2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015 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"
34 #include "dynamic-string.h"
35 #include "fail-open.h"
36 #include "guarded-list.h"
40 #include "mac-learning.h"
41 #include "mcast-snooping.h"
42 #include "meta-flow.h"
43 #include "multipath.h"
44 #include "netdev-vport.h"
49 #include "odp-execute.h"
52 #include "ofp-actions.h"
53 #include "ofp-parse.h"
54 #include "ofp-print.h"
55 #include "ofproto-dpif-ipfix.h"
56 #include "ofproto-dpif-mirror.h"
57 #include "ofproto-dpif-monitor.h"
58 #include "ofproto-dpif-rid.h"
59 #include "ofproto-dpif-sflow.h"
60 #include "ofproto-dpif-upcall.h"
61 #include "ofproto-dpif-xlate.h"
62 #include "poll-loop.h"
64 #include "ovs-router.h"
70 #include "unaligned.h"
72 #include "vlan-bitmap.h"
73 #include "openvswitch/vlog.h"
75 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
77 COVERAGE_DEFINE(ofproto_dpif_expired);
78 COVERAGE_DEFINE(packet_in_overflow);
87 * - Do include packets and bytes from datapath flows which have not
88 * recently been processed by a revalidator. */
89 struct ovs_mutex stats_mutex;
90 struct dpif_flow_stats stats OVS_GUARDED;
92 /* In non-NULL, will point to a new rule (for which a reference is held) to
93 * which all the stats updates should be forwarded. This exists only
94 * transitionally when flows are replaced.
96 * Protected by stats_mutex. If both 'rule->stats_mutex' and
97 * 'rule->new_rule->stats_mutex' must be held together, acquire them in that
99 struct rule_dpif *new_rule OVS_GUARDED;
101 /* If non-zero then the recirculation id that has
102 * been allocated for use with this rule.
103 * The recirculation id and associated internal flow should
104 * be freed when the rule is freed */
108 /* RULE_CAST() depends on this. */
109 BUILD_ASSERT_DECL(offsetof(struct rule_dpif, up) == 0);
111 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes,
112 long long int *used);
113 static struct rule_dpif *rule_dpif_cast(const struct rule *);
114 static void rule_expire(struct rule_dpif *);
121 * - Do include packets and bytes from datapath flows which have not
122 * recently been processed by a revalidator. */
123 struct ovs_mutex stats_mutex;
124 uint64_t packet_count OVS_GUARDED; /* Number of packets received. */
125 uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */
129 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
130 struct ofproto_dpif *ofproto; /* Owning ofproto. */
131 void *aux; /* Key supplied by ofproto's client. */
132 char *name; /* Identifier for log messages. */
135 struct ovs_list ports; /* Contains "struct ofport"s. */
136 enum port_vlan_mode vlan_mode; /* VLAN mode */
137 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
138 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
139 * NULL if all VLANs are trunked. */
140 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
141 struct bond *bond; /* Nonnull iff more than one port. */
142 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
145 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
148 static void bundle_remove(struct ofport *);
149 static void bundle_update(struct ofbundle *);
150 static void bundle_destroy(struct ofbundle *);
151 static void bundle_del_port(struct ofport_dpif *);
152 static void bundle_run(struct ofbundle *);
153 static void bundle_wait(struct ofbundle *);
154 static void bundle_flush_macs(struct ofbundle *, bool);
155 static void bundle_move(struct ofbundle *, struct ofbundle *);
157 static void stp_run(struct ofproto_dpif *ofproto);
158 static void stp_wait(struct ofproto_dpif *ofproto);
159 static int set_stp_port(struct ofport *,
160 const struct ofproto_port_stp_settings *);
162 static void rstp_run(struct ofproto_dpif *ofproto);
163 static void set_rstp_port(struct ofport *,
164 const struct ofproto_port_rstp_settings *);
167 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
171 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
172 struct ovs_list bundle_node;/* In struct ofbundle's "ports" list. */
173 struct cfm *cfm; /* Connectivity Fault Management, if any. */
174 struct bfd *bfd; /* BFD, if any. */
175 struct lldp *lldp; /* lldp, if any. */
176 bool may_enable; /* May be enabled in bonds. */
177 bool is_tunnel; /* This port is a tunnel. */
178 bool is_layer3; /* This is a layer 3 port. */
179 long long int carrier_seq; /* Carrier status changes. */
180 struct ofport_dpif *peer; /* Peer if patch port. */
183 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
184 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
185 long long int stp_state_entered;
187 /* Rapid Spanning Tree. */
188 struct rstp_port *rstp_port; /* Rapid Spanning Tree Protocol, if any. */
189 enum rstp_state rstp_state; /* Always RSTP_DISABLED if RSTP not in use. */
191 /* Queue to DSCP mapping. */
192 struct ofproto_port_queue *qdscp;
195 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
197 * This is deprecated. It is only for compatibility with broken device
198 * drivers in old versions of Linux that do not properly support VLANs when
199 * VLAN devices are not used. When broken device drivers are no longer in
200 * widespread use, we will delete these interfaces. */
201 ofp_port_t realdev_ofp_port;
205 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
207 * This is deprecated. It is only for compatibility with broken device drivers
208 * in old versions of Linux that do not properly support VLANs when VLAN
209 * devices are not used. When broken device drivers are no longer in
210 * widespread use, we will delete these interfaces. */
211 struct vlan_splinter {
212 struct hmap_node realdev_vid_node;
213 struct hmap_node vlandev_node;
214 ofp_port_t realdev_ofp_port;
215 ofp_port_t vlandev_ofp_port;
219 static void vsp_remove(struct ofport_dpif *);
220 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
222 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
225 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
228 static struct ofport_dpif *
229 ofport_dpif_cast(const struct ofport *ofport)
231 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
234 static void port_run(struct ofport_dpif *);
235 static int set_bfd(struct ofport *, const struct smap *);
236 static int set_cfm(struct ofport *, const struct cfm_settings *);
237 static int set_lldp(struct ofport *ofport_, const struct smap *cfg);
238 static void ofport_update_peer(struct ofport_dpif *);
240 /* Reasons that we might need to revalidate every datapath flow, and
241 * corresponding coverage counters.
243 * A value of 0 means that there is no need to revalidate.
245 * It would be nice to have some cleaner way to integrate with coverage
246 * counters, but with only a few reasons I guess this is good enough for
248 enum revalidate_reason {
249 REV_RECONFIGURE = 1, /* Switch configuration changed. */
250 REV_STP, /* Spanning tree protocol port status change. */
251 REV_RSTP, /* RSTP port status change. */
252 REV_BOND, /* Bonding changed. */
253 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
254 REV_FLOW_TABLE, /* Flow table changed. */
255 REV_MAC_LEARNING, /* Mac learning changed. */
256 REV_MCAST_SNOOPING, /* Multicast snooping changed. */
258 COVERAGE_DEFINE(rev_reconfigure);
259 COVERAGE_DEFINE(rev_stp);
260 COVERAGE_DEFINE(rev_rstp);
261 COVERAGE_DEFINE(rev_bond);
262 COVERAGE_DEFINE(rev_port_toggled);
263 COVERAGE_DEFINE(rev_flow_table);
264 COVERAGE_DEFINE(rev_mac_learning);
265 COVERAGE_DEFINE(rev_mcast_snooping);
267 /* All datapaths of a given type share a single dpif backer instance. */
274 struct ovs_rwlock odp_to_ofport_lock;
275 struct hmap odp_to_ofport_map OVS_GUARDED; /* Contains "struct ofport"s. */
277 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
279 enum revalidate_reason need_revalidate; /* Revalidate all flows. */
281 bool recv_set_enable; /* Enables or disables receiving packets. */
283 /* Version string of the datapath stored in OVSDB. */
284 char *dp_version_string;
286 /* Datapath feature support. */
287 struct dpif_backer_support support;
288 struct atomic_count tnl_count;
291 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
292 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
294 struct ofproto_dpif {
295 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
297 struct dpif_backer *backer;
299 ATOMIC(cls_version_t) tables_version; /* For classifier lookups. */
301 uint64_t dump_seq; /* Last read of udpif_dump_seq(). */
303 /* Special OpenFlow rules. */
304 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
305 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
306 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
309 struct netflow *netflow;
310 struct dpif_sflow *sflow;
311 struct dpif_ipfix *ipfix;
312 struct hmap bundles; /* Contains "struct ofbundle"s. */
313 struct mac_learning *ml;
314 struct mcast_snooping *ms;
315 bool has_bonded_bundles;
317 struct mbridge *mbridge;
319 struct ovs_mutex stats_mutex;
320 struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
321 * consumed in userspace. */
325 long long int stp_last_tick;
327 /* Rapid Spanning Tree. */
329 long long int rstp_last_tick;
331 /* VLAN splinters. */
332 struct ovs_mutex vsp_mutex;
333 struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
334 struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */
337 struct sset ports; /* Set of standard port names. */
338 struct sset ghost_ports; /* Ports with no datapath port. */
339 struct sset port_poll_set; /* Queued names for port_poll() reply. */
340 int port_poll_errno; /* Last errno for port_poll() reply. */
341 uint64_t change_seq; /* Connectivity status changes. */
344 struct guarded_list pins; /* Contains "struct ofputil_packet_in"s. */
345 struct seq *pins_seq; /* For notifying 'pins' reception. */
349 /* All existing ofproto_dpif instances, indexed by ->up.name. */
350 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
352 static bool ofproto_use_tnl_push_pop = true;
353 static void ofproto_unixctl_init(void);
355 static inline struct ofproto_dpif *
356 ofproto_dpif_cast(const struct ofproto *ofproto)
358 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
359 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
363 ofproto_dpif_get_max_mpls_depth(const struct ofproto_dpif *ofproto)
365 return ofproto->backer->support.max_mpls_depth;
369 ofproto_dpif_get_enable_recirc(const struct ofproto_dpif *ofproto)
371 return ofproto->backer->support.recirc;
375 ofproto_dpif_get_enable_ufid(struct dpif_backer *backer)
377 return backer->support.ufid;
380 static void ofproto_trace(struct ofproto_dpif *, struct flow *,
381 const struct dp_packet *packet,
382 const struct ofpact[], size_t ofpacts_len,
385 /* Global variables. */
386 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
388 /* Initial mappings of port to bridge mappings. */
389 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
391 /* Executes 'fm'. The caller retains ownership of 'fm' and everything in
394 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
395 struct ofputil_flow_mod *fm)
397 ofproto_flow_mod(&ofproto->up, fm);
400 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
401 * Takes ownership of 'pin' and pin->packet. */
403 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
404 struct ofproto_packet_in *pin)
406 if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
407 COVERAGE_INC(packet_in_overflow);
408 free(CONST_CAST(void *, pin->up.packet));
412 /* Wakes up main thread for packet-in I/O. */
413 seq_change(ofproto->pins_seq);
416 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
417 * packet rather than to send the packet to the controller.
419 * This function returns false to indicate that a packet_in message
420 * for a "table-miss" should be sent to at least one controller.
421 * False otherwise. */
423 ofproto_dpif_wants_packet_in_on_miss(struct ofproto_dpif *ofproto)
425 return connmgr_wants_packet_in_on_miss(ofproto->up.connmgr);
428 /* Factory functions. */
431 init(const struct shash *iface_hints)
433 struct shash_node *node;
435 /* Make a local copy, since we don't own 'iface_hints' elements. */
436 SHASH_FOR_EACH(node, iface_hints) {
437 const struct iface_hint *orig_hint = node->data;
438 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
440 new_hint->br_name = xstrdup(orig_hint->br_name);
441 new_hint->br_type = xstrdup(orig_hint->br_type);
442 new_hint->ofp_port = orig_hint->ofp_port;
444 shash_add(&init_ofp_ports, node->name, new_hint);
449 enumerate_types(struct sset *types)
451 dp_enumerate_types(types);
455 enumerate_names(const char *type, struct sset *names)
457 struct ofproto_dpif *ofproto;
460 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
461 if (strcmp(type, ofproto->up.type)) {
464 sset_add(names, ofproto->up.name);
471 del(const char *type, const char *name)
476 error = dpif_open(name, type, &dpif);
478 error = dpif_delete(dpif);
485 port_open_type(const char *datapath_type, const char *port_type)
487 return dpif_port_open_type(datapath_type, port_type);
490 /* Type functions. */
492 static void process_dpif_port_changes(struct dpif_backer *);
493 static void process_dpif_all_ports_changed(struct dpif_backer *);
494 static void process_dpif_port_change(struct dpif_backer *,
495 const char *devname);
496 static void process_dpif_port_error(struct dpif_backer *, int error);
498 static struct ofproto_dpif *
499 lookup_ofproto_dpif_by_port_name(const char *name)
501 struct ofproto_dpif *ofproto;
503 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
504 if (sset_contains(&ofproto->ports, name)) {
513 type_run(const char *type)
515 struct dpif_backer *backer;
517 backer = shash_find_data(&all_dpif_backers, type);
519 /* This is not necessarily a problem, since backers are only
520 * created on demand. */
525 if (dpif_run(backer->dpif)) {
526 backer->need_revalidate = REV_RECONFIGURE;
529 udpif_run(backer->udpif);
531 /* If vswitchd started with other_config:flow_restore_wait set as "true",
532 * and the configuration has now changed to "false", enable receiving
533 * packets from the datapath. */
534 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
537 backer->recv_set_enable = true;
539 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
541 VLOG_ERR("Failed to enable receiving packets in dpif.");
544 dpif_flow_flush(backer->dpif);
545 backer->need_revalidate = REV_RECONFIGURE;
548 if (backer->recv_set_enable) {
549 udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
552 dpif_poll_threads_set(backer->dpif, n_dpdk_rxqs, pmd_cpu_mask);
554 if (backer->need_revalidate) {
555 struct ofproto_dpif *ofproto;
556 struct simap_node *node;
557 struct simap tmp_backers;
559 /* Handle tunnel garbage collection. */
560 simap_init(&tmp_backers);
561 simap_swap(&backer->tnl_backers, &tmp_backers);
563 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
564 struct ofport_dpif *iter;
566 if (backer != ofproto->backer) {
570 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
571 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
574 if (!iter->is_tunnel) {
578 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
579 namebuf, sizeof namebuf);
580 node = simap_find(&tmp_backers, dp_port);
582 simap_put(&backer->tnl_backers, dp_port, node->data);
583 simap_delete(&tmp_backers, node);
584 node = simap_find(&backer->tnl_backers, dp_port);
586 node = simap_find(&backer->tnl_backers, dp_port);
588 odp_port_t odp_port = ODPP_NONE;
590 if (!dpif_port_add(backer->dpif, iter->up.netdev,
592 simap_put(&backer->tnl_backers, dp_port,
593 odp_to_u32(odp_port));
594 node = simap_find(&backer->tnl_backers, dp_port);
599 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
600 if (tnl_port_reconfigure(iter, iter->up.netdev,
602 ovs_native_tunneling_is_on(ofproto), dp_port)) {
603 backer->need_revalidate = REV_RECONFIGURE;
608 SIMAP_FOR_EACH (node, &tmp_backers) {
609 dpif_port_del(backer->dpif, u32_to_odp(node->data));
611 simap_destroy(&tmp_backers);
613 switch (backer->need_revalidate) {
614 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
615 case REV_STP: COVERAGE_INC(rev_stp); break;
616 case REV_RSTP: COVERAGE_INC(rev_rstp); break;
617 case REV_BOND: COVERAGE_INC(rev_bond); break;
618 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
619 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
620 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
621 case REV_MCAST_SNOOPING: COVERAGE_INC(rev_mcast_snooping); break;
623 backer->need_revalidate = 0;
625 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
626 struct ofport_dpif *ofport;
627 struct ofbundle *bundle;
629 if (ofproto->backer != backer) {
634 xlate_ofproto_set(ofproto, ofproto->up.name,
635 ofproto->backer->dpif, ofproto->ml,
636 ofproto->stp, ofproto->rstp, ofproto->ms,
637 ofproto->mbridge, ofproto->sflow, ofproto->ipfix,
639 ofproto->up.forward_bpdu,
640 connmgr_has_in_band(ofproto->up.connmgr),
641 &ofproto->backer->support);
643 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
644 xlate_bundle_set(ofproto, bundle, bundle->name,
645 bundle->vlan_mode, bundle->vlan,
646 bundle->trunks, bundle->use_priority_tags,
647 bundle->bond, bundle->lacp,
651 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
652 int stp_port = ofport->stp_port
653 ? stp_port_no(ofport->stp_port)
655 xlate_ofport_set(ofproto, ofport->bundle, ofport,
656 ofport->up.ofp_port, ofport->odp_port,
657 ofport->up.netdev, ofport->cfm, ofport->bfd,
658 ofport->lldp, ofport->peer, stp_port,
659 ofport->rstp_port, ofport->qdscp,
660 ofport->n_qdscp, ofport->up.pp.config,
661 ofport->up.pp.state, ofport->is_tunnel,
667 udpif_revalidate(backer->udpif);
670 process_dpif_port_changes(backer);
675 /* Check for and handle port changes in 'backer''s dpif. */
677 process_dpif_port_changes(struct dpif_backer *backer)
683 error = dpif_port_poll(backer->dpif, &devname);
689 process_dpif_all_ports_changed(backer);
693 process_dpif_port_change(backer, devname);
698 process_dpif_port_error(backer, error);
705 process_dpif_all_ports_changed(struct dpif_backer *backer)
707 struct ofproto_dpif *ofproto;
708 struct dpif_port dpif_port;
709 struct dpif_port_dump dump;
710 struct sset devnames;
713 sset_init(&devnames);
714 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
715 if (ofproto->backer == backer) {
716 struct ofport *ofport;
718 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
719 sset_add(&devnames, netdev_get_name(ofport->netdev));
723 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
724 sset_add(&devnames, dpif_port.name);
727 SSET_FOR_EACH (devname, &devnames) {
728 process_dpif_port_change(backer, devname);
730 sset_destroy(&devnames);
734 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
736 struct ofproto_dpif *ofproto;
737 struct dpif_port port;
739 /* Don't report on the datapath's device. */
740 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
744 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
745 &all_ofproto_dpifs) {
746 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
751 ofproto = lookup_ofproto_dpif_by_port_name(devname);
752 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
753 /* The port was removed. If we know the datapath,
754 * report it through poll_set(). If we don't, it may be
755 * notifying us of a removal we initiated, so ignore it.
756 * If there's a pending ENOBUFS, let it stand, since
757 * everything will be reevaluated. */
758 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
759 sset_add(&ofproto->port_poll_set, devname);
760 ofproto->port_poll_errno = 0;
762 } else if (!ofproto) {
763 /* The port was added, but we don't know with which
764 * ofproto we should associate it. Delete it. */
765 dpif_port_del(backer->dpif, port.port_no);
767 struct ofport_dpif *ofport;
769 ofport = ofport_dpif_cast(shash_find_data(
770 &ofproto->up.port_by_name, devname));
772 && ofport->odp_port != port.port_no
773 && !odp_port_to_ofport(backer, port.port_no))
775 /* 'ofport''s datapath port number has changed from
776 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
777 * structures to match. */
778 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
779 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
780 ofport->odp_port = port.port_no;
781 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
782 hash_odp_port(port.port_no));
783 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
784 backer->need_revalidate = REV_RECONFIGURE;
787 dpif_port_destroy(&port);
790 /* Propagate 'error' to all ofprotos based on 'backer'. */
792 process_dpif_port_error(struct dpif_backer *backer, int error)
794 struct ofproto_dpif *ofproto;
796 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
797 if (ofproto->backer == backer) {
798 sset_clear(&ofproto->port_poll_set);
799 ofproto->port_poll_errno = error;
805 type_wait(const char *type)
807 struct dpif_backer *backer;
809 backer = shash_find_data(&all_dpif_backers, type);
811 /* This is not necessarily a problem, since backers are only
812 * created on demand. */
816 dpif_wait(backer->dpif);
819 /* Basic life-cycle. */
821 static int add_internal_flows(struct ofproto_dpif *);
823 static struct ofproto *
826 struct ofproto_dpif *ofproto = xzalloc(sizeof *ofproto);
831 dealloc(struct ofproto *ofproto_)
833 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
838 close_dpif_backer(struct dpif_backer *backer)
840 ovs_assert(backer->refcount > 0);
842 if (--backer->refcount) {
846 udpif_destroy(backer->udpif);
848 simap_destroy(&backer->tnl_backers);
849 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
850 hmap_destroy(&backer->odp_to_ofport_map);
851 shash_find_and_delete(&all_dpif_backers, backer->type);
853 free(backer->dp_version_string);
854 dpif_close(backer->dpif);
858 /* Datapath port slated for removal from datapath. */
860 struct ovs_list list_node;
864 static bool check_variable_length_userdata(struct dpif_backer *backer);
865 static void check_support(struct dpif_backer *backer);
868 open_dpif_backer(const char *type, struct dpif_backer **backerp)
870 struct dpif_backer *backer;
871 struct dpif_port_dump port_dump;
872 struct dpif_port port;
873 struct shash_node *node;
874 struct ovs_list garbage_list;
875 struct odp_garbage *garbage;
884 backer = shash_find_data(&all_dpif_backers, type);
891 backer_name = xasprintf("ovs-%s", type);
893 /* Remove any existing datapaths, since we assume we're the only
894 * userspace controlling the datapath. */
896 dp_enumerate_names(type, &names);
897 SSET_FOR_EACH(name, &names) {
898 struct dpif *old_dpif;
900 /* Don't remove our backer if it exists. */
901 if (!strcmp(name, backer_name)) {
905 if (dpif_open(name, type, &old_dpif)) {
906 VLOG_WARN("couldn't open old datapath %s to remove it", name);
908 dpif_delete(old_dpif);
909 dpif_close(old_dpif);
912 sset_destroy(&names);
914 backer = xmalloc(sizeof *backer);
916 error = dpif_create_and_open(backer_name, type, &backer->dpif);
919 VLOG_ERR("failed to open datapath of type %s: %s", type,
920 ovs_strerror(error));
924 backer->udpif = udpif_create(backer, backer->dpif);
926 backer->type = xstrdup(type);
927 backer->refcount = 1;
928 hmap_init(&backer->odp_to_ofport_map);
929 ovs_rwlock_init(&backer->odp_to_ofport_lock);
930 backer->need_revalidate = 0;
931 simap_init(&backer->tnl_backers);
932 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
935 if (backer->recv_set_enable) {
936 dpif_flow_flush(backer->dpif);
939 /* Loop through the ports already on the datapath and remove any
940 * that we don't need anymore. */
941 list_init(&garbage_list);
942 dpif_port_dump_start(&port_dump, backer->dpif);
943 while (dpif_port_dump_next(&port_dump, &port)) {
944 node = shash_find(&init_ofp_ports, port.name);
945 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
946 garbage = xmalloc(sizeof *garbage);
947 garbage->odp_port = port.port_no;
948 list_push_front(&garbage_list, &garbage->list_node);
951 dpif_port_dump_done(&port_dump);
953 LIST_FOR_EACH_POP (garbage, list_node, &garbage_list) {
954 dpif_port_del(backer->dpif, garbage->odp_port);
958 shash_add(&all_dpif_backers, type, backer);
960 check_support(backer);
961 atomic_count_init(&backer->tnl_count, 0);
963 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
965 VLOG_ERR("failed to listen on datapath of type %s: %s",
966 type, ovs_strerror(error));
967 close_dpif_backer(backer);
971 if (backer->recv_set_enable) {
972 udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
975 /* This check fails if performed before udpif threads have been set,
976 * as the kernel module checks that the 'pid' in userspace action
978 backer->support.variable_length_userdata
979 = check_variable_length_userdata(backer);
980 backer->dp_version_string = dpif_get_dp_version(backer->dpif);
986 ovs_native_tunneling_is_on(struct ofproto_dpif *ofproto)
988 return ofproto_use_tnl_push_pop && ofproto->backer->support.tnl_push_pop &&
989 atomic_count_get(&ofproto->backer->tnl_count);
992 /* Tests whether 'backer''s datapath supports recirculation. Only newer
993 * datapaths support OVS_KEY_ATTR_RECIRC_ID in keys. We need to disable some
994 * features on older datapaths that don't support this feature.
996 * Returns false if 'backer' definitely does not support recirculation, true if
997 * it seems to support recirculation or if at least the error we get is
1000 check_recirc(struct dpif_backer *backer)
1003 struct odputil_keybuf keybuf;
1007 memset(&flow, 0, sizeof flow);
1011 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1012 odp_flow_key_from_flow(&key, &flow, NULL, 0, true);
1013 enable_recirc = dpif_probe_feature(backer->dpif, "recirculation", &key,
1016 if (enable_recirc) {
1017 VLOG_INFO("%s: Datapath supports recirculation",
1018 dpif_name(backer->dpif));
1020 VLOG_INFO("%s: Datapath does not support recirculation",
1021 dpif_name(backer->dpif));
1024 return enable_recirc;
1027 /* Tests whether 'dpif' supports unique flow ids. We can skip serializing
1028 * some flow attributes for datapaths that support this feature.
1030 * Returns true if 'dpif' supports UFID for flow operations.
1031 * Returns false if 'dpif' does not support UFID. */
1033 check_ufid(struct dpif_backer *backer)
1036 struct odputil_keybuf keybuf;
1041 memset(&flow, 0, sizeof flow);
1042 flow.dl_type = htons(0x1234);
1044 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1045 odp_flow_key_from_flow(&key, &flow, NULL, 0, true);
1046 dpif_flow_hash(backer->dpif, key.data, key.size, &ufid);
1048 enable_ufid = dpif_probe_feature(backer->dpif, "UFID", &key, &ufid);
1051 VLOG_INFO("%s: Datapath supports unique flow ids",
1052 dpif_name(backer->dpif));
1054 VLOG_INFO("%s: Datapath does not support unique flow ids",
1055 dpif_name(backer->dpif));
1060 /* Tests whether 'backer''s datapath supports variable-length
1061 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions. We need
1062 * to disable some features on older datapaths that don't support this
1065 * Returns false if 'backer' definitely does not support variable-length
1066 * userdata, true if it seems to support them or if at least the error we get
1069 check_variable_length_userdata(struct dpif_backer *backer)
1071 struct eth_header *eth;
1072 struct ofpbuf actions;
1073 struct dpif_execute execute;
1074 struct dp_packet packet;
1078 /* Compose a userspace action that will cause an ERANGE error on older
1079 * datapaths that don't support variable-length userdata.
1081 * We really test for using userdata longer than 8 bytes, but older
1082 * datapaths accepted these, silently truncating the userdata to 8 bytes.
1083 * The same older datapaths rejected userdata shorter than 8 bytes, so we
1084 * test for that instead as a proxy for longer userdata support. */
1085 ofpbuf_init(&actions, 64);
1086 start = nl_msg_start_nested(&actions, OVS_ACTION_ATTR_USERSPACE);
1087 nl_msg_put_u32(&actions, OVS_USERSPACE_ATTR_PID,
1088 dpif_port_get_pid(backer->dpif, ODPP_NONE, 0));
1089 nl_msg_put_unspec_zero(&actions, OVS_USERSPACE_ATTR_USERDATA, 4);
1090 nl_msg_end_nested(&actions, start);
1092 /* Compose a dummy ethernet packet. */
1093 dp_packet_init(&packet, ETH_HEADER_LEN);
1094 eth = dp_packet_put_zeros(&packet, ETH_HEADER_LEN);
1095 eth->eth_type = htons(0x1234);
1097 /* Execute the actions. On older datapaths this fails with ERANGE, on
1098 * newer datapaths it succeeds. */
1099 execute.actions = actions.data;
1100 execute.actions_len = actions.size;
1101 execute.packet = &packet;
1102 execute.needs_help = false;
1103 execute.probe = true;
1105 error = dpif_execute(backer->dpif, &execute);
1107 dp_packet_uninit(&packet);
1108 ofpbuf_uninit(&actions);
1115 /* Variable-length userdata is not supported. */
1116 VLOG_WARN("%s: datapath does not support variable-length userdata "
1117 "feature (needs Linux 3.10+ or kernel module from OVS "
1118 "1..11+). The NXAST_SAMPLE action will be ignored.",
1119 dpif_name(backer->dpif));
1123 /* Something odd happened. We're not sure whether variable-length
1124 * userdata is supported. Default to "yes". */
1125 VLOG_WARN("%s: variable-length userdata feature probe failed (%s)",
1126 dpif_name(backer->dpif), ovs_strerror(error));
1131 /* Tests the MPLS label stack depth supported by 'backer''s datapath.
1133 * Returns the number of elements in a struct flow's mpls_lse field
1134 * if the datapath supports at least that many entries in an
1136 * Otherwise returns the number of MPLS push actions supported by
1139 check_max_mpls_depth(struct dpif_backer *backer)
1144 for (n = 0; n < FLOW_MAX_MPLS_LABELS; n++) {
1145 struct odputil_keybuf keybuf;
1148 memset(&flow, 0, sizeof flow);
1149 flow.dl_type = htons(ETH_TYPE_MPLS);
1150 flow_set_mpls_bos(&flow, n, 1);
1152 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1153 odp_flow_key_from_flow(&key, &flow, NULL, 0, false);
1154 if (!dpif_probe_feature(backer->dpif, "MPLS", &key, NULL)) {
1159 VLOG_INFO("%s: MPLS label stack length probed as %d",
1160 dpif_name(backer->dpif), n);
1164 /* Tests whether 'backer''s datapath supports masked data in
1165 * OVS_ACTION_ATTR_SET actions. We need to disable some features on older
1166 * datapaths that don't support this feature. */
1168 check_masked_set_action(struct dpif_backer *backer)
1170 struct eth_header *eth;
1171 struct ofpbuf actions;
1172 struct dpif_execute execute;
1173 struct dp_packet packet;
1175 struct ovs_key_ethernet key, mask;
1177 /* Compose a set action that will cause an EINVAL error on older
1178 * datapaths that don't support masked set actions.
1179 * Avoid using a full mask, as it could be translated to a non-masked
1180 * set action instead. */
1181 ofpbuf_init(&actions, 64);
1182 memset(&key, 0x53, sizeof key);
1183 memset(&mask, 0x7f, sizeof mask);
1184 commit_masked_set_action(&actions, OVS_KEY_ATTR_ETHERNET, &key, &mask,
1187 /* Compose a dummy ethernet packet. */
1188 dp_packet_init(&packet, ETH_HEADER_LEN);
1189 eth = dp_packet_put_zeros(&packet, ETH_HEADER_LEN);
1190 eth->eth_type = htons(0x1234);
1192 /* Execute the actions. On older datapaths this fails with EINVAL, on
1193 * newer datapaths it succeeds. */
1194 execute.actions = actions.data;
1195 execute.actions_len = actions.size;
1196 execute.packet = &packet;
1197 execute.needs_help = false;
1198 execute.probe = true;
1200 error = dpif_execute(backer->dpif, &execute);
1202 dp_packet_uninit(&packet);
1203 ofpbuf_uninit(&actions);
1206 /* Masked set action is not supported. */
1207 VLOG_INFO("%s: datapath does not support masked set action feature.",
1208 dpif_name(backer->dpif));
1214 check_support(struct dpif_backer *backer)
1216 /* This feature needs to be tested after udpif threads are set. */
1217 backer->support.variable_length_userdata = false;
1219 backer->support.recirc = check_recirc(backer);
1220 backer->support.max_mpls_depth = check_max_mpls_depth(backer);
1221 backer->support.masked_set_action = check_masked_set_action(backer);
1222 backer->support.ufid = check_ufid(backer);
1223 backer->support.tnl_push_pop = dpif_supports_tnl_push_pop(backer->dpif);
1227 construct(struct ofproto *ofproto_)
1229 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1230 struct shash_node *node, *next;
1233 /* Tunnel module can get used right after the udpif threads are running. */
1234 ofproto_tunnel_init();
1236 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1241 atomic_init(&ofproto->tables_version, CLS_MIN_VERSION);
1242 ofproto->netflow = NULL;
1243 ofproto->sflow = NULL;
1244 ofproto->ipfix = NULL;
1245 ofproto->stp = NULL;
1246 ofproto->rstp = NULL;
1247 ofproto->dump_seq = 0;
1248 hmap_init(&ofproto->bundles);
1249 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1251 ofproto->mbridge = mbridge_create();
1252 ofproto->has_bonded_bundles = false;
1253 ofproto->lacp_enabled = false;
1254 ovs_mutex_init_adaptive(&ofproto->stats_mutex);
1255 ovs_mutex_init(&ofproto->vsp_mutex);
1257 guarded_list_init(&ofproto->pins);
1259 ofproto_unixctl_init();
1261 hmap_init(&ofproto->vlandev_map);
1262 hmap_init(&ofproto->realdev_vid_map);
1264 sset_init(&ofproto->ports);
1265 sset_init(&ofproto->ghost_ports);
1266 sset_init(&ofproto->port_poll_set);
1267 ofproto->port_poll_errno = 0;
1268 ofproto->change_seq = 0;
1269 ofproto->pins_seq = seq_create();
1270 ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1273 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1274 struct iface_hint *iface_hint = node->data;
1276 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1277 /* Check if the datapath already has this port. */
1278 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1279 sset_add(&ofproto->ports, node->name);
1282 free(iface_hint->br_name);
1283 free(iface_hint->br_type);
1285 shash_delete(&init_ofp_ports, node);
1289 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1290 hash_string(ofproto->up.name, 0));
1291 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1293 ofproto_init_tables(ofproto_, N_TABLES);
1294 error = add_internal_flows(ofproto);
1296 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1302 add_internal_miss_flow(struct ofproto_dpif *ofproto, int id,
1303 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1309 match_init_catchall(&match);
1310 match_set_reg(&match, 0, id);
1312 error = ofproto_dpif_add_internal_flow(ofproto, &match, 0, 0, ofpacts,
1314 *rulep = error ? NULL : rule_dpif_cast(rule);
1320 add_internal_flows(struct ofproto_dpif *ofproto)
1322 struct ofpact_controller *controller;
1323 uint64_t ofpacts_stub[128 / 8];
1324 struct ofpbuf ofpacts;
1325 struct rule *unused_rulep OVS_UNUSED;
1330 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1333 controller = ofpact_put_CONTROLLER(&ofpacts);
1334 controller->max_len = UINT16_MAX;
1335 controller->controller_id = 0;
1336 controller->reason = OFPR_NO_MATCH;
1337 ofpact_pad(&ofpacts);
1339 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1340 &ofproto->miss_rule);
1345 ofpbuf_clear(&ofpacts);
1346 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1347 &ofproto->no_packet_in_rule);
1352 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1353 &ofproto->drop_frags_rule);
1358 /* Drop any run away non-recirc rule lookups. Recirc_id has to be
1359 * zero when reaching this rule.
1361 * (priority=2), recirc_id=0, actions=drop
1363 ofpbuf_clear(&ofpacts);
1364 match_init_catchall(&match);
1365 match_set_recirc_id(&match, 0);
1366 error = ofproto_dpif_add_internal_flow(ofproto, &match, 2, 0, &ofpacts,
1372 destruct(struct ofproto *ofproto_)
1374 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1375 struct ofproto_packet_in *pin;
1376 struct rule_dpif *rule;
1377 struct oftable *table;
1378 struct ovs_list pins;
1380 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1382 xlate_remove_ofproto(ofproto);
1385 /* Ensure that the upcall processing threads have no remaining references
1386 * to the ofproto or anything in it. */
1387 udpif_synchronize(ofproto->backer->udpif);
1389 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1391 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1392 CLS_FOR_EACH (rule, up.cr, &table->cls) {
1393 ofproto_rule_delete(&ofproto->up, &rule->up);
1396 ofproto_group_delete_all(&ofproto->up);
1398 guarded_list_pop_all(&ofproto->pins, &pins);
1399 LIST_FOR_EACH_POP (pin, list_node, &pins) {
1400 free(CONST_CAST(void *, pin->up.packet));
1403 guarded_list_destroy(&ofproto->pins);
1405 recirc_free_ofproto(ofproto, ofproto->up.name);
1407 mbridge_unref(ofproto->mbridge);
1409 netflow_unref(ofproto->netflow);
1410 dpif_sflow_unref(ofproto->sflow);
1411 dpif_ipfix_unref(ofproto->ipfix);
1412 hmap_destroy(&ofproto->bundles);
1413 mac_learning_unref(ofproto->ml);
1414 mcast_snooping_unref(ofproto->ms);
1416 hmap_destroy(&ofproto->vlandev_map);
1417 hmap_destroy(&ofproto->realdev_vid_map);
1419 sset_destroy(&ofproto->ports);
1420 sset_destroy(&ofproto->ghost_ports);
1421 sset_destroy(&ofproto->port_poll_set);
1423 ovs_mutex_destroy(&ofproto->stats_mutex);
1424 ovs_mutex_destroy(&ofproto->vsp_mutex);
1426 seq_destroy(ofproto->pins_seq);
1428 close_dpif_backer(ofproto->backer);
1432 run(struct ofproto *ofproto_)
1434 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1435 uint64_t new_seq, new_dump_seq;
1437 if (mbridge_need_revalidate(ofproto->mbridge)) {
1438 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1439 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1440 mac_learning_flush(ofproto->ml);
1441 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1442 mcast_snooping_mdb_flush(ofproto->ms);
1445 /* Always updates the ofproto->pins_seqno to avoid frequent wakeup during
1446 * flow restore. Even though nothing is processed during flow restore,
1447 * all queued 'pins' will be handled immediately when flow restore
1449 ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1451 /* Do not perform any periodic activity required by 'ofproto' while
1452 * waiting for flow restore to complete. */
1453 if (!ofproto_get_flow_restore_wait()) {
1454 struct ofproto_packet_in *pin;
1455 struct ovs_list pins;
1457 guarded_list_pop_all(&ofproto->pins, &pins);
1458 LIST_FOR_EACH_POP (pin, list_node, &pins) {
1459 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1460 free(CONST_CAST(void *, pin->up.packet));
1465 if (ofproto->netflow) {
1466 netflow_run(ofproto->netflow);
1468 if (ofproto->sflow) {
1469 dpif_sflow_run(ofproto->sflow);
1471 if (ofproto->ipfix) {
1472 dpif_ipfix_run(ofproto->ipfix);
1475 new_seq = seq_read(connectivity_seq_get());
1476 if (ofproto->change_seq != new_seq) {
1477 struct ofport_dpif *ofport;
1479 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1483 ofproto->change_seq = new_seq;
1485 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1486 struct ofbundle *bundle;
1488 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1495 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1496 if (mac_learning_run(ofproto->ml)) {
1497 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1499 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1501 if (mcast_snooping_run(ofproto->ms)) {
1502 ofproto->backer->need_revalidate = REV_MCAST_SNOOPING;
1505 new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
1506 if (ofproto->dump_seq != new_dump_seq) {
1507 struct rule *rule, *next_rule;
1509 /* We know stats are relatively fresh, so now is a good time to do some
1511 ofproto->dump_seq = new_dump_seq;
1513 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1515 ovs_mutex_lock(&ofproto_mutex);
1516 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
1517 &ofproto->up.expirable) {
1518 rule_expire(rule_dpif_cast(rule));
1520 ovs_mutex_unlock(&ofproto_mutex);
1522 /* All outstanding data in existing flows has been accounted, so it's a
1523 * good time to do bond rebalancing. */
1524 if (ofproto->has_bonded_bundles) {
1525 struct ofbundle *bundle;
1527 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1529 bond_rebalance(bundle->bond);
1538 wait(struct ofproto *ofproto_)
1540 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1542 if (ofproto_get_flow_restore_wait()) {
1546 if (ofproto->sflow) {
1547 dpif_sflow_wait(ofproto->sflow);
1549 if (ofproto->ipfix) {
1550 dpif_ipfix_wait(ofproto->ipfix);
1552 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1553 struct ofbundle *bundle;
1555 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1556 bundle_wait(bundle);
1559 if (ofproto->netflow) {
1560 netflow_wait(ofproto->netflow);
1562 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1563 mac_learning_wait(ofproto->ml);
1564 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1565 mcast_snooping_wait(ofproto->ms);
1567 if (ofproto->backer->need_revalidate) {
1568 /* Shouldn't happen, but if it does just go around again. */
1569 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1570 poll_immediate_wake();
1573 seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
1574 seq_wait(ofproto->pins_seq, ofproto->pins_seqno);
1578 type_get_memory_usage(const char *type, struct simap *usage)
1580 struct dpif_backer *backer;
1582 backer = shash_find_data(&all_dpif_backers, type);
1584 udpif_get_memory_usage(backer->udpif, usage);
1589 flush(struct ofproto *ofproto_)
1591 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1592 struct dpif_backer *backer = ofproto->backer;
1595 udpif_flush(backer->udpif);
1600 query_tables(struct ofproto *ofproto,
1601 struct ofputil_table_features *features,
1602 struct ofputil_table_stats *stats)
1604 strcpy(features->name, "classifier");
1609 for (i = 0; i < ofproto->n_tables; i++) {
1610 unsigned long missed, matched;
1612 atomic_read_relaxed(&ofproto->tables[i].n_matched, &matched);
1613 atomic_read_relaxed(&ofproto->tables[i].n_missed, &missed);
1615 stats[i].matched_count = matched;
1616 stats[i].lookup_count = matched + missed;
1622 set_tables_version(struct ofproto *ofproto_, cls_version_t version)
1624 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1626 atomic_store_relaxed(&ofproto->tables_version, version);
1630 static struct ofport *
1633 struct ofport_dpif *port = xzalloc(sizeof *port);
1638 port_dealloc(struct ofport *port_)
1640 struct ofport_dpif *port = ofport_dpif_cast(port_);
1645 port_construct(struct ofport *port_)
1647 struct ofport_dpif *port = ofport_dpif_cast(port_);
1648 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1649 const struct netdev *netdev = port->up.netdev;
1650 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1651 const char *dp_port_name;
1652 struct dpif_port dpif_port;
1655 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1656 port->bundle = NULL;
1660 port->may_enable = false;
1661 port->stp_port = NULL;
1662 port->stp_state = STP_DISABLED;
1663 port->rstp_port = NULL;
1664 port->rstp_state = RSTP_DISABLED;
1665 port->is_tunnel = false;
1669 port->realdev_ofp_port = 0;
1670 port->vlandev_vid = 0;
1671 port->carrier_seq = netdev_get_carrier_resets(netdev);
1672 port->is_layer3 = netdev_vport_is_layer3(netdev);
1674 if (netdev_vport_is_patch(netdev)) {
1675 /* By bailing out here, we don't submit the port to the sFlow module
1676 * to be considered for counter polling export. This is correct
1677 * because the patch port represents an interface that sFlow considers
1678 * to be "internal" to the switch as a whole, and therefore not a
1679 * candidate for counter polling. */
1680 port->odp_port = ODPP_NONE;
1681 ofport_update_peer(port);
1685 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
1686 error = dpif_port_query_by_name(ofproto->backer->dpif, dp_port_name,
1692 port->odp_port = dpif_port.port_no;
1694 if (netdev_get_tunnel_config(netdev)) {
1695 atomic_count_inc(&ofproto->backer->tnl_count);
1696 error = tnl_port_add(port, port->up.netdev, port->odp_port,
1697 ovs_native_tunneling_is_on(ofproto), dp_port_name);
1699 atomic_count_dec(&ofproto->backer->tnl_count);
1700 dpif_port_destroy(&dpif_port);
1704 port->is_tunnel = true;
1705 if (ofproto->ipfix) {
1706 dpif_ipfix_add_tunnel_port(ofproto->ipfix, port_, port->odp_port);
1709 /* Sanity-check that a mapping doesn't already exist. This
1710 * shouldn't happen for non-tunnel ports. */
1711 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1712 VLOG_ERR("port %s already has an OpenFlow port number",
1714 dpif_port_destroy(&dpif_port);
1718 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1719 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1720 hash_odp_port(port->odp_port));
1721 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1723 dpif_port_destroy(&dpif_port);
1725 if (ofproto->sflow) {
1726 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1733 port_destruct(struct ofport *port_)
1735 struct ofport_dpif *port = ofport_dpif_cast(port_);
1736 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1737 const char *devname = netdev_get_name(port->up.netdev);
1738 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1739 const char *dp_port_name;
1741 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1743 xlate_ofport_remove(port);
1746 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1748 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1749 /* The underlying device is still there, so delete it. This
1750 * happens when the ofproto is being destroyed, since the caller
1751 * assumes that removal of attached ports will happen as part of
1753 if (!port->is_tunnel) {
1754 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1759 port->peer->peer = NULL;
1763 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1764 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1765 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1766 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1769 if (port->is_tunnel) {
1770 atomic_count_dec(&ofproto->backer->tnl_count);
1773 if (port->is_tunnel && ofproto->ipfix) {
1774 dpif_ipfix_del_tunnel_port(ofproto->ipfix, port->odp_port);
1778 sset_find_and_delete(&ofproto->ports, devname);
1779 sset_find_and_delete(&ofproto->ghost_ports, devname);
1780 bundle_remove(port_);
1781 set_cfm(port_, NULL);
1782 set_bfd(port_, NULL);
1783 set_lldp(port_, NULL);
1784 if (port->stp_port) {
1785 stp_port_disable(port->stp_port);
1787 set_rstp_port(port_, NULL);
1788 if (ofproto->sflow) {
1789 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1796 port_modified(struct ofport *port_)
1798 struct ofport_dpif *port = ofport_dpif_cast(port_);
1799 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1800 const char *dp_port_name;
1801 struct netdev *netdev = port->up.netdev;
1803 if (port->bundle && port->bundle->bond) {
1804 bond_slave_set_netdev(port->bundle->bond, port, netdev);
1808 cfm_set_netdev(port->cfm, netdev);
1812 bfd_set_netdev(port->bfd, netdev);
1815 ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1816 port->lldp, port->up.pp.hw_addr);
1818 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
1820 if (port->is_tunnel) {
1821 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1823 if (tnl_port_reconfigure(port, netdev, port->odp_port,
1824 ovs_native_tunneling_is_on(ofproto),
1826 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1830 ofport_update_peer(port);
1834 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1836 struct ofport_dpif *port = ofport_dpif_cast(port_);
1837 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1838 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1840 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1841 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1842 OFPUTIL_PC_NO_PACKET_IN)) {
1843 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1845 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1846 bundle_update(port->bundle);
1852 set_sflow(struct ofproto *ofproto_,
1853 const struct ofproto_sflow_options *sflow_options)
1855 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1856 struct dpif_sflow *ds = ofproto->sflow;
1858 if (sflow_options) {
1859 uint32_t old_probability = ds ? dpif_sflow_get_probability(ds) : 0;
1861 struct ofport_dpif *ofport;
1863 ds = ofproto->sflow = dpif_sflow_create();
1864 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1865 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1868 dpif_sflow_set_options(ds, sflow_options);
1869 if (dpif_sflow_get_probability(ds) != old_probability) {
1870 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1874 dpif_sflow_unref(ds);
1875 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1876 ofproto->sflow = NULL;
1884 struct ofproto *ofproto_,
1885 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1886 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1887 size_t n_flow_exporters_options)
1889 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1890 struct dpif_ipfix *di = ofproto->ipfix;
1891 bool has_options = bridge_exporter_options || flow_exporters_options;
1892 bool new_di = false;
1894 if (has_options && !di) {
1895 di = ofproto->ipfix = dpif_ipfix_create();
1900 /* Call set_options in any case to cleanly flush the flow
1901 * caches in the last exporters that are to be destroyed. */
1902 dpif_ipfix_set_options(
1903 di, bridge_exporter_options, flow_exporters_options,
1904 n_flow_exporters_options);
1906 /* Add tunnel ports only when a new ipfix created */
1907 if (new_di == true) {
1908 struct ofport_dpif *ofport;
1909 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1910 if (ofport->is_tunnel == true) {
1911 dpif_ipfix_add_tunnel_port(di, &ofport->up, ofport->odp_port);
1917 dpif_ipfix_unref(di);
1918 ofproto->ipfix = NULL;
1926 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1928 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1929 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1930 struct cfm *old = ofport->cfm;
1935 ofport->cfm = cfm_create(ofport->up.netdev);
1938 if (cfm_configure(ofport->cfm, s)) {
1945 cfm_unref(ofport->cfm);
1948 if (ofport->cfm != old) {
1949 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1951 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1952 ofport->lldp, ofport->up.pp.hw_addr);
1957 cfm_status_changed(struct ofport *ofport_)
1959 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1961 return ofport->cfm ? cfm_check_status_change(ofport->cfm) : true;
1965 get_cfm_status(const struct ofport *ofport_,
1966 struct cfm_status *status)
1968 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1972 cfm_get_status(ofport->cfm, status);
1981 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1983 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1984 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1988 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1989 cfg, ofport->up.netdev);
1990 if (ofport->bfd != old) {
1991 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1993 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1994 ofport->lldp, ofport->up.pp.hw_addr);
1999 bfd_status_changed(struct ofport *ofport_)
2001 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2003 return ofport->bfd ? bfd_check_status_change(ofport->bfd) : true;
2007 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2009 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2013 bfd_get_status(ofport->bfd, smap);
2022 set_lldp(struct ofport *ofport_,
2023 const struct smap *cfg)
2025 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2029 if (!ofport->lldp) {
2030 struct ofproto_dpif *ofproto;
2032 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2033 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2034 ofport->lldp = lldp_create(ofport->up.netdev, ofport_->mtu, cfg);
2037 if (!lldp_configure(ofport->lldp, cfg)) {
2042 lldp_unref(ofport->lldp);
2043 ofport->lldp = NULL;
2046 ofproto_dpif_monitor_port_update(ofport,
2050 ofport->up.pp.hw_addr);
2055 get_lldp_status(const struct ofport *ofport_,
2056 struct lldp_status *status OVS_UNUSED)
2058 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2060 return ofport->lldp ? true : false;
2064 set_aa(struct ofproto *ofproto OVS_UNUSED,
2065 const struct aa_settings *s)
2067 return aa_configure(s);
2071 aa_mapping_set(struct ofproto *ofproto_ OVS_UNUSED, void *aux,
2072 const struct aa_mapping_settings *s)
2074 return aa_mapping_register(aux, s);
2078 aa_mapping_unset(struct ofproto *ofproto OVS_UNUSED, void *aux)
2080 return aa_mapping_unregister(aux);
2084 aa_vlan_get_queued(struct ofproto *ofproto OVS_UNUSED, struct ovs_list *list)
2086 return aa_get_vlan_queued(list);
2090 aa_vlan_get_queue_size(struct ofproto *ofproto OVS_UNUSED)
2092 return aa_get_vlan_queue_size();
2096 /* Spanning Tree. */
2098 /* Called while rstp_mutex is held. */
2100 rstp_send_bpdu_cb(struct dp_packet *pkt, void *ofport_, void *ofproto_)
2102 struct ofproto_dpif *ofproto = ofproto_;
2103 struct ofport_dpif *ofport = ofport_;
2104 struct eth_header *eth = dp_packet_l2(pkt);
2106 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2107 if (eth_addr_is_zero(eth->eth_src)) {
2108 VLOG_WARN_RL(&rl, "%s port %d: cannot send RSTP BPDU on a port which "
2109 "does not have a configured source MAC address.",
2110 ofproto->up.name, ofp_to_u16(ofport->up.ofp_port));
2112 ofproto_dpif_send_packet(ofport, pkt);
2114 dp_packet_delete(pkt);
2118 send_bpdu_cb(struct dp_packet *pkt, int port_num, void *ofproto_)
2120 struct ofproto_dpif *ofproto = ofproto_;
2121 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2122 struct ofport_dpif *ofport;
2124 ofport = stp_port_get_aux(sp);
2126 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2127 ofproto->up.name, port_num);
2129 struct eth_header *eth = dp_packet_l2(pkt);
2131 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2132 if (eth_addr_is_zero(eth->eth_src)) {
2133 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2134 "with unknown MAC", ofproto->up.name, port_num);
2136 ofproto_dpif_send_packet(ofport, pkt);
2139 dp_packet_delete(pkt);
2142 /* Configure RSTP on 'ofproto_' using the settings defined in 's'. */
2144 set_rstp(struct ofproto *ofproto_, const struct ofproto_rstp_settings *s)
2146 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2148 /* Only revalidate flows if the configuration changed. */
2149 if (!s != !ofproto->rstp) {
2150 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2154 if (!ofproto->rstp) {
2155 ofproto->rstp = rstp_create(ofproto_->name, s->address,
2156 rstp_send_bpdu_cb, ofproto);
2157 ofproto->rstp_last_tick = time_msec();
2159 rstp_set_bridge_address(ofproto->rstp, s->address);
2160 rstp_set_bridge_priority(ofproto->rstp, s->priority);
2161 rstp_set_bridge_ageing_time(ofproto->rstp, s->ageing_time);
2162 rstp_set_bridge_force_protocol_version(ofproto->rstp,
2163 s->force_protocol_version);
2164 rstp_set_bridge_max_age(ofproto->rstp, s->bridge_max_age);
2165 rstp_set_bridge_forward_delay(ofproto->rstp, s->bridge_forward_delay);
2166 rstp_set_bridge_transmit_hold_count(ofproto->rstp,
2167 s->transmit_hold_count);
2169 struct ofport *ofport;
2170 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2171 set_rstp_port(ofport, NULL);
2173 rstp_unref(ofproto->rstp);
2174 ofproto->rstp = NULL;
2179 get_rstp_status(struct ofproto *ofproto_, struct ofproto_rstp_status *s)
2181 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2183 if (ofproto->rstp) {
2185 s->root_id = rstp_get_root_id(ofproto->rstp);
2186 s->bridge_id = rstp_get_bridge_id(ofproto->rstp);
2187 s->designated_id = rstp_get_designated_id(ofproto->rstp);
2188 s->root_path_cost = rstp_get_root_path_cost(ofproto->rstp);
2189 s->designated_port_id = rstp_get_designated_port_id(ofproto->rstp);
2190 s->bridge_port_id = rstp_get_bridge_port_id(ofproto->rstp);
2197 update_rstp_port_state(struct ofport_dpif *ofport)
2199 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2200 enum rstp_state state;
2202 /* Figure out new state. */
2203 state = ofport->rstp_port ? rstp_port_get_state(ofport->rstp_port)
2207 if (ofport->rstp_state != state) {
2208 enum ofputil_port_state of_state;
2211 VLOG_DBG("port %s: RSTP state changed from %s to %s",
2212 netdev_get_name(ofport->up.netdev),
2213 rstp_state_name(ofport->rstp_state),
2214 rstp_state_name(state));
2216 if (rstp_learn_in_state(ofport->rstp_state)
2217 != rstp_learn_in_state(state)) {
2218 /* XXX: Learning action flows should also be flushed. */
2219 if (ofport->bundle) {
2220 if (!rstp_shift_root_learned_address(ofproto->rstp)
2221 || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2222 bundle_flush_macs(ofport->bundle, false);
2226 fwd_change = rstp_forward_in_state(ofport->rstp_state)
2227 != rstp_forward_in_state(state);
2229 ofproto->backer->need_revalidate = REV_RSTP;
2230 ofport->rstp_state = state;
2232 if (fwd_change && ofport->bundle) {
2233 bundle_update(ofport->bundle);
2236 /* Update the RSTP state bits in the OpenFlow port description. */
2237 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2238 of_state |= (state == RSTP_LEARNING ? OFPUTIL_PS_STP_LEARN
2239 : state == RSTP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2240 : state == RSTP_DISCARDING ? OFPUTIL_PS_STP_LISTEN
2242 ofproto_port_set_state(&ofport->up, of_state);
2247 rstp_run(struct ofproto_dpif *ofproto)
2249 if (ofproto->rstp) {
2250 long long int now = time_msec();
2251 long long int elapsed = now - ofproto->rstp_last_tick;
2252 struct rstp_port *rp;
2253 struct ofport_dpif *ofport;
2255 /* Every second, decrease the values of the timers. */
2256 if (elapsed >= 1000) {
2257 rstp_tick_timers(ofproto->rstp);
2258 ofproto->rstp_last_tick = now;
2261 while ((ofport = rstp_get_next_changed_port_aux(ofproto->rstp, &rp))) {
2262 update_rstp_port_state(ofport);
2266 /* FIXME: This check should be done on-event (i.e., when setting
2267 * p->fdb_flush) and not periodically.
2269 while ((ofport = rstp_check_and_reset_fdb_flush(ofproto->rstp, &rp))) {
2270 if (!rstp_shift_root_learned_address(ofproto->rstp)
2271 || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2272 bundle_flush_macs(ofport->bundle, false);
2276 if (rstp_shift_root_learned_address(ofproto->rstp)) {
2277 bundle_move(((struct ofport_dpif *)rstp_get_old_root_aux(ofproto->rstp))->bundle,
2278 ((struct ofport_dpif *)rstp_get_new_root_aux(ofproto->rstp))->bundle);
2279 rstp_reset_root_changed(ofproto->rstp);
2284 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2286 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2288 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2290 /* Only revalidate flows if the configuration changed. */
2291 if (!s != !ofproto->stp) {
2292 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2296 if (!ofproto->stp) {
2297 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2298 send_bpdu_cb, ofproto);
2299 ofproto->stp_last_tick = time_msec();
2302 stp_set_bridge_id(ofproto->stp, s->system_id);
2303 stp_set_bridge_priority(ofproto->stp, s->priority);
2304 stp_set_hello_time(ofproto->stp, s->hello_time);
2305 stp_set_max_age(ofproto->stp, s->max_age);
2306 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2308 struct ofport *ofport;
2310 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2311 set_stp_port(ofport, NULL);
2314 stp_unref(ofproto->stp);
2315 ofproto->stp = NULL;
2322 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2324 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2328 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2329 s->designated_root = stp_get_designated_root(ofproto->stp);
2330 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2339 update_stp_port_state(struct ofport_dpif *ofport)
2341 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2342 enum stp_state state;
2344 /* Figure out new state. */
2345 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2349 if (ofport->stp_state != state) {
2350 enum ofputil_port_state of_state;
2353 VLOG_DBG("port %s: STP state changed from %s to %s",
2354 netdev_get_name(ofport->up.netdev),
2355 stp_state_name(ofport->stp_state),
2356 stp_state_name(state));
2357 if (stp_learn_in_state(ofport->stp_state)
2358 != stp_learn_in_state(state)) {
2359 /* xxx Learning action flows should also be flushed. */
2360 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2361 mac_learning_flush(ofproto->ml);
2362 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2363 mcast_snooping_mdb_flush(ofproto->ms);
2365 fwd_change = stp_forward_in_state(ofport->stp_state)
2366 != stp_forward_in_state(state);
2368 ofproto->backer->need_revalidate = REV_STP;
2369 ofport->stp_state = state;
2370 ofport->stp_state_entered = time_msec();
2372 if (fwd_change && ofport->bundle) {
2373 bundle_update(ofport->bundle);
2376 /* Update the STP state bits in the OpenFlow port description. */
2377 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2378 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2379 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2380 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2381 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2383 ofproto_port_set_state(&ofport->up, of_state);
2387 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2388 * caller is responsible for assigning STP port numbers and ensuring
2389 * there are no duplicates. */
2391 set_stp_port(struct ofport *ofport_,
2392 const struct ofproto_port_stp_settings *s)
2394 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2395 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2396 struct stp_port *sp = ofport->stp_port;
2398 if (!s || !s->enable) {
2400 ofport->stp_port = NULL;
2401 stp_port_disable(sp);
2402 update_stp_port_state(ofport);
2405 } else if (sp && stp_port_no(sp) != s->port_num
2406 && ofport == stp_port_get_aux(sp)) {
2407 /* The port-id changed, so disable the old one if it's not
2408 * already in use by another port. */
2409 stp_port_disable(sp);
2412 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2414 /* Set name before enabling the port so that debugging messages can print
2416 stp_port_set_name(sp, netdev_get_name(ofport->up.netdev));
2417 stp_port_enable(sp);
2419 stp_port_set_aux(sp, ofport);
2420 stp_port_set_priority(sp, s->priority);
2421 stp_port_set_path_cost(sp, s->path_cost);
2423 update_stp_port_state(ofport);
2429 get_stp_port_status(struct ofport *ofport_,
2430 struct ofproto_port_stp_status *s)
2432 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2433 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2434 struct stp_port *sp = ofport->stp_port;
2436 if (!ofproto->stp || !sp) {
2442 s->port_id = stp_port_get_id(sp);
2443 s->state = stp_port_get_state(sp);
2444 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2445 s->role = stp_port_get_role(sp);
2451 get_stp_port_stats(struct ofport *ofport_,
2452 struct ofproto_port_stp_stats *s)
2454 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2455 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2456 struct stp_port *sp = ofport->stp_port;
2458 if (!ofproto->stp || !sp) {
2464 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2470 stp_run(struct ofproto_dpif *ofproto)
2473 long long int now = time_msec();
2474 long long int elapsed = now - ofproto->stp_last_tick;
2475 struct stp_port *sp;
2478 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2479 ofproto->stp_last_tick = now;
2481 while (stp_get_changed_port(ofproto->stp, &sp)) {
2482 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2485 update_stp_port_state(ofport);
2489 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2490 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2491 mac_learning_flush(ofproto->ml);
2492 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2493 mcast_snooping_mdb_flush(ofproto->ms);
2499 stp_wait(struct ofproto_dpif *ofproto)
2502 poll_timer_wait(1000);
2506 /* Configures RSTP on 'ofport_' using the settings defined in 's'. The
2507 * caller is responsible for assigning RSTP port numbers and ensuring
2508 * there are no duplicates. */
2510 set_rstp_port(struct ofport *ofport_,
2511 const struct ofproto_port_rstp_settings *s)
2513 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2514 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2515 struct rstp_port *rp = ofport->rstp_port;
2517 if (!s || !s->enable) {
2519 rstp_port_unref(rp);
2520 ofport->rstp_port = NULL;
2521 update_rstp_port_state(ofport);
2526 /* Check if need to add a new port. */
2528 rp = ofport->rstp_port = rstp_add_port(ofproto->rstp);
2531 rstp_port_set(rp, s->port_num, s->priority, s->path_cost,
2532 s->admin_edge_port, s->auto_edge,
2533 s->admin_p2p_mac_state, s->admin_port_state, s->mcheck,
2535 update_rstp_port_state(ofport);
2536 /* Synchronize operational status. */
2537 rstp_port_set_mac_operational(rp, ofport->may_enable);
2541 get_rstp_port_status(struct ofport *ofport_,
2542 struct ofproto_port_rstp_status *s)
2544 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2545 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2546 struct rstp_port *rp = ofport->rstp_port;
2548 if (!ofproto->rstp || !rp) {
2554 rstp_port_get_status(rp, &s->port_id, &s->state, &s->role,
2555 &s->designated_bridge_id, &s->designated_port_id,
2556 &s->designated_path_cost, &s->tx_count,
2557 &s->rx_count, &s->error_count, &s->uptime);
2562 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2565 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2566 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2568 if (ofport->n_qdscp != n_qdscp
2569 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2570 n_qdscp * sizeof *qdscp))) {
2571 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2572 free(ofport->qdscp);
2573 ofport->qdscp = n_qdscp
2574 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2576 ofport->n_qdscp = n_qdscp;
2584 /* Expires all MAC learning entries associated with 'bundle' and forces its
2585 * ofproto to revalidate every flow.
2587 * Normally MAC learning entries are removed only from the ofproto associated
2588 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2589 * are removed from every ofproto. When patch ports and SLB bonds are in use
2590 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2591 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2592 * with the host from which it migrated. */
2594 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2596 struct ofproto_dpif *ofproto = bundle->ofproto;
2597 struct mac_learning *ml = ofproto->ml;
2598 struct mac_entry *mac, *next_mac;
2600 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2601 ovs_rwlock_wrlock(&ml->rwlock);
2602 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2603 if (mac_entry_get_port(ml, mac) == bundle) {
2605 struct ofproto_dpif *o;
2607 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2609 struct mac_entry *e;
2611 ovs_rwlock_wrlock(&o->ml->rwlock);
2612 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2614 mac_learning_expire(o->ml, e);
2616 ovs_rwlock_unlock(&o->ml->rwlock);
2621 mac_learning_expire(ml, mac);
2624 ovs_rwlock_unlock(&ml->rwlock);
2628 bundle_move(struct ofbundle *old, struct ofbundle *new)
2630 struct ofproto_dpif *ofproto = old->ofproto;
2631 struct mac_learning *ml = ofproto->ml;
2632 struct mac_entry *mac, *next_mac;
2634 ovs_assert(new->ofproto == old->ofproto);
2636 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2637 ovs_rwlock_wrlock(&ml->rwlock);
2638 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2639 if (mac_entry_get_port(ml, mac) == old) {
2640 mac_entry_set_port(ml, mac, new);
2643 ovs_rwlock_unlock(&ml->rwlock);
2646 static struct ofbundle *
2647 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2649 struct ofbundle *bundle;
2651 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2652 &ofproto->bundles) {
2653 if (bundle->aux == aux) {
2661 bundle_update(struct ofbundle *bundle)
2663 struct ofport_dpif *port;
2665 bundle->floodable = true;
2666 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2667 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2669 || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2670 || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2671 bundle->floodable = false;
2678 bundle_del_port(struct ofport_dpif *port)
2680 struct ofbundle *bundle = port->bundle;
2682 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2684 list_remove(&port->bundle_node);
2685 port->bundle = NULL;
2688 lacp_slave_unregister(bundle->lacp, port);
2691 bond_slave_unregister(bundle->bond, port);
2694 bundle_update(bundle);
2698 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2699 struct lacp_slave_settings *lacp)
2701 struct ofport_dpif *port;
2703 port = ofp_port_to_ofport(bundle->ofproto, ofp_port);
2708 if (port->bundle != bundle) {
2709 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2711 bundle_remove(&port->up);
2714 port->bundle = bundle;
2715 list_push_back(&bundle->ports, &port->bundle_node);
2716 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2718 || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2719 || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2720 bundle->floodable = false;
2724 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2725 lacp_slave_register(bundle->lacp, port, lacp);
2732 bundle_destroy(struct ofbundle *bundle)
2734 struct ofproto_dpif *ofproto;
2735 struct ofport_dpif *port, *next_port;
2741 ofproto = bundle->ofproto;
2742 mbridge_unregister_bundle(ofproto->mbridge, bundle);
2745 xlate_bundle_remove(bundle);
2748 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2749 bundle_del_port(port);
2752 bundle_flush_macs(bundle, true);
2753 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2755 free(bundle->trunks);
2756 lacp_unref(bundle->lacp);
2757 bond_unref(bundle->bond);
2762 bundle_set(struct ofproto *ofproto_, void *aux,
2763 const struct ofproto_bundle_settings *s)
2765 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2766 bool need_flush = false;
2767 struct ofport_dpif *port;
2768 struct ofbundle *bundle;
2769 unsigned long *trunks;
2775 bundle_destroy(bundle_lookup(ofproto, aux));
2779 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2780 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2782 bundle = bundle_lookup(ofproto, aux);
2784 bundle = xmalloc(sizeof *bundle);
2786 bundle->ofproto = ofproto;
2787 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2788 hash_pointer(aux, 0));
2790 bundle->name = NULL;
2792 list_init(&bundle->ports);
2793 bundle->vlan_mode = PORT_VLAN_TRUNK;
2795 bundle->trunks = NULL;
2796 bundle->use_priority_tags = s->use_priority_tags;
2797 bundle->lacp = NULL;
2798 bundle->bond = NULL;
2800 bundle->floodable = true;
2801 mbridge_register_bundle(ofproto->mbridge, bundle);
2804 if (!bundle->name || strcmp(s->name, bundle->name)) {
2806 bundle->name = xstrdup(s->name);
2811 ofproto->lacp_enabled = true;
2812 if (!bundle->lacp) {
2813 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2814 bundle->lacp = lacp_create();
2816 lacp_configure(bundle->lacp, s->lacp);
2818 lacp_unref(bundle->lacp);
2819 bundle->lacp = NULL;
2822 /* Update set of ports. */
2824 for (i = 0; i < s->n_slaves; i++) {
2825 if (!bundle_add_port(bundle, s->slaves[i],
2826 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2830 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2831 struct ofport_dpif *next_port;
2833 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2834 for (i = 0; i < s->n_slaves; i++) {
2835 if (s->slaves[i] == port->up.ofp_port) {
2840 bundle_del_port(port);
2844 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2846 if (list_is_empty(&bundle->ports)) {
2847 bundle_destroy(bundle);
2851 /* Set VLAN tagging mode */
2852 if (s->vlan_mode != bundle->vlan_mode
2853 || s->use_priority_tags != bundle->use_priority_tags) {
2854 bundle->vlan_mode = s->vlan_mode;
2855 bundle->use_priority_tags = s->use_priority_tags;
2860 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2861 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2863 if (vlan != bundle->vlan) {
2864 bundle->vlan = vlan;
2868 /* Get trunked VLANs. */
2869 switch (s->vlan_mode) {
2870 case PORT_VLAN_ACCESS:
2874 case PORT_VLAN_TRUNK:
2875 trunks = CONST_CAST(unsigned long *, s->trunks);
2878 case PORT_VLAN_NATIVE_UNTAGGED:
2879 case PORT_VLAN_NATIVE_TAGGED:
2880 if (vlan != 0 && (!s->trunks
2881 || !bitmap_is_set(s->trunks, vlan)
2882 || bitmap_is_set(s->trunks, 0))) {
2883 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2885 trunks = bitmap_clone(s->trunks, 4096);
2887 trunks = bitmap_allocate1(4096);
2889 bitmap_set1(trunks, vlan);
2890 bitmap_set0(trunks, 0);
2892 trunks = CONST_CAST(unsigned long *, s->trunks);
2899 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2900 free(bundle->trunks);
2901 if (trunks == s->trunks) {
2902 bundle->trunks = vlan_bitmap_clone(trunks);
2904 bundle->trunks = trunks;
2909 if (trunks != s->trunks) {
2914 if (!list_is_short(&bundle->ports)) {
2915 bundle->ofproto->has_bonded_bundles = true;
2917 if (bond_reconfigure(bundle->bond, s->bond)) {
2918 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2921 bundle->bond = bond_create(s->bond, ofproto);
2922 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2925 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2926 bond_slave_register(bundle->bond, port,
2927 port->up.ofp_port, port->up.netdev);
2930 bond_unref(bundle->bond);
2931 bundle->bond = NULL;
2934 /* If we changed something that would affect MAC learning, un-learn
2935 * everything on this port and force flow revalidation. */
2937 bundle_flush_macs(bundle, false);
2944 bundle_remove(struct ofport *port_)
2946 struct ofport_dpif *port = ofport_dpif_cast(port_);
2947 struct ofbundle *bundle = port->bundle;
2950 bundle_del_port(port);
2951 if (list_is_empty(&bundle->ports)) {
2952 bundle_destroy(bundle);
2953 } else if (list_is_short(&bundle->ports)) {
2954 bond_unref(bundle->bond);
2955 bundle->bond = NULL;
2961 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2963 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2964 struct ofport_dpif *port = port_;
2965 uint8_t ea[ETH_ADDR_LEN];
2968 error = netdev_get_etheraddr(port->up.netdev, ea);
2970 struct dp_packet packet;
2973 dp_packet_init(&packet, 0);
2974 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2976 memcpy(packet_pdu, pdu, pdu_size);
2978 ofproto_dpif_send_packet(port, &packet);
2979 dp_packet_uninit(&packet);
2981 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2982 "%s (%s)", port->bundle->name,
2983 netdev_get_name(port->up.netdev), ovs_strerror(error));
2988 bundle_send_learning_packets(struct ofbundle *bundle)
2990 struct ofproto_dpif *ofproto = bundle->ofproto;
2991 int error, n_packets, n_errors;
2992 struct mac_entry *e;
2994 struct ovs_list list_node;
2995 struct ofport_dpif *port;
2996 struct dp_packet *pkt;
2998 struct ovs_list packets;
3000 list_init(&packets);
3001 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
3002 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3003 if (mac_entry_get_port(ofproto->ml, e) != bundle) {
3004 pkt_node = xmalloc(sizeof *pkt_node);
3005 pkt_node->pkt = bond_compose_learning_packet(bundle->bond,
3007 (void **)&pkt_node->port);
3008 list_push_back(&packets, &pkt_node->list_node);
3011 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3013 error = n_packets = n_errors = 0;
3014 LIST_FOR_EACH_POP (pkt_node, list_node, &packets) {
3017 ret = ofproto_dpif_send_packet(pkt_node->port, pkt_node->pkt);
3018 dp_packet_delete(pkt_node->pkt);
3028 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3029 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3030 "packets, last error was: %s",
3031 bundle->name, n_errors, n_packets, ovs_strerror(error));
3033 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3034 bundle->name, n_packets);
3039 bundle_run(struct ofbundle *bundle)
3042 lacp_run(bundle->lacp, send_pdu_cb);
3045 struct ofport_dpif *port;
3047 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
3048 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
3051 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
3052 bundle->ofproto->backer->need_revalidate = REV_BOND;
3055 if (bond_should_send_learning_packets(bundle->bond)) {
3056 bundle_send_learning_packets(bundle);
3062 bundle_wait(struct ofbundle *bundle)
3065 lacp_wait(bundle->lacp);
3068 bond_wait(bundle->bond);
3075 mirror_set__(struct ofproto *ofproto_, void *aux,
3076 const struct ofproto_mirror_settings *s)
3078 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3079 struct ofbundle **srcs, **dsts;
3084 mirror_destroy(ofproto->mbridge, aux);
3088 srcs = xmalloc(s->n_srcs * sizeof *srcs);
3089 dsts = xmalloc(s->n_dsts * sizeof *dsts);
3091 for (i = 0; i < s->n_srcs; i++) {
3092 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
3095 for (i = 0; i < s->n_dsts; i++) {
3096 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
3099 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
3100 s->n_dsts, s->src_vlans,
3101 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
3108 mirror_get_stats__(struct ofproto *ofproto, void *aux,
3109 uint64_t *packets, uint64_t *bytes)
3111 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
3116 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3118 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3119 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3120 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3121 mac_learning_flush(ofproto->ml);
3123 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3128 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3130 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3131 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3132 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
3136 forward_bpdu_changed(struct ofproto *ofproto_)
3138 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3139 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3143 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3146 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3147 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3148 mac_learning_set_idle_time(ofproto->ml, idle_time);
3149 mac_learning_set_max_entries(ofproto->ml, max_entries);
3150 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3153 /* Configures multicast snooping on 'ofport' using the settings
3154 * defined in 's'. */
3156 set_mcast_snooping(struct ofproto *ofproto_,
3157 const struct ofproto_mcast_snooping_settings *s)
3159 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3161 /* Only revalidate flows if the configuration changed. */
3162 if (!s != !ofproto->ms) {
3163 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3168 ofproto->ms = mcast_snooping_create();
3171 ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3172 mcast_snooping_set_idle_time(ofproto->ms, s->idle_time);
3173 mcast_snooping_set_max_entries(ofproto->ms, s->max_entries);
3174 if (mcast_snooping_set_flood_unreg(ofproto->ms, s->flood_unreg)) {
3175 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3177 ovs_rwlock_unlock(&ofproto->ms->rwlock);
3179 mcast_snooping_unref(ofproto->ms);
3186 /* Configures multicast snooping port's flood settings on 'ofproto'. */
3188 set_mcast_snooping_port(struct ofproto *ofproto_, void *aux,
3189 const struct ofproto_mcast_snooping_port_settings *s)
3191 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3192 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3194 if (ofproto->ms && s) {
3195 ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3196 mcast_snooping_set_port_flood(ofproto->ms, bundle, s->flood);
3197 mcast_snooping_set_port_flood_reports(ofproto->ms, bundle,
3199 ovs_rwlock_unlock(&ofproto->ms->rwlock);
3207 struct ofport_dpif *
3208 ofp_port_to_ofport(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
3210 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3211 return ofport ? ofport_dpif_cast(ofport) : NULL;
3215 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3216 struct ofproto_port *ofproto_port,
3217 struct dpif_port *dpif_port)
3219 ofproto_port->name = dpif_port->name;
3220 ofproto_port->type = dpif_port->type;
3221 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3225 ofport_update_peer(struct ofport_dpif *ofport)
3227 const struct ofproto_dpif *ofproto;
3228 struct dpif_backer *backer;
3231 if (!netdev_vport_is_patch(ofport->up.netdev)) {
3235 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
3236 backer->need_revalidate = REV_RECONFIGURE;
3239 ofport->peer->peer = NULL;
3240 ofport->peer = NULL;
3243 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
3248 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3249 struct ofport *peer_ofport;
3250 struct ofport_dpif *peer;
3253 if (ofproto->backer != backer) {
3257 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
3262 peer = ofport_dpif_cast(peer_ofport);
3263 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
3264 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
3266 ofport->peer = peer;
3267 ofport->peer->peer = ofport;
3277 port_run(struct ofport_dpif *ofport)
3279 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3280 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3281 bool enable = netdev_get_carrier(ofport->up.netdev);
3282 bool cfm_enable = false;
3283 bool bfd_enable = false;
3285 ofport->carrier_seq = carrier_seq;
3288 int cfm_opup = cfm_get_opup(ofport->cfm);
3290 cfm_enable = !cfm_get_fault(ofport->cfm);
3292 if (cfm_opup >= 0) {
3293 cfm_enable = cfm_enable && cfm_opup;
3298 bfd_enable = bfd_forwarding(ofport->bfd);
3301 if (ofport->bfd || ofport->cfm) {
3302 enable = enable && (cfm_enable || bfd_enable);
3305 if (ofport->bundle) {
3306 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3307 if (carrier_changed) {
3308 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3312 if (ofport->may_enable != enable) {
3313 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3315 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3317 if (ofport->rstp_port) {
3318 rstp_port_set_mac_operational(ofport->rstp_port, enable);
3322 ofport->may_enable = enable;
3326 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3327 struct ofproto_port *ofproto_port)
3329 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3330 struct dpif_port dpif_port;
3333 if (sset_contains(&ofproto->ghost_ports, devname)) {
3334 const char *type = netdev_get_type_from_name(devname);
3336 /* We may be called before ofproto->up.port_by_name is populated with
3337 * the appropriate ofport. For this reason, we must get the name and
3338 * type from the netdev layer directly. */
3340 const struct ofport *ofport;
3342 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3343 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3344 ofproto_port->name = xstrdup(devname);
3345 ofproto_port->type = xstrdup(type);
3351 if (!sset_contains(&ofproto->ports, devname)) {
3354 error = dpif_port_query_by_name(ofproto->backer->dpif,
3355 devname, &dpif_port);
3357 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3363 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3365 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3366 const char *devname = netdev_get_name(netdev);
3367 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3368 const char *dp_port_name;
3370 if (netdev_vport_is_patch(netdev)) {
3371 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3375 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3376 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3377 odp_port_t port_no = ODPP_NONE;
3380 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3384 if (netdev_get_tunnel_config(netdev)) {
3385 simap_put(&ofproto->backer->tnl_backers,
3386 dp_port_name, odp_to_u32(port_no));
3390 if (netdev_get_tunnel_config(netdev)) {
3391 sset_add(&ofproto->ghost_ports, devname);
3393 sset_add(&ofproto->ports, devname);
3399 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3401 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3402 struct ofport_dpif *ofport = ofp_port_to_ofport(ofproto, ofp_port);
3409 sset_find_and_delete(&ofproto->ghost_ports,
3410 netdev_get_name(ofport->up.netdev));
3411 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3412 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3413 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3415 /* The caller is going to close ofport->up.netdev. If this is a
3416 * bonded port, then the bond is using that netdev, so remove it
3417 * from the bond. The client will need to reconfigure everything
3418 * after deleting ports, so then the slave will get re-added. */
3419 bundle_remove(&ofport->up);
3426 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3428 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3431 error = netdev_get_stats(ofport->up.netdev, stats);
3433 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3434 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3436 ovs_mutex_lock(&ofproto->stats_mutex);
3437 /* ofproto->stats.tx_packets represents packets that we created
3438 * internally and sent to some port (e.g. packets sent with
3439 * ofproto_dpif_send_packet()). Account for them as if they had
3440 * come from OFPP_LOCAL and got forwarded. */
3442 if (stats->rx_packets != UINT64_MAX) {
3443 stats->rx_packets += ofproto->stats.tx_packets;
3446 if (stats->rx_bytes != UINT64_MAX) {
3447 stats->rx_bytes += ofproto->stats.tx_bytes;
3450 /* ofproto->stats.rx_packets represents packets that were received on
3451 * some port and we processed internally and dropped (e.g. STP).
3452 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3454 if (stats->tx_packets != UINT64_MAX) {
3455 stats->tx_packets += ofproto->stats.rx_packets;
3458 if (stats->tx_bytes != UINT64_MAX) {
3459 stats->tx_bytes += ofproto->stats.rx_bytes;
3461 ovs_mutex_unlock(&ofproto->stats_mutex);
3468 port_get_lacp_stats(const struct ofport *ofport_, struct lacp_slave_stats *stats)
3470 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3471 if (ofport->bundle && ofport->bundle->lacp) {
3472 if (lacp_get_slave_stats(ofport->bundle->lacp, ofport, stats)) {
3479 struct port_dump_state {
3484 struct ofproto_port port;
3489 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3491 *statep = xzalloc(sizeof(struct port_dump_state));
3496 port_dump_next(const struct ofproto *ofproto_, void *state_,
3497 struct ofproto_port *port)
3499 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3500 struct port_dump_state *state = state_;
3501 const struct sset *sset;
3502 struct sset_node *node;
3504 if (state->has_port) {
3505 ofproto_port_destroy(&state->port);
3506 state->has_port = false;
3508 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3509 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3512 error = port_query_by_name(ofproto_, node->name, &state->port);
3514 *port = state->port;
3515 state->has_port = true;
3517 } else if (error != ENODEV) {
3522 if (!state->ghost) {
3523 state->ghost = true;
3526 return port_dump_next(ofproto_, state_, port);
3533 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3535 struct port_dump_state *state = state_;
3537 if (state->has_port) {
3538 ofproto_port_destroy(&state->port);
3545 port_poll(const struct ofproto *ofproto_, char **devnamep)
3547 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3549 if (ofproto->port_poll_errno) {
3550 int error = ofproto->port_poll_errno;
3551 ofproto->port_poll_errno = 0;
3555 if (sset_is_empty(&ofproto->port_poll_set)) {
3559 *devnamep = sset_pop(&ofproto->port_poll_set);
3564 port_poll_wait(const struct ofproto *ofproto_)
3566 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3567 dpif_port_poll_wait(ofproto->backer->dpif);
3571 port_is_lacp_current(const struct ofport *ofport_)
3573 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3574 return (ofport->bundle && ofport->bundle->lacp
3575 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3579 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3580 * then delete it entirely. */
3582 rule_expire(struct rule_dpif *rule)
3583 OVS_REQUIRES(ofproto_mutex)
3585 uint16_t hard_timeout, idle_timeout;
3586 long long int now = time_msec();
3589 hard_timeout = rule->up.hard_timeout;
3590 idle_timeout = rule->up.idle_timeout;
3592 /* Has 'rule' expired? */
3594 long long int modified;
3596 ovs_mutex_lock(&rule->up.mutex);
3597 modified = rule->up.modified;
3598 ovs_mutex_unlock(&rule->up.mutex);
3600 if (now > modified + hard_timeout * 1000) {
3601 reason = OFPRR_HARD_TIMEOUT;
3605 if (reason < 0 && idle_timeout) {
3608 ovs_mutex_lock(&rule->stats_mutex);
3609 used = rule->stats.used;
3610 ovs_mutex_unlock(&rule->stats_mutex);
3612 if (now > used + idle_timeout * 1000) {
3613 reason = OFPRR_IDLE_TIMEOUT;
3618 COVERAGE_INC(ofproto_dpif_expired);
3619 ofproto_rule_expire(&rule->up, reason);
3623 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3624 * 'flow' must reflect the data in 'packet'. */
3626 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3627 const struct flow *flow,
3628 struct rule_dpif *rule,
3629 const struct ofpact *ofpacts, size_t ofpacts_len,
3630 struct dp_packet *packet)
3632 struct dpif_flow_stats stats;
3633 struct xlate_out xout;
3634 struct xlate_in xin;
3636 struct dpif_execute execute;
3639 ovs_assert((rule != NULL) != (ofpacts != NULL));
3641 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3644 rule_dpif_credit_stats(rule, &stats);
3647 xlate_in_init(&xin, ofproto, flow, flow->in_port.ofp_port, rule,
3648 stats.tcp_flags, packet);
3649 xin.ofpacts = ofpacts;
3650 xin.ofpacts_len = ofpacts_len;
3651 xin.resubmit_stats = &stats;
3652 xlate_actions(&xin, &xout);
3654 execute.actions = xout.odp_actions->data;
3655 execute.actions_len = xout.odp_actions->size;
3657 pkt_metadata_from_flow(&packet->md, flow);
3658 execute.packet = packet;
3659 execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
3660 execute.probe = false;
3662 /* Fix up in_port. */
3663 in_port = flow->in_port.ofp_port;
3664 if (in_port == OFPP_NONE) {
3665 in_port = OFPP_LOCAL;
3667 execute.packet->md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
3669 error = dpif_execute(ofproto->backer->dpif, &execute);
3671 xlate_out_uninit(&xout);
3677 rule_dpif_credit_stats(struct rule_dpif *rule,
3678 const struct dpif_flow_stats *stats)
3680 ovs_mutex_lock(&rule->stats_mutex);
3681 if (OVS_UNLIKELY(rule->new_rule)) {
3682 rule_dpif_credit_stats(rule->new_rule, stats);
3684 rule->stats.n_packets += stats->n_packets;
3685 rule->stats.n_bytes += stats->n_bytes;
3686 rule->stats.used = MAX(rule->stats.used, stats->used);
3688 ovs_mutex_unlock(&rule->stats_mutex);
3692 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
3693 OVS_REQUIRES(rule->up.mutex)
3695 return rule->up.flow_cookie;
3699 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
3700 uint16_t hard_timeout)
3702 ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
3705 /* Returns 'rule''s actions. The returned actions are RCU-protected, and can
3706 * be read until the calling thread quiesces. */
3707 const struct rule_actions *
3708 rule_dpif_get_actions(const struct rule_dpif *rule)
3710 return rule_get_actions(&rule->up);
3713 /* Sets 'rule''s recirculation id. */
3715 rule_dpif_set_recirc_id(struct rule_dpif *rule, uint32_t id)
3716 OVS_REQUIRES(rule->up.mutex)
3718 ovs_assert(!rule->recirc_id || rule->recirc_id == id);
3719 if (rule->recirc_id == id) {
3720 /* Release the new reference to the same id. */
3723 rule->recirc_id = id;
3727 /* Sets 'rule''s recirculation id. */
3729 rule_set_recirc_id(struct rule *rule_, uint32_t id)
3731 struct rule_dpif *rule = rule_dpif_cast(rule_);
3733 ovs_mutex_lock(&rule->up.mutex);
3734 rule_dpif_set_recirc_id(rule, id);
3735 ovs_mutex_unlock(&rule->up.mutex);
3739 ofproto_dpif_get_tables_version(struct ofproto_dpif *ofproto OVS_UNUSED)
3741 cls_version_t version;
3743 atomic_read_relaxed(&ofproto->tables_version, &version);
3748 /* The returned rule (if any) is valid at least until the next RCU quiescent
3749 * period. If the rule needs to stay around longer, a non-zero 'take_ref'
3750 * must be passed in to cause a reference to be taken on it.
3752 * 'flow' is non-const to allow for temporary modifications during the lookup.
3753 * Any changes are restored before returning. */
3754 static struct rule_dpif *
3755 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, cls_version_t version,
3756 uint8_t table_id, struct flow *flow,
3757 struct flow_wildcards *wc, bool take_ref)
3759 struct classifier *cls = &ofproto->up.tables[table_id].cls;
3760 const struct cls_rule *cls_rule;
3761 struct rule_dpif *rule;
3764 cls_rule = classifier_lookup(cls, version, flow, wc);
3766 rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
3768 /* Try again if the rule was released before we get the reference. */
3769 } while (rule && take_ref && !rule_dpif_try_ref(rule));
3774 /* Look up 'flow' in 'ofproto''s classifier version 'version', starting from
3775 * table '*table_id'. Returns the rule that was found, which may be one of the
3776 * special rules according to packet miss hadling. If 'may_packet_in' is
3777 * false, returning of the miss_rule (which issues packet ins for the
3778 * controller) is avoided. Updates 'wc', if nonnull, to reflect the fields
3779 * that were used during the lookup.
3781 * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3782 * if none is found then the table miss configuration for that table is
3783 * honored, which can result in additional lookups in other OpenFlow tables.
3784 * In this case the function updates '*table_id' to reflect the final OpenFlow
3785 * table that was searched.
3787 * If 'honor_table_miss' is false, then only one table lookup occurs, in
3790 * The rule is returned in '*rule', which is valid at least until the next
3791 * RCU quiescent period. If the '*rule' needs to stay around longer,
3792 * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3793 * on it before this returns.
3795 * 'in_port' allows the lookup to take place as if the in port had the value
3796 * 'in_port'. This is needed for resubmit action support.
3798 * 'flow' is non-const to allow for temporary modifications during the lookup.
3799 * Any changes are restored before returning. */
3801 rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto,
3802 cls_version_t version, struct flow *flow,
3803 struct flow_wildcards *wc, bool take_ref,
3804 const struct dpif_flow_stats *stats,
3805 uint8_t *table_id, ofp_port_t in_port,
3806 bool may_packet_in, bool honor_table_miss)
3808 ovs_be16 old_tp_src = flow->tp_src, old_tp_dst = flow->tp_dst;
3809 ofp_port_t old_in_port = flow->in_port.ofp_port;
3810 enum ofputil_table_miss miss_config;
3811 struct rule_dpif *rule;
3814 /* We always unwildcard nw_frag (for IP), so they
3815 * need not be unwildcarded here. */
3816 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3817 && ofproto->up.frag_handling != OFPC_FRAG_NX_MATCH) {
3818 if (ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3819 /* We must pretend that transport ports are unavailable. */
3820 flow->tp_src = htons(0);
3821 flow->tp_dst = htons(0);
3823 /* Must be OFPC_FRAG_DROP (we don't have OFPC_FRAG_REASM).
3824 * Use the drop_frags_rule (which cannot disappear). */
3825 rule = ofproto->drop_frags_rule;
3827 rule_dpif_ref(rule);
3830 struct oftable *tbl = &ofproto->up.tables[*table_id];
3833 atomic_add_relaxed(&tbl->n_matched, stats->n_packets, &orig);
3839 /* Look up a flow with 'in_port' as the input port. Then restore the
3840 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
3841 * have surprising behavior). */
3842 flow->in_port.ofp_port = in_port;
3844 /* Our current implementation depends on n_tables == N_TABLES, and
3845 * TBL_INTERNAL being the last table. */
3846 BUILD_ASSERT_DECL(N_TABLES == TBL_INTERNAL + 1);
3848 miss_config = OFPUTIL_TABLE_MISS_CONTINUE;
3850 for (next_id = *table_id;
3851 next_id < ofproto->up.n_tables;
3852 next_id++, next_id += (next_id == TBL_INTERNAL))
3854 *table_id = next_id;
3855 rule = rule_dpif_lookup_in_table(ofproto, version, next_id, flow, wc,
3858 struct oftable *tbl = &ofproto->up.tables[next_id];
3861 atomic_add_relaxed(rule ? &tbl->n_matched : &tbl->n_missed,
3862 stats->n_packets, &orig);
3865 goto out; /* Match. */
3867 if (honor_table_miss) {
3868 miss_config = ofproto_table_get_miss_config(&ofproto->up,
3870 if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE) {
3877 rule = ofproto->no_packet_in_rule;
3878 if (may_packet_in) {
3879 if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE
3880 || miss_config == OFPUTIL_TABLE_MISS_CONTROLLER) {
3881 struct ofport_dpif *port;
3883 port = ofp_port_to_ofport(ofproto, old_in_port);
3885 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
3887 } else if (!(port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN)) {
3888 rule = ofproto->miss_rule;
3890 } else if (miss_config == OFPUTIL_TABLE_MISS_DEFAULT &&
3891 connmgr_wants_packet_in_on_miss(ofproto->up.connmgr)) {
3892 rule = ofproto->miss_rule;
3896 rule_dpif_ref(rule);
3899 /* Restore port numbers, as they may have been modified above. */
3900 flow->tp_src = old_tp_src;
3901 flow->tp_dst = old_tp_dst;
3902 /* Restore the old in port. */
3903 flow->in_port.ofp_port = old_in_port;
3909 complete_operation(struct rule_dpif *rule)
3910 OVS_REQUIRES(ofproto_mutex)
3912 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3914 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3917 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
3919 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
3922 static struct rule *
3925 struct rule_dpif *rule = xzalloc(sizeof *rule);
3930 rule_dealloc(struct rule *rule_)
3932 struct rule_dpif *rule = rule_dpif_cast(rule_);
3937 rule_construct(struct rule *rule_)
3938 OVS_NO_THREAD_SAFETY_ANALYSIS
3940 struct rule_dpif *rule = rule_dpif_cast(rule_);
3941 ovs_mutex_init_adaptive(&rule->stats_mutex);
3942 rule->stats.n_packets = 0;
3943 rule->stats.n_bytes = 0;
3944 rule->stats.used = rule->up.modified;
3945 rule->recirc_id = 0;
3946 rule->new_rule = NULL;
3952 rule_insert(struct rule *rule_, struct rule *old_rule_, bool forward_stats)
3953 OVS_REQUIRES(ofproto_mutex)
3955 struct rule_dpif *rule = rule_dpif_cast(rule_);
3957 if (old_rule_ && forward_stats) {
3958 struct rule_dpif *old_rule = rule_dpif_cast(old_rule_);
3960 ovs_assert(!old_rule->new_rule);
3962 /* Take a reference to the new rule, and refer all stats updates from
3963 * the old rule to the new rule. */
3964 rule_dpif_ref(rule);
3966 ovs_mutex_lock(&old_rule->stats_mutex);
3967 ovs_mutex_lock(&rule->stats_mutex);
3968 old_rule->new_rule = rule; /* Forward future stats. */
3969 rule->stats = old_rule->stats; /* Transfer stats to the new rule. */
3970 ovs_mutex_unlock(&rule->stats_mutex);
3971 ovs_mutex_unlock(&old_rule->stats_mutex);
3974 complete_operation(rule);
3978 rule_delete(struct rule *rule_)
3979 OVS_REQUIRES(ofproto_mutex)
3981 struct rule_dpif *rule = rule_dpif_cast(rule_);
3982 complete_operation(rule);
3986 rule_destruct(struct rule *rule_)
3987 OVS_NO_THREAD_SAFETY_ANALYSIS
3989 struct rule_dpif *rule = rule_dpif_cast(rule_);
3991 ovs_mutex_destroy(&rule->stats_mutex);
3992 /* Release reference to the new rule, if any. */
3993 if (rule->new_rule) {
3994 rule_dpif_unref(rule->new_rule);
3996 if (rule->recirc_id) {
3997 recirc_free_id(rule->recirc_id);
4002 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
4003 long long int *used)
4005 struct rule_dpif *rule = rule_dpif_cast(rule_);
4007 ovs_mutex_lock(&rule->stats_mutex);
4008 if (OVS_UNLIKELY(rule->new_rule)) {
4009 rule_get_stats(&rule->new_rule->up, packets, bytes, used);
4011 *packets = rule->stats.n_packets;
4012 *bytes = rule->stats.n_bytes;
4013 *used = rule->stats.used;
4015 ovs_mutex_unlock(&rule->stats_mutex);
4019 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4020 struct dp_packet *packet)
4022 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4024 ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
4028 rule_execute(struct rule *rule, const struct flow *flow,
4029 struct dp_packet *packet)
4031 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4032 dp_packet_delete(packet);
4036 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
4038 return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
4041 static struct ofgroup *
4044 struct group_dpif *group = xzalloc(sizeof *group);
4049 group_dealloc(struct ofgroup *group_)
4051 struct group_dpif *group = group_dpif_cast(group_);
4056 group_construct_stats(struct group_dpif *group)
4057 OVS_REQUIRES(group->stats_mutex)
4059 struct ofputil_bucket *bucket;
4060 const struct ovs_list *buckets;
4062 group->packet_count = 0;
4063 group->byte_count = 0;
4065 group_dpif_get_buckets(group, &buckets);
4066 LIST_FOR_EACH (bucket, list_node, buckets) {
4067 bucket->stats.packet_count = 0;
4068 bucket->stats.byte_count = 0;
4073 group_dpif_credit_stats(struct group_dpif *group,
4074 struct ofputil_bucket *bucket,
4075 const struct dpif_flow_stats *stats)
4077 ovs_mutex_lock(&group->stats_mutex);
4078 group->packet_count += stats->n_packets;
4079 group->byte_count += stats->n_bytes;
4081 bucket->stats.packet_count += stats->n_packets;
4082 bucket->stats.byte_count += stats->n_bytes;
4083 } else { /* Credit to all buckets */
4084 const struct ovs_list *buckets;
4086 group_dpif_get_buckets(group, &buckets);
4087 LIST_FOR_EACH (bucket, list_node, buckets) {
4088 bucket->stats.packet_count += stats->n_packets;
4089 bucket->stats.byte_count += stats->n_bytes;
4092 ovs_mutex_unlock(&group->stats_mutex);
4096 group_construct(struct ofgroup *group_)
4098 struct group_dpif *group = group_dpif_cast(group_);
4099 const struct ofputil_bucket *bucket;
4101 /* Prevent group chaining because our locking structure makes it hard to
4102 * implement deadlock-free. (See xlate_group_resource_check().) */
4103 LIST_FOR_EACH (bucket, list_node, &group->up.buckets) {
4104 const struct ofpact *a;
4106 OFPACT_FOR_EACH (a, bucket->ofpacts, bucket->ofpacts_len) {
4107 if (a->type == OFPACT_GROUP) {
4108 return OFPERR_OFPGMFC_CHAINING_UNSUPPORTED;
4113 ovs_mutex_init_adaptive(&group->stats_mutex);
4114 ovs_mutex_lock(&group->stats_mutex);
4115 group_construct_stats(group);
4116 ovs_mutex_unlock(&group->stats_mutex);
4121 group_destruct(struct ofgroup *group_)
4123 struct group_dpif *group = group_dpif_cast(group_);
4124 ovs_mutex_destroy(&group->stats_mutex);
4128 group_modify(struct ofgroup *group_)
4130 struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
4132 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4138 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
4140 struct group_dpif *group = group_dpif_cast(group_);
4141 struct ofputil_bucket *bucket;
4142 const struct ovs_list *buckets;
4143 struct bucket_counter *bucket_stats;
4145 ovs_mutex_lock(&group->stats_mutex);
4146 ogs->packet_count = group->packet_count;
4147 ogs->byte_count = group->byte_count;
4149 group_dpif_get_buckets(group, &buckets);
4150 bucket_stats = ogs->bucket_stats;
4151 LIST_FOR_EACH (bucket, list_node, buckets) {
4152 bucket_stats->packet_count = bucket->stats.packet_count;
4153 bucket_stats->byte_count = bucket->stats.byte_count;
4156 ovs_mutex_unlock(&group->stats_mutex);
4161 /* If the group exists, this function increments the groups's reference count.
4163 * Make sure to call group_dpif_unref() after no longer needing to maintain
4164 * a reference to the group. */
4166 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
4167 struct group_dpif **group)
4169 struct ofgroup *ofgroup;
4172 found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
4173 *group = found ? group_dpif_cast(ofgroup) : NULL;
4179 group_dpif_get_buckets(const struct group_dpif *group,
4180 const struct ovs_list **buckets)
4182 *buckets = &group->up.buckets;
4185 enum ofp11_group_type
4186 group_dpif_get_type(const struct group_dpif *group)
4188 return group->up.type;
4192 group_dpif_get_selection_method(const struct group_dpif *group)
4194 return group->up.props.selection_method;
4197 /* Sends 'packet' out 'ofport'.
4198 * May modify 'packet'.
4199 * Returns 0 if successful, otherwise a positive errno value. */
4201 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct dp_packet *packet)
4203 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4206 error = xlate_send_packet(ofport, packet);
4208 ovs_mutex_lock(&ofproto->stats_mutex);
4209 ofproto->stats.tx_packets++;
4210 ofproto->stats.tx_bytes += dp_packet_size(packet);
4211 ovs_mutex_unlock(&ofproto->stats_mutex);
4216 group_dpif_get_selection_method_param(const struct group_dpif *group)
4218 return group->up.props.selection_method_param;
4221 const struct field_array *
4222 group_dpif_get_fields(const struct group_dpif *group)
4224 return &group->up.props.fields;
4227 /* Return the version string of the datapath that backs up
4231 get_datapath_version(const struct ofproto *ofproto_)
4233 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4235 return ofproto->backer->dp_version_string;
4239 set_frag_handling(struct ofproto *ofproto_,
4240 enum ofp_config_flags frag_handling)
4242 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4243 if (frag_handling != OFPC_FRAG_REASM) {
4244 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4252 packet_out(struct ofproto *ofproto_, struct dp_packet *packet,
4253 const struct flow *flow,
4254 const struct ofpact *ofpacts, size_t ofpacts_len)
4256 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4258 ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
4259 ofpacts_len, packet);
4266 set_netflow(struct ofproto *ofproto_,
4267 const struct netflow_options *netflow_options)
4269 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4271 if (netflow_options) {
4272 if (!ofproto->netflow) {
4273 ofproto->netflow = netflow_create();
4274 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4276 return netflow_set_options(ofproto->netflow, netflow_options);
4277 } else if (ofproto->netflow) {
4278 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4279 netflow_unref(ofproto->netflow);
4280 ofproto->netflow = NULL;
4287 get_netflow_ids(const struct ofproto *ofproto_,
4288 uint8_t *engine_type, uint8_t *engine_id)
4290 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4292 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
4295 static struct ofproto_dpif *
4296 ofproto_dpif_lookup(const char *name)
4298 struct ofproto_dpif *ofproto;
4300 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
4301 hash_string(name, 0), &all_ofproto_dpifs) {
4302 if (!strcmp(ofproto->up.name, name)) {
4310 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
4311 const char *argv[], void *aux OVS_UNUSED)
4313 struct ofproto_dpif *ofproto;
4316 ofproto = ofproto_dpif_lookup(argv[1]);
4318 unixctl_command_reply_error(conn, "no such bridge");
4321 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4322 mac_learning_flush(ofproto->ml);
4323 ovs_rwlock_unlock(&ofproto->ml->rwlock);
4325 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4326 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4327 mac_learning_flush(ofproto->ml);
4328 ovs_rwlock_unlock(&ofproto->ml->rwlock);
4332 unixctl_command_reply(conn, "table successfully flushed");
4336 ofproto_unixctl_mcast_snooping_flush(struct unixctl_conn *conn, int argc,
4337 const char *argv[], void *aux OVS_UNUSED)
4339 struct ofproto_dpif *ofproto;
4342 ofproto = ofproto_dpif_lookup(argv[1]);
4344 unixctl_command_reply_error(conn, "no such bridge");
4348 if (!mcast_snooping_enabled(ofproto->ms)) {
4349 unixctl_command_reply_error(conn, "multicast snooping is disabled");
4352 mcast_snooping_mdb_flush(ofproto->ms);
4354 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4355 if (!mcast_snooping_enabled(ofproto->ms)) {
4358 mcast_snooping_mdb_flush(ofproto->ms);
4362 unixctl_command_reply(conn, "table successfully flushed");
4365 static struct ofport_dpif *
4366 ofbundle_get_a_port(const struct ofbundle *bundle)
4368 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
4373 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4374 const char *argv[], void *aux OVS_UNUSED)
4376 struct ds ds = DS_EMPTY_INITIALIZER;
4377 const struct ofproto_dpif *ofproto;
4378 const struct mac_entry *e;
4380 ofproto = ofproto_dpif_lookup(argv[1]);
4382 unixctl_command_reply_error(conn, "no such bridge");
4386 ds_put_cstr(&ds, " port VLAN MAC Age\n");
4387 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
4388 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
4389 struct ofbundle *bundle = mac_entry_get_port(ofproto->ml, e);
4390 char name[OFP_MAX_PORT_NAME_LEN];
4392 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4394 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
4395 name, e->vlan, ETH_ADDR_ARGS(e->mac),
4396 mac_entry_age(ofproto->ml, e));
4398 ovs_rwlock_unlock(&ofproto->ml->rwlock);
4399 unixctl_command_reply(conn, ds_cstr(&ds));
4404 ofproto_unixctl_mcast_snooping_show(struct unixctl_conn *conn,
4405 int argc OVS_UNUSED,
4407 void *aux OVS_UNUSED)
4409 struct ds ds = DS_EMPTY_INITIALIZER;
4410 const struct ofproto_dpif *ofproto;
4411 const struct ofbundle *bundle;
4412 const struct mcast_group *grp;
4413 struct mcast_group_bundle *b;
4414 struct mcast_mrouter_bundle *mrouter;
4416 ofproto = ofproto_dpif_lookup(argv[1]);
4418 unixctl_command_reply_error(conn, "no such bridge");
4422 if (!mcast_snooping_enabled(ofproto->ms)) {
4423 unixctl_command_reply_error(conn, "multicast snooping is disabled");
4427 ds_put_cstr(&ds, " port VLAN GROUP Age\n");
4428 ovs_rwlock_rdlock(&ofproto->ms->rwlock);
4429 LIST_FOR_EACH (grp, group_node, &ofproto->ms->group_lru) {
4430 LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
4431 char name[OFP_MAX_PORT_NAME_LEN];
4434 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4436 ds_put_format(&ds, "%5s %4d "IP_FMT" %3d\n",
4437 name, grp->vlan, IP_ARGS(grp->ip4),
4438 mcast_bundle_age(ofproto->ms, b));
4442 /* ports connected to multicast routers */
4443 LIST_FOR_EACH(mrouter, mrouter_node, &ofproto->ms->mrouter_lru) {
4444 char name[OFP_MAX_PORT_NAME_LEN];
4446 bundle = mrouter->port;
4447 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4449 ds_put_format(&ds, "%5s %4d querier %3d\n",
4450 name, mrouter->vlan,
4451 mcast_mrouter_age(ofproto->ms, mrouter));
4453 ovs_rwlock_unlock(&ofproto->ms->rwlock);
4454 unixctl_command_reply(conn, ds_cstr(&ds));
4459 struct xlate_out xout;
4460 struct xlate_in xin;
4461 const struct flow *key;
4463 struct flow_wildcards wc;
4468 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
4470 const struct rule_actions *actions;
4473 ds_put_char_multiple(result, '\t', level);
4475 ds_put_cstr(result, "No match\n");
4479 ovs_mutex_lock(&rule->up.mutex);
4480 cookie = rule->up.flow_cookie;
4481 ovs_mutex_unlock(&rule->up.mutex);
4483 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
4484 rule ? rule->up.table_id : 0, ntohll(cookie));
4485 cls_rule_format(&rule->up.cr, result);
4486 ds_put_char(result, '\n');
4488 actions = rule_dpif_get_actions(rule);
4490 ds_put_char_multiple(result, '\t', level);
4491 ds_put_cstr(result, "OpenFlow actions=");
4492 ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
4493 ds_put_char(result, '\n');
4497 trace_format_flow(struct ds *result, int level, const char *title,
4498 struct trace_ctx *trace)
4500 ds_put_char_multiple(result, '\t', level);
4501 ds_put_format(result, "%s: ", title);
4502 /* Do not report unchanged flows for resubmits. */
4503 if ((level > 0 && flow_equal(&trace->xin.flow, &trace->flow))
4504 || (level == 0 && flow_equal(&trace->xin.flow, trace->key))) {
4505 ds_put_cstr(result, "unchanged");
4507 flow_format(result, &trace->xin.flow);
4508 trace->flow = trace->xin.flow;
4510 ds_put_char(result, '\n');
4514 trace_format_regs(struct ds *result, int level, const char *title,
4515 struct trace_ctx *trace)
4519 ds_put_char_multiple(result, '\t', level);
4520 ds_put_format(result, "%s:", title);
4521 for (i = 0; i < FLOW_N_REGS; i++) {
4522 ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
4524 ds_put_char(result, '\n');
4528 trace_format_odp(struct ds *result, int level, const char *title,
4529 struct trace_ctx *trace)
4531 struct ofpbuf *odp_actions = trace->xout.odp_actions;
4533 ds_put_char_multiple(result, '\t', level);
4534 ds_put_format(result, "%s: ", title);
4535 format_odp_actions(result, odp_actions->data, odp_actions->size);
4536 ds_put_char(result, '\n');
4540 trace_format_megaflow(struct ds *result, int level, const char *title,
4541 struct trace_ctx *trace)
4545 ds_put_char_multiple(result, '\t', level);
4546 ds_put_format(result, "%s: ", title);
4547 flow_wildcards_or(&trace->wc, &trace->xout.wc, &trace->wc);
4548 match_init(&match, trace->key, &trace->wc);
4549 match_format(&match, result, OFP_DEFAULT_PRIORITY);
4550 ds_put_char(result, '\n');
4553 static void trace_report(struct xlate_in *xin, const char *s, int recurse);
4556 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
4558 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4559 struct ds *result = trace->result;
4562 if (rule == xin->ofproto->miss_rule) {
4563 trace_report(xin, "No match, flow generates \"packet in\"s.",
4565 } else if (rule == xin->ofproto->no_packet_in_rule) {
4566 trace_report(xin, "No match, packets dropped because "
4567 "OFPPC_NO_PACKET_IN is set on in_port.", recurse);
4568 } else if (rule == xin->ofproto->drop_frags_rule) {
4569 trace_report(xin, "Packets dropped because they are IP "
4570 "fragments and the fragment handling mode is "
4571 "\"drop\".", recurse);
4575 ds_put_char(result, '\n');
4577 trace_format_flow(result, recurse, "Resubmitted flow", trace);
4578 trace_format_regs(result, recurse, "Resubmitted regs", trace);
4579 trace_format_odp(result, recurse, "Resubmitted odp", trace);
4580 trace_format_megaflow(result, recurse, "Resubmitted megaflow", trace);
4582 trace_format_rule(result, recurse, rule);
4586 trace_report(struct xlate_in *xin, const char *s, int recurse)
4588 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4589 struct ds *result = trace->result;
4591 ds_put_char_multiple(result, '\t', recurse);
4592 ds_put_cstr(result, s);
4593 ds_put_char(result, '\n');
4596 /* Parses the 'argc' elements of 'argv', ignoring argv[0]. The following
4597 * forms are supported:
4599 * - [dpname] odp_flow [-generate | packet]
4600 * - bridge br_flow [-generate | packet]
4602 * On success, initializes '*ofprotop' and 'flow' and returns NULL. On failure
4603 * returns a nonnull malloced error message. */
4604 static char * OVS_WARN_UNUSED_RESULT
4605 parse_flow_and_packet(int argc, const char *argv[],
4606 struct ofproto_dpif **ofprotop, struct flow *flow,
4607 struct dp_packet **packetp)
4609 const struct dpif_backer *backer = NULL;
4610 const char *error = NULL;
4612 struct simap port_names = SIMAP_INITIALIZER(&port_names);
4613 struct dp_packet *packet;
4614 struct ofpbuf odp_key;
4615 struct ofpbuf odp_mask;
4617 ofpbuf_init(&odp_key, 0);
4618 ofpbuf_init(&odp_mask, 0);
4620 /* Handle "-generate" or a hex string as the last argument. */
4621 if (!strcmp(argv[argc - 1], "-generate")) {
4622 packet = dp_packet_new(0);
4625 error = eth_from_hex(argv[argc - 1], &packet);
4628 } else if (argc == 4) {
4629 /* The 3-argument form must end in "-generate' or a hex string. */
4635 /* odp_flow can have its in_port specified as a name instead of port no.
4636 * We do not yet know whether a given flow is a odp_flow or a br_flow.
4637 * But, to know whether a flow is odp_flow through odp_flow_from_string(),
4638 * we need to create a simap of name to port no. */
4640 const char *dp_type;
4641 if (!strncmp(argv[1], "ovs-", 4)) {
4642 dp_type = argv[1] + 4;
4646 backer = shash_find_data(&all_dpif_backers, dp_type);
4647 } else if (argc == 2) {
4648 struct shash_node *node;
4649 if (shash_count(&all_dpif_backers) == 1) {
4650 node = shash_first(&all_dpif_backers);
4651 backer = node->data;
4654 error = "Syntax error";
4657 if (backer && backer->dpif) {
4658 struct dpif_port dpif_port;
4659 struct dpif_port_dump port_dump;
4660 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
4661 simap_put(&port_names, dpif_port.name,
4662 odp_to_u32(dpif_port.port_no));
4666 /* Parse the flow and determine whether a datapath or
4667 * bridge is specified. If function odp_flow_key_from_string()
4668 * returns 0, the flow is a odp_flow. If function
4669 * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
4670 if (!odp_flow_from_string(argv[argc - 1], &port_names,
4671 &odp_key, &odp_mask)) {
4673 error = "Cannot find the datapath";
4677 if (odp_flow_key_to_flow(odp_key.data, odp_key.size, flow) == ODP_FIT_ERROR) {
4678 error = "Failed to parse datapath flow key";
4682 *ofprotop = xlate_lookup_ofproto(backer, flow,
4683 &flow->in_port.ofp_port);
4684 if (*ofprotop == NULL) {
4685 error = "Invalid datapath flow";
4689 vsp_adjust_flow(*ofprotop, flow, NULL);
4692 char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
4695 m_err = xasprintf("Bad openflow flow syntax: %s", err);
4700 error = "Must specify bridge name";
4704 *ofprotop = ofproto_dpif_lookup(argv[1]);
4706 error = "Unknown bridge name";
4712 /* Generate a packet, if requested. */
4714 if (!dp_packet_size(packet)) {
4715 flow_compose(packet, flow);
4717 /* Use the metadata from the flow and the packet argument
4718 * to reconstruct the flow. */
4719 pkt_metadata_from_flow(&packet->md, flow);
4720 flow_extract(packet, flow);
4725 if (error && !m_err) {
4726 m_err = xstrdup(error);
4729 dp_packet_delete(packet);
4733 ofpbuf_uninit(&odp_key);
4734 ofpbuf_uninit(&odp_mask);
4735 simap_destroy(&port_names);
4740 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
4741 void *aux OVS_UNUSED)
4743 struct ofproto_dpif *ofproto;
4744 struct dp_packet *packet;
4748 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4753 ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
4754 unixctl_command_reply(conn, ds_cstr(&result));
4755 ds_destroy(&result);
4756 dp_packet_delete(packet);
4758 unixctl_command_reply_error(conn, error);
4764 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
4765 const char *argv[], void *aux OVS_UNUSED)
4767 enum ofputil_protocol usable_protocols;
4768 struct ofproto_dpif *ofproto;
4769 bool enforce_consistency;
4770 struct ofpbuf ofpacts;
4771 struct dp_packet *packet;
4776 /* Three kinds of error return values! */
4782 ofpbuf_init(&ofpacts, 0);
4784 /* Parse actions. */
4785 error = ofpacts_parse_actions(argv[--argc], &ofpacts, &usable_protocols);
4787 unixctl_command_reply_error(conn, error);
4792 /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
4793 * consistency between the flow and the actions. With -consistent, we
4794 * enforce consistency even for a flow supported in OpenFlow 1.0. */
4795 if (!strcmp(argv[1], "-consistent")) {
4796 enforce_consistency = true;
4800 enforce_consistency = false;
4803 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4805 unixctl_command_reply_error(conn, error);
4810 /* Do the same checks as handle_packet_out() in ofproto.c.
4812 * We pass a 'table_id' of 0 to ofpacts_check(), which isn't
4813 * strictly correct because these actions aren't in any table, but it's OK
4814 * because it 'table_id' is used only to check goto_table instructions, but
4815 * packet-outs take a list of actions and therefore it can't include
4818 * We skip the "meter" check here because meter is an instruction, not an
4819 * action, and thus cannot appear in ofpacts. */
4820 in_port = ofp_to_u16(flow.in_port.ofp_port);
4821 if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
4822 unixctl_command_reply_error(conn, "invalid in_port");
4825 if (enforce_consistency) {
4826 retval = ofpacts_check_consistency(ofpacts.data, ofpacts.size,
4827 &flow, u16_to_ofp(ofproto->up.max_ports),
4828 0, 0, usable_protocols);
4830 retval = ofpacts_check(ofpacts.data, ofpacts.size, &flow,
4831 u16_to_ofp(ofproto->up.max_ports), 0, 0,
4837 ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
4838 unixctl_command_reply_error(conn, ds_cstr(&result));
4842 ofproto_trace(ofproto, &flow, packet,
4843 ofpacts.data, ofpacts.size, &result);
4844 unixctl_command_reply(conn, ds_cstr(&result));
4847 ds_destroy(&result);
4848 dp_packet_delete(packet);
4849 ofpbuf_uninit(&ofpacts);
4852 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
4853 * description of the results to 'ds'.
4855 * The trace follows a packet with the specified 'flow' through the flow
4856 * table. 'packet' may be nonnull to trace an actual packet, with consequent
4857 * side effects (if it is nonnull then its flow must be 'flow').
4859 * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
4860 * trace, otherwise the actions are determined by a flow table lookup. */
4862 ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
4863 const struct dp_packet *packet,
4864 const struct ofpact ofpacts[], size_t ofpacts_len,
4867 struct trace_ctx trace;
4869 ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
4870 ds_put_cstr(ds, "Flow: ");
4871 flow_format(ds, flow);
4872 ds_put_char(ds, '\n');
4874 flow_wildcards_init_catchall(&trace.wc);
4877 trace.key = flow; /* Original flow key, used for megaflow. */
4878 trace.flow = *flow; /* May be modified by actions. */
4879 xlate_in_init(&trace.xin, ofproto, flow, flow->in_port.ofp_port, NULL,
4880 ntohs(flow->tcp_flags), packet);
4881 trace.xin.ofpacts = ofpacts;
4882 trace.xin.ofpacts_len = ofpacts_len;
4883 trace.xin.resubmit_hook = trace_resubmit;
4884 trace.xin.report_hook = trace_report;
4886 xlate_actions(&trace.xin, &trace.xout);
4888 ds_put_char(ds, '\n');
4889 trace_format_flow(ds, 0, "Final flow", &trace);
4890 trace_format_megaflow(ds, 0, "Megaflow", &trace);
4892 ds_put_cstr(ds, "Datapath actions: ");
4893 format_odp_actions(ds, trace.xout.odp_actions->data,
4894 trace.xout.odp_actions->size);
4896 if (trace.xout.slow) {
4897 enum slow_path_reason slow;
4899 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
4900 "slow path because it:");
4902 slow = trace.xout.slow;
4904 enum slow_path_reason bit = rightmost_1bit(slow);
4906 ds_put_format(ds, "\n\t- %s.",
4907 slow_path_reason_to_explanation(bit));
4913 xlate_out_uninit(&trace.xout);
4916 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
4917 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
4918 * to destroy 'ofproto_shash' and free the returned value. */
4919 static const struct shash_node **
4920 get_ofprotos(struct shash *ofproto_shash)
4922 const struct ofproto_dpif *ofproto;
4924 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4925 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
4926 shash_add_nocopy(ofproto_shash, name, ofproto);
4929 return shash_sort(ofproto_shash);
4933 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
4934 const char *argv[] OVS_UNUSED,
4935 void *aux OVS_UNUSED)
4937 struct ds ds = DS_EMPTY_INITIALIZER;
4938 struct shash ofproto_shash;
4939 const struct shash_node **sorted_ofprotos;
4942 shash_init(&ofproto_shash);
4943 sorted_ofprotos = get_ofprotos(&ofproto_shash);
4944 for (i = 0; i < shash_count(&ofproto_shash); i++) {
4945 const struct shash_node *node = sorted_ofprotos[i];
4946 ds_put_format(&ds, "%s\n", node->name);
4949 shash_destroy(&ofproto_shash);
4950 free(sorted_ofprotos);
4952 unixctl_command_reply(conn, ds_cstr(&ds));
4957 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
4959 const struct shash_node **ofprotos;
4960 struct dpif_dp_stats dp_stats;
4961 struct shash ofproto_shash;
4964 dpif_get_dp_stats(backer->dpif, &dp_stats);
4966 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
4967 dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
4969 shash_init(&ofproto_shash);
4970 ofprotos = get_ofprotos(&ofproto_shash);
4971 for (i = 0; i < shash_count(&ofproto_shash); i++) {
4972 struct ofproto_dpif *ofproto = ofprotos[i]->data;
4973 const struct shash_node **ports;
4976 if (ofproto->backer != backer) {
4980 ds_put_format(ds, "\t%s:\n", ofproto->up.name);
4982 ports = shash_sort(&ofproto->up.port_by_name);
4983 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
4984 const struct shash_node *node = ports[j];
4985 struct ofport *ofport = node->data;
4987 odp_port_t odp_port;
4989 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
4992 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
4993 if (odp_port != ODPP_NONE) {
4994 ds_put_format(ds, "%"PRIu32":", odp_port);
4996 ds_put_cstr(ds, "none:");
4999 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5002 if (!netdev_get_config(ofport->netdev, &config)) {
5003 const struct smap_node **nodes;
5006 nodes = smap_sort(&config);
5007 for (i = 0; i < smap_count(&config); i++) {
5008 const struct smap_node *node = nodes[i];
5009 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5010 node->key, node->value);
5014 smap_destroy(&config);
5016 ds_put_char(ds, ')');
5017 ds_put_char(ds, '\n');
5021 shash_destroy(&ofproto_shash);
5026 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5027 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5029 struct ds ds = DS_EMPTY_INITIALIZER;
5030 const struct shash_node **backers;
5033 backers = shash_sort(&all_dpif_backers);
5034 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5035 dpif_show_backer(backers[i]->data, &ds);
5039 unixctl_command_reply(conn, ds_cstr(&ds));
5044 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5045 int argc OVS_UNUSED, const char *argv[],
5046 void *aux OVS_UNUSED)
5048 const struct ofproto_dpif *ofproto;
5050 struct ds ds = DS_EMPTY_INITIALIZER;
5051 bool verbosity = false;
5053 struct dpif_port dpif_port;
5054 struct dpif_port_dump port_dump;
5055 struct hmap portno_names;
5057 struct dpif_flow_dump *flow_dump;
5058 struct dpif_flow_dump_thread *flow_dump_thread;
5062 ofproto = ofproto_dpif_lookup(argv[argc - 1]);
5064 unixctl_command_reply_error(conn, "no such bridge");
5068 if (argc > 2 && !strcmp(argv[1], "-m")) {
5072 hmap_init(&portno_names);
5073 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
5074 odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
5078 flow_dump = dpif_flow_dump_create(ofproto->backer->dpif, false);
5079 flow_dump_thread = dpif_flow_dump_thread_create(flow_dump);
5080 while (dpif_flow_dump_next(flow_dump_thread, &f, 1)) {
5083 if (odp_flow_key_to_flow(f.key, f.key_len, &flow) == ODP_FIT_ERROR
5084 || xlate_lookup_ofproto(ofproto->backer, &flow, NULL) != ofproto) {
5089 odp_format_ufid(&f.ufid, &ds);
5090 ds_put_cstr(&ds, " ");
5092 odp_flow_format(f.key, f.key_len, f.mask, f.mask_len,
5093 &portno_names, &ds, verbosity);
5094 ds_put_cstr(&ds, ", ");
5095 dpif_flow_stats_format(&f.stats, &ds);
5096 ds_put_cstr(&ds, ", actions:");
5097 format_odp_actions(&ds, f.actions, f.actions_len);
5098 ds_put_char(&ds, '\n');
5100 dpif_flow_dump_thread_destroy(flow_dump_thread);
5101 error = dpif_flow_dump_destroy(flow_dump);
5105 ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
5106 unixctl_command_reply_error(conn, ds_cstr(&ds));
5108 unixctl_command_reply(conn, ds_cstr(&ds));
5110 odp_portno_names_destroy(&portno_names);
5111 hmap_destroy(&portno_names);
5116 ofproto_revalidate_all_backers(void)
5118 const struct shash_node **backers;
5121 backers = shash_sort(&all_dpif_backers);
5122 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5123 struct dpif_backer *backer = backers[i]->data;
5124 backer->need_revalidate = REV_RECONFIGURE;
5130 disable_tnl_push_pop(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5131 const char *argv[], void *aux OVS_UNUSED)
5133 if (!strcasecmp(argv[1], "off")) {
5134 ofproto_use_tnl_push_pop = false;
5135 unixctl_command_reply(conn, "Tunnel push-pop off");
5136 ofproto_revalidate_all_backers();
5137 } else if (!strcasecmp(argv[1], "on")) {
5138 ofproto_use_tnl_push_pop = true;
5139 unixctl_command_reply(conn, "Tunnel push-pop on");
5140 ofproto_revalidate_all_backers();
5145 ofproto_unixctl_init(void)
5147 static bool registered;
5153 unixctl_command_register(
5155 "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
5156 1, 3, ofproto_unixctl_trace, NULL);
5157 unixctl_command_register(
5158 "ofproto/trace-packet-out",
5159 "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
5160 2, 6, ofproto_unixctl_trace_actions, NULL);
5161 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5162 ofproto_unixctl_fdb_flush, NULL);
5163 unixctl_command_register("fdb/show", "bridge", 1, 1,
5164 ofproto_unixctl_fdb_show, NULL);
5165 unixctl_command_register("mdb/flush", "[bridge]", 0, 1,
5166 ofproto_unixctl_mcast_snooping_flush, NULL);
5167 unixctl_command_register("mdb/show", "bridge", 1, 1,
5168 ofproto_unixctl_mcast_snooping_show, NULL);
5169 unixctl_command_register("dpif/dump-dps", "", 0, 0,
5170 ofproto_unixctl_dpif_dump_dps, NULL);
5171 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5173 unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
5174 ofproto_unixctl_dpif_dump_flows, NULL);
5176 unixctl_command_register("ofproto/tnl-push-pop", "[on]|[off]", 1, 1,
5177 disable_tnl_push_pop, NULL);
5180 /* Returns true if 'table' is the table used for internal rules,
5181 * false otherwise. */
5183 table_is_internal(uint8_t table_id)
5185 return table_id == TBL_INTERNAL;
5188 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5190 * This is deprecated. It is only for compatibility with broken device drivers
5191 * in old versions of Linux that do not properly support VLANs when VLAN
5192 * devices are not used. When broken device drivers are no longer in
5193 * widespread use, we will delete these interfaces. */
5196 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
5198 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5199 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5201 if (realdev_ofp_port == ofport->realdev_ofp_port
5202 && vid == ofport->vlandev_vid) {
5206 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5208 if (ofport->realdev_ofp_port) {
5211 if (realdev_ofp_port && ofport->bundle) {
5212 /* vlandevs are enslaved to their realdevs, so they are not allowed to
5213 * themselves be part of a bundle. */
5214 bundle_set(ofport_->ofproto, ofport->bundle, NULL);
5217 ofport->realdev_ofp_port = realdev_ofp_port;
5218 ofport->vlandev_vid = vid;
5220 if (realdev_ofp_port) {
5221 vsp_add(ofport, realdev_ofp_port, vid);
5228 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
5230 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
5234 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
5235 OVS_EXCLUDED(ofproto->vsp_mutex)
5237 /* hmap_is_empty is thread safe. */
5238 return !hmap_is_empty(&ofproto->realdev_vid_map);
5243 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
5244 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5245 OVS_REQUIRES(ofproto->vsp_mutex)
5247 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5248 int vid = vlan_tci_to_vid(vlan_tci);
5249 const struct vlan_splinter *vsp;
5251 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5252 hash_realdev_vid(realdev_ofp_port, vid),
5253 &ofproto->realdev_vid_map) {
5254 if (vsp->realdev_ofp_port == realdev_ofp_port
5255 && vsp->vid == vid) {
5256 return vsp->vlandev_ofp_port;
5260 return realdev_ofp_port;
5263 /* Returns the OFP port number of the Linux VLAN device that corresponds to
5264 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
5265 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
5266 * 'vlan_tci' 9, it would return the port number of eth0.9.
5268 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
5269 * function just returns its 'realdev_ofp_port' argument. */
5271 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5272 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5273 OVS_EXCLUDED(ofproto->vsp_mutex)
5277 /* hmap_is_empty is thread safe, see if we can return immediately. */
5278 if (hmap_is_empty(&ofproto->realdev_vid_map)) {
5279 return realdev_ofp_port;
5281 ovs_mutex_lock(&ofproto->vsp_mutex);
5282 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
5283 ovs_mutex_unlock(&ofproto->vsp_mutex);
5287 static struct vlan_splinter *
5288 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
5290 struct vlan_splinter *vsp;
5292 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
5293 hash_ofp_port(vlandev_ofp_port),
5294 &ofproto->vlandev_map) {
5295 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5303 /* Returns the OpenFlow port number of the "real" device underlying the Linux
5304 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
5305 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
5306 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
5307 * eth0 and store 9 in '*vid'.
5309 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
5310 * VLAN device. Unless VLAN splinters are enabled, this is what this function
5313 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5314 ofp_port_t vlandev_ofp_port, int *vid)
5315 OVS_REQUIRES(ofproto->vsp_mutex)
5317 if (!hmap_is_empty(&ofproto->vlandev_map)) {
5318 const struct vlan_splinter *vsp;
5320 vsp = vlandev_find(ofproto, vlandev_ofp_port);
5325 return vsp->realdev_ofp_port;
5331 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
5332 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
5333 * 'flow->in_port' to the "real" device backing the VLAN device, sets
5334 * 'flow->vlan_tci' to the VLAN VID, and returns true. Optionally pushes the
5335 * appropriate VLAN on 'packet' if provided. Otherwise (which is always the
5336 * case unless VLAN splinters are enabled), returns false without making any
5339 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow,
5340 struct dp_packet *packet)
5341 OVS_EXCLUDED(ofproto->vsp_mutex)
5346 /* hmap_is_empty is thread safe. */
5347 if (hmap_is_empty(&ofproto->vlandev_map)) {
5351 ovs_mutex_lock(&ofproto->vsp_mutex);
5352 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
5353 ovs_mutex_unlock(&ofproto->vsp_mutex);
5358 /* Cause the flow to be processed as if it came in on the real device with
5359 * the VLAN device's VLAN ID. */
5360 flow->in_port.ofp_port = realdev;
5361 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
5364 /* Make the packet resemble the flow, so that it gets sent to an
5365 * OpenFlow controller properly, so that it looks correct for sFlow,
5366 * and so that flow_extract() will get the correct vlan_tci if it is
5367 * called on 'packet'. */
5368 eth_push_vlan(packet, htons(ETH_TYPE_VLAN), flow->vlan_tci);
5375 vsp_remove(struct ofport_dpif *port)
5377 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5378 struct vlan_splinter *vsp;
5380 ovs_mutex_lock(&ofproto->vsp_mutex);
5381 vsp = vlandev_find(ofproto, port->up.ofp_port);
5383 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
5384 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
5387 port->realdev_ofp_port = 0;
5389 VLOG_ERR("missing vlan device record");
5391 ovs_mutex_unlock(&ofproto->vsp_mutex);
5395 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
5397 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5399 ovs_mutex_lock(&ofproto->vsp_mutex);
5400 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
5401 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
5402 == realdev_ofp_port)) {
5403 struct vlan_splinter *vsp;
5405 vsp = xmalloc(sizeof *vsp);
5406 vsp->realdev_ofp_port = realdev_ofp_port;
5407 vsp->vlandev_ofp_port = port->up.ofp_port;
5410 port->realdev_ofp_port = realdev_ofp_port;
5412 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
5413 hash_ofp_port(port->up.ofp_port));
5414 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
5415 hash_realdev_vid(realdev_ofp_port, vid));
5417 VLOG_ERR("duplicate vlan device record");
5419 ovs_mutex_unlock(&ofproto->vsp_mutex);
5423 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
5425 const struct ofport_dpif *ofport = ofp_port_to_ofport(ofproto, ofp_port);
5426 return ofport ? ofport->odp_port : ODPP_NONE;
5429 struct ofport_dpif *
5430 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
5432 struct ofport_dpif *port;
5434 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
5435 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
5436 &backer->odp_to_ofport_map) {
5437 if (port->odp_port == odp_port) {
5438 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5443 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5448 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
5450 struct ofport_dpif *port;
5452 port = odp_port_to_ofport(ofproto->backer, odp_port);
5453 if (port && &ofproto->up == port->up.ofproto) {
5454 return port->up.ofp_port;
5461 ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
5462 const struct match *match, int priority,
5463 uint16_t idle_timeout,
5464 const struct ofpbuf *ofpacts,
5465 struct rule **rulep)
5467 struct ofputil_flow_mod fm;
5468 struct rule_dpif *rule;
5472 fm.priority = priority;
5473 fm.new_cookie = htonll(0);
5474 fm.cookie = htonll(0);
5475 fm.cookie_mask = htonll(0);
5476 fm.modify_cookie = false;
5477 fm.table_id = TBL_INTERNAL;
5478 fm.command = OFPFC_ADD;
5479 fm.idle_timeout = idle_timeout;
5480 fm.hard_timeout = 0;
5484 fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5485 fm.ofpacts = ofpacts->data;
5486 fm.ofpacts_len = ofpacts->size;
5488 error = ofproto_flow_mod(&ofproto->up, &fm);
5490 VLOG_ERR_RL(&rl, "failed to add internal flow (%s)",
5491 ofperr_to_string(error));
5496 rule = rule_dpif_lookup_in_table(ofproto,
5497 ofproto_dpif_get_tables_version(ofproto),
5498 TBL_INTERNAL, &fm.match.flow,
5499 &fm.match.wc, false);
5509 ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
5510 struct match *match, int priority)
5512 struct ofputil_flow_mod fm;
5516 fm.priority = priority;
5517 fm.new_cookie = htonll(0);
5518 fm.cookie = htonll(0);
5519 fm.cookie_mask = htonll(0);
5520 fm.modify_cookie = false;
5521 fm.table_id = TBL_INTERNAL;
5522 fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5523 fm.command = OFPFC_DELETE_STRICT;
5525 error = ofproto_flow_mod(&ofproto->up, &fm);
5527 VLOG_ERR_RL(&rl, "failed to delete internal flow (%s)",
5528 ofperr_to_string(error));
5535 const struct ofproto_class ofproto_dpif_class = {
5549 NULL, /* get_memory_usage. */
5550 type_get_memory_usage,
5569 port_is_lacp_current,
5570 port_get_lacp_stats,
5571 NULL, /* rule_choose_table */
5595 aa_vlan_get_queue_size,
5602 get_stp_port_status,
5607 get_rstp_port_status,
5614 is_mirror_output_bundle,
5615 forward_bpdu_changed,
5616 set_mac_table_config,
5618 set_mcast_snooping_port,
5620 NULL, /* meter_get_features */
5621 NULL, /* meter_set */
5622 NULL, /* meter_get */
5623 NULL, /* meter_del */
5624 group_alloc, /* group_alloc */
5625 group_construct, /* group_construct */
5626 group_destruct, /* group_destruct */
5627 group_dealloc, /* group_dealloc */
5628 group_modify, /* group_modify */
5629 group_get_stats, /* group_get_stats */
5630 get_datapath_version, /* get_datapath_version */