 verdict
= rule_dpif_lookup_from_table(ofproto
, flow
, wc
, true,
3228 &table_id
, rule
, take_ref
);
3231 case RULE_DPIF_LOOKUP_VERDICT_MATCH
:
3233 case RULE_DPIF_LOOKUP_VERDICT_CONTROLLER
: {
3234 struct ofport_dpif
*port
;
3236 port
= get_ofp_port(ofproto
, flow
->in_port
.ofp_port
);
3238 VLOG_WARN_RL(&rl
, "packet-in on unknown OpenFlow port %"PRIu16
,
3239 flow
->in_port
.ofp_port
);
3241 config
= port
? port
->up
.pp
.config
: 0;
3244 case RULE_DPIF_LOOKUP_VERDICT_DROP
:
3245 config
= OFPUTIL_PC_NO_PACKET_IN
;
3247 case RULE_DPIF_LOOKUP_VERDICT_DEFAULT
:
3248 if (!connmgr_wants_packet_in_on_miss(ofproto
->up
.connmgr
)) {
3249 config
= OFPUTIL_PC_NO_PACKET_IN
;
3256 choose_miss_rule(config
, ofproto
->miss_rule
,
3257 ofproto
->no_packet_in_rule
, rule
, take_ref
);
3261 /* The returned rule is valid at least until the next RCU quiescent period.
3262 * If the '*rule' needs to stay around longer, a non-zero 'take_ref' must be
3263 * passed in to cause a reference to be taken on it before this returns. */
3264 static struct rule_dpif
*
3265 rule_dpif_lookup_in_table(struct ofproto_dpif
*ofproto
, uint8_t table_id
,
3266 const struct flow
*flow
, struct flow_wildcards
*wc
,
3269 struct classifier
*cls
= &ofproto
->up
.tables
[table_id
].cls
;
3270 const struct cls_rule
*cls_rule
;
3271 struct rule_dpif
*rule
;
3273 fat_rwlock_rdlock(&cls
->rwlock
);
3274 if (ofproto
->up
.frag_handling
!= OFPC_FRAG_NX_MATCH
) {
3276 memset(&wc
->masks
.dl_type
, 0xff, sizeof wc
->masks
.dl_type
);
3277 if (is_ip_any(flow
)) {
3278 wc
->masks
.nw_frag
|= FLOW_NW_FRAG_MASK
;
3282 if (flow
->nw_frag
& FLOW_NW_FRAG_ANY
) {
3283 if (ofproto
->up
.frag_handling
== OFPC_FRAG_NORMAL
) {
3284 /* We must pretend that transport ports are unavailable. */
3285 struct flow ofpc_normal_flow
= *flow
;
3286 ofpc_normal_flow
.tp_src
= htons(0);
3287 ofpc_normal_flow
.tp_dst
= htons(0);
3288 cls_rule
= classifier_lookup(cls
, &ofpc_normal_flow
, wc
);
3290 /* Must be OFPC_FRAG_DROP (we don't have OFPC_FRAG_REASM). */
3291 cls_rule
= &ofproto
->drop_frags_rule
->up
.cr
;
3294 cls_rule
= classifier_lookup(cls
, flow
, wc
);
3297 cls_rule
= classifier_lookup(cls
, flow
, wc
);
3300 rule
= rule_dpif_cast(rule_from_cls_rule(cls_rule
));
3302 rule_dpif_ref(rule
);
3304 fat_rwlock_unlock(&cls
->rwlock
);
3309 /* Look up 'flow' in 'ofproto''s classifier starting from table '*table_id'.
3310 * Stores the rule that was found in '*rule', or NULL if none was found.
3311 * Updates 'wc', if nonnull, to reflect the fields that were used during the
3314 * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3315 * if none is found then the table miss configuration for that table is
3316 * honored, which can result in additional lookups in other OpenFlow tables.
3317 * In this case the function updates '*table_id' to reflect the final OpenFlow
3318 * table that was searched.
3320 * If 'honor_table_miss' is false, then only one table lookup occurs, in
3325 * - RULE_DPIF_LOOKUP_VERDICT_MATCH if a rule (in '*rule') was found.
3327 * - RULE_OFPTC_TABLE_MISS_CONTROLLER if no rule was found and either:
3328 * + 'honor_table_miss' is false
3329 * + a table miss configuration specified that the packet should be
3330 * sent to the controller in this case.
3332 * - RULE_DPIF_LOOKUP_VERDICT_DROP if no rule was found, 'honor_table_miss'
3333 * is true and a table miss configuration specified that the packet
3334 * should be dropped in this case.
3336 * - RULE_DPIF_LOOKUP_VERDICT_DEFAULT if no rule was found,
3337 * 'honor_table_miss' is true and a table miss configuration has
3338 * not been specified in this case.
3340 * The rule is returned in '*rule', which is valid at least until the next
3341 * RCU quiescent period. If the '*rule' needs to stay around longer,
3342 * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3343 * on it before this returns. */
3344 enum rule_dpif_lookup_verdict
3345 rule_dpif_lookup_from_table(struct ofproto_dpif
*ofproto
,
3346 const struct flow
*flow
,
3347 struct flow_wildcards
*wc
,
3348 bool honor_table_miss
,
3349 uint8_t *table_id
, struct rule_dpif
**rule
,
3354 for (next_id
= *table_id
;
3355 next_id
< ofproto
->up
.n_tables
;
3356 next_id
++, next_id
+= (next_id
== TBL_INTERNAL
))
3358 *table_id
= next_id
;
3359 *rule
= rule_dpif_lookup_in_table(ofproto
, *table_id
, flow
, wc
,
3362 return RULE_DPIF_LOOKUP_VERDICT_MATCH
;
3363 } else if (!honor_table_miss
) {
3364 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER
;
3366 switch (ofproto_table_get_config(&ofproto
->up
, *table_id
)) {
3367 case OFPROTO_TABLE_MISS_CONTINUE
:
3370 case OFPROTO_TABLE_MISS_CONTROLLER
:
3371 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER
;
3373 case OFPROTO_TABLE_MISS_DROP
:
3374 return RULE_DPIF_LOOKUP_VERDICT_DROP
;
3376 case OFPROTO_TABLE_MISS_DEFAULT
:
3377 return RULE_DPIF_LOOKUP_VERDICT_DEFAULT
;
3382 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER
;
3385 /* Given a port configuration (specified as zero if there's no port), chooses
3386 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
3389 * The rule is returned in '*rule', which is valid at least until the next
3390 * RCU quiescent period. If the '*rule' needs to stay around longer,
3391 * a reference must be taken on it (rule_dpif_ref()).
3394 choose_miss_rule(enum ofputil_port_config config
, struct rule_dpif
*miss_rule
,
3395 struct rule_dpif
*no_packet_in_rule
, struct rule_dpif
**rule
,
3398 *rule
= config
& OFPUTIL_PC_NO_PACKET_IN
? no_packet_in_rule
: miss_rule
;
3400 rule_dpif_ref(*rule
);
3405 rule_dpif_ref(struct rule_dpif
*rule
)
3408 ofproto_rule_ref(&rule
->up
);
3413 rule_dpif_unref(struct rule_dpif
*rule
)
3416 ofproto_rule_unref(&rule
->up
);
3421 complete_operation(struct rule_dpif
*rule
)
3422 OVS_REQUIRES(ofproto_mutex
)
3424 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
3426 ofproto
->backer
->need_revalidate
= REV_FLOW_TABLE
;
3427 ofoperation_complete(rule
->up
.pending
, 0);
3430 static struct rule_dpif
*rule_dpif_cast(const struct rule
*rule
)
3432 return rule
? CONTAINER_OF(rule
, struct rule_dpif
, up
) : NULL
;
3435 static struct rule
*
3438 struct rule_dpif
*rule
= xmalloc(sizeof *rule
);
3443 rule_dealloc(struct rule
*rule_
)
3445 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
3450 rule_construct(struct rule
*rule_
)
3451 OVS_NO_THREAD_SAFETY_ANALYSIS
3453 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
3454 ovs_mutex_init_adaptive(&rule
->stats_mutex
);
3455 rule
->stats
.n_packets
= 0;
3456 rule
->stats
.n_bytes
= 0;
3457 rule
->stats
.used
= rule
->up
.modified
;
3462 rule_insert(struct rule
*rule_
)
3463 OVS_REQUIRES(ofproto_mutex
)
3465 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
3466 complete_operation(rule
);
3470 rule_delete(struct rule
*rule_
)
3471 OVS_REQUIRES(ofproto_mutex
)
3473 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
3474 complete_operation(rule
);
3478 rule_destruct(struct rule
*rule_
)
3480 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
3481 ovs_mutex_destroy(&rule
->stats_mutex
);
3485 rule_get_stats(struct rule
*rule_
, uint64_t *packets
, uint64_t *bytes
,
3486 long long int *used
)
3488 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
3490 ovs_mutex_lock(&rule
->stats_mutex
);
3491 *packets
= rule
->stats
.n_packets
;
3492 *bytes
= rule
->stats
.n_bytes
;
3493 *used
= rule
->stats
.used
;
3494 ovs_mutex_unlock(&rule
->stats_mutex
);
3498 rule_dpif_execute(struct rule_dpif
*rule
, const struct flow
*flow
,
3499 struct ofpbuf
*packet
)
3501 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
3503 ofproto_dpif_execute_actions(ofproto
, flow
, rule
, NULL
, 0, packet
);
3507 rule_execute(struct rule
*rule
, const struct flow
*flow
,
3508 struct ofpbuf
*packet
)
3510 rule_dpif_execute(rule_dpif_cast(rule
), flow
, packet
);
3511 ofpbuf_delete(packet
);
3516 rule_modify_actions(struct rule
*rule_
, bool reset_counters
)
3517 OVS_REQUIRES(ofproto_mutex
)
3519 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
3521 if (reset_counters
) {
3522 ovs_mutex_lock(&rule
->stats_mutex
);
3523 rule
->stats
.n_packets
= 0;
3524 rule
->stats
.n_bytes
= 0;
3525 ovs_mutex_unlock(&rule
->stats_mutex
);
3528 complete_operation(rule
);
3531 static struct group_dpif
*group_dpif_cast(const struct ofgroup
*group
)
3533 return group
? CONTAINER_OF(group
, struct group_dpif
, up
) : NULL
;
3536 static struct ofgroup
*
3539 struct group_dpif
*group
= xzalloc(sizeof *group
);
3544 group_dealloc(struct ofgroup
*group_
)
3546 struct group_dpif
*group
= group_dpif_cast(group_
);
3551 group_construct_stats(struct group_dpif
*group
)
3552 OVS_REQUIRES(group
->stats_mutex
)
3554 group
->packet_count
= 0;
3555 group
->byte_count
= 0;
3556 if (!group
->bucket_stats
) {
3557 group
->bucket_stats
= xcalloc(group
->up
.n_buckets
,
3558 sizeof *group
->bucket_stats
);
3560 memset(group
->bucket_stats
, 0, group
->up
.n_buckets
*
3561 sizeof *group
->bucket_stats
);
3566 group_construct(struct ofgroup
*group_
)
3568 struct group_dpif
*group
= group_dpif_cast(group_
);
3569 const struct ofputil_bucket
*bucket
;
3571 /* Prevent group chaining because our locking structure makes it hard to
3572 * implement deadlock-free. (See xlate_group_resource_check().) */
3573 LIST_FOR_EACH (bucket
, list_node
, &group
->up
.buckets
) {
3574 const struct ofpact
*a
;
3576 OFPACT_FOR_EACH (a
, bucket
->ofpacts
, bucket
->ofpacts_len
) {
3577 if (a
->type
== OFPACT_GROUP
) {
3578 return OFPERR_OFPGMFC_CHAINING_UNSUPPORTED
;
3583 ovs_mutex_init_adaptive(&group
->stats_mutex
);
3584 ovs_mutex_lock(&group
->stats_mutex
);
3585 group_construct_stats(group
);
3586 ovs_mutex_unlock(&group
->stats_mutex
);
3591 group_destruct__(struct group_dpif
*group
)
3592 OVS_REQUIRES(group
->stats_mutex
)
3594 free(group
->bucket_stats
);
3595 group
->bucket_stats
= NULL
;
3599 group_destruct(struct ofgroup
*group_
)
3601 struct group_dpif
*group
= group_dpif_cast(group_
);
3602 ovs_mutex_lock(&group
->stats_mutex
);
3603 group_destruct__(group
);
3604 ovs_mutex_unlock(&group
->stats_mutex
);
3605 ovs_mutex_destroy(&group
->stats_mutex
);
3609 group_modify(struct ofgroup
*group_
, struct ofgroup
*victim_
)
3611 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(group_
->ofproto
);
3612 struct group_dpif
*group
= group_dpif_cast(group_
);
3613 struct group_dpif
*victim
= group_dpif_cast(victim_
);
3615 ovs_mutex_lock(&group
->stats_mutex
);
3616 if (victim
->up
.n_buckets
< group
->up
.n_buckets
) {
3617 group_destruct__(group
);
3619 group_construct_stats(group
);
3620 ovs_mutex_unlock(&group
->stats_mutex
);
3622 ofproto
->backer
->need_revalidate
= REV_FLOW_TABLE
;
3628 group_get_stats(const struct ofgroup
*group_
, struct ofputil_group_stats
*ogs
)
3630 struct group_dpif
*group
= group_dpif_cast(group_
);
3632 ovs_mutex_lock(&group
->stats_mutex
);
3633 ogs
->packet_count
= group
->packet_count
;
3634 ogs
->byte_count
= group
->byte_count
;
3635 memcpy(ogs
->bucket_stats
, group
->bucket_stats
,
3636 group
->up
.n_buckets
* sizeof *group
->bucket_stats
);
3637 ovs_mutex_unlock(&group
->stats_mutex
);
3643 group_dpif_lookup(struct ofproto_dpif
*ofproto
, uint32_t group_id
,
3644 struct group_dpif
**group
)
3645 OVS_TRY_RDLOCK(true, (*group
)->up
.rwlock
)
3647 struct ofgroup
*ofgroup
;
3651 found
= ofproto_group_lookup(&ofproto
->up
, group_id
, &ofgroup
);
3652 *group
= found
? group_dpif_cast(ofgroup
) : NULL
;
3658 group_dpif_release(struct group_dpif
*group
)
3659 OVS_RELEASES(group
->up
.rwlock
)
3661 ofproto_group_release(&group
->up
);
3665 group_dpif_get_buckets(const struct group_dpif
*group
,
3666 const struct list
**buckets
)
3668 *buckets
= &group
->up
.buckets
;
3671 enum ofp11_group_type
3672 group_dpif_get_type(const struct group_dpif
*group
)
3674 return group
->up
.type
;
3677 /* Sends 'packet' out 'ofport'.
3678 * May modify 'packet'.
3679 * Returns 0 if successful, otherwise a positive errno value. */
3681 ofproto_dpif_send_packet(const struct ofport_dpif
*ofport
, struct ofpbuf
*packet
)
3683 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
3686 error
= xlate_send_packet(ofport
, packet
);
3688 ovs_mutex_lock(&ofproto
->stats_mutex
);
3689 ofproto
->stats
.tx_packets
++;
3690 ofproto
->stats
.tx_bytes
+= ofpbuf_size(packet
);
3691 ovs_mutex_unlock(&ofproto
->stats_mutex
);
3696 set_frag_handling(struct ofproto
*ofproto_
,
3697 enum ofp_config_flags frag_handling
)
3699 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3700 if (frag_handling
!= OFPC_FRAG_REASM
) {
3701 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3709 packet_out(struct ofproto
*ofproto_
, struct ofpbuf
*packet
,
3710 const struct flow
*flow
,
3711 const struct ofpact
*ofpacts
, size_t ofpacts_len
)
3713 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3715 ofproto_dpif_execute_actions(ofproto
, flow
, NULL
, ofpacts
,
3716 ofpacts_len
, packet
);
3723 set_netflow(struct ofproto
*ofproto_
,
3724 const struct netflow_options
*netflow_options
)
3726 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3728 if (netflow_options
) {
3729 if (!ofproto
->netflow
) {
3730 ofproto
->netflow
= netflow_create();
3731 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3733 return netflow_set_options(ofproto
->netflow
, netflow_options
);
3734 } else if (ofproto
->netflow
) {
3735 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3736 netflow_unref(ofproto
->netflow
);
3737 ofproto
->netflow
= NULL
;
3744 get_netflow_ids(const struct ofproto
*ofproto_
,
3745 uint8_t *engine_type
, uint8_t *engine_id
)
3747 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3749 dpif_get_netflow_ids(ofproto
->backer
->dpif
, engine_type
, engine_id
);
3752 static struct ofproto_dpif
*
3753 ofproto_dpif_lookup(const char *name
)
3755 struct ofproto_dpif
*ofproto
;
3757 HMAP_FOR_EACH_WITH_HASH (ofproto
, all_ofproto_dpifs_node
,
3758 hash_string(name
, 0), &all_ofproto_dpifs
) {
3759 if (!strcmp(ofproto
->up
.name
, name
)) {
3767 ofproto_unixctl_fdb_flush(struct unixctl_conn
*conn
, int argc
,
3768 const char *argv
[], void *aux OVS_UNUSED
)
3770 struct ofproto_dpif
*ofproto
;
3773 ofproto
= ofproto_dpif_lookup(argv
[1]);
3775 unixctl_command_reply_error(conn
, "no such bridge");
3778 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
3779 mac_learning_flush(ofproto
->ml
);
3780 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
3782 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
3783 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
3784 mac_learning_flush(ofproto
->ml
);
3785 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
3789 unixctl_command_reply(conn
, "table successfully flushed");
3792 static struct ofport_dpif
*
3793 ofbundle_get_a_port(const struct ofbundle
*bundle
)
3795 return CONTAINER_OF(list_front(&bundle
->ports
), struct ofport_dpif
,
3800 ofproto_unixctl_fdb_show(struct unixctl_conn
*conn
, int argc OVS_UNUSED
,
3801 const char *argv
[], void *aux OVS_UNUSED
)
3803 struct ds ds
= DS_EMPTY_INITIALIZER
;
3804 const struct ofproto_dpif
*ofproto
;
3805 const struct mac_entry
*e
;
3807 ofproto
= ofproto_dpif_lookup(argv
[1]);
3809 unixctl_command_reply_error(conn
, "no such bridge");
3813 ds_put_cstr(&ds
, " port VLAN MAC Age\n");
3814 ovs_rwlock_rdlock(&ofproto
->ml
->rwlock
);
3815 LIST_FOR_EACH (e
, lru_node
, &ofproto
->ml
->lrus
) {
3816 struct ofbundle
*bundle
= e
->port
.p
;
3817 char name
[OFP_MAX_PORT_NAME_LEN
];
3819 ofputil_port_to_string(ofbundle_get_a_port(bundle
)->up
.ofp_port
,
3821 ds_put_format(&ds
, "%5s %4d "ETH_ADDR_FMT
" %3d\n",
3822 name
, e
->vlan
, ETH_ADDR_ARGS(e
->mac
),
3823 mac_entry_age(ofproto
->ml
, e
));
3825 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
3826 unixctl_command_reply(conn
, ds_cstr(&ds
));
3831 struct xlate_out xout
;
3832 struct xlate_in xin
;
3833 const struct flow
*key
;
3835 struct flow_wildcards wc
;
3840 trace_format_rule(struct ds
*result
, int level
, const struct rule_dpif
*rule
)
3842 struct rule_actions
*actions
;
3845 ds_put_char_multiple(result
, '\t', level
);
3847 ds_put_cstr(result
, "No match\n");
3851 ovs_mutex_lock(&rule
->up
.mutex
);
3852 cookie
= rule
->up
.flow_cookie
;
3853 ovs_mutex_unlock(&rule
->up
.mutex
);
3855 ds_put_format(result
, "Rule: table=%"PRIu8
" cookie=%#"PRIx64
" ",
3856 rule
? rule
->up
.table_id
: 0, ntohll(cookie
));
3857 cls_rule_format(&rule
->up
.cr
, result
);
3858 ds_put_char(result
, '\n');
3860 actions
= rule_dpif_get_actions(rule
);
3862 ds_put_char_multiple(result
, '\t', level
);
3863 ds_put_cstr(result
, "OpenFlow actions=");
3864 ofpacts_format(actions
->ofpacts
, actions
->ofpacts_len
, result
);
3865 ds_put_char(result
, '\n');
3869 trace_format_flow(struct ds
*result
, int level
, const char *title
,
3870 struct trace_ctx
*trace
)
3872 ds_put_char_multiple(result
, '\t', level
);
3873 ds_put_format(result
, "%s: ", title
);
3874 /* Do not report unchanged flows for resubmits. */
3875 if ((level
> 0 && flow_equal(&trace
->xin
.flow
, &trace
->flow
))
3876 || (level
== 0 && flow_equal(&trace
->xin
.flow
, trace
->key
))) {
3877 ds_put_cstr(result
, "unchanged");
3879 flow_format(result
, &trace
->xin
.flow
);
3880 trace
->flow
= trace
->xin
.flow
;
3882 ds_put_char(result
, '\n');
3886 trace_format_regs(struct ds
*result
, int level
, const char *title
,
3887 struct trace_ctx
*trace
)
3891 ds_put_char_multiple(result
, '\t', level
);
3892 ds_put_format(result
, "%s:", title
);
3893 for (i
= 0; i
< FLOW_N_REGS
; i
++) {
3894 ds_put_format(result
, " reg%"PRIuSIZE
"=0x%"PRIx32
, i
, trace
->flow
.regs
[i
]);
3896 ds_put_char(result
, '\n');
3900 trace_format_odp(struct ds
*result
, int level
, const char *title
,
3901 struct trace_ctx
*trace
)
3903 struct ofpbuf
*odp_actions
= &trace
->xout
.odp_actions
;
3905 ds_put_char_multiple(result
, '\t', level
);
3906 ds_put_format(result
, "%s: ", title
);
3907 format_odp_actions(result
, ofpbuf_data(odp_actions
),
3908 ofpbuf_size(odp_actions
));
3909 ds_put_char(result
, '\n');
3913 trace_format_megaflow(struct ds
*result
, int level
, const char *title
,
3914 struct trace_ctx
*trace
)
3918 ds_put_char_multiple(result
, '\t', level
);
3919 ds_put_format(result
, "%s: ", title
);
3920 flow_wildcards_or(&trace
->wc
, &trace
->xout
.wc
, &trace
->wc
);
3921 match_init(&match
, trace
->key
, &trace
->wc
);
3922 match_format(&match
, result
, OFP_DEFAULT_PRIORITY
);
3923 ds_put_char(result
, '\n');
3927 trace_resubmit(struct xlate_in
*xin
, struct rule_dpif
*rule
, int recurse
)
3929 struct trace_ctx
*trace
= CONTAINER_OF(xin
, struct trace_ctx
, xin
);
3930 struct ds
*result
= trace
->result
;
3932 ds_put_char(result
, '\n');
3933 trace_format_flow(result
, recurse
+ 1, "Resubmitted flow", trace
);
3934 trace_format_regs(result
, recurse
+ 1, "Resubmitted regs", trace
);
3935 trace_format_odp(result
, recurse
+ 1, "Resubmitted odp", trace
);
3936 trace_format_megaflow(result
, recurse
+ 1, "Resubmitted megaflow", trace
);
3937 trace_format_rule(result
, recurse
+ 1, rule
);
3941 trace_report(struct xlate_in
*xin
, const char *s
, int recurse
)
3943 struct trace_ctx
*trace
= CONTAINER_OF(xin
, struct trace_ctx
, xin
);
3944 struct ds
*result
= trace
->result
;
3946 ds_put_char_multiple(result
, '\t', recurse
);
3947 ds_put_cstr(result
, s
);
3948 ds_put_char(result
, '\n');
3951 /* Parses the 'argc' elements of 'argv', ignoring argv[0]. The following
3952 * forms are supported:
3954 * - [dpname] odp_flow [-generate | packet]
3955 * - bridge br_flow [-generate | packet]
3957 * On success, initializes '*ofprotop' and 'flow' and returns NULL. On failure
3958 * returns a nonnull malloced error message. */
3959 static char * WARN_UNUSED_RESULT
3960 parse_flow_and_packet(int argc
, const char *argv
[],
3961 struct ofproto_dpif
**ofprotop
, struct flow
*flow
,
3962 struct ofpbuf
**packetp
)
3964 const struct dpif_backer
*backer
= NULL
;
3965 const char *error
= NULL
;
3967 struct simap port_names
= SIMAP_INITIALIZER(&port_names
);
3968 struct ofpbuf
*packet
;
3969 struct ofpbuf odp_key
;
3970 struct ofpbuf odp_mask
;
3972 ofpbuf_init(&odp_key
, 0);
3973 ofpbuf_init(&odp_mask
, 0);
3975 /* Handle "-generate" or a hex string as the last argument. */
3976 if (!strcmp(argv
[argc
- 1], "-generate")) {
3977 packet
= ofpbuf_new(0);
3980 error
= eth_from_hex(argv
[argc
- 1], &packet
);
3983 } else if (argc
== 4) {
3984 /* The 3-argument form must end in "-generate' or a hex string. */
3990 /* odp_flow can have its in_port specified as a name instead of port no.
3991 * We do not yet know whether a given flow is a odp_flow or a br_flow.
3992 * But, to know whether a flow is odp_flow through odp_flow_from_string(),
3993 * we need to create a simap of name to port no. */
3995 const char *dp_type
;
3996 if (!strncmp(argv
[1], "ovs-", 4)) {
3997 dp_type
= argv
[1] + 4;
4001 backer
= shash_find_data(&all_dpif_backers
, dp_type
);
4002 } else if (argc
== 2) {
4003 struct shash_node
*node
;
4004 if (shash_count(&all_dpif_backers
) == 1) {
4005 node
= shash_first(&all_dpif_backers
);
4006 backer
= node
->data
;
4009 error
= "Syntax error";
4012 if (backer
&& backer
->dpif
) {
4013 struct dpif_port dpif_port
;
4014 struct dpif_port_dump port_dump
;
4015 DPIF_PORT_FOR_EACH (&dpif_port
, &port_dump
, backer
->dpif
) {
4016 simap_put(&port_names
, dpif_port
.name
,
4017 odp_to_u32(dpif_port
.port_no
));
4021 /* Parse the flow and determine whether a datapath or
4022 * bridge is specified. If function odp_flow_key_from_string()
4023 * returns 0, the flow is a odp_flow. If function
4024 * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
4025 if (!odp_flow_from_string(argv
[argc
- 1], &port_names
,
4026 &odp_key
, &odp_mask
)) {
4028 error
= "Cannot find the datapath";
4032 if (xlate_receive(backer
, NULL
, ofpbuf_data(&odp_key
),
4033 ofpbuf_size(&odp_key
), flow
,
4034 ofprotop
, NULL
, NULL
, NULL
, NULL
)) {
4035 error
= "Invalid datapath flow";
4039 char *err
= parse_ofp_exact_flow(flow
, NULL
, argv
[argc
- 1], NULL
);
4042 m_err
= xasprintf("Bad flow syntax: %s", err
);
4047 error
= "Must specify bridge name";
4051 *ofprotop
= ofproto_dpif_lookup(argv
[1]);
4053 error
= "Unknown bridge name";
4059 /* Generate a packet, if requested. */
4061 if (!ofpbuf_size(packet
)) {
4062 flow_compose(packet
, flow
);
4064 struct pkt_metadata md
= pkt_metadata_from_flow(flow
);
4066 /* Use the metadata from the flow and the packet argument
4067 * to reconstruct the flow. */
4068 flow_extract(packet
, &md
, flow
);
4073 if (error
&& !m_err
) {
4074 m_err
= xstrdup(error
);
4077 ofpbuf_delete(packet
);
4081 ofpbuf_uninit(&odp_key
);
4082 ofpbuf_uninit(&odp_mask
);
4083 simap_destroy(&port_names
);
4088 ofproto_unixctl_trace(struct unixctl_conn
*conn
, int argc
, const char *argv
[],
4089 void *aux OVS_UNUSED
)
4091 struct ofproto_dpif
*ofproto
;
4092 struct ofpbuf
*packet
;
4096 error
= parse_flow_and_packet(argc
, argv
, &ofproto
, &flow
, &packet
);
4101 ofproto_trace(ofproto
, &flow
, packet
, NULL
, 0, &result
);
4102 unixctl_command_reply(conn
, ds_cstr(&result
));
4103 ds_destroy(&result
);
4104 ofpbuf_delete(packet
);
4106 unixctl_command_reply_error(conn
, error
);
4112 ofproto_unixctl_trace_actions(struct unixctl_conn
*conn
, int argc
,
4113 const char *argv
[], void *aux OVS_UNUSED
)
4115 enum ofputil_protocol usable_protocols
;
4116 struct ofproto_dpif
*ofproto
;
4117 bool enforce_consistency
;
4118 struct ofpbuf ofpacts
;
4119 struct ofpbuf
*packet
;
4124 /* Three kinds of error return values! */
4130 ofpbuf_init(&ofpacts
, 0);
4132 /* Parse actions. */
4133 error
= parse_ofpacts(argv
[--argc
], &ofpacts
, &usable_protocols
);
4135 unixctl_command_reply_error(conn
, error
);
4140 /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
4141 * consistency between the flow and the actions. With -consistent, we
4142 * enforce consistency even for a flow supported in OpenFlow 1.0. */
4143 if (!strcmp(argv
[1], "-consistent")) {
4144 enforce_consistency
= true;
4148 enforce_consistency
= false;
4151 error
= parse_flow_and_packet(argc
, argv
, &ofproto
, &flow
, &packet
);
4153 unixctl_command_reply_error(conn
, error
);
4158 /* Do the same checks as handle_packet_out() in ofproto.c.
4160 * We pass a 'table_id' of 0 to ofproto_check_ofpacts(), which isn't
4161 * strictly correct because these actions aren't in any table, but it's OK
4162 * because it 'table_id' is used only to check goto_table instructions, but
4163 * packet-outs take a list of actions and therefore it can't include
4166 * We skip the "meter" check here because meter is an instruction, not an
4167 * action, and thus cannot appear in ofpacts. */
4168 in_port
= ofp_to_u16(flow
.in_port
.ofp_port
);
4169 if (in_port
>= ofproto
->up
.max_ports
&& in_port
< ofp_to_u16(OFPP_MAX
)) {
4170 unixctl_command_reply_error(conn
, "invalid in_port");
4173 if (enforce_consistency
) {
4174 retval
= ofpacts_check_consistency(ofpbuf_data(&ofpacts
), ofpbuf_size(&ofpacts
),
4175 &flow
, u16_to_ofp(ofproto
->up
.max_ports
),
4176 0, 0, usable_protocols
);
4178 retval
= ofpacts_check(ofpbuf_data(&ofpacts
), ofpbuf_size(&ofpacts
), &flow
,
4179 u16_to_ofp(ofproto
->up
.max_ports
), 0, 0,
4185 ds_put_format(&result
, "Bad actions: %s", ofperr_to_string(retval
));
4186 unixctl_command_reply_error(conn
, ds_cstr(&result
));
4190 ofproto_trace(ofproto
, &flow
, packet
,
4191 ofpbuf_data(&ofpacts
), ofpbuf_size(&ofpacts
), &result
);
4192 unixctl_command_reply(conn
, ds_cstr(&result
));
4195 ds_destroy(&result
);
4196 ofpbuf_delete(packet
);
4197 ofpbuf_uninit(&ofpacts
);
4200 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
4201 * description of the results to 'ds'.
4203 * The trace follows a packet with the specified 'flow' through the flow
4204 * table. 'packet' may be nonnull to trace an actual packet, with consequent
4205 * side effects (if it is nonnull then its flow must be 'flow').
4207 * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
4208 * trace, otherwise the actions are determined by a flow table lookup. */
4210 ofproto_trace(struct ofproto_dpif
*ofproto
, struct flow
*flow
,
4211 const struct ofpbuf
*packet
,
4212 const struct ofpact ofpacts
[], size_t ofpacts_len
,
4215 struct rule_dpif
*rule
;
4216 struct trace_ctx trace
;
4218 ds_put_format(ds
, "Bridge: %s\n", ofproto
->up
.name
);
4219 ds_put_cstr(ds
, "Flow: ");
4220 flow_format(ds
, flow
);
4221 ds_put_char(ds
, '\n');
4223 flow_wildcards_init_catchall(&trace
.wc
);
4227 rule_dpif_lookup(ofproto
, flow
, &trace
.wc
, &rule
, false);
4229 trace_format_rule(ds
, 0, rule
);
4230 if (rule
== ofproto
->miss_rule
) {
4231 ds_put_cstr(ds
, "\nNo match, flow generates \"packet in\"s.\n");
4232 } else if (rule
== ofproto
->no_packet_in_rule
) {
4233 ds_put_cstr(ds
, "\nNo match, packets dropped because "
4234 "OFPPC_NO_PACKET_IN is set on in_port.\n");
4235 } else if (rule
== ofproto
->drop_frags_rule
) {
4236 ds_put_cstr(ds
, "\nPackets dropped because they are IP fragments "
4237 "and the fragment handling mode is \"drop\".\n");
4241 if (rule
|| ofpacts
) {
4243 trace
.key
= flow
; /* Original flow key, used for megaflow. */
4244 trace
.flow
= *flow
; /* May be modified by actions. */
4245 xlate_in_init(&trace
.xin
, ofproto
, flow
, rule
, ntohs(flow
->tcp_flags
),
4248 trace
.xin
.ofpacts
= ofpacts
;
4249 trace
.xin
.ofpacts_len
= ofpacts_len
;
4251 trace
.xin
.resubmit_hook
= trace_resubmit
;
4252 trace
.xin
.report_hook
= trace_report
;
4254 xlate_actions(&trace
.xin
, &trace
.xout
);
4256 ds_put_char(ds
, '\n');
4257 trace_format_flow(ds
, 0, "Final flow", &trace
);
4258 trace_format_megaflow(ds
, 0, "Megaflow", &trace
);
4260 ds_put_cstr(ds
, "Datapath actions: ");
4261 format_odp_actions(ds
, ofpbuf_data(&trace
.xout
.odp_actions
),
4262 ofpbuf_size(&trace
.xout
.odp_actions
));
4264 if (trace
.xout
.slow
) {
4265 enum slow_path_reason slow
;
4267 ds_put_cstr(ds
, "\nThis flow is handled by the userspace "
4268 "slow path because it:");
4270 slow
= trace
.xout
.slow
;
4272 enum slow_path_reason bit
= rightmost_1bit(slow
);
4274 ds_put_format(ds
, "\n\t- %s.",
4275 slow_path_reason_to_explanation(bit
));
4281 xlate_out_uninit(&trace
.xout
);
4285 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
4286 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
4287 * to destroy 'ofproto_shash' and free the returned value. */
4288 static const struct shash_node
**
4289 get_ofprotos(struct shash
*ofproto_shash
)
4291 const struct ofproto_dpif
*ofproto
;
4293 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
4294 char *name
= xasprintf("%s@%s", ofproto
->up
.type
, ofproto
->up
.name
);
4295 shash_add_nocopy(ofproto_shash
, name
, ofproto
);
4298 return shash_sort(ofproto_shash
);
4302 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn
*conn
, int argc OVS_UNUSED
,
4303 const char *argv
[] OVS_UNUSED
,
4304 void *aux OVS_UNUSED
)
4306 struct ds ds
= DS_EMPTY_INITIALIZER
;
4307 struct shash ofproto_shash
;
4308 const struct shash_node
**sorted_ofprotos
;
4311 shash_init(&ofproto_shash
);
4312 sorted_ofprotos
= get_ofprotos(&ofproto_shash
);
4313 for (i
= 0; i
< shash_count(&ofproto_shash
); i
++) {
4314 const struct shash_node
*node
= sorted_ofprotos
[i
];
4315 ds_put_format(&ds
, "%s\n", node
->name
);
4318 shash_destroy(&ofproto_shash
);
4319 free(sorted_ofprotos
);
4321 unixctl_command_reply(conn
, ds_cstr(&ds
));
4326 dpif_show_backer(const struct dpif_backer
*backer
, struct ds
*ds
)
4328 const struct shash_node
**ofprotos
;
4329 struct dpif_dp_stats dp_stats
;
4330 struct shash ofproto_shash
;
4333 dpif_get_dp_stats(backer
->dpif
, &dp_stats
);
4335 ds_put_format(ds
, "%s: hit:%"PRIu64
" missed:%"PRIu64
"\n",
4336 dpif_name(backer
->dpif
), dp_stats
.n_hit
, dp_stats
.n_missed
);
4338 shash_init(&ofproto_shash
);
4339 ofprotos
= get_ofprotos(&ofproto_shash
);
4340 for (i
= 0; i
< shash_count(&ofproto_shash
); i
++) {
4341 struct ofproto_dpif
*ofproto
= ofprotos
[i
]->data
;
4342 const struct shash_node
**ports
;
4345 if (ofproto
->backer
!= backer
) {
4349 ds_put_format(ds
, "\t%s:\n", ofproto
->up
.name
);
4351 ports
= shash_sort(&ofproto
->up
.port_by_name
);
4352 for (j
= 0; j
< shash_count(&ofproto
->up
.port_by_name
); j
++) {
4353 const struct shash_node
*node
= ports
[j
];
4354 struct ofport
*ofport
= node
->data
;
4356 odp_port_t odp_port
;
4358 ds_put_format(ds
, "\t\t%s %u/", netdev_get_name(ofport
->netdev
),
4361 odp_port
= ofp_port_to_odp_port(ofproto
, ofport
->ofp_port
);
4362 if (odp_port
!= ODPP_NONE
) {
4363 ds_put_format(ds
, "%"PRIu32
":", odp_port
);
4365 ds_put_cstr(ds
, "none:");
4368 ds_put_format(ds
, " (%s", netdev_get_type(ofport
->netdev
));
4371 if (!netdev_get_config(ofport
->netdev
, &config
)) {
4372 const struct smap_node
**nodes
;
4375 nodes
= smap_sort(&config
);
4376 for (i
= 0; i
< smap_count(&config
); i
++) {
4377 const struct smap_node
*node
= nodes
[i
];
4378 ds_put_format(ds
, "%c %s=%s", i
? ',' : ':',
4379 node
->key
, node
->value
);
4383 smap_destroy(&config
);
4385 ds_put_char(ds
, ')');
4386 ds_put_char(ds
, '\n');
4390 shash_destroy(&ofproto_shash
);
4395 ofproto_unixctl_dpif_show(struct unixctl_conn
*conn
, int argc OVS_UNUSED
,
4396 const char *argv
[] OVS_UNUSED
, void *aux OVS_UNUSED
)
4398 struct ds ds
= DS_EMPTY_INITIALIZER
;
4399 const struct shash_node
**backers
;
4402 backers
= shash_sort(&all_dpif_backers
);
4403 for (i
= 0; i
< shash_count(&all_dpif_backers
); i
++) {
4404 dpif_show_backer(backers
[i
]->data
, &ds
);
4408 unixctl_command_reply(conn
, ds_cstr(&ds
));
4413 ofproto_dpif_contains_flow(const struct ofproto_dpif
*ofproto
,
4414 const struct nlattr
*key
, size_t key_len
)
4416 struct ofproto_dpif
*ofp
;
4419 xlate_receive(ofproto
->backer
, NULL
, key
, key_len
, &flow
, &ofp
,
4420 NULL
, NULL
, NULL
, NULL
);
4421 return ofp
== ofproto
;
4425 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn
*conn
,
4426 int argc OVS_UNUSED
, const char *argv
[],
4427 void *aux OVS_UNUSED
)
4429 struct ds ds
= DS_EMPTY_INITIALIZER
;
4430 const struct dpif_flow_stats
*stats
;
4431 const struct ofproto_dpif
*ofproto
;
4432 struct dpif_flow_dump flow_dump
;
4433 const struct nlattr
*actions
;
4434 const struct nlattr
*mask
;
4435 const struct nlattr
*key
;
4439 bool verbosity
= false;
4440 struct dpif_port dpif_port
;
4441 struct dpif_port_dump port_dump
;
4442 struct hmap portno_names
;
4446 ofproto
= ofproto_dpif_lookup(argv
[argc
- 1]);
4448 unixctl_command_reply_error(conn
, "no such bridge");
4452 if (argc
> 2 && !strcmp(argv
[1], "-m")) {
4456 hmap_init(&portno_names
);
4457 DPIF_PORT_FOR_EACH (&dpif_port
, &port_dump
, ofproto
->backer
->dpif
) {
4458 odp_portno_names_set(&portno_names
, dpif_port
.port_no
, dpif_port
.name
);
4462 error
= dpif_flow_dump_start(&flow_dump
, ofproto
->backer
->dpif
);
4466 dpif_flow_dump_state_init(ofproto
->backer
->dpif
, &state
);
4467 while (dpif_flow_dump_next(&flow_dump
, state
, &key
, &key_len
,
4468 &mask
, &mask_len
, &actions
, &actions_len
,
4470 if (!ofproto_dpif_contains_flow(ofproto
, key
, key_len
)) {
4474 odp_flow_format(key
, key_len
, mask
, mask_len
, &portno_names
, &ds
,
4476 ds_put_cstr(&ds
, ", ");
4477 dpif_flow_stats_format(stats
, &ds
);
4478 ds_put_cstr(&ds
, ", actions:");
4479 format_odp_actions(&ds
, actions
, actions_len
);
4480 ds_put_char(&ds
, '\n');
4482 dpif_flow_dump_state_uninit(ofproto
->backer
->dpif
, state
);
4483 error
= dpif_flow_dump_done(&flow_dump
);
4488 ds_put_format(&ds
, "dpif/dump_flows failed: %s", ovs_strerror(errno
));
4489 unixctl_command_reply_error(conn
, ds_cstr(&ds
));
4491 unixctl_command_reply(conn
, ds_cstr(&ds
));
4493 odp_portno_names_destroy(&portno_names
);
4494 hmap_destroy(&portno_names
);
4499 ofproto_dpif_unixctl_init(void)
4501 static bool registered
;
4507 unixctl_command_register(
4509 "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
4510 1, 3, ofproto_unixctl_trace
, NULL
);
4511 unixctl_command_register(
4512 "ofproto/trace-packet-out",
4513 "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
4514 2, 6, ofproto_unixctl_trace_actions
, NULL
);
4515 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
4516 ofproto_unixctl_fdb_flush
, NULL
);
4517 unixctl_command_register("fdb/show", "bridge", 1, 1,
4518 ofproto_unixctl_fdb_show
, NULL
);
4519 unixctl_command_register("dpif/dump-dps", "", 0, 0,
4520 ofproto_unixctl_dpif_dump_dps
, NULL
);
4521 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show
,
4523 unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
4524 ofproto_unixctl_dpif_dump_flows
, NULL
);
4528 /* Returns true if 'rule' is an internal rule, false otherwise. */
4530 rule_is_internal(const struct rule
*rule
)
4532 return rule
->table_id
== TBL_INTERNAL
;
4535 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4537 * This is deprecated. It is only for compatibility with broken device drivers
4538 * in old versions of Linux that do not properly support VLANs when VLAN
4539 * devices are not used. When broken device drivers are no longer in
4540 * widespread use, we will delete these interfaces. */
4543 set_realdev(struct ofport
*ofport_
, ofp_port_t realdev_ofp_port
, int vid
)
4545 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport_
->ofproto
);
4546 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
4548 if (realdev_ofp_port
== ofport
->realdev_ofp_port
4549 && vid
== ofport
->vlandev_vid
) {
4553 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
4555 if (ofport
->realdev_ofp_port
) {
4558 if (realdev_ofp_port
&& ofport
->bundle
) {
4559 /* vlandevs are enslaved to their realdevs, so they are not allowed to
4560 * themselves be part of a bundle. */
4561 bundle_set(ofport_
->ofproto
, ofport
->bundle
, NULL
);
4564 ofport
->realdev_ofp_port
= realdev_ofp_port
;
4565 ofport
->vlandev_vid
= vid
;
4567 if (realdev_ofp_port
) {
4568 vsp_add(ofport
, realdev_ofp_port
, vid
);
4575 hash_realdev_vid(ofp_port_t realdev_ofp_port
, int vid
)
4577 return hash_2words(ofp_to_u16(realdev_ofp_port
), vid
);
4581 ofproto_has_vlan_splinters(const struct ofproto_dpif
*ofproto
)
4582 OVS_EXCLUDED(ofproto
->vsp_mutex
)
4584 /* hmap_is_empty is thread safe. */
4585 return !hmap_is_empty(&ofproto
->realdev_vid_map
);
4590 vsp_realdev_to_vlandev__(const struct ofproto_dpif
*ofproto
,
4591 ofp_port_t realdev_ofp_port
, ovs_be16 vlan_tci
)
4592 OVS_REQUIRES(ofproto
->vsp_mutex
)
4594 if (!hmap_is_empty(&ofproto
->realdev_vid_map
)) {
4595 int vid
= vlan_tci_to_vid(vlan_tci
);
4596 const struct vlan_splinter
*vsp
;
4598 HMAP_FOR_EACH_WITH_HASH (vsp
, realdev_vid_node
,
4599 hash_realdev_vid(realdev_ofp_port
, vid
),
4600 &ofproto
->realdev_vid_map
) {
4601 if (vsp
->realdev_ofp_port
== realdev_ofp_port
4602 && vsp
->vid
== vid
) {
4603 return vsp
->vlandev_ofp_port
;
4607 return realdev_ofp_port
;
4610 /* Returns the OFP port number of the Linux VLAN device that corresponds to
4611 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
4612 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
4613 * 'vlan_tci' 9, it would return the port number of eth0.9.
4615 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
4616 * function just returns its 'realdev_ofp_port' argument. */
4618 vsp_realdev_to_vlandev(const struct ofproto_dpif
*ofproto
,
4619 ofp_port_t realdev_ofp_port
, ovs_be16 vlan_tci
)
4620 OVS_EXCLUDED(ofproto
->vsp_mutex
)
4624 /* hmap_is_empty is thread safe, see if we can return immediately. */
4625 if (hmap_is_empty(&ofproto
->realdev_vid_map
)) {
4626 return realdev_ofp_port
;
4628 ovs_mutex_lock(&ofproto
->vsp_mutex
);
4629 ret
= vsp_realdev_to_vlandev__(ofproto
, realdev_ofp_port
, vlan_tci
);
4630 ovs_mutex_unlock(&ofproto
->vsp_mutex
);
4634 static struct vlan_splinter
*
4635 vlandev_find(const struct ofproto_dpif
*ofproto
, ofp_port_t vlandev_ofp_port
)
4637 struct vlan_splinter
*vsp
;
4639 HMAP_FOR_EACH_WITH_HASH (vsp
, vlandev_node
,
4640 hash_ofp_port(vlandev_ofp_port
),
4641 &ofproto
->vlandev_map
) {
4642 if (vsp
->vlandev_ofp_port
== vlandev_ofp_port
) {
4650 /* Returns the OpenFlow port number of the "real" device underlying the Linux
4651 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
4652 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
4653 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
4654 * eth0 and store 9 in '*vid'.
4656 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
4657 * VLAN device. Unless VLAN splinters are enabled, this is what this function
4660 vsp_vlandev_to_realdev(const struct ofproto_dpif
*ofproto
,
4661 ofp_port_t vlandev_ofp_port
, int *vid
)
4662 OVS_REQUIRES(ofproto
->vsp_mutex
)
4664 if (!hmap_is_empty(&ofproto
->vlandev_map
)) {
4665 const struct vlan_splinter
*vsp
;
4667 vsp
= vlandev_find(ofproto
, vlandev_ofp_port
);
4672 return vsp
->realdev_ofp_port
;
4678 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
4679 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
4680 * 'flow->in_port' to the "real" device backing the VLAN device, sets
4681 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
4682 * always the case unless VLAN splinters are enabled), returns false without
4683 * making any changes. */
4685 vsp_adjust_flow(const struct ofproto_dpif
*ofproto
, struct flow
*flow
)
4686 OVS_EXCLUDED(ofproto
->vsp_mutex
)
4691 /* hmap_is_empty is thread safe. */
4692 if (hmap_is_empty(&ofproto
->vlandev_map
)) {
4696 ovs_mutex_lock(&ofproto
->vsp_mutex
);
4697 realdev
= vsp_vlandev_to_realdev(ofproto
, flow
->in_port
.ofp_port
, &vid
);
4698 ovs_mutex_unlock(&ofproto
->vsp_mutex
);
4703 /* Cause the flow to be processed as if it came in on the real device with
4704 * the VLAN device's VLAN ID. */
4705 flow
->in_port
.ofp_port
= realdev
;
4706 flow
->vlan_tci
= htons((vid
& VLAN_VID_MASK
) | VLAN_CFI
);
4711 vsp_remove(struct ofport_dpif
*port
)
4713 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
4714 struct vlan_splinter
*vsp
;
4716 ovs_mutex_lock(&ofproto
->vsp_mutex
);
4717 vsp
= vlandev_find(ofproto
, port
->up
.ofp_port
);
4719 hmap_remove(&ofproto
->vlandev_map
, &vsp
->vlandev_node
);
4720 hmap_remove(&ofproto
->realdev_vid_map
, &vsp
->realdev_vid_node
);
4723 port
->realdev_ofp_port
= 0;
4725 VLOG_ERR("missing vlan device record");
4727 ovs_mutex_unlock(&ofproto
->vsp_mutex
);
4731 vsp_add(struct ofport_dpif
*port
, ofp_port_t realdev_ofp_port
, int vid
)
4733 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
4735 ovs_mutex_lock(&ofproto
->vsp_mutex
);
4736 if (!vsp_vlandev_to_realdev(ofproto
, port
->up
.ofp_port
, NULL
)
4737 && (vsp_realdev_to_vlandev__(ofproto
, realdev_ofp_port
, htons(vid
))
4738 == realdev_ofp_port
)) {
4739 struct vlan_splinter
*vsp
;
4741 vsp
= xmalloc(sizeof *vsp
);
4742 vsp
->realdev_ofp_port
= realdev_ofp_port
;
4743 vsp
->vlandev_ofp_port
= port
->up
.ofp_port
;
4746 port
->realdev_ofp_port
= realdev_ofp_port
;
4748 hmap_insert(&ofproto
->vlandev_map
, &vsp
->vlandev_node
,
4749 hash_ofp_port(port
->up
.ofp_port
));
4750 hmap_insert(&ofproto
->realdev_vid_map
, &vsp
->realdev_vid_node
,
4751 hash_realdev_vid(realdev_ofp_port
, vid
));
4753 VLOG_ERR("duplicate vlan device record");
4755 ovs_mutex_unlock(&ofproto
->vsp_mutex
);
4759 ofp_port_to_odp_port(const struct ofproto_dpif
*ofproto
, ofp_port_t ofp_port
)
4761 const struct ofport_dpif
*ofport
= get_ofp_port(ofproto
, ofp_port
);
4762 return ofport
? ofport
->odp_port
: ODPP_NONE
;
4765 struct ofport_dpif
*
4766 odp_port_to_ofport(const struct dpif_backer
*backer
, odp_port_t odp_port
)
4768 struct ofport_dpif
*port
;
4770 ovs_rwlock_rdlock(&backer
->odp_to_ofport_lock
);
4771 HMAP_FOR_EACH_IN_BUCKET (port
, odp_port_node
, hash_odp_port(odp_port
),
4772 &backer
->odp_to_ofport_map
) {
4773 if (port
->odp_port
== odp_port
) {
4774 ovs_rwlock_unlock(&backer
->odp_to_ofport_lock
);
4779 ovs_rwlock_unlock(&backer
->odp_to_ofport_lock
);
4784 odp_port_to_ofp_port(const struct ofproto_dpif
*ofproto
, odp_port_t odp_port
)
4786 struct ofport_dpif
*port
;
4788 port
= odp_port_to_ofport(ofproto
->backer
, odp_port
);
4789 if (port
&& &ofproto
->up
== port
->up
.ofproto
) {
4790 return port
->up
.ofp_port
;
4797 ofproto_dpif_alloc_recirc_id(struct ofproto_dpif
*ofproto
)
4799 struct dpif_backer
*backer
= ofproto
->backer
;
4801 return recirc_id_alloc(backer
->rid_pool
);
4805 ofproto_dpif_free_recirc_id(struct ofproto_dpif
*ofproto
, uint32_t recirc_id
)
4807 struct dpif_backer
*backer
= ofproto
->backer
;
4809 recirc_id_free(backer
->rid_pool
, recirc_id
);
4813 ofproto_dpif_add_internal_flow(struct ofproto_dpif
*ofproto
,
4814 struct match
*match
, int priority
,
4815 const struct ofpbuf
*ofpacts
,
4816 struct rule
**rulep
)
4818 struct ofputil_flow_mod fm
;
4819 struct rule_dpif
*rule
;
4823 fm
.priority
= priority
;
4824 fm
.new_cookie
= htonll(0);
4825 fm
.cookie
= htonll(0);
4826 fm
.cookie_mask
= htonll(0);
4827 fm
.modify_cookie
= false;
4828 fm
.table_id
= TBL_INTERNAL
;
4829 fm
.command
= OFPFC_ADD
;
4830 fm
.idle_timeout
= 0;
4831 fm
.hard_timeout
= 0;
4834 fm
.flags
= OFPUTIL_FF_HIDDEN_FIELDS
| OFPUTIL_FF_NO_READONLY
;
4835 fm
.ofpacts
= ofpbuf_data(ofpacts
);
4836 fm
.ofpacts_len
= ofpbuf_size(ofpacts
);
4838 error
= ofproto_flow_mod(&ofproto
->up
, &fm
);
4840 VLOG_ERR_RL(&rl
, "failed to add internal flow (%s)",
4841 ofperr_to_string(error
));
4846 rule
= rule_dpif_lookup_in_table(ofproto
, TBL_INTERNAL
, &match
->flow
,
4857 ofproto_dpif_delete_internal_flow(struct ofproto_dpif
*ofproto
,
4858 struct match
*match
, int priority
)
4860 struct ofputil_flow_mod fm
;
4864 fm
.priority
= priority
;
4865 fm
.new_cookie
= htonll(0);
4866 fm
.cookie
= htonll(0);
4867 fm
.cookie_mask
= htonll(0);
4868 fm
.modify_cookie
= false;
4869 fm
.table_id
= TBL_INTERNAL
;
4870 fm
.flags
= OFPUTIL_FF_HIDDEN_FIELDS
| OFPUTIL_FF_NO_READONLY
;
4871 fm
.command
= OFPFC_DELETE_STRICT
;
4873 error
= ofproto_flow_mod(&ofproto
->up
, &fm
);
4875 VLOG_ERR_RL(&rl
, "failed to delete internal flow (%s)",
4876 ofperr_to_string(error
));
4883 const struct ofproto_class ofproto_dpif_class
= {
4897 NULL
, /* get_memory_usage. */
4898 type_get_memory_usage
,
4917 port_is_lacp_current
,
4918 NULL
, /* rule_choose_table */
4927 rule_modify_actions
,
4941 get_stp_port_status
,
4949 is_mirror_output_bundle
,
4950 forward_bpdu_changed
,
4951 set_mac_table_config
,
4953 NULL
, /* meter_get_features */
4954 NULL
, /* meter_set */
4955 NULL
, /* meter_get */
4956 NULL
, /* meter_del */
4957 group_alloc
, /* group_alloc */
4958 group_construct
, /* group_construct */
4959 group_destruct
, /* group_destruct */
4960 group_dealloc
, /* group_dealloc */
4961 group_modify
, /* group_modify */
4962 group_get_stats
, /* group_get_stats */