2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-dpif.h"
20 #include "ofproto/ofproto-provider.h"
27 #include "byte-order.h"
28 #include "connectivity.h"
33 #include "dynamic-string.h"
34 #include "fail-open.h"
35 #include "guarded-list.h"
39 #include "mac-learning.h"
40 #include "meta-flow.h"
41 #include "multipath.h"
42 #include "netdev-vport.h"
47 #include "odp-execute.h"
50 #include "ofp-actions.h"
51 #include "ofp-parse.h"
52 #include "ofp-print.h"
53 #include "ofproto-dpif-ipfix.h"
54 #include "ofproto-dpif-mirror.h"
55 #include "ofproto-dpif-monitor.h"
56 #include "ofproto-dpif-rid.h"
57 #include "ofproto-dpif-sflow.h"
58 #include "ofproto-dpif-upcall.h"
59 #include "ofproto-dpif-xlate.h"
60 #include "poll-loop.h"
66 #include "unaligned.h"
68 #include "vlan-bitmap.h"
71 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
73 COVERAGE_DEFINE(ofproto_dpif_expired);
74 COVERAGE_DEFINE(packet_in_overflow);
76 /* No bfd/cfm status change. */
77 #define NO_STATUS_CHANGE -1
86 * - Do include packets and bytes from datapath flows which have not
87 * recently been processed by a revalidator. */
88 struct ovs_mutex stats_mutex;
89 struct dpif_flow_stats stats OVS_GUARDED;
92 /* RULE_CAST() depends on this. */
93 BUILD_ASSERT_DECL(offsetof(struct rule_dpif, up) == 0);
95 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes,
97 static struct rule_dpif *rule_dpif_cast(const struct rule *);
98 static void rule_expire(struct rule_dpif *);
105 * - Do include packets and bytes from datapath flows which have not
106 * recently been processed by a revalidator. */
107 struct ovs_mutex stats_mutex;
108 uint64_t packet_count OVS_GUARDED; /* Number of packets received. */
109 uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */
110 struct bucket_counter *bucket_stats OVS_GUARDED; /* Bucket statistics. */
114 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
115 struct ofproto_dpif *ofproto; /* Owning ofproto. */
116 void *aux; /* Key supplied by ofproto's client. */
117 char *name; /* Identifier for log messages. */
120 struct list ports; /* Contains "struct ofport"s. */
121 enum port_vlan_mode vlan_mode; /* VLAN mode */
122 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
123 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
124 * NULL if all VLANs are trunked. */
125 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
126 struct bond *bond; /* Nonnull iff more than one port. */
127 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
130 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
133 static void bundle_remove(struct ofport *);
134 static void bundle_update(struct ofbundle *);
135 static void bundle_destroy(struct ofbundle *);
136 static void bundle_del_port(struct ofport_dpif *);
137 static void bundle_run(struct ofbundle *);
138 static void bundle_wait(struct ofbundle *);
140 static void stp_run(struct ofproto_dpif *ofproto);
141 static void stp_wait(struct ofproto_dpif *ofproto);
142 static int set_stp_port(struct ofport *,
143 const struct ofproto_port_stp_settings *);
146 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
150 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
151 struct list bundle_node; /* In struct ofbundle's "ports" list. */
152 struct cfm *cfm; /* Connectivity Fault Management, if any. */
153 struct bfd *bfd; /* BFD, if any. */
154 bool may_enable; /* May be enabled in bonds. */
155 bool is_tunnel; /* This port is a tunnel. */
156 bool is_layer3; /* This is a layer 3 port. */
157 long long int carrier_seq; /* Carrier status changes. */
158 struct ofport_dpif *peer; /* Peer if patch port. */
161 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
162 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
163 long long int stp_state_entered;
165 /* Queue to DSCP mapping. */
166 struct ofproto_port_queue *qdscp;
169 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
171 * This is deprecated. It is only for compatibility with broken device
172 * drivers in old versions of Linux that do not properly support VLANs when
173 * VLAN devices are not used. When broken device drivers are no longer in
174 * widespread use, we will delete these interfaces. */
175 ofp_port_t realdev_ofp_port;
179 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
181 * This is deprecated. It is only for compatibility with broken device drivers
182 * in old versions of Linux that do not properly support VLANs when VLAN
183 * devices are not used. When broken device drivers are no longer in
184 * widespread use, we will delete these interfaces. */
185 struct vlan_splinter {
186 struct hmap_node realdev_vid_node;
187 struct hmap_node vlandev_node;
188 ofp_port_t realdev_ofp_port;
189 ofp_port_t vlandev_ofp_port;
193 static void vsp_remove(struct ofport_dpif *);
194 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
196 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
199 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
202 static struct ofport_dpif *
203 ofport_dpif_cast(const struct ofport *ofport)
205 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
208 static void port_run(struct ofport_dpif *);
209 static int set_bfd(struct ofport *, const struct smap *);
210 static int set_cfm(struct ofport *, const struct cfm_settings *);
211 static void ofport_update_peer(struct ofport_dpif *);
213 struct dpif_completion {
214 struct list list_node;
215 struct ofoperation *op;
218 /* Reasons that we might need to revalidate every datapath flow, and
219 * corresponding coverage counters.
221 * A value of 0 means that there is no need to revalidate.
223 * It would be nice to have some cleaner way to integrate with coverage
224 * counters, but with only a few reasons I guess this is good enough for
226 enum revalidate_reason {
227 REV_RECONFIGURE = 1, /* Switch configuration changed. */
228 REV_STP, /* Spanning tree protocol port status change. */
229 REV_BOND, /* Bonding changed. */
230 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
231 REV_FLOW_TABLE, /* Flow table changed. */
232 REV_MAC_LEARNING, /* Mac learning changed. */
234 COVERAGE_DEFINE(rev_reconfigure);
235 COVERAGE_DEFINE(rev_stp);
236 COVERAGE_DEFINE(rev_bond);
237 COVERAGE_DEFINE(rev_port_toggled);
238 COVERAGE_DEFINE(rev_flow_table);
239 COVERAGE_DEFINE(rev_mac_learning);
241 /* All datapaths of a given type share a single dpif backer instance. */
248 struct ovs_rwlock odp_to_ofport_lock;
249 struct hmap odp_to_ofport_map OVS_GUARDED; /* Contains "struct ofport"s. */
251 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
253 enum revalidate_reason need_revalidate; /* Revalidate all flows. */
255 bool recv_set_enable; /* Enables or disables receiving packets. */
258 struct recirc_id_pool *rid_pool; /* Recirculation ID pool. */
259 bool enable_recirc; /* True if the datapath supports recirculation */
261 /* True if the datapath supports variable-length
262 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.
263 * False if the datapath supports only 8-byte (or shorter) userdata. */
264 bool variable_length_userdata;
266 /* Maximum number of MPLS label stack entries that the datapath supports
268 size_t max_mpls_depth;
271 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
272 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
274 struct ofproto_dpif {
275 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
277 struct dpif_backer *backer;
279 uint64_t dump_seq; /* Last read of udpif_dump_seq(). */
281 /* Special OpenFlow rules. */
282 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
283 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
284 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
287 struct netflow *netflow;
288 struct dpif_sflow *sflow;
289 struct dpif_ipfix *ipfix;
290 struct hmap bundles; /* Contains "struct ofbundle"s. */
291 struct mac_learning *ml;
292 bool has_bonded_bundles;
294 struct mbridge *mbridge;
296 struct ovs_mutex stats_mutex;
297 struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
298 * consumed in userspace. */
302 long long int stp_last_tick;
304 /* VLAN splinters. */
305 struct ovs_mutex vsp_mutex;
306 struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
307 struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */
310 struct sset ports; /* Set of standard port names. */
311 struct sset ghost_ports; /* Ports with no datapath port. */
312 struct sset port_poll_set; /* Queued names for port_poll() reply. */
313 int port_poll_errno; /* Last errno for port_poll() reply. */
314 uint64_t change_seq; /* Connectivity status changes. */
317 struct guarded_list pins; /* Contains "struct ofputil_packet_in"s. */
318 struct seq *pins_seq; /* For notifying 'pins' reception. */
322 /* All existing ofproto_dpif instances, indexed by ->up.name. */
323 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
325 static void ofproto_dpif_unixctl_init(void);
327 static inline struct ofproto_dpif *
328 ofproto_dpif_cast(const struct ofproto *ofproto)
330 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
331 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
335 ofproto_dpif_get_max_mpls_depth(const struct ofproto_dpif *ofproto)
337 return ofproto->backer->max_mpls_depth;
341 ofproto_dpif_get_enable_recirc(const struct ofproto_dpif *ofproto)
343 return ofproto->backer->enable_recirc;
346 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
347 ofp_port_t ofp_port);
348 static void ofproto_trace(struct ofproto_dpif *, struct flow *,
349 const struct ofpbuf *packet,
350 const struct ofpact[], size_t ofpacts_len,
353 /* Global variables. */
354 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
356 /* Initial mappings of port to bridge mappings. */
357 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
359 /* Executes 'fm'. The caller retains ownership of 'fm' and everything in
362 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
363 struct ofputil_flow_mod *fm)
365 ofproto_flow_mod(&ofproto->up, fm);
368 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
369 * Takes ownership of 'pin' and pin->packet. */
371 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
372 struct ofproto_packet_in *pin)
374 if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
375 COVERAGE_INC(packet_in_overflow);
376 free(CONST_CAST(void *, pin->up.packet));
380 /* Wakes up main thread for packet-in I/O. */
381 seq_change(ofproto->pins_seq);
384 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
385 * packet rather than to send the packet to the controller.
387 * This function returns false to indicate that a packet_in message
388 * for a "table-miss" should be sent to at least one controller.
389 * False otherwise. */
391 ofproto_dpif_wants_packet_in_on_miss(struct ofproto_dpif *ofproto)
393 return connmgr_wants_packet_in_on_miss(ofproto->up.connmgr);
396 /* Factory functions. */
399 init(const struct shash *iface_hints)
401 struct shash_node *node;
403 /* Make a local copy, since we don't own 'iface_hints' elements. */
404 SHASH_FOR_EACH(node, iface_hints) {
405 const struct iface_hint *orig_hint = node->data;
406 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
408 new_hint->br_name = xstrdup(orig_hint->br_name);
409 new_hint->br_type = xstrdup(orig_hint->br_type);
410 new_hint->ofp_port = orig_hint->ofp_port;
412 shash_add(&init_ofp_ports, node->name, new_hint);
417 enumerate_types(struct sset *types)
419 dp_enumerate_types(types);
423 enumerate_names(const char *type, struct sset *names)
425 struct ofproto_dpif *ofproto;
428 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
429 if (strcmp(type, ofproto->up.type)) {
432 sset_add(names, ofproto->up.name);
439 del(const char *type, const char *name)
444 error = dpif_open(name, type, &dpif);
446 error = dpif_delete(dpif);
453 port_open_type(const char *datapath_type, const char *port_type)
455 return dpif_port_open_type(datapath_type, port_type);
458 /* Type functions. */
460 static void process_dpif_port_changes(struct dpif_backer *);
461 static void process_dpif_all_ports_changed(struct dpif_backer *);
462 static void process_dpif_port_change(struct dpif_backer *,
463 const char *devname);
464 static void process_dpif_port_error(struct dpif_backer *, int error);
466 static struct ofproto_dpif *
467 lookup_ofproto_dpif_by_port_name(const char *name)
469 struct ofproto_dpif *ofproto;
471 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
472 if (sset_contains(&ofproto->ports, name)) {
481 type_run(const char *type)
483 struct dpif_backer *backer;
485 backer = shash_find_data(&all_dpif_backers, type);
487 /* This is not necessarily a problem, since backers are only
488 * created on demand. */
492 dpif_run(backer->dpif);
494 /* If vswitchd started with other_config:flow_restore_wait set as "true",
495 * and the configuration has now changed to "false", enable receiving
496 * packets from the datapath. */
497 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
500 backer->recv_set_enable = true;
502 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
504 VLOG_ERR("Failed to enable receiving packets in dpif.");
507 dpif_flow_flush(backer->dpif);
508 backer->need_revalidate = REV_RECONFIGURE;
511 if (backer->recv_set_enable) {
512 udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
515 if (backer->need_revalidate) {
516 struct ofproto_dpif *ofproto;
517 struct simap_node *node;
518 struct simap tmp_backers;
520 /* Handle tunnel garbage collection. */
521 simap_init(&tmp_backers);
522 simap_swap(&backer->tnl_backers, &tmp_backers);
524 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
525 struct ofport_dpif *iter;
527 if (backer != ofproto->backer) {
531 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
532 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
535 if (!iter->is_tunnel) {
539 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
540 namebuf, sizeof namebuf);
541 node = simap_find(&tmp_backers, dp_port);
543 simap_put(&backer->tnl_backers, dp_port, node->data);
544 simap_delete(&tmp_backers, node);
545 node = simap_find(&backer->tnl_backers, dp_port);
547 node = simap_find(&backer->tnl_backers, dp_port);
549 odp_port_t odp_port = ODPP_NONE;
551 if (!dpif_port_add(backer->dpif, iter->up.netdev,
553 simap_put(&backer->tnl_backers, dp_port,
554 odp_to_u32(odp_port));
555 node = simap_find(&backer->tnl_backers, dp_port);
560 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
561 if (tnl_port_reconfigure(iter, iter->up.netdev,
563 backer->need_revalidate = REV_RECONFIGURE;
568 SIMAP_FOR_EACH (node, &tmp_backers) {
569 dpif_port_del(backer->dpif, u32_to_odp(node->data));
571 simap_destroy(&tmp_backers);
573 switch (backer->need_revalidate) {
574 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
575 case REV_STP: COVERAGE_INC(rev_stp); break;
576 case REV_BOND: COVERAGE_INC(rev_bond); break;
577 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
578 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
579 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
581 backer->need_revalidate = 0;
583 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
584 struct ofport_dpif *ofport;
585 struct ofbundle *bundle;
587 if (ofproto->backer != backer) {
591 ovs_rwlock_wrlock(&xlate_rwlock);
592 xlate_ofproto_set(ofproto, ofproto->up.name,
593 ofproto->backer->dpif, ofproto->miss_rule,
594 ofproto->no_packet_in_rule, ofproto->ml,
595 ofproto->stp, ofproto->mbridge,
596 ofproto->sflow, ofproto->ipfix,
597 ofproto->netflow, ofproto->up.frag_handling,
598 ofproto->up.forward_bpdu,
599 connmgr_has_in_band(ofproto->up.connmgr),
600 ofproto->backer->enable_recirc,
601 ofproto->backer->variable_length_userdata,
602 ofproto->backer->max_mpls_depth);
604 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
605 xlate_bundle_set(ofproto, bundle, bundle->name,
606 bundle->vlan_mode, bundle->vlan,
607 bundle->trunks, bundle->use_priority_tags,
608 bundle->bond, bundle->lacp,
612 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
613 int stp_port = ofport->stp_port
614 ? stp_port_no(ofport->stp_port)
616 xlate_ofport_set(ofproto, ofport->bundle, ofport,
617 ofport->up.ofp_port, ofport->odp_port,
618 ofport->up.netdev, ofport->cfm,
619 ofport->bfd, ofport->peer, stp_port,
620 ofport->qdscp, ofport->n_qdscp,
621 ofport->up.pp.config, ofport->up.pp.state,
622 ofport->is_tunnel, ofport->may_enable);
624 ovs_rwlock_unlock(&xlate_rwlock);
627 udpif_revalidate(backer->udpif);
630 process_dpif_port_changes(backer);
635 /* Check for and handle port changes in 'backer''s dpif. */
637 process_dpif_port_changes(struct dpif_backer *backer)
643 error = dpif_port_poll(backer->dpif, &devname);
649 process_dpif_all_ports_changed(backer);
653 process_dpif_port_change(backer, devname);
658 process_dpif_port_error(backer, error);
665 process_dpif_all_ports_changed(struct dpif_backer *backer)
667 struct ofproto_dpif *ofproto;
668 struct dpif_port dpif_port;
669 struct dpif_port_dump dump;
670 struct sset devnames;
673 sset_init(&devnames);
674 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
675 if (ofproto->backer == backer) {
676 struct ofport *ofport;
678 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
679 sset_add(&devnames, netdev_get_name(ofport->netdev));
683 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
684 sset_add(&devnames, dpif_port.name);
687 SSET_FOR_EACH (devname, &devnames) {
688 process_dpif_port_change(backer, devname);
690 sset_destroy(&devnames);
694 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
696 struct ofproto_dpif *ofproto;
697 struct dpif_port port;
699 /* Don't report on the datapath's device. */
700 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
704 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
705 &all_ofproto_dpifs) {
706 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
711 ofproto = lookup_ofproto_dpif_by_port_name(devname);
712 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
713 /* The port was removed. If we know the datapath,
714 * report it through poll_set(). If we don't, it may be
715 * notifying us of a removal we initiated, so ignore it.
716 * If there's a pending ENOBUFS, let it stand, since
717 * everything will be reevaluated. */
718 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
719 sset_add(&ofproto->port_poll_set, devname);
720 ofproto->port_poll_errno = 0;
722 } else if (!ofproto) {
723 /* The port was added, but we don't know with which
724 * ofproto we should associate it. Delete it. */
725 dpif_port_del(backer->dpif, port.port_no);
727 struct ofport_dpif *ofport;
729 ofport = ofport_dpif_cast(shash_find_data(
730 &ofproto->up.port_by_name, devname));
732 && ofport->odp_port != port.port_no
733 && !odp_port_to_ofport(backer, port.port_no))
735 /* 'ofport''s datapath port number has changed from
736 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
737 * structures to match. */
738 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
739 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
740 ofport->odp_port = port.port_no;
741 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
742 hash_odp_port(port.port_no));
743 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
744 backer->need_revalidate = REV_RECONFIGURE;
747 dpif_port_destroy(&port);
750 /* Propagate 'error' to all ofprotos based on 'backer'. */
752 process_dpif_port_error(struct dpif_backer *backer, int error)
754 struct ofproto_dpif *ofproto;
756 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
757 if (ofproto->backer == backer) {
758 sset_clear(&ofproto->port_poll_set);
759 ofproto->port_poll_errno = error;
765 type_wait(const char *type)
767 struct dpif_backer *backer;
769 backer = shash_find_data(&all_dpif_backers, type);
771 /* This is not necessarily a problem, since backers are only
772 * created on demand. */
776 dpif_wait(backer->dpif);
779 /* Basic life-cycle. */
781 static int add_internal_flows(struct ofproto_dpif *);
783 static struct ofproto *
786 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
791 dealloc(struct ofproto *ofproto_)
793 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
798 close_dpif_backer(struct dpif_backer *backer)
800 ovs_assert(backer->refcount > 0);
802 if (--backer->refcount) {
806 udpif_destroy(backer->udpif);
808 simap_destroy(&backer->tnl_backers);
809 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
810 hmap_destroy(&backer->odp_to_ofport_map);
811 shash_find_and_delete(&all_dpif_backers, backer->type);
812 recirc_id_pool_destroy(backer->rid_pool);
814 dpif_close(backer->dpif);
818 /* Datapath port slated for removal from datapath. */
820 struct list list_node;
824 static bool check_variable_length_userdata(struct dpif_backer *backer);
825 static size_t check_max_mpls_depth(struct dpif_backer *backer);
826 static bool check_recirc(struct dpif_backer *backer);
829 open_dpif_backer(const char *type, struct dpif_backer **backerp)
831 struct dpif_backer *backer;
832 struct dpif_port_dump port_dump;
833 struct dpif_port port;
834 struct shash_node *node;
835 struct list garbage_list;
836 struct odp_garbage *garbage, *next;
843 backer = shash_find_data(&all_dpif_backers, type);
850 backer_name = xasprintf("ovs-%s", type);
852 /* Remove any existing datapaths, since we assume we're the only
853 * userspace controlling the datapath. */
855 dp_enumerate_names(type, &names);
856 SSET_FOR_EACH(name, &names) {
857 struct dpif *old_dpif;
859 /* Don't remove our backer if it exists. */
860 if (!strcmp(name, backer_name)) {
864 if (dpif_open(name, type, &old_dpif)) {
865 VLOG_WARN("couldn't open old datapath %s to remove it", name);
867 dpif_delete(old_dpif);
868 dpif_close(old_dpif);
871 sset_destroy(&names);
873 backer = xmalloc(sizeof *backer);
875 error = dpif_create_and_open(backer_name, type, &backer->dpif);
878 VLOG_ERR("failed to open datapath of type %s: %s", type,
879 ovs_strerror(error));
883 backer->udpif = udpif_create(backer, backer->dpif);
885 backer->type = xstrdup(type);
886 backer->refcount = 1;
887 hmap_init(&backer->odp_to_ofport_map);
888 ovs_rwlock_init(&backer->odp_to_ofport_lock);
889 backer->need_revalidate = 0;
890 simap_init(&backer->tnl_backers);
891 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
894 if (backer->recv_set_enable) {
895 dpif_flow_flush(backer->dpif);
898 /* Loop through the ports already on the datapath and remove any
899 * that we don't need anymore. */
900 list_init(&garbage_list);
901 dpif_port_dump_start(&port_dump, backer->dpif);
902 while (dpif_port_dump_next(&port_dump, &port)) {
903 node = shash_find(&init_ofp_ports, port.name);
904 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
905 garbage = xmalloc(sizeof *garbage);
906 garbage->odp_port = port.port_no;
907 list_push_front(&garbage_list, &garbage->list_node);
910 dpif_port_dump_done(&port_dump);
912 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
913 dpif_port_del(backer->dpif, garbage->odp_port);
914 list_remove(&garbage->list_node);
918 shash_add(&all_dpif_backers, type, backer);
920 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
922 VLOG_ERR("failed to listen on datapath of type %s: %s",
923 type, ovs_strerror(error));
924 close_dpif_backer(backer);
927 backer->enable_recirc = check_recirc(backer);
928 backer->variable_length_userdata = check_variable_length_userdata(backer);
929 backer->max_mpls_depth = check_max_mpls_depth(backer);
930 backer->rid_pool = recirc_id_pool_create();
932 if (backer->recv_set_enable) {
933 udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
939 /* Tests whether 'backer''s datapath supports recirculation. Only newer
940 * datapaths support OVS_KEY_ATTR_RECIRC_ID in keys. We need to disable some
941 * features on older datapaths that don't support this feature.
943 * Returns false if 'backer' definitely does not support recirculation, true if
944 * it seems to support recirculation or if at least the error we get is
947 check_recirc(struct dpif_backer *backer)
950 struct odputil_keybuf keybuf;
953 bool enable_recirc = false;
955 memset(&flow, 0, sizeof flow);
959 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
960 odp_flow_key_from_flow(&key, &flow, NULL, 0);
962 error = dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
963 ofpbuf_data(&key), ofpbuf_size(&key), NULL, 0, NULL,
965 if (error && error != EEXIST) {
966 if (error != EINVAL) {
967 VLOG_WARN("%s: Reciculation flow probe failed (%s)",
968 dpif_name(backer->dpif), ovs_strerror(error));
973 error = dpif_flow_del(backer->dpif, ofpbuf_data(&key), ofpbuf_size(&key),
976 VLOG_WARN("%s: failed to delete recirculation feature probe flow",
977 dpif_name(backer->dpif));
980 enable_recirc = true;
984 VLOG_INFO("%s: Datapath supports recirculation",
985 dpif_name(backer->dpif));
987 VLOG_INFO("%s: Datapath does not support recirculation",
988 dpif_name(backer->dpif));
991 return enable_recirc;
994 /* Tests whether 'backer''s datapath supports variable-length
995 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions. We need
996 * to disable some features on older datapaths that don't support this
999 * Returns false if 'backer' definitely does not support variable-length
1000 * userdata, true if it seems to support them or if at least the error we get
1003 check_variable_length_userdata(struct dpif_backer *backer)
1005 struct eth_header *eth;
1006 struct ofpbuf actions;
1007 struct dpif_execute execute;
1008 struct ofpbuf packet;
1012 /* Compose a userspace action that will cause an ERANGE error on older
1013 * datapaths that don't support variable-length userdata.
1015 * We really test for using userdata longer than 8 bytes, but older
1016 * datapaths accepted these, silently truncating the userdata to 8 bytes.
1017 * The same older datapaths rejected userdata shorter than 8 bytes, so we
1018 * test for that instead as a proxy for longer userdata support. */
1019 ofpbuf_init(&actions, 64);
1020 start = nl_msg_start_nested(&actions, OVS_ACTION_ATTR_USERSPACE);
1021 nl_msg_put_u32(&actions, OVS_USERSPACE_ATTR_PID,
1022 dpif_port_get_pid(backer->dpif, ODPP_NONE, 0));
1023 nl_msg_put_unspec_zero(&actions, OVS_USERSPACE_ATTR_USERDATA, 4);
1024 nl_msg_end_nested(&actions, start);
1026 /* Compose a dummy ethernet packet. */
1027 ofpbuf_init(&packet, ETH_HEADER_LEN);
1028 eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
1029 eth->eth_type = htons(0x1234);
1031 /* Execute the actions. On older datapaths this fails with ERANGE, on
1032 * newer datapaths it succeeds. */
1033 execute.actions = ofpbuf_data(&actions);
1034 execute.actions_len = ofpbuf_size(&actions);
1035 execute.packet = &packet;
1036 execute.md = PKT_METADATA_INITIALIZER(0);
1037 execute.needs_help = false;
1039 error = dpif_execute(backer->dpif, &execute);
1041 ofpbuf_uninit(&packet);
1042 ofpbuf_uninit(&actions);
1046 /* Variable-length userdata is supported.
1048 * Purge received packets to avoid processing the nonsense packet we
1049 * sent to userspace, then report success. */
1050 dpif_recv_purge(backer->dpif);
1054 /* Variable-length userdata is not supported. */
1055 VLOG_WARN("%s: datapath does not support variable-length userdata "
1056 "feature (needs Linux 3.10+ or kernel module from OVS "
1057 "1..11+). The NXAST_SAMPLE action will be ignored.",
1058 dpif_name(backer->dpif));
1062 /* Something odd happened. We're not sure whether variable-length
1063 * userdata is supported. Default to "yes". */
1064 VLOG_WARN("%s: variable-length userdata feature probe failed (%s)",
1065 dpif_name(backer->dpif), ovs_strerror(error));
1070 /* Tests the MPLS label stack depth supported by 'backer''s datapath.
1072 * Returns the number of elements in a struct flow's mpls_lse field
1073 * if the datapath supports at least that many entries in an
1075 * Otherwise returns the number of MPLS push actions supported by
1078 check_max_mpls_depth(struct dpif_backer *backer)
1083 for (n = 0; n < FLOW_MAX_MPLS_LABELS; n++) {
1084 struct odputil_keybuf keybuf;
1088 memset(&flow, 0, sizeof flow);
1089 flow.dl_type = htons(ETH_TYPE_MPLS);
1090 flow_set_mpls_bos(&flow, n, 1);
1092 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1093 odp_flow_key_from_flow(&key, &flow, NULL, 0);
1095 error = dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
1096 ofpbuf_data(&key), ofpbuf_size(&key), NULL, 0, NULL, 0, NULL);
1097 if (error && error != EEXIST) {
1098 if (error != EINVAL) {
1099 VLOG_WARN("%s: MPLS stack length feature probe failed (%s)",
1100 dpif_name(backer->dpif), ovs_strerror(error));
1105 error = dpif_flow_del(backer->dpif, ofpbuf_data(&key), ofpbuf_size(&key), NULL);
1107 VLOG_WARN("%s: failed to delete MPLS feature probe flow",
1108 dpif_name(backer->dpif));
1112 VLOG_INFO("%s: MPLS label stack length probed as %d",
1113 dpif_name(backer->dpif), n);
1118 construct(struct ofproto *ofproto_)
1120 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1121 struct shash_node *node, *next;
1124 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1129 ofproto->netflow = NULL;
1130 ofproto->sflow = NULL;
1131 ofproto->ipfix = NULL;
1132 ofproto->stp = NULL;
1133 ofproto->dump_seq = 0;
1134 hmap_init(&ofproto->bundles);
1135 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1136 ofproto->mbridge = mbridge_create();
1137 ofproto->has_bonded_bundles = false;
1138 ofproto->lacp_enabled = false;
1139 ovs_mutex_init_adaptive(&ofproto->stats_mutex);
1140 ovs_mutex_init(&ofproto->vsp_mutex);
1142 guarded_list_init(&ofproto->pins);
1144 ofproto_dpif_unixctl_init();
1146 hmap_init(&ofproto->vlandev_map);
1147 hmap_init(&ofproto->realdev_vid_map);
1149 sset_init(&ofproto->ports);
1150 sset_init(&ofproto->ghost_ports);
1151 sset_init(&ofproto->port_poll_set);
1152 ofproto->port_poll_errno = 0;
1153 ofproto->change_seq = 0;
1154 ofproto->pins_seq = seq_create();
1155 ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1158 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1159 struct iface_hint *iface_hint = node->data;
1161 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1162 /* Check if the datapath already has this port. */
1163 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1164 sset_add(&ofproto->ports, node->name);
1167 free(iface_hint->br_name);
1168 free(iface_hint->br_type);
1170 shash_delete(&init_ofp_ports, node);
1174 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1175 hash_string(ofproto->up.name, 0));
1176 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1178 ofproto_init_tables(ofproto_, N_TABLES);
1179 error = add_internal_flows(ofproto);
1181 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1187 add_internal_miss_flow(struct ofproto_dpif *ofproto, int id,
1188 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1194 match_init_catchall(&match);
1195 match_set_reg(&match, 0, id);
1197 error = ofproto_dpif_add_internal_flow(ofproto, &match, 0, ofpacts, &rule);
1198 *rulep = error ? NULL : rule_dpif_cast(rule);
1204 add_internal_flows(struct ofproto_dpif *ofproto)
1206 struct ofpact_controller *controller;
1207 uint64_t ofpacts_stub[128 / 8];
1208 struct ofpbuf ofpacts;
1209 struct rule *unused_rulep OVS_UNUSED;
1210 struct ofpact_resubmit *resubmit;
1215 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1218 controller = ofpact_put_CONTROLLER(&ofpacts);
1219 controller->max_len = UINT16_MAX;
1220 controller->controller_id = 0;
1221 controller->reason = OFPR_NO_MATCH;
1222 ofpact_pad(&ofpacts);
1224 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1225 &ofproto->miss_rule);
1230 ofpbuf_clear(&ofpacts);
1231 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1232 &ofproto->no_packet_in_rule);
1237 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1238 &ofproto->drop_frags_rule);
1243 /* Continue non-recirculation rule lookups from table 0.
1245 * (priority=2), recirc=0, actions=resubmit(, 0)
1247 resubmit = ofpact_put_RESUBMIT(&ofpacts);
1248 resubmit->ofpact.compat = 0;
1249 resubmit->in_port = OFPP_IN_PORT;
1250 resubmit->table_id = 0;
1252 match_init_catchall(&match);
1253 match_set_recirc_id(&match, 0);
1255 error = ofproto_dpif_add_internal_flow(ofproto, &match, 2, &ofpacts,
1261 /* Drop any run away recirc rule lookups. Recirc_id has to be
1262 * non-zero when reaching this rule.
1264 * (priority=1), *, actions=drop
1266 ofpbuf_clear(&ofpacts);
1267 match_init_catchall(&match);
1268 error = ofproto_dpif_add_internal_flow(ofproto, &match, 1, &ofpacts,
1275 destruct(struct ofproto *ofproto_)
1277 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1278 struct rule_dpif *rule, *next_rule;
1279 struct ofproto_packet_in *pin, *next_pin;
1280 struct oftable *table;
1283 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1284 ovs_rwlock_wrlock(&xlate_rwlock);
1285 xlate_remove_ofproto(ofproto);
1286 ovs_rwlock_unlock(&xlate_rwlock);
1288 /* Ensure that the upcall processing threads have no remaining references
1289 * to the ofproto or anything in it. */
1290 udpif_synchronize(ofproto->backer->udpif);
1292 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1294 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1295 struct cls_cursor cursor;
1297 fat_rwlock_rdlock(&table->cls.rwlock);
1298 cls_cursor_init(&cursor, &table->cls, NULL);
1299 fat_rwlock_unlock(&table->cls.rwlock);
1300 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1301 ofproto_rule_delete(&ofproto->up, &rule->up);
1305 guarded_list_pop_all(&ofproto->pins, &pins);
1306 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1307 list_remove(&pin->list_node);
1308 free(CONST_CAST(void *, pin->up.packet));
1311 guarded_list_destroy(&ofproto->pins);
1313 mbridge_unref(ofproto->mbridge);
1315 netflow_unref(ofproto->netflow);
1316 dpif_sflow_unref(ofproto->sflow);
1317 hmap_destroy(&ofproto->bundles);
1318 mac_learning_unref(ofproto->ml);
1320 hmap_destroy(&ofproto->vlandev_map);
1321 hmap_destroy(&ofproto->realdev_vid_map);
1323 sset_destroy(&ofproto->ports);
1324 sset_destroy(&ofproto->ghost_ports);
1325 sset_destroy(&ofproto->port_poll_set);
1327 ovs_mutex_destroy(&ofproto->stats_mutex);
1328 ovs_mutex_destroy(&ofproto->vsp_mutex);
1330 seq_destroy(ofproto->pins_seq);
1332 close_dpif_backer(ofproto->backer);
1336 run(struct ofproto *ofproto_)
1338 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1339 uint64_t new_seq, new_dump_seq;
1341 if (mbridge_need_revalidate(ofproto->mbridge)) {
1342 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1343 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1344 mac_learning_flush(ofproto->ml);
1345 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1348 /* Always updates the ofproto->pins_seqno to avoid frequent wakeup during
1349 * flow restore. Even though nothing is processed during flow restore,
1350 * all queued 'pins' will be handled immediately when flow restore
1352 ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1354 /* Do not perform any periodic activity required by 'ofproto' while
1355 * waiting for flow restore to complete. */
1356 if (!ofproto_get_flow_restore_wait()) {
1357 struct ofproto_packet_in *pin, *next_pin;
1360 guarded_list_pop_all(&ofproto->pins, &pins);
1361 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1362 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1363 list_remove(&pin->list_node);
1364 free(CONST_CAST(void *, pin->up.packet));
1369 if (ofproto->netflow) {
1370 netflow_run(ofproto->netflow);
1372 if (ofproto->sflow) {
1373 dpif_sflow_run(ofproto->sflow);
1375 if (ofproto->ipfix) {
1376 dpif_ipfix_run(ofproto->ipfix);
1379 new_seq = seq_read(connectivity_seq_get());
1380 if (ofproto->change_seq != new_seq) {
1381 struct ofport_dpif *ofport;
1383 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1387 ofproto->change_seq = new_seq;
1389 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1390 struct ofbundle *bundle;
1392 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1398 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1399 if (mac_learning_run(ofproto->ml)) {
1400 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1402 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1404 new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
1405 if (ofproto->dump_seq != new_dump_seq) {
1406 struct rule *rule, *next_rule;
1408 /* We know stats are relatively fresh, so now is a good time to do some
1410 ofproto->dump_seq = new_dump_seq;
1412 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1414 ovs_mutex_lock(&ofproto_mutex);
1415 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
1416 &ofproto->up.expirable) {
1417 rule_expire(rule_dpif_cast(rule));
1419 ovs_mutex_unlock(&ofproto_mutex);
1421 /* All outstanding data in existing flows has been accounted, so it's a
1422 * good time to do bond rebalancing. */
1423 if (ofproto->has_bonded_bundles) {
1424 struct ofbundle *bundle;
1426 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1428 bond_rebalance(bundle->bond);
1438 wait(struct ofproto *ofproto_)
1440 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1442 if (ofproto_get_flow_restore_wait()) {
1446 if (ofproto->sflow) {
1447 dpif_sflow_wait(ofproto->sflow);
1449 if (ofproto->ipfix) {
1450 dpif_ipfix_wait(ofproto->ipfix);
1452 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1453 struct ofbundle *bundle;
1455 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1456 bundle_wait(bundle);
1459 if (ofproto->netflow) {
1460 netflow_wait(ofproto->netflow);
1462 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1463 mac_learning_wait(ofproto->ml);
1464 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1466 if (ofproto->backer->need_revalidate) {
1467 /* Shouldn't happen, but if it does just go around again. */
1468 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1469 poll_immediate_wake();
1472 seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
1473 seq_wait(ofproto->pins_seq, ofproto->pins_seqno);
1477 type_get_memory_usage(const char *type, struct simap *usage)
1479 struct dpif_backer *backer;
1481 backer = shash_find_data(&all_dpif_backers, type);
1483 udpif_get_memory_usage(backer->udpif, usage);
1488 flush(struct ofproto *ofproto_)
1490 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1491 struct dpif_backer *backer = ofproto->backer;
1494 udpif_flush(backer->udpif);
1499 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1500 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1502 *arp_match_ip = true;
1503 *actions = (OFPUTIL_A_OUTPUT |
1504 OFPUTIL_A_SET_VLAN_VID |
1505 OFPUTIL_A_SET_VLAN_PCP |
1506 OFPUTIL_A_STRIP_VLAN |
1507 OFPUTIL_A_SET_DL_SRC |
1508 OFPUTIL_A_SET_DL_DST |
1509 OFPUTIL_A_SET_NW_SRC |
1510 OFPUTIL_A_SET_NW_DST |
1511 OFPUTIL_A_SET_NW_TOS |
1512 OFPUTIL_A_SET_TP_SRC |
1513 OFPUTIL_A_SET_TP_DST |
1518 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1520 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1521 struct dpif_dp_stats s;
1522 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1526 strcpy(ots->name, "classifier");
1528 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1529 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes, &used);
1530 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes,
1532 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes,
1534 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1535 ots->lookup_count = htonll(n_lookup);
1536 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1539 static struct ofport *
1542 struct ofport_dpif *port = xmalloc(sizeof *port);
1547 port_dealloc(struct ofport *port_)
1549 struct ofport_dpif *port = ofport_dpif_cast(port_);
1554 port_construct(struct ofport *port_)
1556 struct ofport_dpif *port = ofport_dpif_cast(port_);
1557 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1558 const struct netdev *netdev = port->up.netdev;
1559 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1560 struct dpif_port dpif_port;
1563 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1564 port->bundle = NULL;
1567 port->may_enable = true;
1568 port->stp_port = NULL;
1569 port->stp_state = STP_DISABLED;
1570 port->is_tunnel = false;
1574 port->realdev_ofp_port = 0;
1575 port->vlandev_vid = 0;
1576 port->carrier_seq = netdev_get_carrier_resets(netdev);
1577 port->is_layer3 = netdev_vport_is_layer3(netdev);
1579 if (netdev_vport_is_patch(netdev)) {
1580 /* By bailing out here, we don't submit the port to the sFlow module
1581 * to be considered for counter polling export. This is correct
1582 * because the patch port represents an interface that sFlow considers
1583 * to be "internal" to the switch as a whole, and therefore not an
1584 * candidate for counter polling. */
1585 port->odp_port = ODPP_NONE;
1586 ofport_update_peer(port);
1590 error = dpif_port_query_by_name(ofproto->backer->dpif,
1591 netdev_vport_get_dpif_port(netdev, namebuf,
1598 port->odp_port = dpif_port.port_no;
1600 if (netdev_get_tunnel_config(netdev)) {
1601 tnl_port_add(port, port->up.netdev, port->odp_port);
1602 port->is_tunnel = true;
1604 /* Sanity-check that a mapping doesn't already exist. This
1605 * shouldn't happen for non-tunnel ports. */
1606 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1607 VLOG_ERR("port %s already has an OpenFlow port number",
1609 dpif_port_destroy(&dpif_port);
1613 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1614 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1615 hash_odp_port(port->odp_port));
1616 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1618 dpif_port_destroy(&dpif_port);
1620 if (ofproto->sflow) {
1621 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1628 port_destruct(struct ofport *port_)
1630 struct ofport_dpif *port = ofport_dpif_cast(port_);
1631 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1632 const char *devname = netdev_get_name(port->up.netdev);
1633 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1634 const char *dp_port_name;
1636 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1637 ovs_rwlock_wrlock(&xlate_rwlock);
1638 xlate_ofport_remove(port);
1639 ovs_rwlock_unlock(&xlate_rwlock);
1641 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1643 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1644 /* The underlying device is still there, so delete it. This
1645 * happens when the ofproto is being destroyed, since the caller
1646 * assumes that removal of attached ports will happen as part of
1648 if (!port->is_tunnel) {
1649 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1654 port->peer->peer = NULL;
1658 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1659 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1660 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1661 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1665 sset_find_and_delete(&ofproto->ports, devname);
1666 sset_find_and_delete(&ofproto->ghost_ports, devname);
1667 bundle_remove(port_);
1668 set_cfm(port_, NULL);
1669 set_bfd(port_, NULL);
1670 if (port->stp_port) {
1671 stp_port_disable(port->stp_port);
1673 if (ofproto->sflow) {
1674 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1681 port_modified(struct ofport *port_)
1683 struct ofport_dpif *port = ofport_dpif_cast(port_);
1685 if (port->bundle && port->bundle->bond) {
1686 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1690 cfm_set_netdev(port->cfm, port->up.netdev);
1694 bfd_set_netdev(port->bfd, port->up.netdev);
1697 ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1698 port->up.pp.hw_addr);
1700 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1702 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1706 ofport_update_peer(port);
1710 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1712 struct ofport_dpif *port = ofport_dpif_cast(port_);
1713 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1714 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1716 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1717 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1718 OFPUTIL_PC_NO_PACKET_IN)) {
1719 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1721 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1722 bundle_update(port->bundle);
1728 set_sflow(struct ofproto *ofproto_,
1729 const struct ofproto_sflow_options *sflow_options)
1731 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1732 struct dpif_sflow *ds = ofproto->sflow;
1734 if (sflow_options) {
1736 struct ofport_dpif *ofport;
1738 ds = ofproto->sflow = dpif_sflow_create();
1739 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1740 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1742 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1744 dpif_sflow_set_options(ds, sflow_options);
1747 dpif_sflow_unref(ds);
1748 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1749 ofproto->sflow = NULL;
1757 struct ofproto *ofproto_,
1758 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1759 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1760 size_t n_flow_exporters_options)
1762 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1763 struct dpif_ipfix *di = ofproto->ipfix;
1764 bool has_options = bridge_exporter_options || flow_exporters_options;
1766 if (has_options && !di) {
1767 di = ofproto->ipfix = dpif_ipfix_create();
1771 /* Call set_options in any case to cleanly flush the flow
1772 * caches in the last exporters that are to be destroyed. */
1773 dpif_ipfix_set_options(
1774 di, bridge_exporter_options, flow_exporters_options,
1775 n_flow_exporters_options);
1778 dpif_ipfix_unref(di);
1779 ofproto->ipfix = NULL;
1787 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1789 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1794 struct ofproto_dpif *ofproto;
1796 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1797 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1798 ofport->cfm = cfm_create(ofport->up.netdev);
1801 if (cfm_configure(ofport->cfm, s)) {
1808 cfm_unref(ofport->cfm);
1811 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1812 ofport->up.pp.hw_addr);
1817 get_cfm_status(const struct ofport *ofport_, bool force,
1818 struct ofproto_cfm_status *status)
1820 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1824 if (cfm_check_status_change(ofport->cfm) || force) {
1825 status->faults = cfm_get_fault(ofport->cfm);
1826 status->flap_count = cfm_get_flap_count(ofport->cfm);
1827 status->remote_opstate = cfm_get_opup(ofport->cfm);
1828 status->health = cfm_get_health(ofport->cfm);
1829 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1831 ret = NO_STATUS_CHANGE;
1841 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1843 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1844 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1848 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1849 cfg, ofport->up.netdev);
1850 if (ofport->bfd != old) {
1851 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1853 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1854 ofport->up.pp.hw_addr);
1859 get_bfd_status(struct ofport *ofport_, bool force, struct smap *smap)
1861 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1865 if (bfd_check_status_change(ofport->bfd) || force) {
1866 bfd_get_status(ofport->bfd, smap);
1868 ret = NO_STATUS_CHANGE;
1877 /* Spanning Tree. */
1880 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1882 struct ofproto_dpif *ofproto = ofproto_;
1883 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1884 struct ofport_dpif *ofport;
1886 ofport = stp_port_get_aux(sp);
1888 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1889 ofproto->up.name, port_num);
1891 struct eth_header *eth = ofpbuf_l2(pkt);
1893 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1894 if (eth_addr_is_zero(eth->eth_src)) {
1895 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1896 "with unknown MAC", ofproto->up.name, port_num);
1898 ofproto_dpif_send_packet(ofport, pkt);
1904 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1906 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1908 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1910 /* Only revalidate flows if the configuration changed. */
1911 if (!s != !ofproto->stp) {
1912 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1916 if (!ofproto->stp) {
1917 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1918 send_bpdu_cb, ofproto);
1919 ofproto->stp_last_tick = time_msec();
1922 stp_set_bridge_id(ofproto->stp, s->system_id);
1923 stp_set_bridge_priority(ofproto->stp, s->priority);
1924 stp_set_hello_time(ofproto->stp, s->hello_time);
1925 stp_set_max_age(ofproto->stp, s->max_age);
1926 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1928 struct ofport *ofport;
1930 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1931 set_stp_port(ofport, NULL);
1934 stp_unref(ofproto->stp);
1935 ofproto->stp = NULL;
1942 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1944 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1948 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1949 s->designated_root = stp_get_designated_root(ofproto->stp);
1950 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1959 update_stp_port_state(struct ofport_dpif *ofport)
1961 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1962 enum stp_state state;
1964 /* Figure out new state. */
1965 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1969 if (ofport->stp_state != state) {
1970 enum ofputil_port_state of_state;
1973 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1974 netdev_get_name(ofport->up.netdev),
1975 stp_state_name(ofport->stp_state),
1976 stp_state_name(state));
1977 if (stp_learn_in_state(ofport->stp_state)
1978 != stp_learn_in_state(state)) {
1979 /* xxx Learning action flows should also be flushed. */
1980 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1981 mac_learning_flush(ofproto->ml);
1982 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1984 fwd_change = stp_forward_in_state(ofport->stp_state)
1985 != stp_forward_in_state(state);
1987 ofproto->backer->need_revalidate = REV_STP;
1988 ofport->stp_state = state;
1989 ofport->stp_state_entered = time_msec();
1991 if (fwd_change && ofport->bundle) {
1992 bundle_update(ofport->bundle);
1995 /* Update the STP state bits in the OpenFlow port description. */
1996 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1997 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1998 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1999 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2000 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2002 ofproto_port_set_state(&ofport->up, of_state);
2006 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2007 * caller is responsible for assigning STP port numbers and ensuring
2008 * there are no duplicates. */
2010 set_stp_port(struct ofport *ofport_,
2011 const struct ofproto_port_stp_settings *s)
2013 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2014 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2015 struct stp_port *sp = ofport->stp_port;
2017 if (!s || !s->enable) {
2019 ofport->stp_port = NULL;
2020 stp_port_disable(sp);
2021 update_stp_port_state(ofport);
2024 } else if (sp && stp_port_no(sp) != s->port_num
2025 && ofport == stp_port_get_aux(sp)) {
2026 /* The port-id changed, so disable the old one if it's not
2027 * already in use by another port. */
2028 stp_port_disable(sp);
2031 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2032 stp_port_enable(sp);
2034 stp_port_set_aux(sp, ofport);
2035 stp_port_set_priority(sp, s->priority);
2036 stp_port_set_path_cost(sp, s->path_cost);
2038 update_stp_port_state(ofport);
2044 get_stp_port_status(struct ofport *ofport_,
2045 struct ofproto_port_stp_status *s)
2047 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2048 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2049 struct stp_port *sp = ofport->stp_port;
2051 if (!ofproto->stp || !sp) {
2057 s->port_id = stp_port_get_id(sp);
2058 s->state = stp_port_get_state(sp);
2059 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2060 s->role = stp_port_get_role(sp);
2066 get_stp_port_stats(struct ofport *ofport_,
2067 struct ofproto_port_stp_stats *s)
2069 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2070 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2071 struct stp_port *sp = ofport->stp_port;
2073 if (!ofproto->stp || !sp) {
2079 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2085 stp_run(struct ofproto_dpif *ofproto)
2088 long long int now = time_msec();
2089 long long int elapsed = now - ofproto->stp_last_tick;
2090 struct stp_port *sp;
2093 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2094 ofproto->stp_last_tick = now;
2096 while (stp_get_changed_port(ofproto->stp, &sp)) {
2097 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2100 update_stp_port_state(ofport);
2104 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2105 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2106 mac_learning_flush(ofproto->ml);
2107 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2113 stp_wait(struct ofproto_dpif *ofproto)
2116 poll_timer_wait(1000);
2121 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2124 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2125 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2127 if (ofport->n_qdscp != n_qdscp
2128 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2129 n_qdscp * sizeof *qdscp))) {
2130 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2131 free(ofport->qdscp);
2132 ofport->qdscp = n_qdscp
2133 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2135 ofport->n_qdscp = n_qdscp;
2143 /* Expires all MAC learning entries associated with 'bundle' and forces its
2144 * ofproto to revalidate every flow.
2146 * Normally MAC learning entries are removed only from the ofproto associated
2147 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2148 * are removed from every ofproto. When patch ports and SLB bonds are in use
2149 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2150 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2151 * with the host from which it migrated. */
2153 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2155 struct ofproto_dpif *ofproto = bundle->ofproto;
2156 struct mac_learning *ml = ofproto->ml;
2157 struct mac_entry *mac, *next_mac;
2159 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2160 ovs_rwlock_wrlock(&ml->rwlock);
2161 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2162 if (mac->port.p == bundle) {
2164 struct ofproto_dpif *o;
2166 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2168 struct mac_entry *e;
2170 ovs_rwlock_wrlock(&o->ml->rwlock);
2171 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2173 mac_learning_expire(o->ml, e);
2175 ovs_rwlock_unlock(&o->ml->rwlock);
2180 mac_learning_expire(ml, mac);
2183 ovs_rwlock_unlock(&ml->rwlock);
2186 static struct ofbundle *
2187 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2189 struct ofbundle *bundle;
2191 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2192 &ofproto->bundles) {
2193 if (bundle->aux == aux) {
2201 bundle_update(struct ofbundle *bundle)
2203 struct ofport_dpif *port;
2205 bundle->floodable = true;
2206 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2207 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2209 || !stp_forward_in_state(port->stp_state)) {
2210 bundle->floodable = false;
2217 bundle_del_port(struct ofport_dpif *port)
2219 struct ofbundle *bundle = port->bundle;
2221 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2223 list_remove(&port->bundle_node);
2224 port->bundle = NULL;
2227 lacp_slave_unregister(bundle->lacp, port);
2230 bond_slave_unregister(bundle->bond, port);
2233 bundle_update(bundle);
2237 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2238 struct lacp_slave_settings *lacp)
2240 struct ofport_dpif *port;
2242 port = get_ofp_port(bundle->ofproto, ofp_port);
2247 if (port->bundle != bundle) {
2248 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2250 bundle_remove(&port->up);
2253 port->bundle = bundle;
2254 list_push_back(&bundle->ports, &port->bundle_node);
2255 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2257 || !stp_forward_in_state(port->stp_state)) {
2258 bundle->floodable = false;
2262 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2263 lacp_slave_register(bundle->lacp, port, lacp);
2270 bundle_destroy(struct ofbundle *bundle)
2272 struct ofproto_dpif *ofproto;
2273 struct ofport_dpif *port, *next_port;
2279 ofproto = bundle->ofproto;
2280 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2282 ovs_rwlock_wrlock(&xlate_rwlock);
2283 xlate_bundle_remove(bundle);
2284 ovs_rwlock_unlock(&xlate_rwlock);
2286 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2287 bundle_del_port(port);
2290 bundle_flush_macs(bundle, true);
2291 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2293 free(bundle->trunks);
2294 lacp_unref(bundle->lacp);
2295 bond_unref(bundle->bond);
2300 bundle_set(struct ofproto *ofproto_, void *aux,
2301 const struct ofproto_bundle_settings *s)
2303 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2304 bool need_flush = false;
2305 struct ofport_dpif *port;
2306 struct ofbundle *bundle;
2307 unsigned long *trunks;
2313 bundle_destroy(bundle_lookup(ofproto, aux));
2317 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2318 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2320 bundle = bundle_lookup(ofproto, aux);
2322 bundle = xmalloc(sizeof *bundle);
2324 bundle->ofproto = ofproto;
2325 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2326 hash_pointer(aux, 0));
2328 bundle->name = NULL;
2330 list_init(&bundle->ports);
2331 bundle->vlan_mode = PORT_VLAN_TRUNK;
2333 bundle->trunks = NULL;
2334 bundle->use_priority_tags = s->use_priority_tags;
2335 bundle->lacp = NULL;
2336 bundle->bond = NULL;
2338 bundle->floodable = true;
2339 mbridge_register_bundle(ofproto->mbridge, bundle);
2342 if (!bundle->name || strcmp(s->name, bundle->name)) {
2344 bundle->name = xstrdup(s->name);
2349 ofproto->lacp_enabled = true;
2350 if (!bundle->lacp) {
2351 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2352 bundle->lacp = lacp_create();
2354 lacp_configure(bundle->lacp, s->lacp);
2356 lacp_unref(bundle->lacp);
2357 bundle->lacp = NULL;
2360 /* Update set of ports. */
2362 for (i = 0; i < s->n_slaves; i++) {
2363 if (!bundle_add_port(bundle, s->slaves[i],
2364 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2368 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2369 struct ofport_dpif *next_port;
2371 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2372 for (i = 0; i < s->n_slaves; i++) {
2373 if (s->slaves[i] == port->up.ofp_port) {
2378 bundle_del_port(port);
2382 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2384 if (list_is_empty(&bundle->ports)) {
2385 bundle_destroy(bundle);
2389 /* Set VLAN tagging mode */
2390 if (s->vlan_mode != bundle->vlan_mode
2391 || s->use_priority_tags != bundle->use_priority_tags) {
2392 bundle->vlan_mode = s->vlan_mode;
2393 bundle->use_priority_tags = s->use_priority_tags;
2398 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2399 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2401 if (vlan != bundle->vlan) {
2402 bundle->vlan = vlan;
2406 /* Get trunked VLANs. */
2407 switch (s->vlan_mode) {
2408 case PORT_VLAN_ACCESS:
2412 case PORT_VLAN_TRUNK:
2413 trunks = CONST_CAST(unsigned long *, s->trunks);
2416 case PORT_VLAN_NATIVE_UNTAGGED:
2417 case PORT_VLAN_NATIVE_TAGGED:
2418 if (vlan != 0 && (!s->trunks
2419 || !bitmap_is_set(s->trunks, vlan)
2420 || bitmap_is_set(s->trunks, 0))) {
2421 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2423 trunks = bitmap_clone(s->trunks, 4096);
2425 trunks = bitmap_allocate1(4096);
2427 bitmap_set1(trunks, vlan);
2428 bitmap_set0(trunks, 0);
2430 trunks = CONST_CAST(unsigned long *, s->trunks);
2437 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2438 free(bundle->trunks);
2439 if (trunks == s->trunks) {
2440 bundle->trunks = vlan_bitmap_clone(trunks);
2442 bundle->trunks = trunks;
2447 if (trunks != s->trunks) {
2452 if (!list_is_short(&bundle->ports)) {
2453 bundle->ofproto->has_bonded_bundles = true;
2455 if (bond_reconfigure(bundle->bond, s->bond)) {
2456 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2459 bundle->bond = bond_create(s->bond, ofproto);
2460 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2463 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2464 bond_slave_register(bundle->bond, port,
2465 port->up.ofp_port, port->up.netdev);
2468 bond_unref(bundle->bond);
2469 bundle->bond = NULL;
2472 /* If we changed something that would affect MAC learning, un-learn
2473 * everything on this port and force flow revalidation. */
2475 bundle_flush_macs(bundle, false);
2482 bundle_remove(struct ofport *port_)
2484 struct ofport_dpif *port = ofport_dpif_cast(port_);
2485 struct ofbundle *bundle = port->bundle;
2488 bundle_del_port(port);
2489 if (list_is_empty(&bundle->ports)) {
2490 bundle_destroy(bundle);
2491 } else if (list_is_short(&bundle->ports)) {
2492 bond_unref(bundle->bond);
2493 bundle->bond = NULL;
2499 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2501 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2502 struct ofport_dpif *port = port_;
2503 uint8_t ea[ETH_ADDR_LEN];
2506 error = netdev_get_etheraddr(port->up.netdev, ea);
2508 struct ofpbuf packet;
2511 ofpbuf_init(&packet, 0);
2512 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2514 memcpy(packet_pdu, pdu, pdu_size);
2516 ofproto_dpif_send_packet(port, &packet);
2517 ofpbuf_uninit(&packet);
2519 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2520 "%s (%s)", port->bundle->name,
2521 netdev_get_name(port->up.netdev), ovs_strerror(error));
2526 bundle_send_learning_packets(struct ofbundle *bundle)
2528 struct ofproto_dpif *ofproto = bundle->ofproto;
2529 struct ofpbuf *learning_packet;
2530 int error, n_packets, n_errors;
2531 struct mac_entry *e;
2532 struct list packets;
2534 list_init(&packets);
2535 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2536 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2537 if (e->port.p != bundle) {
2540 learning_packet = bond_compose_learning_packet(bundle->bond,
2543 /* Temporarily use 'frame' as a private pointer (see below). */
2544 ovs_assert(learning_packet->frame == ofpbuf_data(learning_packet));
2545 learning_packet->frame = port_void;
2546 list_push_back(&packets, &learning_packet->list_node);
2549 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2551 error = n_packets = n_errors = 0;
2552 LIST_FOR_EACH (learning_packet, list_node, &packets) {
2554 void *port_void = learning_packet->frame;
2556 /* Restore 'frame'. */
2557 learning_packet->frame = ofpbuf_data(learning_packet);
2558 ret = ofproto_dpif_send_packet(port_void, learning_packet);
2565 ofpbuf_list_delete(&packets);
2568 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2569 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2570 "packets, last error was: %s",
2571 bundle->name, n_errors, n_packets, ovs_strerror(error));
2573 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2574 bundle->name, n_packets);
2579 bundle_run(struct ofbundle *bundle)
2582 lacp_run(bundle->lacp, send_pdu_cb);
2585 struct ofport_dpif *port;
2587 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2588 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2591 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2592 bundle->ofproto->backer->need_revalidate = REV_BOND;
2595 if (bond_should_send_learning_packets(bundle->bond)) {
2596 bundle_send_learning_packets(bundle);
2602 bundle_wait(struct ofbundle *bundle)
2605 lacp_wait(bundle->lacp);
2608 bond_wait(bundle->bond);
2615 mirror_set__(struct ofproto *ofproto_, void *aux,
2616 const struct ofproto_mirror_settings *s)
2618 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2619 struct ofbundle **srcs, **dsts;
2624 mirror_destroy(ofproto->mbridge, aux);
2628 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2629 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2631 for (i = 0; i < s->n_srcs; i++) {
2632 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2635 for (i = 0; i < s->n_dsts; i++) {
2636 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2639 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2640 s->n_dsts, s->src_vlans,
2641 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2648 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2649 uint64_t *packets, uint64_t *bytes)
2651 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2656 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2658 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2659 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2660 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2661 mac_learning_flush(ofproto->ml);
2663 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2668 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2670 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2671 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2672 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2676 forward_bpdu_changed(struct ofproto *ofproto_)
2678 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2679 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2683 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2686 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2687 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2688 mac_learning_set_idle_time(ofproto->ml, idle_time);
2689 mac_learning_set_max_entries(ofproto->ml, max_entries);
2690 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2695 static struct ofport_dpif *
2696 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2698 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2699 return ofport ? ofport_dpif_cast(ofport) : NULL;
2703 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2704 struct ofproto_port *ofproto_port,
2705 struct dpif_port *dpif_port)
2707 ofproto_port->name = dpif_port->name;
2708 ofproto_port->type = dpif_port->type;
2709 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2713 ofport_update_peer(struct ofport_dpif *ofport)
2715 const struct ofproto_dpif *ofproto;
2716 struct dpif_backer *backer;
2719 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2723 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2724 backer->need_revalidate = REV_RECONFIGURE;
2727 ofport->peer->peer = NULL;
2728 ofport->peer = NULL;
2731 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2736 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2737 struct ofport *peer_ofport;
2738 struct ofport_dpif *peer;
2741 if (ofproto->backer != backer) {
2745 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2750 peer = ofport_dpif_cast(peer_ofport);
2751 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2752 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2754 ofport->peer = peer;
2755 ofport->peer->peer = ofport;
2765 port_run(struct ofport_dpif *ofport)
2767 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2768 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2769 bool enable = netdev_get_carrier(ofport->up.netdev);
2770 bool cfm_enable = false;
2771 bool bfd_enable = false;
2773 ofport->carrier_seq = carrier_seq;
2776 int cfm_opup = cfm_get_opup(ofport->cfm);
2778 cfm_enable = !cfm_get_fault(ofport->cfm);
2780 if (cfm_opup >= 0) {
2781 cfm_enable = cfm_enable && cfm_opup;
2786 bfd_enable = bfd_forwarding(ofport->bfd);
2789 if (ofport->bfd || ofport->cfm) {
2790 enable = enable && (cfm_enable || bfd_enable);
2793 if (ofport->bundle) {
2794 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2795 if (carrier_changed) {
2796 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2800 if (ofport->may_enable != enable) {
2801 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2802 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2805 ofport->may_enable = enable;
2809 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2810 struct ofproto_port *ofproto_port)
2812 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2813 struct dpif_port dpif_port;
2816 if (sset_contains(&ofproto->ghost_ports, devname)) {
2817 const char *type = netdev_get_type_from_name(devname);
2819 /* We may be called before ofproto->up.port_by_name is populated with
2820 * the appropriate ofport. For this reason, we must get the name and
2821 * type from the netdev layer directly. */
2823 const struct ofport *ofport;
2825 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2826 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2827 ofproto_port->name = xstrdup(devname);
2828 ofproto_port->type = xstrdup(type);
2834 if (!sset_contains(&ofproto->ports, devname)) {
2837 error = dpif_port_query_by_name(ofproto->backer->dpif,
2838 devname, &dpif_port);
2840 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2846 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2848 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2849 const char *devname = netdev_get_name(netdev);
2850 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2851 const char *dp_port_name;
2853 if (netdev_vport_is_patch(netdev)) {
2854 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2858 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2859 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2860 odp_port_t port_no = ODPP_NONE;
2863 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2867 if (netdev_get_tunnel_config(netdev)) {
2868 simap_put(&ofproto->backer->tnl_backers,
2869 dp_port_name, odp_to_u32(port_no));
2873 if (netdev_get_tunnel_config(netdev)) {
2874 sset_add(&ofproto->ghost_ports, devname);
2876 sset_add(&ofproto->ports, devname);
2882 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
2884 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2885 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2892 sset_find_and_delete(&ofproto->ghost_ports,
2893 netdev_get_name(ofport->up.netdev));
2894 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2895 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
2896 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
2898 /* The caller is going to close ofport->up.netdev. If this is a
2899 * bonded port, then the bond is using that netdev, so remove it
2900 * from the bond. The client will need to reconfigure everything
2901 * after deleting ports, so then the slave will get re-added. */
2902 bundle_remove(&ofport->up);
2909 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2911 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2914 error = netdev_get_stats(ofport->up.netdev, stats);
2916 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
2917 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2919 ovs_mutex_lock(&ofproto->stats_mutex);
2920 /* ofproto->stats.tx_packets represents packets that we created
2921 * internally and sent to some port (e.g. packets sent with
2922 * ofproto_dpif_send_packet()). Account for them as if they had
2923 * come from OFPP_LOCAL and got forwarded. */
2925 if (stats->rx_packets != UINT64_MAX) {
2926 stats->rx_packets += ofproto->stats.tx_packets;
2929 if (stats->rx_bytes != UINT64_MAX) {
2930 stats->rx_bytes += ofproto->stats.tx_bytes;
2933 /* ofproto->stats.rx_packets represents packets that were received on
2934 * some port and we processed internally and dropped (e.g. STP).
2935 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2937 if (stats->tx_packets != UINT64_MAX) {
2938 stats->tx_packets += ofproto->stats.rx_packets;
2941 if (stats->tx_bytes != UINT64_MAX) {
2942 stats->tx_bytes += ofproto->stats.rx_bytes;
2944 ovs_mutex_unlock(&ofproto->stats_mutex);
2950 struct port_dump_state {
2955 struct ofproto_port port;
2960 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
2962 *statep = xzalloc(sizeof(struct port_dump_state));
2967 port_dump_next(const struct ofproto *ofproto_, void *state_,
2968 struct ofproto_port *port)
2970 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2971 struct port_dump_state *state = state_;
2972 const struct sset *sset;
2973 struct sset_node *node;
2975 if (state->has_port) {
2976 ofproto_port_destroy(&state->port);
2977 state->has_port = false;
2979 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
2980 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
2983 error = port_query_by_name(ofproto_, node->name, &state->port);
2985 *port = state->port;
2986 state->has_port = true;
2988 } else if (error != ENODEV) {
2993 if (!state->ghost) {
2994 state->ghost = true;
2997 return port_dump_next(ofproto_, state_, port);
3004 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3006 struct port_dump_state *state = state_;
3008 if (state->has_port) {
3009 ofproto_port_destroy(&state->port);
3016 port_poll(const struct ofproto *ofproto_, char **devnamep)
3018 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3020 if (ofproto->port_poll_errno) {
3021 int error = ofproto->port_poll_errno;
3022 ofproto->port_poll_errno = 0;
3026 if (sset_is_empty(&ofproto->port_poll_set)) {
3030 *devnamep = sset_pop(&ofproto->port_poll_set);
3035 port_poll_wait(const struct ofproto *ofproto_)
3037 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3038 dpif_port_poll_wait(ofproto->backer->dpif);
3042 port_is_lacp_current(const struct ofport *ofport_)
3044 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3045 return (ofport->bundle && ofport->bundle->lacp
3046 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3050 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3051 * then delete it entirely. */
3053 rule_expire(struct rule_dpif *rule)
3054 OVS_REQUIRES(ofproto_mutex)
3056 uint16_t hard_timeout, idle_timeout;
3057 long long int now = time_msec();
3060 ovs_assert(!rule->up.pending);
3062 hard_timeout = rule->up.hard_timeout;
3063 idle_timeout = rule->up.idle_timeout;
3065 /* Has 'rule' expired? */
3067 long long int modified;
3069 ovs_mutex_lock(&rule->up.mutex);
3070 modified = rule->up.modified;
3071 ovs_mutex_unlock(&rule->up.mutex);
3073 if (now > modified + hard_timeout * 1000) {
3074 reason = OFPRR_HARD_TIMEOUT;
3078 if (reason < 0 && idle_timeout) {
3081 ovs_mutex_lock(&rule->stats_mutex);
3082 used = rule->stats.used;
3083 ovs_mutex_unlock(&rule->stats_mutex);
3085 if (now > used + idle_timeout * 1000) {
3086 reason = OFPRR_IDLE_TIMEOUT;
3091 COVERAGE_INC(ofproto_dpif_expired);
3092 ofproto_rule_expire(&rule->up, reason);
3096 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3097 * 'flow' must reflect the data in 'packet'. */
3099 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3100 const struct flow *flow,
3101 struct rule_dpif *rule,
3102 const struct ofpact *ofpacts, size_t ofpacts_len,
3103 struct ofpbuf *packet)
3105 struct dpif_flow_stats stats;
3106 struct xlate_out xout;
3107 struct xlate_in xin;
3109 struct dpif_execute execute;
3112 ovs_assert((rule != NULL) != (ofpacts != NULL));
3114 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3117 rule_dpif_credit_stats(rule, &stats);
3120 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
3121 xin.ofpacts = ofpacts;
3122 xin.ofpacts_len = ofpacts_len;
3123 xin.resubmit_stats = &stats;
3124 xlate_actions(&xin, &xout);
3126 execute.actions = ofpbuf_data(&xout.odp_actions);
3127 execute.actions_len = ofpbuf_size(&xout.odp_actions);
3128 execute.packet = packet;
3129 execute.md = pkt_metadata_from_flow(flow);
3130 execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
3132 /* Fix up in_port. */
3133 in_port = flow->in_port.ofp_port;
3134 if (in_port == OFPP_NONE) {
3135 in_port = OFPP_LOCAL;
3137 execute.md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
3139 error = dpif_execute(ofproto->backer->dpif, &execute);
3141 xlate_out_uninit(&xout);
3147 rule_dpif_credit_stats(struct rule_dpif *rule,
3148 const struct dpif_flow_stats *stats)
3150 ovs_mutex_lock(&rule->stats_mutex);
3151 rule->stats.n_packets += stats->n_packets;
3152 rule->stats.n_bytes += stats->n_bytes;
3153 rule->stats.used = MAX(rule->stats.used, stats->used);
3154 ovs_mutex_unlock(&rule->stats_mutex);
3158 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
3159 OVS_REQUIRES(rule->up.mutex)
3161 return rule->up.flow_cookie;
3165 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
3166 uint16_t hard_timeout)
3168 ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
3171 /* Returns 'rule''s actions. The caller owns a reference on the returned
3172 * actions and must eventually release it (with rule_actions_unref()) to avoid
3174 const struct rule_actions *
3175 rule_dpif_get_actions(const struct rule_dpif *rule)
3177 return rule_get_actions(&rule->up);
3180 /* Lookup 'flow' in table 0 of 'ofproto''s classifier.
3181 * If 'wc' is non-null, sets the fields that were relevant as part of
3182 * the lookup. Returns the table_id where a match or miss occurred.
3184 * The return value will be zero unless there was a miss and
3185 * OFPTC11_TABLE_MISS_CONTINUE is in effect for the sequence of tables
3186 * where misses occur.
3188 * The rule is returned in '*rule', which is valid at least until the next
3189 * RCU quiescent period. If the '*rule' needs to stay around longer,
3190 * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3191 * on it before this returns. */
3193 rule_dpif_lookup(struct ofproto_dpif *ofproto, struct flow *flow,
3194 struct flow_wildcards *wc, struct rule_dpif **rule,
3197 enum rule_dpif_lookup_verdict verdict;
3198 enum ofputil_port_config config = 0;
3201 if (ofproto_dpif_get_enable_recirc(ofproto)) {
3202 /* Always exactly match recirc_id since datapath supports
3205 wc->masks.recirc_id = UINT32_MAX;
3208 /* Start looking up from internal table for post recirculation flows
3209 * or packets. We can also simply send all, including normal flows
3210 * or packets to the internal table. They will not match any post
3211 * recirculation rules except the 'catch all' rule that resubmit
3214 * As an optimization, we send normal flows and packets to table 0
3215 * directly, saving one table lookup. */
3216 table_id = flow->recirc_id ? TBL_INTERNAL : 0;
3221 verdict = rule_dpif_lookup_from_table(ofproto, flow, wc, true,
3222 &table_id, rule, take_ref);
3225 case RULE_DPIF_LOOKUP_VERDICT_MATCH:
3227 case RULE_DPIF_LOOKUP_VERDICT_CONTROLLER: {
3228 struct ofport_dpif *port;
3230 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
3232 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
3233 flow->in_port.ofp_port);
3235 config = port ? port->up.pp.config : 0;
3238 case RULE_DPIF_LOOKUP_VERDICT_DROP:
3239 config = OFPUTIL_PC_NO_PACKET_IN;
3241 case RULE_DPIF_LOOKUP_VERDICT_DEFAULT:
3242 if (!connmgr_wants_packet_in_on_miss(ofproto->up.connmgr)) {
3243 config = OFPUTIL_PC_NO_PACKET_IN;
3250 choose_miss_rule(config, ofproto->miss_rule,
3251 ofproto->no_packet_in_rule, rule, take_ref);
3255 /* The returned rule is valid at least until the next RCU quiescent period.
3256 * If the '*rule' needs to stay around longer, a non-zero 'take_ref' must be
3257 * passed in to cause a reference to be taken on it before this returns. */
3258 static struct rule_dpif *
3259 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, uint8_t table_id,
3260 const struct flow *flow, struct flow_wildcards *wc,
3263 struct classifier *cls = &ofproto->up.tables[table_id].cls;
3264 const struct cls_rule *cls_rule;
3265 struct rule_dpif *rule;
3267 fat_rwlock_rdlock(&cls->rwlock);
3268 if (ofproto->up.frag_handling != OFPC_FRAG_NX_MATCH) {
3270 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
3271 if (is_ip_any(flow)) {
3272 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
3276 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
3277 if (ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3278 /* We must pretend that transport ports are unavailable. */
3279 struct flow ofpc_normal_flow = *flow;
3280 ofpc_normal_flow.tp_src = htons(0);
3281 ofpc_normal_flow.tp_dst = htons(0);
3282 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
3284 /* Must be OFPC_FRAG_DROP (we don't have OFPC_FRAG_REASM). */
3285 cls_rule = &ofproto->drop_frags_rule->up.cr;
3288 cls_rule = classifier_lookup(cls, flow, wc);
3291 cls_rule = classifier_lookup(cls, flow, wc);
3294 rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
3296 rule_dpif_ref(rule);
3298 fat_rwlock_unlock(&cls->rwlock);
3303 /* Look up 'flow' in 'ofproto''s classifier starting from table '*table_id'.
3304 * Stores the rule that was found in '*rule', or NULL if none was found.
3305 * Updates 'wc', if nonnull, to reflect the fields that were used during the
3308 * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3309 * if none is found then the table miss configuration for that table is
3310 * honored, which can result in additional lookups in other OpenFlow tables.
3311 * In this case the function updates '*table_id' to reflect the final OpenFlow
3312 * table that was searched.
3314 * If 'honor_table_miss' is false, then only one table lookup occurs, in
3319 * - RULE_DPIF_LOOKUP_VERDICT_MATCH if a rule (in '*rule') was found.
3321 * - RULE_OFPTC_TABLE_MISS_CONTROLLER if no rule was found and either:
3322 * + 'honor_table_miss' is false
3323 * + a table miss configuration specified that the packet should be
3324 * sent to the controller in this case.
3326 * - RULE_DPIF_LOOKUP_VERDICT_DROP if no rule was found, 'honor_table_miss'
3327 * is true and a table miss configuration specified that the packet
3328 * should be dropped in this case.
3330 * - RULE_DPIF_LOOKUP_VERDICT_DEFAULT if no rule was found,
3331 * 'honor_table_miss' is true and a table miss configuration has
3332 * not been specified in this case.
3334 * The rule is returned in '*rule', which is valid at least until the next
3335 * RCU quiescent period. If the '*rule' needs to stay around longer,
3336 * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3337 * on it before this returns. */
3338 enum rule_dpif_lookup_verdict
3339 rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto,
3340 const struct flow *flow,
3341 struct flow_wildcards *wc,
3342 bool honor_table_miss,
3343 uint8_t *table_id, struct rule_dpif **rule,
3348 for (next_id = *table_id;
3349 next_id < ofproto->up.n_tables;
3350 next_id++, next_id += (next_id == TBL_INTERNAL))
3352 *table_id = next_id;
3353 *rule = rule_dpif_lookup_in_table(ofproto, *table_id, flow, wc,
3356 return RULE_DPIF_LOOKUP_VERDICT_MATCH;
3357 } else if (!honor_table_miss) {
3358 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3360 switch (ofproto_table_get_config(&ofproto->up, *table_id)) {
3361 case OFPROTO_TABLE_MISS_CONTINUE:
3364 case OFPROTO_TABLE_MISS_CONTROLLER:
3365 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3367 case OFPROTO_TABLE_MISS_DROP:
3368 return RULE_DPIF_LOOKUP_VERDICT_DROP;
3370 case OFPROTO_TABLE_MISS_DEFAULT:
3371 return RULE_DPIF_LOOKUP_VERDICT_DEFAULT;
3376 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3379 /* Given a port configuration (specified as zero if there's no port), chooses
3380 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
3383 * The rule is returned in '*rule', which is valid at least until the next
3384 * RCU quiescent period. If the '*rule' needs to stay around longer,
3385 * a reference must be taken on it (rule_dpif_ref()).
3388 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
3389 struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule,
3392 *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
3394 rule_dpif_ref(*rule);
3399 complete_operation(struct rule_dpif *rule)
3400 OVS_REQUIRES(ofproto_mutex)
3402 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3404 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3405 ofoperation_complete(rule->up.pending, 0);
3408 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
3410 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
3413 static struct rule *
3416 struct rule_dpif *rule = xmalloc(sizeof *rule);
3421 rule_dealloc(struct rule *rule_)
3423 struct rule_dpif *rule = rule_dpif_cast(rule_);
3428 rule_construct(struct rule *rule_)
3429 OVS_NO_THREAD_SAFETY_ANALYSIS
3431 struct rule_dpif *rule = rule_dpif_cast(rule_);
3432 ovs_mutex_init_adaptive(&rule->stats_mutex);
3433 rule->stats.n_packets = 0;
3434 rule->stats.n_bytes = 0;
3435 rule->stats.used = rule->up.modified;
3440 rule_insert(struct rule *rule_)
3441 OVS_REQUIRES(ofproto_mutex)
3443 struct rule_dpif *rule = rule_dpif_cast(rule_);
3444 complete_operation(rule);
3448 rule_delete(struct rule *rule_)
3449 OVS_REQUIRES(ofproto_mutex)
3451 struct rule_dpif *rule = rule_dpif_cast(rule_);
3452 complete_operation(rule);
3456 rule_destruct(struct rule *rule_)
3458 struct rule_dpif *rule = rule_dpif_cast(rule_);
3459 ovs_mutex_destroy(&rule->stats_mutex);
3463 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
3464 long long int *used)
3466 struct rule_dpif *rule = rule_dpif_cast(rule_);
3468 ovs_mutex_lock(&rule->stats_mutex);
3469 *packets = rule->stats.n_packets;
3470 *bytes = rule->stats.n_bytes;
3471 *used = rule->stats.used;
3472 ovs_mutex_unlock(&rule->stats_mutex);
3476 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
3477 struct ofpbuf *packet)
3479 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3481 ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
3485 rule_execute(struct rule *rule, const struct flow *flow,
3486 struct ofpbuf *packet)
3488 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
3489 ofpbuf_delete(packet);
3494 rule_modify_actions(struct rule *rule_, bool reset_counters)
3495 OVS_REQUIRES(ofproto_mutex)
3497 struct rule_dpif *rule = rule_dpif_cast(rule_);
3499 if (reset_counters) {
3500 ovs_mutex_lock(&rule->stats_mutex);
3501 rule->stats.n_packets = 0;
3502 rule->stats.n_bytes = 0;
3503 ovs_mutex_unlock(&rule->stats_mutex);
3506 complete_operation(rule);
3509 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
3511 return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
3514 static struct ofgroup *
3517 struct group_dpif *group = xzalloc(sizeof *group);
3522 group_dealloc(struct ofgroup *group_)
3524 struct group_dpif *group = group_dpif_cast(group_);
3529 group_construct_stats(struct group_dpif *group)
3530 OVS_REQUIRES(group->stats_mutex)
3532 group->packet_count = 0;
3533 group->byte_count = 0;
3534 if (!group->bucket_stats) {
3535 group->bucket_stats = xcalloc(group->up.n_buckets,
3536 sizeof *group->bucket_stats);
3538 memset(group->bucket_stats, 0, group->up.n_buckets *
3539 sizeof *group->bucket_stats);
3544 group_construct(struct ofgroup *group_)
3546 struct group_dpif *group = group_dpif_cast(group_);
3547 const struct ofputil_bucket *bucket;
3549 /* Prevent group chaining because our locking structure makes it hard to
3550 * implement deadlock-free. (See xlate_group_resource_check().) */
3551 LIST_FOR_EACH (bucket, list_node, &group->up.buckets) {
3552 const struct ofpact *a;
3554 OFPACT_FOR_EACH (a, bucket->ofpacts, bucket->ofpacts_len) {
3555 if (a->type == OFPACT_GROUP) {
3556 return OFPERR_OFPGMFC_CHAINING_UNSUPPORTED;
3561 ovs_mutex_init_adaptive(&group->stats_mutex);
3562 ovs_mutex_lock(&group->stats_mutex);
3563 group_construct_stats(group);
3564 ovs_mutex_unlock(&group->stats_mutex);
3569 group_destruct__(struct group_dpif *group)
3570 OVS_REQUIRES(group->stats_mutex)
3572 free(group->bucket_stats);
3573 group->bucket_stats = NULL;
3577 group_destruct(struct ofgroup *group_)
3579 struct group_dpif *group = group_dpif_cast(group_);
3580 ovs_mutex_lock(&group->stats_mutex);
3581 group_destruct__(group);
3582 ovs_mutex_unlock(&group->stats_mutex);
3583 ovs_mutex_destroy(&group->stats_mutex);
3587 group_modify(struct ofgroup *group_, struct ofgroup *victim_)
3589 struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
3590 struct group_dpif *group = group_dpif_cast(group_);
3591 struct group_dpif *victim = group_dpif_cast(victim_);
3593 ovs_mutex_lock(&group->stats_mutex);
3594 if (victim->up.n_buckets < group->up.n_buckets) {
3595 group_destruct__(group);
3597 group_construct_stats(group);
3598 ovs_mutex_unlock(&group->stats_mutex);
3600 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3606 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
3608 struct group_dpif *group = group_dpif_cast(group_);
3610 ovs_mutex_lock(&group->stats_mutex);
3611 ogs->packet_count = group->packet_count;
3612 ogs->byte_count = group->byte_count;
3613 memcpy(ogs->bucket_stats, group->bucket_stats,
3614 group->up.n_buckets * sizeof *group->bucket_stats);
3615 ovs_mutex_unlock(&group->stats_mutex);
3621 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
3622 struct group_dpif **group)
3623 OVS_TRY_RDLOCK(true, (*group)->up.rwlock)
3625 struct ofgroup *ofgroup;
3629 found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
3630 *group = found ? group_dpif_cast(ofgroup) : NULL;
3636 group_dpif_release(struct group_dpif *group)
3637 OVS_RELEASES(group->up.rwlock)
3639 ofproto_group_release(&group->up);
3643 group_dpif_get_buckets(const struct group_dpif *group,
3644 const struct list **buckets)
3646 *buckets = &group->up.buckets;
3649 enum ofp11_group_type
3650 group_dpif_get_type(const struct group_dpif *group)
3652 return group->up.type;
3655 /* Sends 'packet' out 'ofport'.
3656 * May modify 'packet'.
3657 * Returns 0 if successful, otherwise a positive errno value. */
3659 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
3661 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3664 error = xlate_send_packet(ofport, packet);
3666 ovs_mutex_lock(&ofproto->stats_mutex);
3667 ofproto->stats.tx_packets++;
3668 ofproto->stats.tx_bytes += ofpbuf_size(packet);
3669 ovs_mutex_unlock(&ofproto->stats_mutex);
3674 set_frag_handling(struct ofproto *ofproto_,
3675 enum ofp_config_flags frag_handling)
3677 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3678 if (frag_handling != OFPC_FRAG_REASM) {
3679 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3687 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
3688 const struct flow *flow,
3689 const struct ofpact *ofpacts, size_t ofpacts_len)
3691 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3693 ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
3694 ofpacts_len, packet);
3701 set_netflow(struct ofproto *ofproto_,
3702 const struct netflow_options *netflow_options)
3704 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3706 if (netflow_options) {
3707 if (!ofproto->netflow) {
3708 ofproto->netflow = netflow_create();
3709 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3711 return netflow_set_options(ofproto->netflow, netflow_options);
3712 } else if (ofproto->netflow) {
3713 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3714 netflow_unref(ofproto->netflow);
3715 ofproto->netflow = NULL;
3722 get_netflow_ids(const struct ofproto *ofproto_,
3723 uint8_t *engine_type, uint8_t *engine_id)
3725 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3727 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
3730 static struct ofproto_dpif *
3731 ofproto_dpif_lookup(const char *name)
3733 struct ofproto_dpif *ofproto;
3735 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
3736 hash_string(name, 0), &all_ofproto_dpifs) {
3737 if (!strcmp(ofproto->up.name, name)) {
3745 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
3746 const char *argv[], void *aux OVS_UNUSED)
3748 struct ofproto_dpif *ofproto;
3751 ofproto = ofproto_dpif_lookup(argv[1]);
3753 unixctl_command_reply_error(conn, "no such bridge");
3756 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3757 mac_learning_flush(ofproto->ml);
3758 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3760 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3761 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3762 mac_learning_flush(ofproto->ml);
3763 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3767 unixctl_command_reply(conn, "table successfully flushed");
3770 static struct ofport_dpif *
3771 ofbundle_get_a_port(const struct ofbundle *bundle)
3773 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
3778 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
3779 const char *argv[], void *aux OVS_UNUSED)
3781 struct ds ds = DS_EMPTY_INITIALIZER;
3782 const struct ofproto_dpif *ofproto;
3783 const struct mac_entry *e;
3785 ofproto = ofproto_dpif_lookup(argv[1]);
3787 unixctl_command_reply_error(conn, "no such bridge");
3791 ds_put_cstr(&ds, " port VLAN MAC Age\n");
3792 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
3793 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3794 struct ofbundle *bundle = e->port.p;
3795 char name[OFP_MAX_PORT_NAME_LEN];
3797 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
3799 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
3800 name, e->vlan, ETH_ADDR_ARGS(e->mac),
3801 mac_entry_age(ofproto->ml, e));
3803 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3804 unixctl_command_reply(conn, ds_cstr(&ds));
3809 struct xlate_out xout;
3810 struct xlate_in xin;
3811 const struct flow *key;
3813 struct flow_wildcards wc;
3818 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
3820 const struct rule_actions *actions;
3823 ds_put_char_multiple(result, '\t', level);
3825 ds_put_cstr(result, "No match\n");
3829 ovs_mutex_lock(&rule->up.mutex);
3830 cookie = rule->up.flow_cookie;
3831 ovs_mutex_unlock(&rule->up.mutex);
3833 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
3834 rule ? rule->up.table_id : 0, ntohll(cookie));
3835 cls_rule_format(&rule->up.cr, result);
3836 ds_put_char(result, '\n');
3838 actions = rule_dpif_get_actions(rule);
3840 ds_put_char_multiple(result, '\t', level);
3841 ds_put_cstr(result, "OpenFlow actions=");
3842 ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
3843 ds_put_char(result, '\n');
3847 trace_format_flow(struct ds *result, int level, const char *title,
3848 struct trace_ctx *trace)
3850 ds_put_char_multiple(result, '\t', level);
3851 ds_put_format(result, "%s: ", title);
3852 /* Do not report unchanged flows for resubmits. */
3853 if ((level > 0 && flow_equal(&trace->xin.flow, &trace->flow))
3854 || (level == 0 && flow_equal(&trace->xin.flow, trace->key))) {
3855 ds_put_cstr(result, "unchanged");
3857 flow_format(result, &trace->xin.flow);
3858 trace->flow = trace->xin.flow;
3860 ds_put_char(result, '\n');
3864 trace_format_regs(struct ds *result, int level, const char *title,
3865 struct trace_ctx *trace)
3869 ds_put_char_multiple(result, '\t', level);
3870 ds_put_format(result, "%s:", title);
3871 for (i = 0; i < FLOW_N_REGS; i++) {
3872 ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
3874 ds_put_char(result, '\n');
3878 trace_format_odp(struct ds *result, int level, const char *title,
3879 struct trace_ctx *trace)
3881 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
3883 ds_put_char_multiple(result, '\t', level);
3884 ds_put_format(result, "%s: ", title);
3885 format_odp_actions(result, ofpbuf_data(odp_actions),
3886 ofpbuf_size(odp_actions));
3887 ds_put_char(result, '\n');
3891 trace_format_megaflow(struct ds *result, int level, const char *title,
3892 struct trace_ctx *trace)
3896 ds_put_char_multiple(result, '\t', level);
3897 ds_put_format(result, "%s: ", title);
3898 flow_wildcards_or(&trace->wc, &trace->xout.wc, &trace->wc);
3899 match_init(&match, trace->key, &trace->wc);
3900 match_format(&match, result, OFP_DEFAULT_PRIORITY);
3901 ds_put_char(result, '\n');
3905 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
3907 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
3908 struct ds *result = trace->result;
3910 ds_put_char(result, '\n');
3911 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
3912 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
3913 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
3914 trace_format_megaflow(result, recurse + 1, "Resubmitted megaflow", trace);
3915 trace_format_rule(result, recurse + 1, rule);
3919 trace_report(struct xlate_in *xin, const char *s, int recurse)
3921 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
3922 struct ds *result = trace->result;
3924 ds_put_char_multiple(result, '\t', recurse);
3925 ds_put_cstr(result, s);
3926 ds_put_char(result, '\n');
3929 /* Parses the 'argc' elements of 'argv', ignoring argv[0]. The following
3930 * forms are supported:
3932 * - [dpname] odp_flow [-generate | packet]
3933 * - bridge br_flow [-generate | packet]
3935 * On success, initializes '*ofprotop' and 'flow' and returns NULL. On failure
3936 * returns a nonnull malloced error message. */
3937 static char * WARN_UNUSED_RESULT
3938 parse_flow_and_packet(int argc, const char *argv[],
3939 struct ofproto_dpif **ofprotop, struct flow *flow,
3940 struct ofpbuf **packetp)
3942 const struct dpif_backer *backer = NULL;
3943 const char *error = NULL;
3945 struct simap port_names = SIMAP_INITIALIZER(&port_names);
3946 struct ofpbuf *packet;
3947 struct ofpbuf odp_key;
3948 struct ofpbuf odp_mask;
3950 ofpbuf_init(&odp_key, 0);
3951 ofpbuf_init(&odp_mask, 0);
3953 /* Handle "-generate" or a hex string as the last argument. */
3954 if (!strcmp(argv[argc - 1], "-generate")) {
3955 packet = ofpbuf_new(0);
3958 error = eth_from_hex(argv[argc - 1], &packet);
3961 } else if (argc == 4) {
3962 /* The 3-argument form must end in "-generate' or a hex string. */
3968 /* odp_flow can have its in_port specified as a name instead of port no.
3969 * We do not yet know whether a given flow is a odp_flow or a br_flow.
3970 * But, to know whether a flow is odp_flow through odp_flow_from_string(),
3971 * we need to create a simap of name to port no. */
3973 const char *dp_type;
3974 if (!strncmp(argv[1], "ovs-", 4)) {
3975 dp_type = argv[1] + 4;
3979 backer = shash_find_data(&all_dpif_backers, dp_type);
3980 } else if (argc == 2) {
3981 struct shash_node *node;
3982 if (shash_count(&all_dpif_backers) == 1) {
3983 node = shash_first(&all_dpif_backers);
3984 backer = node->data;
3987 error = "Syntax error";
3990 if (backer && backer->dpif) {
3991 struct dpif_port dpif_port;
3992 struct dpif_port_dump port_dump;
3993 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
3994 simap_put(&port_names, dpif_port.name,
3995 odp_to_u32(dpif_port.port_no));
3999 /* Parse the flow and determine whether a datapath or
4000 * bridge is specified. If function odp_flow_key_from_string()
4001 * returns 0, the flow is a odp_flow. If function
4002 * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
4003 if (!odp_flow_from_string(argv[argc - 1], &port_names,
4004 &odp_key, &odp_mask)) {
4006 error = "Cannot find the datapath";
4010 if (xlate_receive(backer, NULL, ofpbuf_data(&odp_key),
4011 ofpbuf_size(&odp_key), flow,
4012 ofprotop, NULL, NULL, NULL, NULL)) {
4013 error = "Invalid datapath flow";
4017 char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
4020 m_err = xasprintf("Bad flow syntax: %s", err);
4025 error = "Must specify bridge name";
4029 *ofprotop = ofproto_dpif_lookup(argv[1]);
4031 error = "Unknown bridge name";
4037 /* Generate a packet, if requested. */
4039 if (!ofpbuf_size(packet)) {
4040 flow_compose(packet, flow);
4042 struct pkt_metadata md = pkt_metadata_from_flow(flow);
4044 /* Use the metadata from the flow and the packet argument
4045 * to reconstruct the flow. */
4046 flow_extract(packet, &md, flow);
4051 if (error && !m_err) {
4052 m_err = xstrdup(error);
4055 ofpbuf_delete(packet);
4059 ofpbuf_uninit(&odp_key);
4060 ofpbuf_uninit(&odp_mask);
4061 simap_destroy(&port_names);
4066 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
4067 void *aux OVS_UNUSED)
4069 struct ofproto_dpif *ofproto;
4070 struct ofpbuf *packet;
4074 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4079 ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
4080 unixctl_command_reply(conn, ds_cstr(&result));
4081 ds_destroy(&result);
4082 ofpbuf_delete(packet);
4084 unixctl_command_reply_error(conn, error);
4090 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
4091 const char *argv[], void *aux OVS_UNUSED)
4093 enum ofputil_protocol usable_protocols;
4094 struct ofproto_dpif *ofproto;
4095 bool enforce_consistency;
4096 struct ofpbuf ofpacts;
4097 struct ofpbuf *packet;
4102 /* Three kinds of error return values! */
4108 ofpbuf_init(&ofpacts, 0);
4110 /* Parse actions. */
4111 error = parse_ofpacts(argv[--argc], &ofpacts, &usable_protocols);
4113 unixctl_command_reply_error(conn, error);
4118 /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
4119 * consistency between the flow and the actions. With -consistent, we
4120 * enforce consistency even for a flow supported in OpenFlow 1.0. */
4121 if (!strcmp(argv[1], "-consistent")) {
4122 enforce_consistency = true;
4126 enforce_consistency = false;
4129 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4131 unixctl_command_reply_error(conn, error);
4136 /* Do the same checks as handle_packet_out() in ofproto.c.
4138 * We pass a 'table_id' of 0 to ofproto_check_ofpacts(), which isn't
4139 * strictly correct because these actions aren't in any table, but it's OK
4140 * because it 'table_id' is used only to check goto_table instructions, but
4141 * packet-outs take a list of actions and therefore it can't include
4144 * We skip the "meter" check here because meter is an instruction, not an
4145 * action, and thus cannot appear in ofpacts. */
4146 in_port = ofp_to_u16(flow.in_port.ofp_port);
4147 if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
4148 unixctl_command_reply_error(conn, "invalid in_port");
4151 if (enforce_consistency) {
4152 retval = ofpacts_check_consistency(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts),
4153 &flow, u16_to_ofp(ofproto->up.max_ports),
4154 0, 0, usable_protocols);
4156 retval = ofpacts_check(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &flow,
4157 u16_to_ofp(ofproto->up.max_ports), 0, 0,
4163 ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
4164 unixctl_command_reply_error(conn, ds_cstr(&result));
4168 ofproto_trace(ofproto, &flow, packet,
4169 ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &result);
4170 unixctl_command_reply(conn, ds_cstr(&result));
4173 ds_destroy(&result);
4174 ofpbuf_delete(packet);
4175 ofpbuf_uninit(&ofpacts);
4178 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
4179 * description of the results to 'ds'.
4181 * The trace follows a packet with the specified 'flow' through the flow
4182 * table. 'packet' may be nonnull to trace an actual packet, with consequent
4183 * side effects (if it is nonnull then its flow must be 'flow').
4185 * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
4186 * trace, otherwise the actions are determined by a flow table lookup. */
4188 ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
4189 const struct ofpbuf *packet,
4190 const struct ofpact ofpacts[], size_t ofpacts_len,
4193 struct rule_dpif *rule;
4194 struct trace_ctx trace;
4196 ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
4197 ds_put_cstr(ds, "Flow: ");
4198 flow_format(ds, flow);
4199 ds_put_char(ds, '\n');
4201 flow_wildcards_init_catchall(&trace.wc);
4205 rule_dpif_lookup(ofproto, flow, &trace.wc, &rule, false);
4207 trace_format_rule(ds, 0, rule);
4208 if (rule == ofproto->miss_rule) {
4209 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
4210 } else if (rule == ofproto->no_packet_in_rule) {
4211 ds_put_cstr(ds, "\nNo match, packets dropped because "
4212 "OFPPC_NO_PACKET_IN is set on in_port.\n");
4213 } else if (rule == ofproto->drop_frags_rule) {
4214 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
4215 "and the fragment handling mode is \"drop\".\n");
4219 if (rule || ofpacts) {
4221 trace.key = flow; /* Original flow key, used for megaflow. */
4222 trace.flow = *flow; /* May be modified by actions. */
4223 xlate_in_init(&trace.xin, ofproto, flow, rule, ntohs(flow->tcp_flags),
4226 trace.xin.ofpacts = ofpacts;
4227 trace.xin.ofpacts_len = ofpacts_len;
4229 trace.xin.resubmit_hook = trace_resubmit;
4230 trace.xin.report_hook = trace_report;
4232 xlate_actions(&trace.xin, &trace.xout);
4234 ds_put_char(ds, '\n');
4235 trace_format_flow(ds, 0, "Final flow", &trace);
4236 trace_format_megaflow(ds, 0, "Megaflow", &trace);
4238 ds_put_cstr(ds, "Datapath actions: ");
4239 format_odp_actions(ds, ofpbuf_data(&trace.xout.odp_actions),
4240 ofpbuf_size(&trace.xout.odp_actions));
4242 if (trace.xout.slow) {
4243 enum slow_path_reason slow;
4245 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
4246 "slow path because it:");
4248 slow = trace.xout.slow;
4250 enum slow_path_reason bit = rightmost_1bit(slow);
4252 ds_put_format(ds, "\n\t- %s.",
4253 slow_path_reason_to_explanation(bit));
4259 xlate_out_uninit(&trace.xout);
4263 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
4264 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
4265 * to destroy 'ofproto_shash' and free the returned value. */
4266 static const struct shash_node **
4267 get_ofprotos(struct shash *ofproto_shash)
4269 const struct ofproto_dpif *ofproto;
4271 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4272 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
4273 shash_add_nocopy(ofproto_shash, name, ofproto);
4276 return shash_sort(ofproto_shash);
4280 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
4281 const char *argv[] OVS_UNUSED,
4282 void *aux OVS_UNUSED)
4284 struct ds ds = DS_EMPTY_INITIALIZER;
4285 struct shash ofproto_shash;
4286 const struct shash_node **sorted_ofprotos;
4289 shash_init(&ofproto_shash);
4290 sorted_ofprotos = get_ofprotos(&ofproto_shash);
4291 for (i = 0; i < shash_count(&ofproto_shash); i++) {
4292 const struct shash_node *node = sorted_ofprotos[i];
4293 ds_put_format(&ds, "%s\n", node->name);
4296 shash_destroy(&ofproto_shash);
4297 free(sorted_ofprotos);
4299 unixctl_command_reply(conn, ds_cstr(&ds));
4304 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
4306 const struct shash_node **ofprotos;
4307 struct dpif_dp_stats dp_stats;
4308 struct shash ofproto_shash;
4311 dpif_get_dp_stats(backer->dpif, &dp_stats);
4313 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
4314 dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
4316 shash_init(&ofproto_shash);
4317 ofprotos = get_ofprotos(&ofproto_shash);
4318 for (i = 0; i < shash_count(&ofproto_shash); i++) {
4319 struct ofproto_dpif *ofproto = ofprotos[i]->data;
4320 const struct shash_node **ports;
4323 if (ofproto->backer != backer) {
4327 ds_put_format(ds, "\t%s:\n", ofproto->up.name);
4329 ports = shash_sort(&ofproto->up.port_by_name);
4330 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
4331 const struct shash_node *node = ports[j];
4332 struct ofport *ofport = node->data;
4334 odp_port_t odp_port;
4336 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
4339 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
4340 if (odp_port != ODPP_NONE) {
4341 ds_put_format(ds, "%"PRIu32":", odp_port);
4343 ds_put_cstr(ds, "none:");
4346 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
4349 if (!netdev_get_config(ofport->netdev, &config)) {
4350 const struct smap_node **nodes;
4353 nodes = smap_sort(&config);
4354 for (i = 0; i < smap_count(&config); i++) {
4355 const struct smap_node *node = nodes[i];
4356 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
4357 node->key, node->value);
4361 smap_destroy(&config);
4363 ds_put_char(ds, ')');
4364 ds_put_char(ds, '\n');
4368 shash_destroy(&ofproto_shash);
4373 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4374 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4376 struct ds ds = DS_EMPTY_INITIALIZER;
4377 const struct shash_node **backers;
4380 backers = shash_sort(&all_dpif_backers);
4381 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
4382 dpif_show_backer(backers[i]->data, &ds);
4386 unixctl_command_reply(conn, ds_cstr(&ds));
4391 ofproto_dpif_contains_flow(const struct ofproto_dpif *ofproto,
4392 const struct nlattr *key, size_t key_len)
4394 struct ofproto_dpif *ofp;
4397 xlate_receive(ofproto->backer, NULL, key, key_len, &flow, &ofp,
4398 NULL, NULL, NULL, NULL);
4399 return ofp == ofproto;
4403 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
4404 int argc OVS_UNUSED, const char *argv[],
4405 void *aux OVS_UNUSED)
4407 struct ds ds = DS_EMPTY_INITIALIZER;
4408 const struct dpif_flow_stats *stats;
4409 const struct ofproto_dpif *ofproto;
4410 struct dpif_flow_dump flow_dump;
4411 const struct nlattr *actions;
4412 const struct nlattr *mask;
4413 const struct nlattr *key;
4417 bool verbosity = false;
4418 struct dpif_port dpif_port;
4419 struct dpif_port_dump port_dump;
4420 struct hmap portno_names;
4424 ofproto = ofproto_dpif_lookup(argv[argc - 1]);
4426 unixctl_command_reply_error(conn, "no such bridge");
4430 if (argc > 2 && !strcmp(argv[1], "-m")) {
4434 hmap_init(&portno_names);
4435 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
4436 odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
4440 error = dpif_flow_dump_start(&flow_dump, ofproto->backer->dpif);
4444 dpif_flow_dump_state_init(ofproto->backer->dpif, &state);
4445 while (dpif_flow_dump_next(&flow_dump, state, &key, &key_len,
4446 &mask, &mask_len, &actions, &actions_len,
4448 if (!ofproto_dpif_contains_flow(ofproto, key, key_len)) {
4452 odp_flow_format(key, key_len, mask, mask_len, &portno_names, &ds,
4454 ds_put_cstr(&ds, ", ");
4455 dpif_flow_stats_format(stats, &ds);
4456 ds_put_cstr(&ds, ", actions:");
4457 format_odp_actions(&ds, actions, actions_len);
4458 ds_put_char(&ds, '\n');
4460 dpif_flow_dump_state_uninit(ofproto->backer->dpif, state);
4461 error = dpif_flow_dump_done(&flow_dump);
4466 ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
4467 unixctl_command_reply_error(conn, ds_cstr(&ds));
4469 unixctl_command_reply(conn, ds_cstr(&ds));
4471 odp_portno_names_destroy(&portno_names);
4472 hmap_destroy(&portno_names);
4477 ofproto_dpif_unixctl_init(void)
4479 static bool registered;
4485 unixctl_command_register(
4487 "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
4488 1, 3, ofproto_unixctl_trace, NULL);
4489 unixctl_command_register(
4490 "ofproto/trace-packet-out",
4491 "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
4492 2, 6, ofproto_unixctl_trace_actions, NULL);
4493 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
4494 ofproto_unixctl_fdb_flush, NULL);
4495 unixctl_command_register("fdb/show", "bridge", 1, 1,
4496 ofproto_unixctl_fdb_show, NULL);
4497 unixctl_command_register("dpif/dump-dps", "", 0, 0,
4498 ofproto_unixctl_dpif_dump_dps, NULL);
4499 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
4501 unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
4502 ofproto_unixctl_dpif_dump_flows, NULL);
4505 /* Returns true if 'table' is the table used for internal rules,
4506 * false otherwise. */
4508 table_is_internal(uint8_t table_id)
4510 return table_id == TBL_INTERNAL;
4513 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4515 * This is deprecated. It is only for compatibility with broken device drivers
4516 * in old versions of Linux that do not properly support VLANs when VLAN
4517 * devices are not used. When broken device drivers are no longer in
4518 * widespread use, we will delete these interfaces. */
4521 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
4523 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
4524 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
4526 if (realdev_ofp_port == ofport->realdev_ofp_port
4527 && vid == ofport->vlandev_vid) {
4531 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4533 if (ofport->realdev_ofp_port) {
4536 if (realdev_ofp_port && ofport->bundle) {
4537 /* vlandevs are enslaved to their realdevs, so they are not allowed to
4538 * themselves be part of a bundle. */
4539 bundle_set(ofport_->ofproto, ofport->bundle, NULL);
4542 ofport->realdev_ofp_port = realdev_ofp_port;
4543 ofport->vlandev_vid = vid;
4545 if (realdev_ofp_port) {
4546 vsp_add(ofport, realdev_ofp_port, vid);
4553 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
4555 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
4559 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
4560 OVS_EXCLUDED(ofproto->vsp_mutex)
4562 /* hmap_is_empty is thread safe. */
4563 return !hmap_is_empty(&ofproto->realdev_vid_map);
4568 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
4569 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
4570 OVS_REQUIRES(ofproto->vsp_mutex)
4572 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
4573 int vid = vlan_tci_to_vid(vlan_tci);
4574 const struct vlan_splinter *vsp;
4576 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
4577 hash_realdev_vid(realdev_ofp_port, vid),
4578 &ofproto->realdev_vid_map) {
4579 if (vsp->realdev_ofp_port == realdev_ofp_port
4580 && vsp->vid == vid) {
4581 return vsp->vlandev_ofp_port;
4585 return realdev_ofp_port;
4588 /* Returns the OFP port number of the Linux VLAN device that corresponds to
4589 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
4590 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
4591 * 'vlan_tci' 9, it would return the port number of eth0.9.
4593 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
4594 * function just returns its 'realdev_ofp_port' argument. */
4596 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
4597 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
4598 OVS_EXCLUDED(ofproto->vsp_mutex)
4602 /* hmap_is_empty is thread safe, see if we can return immediately. */
4603 if (hmap_is_empty(&ofproto->realdev_vid_map)) {
4604 return realdev_ofp_port;
4606 ovs_mutex_lock(&ofproto->vsp_mutex);
4607 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
4608 ovs_mutex_unlock(&ofproto->vsp_mutex);
4612 static struct vlan_splinter *
4613 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
4615 struct vlan_splinter *vsp;
4617 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
4618 hash_ofp_port(vlandev_ofp_port),
4619 &ofproto->vlandev_map) {
4620 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
4628 /* Returns the OpenFlow port number of the "real" device underlying the Linux
4629 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
4630 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
4631 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
4632 * eth0 and store 9 in '*vid'.
4634 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
4635 * VLAN device. Unless VLAN splinters are enabled, this is what this function
4638 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
4639 ofp_port_t vlandev_ofp_port, int *vid)
4640 OVS_REQUIRES(ofproto->vsp_mutex)
4642 if (!hmap_is_empty(&ofproto->vlandev_map)) {
4643 const struct vlan_splinter *vsp;
4645 vsp = vlandev_find(ofproto, vlandev_ofp_port);
4650 return vsp->realdev_ofp_port;
4656 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
4657 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
4658 * 'flow->in_port' to the "real" device backing the VLAN device, sets
4659 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
4660 * always the case unless VLAN splinters are enabled), returns false without
4661 * making any changes. */
4663 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
4664 OVS_EXCLUDED(ofproto->vsp_mutex)
4669 /* hmap_is_empty is thread safe. */
4670 if (hmap_is_empty(&ofproto->vlandev_map)) {
4674 ovs_mutex_lock(&ofproto->vsp_mutex);
4675 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
4676 ovs_mutex_unlock(&ofproto->vsp_mutex);
4681 /* Cause the flow to be processed as if it came in on the real device with
4682 * the VLAN device's VLAN ID. */
4683 flow->in_port.ofp_port = realdev;
4684 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
4689 vsp_remove(struct ofport_dpif *port)
4691 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
4692 struct vlan_splinter *vsp;
4694 ovs_mutex_lock(&ofproto->vsp_mutex);
4695 vsp = vlandev_find(ofproto, port->up.ofp_port);
4697 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
4698 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
4701 port->realdev_ofp_port = 0;
4703 VLOG_ERR("missing vlan device record");
4705 ovs_mutex_unlock(&ofproto->vsp_mutex);
4709 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
4711 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
4713 ovs_mutex_lock(&ofproto->vsp_mutex);
4714 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
4715 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
4716 == realdev_ofp_port)) {
4717 struct vlan_splinter *vsp;
4719 vsp = xmalloc(sizeof *vsp);
4720 vsp->realdev_ofp_port = realdev_ofp_port;
4721 vsp->vlandev_ofp_port = port->up.ofp_port;
4724 port->realdev_ofp_port = realdev_ofp_port;
4726 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
4727 hash_ofp_port(port->up.ofp_port));
4728 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
4729 hash_realdev_vid(realdev_ofp_port, vid));
4731 VLOG_ERR("duplicate vlan device record");
4733 ovs_mutex_unlock(&ofproto->vsp_mutex);
4737 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
4739 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
4740 return ofport ? ofport->odp_port : ODPP_NONE;
4743 struct ofport_dpif *
4744 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
4746 struct ofport_dpif *port;
4748 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
4749 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
4750 &backer->odp_to_ofport_map) {
4751 if (port->odp_port == odp_port) {
4752 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
4757 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
4762 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
4764 struct ofport_dpif *port;
4766 port = odp_port_to_ofport(ofproto->backer, odp_port);
4767 if (port && &ofproto->up == port->up.ofproto) {
4768 return port->up.ofp_port;
4775 ofproto_dpif_alloc_recirc_id(struct ofproto_dpif *ofproto)
4777 struct dpif_backer *backer = ofproto->backer;
4779 return recirc_id_alloc(backer->rid_pool);
4783 ofproto_dpif_free_recirc_id(struct ofproto_dpif *ofproto, uint32_t recirc_id)
4785 struct dpif_backer *backer = ofproto->backer;
4787 recirc_id_free(backer->rid_pool, recirc_id);
4791 ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
4792 const struct match *match, int priority,
4793 const struct ofpbuf *ofpacts,
4794 struct rule **rulep)
4796 struct ofputil_flow_mod fm;
4797 struct rule_dpif *rule;
4801 fm.priority = priority;
4802 fm.new_cookie = htonll(0);
4803 fm.cookie = htonll(0);
4804 fm.cookie_mask = htonll(0);
4805 fm.modify_cookie = false;
4806 fm.table_id = TBL_INTERNAL;
4807 fm.command = OFPFC_ADD;
4808 fm.idle_timeout = 0;
4809 fm.hard_timeout = 0;
4812 fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
4813 fm.ofpacts = ofpbuf_data(ofpacts);
4814 fm.ofpacts_len = ofpbuf_size(ofpacts);
4816 error = ofproto_flow_mod(&ofproto->up, &fm);
4818 VLOG_ERR_RL(&rl, "failed to add internal flow (%s)",
4819 ofperr_to_string(error));
4824 rule = rule_dpif_lookup_in_table(ofproto, TBL_INTERNAL, &fm.match.flow,
4825 &fm.match.wc, false);
4835 ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
4836 struct match *match, int priority)
4838 struct ofputil_flow_mod fm;
4842 fm.priority = priority;
4843 fm.new_cookie = htonll(0);
4844 fm.cookie = htonll(0);
4845 fm.cookie_mask = htonll(0);
4846 fm.modify_cookie = false;
4847 fm.table_id = TBL_INTERNAL;
4848 fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
4849 fm.command = OFPFC_DELETE_STRICT;
4851 error = ofproto_flow_mod(&ofproto->up, &fm);
4853 VLOG_ERR_RL(&rl, "failed to delete internal flow (%s)",
4854 ofperr_to_string(error));
4861 const struct ofproto_class ofproto_dpif_class = {
4875 NULL, /* get_memory_usage. */
4876 type_get_memory_usage,
4895 port_is_lacp_current,
4896 NULL, /* rule_choose_table */
4905 rule_modify_actions,
4919 get_stp_port_status,
4927 is_mirror_output_bundle,
4928 forward_bpdu_changed,
4929 set_mac_table_config,
4931 NULL, /* meter_get_features */
4932 NULL, /* meter_set */
4933 NULL, /* meter_get */
4934 NULL, /* meter_del */
4935 group_alloc, /* group_alloc */
4936 group_construct, /* group_construct */
4937 group_destruct, /* group_destruct */
4938 group_dealloc, /* group_dealloc */
4939 group_modify, /* group_modify */
4940 group_get_stats, /* group_get_stats */