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 "mcast-snooping.h"
41 #include "meta-flow.h"
42 #include "multipath.h"
43 #include "netdev-vport.h"
48 #include "odp-execute.h"
51 #include "ofp-actions.h"
52 #include "ofp-parse.h"
53 #include "ofp-print.h"
54 #include "ofproto-dpif-ipfix.h"
55 #include "ofproto-dpif-mirror.h"
56 #include "ofproto-dpif-monitor.h"
57 #include "ofproto-dpif-rid.h"
58 #include "ofproto-dpif-sflow.h"
59 #include "ofproto-dpif-upcall.h"
60 #include "ofproto-dpif-xlate.h"
61 #include "poll-loop.h"
67 #include "unaligned.h"
69 #include "vlan-bitmap.h"
72 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
74 COVERAGE_DEFINE(ofproto_dpif_expired);
75 COVERAGE_DEFINE(packet_in_overflow);
84 * - Do include packets and bytes from datapath flows which have not
85 * recently been processed by a revalidator. */
86 struct ovs_mutex stats_mutex;
87 struct dpif_flow_stats stats OVS_GUARDED;
89 /* If non-zero then the recirculation id that has
90 * been allocated for use with this rule.
91 * The recirculation id and associated internal flow should
92 * be freed when the rule is freed */
96 /* RULE_CAST() depends on this. */
97 BUILD_ASSERT_DECL(offsetof(struct rule_dpif, up) == 0);
99 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes,
100 long long int *used);
101 static struct rule_dpif *rule_dpif_cast(const struct rule *);
102 static void rule_expire(struct rule_dpif *);
109 * - Do include packets and bytes from datapath flows which have not
110 * recently been processed by a revalidator. */
111 struct ovs_mutex stats_mutex;
112 uint64_t packet_count OVS_GUARDED; /* Number of packets received. */
113 uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */
117 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
118 struct ofproto_dpif *ofproto; /* Owning ofproto. */
119 void *aux; /* Key supplied by ofproto's client. */
120 char *name; /* Identifier for log messages. */
123 struct list ports; /* Contains "struct ofport"s. */
124 enum port_vlan_mode vlan_mode; /* VLAN mode */
125 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
126 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
127 * NULL if all VLANs are trunked. */
128 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
129 struct bond *bond; /* Nonnull iff more than one port. */
130 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
133 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
136 static void bundle_remove(struct ofport *);
137 static void bundle_update(struct ofbundle *);
138 static void bundle_destroy(struct ofbundle *);
139 static void bundle_del_port(struct ofport_dpif *);
140 static void bundle_run(struct ofbundle *);
141 static void bundle_wait(struct ofbundle *);
143 static void stp_run(struct ofproto_dpif *ofproto);
144 static void stp_wait(struct ofproto_dpif *ofproto);
145 static int set_stp_port(struct ofport *,
146 const struct ofproto_port_stp_settings *);
149 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
153 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
154 struct list bundle_node; /* In struct ofbundle's "ports" list. */
155 struct cfm *cfm; /* Connectivity Fault Management, if any. */
156 struct bfd *bfd; /* BFD, if any. */
157 bool may_enable; /* May be enabled in bonds. */
158 bool is_tunnel; /* This port is a tunnel. */
159 bool is_layer3; /* This is a layer 3 port. */
160 long long int carrier_seq; /* Carrier status changes. */
161 struct ofport_dpif *peer; /* Peer if patch port. */
164 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
165 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
166 long long int stp_state_entered;
168 /* Queue to DSCP mapping. */
169 struct ofproto_port_queue *qdscp;
172 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
174 * This is deprecated. It is only for compatibility with broken device
175 * drivers in old versions of Linux that do not properly support VLANs when
176 * VLAN devices are not used. When broken device drivers are no longer in
177 * widespread use, we will delete these interfaces. */
178 ofp_port_t realdev_ofp_port;
182 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
184 * This is deprecated. It is only for compatibility with broken device drivers
185 * in old versions of Linux that do not properly support VLANs when VLAN
186 * devices are not used. When broken device drivers are no longer in
187 * widespread use, we will delete these interfaces. */
188 struct vlan_splinter {
189 struct hmap_node realdev_vid_node;
190 struct hmap_node vlandev_node;
191 ofp_port_t realdev_ofp_port;
192 ofp_port_t vlandev_ofp_port;
196 static void vsp_remove(struct ofport_dpif *);
197 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
199 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
202 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
205 static struct ofport_dpif *
206 ofport_dpif_cast(const struct ofport *ofport)
208 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
211 static void port_run(struct ofport_dpif *);
212 static int set_bfd(struct ofport *, const struct smap *);
213 static int set_cfm(struct ofport *, const struct cfm_settings *);
214 static void ofport_update_peer(struct ofport_dpif *);
216 /* Reasons that we might need to revalidate every datapath flow, and
217 * corresponding coverage counters.
219 * A value of 0 means that there is no need to revalidate.
221 * It would be nice to have some cleaner way to integrate with coverage
222 * counters, but with only a few reasons I guess this is good enough for
224 enum revalidate_reason {
225 REV_RECONFIGURE = 1, /* Switch configuration changed. */
226 REV_STP, /* Spanning tree protocol port status change. */
227 REV_BOND, /* Bonding changed. */
228 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
229 REV_FLOW_TABLE, /* Flow table changed. */
230 REV_MAC_LEARNING, /* Mac learning changed. */
231 REV_MCAST_SNOOPING, /* Multicast snooping 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);
239 COVERAGE_DEFINE(rev_mcast_snooping);
241 /* All datapaths of a given type share a single dpif backer instance. */
248 struct ovs_rwlock odp_to_ofport_lock;
249 struct hmap odp_to_ofport_map OVS_GUARDED; /* Contains "struct ofport"s. */
251 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
253 enum revalidate_reason need_revalidate; /* Revalidate all flows. */
255 bool recv_set_enable; /* Enables or disables receiving packets. */
258 struct recirc_id_pool *rid_pool; /* Recirculation ID pool. */
259 bool enable_recirc; /* True if the datapath supports recirculation */
261 /* True if the datapath supports variable-length
262 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.
263 * False if the datapath supports only 8-byte (or shorter) userdata. */
264 bool variable_length_userdata;
266 /* Maximum number of MPLS label stack entries that the datapath supports
268 size_t max_mpls_depth;
271 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
272 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
274 struct ofproto_dpif {
275 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
277 struct dpif_backer *backer;
279 uint64_t dump_seq; /* Last read of udpif_dump_seq(). */
281 /* Special OpenFlow rules. */
282 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
283 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
284 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
287 struct netflow *netflow;
288 struct dpif_sflow *sflow;
289 struct dpif_ipfix *ipfix;
290 struct hmap bundles; /* Contains "struct ofbundle"s. */
291 struct mac_learning *ml;
292 struct mcast_snooping *ms;
293 bool has_bonded_bundles;
295 struct mbridge *mbridge;
297 struct ovs_mutex stats_mutex;
298 struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
299 * consumed in userspace. */
303 long long int stp_last_tick;
305 /* VLAN splinters. */
306 struct ovs_mutex vsp_mutex;
307 struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
308 struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */
311 struct sset ports; /* Set of standard port names. */
312 struct sset ghost_ports; /* Ports with no datapath port. */
313 struct sset port_poll_set; /* Queued names for port_poll() reply. */
314 int port_poll_errno; /* Last errno for port_poll() reply. */
315 uint64_t change_seq; /* Connectivity status changes. */
318 struct guarded_list pins; /* Contains "struct ofputil_packet_in"s. */
319 struct seq *pins_seq; /* For notifying 'pins' reception. */
323 /* All existing ofproto_dpif instances, indexed by ->up.name. */
324 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
326 static void ofproto_dpif_unixctl_init(void);
328 static inline struct ofproto_dpif *
329 ofproto_dpif_cast(const struct ofproto *ofproto)
331 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
332 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
336 ofproto_dpif_get_max_mpls_depth(const struct ofproto_dpif *ofproto)
338 return ofproto->backer->max_mpls_depth;
342 ofproto_dpif_get_enable_recirc(const struct ofproto_dpif *ofproto)
344 return ofproto->backer->enable_recirc;
347 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
348 ofp_port_t ofp_port);
349 static void ofproto_trace(struct ofproto_dpif *, struct flow *,
350 const struct ofpbuf *packet,
351 const struct ofpact[], size_t ofpacts_len,
354 /* Global variables. */
355 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
357 /* Initial mappings of port to bridge mappings. */
358 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
360 /* Executes 'fm'. The caller retains ownership of 'fm' and everything in
363 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
364 struct ofputil_flow_mod *fm)
366 ofproto_flow_mod(&ofproto->up, fm);
369 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
370 * Takes ownership of 'pin' and pin->packet. */
372 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
373 struct ofproto_packet_in *pin)
375 if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
376 COVERAGE_INC(packet_in_overflow);
377 free(CONST_CAST(void *, pin->up.packet));
381 /* Wakes up main thread for packet-in I/O. */
382 seq_change(ofproto->pins_seq);
385 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
386 * packet rather than to send the packet to the controller.
388 * This function returns false to indicate that a packet_in message
389 * for a "table-miss" should be sent to at least one controller.
390 * False otherwise. */
392 ofproto_dpif_wants_packet_in_on_miss(struct ofproto_dpif *ofproto)
394 return connmgr_wants_packet_in_on_miss(ofproto->up.connmgr);
397 /* Factory functions. */
400 init(const struct shash *iface_hints)
402 struct shash_node *node;
404 /* Make a local copy, since we don't own 'iface_hints' elements. */
405 SHASH_FOR_EACH(node, iface_hints) {
406 const struct iface_hint *orig_hint = node->data;
407 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
409 new_hint->br_name = xstrdup(orig_hint->br_name);
410 new_hint->br_type = xstrdup(orig_hint->br_type);
411 new_hint->ofp_port = orig_hint->ofp_port;
413 shash_add(&init_ofp_ports, node->name, new_hint);
418 enumerate_types(struct sset *types)
420 dp_enumerate_types(types);
424 enumerate_names(const char *type, struct sset *names)
426 struct ofproto_dpif *ofproto;
429 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
430 if (strcmp(type, ofproto->up.type)) {
433 sset_add(names, ofproto->up.name);
440 del(const char *type, const char *name)
445 error = dpif_open(name, type, &dpif);
447 error = dpif_delete(dpif);
454 port_open_type(const char *datapath_type, const char *port_type)
456 return dpif_port_open_type(datapath_type, port_type);
459 /* Type functions. */
461 static void process_dpif_port_changes(struct dpif_backer *);
462 static void process_dpif_all_ports_changed(struct dpif_backer *);
463 static void process_dpif_port_change(struct dpif_backer *,
464 const char *devname);
465 static void process_dpif_port_error(struct dpif_backer *, int error);
467 static struct ofproto_dpif *
468 lookup_ofproto_dpif_by_port_name(const char *name)
470 struct ofproto_dpif *ofproto;
472 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
473 if (sset_contains(&ofproto->ports, name)) {
482 type_run(const char *type)
484 struct dpif_backer *backer;
486 backer = shash_find_data(&all_dpif_backers, type);
488 /* This is not necessarily a problem, since backers are only
489 * created on demand. */
493 dpif_run(backer->dpif);
494 udpif_run(backer->udpif);
496 /* If vswitchd started with other_config:flow_restore_wait set as "true",
497 * and the configuration has now changed to "false", enable receiving
498 * packets from the datapath. */
499 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
502 backer->recv_set_enable = true;
504 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
506 VLOG_ERR("Failed to enable receiving packets in dpif.");
509 dpif_flow_flush(backer->dpif);
510 backer->need_revalidate = REV_RECONFIGURE;
513 if (backer->recv_set_enable) {
514 udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
517 if (backer->need_revalidate) {
518 struct ofproto_dpif *ofproto;
519 struct simap_node *node;
520 struct simap tmp_backers;
522 /* Handle tunnel garbage collection. */
523 simap_init(&tmp_backers);
524 simap_swap(&backer->tnl_backers, &tmp_backers);
526 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
527 struct ofport_dpif *iter;
529 if (backer != ofproto->backer) {
533 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
534 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
537 if (!iter->is_tunnel) {
541 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
542 namebuf, sizeof namebuf);
543 node = simap_find(&tmp_backers, dp_port);
545 simap_put(&backer->tnl_backers, dp_port, node->data);
546 simap_delete(&tmp_backers, node);
547 node = simap_find(&backer->tnl_backers, dp_port);
549 node = simap_find(&backer->tnl_backers, dp_port);
551 odp_port_t odp_port = ODPP_NONE;
553 if (!dpif_port_add(backer->dpif, iter->up.netdev,
555 simap_put(&backer->tnl_backers, dp_port,
556 odp_to_u32(odp_port));
557 node = simap_find(&backer->tnl_backers, dp_port);
562 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
563 if (tnl_port_reconfigure(iter, iter->up.netdev,
565 backer->need_revalidate = REV_RECONFIGURE;
570 SIMAP_FOR_EACH (node, &tmp_backers) {
571 dpif_port_del(backer->dpif, u32_to_odp(node->data));
573 simap_destroy(&tmp_backers);
575 switch (backer->need_revalidate) {
576 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
577 case REV_STP: COVERAGE_INC(rev_stp); break;
578 case REV_BOND: COVERAGE_INC(rev_bond); break;
579 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
580 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
581 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
582 case REV_MCAST_SNOOPING: COVERAGE_INC(rev_mcast_snooping); break;
584 backer->need_revalidate = 0;
586 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
587 struct ofport_dpif *ofport;
588 struct ofbundle *bundle;
590 if (ofproto->backer != backer) {
595 xlate_ofproto_set(ofproto, ofproto->up.name,
596 ofproto->backer->dpif, ofproto->miss_rule,
597 ofproto->no_packet_in_rule, ofproto->ml,
598 ofproto->stp, ofproto->ms, ofproto->mbridge,
599 ofproto->sflow, ofproto->ipfix,
600 ofproto->netflow, ofproto->up.frag_handling,
601 ofproto->up.forward_bpdu,
602 connmgr_has_in_band(ofproto->up.connmgr),
603 ofproto->backer->enable_recirc,
604 ofproto->backer->variable_length_userdata,
605 ofproto->backer->max_mpls_depth);
607 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
608 xlate_bundle_set(ofproto, bundle, bundle->name,
609 bundle->vlan_mode, bundle->vlan,
610 bundle->trunks, bundle->use_priority_tags,
611 bundle->bond, bundle->lacp,
615 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
616 int stp_port = ofport->stp_port
617 ? stp_port_no(ofport->stp_port)
619 xlate_ofport_set(ofproto, ofport->bundle, ofport,
620 ofport->up.ofp_port, ofport->odp_port,
621 ofport->up.netdev, ofport->cfm,
622 ofport->bfd, ofport->peer, stp_port,
623 ofport->qdscp, ofport->n_qdscp,
624 ofport->up.pp.config, ofport->up.pp.state,
625 ofport->is_tunnel, ofport->may_enable);
630 udpif_revalidate(backer->udpif);
633 process_dpif_port_changes(backer);
638 /* Check for and handle port changes in 'backer''s dpif. */
640 process_dpif_port_changes(struct dpif_backer *backer)
646 error = dpif_port_poll(backer->dpif, &devname);
652 process_dpif_all_ports_changed(backer);
656 process_dpif_port_change(backer, devname);
661 process_dpif_port_error(backer, error);
668 process_dpif_all_ports_changed(struct dpif_backer *backer)
670 struct ofproto_dpif *ofproto;
671 struct dpif_port dpif_port;
672 struct dpif_port_dump dump;
673 struct sset devnames;
676 sset_init(&devnames);
677 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
678 if (ofproto->backer == backer) {
679 struct ofport *ofport;
681 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
682 sset_add(&devnames, netdev_get_name(ofport->netdev));
686 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
687 sset_add(&devnames, dpif_port.name);
690 SSET_FOR_EACH (devname, &devnames) {
691 process_dpif_port_change(backer, devname);
693 sset_destroy(&devnames);
697 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
699 struct ofproto_dpif *ofproto;
700 struct dpif_port port;
702 /* Don't report on the datapath's device. */
703 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
707 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
708 &all_ofproto_dpifs) {
709 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
714 ofproto = lookup_ofproto_dpif_by_port_name(devname);
715 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
716 /* The port was removed. If we know the datapath,
717 * report it through poll_set(). If we don't, it may be
718 * notifying us of a removal we initiated, so ignore it.
719 * If there's a pending ENOBUFS, let it stand, since
720 * everything will be reevaluated. */
721 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
722 sset_add(&ofproto->port_poll_set, devname);
723 ofproto->port_poll_errno = 0;
725 } else if (!ofproto) {
726 /* The port was added, but we don't know with which
727 * ofproto we should associate it. Delete it. */
728 dpif_port_del(backer->dpif, port.port_no);
730 struct ofport_dpif *ofport;
732 ofport = ofport_dpif_cast(shash_find_data(
733 &ofproto->up.port_by_name, devname));
735 && ofport->odp_port != port.port_no
736 && !odp_port_to_ofport(backer, port.port_no))
738 /* 'ofport''s datapath port number has changed from
739 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
740 * structures to match. */
741 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
742 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
743 ofport->odp_port = port.port_no;
744 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
745 hash_odp_port(port.port_no));
746 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
747 backer->need_revalidate = REV_RECONFIGURE;
750 dpif_port_destroy(&port);
753 /* Propagate 'error' to all ofprotos based on 'backer'. */
755 process_dpif_port_error(struct dpif_backer *backer, int error)
757 struct ofproto_dpif *ofproto;
759 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
760 if (ofproto->backer == backer) {
761 sset_clear(&ofproto->port_poll_set);
762 ofproto->port_poll_errno = error;
768 type_wait(const char *type)
770 struct dpif_backer *backer;
772 backer = shash_find_data(&all_dpif_backers, type);
774 /* This is not necessarily a problem, since backers are only
775 * created on demand. */
779 dpif_wait(backer->dpif);
782 /* Basic life-cycle. */
784 static int add_internal_flows(struct ofproto_dpif *);
786 static struct ofproto *
789 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
794 dealloc(struct ofproto *ofproto_)
796 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
801 close_dpif_backer(struct dpif_backer *backer)
803 ovs_assert(backer->refcount > 0);
805 if (--backer->refcount) {
809 udpif_destroy(backer->udpif);
811 simap_destroy(&backer->tnl_backers);
812 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
813 hmap_destroy(&backer->odp_to_ofport_map);
814 shash_find_and_delete(&all_dpif_backers, backer->type);
815 recirc_id_pool_destroy(backer->rid_pool);
817 dpif_close(backer->dpif);
821 /* Datapath port slated for removal from datapath. */
823 struct list list_node;
827 static bool check_variable_length_userdata(struct dpif_backer *backer);
828 static size_t check_max_mpls_depth(struct dpif_backer *backer);
829 static bool check_recirc(struct dpif_backer *backer);
832 open_dpif_backer(const char *type, struct dpif_backer **backerp)
834 struct dpif_backer *backer;
835 struct dpif_port_dump port_dump;
836 struct dpif_port port;
837 struct shash_node *node;
838 struct list garbage_list;
839 struct odp_garbage *garbage, *next;
846 backer = shash_find_data(&all_dpif_backers, type);
853 backer_name = xasprintf("ovs-%s", type);
855 /* Remove any existing datapaths, since we assume we're the only
856 * userspace controlling the datapath. */
858 dp_enumerate_names(type, &names);
859 SSET_FOR_EACH(name, &names) {
860 struct dpif *old_dpif;
862 /* Don't remove our backer if it exists. */
863 if (!strcmp(name, backer_name)) {
867 if (dpif_open(name, type, &old_dpif)) {
868 VLOG_WARN("couldn't open old datapath %s to remove it", name);
870 dpif_delete(old_dpif);
871 dpif_close(old_dpif);
874 sset_destroy(&names);
876 backer = xmalloc(sizeof *backer);
878 error = dpif_create_and_open(backer_name, type, &backer->dpif);
881 VLOG_ERR("failed to open datapath of type %s: %s", type,
882 ovs_strerror(error));
886 backer->udpif = udpif_create(backer, backer->dpif);
888 backer->type = xstrdup(type);
889 backer->refcount = 1;
890 hmap_init(&backer->odp_to_ofport_map);
891 ovs_rwlock_init(&backer->odp_to_ofport_lock);
892 backer->need_revalidate = 0;
893 simap_init(&backer->tnl_backers);
894 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
897 if (backer->recv_set_enable) {
898 dpif_flow_flush(backer->dpif);
901 /* Loop through the ports already on the datapath and remove any
902 * that we don't need anymore. */
903 list_init(&garbage_list);
904 dpif_port_dump_start(&port_dump, backer->dpif);
905 while (dpif_port_dump_next(&port_dump, &port)) {
906 node = shash_find(&init_ofp_ports, port.name);
907 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
908 garbage = xmalloc(sizeof *garbage);
909 garbage->odp_port = port.port_no;
910 list_push_front(&garbage_list, &garbage->list_node);
913 dpif_port_dump_done(&port_dump);
915 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
916 dpif_port_del(backer->dpif, garbage->odp_port);
917 list_remove(&garbage->list_node);
921 shash_add(&all_dpif_backers, type, backer);
923 backer->enable_recirc = check_recirc(backer);
924 backer->variable_length_userdata = check_variable_length_userdata(backer);
925 backer->max_mpls_depth = check_max_mpls_depth(backer);
926 backer->rid_pool = recirc_id_pool_create();
928 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
930 VLOG_ERR("failed to listen on datapath of type %s: %s",
931 type, ovs_strerror(error));
932 close_dpif_backer(backer);
936 if (backer->recv_set_enable) {
937 udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
943 /* Tests whether 'backer''s datapath supports recirculation. Only newer
944 * datapaths support OVS_KEY_ATTR_RECIRC_ID in keys. We need to disable some
945 * features on older datapaths that don't support this feature.
947 * Returns false if 'backer' definitely does not support recirculation, true if
948 * it seems to support recirculation or if at least the error we get is
951 check_recirc(struct dpif_backer *backer)
954 struct odputil_keybuf keybuf;
957 bool enable_recirc = false;
959 memset(&flow, 0, sizeof flow);
963 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
964 odp_flow_key_from_flow(&key, &flow, NULL, 0, true);
966 error = dpif_flow_put(backer->dpif, DPIF_FP_CREATE,
967 ofpbuf_data(&key), ofpbuf_size(&key), NULL, 0, NULL,
969 if (error && error != EEXIST) {
970 if (error != EINVAL) {
971 VLOG_WARN("%s: Reciculation flow probe failed (%s)",
972 dpif_name(backer->dpif), ovs_strerror(error));
977 error = dpif_flow_del(backer->dpif, ofpbuf_data(&key), ofpbuf_size(&key),
980 VLOG_WARN("%s: failed to delete recirculation feature probe flow",
981 dpif_name(backer->dpif));
984 enable_recirc = true;
988 VLOG_INFO("%s: Datapath supports recirculation",
989 dpif_name(backer->dpif));
991 VLOG_INFO("%s: Datapath does not support recirculation",
992 dpif_name(backer->dpif));
995 return enable_recirc;
998 /* Tests whether 'backer''s datapath supports variable-length
999 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions. We need
1000 * to disable some features on older datapaths that don't support this
1003 * Returns false if 'backer' definitely does not support variable-length
1004 * userdata, true if it seems to support them or if at least the error we get
1007 check_variable_length_userdata(struct dpif_backer *backer)
1009 struct eth_header *eth;
1010 struct ofpbuf actions;
1011 struct dpif_execute execute;
1012 struct ofpbuf packet;
1016 /* Compose a userspace action that will cause an ERANGE error on older
1017 * datapaths that don't support variable-length userdata.
1019 * We really test for using userdata longer than 8 bytes, but older
1020 * datapaths accepted these, silently truncating the userdata to 8 bytes.
1021 * The same older datapaths rejected userdata shorter than 8 bytes, so we
1022 * test for that instead as a proxy for longer userdata support. */
1023 ofpbuf_init(&actions, 64);
1024 start = nl_msg_start_nested(&actions, OVS_ACTION_ATTR_USERSPACE);
1025 nl_msg_put_u32(&actions, OVS_USERSPACE_ATTR_PID,
1026 dpif_port_get_pid(backer->dpif, ODPP_NONE, 0));
1027 nl_msg_put_unspec_zero(&actions, OVS_USERSPACE_ATTR_USERDATA, 4);
1028 nl_msg_end_nested(&actions, start);
1030 /* Compose a dummy ethernet packet. */
1031 ofpbuf_init(&packet, ETH_HEADER_LEN);
1032 eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
1033 eth->eth_type = htons(0x1234);
1035 /* Execute the actions. On older datapaths this fails with ERANGE, on
1036 * newer datapaths it succeeds. */
1037 execute.actions = ofpbuf_data(&actions);
1038 execute.actions_len = ofpbuf_size(&actions);
1039 execute.packet = &packet;
1040 execute.md = PKT_METADATA_INITIALIZER(0);
1041 execute.needs_help = false;
1043 error = dpif_execute(backer->dpif, &execute);
1045 ofpbuf_uninit(&packet);
1046 ofpbuf_uninit(&actions);
1053 /* Variable-length userdata is not supported. */
1054 VLOG_WARN("%s: datapath does not support variable-length userdata "
1055 "feature (needs Linux 3.10+ or kernel module from OVS "
1056 "1..11+). The NXAST_SAMPLE action will be ignored.",
1057 dpif_name(backer->dpif));
1061 /* Something odd happened. We're not sure whether variable-length
1062 * userdata is supported. Default to "yes". */
1063 VLOG_WARN("%s: variable-length userdata feature probe failed (%s)",
1064 dpif_name(backer->dpif), ovs_strerror(error));
1069 /* Tests the MPLS label stack depth supported by 'backer''s datapath.
1071 * Returns the number of elements in a struct flow's mpls_lse field
1072 * if the datapath supports at least that many entries in an
1074 * Otherwise returns the number of MPLS push actions supported by
1077 check_max_mpls_depth(struct dpif_backer *backer)
1082 for (n = 0; n < FLOW_MAX_MPLS_LABELS; n++) {
1083 struct odputil_keybuf keybuf;
1087 memset(&flow, 0, sizeof flow);
1088 flow.dl_type = htons(ETH_TYPE_MPLS);
1089 flow_set_mpls_bos(&flow, n, 1);
1091 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1092 odp_flow_key_from_flow(&key, &flow, NULL, 0, false);
1094 error = dpif_flow_put(backer->dpif, DPIF_FP_CREATE,
1095 ofpbuf_data(&key), ofpbuf_size(&key), NULL, 0, NULL, 0, NULL);
1096 if (error && error != EEXIST) {
1097 if (error != EINVAL) {
1098 VLOG_WARN("%s: MPLS stack length feature probe failed (%s)",
1099 dpif_name(backer->dpif), ovs_strerror(error));
1104 error = dpif_flow_del(backer->dpif, ofpbuf_data(&key), ofpbuf_size(&key), NULL);
1106 VLOG_WARN("%s: failed to delete MPLS feature probe flow",
1107 dpif_name(backer->dpif));
1111 VLOG_INFO("%s: MPLS label stack length probed as %d",
1112 dpif_name(backer->dpif), n);
1117 construct(struct ofproto *ofproto_)
1119 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1120 struct shash_node *node, *next;
1123 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1128 ofproto->netflow = NULL;
1129 ofproto->sflow = NULL;
1130 ofproto->ipfix = NULL;
1131 ofproto->stp = NULL;
1132 ofproto->dump_seq = 0;
1133 hmap_init(&ofproto->bundles);
1134 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1136 ofproto->mbridge = mbridge_create();
1137 ofproto->has_bonded_bundles = false;
1138 ofproto->lacp_enabled = false;
1139 ovs_mutex_init_adaptive(&ofproto->stats_mutex);
1140 ovs_mutex_init(&ofproto->vsp_mutex);
1142 guarded_list_init(&ofproto->pins);
1144 ofproto_dpif_unixctl_init();
1146 hmap_init(&ofproto->vlandev_map);
1147 hmap_init(&ofproto->realdev_vid_map);
1149 sset_init(&ofproto->ports);
1150 sset_init(&ofproto->ghost_ports);
1151 sset_init(&ofproto->port_poll_set);
1152 ofproto->port_poll_errno = 0;
1153 ofproto->change_seq = 0;
1154 ofproto->pins_seq = seq_create();
1155 ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1158 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1159 struct iface_hint *iface_hint = node->data;
1161 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1162 /* Check if the datapath already has this port. */
1163 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1164 sset_add(&ofproto->ports, node->name);
1167 free(iface_hint->br_name);
1168 free(iface_hint->br_type);
1170 shash_delete(&init_ofp_ports, node);
1174 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1175 hash_string(ofproto->up.name, 0));
1176 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1178 ofproto_init_tables(ofproto_, N_TABLES);
1179 error = add_internal_flows(ofproto);
1181 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1187 add_internal_miss_flow(struct ofproto_dpif *ofproto, int id,
1188 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1194 match_init_catchall(&match);
1195 match_set_reg(&match, 0, id);
1197 error = ofproto_dpif_add_internal_flow(ofproto, &match, 0, 0, ofpacts,
1199 *rulep = error ? NULL : rule_dpif_cast(rule);
1205 add_internal_flows(struct ofproto_dpif *ofproto)
1207 struct ofpact_controller *controller;
1208 uint64_t ofpacts_stub[128 / 8];
1209 struct ofpbuf ofpacts;
1210 struct rule *unused_rulep OVS_UNUSED;
1211 struct ofpact_resubmit *resubmit;
1216 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1219 controller = ofpact_put_CONTROLLER(&ofpacts);
1220 controller->max_len = UINT16_MAX;
1221 controller->controller_id = 0;
1222 controller->reason = OFPR_NO_MATCH;
1223 ofpact_pad(&ofpacts);
1225 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1226 &ofproto->miss_rule);
1231 ofpbuf_clear(&ofpacts);
1232 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1233 &ofproto->no_packet_in_rule);
1238 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1239 &ofproto->drop_frags_rule);
1244 /* Continue non-recirculation rule lookups from table 0.
1246 * (priority=2), recirc=0, actions=resubmit(, 0)
1248 resubmit = ofpact_put_RESUBMIT(&ofpacts);
1249 resubmit->ofpact.compat = 0;
1250 resubmit->in_port = OFPP_IN_PORT;
1251 resubmit->table_id = 0;
1253 match_init_catchall(&match);
1254 match_set_recirc_id(&match, 0);
1256 error = ofproto_dpif_add_internal_flow(ofproto, &match, 2, 0, &ofpacts,
1262 /* Drop any run away recirc rule lookups. Recirc_id has to be
1263 * non-zero when reaching this rule.
1265 * (priority=1), *, actions=drop
1267 ofpbuf_clear(&ofpacts);
1268 match_init_catchall(&match);
1269 error = ofproto_dpif_add_internal_flow(ofproto, &match, 1, 0, &ofpacts,
1276 destruct(struct ofproto *ofproto_)
1278 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1279 struct ofproto_packet_in *pin, *next_pin;
1280 struct rule_dpif *rule;
1281 struct oftable *table;
1284 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1286 xlate_remove_ofproto(ofproto);
1289 /* Ensure that the upcall processing threads have no remaining references
1290 * to the ofproto or anything in it. */
1291 udpif_synchronize(ofproto->backer->udpif);
1293 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1295 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1296 CLS_FOR_EACH_SAFE (rule, up.cr, &table->cls) {
1297 ofproto_rule_delete(&ofproto->up, &rule->up);
1301 guarded_list_pop_all(&ofproto->pins, &pins);
1302 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1303 list_remove(&pin->list_node);
1304 free(CONST_CAST(void *, pin->up.packet));
1307 guarded_list_destroy(&ofproto->pins);
1309 mbridge_unref(ofproto->mbridge);
1311 netflow_unref(ofproto->netflow);
1312 dpif_sflow_unref(ofproto->sflow);
1313 hmap_destroy(&ofproto->bundles);
1314 mac_learning_unref(ofproto->ml);
1315 mcast_snooping_unref(ofproto->ms);
1317 hmap_destroy(&ofproto->vlandev_map);
1318 hmap_destroy(&ofproto->realdev_vid_map);
1320 sset_destroy(&ofproto->ports);
1321 sset_destroy(&ofproto->ghost_ports);
1322 sset_destroy(&ofproto->port_poll_set);
1324 ovs_mutex_destroy(&ofproto->stats_mutex);
1325 ovs_mutex_destroy(&ofproto->vsp_mutex);
1327 seq_destroy(ofproto->pins_seq);
1329 close_dpif_backer(ofproto->backer);
1333 run(struct ofproto *ofproto_)
1335 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1336 uint64_t new_seq, new_dump_seq;
1338 if (mbridge_need_revalidate(ofproto->mbridge)) {
1339 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1340 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1341 mac_learning_flush(ofproto->ml);
1342 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1343 mcast_snooping_mdb_flush(ofproto->ms);
1346 /* Always updates the ofproto->pins_seqno to avoid frequent wakeup during
1347 * flow restore. Even though nothing is processed during flow restore,
1348 * all queued 'pins' will be handled immediately when flow restore
1350 ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1352 /* Do not perform any periodic activity required by 'ofproto' while
1353 * waiting for flow restore to complete. */
1354 if (!ofproto_get_flow_restore_wait()) {
1355 struct ofproto_packet_in *pin, *next_pin;
1358 guarded_list_pop_all(&ofproto->pins, &pins);
1359 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1360 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1361 list_remove(&pin->list_node);
1362 free(CONST_CAST(void *, pin->up.packet));
1367 if (ofproto->netflow) {
1368 netflow_run(ofproto->netflow);
1370 if (ofproto->sflow) {
1371 dpif_sflow_run(ofproto->sflow);
1373 if (ofproto->ipfix) {
1374 dpif_ipfix_run(ofproto->ipfix);
1377 new_seq = seq_read(connectivity_seq_get());
1378 if (ofproto->change_seq != new_seq) {
1379 struct ofport_dpif *ofport;
1381 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1385 ofproto->change_seq = new_seq;
1387 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1388 struct ofbundle *bundle;
1390 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1396 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1397 if (mac_learning_run(ofproto->ml)) {
1398 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1400 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1402 if (mcast_snooping_run(ofproto->ms)) {
1403 ofproto->backer->need_revalidate = REV_MCAST_SNOOPING;
1406 new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
1407 if (ofproto->dump_seq != new_dump_seq) {
1408 struct rule *rule, *next_rule;
1410 /* We know stats are relatively fresh, so now is a good time to do some
1412 ofproto->dump_seq = new_dump_seq;
1414 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1416 ovs_mutex_lock(&ofproto_mutex);
1417 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
1418 &ofproto->up.expirable) {
1419 rule_expire(rule_dpif_cast(rule));
1421 ovs_mutex_unlock(&ofproto_mutex);
1423 /* All outstanding data in existing flows has been accounted, so it's a
1424 * good time to do bond rebalancing. */
1425 if (ofproto->has_bonded_bundles) {
1426 struct ofbundle *bundle;
1428 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1430 bond_rebalance(bundle->bond);
1440 wait(struct ofproto *ofproto_)
1442 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1444 if (ofproto_get_flow_restore_wait()) {
1448 if (ofproto->sflow) {
1449 dpif_sflow_wait(ofproto->sflow);
1451 if (ofproto->ipfix) {
1452 dpif_ipfix_wait(ofproto->ipfix);
1454 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1455 struct ofbundle *bundle;
1457 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1458 bundle_wait(bundle);
1461 if (ofproto->netflow) {
1462 netflow_wait(ofproto->netflow);
1464 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1465 mac_learning_wait(ofproto->ml);
1466 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1467 mcast_snooping_wait(ofproto->ms);
1469 if (ofproto->backer->need_revalidate) {
1470 /* Shouldn't happen, but if it does just go around again. */
1471 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1472 poll_immediate_wake();
1475 seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
1476 seq_wait(ofproto->pins_seq, ofproto->pins_seqno);
1480 type_get_memory_usage(const char *type, struct simap *usage)
1482 struct dpif_backer *backer;
1484 backer = shash_find_data(&all_dpif_backers, type);
1486 udpif_get_memory_usage(backer->udpif, usage);
1491 flush(struct ofproto *ofproto_)
1493 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1494 struct dpif_backer *backer = ofproto->backer;
1497 udpif_flush(backer->udpif);
1502 query_tables(struct ofproto *ofproto,
1503 struct ofputil_table_features *features,
1504 struct ofputil_table_stats *stats)
1506 strcpy(features->name, "classifier");
1511 for (i = 0; i < ofproto->n_tables; i++) {
1512 unsigned long missed, matched;
1514 atomic_read(&ofproto->tables[i].n_matched, &matched);
1515 stats[i].matched_count = matched;
1516 atomic_read(&ofproto->tables[i].n_missed, &missed);
1517 stats[i].lookup_count = matched + missed;
1522 static struct ofport *
1525 struct ofport_dpif *port = xmalloc(sizeof *port);
1530 port_dealloc(struct ofport *port_)
1532 struct ofport_dpif *port = ofport_dpif_cast(port_);
1537 port_construct(struct ofport *port_)
1539 struct ofport_dpif *port = ofport_dpif_cast(port_);
1540 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1541 const struct netdev *netdev = port->up.netdev;
1542 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1543 struct dpif_port dpif_port;
1546 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1547 port->bundle = NULL;
1550 port->may_enable = true;
1551 port->stp_port = NULL;
1552 port->stp_state = STP_DISABLED;
1553 port->is_tunnel = false;
1557 port->realdev_ofp_port = 0;
1558 port->vlandev_vid = 0;
1559 port->carrier_seq = netdev_get_carrier_resets(netdev);
1560 port->is_layer3 = netdev_vport_is_layer3(netdev);
1562 if (netdev_vport_is_patch(netdev)) {
1563 /* By bailing out here, we don't submit the port to the sFlow module
1564 * to be considered for counter polling export. This is correct
1565 * because the patch port represents an interface that sFlow considers
1566 * to be "internal" to the switch as a whole, and therefore not an
1567 * candidate for counter polling. */
1568 port->odp_port = ODPP_NONE;
1569 ofport_update_peer(port);
1573 error = dpif_port_query_by_name(ofproto->backer->dpif,
1574 netdev_vport_get_dpif_port(netdev, namebuf,
1581 port->odp_port = dpif_port.port_no;
1583 if (netdev_get_tunnel_config(netdev)) {
1584 tnl_port_add(port, port->up.netdev, port->odp_port);
1585 port->is_tunnel = true;
1587 /* Sanity-check that a mapping doesn't already exist. This
1588 * shouldn't happen for non-tunnel ports. */
1589 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1590 VLOG_ERR("port %s already has an OpenFlow port number",
1592 dpif_port_destroy(&dpif_port);
1596 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1597 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1598 hash_odp_port(port->odp_port));
1599 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1601 dpif_port_destroy(&dpif_port);
1603 if (ofproto->sflow) {
1604 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1611 port_destruct(struct ofport *port_)
1613 struct ofport_dpif *port = ofport_dpif_cast(port_);
1614 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1615 const char *devname = netdev_get_name(port->up.netdev);
1616 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1617 const char *dp_port_name;
1619 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1621 xlate_ofport_remove(port);
1624 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1626 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1627 /* The underlying device is still there, so delete it. This
1628 * happens when the ofproto is being destroyed, since the caller
1629 * assumes that removal of attached ports will happen as part of
1631 if (!port->is_tunnel) {
1632 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1637 port->peer->peer = NULL;
1641 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1642 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1643 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1644 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1648 sset_find_and_delete(&ofproto->ports, devname);
1649 sset_find_and_delete(&ofproto->ghost_ports, devname);
1650 bundle_remove(port_);
1651 set_cfm(port_, NULL);
1652 set_bfd(port_, NULL);
1653 if (port->stp_port) {
1654 stp_port_disable(port->stp_port);
1656 if (ofproto->sflow) {
1657 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1664 port_modified(struct ofport *port_)
1666 struct ofport_dpif *port = ofport_dpif_cast(port_);
1668 if (port->bundle && port->bundle->bond) {
1669 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1673 cfm_set_netdev(port->cfm, port->up.netdev);
1677 bfd_set_netdev(port->bfd, port->up.netdev);
1680 ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1681 port->up.pp.hw_addr);
1683 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1685 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1689 ofport_update_peer(port);
1693 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1695 struct ofport_dpif *port = ofport_dpif_cast(port_);
1696 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1697 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1699 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1700 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1701 OFPUTIL_PC_NO_PACKET_IN)) {
1702 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1704 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1705 bundle_update(port->bundle);
1711 set_sflow(struct ofproto *ofproto_,
1712 const struct ofproto_sflow_options *sflow_options)
1714 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1715 struct dpif_sflow *ds = ofproto->sflow;
1717 if (sflow_options) {
1719 struct ofport_dpif *ofport;
1721 ds = ofproto->sflow = dpif_sflow_create();
1722 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1723 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1725 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1727 dpif_sflow_set_options(ds, sflow_options);
1730 dpif_sflow_unref(ds);
1731 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1732 ofproto->sflow = NULL;
1740 struct ofproto *ofproto_,
1741 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1742 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1743 size_t n_flow_exporters_options)
1745 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1746 struct dpif_ipfix *di = ofproto->ipfix;
1747 bool has_options = bridge_exporter_options || flow_exporters_options;
1749 if (has_options && !di) {
1750 di = ofproto->ipfix = dpif_ipfix_create();
1754 /* Call set_options in any case to cleanly flush the flow
1755 * caches in the last exporters that are to be destroyed. */
1756 dpif_ipfix_set_options(
1757 di, bridge_exporter_options, flow_exporters_options,
1758 n_flow_exporters_options);
1761 dpif_ipfix_unref(di);
1762 ofproto->ipfix = NULL;
1770 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1772 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1777 struct ofproto_dpif *ofproto;
1779 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1780 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1781 ofport->cfm = cfm_create(ofport->up.netdev);
1784 if (cfm_configure(ofport->cfm, s)) {
1791 cfm_unref(ofport->cfm);
1794 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1795 ofport->up.pp.hw_addr);
1800 cfm_status_changed(struct ofport *ofport_)
1802 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1804 return ofport->cfm ? cfm_check_status_change(ofport->cfm) : true;
1808 get_cfm_status(const struct ofport *ofport_,
1809 struct cfm_status *status)
1811 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1815 cfm_get_status(ofport->cfm, status);
1824 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1826 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1827 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1831 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1832 cfg, ofport->up.netdev);
1833 if (ofport->bfd != old) {
1834 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1836 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1837 ofport->up.pp.hw_addr);
1842 bfd_status_changed(struct ofport *ofport_)
1844 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1846 return ofport->bfd ? bfd_check_status_change(ofport->bfd) : true;
1850 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1852 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1856 bfd_get_status(ofport->bfd, smap);
1864 /* Spanning Tree. */
1867 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1869 struct ofproto_dpif *ofproto = ofproto_;
1870 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1871 struct ofport_dpif *ofport;
1873 ofport = stp_port_get_aux(sp);
1875 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1876 ofproto->up.name, port_num);
1878 struct eth_header *eth = ofpbuf_l2(pkt);
1880 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1881 if (eth_addr_is_zero(eth->eth_src)) {
1882 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1883 "with unknown MAC", ofproto->up.name, port_num);
1885 ofproto_dpif_send_packet(ofport, pkt);
1891 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1893 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1895 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1897 /* Only revalidate flows if the configuration changed. */
1898 if (!s != !ofproto->stp) {
1899 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1903 if (!ofproto->stp) {
1904 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1905 send_bpdu_cb, ofproto);
1906 ofproto->stp_last_tick = time_msec();
1909 stp_set_bridge_id(ofproto->stp, s->system_id);
1910 stp_set_bridge_priority(ofproto->stp, s->priority);
1911 stp_set_hello_time(ofproto->stp, s->hello_time);
1912 stp_set_max_age(ofproto->stp, s->max_age);
1913 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1915 struct ofport *ofport;
1917 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1918 set_stp_port(ofport, NULL);
1921 stp_unref(ofproto->stp);
1922 ofproto->stp = NULL;
1929 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1931 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1935 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1936 s->designated_root = stp_get_designated_root(ofproto->stp);
1937 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1946 update_stp_port_state(struct ofport_dpif *ofport)
1948 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1949 enum stp_state state;
1951 /* Figure out new state. */
1952 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1956 if (ofport->stp_state != state) {
1957 enum ofputil_port_state of_state;
1960 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1961 netdev_get_name(ofport->up.netdev),
1962 stp_state_name(ofport->stp_state),
1963 stp_state_name(state));
1964 if (stp_learn_in_state(ofport->stp_state)
1965 != stp_learn_in_state(state)) {
1966 /* xxx Learning action flows should also be flushed. */
1967 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1968 mac_learning_flush(ofproto->ml);
1969 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1970 mcast_snooping_mdb_flush(ofproto->ms);
1972 fwd_change = stp_forward_in_state(ofport->stp_state)
1973 != stp_forward_in_state(state);
1975 ofproto->backer->need_revalidate = REV_STP;
1976 ofport->stp_state = state;
1977 ofport->stp_state_entered = time_msec();
1979 if (fwd_change && ofport->bundle) {
1980 bundle_update(ofport->bundle);
1983 /* Update the STP state bits in the OpenFlow port description. */
1984 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1985 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1986 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1987 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1988 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1990 ofproto_port_set_state(&ofport->up, of_state);
1994 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1995 * caller is responsible for assigning STP port numbers and ensuring
1996 * there are no duplicates. */
1998 set_stp_port(struct ofport *ofport_,
1999 const struct ofproto_port_stp_settings *s)
2001 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2002 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2003 struct stp_port *sp = ofport->stp_port;
2005 if (!s || !s->enable) {
2007 ofport->stp_port = NULL;
2008 stp_port_disable(sp);
2009 update_stp_port_state(ofport);
2012 } else if (sp && stp_port_no(sp) != s->port_num
2013 && ofport == stp_port_get_aux(sp)) {
2014 /* The port-id changed, so disable the old one if it's not
2015 * already in use by another port. */
2016 stp_port_disable(sp);
2019 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2020 stp_port_enable(sp);
2022 stp_port_set_name(sp, netdev_get_name(ofport->up.netdev));
2023 stp_port_set_aux(sp, ofport);
2024 stp_port_set_priority(sp, s->priority);
2025 stp_port_set_path_cost(sp, s->path_cost);
2027 update_stp_port_state(ofport);
2033 get_stp_port_status(struct ofport *ofport_,
2034 struct ofproto_port_stp_status *s)
2036 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2037 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2038 struct stp_port *sp = ofport->stp_port;
2040 if (!ofproto->stp || !sp) {
2046 s->port_id = stp_port_get_id(sp);
2047 s->state = stp_port_get_state(sp);
2048 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2049 s->role = stp_port_get_role(sp);
2055 get_stp_port_stats(struct ofport *ofport_,
2056 struct ofproto_port_stp_stats *s)
2058 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2059 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2060 struct stp_port *sp = ofport->stp_port;
2062 if (!ofproto->stp || !sp) {
2068 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2074 stp_run(struct ofproto_dpif *ofproto)
2077 long long int now = time_msec();
2078 long long int elapsed = now - ofproto->stp_last_tick;
2079 struct stp_port *sp;
2082 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2083 ofproto->stp_last_tick = now;
2085 while (stp_get_changed_port(ofproto->stp, &sp)) {
2086 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2089 update_stp_port_state(ofport);
2093 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2094 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2095 mac_learning_flush(ofproto->ml);
2096 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2097 mcast_snooping_mdb_flush(ofproto->ms);
2103 stp_wait(struct ofproto_dpif *ofproto)
2106 poll_timer_wait(1000);
2111 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2114 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2115 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2117 if (ofport->n_qdscp != n_qdscp
2118 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2119 n_qdscp * sizeof *qdscp))) {
2120 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2121 free(ofport->qdscp);
2122 ofport->qdscp = n_qdscp
2123 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2125 ofport->n_qdscp = n_qdscp;
2133 /* Expires all MAC learning entries associated with 'bundle' and forces its
2134 * ofproto to revalidate every flow.
2136 * Normally MAC learning entries are removed only from the ofproto associated
2137 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2138 * are removed from every ofproto. When patch ports and SLB bonds are in use
2139 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2140 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2141 * with the host from which it migrated. */
2143 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2145 struct ofproto_dpif *ofproto = bundle->ofproto;
2146 struct mac_learning *ml = ofproto->ml;
2147 struct mac_entry *mac, *next_mac;
2149 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2150 ovs_rwlock_wrlock(&ml->rwlock);
2151 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2152 if (mac->port.p == bundle) {
2154 struct ofproto_dpif *o;
2156 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2158 struct mac_entry *e;
2160 ovs_rwlock_wrlock(&o->ml->rwlock);
2161 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2163 mac_learning_expire(o->ml, e);
2165 ovs_rwlock_unlock(&o->ml->rwlock);
2170 mac_learning_expire(ml, mac);
2173 ovs_rwlock_unlock(&ml->rwlock);
2176 static struct ofbundle *
2177 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2179 struct ofbundle *bundle;
2181 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2182 &ofproto->bundles) {
2183 if (bundle->aux == aux) {
2191 bundle_update(struct ofbundle *bundle)
2193 struct ofport_dpif *port;
2195 bundle->floodable = true;
2196 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2197 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2199 || !stp_forward_in_state(port->stp_state)) {
2200 bundle->floodable = false;
2207 bundle_del_port(struct ofport_dpif *port)
2209 struct ofbundle *bundle = port->bundle;
2211 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2213 list_remove(&port->bundle_node);
2214 port->bundle = NULL;
2217 lacp_slave_unregister(bundle->lacp, port);
2220 bond_slave_unregister(bundle->bond, port);
2223 bundle_update(bundle);
2227 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2228 struct lacp_slave_settings *lacp)
2230 struct ofport_dpif *port;
2232 port = get_ofp_port(bundle->ofproto, ofp_port);
2237 if (port->bundle != bundle) {
2238 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2240 bundle_remove(&port->up);
2243 port->bundle = bundle;
2244 list_push_back(&bundle->ports, &port->bundle_node);
2245 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2247 || !stp_forward_in_state(port->stp_state)) {
2248 bundle->floodable = false;
2252 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2253 lacp_slave_register(bundle->lacp, port, lacp);
2260 bundle_destroy(struct ofbundle *bundle)
2262 struct ofproto_dpif *ofproto;
2263 struct ofport_dpif *port, *next_port;
2269 ofproto = bundle->ofproto;
2270 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2273 xlate_bundle_remove(bundle);
2276 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2277 bundle_del_port(port);
2280 bundle_flush_macs(bundle, true);
2281 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2283 free(bundle->trunks);
2284 lacp_unref(bundle->lacp);
2285 bond_unref(bundle->bond);
2290 bundle_set(struct ofproto *ofproto_, void *aux,
2291 const struct ofproto_bundle_settings *s)
2293 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2294 bool need_flush = false;
2295 struct ofport_dpif *port;
2296 struct ofbundle *bundle;
2297 unsigned long *trunks;
2303 bundle_destroy(bundle_lookup(ofproto, aux));
2307 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2308 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2310 bundle = bundle_lookup(ofproto, aux);
2312 bundle = xmalloc(sizeof *bundle);
2314 bundle->ofproto = ofproto;
2315 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2316 hash_pointer(aux, 0));
2318 bundle->name = NULL;
2320 list_init(&bundle->ports);
2321 bundle->vlan_mode = PORT_VLAN_TRUNK;
2323 bundle->trunks = NULL;
2324 bundle->use_priority_tags = s->use_priority_tags;
2325 bundle->lacp = NULL;
2326 bundle->bond = NULL;
2328 bundle->floodable = true;
2329 mbridge_register_bundle(ofproto->mbridge, bundle);
2332 if (!bundle->name || strcmp(s->name, bundle->name)) {
2334 bundle->name = xstrdup(s->name);
2339 ofproto->lacp_enabled = true;
2340 if (!bundle->lacp) {
2341 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2342 bundle->lacp = lacp_create();
2344 lacp_configure(bundle->lacp, s->lacp);
2346 lacp_unref(bundle->lacp);
2347 bundle->lacp = NULL;
2350 /* Update set of ports. */
2352 for (i = 0; i < s->n_slaves; i++) {
2353 if (!bundle_add_port(bundle, s->slaves[i],
2354 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2358 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2359 struct ofport_dpif *next_port;
2361 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2362 for (i = 0; i < s->n_slaves; i++) {
2363 if (s->slaves[i] == port->up.ofp_port) {
2368 bundle_del_port(port);
2372 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2374 if (list_is_empty(&bundle->ports)) {
2375 bundle_destroy(bundle);
2379 /* Set VLAN tagging mode */
2380 if (s->vlan_mode != bundle->vlan_mode
2381 || s->use_priority_tags != bundle->use_priority_tags) {
2382 bundle->vlan_mode = s->vlan_mode;
2383 bundle->use_priority_tags = s->use_priority_tags;
2388 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2389 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2391 if (vlan != bundle->vlan) {
2392 bundle->vlan = vlan;
2396 /* Get trunked VLANs. */
2397 switch (s->vlan_mode) {
2398 case PORT_VLAN_ACCESS:
2402 case PORT_VLAN_TRUNK:
2403 trunks = CONST_CAST(unsigned long *, s->trunks);
2406 case PORT_VLAN_NATIVE_UNTAGGED:
2407 case PORT_VLAN_NATIVE_TAGGED:
2408 if (vlan != 0 && (!s->trunks
2409 || !bitmap_is_set(s->trunks, vlan)
2410 || bitmap_is_set(s->trunks, 0))) {
2411 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2413 trunks = bitmap_clone(s->trunks, 4096);
2415 trunks = bitmap_allocate1(4096);
2417 bitmap_set1(trunks, vlan);
2418 bitmap_set0(trunks, 0);
2420 trunks = CONST_CAST(unsigned long *, s->trunks);
2427 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2428 free(bundle->trunks);
2429 if (trunks == s->trunks) {
2430 bundle->trunks = vlan_bitmap_clone(trunks);
2432 bundle->trunks = trunks;
2437 if (trunks != s->trunks) {
2442 if (!list_is_short(&bundle->ports)) {
2443 bundle->ofproto->has_bonded_bundles = true;
2445 if (bond_reconfigure(bundle->bond, s->bond)) {
2446 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2449 bundle->bond = bond_create(s->bond, ofproto);
2450 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2453 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2454 bond_slave_register(bundle->bond, port,
2455 port->up.ofp_port, port->up.netdev);
2458 bond_unref(bundle->bond);
2459 bundle->bond = NULL;
2462 /* If we changed something that would affect MAC learning, un-learn
2463 * everything on this port and force flow revalidation. */
2465 bundle_flush_macs(bundle, false);
2472 bundle_remove(struct ofport *port_)
2474 struct ofport_dpif *port = ofport_dpif_cast(port_);
2475 struct ofbundle *bundle = port->bundle;
2478 bundle_del_port(port);
2479 if (list_is_empty(&bundle->ports)) {
2480 bundle_destroy(bundle);
2481 } else if (list_is_short(&bundle->ports)) {
2482 bond_unref(bundle->bond);
2483 bundle->bond = NULL;
2489 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2491 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2492 struct ofport_dpif *port = port_;
2493 uint8_t ea[ETH_ADDR_LEN];
2496 error = netdev_get_etheraddr(port->up.netdev, ea);
2498 struct ofpbuf packet;
2501 ofpbuf_init(&packet, 0);
2502 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2504 memcpy(packet_pdu, pdu, pdu_size);
2506 ofproto_dpif_send_packet(port, &packet);
2507 ofpbuf_uninit(&packet);
2509 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2510 "%s (%s)", port->bundle->name,
2511 netdev_get_name(port->up.netdev), ovs_strerror(error));
2516 bundle_send_learning_packets(struct ofbundle *bundle)
2518 struct ofproto_dpif *ofproto = bundle->ofproto;
2519 struct ofpbuf *learning_packet;
2520 int error, n_packets, n_errors;
2521 struct mac_entry *e;
2522 struct list packets;
2524 list_init(&packets);
2525 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2526 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2527 if (e->port.p != bundle) {
2530 learning_packet = bond_compose_learning_packet(bundle->bond,
2533 /* Temporarily use 'frame' as a private pointer (see below). */
2534 ovs_assert(learning_packet->frame == ofpbuf_data(learning_packet));
2535 learning_packet->frame = port_void;
2536 list_push_back(&packets, &learning_packet->list_node);
2539 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2541 error = n_packets = n_errors = 0;
2542 LIST_FOR_EACH (learning_packet, list_node, &packets) {
2544 void *port_void = learning_packet->frame;
2546 /* Restore 'frame'. */
2547 learning_packet->frame = ofpbuf_data(learning_packet);
2548 ret = ofproto_dpif_send_packet(port_void, learning_packet);
2555 ofpbuf_list_delete(&packets);
2558 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2559 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2560 "packets, last error was: %s",
2561 bundle->name, n_errors, n_packets, ovs_strerror(error));
2563 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2564 bundle->name, n_packets);
2569 bundle_run(struct ofbundle *bundle)
2572 lacp_run(bundle->lacp, send_pdu_cb);
2575 struct ofport_dpif *port;
2577 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2578 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2581 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2582 bundle->ofproto->backer->need_revalidate = REV_BOND;
2585 if (bond_should_send_learning_packets(bundle->bond)) {
2586 bundle_send_learning_packets(bundle);
2592 bundle_wait(struct ofbundle *bundle)
2595 lacp_wait(bundle->lacp);
2598 bond_wait(bundle->bond);
2605 mirror_set__(struct ofproto *ofproto_, void *aux,
2606 const struct ofproto_mirror_settings *s)
2608 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2609 struct ofbundle **srcs, **dsts;
2614 mirror_destroy(ofproto->mbridge, aux);
2618 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2619 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2621 for (i = 0; i < s->n_srcs; i++) {
2622 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2625 for (i = 0; i < s->n_dsts; i++) {
2626 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2629 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2630 s->n_dsts, s->src_vlans,
2631 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2638 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2639 uint64_t *packets, uint64_t *bytes)
2641 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2646 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2648 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2649 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2650 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2651 mac_learning_flush(ofproto->ml);
2653 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2658 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2660 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2661 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2662 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2666 forward_bpdu_changed(struct ofproto *ofproto_)
2668 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2669 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2673 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2676 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2677 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2678 mac_learning_set_idle_time(ofproto->ml, idle_time);
2679 mac_learning_set_max_entries(ofproto->ml, max_entries);
2680 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2683 /* Configures multicast snooping on 'ofport' using the settings
2684 * defined in 's'. */
2686 set_mcast_snooping(struct ofproto *ofproto_,
2687 const struct ofproto_mcast_snooping_settings *s)
2689 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2691 /* Only revalidate flows if the configuration changed. */
2692 if (!s != !ofproto->ms) {
2693 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2698 ofproto->ms = mcast_snooping_create();
2701 ovs_rwlock_wrlock(&ofproto->ms->rwlock);
2702 mcast_snooping_set_idle_time(ofproto->ms, s->idle_time);
2703 mcast_snooping_set_max_entries(ofproto->ms, s->max_entries);
2704 if (mcast_snooping_set_flood_unreg(ofproto->ms, s->flood_unreg)) {
2705 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2707 ovs_rwlock_unlock(&ofproto->ms->rwlock);
2709 mcast_snooping_unref(ofproto->ms);
2716 /* Configures multicast snooping port's flood setting on 'ofproto'. */
2718 set_mcast_snooping_port(struct ofproto *ofproto_, void *aux, bool flood)
2720 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2721 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2724 ovs_rwlock_wrlock(&ofproto->ms->rwlock);
2725 mcast_snooping_set_port_flood(ofproto->ms, bundle->vlan, bundle,
2727 ovs_rwlock_unlock(&ofproto->ms->rwlock);
2735 static struct ofport_dpif *
2736 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2738 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2739 return ofport ? ofport_dpif_cast(ofport) : NULL;
2743 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2744 struct ofproto_port *ofproto_port,
2745 struct dpif_port *dpif_port)
2747 ofproto_port->name = dpif_port->name;
2748 ofproto_port->type = dpif_port->type;
2749 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2753 ofport_update_peer(struct ofport_dpif *ofport)
2755 const struct ofproto_dpif *ofproto;
2756 struct dpif_backer *backer;
2759 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2763 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2764 backer->need_revalidate = REV_RECONFIGURE;
2767 ofport->peer->peer = NULL;
2768 ofport->peer = NULL;
2771 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2776 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2777 struct ofport *peer_ofport;
2778 struct ofport_dpif *peer;
2781 if (ofproto->backer != backer) {
2785 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2790 peer = ofport_dpif_cast(peer_ofport);
2791 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2792 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2794 ofport->peer = peer;
2795 ofport->peer->peer = ofport;
2805 port_run(struct ofport_dpif *ofport)
2807 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2808 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2809 bool enable = netdev_get_carrier(ofport->up.netdev);
2810 bool cfm_enable = false;
2811 bool bfd_enable = false;
2813 ofport->carrier_seq = carrier_seq;
2816 int cfm_opup = cfm_get_opup(ofport->cfm);
2818 cfm_enable = !cfm_get_fault(ofport->cfm);
2820 if (cfm_opup >= 0) {
2821 cfm_enable = cfm_enable && cfm_opup;
2826 bfd_enable = bfd_forwarding(ofport->bfd);
2829 if (ofport->bfd || ofport->cfm) {
2830 enable = enable && (cfm_enable || bfd_enable);
2833 if (ofport->bundle) {
2834 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2835 if (carrier_changed) {
2836 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2840 if (ofport->may_enable != enable) {
2841 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2842 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2845 ofport->may_enable = enable;
2849 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2850 struct ofproto_port *ofproto_port)
2852 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2853 struct dpif_port dpif_port;
2856 if (sset_contains(&ofproto->ghost_ports, devname)) {
2857 const char *type = netdev_get_type_from_name(devname);
2859 /* We may be called before ofproto->up.port_by_name is populated with
2860 * the appropriate ofport. For this reason, we must get the name and
2861 * type from the netdev layer directly. */
2863 const struct ofport *ofport;
2865 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2866 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2867 ofproto_port->name = xstrdup(devname);
2868 ofproto_port->type = xstrdup(type);
2874 if (!sset_contains(&ofproto->ports, devname)) {
2877 error = dpif_port_query_by_name(ofproto->backer->dpif,
2878 devname, &dpif_port);
2880 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2886 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2888 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2889 const char *devname = netdev_get_name(netdev);
2890 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2891 const char *dp_port_name;
2893 if (netdev_vport_is_patch(netdev)) {
2894 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2898 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2899 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2900 odp_port_t port_no = ODPP_NONE;
2903 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2907 if (netdev_get_tunnel_config(netdev)) {
2908 simap_put(&ofproto->backer->tnl_backers,
2909 dp_port_name, odp_to_u32(port_no));
2913 if (netdev_get_tunnel_config(netdev)) {
2914 sset_add(&ofproto->ghost_ports, devname);
2916 sset_add(&ofproto->ports, devname);
2922 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
2924 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2925 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2932 sset_find_and_delete(&ofproto->ghost_ports,
2933 netdev_get_name(ofport->up.netdev));
2934 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2935 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
2936 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
2938 /* The caller is going to close ofport->up.netdev. If this is a
2939 * bonded port, then the bond is using that netdev, so remove it
2940 * from the bond. The client will need to reconfigure everything
2941 * after deleting ports, so then the slave will get re-added. */
2942 bundle_remove(&ofport->up);
2949 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2951 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2954 error = netdev_get_stats(ofport->up.netdev, stats);
2956 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
2957 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2959 ovs_mutex_lock(&ofproto->stats_mutex);
2960 /* ofproto->stats.tx_packets represents packets that we created
2961 * internally and sent to some port (e.g. packets sent with
2962 * ofproto_dpif_send_packet()). Account for them as if they had
2963 * come from OFPP_LOCAL and got forwarded. */
2965 if (stats->rx_packets != UINT64_MAX) {
2966 stats->rx_packets += ofproto->stats.tx_packets;
2969 if (stats->rx_bytes != UINT64_MAX) {
2970 stats->rx_bytes += ofproto->stats.tx_bytes;
2973 /* ofproto->stats.rx_packets represents packets that were received on
2974 * some port and we processed internally and dropped (e.g. STP).
2975 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2977 if (stats->tx_packets != UINT64_MAX) {
2978 stats->tx_packets += ofproto->stats.rx_packets;
2981 if (stats->tx_bytes != UINT64_MAX) {
2982 stats->tx_bytes += ofproto->stats.rx_bytes;
2984 ovs_mutex_unlock(&ofproto->stats_mutex);
2990 struct port_dump_state {
2995 struct ofproto_port port;
3000 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3002 *statep = xzalloc(sizeof(struct port_dump_state));
3007 port_dump_next(const struct ofproto *ofproto_, void *state_,
3008 struct ofproto_port *port)
3010 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3011 struct port_dump_state *state = state_;
3012 const struct sset *sset;
3013 struct sset_node *node;
3015 if (state->has_port) {
3016 ofproto_port_destroy(&state->port);
3017 state->has_port = false;
3019 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3020 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3023 error = port_query_by_name(ofproto_, node->name, &state->port);
3025 *port = state->port;
3026 state->has_port = true;
3028 } else if (error != ENODEV) {
3033 if (!state->ghost) {
3034 state->ghost = true;
3037 return port_dump_next(ofproto_, state_, port);
3044 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3046 struct port_dump_state *state = state_;
3048 if (state->has_port) {
3049 ofproto_port_destroy(&state->port);
3056 port_poll(const struct ofproto *ofproto_, char **devnamep)
3058 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3060 if (ofproto->port_poll_errno) {
3061 int error = ofproto->port_poll_errno;
3062 ofproto->port_poll_errno = 0;
3066 if (sset_is_empty(&ofproto->port_poll_set)) {
3070 *devnamep = sset_pop(&ofproto->port_poll_set);
3075 port_poll_wait(const struct ofproto *ofproto_)
3077 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3078 dpif_port_poll_wait(ofproto->backer->dpif);
3082 port_is_lacp_current(const struct ofport *ofport_)
3084 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3085 return (ofport->bundle && ofport->bundle->lacp
3086 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3090 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3091 * then delete it entirely. */
3093 rule_expire(struct rule_dpif *rule)
3094 OVS_REQUIRES(ofproto_mutex)
3096 uint16_t hard_timeout, idle_timeout;
3097 long long int now = time_msec();
3100 hard_timeout = rule->up.hard_timeout;
3101 idle_timeout = rule->up.idle_timeout;
3103 /* Has 'rule' expired? */
3105 long long int modified;
3107 ovs_mutex_lock(&rule->up.mutex);
3108 modified = rule->up.modified;
3109 ovs_mutex_unlock(&rule->up.mutex);
3111 if (now > modified + hard_timeout * 1000) {
3112 reason = OFPRR_HARD_TIMEOUT;
3116 if (reason < 0 && idle_timeout) {
3119 ovs_mutex_lock(&rule->stats_mutex);
3120 used = rule->stats.used;
3121 ovs_mutex_unlock(&rule->stats_mutex);
3123 if (now > used + idle_timeout * 1000) {
3124 reason = OFPRR_IDLE_TIMEOUT;
3129 COVERAGE_INC(ofproto_dpif_expired);
3130 ofproto_rule_expire(&rule->up, reason);
3134 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3135 * 'flow' must reflect the data in 'packet'. */
3137 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3138 const struct flow *flow,
3139 struct rule_dpif *rule,
3140 const struct ofpact *ofpacts, size_t ofpacts_len,
3141 struct ofpbuf *packet)
3143 struct dpif_flow_stats stats;
3144 struct xlate_out xout;
3145 struct xlate_in xin;
3147 struct dpif_execute execute;
3150 ovs_assert((rule != NULL) != (ofpacts != NULL));
3152 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3155 rule_dpif_credit_stats(rule, &stats);
3158 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
3159 xin.ofpacts = ofpacts;
3160 xin.ofpacts_len = ofpacts_len;
3161 xin.resubmit_stats = &stats;
3162 xlate_actions(&xin, &xout);
3164 execute.actions = ofpbuf_data(&xout.odp_actions);
3165 execute.actions_len = ofpbuf_size(&xout.odp_actions);
3166 execute.packet = packet;
3167 execute.md = pkt_metadata_from_flow(flow);
3168 execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
3170 /* Fix up in_port. */
3171 in_port = flow->in_port.ofp_port;
3172 if (in_port == OFPP_NONE) {
3173 in_port = OFPP_LOCAL;
3175 execute.md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
3177 error = dpif_execute(ofproto->backer->dpif, &execute);
3179 xlate_out_uninit(&xout);
3185 rule_dpif_credit_stats(struct rule_dpif *rule,
3186 const struct dpif_flow_stats *stats)
3188 ovs_mutex_lock(&rule->stats_mutex);
3189 rule->stats.n_packets += stats->n_packets;
3190 rule->stats.n_bytes += stats->n_bytes;
3191 rule->stats.used = MAX(rule->stats.used, stats->used);
3192 ovs_mutex_unlock(&rule->stats_mutex);
3196 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
3197 OVS_REQUIRES(rule->up.mutex)
3199 return rule->up.flow_cookie;
3203 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
3204 uint16_t hard_timeout)
3206 ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
3209 /* Returns 'rule''s actions. The caller owns a reference on the returned
3210 * actions and must eventually release it (with rule_actions_unref()) to avoid
3212 const struct rule_actions *
3213 rule_dpif_get_actions(const struct rule_dpif *rule)
3215 return rule_get_actions(&rule->up);
3218 /* Sets 'rule''s recirculation id. */
3220 rule_dpif_set_recirc_id(struct rule_dpif *rule, uint32_t id)
3221 OVS_REQUIRES(rule->up.mutex)
3223 ovs_assert(!rule->recirc_id);
3224 rule->recirc_id = id;
3227 /* Returns 'rule''s recirculation id. */
3229 rule_dpif_get_recirc_id(struct rule_dpif *rule)
3230 OVS_REQUIRES(rule->up.mutex)
3232 if (!rule->recirc_id) {
3233 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3235 rule_dpif_set_recirc_id(rule, ofproto_dpif_alloc_recirc_id(ofproto));
3237 return rule->recirc_id;
3240 /* Sets 'rule''s recirculation id. */
3242 rule_set_recirc_id(struct rule *rule_, uint32_t id)
3244 struct rule_dpif *rule = rule_dpif_cast(rule_);
3246 ovs_mutex_lock(&rule->up.mutex);
3247 rule_dpif_set_recirc_id(rule, id);
3248 ovs_mutex_unlock(&rule->up.mutex);
3251 /* Lookup 'flow' in table 0 of 'ofproto''s classifier.
3252 * If 'wc' is non-null, sets the fields that were relevant as part of
3253 * the lookup. Returns the table_id where a match or miss occurred.
3255 * The return value will be zero unless there was a miss and
3256 * OFPTC11_TABLE_MISS_CONTINUE is in effect for the sequence of tables
3257 * where misses occur.
3259 * The rule is returned in '*rule', which is valid at least until the next
3260 * RCU quiescent period. If the '*rule' needs to stay around longer,
3261 * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3262 * on it before this returns. */
3264 rule_dpif_lookup(struct ofproto_dpif *ofproto, struct flow *flow,
3265 struct flow_wildcards *wc, struct rule_dpif **rule,
3266 bool take_ref, const struct dpif_flow_stats *stats)
3268 enum rule_dpif_lookup_verdict verdict;
3269 enum ofputil_port_config config = 0;
3272 if (ofproto_dpif_get_enable_recirc(ofproto)) {
3273 /* Always exactly match recirc_id since datapath supports
3276 wc->masks.recirc_id = UINT32_MAX;
3279 /* Start looking up from internal table for post recirculation flows
3280 * or packets. We can also simply send all, including normal flows
3281 * or packets to the internal table. They will not match any post
3282 * recirculation rules except the 'catch all' rule that resubmit
3285 * As an optimization, we send normal flows and packets to table 0
3286 * directly, saving one table lookup. */
3287 table_id = flow->recirc_id ? TBL_INTERNAL : 0;
3292 verdict = rule_dpif_lookup_from_table(ofproto, flow, wc, true,
3293 &table_id, rule, take_ref, stats);
3296 case RULE_DPIF_LOOKUP_VERDICT_MATCH:
3298 case RULE_DPIF_LOOKUP_VERDICT_CONTROLLER: {
3299 struct ofport_dpif *port;
3301 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
3303 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
3304 flow->in_port.ofp_port);
3306 config = port ? port->up.pp.config : 0;
3309 case RULE_DPIF_LOOKUP_VERDICT_DROP:
3310 config = OFPUTIL_PC_NO_PACKET_IN;
3312 case RULE_DPIF_LOOKUP_VERDICT_DEFAULT:
3313 if (!connmgr_wants_packet_in_on_miss(ofproto->up.connmgr)) {
3314 config = OFPUTIL_PC_NO_PACKET_IN;
3321 choose_miss_rule(config, ofproto->miss_rule,
3322 ofproto->no_packet_in_rule, rule, take_ref);
3326 /* The returned rule is valid at least until the next RCU quiescent period.
3327 * If the '*rule' needs to stay around longer, a non-zero 'take_ref' must be
3328 * passed in to cause a reference to be taken on it before this returns. */
3329 static struct rule_dpif *
3330 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, uint8_t table_id,
3331 const struct flow *flow, struct flow_wildcards *wc,
3334 struct classifier *cls = &ofproto->up.tables[table_id].cls;
3335 const struct cls_rule *cls_rule;
3336 struct rule_dpif *rule;
3337 struct flow ofpc_normal_flow;
3339 if (ofproto->up.frag_handling != OFPC_FRAG_NX_MATCH) {
3340 /* We always unwildcard dl_type and nw_frag (for IP), so they
3341 * need not be unwildcarded here. */
3343 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
3344 if (ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3345 /* We must pretend that transport ports are unavailable. */
3346 ofpc_normal_flow = *flow;
3347 ofpc_normal_flow.tp_src = htons(0);
3348 ofpc_normal_flow.tp_dst = htons(0);
3349 flow = &ofpc_normal_flow;
3351 /* Must be OFPC_FRAG_DROP (we don't have OFPC_FRAG_REASM).
3352 * Use the drop_frags_rule (which cannot disappear). */
3353 cls_rule = &ofproto->drop_frags_rule->up.cr;
3354 rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
3356 rule_dpif_ref(rule);
3364 cls_rule = classifier_lookup(cls, flow, wc);
3366 rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
3368 /* Try again if the rule was released before we get the reference. */
3369 } while (rule && take_ref && !rule_dpif_try_ref(rule));
3374 /* Look up 'flow' in 'ofproto''s classifier starting from table '*table_id'.
3375 * Stores the rule that was found in '*rule', or NULL if none was found.
3376 * Updates 'wc', if nonnull, to reflect the fields that were used during the
3379 * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3380 * if none is found then the table miss configuration for that table is
3381 * honored, which can result in additional lookups in other OpenFlow tables.
3382 * In this case the function updates '*table_id' to reflect the final OpenFlow
3383 * table that was searched.
3385 * If 'honor_table_miss' is false, then only one table lookup occurs, in
3390 * - RULE_DPIF_LOOKUP_VERDICT_MATCH if a rule (in '*rule') was found.
3392 * - RULE_OFPTC_TABLE_MISS_CONTROLLER if no rule was found and either:
3393 * + 'honor_table_miss' is false
3394 * + a table miss configuration specified that the packet should be
3395 * sent to the controller in this case.
3397 * - RULE_DPIF_LOOKUP_VERDICT_DROP if no rule was found, 'honor_table_miss'
3398 * is true and a table miss configuration specified that the packet
3399 * should be dropped in this case.
3401 * - RULE_DPIF_LOOKUP_VERDICT_DEFAULT if no rule was found,
3402 * 'honor_table_miss' is true and a table miss configuration has
3403 * not been specified in this case.
3405 * The rule is returned in '*rule', which is valid at least until the next
3406 * RCU quiescent period. If the '*rule' needs to stay around longer,
3407 * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3408 * on it before this returns. */
3409 enum rule_dpif_lookup_verdict
3410 rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto,
3411 const struct flow *flow,
3412 struct flow_wildcards *wc,
3413 bool honor_table_miss,
3414 uint8_t *table_id, struct rule_dpif **rule,
3415 bool take_ref, const struct dpif_flow_stats *stats)
3419 for (next_id = *table_id;
3420 next_id < ofproto->up.n_tables;
3421 next_id++, next_id += (next_id == TBL_INTERNAL))
3423 *table_id = next_id;
3424 *rule = rule_dpif_lookup_in_table(ofproto, *table_id, flow, wc,
3427 struct oftable *tbl = &ofproto->up.tables[next_id];
3428 atomic_ulong *stat = *rule ? &tbl->n_matched : &tbl->n_missed;
3430 atomic_add(stat, stats->n_packets, &orig);
3433 return RULE_DPIF_LOOKUP_VERDICT_MATCH;
3434 } else if (!honor_table_miss) {
3435 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3437 switch (ofproto_table_get_miss_config(&ofproto->up, *table_id)) {
3438 case OFPUTIL_TABLE_MISS_CONTINUE:
3441 case OFPUTIL_TABLE_MISS_CONTROLLER:
3442 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3444 case OFPUTIL_TABLE_MISS_DROP:
3445 return RULE_DPIF_LOOKUP_VERDICT_DROP;
3447 case OFPUTIL_TABLE_MISS_DEFAULT:
3448 return RULE_DPIF_LOOKUP_VERDICT_DEFAULT;
3453 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3456 /* Given a port configuration (specified as zero if there's no port), chooses
3457 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
3460 * The rule is returned in '*rule', which is valid at least until the next
3461 * RCU quiescent period. If the '*rule' needs to stay around longer,
3462 * a reference must be taken on it (rule_dpif_ref()).
3465 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
3466 struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule,
3469 *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
3471 rule_dpif_ref(*rule);
3476 complete_operation(struct rule_dpif *rule)
3477 OVS_REQUIRES(ofproto_mutex)
3479 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3481 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3484 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
3486 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
3489 static struct rule *
3492 struct rule_dpif *rule = xmalloc(sizeof *rule);
3497 rule_dealloc(struct rule *rule_)
3499 struct rule_dpif *rule = rule_dpif_cast(rule_);
3504 rule_construct(struct rule *rule_)
3505 OVS_NO_THREAD_SAFETY_ANALYSIS
3507 struct rule_dpif *rule = rule_dpif_cast(rule_);
3508 ovs_mutex_init_adaptive(&rule->stats_mutex);
3509 rule->stats.n_packets = 0;
3510 rule->stats.n_bytes = 0;
3511 rule->stats.used = rule->up.modified;
3512 rule->recirc_id = 0;
3518 rule_insert(struct rule *rule_)
3519 OVS_REQUIRES(ofproto_mutex)
3521 struct rule_dpif *rule = rule_dpif_cast(rule_);
3522 complete_operation(rule);
3527 rule_delete(struct rule *rule_)
3528 OVS_REQUIRES(ofproto_mutex)
3530 struct rule_dpif *rule = rule_dpif_cast(rule_);
3531 complete_operation(rule);
3535 rule_destruct(struct rule *rule_)
3537 struct rule_dpif *rule = rule_dpif_cast(rule_);
3539 ovs_mutex_destroy(&rule->stats_mutex);
3540 if (rule->recirc_id) {
3541 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3543 ofproto_dpif_free_recirc_id(ofproto, rule->recirc_id);
3548 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
3549 long long int *used)
3551 struct rule_dpif *rule = rule_dpif_cast(rule_);
3553 ovs_mutex_lock(&rule->stats_mutex);
3554 *packets = rule->stats.n_packets;
3555 *bytes = rule->stats.n_bytes;
3556 *used = rule->stats.used;
3557 ovs_mutex_unlock(&rule->stats_mutex);
3561 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
3562 struct ofpbuf *packet)
3564 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3566 ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
3570 rule_execute(struct rule *rule, const struct flow *flow,
3571 struct ofpbuf *packet)
3573 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
3574 ofpbuf_delete(packet);
3579 rule_modify_actions(struct rule *rule_, bool reset_counters)
3580 OVS_REQUIRES(ofproto_mutex)
3582 struct rule_dpif *rule = rule_dpif_cast(rule_);
3584 if (reset_counters) {
3585 ovs_mutex_lock(&rule->stats_mutex);
3586 rule->stats.n_packets = 0;
3587 rule->stats.n_bytes = 0;
3588 ovs_mutex_unlock(&rule->stats_mutex);
3591 complete_operation(rule);
3594 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
3596 return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
3599 static struct ofgroup *
3602 struct group_dpif *group = xzalloc(sizeof *group);
3607 group_dealloc(struct ofgroup *group_)
3609 struct group_dpif *group = group_dpif_cast(group_);
3614 group_construct_stats(struct group_dpif *group)
3615 OVS_REQUIRES(group->stats_mutex)
3617 struct ofputil_bucket *bucket;
3618 const struct list *buckets;
3620 group->packet_count = 0;
3621 group->byte_count = 0;
3623 group_dpif_get_buckets(group, &buckets);
3624 LIST_FOR_EACH (bucket, list_node, buckets) {
3625 bucket->stats.packet_count = 0;
3626 bucket->stats.byte_count = 0;
3631 group_dpif_credit_stats(struct group_dpif *group,
3632 struct ofputil_bucket *bucket,
3633 const struct dpif_flow_stats *stats)
3635 ovs_mutex_lock(&group->stats_mutex);
3636 group->packet_count += stats->n_packets;
3637 group->byte_count += stats->n_bytes;
3639 bucket->stats.packet_count += stats->n_packets;
3640 bucket->stats.byte_count += stats->n_bytes;
3641 } else { /* Credit to all buckets */
3642 const struct list *buckets;
3644 group_dpif_get_buckets(group, &buckets);
3645 LIST_FOR_EACH (bucket, list_node, buckets) {
3646 bucket->stats.packet_count += stats->n_packets;
3647 bucket->stats.byte_count += stats->n_bytes;
3650 ovs_mutex_unlock(&group->stats_mutex);
3654 group_construct(struct ofgroup *group_)
3656 struct group_dpif *group = group_dpif_cast(group_);
3657 const struct ofputil_bucket *bucket;
3659 /* Prevent group chaining because our locking structure makes it hard to
3660 * implement deadlock-free. (See xlate_group_resource_check().) */
3661 LIST_FOR_EACH (bucket, list_node, &group->up.buckets) {
3662 const struct ofpact *a;
3664 OFPACT_FOR_EACH (a, bucket->ofpacts, bucket->ofpacts_len) {
3665 if (a->type == OFPACT_GROUP) {
3666 return OFPERR_OFPGMFC_CHAINING_UNSUPPORTED;
3671 ovs_mutex_init_adaptive(&group->stats_mutex);
3672 ovs_mutex_lock(&group->stats_mutex);
3673 group_construct_stats(group);
3674 ovs_mutex_unlock(&group->stats_mutex);
3679 group_destruct(struct ofgroup *group_)
3681 struct group_dpif *group = group_dpif_cast(group_);
3682 ovs_mutex_destroy(&group->stats_mutex);
3686 group_modify(struct ofgroup *group_)
3688 struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
3690 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3696 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
3698 struct group_dpif *group = group_dpif_cast(group_);
3699 struct ofputil_bucket *bucket;
3700 const struct list *buckets;
3701 struct bucket_counter *bucket_stats;
3703 ovs_mutex_lock(&group->stats_mutex);
3704 ogs->packet_count = group->packet_count;
3705 ogs->byte_count = group->byte_count;
3707 group_dpif_get_buckets(group, &buckets);
3708 bucket_stats = ogs->bucket_stats;
3709 LIST_FOR_EACH (bucket, list_node, buckets) {
3710 bucket_stats->packet_count = bucket->stats.packet_count;
3711 bucket_stats->byte_count = bucket->stats.byte_count;
3714 ovs_mutex_unlock(&group->stats_mutex);
3719 /* If the group exists, this function increments the groups's reference count.
3721 * Make sure to call group_dpif_unref() after no longer needing to maintain
3722 * a reference to the group. */
3724 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
3725 struct group_dpif **group)
3727 struct ofgroup *ofgroup;
3730 found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
3731 *group = found ? group_dpif_cast(ofgroup) : NULL;
3737 group_dpif_get_buckets(const struct group_dpif *group,
3738 const struct list **buckets)
3740 *buckets = &group->up.buckets;
3743 enum ofp11_group_type
3744 group_dpif_get_type(const struct group_dpif *group)
3746 return group->up.type;
3749 /* Sends 'packet' out 'ofport'.
3750 * May modify 'packet'.
3751 * Returns 0 if successful, otherwise a positive errno value. */
3753 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
3755 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3758 error = xlate_send_packet(ofport, packet);
3760 ovs_mutex_lock(&ofproto->stats_mutex);
3761 ofproto->stats.tx_packets++;
3762 ofproto->stats.tx_bytes += ofpbuf_size(packet);
3763 ovs_mutex_unlock(&ofproto->stats_mutex);
3768 set_frag_handling(struct ofproto *ofproto_,
3769 enum ofp_config_flags frag_handling)
3771 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3772 if (frag_handling != OFPC_FRAG_REASM) {
3773 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3781 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
3782 const struct flow *flow,
3783 const struct ofpact *ofpacts, size_t ofpacts_len)
3785 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3787 ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
3788 ofpacts_len, packet);
3795 set_netflow(struct ofproto *ofproto_,
3796 const struct netflow_options *netflow_options)
3798 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3800 if (netflow_options) {
3801 if (!ofproto->netflow) {
3802 ofproto->netflow = netflow_create();
3803 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3805 return netflow_set_options(ofproto->netflow, netflow_options);
3806 } else if (ofproto->netflow) {
3807 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3808 netflow_unref(ofproto->netflow);
3809 ofproto->netflow = NULL;
3816 get_netflow_ids(const struct ofproto *ofproto_,
3817 uint8_t *engine_type, uint8_t *engine_id)
3819 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3821 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
3824 static struct ofproto_dpif *
3825 ofproto_dpif_lookup(const char *name)
3827 struct ofproto_dpif *ofproto;
3829 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
3830 hash_string(name, 0), &all_ofproto_dpifs) {
3831 if (!strcmp(ofproto->up.name, name)) {
3839 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
3840 const char *argv[], void *aux OVS_UNUSED)
3842 struct ofproto_dpif *ofproto;
3845 ofproto = ofproto_dpif_lookup(argv[1]);
3847 unixctl_command_reply_error(conn, "no such bridge");
3850 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3851 mac_learning_flush(ofproto->ml);
3852 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3854 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3855 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3856 mac_learning_flush(ofproto->ml);
3857 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3861 unixctl_command_reply(conn, "table successfully flushed");
3865 ofproto_unixctl_mcast_snooping_flush(struct unixctl_conn *conn, int argc,
3866 const char *argv[], void *aux OVS_UNUSED)
3868 struct ofproto_dpif *ofproto;
3871 ofproto = ofproto_dpif_lookup(argv[1]);
3873 unixctl_command_reply_error(conn, "no such bridge");
3877 if (!mcast_snooping_enabled(ofproto->ms)) {
3878 unixctl_command_reply_error(conn, "multicast snooping is disabled");
3881 mcast_snooping_mdb_flush(ofproto->ms);
3883 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3884 if (!mcast_snooping_enabled(ofproto->ms)) {
3887 mcast_snooping_mdb_flush(ofproto->ms);
3891 unixctl_command_reply(conn, "table successfully flushed");
3894 static struct ofport_dpif *
3895 ofbundle_get_a_port(const struct ofbundle *bundle)
3897 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
3902 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
3903 const char *argv[], void *aux OVS_UNUSED)
3905 struct ds ds = DS_EMPTY_INITIALIZER;
3906 const struct ofproto_dpif *ofproto;
3907 const struct mac_entry *e;
3909 ofproto = ofproto_dpif_lookup(argv[1]);
3911 unixctl_command_reply_error(conn, "no such bridge");
3915 ds_put_cstr(&ds, " port VLAN MAC Age\n");
3916 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
3917 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3918 struct ofbundle *bundle = e->port.p;
3919 char name[OFP_MAX_PORT_NAME_LEN];
3921 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
3923 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
3924 name, e->vlan, ETH_ADDR_ARGS(e->mac),
3925 mac_entry_age(ofproto->ml, e));
3927 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3928 unixctl_command_reply(conn, ds_cstr(&ds));
3933 ofproto_unixctl_mcast_snooping_show(struct unixctl_conn *conn,
3934 int argc OVS_UNUSED,
3936 void *aux OVS_UNUSED)
3938 struct ds ds = DS_EMPTY_INITIALIZER;
3939 const struct ofproto_dpif *ofproto;
3940 const struct ofbundle *bundle;
3941 const struct mcast_group *grp;
3942 struct mcast_group_bundle *b;
3943 struct mcast_mrouter_bundle *mrouter;
3945 ofproto = ofproto_dpif_lookup(argv[1]);
3947 unixctl_command_reply_error(conn, "no such bridge");
3951 if (!mcast_snooping_enabled(ofproto->ms)) {
3952 unixctl_command_reply_error(conn, "multicast snooping is disabled");
3956 ds_put_cstr(&ds, " port VLAN GROUP Age\n");
3957 ovs_rwlock_rdlock(&ofproto->ms->rwlock);
3958 LIST_FOR_EACH (grp, group_node, &ofproto->ms->group_lru) {
3959 LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
3960 char name[OFP_MAX_PORT_NAME_LEN];
3963 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
3965 ds_put_format(&ds, "%5s %4d "IP_FMT" %3d\n",
3966 name, grp->vlan, IP_ARGS(grp->ip4),
3967 mcast_bundle_age(ofproto->ms, b));
3971 /* ports connected to multicast routers */
3972 LIST_FOR_EACH(mrouter, mrouter_node, &ofproto->ms->mrouter_lru) {
3973 char name[OFP_MAX_PORT_NAME_LEN];
3975 bundle = mrouter->port;
3976 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
3978 ds_put_format(&ds, "%5s %4d querier %3d\n",
3979 name, mrouter->vlan,
3980 mcast_mrouter_age(ofproto->ms, mrouter));
3982 ovs_rwlock_unlock(&ofproto->ms->rwlock);
3983 unixctl_command_reply(conn, ds_cstr(&ds));
3988 struct xlate_out xout;
3989 struct xlate_in xin;
3990 const struct flow *key;
3992 struct flow_wildcards wc;
3997 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
3999 const struct rule_actions *actions;
4002 ds_put_char_multiple(result, '\t', level);
4004 ds_put_cstr(result, "No match\n");
4008 ovs_mutex_lock(&rule->up.mutex);
4009 cookie = rule->up.flow_cookie;
4010 ovs_mutex_unlock(&rule->up.mutex);
4012 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
4013 rule ? rule->up.table_id : 0, ntohll(cookie));
4014 cls_rule_format(&rule->up.cr, result);
4015 ds_put_char(result, '\n');
4017 actions = rule_dpif_get_actions(rule);
4019 ds_put_char_multiple(result, '\t', level);
4020 ds_put_cstr(result, "OpenFlow actions=");
4021 ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
4022 ds_put_char(result, '\n');
4026 trace_format_flow(struct ds *result, int level, const char *title,
4027 struct trace_ctx *trace)
4029 ds_put_char_multiple(result, '\t', level);
4030 ds_put_format(result, "%s: ", title);
4031 /* Do not report unchanged flows for resubmits. */
4032 if ((level > 0 && flow_equal(&trace->xin.flow, &trace->flow))
4033 || (level == 0 && flow_equal(&trace->xin.flow, trace->key))) {
4034 ds_put_cstr(result, "unchanged");
4036 flow_format(result, &trace->xin.flow);
4037 trace->flow = trace->xin.flow;
4039 ds_put_char(result, '\n');
4043 trace_format_regs(struct ds *result, int level, const char *title,
4044 struct trace_ctx *trace)
4048 ds_put_char_multiple(result, '\t', level);
4049 ds_put_format(result, "%s:", title);
4050 for (i = 0; i < FLOW_N_REGS; i++) {
4051 ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
4053 ds_put_char(result, '\n');
4057 trace_format_odp(struct ds *result, int level, const char *title,
4058 struct trace_ctx *trace)
4060 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
4062 ds_put_char_multiple(result, '\t', level);
4063 ds_put_format(result, "%s: ", title);
4064 format_odp_actions(result, ofpbuf_data(odp_actions),
4065 ofpbuf_size(odp_actions));
4066 ds_put_char(result, '\n');
4070 trace_format_megaflow(struct ds *result, int level, const char *title,
4071 struct trace_ctx *trace)
4075 ds_put_char_multiple(result, '\t', level);
4076 ds_put_format(result, "%s: ", title);
4077 flow_wildcards_or(&trace->wc, &trace->xout.wc, &trace->wc);
4078 match_init(&match, trace->key, &trace->wc);
4079 match_format(&match, result, OFP_DEFAULT_PRIORITY);
4080 ds_put_char(result, '\n');
4084 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
4086 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4087 struct ds *result = trace->result;
4089 ds_put_char(result, '\n');
4090 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
4091 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
4092 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
4093 trace_format_megaflow(result, recurse + 1, "Resubmitted megaflow", trace);
4094 trace_format_rule(result, recurse + 1, rule);
4098 trace_report(struct xlate_in *xin, const char *s, int recurse)
4100 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4101 struct ds *result = trace->result;
4103 ds_put_char_multiple(result, '\t', recurse);
4104 ds_put_cstr(result, s);
4105 ds_put_char(result, '\n');
4108 /* Parses the 'argc' elements of 'argv', ignoring argv[0]. The following
4109 * forms are supported:
4111 * - [dpname] odp_flow [-generate | packet]
4112 * - bridge br_flow [-generate | packet]
4114 * On success, initializes '*ofprotop' and 'flow' and returns NULL. On failure
4115 * returns a nonnull malloced error message. */
4116 static char * WARN_UNUSED_RESULT
4117 parse_flow_and_packet(int argc, const char *argv[],
4118 struct ofproto_dpif **ofprotop, struct flow *flow,
4119 struct ofpbuf **packetp)
4121 const struct dpif_backer *backer = NULL;
4122 const char *error = NULL;
4124 struct simap port_names = SIMAP_INITIALIZER(&port_names);
4125 struct ofpbuf *packet;
4126 struct ofpbuf odp_key;
4127 struct ofpbuf odp_mask;
4129 ofpbuf_init(&odp_key, 0);
4130 ofpbuf_init(&odp_mask, 0);
4132 /* Handle "-generate" or a hex string as the last argument. */
4133 if (!strcmp(argv[argc - 1], "-generate")) {
4134 packet = ofpbuf_new(0);
4137 error = eth_from_hex(argv[argc - 1], &packet);
4140 } else if (argc == 4) {
4141 /* The 3-argument form must end in "-generate' or a hex string. */
4147 /* odp_flow can have its in_port specified as a name instead of port no.
4148 * We do not yet know whether a given flow is a odp_flow or a br_flow.
4149 * But, to know whether a flow is odp_flow through odp_flow_from_string(),
4150 * we need to create a simap of name to port no. */
4152 const char *dp_type;
4153 if (!strncmp(argv[1], "ovs-", 4)) {
4154 dp_type = argv[1] + 4;
4158 backer = shash_find_data(&all_dpif_backers, dp_type);
4159 } else if (argc == 2) {
4160 struct shash_node *node;
4161 if (shash_count(&all_dpif_backers) == 1) {
4162 node = shash_first(&all_dpif_backers);
4163 backer = node->data;
4166 error = "Syntax error";
4169 if (backer && backer->dpif) {
4170 struct dpif_port dpif_port;
4171 struct dpif_port_dump port_dump;
4172 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
4173 simap_put(&port_names, dpif_port.name,
4174 odp_to_u32(dpif_port.port_no));
4178 /* Parse the flow and determine whether a datapath or
4179 * bridge is specified. If function odp_flow_key_from_string()
4180 * returns 0, the flow is a odp_flow. If function
4181 * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
4182 if (!odp_flow_from_string(argv[argc - 1], &port_names,
4183 &odp_key, &odp_mask)) {
4185 error = "Cannot find the datapath";
4189 if (xlate_receive(backer, NULL, ofpbuf_data(&odp_key),
4190 ofpbuf_size(&odp_key), flow,
4191 ofprotop, NULL, NULL, NULL, NULL)) {
4192 error = "Invalid datapath flow";
4196 char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
4199 m_err = xasprintf("Bad flow syntax: %s", err);
4204 error = "Must specify bridge name";
4208 *ofprotop = ofproto_dpif_lookup(argv[1]);
4210 error = "Unknown bridge name";
4216 /* Generate a packet, if requested. */
4218 if (!ofpbuf_size(packet)) {
4219 flow_compose(packet, flow);
4221 struct pkt_metadata md = pkt_metadata_from_flow(flow);
4223 /* Use the metadata from the flow and the packet argument
4224 * to reconstruct the flow. */
4225 flow_extract(packet, &md, flow);
4230 if (error && !m_err) {
4231 m_err = xstrdup(error);
4234 ofpbuf_delete(packet);
4238 ofpbuf_uninit(&odp_key);
4239 ofpbuf_uninit(&odp_mask);
4240 simap_destroy(&port_names);
4245 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
4246 void *aux OVS_UNUSED)
4248 struct ofproto_dpif *ofproto;
4249 struct ofpbuf *packet;
4253 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4258 ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
4259 unixctl_command_reply(conn, ds_cstr(&result));
4260 ds_destroy(&result);
4261 ofpbuf_delete(packet);
4263 unixctl_command_reply_error(conn, error);
4269 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
4270 const char *argv[], void *aux OVS_UNUSED)
4272 enum ofputil_protocol usable_protocols;
4273 struct ofproto_dpif *ofproto;
4274 bool enforce_consistency;
4275 struct ofpbuf ofpacts;
4276 struct ofpbuf *packet;
4281 /* Three kinds of error return values! */
4287 ofpbuf_init(&ofpacts, 0);
4289 /* Parse actions. */
4290 error = parse_ofpacts(argv[--argc], &ofpacts, &usable_protocols);
4292 unixctl_command_reply_error(conn, error);
4297 /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
4298 * consistency between the flow and the actions. With -consistent, we
4299 * enforce consistency even for a flow supported in OpenFlow 1.0. */
4300 if (!strcmp(argv[1], "-consistent")) {
4301 enforce_consistency = true;
4305 enforce_consistency = false;
4308 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4310 unixctl_command_reply_error(conn, error);
4315 /* Do the same checks as handle_packet_out() in ofproto.c.
4317 * We pass a 'table_id' of 0 to ofproto_check_ofpacts(), which isn't
4318 * strictly correct because these actions aren't in any table, but it's OK
4319 * because it 'table_id' is used only to check goto_table instructions, but
4320 * packet-outs take a list of actions and therefore it can't include
4323 * We skip the "meter" check here because meter is an instruction, not an
4324 * action, and thus cannot appear in ofpacts. */
4325 in_port = ofp_to_u16(flow.in_port.ofp_port);
4326 if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
4327 unixctl_command_reply_error(conn, "invalid in_port");
4330 if (enforce_consistency) {
4331 retval = ofpacts_check_consistency(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts),
4332 &flow, u16_to_ofp(ofproto->up.max_ports),
4333 0, 0, usable_protocols);
4335 retval = ofpacts_check(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &flow,
4336 u16_to_ofp(ofproto->up.max_ports), 0, 0,
4342 ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
4343 unixctl_command_reply_error(conn, ds_cstr(&result));
4347 ofproto_trace(ofproto, &flow, packet,
4348 ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &result);
4349 unixctl_command_reply(conn, ds_cstr(&result));
4352 ds_destroy(&result);
4353 ofpbuf_delete(packet);
4354 ofpbuf_uninit(&ofpacts);
4357 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
4358 * description of the results to 'ds'.
4360 * The trace follows a packet with the specified 'flow' through the flow
4361 * table. 'packet' may be nonnull to trace an actual packet, with consequent
4362 * side effects (if it is nonnull then its flow must be 'flow').
4364 * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
4365 * trace, otherwise the actions are determined by a flow table lookup. */
4367 ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
4368 const struct ofpbuf *packet,
4369 const struct ofpact ofpacts[], size_t ofpacts_len,
4372 struct rule_dpif *rule;
4373 struct trace_ctx trace;
4375 ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
4376 ds_put_cstr(ds, "Flow: ");
4377 flow_format(ds, flow);
4378 ds_put_char(ds, '\n');
4380 flow_wildcards_init_catchall(&trace.wc);
4384 rule_dpif_lookup(ofproto, flow, &trace.wc, &rule, false, NULL);
4386 trace_format_rule(ds, 0, rule);
4387 if (rule == ofproto->miss_rule) {
4388 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
4389 } else if (rule == ofproto->no_packet_in_rule) {
4390 ds_put_cstr(ds, "\nNo match, packets dropped because "
4391 "OFPPC_NO_PACKET_IN is set on in_port.\n");
4392 } else if (rule == ofproto->drop_frags_rule) {
4393 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
4394 "and the fragment handling mode is \"drop\".\n");
4398 if (rule || ofpacts) {
4400 trace.key = flow; /* Original flow key, used for megaflow. */
4401 trace.flow = *flow; /* May be modified by actions. */
4402 xlate_in_init(&trace.xin, ofproto, flow, rule, ntohs(flow->tcp_flags),
4405 trace.xin.ofpacts = ofpacts;
4406 trace.xin.ofpacts_len = ofpacts_len;
4408 trace.xin.resubmit_hook = trace_resubmit;
4409 trace.xin.report_hook = trace_report;
4411 xlate_actions(&trace.xin, &trace.xout);
4413 ds_put_char(ds, '\n');
4414 trace_format_flow(ds, 0, "Final flow", &trace);
4415 trace_format_megaflow(ds, 0, "Megaflow", &trace);
4417 ds_put_cstr(ds, "Datapath actions: ");
4418 format_odp_actions(ds, ofpbuf_data(&trace.xout.odp_actions),
4419 ofpbuf_size(&trace.xout.odp_actions));
4421 if (trace.xout.slow) {
4422 enum slow_path_reason slow;
4424 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
4425 "slow path because it:");
4427 slow = trace.xout.slow;
4429 enum slow_path_reason bit = rightmost_1bit(slow);
4431 ds_put_format(ds, "\n\t- %s.",
4432 slow_path_reason_to_explanation(bit));
4438 xlate_out_uninit(&trace.xout);
4442 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
4443 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
4444 * to destroy 'ofproto_shash' and free the returned value. */
4445 static const struct shash_node **
4446 get_ofprotos(struct shash *ofproto_shash)
4448 const struct ofproto_dpif *ofproto;
4450 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4451 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
4452 shash_add_nocopy(ofproto_shash, name, ofproto);
4455 return shash_sort(ofproto_shash);
4459 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
4460 const char *argv[] OVS_UNUSED,
4461 void *aux OVS_UNUSED)
4463 struct ds ds = DS_EMPTY_INITIALIZER;
4464 struct shash ofproto_shash;
4465 const struct shash_node **sorted_ofprotos;
4468 shash_init(&ofproto_shash);
4469 sorted_ofprotos = get_ofprotos(&ofproto_shash);
4470 for (i = 0; i < shash_count(&ofproto_shash); i++) {
4471 const struct shash_node *node = sorted_ofprotos[i];
4472 ds_put_format(&ds, "%s\n", node->name);
4475 shash_destroy(&ofproto_shash);
4476 free(sorted_ofprotos);
4478 unixctl_command_reply(conn, ds_cstr(&ds));
4483 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
4485 const struct shash_node **ofprotos;
4486 struct dpif_dp_stats dp_stats;
4487 struct shash ofproto_shash;
4490 dpif_get_dp_stats(backer->dpif, &dp_stats);
4492 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
4493 dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
4495 shash_init(&ofproto_shash);
4496 ofprotos = get_ofprotos(&ofproto_shash);
4497 for (i = 0; i < shash_count(&ofproto_shash); i++) {
4498 struct ofproto_dpif *ofproto = ofprotos[i]->data;
4499 const struct shash_node **ports;
4502 if (ofproto->backer != backer) {
4506 ds_put_format(ds, "\t%s:\n", ofproto->up.name);
4508 ports = shash_sort(&ofproto->up.port_by_name);
4509 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
4510 const struct shash_node *node = ports[j];
4511 struct ofport *ofport = node->data;
4513 odp_port_t odp_port;
4515 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
4518 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
4519 if (odp_port != ODPP_NONE) {
4520 ds_put_format(ds, "%"PRIu32":", odp_port);
4522 ds_put_cstr(ds, "none:");
4525 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
4528 if (!netdev_get_config(ofport->netdev, &config)) {
4529 const struct smap_node **nodes;
4532 nodes = smap_sort(&config);
4533 for (i = 0; i < smap_count(&config); i++) {
4534 const struct smap_node *node = nodes[i];
4535 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
4536 node->key, node->value);
4540 smap_destroy(&config);
4542 ds_put_char(ds, ')');
4543 ds_put_char(ds, '\n');
4547 shash_destroy(&ofproto_shash);
4552 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4553 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4555 struct ds ds = DS_EMPTY_INITIALIZER;
4556 const struct shash_node **backers;
4559 backers = shash_sort(&all_dpif_backers);
4560 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
4561 dpif_show_backer(backers[i]->data, &ds);
4565 unixctl_command_reply(conn, ds_cstr(&ds));
4570 ofproto_dpif_contains_flow(const struct ofproto_dpif *ofproto,
4571 const struct nlattr *key, size_t key_len)
4573 struct ofproto_dpif *ofp;
4576 xlate_receive(ofproto->backer, NULL, key, key_len, &flow, &ofp,
4577 NULL, NULL, NULL, NULL);
4578 return ofp == ofproto;
4582 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
4583 int argc OVS_UNUSED, const char *argv[],
4584 void *aux OVS_UNUSED)
4586 const struct ofproto_dpif *ofproto;
4588 struct ds ds = DS_EMPTY_INITIALIZER;
4589 bool verbosity = false;
4591 struct dpif_port dpif_port;
4592 struct dpif_port_dump port_dump;
4593 struct hmap portno_names;
4595 struct dpif_flow_dump *flow_dump;
4596 struct dpif_flow_dump_thread *flow_dump_thread;
4600 ofproto = ofproto_dpif_lookup(argv[argc - 1]);
4602 unixctl_command_reply_error(conn, "no such bridge");
4606 if (argc > 2 && !strcmp(argv[1], "-m")) {
4610 hmap_init(&portno_names);
4611 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
4612 odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
4616 flow_dump = dpif_flow_dump_create(ofproto->backer->dpif);
4617 flow_dump_thread = dpif_flow_dump_thread_create(flow_dump);
4618 while (dpif_flow_dump_next(flow_dump_thread, &f, 1)) {
4619 if (!ofproto_dpif_contains_flow(ofproto, f.key, f.key_len)) {
4623 odp_flow_format(f.key, f.key_len, f.mask, f.mask_len,
4624 &portno_names, &ds, verbosity);
4625 ds_put_cstr(&ds, ", ");
4626 dpif_flow_stats_format(&f.stats, &ds);
4627 ds_put_cstr(&ds, ", actions:");
4628 format_odp_actions(&ds, f.actions, f.actions_len);
4629 ds_put_char(&ds, '\n');
4631 dpif_flow_dump_thread_destroy(flow_dump_thread);
4632 error = dpif_flow_dump_destroy(flow_dump);
4636 ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
4637 unixctl_command_reply_error(conn, ds_cstr(&ds));
4639 unixctl_command_reply(conn, ds_cstr(&ds));
4641 odp_portno_names_destroy(&portno_names);
4642 hmap_destroy(&portno_names);
4647 ofproto_dpif_unixctl_init(void)
4649 static bool registered;
4655 unixctl_command_register(
4657 "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
4658 1, 3, ofproto_unixctl_trace, NULL);
4659 unixctl_command_register(
4660 "ofproto/trace-packet-out",
4661 "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
4662 2, 6, ofproto_unixctl_trace_actions, NULL);
4663 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
4664 ofproto_unixctl_fdb_flush, NULL);
4665 unixctl_command_register("fdb/show", "bridge", 1, 1,
4666 ofproto_unixctl_fdb_show, NULL);
4667 unixctl_command_register("mdb/flush", "[bridge]", 0, 1,
4668 ofproto_unixctl_mcast_snooping_flush, NULL);
4669 unixctl_command_register("mdb/show", "bridge", 1, 1,
4670 ofproto_unixctl_mcast_snooping_show, NULL);
4671 unixctl_command_register("dpif/dump-dps", "", 0, 0,
4672 ofproto_unixctl_dpif_dump_dps, NULL);
4673 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
4675 unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
4676 ofproto_unixctl_dpif_dump_flows, NULL);
4679 /* Returns true if 'table' is the table used for internal rules,
4680 * false otherwise. */
4682 table_is_internal(uint8_t table_id)
4684 return table_id == TBL_INTERNAL;
4687 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4689 * This is deprecated. It is only for compatibility with broken device drivers
4690 * in old versions of Linux that do not properly support VLANs when VLAN
4691 * devices are not used. When broken device drivers are no longer in
4692 * widespread use, we will delete these interfaces. */
4695 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
4697 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
4698 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
4700 if (realdev_ofp_port == ofport->realdev_ofp_port
4701 && vid == ofport->vlandev_vid) {
4705 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4707 if (ofport->realdev_ofp_port) {
4710 if (realdev_ofp_port && ofport->bundle) {
4711 /* vlandevs are enslaved to their realdevs, so they are not allowed to
4712 * themselves be part of a bundle. */
4713 bundle_set(ofport_->ofproto, ofport->bundle, NULL);
4716 ofport->realdev_ofp_port = realdev_ofp_port;
4717 ofport->vlandev_vid = vid;
4719 if (realdev_ofp_port) {
4720 vsp_add(ofport, realdev_ofp_port, vid);
4727 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
4729 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
4733 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
4734 OVS_EXCLUDED(ofproto->vsp_mutex)
4736 /* hmap_is_empty is thread safe. */
4737 return !hmap_is_empty(&ofproto->realdev_vid_map);
4742 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
4743 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
4744 OVS_REQUIRES(ofproto->vsp_mutex)
4746 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
4747 int vid = vlan_tci_to_vid(vlan_tci);
4748 const struct vlan_splinter *vsp;
4750 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
4751 hash_realdev_vid(realdev_ofp_port, vid),
4752 &ofproto->realdev_vid_map) {
4753 if (vsp->realdev_ofp_port == realdev_ofp_port
4754 && vsp->vid == vid) {
4755 return vsp->vlandev_ofp_port;
4759 return realdev_ofp_port;
4762 /* Returns the OFP port number of the Linux VLAN device that corresponds to
4763 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
4764 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
4765 * 'vlan_tci' 9, it would return the port number of eth0.9.
4767 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
4768 * function just returns its 'realdev_ofp_port' argument. */
4770 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
4771 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
4772 OVS_EXCLUDED(ofproto->vsp_mutex)
4776 /* hmap_is_empty is thread safe, see if we can return immediately. */
4777 if (hmap_is_empty(&ofproto->realdev_vid_map)) {
4778 return realdev_ofp_port;
4780 ovs_mutex_lock(&ofproto->vsp_mutex);
4781 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
4782 ovs_mutex_unlock(&ofproto->vsp_mutex);
4786 static struct vlan_splinter *
4787 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
4789 struct vlan_splinter *vsp;
4791 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
4792 hash_ofp_port(vlandev_ofp_port),
4793 &ofproto->vlandev_map) {
4794 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
4802 /* Returns the OpenFlow port number of the "real" device underlying the Linux
4803 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
4804 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
4805 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
4806 * eth0 and store 9 in '*vid'.
4808 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
4809 * VLAN device. Unless VLAN splinters are enabled, this is what this function
4812 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
4813 ofp_port_t vlandev_ofp_port, int *vid)
4814 OVS_REQUIRES(ofproto->vsp_mutex)
4816 if (!hmap_is_empty(&ofproto->vlandev_map)) {
4817 const struct vlan_splinter *vsp;
4819 vsp = vlandev_find(ofproto, vlandev_ofp_port);
4824 return vsp->realdev_ofp_port;
4830 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
4831 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
4832 * 'flow->in_port' to the "real" device backing the VLAN device, sets
4833 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
4834 * always the case unless VLAN splinters are enabled), returns false without
4835 * making any changes. */
4837 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
4838 OVS_EXCLUDED(ofproto->vsp_mutex)
4843 /* hmap_is_empty is thread safe. */
4844 if (hmap_is_empty(&ofproto->vlandev_map)) {
4848 ovs_mutex_lock(&ofproto->vsp_mutex);
4849 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
4850 ovs_mutex_unlock(&ofproto->vsp_mutex);
4855 /* Cause the flow to be processed as if it came in on the real device with
4856 * the VLAN device's VLAN ID. */
4857 flow->in_port.ofp_port = realdev;
4858 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
4863 vsp_remove(struct ofport_dpif *port)
4865 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
4866 struct vlan_splinter *vsp;
4868 ovs_mutex_lock(&ofproto->vsp_mutex);
4869 vsp = vlandev_find(ofproto, port->up.ofp_port);
4871 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
4872 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
4875 port->realdev_ofp_port = 0;
4877 VLOG_ERR("missing vlan device record");
4879 ovs_mutex_unlock(&ofproto->vsp_mutex);
4883 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
4885 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
4887 ovs_mutex_lock(&ofproto->vsp_mutex);
4888 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
4889 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
4890 == realdev_ofp_port)) {
4891 struct vlan_splinter *vsp;
4893 vsp = xmalloc(sizeof *vsp);
4894 vsp->realdev_ofp_port = realdev_ofp_port;
4895 vsp->vlandev_ofp_port = port->up.ofp_port;
4898 port->realdev_ofp_port = realdev_ofp_port;
4900 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
4901 hash_ofp_port(port->up.ofp_port));
4902 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
4903 hash_realdev_vid(realdev_ofp_port, vid));
4905 VLOG_ERR("duplicate vlan device record");
4907 ovs_mutex_unlock(&ofproto->vsp_mutex);
4911 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
4913 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
4914 return ofport ? ofport->odp_port : ODPP_NONE;
4917 struct ofport_dpif *
4918 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
4920 struct ofport_dpif *port;
4922 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
4923 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
4924 &backer->odp_to_ofport_map) {
4925 if (port->odp_port == odp_port) {
4926 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
4931 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
4936 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
4938 struct ofport_dpif *port;
4940 port = odp_port_to_ofport(ofproto->backer, odp_port);
4941 if (port && &ofproto->up == port->up.ofproto) {
4942 return port->up.ofp_port;
4949 ofproto_dpif_alloc_recirc_id(struct ofproto_dpif *ofproto)
4951 struct dpif_backer *backer = ofproto->backer;
4953 return recirc_id_alloc(backer->rid_pool);
4957 ofproto_dpif_free_recirc_id(struct ofproto_dpif *ofproto, uint32_t recirc_id)
4959 struct dpif_backer *backer = ofproto->backer;
4961 recirc_id_free(backer->rid_pool, recirc_id);
4965 ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
4966 const struct match *match, int priority,
4967 uint16_t idle_timeout,
4968 const struct ofpbuf *ofpacts,
4969 struct rule **rulep)
4971 struct ofputil_flow_mod fm;
4972 struct rule_dpif *rule;
4976 fm.priority = priority;
4977 fm.new_cookie = htonll(0);
4978 fm.cookie = htonll(0);
4979 fm.cookie_mask = htonll(0);
4980 fm.modify_cookie = false;
4981 fm.table_id = TBL_INTERNAL;
4982 fm.command = OFPFC_ADD;
4983 fm.idle_timeout = idle_timeout;
4984 fm.hard_timeout = 0;
4987 fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
4988 fm.ofpacts = ofpbuf_data(ofpacts);
4989 fm.ofpacts_len = ofpbuf_size(ofpacts);
4991 error = ofproto_flow_mod(&ofproto->up, &fm);
4993 VLOG_ERR_RL(&rl, "failed to add internal flow (%s)",
4994 ofperr_to_string(error));
4999 rule = rule_dpif_lookup_in_table(ofproto, TBL_INTERNAL, &fm.match.flow,
5000 &fm.match.wc, false);
5010 ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
5011 struct match *match, int priority)
5013 struct ofputil_flow_mod fm;
5017 fm.priority = priority;
5018 fm.new_cookie = htonll(0);
5019 fm.cookie = htonll(0);
5020 fm.cookie_mask = htonll(0);
5021 fm.modify_cookie = false;
5022 fm.table_id = TBL_INTERNAL;
5023 fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5024 fm.command = OFPFC_DELETE_STRICT;
5026 error = ofproto_flow_mod(&ofproto->up, &fm);
5028 VLOG_ERR_RL(&rl, "failed to delete internal flow (%s)",
5029 ofperr_to_string(error));
5036 const struct ofproto_class ofproto_dpif_class = {
5050 NULL, /* get_memory_usage. */
5051 type_get_memory_usage,
5069 port_is_lacp_current,
5070 NULL, /* rule_choose_table */
5079 NULL, /* rule_premodify_actions */
5080 rule_modify_actions,
5096 get_stp_port_status,
5104 is_mirror_output_bundle,
5105 forward_bpdu_changed,
5106 set_mac_table_config,
5108 set_mcast_snooping_port,
5110 NULL, /* meter_get_features */
5111 NULL, /* meter_set */
5112 NULL, /* meter_get */
5113 NULL, /* meter_del */
5114 group_alloc, /* group_alloc */
5115 group_construct, /* group_construct */
5116 group_destruct, /* group_destruct */
5117 group_dealloc, /* group_dealloc */
5118 group_modify, /* group_modify */
5119 group_get_stats, /* group_get_stats */