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 /* No bfd/cfm status change. */
77 #define NO_STATUS_CHANGE -1
86 * - Do include packets and bytes from datapath flows which have not
87 * recently been processed by a revalidator. */
88 struct ovs_mutex stats_mutex
;
89 struct dpif_flow_stats stats OVS_GUARDED
;
92 /* RULE_CAST() depends on this. */
93 BUILD_ASSERT_DECL(offsetof(struct rule_dpif
, up
) == 0);
95 static void rule_get_stats(struct rule
*, uint64_t *packets
, uint64_t *bytes
,
97 static struct rule_dpif
*rule_dpif_cast(const struct rule
*);
98 static void rule_expire(struct rule_dpif
*);
105 * - Do include packets and bytes from datapath flows which have not
106 * recently been processed by a revalidator. */
107 struct ovs_mutex stats_mutex
;
108 uint64_t packet_count OVS_GUARDED
; /* Number of packets received. */
109 uint64_t byte_count OVS_GUARDED
; /* Number of bytes received. */
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. */
317 struct seq
*pins_seq
; /* For notifying 'pins' reception. */
321 /* All existing ofproto_dpif instances, indexed by ->up.name. */
322 static struct hmap all_ofproto_dpifs
= HMAP_INITIALIZER(&all_ofproto_dpifs
);
324 static void ofproto_dpif_unixctl_init(void);
326 static inline struct ofproto_dpif
*
327 ofproto_dpif_cast(const struct ofproto
*ofproto
)
329 ovs_assert(ofproto
->ofproto_class
== &ofproto_dpif_class
);
330 return CONTAINER_OF(ofproto
, struct ofproto_dpif
, up
);
334 ofproto_dpif_get_max_mpls_depth(const struct ofproto_dpif
*ofproto
)
336 return ofproto
->backer
->max_mpls_depth
;
340 ofproto_dpif_get_enable_recirc(const struct ofproto_dpif
*ofproto
)
342 return ofproto
->backer
->enable_recirc
;
345 static struct ofport_dpif
*get_ofp_port(const struct ofproto_dpif
*ofproto
,
346 ofp_port_t ofp_port
);
347 static void ofproto_trace(struct ofproto_dpif
*, struct flow
*,
348 const struct ofpbuf
*packet
,
349 const struct ofpact
[], size_t ofpacts_len
,
352 /* Global variables. */
353 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
355 /* Initial mappings of port to bridge mappings. */
356 static struct shash init_ofp_ports
= SHASH_INITIALIZER(&init_ofp_ports
);
358 /* Executes 'fm'. The caller retains ownership of 'fm' and everything in
361 ofproto_dpif_flow_mod(struct ofproto_dpif
*ofproto
,
362 struct ofputil_flow_mod
*fm
)
364 ofproto_flow_mod(&ofproto
->up
, fm
);
367 /* Resets the modified time for 'rule' or an equivalent rule. If 'rule' is not
368 * in the classifier, but an equivalent rule is, unref 'rule' and ref the new
369 * rule. Otherwise if 'rule' is no longer installed in the classifier,
372 * Returns the rule whose modified time has been reset. */
374 ofproto_dpif_refresh_rule(struct rule_dpif
*rule
)
376 return rule_dpif_cast(ofproto_refresh_rule(&rule
->up
));
379 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
380 * Takes ownership of 'pin' and pin->packet. */
382 ofproto_dpif_send_packet_in(struct ofproto_dpif
*ofproto
,
383 struct ofproto_packet_in
*pin
)
385 if (!guarded_list_push_back(&ofproto
->pins
, &pin
->list_node
, 1024)) {
386 COVERAGE_INC(packet_in_overflow
);
387 free(CONST_CAST(void *, pin
->up
.packet
));
391 /* Wakes up main thread for packet-in I/O. */
392 seq_change(ofproto
->pins_seq
);
395 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
396 * packet rather than to send the packet to the controller.
398 * This function returns false to indicate that a packet_in message
399 * for a "table-miss" should be sent to at least one controller.
400 * False otherwise. */
402 ofproto_dpif_wants_packet_in_on_miss(struct ofproto_dpif
*ofproto
)
404 return connmgr_wants_packet_in_on_miss(ofproto
->up
.connmgr
);
407 /* Factory functions. */
410 init(const struct shash
*iface_hints
)
412 struct shash_node
*node
;
414 /* Make a local copy, since we don't own 'iface_hints' elements. */
415 SHASH_FOR_EACH(node
, iface_hints
) {
416 const struct iface_hint
*orig_hint
= node
->data
;
417 struct iface_hint
*new_hint
= xmalloc(sizeof *new_hint
);
419 new_hint
->br_name
= xstrdup(orig_hint
->br_name
);
420 new_hint
->br_type
= xstrdup(orig_hint
->br_type
);
421 new_hint
->ofp_port
= orig_hint
->ofp_port
;
423 shash_add(&init_ofp_ports
, node
->name
, new_hint
);
428 enumerate_types(struct sset
*types
)
430 dp_enumerate_types(types
);
434 enumerate_names(const char *type
, struct sset
*names
)
436 struct ofproto_dpif
*ofproto
;
439 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
440 if (strcmp(type
, ofproto
->up
.type
)) {
443 sset_add(names
, ofproto
->up
.name
);
450 del(const char *type
, const char *name
)
455 error
= dpif_open(name
, type
, &dpif
);
457 error
= dpif_delete(dpif
);
464 port_open_type(const char *datapath_type
, const char *port_type
)
466 return dpif_port_open_type(datapath_type
, port_type
);
469 /* Type functions. */
471 static void process_dpif_port_changes(struct dpif_backer
*);
472 static void process_dpif_all_ports_changed(struct dpif_backer
*);
473 static void process_dpif_port_change(struct dpif_backer
*,
474 const char *devname
);
475 static void process_dpif_port_error(struct dpif_backer
*, int error
);
477 static struct ofproto_dpif
*
478 lookup_ofproto_dpif_by_port_name(const char *name
)
480 struct ofproto_dpif
*ofproto
;
482 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
483 if (sset_contains(&ofproto
->ports
, name
)) {
492 type_run(const char *type
)
494 struct dpif_backer
*backer
;
496 backer
= shash_find_data(&all_dpif_backers
, type
);
498 /* This is not necessarily a problem, since backers are only
499 * created on demand. */
503 dpif_run(backer
->dpif
);
505 /* If vswitchd started with other_config:flow_restore_wait set as "true",
506 * and the configuration has now changed to "false", enable receiving
507 * packets from the datapath. */
508 if (!backer
->recv_set_enable
&& !ofproto_get_flow_restore_wait()) {
511 backer
->recv_set_enable
= true;
513 error
= dpif_recv_set(backer
->dpif
, backer
->recv_set_enable
);
515 VLOG_ERR("Failed to enable receiving packets in dpif.");
518 dpif_flow_flush(backer
->dpif
);
519 backer
->need_revalidate
= REV_RECONFIGURE
;
522 if (backer
->recv_set_enable
) {
523 udpif_set_threads(backer
->udpif
, n_handlers
, n_revalidators
);
526 if (backer
->need_revalidate
) {
527 struct ofproto_dpif
*ofproto
;
528 struct simap_node
*node
;
529 struct simap tmp_backers
;
531 /* Handle tunnel garbage collection. */
532 simap_init(&tmp_backers
);
533 simap_swap(&backer
->tnl_backers
, &tmp_backers
);
535 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
536 struct ofport_dpif
*iter
;
538 if (backer
!= ofproto
->backer
) {
542 HMAP_FOR_EACH (iter
, up
.hmap_node
, &ofproto
->up
.ports
) {
543 char namebuf
[NETDEV_VPORT_NAME_BUFSIZE
];
546 if (!iter
->is_tunnel
) {
550 dp_port
= netdev_vport_get_dpif_port(iter
->up
.netdev
,
551 namebuf
, sizeof namebuf
);
552 node
= simap_find(&tmp_backers
, dp_port
);
554 simap_put(&backer
->tnl_backers
, dp_port
, node
->data
);
555 simap_delete(&tmp_backers
, node
);
556 node
= simap_find(&backer
->tnl_backers
, dp_port
);
558 node
= simap_find(&backer
->tnl_backers
, dp_port
);
560 odp_port_t odp_port
= ODPP_NONE
;
562 if (!dpif_port_add(backer
->dpif
, iter
->up
.netdev
,
564 simap_put(&backer
->tnl_backers
, dp_port
,
565 odp_to_u32(odp_port
));
566 node
= simap_find(&backer
->tnl_backers
, dp_port
);
571 iter
->odp_port
= node
? u32_to_odp(node
->data
) : ODPP_NONE
;
572 if (tnl_port_reconfigure(iter
, iter
->up
.netdev
,
574 backer
->need_revalidate
= REV_RECONFIGURE
;
579 SIMAP_FOR_EACH (node
, &tmp_backers
) {
580 dpif_port_del(backer
->dpif
, u32_to_odp(node
->data
));
582 simap_destroy(&tmp_backers
);
584 switch (backer
->need_revalidate
) {
585 case REV_RECONFIGURE
: COVERAGE_INC(rev_reconfigure
); break;
586 case REV_STP
: COVERAGE_INC(rev_stp
); break;
587 case REV_BOND
: COVERAGE_INC(rev_bond
); break;
588 case REV_PORT_TOGGLED
: COVERAGE_INC(rev_port_toggled
); break;
589 case REV_FLOW_TABLE
: COVERAGE_INC(rev_flow_table
); break;
590 case REV_MAC_LEARNING
: COVERAGE_INC(rev_mac_learning
); break;
592 backer
->need_revalidate
= 0;
594 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
595 struct ofport_dpif
*ofport
;
596 struct ofbundle
*bundle
;
598 if (ofproto
->backer
!= backer
) {
603 xlate_ofproto_set(ofproto
, ofproto
->up
.name
,
604 ofproto
->backer
->dpif
, ofproto
->miss_rule
,
605 ofproto
->no_packet_in_rule
, ofproto
->ml
,
606 ofproto
->stp
, ofproto
->mbridge
,
607 ofproto
->sflow
, ofproto
->ipfix
,
608 ofproto
->netflow
, ofproto
->up
.frag_handling
,
609 ofproto
->up
.forward_bpdu
,
610 connmgr_has_in_band(ofproto
->up
.connmgr
),
611 ofproto
->backer
->enable_recirc
,
612 ofproto
->backer
->variable_length_userdata
,
613 ofproto
->backer
->max_mpls_depth
);
615 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
616 xlate_bundle_set(ofproto
, bundle
, bundle
->name
,
617 bundle
->vlan_mode
, bundle
->vlan
,
618 bundle
->trunks
, bundle
->use_priority_tags
,
619 bundle
->bond
, bundle
->lacp
,
623 HMAP_FOR_EACH (ofport
, up
.hmap_node
, &ofproto
->up
.ports
) {
624 int stp_port
= ofport
->stp_port
625 ? stp_port_no(ofport
->stp_port
)
627 xlate_ofport_set(ofproto
, ofport
->bundle
, ofport
,
628 ofport
->up
.ofp_port
, ofport
->odp_port
,
629 ofport
->up
.netdev
, ofport
->cfm
,
630 ofport
->bfd
, ofport
->peer
, stp_port
,
631 ofport
->qdscp
, ofport
->n_qdscp
,
632 ofport
->up
.pp
.config
, ofport
->up
.pp
.state
,
633 ofport
->is_tunnel
, ofport
->may_enable
);
638 udpif_revalidate(backer
->udpif
);
641 process_dpif_port_changes(backer
);
646 /* Check for and handle port changes in 'backer''s dpif. */
648 process_dpif_port_changes(struct dpif_backer
*backer
)
654 error
= dpif_port_poll(backer
->dpif
, &devname
);
660 process_dpif_all_ports_changed(backer
);
664 process_dpif_port_change(backer
, devname
);
669 process_dpif_port_error(backer
, error
);
676 process_dpif_all_ports_changed(struct dpif_backer
*backer
)
678 struct ofproto_dpif
*ofproto
;
679 struct dpif_port dpif_port
;
680 struct dpif_port_dump dump
;
681 struct sset devnames
;
684 sset_init(&devnames
);
685 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
686 if (ofproto
->backer
== backer
) {
687 struct ofport
*ofport
;
689 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->up
.ports
) {
690 sset_add(&devnames
, netdev_get_name(ofport
->netdev
));
694 DPIF_PORT_FOR_EACH (&dpif_port
, &dump
, backer
->dpif
) {
695 sset_add(&devnames
, dpif_port
.name
);
698 SSET_FOR_EACH (devname
, &devnames
) {
699 process_dpif_port_change(backer
, devname
);
701 sset_destroy(&devnames
);
705 process_dpif_port_change(struct dpif_backer
*backer
, const char *devname
)
707 struct ofproto_dpif
*ofproto
;
708 struct dpif_port port
;
710 /* Don't report on the datapath's device. */
711 if (!strcmp(devname
, dpif_base_name(backer
->dpif
))) {
715 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
,
716 &all_ofproto_dpifs
) {
717 if (simap_contains(&ofproto
->backer
->tnl_backers
, devname
)) {
722 ofproto
= lookup_ofproto_dpif_by_port_name(devname
);
723 if (dpif_port_query_by_name(backer
->dpif
, devname
, &port
)) {
724 /* The port was removed. If we know the datapath,
725 * report it through poll_set(). If we don't, it may be
726 * notifying us of a removal we initiated, so ignore it.
727 * If there's a pending ENOBUFS, let it stand, since
728 * everything will be reevaluated. */
729 if (ofproto
&& ofproto
->port_poll_errno
!= ENOBUFS
) {
730 sset_add(&ofproto
->port_poll_set
, devname
);
731 ofproto
->port_poll_errno
= 0;
733 } else if (!ofproto
) {
734 /* The port was added, but we don't know with which
735 * ofproto we should associate it. Delete it. */
736 dpif_port_del(backer
->dpif
, port
.port_no
);
738 struct ofport_dpif
*ofport
;
740 ofport
= ofport_dpif_cast(shash_find_data(
741 &ofproto
->up
.port_by_name
, devname
));
743 && ofport
->odp_port
!= port
.port_no
744 && !odp_port_to_ofport(backer
, port
.port_no
))
746 /* 'ofport''s datapath port number has changed from
747 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
748 * structures to match. */
749 ovs_rwlock_wrlock(&backer
->odp_to_ofport_lock
);
750 hmap_remove(&backer
->odp_to_ofport_map
, &ofport
->odp_port_node
);
751 ofport
->odp_port
= port
.port_no
;
752 hmap_insert(&backer
->odp_to_ofport_map
, &ofport
->odp_port_node
,
753 hash_odp_port(port
.port_no
));
754 ovs_rwlock_unlock(&backer
->odp_to_ofport_lock
);
755 backer
->need_revalidate
= REV_RECONFIGURE
;
758 dpif_port_destroy(&port
);
761 /* Propagate 'error' to all ofprotos based on 'backer'. */
763 process_dpif_port_error(struct dpif_backer
*backer
, int error
)
765 struct ofproto_dpif
*ofproto
;
767 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
768 if (ofproto
->backer
== backer
) {
769 sset_clear(&ofproto
->port_poll_set
);
770 ofproto
->port_poll_errno
= error
;
776 type_wait(const char *type
)
778 struct dpif_backer
*backer
;
780 backer
= shash_find_data(&all_dpif_backers
, type
);
782 /* This is not necessarily a problem, since backers are only
783 * created on demand. */
787 dpif_wait(backer
->dpif
);
790 /* Basic life-cycle. */
792 static int add_internal_flows(struct ofproto_dpif
*);
794 static struct ofproto
*
797 struct ofproto_dpif
*ofproto
= xmalloc(sizeof *ofproto
);
802 dealloc(struct ofproto
*ofproto_
)
804 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
809 close_dpif_backer(struct dpif_backer
*backer
)
811 ovs_assert(backer
->refcount
> 0);
813 if (--backer
->refcount
) {
817 udpif_destroy(backer
->udpif
);
819 simap_destroy(&backer
->tnl_backers
);
820 ovs_rwlock_destroy(&backer
->odp_to_ofport_lock
);
821 hmap_destroy(&backer
->odp_to_ofport_map
);
822 shash_find_and_delete(&all_dpif_backers
, backer
->type
);
823 recirc_id_pool_destroy(backer
->rid_pool
);
825 dpif_close(backer
->dpif
);
829 /* Datapath port slated for removal from datapath. */
831 struct list list_node
;
835 static bool check_variable_length_userdata(struct dpif_backer
*backer
);
836 static size_t check_max_mpls_depth(struct dpif_backer
*backer
);
837 static bool check_recirc(struct dpif_backer
*backer
);
840 open_dpif_backer(const char *type
, struct dpif_backer
**backerp
)
842 struct dpif_backer
*backer
;
843 struct dpif_port_dump port_dump
;
844 struct dpif_port port
;
845 struct shash_node
*node
;
846 struct list garbage_list
;
847 struct odp_garbage
*garbage
, *next
;
854 backer
= shash_find_data(&all_dpif_backers
, type
);
861 backer_name
= xasprintf("ovs-%s", type
);
863 /* Remove any existing datapaths, since we assume we're the only
864 * userspace controlling the datapath. */
866 dp_enumerate_names(type
, &names
);
867 SSET_FOR_EACH(name
, &names
) {
868 struct dpif
*old_dpif
;
870 /* Don't remove our backer if it exists. */
871 if (!strcmp(name
, backer_name
)) {
875 if (dpif_open(name
, type
, &old_dpif
)) {
876 VLOG_WARN("couldn't open old datapath %s to remove it", name
);
878 dpif_delete(old_dpif
);
879 dpif_close(old_dpif
);
882 sset_destroy(&names
);
884 backer
= xmalloc(sizeof *backer
);
886 error
= dpif_create_and_open(backer_name
, type
, &backer
->dpif
);
889 VLOG_ERR("failed to open datapath of type %s: %s", type
,
890 ovs_strerror(error
));
894 backer
->udpif
= udpif_create(backer
, backer
->dpif
);
896 backer
->type
= xstrdup(type
);
897 backer
->refcount
= 1;
898 hmap_init(&backer
->odp_to_ofport_map
);
899 ovs_rwlock_init(&backer
->odp_to_ofport_lock
);
900 backer
->need_revalidate
= 0;
901 simap_init(&backer
->tnl_backers
);
902 backer
->recv_set_enable
= !ofproto_get_flow_restore_wait();
905 if (backer
->recv_set_enable
) {
906 dpif_flow_flush(backer
->dpif
);
909 /* Loop through the ports already on the datapath and remove any
910 * that we don't need anymore. */
911 list_init(&garbage_list
);
912 dpif_port_dump_start(&port_dump
, backer
->dpif
);
913 while (dpif_port_dump_next(&port_dump
, &port
)) {
914 node
= shash_find(&init_ofp_ports
, port
.name
);
915 if (!node
&& strcmp(port
.name
, dpif_base_name(backer
->dpif
))) {
916 garbage
= xmalloc(sizeof *garbage
);
917 garbage
->odp_port
= port
.port_no
;
918 list_push_front(&garbage_list
, &garbage
->list_node
);
921 dpif_port_dump_done(&port_dump
);
923 LIST_FOR_EACH_SAFE (garbage
, next
, list_node
, &garbage_list
) {
924 dpif_port_del(backer
->dpif
, garbage
->odp_port
);
925 list_remove(&garbage
->list_node
);
929 shash_add(&all_dpif_backers
, type
, backer
);
931 error
= dpif_recv_set(backer
->dpif
, backer
->recv_set_enable
);
933 VLOG_ERR("failed to listen on datapath of type %s: %s",
934 type
, ovs_strerror(error
));
935 close_dpif_backer(backer
);
938 backer
->enable_recirc
= check_recirc(backer
);
939 backer
->variable_length_userdata
= check_variable_length_userdata(backer
);
940 backer
->max_mpls_depth
= check_max_mpls_depth(backer
);
941 backer
->rid_pool
= recirc_id_pool_create();
943 if (backer
->recv_set_enable
) {
944 udpif_set_threads(backer
->udpif
, n_handlers
, n_revalidators
);
950 /* Tests whether 'backer''s datapath supports recirculation. Only newer
951 * datapaths support OVS_KEY_ATTR_RECIRC_ID in keys. We need to disable some
952 * features on older datapaths that don't support this feature.
954 * Returns false if 'backer' definitely does not support recirculation, true if
955 * it seems to support recirculation or if at least the error we get is
958 check_recirc(struct dpif_backer
*backer
)
961 struct odputil_keybuf keybuf
;
964 bool enable_recirc
= false;
966 memset(&flow
, 0, sizeof flow
);
970 ofpbuf_use_stack(&key
, &keybuf
, sizeof keybuf
);
971 odp_flow_key_from_flow(&key
, &flow
, NULL
, 0, true);
973 error
= dpif_flow_put(backer
->dpif
, DPIF_FP_CREATE
| DPIF_FP_MODIFY
,
974 ofpbuf_data(&key
), ofpbuf_size(&key
), NULL
, 0, NULL
,
976 if (error
&& error
!= EEXIST
) {
977 if (error
!= EINVAL
) {
978 VLOG_WARN("%s: Reciculation flow probe failed (%s)",
979 dpif_name(backer
->dpif
), ovs_strerror(error
));
984 error
= dpif_flow_del(backer
->dpif
, ofpbuf_data(&key
), ofpbuf_size(&key
),
987 VLOG_WARN("%s: failed to delete recirculation feature probe flow",
988 dpif_name(backer
->dpif
));
991 enable_recirc
= true;
995 VLOG_INFO("%s: Datapath supports recirculation",
996 dpif_name(backer
->dpif
));
998 VLOG_INFO("%s: Datapath does not support recirculation",
999 dpif_name(backer
->dpif
));
1002 return enable_recirc
;
1005 /* Tests whether 'backer''s datapath supports variable-length
1006 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions. We need
1007 * to disable some features on older datapaths that don't support this
1010 * Returns false if 'backer' definitely does not support variable-length
1011 * userdata, true if it seems to support them or if at least the error we get
1014 check_variable_length_userdata(struct dpif_backer
*backer
)
1016 struct eth_header
*eth
;
1017 struct ofpbuf actions
;
1018 struct dpif_execute execute
;
1019 struct ofpbuf packet
;
1023 /* Compose a userspace action that will cause an ERANGE error on older
1024 * datapaths that don't support variable-length userdata.
1026 * We really test for using userdata longer than 8 bytes, but older
1027 * datapaths accepted these, silently truncating the userdata to 8 bytes.
1028 * The same older datapaths rejected userdata shorter than 8 bytes, so we
1029 * test for that instead as a proxy for longer userdata support. */
1030 ofpbuf_init(&actions
, 64);
1031 start
= nl_msg_start_nested(&actions
, OVS_ACTION_ATTR_USERSPACE
);
1032 nl_msg_put_u32(&actions
, OVS_USERSPACE_ATTR_PID
,
1033 dpif_port_get_pid(backer
->dpif
, ODPP_NONE
, 0));
1034 nl_msg_put_unspec_zero(&actions
, OVS_USERSPACE_ATTR_USERDATA
, 4);
1035 nl_msg_end_nested(&actions
, start
);
1037 /* Compose a dummy ethernet packet. */
1038 ofpbuf_init(&packet
, ETH_HEADER_LEN
);
1039 eth
= ofpbuf_put_zeros(&packet
, ETH_HEADER_LEN
);
1040 eth
->eth_type
= htons(0x1234);
1042 /* Execute the actions. On older datapaths this fails with ERANGE, on
1043 * newer datapaths it succeeds. */
1044 execute
.actions
= ofpbuf_data(&actions
);
1045 execute
.actions_len
= ofpbuf_size(&actions
);
1046 execute
.packet
= &packet
;
1047 execute
.md
= PKT_METADATA_INITIALIZER(0);
1048 execute
.needs_help
= false;
1050 error
= dpif_execute(backer
->dpif
, &execute
);
1052 ofpbuf_uninit(&packet
);
1053 ofpbuf_uninit(&actions
);
1057 /* Variable-length userdata is supported.
1059 * Purge received packets to avoid processing the nonsense packet we
1060 * sent to userspace, then report success. */
1061 dpif_recv_purge(backer
->dpif
);
1065 /* Variable-length userdata is not supported. */
1066 VLOG_WARN("%s: datapath does not support variable-length userdata "
1067 "feature (needs Linux 3.10+ or kernel module from OVS "
1068 "1..11+). The NXAST_SAMPLE action will be ignored.",
1069 dpif_name(backer
->dpif
));
1073 /* Something odd happened. We're not sure whether variable-length
1074 * userdata is supported. Default to "yes". */
1075 VLOG_WARN("%s: variable-length userdata feature probe failed (%s)",
1076 dpif_name(backer
->dpif
), ovs_strerror(error
));
1081 /* Tests the MPLS label stack depth supported by 'backer''s datapath.
1083 * Returns the number of elements in a struct flow's mpls_lse field
1084 * if the datapath supports at least that many entries in an
1086 * Otherwise returns the number of MPLS push actions supported by
1089 check_max_mpls_depth(struct dpif_backer
*backer
)
1094 for (n
= 0; n
< FLOW_MAX_MPLS_LABELS
; n
++) {
1095 struct odputil_keybuf keybuf
;
1099 memset(&flow
, 0, sizeof flow
);
1100 flow
.dl_type
= htons(ETH_TYPE_MPLS
);
1101 flow_set_mpls_bos(&flow
, n
, 1);
1103 ofpbuf_use_stack(&key
, &keybuf
, sizeof keybuf
);
1104 odp_flow_key_from_flow(&key
, &flow
, NULL
, 0, false);
1106 error
= dpif_flow_put(backer
->dpif
, DPIF_FP_CREATE
| DPIF_FP_MODIFY
,
1107 ofpbuf_data(&key
), ofpbuf_size(&key
), NULL
, 0, NULL
, 0, NULL
);
1108 if (error
&& error
!= EEXIST
) {
1109 if (error
!= EINVAL
) {
1110 VLOG_WARN("%s: MPLS stack length feature probe failed (%s)",
1111 dpif_name(backer
->dpif
), ovs_strerror(error
));
1116 error
= dpif_flow_del(backer
->dpif
, ofpbuf_data(&key
), ofpbuf_size(&key
), NULL
);
1118 VLOG_WARN("%s: failed to delete MPLS feature probe flow",
1119 dpif_name(backer
->dpif
));
1123 VLOG_INFO("%s: MPLS label stack length probed as %d",
1124 dpif_name(backer
->dpif
), n
);
1129 construct(struct ofproto
*ofproto_
)
1131 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1132 struct shash_node
*node
, *next
;
1135 error
= open_dpif_backer(ofproto
->up
.type
, &ofproto
->backer
);
1140 ofproto
->netflow
= NULL
;
1141 ofproto
->sflow
= NULL
;
1142 ofproto
->ipfix
= NULL
;
1143 ofproto
->stp
= NULL
;
1144 ofproto
->dump_seq
= 0;
1145 hmap_init(&ofproto
->bundles
);
1146 ofproto
->ml
= mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME
);
1147 ofproto
->mbridge
= mbridge_create();
1148 ofproto
->has_bonded_bundles
= false;
1149 ofproto
->lacp_enabled
= false;
1150 ovs_mutex_init_adaptive(&ofproto
->stats_mutex
);
1151 ovs_mutex_init(&ofproto
->vsp_mutex
);
1153 guarded_list_init(&ofproto
->pins
);
1155 ofproto_dpif_unixctl_init();
1157 hmap_init(&ofproto
->vlandev_map
);
1158 hmap_init(&ofproto
->realdev_vid_map
);
1160 sset_init(&ofproto
->ports
);
1161 sset_init(&ofproto
->ghost_ports
);
1162 sset_init(&ofproto
->port_poll_set
);
1163 ofproto
->port_poll_errno
= 0;
1164 ofproto
->change_seq
= 0;
1165 ofproto
->pins_seq
= seq_create();
1166 ofproto
->pins_seqno
= seq_read(ofproto
->pins_seq
);
1169 SHASH_FOR_EACH_SAFE (node
, next
, &init_ofp_ports
) {
1170 struct iface_hint
*iface_hint
= node
->data
;
1172 if (!strcmp(iface_hint
->br_name
, ofproto
->up
.name
)) {
1173 /* Check if the datapath already has this port. */
1174 if (dpif_port_exists(ofproto
->backer
->dpif
, node
->name
)) {
1175 sset_add(&ofproto
->ports
, node
->name
);
1178 free(iface_hint
->br_name
);
1179 free(iface_hint
->br_type
);
1181 shash_delete(&init_ofp_ports
, node
);
1185 hmap_insert(&all_ofproto_dpifs
, &ofproto
->all_ofproto_dpifs_node
,
1186 hash_string(ofproto
->up
.name
, 0));
1187 memset(&ofproto
->stats
, 0, sizeof ofproto
->stats
);
1189 ofproto_init_tables(ofproto_
, N_TABLES
);
1190 error
= add_internal_flows(ofproto
);
1192 ofproto
->up
.tables
[TBL_INTERNAL
].flags
= OFTABLE_HIDDEN
| OFTABLE_READONLY
;
1198 add_internal_miss_flow(struct ofproto_dpif
*ofproto
, int id
,
1199 const struct ofpbuf
*ofpacts
, struct rule_dpif
**rulep
)
1205 match_init_catchall(&match
);
1206 match_set_reg(&match
, 0, id
);
1208 error
= ofproto_dpif_add_internal_flow(ofproto
, &match
, 0, ofpacts
, &rule
);
1209 *rulep
= error
? NULL
: rule_dpif_cast(rule
);
1215 add_internal_flows(struct ofproto_dpif
*ofproto
)
1217 struct ofpact_controller
*controller
;
1218 uint64_t ofpacts_stub
[128 / 8];
1219 struct ofpbuf ofpacts
;
1220 struct rule
*unused_rulep OVS_UNUSED
;
1221 struct ofpact_resubmit
*resubmit
;
1226 ofpbuf_use_stack(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
1229 controller
= ofpact_put_CONTROLLER(&ofpacts
);
1230 controller
->max_len
= UINT16_MAX
;
1231 controller
->controller_id
= 0;
1232 controller
->reason
= OFPR_NO_MATCH
;
1233 ofpact_pad(&ofpacts
);
1235 error
= add_internal_miss_flow(ofproto
, id
++, &ofpacts
,
1236 &ofproto
->miss_rule
);
1241 ofpbuf_clear(&ofpacts
);
1242 error
= add_internal_miss_flow(ofproto
, id
++, &ofpacts
,
1243 &ofproto
->no_packet_in_rule
);
1248 error
= add_internal_miss_flow(ofproto
, id
++, &ofpacts
,
1249 &ofproto
->drop_frags_rule
);
1254 /* Continue non-recirculation rule lookups from table 0.
1256 * (priority=2), recirc=0, actions=resubmit(, 0)
1258 resubmit
= ofpact_put_RESUBMIT(&ofpacts
);
1259 resubmit
->ofpact
.compat
= 0;
1260 resubmit
->in_port
= OFPP_IN_PORT
;
1261 resubmit
->table_id
= 0;
1263 match_init_catchall(&match
);
1264 match_set_recirc_id(&match
, 0);
1266 error
= ofproto_dpif_add_internal_flow(ofproto
, &match
, 2, &ofpacts
,
1272 /* Drop any run away recirc rule lookups. Recirc_id has to be
1273 * non-zero when reaching this rule.
1275 * (priority=1), *, actions=drop
1277 ofpbuf_clear(&ofpacts
);
1278 match_init_catchall(&match
);
1279 error
= ofproto_dpif_add_internal_flow(ofproto
, &match
, 1, &ofpacts
,
1286 destruct(struct ofproto
*ofproto_
)
1288 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1289 struct rule_dpif
*rule
, *next_rule
;
1290 struct ofproto_packet_in
*pin
, *next_pin
;
1291 struct oftable
*table
;
1294 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1296 xlate_remove_ofproto(ofproto
);
1299 /* Ensure that the upcall processing threads have no remaining references
1300 * to the ofproto or anything in it. */
1301 udpif_synchronize(ofproto
->backer
->udpif
);
1303 hmap_remove(&all_ofproto_dpifs
, &ofproto
->all_ofproto_dpifs_node
);
1305 OFPROTO_FOR_EACH_TABLE (table
, &ofproto
->up
) {
1306 struct cls_cursor cursor
;
1308 fat_rwlock_rdlock(&table
->cls
.rwlock
);
1309 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
1310 fat_rwlock_unlock(&table
->cls
.rwlock
);
1311 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, up
.cr
, &cursor
) {
1312 ofproto_rule_delete(&ofproto
->up
, &rule
->up
);
1316 guarded_list_pop_all(&ofproto
->pins
, &pins
);
1317 LIST_FOR_EACH_SAFE (pin
, next_pin
, list_node
, &pins
) {
1318 list_remove(&pin
->list_node
);
1319 free(CONST_CAST(void *, pin
->up
.packet
));
1322 guarded_list_destroy(&ofproto
->pins
);
1324 mbridge_unref(ofproto
->mbridge
);
1326 netflow_unref(ofproto
->netflow
);
1327 dpif_sflow_unref(ofproto
->sflow
);
1328 hmap_destroy(&ofproto
->bundles
);
1329 mac_learning_unref(ofproto
->ml
);
1331 hmap_destroy(&ofproto
->vlandev_map
);
1332 hmap_destroy(&ofproto
->realdev_vid_map
);
1334 sset_destroy(&ofproto
->ports
);
1335 sset_destroy(&ofproto
->ghost_ports
);
1336 sset_destroy(&ofproto
->port_poll_set
);
1338 ovs_mutex_destroy(&ofproto
->stats_mutex
);
1339 ovs_mutex_destroy(&ofproto
->vsp_mutex
);
1341 seq_destroy(ofproto
->pins_seq
);
1343 close_dpif_backer(ofproto
->backer
);
1347 run(struct ofproto
*ofproto_
)
1349 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1350 uint64_t new_seq
, new_dump_seq
;
1352 if (mbridge_need_revalidate(ofproto
->mbridge
)) {
1353 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1354 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
1355 mac_learning_flush(ofproto
->ml
);
1356 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
1359 /* Always updates the ofproto->pins_seqno to avoid frequent wakeup during
1360 * flow restore. Even though nothing is processed during flow restore,
1361 * all queued 'pins' will be handled immediately when flow restore
1363 ofproto
->pins_seqno
= seq_read(ofproto
->pins_seq
);
1365 /* Do not perform any periodic activity required by 'ofproto' while
1366 * waiting for flow restore to complete. */
1367 if (!ofproto_get_flow_restore_wait()) {
1368 struct ofproto_packet_in
*pin
, *next_pin
;
1371 guarded_list_pop_all(&ofproto
->pins
, &pins
);
1372 LIST_FOR_EACH_SAFE (pin
, next_pin
, list_node
, &pins
) {
1373 connmgr_send_packet_in(ofproto
->up
.connmgr
, pin
);
1374 list_remove(&pin
->list_node
);
1375 free(CONST_CAST(void *, pin
->up
.packet
));
1380 if (ofproto
->netflow
) {
1381 netflow_run(ofproto
->netflow
);
1383 if (ofproto
->sflow
) {
1384 dpif_sflow_run(ofproto
->sflow
);
1386 if (ofproto
->ipfix
) {
1387 dpif_ipfix_run(ofproto
->ipfix
);
1390 new_seq
= seq_read(connectivity_seq_get());
1391 if (ofproto
->change_seq
!= new_seq
) {
1392 struct ofport_dpif
*ofport
;
1394 HMAP_FOR_EACH (ofport
, up
.hmap_node
, &ofproto
->up
.ports
) {
1398 ofproto
->change_seq
= new_seq
;
1400 if (ofproto
->lacp_enabled
|| ofproto
->has_bonded_bundles
) {
1401 struct ofbundle
*bundle
;
1403 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
1409 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
1410 if (mac_learning_run(ofproto
->ml
)) {
1411 ofproto
->backer
->need_revalidate
= REV_MAC_LEARNING
;
1413 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
1415 new_dump_seq
= seq_read(udpif_dump_seq(ofproto
->backer
->udpif
));
1416 if (ofproto
->dump_seq
!= new_dump_seq
) {
1417 struct rule
*rule
, *next_rule
;
1419 /* We know stats are relatively fresh, so now is a good time to do some
1421 ofproto
->dump_seq
= new_dump_seq
;
1423 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1425 ovs_mutex_lock(&ofproto_mutex
);
1426 LIST_FOR_EACH_SAFE (rule
, next_rule
, expirable
,
1427 &ofproto
->up
.expirable
) {
1428 rule_expire(rule_dpif_cast(rule
));
1430 ovs_mutex_unlock(&ofproto_mutex
);
1432 /* All outstanding data in existing flows has been accounted, so it's a
1433 * good time to do bond rebalancing. */
1434 if (ofproto
->has_bonded_bundles
) {
1435 struct ofbundle
*bundle
;
1437 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
1439 bond_rebalance(bundle
->bond
);
1449 wait(struct ofproto
*ofproto_
)
1451 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1453 if (ofproto_get_flow_restore_wait()) {
1457 if (ofproto
->sflow
) {
1458 dpif_sflow_wait(ofproto
->sflow
);
1460 if (ofproto
->ipfix
) {
1461 dpif_ipfix_wait(ofproto
->ipfix
);
1463 if (ofproto
->lacp_enabled
|| ofproto
->has_bonded_bundles
) {
1464 struct ofbundle
*bundle
;
1466 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
1467 bundle_wait(bundle
);
1470 if (ofproto
->netflow
) {
1471 netflow_wait(ofproto
->netflow
);
1473 ovs_rwlock_rdlock(&ofproto
->ml
->rwlock
);
1474 mac_learning_wait(ofproto
->ml
);
1475 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
1477 if (ofproto
->backer
->need_revalidate
) {
1478 /* Shouldn't happen, but if it does just go around again. */
1479 VLOG_DBG_RL(&rl
, "need revalidate in ofproto_wait_cb()");
1480 poll_immediate_wake();
1483 seq_wait(udpif_dump_seq(ofproto
->backer
->udpif
), ofproto
->dump_seq
);
1484 seq_wait(ofproto
->pins_seq
, ofproto
->pins_seqno
);
1488 type_get_memory_usage(const char *type
, struct simap
*usage
)
1490 struct dpif_backer
*backer
;
1492 backer
= shash_find_data(&all_dpif_backers
, type
);
1494 udpif_get_memory_usage(backer
->udpif
, usage
);
1499 flush(struct ofproto
*ofproto_
)
1501 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1502 struct dpif_backer
*backer
= ofproto
->backer
;
1505 udpif_flush(backer
->udpif
);
1510 get_features(struct ofproto
*ofproto_ OVS_UNUSED
,
1511 bool *arp_match_ip
, enum ofputil_action_bitmap
*actions
)
1513 *arp_match_ip
= true;
1514 *actions
= (OFPUTIL_A_OUTPUT
|
1515 OFPUTIL_A_SET_VLAN_VID
|
1516 OFPUTIL_A_SET_VLAN_PCP
|
1517 OFPUTIL_A_STRIP_VLAN
|
1518 OFPUTIL_A_SET_DL_SRC
|
1519 OFPUTIL_A_SET_DL_DST
|
1520 OFPUTIL_A_SET_NW_SRC
|
1521 OFPUTIL_A_SET_NW_DST
|
1522 OFPUTIL_A_SET_NW_TOS
|
1523 OFPUTIL_A_SET_TP_SRC
|
1524 OFPUTIL_A_SET_TP_DST
|
1529 get_tables(struct ofproto
*ofproto
, struct ofp12_table_stats
*ots
)
1533 strcpy(ots
->name
, "classifier");
1535 for (i
= 0; i
< ofproto
->n_tables
; i
++) {
1536 unsigned long missed
, matched
;
1538 atomic_read(&ofproto
->tables
[i
].n_matched
, &matched
);
1539 ots
[i
].matched_count
= htonll(matched
);
1540 atomic_read(&ofproto
->tables
[i
].n_missed
, &missed
);
1541 ots
[i
].lookup_count
= htonll(matched
+ missed
);
1545 static struct ofport
*
1548 struct ofport_dpif
*port
= xmalloc(sizeof *port
);
1553 port_dealloc(struct ofport
*port_
)
1555 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
1560 port_construct(struct ofport
*port_
)
1562 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
1563 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
1564 const struct netdev
*netdev
= port
->up
.netdev
;
1565 char namebuf
[NETDEV_VPORT_NAME_BUFSIZE
];
1566 struct dpif_port dpif_port
;
1569 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1570 port
->bundle
= NULL
;
1573 port
->may_enable
= true;
1574 port
->stp_port
= NULL
;
1575 port
->stp_state
= STP_DISABLED
;
1576 port
->is_tunnel
= false;
1580 port
->realdev_ofp_port
= 0;
1581 port
->vlandev_vid
= 0;
1582 port
->carrier_seq
= netdev_get_carrier_resets(netdev
);
1583 port
->is_layer3
= netdev_vport_is_layer3(netdev
);
1585 if (netdev_vport_is_patch(netdev
)) {
1586 /* By bailing out here, we don't submit the port to the sFlow module
1587 * to be considered for counter polling export. This is correct
1588 * because the patch port represents an interface that sFlow considers
1589 * to be "internal" to the switch as a whole, and therefore not an
1590 * candidate for counter polling. */
1591 port
->odp_port
= ODPP_NONE
;
1592 ofport_update_peer(port
);
1596 error
= dpif_port_query_by_name(ofproto
->backer
->dpif
,
1597 netdev_vport_get_dpif_port(netdev
, namebuf
,
1604 port
->odp_port
= dpif_port
.port_no
;
1606 if (netdev_get_tunnel_config(netdev
)) {
1607 tnl_port_add(port
, port
->up
.netdev
, port
->odp_port
);
1608 port
->is_tunnel
= true;
1610 /* Sanity-check that a mapping doesn't already exist. This
1611 * shouldn't happen for non-tunnel ports. */
1612 if (odp_port_to_ofp_port(ofproto
, port
->odp_port
) != OFPP_NONE
) {
1613 VLOG_ERR("port %s already has an OpenFlow port number",
1615 dpif_port_destroy(&dpif_port
);
1619 ovs_rwlock_wrlock(&ofproto
->backer
->odp_to_ofport_lock
);
1620 hmap_insert(&ofproto
->backer
->odp_to_ofport_map
, &port
->odp_port_node
,
1621 hash_odp_port(port
->odp_port
));
1622 ovs_rwlock_unlock(&ofproto
->backer
->odp_to_ofport_lock
);
1624 dpif_port_destroy(&dpif_port
);
1626 if (ofproto
->sflow
) {
1627 dpif_sflow_add_port(ofproto
->sflow
, port_
, port
->odp_port
);
1634 port_destruct(struct ofport
*port_
)
1636 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
1637 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
1638 const char *devname
= netdev_get_name(port
->up
.netdev
);
1639 char namebuf
[NETDEV_VPORT_NAME_BUFSIZE
];
1640 const char *dp_port_name
;
1642 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1644 xlate_ofport_remove(port
);
1647 dp_port_name
= netdev_vport_get_dpif_port(port
->up
.netdev
, namebuf
,
1649 if (dpif_port_exists(ofproto
->backer
->dpif
, dp_port_name
)) {
1650 /* The underlying device is still there, so delete it. This
1651 * happens when the ofproto is being destroyed, since the caller
1652 * assumes that removal of attached ports will happen as part of
1654 if (!port
->is_tunnel
) {
1655 dpif_port_del(ofproto
->backer
->dpif
, port
->odp_port
);
1660 port
->peer
->peer
= NULL
;
1664 if (port
->odp_port
!= ODPP_NONE
&& !port
->is_tunnel
) {
1665 ovs_rwlock_wrlock(&ofproto
->backer
->odp_to_ofport_lock
);
1666 hmap_remove(&ofproto
->backer
->odp_to_ofport_map
, &port
->odp_port_node
);
1667 ovs_rwlock_unlock(&ofproto
->backer
->odp_to_ofport_lock
);
1671 sset_find_and_delete(&ofproto
->ports
, devname
);
1672 sset_find_and_delete(&ofproto
->ghost_ports
, devname
);
1673 bundle_remove(port_
);
1674 set_cfm(port_
, NULL
);
1675 set_bfd(port_
, NULL
);
1676 if (port
->stp_port
) {
1677 stp_port_disable(port
->stp_port
);
1679 if (ofproto
->sflow
) {
1680 dpif_sflow_del_port(ofproto
->sflow
, port
->odp_port
);
1687 port_modified(struct ofport
*port_
)
1689 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
1691 if (port
->bundle
&& port
->bundle
->bond
) {
1692 bond_slave_set_netdev(port
->bundle
->bond
, port
, port
->up
.netdev
);
1696 cfm_set_netdev(port
->cfm
, port
->up
.netdev
);
1700 bfd_set_netdev(port
->bfd
, port
->up
.netdev
);
1703 ofproto_dpif_monitor_port_update(port
, port
->bfd
, port
->cfm
,
1704 port
->up
.pp
.hw_addr
);
1706 if (port
->is_tunnel
&& tnl_port_reconfigure(port
, port
->up
.netdev
,
1708 ofproto_dpif_cast(port
->up
.ofproto
)->backer
->need_revalidate
=
1712 ofport_update_peer(port
);
1716 port_reconfigured(struct ofport
*port_
, enum ofputil_port_config old_config
)
1718 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
1719 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
1720 enum ofputil_port_config changed
= old_config
^ port
->up
.pp
.config
;
1722 if (changed
& (OFPUTIL_PC_NO_RECV
| OFPUTIL_PC_NO_RECV_STP
|
1723 OFPUTIL_PC_NO_FWD
| OFPUTIL_PC_NO_FLOOD
|
1724 OFPUTIL_PC_NO_PACKET_IN
)) {
1725 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1727 if (changed
& OFPUTIL_PC_NO_FLOOD
&& port
->bundle
) {
1728 bundle_update(port
->bundle
);
1734 set_sflow(struct ofproto
*ofproto_
,
1735 const struct ofproto_sflow_options
*sflow_options
)
1737 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1738 struct dpif_sflow
*ds
= ofproto
->sflow
;
1740 if (sflow_options
) {
1742 struct ofport_dpif
*ofport
;
1744 ds
= ofproto
->sflow
= dpif_sflow_create();
1745 HMAP_FOR_EACH (ofport
, up
.hmap_node
, &ofproto
->up
.ports
) {
1746 dpif_sflow_add_port(ds
, &ofport
->up
, ofport
->odp_port
);
1748 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1750 dpif_sflow_set_options(ds
, sflow_options
);
1753 dpif_sflow_unref(ds
);
1754 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1755 ofproto
->sflow
= NULL
;
1763 struct ofproto
*ofproto_
,
1764 const struct ofproto_ipfix_bridge_exporter_options
*bridge_exporter_options
,
1765 const struct ofproto_ipfix_flow_exporter_options
*flow_exporters_options
,
1766 size_t n_flow_exporters_options
)
1768 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1769 struct dpif_ipfix
*di
= ofproto
->ipfix
;
1770 bool has_options
= bridge_exporter_options
|| flow_exporters_options
;
1772 if (has_options
&& !di
) {
1773 di
= ofproto
->ipfix
= dpif_ipfix_create();
1777 /* Call set_options in any case to cleanly flush the flow
1778 * caches in the last exporters that are to be destroyed. */
1779 dpif_ipfix_set_options(
1780 di
, bridge_exporter_options
, flow_exporters_options
,
1781 n_flow_exporters_options
);
1784 dpif_ipfix_unref(di
);
1785 ofproto
->ipfix
= NULL
;
1793 set_cfm(struct ofport
*ofport_
, const struct cfm_settings
*s
)
1795 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
1800 struct ofproto_dpif
*ofproto
;
1802 ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
1803 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1804 ofport
->cfm
= cfm_create(ofport
->up
.netdev
);
1807 if (cfm_configure(ofport
->cfm
, s
)) {
1814 cfm_unref(ofport
->cfm
);
1817 ofproto_dpif_monitor_port_update(ofport
, ofport
->bfd
, ofport
->cfm
,
1818 ofport
->up
.pp
.hw_addr
);
1823 get_cfm_status(const struct ofport
*ofport_
,
1824 struct ofproto_cfm_status
*status
)
1826 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
1830 if (cfm_check_status_change(ofport
->cfm
)) {
1831 status
->faults
= cfm_get_fault(ofport
->cfm
);
1832 status
->flap_count
= cfm_get_flap_count(ofport
->cfm
);
1833 status
->remote_opstate
= cfm_get_opup(ofport
->cfm
);
1834 status
->health
= cfm_get_health(ofport
->cfm
);
1835 cfm_get_remote_mpids(ofport
->cfm
, &status
->rmps
, &status
->n_rmps
);
1837 ret
= NO_STATUS_CHANGE
;
1847 set_bfd(struct ofport
*ofport_
, const struct smap
*cfg
)
1849 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport_
->ofproto
);
1850 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
1854 ofport
->bfd
= bfd_configure(old
, netdev_get_name(ofport
->up
.netdev
),
1855 cfg
, ofport
->up
.netdev
);
1856 if (ofport
->bfd
!= old
) {
1857 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1859 ofproto_dpif_monitor_port_update(ofport
, ofport
->bfd
, ofport
->cfm
,
1860 ofport
->up
.pp
.hw_addr
);
1865 get_bfd_status(struct ofport
*ofport_
, struct smap
*smap
)
1867 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
1871 if (bfd_check_status_change(ofport
->bfd
)) {
1872 bfd_get_status(ofport
->bfd
, smap
);
1874 ret
= NO_STATUS_CHANGE
;
1883 /* Spanning Tree. */
1886 send_bpdu_cb(struct ofpbuf
*pkt
, int port_num
, void *ofproto_
)
1888 struct ofproto_dpif
*ofproto
= ofproto_
;
1889 struct stp_port
*sp
= stp_get_port(ofproto
->stp
, port_num
);
1890 struct ofport_dpif
*ofport
;
1892 ofport
= stp_port_get_aux(sp
);
1894 VLOG_WARN_RL(&rl
, "%s: cannot send BPDU on unknown port %d",
1895 ofproto
->up
.name
, port_num
);
1897 struct eth_header
*eth
= ofpbuf_l2(pkt
);
1899 netdev_get_etheraddr(ofport
->up
.netdev
, eth
->eth_src
);
1900 if (eth_addr_is_zero(eth
->eth_src
)) {
1901 VLOG_WARN_RL(&rl
, "%s: cannot send BPDU on port %d "
1902 "with unknown MAC", ofproto
->up
.name
, port_num
);
1904 ofproto_dpif_send_packet(ofport
, pkt
);
1910 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1912 set_stp(struct ofproto
*ofproto_
, const struct ofproto_stp_settings
*s
)
1914 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1916 /* Only revalidate flows if the configuration changed. */
1917 if (!s
!= !ofproto
->stp
) {
1918 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1922 if (!ofproto
->stp
) {
1923 ofproto
->stp
= stp_create(ofproto_
->name
, s
->system_id
,
1924 send_bpdu_cb
, ofproto
);
1925 ofproto
->stp_last_tick
= time_msec();
1928 stp_set_bridge_id(ofproto
->stp
, s
->system_id
);
1929 stp_set_bridge_priority(ofproto
->stp
, s
->priority
);
1930 stp_set_hello_time(ofproto
->stp
, s
->hello_time
);
1931 stp_set_max_age(ofproto
->stp
, s
->max_age
);
1932 stp_set_forward_delay(ofproto
->stp
, s
->fwd_delay
);
1934 struct ofport
*ofport
;
1936 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->up
.ports
) {
1937 set_stp_port(ofport
, NULL
);
1940 stp_unref(ofproto
->stp
);
1941 ofproto
->stp
= NULL
;
1948 get_stp_status(struct ofproto
*ofproto_
, struct ofproto_stp_status
*s
)
1950 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1954 s
->bridge_id
= stp_get_bridge_id(ofproto
->stp
);
1955 s
->designated_root
= stp_get_designated_root(ofproto
->stp
);
1956 s
->root_path_cost
= stp_get_root_path_cost(ofproto
->stp
);
1965 update_stp_port_state(struct ofport_dpif
*ofport
)
1967 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
1968 enum stp_state state
;
1970 /* Figure out new state. */
1971 state
= ofport
->stp_port
? stp_port_get_state(ofport
->stp_port
)
1975 if (ofport
->stp_state
!= state
) {
1976 enum ofputil_port_state of_state
;
1979 VLOG_DBG_RL(&rl
, "port %s: STP state changed from %s to %s",
1980 netdev_get_name(ofport
->up
.netdev
),
1981 stp_state_name(ofport
->stp_state
),
1982 stp_state_name(state
));
1983 if (stp_learn_in_state(ofport
->stp_state
)
1984 != stp_learn_in_state(state
)) {
1985 /* xxx Learning action flows should also be flushed. */
1986 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
1987 mac_learning_flush(ofproto
->ml
);
1988 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
1990 fwd_change
= stp_forward_in_state(ofport
->stp_state
)
1991 != stp_forward_in_state(state
);
1993 ofproto
->backer
->need_revalidate
= REV_STP
;
1994 ofport
->stp_state
= state
;
1995 ofport
->stp_state_entered
= time_msec();
1997 if (fwd_change
&& ofport
->bundle
) {
1998 bundle_update(ofport
->bundle
);
2001 /* Update the STP state bits in the OpenFlow port description. */
2002 of_state
= ofport
->up
.pp
.state
& ~OFPUTIL_PS_STP_MASK
;
2003 of_state
|= (state
== STP_LISTENING
? OFPUTIL_PS_STP_LISTEN
2004 : state
== STP_LEARNING
? OFPUTIL_PS_STP_LEARN
2005 : state
== STP_FORWARDING
? OFPUTIL_PS_STP_FORWARD
2006 : state
== STP_BLOCKING
? OFPUTIL_PS_STP_BLOCK
2008 ofproto_port_set_state(&ofport
->up
, of_state
);
2012 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2013 * caller is responsible for assigning STP port numbers and ensuring
2014 * there are no duplicates. */
2016 set_stp_port(struct ofport
*ofport_
,
2017 const struct ofproto_port_stp_settings
*s
)
2019 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2020 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2021 struct stp_port
*sp
= ofport
->stp_port
;
2023 if (!s
|| !s
->enable
) {
2025 ofport
->stp_port
= NULL
;
2026 stp_port_disable(sp
);
2027 update_stp_port_state(ofport
);
2030 } else if (sp
&& stp_port_no(sp
) != s
->port_num
2031 && ofport
== stp_port_get_aux(sp
)) {
2032 /* The port-id changed, so disable the old one if it's not
2033 * already in use by another port. */
2034 stp_port_disable(sp
);
2037 sp
= ofport
->stp_port
= stp_get_port(ofproto
->stp
, s
->port_num
);
2038 stp_port_enable(sp
);
2040 stp_port_set_aux(sp
, ofport
);
2041 stp_port_set_priority(sp
, s
->priority
);
2042 stp_port_set_path_cost(sp
, s
->path_cost
);
2044 update_stp_port_state(ofport
);
2050 get_stp_port_status(struct ofport
*ofport_
,
2051 struct ofproto_port_stp_status
*s
)
2053 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2054 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2055 struct stp_port
*sp
= ofport
->stp_port
;
2057 if (!ofproto
->stp
|| !sp
) {
2063 s
->port_id
= stp_port_get_id(sp
);
2064 s
->state
= stp_port_get_state(sp
);
2065 s
->sec_in_state
= (time_msec() - ofport
->stp_state_entered
) / 1000;
2066 s
->role
= stp_port_get_role(sp
);
2072 get_stp_port_stats(struct ofport
*ofport_
,
2073 struct ofproto_port_stp_stats
*s
)
2075 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2076 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2077 struct stp_port
*sp
= ofport
->stp_port
;
2079 if (!ofproto
->stp
|| !sp
) {
2085 stp_port_get_counts(sp
, &s
->tx_count
, &s
->rx_count
, &s
->error_count
);
2091 stp_run(struct ofproto_dpif
*ofproto
)
2094 long long int now
= time_msec();
2095 long long int elapsed
= now
- ofproto
->stp_last_tick
;
2096 struct stp_port
*sp
;
2099 stp_tick(ofproto
->stp
, MIN(INT_MAX
, elapsed
));
2100 ofproto
->stp_last_tick
= now
;
2102 while (stp_get_changed_port(ofproto
->stp
, &sp
)) {
2103 struct ofport_dpif
*ofport
= stp_port_get_aux(sp
);
2106 update_stp_port_state(ofport
);
2110 if (stp_check_and_reset_fdb_flush(ofproto
->stp
)) {
2111 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
2112 mac_learning_flush(ofproto
->ml
);
2113 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
2119 stp_wait(struct ofproto_dpif
*ofproto
)
2122 poll_timer_wait(1000);
2127 set_queues(struct ofport
*ofport_
, const struct ofproto_port_queue
*qdscp
,
2130 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2131 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2133 if (ofport
->n_qdscp
!= n_qdscp
2134 || (n_qdscp
&& memcmp(ofport
->qdscp
, qdscp
,
2135 n_qdscp
* sizeof *qdscp
))) {
2136 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2137 free(ofport
->qdscp
);
2138 ofport
->qdscp
= n_qdscp
2139 ? xmemdup(qdscp
, n_qdscp
* sizeof *qdscp
)
2141 ofport
->n_qdscp
= n_qdscp
;
2149 /* Expires all MAC learning entries associated with 'bundle' and forces its
2150 * ofproto to revalidate every flow.
2152 * Normally MAC learning entries are removed only from the ofproto associated
2153 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2154 * are removed from every ofproto. When patch ports and SLB bonds are in use
2155 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2156 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2157 * with the host from which it migrated. */
2159 bundle_flush_macs(struct ofbundle
*bundle
, bool all_ofprotos
)
2161 struct ofproto_dpif
*ofproto
= bundle
->ofproto
;
2162 struct mac_learning
*ml
= ofproto
->ml
;
2163 struct mac_entry
*mac
, *next_mac
;
2165 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2166 ovs_rwlock_wrlock(&ml
->rwlock
);
2167 LIST_FOR_EACH_SAFE (mac
, next_mac
, lru_node
, &ml
->lrus
) {
2168 if (mac
->port
.p
== bundle
) {
2170 struct ofproto_dpif
*o
;
2172 HMAP_FOR_EACH (o
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
2174 struct mac_entry
*e
;
2176 ovs_rwlock_wrlock(&o
->ml
->rwlock
);
2177 e
= mac_learning_lookup(o
->ml
, mac
->mac
, mac
->vlan
);
2179 mac_learning_expire(o
->ml
, e
);
2181 ovs_rwlock_unlock(&o
->ml
->rwlock
);
2186 mac_learning_expire(ml
, mac
);
2189 ovs_rwlock_unlock(&ml
->rwlock
);
2192 static struct ofbundle
*
2193 bundle_lookup(const struct ofproto_dpif
*ofproto
, void *aux
)
2195 struct ofbundle
*bundle
;
2197 HMAP_FOR_EACH_IN_BUCKET (bundle
, hmap_node
, hash_pointer(aux
, 0),
2198 &ofproto
->bundles
) {
2199 if (bundle
->aux
== aux
) {
2207 bundle_update(struct ofbundle
*bundle
)
2209 struct ofport_dpif
*port
;
2211 bundle
->floodable
= true;
2212 LIST_FOR_EACH (port
, bundle_node
, &bundle
->ports
) {
2213 if (port
->up
.pp
.config
& OFPUTIL_PC_NO_FLOOD
2215 || !stp_forward_in_state(port
->stp_state
)) {
2216 bundle
->floodable
= false;
2223 bundle_del_port(struct ofport_dpif
*port
)
2225 struct ofbundle
*bundle
= port
->bundle
;
2227 bundle
->ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2229 list_remove(&port
->bundle_node
);
2230 port
->bundle
= NULL
;
2233 lacp_slave_unregister(bundle
->lacp
, port
);
2236 bond_slave_unregister(bundle
->bond
, port
);
2239 bundle_update(bundle
);
2243 bundle_add_port(struct ofbundle
*bundle
, ofp_port_t ofp_port
,
2244 struct lacp_slave_settings
*lacp
)
2246 struct ofport_dpif
*port
;
2248 port
= get_ofp_port(bundle
->ofproto
, ofp_port
);
2253 if (port
->bundle
!= bundle
) {
2254 bundle
->ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2256 bundle_remove(&port
->up
);
2259 port
->bundle
= bundle
;
2260 list_push_back(&bundle
->ports
, &port
->bundle_node
);
2261 if (port
->up
.pp
.config
& OFPUTIL_PC_NO_FLOOD
2263 || !stp_forward_in_state(port
->stp_state
)) {
2264 bundle
->floodable
= false;
2268 bundle
->ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2269 lacp_slave_register(bundle
->lacp
, port
, lacp
);
2276 bundle_destroy(struct ofbundle
*bundle
)
2278 struct ofproto_dpif
*ofproto
;
2279 struct ofport_dpif
*port
, *next_port
;
2285 ofproto
= bundle
->ofproto
;
2286 mbridge_unregister_bundle(ofproto
->mbridge
, bundle
->aux
);
2289 xlate_bundle_remove(bundle
);
2292 LIST_FOR_EACH_SAFE (port
, next_port
, bundle_node
, &bundle
->ports
) {
2293 bundle_del_port(port
);
2296 bundle_flush_macs(bundle
, true);
2297 hmap_remove(&ofproto
->bundles
, &bundle
->hmap_node
);
2299 free(bundle
->trunks
);
2300 lacp_unref(bundle
->lacp
);
2301 bond_unref(bundle
->bond
);
2306 bundle_set(struct ofproto
*ofproto_
, void *aux
,
2307 const struct ofproto_bundle_settings
*s
)
2309 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2310 bool need_flush
= false;
2311 struct ofport_dpif
*port
;
2312 struct ofbundle
*bundle
;
2313 unsigned long *trunks
;
2319 bundle_destroy(bundle_lookup(ofproto
, aux
));
2323 ovs_assert(s
->n_slaves
== 1 || s
->bond
!= NULL
);
2324 ovs_assert((s
->lacp
!= NULL
) == (s
->lacp_slaves
!= NULL
));
2326 bundle
= bundle_lookup(ofproto
, aux
);
2328 bundle
= xmalloc(sizeof *bundle
);
2330 bundle
->ofproto
= ofproto
;
2331 hmap_insert(&ofproto
->bundles
, &bundle
->hmap_node
,
2332 hash_pointer(aux
, 0));
2334 bundle
->name
= NULL
;
2336 list_init(&bundle
->ports
);
2337 bundle
->vlan_mode
= PORT_VLAN_TRUNK
;
2339 bundle
->trunks
= NULL
;
2340 bundle
->use_priority_tags
= s
->use_priority_tags
;
2341 bundle
->lacp
= NULL
;
2342 bundle
->bond
= NULL
;
2344 bundle
->floodable
= true;
2345 mbridge_register_bundle(ofproto
->mbridge
, bundle
);
2348 if (!bundle
->name
|| strcmp(s
->name
, bundle
->name
)) {
2350 bundle
->name
= xstrdup(s
->name
);
2355 ofproto
->lacp_enabled
= true;
2356 if (!bundle
->lacp
) {
2357 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2358 bundle
->lacp
= lacp_create();
2360 lacp_configure(bundle
->lacp
, s
->lacp
);
2362 lacp_unref(bundle
->lacp
);
2363 bundle
->lacp
= NULL
;
2366 /* Update set of ports. */
2368 for (i
= 0; i
< s
->n_slaves
; i
++) {
2369 if (!bundle_add_port(bundle
, s
->slaves
[i
],
2370 s
->lacp
? &s
->lacp_slaves
[i
] : NULL
)) {
2374 if (!ok
|| list_size(&bundle
->ports
) != s
->n_slaves
) {
2375 struct ofport_dpif
*next_port
;
2377 LIST_FOR_EACH_SAFE (port
, next_port
, bundle_node
, &bundle
->ports
) {
2378 for (i
= 0; i
< s
->n_slaves
; i
++) {
2379 if (s
->slaves
[i
] == port
->up
.ofp_port
) {
2384 bundle_del_port(port
);
2388 ovs_assert(list_size(&bundle
->ports
) <= s
->n_slaves
);
2390 if (list_is_empty(&bundle
->ports
)) {
2391 bundle_destroy(bundle
);
2395 /* Set VLAN tagging mode */
2396 if (s
->vlan_mode
!= bundle
->vlan_mode
2397 || s
->use_priority_tags
!= bundle
->use_priority_tags
) {
2398 bundle
->vlan_mode
= s
->vlan_mode
;
2399 bundle
->use_priority_tags
= s
->use_priority_tags
;
2404 vlan
= (s
->vlan_mode
== PORT_VLAN_TRUNK
? -1
2405 : s
->vlan
>= 0 && s
->vlan
<= 4095 ? s
->vlan
2407 if (vlan
!= bundle
->vlan
) {
2408 bundle
->vlan
= vlan
;
2412 /* Get trunked VLANs. */
2413 switch (s
->vlan_mode
) {
2414 case PORT_VLAN_ACCESS
:
2418 case PORT_VLAN_TRUNK
:
2419 trunks
= CONST_CAST(unsigned long *, s
->trunks
);
2422 case PORT_VLAN_NATIVE_UNTAGGED
:
2423 case PORT_VLAN_NATIVE_TAGGED
:
2424 if (vlan
!= 0 && (!s
->trunks
2425 || !bitmap_is_set(s
->trunks
, vlan
)
2426 || bitmap_is_set(s
->trunks
, 0))) {
2427 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2429 trunks
= bitmap_clone(s
->trunks
, 4096);
2431 trunks
= bitmap_allocate1(4096);
2433 bitmap_set1(trunks
, vlan
);
2434 bitmap_set0(trunks
, 0);
2436 trunks
= CONST_CAST(unsigned long *, s
->trunks
);
2443 if (!vlan_bitmap_equal(trunks
, bundle
->trunks
)) {
2444 free(bundle
->trunks
);
2445 if (trunks
== s
->trunks
) {
2446 bundle
->trunks
= vlan_bitmap_clone(trunks
);
2448 bundle
->trunks
= trunks
;
2453 if (trunks
!= s
->trunks
) {
2458 if (!list_is_short(&bundle
->ports
)) {
2459 bundle
->ofproto
->has_bonded_bundles
= true;
2461 if (bond_reconfigure(bundle
->bond
, s
->bond
)) {
2462 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2465 bundle
->bond
= bond_create(s
->bond
, ofproto
);
2466 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2469 LIST_FOR_EACH (port
, bundle_node
, &bundle
->ports
) {
2470 bond_slave_register(bundle
->bond
, port
,
2471 port
->up
.ofp_port
, port
->up
.netdev
);
2474 bond_unref(bundle
->bond
);
2475 bundle
->bond
= NULL
;
2478 /* If we changed something that would affect MAC learning, un-learn
2479 * everything on this port and force flow revalidation. */
2481 bundle_flush_macs(bundle
, false);
2488 bundle_remove(struct ofport
*port_
)
2490 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
2491 struct ofbundle
*bundle
= port
->bundle
;
2494 bundle_del_port(port
);
2495 if (list_is_empty(&bundle
->ports
)) {
2496 bundle_destroy(bundle
);
2497 } else if (list_is_short(&bundle
->ports
)) {
2498 bond_unref(bundle
->bond
);
2499 bundle
->bond
= NULL
;
2505 send_pdu_cb(void *port_
, const void *pdu
, size_t pdu_size
)
2507 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 10);
2508 struct ofport_dpif
*port
= port_
;
2509 uint8_t ea
[ETH_ADDR_LEN
];
2512 error
= netdev_get_etheraddr(port
->up
.netdev
, ea
);
2514 struct ofpbuf packet
;
2517 ofpbuf_init(&packet
, 0);
2518 packet_pdu
= eth_compose(&packet
, eth_addr_lacp
, ea
, ETH_TYPE_LACP
,
2520 memcpy(packet_pdu
, pdu
, pdu_size
);
2522 ofproto_dpif_send_packet(port
, &packet
);
2523 ofpbuf_uninit(&packet
);
2525 VLOG_ERR_RL(&rl
, "port %s: cannot obtain Ethernet address of iface "
2526 "%s (%s)", port
->bundle
->name
,
2527 netdev_get_name(port
->up
.netdev
), ovs_strerror(error
));
2532 bundle_send_learning_packets(struct ofbundle
*bundle
)
2534 struct ofproto_dpif
*ofproto
= bundle
->ofproto
;
2535 struct ofpbuf
*learning_packet
;
2536 int error
, n_packets
, n_errors
;
2537 struct mac_entry
*e
;
2538 struct list packets
;
2540 list_init(&packets
);
2541 ovs_rwlock_rdlock(&ofproto
->ml
->rwlock
);
2542 LIST_FOR_EACH (e
, lru_node
, &ofproto
->ml
->lrus
) {
2543 if (e
->port
.p
!= bundle
) {
2546 learning_packet
= bond_compose_learning_packet(bundle
->bond
,
2549 /* Temporarily use 'frame' as a private pointer (see below). */
2550 ovs_assert(learning_packet
->frame
== ofpbuf_data(learning_packet
));
2551 learning_packet
->frame
= port_void
;
2552 list_push_back(&packets
, &learning_packet
->list_node
);
2555 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
2557 error
= n_packets
= n_errors
= 0;
2558 LIST_FOR_EACH (learning_packet
, list_node
, &packets
) {
2560 void *port_void
= learning_packet
->frame
;
2562 /* Restore 'frame'. */
2563 learning_packet
->frame
= ofpbuf_data(learning_packet
);
2564 ret
= ofproto_dpif_send_packet(port_void
, learning_packet
);
2571 ofpbuf_list_delete(&packets
);
2574 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
2575 VLOG_WARN_RL(&rl
, "bond %s: %d errors sending %d gratuitous learning "
2576 "packets, last error was: %s",
2577 bundle
->name
, n_errors
, n_packets
, ovs_strerror(error
));
2579 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2580 bundle
->name
, n_packets
);
2585 bundle_run(struct ofbundle
*bundle
)
2588 lacp_run(bundle
->lacp
, send_pdu_cb
);
2591 struct ofport_dpif
*port
;
2593 LIST_FOR_EACH (port
, bundle_node
, &bundle
->ports
) {
2594 bond_slave_set_may_enable(bundle
->bond
, port
, port
->may_enable
);
2597 if (bond_run(bundle
->bond
, lacp_status(bundle
->lacp
))) {
2598 bundle
->ofproto
->backer
->need_revalidate
= REV_BOND
;
2601 if (bond_should_send_learning_packets(bundle
->bond
)) {
2602 bundle_send_learning_packets(bundle
);
2608 bundle_wait(struct ofbundle
*bundle
)
2611 lacp_wait(bundle
->lacp
);
2614 bond_wait(bundle
->bond
);
2621 mirror_set__(struct ofproto
*ofproto_
, void *aux
,
2622 const struct ofproto_mirror_settings
*s
)
2624 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2625 struct ofbundle
**srcs
, **dsts
;
2630 mirror_destroy(ofproto
->mbridge
, aux
);
2634 srcs
= xmalloc(s
->n_srcs
* sizeof *srcs
);
2635 dsts
= xmalloc(s
->n_dsts
* sizeof *dsts
);
2637 for (i
= 0; i
< s
->n_srcs
; i
++) {
2638 srcs
[i
] = bundle_lookup(ofproto
, s
->srcs
[i
]);
2641 for (i
= 0; i
< s
->n_dsts
; i
++) {
2642 dsts
[i
] = bundle_lookup(ofproto
, s
->dsts
[i
]);
2645 error
= mirror_set(ofproto
->mbridge
, aux
, s
->name
, srcs
, s
->n_srcs
, dsts
,
2646 s
->n_dsts
, s
->src_vlans
,
2647 bundle_lookup(ofproto
, s
->out_bundle
), s
->out_vlan
);
2654 mirror_get_stats__(struct ofproto
*ofproto
, void *aux
,
2655 uint64_t *packets
, uint64_t *bytes
)
2657 return mirror_get_stats(ofproto_dpif_cast(ofproto
)->mbridge
, aux
, packets
,
2662 set_flood_vlans(struct ofproto
*ofproto_
, unsigned long *flood_vlans
)
2664 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2665 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
2666 if (mac_learning_set_flood_vlans(ofproto
->ml
, flood_vlans
)) {
2667 mac_learning_flush(ofproto
->ml
);
2669 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
2674 is_mirror_output_bundle(const struct ofproto
*ofproto_
, void *aux
)
2676 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2677 struct ofbundle
*bundle
= bundle_lookup(ofproto
, aux
);
2678 return bundle
&& mirror_bundle_out(ofproto
->mbridge
, bundle
) != 0;
2682 forward_bpdu_changed(struct ofproto
*ofproto_
)
2684 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2685 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2689 set_mac_table_config(struct ofproto
*ofproto_
, unsigned int idle_time
,
2692 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2693 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
2694 mac_learning_set_idle_time(ofproto
->ml
, idle_time
);
2695 mac_learning_set_max_entries(ofproto
->ml
, max_entries
);
2696 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
2701 static struct ofport_dpif
*
2702 get_ofp_port(const struct ofproto_dpif
*ofproto
, ofp_port_t ofp_port
)
2704 struct ofport
*ofport
= ofproto_get_port(&ofproto
->up
, ofp_port
);
2705 return ofport
? ofport_dpif_cast(ofport
) : NULL
;
2709 ofproto_port_from_dpif_port(struct ofproto_dpif
*ofproto
,
2710 struct ofproto_port
*ofproto_port
,
2711 struct dpif_port
*dpif_port
)
2713 ofproto_port
->name
= dpif_port
->name
;
2714 ofproto_port
->type
= dpif_port
->type
;
2715 ofproto_port
->ofp_port
= odp_port_to_ofp_port(ofproto
, dpif_port
->port_no
);
2719 ofport_update_peer(struct ofport_dpif
*ofport
)
2721 const struct ofproto_dpif
*ofproto
;
2722 struct dpif_backer
*backer
;
2725 if (!netdev_vport_is_patch(ofport
->up
.netdev
)) {
2729 backer
= ofproto_dpif_cast(ofport
->up
.ofproto
)->backer
;
2730 backer
->need_revalidate
= REV_RECONFIGURE
;
2733 ofport
->peer
->peer
= NULL
;
2734 ofport
->peer
= NULL
;
2737 peer_name
= netdev_vport_patch_peer(ofport
->up
.netdev
);
2742 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
2743 struct ofport
*peer_ofport
;
2744 struct ofport_dpif
*peer
;
2747 if (ofproto
->backer
!= backer
) {
2751 peer_ofport
= shash_find_data(&ofproto
->up
.port_by_name
, peer_name
);
2756 peer
= ofport_dpif_cast(peer_ofport
);
2757 peer_peer
= netdev_vport_patch_peer(peer
->up
.netdev
);
2758 if (peer_peer
&& !strcmp(netdev_get_name(ofport
->up
.netdev
),
2760 ofport
->peer
= peer
;
2761 ofport
->peer
->peer
= ofport
;
2771 port_run(struct ofport_dpif
*ofport
)
2773 long long int carrier_seq
= netdev_get_carrier_resets(ofport
->up
.netdev
);
2774 bool carrier_changed
= carrier_seq
!= ofport
->carrier_seq
;
2775 bool enable
= netdev_get_carrier(ofport
->up
.netdev
);
2776 bool cfm_enable
= false;
2777 bool bfd_enable
= false;
2779 ofport
->carrier_seq
= carrier_seq
;
2782 int cfm_opup
= cfm_get_opup(ofport
->cfm
);
2784 cfm_enable
= !cfm_get_fault(ofport
->cfm
);
2786 if (cfm_opup
>= 0) {
2787 cfm_enable
= cfm_enable
&& cfm_opup
;
2792 bfd_enable
= bfd_forwarding(ofport
->bfd
);
2795 if (ofport
->bfd
|| ofport
->cfm
) {
2796 enable
= enable
&& (cfm_enable
|| bfd_enable
);
2799 if (ofport
->bundle
) {
2800 enable
= enable
&& lacp_slave_may_enable(ofport
->bundle
->lacp
, ofport
);
2801 if (carrier_changed
) {
2802 lacp_slave_carrier_changed(ofport
->bundle
->lacp
, ofport
);
2806 if (ofport
->may_enable
!= enable
) {
2807 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2808 ofproto
->backer
->need_revalidate
= REV_PORT_TOGGLED
;
2811 ofport
->may_enable
= enable
;
2815 port_query_by_name(const struct ofproto
*ofproto_
, const char *devname
,
2816 struct ofproto_port
*ofproto_port
)
2818 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2819 struct dpif_port dpif_port
;
2822 if (sset_contains(&ofproto
->ghost_ports
, devname
)) {
2823 const char *type
= netdev_get_type_from_name(devname
);
2825 /* We may be called before ofproto->up.port_by_name is populated with
2826 * the appropriate ofport. For this reason, we must get the name and
2827 * type from the netdev layer directly. */
2829 const struct ofport
*ofport
;
2831 ofport
= shash_find_data(&ofproto
->up
.port_by_name
, devname
);
2832 ofproto_port
->ofp_port
= ofport
? ofport
->ofp_port
: OFPP_NONE
;
2833 ofproto_port
->name
= xstrdup(devname
);
2834 ofproto_port
->type
= xstrdup(type
);
2840 if (!sset_contains(&ofproto
->ports
, devname
)) {
2843 error
= dpif_port_query_by_name(ofproto
->backer
->dpif
,
2844 devname
, &dpif_port
);
2846 ofproto_port_from_dpif_port(ofproto
, ofproto_port
, &dpif_port
);
2852 port_add(struct ofproto
*ofproto_
, struct netdev
*netdev
)
2854 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2855 const char *devname
= netdev_get_name(netdev
);
2856 char namebuf
[NETDEV_VPORT_NAME_BUFSIZE
];
2857 const char *dp_port_name
;
2859 if (netdev_vport_is_patch(netdev
)) {
2860 sset_add(&ofproto
->ghost_ports
, netdev_get_name(netdev
));
2864 dp_port_name
= netdev_vport_get_dpif_port(netdev
, namebuf
, sizeof namebuf
);
2865 if (!dpif_port_exists(ofproto
->backer
->dpif
, dp_port_name
)) {
2866 odp_port_t port_no
= ODPP_NONE
;
2869 error
= dpif_port_add(ofproto
->backer
->dpif
, netdev
, &port_no
);
2873 if (netdev_get_tunnel_config(netdev
)) {
2874 simap_put(&ofproto
->backer
->tnl_backers
,
2875 dp_port_name
, odp_to_u32(port_no
));
2879 if (netdev_get_tunnel_config(netdev
)) {
2880 sset_add(&ofproto
->ghost_ports
, devname
);
2882 sset_add(&ofproto
->ports
, devname
);
2888 port_del(struct ofproto
*ofproto_
, ofp_port_t ofp_port
)
2890 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2891 struct ofport_dpif
*ofport
= get_ofp_port(ofproto
, ofp_port
);
2898 sset_find_and_delete(&ofproto
->ghost_ports
,
2899 netdev_get_name(ofport
->up
.netdev
));
2900 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2901 if (!ofport
->is_tunnel
&& !netdev_vport_is_patch(ofport
->up
.netdev
)) {
2902 error
= dpif_port_del(ofproto
->backer
->dpif
, ofport
->odp_port
);
2904 /* The caller is going to close ofport->up.netdev. If this is a
2905 * bonded port, then the bond is using that netdev, so remove it
2906 * from the bond. The client will need to reconfigure everything
2907 * after deleting ports, so then the slave will get re-added. */
2908 bundle_remove(&ofport
->up
);
2915 port_get_stats(const struct ofport
*ofport_
, struct netdev_stats
*stats
)
2917 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2920 error
= netdev_get_stats(ofport
->up
.netdev
, stats
);
2922 if (!error
&& ofport_
->ofp_port
== OFPP_LOCAL
) {
2923 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2925 ovs_mutex_lock(&ofproto
->stats_mutex
);
2926 /* ofproto->stats.tx_packets represents packets that we created
2927 * internally and sent to some port (e.g. packets sent with
2928 * ofproto_dpif_send_packet()). Account for them as if they had
2929 * come from OFPP_LOCAL and got forwarded. */
2931 if (stats
->rx_packets
!= UINT64_MAX
) {
2932 stats
->rx_packets
+= ofproto
->stats
.tx_packets
;
2935 if (stats
->rx_bytes
!= UINT64_MAX
) {
2936 stats
->rx_bytes
+= ofproto
->stats
.tx_bytes
;
2939 /* ofproto->stats.rx_packets represents packets that were received on
2940 * some port and we processed internally and dropped (e.g. STP).
2941 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2943 if (stats
->tx_packets
!= UINT64_MAX
) {
2944 stats
->tx_packets
+= ofproto
->stats
.rx_packets
;
2947 if (stats
->tx_bytes
!= UINT64_MAX
) {
2948 stats
->tx_bytes
+= ofproto
->stats
.rx_bytes
;
2950 ovs_mutex_unlock(&ofproto
->stats_mutex
);
2956 struct port_dump_state
{
2961 struct ofproto_port port
;
2966 port_dump_start(const struct ofproto
*ofproto_ OVS_UNUSED
, void **statep
)
2968 *statep
= xzalloc(sizeof(struct port_dump_state
));
2973 port_dump_next(const struct ofproto
*ofproto_
, void *state_
,
2974 struct ofproto_port
*port
)
2976 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2977 struct port_dump_state
*state
= state_
;
2978 const struct sset
*sset
;
2979 struct sset_node
*node
;
2981 if (state
->has_port
) {
2982 ofproto_port_destroy(&state
->port
);
2983 state
->has_port
= false;
2985 sset
= state
->ghost
? &ofproto
->ghost_ports
: &ofproto
->ports
;
2986 while ((node
= sset_at_position(sset
, &state
->bucket
, &state
->offset
))) {
2989 error
= port_query_by_name(ofproto_
, node
->name
, &state
->port
);
2991 *port
= state
->port
;
2992 state
->has_port
= true;
2994 } else if (error
!= ENODEV
) {
2999 if (!state
->ghost
) {
3000 state
->ghost
= true;
3003 return port_dump_next(ofproto_
, state_
, port
);
3010 port_dump_done(const struct ofproto
*ofproto_ OVS_UNUSED
, void *state_
)
3012 struct port_dump_state
*state
= state_
;
3014 if (state
->has_port
) {
3015 ofproto_port_destroy(&state
->port
);
3022 port_poll(const struct ofproto
*ofproto_
, char **devnamep
)
3024 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3026 if (ofproto
->port_poll_errno
) {
3027 int error
= ofproto
->port_poll_errno
;
3028 ofproto
->port_poll_errno
= 0;
3032 if (sset_is_empty(&ofproto
->port_poll_set
)) {
3036 *devnamep
= sset_pop(&ofproto
->port_poll_set
);
3041 port_poll_wait(const struct ofproto
*ofproto_
)
3043 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3044 dpif_port_poll_wait(ofproto
->backer
->dpif
);
3048 port_is_lacp_current(const struct ofport
*ofport_
)
3050 const struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
3051 return (ofport
->bundle
&& ofport
->bundle
->lacp
3052 ? lacp_slave_is_current(ofport
->bundle
->lacp
, ofport
)
3056 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3057 * then delete it entirely. */
3059 rule_expire(struct rule_dpif
*rule
)
3060 OVS_REQUIRES(ofproto_mutex
)
3062 uint16_t hard_timeout
, idle_timeout
;
3063 long long int now
= time_msec();
3066 ovs_assert(!rule
->up
.pending
);
3068 hard_timeout
= rule
->up
.hard_timeout
;
3069 idle_timeout
= rule
->up
.idle_timeout
;
3071 /* Has 'rule' expired? */
3073 long long int modified
;
3075 ovs_mutex_lock(&rule
->up
.mutex
);
3076 modified
= rule
->up
.modified
;
3077 ovs_mutex_unlock(&rule
->up
.mutex
);
3079 if (now
> modified
+ hard_timeout
* 1000) {
3080 reason
= OFPRR_HARD_TIMEOUT
;
3084 if (reason
< 0 && idle_timeout
) {
3087 ovs_mutex_lock(&rule
->stats_mutex
);
3088 used
= rule
->stats
.used
;
3089 ovs_mutex_unlock(&rule
->stats_mutex
);
3091 if (now
> used
+ idle_timeout
* 1000) {
3092 reason
= OFPRR_IDLE_TIMEOUT
;
3097 COVERAGE_INC(ofproto_dpif_expired
);
3098 ofproto_rule_expire(&rule
->up
, reason
);
3102 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3103 * 'flow' must reflect the data in 'packet'. */
3105 ofproto_dpif_execute_actions(struct ofproto_dpif
*ofproto
,
3106 const struct flow
*flow
,
3107 struct rule_dpif
*rule
,
3108 const struct ofpact
*ofpacts
, size_t ofpacts_len
,
3109 struct ofpbuf
*packet
)
3111 struct dpif_flow_stats stats
;
3112 struct xlate_out xout
;
3113 struct xlate_in xin
;
3115 struct dpif_execute execute
;
3118 ovs_assert((rule
!= NULL
) != (ofpacts
!= NULL
));
3120 dpif_flow_stats_extract(flow
, packet
, time_msec(), &stats
);
3123 rule_dpif_credit_stats(rule
, &stats
);
3126 xlate_in_init(&xin
, ofproto
, flow
, rule
, stats
.tcp_flags
, packet
);
3127 xin
.ofpacts
= ofpacts
;
3128 xin
.ofpacts_len
= ofpacts_len
;
3129 xin
.resubmit_stats
= &stats
;
3130 xlate_actions(&xin
, &xout
);
3132 in_port
= flow
->in_port
.ofp_port
;
3133 if (in_port
== OFPP_NONE
) {
3134 in_port
= OFPP_LOCAL
;
3136 execute
.actions
= ofpbuf_data(&xout
.odp_actions
);
3137 execute
.actions_len
= ofpbuf_size(&xout
.odp_actions
);
3138 execute
.packet
= packet
;
3139 execute
.md
.tunnel
= flow
->tunnel
;
3140 execute
.md
.skb_priority
= flow
->skb_priority
;
3141 execute
.md
.pkt_mark
= flow
->pkt_mark
;
3142 execute
.md
.in_port
.odp_port
= ofp_port_to_odp_port(ofproto
, in_port
);
3143 execute
.needs_help
= (xout
.slow
& SLOW_ACTION
) != 0;
3145 error
= dpif_execute(ofproto
->backer
->dpif
, &execute
);
3147 xlate_out_uninit(&xout
);
3153 rule_dpif_credit_stats(struct rule_dpif
*rule
,
3154 const struct dpif_flow_stats
*stats
)
3156 ovs_mutex_lock(&rule
->stats_mutex
);
3157 rule
->stats
.n_packets
+= stats
->n_packets
;
3158 rule
->stats
.n_bytes
+= stats
->n_bytes
;
3159 rule
->stats
.used
= MAX(rule
->stats
.used
, stats
->used
);
3160 ovs_mutex_unlock(&rule
->stats_mutex
);
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 const 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
,
3201 bool take_ref
, const struct dpif_flow_stats
*stats
)
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
, stats
);
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
,
3350 bool take_ref
, const struct dpif_flow_stats
*stats
)
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 struct oftable
*tbl
= &ofproto
->up
.tables
[next_id
];
3363 atomic_ulong
*stat
= *rule
? &tbl
->n_matched
: &tbl
->n_missed
;
3365 atomic_add(stat
, stats
->n_packets
, &orig
);
3368 return RULE_DPIF_LOOKUP_VERDICT_MATCH
;
3369 } else if (!honor_table_miss
) {
3370 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER
;
3372 switch (ofproto_table_get_config(&ofproto
->up
, *table_id
)) {
3373 case OFPROTO_TABLE_MISS_CONTINUE
:
3376 case OFPROTO_TABLE_MISS_CONTROLLER
:
3377 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER
;
3379 case OFPROTO_TABLE_MISS_DROP
:
3380 return RULE_DPIF_LOOKUP_VERDICT_DROP
;
3382 case OFPROTO_TABLE_MISS_DEFAULT
:
3383 return RULE_DPIF_LOOKUP_VERDICT_DEFAULT
;
3388 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER
;
3391 /* Given a port configuration (specified as zero if there's no port), chooses
3392 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
3395 * The rule is returned in '*rule', which is valid at least until the next
3396 * RCU quiescent period. If the '*rule' needs to stay around longer,
3397 * a reference must be taken on it (rule_dpif_ref()).
3400 choose_miss_rule(enum ofputil_port_config config
, struct rule_dpif
*miss_rule
,
3401 struct rule_dpif
*no_packet_in_rule
, struct rule_dpif
**rule
,
3404 *rule
= config
& OFPUTIL_PC_NO_PACKET_IN
? no_packet_in_rule
: miss_rule
;
3406 rule_dpif_ref(*rule
);
3411 complete_operation(struct rule_dpif
*rule
)
3412 OVS_REQUIRES(ofproto_mutex
)
3414 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
3416 ofproto
->backer
->need_revalidate
= REV_FLOW_TABLE
;
3417 ofoperation_complete(rule
->up
.pending
, 0);
3420 static struct rule_dpif
*rule_dpif_cast(const struct rule
*rule
)
3422 return rule
? CONTAINER_OF(rule
, struct rule_dpif
, up
) : NULL
;
3425 static struct rule
*
3428 struct rule_dpif
*rule
= xmalloc(sizeof *rule
);
3433 rule_dealloc(struct rule
*rule_
)
3435 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
3440 rule_construct(struct rule
*rule_
)
3441 OVS_NO_THREAD_SAFETY_ANALYSIS
3443 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
3444 ovs_mutex_init_adaptive(&rule
->stats_mutex
);
3445 rule
->stats
.n_packets
= 0;
3446 rule
->stats
.n_bytes
= 0;
3447 rule
->stats
.used
= rule
->up
.modified
;
3452 rule_insert(struct rule
*rule_
)
3453 OVS_REQUIRES(ofproto_mutex
)
3455 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
3456 complete_operation(rule
);
3460 rule_delete(struct rule
*rule_
)
3461 OVS_REQUIRES(ofproto_mutex
)
3463 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
3464 complete_operation(rule
);
3468 rule_destruct(struct rule
*rule_
)
3470 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
3471 ovs_mutex_destroy(&rule
->stats_mutex
);
3475 rule_get_stats(struct rule
*rule_
, uint64_t *packets
, uint64_t *bytes
,
3476 long long int *used
)
3478 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
3480 ovs_mutex_lock(&rule
->stats_mutex
);
3481 *packets
= rule
->stats
.n_packets
;
3482 *bytes
= rule
->stats
.n_bytes
;
3483 *used
= rule
->stats
.used
;
3484 ovs_mutex_unlock(&rule
->stats_mutex
);
3488 rule_dpif_execute(struct rule_dpif
*rule
, const struct flow
*flow
,
3489 struct ofpbuf
*packet
)
3491 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
3493 ofproto_dpif_execute_actions(ofproto
, flow
, rule
, NULL
, 0, packet
);
3497 rule_execute(struct rule
*rule
, const struct flow
*flow
,
3498 struct ofpbuf
*packet
)
3500 rule_dpif_execute(rule_dpif_cast(rule
), flow
, packet
);
3501 ofpbuf_delete(packet
);
3506 rule_modify_actions(struct rule
*rule_
, bool reset_counters
)
3507 OVS_REQUIRES(ofproto_mutex
)
3509 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
3511 if (reset_counters
) {
3512 ovs_mutex_lock(&rule
->stats_mutex
);
3513 rule
->stats
.n_packets
= 0;
3514 rule
->stats
.n_bytes
= 0;
3515 ovs_mutex_unlock(&rule
->stats_mutex
);
3518 complete_operation(rule
);
3521 static struct group_dpif
*group_dpif_cast(const struct ofgroup
*group
)
3523 return group
? CONTAINER_OF(group
, struct group_dpif
, up
) : NULL
;
3526 static struct ofgroup
*
3529 struct group_dpif
*group
= xzalloc(sizeof *group
);
3534 group_dealloc(struct ofgroup
*group_
)
3536 struct group_dpif
*group
= group_dpif_cast(group_
);
3541 group_construct_stats(struct group_dpif
*group
)
3542 OVS_REQUIRES(group
->stats_mutex
)
3544 struct ofputil_bucket
*bucket
;
3545 const struct list
*buckets
;
3547 group
->packet_count
= 0;
3548 group
->byte_count
= 0;
3550 group_dpif_get_buckets(group
, &buckets
);
3551 LIST_FOR_EACH (bucket
, list_node
, buckets
) {
3552 bucket
->stats
.packet_count
= 0;
3553 bucket
->stats
.byte_count
= 0;
3558 group_dpif_credit_stats(struct group_dpif
*group
,
3559 struct ofputil_bucket
*bucket
,
3560 const struct dpif_flow_stats
*stats
)
3562 ovs_mutex_lock(&group
->stats_mutex
);
3563 group
->packet_count
+= stats
->n_packets
;
3564 group
->byte_count
+= stats
->n_bytes
;
3566 bucket
->stats
.packet_count
+= stats
->n_packets
;
3567 bucket
->stats
.byte_count
+= stats
->n_bytes
;
3568 } else { /* Credit to all buckets */
3569 const struct list
*buckets
;
3571 group_dpif_get_buckets(group
, &buckets
);
3572 LIST_FOR_EACH (bucket
, list_node
, buckets
) {
3573 bucket
->stats
.packet_count
+= stats
->n_packets
;
3574 bucket
->stats
.byte_count
+= stats
->n_bytes
;
3577 ovs_mutex_unlock(&group
->stats_mutex
);
3581 group_construct(struct ofgroup
*group_
)
3583 struct group_dpif
*group
= group_dpif_cast(group_
);
3584 const struct ofputil_bucket
*bucket
;
3586 /* Prevent group chaining because our locking structure makes it hard to
3587 * implement deadlock-free. (See xlate_group_resource_check().) */
3588 LIST_FOR_EACH (bucket
, list_node
, &group
->up
.buckets
) {
3589 const struct ofpact
*a
;
3591 OFPACT_FOR_EACH (a
, bucket
->ofpacts
, bucket
->ofpacts_len
) {
3592 if (a
->type
== OFPACT_GROUP
) {
3593 return OFPERR_OFPGMFC_CHAINING_UNSUPPORTED
;
3598 ovs_mutex_init_adaptive(&group
->stats_mutex
);
3599 ovs_mutex_lock(&group
->stats_mutex
);
3600 group_construct_stats(group
);
3601 ovs_mutex_unlock(&group
->stats_mutex
);
3606 group_destruct(struct ofgroup
*group_
)
3608 struct group_dpif
*group
= group_dpif_cast(group_
);
3609 ovs_mutex_destroy(&group
->stats_mutex
);
3613 group_modify(struct ofgroup
*group_
)
3615 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(group_
->ofproto
);
3617 ofproto
->backer
->need_revalidate
= REV_FLOW_TABLE
;
3623 group_get_stats(const struct ofgroup
*group_
, struct ofputil_group_stats
*ogs
)
3625 struct group_dpif
*group
= group_dpif_cast(group_
);
3626 struct ofputil_bucket
*bucket
;
3627 const struct list
*buckets
;
3628 struct bucket_counter
*bucket_stats
;
3630 ovs_mutex_lock(&group
->stats_mutex
);
3631 ogs
->packet_count
= group
->packet_count
;
3632 ogs
->byte_count
= group
->byte_count
;
3634 group_dpif_get_buckets(group
, &buckets
);
3635 bucket_stats
= ogs
->bucket_stats
;
3636 LIST_FOR_EACH (bucket
, list_node
, buckets
) {
3637 bucket_stats
->packet_count
= bucket
->stats
.packet_count
;
3638 bucket_stats
->byte_count
= bucket
->stats
.byte_count
;
3641 ovs_mutex_unlock(&group
->stats_mutex
);
3646 /* If the group exists, this function increments the groups's reference count.
3648 * Make sure to call group_dpif_unref() after no longer needing to maintain
3649 * a reference to the group. */
3651 group_dpif_lookup(struct ofproto_dpif
*ofproto
, uint32_t group_id
,
3652 struct group_dpif
**group
)
3654 struct ofgroup
*ofgroup
;
3657 found
= ofproto_group_lookup(&ofproto
->up
, group_id
, &ofgroup
);
3658 *group
= found
? group_dpif_cast(ofgroup
) : NULL
;
3664 group_dpif_get_buckets(const struct group_dpif
*group
,
3665 const struct list
**buckets
)
3667 *buckets
= &group
->up
.buckets
;
3670 enum ofp11_group_type
3671 group_dpif_get_type(const struct group_dpif
*group
)
3673 return group
->up
.type
;
3676 /* Sends 'packet' out 'ofport'.
3677 * May modify 'packet'.
3678 * Returns 0 if successful, otherwise a positive errno value. */
3680 ofproto_dpif_send_packet(const struct ofport_dpif
*ofport
, struct ofpbuf
*packet
)
3682 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
3685 error
= xlate_send_packet(ofport
, packet
);
3687 ovs_mutex_lock(&ofproto
->stats_mutex
);
3688 ofproto
->stats
.tx_packets
++;
3689 ofproto
->stats
.tx_bytes
+= ofpbuf_size(packet
);
3690 ovs_mutex_unlock(&ofproto
->stats_mutex
);
3695 set_frag_handling(struct ofproto
*ofproto_
,
3696 enum ofp_config_flags frag_handling
)
3698 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3699 if (frag_handling
!= OFPC_FRAG_REASM
) {
3700 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3708 packet_out(struct ofproto
*ofproto_
, struct ofpbuf
*packet
,
3709 const struct flow
*flow
,
3710 const struct ofpact
*ofpacts
, size_t ofpacts_len
)
3712 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3714 ofproto_dpif_execute_actions(ofproto
, flow
, NULL
, ofpacts
,
3715 ofpacts_len
, packet
);
3722 set_netflow(struct ofproto
*ofproto_
,
3723 const struct netflow_options
*netflow_options
)
3725 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3727 if (netflow_options
) {
3728 if (!ofproto
->netflow
) {
3729 ofproto
->netflow
= netflow_create();
3730 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3732 return netflow_set_options(ofproto
->netflow
, netflow_options
);
3733 } else if (ofproto
->netflow
) {
3734 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3735 netflow_unref(ofproto
->netflow
);
3736 ofproto
->netflow
= NULL
;
3743 get_netflow_ids(const struct ofproto
*ofproto_
,
3744 uint8_t *engine_type
, uint8_t *engine_id
)
3746 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3748 dpif_get_netflow_ids(ofproto
->backer
->dpif
, engine_type
, engine_id
);
3751 static struct ofproto_dpif
*
3752 ofproto_dpif_lookup(const char *name
)
3754 struct ofproto_dpif
*ofproto
;
3756 HMAP_FOR_EACH_WITH_HASH (ofproto
, all_ofproto_dpifs_node
,
3757 hash_string(name
, 0), &all_ofproto_dpifs
) {
3758 if (!strcmp(ofproto
->up
.name
, name
)) {
3766 ofproto_unixctl_fdb_flush(struct unixctl_conn
*conn
, int argc
,
3767 const char *argv
[], void *aux OVS_UNUSED
)
3769 struct ofproto_dpif
*ofproto
;
3772 ofproto
= ofproto_dpif_lookup(argv
[1]);
3774 unixctl_command_reply_error(conn
, "no such bridge");
3777 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
3778 mac_learning_flush(ofproto
->ml
);
3779 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
3781 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
3782 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
3783 mac_learning_flush(ofproto
->ml
);
3784 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
3788 unixctl_command_reply(conn
, "table successfully flushed");
3791 static struct ofport_dpif
*
3792 ofbundle_get_a_port(const struct ofbundle
*bundle
)
3794 return CONTAINER_OF(list_front(&bundle
->ports
), struct ofport_dpif
,
3799 ofproto_unixctl_fdb_show(struct unixctl_conn
*conn
, int argc OVS_UNUSED
,
3800 const char *argv
[], void *aux OVS_UNUSED
)
3802 struct ds ds
= DS_EMPTY_INITIALIZER
;
3803 const struct ofproto_dpif
*ofproto
;
3804 const struct mac_entry
*e
;
3806 ofproto
= ofproto_dpif_lookup(argv
[1]);
3808 unixctl_command_reply_error(conn
, "no such bridge");
3812 ds_put_cstr(&ds
, " port VLAN MAC Age\n");
3813 ovs_rwlock_rdlock(&ofproto
->ml
->rwlock
);
3814 LIST_FOR_EACH (e
, lru_node
, &ofproto
->ml
->lrus
) {
3815 struct ofbundle
*bundle
= e
->port
.p
;
3816 char name
[OFP_MAX_PORT_NAME_LEN
];
3818 ofputil_port_to_string(ofbundle_get_a_port(bundle
)->up
.ofp_port
,
3820 ds_put_format(&ds
, "%5s %4d "ETH_ADDR_FMT
" %3d\n",
3821 name
, e
->vlan
, ETH_ADDR_ARGS(e
->mac
),
3822 mac_entry_age(ofproto
->ml
, e
));
3824 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
3825 unixctl_command_reply(conn
, ds_cstr(&ds
));
3830 struct xlate_out xout
;
3831 struct xlate_in xin
;
3832 const struct flow
*key
;
3834 struct flow_wildcards wc
;
3839 trace_format_rule(struct ds
*result
, int level
, const struct rule_dpif
*rule
)
3841 const struct rule_actions
*actions
;
3844 ds_put_char_multiple(result
, '\t', level
);
3846 ds_put_cstr(result
, "No match\n");
3850 ovs_mutex_lock(&rule
->up
.mutex
);
3851 cookie
= rule
->up
.flow_cookie
;
3852 ovs_mutex_unlock(&rule
->up
.mutex
);
3854 ds_put_format(result
, "Rule: table=%"PRIu8
" cookie=%#"PRIx64
" ",
3855 rule
? rule
->up
.table_id
: 0, ntohll(cookie
));
3856 cls_rule_format(&rule
->up
.cr
, result
);
3857 ds_put_char(result
, '\n');
3859 actions
= rule_dpif_get_actions(rule
);
3861 ds_put_char_multiple(result
, '\t', level
);
3862 ds_put_cstr(result
, "OpenFlow actions=");
3863 ofpacts_format(actions
->ofpacts
, actions
->ofpacts_len
, result
);
3864 ds_put_char(result
, '\n');
3868 trace_format_flow(struct ds
*result
, int level
, const char *title
,
3869 struct trace_ctx
*trace
)
3871 ds_put_char_multiple(result
, '\t', level
);
3872 ds_put_format(result
, "%s: ", title
);
3873 /* Do not report unchanged flows for resubmits. */
3874 if ((level
> 0 && flow_equal(&trace
->xin
.flow
, &trace
->flow
))
3875 || (level
== 0 && flow_equal(&trace
->xin
.flow
, trace
->key
))) {
3876 ds_put_cstr(result
, "unchanged");
3878 flow_format(result
, &trace
->xin
.flow
);
3879 trace
->flow
= trace
->xin
.flow
;
3881 ds_put_char(result
, '\n');
3885 trace_format_regs(struct ds
*result
, int level
, const char *title
,
3886 struct trace_ctx
*trace
)
3890 ds_put_char_multiple(result
, '\t', level
);
3891 ds_put_format(result
, "%s:", title
);
3892 for (i
= 0; i
< FLOW_N_REGS
; i
++) {
3893 ds_put_format(result
, " reg%"PRIuSIZE
"=0x%"PRIx32
, i
, trace
->flow
.regs
[i
]);
3895 ds_put_char(result
, '\n');
3899 trace_format_odp(struct ds
*result
, int level
, const char *title
,
3900 struct trace_ctx
*trace
)
3902 struct ofpbuf
*odp_actions
= &trace
->xout
.odp_actions
;
3904 ds_put_char_multiple(result
, '\t', level
);
3905 ds_put_format(result
, "%s: ", title
);
3906 format_odp_actions(result
, ofpbuf_data(odp_actions
),
3907 ofpbuf_size(odp_actions
));
3908 ds_put_char(result
, '\n');
3912 trace_format_megaflow(struct ds
*result
, int level
, const char *title
,
3913 struct trace_ctx
*trace
)
3917 ds_put_char_multiple(result
, '\t', level
);
3918 ds_put_format(result
, "%s: ", title
);
3919 flow_wildcards_or(&trace
->wc
, &trace
->xout
.wc
, &trace
->wc
);
3920 match_init(&match
, trace
->key
, &trace
->wc
);
3921 match_format(&match
, result
, OFP_DEFAULT_PRIORITY
);
3922 ds_put_char(result
, '\n');
3926 trace_resubmit(struct xlate_in
*xin
, struct rule_dpif
*rule
, int recurse
)
3928 struct trace_ctx
*trace
= CONTAINER_OF(xin
, struct trace_ctx
, xin
);
3929 struct ds
*result
= trace
->result
;
3931 ds_put_char(result
, '\n');
3932 trace_format_flow(result
, recurse
+ 1, "Resubmitted flow", trace
);
3933 trace_format_regs(result
, recurse
+ 1, "Resubmitted regs", trace
);
3934 trace_format_odp(result
, recurse
+ 1, "Resubmitted odp", trace
);
3935 trace_format_megaflow(result
, recurse
+ 1, "Resubmitted megaflow", trace
);
3936 trace_format_rule(result
, recurse
+ 1, rule
);
3940 trace_report(struct xlate_in
*xin
, const char *s
, int recurse
)
3942 struct trace_ctx
*trace
= CONTAINER_OF(xin
, struct trace_ctx
, xin
);
3943 struct ds
*result
= trace
->result
;
3945 ds_put_char_multiple(result
, '\t', recurse
);
3946 ds_put_cstr(result
, s
);
3947 ds_put_char(result
, '\n');
3950 /* Parses the 'argc' elements of 'argv', ignoring argv[0]. The following
3951 * forms are supported:
3953 * - [dpname] odp_flow [-generate | packet]
3954 * - bridge br_flow [-generate | packet]
3956 * On success, initializes '*ofprotop' and 'flow' and returns NULL. On failure
3957 * returns a nonnull malloced error message. */
3958 static char * WARN_UNUSED_RESULT
3959 parse_flow_and_packet(int argc
, const char *argv
[],
3960 struct ofproto_dpif
**ofprotop
, struct flow
*flow
,
3961 struct ofpbuf
**packetp
)
3963 const struct dpif_backer
*backer
= NULL
;
3964 const char *error
= NULL
;
3966 struct simap port_names
= SIMAP_INITIALIZER(&port_names
);
3967 struct ofpbuf
*packet
;
3968 struct ofpbuf odp_key
;
3969 struct ofpbuf odp_mask
;
3971 ofpbuf_init(&odp_key
, 0);
3972 ofpbuf_init(&odp_mask
, 0);
3974 /* Handle "-generate" or a hex string as the last argument. */
3975 if (!strcmp(argv
[argc
- 1], "-generate")) {
3976 packet
= ofpbuf_new(0);
3979 error
= eth_from_hex(argv
[argc
- 1], &packet
);
3982 } else if (argc
== 4) {
3983 /* The 3-argument form must end in "-generate' or a hex string. */
3989 /* odp_flow can have its in_port specified as a name instead of port no.
3990 * We do not yet know whether a given flow is a odp_flow or a br_flow.
3991 * But, to know whether a flow is odp_flow through odp_flow_from_string(),
3992 * we need to create a simap of name to port no. */
3994 const char *dp_type
;
3995 if (!strncmp(argv
[1], "ovs-", 4)) {
3996 dp_type
= argv
[1] + 4;
4000 backer
= shash_find_data(&all_dpif_backers
, dp_type
);
4001 } else if (argc
== 2) {
4002 struct shash_node
*node
;
4003 if (shash_count(&all_dpif_backers
) == 1) {
4004 node
= shash_first(&all_dpif_backers
);
4005 backer
= node
->data
;
4008 error
= "Syntax error";
4011 if (backer
&& backer
->dpif
) {
4012 struct dpif_port dpif_port
;
4013 struct dpif_port_dump port_dump
;
4014 DPIF_PORT_FOR_EACH (&dpif_port
, &port_dump
, backer
->dpif
) {
4015 simap_put(&port_names
, dpif_port
.name
,
4016 odp_to_u32(dpif_port
.port_no
));
4020 /* Parse the flow and determine whether a datapath or
4021 * bridge is specified. If function odp_flow_key_from_string()
4022 * returns 0, the flow is a odp_flow. If function
4023 * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
4024 if (!odp_flow_from_string(argv
[argc
- 1], &port_names
,
4025 &odp_key
, &odp_mask
)) {
4027 error
= "Cannot find the datapath";
4031 if (xlate_receive(backer
, NULL
, ofpbuf_data(&odp_key
),
4032 ofpbuf_size(&odp_key
), flow
,
4033 ofprotop
, NULL
, NULL
, NULL
, NULL
)) {
4034 error
= "Invalid datapath flow";
4038 char *err
= parse_ofp_exact_flow(flow
, NULL
, argv
[argc
- 1], NULL
);
4041 m_err
= xasprintf("Bad flow syntax: %s", err
);
4046 error
= "Must specify bridge name";
4050 *ofprotop
= ofproto_dpif_lookup(argv
[1]);
4052 error
= "Unknown bridge name";
4058 /* Generate a packet, if requested. */
4060 if (!ofpbuf_size(packet
)) {
4061 flow_compose(packet
, flow
);
4063 struct pkt_metadata md
= pkt_metadata_from_flow(flow
);
4065 /* Use the metadata from the flow and the packet argument
4066 * to reconstruct the flow. */
4067 flow_extract(packet
, &md
, flow
);
4072 if (error
&& !m_err
) {
4073 m_err
= xstrdup(error
);
4076 ofpbuf_delete(packet
);
4080 ofpbuf_uninit(&odp_key
);
4081 ofpbuf_uninit(&odp_mask
);
4082 simap_destroy(&port_names
);
4087 ofproto_unixctl_trace(struct unixctl_conn
*conn
, int argc
, const char *argv
[],
4088 void *aux OVS_UNUSED
)
4090 struct ofproto_dpif
*ofproto
;
4091 struct ofpbuf
*packet
;
4095 error
= parse_flow_and_packet(argc
, argv
, &ofproto
, &flow
, &packet
);
4100 ofproto_trace(ofproto
, &flow
, packet
, NULL
, 0, &result
);
4101 unixctl_command_reply(conn
, ds_cstr(&result
));
4102 ds_destroy(&result
);
4103 ofpbuf_delete(packet
);
4105 unixctl_command_reply_error(conn
, error
);
4111 ofproto_unixctl_trace_actions(struct unixctl_conn
*conn
, int argc
,
4112 const char *argv
[], void *aux OVS_UNUSED
)
4114 enum ofputil_protocol usable_protocols
;
4115 struct ofproto_dpif
*ofproto
;
4116 bool enforce_consistency
;
4117 struct ofpbuf ofpacts
;
4118 struct ofpbuf
*packet
;
4123 /* Three kinds of error return values! */
4129 ofpbuf_init(&ofpacts
, 0);
4131 /* Parse actions. */
4132 error
= parse_ofpacts(argv
[--argc
], &ofpacts
, &usable_protocols
);
4134 unixctl_command_reply_error(conn
, error
);
4139 /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
4140 * consistency between the flow and the actions. With -consistent, we
4141 * enforce consistency even for a flow supported in OpenFlow 1.0. */
4142 if (!strcmp(argv
[1], "-consistent")) {
4143 enforce_consistency
= true;
4147 enforce_consistency
= false;
4150 error
= parse_flow_and_packet(argc
, argv
, &ofproto
, &flow
, &packet
);
4152 unixctl_command_reply_error(conn
, error
);
4157 /* Do the same checks as handle_packet_out() in ofproto.c.
4159 * We pass a 'table_id' of 0 to ofproto_check_ofpacts(), which isn't
4160 * strictly correct because these actions aren't in any table, but it's OK
4161 * because it 'table_id' is used only to check goto_table instructions, but
4162 * packet-outs take a list of actions and therefore it can't include
4165 * We skip the "meter" check here because meter is an instruction, not an
4166 * action, and thus cannot appear in ofpacts. */
4167 in_port
= ofp_to_u16(flow
.in_port
.ofp_port
);
4168 if (in_port
>= ofproto
->up
.max_ports
&& in_port
< ofp_to_u16(OFPP_MAX
)) {
4169 unixctl_command_reply_error(conn
, "invalid in_port");
4172 if (enforce_consistency
) {
4173 retval
= ofpacts_check_consistency(ofpbuf_data(&ofpacts
), ofpbuf_size(&ofpacts
),
4174 &flow
, u16_to_ofp(ofproto
->up
.max_ports
),
4175 0, 0, usable_protocols
);
4177 retval
= ofpacts_check(ofpbuf_data(&ofpacts
), ofpbuf_size(&ofpacts
), &flow
,
4178 u16_to_ofp(ofproto
->up
.max_ports
), 0, 0,
4184 ds_put_format(&result
, "Bad actions: %s", ofperr_to_string(retval
));
4185 unixctl_command_reply_error(conn
, ds_cstr(&result
));
4189 ofproto_trace(ofproto
, &flow
, packet
,
4190 ofpbuf_data(&ofpacts
), ofpbuf_size(&ofpacts
), &result
);
4191 unixctl_command_reply(conn
, ds_cstr(&result
));
4194 ds_destroy(&result
);
4195 ofpbuf_delete(packet
);
4196 ofpbuf_uninit(&ofpacts
);
4199 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
4200 * description of the results to 'ds'.
4202 * The trace follows a packet with the specified 'flow' through the flow
4203 * table. 'packet' may be nonnull to trace an actual packet, with consequent
4204 * side effects (if it is nonnull then its flow must be 'flow').
4206 * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
4207 * trace, otherwise the actions are determined by a flow table lookup. */
4209 ofproto_trace(struct ofproto_dpif
*ofproto
, struct flow
*flow
,
4210 const struct ofpbuf
*packet
,
4211 const struct ofpact ofpacts
[], size_t ofpacts_len
,
4214 struct rule_dpif
*rule
;
4215 struct trace_ctx trace
;
4217 ds_put_format(ds
, "Bridge: %s\n", ofproto
->up
.name
);
4218 ds_put_cstr(ds
, "Flow: ");
4219 flow_format(ds
, flow
);
4220 ds_put_char(ds
, '\n');
4222 flow_wildcards_init_catchall(&trace
.wc
);
4226 rule_dpif_lookup(ofproto
, flow
, &trace
.wc
, &rule
, false, NULL
);
4228 trace_format_rule(ds
, 0, rule
);
4229 if (rule
== ofproto
->miss_rule
) {
4230 ds_put_cstr(ds
, "\nNo match, flow generates \"packet in\"s.\n");
4231 } else if (rule
== ofproto
->no_packet_in_rule
) {
4232 ds_put_cstr(ds
, "\nNo match, packets dropped because "
4233 "OFPPC_NO_PACKET_IN is set on in_port.\n");
4234 } else if (rule
== ofproto
->drop_frags_rule
) {
4235 ds_put_cstr(ds
, "\nPackets dropped because they are IP fragments "
4236 "and the fragment handling mode is \"drop\".\n");
4240 if (rule
|| ofpacts
) {
4242 trace
.key
= flow
; /* Original flow key, used for megaflow. */
4243 trace
.flow
= *flow
; /* May be modified by actions. */
4244 xlate_in_init(&trace
.xin
, ofproto
, flow
, rule
, ntohs(flow
->tcp_flags
),
4247 trace
.xin
.ofpacts
= ofpacts
;
4248 trace
.xin
.ofpacts_len
= ofpacts_len
;
4250 trace
.xin
.resubmit_hook
= trace_resubmit
;
4251 trace
.xin
.report_hook
= trace_report
;
4253 xlate_actions(&trace
.xin
, &trace
.xout
);
4255 ds_put_char(ds
, '\n');
4256 trace_format_flow(ds
, 0, "Final flow", &trace
);
4257 trace_format_megaflow(ds
, 0, "Megaflow", &trace
);
4259 ds_put_cstr(ds
, "Datapath actions: ");
4260 format_odp_actions(ds
, ofpbuf_data(&trace
.xout
.odp_actions
),
4261 ofpbuf_size(&trace
.xout
.odp_actions
));
4263 if (trace
.xout
.slow
) {
4264 enum slow_path_reason slow
;
4266 ds_put_cstr(ds
, "\nThis flow is handled by the userspace "
4267 "slow path because it:");
4269 slow
= trace
.xout
.slow
;
4271 enum slow_path_reason bit
= rightmost_1bit(slow
);
4273 ds_put_format(ds
, "\n\t- %s.",
4274 slow_path_reason_to_explanation(bit
));
4280 xlate_out_uninit(&trace
.xout
);
4284 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
4285 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
4286 * to destroy 'ofproto_shash' and free the returned value. */
4287 static const struct shash_node
**
4288 get_ofprotos(struct shash
*ofproto_shash
)
4290 const struct ofproto_dpif
*ofproto
;
4292 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
4293 char *name
= xasprintf("%s@%s", ofproto
->up
.type
, ofproto
->up
.name
);
4294 shash_add_nocopy(ofproto_shash
, name
, ofproto
);
4297 return shash_sort(ofproto_shash
);
4301 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn
*conn
, int argc OVS_UNUSED
,
4302 const char *argv
[] OVS_UNUSED
,
4303 void *aux OVS_UNUSED
)
4305 struct ds ds
= DS_EMPTY_INITIALIZER
;
4306 struct shash ofproto_shash
;
4307 const struct shash_node
**sorted_ofprotos
;
4310 shash_init(&ofproto_shash
);
4311 sorted_ofprotos
= get_ofprotos(&ofproto_shash
);
4312 for (i
= 0; i
< shash_count(&ofproto_shash
); i
++) {
4313 const struct shash_node
*node
= sorted_ofprotos
[i
];
4314 ds_put_format(&ds
, "%s\n", node
->name
);
4317 shash_destroy(&ofproto_shash
);
4318 free(sorted_ofprotos
);
4320 unixctl_command_reply(conn
, ds_cstr(&ds
));
4325 dpif_show_backer(const struct dpif_backer
*backer
, struct ds
*ds
)
4327 const struct shash_node
**ofprotos
;
4328 struct dpif_dp_stats dp_stats
;
4329 struct shash ofproto_shash
;
4332 dpif_get_dp_stats(backer
->dpif
, &dp_stats
);
4334 ds_put_format(ds
, "%s: hit:%"PRIu64
" missed:%"PRIu64
"\n",
4335 dpif_name(backer
->dpif
), dp_stats
.n_hit
, dp_stats
.n_missed
);
4337 shash_init(&ofproto_shash
);
4338 ofprotos
= get_ofprotos(&ofproto_shash
);
4339 for (i
= 0; i
< shash_count(&ofproto_shash
); i
++) {
4340 struct ofproto_dpif
*ofproto
= ofprotos
[i
]->data
;
4341 const struct shash_node
**ports
;
4344 if (ofproto
->backer
!= backer
) {
4348 ds_put_format(ds
, "\t%s:\n", ofproto
->up
.name
);
4350 ports
= shash_sort(&ofproto
->up
.port_by_name
);
4351 for (j
= 0; j
< shash_count(&ofproto
->up
.port_by_name
); j
++) {
4352 const struct shash_node
*node
= ports
[j
];
4353 struct ofport
*ofport
= node
->data
;
4355 odp_port_t odp_port
;
4357 ds_put_format(ds
, "\t\t%s %u/", netdev_get_name(ofport
->netdev
),
4360 odp_port
= ofp_port_to_odp_port(ofproto
, ofport
->ofp_port
);
4361 if (odp_port
!= ODPP_NONE
) {
4362 ds_put_format(ds
, "%"PRIu32
":", odp_port
);
4364 ds_put_cstr(ds
, "none:");
4367 ds_put_format(ds
, " (%s", netdev_get_type(ofport
->netdev
));
4370 if (!netdev_get_config(ofport
->netdev
, &config
)) {
4371 const struct smap_node
**nodes
;
4374 nodes
= smap_sort(&config
);
4375 for (i
= 0; i
< smap_count(&config
); i
++) {
4376 const struct smap_node
*node
= nodes
[i
];
4377 ds_put_format(ds
, "%c %s=%s", i
? ',' : ':',
4378 node
->key
, node
->value
);
4382 smap_destroy(&config
);
4384 ds_put_char(ds
, ')');
4385 ds_put_char(ds
, '\n');
4389 shash_destroy(&ofproto_shash
);
4394 ofproto_unixctl_dpif_show(struct unixctl_conn
*conn
, int argc OVS_UNUSED
,
4395 const char *argv
[] OVS_UNUSED
, void *aux OVS_UNUSED
)
4397 struct ds ds
= DS_EMPTY_INITIALIZER
;
4398 const struct shash_node
**backers
;
4401 backers
= shash_sort(&all_dpif_backers
);
4402 for (i
= 0; i
< shash_count(&all_dpif_backers
); i
++) {
4403 dpif_show_backer(backers
[i
]->data
, &ds
);
4407 unixctl_command_reply(conn
, ds_cstr(&ds
));
4412 ofproto_dpif_contains_flow(const struct ofproto_dpif
*ofproto
,
4413 const struct nlattr
*key
, size_t key_len
)
4415 struct ofproto_dpif
*ofp
;
4418 xlate_receive(ofproto
->backer
, NULL
, key
, key_len
, &flow
, &ofp
,
4419 NULL
, NULL
, NULL
, NULL
);
4420 return ofp
== ofproto
;
4424 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn
*conn
,
4425 int argc OVS_UNUSED
, const char *argv
[],
4426 void *aux OVS_UNUSED
)
4428 const struct ofproto_dpif
*ofproto
;
4430 struct ds ds
= DS_EMPTY_INITIALIZER
;
4431 bool verbosity
= false;
4433 struct dpif_port dpif_port
;
4434 struct dpif_port_dump port_dump
;
4435 struct hmap portno_names
;
4437 struct dpif_flow_dump
*flow_dump
;
4438 struct dpif_flow_dump_thread
*flow_dump_thread
;
4442 ofproto
= ofproto_dpif_lookup(argv
[argc
- 1]);
4444 unixctl_command_reply_error(conn
, "no such bridge");
4448 if (argc
> 2 && !strcmp(argv
[1], "-m")) {
4452 hmap_init(&portno_names
);
4453 DPIF_PORT_FOR_EACH (&dpif_port
, &port_dump
, ofproto
->backer
->dpif
) {
4454 odp_portno_names_set(&portno_names
, dpif_port
.port_no
, dpif_port
.name
);
4458 flow_dump
= dpif_flow_dump_create(ofproto
->backer
->dpif
);
4459 flow_dump_thread
= dpif_flow_dump_thread_create(flow_dump
);
4460 while (dpif_flow_dump_next(flow_dump_thread
, &f
, 1)) {
4461 if (!ofproto_dpif_contains_flow(ofproto
, f
.key
, f
.key_len
)) {
4465 odp_flow_format(f
.key
, f
.key_len
, f
.mask
, f
.mask_len
,
4466 &portno_names
, &ds
, verbosity
);
4467 ds_put_cstr(&ds
, ", ");
4468 dpif_flow_stats_format(&f
.stats
, &ds
);
4469 ds_put_cstr(&ds
, ", actions:");
4470 format_odp_actions(&ds
, f
.actions
, f
.actions_len
);
4471 ds_put_char(&ds
, '\n');
4473 dpif_flow_dump_thread_destroy(flow_dump_thread
);
4474 error
= dpif_flow_dump_destroy(flow_dump
);
4478 ds_put_format(&ds
, "dpif/dump_flows failed: %s", ovs_strerror(errno
));
4479 unixctl_command_reply_error(conn
, ds_cstr(&ds
));
4481 unixctl_command_reply(conn
, ds_cstr(&ds
));
4483 odp_portno_names_destroy(&portno_names
);
4484 hmap_destroy(&portno_names
);
4489 ofproto_dpif_unixctl_init(void)
4491 static bool registered
;
4497 unixctl_command_register(
4499 "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
4500 1, 3, ofproto_unixctl_trace
, NULL
);
4501 unixctl_command_register(
4502 "ofproto/trace-packet-out",
4503 "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
4504 2, 6, ofproto_unixctl_trace_actions
, NULL
);
4505 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
4506 ofproto_unixctl_fdb_flush
, NULL
);
4507 unixctl_command_register("fdb/show", "bridge", 1, 1,
4508 ofproto_unixctl_fdb_show
, NULL
);
4509 unixctl_command_register("dpif/dump-dps", "", 0, 0,
4510 ofproto_unixctl_dpif_dump_dps
, NULL
);
4511 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show
,
4513 unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
4514 ofproto_unixctl_dpif_dump_flows
, NULL
);
4517 /* Returns true if 'table' is the table used for internal rules,
4518 * false otherwise. */
4520 table_is_internal(uint8_t table_id
)
4522 return table_id
== TBL_INTERNAL
;
4525 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4527 * This is deprecated. It is only for compatibility with broken device drivers
4528 * in old versions of Linux that do not properly support VLANs when VLAN
4529 * devices are not used. When broken device drivers are no longer in
4530 * widespread use, we will delete these interfaces. */
4533 set_realdev(struct ofport
*ofport_
, ofp_port_t realdev_ofp_port
, int vid
)
4535 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport_
->ofproto
);
4536 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
4538 if (realdev_ofp_port
== ofport
->realdev_ofp_port
4539 && vid
== ofport
->vlandev_vid
) {
4543 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
4545 if (ofport
->realdev_ofp_port
) {
4548 if (realdev_ofp_port
&& ofport
->bundle
) {
4549 /* vlandevs are enslaved to their realdevs, so they are not allowed to
4550 * themselves be part of a bundle. */
4551 bundle_set(ofport_
->ofproto
, ofport
->bundle
, NULL
);
4554 ofport
->realdev_ofp_port
= realdev_ofp_port
;
4555 ofport
->vlandev_vid
= vid
;
4557 if (realdev_ofp_port
) {
4558 vsp_add(ofport
, realdev_ofp_port
, vid
);
4565 hash_realdev_vid(ofp_port_t realdev_ofp_port
, int vid
)
4567 return hash_2words(ofp_to_u16(realdev_ofp_port
), vid
);
4571 ofproto_has_vlan_splinters(const struct ofproto_dpif
*ofproto
)
4572 OVS_EXCLUDED(ofproto
->vsp_mutex
)
4574 /* hmap_is_empty is thread safe. */
4575 return !hmap_is_empty(&ofproto
->realdev_vid_map
);
4580 vsp_realdev_to_vlandev__(const struct ofproto_dpif
*ofproto
,
4581 ofp_port_t realdev_ofp_port
, ovs_be16 vlan_tci
)
4582 OVS_REQUIRES(ofproto
->vsp_mutex
)
4584 if (!hmap_is_empty(&ofproto
->realdev_vid_map
)) {
4585 int vid
= vlan_tci_to_vid(vlan_tci
);
4586 const struct vlan_splinter
*vsp
;
4588 HMAP_FOR_EACH_WITH_HASH (vsp
, realdev_vid_node
,
4589 hash_realdev_vid(realdev_ofp_port
, vid
),
4590 &ofproto
->realdev_vid_map
) {
4591 if (vsp
->realdev_ofp_port
== realdev_ofp_port
4592 && vsp
->vid
== vid
) {
4593 return vsp
->vlandev_ofp_port
;
4597 return realdev_ofp_port
;
4600 /* Returns the OFP port number of the Linux VLAN device that corresponds to
4601 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
4602 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
4603 * 'vlan_tci' 9, it would return the port number of eth0.9.
4605 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
4606 * function just returns its 'realdev_ofp_port' argument. */
4608 vsp_realdev_to_vlandev(const struct ofproto_dpif
*ofproto
,
4609 ofp_port_t realdev_ofp_port
, ovs_be16 vlan_tci
)
4610 OVS_EXCLUDED(ofproto
->vsp_mutex
)
4614 /* hmap_is_empty is thread safe, see if we can return immediately. */
4615 if (hmap_is_empty(&ofproto
->realdev_vid_map
)) {
4616 return realdev_ofp_port
;
4618 ovs_mutex_lock(&ofproto
->vsp_mutex
);
4619 ret
= vsp_realdev_to_vlandev__(ofproto
, realdev_ofp_port
, vlan_tci
);
4620 ovs_mutex_unlock(&ofproto
->vsp_mutex
);
4624 static struct vlan_splinter
*
4625 vlandev_find(const struct ofproto_dpif
*ofproto
, ofp_port_t vlandev_ofp_port
)
4627 struct vlan_splinter
*vsp
;
4629 HMAP_FOR_EACH_WITH_HASH (vsp
, vlandev_node
,
4630 hash_ofp_port(vlandev_ofp_port
),
4631 &ofproto
->vlandev_map
) {
4632 if (vsp
->vlandev_ofp_port
== vlandev_ofp_port
) {
4640 /* Returns the OpenFlow port number of the "real" device underlying the Linux
4641 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
4642 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
4643 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
4644 * eth0 and store 9 in '*vid'.
4646 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
4647 * VLAN device. Unless VLAN splinters are enabled, this is what this function
4650 vsp_vlandev_to_realdev(const struct ofproto_dpif
*ofproto
,
4651 ofp_port_t vlandev_ofp_port
, int *vid
)
4652 OVS_REQUIRES(ofproto
->vsp_mutex
)
4654 if (!hmap_is_empty(&ofproto
->vlandev_map
)) {
4655 const struct vlan_splinter
*vsp
;
4657 vsp
= vlandev_find(ofproto
, vlandev_ofp_port
);
4662 return vsp
->realdev_ofp_port
;
4668 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
4669 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
4670 * 'flow->in_port' to the "real" device backing the VLAN device, sets
4671 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
4672 * always the case unless VLAN splinters are enabled), returns false without
4673 * making any changes. */
4675 vsp_adjust_flow(const struct ofproto_dpif
*ofproto
, struct flow
*flow
)
4676 OVS_EXCLUDED(ofproto
->vsp_mutex
)
4681 /* hmap_is_empty is thread safe. */
4682 if (hmap_is_empty(&ofproto
->vlandev_map
)) {
4686 ovs_mutex_lock(&ofproto
->vsp_mutex
);
4687 realdev
= vsp_vlandev_to_realdev(ofproto
, flow
->in_port
.ofp_port
, &vid
);
4688 ovs_mutex_unlock(&ofproto
->vsp_mutex
);
4693 /* Cause the flow to be processed as if it came in on the real device with
4694 * the VLAN device's VLAN ID. */
4695 flow
->in_port
.ofp_port
= realdev
;
4696 flow
->vlan_tci
= htons((vid
& VLAN_VID_MASK
) | VLAN_CFI
);
4701 vsp_remove(struct ofport_dpif
*port
)
4703 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
4704 struct vlan_splinter
*vsp
;
4706 ovs_mutex_lock(&ofproto
->vsp_mutex
);
4707 vsp
= vlandev_find(ofproto
, port
->up
.ofp_port
);
4709 hmap_remove(&ofproto
->vlandev_map
, &vsp
->vlandev_node
);
4710 hmap_remove(&ofproto
->realdev_vid_map
, &vsp
->realdev_vid_node
);
4713 port
->realdev_ofp_port
= 0;
4715 VLOG_ERR("missing vlan device record");
4717 ovs_mutex_unlock(&ofproto
->vsp_mutex
);
4721 vsp_add(struct ofport_dpif
*port
, ofp_port_t realdev_ofp_port
, int vid
)
4723 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
4725 ovs_mutex_lock(&ofproto
->vsp_mutex
);
4726 if (!vsp_vlandev_to_realdev(ofproto
, port
->up
.ofp_port
, NULL
)
4727 && (vsp_realdev_to_vlandev__(ofproto
, realdev_ofp_port
, htons(vid
))
4728 == realdev_ofp_port
)) {
4729 struct vlan_splinter
*vsp
;
4731 vsp
= xmalloc(sizeof *vsp
);
4732 vsp
->realdev_ofp_port
= realdev_ofp_port
;
4733 vsp
->vlandev_ofp_port
= port
->up
.ofp_port
;
4736 port
->realdev_ofp_port
= realdev_ofp_port
;
4738 hmap_insert(&ofproto
->vlandev_map
, &vsp
->vlandev_node
,
4739 hash_ofp_port(port
->up
.ofp_port
));
4740 hmap_insert(&ofproto
->realdev_vid_map
, &vsp
->realdev_vid_node
,
4741 hash_realdev_vid(realdev_ofp_port
, vid
));
4743 VLOG_ERR("duplicate vlan device record");
4745 ovs_mutex_unlock(&ofproto
->vsp_mutex
);
4749 ofp_port_to_odp_port(const struct ofproto_dpif
*ofproto
, ofp_port_t ofp_port
)
4751 const struct ofport_dpif
*ofport
= get_ofp_port(ofproto
, ofp_port
);
4752 return ofport
? ofport
->odp_port
: ODPP_NONE
;
4755 struct ofport_dpif
*
4756 odp_port_to_ofport(const struct dpif_backer
*backer
, odp_port_t odp_port
)
4758 struct ofport_dpif
*port
;
4760 ovs_rwlock_rdlock(&backer
->odp_to_ofport_lock
);
4761 HMAP_FOR_EACH_IN_BUCKET (port
, odp_port_node
, hash_odp_port(odp_port
),
4762 &backer
->odp_to_ofport_map
) {
4763 if (port
->odp_port
== odp_port
) {
4764 ovs_rwlock_unlock(&backer
->odp_to_ofport_lock
);
4769 ovs_rwlock_unlock(&backer
->odp_to_ofport_lock
);
4774 odp_port_to_ofp_port(const struct ofproto_dpif
*ofproto
, odp_port_t odp_port
)
4776 struct ofport_dpif
*port
;
4778 port
= odp_port_to_ofport(ofproto
->backer
, odp_port
);
4779 if (port
&& &ofproto
->up
== port
->up
.ofproto
) {
4780 return port
->up
.ofp_port
;
4787 ofproto_dpif_alloc_recirc_id(struct ofproto_dpif
*ofproto
)
4789 struct dpif_backer
*backer
= ofproto
->backer
;
4791 return recirc_id_alloc(backer
->rid_pool
);
4795 ofproto_dpif_free_recirc_id(struct ofproto_dpif
*ofproto
, uint32_t recirc_id
)
4797 struct dpif_backer
*backer
= ofproto
->backer
;
4799 recirc_id_free(backer
->rid_pool
, recirc_id
);
4803 ofproto_dpif_add_internal_flow(struct ofproto_dpif
*ofproto
,
4804 const struct match
*match
, int priority
,
4805 const struct ofpbuf
*ofpacts
,
4806 struct rule
**rulep
)
4808 struct ofputil_flow_mod fm
;
4809 struct rule_dpif
*rule
;
4813 fm
.priority
= priority
;
4814 fm
.new_cookie
= htonll(0);
4815 fm
.cookie
= htonll(0);
4816 fm
.cookie_mask
= htonll(0);
4817 fm
.modify_cookie
= false;
4818 fm
.table_id
= TBL_INTERNAL
;
4819 fm
.command
= OFPFC_ADD
;
4820 fm
.idle_timeout
= 0;
4821 fm
.hard_timeout
= 0;
4824 fm
.flags
= OFPUTIL_FF_HIDDEN_FIELDS
| OFPUTIL_FF_NO_READONLY
;
4825 fm
.ofpacts
= ofpbuf_data(ofpacts
);
4826 fm
.ofpacts_len
= ofpbuf_size(ofpacts
);
4828 error
= ofproto_flow_mod(&ofproto
->up
, &fm
);
4830 VLOG_ERR_RL(&rl
, "failed to add internal flow (%s)",
4831 ofperr_to_string(error
));
4836 rule
= rule_dpif_lookup_in_table(ofproto
, TBL_INTERNAL
, &fm
.match
.flow
,
4837 &fm
.match
.wc
, false);
4847 ofproto_dpif_delete_internal_flow(struct ofproto_dpif
*ofproto
,
4848 struct match
*match
, int priority
)
4850 struct ofputil_flow_mod fm
;
4854 fm
.priority
= priority
;
4855 fm
.new_cookie
= htonll(0);
4856 fm
.cookie
= htonll(0);
4857 fm
.cookie_mask
= htonll(0);
4858 fm
.modify_cookie
= false;
4859 fm
.table_id
= TBL_INTERNAL
;
4860 fm
.flags
= OFPUTIL_FF_HIDDEN_FIELDS
| OFPUTIL_FF_NO_READONLY
;
4861 fm
.command
= OFPFC_DELETE_STRICT
;
4863 error
= ofproto_flow_mod(&ofproto
->up
, &fm
);
4865 VLOG_ERR_RL(&rl
, "failed to delete internal flow (%s)",
4866 ofperr_to_string(error
));
4873 const struct ofproto_class ofproto_dpif_class
= {
4887 NULL
, /* get_memory_usage. */
4888 type_get_memory_usage
,
4907 port_is_lacp_current
,
4908 NULL
, /* rule_choose_table */
4917 rule_modify_actions
,
4931 get_stp_port_status
,
4939 is_mirror_output_bundle
,
4940 forward_bpdu_changed
,
4941 set_mac_table_config
,
4943 NULL
, /* meter_get_features */
4944 NULL
, /* meter_set */
4945 NULL
, /* meter_get */
4946 NULL
, /* meter_del */
4947 group_alloc
, /* group_alloc */
4948 group_construct
, /* group_construct */
4949 group_destruct
, /* group_destruct */
4950 group_dealloc
, /* group_dealloc */
4951 group_modify
, /* group_modify */
4952 group_get_stats
, /* group_get_stats */