2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-dpif.h"
20 #include "ofproto/ofproto-provider.h"
27 #include "byte-order.h"
28 #include "connectivity.h"
33 #include "dynamic-string.h"
34 #include "fail-open.h"
35 #include "guarded-list.h"
39 #include "mac-learning.h"
40 #include "meta-flow.h"
41 #include "multipath.h"
42 #include "netdev-vport.h"
47 #include "odp-execute.h"
50 #include "ofp-actions.h"
51 #include "ofp-parse.h"
52 #include "ofp-print.h"
53 #include "ofproto-dpif-ipfix.h"
54 #include "ofproto-dpif-mirror.h"
55 #include "ofproto-dpif-monitor.h"
56 #include "ofproto-dpif-sflow.h"
57 #include "ofproto-dpif-upcall.h"
58 #include "ofproto-dpif-xlate.h"
59 #include "poll-loop.h"
65 #include "unaligned.h"
67 #include "vlan-bitmap.h"
70 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
72 COVERAGE_DEFINE(ofproto_dpif_expired);
73 COVERAGE_DEFINE(packet_in_overflow);
75 /* Number of implemented OpenFlow tables. */
76 enum { N_TABLES = 255 };
77 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
78 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
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;
93 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes,
95 static struct rule_dpif *rule_dpif_cast(const struct rule *);
96 static void rule_expire(struct rule_dpif *);
103 * - Do include packets and bytes from datapath flows which have not
104 * recently been processed by a revalidator. */
105 struct ovs_mutex stats_mutex;
106 uint64_t packet_count OVS_GUARDED; /* Number of packets received. */
107 uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */
108 struct bucket_counter *bucket_stats OVS_GUARDED; /* Bucket statistics. */
112 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
113 struct ofproto_dpif *ofproto; /* Owning ofproto. */
114 void *aux; /* Key supplied by ofproto's client. */
115 char *name; /* Identifier for log messages. */
118 struct list ports; /* Contains "struct ofport"s. */
119 enum port_vlan_mode vlan_mode; /* VLAN mode */
120 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
121 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
122 * NULL if all VLANs are trunked. */
123 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
124 struct bond *bond; /* Nonnull iff more than one port. */
125 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
128 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
131 static void bundle_remove(struct ofport *);
132 static void bundle_update(struct ofbundle *);
133 static void bundle_destroy(struct ofbundle *);
134 static void bundle_del_port(struct ofport_dpif *);
135 static void bundle_run(struct ofbundle *);
136 static void bundle_wait(struct ofbundle *);
138 static void stp_run(struct ofproto_dpif *ofproto);
139 static void stp_wait(struct ofproto_dpif *ofproto);
140 static int set_stp_port(struct ofport *,
141 const struct ofproto_port_stp_settings *);
144 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
148 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
149 struct list bundle_node; /* In struct ofbundle's "ports" list. */
150 struct cfm *cfm; /* Connectivity Fault Management, if any. */
151 struct bfd *bfd; /* BFD, if any. */
152 bool may_enable; /* May be enabled in bonds. */
153 bool is_tunnel; /* This port is a tunnel. */
154 bool is_layer3; /* This is a layer 3 port. */
155 long long int carrier_seq; /* Carrier status changes. */
156 struct ofport_dpif *peer; /* Peer if patch port. */
159 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
160 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
161 long long int stp_state_entered;
163 /* Queue to DSCP mapping. */
164 struct ofproto_port_queue *qdscp;
167 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
169 * This is deprecated. It is only for compatibility with broken device
170 * drivers in old versions of Linux that do not properly support VLANs when
171 * VLAN devices are not used. When broken device drivers are no longer in
172 * widespread use, we will delete these interfaces. */
173 ofp_port_t realdev_ofp_port;
177 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
179 * This is deprecated. It is only for compatibility with broken device drivers
180 * in old versions of Linux that do not properly support VLANs when VLAN
181 * devices are not used. When broken device drivers are no longer in
182 * widespread use, we will delete these interfaces. */
183 struct vlan_splinter {
184 struct hmap_node realdev_vid_node;
185 struct hmap_node vlandev_node;
186 ofp_port_t realdev_ofp_port;
187 ofp_port_t vlandev_ofp_port;
191 static void vsp_remove(struct ofport_dpif *);
192 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
194 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
197 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
200 static struct ofport_dpif *
201 ofport_dpif_cast(const struct ofport *ofport)
203 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
206 static void port_run(struct ofport_dpif *);
207 static int set_bfd(struct ofport *, const struct smap *);
208 static int set_cfm(struct ofport *, const struct cfm_settings *);
209 static void ofport_update_peer(struct ofport_dpif *);
211 struct dpif_completion {
212 struct list list_node;
213 struct ofoperation *op;
216 /* Reasons that we might need to revalidate every datapath flow, and
217 * corresponding coverage counters.
219 * A value of 0 means that there is no need to revalidate.
221 * It would be nice to have some cleaner way to integrate with coverage
222 * counters, but with only a few reasons I guess this is good enough for
224 enum revalidate_reason {
225 REV_RECONFIGURE = 1, /* Switch configuration changed. */
226 REV_STP, /* Spanning tree protocol port status change. */
227 REV_BOND, /* Bonding changed. */
228 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
229 REV_FLOW_TABLE, /* Flow table changed. */
230 REV_MAC_LEARNING, /* Mac learning changed. */
232 COVERAGE_DEFINE(rev_reconfigure);
233 COVERAGE_DEFINE(rev_stp);
234 COVERAGE_DEFINE(rev_bond);
235 COVERAGE_DEFINE(rev_port_toggled);
236 COVERAGE_DEFINE(rev_flow_table);
237 COVERAGE_DEFINE(rev_mac_learning);
239 /* All datapaths of a given type share a single dpif backer instance. */
246 struct ovs_rwlock odp_to_ofport_lock;
247 struct hmap odp_to_ofport_map OVS_GUARDED; /* Contains "struct ofport"s. */
249 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
251 enum revalidate_reason need_revalidate; /* Revalidate all flows. */
253 bool recv_set_enable; /* Enables or disables receiving packets. */
255 /* True if the datapath supports variable-length
256 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.
257 * False if the datapath supports only 8-byte (or shorter) userdata. */
258 bool variable_length_userdata;
260 /* Maximum number of MPLS label stack entries that the datapath supports
262 size_t max_mpls_depth;
265 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
266 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
268 struct ofproto_dpif {
269 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
271 struct dpif_backer *backer;
273 uint64_t dump_seq; /* Last read of udpif_dump_seq(). */
275 /* Special OpenFlow rules. */
276 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
277 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
278 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
281 struct netflow *netflow;
282 struct dpif_sflow *sflow;
283 struct dpif_ipfix *ipfix;
284 struct hmap bundles; /* Contains "struct ofbundle"s. */
285 struct mac_learning *ml;
286 bool has_bonded_bundles;
288 struct mbridge *mbridge;
290 struct ovs_mutex stats_mutex;
291 struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
292 * consumed in userspace. */
296 long long int stp_last_tick;
298 /* VLAN splinters. */
299 struct ovs_mutex vsp_mutex;
300 struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
301 struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */
304 struct sset ports; /* Set of standard port names. */
305 struct sset ghost_ports; /* Ports with no datapath port. */
306 struct sset port_poll_set; /* Queued names for port_poll() reply. */
307 int port_poll_errno; /* Last errno for port_poll() reply. */
308 uint64_t change_seq; /* Connectivity status changes. */
311 struct guarded_list pins; /* Contains "struct ofputil_packet_in"s. */
314 /* All existing ofproto_dpif instances, indexed by ->up.name. */
315 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
317 static void ofproto_dpif_unixctl_init(void);
319 static inline struct ofproto_dpif *
320 ofproto_dpif_cast(const struct ofproto *ofproto)
322 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
323 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
327 ofproto_dpif_get_max_mpls_depth(const struct ofproto_dpif *ofproto)
329 return ofproto->backer->max_mpls_depth;
332 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
333 ofp_port_t ofp_port);
334 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
335 const struct ofpbuf *packet,
336 const struct ofpact[], size_t ofpacts_len,
339 /* Global variables. */
340 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
342 /* Initial mappings of port to bridge mappings. */
343 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
345 /* Executes 'fm'. The caller retains ownership of 'fm' and everything in
348 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
349 struct ofputil_flow_mod *fm)
351 ofproto_flow_mod(&ofproto->up, fm);
354 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
355 * Takes ownership of 'pin' and pin->packet. */
357 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
358 struct ofproto_packet_in *pin)
360 if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
361 COVERAGE_INC(packet_in_overflow);
362 free(CONST_CAST(void *, pin->up.packet));
367 /* Factory functions. */
370 init(const struct shash *iface_hints)
372 struct shash_node *node;
374 /* Make a local copy, since we don't own 'iface_hints' elements. */
375 SHASH_FOR_EACH(node, iface_hints) {
376 const struct iface_hint *orig_hint = node->data;
377 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
379 new_hint->br_name = xstrdup(orig_hint->br_name);
380 new_hint->br_type = xstrdup(orig_hint->br_type);
381 new_hint->ofp_port = orig_hint->ofp_port;
383 shash_add(&init_ofp_ports, node->name, new_hint);
388 enumerate_types(struct sset *types)
390 dp_enumerate_types(types);
394 enumerate_names(const char *type, struct sset *names)
396 struct ofproto_dpif *ofproto;
399 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
400 if (strcmp(type, ofproto->up.type)) {
403 sset_add(names, ofproto->up.name);
410 del(const char *type, const char *name)
415 error = dpif_open(name, type, &dpif);
417 error = dpif_delete(dpif);
424 port_open_type(const char *datapath_type, const char *port_type)
426 return dpif_port_open_type(datapath_type, port_type);
429 /* Type functions. */
431 static void process_dpif_port_changes(struct dpif_backer *);
432 static void process_dpif_all_ports_changed(struct dpif_backer *);
433 static void process_dpif_port_change(struct dpif_backer *,
434 const char *devname);
435 static void process_dpif_port_error(struct dpif_backer *, int error);
437 static struct ofproto_dpif *
438 lookup_ofproto_dpif_by_port_name(const char *name)
440 struct ofproto_dpif *ofproto;
442 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
443 if (sset_contains(&ofproto->ports, name)) {
452 type_run(const char *type)
454 struct dpif_backer *backer;
456 backer = shash_find_data(&all_dpif_backers, type);
458 /* This is not necessarily a problem, since backers are only
459 * created on demand. */
463 dpif_run(backer->dpif);
465 /* If vswitchd started with other_config:flow_restore_wait set as "true",
466 * and the configuration has now changed to "false", enable receiving
467 * packets from the datapath. */
468 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
471 backer->recv_set_enable = true;
473 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
475 VLOG_ERR("Failed to enable receiving packets in dpif.");
478 dpif_flow_flush(backer->dpif);
479 backer->need_revalidate = REV_RECONFIGURE;
482 if (backer->recv_set_enable) {
483 udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
486 if (backer->need_revalidate) {
487 struct ofproto_dpif *ofproto;
488 struct simap_node *node;
489 struct simap tmp_backers;
491 /* Handle tunnel garbage collection. */
492 simap_init(&tmp_backers);
493 simap_swap(&backer->tnl_backers, &tmp_backers);
495 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
496 struct ofport_dpif *iter;
498 if (backer != ofproto->backer) {
502 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
503 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
506 if (!iter->is_tunnel) {
510 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
511 namebuf, sizeof namebuf);
512 node = simap_find(&tmp_backers, dp_port);
514 simap_put(&backer->tnl_backers, dp_port, node->data);
515 simap_delete(&tmp_backers, node);
516 node = simap_find(&backer->tnl_backers, dp_port);
518 node = simap_find(&backer->tnl_backers, dp_port);
520 odp_port_t odp_port = ODPP_NONE;
522 if (!dpif_port_add(backer->dpif, iter->up.netdev,
524 simap_put(&backer->tnl_backers, dp_port,
525 odp_to_u32(odp_port));
526 node = simap_find(&backer->tnl_backers, dp_port);
531 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
532 if (tnl_port_reconfigure(iter, iter->up.netdev,
534 backer->need_revalidate = REV_RECONFIGURE;
539 SIMAP_FOR_EACH (node, &tmp_backers) {
540 dpif_port_del(backer->dpif, u32_to_odp(node->data));
542 simap_destroy(&tmp_backers);
544 switch (backer->need_revalidate) {
545 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
546 case REV_STP: COVERAGE_INC(rev_stp); break;
547 case REV_BOND: COVERAGE_INC(rev_bond); break;
548 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
549 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
550 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
552 backer->need_revalidate = 0;
554 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
555 struct ofport_dpif *ofport;
556 struct ofbundle *bundle;
558 if (ofproto->backer != backer) {
562 ovs_rwlock_wrlock(&xlate_rwlock);
563 xlate_ofproto_set(ofproto, ofproto->up.name,
564 ofproto->backer->dpif, ofproto->miss_rule,
565 ofproto->no_packet_in_rule, ofproto->ml,
566 ofproto->stp, ofproto->mbridge,
567 ofproto->sflow, ofproto->ipfix,
568 ofproto->netflow, ofproto->up.frag_handling,
569 ofproto->up.forward_bpdu,
570 connmgr_has_in_band(ofproto->up.connmgr),
571 ofproto->backer->variable_length_userdata,
572 ofproto->backer->max_mpls_depth);
574 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
575 xlate_bundle_set(ofproto, bundle, bundle->name,
576 bundle->vlan_mode, bundle->vlan,
577 bundle->trunks, bundle->use_priority_tags,
578 bundle->bond, bundle->lacp,
582 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
583 int stp_port = ofport->stp_port
584 ? stp_port_no(ofport->stp_port)
586 xlate_ofport_set(ofproto, ofport->bundle, ofport,
587 ofport->up.ofp_port, ofport->odp_port,
588 ofport->up.netdev, ofport->cfm,
589 ofport->bfd, ofport->peer, stp_port,
590 ofport->qdscp, ofport->n_qdscp,
591 ofport->up.pp.config, ofport->up.pp.state,
592 ofport->is_tunnel, ofport->may_enable);
594 ovs_rwlock_unlock(&xlate_rwlock);
597 udpif_revalidate(backer->udpif);
600 process_dpif_port_changes(backer);
605 /* Check for and handle port changes in 'backer''s dpif. */
607 process_dpif_port_changes(struct dpif_backer *backer)
613 error = dpif_port_poll(backer->dpif, &devname);
619 process_dpif_all_ports_changed(backer);
623 process_dpif_port_change(backer, devname);
628 process_dpif_port_error(backer, error);
635 process_dpif_all_ports_changed(struct dpif_backer *backer)
637 struct ofproto_dpif *ofproto;
638 struct dpif_port dpif_port;
639 struct dpif_port_dump dump;
640 struct sset devnames;
643 sset_init(&devnames);
644 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
645 if (ofproto->backer == backer) {
646 struct ofport *ofport;
648 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
649 sset_add(&devnames, netdev_get_name(ofport->netdev));
653 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
654 sset_add(&devnames, dpif_port.name);
657 SSET_FOR_EACH (devname, &devnames) {
658 process_dpif_port_change(backer, devname);
660 sset_destroy(&devnames);
664 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
666 struct ofproto_dpif *ofproto;
667 struct dpif_port port;
669 /* Don't report on the datapath's device. */
670 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
674 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
675 &all_ofproto_dpifs) {
676 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
681 ofproto = lookup_ofproto_dpif_by_port_name(devname);
682 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
683 /* The port was removed. If we know the datapath,
684 * report it through poll_set(). If we don't, it may be
685 * notifying us of a removal we initiated, so ignore it.
686 * If there's a pending ENOBUFS, let it stand, since
687 * everything will be reevaluated. */
688 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
689 sset_add(&ofproto->port_poll_set, devname);
690 ofproto->port_poll_errno = 0;
692 } else if (!ofproto) {
693 /* The port was added, but we don't know with which
694 * ofproto we should associate it. Delete it. */
695 dpif_port_del(backer->dpif, port.port_no);
697 struct ofport_dpif *ofport;
699 ofport = ofport_dpif_cast(shash_find_data(
700 &ofproto->up.port_by_name, devname));
702 && ofport->odp_port != port.port_no
703 && !odp_port_to_ofport(backer, port.port_no))
705 /* 'ofport''s datapath port number has changed from
706 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
707 * structures to match. */
708 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
709 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
710 ofport->odp_port = port.port_no;
711 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
712 hash_odp_port(port.port_no));
713 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
714 backer->need_revalidate = REV_RECONFIGURE;
717 dpif_port_destroy(&port);
720 /* Propagate 'error' to all ofprotos based on 'backer'. */
722 process_dpif_port_error(struct dpif_backer *backer, int error)
724 struct ofproto_dpif *ofproto;
726 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
727 if (ofproto->backer == backer) {
728 sset_clear(&ofproto->port_poll_set);
729 ofproto->port_poll_errno = error;
735 type_wait(const char *type)
737 struct dpif_backer *backer;
739 backer = shash_find_data(&all_dpif_backers, type);
741 /* This is not necessarily a problem, since backers are only
742 * created on demand. */
746 dpif_wait(backer->dpif);
749 /* Basic life-cycle. */
751 static int add_internal_flows(struct ofproto_dpif *);
753 static struct ofproto *
756 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
761 dealloc(struct ofproto *ofproto_)
763 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
768 close_dpif_backer(struct dpif_backer *backer)
770 ovs_assert(backer->refcount > 0);
772 if (--backer->refcount) {
776 udpif_destroy(backer->udpif);
778 simap_destroy(&backer->tnl_backers);
779 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
780 hmap_destroy(&backer->odp_to_ofport_map);
781 shash_find_and_delete(&all_dpif_backers, backer->type);
783 dpif_close(backer->dpif);
788 /* Datapath port slated for removal from datapath. */
790 struct list list_node;
794 static bool check_variable_length_userdata(struct dpif_backer *backer);
795 static size_t check_max_mpls_depth(struct dpif_backer *backer);
798 open_dpif_backer(const char *type, struct dpif_backer **backerp)
800 struct dpif_backer *backer;
801 struct dpif_port_dump port_dump;
802 struct dpif_port port;
803 struct shash_node *node;
804 struct list garbage_list;
805 struct odp_garbage *garbage, *next;
811 backer = shash_find_data(&all_dpif_backers, type);
818 backer_name = xasprintf("ovs-%s", type);
820 /* Remove any existing datapaths, since we assume we're the only
821 * userspace controlling the datapath. */
823 dp_enumerate_names(type, &names);
824 SSET_FOR_EACH(name, &names) {
825 struct dpif *old_dpif;
827 /* Don't remove our backer if it exists. */
828 if (!strcmp(name, backer_name)) {
832 if (dpif_open(name, type, &old_dpif)) {
833 VLOG_WARN("couldn't open old datapath %s to remove it", name);
835 dpif_delete(old_dpif);
836 dpif_close(old_dpif);
839 sset_destroy(&names);
841 backer = xmalloc(sizeof *backer);
843 error = dpif_create_and_open(backer_name, type, &backer->dpif);
846 VLOG_ERR("failed to open datapath of type %s: %s", type,
847 ovs_strerror(error));
851 backer->udpif = udpif_create(backer, backer->dpif);
853 backer->type = xstrdup(type);
854 backer->refcount = 1;
855 hmap_init(&backer->odp_to_ofport_map);
856 ovs_rwlock_init(&backer->odp_to_ofport_lock);
857 backer->need_revalidate = 0;
858 simap_init(&backer->tnl_backers);
859 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
862 if (backer->recv_set_enable) {
863 dpif_flow_flush(backer->dpif);
866 /* Loop through the ports already on the datapath and remove any
867 * that we don't need anymore. */
868 list_init(&garbage_list);
869 dpif_port_dump_start(&port_dump, backer->dpif);
870 while (dpif_port_dump_next(&port_dump, &port)) {
871 node = shash_find(&init_ofp_ports, port.name);
872 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
873 garbage = xmalloc(sizeof *garbage);
874 garbage->odp_port = port.port_no;
875 list_push_front(&garbage_list, &garbage->list_node);
878 dpif_port_dump_done(&port_dump);
880 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
881 dpif_port_del(backer->dpif, garbage->odp_port);
882 list_remove(&garbage->list_node);
886 shash_add(&all_dpif_backers, type, backer);
888 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
890 VLOG_ERR("failed to listen on datapath of type %s: %s",
891 type, ovs_strerror(error));
892 close_dpif_backer(backer);
895 backer->variable_length_userdata = check_variable_length_userdata(backer);
896 backer->max_mpls_depth = check_max_mpls_depth(backer);
898 if (backer->recv_set_enable) {
899 udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
905 /* Tests whether 'backer''s datapath supports variable-length
906 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions. We need
907 * to disable some features on older datapaths that don't support this
910 * Returns false if 'backer' definitely does not support variable-length
911 * userdata, true if it seems to support them or if at least the error we get
914 check_variable_length_userdata(struct dpif_backer *backer)
916 struct eth_header *eth;
917 struct ofpbuf actions;
918 struct dpif_execute execute;
919 struct ofpbuf packet;
923 /* Compose a userspace action that will cause an ERANGE error on older
924 * datapaths that don't support variable-length userdata.
926 * We really test for using userdata longer than 8 bytes, but older
927 * datapaths accepted these, silently truncating the userdata to 8 bytes.
928 * The same older datapaths rejected userdata shorter than 8 bytes, so we
929 * test for that instead as a proxy for longer userdata support. */
930 ofpbuf_init(&actions, 64);
931 start = nl_msg_start_nested(&actions, OVS_ACTION_ATTR_USERSPACE);
932 nl_msg_put_u32(&actions, OVS_USERSPACE_ATTR_PID,
933 dpif_port_get_pid(backer->dpif, ODPP_NONE, 0));
934 nl_msg_put_unspec_zero(&actions, OVS_USERSPACE_ATTR_USERDATA, 4);
935 nl_msg_end_nested(&actions, start);
937 /* Compose a dummy ethernet packet. */
938 ofpbuf_init(&packet, ETH_HEADER_LEN);
939 eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
940 eth->eth_type = htons(0x1234);
942 /* Execute the actions. On older datapaths this fails with ERANGE, on
943 * newer datapaths it succeeds. */
944 execute.actions = actions.data;
945 execute.actions_len = actions.size;
946 execute.packet = &packet;
947 execute.md = PKT_METADATA_INITIALIZER(0);
948 execute.needs_help = false;
950 error = dpif_execute(backer->dpif, &execute);
952 ofpbuf_uninit(&packet);
953 ofpbuf_uninit(&actions);
957 /* Variable-length userdata is supported.
959 * Purge received packets to avoid processing the nonsense packet we
960 * sent to userspace, then report success. */
961 dpif_recv_purge(backer->dpif);
965 /* Variable-length userdata is not supported. */
966 VLOG_WARN("%s: datapath does not support variable-length userdata "
967 "feature (needs Linux 3.10+ or kernel module from OVS "
968 "1..11+). The NXAST_SAMPLE action will be ignored.",
969 dpif_name(backer->dpif));
973 /* Something odd happened. We're not sure whether variable-length
974 * userdata is supported. Default to "yes". */
975 VLOG_WARN("%s: variable-length userdata feature probe failed (%s)",
976 dpif_name(backer->dpif), ovs_strerror(error));
981 /* Tests the MPLS label stack depth supported by 'backer''s datapath.
983 * Returns the number of elements in a struct flow's mpls_lse field
984 * if the datapath supports at least that many entries in an
986 * Otherwise returns the number of MPLS push actions supported by
989 check_max_mpls_depth(struct dpif_backer *backer)
994 for (n = 0; n < FLOW_MAX_MPLS_LABELS; n++) {
995 struct odputil_keybuf keybuf;
999 memset(&flow, 0, sizeof flow);
1000 flow.dl_type = htons(ETH_TYPE_MPLS);
1001 flow_set_mpls_bos(&flow, n, 1);
1003 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1004 odp_flow_key_from_flow(&key, &flow, 0);
1006 error = dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
1007 key.data, key.size, NULL, 0, NULL, 0, NULL);
1008 if (error && error != EEXIST) {
1009 if (error != EINVAL) {
1010 VLOG_WARN("%s: MPLS stack length feature probe failed (%s)",
1011 dpif_name(backer->dpif), ovs_strerror(error));
1016 error = dpif_flow_del(backer->dpif, key.data, key.size, NULL);
1018 VLOG_WARN("%s: failed to delete MPLS feature probe flow",
1019 dpif_name(backer->dpif));
1023 VLOG_INFO("%s: MPLS label stack length probed as %d",
1024 dpif_name(backer->dpif), n);
1029 construct(struct ofproto *ofproto_)
1031 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1032 struct shash_node *node, *next;
1035 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1040 ofproto->netflow = NULL;
1041 ofproto->sflow = NULL;
1042 ofproto->ipfix = NULL;
1043 ofproto->stp = NULL;
1044 ofproto->dump_seq = 0;
1045 hmap_init(&ofproto->bundles);
1046 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1047 ofproto->mbridge = mbridge_create();
1048 ofproto->has_bonded_bundles = false;
1049 ofproto->lacp_enabled = false;
1050 ovs_mutex_init_adaptive(&ofproto->stats_mutex);
1051 ovs_mutex_init(&ofproto->vsp_mutex);
1053 guarded_list_init(&ofproto->pins);
1055 ofproto_dpif_unixctl_init();
1057 hmap_init(&ofproto->vlandev_map);
1058 hmap_init(&ofproto->realdev_vid_map);
1060 sset_init(&ofproto->ports);
1061 sset_init(&ofproto->ghost_ports);
1062 sset_init(&ofproto->port_poll_set);
1063 ofproto->port_poll_errno = 0;
1064 ofproto->change_seq = 0;
1066 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1067 struct iface_hint *iface_hint = node->data;
1069 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1070 /* Check if the datapath already has this port. */
1071 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1072 sset_add(&ofproto->ports, node->name);
1075 free(iface_hint->br_name);
1076 free(iface_hint->br_type);
1078 shash_delete(&init_ofp_ports, node);
1082 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1083 hash_string(ofproto->up.name, 0));
1084 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1086 ofproto_init_tables(ofproto_, N_TABLES);
1087 error = add_internal_flows(ofproto);
1088 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1094 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1095 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1097 struct ofputil_flow_mod fm;
1098 struct classifier *cls;
1101 match_init_catchall(&fm.match);
1103 match_set_reg(&fm.match, 0, id);
1104 fm.new_cookie = htonll(0);
1105 fm.cookie = htonll(0);
1106 fm.cookie_mask = htonll(0);
1107 fm.modify_cookie = false;
1108 fm.table_id = TBL_INTERNAL;
1109 fm.command = OFPFC_ADD;
1110 fm.idle_timeout = 0;
1111 fm.hard_timeout = 0;
1115 fm.ofpacts = ofpacts->data;
1116 fm.ofpacts_len = ofpacts->size;
1118 error = ofproto_flow_mod(&ofproto->up, &fm);
1120 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1121 id, ofperr_to_string(error));
1125 cls = &ofproto->up.tables[TBL_INTERNAL].cls;
1126 fat_rwlock_rdlock(&cls->rwlock);
1127 *rulep = rule_dpif_cast(rule_from_cls_rule(
1128 classifier_lookup(cls, &fm.match.flow, NULL)));
1129 ovs_assert(*rulep != NULL);
1130 fat_rwlock_unlock(&cls->rwlock);
1136 add_internal_flows(struct ofproto_dpif *ofproto)
1138 struct ofpact_controller *controller;
1139 uint64_t ofpacts_stub[128 / 8];
1140 struct ofpbuf ofpacts;
1144 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1147 controller = ofpact_put_CONTROLLER(&ofpacts);
1148 controller->max_len = UINT16_MAX;
1149 controller->controller_id = 0;
1150 controller->reason = OFPR_NO_MATCH;
1151 ofpact_pad(&ofpacts);
1153 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1158 ofpbuf_clear(&ofpacts);
1159 error = add_internal_flow(ofproto, id++, &ofpacts,
1160 &ofproto->no_packet_in_rule);
1165 error = add_internal_flow(ofproto, id++, &ofpacts,
1166 &ofproto->drop_frags_rule);
1171 destruct(struct ofproto *ofproto_)
1173 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1174 struct rule_dpif *rule, *next_rule;
1175 struct ofproto_packet_in *pin, *next_pin;
1176 struct oftable *table;
1179 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1180 ovs_rwlock_wrlock(&xlate_rwlock);
1181 xlate_remove_ofproto(ofproto);
1182 ovs_rwlock_unlock(&xlate_rwlock);
1184 /* Ensure that the upcall processing threads have no remaining references
1185 * to the ofproto or anything in it. */
1186 udpif_synchronize(ofproto->backer->udpif);
1188 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1190 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1191 struct cls_cursor cursor;
1193 fat_rwlock_rdlock(&table->cls.rwlock);
1194 cls_cursor_init(&cursor, &table->cls, NULL);
1195 fat_rwlock_unlock(&table->cls.rwlock);
1196 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1197 ofproto_rule_delete(&ofproto->up, &rule->up);
1201 guarded_list_pop_all(&ofproto->pins, &pins);
1202 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1203 list_remove(&pin->list_node);
1204 free(CONST_CAST(void *, pin->up.packet));
1207 guarded_list_destroy(&ofproto->pins);
1209 mbridge_unref(ofproto->mbridge);
1211 netflow_unref(ofproto->netflow);
1212 dpif_sflow_unref(ofproto->sflow);
1213 hmap_destroy(&ofproto->bundles);
1214 mac_learning_unref(ofproto->ml);
1216 hmap_destroy(&ofproto->vlandev_map);
1217 hmap_destroy(&ofproto->realdev_vid_map);
1219 sset_destroy(&ofproto->ports);
1220 sset_destroy(&ofproto->ghost_ports);
1221 sset_destroy(&ofproto->port_poll_set);
1223 ovs_mutex_destroy(&ofproto->stats_mutex);
1224 ovs_mutex_destroy(&ofproto->vsp_mutex);
1226 close_dpif_backer(ofproto->backer);
1230 run(struct ofproto *ofproto_)
1232 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1233 uint64_t new_seq, new_dump_seq;
1235 if (mbridge_need_revalidate(ofproto->mbridge)) {
1236 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1237 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1238 mac_learning_flush(ofproto->ml);
1239 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1242 /* Do not perform any periodic activity required by 'ofproto' while
1243 * waiting for flow restore to complete. */
1244 if (!ofproto_get_flow_restore_wait()) {
1245 struct ofproto_packet_in *pin, *next_pin;
1248 guarded_list_pop_all(&ofproto->pins, &pins);
1249 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1250 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1251 list_remove(&pin->list_node);
1252 free(CONST_CAST(void *, pin->up.packet));
1257 if (ofproto->netflow) {
1258 netflow_run(ofproto->netflow);
1260 if (ofproto->sflow) {
1261 dpif_sflow_run(ofproto->sflow);
1263 if (ofproto->ipfix) {
1264 dpif_ipfix_run(ofproto->ipfix);
1267 new_seq = seq_read(connectivity_seq_get());
1268 if (ofproto->change_seq != new_seq) {
1269 struct ofport_dpif *ofport;
1271 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1275 ofproto->change_seq = new_seq;
1277 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1278 struct ofbundle *bundle;
1280 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1286 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1287 if (mac_learning_run(ofproto->ml)) {
1288 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1290 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1292 new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
1293 if (ofproto->dump_seq != new_dump_seq) {
1294 struct rule *rule, *next_rule;
1296 /* We know stats are relatively fresh, so now is a good time to do some
1298 ofproto->dump_seq = new_dump_seq;
1300 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1302 ovs_mutex_lock(&ofproto_mutex);
1303 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
1304 &ofproto->up.expirable) {
1305 rule_expire(rule_dpif_cast(rule));
1307 ovs_mutex_unlock(&ofproto_mutex);
1309 /* All outstanding data in existing flows has been accounted, so it's a
1310 * good time to do bond rebalancing. */
1311 if (ofproto->has_bonded_bundles) {
1312 struct ofbundle *bundle;
1314 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1316 bond_rebalance(bundle->bond);
1326 wait(struct ofproto *ofproto_)
1328 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1330 if (ofproto_get_flow_restore_wait()) {
1334 if (ofproto->sflow) {
1335 dpif_sflow_wait(ofproto->sflow);
1337 if (ofproto->ipfix) {
1338 dpif_ipfix_wait(ofproto->ipfix);
1340 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1341 struct ofbundle *bundle;
1343 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1344 bundle_wait(bundle);
1347 if (ofproto->netflow) {
1348 netflow_wait(ofproto->netflow);
1350 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1351 mac_learning_wait(ofproto->ml);
1352 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1354 if (ofproto->backer->need_revalidate) {
1355 /* Shouldn't happen, but if it does just go around again. */
1356 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1357 poll_immediate_wake();
1360 seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
1364 type_get_memory_usage(const char *type, struct simap *usage)
1366 struct dpif_backer *backer;
1368 backer = shash_find_data(&all_dpif_backers, type);
1370 udpif_get_memory_usage(backer->udpif, usage);
1375 flush(struct ofproto *ofproto_)
1377 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1378 struct dpif_backer *backer = ofproto->backer;
1381 udpif_flush(backer->udpif);
1386 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1387 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1389 *arp_match_ip = true;
1390 *actions = (OFPUTIL_A_OUTPUT |
1391 OFPUTIL_A_SET_VLAN_VID |
1392 OFPUTIL_A_SET_VLAN_PCP |
1393 OFPUTIL_A_STRIP_VLAN |
1394 OFPUTIL_A_SET_DL_SRC |
1395 OFPUTIL_A_SET_DL_DST |
1396 OFPUTIL_A_SET_NW_SRC |
1397 OFPUTIL_A_SET_NW_DST |
1398 OFPUTIL_A_SET_NW_TOS |
1399 OFPUTIL_A_SET_TP_SRC |
1400 OFPUTIL_A_SET_TP_DST |
1405 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1407 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1408 struct dpif_dp_stats s;
1409 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1413 strcpy(ots->name, "classifier");
1415 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1416 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes, &used);
1417 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes,
1419 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes,
1421 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1422 ots->lookup_count = htonll(n_lookup);
1423 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1426 static struct ofport *
1429 struct ofport_dpif *port = xmalloc(sizeof *port);
1434 port_dealloc(struct ofport *port_)
1436 struct ofport_dpif *port = ofport_dpif_cast(port_);
1441 port_construct(struct ofport *port_)
1443 struct ofport_dpif *port = ofport_dpif_cast(port_);
1444 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1445 const struct netdev *netdev = port->up.netdev;
1446 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1447 struct dpif_port dpif_port;
1450 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1451 port->bundle = NULL;
1454 port->may_enable = true;
1455 port->stp_port = NULL;
1456 port->stp_state = STP_DISABLED;
1457 port->is_tunnel = false;
1461 port->realdev_ofp_port = 0;
1462 port->vlandev_vid = 0;
1463 port->carrier_seq = netdev_get_carrier_resets(netdev);
1464 port->is_layer3 = netdev_vport_is_layer3(netdev);
1466 if (netdev_vport_is_patch(netdev)) {
1467 /* By bailing out here, we don't submit the port to the sFlow module
1468 * to be considered for counter polling export. This is correct
1469 * because the patch port represents an interface that sFlow considers
1470 * to be "internal" to the switch as a whole, and therefore not an
1471 * candidate for counter polling. */
1472 port->odp_port = ODPP_NONE;
1473 ofport_update_peer(port);
1477 error = dpif_port_query_by_name(ofproto->backer->dpif,
1478 netdev_vport_get_dpif_port(netdev, namebuf,
1485 port->odp_port = dpif_port.port_no;
1487 if (netdev_get_tunnel_config(netdev)) {
1488 tnl_port_add(port, port->up.netdev, port->odp_port);
1489 port->is_tunnel = true;
1491 /* Sanity-check that a mapping doesn't already exist. This
1492 * shouldn't happen for non-tunnel ports. */
1493 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1494 VLOG_ERR("port %s already has an OpenFlow port number",
1496 dpif_port_destroy(&dpif_port);
1500 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1501 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1502 hash_odp_port(port->odp_port));
1503 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1505 dpif_port_destroy(&dpif_port);
1507 if (ofproto->sflow) {
1508 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1515 port_destruct(struct ofport *port_)
1517 struct ofport_dpif *port = ofport_dpif_cast(port_);
1518 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1519 const char *devname = netdev_get_name(port->up.netdev);
1520 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1521 const char *dp_port_name;
1523 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1524 ovs_rwlock_wrlock(&xlate_rwlock);
1525 xlate_ofport_remove(port);
1526 ovs_rwlock_unlock(&xlate_rwlock);
1528 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1530 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1531 /* The underlying device is still there, so delete it. This
1532 * happens when the ofproto is being destroyed, since the caller
1533 * assumes that removal of attached ports will happen as part of
1535 if (!port->is_tunnel) {
1536 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1541 port->peer->peer = NULL;
1545 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1546 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1547 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1548 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1552 sset_find_and_delete(&ofproto->ports, devname);
1553 sset_find_and_delete(&ofproto->ghost_ports, devname);
1554 bundle_remove(port_);
1555 set_cfm(port_, NULL);
1556 set_bfd(port_, NULL);
1557 if (port->stp_port) {
1558 stp_port_disable(port->stp_port);
1560 if (ofproto->sflow) {
1561 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1568 port_modified(struct ofport *port_)
1570 struct ofport_dpif *port = ofport_dpif_cast(port_);
1572 if (port->bundle && port->bundle->bond) {
1573 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1577 cfm_set_netdev(port->cfm, port->up.netdev);
1581 bfd_set_netdev(port->bfd, port->up.netdev);
1584 ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1585 port->up.pp.hw_addr);
1587 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1589 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1593 ofport_update_peer(port);
1597 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1599 struct ofport_dpif *port = ofport_dpif_cast(port_);
1600 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1601 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1603 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1604 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1605 OFPUTIL_PC_NO_PACKET_IN)) {
1606 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1608 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1609 bundle_update(port->bundle);
1615 set_sflow(struct ofproto *ofproto_,
1616 const struct ofproto_sflow_options *sflow_options)
1618 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1619 struct dpif_sflow *ds = ofproto->sflow;
1621 if (sflow_options) {
1623 struct ofport_dpif *ofport;
1625 ds = ofproto->sflow = dpif_sflow_create();
1626 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1627 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1629 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1631 dpif_sflow_set_options(ds, sflow_options);
1634 dpif_sflow_unref(ds);
1635 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1636 ofproto->sflow = NULL;
1644 struct ofproto *ofproto_,
1645 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1646 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1647 size_t n_flow_exporters_options)
1649 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1650 struct dpif_ipfix *di = ofproto->ipfix;
1651 bool has_options = bridge_exporter_options || flow_exporters_options;
1653 if (has_options && !di) {
1654 di = ofproto->ipfix = dpif_ipfix_create();
1658 /* Call set_options in any case to cleanly flush the flow
1659 * caches in the last exporters that are to be destroyed. */
1660 dpif_ipfix_set_options(
1661 di, bridge_exporter_options, flow_exporters_options,
1662 n_flow_exporters_options);
1665 dpif_ipfix_unref(di);
1666 ofproto->ipfix = NULL;
1674 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1676 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1681 struct ofproto_dpif *ofproto;
1683 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1684 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1685 ofport->cfm = cfm_create(ofport->up.netdev);
1688 if (cfm_configure(ofport->cfm, s)) {
1695 cfm_unref(ofport->cfm);
1698 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1699 ofport->up.pp.hw_addr);
1704 get_cfm_status(const struct ofport *ofport_,
1705 struct ofproto_cfm_status *status)
1707 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1710 status->faults = cfm_get_fault(ofport->cfm);
1711 status->flap_count = cfm_get_flap_count(ofport->cfm);
1712 status->remote_opstate = cfm_get_opup(ofport->cfm);
1713 status->health = cfm_get_health(ofport->cfm);
1714 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1722 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1724 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1725 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1729 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1730 cfg, ofport->up.netdev);
1731 if (ofport->bfd != old) {
1732 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1734 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1735 ofport->up.pp.hw_addr);
1740 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1742 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1745 bfd_get_status(ofport->bfd, smap);
1752 /* Spanning Tree. */
1755 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1757 struct ofproto_dpif *ofproto = ofproto_;
1758 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1759 struct ofport_dpif *ofport;
1761 ofport = stp_port_get_aux(sp);
1763 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1764 ofproto->up.name, port_num);
1766 struct eth_header *eth = pkt->l2;
1768 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1769 if (eth_addr_is_zero(eth->eth_src)) {
1770 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1771 "with unknown MAC", ofproto->up.name, port_num);
1773 ofproto_dpif_send_packet(ofport, pkt);
1779 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1781 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1783 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1785 /* Only revalidate flows if the configuration changed. */
1786 if (!s != !ofproto->stp) {
1787 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1791 if (!ofproto->stp) {
1792 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1793 send_bpdu_cb, ofproto);
1794 ofproto->stp_last_tick = time_msec();
1797 stp_set_bridge_id(ofproto->stp, s->system_id);
1798 stp_set_bridge_priority(ofproto->stp, s->priority);
1799 stp_set_hello_time(ofproto->stp, s->hello_time);
1800 stp_set_max_age(ofproto->stp, s->max_age);
1801 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1803 struct ofport *ofport;
1805 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1806 set_stp_port(ofport, NULL);
1809 stp_unref(ofproto->stp);
1810 ofproto->stp = NULL;
1817 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1819 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1823 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1824 s->designated_root = stp_get_designated_root(ofproto->stp);
1825 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1834 update_stp_port_state(struct ofport_dpif *ofport)
1836 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1837 enum stp_state state;
1839 /* Figure out new state. */
1840 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1844 if (ofport->stp_state != state) {
1845 enum ofputil_port_state of_state;
1848 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1849 netdev_get_name(ofport->up.netdev),
1850 stp_state_name(ofport->stp_state),
1851 stp_state_name(state));
1852 if (stp_learn_in_state(ofport->stp_state)
1853 != stp_learn_in_state(state)) {
1854 /* xxx Learning action flows should also be flushed. */
1855 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1856 mac_learning_flush(ofproto->ml);
1857 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1859 fwd_change = stp_forward_in_state(ofport->stp_state)
1860 != stp_forward_in_state(state);
1862 ofproto->backer->need_revalidate = REV_STP;
1863 ofport->stp_state = state;
1864 ofport->stp_state_entered = time_msec();
1866 if (fwd_change && ofport->bundle) {
1867 bundle_update(ofport->bundle);
1870 /* Update the STP state bits in the OpenFlow port description. */
1871 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1872 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1873 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1874 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1875 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1877 ofproto_port_set_state(&ofport->up, of_state);
1881 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1882 * caller is responsible for assigning STP port numbers and ensuring
1883 * there are no duplicates. */
1885 set_stp_port(struct ofport *ofport_,
1886 const struct ofproto_port_stp_settings *s)
1888 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1889 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1890 struct stp_port *sp = ofport->stp_port;
1892 if (!s || !s->enable) {
1894 ofport->stp_port = NULL;
1895 stp_port_disable(sp);
1896 update_stp_port_state(ofport);
1899 } else if (sp && stp_port_no(sp) != s->port_num
1900 && ofport == stp_port_get_aux(sp)) {
1901 /* The port-id changed, so disable the old one if it's not
1902 * already in use by another port. */
1903 stp_port_disable(sp);
1906 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1907 stp_port_enable(sp);
1909 stp_port_set_aux(sp, ofport);
1910 stp_port_set_priority(sp, s->priority);
1911 stp_port_set_path_cost(sp, s->path_cost);
1913 update_stp_port_state(ofport);
1919 get_stp_port_status(struct ofport *ofport_,
1920 struct ofproto_port_stp_status *s)
1922 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1923 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1924 struct stp_port *sp = ofport->stp_port;
1926 if (!ofproto->stp || !sp) {
1932 s->port_id = stp_port_get_id(sp);
1933 s->state = stp_port_get_state(sp);
1934 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1935 s->role = stp_port_get_role(sp);
1941 get_stp_port_stats(struct ofport *ofport_,
1942 struct ofproto_port_stp_stats *s)
1944 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1945 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1946 struct stp_port *sp = ofport->stp_port;
1948 if (!ofproto->stp || !sp) {
1954 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1960 stp_run(struct ofproto_dpif *ofproto)
1963 long long int now = time_msec();
1964 long long int elapsed = now - ofproto->stp_last_tick;
1965 struct stp_port *sp;
1968 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1969 ofproto->stp_last_tick = now;
1971 while (stp_get_changed_port(ofproto->stp, &sp)) {
1972 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1975 update_stp_port_state(ofport);
1979 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1980 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1981 mac_learning_flush(ofproto->ml);
1982 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1988 stp_wait(struct ofproto_dpif *ofproto)
1991 poll_timer_wait(1000);
1996 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
1999 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2000 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2002 if (ofport->n_qdscp != n_qdscp
2003 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2004 n_qdscp * sizeof *qdscp))) {
2005 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2006 free(ofport->qdscp);
2007 ofport->qdscp = n_qdscp
2008 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2010 ofport->n_qdscp = n_qdscp;
2018 /* Expires all MAC learning entries associated with 'bundle' and forces its
2019 * ofproto to revalidate every flow.
2021 * Normally MAC learning entries are removed only from the ofproto associated
2022 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2023 * are removed from every ofproto. When patch ports and SLB bonds are in use
2024 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2025 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2026 * with the host from which it migrated. */
2028 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2030 struct ofproto_dpif *ofproto = bundle->ofproto;
2031 struct mac_learning *ml = ofproto->ml;
2032 struct mac_entry *mac, *next_mac;
2034 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2035 ovs_rwlock_wrlock(&ml->rwlock);
2036 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2037 if (mac->port.p == bundle) {
2039 struct ofproto_dpif *o;
2041 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2043 struct mac_entry *e;
2045 ovs_rwlock_wrlock(&o->ml->rwlock);
2046 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2048 mac_learning_expire(o->ml, e);
2050 ovs_rwlock_unlock(&o->ml->rwlock);
2055 mac_learning_expire(ml, mac);
2058 ovs_rwlock_unlock(&ml->rwlock);
2061 static struct ofbundle *
2062 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2064 struct ofbundle *bundle;
2066 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2067 &ofproto->bundles) {
2068 if (bundle->aux == aux) {
2076 bundle_update(struct ofbundle *bundle)
2078 struct ofport_dpif *port;
2080 bundle->floodable = true;
2081 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2082 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2084 || !stp_forward_in_state(port->stp_state)) {
2085 bundle->floodable = false;
2092 bundle_del_port(struct ofport_dpif *port)
2094 struct ofbundle *bundle = port->bundle;
2096 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2098 list_remove(&port->bundle_node);
2099 port->bundle = NULL;
2102 lacp_slave_unregister(bundle->lacp, port);
2105 bond_slave_unregister(bundle->bond, port);
2108 bundle_update(bundle);
2112 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2113 struct lacp_slave_settings *lacp)
2115 struct ofport_dpif *port;
2117 port = get_ofp_port(bundle->ofproto, ofp_port);
2122 if (port->bundle != bundle) {
2123 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2125 bundle_remove(&port->up);
2128 port->bundle = bundle;
2129 list_push_back(&bundle->ports, &port->bundle_node);
2130 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2132 || !stp_forward_in_state(port->stp_state)) {
2133 bundle->floodable = false;
2137 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2138 lacp_slave_register(bundle->lacp, port, lacp);
2145 bundle_destroy(struct ofbundle *bundle)
2147 struct ofproto_dpif *ofproto;
2148 struct ofport_dpif *port, *next_port;
2154 ofproto = bundle->ofproto;
2155 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2157 ovs_rwlock_wrlock(&xlate_rwlock);
2158 xlate_bundle_remove(bundle);
2159 ovs_rwlock_unlock(&xlate_rwlock);
2161 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2162 bundle_del_port(port);
2165 bundle_flush_macs(bundle, true);
2166 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2168 free(bundle->trunks);
2169 lacp_unref(bundle->lacp);
2170 bond_unref(bundle->bond);
2175 bundle_set(struct ofproto *ofproto_, void *aux,
2176 const struct ofproto_bundle_settings *s)
2178 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2179 bool need_flush = false;
2180 struct ofport_dpif *port;
2181 struct ofbundle *bundle;
2182 unsigned long *trunks;
2188 bundle_destroy(bundle_lookup(ofproto, aux));
2192 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2193 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2195 bundle = bundle_lookup(ofproto, aux);
2197 bundle = xmalloc(sizeof *bundle);
2199 bundle->ofproto = ofproto;
2200 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2201 hash_pointer(aux, 0));
2203 bundle->name = NULL;
2205 list_init(&bundle->ports);
2206 bundle->vlan_mode = PORT_VLAN_TRUNK;
2208 bundle->trunks = NULL;
2209 bundle->use_priority_tags = s->use_priority_tags;
2210 bundle->lacp = NULL;
2211 bundle->bond = NULL;
2213 bundle->floodable = true;
2214 mbridge_register_bundle(ofproto->mbridge, bundle);
2217 if (!bundle->name || strcmp(s->name, bundle->name)) {
2219 bundle->name = xstrdup(s->name);
2224 ofproto->lacp_enabled = true;
2225 if (!bundle->lacp) {
2226 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2227 bundle->lacp = lacp_create();
2229 lacp_configure(bundle->lacp, s->lacp);
2231 lacp_unref(bundle->lacp);
2232 bundle->lacp = NULL;
2235 /* Update set of ports. */
2237 for (i = 0; i < s->n_slaves; i++) {
2238 if (!bundle_add_port(bundle, s->slaves[i],
2239 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2243 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2244 struct ofport_dpif *next_port;
2246 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2247 for (i = 0; i < s->n_slaves; i++) {
2248 if (s->slaves[i] == port->up.ofp_port) {
2253 bundle_del_port(port);
2257 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2259 if (list_is_empty(&bundle->ports)) {
2260 bundle_destroy(bundle);
2264 /* Set VLAN tagging mode */
2265 if (s->vlan_mode != bundle->vlan_mode
2266 || s->use_priority_tags != bundle->use_priority_tags) {
2267 bundle->vlan_mode = s->vlan_mode;
2268 bundle->use_priority_tags = s->use_priority_tags;
2273 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2274 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2276 if (vlan != bundle->vlan) {
2277 bundle->vlan = vlan;
2281 /* Get trunked VLANs. */
2282 switch (s->vlan_mode) {
2283 case PORT_VLAN_ACCESS:
2287 case PORT_VLAN_TRUNK:
2288 trunks = CONST_CAST(unsigned long *, s->trunks);
2291 case PORT_VLAN_NATIVE_UNTAGGED:
2292 case PORT_VLAN_NATIVE_TAGGED:
2293 if (vlan != 0 && (!s->trunks
2294 || !bitmap_is_set(s->trunks, vlan)
2295 || bitmap_is_set(s->trunks, 0))) {
2296 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2298 trunks = bitmap_clone(s->trunks, 4096);
2300 trunks = bitmap_allocate1(4096);
2302 bitmap_set1(trunks, vlan);
2303 bitmap_set0(trunks, 0);
2305 trunks = CONST_CAST(unsigned long *, s->trunks);
2312 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2313 free(bundle->trunks);
2314 if (trunks == s->trunks) {
2315 bundle->trunks = vlan_bitmap_clone(trunks);
2317 bundle->trunks = trunks;
2322 if (trunks != s->trunks) {
2327 if (!list_is_short(&bundle->ports)) {
2328 bundle->ofproto->has_bonded_bundles = true;
2330 if (bond_reconfigure(bundle->bond, s->bond)) {
2331 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2334 bundle->bond = bond_create(s->bond);
2335 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2338 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2339 bond_slave_register(bundle->bond, port, port->up.netdev);
2342 bond_unref(bundle->bond);
2343 bundle->bond = NULL;
2346 /* If we changed something that would affect MAC learning, un-learn
2347 * everything on this port and force flow revalidation. */
2349 bundle_flush_macs(bundle, false);
2356 bundle_remove(struct ofport *port_)
2358 struct ofport_dpif *port = ofport_dpif_cast(port_);
2359 struct ofbundle *bundle = port->bundle;
2362 bundle_del_port(port);
2363 if (list_is_empty(&bundle->ports)) {
2364 bundle_destroy(bundle);
2365 } else if (list_is_short(&bundle->ports)) {
2366 bond_unref(bundle->bond);
2367 bundle->bond = NULL;
2373 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2375 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2376 struct ofport_dpif *port = port_;
2377 uint8_t ea[ETH_ADDR_LEN];
2380 error = netdev_get_etheraddr(port->up.netdev, ea);
2382 struct ofpbuf packet;
2385 ofpbuf_init(&packet, 0);
2386 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2388 memcpy(packet_pdu, pdu, pdu_size);
2390 ofproto_dpif_send_packet(port, &packet);
2391 ofpbuf_uninit(&packet);
2393 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2394 "%s (%s)", port->bundle->name,
2395 netdev_get_name(port->up.netdev), ovs_strerror(error));
2400 bundle_send_learning_packets(struct ofbundle *bundle)
2402 struct ofproto_dpif *ofproto = bundle->ofproto;
2403 struct ofpbuf *learning_packet;
2404 int error, n_packets, n_errors;
2405 struct mac_entry *e;
2406 struct list packets;
2408 list_init(&packets);
2409 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2410 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2411 if (e->port.p != bundle) {
2414 learning_packet = bond_compose_learning_packet(bundle->bond,
2417 learning_packet->private_p = port_void;
2418 list_push_back(&packets, &learning_packet->list_node);
2421 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2423 error = n_packets = n_errors = 0;
2424 LIST_FOR_EACH (learning_packet, list_node, &packets) {
2427 ret = ofproto_dpif_send_packet(learning_packet->private_p, learning_packet);
2434 ofpbuf_list_delete(&packets);
2437 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2438 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2439 "packets, last error was: %s",
2440 bundle->name, n_errors, n_packets, ovs_strerror(error));
2442 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2443 bundle->name, n_packets);
2448 bundle_run(struct ofbundle *bundle)
2451 lacp_run(bundle->lacp, send_pdu_cb);
2454 struct ofport_dpif *port;
2456 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2457 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2460 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2461 bundle->ofproto->backer->need_revalidate = REV_BOND;
2464 if (bond_should_send_learning_packets(bundle->bond)) {
2465 bundle_send_learning_packets(bundle);
2471 bundle_wait(struct ofbundle *bundle)
2474 lacp_wait(bundle->lacp);
2477 bond_wait(bundle->bond);
2484 mirror_set__(struct ofproto *ofproto_, void *aux,
2485 const struct ofproto_mirror_settings *s)
2487 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2488 struct ofbundle **srcs, **dsts;
2493 mirror_destroy(ofproto->mbridge, aux);
2497 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2498 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2500 for (i = 0; i < s->n_srcs; i++) {
2501 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2504 for (i = 0; i < s->n_dsts; i++) {
2505 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2508 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2509 s->n_dsts, s->src_vlans,
2510 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2517 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2518 uint64_t *packets, uint64_t *bytes)
2520 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2525 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2527 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2528 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2529 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2530 mac_learning_flush(ofproto->ml);
2532 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2537 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2539 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2540 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2541 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2545 forward_bpdu_changed(struct ofproto *ofproto_)
2547 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2548 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2552 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2555 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2556 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2557 mac_learning_set_idle_time(ofproto->ml, idle_time);
2558 mac_learning_set_max_entries(ofproto->ml, max_entries);
2559 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2564 static struct ofport_dpif *
2565 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2567 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2568 return ofport ? ofport_dpif_cast(ofport) : NULL;
2572 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2573 struct ofproto_port *ofproto_port,
2574 struct dpif_port *dpif_port)
2576 ofproto_port->name = dpif_port->name;
2577 ofproto_port->type = dpif_port->type;
2578 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2582 ofport_update_peer(struct ofport_dpif *ofport)
2584 const struct ofproto_dpif *ofproto;
2585 struct dpif_backer *backer;
2588 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2592 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2593 backer->need_revalidate = REV_RECONFIGURE;
2596 ofport->peer->peer = NULL;
2597 ofport->peer = NULL;
2600 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2605 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2606 struct ofport *peer_ofport;
2607 struct ofport_dpif *peer;
2610 if (ofproto->backer != backer) {
2614 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2619 peer = ofport_dpif_cast(peer_ofport);
2620 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2621 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2623 ofport->peer = peer;
2624 ofport->peer->peer = ofport;
2634 port_run(struct ofport_dpif *ofport)
2636 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2637 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2638 bool enable = netdev_get_carrier(ofport->up.netdev);
2639 bool cfm_enable = false;
2640 bool bfd_enable = false;
2642 ofport->carrier_seq = carrier_seq;
2645 int cfm_opup = cfm_get_opup(ofport->cfm);
2647 cfm_enable = !cfm_get_fault(ofport->cfm);
2649 if (cfm_opup >= 0) {
2650 cfm_enable = cfm_enable && cfm_opup;
2655 bfd_enable = bfd_forwarding(ofport->bfd);
2658 if (ofport->bfd || ofport->cfm) {
2659 enable = enable && (cfm_enable || bfd_enable);
2662 if (ofport->bundle) {
2663 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2664 if (carrier_changed) {
2665 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2669 if (ofport->may_enable != enable) {
2670 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2671 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2674 ofport->may_enable = enable;
2678 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2679 struct ofproto_port *ofproto_port)
2681 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2682 struct dpif_port dpif_port;
2685 if (sset_contains(&ofproto->ghost_ports, devname)) {
2686 const char *type = netdev_get_type_from_name(devname);
2688 /* We may be called before ofproto->up.port_by_name is populated with
2689 * the appropriate ofport. For this reason, we must get the name and
2690 * type from the netdev layer directly. */
2692 const struct ofport *ofport;
2694 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2695 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2696 ofproto_port->name = xstrdup(devname);
2697 ofproto_port->type = xstrdup(type);
2703 if (!sset_contains(&ofproto->ports, devname)) {
2706 error = dpif_port_query_by_name(ofproto->backer->dpif,
2707 devname, &dpif_port);
2709 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2715 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2717 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2718 const char *devname = netdev_get_name(netdev);
2719 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2720 const char *dp_port_name;
2722 if (netdev_vport_is_patch(netdev)) {
2723 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2727 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2728 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2729 odp_port_t port_no = ODPP_NONE;
2732 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2736 if (netdev_get_tunnel_config(netdev)) {
2737 simap_put(&ofproto->backer->tnl_backers,
2738 dp_port_name, odp_to_u32(port_no));
2742 if (netdev_get_tunnel_config(netdev)) {
2743 sset_add(&ofproto->ghost_ports, devname);
2745 sset_add(&ofproto->ports, devname);
2751 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
2753 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2754 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2761 sset_find_and_delete(&ofproto->ghost_ports,
2762 netdev_get_name(ofport->up.netdev));
2763 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2764 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
2765 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
2767 /* The caller is going to close ofport->up.netdev. If this is a
2768 * bonded port, then the bond is using that netdev, so remove it
2769 * from the bond. The client will need to reconfigure everything
2770 * after deleting ports, so then the slave will get re-added. */
2771 bundle_remove(&ofport->up);
2778 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2780 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2783 error = netdev_get_stats(ofport->up.netdev, stats);
2785 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
2786 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2788 ovs_mutex_lock(&ofproto->stats_mutex);
2789 /* ofproto->stats.tx_packets represents packets that we created
2790 * internally and sent to some port (e.g. packets sent with
2791 * ofproto_dpif_send_packet()). Account for them as if they had
2792 * come from OFPP_LOCAL and got forwarded. */
2794 if (stats->rx_packets != UINT64_MAX) {
2795 stats->rx_packets += ofproto->stats.tx_packets;
2798 if (stats->rx_bytes != UINT64_MAX) {
2799 stats->rx_bytes += ofproto->stats.tx_bytes;
2802 /* ofproto->stats.rx_packets represents packets that were received on
2803 * some port and we processed internally and dropped (e.g. STP).
2804 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2806 if (stats->tx_packets != UINT64_MAX) {
2807 stats->tx_packets += ofproto->stats.rx_packets;
2810 if (stats->tx_bytes != UINT64_MAX) {
2811 stats->tx_bytes += ofproto->stats.rx_bytes;
2813 ovs_mutex_unlock(&ofproto->stats_mutex);
2819 struct port_dump_state {
2824 struct ofproto_port port;
2829 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
2831 *statep = xzalloc(sizeof(struct port_dump_state));
2836 port_dump_next(const struct ofproto *ofproto_, void *state_,
2837 struct ofproto_port *port)
2839 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2840 struct port_dump_state *state = state_;
2841 const struct sset *sset;
2842 struct sset_node *node;
2844 if (state->has_port) {
2845 ofproto_port_destroy(&state->port);
2846 state->has_port = false;
2848 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
2849 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
2852 error = port_query_by_name(ofproto_, node->name, &state->port);
2854 *port = state->port;
2855 state->has_port = true;
2857 } else if (error != ENODEV) {
2862 if (!state->ghost) {
2863 state->ghost = true;
2866 return port_dump_next(ofproto_, state_, port);
2873 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2875 struct port_dump_state *state = state_;
2877 if (state->has_port) {
2878 ofproto_port_destroy(&state->port);
2885 port_poll(const struct ofproto *ofproto_, char **devnamep)
2887 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2889 if (ofproto->port_poll_errno) {
2890 int error = ofproto->port_poll_errno;
2891 ofproto->port_poll_errno = 0;
2895 if (sset_is_empty(&ofproto->port_poll_set)) {
2899 *devnamep = sset_pop(&ofproto->port_poll_set);
2904 port_poll_wait(const struct ofproto *ofproto_)
2906 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2907 dpif_port_poll_wait(ofproto->backer->dpif);
2911 port_is_lacp_current(const struct ofport *ofport_)
2913 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2914 return (ofport->bundle && ofport->bundle->lacp
2915 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2919 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
2920 * then delete it entirely. */
2922 rule_expire(struct rule_dpif *rule)
2923 OVS_REQUIRES(ofproto_mutex)
2925 uint16_t hard_timeout, idle_timeout;
2926 long long int now = time_msec();
2929 ovs_assert(!rule->up.pending);
2931 hard_timeout = rule->up.hard_timeout;
2932 idle_timeout = rule->up.idle_timeout;
2934 /* Has 'rule' expired? */
2936 long long int modified;
2938 ovs_mutex_lock(&rule->up.mutex);
2939 modified = rule->up.modified;
2940 ovs_mutex_unlock(&rule->up.mutex);
2942 if (now > modified + hard_timeout * 1000) {
2943 reason = OFPRR_HARD_TIMEOUT;
2947 if (reason < 0 && idle_timeout) {
2950 ovs_mutex_lock(&rule->stats_mutex);
2951 used = rule->stats.used;
2952 ovs_mutex_unlock(&rule->stats_mutex);
2954 if (now > used + idle_timeout * 1000) {
2955 reason = OFPRR_IDLE_TIMEOUT;
2960 COVERAGE_INC(ofproto_dpif_expired);
2961 ofproto_rule_expire(&rule->up, reason);
2965 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
2966 * 'flow' must reflect the data in 'packet'. */
2968 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
2969 const struct flow *flow,
2970 struct rule_dpif *rule,
2971 const struct ofpact *ofpacts, size_t ofpacts_len,
2972 struct ofpbuf *packet)
2974 struct dpif_flow_stats stats;
2975 struct xlate_out xout;
2976 struct xlate_in xin;
2978 struct dpif_execute execute;
2981 ovs_assert((rule != NULL) != (ofpacts != NULL));
2983 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
2985 rule_dpif_credit_stats(rule, &stats);
2988 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
2989 xin.ofpacts = ofpacts;
2990 xin.ofpacts_len = ofpacts_len;
2991 xin.resubmit_stats = &stats;
2992 xlate_actions(&xin, &xout);
2994 in_port = flow->in_port.ofp_port;
2995 if (in_port == OFPP_NONE) {
2996 in_port = OFPP_LOCAL;
2998 execute.actions = xout.odp_actions.data;
2999 execute.actions_len = xout.odp_actions.size;
3000 execute.packet = packet;
3001 execute.md.tunnel = flow->tunnel;
3002 execute.md.skb_priority = flow->skb_priority;
3003 execute.md.pkt_mark = flow->pkt_mark;
3004 execute.md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
3005 execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
3007 error = dpif_execute(ofproto->backer->dpif, &execute);
3009 xlate_out_uninit(&xout);
3015 rule_dpif_credit_stats(struct rule_dpif *rule,
3016 const struct dpif_flow_stats *stats)
3018 ovs_mutex_lock(&rule->stats_mutex);
3019 rule->stats.n_packets += stats->n_packets;
3020 rule->stats.n_bytes += stats->n_bytes;
3021 rule->stats.used = MAX(rule->stats.used, stats->used);
3022 ovs_mutex_unlock(&rule->stats_mutex);
3026 rule_dpif_is_fail_open(const struct rule_dpif *rule)
3028 return is_fail_open_rule(&rule->up);
3032 rule_dpif_is_table_miss(const struct rule_dpif *rule)
3034 return rule_is_table_miss(&rule->up);
3038 rule_dpif_is_internal(const struct rule_dpif *rule)
3040 return rule_is_internal(&rule->up);
3044 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
3045 OVS_REQUIRES(rule->up.mutex)
3047 return rule->up.flow_cookie;
3051 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
3052 uint16_t hard_timeout)
3054 ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
3057 /* Returns 'rule''s actions. The caller owns a reference on the returned
3058 * actions and must eventually release it (with rule_actions_unref()) to avoid
3060 struct rule_actions *
3061 rule_dpif_get_actions(const struct rule_dpif *rule)
3063 return rule_get_actions(&rule->up);
3066 /* Lookup 'flow' in table 0 of 'ofproto''s classifier.
3067 * If 'wc' is non-null, sets the fields that were relevant as part of
3068 * the lookup. Returns the table_id where a match or miss occurred.
3070 * The return value will be zero unless there was a miss and
3071 * OFPTC_TABLE_MISS_CONTINUE is in effect for the sequence of tables
3072 * where misses occur. */
3074 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
3075 struct flow_wildcards *wc, struct rule_dpif **rule)
3077 enum rule_dpif_lookup_verdict verdict;
3078 enum ofputil_port_config config = 0;
3079 uint8_t table_id = 0;
3081 verdict = rule_dpif_lookup_from_table(ofproto, flow, wc, true,
3085 case RULE_DPIF_LOOKUP_VERDICT_MATCH:
3087 case RULE_DPIF_LOOKUP_VERDICT_CONTROLLER: {
3088 struct ofport_dpif *port;
3090 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
3092 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
3093 flow->in_port.ofp_port);
3095 config = port ? port->up.pp.config : 0;
3098 case RULE_DPIF_LOOKUP_VERDICT_DROP:
3099 config = OFPUTIL_PC_NO_PACKET_IN;
3105 choose_miss_rule(config, ofproto->miss_rule,
3106 ofproto->no_packet_in_rule, rule);
3110 static struct rule_dpif *
3111 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, uint8_t table_id,
3112 const struct flow *flow, struct flow_wildcards *wc)
3114 struct classifier *cls = &ofproto->up.tables[table_id].cls;
3115 const struct cls_rule *cls_rule;
3116 struct rule_dpif *rule;
3118 fat_rwlock_rdlock(&cls->rwlock);
3119 if (ofproto->up.frag_handling != OFPC_FRAG_NX_MATCH) {
3121 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
3122 if (is_ip_any(flow)) {
3123 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
3127 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
3128 if (ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3129 /* We must pretend that transport ports are unavailable. */
3130 struct flow ofpc_normal_flow = *flow;
3131 ofpc_normal_flow.tp_src = htons(0);
3132 ofpc_normal_flow.tp_dst = htons(0);
3133 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
3135 /* Must be OFPC_FRAG_DROP (we don't have OFPC_FRAG_REASM). */
3136 cls_rule = &ofproto->drop_frags_rule->up.cr;
3139 cls_rule = classifier_lookup(cls, flow, wc);
3142 cls_rule = classifier_lookup(cls, flow, wc);
3145 rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
3146 rule_dpif_ref(rule);
3147 fat_rwlock_unlock(&cls->rwlock);
3152 /* Look up 'flow' in 'ofproto''s classifier starting from table '*table_id'.
3153 * Stores the rule that was found in '*rule', or NULL if none was found.
3154 * Updates 'wc', if nonnull, to reflect the fields that were used during the
3157 * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3158 * if none is found then the table miss configuration for that table is
3159 * honored, which can result in additional lookups in other OpenFlow tables.
3160 * In this case the function updates '*table_id' to reflect the final OpenFlow
3161 * table that was searched.
3163 * If 'honor_table_miss' is false, then only one table lookup occurs, in
3168 * - RULE_DPIF_LOOKUP_VERDICT_MATCH if a rule (in '*rule') was found.
3170 * - RULE_DPIF_LOOKUP_VERDICT_DROP if no rule was found and a table miss
3171 * configuration specified that the packet should be dropped in this
3172 * case. (This occurs only if 'honor_table_miss' is true, because only in
3173 * this case does the table miss configuration matter.)
3175 * - RULE_DPIF_LOOKUP_VERDICT_CONTROLLER if no rule was found otherwise. */
3176 enum rule_dpif_lookup_verdict
3177 rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto,
3178 const struct flow *flow,
3179 struct flow_wildcards *wc,
3180 bool honor_table_miss,
3181 uint8_t *table_id, struct rule_dpif **rule)
3185 for (next_id = *table_id;
3186 next_id < ofproto->up.n_tables;
3187 next_id++, next_id += (next_id == TBL_INTERNAL))
3189 *table_id = next_id;
3190 *rule = rule_dpif_lookup_in_table(ofproto, *table_id, flow, wc);
3192 return RULE_DPIF_LOOKUP_VERDICT_MATCH;
3193 } else if (!honor_table_miss) {
3194 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3196 switch (table_get_config(&ofproto->up, *table_id)
3197 & OFPTC11_TABLE_MISS_MASK) {
3198 case OFPTC11_TABLE_MISS_CONTINUE:
3201 case OFPTC11_TABLE_MISS_CONTROLLER:
3202 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3204 case OFPTC11_TABLE_MISS_DROP:
3205 return RULE_DPIF_LOOKUP_VERDICT_DROP;
3210 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3213 /* Given a port configuration (specified as zero if there's no port), chooses
3214 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
3215 * flow table miss. */
3217 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
3218 struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
3220 *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
3221 rule_dpif_ref(*rule);
3225 rule_dpif_ref(struct rule_dpif *rule)
3228 ofproto_rule_ref(&rule->up);
3233 rule_dpif_unref(struct rule_dpif *rule)
3236 ofproto_rule_unref(&rule->up);
3241 complete_operation(struct rule_dpif *rule)
3242 OVS_REQUIRES(ofproto_mutex)
3244 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3246 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3247 ofoperation_complete(rule->up.pending, 0);
3250 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
3252 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
3255 static struct rule *
3258 struct rule_dpif *rule = xmalloc(sizeof *rule);
3263 rule_dealloc(struct rule *rule_)
3265 struct rule_dpif *rule = rule_dpif_cast(rule_);
3270 rule_construct(struct rule *rule_)
3271 OVS_NO_THREAD_SAFETY_ANALYSIS
3273 struct rule_dpif *rule = rule_dpif_cast(rule_);
3274 ovs_mutex_init_adaptive(&rule->stats_mutex);
3275 rule->stats.n_packets = 0;
3276 rule->stats.n_bytes = 0;
3277 rule->stats.used = rule->up.modified;
3282 rule_insert(struct rule *rule_)
3283 OVS_REQUIRES(ofproto_mutex)
3285 struct rule_dpif *rule = rule_dpif_cast(rule_);
3286 complete_operation(rule);
3290 rule_delete(struct rule *rule_)
3291 OVS_REQUIRES(ofproto_mutex)
3293 struct rule_dpif *rule = rule_dpif_cast(rule_);
3294 complete_operation(rule);
3298 rule_destruct(struct rule *rule_)
3300 struct rule_dpif *rule = rule_dpif_cast(rule_);
3301 ovs_mutex_destroy(&rule->stats_mutex);
3305 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
3306 long long int *used)
3308 struct rule_dpif *rule = rule_dpif_cast(rule_);
3310 ovs_mutex_lock(&rule->stats_mutex);
3311 *packets = rule->stats.n_packets;
3312 *bytes = rule->stats.n_bytes;
3313 *used = rule->stats.used;
3314 ovs_mutex_unlock(&rule->stats_mutex);
3318 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
3319 struct ofpbuf *packet)
3321 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3323 ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
3327 rule_execute(struct rule *rule, const struct flow *flow,
3328 struct ofpbuf *packet)
3330 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
3331 ofpbuf_delete(packet);
3336 rule_modify_actions(struct rule *rule_, bool reset_counters)
3337 OVS_REQUIRES(ofproto_mutex)
3339 struct rule_dpif *rule = rule_dpif_cast(rule_);
3341 if (reset_counters) {
3342 ovs_mutex_lock(&rule->stats_mutex);
3343 rule->stats.n_packets = 0;
3344 rule->stats.n_bytes = 0;
3345 ovs_mutex_unlock(&rule->stats_mutex);
3348 complete_operation(rule);
3351 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
3353 return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
3356 static struct ofgroup *
3359 struct group_dpif *group = xzalloc(sizeof *group);
3364 group_dealloc(struct ofgroup *group_)
3366 struct group_dpif *group = group_dpif_cast(group_);
3371 group_construct_stats(struct group_dpif *group)
3372 OVS_REQUIRES(group->stats_mutex)
3374 group->packet_count = 0;
3375 group->byte_count = 0;
3376 if (!group->bucket_stats) {
3377 group->bucket_stats = xcalloc(group->up.n_buckets,
3378 sizeof *group->bucket_stats);
3380 memset(group->bucket_stats, 0, group->up.n_buckets *
3381 sizeof *group->bucket_stats);
3386 group_construct(struct ofgroup *group_)
3388 struct group_dpif *group = group_dpif_cast(group_);
3389 const struct ofputil_bucket *bucket;
3391 /* Prevent group chaining because our locking structure makes it hard to
3392 * implement deadlock-free. (See xlate_group_resource_check().) */
3393 LIST_FOR_EACH (bucket, list_node, &group->up.buckets) {
3394 const struct ofpact *a;
3396 OFPACT_FOR_EACH (a, bucket->ofpacts, bucket->ofpacts_len) {
3397 if (a->type == OFPACT_GROUP) {
3398 return OFPERR_OFPGMFC_CHAINING_UNSUPPORTED;
3403 ovs_mutex_init_adaptive(&group->stats_mutex);
3404 ovs_mutex_lock(&group->stats_mutex);
3405 group_construct_stats(group);
3406 ovs_mutex_unlock(&group->stats_mutex);
3411 group_destruct__(struct group_dpif *group)
3412 OVS_REQUIRES(group->stats_mutex)
3414 free(group->bucket_stats);
3415 group->bucket_stats = NULL;
3419 group_destruct(struct ofgroup *group_)
3421 struct group_dpif *group = group_dpif_cast(group_);
3422 ovs_mutex_lock(&group->stats_mutex);
3423 group_destruct__(group);
3424 ovs_mutex_unlock(&group->stats_mutex);
3425 ovs_mutex_destroy(&group->stats_mutex);
3429 group_modify(struct ofgroup *group_, struct ofgroup *victim_)
3431 struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
3432 struct group_dpif *group = group_dpif_cast(group_);
3433 struct group_dpif *victim = group_dpif_cast(victim_);
3435 ovs_mutex_lock(&group->stats_mutex);
3436 if (victim->up.n_buckets < group->up.n_buckets) {
3437 group_destruct__(group);
3439 group_construct_stats(group);
3440 ovs_mutex_unlock(&group->stats_mutex);
3442 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3448 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
3450 struct group_dpif *group = group_dpif_cast(group_);
3452 ovs_mutex_lock(&group->stats_mutex);
3453 ogs->packet_count = group->packet_count;
3454 ogs->byte_count = group->byte_count;
3455 memcpy(ogs->bucket_stats, group->bucket_stats,
3456 group->up.n_buckets * sizeof *group->bucket_stats);
3457 ovs_mutex_unlock(&group->stats_mutex);
3463 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
3464 struct group_dpif **group)
3465 OVS_TRY_RDLOCK(true, (*group)->up.rwlock)
3467 struct ofgroup *ofgroup;
3471 found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
3472 *group = found ? group_dpif_cast(ofgroup) : NULL;
3478 group_dpif_release(struct group_dpif *group)
3479 OVS_RELEASES(group->up.rwlock)
3481 ofproto_group_release(&group->up);
3485 group_dpif_get_buckets(const struct group_dpif *group,
3486 const struct list **buckets)
3488 *buckets = &group->up.buckets;
3491 enum ofp11_group_type
3492 group_dpif_get_type(const struct group_dpif *group)
3494 return group->up.type;
3497 /* Sends 'packet' out 'ofport'.
3498 * May modify 'packet'.
3499 * Returns 0 if successful, otherwise a positive errno value. */
3501 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
3503 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3506 error = xlate_send_packet(ofport, packet);
3508 ovs_mutex_lock(&ofproto->stats_mutex);
3509 ofproto->stats.tx_packets++;
3510 ofproto->stats.tx_bytes += packet->size;
3511 ovs_mutex_unlock(&ofproto->stats_mutex);
3516 set_frag_handling(struct ofproto *ofproto_,
3517 enum ofp_config_flags frag_handling)
3519 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3520 if (frag_handling != OFPC_FRAG_REASM) {
3521 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3529 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
3530 const struct flow *flow,
3531 const struct ofpact *ofpacts, size_t ofpacts_len)
3533 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3535 ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
3536 ofpacts_len, packet);
3543 set_netflow(struct ofproto *ofproto_,
3544 const struct netflow_options *netflow_options)
3546 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3548 if (netflow_options) {
3549 if (!ofproto->netflow) {
3550 ofproto->netflow = netflow_create();
3551 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3553 return netflow_set_options(ofproto->netflow, netflow_options);
3554 } else if (ofproto->netflow) {
3555 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3556 netflow_unref(ofproto->netflow);
3557 ofproto->netflow = NULL;
3564 get_netflow_ids(const struct ofproto *ofproto_,
3565 uint8_t *engine_type, uint8_t *engine_id)
3567 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3569 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
3572 static struct ofproto_dpif *
3573 ofproto_dpif_lookup(const char *name)
3575 struct ofproto_dpif *ofproto;
3577 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
3578 hash_string(name, 0), &all_ofproto_dpifs) {
3579 if (!strcmp(ofproto->up.name, name)) {
3587 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
3588 const char *argv[], void *aux OVS_UNUSED)
3590 struct ofproto_dpif *ofproto;
3593 ofproto = ofproto_dpif_lookup(argv[1]);
3595 unixctl_command_reply_error(conn, "no such bridge");
3598 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3599 mac_learning_flush(ofproto->ml);
3600 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3602 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3603 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3604 mac_learning_flush(ofproto->ml);
3605 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3609 unixctl_command_reply(conn, "table successfully flushed");
3612 static struct ofport_dpif *
3613 ofbundle_get_a_port(const struct ofbundle *bundle)
3615 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
3620 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
3621 const char *argv[], void *aux OVS_UNUSED)
3623 struct ds ds = DS_EMPTY_INITIALIZER;
3624 const struct ofproto_dpif *ofproto;
3625 const struct mac_entry *e;
3627 ofproto = ofproto_dpif_lookup(argv[1]);
3629 unixctl_command_reply_error(conn, "no such bridge");
3633 ds_put_cstr(&ds, " port VLAN MAC Age\n");
3634 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
3635 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3636 struct ofbundle *bundle = e->port.p;
3637 char name[OFP_MAX_PORT_NAME_LEN];
3639 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
3641 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
3642 name, e->vlan, ETH_ADDR_ARGS(e->mac),
3643 mac_entry_age(ofproto->ml, e));
3645 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3646 unixctl_command_reply(conn, ds_cstr(&ds));
3651 struct xlate_out xout;
3652 struct xlate_in xin;
3654 struct flow_wildcards wc;
3659 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
3661 struct rule_actions *actions;
3664 ds_put_char_multiple(result, '\t', level);
3666 ds_put_cstr(result, "No match\n");
3670 ovs_mutex_lock(&rule->up.mutex);
3671 cookie = rule->up.flow_cookie;
3672 ovs_mutex_unlock(&rule->up.mutex);
3674 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
3675 rule ? rule->up.table_id : 0, ntohll(cookie));
3676 cls_rule_format(&rule->up.cr, result);
3677 ds_put_char(result, '\n');
3679 actions = rule_dpif_get_actions(rule);
3681 ds_put_char_multiple(result, '\t', level);
3682 ds_put_cstr(result, "OpenFlow actions=");
3683 ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
3684 ds_put_char(result, '\n');
3688 trace_format_flow(struct ds *result, int level, const char *title,
3689 struct trace_ctx *trace)
3691 ds_put_char_multiple(result, '\t', level);
3692 ds_put_format(result, "%s: ", title);
3693 if (flow_equal(&trace->xin.flow, &trace->flow)) {
3694 ds_put_cstr(result, "unchanged");
3696 flow_format(result, &trace->xin.flow);
3697 trace->flow = trace->xin.flow;
3699 ds_put_char(result, '\n');
3703 trace_format_regs(struct ds *result, int level, const char *title,
3704 struct trace_ctx *trace)
3708 ds_put_char_multiple(result, '\t', level);
3709 ds_put_format(result, "%s:", title);
3710 for (i = 0; i < FLOW_N_REGS; i++) {
3711 ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
3713 ds_put_char(result, '\n');
3717 trace_format_odp(struct ds *result, int level, const char *title,
3718 struct trace_ctx *trace)
3720 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
3722 ds_put_char_multiple(result, '\t', level);
3723 ds_put_format(result, "%s: ", title);
3724 format_odp_actions(result, odp_actions->data, odp_actions->size);
3725 ds_put_char(result, '\n');
3729 trace_format_megaflow(struct ds *result, int level, const char *title,
3730 struct trace_ctx *trace)
3734 ds_put_char_multiple(result, '\t', level);
3735 ds_put_format(result, "%s: ", title);
3736 flow_wildcards_or(&trace->wc, &trace->xout.wc, &trace->wc);
3737 match_init(&match, &trace->flow, &trace->wc);
3738 match_format(&match, result, OFP_DEFAULT_PRIORITY);
3739 ds_put_char(result, '\n');
3743 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
3745 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
3746 struct ds *result = trace->result;
3748 ds_put_char(result, '\n');
3749 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
3750 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
3751 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
3752 trace_format_megaflow(result, recurse + 1, "Resubmitted megaflow", trace);
3753 trace_format_rule(result, recurse + 1, rule);
3757 trace_report(struct xlate_in *xin, const char *s, int recurse)
3759 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
3760 struct ds *result = trace->result;
3762 ds_put_char_multiple(result, '\t', recurse);
3763 ds_put_cstr(result, s);
3764 ds_put_char(result, '\n');
3767 /* Parses the 'argc' elements of 'argv', ignoring argv[0]. The following
3768 * forms are supported:
3770 * - [dpname] odp_flow [-generate | packet]
3771 * - bridge br_flow [-generate | packet]
3773 * On success, initializes '*ofprotop' and 'flow' and returns NULL. On failure
3774 * returns a nonnull malloced error message. */
3775 static char * WARN_UNUSED_RESULT
3776 parse_flow_and_packet(int argc, const char *argv[],
3777 struct ofproto_dpif **ofprotop, struct flow *flow,
3778 struct ofpbuf **packetp)
3780 const struct dpif_backer *backer = NULL;
3781 const char *error = NULL;
3783 struct simap port_names = SIMAP_INITIALIZER(&port_names);
3784 struct ofpbuf *packet;
3785 struct ofpbuf odp_key;
3786 struct ofpbuf odp_mask;
3788 ofpbuf_init(&odp_key, 0);
3789 ofpbuf_init(&odp_mask, 0);
3791 /* Handle "-generate" or a hex string as the last argument. */
3792 if (!strcmp(argv[argc - 1], "-generate")) {
3793 packet = ofpbuf_new(0);
3796 error = eth_from_hex(argv[argc - 1], &packet);
3799 } else if (argc == 4) {
3800 /* The 3-argument form must end in "-generate' or a hex string. */
3806 /* odp_flow can have its in_port specified as a name instead of port no.
3807 * We do not yet know whether a given flow is a odp_flow or a br_flow.
3808 * But, to know whether a flow is odp_flow through odp_flow_from_string(),
3809 * we need to create a simap of name to port no. */
3811 const char *dp_type;
3812 if (!strncmp(argv[1], "ovs-", 4)) {
3813 dp_type = argv[1] + 4;
3817 backer = shash_find_data(&all_dpif_backers, dp_type);
3818 } else if (argc == 2) {
3819 struct shash_node *node;
3820 if (shash_count(&all_dpif_backers) == 1) {
3821 node = shash_first(&all_dpif_backers);
3822 backer = node->data;
3825 error = "Syntax error";
3828 if (backer && backer->dpif) {
3829 struct dpif_port dpif_port;
3830 struct dpif_port_dump port_dump;
3831 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
3832 simap_put(&port_names, dpif_port.name,
3833 odp_to_u32(dpif_port.port_no));
3837 /* Parse the flow and determine whether a datapath or
3838 * bridge is specified. If function odp_flow_key_from_string()
3839 * returns 0, the flow is a odp_flow. If function
3840 * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
3841 if (!odp_flow_from_string(argv[argc - 1], &port_names,
3842 &odp_key, &odp_mask)) {
3844 error = "Cannot find the datapath";
3848 if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, flow,
3849 ofprotop, NULL, NULL, NULL, NULL)) {
3850 error = "Invalid datapath flow";
3854 char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
3857 m_err = xasprintf("Bad flow syntax: %s", err);
3862 error = "Must specify bridge name";
3866 *ofprotop = ofproto_dpif_lookup(argv[1]);
3868 error = "Unknown bridge name";
3874 /* Generate a packet, if requested. */
3876 if (!packet->size) {
3877 flow_compose(packet, flow);
3879 struct pkt_metadata md = pkt_metadata_from_flow(flow);
3881 /* Use the metadata from the flow and the packet argument
3882 * to reconstruct the flow. */
3883 flow_extract(packet, &md, flow);
3888 if (error && !m_err) {
3889 m_err = xstrdup(error);
3892 ofpbuf_delete(packet);
3896 ofpbuf_uninit(&odp_key);
3897 ofpbuf_uninit(&odp_mask);
3898 simap_destroy(&port_names);
3903 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
3904 void *aux OVS_UNUSED)
3906 struct ofproto_dpif *ofproto;
3907 struct ofpbuf *packet;
3911 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
3916 ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
3917 unixctl_command_reply(conn, ds_cstr(&result));
3918 ds_destroy(&result);
3919 ofpbuf_delete(packet);
3921 unixctl_command_reply_error(conn, error);
3927 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
3928 const char *argv[], void *aux OVS_UNUSED)
3930 enum ofputil_protocol usable_protocols;
3931 struct ofproto_dpif *ofproto;
3932 bool enforce_consistency;
3933 struct ofpbuf ofpacts;
3934 struct ofpbuf *packet;
3939 /* Three kinds of error return values! */
3945 ofpbuf_init(&ofpacts, 0);
3947 /* Parse actions. */
3948 error = parse_ofpacts(argv[--argc], &ofpacts, &usable_protocols);
3950 unixctl_command_reply_error(conn, error);
3955 /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
3956 * consistency between the flow and the actions. With -consistent, we
3957 * enforce consistency even for a flow supported in OpenFlow 1.0. */
3958 if (!strcmp(argv[1], "-consistent")) {
3959 enforce_consistency = true;
3963 enforce_consistency = false;
3966 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
3968 unixctl_command_reply_error(conn, error);
3973 /* Do the same checks as handle_packet_out() in ofproto.c.
3975 * We pass a 'table_id' of 0 to ofproto_check_ofpacts(), which isn't
3976 * strictly correct because these actions aren't in any table, but it's OK
3977 * because it 'table_id' is used only to check goto_table instructions, but
3978 * packet-outs take a list of actions and therefore it can't include
3981 * We skip the "meter" check here because meter is an instruction, not an
3982 * action, and thus cannot appear in ofpacts. */
3983 in_port = ofp_to_u16(flow.in_port.ofp_port);
3984 if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
3985 unixctl_command_reply_error(conn, "invalid in_port");
3988 if (enforce_consistency) {
3989 retval = ofpacts_check_consistency(ofpacts.data, ofpacts.size, &flow,
3990 u16_to_ofp(ofproto->up.max_ports),
3991 0, 0, usable_protocols);
3993 retval = ofpacts_check(ofpacts.data, ofpacts.size, &flow,
3994 u16_to_ofp(ofproto->up.max_ports), 0, 0,
4000 ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
4001 unixctl_command_reply_error(conn, ds_cstr(&result));
4005 ofproto_trace(ofproto, &flow, packet, ofpacts.data, ofpacts.size, &result);
4006 unixctl_command_reply(conn, ds_cstr(&result));
4009 ds_destroy(&result);
4010 ofpbuf_delete(packet);
4011 ofpbuf_uninit(&ofpacts);
4014 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
4015 * description of the results to 'ds'.
4017 * The trace follows a packet with the specified 'flow' through the flow
4018 * table. 'packet' may be nonnull to trace an actual packet, with consequent
4019 * side effects (if it is nonnull then its flow must be 'flow').
4021 * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
4022 * trace, otherwise the actions are determined by a flow table lookup. */
4024 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
4025 const struct ofpbuf *packet,
4026 const struct ofpact ofpacts[], size_t ofpacts_len,
4029 struct rule_dpif *rule;
4030 struct trace_ctx trace;
4032 ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
4033 ds_put_cstr(ds, "Flow: ");
4034 flow_format(ds, flow);
4035 ds_put_char(ds, '\n');
4037 flow_wildcards_init_catchall(&trace.wc);
4041 rule_dpif_lookup(ofproto, flow, &trace.wc, &rule);
4043 trace_format_rule(ds, 0, rule);
4044 if (rule == ofproto->miss_rule) {
4045 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
4046 } else if (rule == ofproto->no_packet_in_rule) {
4047 ds_put_cstr(ds, "\nNo match, packets dropped because "
4048 "OFPPC_NO_PACKET_IN is set on in_port.\n");
4049 } else if (rule == ofproto->drop_frags_rule) {
4050 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
4051 "and the fragment handling mode is \"drop\".\n");
4055 if (rule || ofpacts) {
4058 xlate_in_init(&trace.xin, ofproto, flow, rule, ntohs(flow->tcp_flags),
4061 trace.xin.ofpacts = ofpacts;
4062 trace.xin.ofpacts_len = ofpacts_len;
4064 trace.xin.resubmit_hook = trace_resubmit;
4065 trace.xin.report_hook = trace_report;
4067 xlate_actions(&trace.xin, &trace.xout);
4069 ds_put_char(ds, '\n');
4070 trace_format_flow(ds, 0, "Final flow", &trace);
4071 trace_format_megaflow(ds, 0, "Megaflow", &trace);
4073 ds_put_cstr(ds, "Datapath actions: ");
4074 format_odp_actions(ds, trace.xout.odp_actions.data,
4075 trace.xout.odp_actions.size);
4077 if (trace.xout.slow) {
4078 enum slow_path_reason slow;
4080 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
4081 "slow path because it:");
4083 slow = trace.xout.slow;
4085 enum slow_path_reason bit = rightmost_1bit(slow);
4087 ds_put_format(ds, "\n\t- %s.",
4088 slow_path_reason_to_explanation(bit));
4094 xlate_out_uninit(&trace.xout);
4097 rule_dpif_unref(rule);
4100 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
4101 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
4102 * to destroy 'ofproto_shash' and free the returned value. */
4103 static const struct shash_node **
4104 get_ofprotos(struct shash *ofproto_shash)
4106 const struct ofproto_dpif *ofproto;
4108 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4109 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
4110 shash_add_nocopy(ofproto_shash, name, ofproto);
4113 return shash_sort(ofproto_shash);
4117 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
4118 const char *argv[] OVS_UNUSED,
4119 void *aux OVS_UNUSED)
4121 struct ds ds = DS_EMPTY_INITIALIZER;
4122 struct shash ofproto_shash;
4123 const struct shash_node **sorted_ofprotos;
4126 shash_init(&ofproto_shash);
4127 sorted_ofprotos = get_ofprotos(&ofproto_shash);
4128 for (i = 0; i < shash_count(&ofproto_shash); i++) {
4129 const struct shash_node *node = sorted_ofprotos[i];
4130 ds_put_format(&ds, "%s\n", node->name);
4133 shash_destroy(&ofproto_shash);
4134 free(sorted_ofprotos);
4136 unixctl_command_reply(conn, ds_cstr(&ds));
4141 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
4143 const struct shash_node **ofprotos;
4144 struct dpif_dp_stats dp_stats;
4145 struct shash ofproto_shash;
4148 dpif_get_dp_stats(backer->dpif, &dp_stats);
4150 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
4151 dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
4153 shash_init(&ofproto_shash);
4154 ofprotos = get_ofprotos(&ofproto_shash);
4155 for (i = 0; i < shash_count(&ofproto_shash); i++) {
4156 struct ofproto_dpif *ofproto = ofprotos[i]->data;
4157 const struct shash_node **ports;
4160 if (ofproto->backer != backer) {
4164 ds_put_format(ds, "\t%s:\n", ofproto->up.name);
4166 ports = shash_sort(&ofproto->up.port_by_name);
4167 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
4168 const struct shash_node *node = ports[j];
4169 struct ofport *ofport = node->data;
4171 odp_port_t odp_port;
4173 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
4176 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
4177 if (odp_port != ODPP_NONE) {
4178 ds_put_format(ds, "%"PRIu32":", odp_port);
4180 ds_put_cstr(ds, "none:");
4183 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
4186 if (!netdev_get_config(ofport->netdev, &config)) {
4187 const struct smap_node **nodes;
4190 nodes = smap_sort(&config);
4191 for (i = 0; i < smap_count(&config); i++) {
4192 const struct smap_node *node = nodes[i];
4193 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
4194 node->key, node->value);
4198 smap_destroy(&config);
4200 ds_put_char(ds, ')');
4201 ds_put_char(ds, '\n');
4205 shash_destroy(&ofproto_shash);
4210 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4211 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4213 struct ds ds = DS_EMPTY_INITIALIZER;
4214 const struct shash_node **backers;
4217 backers = shash_sort(&all_dpif_backers);
4218 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
4219 dpif_show_backer(backers[i]->data, &ds);
4223 unixctl_command_reply(conn, ds_cstr(&ds));
4228 ofproto_dpif_contains_flow(const struct ofproto_dpif *ofproto,
4229 const struct nlattr *key, size_t key_len)
4231 struct ofproto_dpif *ofp;
4234 xlate_receive(ofproto->backer, NULL, key, key_len, &flow, &ofp,
4235 NULL, NULL, NULL, NULL);
4236 return ofp == ofproto;
4240 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
4241 int argc OVS_UNUSED, const char *argv[],
4242 void *aux OVS_UNUSED)
4244 struct ds ds = DS_EMPTY_INITIALIZER;
4245 const struct dpif_flow_stats *stats;
4246 const struct ofproto_dpif *ofproto;
4247 struct dpif_flow_dump flow_dump;
4248 const struct nlattr *actions;
4249 const struct nlattr *mask;
4250 const struct nlattr *key;
4254 bool verbosity = false;
4255 struct dpif_port dpif_port;
4256 struct dpif_port_dump port_dump;
4257 struct hmap portno_names;
4261 ofproto = ofproto_dpif_lookup(argv[argc - 1]);
4263 unixctl_command_reply_error(conn, "no such bridge");
4267 if (argc > 2 && !strcmp(argv[1], "-m")) {
4271 hmap_init(&portno_names);
4272 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
4273 odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
4277 error = dpif_flow_dump_start(&flow_dump, ofproto->backer->dpif);
4281 dpif_flow_dump_state_init(ofproto->backer->dpif, &state);
4282 while (dpif_flow_dump_next(&flow_dump, state, &key, &key_len,
4283 &mask, &mask_len, &actions, &actions_len,
4285 if (!ofproto_dpif_contains_flow(ofproto, key, key_len)) {
4289 odp_flow_format(key, key_len, mask, mask_len, &portno_names, &ds,
4291 ds_put_cstr(&ds, ", ");
4292 dpif_flow_stats_format(stats, &ds);
4293 ds_put_cstr(&ds, ", actions:");
4294 format_odp_actions(&ds, actions, actions_len);
4295 ds_put_char(&ds, '\n');
4297 dpif_flow_dump_state_uninit(ofproto->backer->dpif, state);
4298 error = dpif_flow_dump_done(&flow_dump);
4303 ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
4304 unixctl_command_reply_error(conn, ds_cstr(&ds));
4306 unixctl_command_reply(conn, ds_cstr(&ds));
4308 odp_portno_names_destroy(&portno_names);
4309 hmap_destroy(&portno_names);
4314 ofproto_dpif_unixctl_init(void)
4316 static bool registered;
4322 unixctl_command_register(
4324 "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
4325 1, 3, ofproto_unixctl_trace, NULL);
4326 unixctl_command_register(
4327 "ofproto/trace-packet-out",
4328 "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
4329 2, 6, ofproto_unixctl_trace_actions, NULL);
4330 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
4331 ofproto_unixctl_fdb_flush, NULL);
4332 unixctl_command_register("fdb/show", "bridge", 1, 1,
4333 ofproto_unixctl_fdb_show, NULL);
4334 unixctl_command_register("dpif/dump-dps", "", 0, 0,
4335 ofproto_unixctl_dpif_dump_dps, NULL);
4336 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
4338 unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
4339 ofproto_unixctl_dpif_dump_flows, NULL);
4343 /* Returns true if 'rule' is an internal rule, false otherwise. */
4345 rule_is_internal(const struct rule *rule)
4347 return rule->table_id == TBL_INTERNAL;
4350 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4352 * This is deprecated. It is only for compatibility with broken device drivers
4353 * in old versions of Linux that do not properly support VLANs when VLAN
4354 * devices are not used. When broken device drivers are no longer in
4355 * widespread use, we will delete these interfaces. */
4358 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
4360 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
4361 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
4363 if (realdev_ofp_port == ofport->realdev_ofp_port
4364 && vid == ofport->vlandev_vid) {
4368 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4370 if (ofport->realdev_ofp_port) {
4373 if (realdev_ofp_port && ofport->bundle) {
4374 /* vlandevs are enslaved to their realdevs, so they are not allowed to
4375 * themselves be part of a bundle. */
4376 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
4379 ofport->realdev_ofp_port = realdev_ofp_port;
4380 ofport->vlandev_vid = vid;
4382 if (realdev_ofp_port) {
4383 vsp_add(ofport, realdev_ofp_port, vid);
4390 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
4392 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
4396 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
4397 OVS_EXCLUDED(ofproto->vsp_mutex)
4399 /* hmap_is_empty is thread safe. */
4400 return !hmap_is_empty(&ofproto->realdev_vid_map);
4404 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
4405 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
4406 OVS_REQUIRES(ofproto->vsp_mutex)
4408 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
4409 int vid = vlan_tci_to_vid(vlan_tci);
4410 const struct vlan_splinter *vsp;
4412 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
4413 hash_realdev_vid(realdev_ofp_port, vid),
4414 &ofproto->realdev_vid_map) {
4415 if (vsp->realdev_ofp_port == realdev_ofp_port
4416 && vsp->vid == vid) {
4417 return vsp->vlandev_ofp_port;
4421 return realdev_ofp_port;
4424 /* Returns the OFP port number of the Linux VLAN device that corresponds to
4425 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
4426 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
4427 * 'vlan_tci' 9, it would return the port number of eth0.9.
4429 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
4430 * function just returns its 'realdev_ofp_port' argument. */
4432 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
4433 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
4434 OVS_EXCLUDED(ofproto->vsp_mutex)
4438 /* hmap_is_empty is thread safe, see if we can return immediately. */
4439 if (hmap_is_empty(&ofproto->realdev_vid_map)) {
4440 return realdev_ofp_port;
4442 ovs_mutex_lock(&ofproto->vsp_mutex);
4443 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
4444 ovs_mutex_unlock(&ofproto->vsp_mutex);
4448 static struct vlan_splinter *
4449 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
4451 struct vlan_splinter *vsp;
4453 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
4454 hash_ofp_port(vlandev_ofp_port),
4455 &ofproto->vlandev_map) {
4456 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
4464 /* Returns the OpenFlow port number of the "real" device underlying the Linux
4465 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
4466 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
4467 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
4468 * eth0 and store 9 in '*vid'.
4470 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
4471 * VLAN device. Unless VLAN splinters are enabled, this is what this function
4474 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
4475 ofp_port_t vlandev_ofp_port, int *vid)
4476 OVS_REQUIRES(ofproto->vsp_mutex)
4478 if (!hmap_is_empty(&ofproto->vlandev_map)) {
4479 const struct vlan_splinter *vsp;
4481 vsp = vlandev_find(ofproto, vlandev_ofp_port);
4486 return vsp->realdev_ofp_port;
4492 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
4493 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
4494 * 'flow->in_port' to the "real" device backing the VLAN device, sets
4495 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
4496 * always the case unless VLAN splinters are enabled), returns false without
4497 * making any changes. */
4499 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
4500 OVS_EXCLUDED(ofproto->vsp_mutex)
4505 /* hmap_is_empty is thread safe. */
4506 if (hmap_is_empty(&ofproto->vlandev_map)) {
4510 ovs_mutex_lock(&ofproto->vsp_mutex);
4511 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
4512 ovs_mutex_unlock(&ofproto->vsp_mutex);
4517 /* Cause the flow to be processed as if it came in on the real device with
4518 * the VLAN device's VLAN ID. */
4519 flow->in_port.ofp_port = realdev;
4520 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
4525 vsp_remove(struct ofport_dpif *port)
4527 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
4528 struct vlan_splinter *vsp;
4530 ovs_mutex_lock(&ofproto->vsp_mutex);
4531 vsp = vlandev_find(ofproto, port->up.ofp_port);
4533 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
4534 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
4537 port->realdev_ofp_port = 0;
4539 VLOG_ERR("missing vlan device record");
4541 ovs_mutex_unlock(&ofproto->vsp_mutex);
4545 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
4547 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
4549 ovs_mutex_lock(&ofproto->vsp_mutex);
4550 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
4551 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
4552 == realdev_ofp_port)) {
4553 struct vlan_splinter *vsp;
4555 vsp = xmalloc(sizeof *vsp);
4556 vsp->realdev_ofp_port = realdev_ofp_port;
4557 vsp->vlandev_ofp_port = port->up.ofp_port;
4560 port->realdev_ofp_port = realdev_ofp_port;
4562 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
4563 hash_ofp_port(port->up.ofp_port));
4564 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
4565 hash_realdev_vid(realdev_ofp_port, vid));
4567 VLOG_ERR("duplicate vlan device record");
4569 ovs_mutex_unlock(&ofproto->vsp_mutex);
4573 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
4575 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
4576 return ofport ? ofport->odp_port : ODPP_NONE;
4579 struct ofport_dpif *
4580 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
4582 struct ofport_dpif *port;
4584 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
4585 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
4586 &backer->odp_to_ofport_map) {
4587 if (port->odp_port == odp_port) {
4588 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
4593 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
4598 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
4600 struct ofport_dpif *port;
4602 port = odp_port_to_ofport(ofproto->backer, odp_port);
4603 if (port && &ofproto->up == port->up.ofproto) {
4604 return port->up.ofp_port;
4610 const struct ofproto_class ofproto_dpif_class = {
4624 NULL, /* get_memory_usage. */
4625 type_get_memory_usage,
4644 port_is_lacp_current,
4645 NULL, /* rule_choose_table */
4654 rule_modify_actions,
4668 get_stp_port_status,
4676 is_mirror_output_bundle,
4677 forward_bpdu_changed,
4678 set_mac_table_config,
4680 NULL, /* meter_get_features */
4681 NULL, /* meter_set */
4682 NULL, /* meter_get */
4683 NULL, /* meter_del */
4684 group_alloc, /* group_alloc */
4685 group_construct, /* group_construct */
4686 group_destruct, /* group_destruct */
4687 group_dealloc, /* group_dealloc */
4688 group_modify, /* group_modify */
4689 group_get_stats, /* group_get_stats */