2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-dpif.h"
20 #include "ofproto/ofproto-provider.h"
27 #include "byte-order.h"
28 #include "connectivity.h"
33 #include "dynamic-string.h"
34 #include "fail-open.h"
35 #include "guarded-list.h"
39 #include "mac-learning.h"
40 #include "mcast-snooping.h"
41 #include "meta-flow.h"
42 #include "multipath.h"
43 #include "netdev-vport.h"
48 #include "odp-execute.h"
51 #include "ofp-actions.h"
52 #include "ofp-parse.h"
53 #include "ofp-print.h"
54 #include "ofproto-dpif-ipfix.h"
55 #include "ofproto-dpif-mirror.h"
56 #include "ofproto-dpif-monitor.h"
57 #include "ofproto-dpif-rid.h"
58 #include "ofproto-dpif-sflow.h"
59 #include "ofproto-dpif-upcall.h"
60 #include "ofproto-dpif-xlate.h"
61 #include "poll-loop.h"
63 #include "ovs-router.h"
69 #include "unaligned.h"
71 #include "vlan-bitmap.h"
72 #include "openvswitch/vlog.h"
74 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
76 COVERAGE_DEFINE(ofproto_dpif_expired);
77 COVERAGE_DEFINE(packet_in_overflow);
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;
91 /* If non-zero then the recirculation id that has
92 * been allocated for use with this rule.
93 * The recirculation id and associated internal flow should
94 * be freed when the rule is freed */
98 /* RULE_CAST() depends on this. */
99 BUILD_ASSERT_DECL(offsetof(struct rule_dpif, up) == 0);
101 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes,
102 long long int *used);
103 static struct rule_dpif *rule_dpif_cast(const struct rule *);
104 static void rule_expire(struct rule_dpif *);
111 * - Do include packets and bytes from datapath flows which have not
112 * recently been processed by a revalidator. */
113 struct ovs_mutex stats_mutex;
114 uint64_t packet_count OVS_GUARDED; /* Number of packets received. */
115 uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */
119 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
120 struct ofproto_dpif *ofproto; /* Owning ofproto. */
121 void *aux; /* Key supplied by ofproto's client. */
122 char *name; /* Identifier for log messages. */
125 struct ovs_list ports; /* Contains "struct ofport"s. */
126 enum port_vlan_mode vlan_mode; /* VLAN mode */
127 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
128 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
129 * NULL if all VLANs are trunked. */
130 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
131 struct bond *bond; /* Nonnull iff more than one port. */
132 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
135 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
138 static void bundle_remove(struct ofport *);
139 static void bundle_update(struct ofbundle *);
140 static void bundle_destroy(struct ofbundle *);
141 static void bundle_del_port(struct ofport_dpif *);
142 static void bundle_run(struct ofbundle *);
143 static void bundle_wait(struct ofbundle *);
144 static void bundle_flush_macs(struct ofbundle *, bool);
145 static void bundle_move(struct ofbundle *, struct ofbundle *);
147 static void stp_run(struct ofproto_dpif *ofproto);
148 static void stp_wait(struct ofproto_dpif *ofproto);
149 static int set_stp_port(struct ofport *,
150 const struct ofproto_port_stp_settings *);
152 static void rstp_run(struct ofproto_dpif *ofproto);
153 static void set_rstp_port(struct ofport *,
154 const struct ofproto_port_rstp_settings *);
157 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
161 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
162 struct ovs_list bundle_node;/* In struct ofbundle's "ports" list. */
163 struct cfm *cfm; /* Connectivity Fault Management, if any. */
164 struct bfd *bfd; /* BFD, if any. */
165 bool may_enable; /* May be enabled in bonds. */
166 bool is_tunnel; /* This port is a tunnel. */
167 bool is_layer3; /* This is a layer 3 port. */
168 long long int carrier_seq; /* Carrier status changes. */
169 struct ofport_dpif *peer; /* Peer if patch port. */
172 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
173 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
174 long long int stp_state_entered;
176 /* Rapid Spanning Tree. */
177 struct rstp_port *rstp_port; /* Rapid Spanning Tree Protocol, if any. */
178 enum rstp_state rstp_state; /* Always RSTP_DISABLED if RSTP not in use. */
180 /* Queue to DSCP mapping. */
181 struct ofproto_port_queue *qdscp;
184 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
186 * This is deprecated. It is only for compatibility with broken device
187 * drivers in old versions of Linux that do not properly support VLANs when
188 * VLAN devices are not used. When broken device drivers are no longer in
189 * widespread use, we will delete these interfaces. */
190 ofp_port_t realdev_ofp_port;
194 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
196 * This is deprecated. It is only for compatibility with broken device drivers
197 * in old versions of Linux that do not properly support VLANs when VLAN
198 * devices are not used. When broken device drivers are no longer in
199 * widespread use, we will delete these interfaces. */
200 struct vlan_splinter {
201 struct hmap_node realdev_vid_node;
202 struct hmap_node vlandev_node;
203 ofp_port_t realdev_ofp_port;
204 ofp_port_t vlandev_ofp_port;
208 static void vsp_remove(struct ofport_dpif *);
209 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
211 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
214 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
217 static struct ofport_dpif *
218 ofport_dpif_cast(const struct ofport *ofport)
220 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
223 static void port_run(struct ofport_dpif *);
224 static int set_bfd(struct ofport *, const struct smap *);
225 static int set_cfm(struct ofport *, const struct cfm_settings *);
226 static void ofport_update_peer(struct ofport_dpif *);
228 /* Reasons that we might need to revalidate every datapath flow, and
229 * corresponding coverage counters.
231 * A value of 0 means that there is no need to revalidate.
233 * It would be nice to have some cleaner way to integrate with coverage
234 * counters, but with only a few reasons I guess this is good enough for
236 enum revalidate_reason {
237 REV_RECONFIGURE = 1, /* Switch configuration changed. */
238 REV_STP, /* Spanning tree protocol port status change. */
239 REV_RSTP, /* RSTP port status change. */
240 REV_BOND, /* Bonding changed. */
241 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
242 REV_FLOW_TABLE, /* Flow table changed. */
243 REV_MAC_LEARNING, /* Mac learning changed. */
244 REV_MCAST_SNOOPING, /* Multicast snooping changed. */
246 COVERAGE_DEFINE(rev_reconfigure);
247 COVERAGE_DEFINE(rev_stp);
248 COVERAGE_DEFINE(rev_rstp);
249 COVERAGE_DEFINE(rev_bond);
250 COVERAGE_DEFINE(rev_port_toggled);
251 COVERAGE_DEFINE(rev_flow_table);
252 COVERAGE_DEFINE(rev_mac_learning);
253 COVERAGE_DEFINE(rev_mcast_snooping);
255 /* Stores mapping between 'recirc_id' and 'ofproto-dpif'. */
256 struct dpif_backer_recirc_node {
257 struct cmap_node cmap_node;
258 struct ofproto_dpif *ofproto;
262 /* All datapaths of a given type share a single dpif backer instance. */
269 struct ovs_rwlock odp_to_ofport_lock;
270 struct hmap odp_to_ofport_map OVS_GUARDED; /* Contains "struct ofport"s. */
272 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
274 enum revalidate_reason need_revalidate; /* Revalidate all flows. */
276 bool recv_set_enable; /* Enables or disables receiving packets. */
279 struct recirc_id_pool *rid_pool; /* Recirculation ID pool. */
280 struct cmap recirc_map; /* Map of 'recirc_id's to 'ofproto's. */
281 struct ovs_mutex recirc_mutex; /* Protects 'recirc_map'. */
282 bool enable_recirc; /* True if the datapath supports recirculation */
284 /* True if the datapath supports unique flow identifiers */
287 /* True if the datapath supports variable-length
288 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.
289 * False if the datapath supports only 8-byte (or shorter) userdata. */
290 bool variable_length_userdata;
292 /* True if the datapath supports masked data in OVS_ACTION_ATTR_SET
294 bool masked_set_action;
296 /* Maximum number of MPLS label stack entries that the datapath supports
298 size_t max_mpls_depth;
300 /* Version string of the datapath stored in OVSDB. */
301 char *dp_version_string;
303 /* True if the datapath supports tnl_push and pop actions. */
304 bool enable_tnl_push_pop;
305 struct atomic_count tnl_count;
308 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
309 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
311 struct ofproto_dpif {
312 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
314 struct dpif_backer *backer;
316 uint64_t dump_seq; /* Last read of udpif_dump_seq(). */
318 /* Special OpenFlow rules. */
319 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
320 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
321 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
324 struct netflow *netflow;
325 struct dpif_sflow *sflow;
326 struct dpif_ipfix *ipfix;
327 struct hmap bundles; /* Contains "struct ofbundle"s. */
328 struct mac_learning *ml;
329 struct mcast_snooping *ms;
330 bool has_bonded_bundles;
332 struct mbridge *mbridge;
334 struct ovs_mutex stats_mutex;
335 struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
336 * consumed in userspace. */
340 long long int stp_last_tick;
342 /* Rapid Spanning Tree. */
344 long long int rstp_last_tick;
346 /* VLAN splinters. */
347 struct ovs_mutex vsp_mutex;
348 struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
349 struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */
352 struct sset ports; /* Set of standard port names. */
353 struct sset ghost_ports; /* Ports with no datapath port. */
354 struct sset port_poll_set; /* Queued names for port_poll() reply. */
355 int port_poll_errno; /* Last errno for port_poll() reply. */
356 uint64_t change_seq; /* Connectivity status changes. */
359 struct guarded_list pins; /* Contains "struct ofputil_packet_in"s. */
360 struct seq *pins_seq; /* For notifying 'pins' reception. */
364 /* All existing ofproto_dpif instances, indexed by ->up.name. */
365 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
367 static bool ofproto_use_tnl_push_pop = true;
368 static void ofproto_unixctl_init(void);
370 static inline struct ofproto_dpif *
371 ofproto_dpif_cast(const struct ofproto *ofproto)
373 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
374 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
378 ofproto_dpif_get_max_mpls_depth(const struct ofproto_dpif *ofproto)
380 return ofproto->backer->max_mpls_depth;
384 ofproto_dpif_get_enable_recirc(const struct ofproto_dpif *ofproto)
386 return ofproto->backer->enable_recirc;
390 ofproto_dpif_get_enable_ufid(struct dpif_backer *backer)
392 return backer->enable_ufid;
395 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
396 ofp_port_t ofp_port);
397 static void ofproto_trace(struct ofproto_dpif *, struct flow *,
398 const struct dp_packet *packet,
399 const struct ofpact[], size_t ofpacts_len,
402 /* Global variables. */
403 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
405 /* Initial mappings of port to bridge mappings. */
406 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
408 /* Executes 'fm'. The caller retains ownership of 'fm' and everything in
411 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
412 struct ofputil_flow_mod *fm)
414 ofproto_flow_mod(&ofproto->up, fm);
417 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
418 * Takes ownership of 'pin' and pin->packet. */
420 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
421 struct ofproto_packet_in *pin)
423 if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
424 COVERAGE_INC(packet_in_overflow);
425 free(CONST_CAST(void *, pin->up.packet));
429 /* Wakes up main thread for packet-in I/O. */
430 seq_change(ofproto->pins_seq);
433 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
434 * packet rather than to send the packet to the controller.
436 * This function returns false to indicate that a packet_in message
437 * for a "table-miss" should be sent to at least one controller.
438 * False otherwise. */
440 ofproto_dpif_wants_packet_in_on_miss(struct ofproto_dpif *ofproto)
442 return connmgr_wants_packet_in_on_miss(ofproto->up.connmgr);
445 /* Factory functions. */
448 init(const struct shash *iface_hints)
450 struct shash_node *node;
452 /* Make a local copy, since we don't own 'iface_hints' elements. */
453 SHASH_FOR_EACH(node, iface_hints) {
454 const struct iface_hint *orig_hint = node->data;
455 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
457 new_hint->br_name = xstrdup(orig_hint->br_name);
458 new_hint->br_type = xstrdup(orig_hint->br_type);
459 new_hint->ofp_port = orig_hint->ofp_port;
461 shash_add(&init_ofp_ports, node->name, new_hint);
466 enumerate_types(struct sset *types)
468 dp_enumerate_types(types);
472 enumerate_names(const char *type, struct sset *names)
474 struct ofproto_dpif *ofproto;
477 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
478 if (strcmp(type, ofproto->up.type)) {
481 sset_add(names, ofproto->up.name);
488 del(const char *type, const char *name)
493 error = dpif_open(name, type, &dpif);
495 error = dpif_delete(dpif);
502 port_open_type(const char *datapath_type, const char *port_type)
504 return dpif_port_open_type(datapath_type, port_type);
507 /* Type functions. */
509 static void process_dpif_port_changes(struct dpif_backer *);
510 static void process_dpif_all_ports_changed(struct dpif_backer *);
511 static void process_dpif_port_change(struct dpif_backer *,
512 const char *devname);
513 static void process_dpif_port_error(struct dpif_backer *, int error);
515 static struct ofproto_dpif *
516 lookup_ofproto_dpif_by_port_name(const char *name)
518 struct ofproto_dpif *ofproto;
520 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
521 if (sset_contains(&ofproto->ports, name)) {
530 type_run(const char *type)
532 struct dpif_backer *backer;
534 backer = shash_find_data(&all_dpif_backers, type);
536 /* This is not necessarily a problem, since backers are only
537 * created on demand. */
542 if (dpif_run(backer->dpif)) {
543 backer->need_revalidate = REV_RECONFIGURE;
546 udpif_run(backer->udpif);
548 /* If vswitchd started with other_config:flow_restore_wait set as "true",
549 * and the configuration has now changed to "false", enable receiving
550 * packets from the datapath. */
551 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
554 backer->recv_set_enable = true;
556 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
558 VLOG_ERR("Failed to enable receiving packets in dpif.");
561 dpif_flow_flush(backer->dpif);
562 backer->need_revalidate = REV_RECONFIGURE;
565 if (backer->recv_set_enable) {
566 udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
569 dpif_poll_threads_set(backer->dpif, n_dpdk_rxqs, pmd_cpu_mask);
571 if (backer->need_revalidate) {
572 struct ofproto_dpif *ofproto;
573 struct simap_node *node;
574 struct simap tmp_backers;
576 /* Handle tunnel garbage collection. */
577 simap_init(&tmp_backers);
578 simap_swap(&backer->tnl_backers, &tmp_backers);
580 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
581 struct ofport_dpif *iter;
583 if (backer != ofproto->backer) {
587 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
588 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
591 if (!iter->is_tunnel) {
595 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
596 namebuf, sizeof namebuf);
597 node = simap_find(&tmp_backers, dp_port);
599 simap_put(&backer->tnl_backers, dp_port, node->data);
600 simap_delete(&tmp_backers, node);
601 node = simap_find(&backer->tnl_backers, dp_port);
603 node = simap_find(&backer->tnl_backers, dp_port);
605 odp_port_t odp_port = ODPP_NONE;
607 if (!dpif_port_add(backer->dpif, iter->up.netdev,
609 simap_put(&backer->tnl_backers, dp_port,
610 odp_to_u32(odp_port));
611 node = simap_find(&backer->tnl_backers, dp_port);
616 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
617 if (tnl_port_reconfigure(iter, iter->up.netdev,
619 ovs_native_tunneling_is_on(ofproto), dp_port)) {
620 backer->need_revalidate = REV_RECONFIGURE;
625 SIMAP_FOR_EACH (node, &tmp_backers) {
626 dpif_port_del(backer->dpif, u32_to_odp(node->data));
628 simap_destroy(&tmp_backers);
630 switch (backer->need_revalidate) {
631 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
632 case REV_STP: COVERAGE_INC(rev_stp); break;
633 case REV_RSTP: COVERAGE_INC(rev_rstp); break;
634 case REV_BOND: COVERAGE_INC(rev_bond); break;
635 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
636 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
637 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
638 case REV_MCAST_SNOOPING: COVERAGE_INC(rev_mcast_snooping); break;
640 backer->need_revalidate = 0;
642 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
643 struct ofport_dpif *ofport;
644 struct ofbundle *bundle;
646 if (ofproto->backer != backer) {
651 xlate_ofproto_set(ofproto, ofproto->up.name,
652 ofproto->backer->dpif, ofproto->ml,
653 ofproto->stp, ofproto->rstp, ofproto->ms,
654 ofproto->mbridge, ofproto->sflow, ofproto->ipfix,
656 ofproto->up.forward_bpdu,
657 connmgr_has_in_band(ofproto->up.connmgr),
658 ofproto->backer->enable_recirc,
659 ofproto->backer->variable_length_userdata,
660 ofproto->backer->max_mpls_depth,
661 ofproto->backer->masked_set_action);
663 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
664 xlate_bundle_set(ofproto, bundle, bundle->name,
665 bundle->vlan_mode, bundle->vlan,
666 bundle->trunks, bundle->use_priority_tags,
667 bundle->bond, bundle->lacp,
671 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
672 int stp_port = ofport->stp_port
673 ? stp_port_no(ofport->stp_port)
675 xlate_ofport_set(ofproto, ofport->bundle, ofport,
676 ofport->up.ofp_port, ofport->odp_port,
677 ofport->up.netdev, ofport->cfm,
678 ofport->bfd, ofport->peer, stp_port,
679 ofport->rstp_port, ofport->qdscp,
680 ofport->n_qdscp, ofport->up.pp.config,
681 ofport->up.pp.state, ofport->is_tunnel,
687 udpif_revalidate(backer->udpif);
690 process_dpif_port_changes(backer);
695 /* Check for and handle port changes in 'backer''s dpif. */
697 process_dpif_port_changes(struct dpif_backer *backer)
703 error = dpif_port_poll(backer->dpif, &devname);
709 process_dpif_all_ports_changed(backer);
713 process_dpif_port_change(backer, devname);
718 process_dpif_port_error(backer, error);
725 process_dpif_all_ports_changed(struct dpif_backer *backer)
727 struct ofproto_dpif *ofproto;
728 struct dpif_port dpif_port;
729 struct dpif_port_dump dump;
730 struct sset devnames;
733 sset_init(&devnames);
734 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
735 if (ofproto->backer == backer) {
736 struct ofport *ofport;
738 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
739 sset_add(&devnames, netdev_get_name(ofport->netdev));
743 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
744 sset_add(&devnames, dpif_port.name);
747 SSET_FOR_EACH (devname, &devnames) {
748 process_dpif_port_change(backer, devname);
750 sset_destroy(&devnames);
754 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
756 struct ofproto_dpif *ofproto;
757 struct dpif_port port;
759 /* Don't report on the datapath's device. */
760 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
764 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
765 &all_ofproto_dpifs) {
766 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
771 ofproto = lookup_ofproto_dpif_by_port_name(devname);
772 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
773 /* The port was removed. If we know the datapath,
774 * report it through poll_set(). If we don't, it may be
775 * notifying us of a removal we initiated, so ignore it.
776 * If there's a pending ENOBUFS, let it stand, since
777 * everything will be reevaluated. */
778 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
779 sset_add(&ofproto->port_poll_set, devname);
780 ofproto->port_poll_errno = 0;
782 } else if (!ofproto) {
783 /* The port was added, but we don't know with which
784 * ofproto we should associate it. Delete it. */
785 dpif_port_del(backer->dpif, port.port_no);
787 struct ofport_dpif *ofport;
789 ofport = ofport_dpif_cast(shash_find_data(
790 &ofproto->up.port_by_name, devname));
792 && ofport->odp_port != port.port_no
793 && !odp_port_to_ofport(backer, port.port_no))
795 /* 'ofport''s datapath port number has changed from
796 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
797 * structures to match. */
798 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
799 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
800 ofport->odp_port = port.port_no;
801 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
802 hash_odp_port(port.port_no));
803 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
804 backer->need_revalidate = REV_RECONFIGURE;
807 dpif_port_destroy(&port);
810 /* Propagate 'error' to all ofprotos based on 'backer'. */
812 process_dpif_port_error(struct dpif_backer *backer, int error)
814 struct ofproto_dpif *ofproto;
816 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
817 if (ofproto->backer == backer) {
818 sset_clear(&ofproto->port_poll_set);
819 ofproto->port_poll_errno = error;
825 type_wait(const char *type)
827 struct dpif_backer *backer;
829 backer = shash_find_data(&all_dpif_backers, type);
831 /* This is not necessarily a problem, since backers are only
832 * created on demand. */
836 dpif_wait(backer->dpif);
839 /* Basic life-cycle. */
841 static int add_internal_flows(struct ofproto_dpif *);
843 static struct ofproto *
846 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
851 dealloc(struct ofproto *ofproto_)
853 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
857 /* Called when 'ofproto' is destructed. Checks for and clears any
860 dpif_backer_recirc_clear_ofproto(struct dpif_backer *backer,
861 struct ofproto_dpif *ofproto)
863 struct dpif_backer_recirc_node *node;
865 ovs_mutex_lock(&backer->recirc_mutex);
866 CMAP_FOR_EACH (node, cmap_node, &backer->recirc_map) {
867 if (node->ofproto == ofproto) {
868 VLOG_ERR("recirc_id %"PRIu32", not freed when ofproto (%s) "
869 "is destructed", node->recirc_id, ofproto->up.name);
870 cmap_remove(&backer->recirc_map, &node->cmap_node,
872 ovsrcu_postpone(free, node);
875 ovs_mutex_unlock(&backer->recirc_mutex);
879 close_dpif_backer(struct dpif_backer *backer)
881 ovs_assert(backer->refcount > 0);
883 if (--backer->refcount) {
887 udpif_destroy(backer->udpif);
889 simap_destroy(&backer->tnl_backers);
890 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
891 hmap_destroy(&backer->odp_to_ofport_map);
892 shash_find_and_delete(&all_dpif_backers, backer->type);
893 recirc_id_pool_destroy(backer->rid_pool);
894 cmap_destroy(&backer->recirc_map);
895 ovs_mutex_destroy(&backer->recirc_mutex);
897 free(backer->dp_version_string);
898 dpif_close(backer->dpif);
902 /* Datapath port slated for removal from datapath. */
904 struct ovs_list list_node;
908 static bool check_variable_length_userdata(struct dpif_backer *backer);
909 static size_t check_max_mpls_depth(struct dpif_backer *backer);
910 static bool check_recirc(struct dpif_backer *backer);
911 static bool check_ufid(struct dpif_backer *backer);
912 static bool check_masked_set_action(struct dpif_backer *backer);
915 open_dpif_backer(const char *type, struct dpif_backer **backerp)
917 struct dpif_backer *backer;
918 struct dpif_port_dump port_dump;
919 struct dpif_port port;
920 struct shash_node *node;
921 struct ovs_list garbage_list;
922 struct odp_garbage *garbage, *next;
929 backer = shash_find_data(&all_dpif_backers, type);
936 backer_name = xasprintf("ovs-%s", type);
938 /* Remove any existing datapaths, since we assume we're the only
939 * userspace controlling the datapath. */
941 dp_enumerate_names(type, &names);
942 SSET_FOR_EACH(name, &names) {
943 struct dpif *old_dpif;
945 /* Don't remove our backer if it exists. */
946 if (!strcmp(name, backer_name)) {
950 if (dpif_open(name, type, &old_dpif)) {
951 VLOG_WARN("couldn't open old datapath %s to remove it", name);
953 dpif_delete(old_dpif);
954 dpif_close(old_dpif);
957 sset_destroy(&names);
959 backer = xmalloc(sizeof *backer);
961 error = dpif_create_and_open(backer_name, type, &backer->dpif);
964 VLOG_ERR("failed to open datapath of type %s: %s", type,
965 ovs_strerror(error));
969 backer->udpif = udpif_create(backer, backer->dpif);
971 backer->type = xstrdup(type);
972 backer->refcount = 1;
973 hmap_init(&backer->odp_to_ofport_map);
974 ovs_rwlock_init(&backer->odp_to_ofport_lock);
975 backer->need_revalidate = 0;
976 simap_init(&backer->tnl_backers);
977 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
980 if (backer->recv_set_enable) {
981 dpif_flow_flush(backer->dpif);
984 /* Loop through the ports already on the datapath and remove any
985 * that we don't need anymore. */
986 list_init(&garbage_list);
987 dpif_port_dump_start(&port_dump, backer->dpif);
988 while (dpif_port_dump_next(&port_dump, &port)) {
989 node = shash_find(&init_ofp_ports, port.name);
990 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
991 garbage = xmalloc(sizeof *garbage);
992 garbage->odp_port = port.port_no;
993 list_push_front(&garbage_list, &garbage->list_node);
996 dpif_port_dump_done(&port_dump);
998 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
999 dpif_port_del(backer->dpif, garbage->odp_port);
1000 list_remove(&garbage->list_node);
1004 shash_add(&all_dpif_backers, type, backer);
1006 backer->enable_recirc = check_recirc(backer);
1007 backer->max_mpls_depth = check_max_mpls_depth(backer);
1008 backer->masked_set_action = check_masked_set_action(backer);
1009 backer->enable_ufid = check_ufid(backer);
1010 backer->rid_pool = recirc_id_pool_create();
1011 ovs_mutex_init(&backer->recirc_mutex);
1012 cmap_init(&backer->recirc_map);
1014 backer->enable_tnl_push_pop = dpif_supports_tnl_push_pop(backer->dpif);
1015 atomic_count_init(&backer->tnl_count, 0);
1017 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1019 VLOG_ERR("failed to listen on datapath of type %s: %s",
1020 type, ovs_strerror(error));
1021 close_dpif_backer(backer);
1025 if (backer->recv_set_enable) {
1026 udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
1029 /* This check fails if performed before udpif threads have been set,
1030 * as the kernel module checks that the 'pid' in userspace action
1032 backer->variable_length_userdata = check_variable_length_userdata(backer);
1033 backer->dp_version_string = dpif_get_dp_version(backer->dpif);
1039 ovs_native_tunneling_is_on(struct ofproto_dpif *ofproto)
1041 return ofproto_use_tnl_push_pop && ofproto->backer->enable_tnl_push_pop &&
1042 atomic_count_get(&ofproto->backer->tnl_count);
1045 /* Tests whether 'backer''s datapath supports recirculation. Only newer
1046 * datapaths support OVS_KEY_ATTR_RECIRC_ID in keys. We need to disable some
1047 * features on older datapaths that don't support this feature.
1049 * Returns false if 'backer' definitely does not support recirculation, true if
1050 * it seems to support recirculation or if at least the error we get is
1053 check_recirc(struct dpif_backer *backer)
1056 struct odputil_keybuf keybuf;
1060 memset(&flow, 0, sizeof flow);
1064 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1065 odp_flow_key_from_flow(&key, &flow, NULL, 0, true);
1066 enable_recirc = dpif_probe_feature(backer->dpif, "recirculation", &key,
1069 if (enable_recirc) {
1070 VLOG_INFO("%s: Datapath supports recirculation",
1071 dpif_name(backer->dpif));
1073 VLOG_INFO("%s: Datapath does not support recirculation",
1074 dpif_name(backer->dpif));
1077 return enable_recirc;
1080 /* Tests whether 'dpif' supports unique flow ids. We can skip serializing
1081 * some flow attributes for datapaths that support this feature.
1083 * Returns true if 'dpif' supports UFID for flow operations.
1084 * Returns false if 'dpif' does not support UFID. */
1086 check_ufid(struct dpif_backer *backer)
1089 struct odputil_keybuf keybuf;
1094 memset(&flow, 0, sizeof flow);
1095 flow.dl_type = htons(0x1234);
1097 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1098 odp_flow_key_from_flow(&key, &flow, NULL, 0, true);
1099 dpif_flow_hash(backer->dpif, key.data, key.size, &ufid);
1101 enable_ufid = dpif_probe_feature(backer->dpif, "UFID", &key, &ufid);
1104 VLOG_INFO("%s: Datapath supports unique flow ids",
1105 dpif_name(backer->dpif));
1107 VLOG_INFO("%s: Datapath does not support unique flow ids",
1108 dpif_name(backer->dpif));
1113 /* Tests whether 'backer''s datapath supports variable-length
1114 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions. We need
1115 * to disable some features on older datapaths that don't support this
1118 * Returns false if 'backer' definitely does not support variable-length
1119 * userdata, true if it seems to support them or if at least the error we get
1122 check_variable_length_userdata(struct dpif_backer *backer)
1124 struct eth_header *eth;
1125 struct ofpbuf actions;
1126 struct dpif_execute execute;
1127 struct dp_packet packet;
1131 /* Compose a userspace action that will cause an ERANGE error on older
1132 * datapaths that don't support variable-length userdata.
1134 * We really test for using userdata longer than 8 bytes, but older
1135 * datapaths accepted these, silently truncating the userdata to 8 bytes.
1136 * The same older datapaths rejected userdata shorter than 8 bytes, so we
1137 * test for that instead as a proxy for longer userdata support. */
1138 ofpbuf_init(&actions, 64);
1139 start = nl_msg_start_nested(&actions, OVS_ACTION_ATTR_USERSPACE);
1140 nl_msg_put_u32(&actions, OVS_USERSPACE_ATTR_PID,
1141 dpif_port_get_pid(backer->dpif, ODPP_NONE, 0));
1142 nl_msg_put_unspec_zero(&actions, OVS_USERSPACE_ATTR_USERDATA, 4);
1143 nl_msg_end_nested(&actions, start);
1145 /* Compose a dummy ethernet packet. */
1146 dp_packet_init(&packet, ETH_HEADER_LEN);
1147 eth = dp_packet_put_zeros(&packet, ETH_HEADER_LEN);
1148 eth->eth_type = htons(0x1234);
1150 /* Execute the actions. On older datapaths this fails with ERANGE, on
1151 * newer datapaths it succeeds. */
1152 execute.actions = actions.data;
1153 execute.actions_len = actions.size;
1154 execute.packet = &packet;
1155 execute.needs_help = false;
1156 execute.probe = true;
1158 error = dpif_execute(backer->dpif, &execute);
1160 dp_packet_uninit(&packet);
1161 ofpbuf_uninit(&actions);
1168 /* Variable-length userdata is not supported. */
1169 VLOG_WARN("%s: datapath does not support variable-length userdata "
1170 "feature (needs Linux 3.10+ or kernel module from OVS "
1171 "1..11+). The NXAST_SAMPLE action will be ignored.",
1172 dpif_name(backer->dpif));
1176 /* Something odd happened. We're not sure whether variable-length
1177 * userdata is supported. Default to "yes". */
1178 VLOG_WARN("%s: variable-length userdata feature probe failed (%s)",
1179 dpif_name(backer->dpif), ovs_strerror(error));
1184 /* Tests the MPLS label stack depth supported by 'backer''s datapath.
1186 * Returns the number of elements in a struct flow's mpls_lse field
1187 * if the datapath supports at least that many entries in an
1189 * Otherwise returns the number of MPLS push actions supported by
1192 check_max_mpls_depth(struct dpif_backer *backer)
1197 for (n = 0; n < FLOW_MAX_MPLS_LABELS; n++) {
1198 struct odputil_keybuf keybuf;
1201 memset(&flow, 0, sizeof flow);
1202 flow.dl_type = htons(ETH_TYPE_MPLS);
1203 flow_set_mpls_bos(&flow, n, 1);
1205 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1206 odp_flow_key_from_flow(&key, &flow, NULL, 0, false);
1207 if (!dpif_probe_feature(backer->dpif, "MPLS", &key, NULL)) {
1212 VLOG_INFO("%s: MPLS label stack length probed as %d",
1213 dpif_name(backer->dpif), n);
1217 /* Tests whether 'backer''s datapath supports masked data in
1218 * OVS_ACTION_ATTR_SET actions. We need to disable some features on older
1219 * datapaths that don't support this feature. */
1221 check_masked_set_action(struct dpif_backer *backer)
1223 struct eth_header *eth;
1224 struct ofpbuf actions;
1225 struct dpif_execute execute;
1226 struct dp_packet packet;
1228 struct ovs_key_ethernet key, mask;
1230 /* Compose a set action that will cause an EINVAL error on older
1231 * datapaths that don't support masked set actions.
1232 * Avoid using a full mask, as it could be translated to a non-masked
1233 * set action instead. */
1234 ofpbuf_init(&actions, 64);
1235 memset(&key, 0x53, sizeof key);
1236 memset(&mask, 0x7f, sizeof mask);
1237 commit_masked_set_action(&actions, OVS_KEY_ATTR_ETHERNET, &key, &mask,
1240 /* Compose a dummy ethernet packet. */
1241 dp_packet_init(&packet, ETH_HEADER_LEN);
1242 eth = dp_packet_put_zeros(&packet, ETH_HEADER_LEN);
1243 eth->eth_type = htons(0x1234);
1245 /* Execute the actions. On older datapaths this fails with EINVAL, on
1246 * newer datapaths it succeeds. */
1247 execute.actions = actions.data;
1248 execute.actions_len = actions.size;
1249 execute.packet = &packet;
1250 execute.needs_help = false;
1251 execute.probe = true;
1253 error = dpif_execute(backer->dpif, &execute);
1255 dp_packet_uninit(&packet);
1256 ofpbuf_uninit(&actions);
1259 /* Masked set action is not supported. */
1260 VLOG_INFO("%s: datapath does not support masked set action feature.",
1261 dpif_name(backer->dpif));
1267 construct(struct ofproto *ofproto_)
1269 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1270 struct shash_node *node, *next;
1273 /* Tunnel module can get used right after the udpif threads are running. */
1274 ofproto_tunnel_init();
1276 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1281 ofproto->netflow = NULL;
1282 ofproto->sflow = NULL;
1283 ofproto->ipfix = NULL;
1284 ofproto->stp = NULL;
1285 ofproto->rstp = NULL;
1286 ofproto->dump_seq = 0;
1287 hmap_init(&ofproto->bundles);
1288 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1290 ofproto->mbridge = mbridge_create();
1291 ofproto->has_bonded_bundles = false;
1292 ofproto->lacp_enabled = false;
1293 ovs_mutex_init_adaptive(&ofproto->stats_mutex);
1294 ovs_mutex_init(&ofproto->vsp_mutex);
1296 guarded_list_init(&ofproto->pins);
1298 ofproto_unixctl_init();
1300 hmap_init(&ofproto->vlandev_map);
1301 hmap_init(&ofproto->realdev_vid_map);
1303 sset_init(&ofproto->ports);
1304 sset_init(&ofproto->ghost_ports);
1305 sset_init(&ofproto->port_poll_set);
1306 ofproto->port_poll_errno = 0;
1307 ofproto->change_seq = 0;
1308 ofproto->pins_seq = seq_create();
1309 ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1312 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1313 struct iface_hint *iface_hint = node->data;
1315 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1316 /* Check if the datapath already has this port. */
1317 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1318 sset_add(&ofproto->ports, node->name);
1321 free(iface_hint->br_name);
1322 free(iface_hint->br_type);
1324 shash_delete(&init_ofp_ports, node);
1328 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1329 hash_string(ofproto->up.name, 0));
1330 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1332 ofproto_init_tables(ofproto_, N_TABLES);
1333 error = add_internal_flows(ofproto);
1335 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1341 add_internal_miss_flow(struct ofproto_dpif *ofproto, int id,
1342 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1348 match_init_catchall(&match);
1349 match_set_reg(&match, 0, id);
1351 error = ofproto_dpif_add_internal_flow(ofproto, &match, 0, 0, ofpacts,
1353 *rulep = error ? NULL : rule_dpif_cast(rule);
1359 add_internal_flows(struct ofproto_dpif *ofproto)
1361 struct ofpact_controller *controller;
1362 uint64_t ofpacts_stub[128 / 8];
1363 struct ofpbuf ofpacts;
1364 struct rule *unused_rulep OVS_UNUSED;
1365 struct ofpact_resubmit *resubmit;
1370 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1373 controller = ofpact_put_CONTROLLER(&ofpacts);
1374 controller->max_len = UINT16_MAX;
1375 controller->controller_id = 0;
1376 controller->reason = OFPR_NO_MATCH;
1377 ofpact_pad(&ofpacts);
1379 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1380 &ofproto->miss_rule);
1385 ofpbuf_clear(&ofpacts);
1386 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1387 &ofproto->no_packet_in_rule);
1392 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1393 &ofproto->drop_frags_rule);
1398 /* Drop any run away non-recirc rule lookups. Recirc_id has to be
1399 * zero when reaching this rule.
1401 * (priority=2), recirc_id=0, actions=drop
1403 ofpbuf_clear(&ofpacts);
1404 match_init_catchall(&match);
1405 match_set_recirc_id(&match, 0);
1406 error = ofproto_dpif_add_internal_flow(ofproto, &match, 2, 0, &ofpacts,
1412 /* Continue rule lookups for not-matched recirc rules from table 0.
1414 * (priority=1), actions=resubmit(, 0)
1416 resubmit = ofpact_put_RESUBMIT(&ofpacts);
1417 resubmit->in_port = OFPP_IN_PORT;
1418 resubmit->table_id = 0;
1420 match_init_catchall(&match);
1421 error = ofproto_dpif_add_internal_flow(ofproto, &match, 1, 0, &ofpacts,
1428 destruct(struct ofproto *ofproto_)
1430 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1431 struct ofproto_packet_in *pin, *next_pin;
1432 struct rule_dpif *rule;
1433 struct oftable *table;
1434 struct ovs_list pins;
1436 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1438 xlate_remove_ofproto(ofproto);
1441 /* Ensure that the upcall processing threads have no remaining references
1442 * to the ofproto or anything in it. */
1443 udpif_synchronize(ofproto->backer->udpif);
1445 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1447 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1448 CLS_FOR_EACH (rule, up.cr, &table->cls) {
1449 ofproto_rule_delete(&ofproto->up, &rule->up);
1453 guarded_list_pop_all(&ofproto->pins, &pins);
1454 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1455 list_remove(&pin->list_node);
1456 free(CONST_CAST(void *, pin->up.packet));
1459 guarded_list_destroy(&ofproto->pins);
1461 dpif_backer_recirc_clear_ofproto(ofproto->backer, ofproto);
1463 mbridge_unref(ofproto->mbridge);
1465 netflow_unref(ofproto->netflow);
1466 dpif_sflow_unref(ofproto->sflow);
1467 dpif_ipfix_unref(ofproto->ipfix);
1468 hmap_destroy(&ofproto->bundles);
1469 mac_learning_unref(ofproto->ml);
1470 mcast_snooping_unref(ofproto->ms);
1472 hmap_destroy(&ofproto->vlandev_map);
1473 hmap_destroy(&ofproto->realdev_vid_map);
1475 sset_destroy(&ofproto->ports);
1476 sset_destroy(&ofproto->ghost_ports);
1477 sset_destroy(&ofproto->port_poll_set);
1479 ovs_mutex_destroy(&ofproto->stats_mutex);
1480 ovs_mutex_destroy(&ofproto->vsp_mutex);
1482 seq_destroy(ofproto->pins_seq);
1484 close_dpif_backer(ofproto->backer);
1488 run(struct ofproto *ofproto_)
1490 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1491 uint64_t new_seq, new_dump_seq;
1493 if (mbridge_need_revalidate(ofproto->mbridge)) {
1494 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1495 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1496 mac_learning_flush(ofproto->ml);
1497 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1498 mcast_snooping_mdb_flush(ofproto->ms);
1501 /* Always updates the ofproto->pins_seqno to avoid frequent wakeup during
1502 * flow restore. Even though nothing is processed during flow restore,
1503 * all queued 'pins' will be handled immediately when flow restore
1505 ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1507 /* Do not perform any periodic activity required by 'ofproto' while
1508 * waiting for flow restore to complete. */
1509 if (!ofproto_get_flow_restore_wait()) {
1510 struct ofproto_packet_in *pin, *next_pin;
1511 struct ovs_list pins;
1513 guarded_list_pop_all(&ofproto->pins, &pins);
1514 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1515 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1516 list_remove(&pin->list_node);
1517 free(CONST_CAST(void *, pin->up.packet));
1522 if (ofproto->netflow) {
1523 netflow_run(ofproto->netflow);
1525 if (ofproto->sflow) {
1526 dpif_sflow_run(ofproto->sflow);
1528 if (ofproto->ipfix) {
1529 dpif_ipfix_run(ofproto->ipfix);
1532 new_seq = seq_read(connectivity_seq_get());
1533 if (ofproto->change_seq != new_seq) {
1534 struct ofport_dpif *ofport;
1536 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1540 ofproto->change_seq = new_seq;
1542 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1543 struct ofbundle *bundle;
1545 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1552 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1553 if (mac_learning_run(ofproto->ml)) {
1554 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1556 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1558 if (mcast_snooping_run(ofproto->ms)) {
1559 ofproto->backer->need_revalidate = REV_MCAST_SNOOPING;
1562 new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
1563 if (ofproto->dump_seq != new_dump_seq) {
1564 struct rule *rule, *next_rule;
1566 /* We know stats are relatively fresh, so now is a good time to do some
1568 ofproto->dump_seq = new_dump_seq;
1570 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1572 ovs_mutex_lock(&ofproto_mutex);
1573 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
1574 &ofproto->up.expirable) {
1575 rule_expire(rule_dpif_cast(rule));
1577 ovs_mutex_unlock(&ofproto_mutex);
1579 /* All outstanding data in existing flows has been accounted, so it's a
1580 * good time to do bond rebalancing. */
1581 if (ofproto->has_bonded_bundles) {
1582 struct ofbundle *bundle;
1584 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1586 bond_rebalance(bundle->bond);
1595 wait(struct ofproto *ofproto_)
1597 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1599 if (ofproto_get_flow_restore_wait()) {
1603 if (ofproto->sflow) {
1604 dpif_sflow_wait(ofproto->sflow);
1606 if (ofproto->ipfix) {
1607 dpif_ipfix_wait(ofproto->ipfix);
1609 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1610 struct ofbundle *bundle;
1612 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1613 bundle_wait(bundle);
1616 if (ofproto->netflow) {
1617 netflow_wait(ofproto->netflow);
1619 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1620 mac_learning_wait(ofproto->ml);
1621 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1622 mcast_snooping_wait(ofproto->ms);
1624 if (ofproto->backer->need_revalidate) {
1625 /* Shouldn't happen, but if it does just go around again. */
1626 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1627 poll_immediate_wake();
1630 seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
1631 seq_wait(ofproto->pins_seq, ofproto->pins_seqno);
1635 type_get_memory_usage(const char *type, struct simap *usage)
1637 struct dpif_backer *backer;
1639 backer = shash_find_data(&all_dpif_backers, type);
1641 udpif_get_memory_usage(backer->udpif, usage);
1646 flush(struct ofproto *ofproto_)
1648 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1649 struct dpif_backer *backer = ofproto->backer;
1652 udpif_flush(backer->udpif);
1657 query_tables(struct ofproto *ofproto,
1658 struct ofputil_table_features *features,
1659 struct ofputil_table_stats *stats)
1661 strcpy(features->name, "classifier");
1666 for (i = 0; i < ofproto->n_tables; i++) {
1667 unsigned long missed, matched;
1669 atomic_read_relaxed(&ofproto->tables[i].n_matched, &matched);
1670 atomic_read_relaxed(&ofproto->tables[i].n_missed, &missed);
1672 stats[i].matched_count = matched;
1673 stats[i].lookup_count = matched + missed;
1678 static struct ofport *
1681 struct ofport_dpif *port = xmalloc(sizeof *port);
1686 port_dealloc(struct ofport *port_)
1688 struct ofport_dpif *port = ofport_dpif_cast(port_);
1693 port_construct(struct ofport *port_)
1695 struct ofport_dpif *port = ofport_dpif_cast(port_);
1696 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1697 const struct netdev *netdev = port->up.netdev;
1698 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1699 struct dpif_port dpif_port;
1702 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1703 port->bundle = NULL;
1706 port->may_enable = false;
1707 port->stp_port = NULL;
1708 port->stp_state = STP_DISABLED;
1709 port->rstp_port = NULL;
1710 port->rstp_state = RSTP_DISABLED;
1711 port->is_tunnel = false;
1715 port->realdev_ofp_port = 0;
1716 port->vlandev_vid = 0;
1717 port->carrier_seq = netdev_get_carrier_resets(netdev);
1718 port->is_layer3 = netdev_vport_is_layer3(netdev);
1720 if (netdev_vport_is_patch(netdev)) {
1721 /* By bailing out here, we don't submit the port to the sFlow module
1722 * to be considered for counter polling export. This is correct
1723 * because the patch port represents an interface that sFlow considers
1724 * to be "internal" to the switch as a whole, and therefore not a
1725 * candidate for counter polling. */
1726 port->odp_port = ODPP_NONE;
1727 ofport_update_peer(port);
1731 error = dpif_port_query_by_name(ofproto->backer->dpif,
1732 netdev_vport_get_dpif_port(netdev, namebuf,
1739 port->odp_port = dpif_port.port_no;
1741 if (netdev_get_tunnel_config(netdev)) {
1742 atomic_count_inc(&ofproto->backer->tnl_count);
1743 tnl_port_add(port, port->up.netdev, port->odp_port,
1744 ovs_native_tunneling_is_on(ofproto), namebuf);
1745 port->is_tunnel = true;
1746 if (ofproto->ipfix) {
1747 dpif_ipfix_add_tunnel_port(ofproto->ipfix, port_, port->odp_port);
1750 /* Sanity-check that a mapping doesn't already exist. This
1751 * shouldn't happen for non-tunnel ports. */
1752 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1753 VLOG_ERR("port %s already has an OpenFlow port number",
1755 dpif_port_destroy(&dpif_port);
1759 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1760 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1761 hash_odp_port(port->odp_port));
1762 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1764 dpif_port_destroy(&dpif_port);
1766 if (ofproto->sflow) {
1767 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1774 port_destruct(struct ofport *port_)
1776 struct ofport_dpif *port = ofport_dpif_cast(port_);
1777 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1778 const char *devname = netdev_get_name(port->up.netdev);
1779 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1780 const char *dp_port_name;
1782 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1784 xlate_ofport_remove(port);
1787 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1789 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1790 /* The underlying device is still there, so delete it. This
1791 * happens when the ofproto is being destroyed, since the caller
1792 * assumes that removal of attached ports will happen as part of
1794 if (!port->is_tunnel) {
1795 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1800 port->peer->peer = NULL;
1804 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1805 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1806 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1807 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1810 if (port->is_tunnel) {
1811 atomic_count_dec(&ofproto->backer->tnl_count);
1814 if (port->is_tunnel && ofproto->ipfix) {
1815 dpif_ipfix_del_tunnel_port(ofproto->ipfix, port->odp_port);
1819 sset_find_and_delete(&ofproto->ports, devname);
1820 sset_find_and_delete(&ofproto->ghost_ports, devname);
1821 bundle_remove(port_);
1822 set_cfm(port_, NULL);
1823 set_bfd(port_, NULL);
1824 if (port->stp_port) {
1825 stp_port_disable(port->stp_port);
1827 set_rstp_port(port_, NULL);
1828 if (ofproto->sflow) {
1829 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1836 port_modified(struct ofport *port_)
1838 struct ofport_dpif *port = ofport_dpif_cast(port_);
1839 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1840 struct netdev *netdev = port->up.netdev;
1842 if (port->bundle && port->bundle->bond) {
1843 bond_slave_set_netdev(port->bundle->bond, port, netdev);
1847 cfm_set_netdev(port->cfm, netdev);
1851 bfd_set_netdev(port->bfd, netdev);
1854 ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1855 port->up.pp.hw_addr);
1857 netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
1859 if (port->is_tunnel) {
1860 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1862 if (tnl_port_reconfigure(port, netdev, port->odp_port,
1863 ovs_native_tunneling_is_on(ofproto), namebuf)) {
1864 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1868 ofport_update_peer(port);
1872 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1874 struct ofport_dpif *port = ofport_dpif_cast(port_);
1875 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1876 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1878 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1879 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1880 OFPUTIL_PC_NO_PACKET_IN)) {
1881 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1883 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1884 bundle_update(port->bundle);
1890 set_sflow(struct ofproto *ofproto_,
1891 const struct ofproto_sflow_options *sflow_options)
1893 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1894 struct dpif_sflow *ds = ofproto->sflow;
1896 if (sflow_options) {
1897 uint32_t old_probability = ds ? dpif_sflow_get_probability(ds) : 0;
1899 struct ofport_dpif *ofport;
1901 ds = ofproto->sflow = dpif_sflow_create();
1902 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1903 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1906 dpif_sflow_set_options(ds, sflow_options);
1907 if (dpif_sflow_get_probability(ds) != old_probability) {
1908 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1912 dpif_sflow_unref(ds);
1913 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1914 ofproto->sflow = NULL;
1922 struct ofproto *ofproto_,
1923 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1924 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1925 size_t n_flow_exporters_options)
1927 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1928 struct dpif_ipfix *di = ofproto->ipfix;
1929 bool has_options = bridge_exporter_options || flow_exporters_options;
1930 bool new_di = false;
1932 if (has_options && !di) {
1933 di = ofproto->ipfix = dpif_ipfix_create();
1938 /* Call set_options in any case to cleanly flush the flow
1939 * caches in the last exporters that are to be destroyed. */
1940 dpif_ipfix_set_options(
1941 di, bridge_exporter_options, flow_exporters_options,
1942 n_flow_exporters_options);
1944 /* Add tunnel ports only when a new ipfix created */
1945 if (new_di == true) {
1946 struct ofport_dpif *ofport;
1947 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1948 if (ofport->is_tunnel == true) {
1949 dpif_ipfix_add_tunnel_port(di, &ofport->up, ofport->odp_port);
1955 dpif_ipfix_unref(di);
1956 ofproto->ipfix = NULL;
1964 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1966 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1971 struct ofproto_dpif *ofproto;
1973 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1974 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1975 ofport->cfm = cfm_create(ofport->up.netdev);
1978 if (cfm_configure(ofport->cfm, s)) {
1985 cfm_unref(ofport->cfm);
1988 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1989 ofport->up.pp.hw_addr);
1994 cfm_status_changed(struct ofport *ofport_)
1996 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1998 return ofport->cfm ? cfm_check_status_change(ofport->cfm) : true;
2002 get_cfm_status(const struct ofport *ofport_,
2003 struct cfm_status *status)
2005 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2009 cfm_get_status(ofport->cfm, status);
2018 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2020 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2021 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2025 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
2026 cfg, ofport->up.netdev);
2027 if (ofport->bfd != old) {
2028 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2030 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
2031 ofport->up.pp.hw_addr);
2036 bfd_status_changed(struct ofport *ofport_)
2038 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2040 return ofport->bfd ? bfd_check_status_change(ofport->bfd) : true;
2044 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2046 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2050 bfd_get_status(ofport->bfd, smap);
2058 /* Spanning Tree. */
2060 /* Called while rstp_mutex is held. */
2062 rstp_send_bpdu_cb(struct dp_packet *pkt, void *ofport_, void *ofproto_)
2064 struct ofproto_dpif *ofproto = ofproto_;
2065 struct ofport_dpif *ofport = ofport_;
2066 struct eth_header *eth = dp_packet_l2(pkt);
2068 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2069 if (eth_addr_is_zero(eth->eth_src)) {
2070 VLOG_WARN_RL(&rl, "%s port %d: cannot send RSTP BPDU on a port which "
2071 "does not have a configured source MAC address.",
2072 ofproto->up.name, ofp_to_u16(ofport->up.ofp_port));
2074 ofproto_dpif_send_packet(ofport, pkt);
2076 dp_packet_delete(pkt);
2080 send_bpdu_cb(struct dp_packet *pkt, int port_num, void *ofproto_)
2082 struct ofproto_dpif *ofproto = ofproto_;
2083 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2084 struct ofport_dpif *ofport;
2086 ofport = stp_port_get_aux(sp);
2088 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2089 ofproto->up.name, port_num);
2091 struct eth_header *eth = dp_packet_l2(pkt);
2093 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2094 if (eth_addr_is_zero(eth->eth_src)) {
2095 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2096 "with unknown MAC", ofproto->up.name, port_num);
2098 ofproto_dpif_send_packet(ofport, pkt);
2101 dp_packet_delete(pkt);
2104 /* Configure RSTP on 'ofproto_' using the settings defined in 's'. */
2106 set_rstp(struct ofproto *ofproto_, const struct ofproto_rstp_settings *s)
2108 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2110 /* Only revalidate flows if the configuration changed. */
2111 if (!s != !ofproto->rstp) {
2112 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2116 if (!ofproto->rstp) {
2117 ofproto->rstp = rstp_create(ofproto_->name, s->address,
2118 rstp_send_bpdu_cb, ofproto);
2119 ofproto->rstp_last_tick = time_msec();
2121 rstp_set_bridge_address(ofproto->rstp, s->address);
2122 rstp_set_bridge_priority(ofproto->rstp, s->priority);
2123 rstp_set_bridge_ageing_time(ofproto->rstp, s->ageing_time);
2124 rstp_set_bridge_force_protocol_version(ofproto->rstp,
2125 s->force_protocol_version);
2126 rstp_set_bridge_max_age(ofproto->rstp, s->bridge_max_age);
2127 rstp_set_bridge_forward_delay(ofproto->rstp, s->bridge_forward_delay);
2128 rstp_set_bridge_transmit_hold_count(ofproto->rstp,
2129 s->transmit_hold_count);
2131 struct ofport *ofport;
2132 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2133 set_rstp_port(ofport, NULL);
2135 rstp_unref(ofproto->rstp);
2136 ofproto->rstp = NULL;
2141 get_rstp_status(struct ofproto *ofproto_, struct ofproto_rstp_status *s)
2143 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2145 if (ofproto->rstp) {
2147 s->root_id = rstp_get_root_id(ofproto->rstp);
2148 s->bridge_id = rstp_get_bridge_id(ofproto->rstp);
2149 s->designated_id = rstp_get_designated_id(ofproto->rstp);
2150 s->root_path_cost = rstp_get_root_path_cost(ofproto->rstp);
2151 s->designated_port_id = rstp_get_designated_port_id(ofproto->rstp);
2152 s->bridge_port_id = rstp_get_bridge_port_id(ofproto->rstp);
2159 update_rstp_port_state(struct ofport_dpif *ofport)
2161 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2162 enum rstp_state state;
2164 /* Figure out new state. */
2165 state = ofport->rstp_port ? rstp_port_get_state(ofport->rstp_port)
2169 if (ofport->rstp_state != state) {
2170 enum ofputil_port_state of_state;
2173 VLOG_DBG("port %s: RSTP state changed from %s to %s",
2174 netdev_get_name(ofport->up.netdev),
2175 rstp_state_name(ofport->rstp_state),
2176 rstp_state_name(state));
2178 if (rstp_learn_in_state(ofport->rstp_state)
2179 != rstp_learn_in_state(state)) {
2180 /* XXX: Learning action flows should also be flushed. */
2181 if (ofport->bundle) {
2182 if (!rstp_shift_root_learned_address(ofproto->rstp)
2183 || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2184 bundle_flush_macs(ofport->bundle, false);
2188 fwd_change = rstp_forward_in_state(ofport->rstp_state)
2189 != rstp_forward_in_state(state);
2191 ofproto->backer->need_revalidate = REV_RSTP;
2192 ofport->rstp_state = state;
2194 if (fwd_change && ofport->bundle) {
2195 bundle_update(ofport->bundle);
2198 /* Update the RSTP state bits in the OpenFlow port description. */
2199 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2200 of_state |= (state == RSTP_LEARNING ? OFPUTIL_PS_STP_LEARN
2201 : state == RSTP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2202 : state == RSTP_DISCARDING ? OFPUTIL_PS_STP_LISTEN
2204 ofproto_port_set_state(&ofport->up, of_state);
2209 rstp_run(struct ofproto_dpif *ofproto)
2211 if (ofproto->rstp) {
2212 long long int now = time_msec();
2213 long long int elapsed = now - ofproto->rstp_last_tick;
2214 struct rstp_port *rp;
2215 struct ofport_dpif *ofport;
2217 /* Every second, decrease the values of the timers. */
2218 if (elapsed >= 1000) {
2219 rstp_tick_timers(ofproto->rstp);
2220 ofproto->rstp_last_tick = now;
2223 while ((ofport = rstp_get_next_changed_port_aux(ofproto->rstp, &rp))) {
2224 update_rstp_port_state(ofport);
2228 /* FIXME: This check should be done on-event (i.e., when setting
2229 * p->fdb_flush) and not periodically.
2231 while ((ofport = rstp_check_and_reset_fdb_flush(ofproto->rstp, &rp))) {
2232 if (!rstp_shift_root_learned_address(ofproto->rstp)
2233 || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2234 bundle_flush_macs(ofport->bundle, false);
2238 if (rstp_shift_root_learned_address(ofproto->rstp)) {
2239 bundle_move(((struct ofport_dpif *)rstp_get_old_root_aux(ofproto->rstp))->bundle,
2240 ((struct ofport_dpif *)rstp_get_new_root_aux(ofproto->rstp))->bundle);
2241 rstp_reset_root_changed(ofproto->rstp);
2246 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2248 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2250 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2252 /* Only revalidate flows if the configuration changed. */
2253 if (!s != !ofproto->stp) {
2254 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2258 if (!ofproto->stp) {
2259 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2260 send_bpdu_cb, ofproto);
2261 ofproto->stp_last_tick = time_msec();
2264 stp_set_bridge_id(ofproto->stp, s->system_id);
2265 stp_set_bridge_priority(ofproto->stp, s->priority);
2266 stp_set_hello_time(ofproto->stp, s->hello_time);
2267 stp_set_max_age(ofproto->stp, s->max_age);
2268 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2270 struct ofport *ofport;
2272 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2273 set_stp_port(ofport, NULL);
2276 stp_unref(ofproto->stp);
2277 ofproto->stp = NULL;
2284 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2286 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2290 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2291 s->designated_root = stp_get_designated_root(ofproto->stp);
2292 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2301 update_stp_port_state(struct ofport_dpif *ofport)
2303 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2304 enum stp_state state;
2306 /* Figure out new state. */
2307 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2311 if (ofport->stp_state != state) {
2312 enum ofputil_port_state of_state;
2315 VLOG_DBG("port %s: STP state changed from %s to %s",
2316 netdev_get_name(ofport->up.netdev),
2317 stp_state_name(ofport->stp_state),
2318 stp_state_name(state));
2319 if (stp_learn_in_state(ofport->stp_state)
2320 != stp_learn_in_state(state)) {
2321 /* xxx Learning action flows should also be flushed. */
2322 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2323 mac_learning_flush(ofproto->ml);
2324 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2325 mcast_snooping_mdb_flush(ofproto->ms);
2327 fwd_change = stp_forward_in_state(ofport->stp_state)
2328 != stp_forward_in_state(state);
2330 ofproto->backer->need_revalidate = REV_STP;
2331 ofport->stp_state = state;
2332 ofport->stp_state_entered = time_msec();
2334 if (fwd_change && ofport->bundle) {
2335 bundle_update(ofport->bundle);
2338 /* Update the STP state bits in the OpenFlow port description. */
2339 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2340 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2341 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2342 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2343 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2345 ofproto_port_set_state(&ofport->up, of_state);
2349 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2350 * caller is responsible for assigning STP port numbers and ensuring
2351 * there are no duplicates. */
2353 set_stp_port(struct ofport *ofport_,
2354 const struct ofproto_port_stp_settings *s)
2356 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2357 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2358 struct stp_port *sp = ofport->stp_port;
2360 if (!s || !s->enable) {
2362 ofport->stp_port = NULL;
2363 stp_port_disable(sp);
2364 update_stp_port_state(ofport);
2367 } else if (sp && stp_port_no(sp) != s->port_num
2368 && ofport == stp_port_get_aux(sp)) {
2369 /* The port-id changed, so disable the old one if it's not
2370 * already in use by another port. */
2371 stp_port_disable(sp);
2374 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2376 /* Set name before enabling the port so that debugging messages can print
2378 stp_port_set_name(sp, netdev_get_name(ofport->up.netdev));
2379 stp_port_enable(sp);
2381 stp_port_set_aux(sp, ofport);
2382 stp_port_set_priority(sp, s->priority);
2383 stp_port_set_path_cost(sp, s->path_cost);
2385 update_stp_port_state(ofport);
2391 get_stp_port_status(struct ofport *ofport_,
2392 struct ofproto_port_stp_status *s)
2394 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2395 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2396 struct stp_port *sp = ofport->stp_port;
2398 if (!ofproto->stp || !sp) {
2404 s->port_id = stp_port_get_id(sp);
2405 s->state = stp_port_get_state(sp);
2406 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2407 s->role = stp_port_get_role(sp);
2413 get_stp_port_stats(struct ofport *ofport_,
2414 struct ofproto_port_stp_stats *s)
2416 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2417 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2418 struct stp_port *sp = ofport->stp_port;
2420 if (!ofproto->stp || !sp) {
2426 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2432 stp_run(struct ofproto_dpif *ofproto)
2435 long long int now = time_msec();
2436 long long int elapsed = now - ofproto->stp_last_tick;
2437 struct stp_port *sp;
2440 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2441 ofproto->stp_last_tick = now;
2443 while (stp_get_changed_port(ofproto->stp, &sp)) {
2444 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2447 update_stp_port_state(ofport);
2451 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2452 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2453 mac_learning_flush(ofproto->ml);
2454 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2455 mcast_snooping_mdb_flush(ofproto->ms);
2461 stp_wait(struct ofproto_dpif *ofproto)
2464 poll_timer_wait(1000);
2468 /* Configures RSTP on 'ofport_' using the settings defined in 's'. The
2469 * caller is responsible for assigning RSTP port numbers and ensuring
2470 * there are no duplicates. */
2472 set_rstp_port(struct ofport *ofport_,
2473 const struct ofproto_port_rstp_settings *s)
2475 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2476 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2477 struct rstp_port *rp = ofport->rstp_port;
2479 if (!s || !s->enable) {
2481 rstp_port_unref(rp);
2482 ofport->rstp_port = NULL;
2483 update_rstp_port_state(ofport);
2488 /* Check if need to add a new port. */
2490 rp = ofport->rstp_port = rstp_add_port(ofproto->rstp);
2493 rstp_port_set(rp, s->port_num, s->priority, s->path_cost,
2494 s->admin_edge_port, s->auto_edge,
2495 s->admin_p2p_mac_state, s->admin_port_state, s->mcheck,
2497 update_rstp_port_state(ofport);
2498 /* Synchronize operational status. */
2499 rstp_port_set_mac_operational(rp, ofport->may_enable);
2503 get_rstp_port_status(struct ofport *ofport_,
2504 struct ofproto_port_rstp_status *s)
2506 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2507 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2508 struct rstp_port *rp = ofport->rstp_port;
2510 if (!ofproto->rstp || !rp) {
2516 rstp_port_get_status(rp, &s->port_id, &s->state, &s->role,
2517 &s->designated_bridge_id, &s->designated_port_id,
2518 &s->designated_path_cost, &s->tx_count,
2519 &s->rx_count, &s->error_count, &s->uptime);
2524 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2527 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2528 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2530 if (ofport->n_qdscp != n_qdscp
2531 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2532 n_qdscp * sizeof *qdscp))) {
2533 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2534 free(ofport->qdscp);
2535 ofport->qdscp = n_qdscp
2536 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2538 ofport->n_qdscp = n_qdscp;
2546 /* Expires all MAC learning entries associated with 'bundle' and forces its
2547 * ofproto to revalidate every flow.
2549 * Normally MAC learning entries are removed only from the ofproto associated
2550 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2551 * are removed from every ofproto. When patch ports and SLB bonds are in use
2552 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2553 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2554 * with the host from which it migrated. */
2556 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2558 struct ofproto_dpif *ofproto = bundle->ofproto;
2559 struct mac_learning *ml = ofproto->ml;
2560 struct mac_entry *mac, *next_mac;
2562 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2563 ovs_rwlock_wrlock(&ml->rwlock);
2564 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2565 if (mac_entry_get_port(ml, mac) == bundle) {
2567 struct ofproto_dpif *o;
2569 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2571 struct mac_entry *e;
2573 ovs_rwlock_wrlock(&o->ml->rwlock);
2574 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2576 mac_learning_expire(o->ml, e);
2578 ovs_rwlock_unlock(&o->ml->rwlock);
2583 mac_learning_expire(ml, mac);
2586 ovs_rwlock_unlock(&ml->rwlock);
2590 bundle_move(struct ofbundle *old, struct ofbundle *new)
2592 struct ofproto_dpif *ofproto = old->ofproto;
2593 struct mac_learning *ml = ofproto->ml;
2594 struct mac_entry *mac, *next_mac;
2596 ovs_assert(new->ofproto == old->ofproto);
2598 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2599 ovs_rwlock_wrlock(&ml->rwlock);
2600 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2601 if (mac_entry_get_port(ml, mac) == old) {
2602 mac_entry_set_port(ml, mac, new);
2605 ovs_rwlock_unlock(&ml->rwlock);
2608 static struct ofbundle *
2609 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2611 struct ofbundle *bundle;
2613 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2614 &ofproto->bundles) {
2615 if (bundle->aux == aux) {
2623 bundle_update(struct ofbundle *bundle)
2625 struct ofport_dpif *port;
2627 bundle->floodable = true;
2628 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2629 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2631 || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2632 || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2633 bundle->floodable = false;
2640 bundle_del_port(struct ofport_dpif *port)
2642 struct ofbundle *bundle = port->bundle;
2644 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2646 list_remove(&port->bundle_node);
2647 port->bundle = NULL;
2650 lacp_slave_unregister(bundle->lacp, port);
2653 bond_slave_unregister(bundle->bond, port);
2656 bundle_update(bundle);
2660 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2661 struct lacp_slave_settings *lacp)
2663 struct ofport_dpif *port;
2665 port = get_ofp_port(bundle->ofproto, ofp_port);
2670 if (port->bundle != bundle) {
2671 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2673 bundle_remove(&port->up);
2676 port->bundle = bundle;
2677 list_push_back(&bundle->ports, &port->bundle_node);
2678 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2680 || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2681 || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2682 bundle->floodable = false;
2686 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2687 lacp_slave_register(bundle->lacp, port, lacp);
2694 bundle_destroy(struct ofbundle *bundle)
2696 struct ofproto_dpif *ofproto;
2697 struct ofport_dpif *port, *next_port;
2703 ofproto = bundle->ofproto;
2704 mbridge_unregister_bundle(ofproto->mbridge, bundle);
2707 xlate_bundle_remove(bundle);
2710 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2711 bundle_del_port(port);
2714 bundle_flush_macs(bundle, true);
2715 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2717 free(bundle->trunks);
2718 lacp_unref(bundle->lacp);
2719 bond_unref(bundle->bond);
2724 bundle_set(struct ofproto *ofproto_, void *aux,
2725 const struct ofproto_bundle_settings *s)
2727 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2728 bool need_flush = false;
2729 struct ofport_dpif *port;
2730 struct ofbundle *bundle;
2731 unsigned long *trunks;
2737 bundle_destroy(bundle_lookup(ofproto, aux));
2741 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2742 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2744 bundle = bundle_lookup(ofproto, aux);
2746 bundle = xmalloc(sizeof *bundle);
2748 bundle->ofproto = ofproto;
2749 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2750 hash_pointer(aux, 0));
2752 bundle->name = NULL;
2754 list_init(&bundle->ports);
2755 bundle->vlan_mode = PORT_VLAN_TRUNK;
2757 bundle->trunks = NULL;
2758 bundle->use_priority_tags = s->use_priority_tags;
2759 bundle->lacp = NULL;
2760 bundle->bond = NULL;
2762 bundle->floodable = true;
2763 mbridge_register_bundle(ofproto->mbridge, bundle);
2766 if (!bundle->name || strcmp(s->name, bundle->name)) {
2768 bundle->name = xstrdup(s->name);
2773 ofproto->lacp_enabled = true;
2774 if (!bundle->lacp) {
2775 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2776 bundle->lacp = lacp_create();
2778 lacp_configure(bundle->lacp, s->lacp);
2780 lacp_unref(bundle->lacp);
2781 bundle->lacp = NULL;
2784 /* Update set of ports. */
2786 for (i = 0; i < s->n_slaves; i++) {
2787 if (!bundle_add_port(bundle, s->slaves[i],
2788 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2792 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2793 struct ofport_dpif *next_port;
2795 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2796 for (i = 0; i < s->n_slaves; i++) {
2797 if (s->slaves[i] == port->up.ofp_port) {
2802 bundle_del_port(port);
2806 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2808 if (list_is_empty(&bundle->ports)) {
2809 bundle_destroy(bundle);
2813 /* Set VLAN tagging mode */
2814 if (s->vlan_mode != bundle->vlan_mode
2815 || s->use_priority_tags != bundle->use_priority_tags) {
2816 bundle->vlan_mode = s->vlan_mode;
2817 bundle->use_priority_tags = s->use_priority_tags;
2822 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2823 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2825 if (vlan != bundle->vlan) {
2826 bundle->vlan = vlan;
2830 /* Get trunked VLANs. */
2831 switch (s->vlan_mode) {
2832 case PORT_VLAN_ACCESS:
2836 case PORT_VLAN_TRUNK:
2837 trunks = CONST_CAST(unsigned long *, s->trunks);
2840 case PORT_VLAN_NATIVE_UNTAGGED:
2841 case PORT_VLAN_NATIVE_TAGGED:
2842 if (vlan != 0 && (!s->trunks
2843 || !bitmap_is_set(s->trunks, vlan)
2844 || bitmap_is_set(s->trunks, 0))) {
2845 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2847 trunks = bitmap_clone(s->trunks, 4096);
2849 trunks = bitmap_allocate1(4096);
2851 bitmap_set1(trunks, vlan);
2852 bitmap_set0(trunks, 0);
2854 trunks = CONST_CAST(unsigned long *, s->trunks);
2861 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2862 free(bundle->trunks);
2863 if (trunks == s->trunks) {
2864 bundle->trunks = vlan_bitmap_clone(trunks);
2866 bundle->trunks = trunks;
2871 if (trunks != s->trunks) {
2876 if (!list_is_short(&bundle->ports)) {
2877 bundle->ofproto->has_bonded_bundles = true;
2879 if (bond_reconfigure(bundle->bond, s->bond)) {
2880 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2883 bundle->bond = bond_create(s->bond, ofproto);
2884 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2887 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2888 bond_slave_register(bundle->bond, port,
2889 port->up.ofp_port, port->up.netdev);
2892 bond_unref(bundle->bond);
2893 bundle->bond = NULL;
2896 /* If we changed something that would affect MAC learning, un-learn
2897 * everything on this port and force flow revalidation. */
2899 bundle_flush_macs(bundle, false);
2906 bundle_remove(struct ofport *port_)
2908 struct ofport_dpif *port = ofport_dpif_cast(port_);
2909 struct ofbundle *bundle = port->bundle;
2912 bundle_del_port(port);
2913 if (list_is_empty(&bundle->ports)) {
2914 bundle_destroy(bundle);
2915 } else if (list_is_short(&bundle->ports)) {
2916 bond_unref(bundle->bond);
2917 bundle->bond = NULL;
2923 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2925 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2926 struct ofport_dpif *port = port_;
2927 uint8_t ea[ETH_ADDR_LEN];
2930 error = netdev_get_etheraddr(port->up.netdev, ea);
2932 struct dp_packet packet;
2935 dp_packet_init(&packet, 0);
2936 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2938 memcpy(packet_pdu, pdu, pdu_size);
2940 ofproto_dpif_send_packet(port, &packet);
2941 dp_packet_uninit(&packet);
2943 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2944 "%s (%s)", port->bundle->name,
2945 netdev_get_name(port->up.netdev), ovs_strerror(error));
2950 bundle_send_learning_packets(struct ofbundle *bundle)
2952 struct ofproto_dpif *ofproto = bundle->ofproto;
2953 struct dp_packet *learning_packet;
2954 int error, n_packets, n_errors;
2955 struct mac_entry *e;
2956 struct ovs_list packets;
2958 list_init(&packets);
2959 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2960 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2961 if (mac_entry_get_port(ofproto->ml, e) != bundle) {
2964 learning_packet = bond_compose_learning_packet(bundle->bond,
2967 /* Temporarily use 'frame' as a private pointer (see below). */
2968 ovs_assert(learning_packet->frame == dp_packet_data(learning_packet));
2969 learning_packet->frame = port_void;
2970 list_push_back(&packets, &learning_packet->list_node);
2973 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2975 error = n_packets = n_errors = 0;
2976 LIST_FOR_EACH (learning_packet, list_node, &packets) {
2978 void *port_void = learning_packet->frame;
2980 /* Restore 'frame'. */
2981 learning_packet->frame = dp_packet_data(learning_packet);
2982 ret = ofproto_dpif_send_packet(port_void, learning_packet);
2989 dp_packet_list_delete(&packets);
2992 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2993 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2994 "packets, last error was: %s",
2995 bundle->name, n_errors, n_packets, ovs_strerror(error));
2997 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2998 bundle->name, n_packets);
3003 bundle_run(struct ofbundle *bundle)
3006 lacp_run(bundle->lacp, send_pdu_cb);
3009 struct ofport_dpif *port;
3011 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
3012 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
3015 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
3016 bundle->ofproto->backer->need_revalidate = REV_BOND;
3019 if (bond_should_send_learning_packets(bundle->bond)) {
3020 bundle_send_learning_packets(bundle);
3026 bundle_wait(struct ofbundle *bundle)
3029 lacp_wait(bundle->lacp);
3032 bond_wait(bundle->bond);
3039 mirror_set__(struct ofproto *ofproto_, void *aux,
3040 const struct ofproto_mirror_settings *s)
3042 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3043 struct ofbundle **srcs, **dsts;
3048 mirror_destroy(ofproto->mbridge, aux);
3052 srcs = xmalloc(s->n_srcs * sizeof *srcs);
3053 dsts = xmalloc(s->n_dsts * sizeof *dsts);
3055 for (i = 0; i < s->n_srcs; i++) {
3056 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
3059 for (i = 0; i < s->n_dsts; i++) {
3060 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
3063 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
3064 s->n_dsts, s->src_vlans,
3065 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
3072 mirror_get_stats__(struct ofproto *ofproto, void *aux,
3073 uint64_t *packets, uint64_t *bytes)
3075 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
3080 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3082 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3083 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3084 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3085 mac_learning_flush(ofproto->ml);
3087 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3092 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3094 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3095 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3096 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
3100 forward_bpdu_changed(struct ofproto *ofproto_)
3102 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3103 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3107 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3110 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3111 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3112 mac_learning_set_idle_time(ofproto->ml, idle_time);
3113 mac_learning_set_max_entries(ofproto->ml, max_entries);
3114 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3117 /* Configures multicast snooping on 'ofport' using the settings
3118 * defined in 's'. */
3120 set_mcast_snooping(struct ofproto *ofproto_,
3121 const struct ofproto_mcast_snooping_settings *s)
3123 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3125 /* Only revalidate flows if the configuration changed. */
3126 if (!s != !ofproto->ms) {
3127 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3132 ofproto->ms = mcast_snooping_create();
3135 ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3136 mcast_snooping_set_idle_time(ofproto->ms, s->idle_time);
3137 mcast_snooping_set_max_entries(ofproto->ms, s->max_entries);
3138 if (mcast_snooping_set_flood_unreg(ofproto->ms, s->flood_unreg)) {
3139 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3141 ovs_rwlock_unlock(&ofproto->ms->rwlock);
3143 mcast_snooping_unref(ofproto->ms);
3150 /* Configures multicast snooping port's flood settings on 'ofproto'. */
3152 set_mcast_snooping_port(struct ofproto *ofproto_, void *aux,
3153 const struct ofproto_mcast_snooping_port_settings *s)
3155 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3156 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3158 if (ofproto->ms && s) {
3159 ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3160 mcast_snooping_set_port_flood(ofproto->ms, bundle, s->flood);
3161 mcast_snooping_set_port_flood_reports(ofproto->ms, bundle,
3163 ovs_rwlock_unlock(&ofproto->ms->rwlock);
3171 static struct ofport_dpif *
3172 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
3174 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3175 return ofport ? ofport_dpif_cast(ofport) : NULL;
3179 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3180 struct ofproto_port *ofproto_port,
3181 struct dpif_port *dpif_port)
3183 ofproto_port->name = dpif_port->name;
3184 ofproto_port->type = dpif_port->type;
3185 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3189 ofport_update_peer(struct ofport_dpif *ofport)
3191 const struct ofproto_dpif *ofproto;
3192 struct dpif_backer *backer;
3195 if (!netdev_vport_is_patch(ofport->up.netdev)) {
3199 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
3200 backer->need_revalidate = REV_RECONFIGURE;
3203 ofport->peer->peer = NULL;
3204 ofport->peer = NULL;
3207 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
3212 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3213 struct ofport *peer_ofport;
3214 struct ofport_dpif *peer;
3217 if (ofproto->backer != backer) {
3221 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
3226 peer = ofport_dpif_cast(peer_ofport);
3227 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
3228 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
3230 ofport->peer = peer;
3231 ofport->peer->peer = ofport;
3241 port_run(struct ofport_dpif *ofport)
3243 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3244 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3245 bool enable = netdev_get_carrier(ofport->up.netdev);
3246 bool cfm_enable = false;
3247 bool bfd_enable = false;
3249 ofport->carrier_seq = carrier_seq;
3252 int cfm_opup = cfm_get_opup(ofport->cfm);
3254 cfm_enable = !cfm_get_fault(ofport->cfm);
3256 if (cfm_opup >= 0) {
3257 cfm_enable = cfm_enable && cfm_opup;
3262 bfd_enable = bfd_forwarding(ofport->bfd);
3265 if (ofport->bfd || ofport->cfm) {
3266 enable = enable && (cfm_enable || bfd_enable);
3269 if (ofport->bundle) {
3270 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3271 if (carrier_changed) {
3272 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3276 if (ofport->may_enable != enable) {
3277 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3279 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3281 if (ofport->rstp_port) {
3282 rstp_port_set_mac_operational(ofport->rstp_port, enable);
3286 ofport->may_enable = enable;
3290 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3291 struct ofproto_port *ofproto_port)
3293 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3294 struct dpif_port dpif_port;
3297 if (sset_contains(&ofproto->ghost_ports, devname)) {
3298 const char *type = netdev_get_type_from_name(devname);
3300 /* We may be called before ofproto->up.port_by_name is populated with
3301 * the appropriate ofport. For this reason, we must get the name and
3302 * type from the netdev layer directly. */
3304 const struct ofport *ofport;
3306 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3307 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3308 ofproto_port->name = xstrdup(devname);
3309 ofproto_port->type = xstrdup(type);
3315 if (!sset_contains(&ofproto->ports, devname)) {
3318 error = dpif_port_query_by_name(ofproto->backer->dpif,
3319 devname, &dpif_port);
3321 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3327 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3329 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3330 const char *devname = netdev_get_name(netdev);
3331 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3332 const char *dp_port_name;
3334 if (netdev_vport_is_patch(netdev)) {
3335 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3339 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3340 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3341 odp_port_t port_no = ODPP_NONE;
3344 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3348 if (netdev_get_tunnel_config(netdev)) {
3349 simap_put(&ofproto->backer->tnl_backers,
3350 dp_port_name, odp_to_u32(port_no));
3354 if (netdev_get_tunnel_config(netdev)) {
3355 sset_add(&ofproto->ghost_ports, devname);
3357 sset_add(&ofproto->ports, devname);
3363 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3365 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3366 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3373 sset_find_and_delete(&ofproto->ghost_ports,
3374 netdev_get_name(ofport->up.netdev));
3375 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3376 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3377 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3379 /* The caller is going to close ofport->up.netdev. If this is a
3380 * bonded port, then the bond is using that netdev, so remove it
3381 * from the bond. The client will need to reconfigure everything
3382 * after deleting ports, so then the slave will get re-added. */
3383 bundle_remove(&ofport->up);
3390 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3392 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3395 error = netdev_get_stats(ofport->up.netdev, stats);
3397 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3398 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3400 ovs_mutex_lock(&ofproto->stats_mutex);
3401 /* ofproto->stats.tx_packets represents packets that we created
3402 * internally and sent to some port (e.g. packets sent with
3403 * ofproto_dpif_send_packet()). Account for them as if they had
3404 * come from OFPP_LOCAL and got forwarded. */
3406 if (stats->rx_packets != UINT64_MAX) {
3407 stats->rx_packets += ofproto->stats.tx_packets;
3410 if (stats->rx_bytes != UINT64_MAX) {
3411 stats->rx_bytes += ofproto->stats.tx_bytes;
3414 /* ofproto->stats.rx_packets represents packets that were received on
3415 * some port and we processed internally and dropped (e.g. STP).
3416 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3418 if (stats->tx_packets != UINT64_MAX) {
3419 stats->tx_packets += ofproto->stats.rx_packets;
3422 if (stats->tx_bytes != UINT64_MAX) {
3423 stats->tx_bytes += ofproto->stats.rx_bytes;
3425 ovs_mutex_unlock(&ofproto->stats_mutex);
3432 port_get_lacp_stats(const struct ofport *ofport_, struct lacp_slave_stats *stats)
3434 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3435 if (ofport->bundle && ofport->bundle->lacp) {
3436 if (lacp_get_slave_stats(ofport->bundle->lacp, ofport, stats)) {
3443 struct port_dump_state {
3448 struct ofproto_port port;
3453 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3455 *statep = xzalloc(sizeof(struct port_dump_state));
3460 port_dump_next(const struct ofproto *ofproto_, void *state_,
3461 struct ofproto_port *port)
3463 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3464 struct port_dump_state *state = state_;
3465 const struct sset *sset;
3466 struct sset_node *node;
3468 if (state->has_port) {
3469 ofproto_port_destroy(&state->port);
3470 state->has_port = false;
3472 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3473 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3476 error = port_query_by_name(ofproto_, node->name, &state->port);
3478 *port = state->port;
3479 state->has_port = true;
3481 } else if (error != ENODEV) {
3486 if (!state->ghost) {
3487 state->ghost = true;
3490 return port_dump_next(ofproto_, state_, port);
3497 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3499 struct port_dump_state *state = state_;
3501 if (state->has_port) {
3502 ofproto_port_destroy(&state->port);
3509 port_poll(const struct ofproto *ofproto_, char **devnamep)
3511 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3513 if (ofproto->port_poll_errno) {
3514 int error = ofproto->port_poll_errno;
3515 ofproto->port_poll_errno = 0;
3519 if (sset_is_empty(&ofproto->port_poll_set)) {
3523 *devnamep = sset_pop(&ofproto->port_poll_set);
3528 port_poll_wait(const struct ofproto *ofproto_)
3530 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3531 dpif_port_poll_wait(ofproto->backer->dpif);
3535 port_is_lacp_current(const struct ofport *ofport_)
3537 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3538 return (ofport->bundle && ofport->bundle->lacp
3539 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3543 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3544 * then delete it entirely. */
3546 rule_expire(struct rule_dpif *rule)
3547 OVS_REQUIRES(ofproto_mutex)
3549 uint16_t hard_timeout, idle_timeout;
3550 long long int now = time_msec();
3553 hard_timeout = rule->up.hard_timeout;
3554 idle_timeout = rule->up.idle_timeout;
3556 /* Has 'rule' expired? */
3558 long long int modified;
3560 ovs_mutex_lock(&rule->up.mutex);
3561 modified = rule->up.modified;
3562 ovs_mutex_unlock(&rule->up.mutex);
3564 if (now > modified + hard_timeout * 1000) {
3565 reason = OFPRR_HARD_TIMEOUT;
3569 if (reason < 0 && idle_timeout) {
3572 ovs_mutex_lock(&rule->stats_mutex);
3573 used = rule->stats.used;
3574 ovs_mutex_unlock(&rule->stats_mutex);
3576 if (now > used + idle_timeout * 1000) {
3577 reason = OFPRR_IDLE_TIMEOUT;
3582 COVERAGE_INC(ofproto_dpif_expired);
3583 ofproto_rule_expire(&rule->up, reason);
3587 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3588 * 'flow' must reflect the data in 'packet'. */
3590 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3591 const struct flow *flow,
3592 struct rule_dpif *rule,
3593 const struct ofpact *ofpacts, size_t ofpacts_len,
3594 struct dp_packet *packet)
3596 struct dpif_flow_stats stats;
3597 struct xlate_out xout;
3598 struct xlate_in xin;
3600 struct dpif_execute execute;
3603 ovs_assert((rule != NULL) != (ofpacts != NULL));
3605 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3608 rule_dpif_credit_stats(rule, &stats);
3611 xlate_in_init(&xin, ofproto, flow, flow->in_port.ofp_port, rule,
3612 stats.tcp_flags, packet);
3613 xin.ofpacts = ofpacts;
3614 xin.ofpacts_len = ofpacts_len;
3615 xin.resubmit_stats = &stats;
3616 xlate_actions(&xin, &xout);
3618 execute.actions = xout.odp_actions->data;
3619 execute.actions_len = xout.odp_actions->size;
3621 pkt_metadata_from_flow(&packet->md, flow);
3622 execute.packet = packet;
3623 execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
3624 execute.probe = false;
3626 /* Fix up in_port. */
3627 in_port = flow->in_port.ofp_port;
3628 if (in_port == OFPP_NONE) {
3629 in_port = OFPP_LOCAL;
3631 execute.packet->md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
3633 error = dpif_execute(ofproto->backer->dpif, &execute);
3635 xlate_out_uninit(&xout);
3641 rule_dpif_credit_stats(struct rule_dpif *rule,
3642 const struct dpif_flow_stats *stats)
3644 ovs_mutex_lock(&rule->stats_mutex);
3645 rule->stats.n_packets += stats->n_packets;
3646 rule->stats.n_bytes += stats->n_bytes;
3647 rule->stats.used = MAX(rule->stats.used, stats->used);
3648 ovs_mutex_unlock(&rule->stats_mutex);
3652 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
3653 OVS_REQUIRES(rule->up.mutex)
3655 return rule->up.flow_cookie;
3659 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
3660 uint16_t hard_timeout)
3662 ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
3665 /* Returns 'rule''s actions. The returned actions are RCU-protected, and can
3666 * be read until the calling thread quiesces. */
3667 const struct rule_actions *
3668 rule_dpif_get_actions(const struct rule_dpif *rule)
3670 return rule_get_actions(&rule->up);
3673 /* Sets 'rule''s recirculation id. */
3675 rule_dpif_set_recirc_id(struct rule_dpif *rule, uint32_t id)
3676 OVS_REQUIRES(rule->up.mutex)
3678 ovs_assert(!rule->recirc_id);
3679 rule->recirc_id = id;
3682 /* Returns 'rule''s recirculation id. */
3684 rule_dpif_get_recirc_id(struct rule_dpif *rule)
3685 OVS_REQUIRES(rule->up.mutex)
3687 if (!rule->recirc_id) {
3688 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3690 rule_dpif_set_recirc_id(rule, ofproto_dpif_alloc_recirc_id(ofproto));
3692 return rule->recirc_id;
3695 /* Sets 'rule''s recirculation id. */
3697 rule_set_recirc_id(struct rule *rule_, uint32_t id)
3699 struct rule_dpif *rule = rule_dpif_cast(rule_);
3701 ovs_mutex_lock(&rule->up.mutex);
3702 rule_dpif_set_recirc_id(rule, id);
3703 ovs_mutex_unlock(&rule->up.mutex);
3706 /* Lookup 'flow' in table 0 of 'ofproto''s classifier.
3707 * If 'wc' is non-null, sets the fields that were relevant as part of
3708 * the lookup. Returns the table id where a match or miss occurred via
3709 * 'table_id'. This will be zero unless there was a miss and
3710 * OFPTC11_TABLE_MISS_CONTINUE is in effect for the sequence of tables
3711 * where misses occur, or TBL_INTERNAL if the rule has a non-zero
3712 * recirculation ID, and a match was found in the internal table, or if
3713 * there was no match and one of the special rules (drop_frags_rule,
3714 * miss_rule, or no_packet_in_rule) was returned.
3716 * The return value is the found rule, which is valid at least until the next
3717 * RCU quiescent period. If the rule needs to stay around longer,
3718 * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3719 * on it before this returns. */
3721 rule_dpif_lookup(struct ofproto_dpif *ofproto, struct flow *flow,
3722 struct flow_wildcards *wc, bool take_ref,
3723 const struct dpif_flow_stats *stats, uint8_t *table_id)
3727 if (ofproto_dpif_get_enable_recirc(ofproto)) {
3728 /* Always exactly match recirc_id since datapath supports
3731 wc->masks.recirc_id = UINT32_MAX;
3733 *table_id = rule_dpif_lookup_get_init_table_id(flow);
3736 return rule_dpif_lookup_from_table(ofproto, flow, wc, take_ref, stats,
3737 table_id, flow->in_port.ofp_port, true,
3741 /* The returned rule (if any) is valid at least until the next RCU quiescent
3742 * period. If the rule needs to stay around longer, a non-zero 'take_ref'
3743 * must be passed in to cause a reference to be taken on it.
3745 * 'flow' is non-const to allow for temporary modifications during the lookup.
3746 * Any changes are restored before returning. */
3747 static struct rule_dpif *
3748 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, uint8_t table_id,
3749 struct flow *flow, struct flow_wildcards *wc,
3752 struct classifier *cls = &ofproto->up.tables[table_id].cls;
3753 const struct cls_rule *cls_rule;
3754 struct rule_dpif *rule;
3757 cls_rule = classifier_lookup(cls, flow, wc);
3759 rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
3761 /* Try again if the rule was released before we get the reference. */
3762 } while (rule && take_ref && !rule_dpif_try_ref(rule));
3767 /* Look up 'flow' in 'ofproto''s classifier starting from table '*table_id'.
3768 * Returns the rule that was found, which may be one of the special rules
3769 * according to packet miss hadling. If 'may_packet_in' is false, returning of
3770 * the miss_rule (which issues packet ins for the controller) is avoided.
3771 * Updates 'wc', if nonnull, to reflect the fields that were used during the
3774 * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3775 * if none is found then the table miss configuration for that table is
3776 * honored, which can result in additional lookups in other OpenFlow tables.
3777 * In this case the function updates '*table_id' to reflect the final OpenFlow
3778 * table that was searched.
3780 * If 'honor_table_miss' is false, then only one table lookup occurs, in
3783 * The rule is returned in '*rule', which is valid at least until the next
3784 * RCU quiescent period. If the '*rule' needs to stay around longer,
3785 * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3786 * on it before this returns.
3788 * 'in_port' allows the lookup to take place as if the in port had the value
3789 * 'in_port'. This is needed for resubmit action support.
3791 * 'flow' is non-const to allow for temporary modifications during the lookup.
3792 * Any changes are restored before returning. */
3794 rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto, struct flow *flow,
3795 struct flow_wildcards *wc, bool take_ref,
3796 const struct dpif_flow_stats *stats,
3797 uint8_t *table_id, ofp_port_t in_port,
3798 bool may_packet_in, bool honor_table_miss)
3800 ovs_be16 old_tp_src = flow->tp_src, old_tp_dst = flow->tp_dst;
3801 ofp_port_t old_in_port = flow->in_port.ofp_port;
3802 enum ofputil_table_miss miss_config;
3803 struct rule_dpif *rule;
3806 /* We always unwildcard nw_frag (for IP), so they
3807 * need not be unwildcarded here. */
3808 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3809 && ofproto->up.frag_handling != OFPC_FRAG_NX_MATCH) {
3810 if (ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3811 /* We must pretend that transport ports are unavailable. */
3812 flow->tp_src = htons(0);
3813 flow->tp_dst = htons(0);
3815 /* Must be OFPC_FRAG_DROP (we don't have OFPC_FRAG_REASM).
3816 * Use the drop_frags_rule (which cannot disappear). */
3817 rule = ofproto->drop_frags_rule;
3819 rule_dpif_ref(rule);
3822 struct oftable *tbl = &ofproto->up.tables[*table_id];
3825 atomic_add_relaxed(&tbl->n_matched, stats->n_packets, &orig);
3831 /* Look up a flow with 'in_port' as the input port. Then restore the
3832 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
3833 * have surprising behavior). */
3834 flow->in_port.ofp_port = in_port;
3836 /* Our current implementation depends on n_tables == N_TABLES, and
3837 * TBL_INTERNAL being the last table. */
3838 BUILD_ASSERT_DECL(N_TABLES == TBL_INTERNAL + 1);
3840 miss_config = OFPUTIL_TABLE_MISS_CONTINUE;
3842 for (next_id = *table_id;
3843 next_id < ofproto->up.n_tables;
3844 next_id++, next_id += (next_id == TBL_INTERNAL))
3846 *table_id = next_id;
3847 rule = rule_dpif_lookup_in_table(ofproto, next_id, flow, wc, take_ref);
3849 struct oftable *tbl = &ofproto->up.tables[next_id];
3852 atomic_add_relaxed(rule ? &tbl->n_matched : &tbl->n_missed,
3853 stats->n_packets, &orig);
3856 goto out; /* Match. */
3858 if (honor_table_miss) {
3859 miss_config = ofproto_table_get_miss_config(&ofproto->up,
3861 if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE) {
3868 rule = ofproto->no_packet_in_rule;
3869 if (may_packet_in) {
3870 if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE
3871 || miss_config == OFPUTIL_TABLE_MISS_CONTROLLER) {
3872 struct ofport_dpif *port;
3874 port = get_ofp_port(ofproto, old_in_port);
3876 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
3878 } else if (!(port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN)) {
3879 rule = ofproto->miss_rule;
3881 } else if (miss_config == OFPUTIL_TABLE_MISS_DEFAULT &&
3882 connmgr_wants_packet_in_on_miss(ofproto->up.connmgr)) {
3883 rule = ofproto->miss_rule;
3887 rule_dpif_ref(rule);
3890 /* Restore port numbers, as they may have been modified above. */
3891 flow->tp_src = old_tp_src;
3892 flow->tp_dst = old_tp_dst;
3893 /* Restore the old in port. */
3894 flow->in_port.ofp_port = old_in_port;
3900 complete_operation(struct rule_dpif *rule)
3901 OVS_REQUIRES(ofproto_mutex)
3903 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3905 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3908 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
3910 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
3913 static struct rule *
3916 struct rule_dpif *rule = xmalloc(sizeof *rule);
3921 rule_dealloc(struct rule *rule_)
3923 struct rule_dpif *rule = rule_dpif_cast(rule_);
3928 rule_construct(struct rule *rule_)
3929 OVS_NO_THREAD_SAFETY_ANALYSIS
3931 struct rule_dpif *rule = rule_dpif_cast(rule_);
3932 ovs_mutex_init_adaptive(&rule->stats_mutex);
3933 rule->stats.n_packets = 0;
3934 rule->stats.n_bytes = 0;
3935 rule->stats.used = rule->up.modified;
3936 rule->recirc_id = 0;
3942 rule_insert(struct rule *rule_)
3943 OVS_REQUIRES(ofproto_mutex)
3945 struct rule_dpif *rule = rule_dpif_cast(rule_);
3946 complete_operation(rule);
3951 rule_delete(struct rule *rule_)
3952 OVS_REQUIRES(ofproto_mutex)
3954 struct rule_dpif *rule = rule_dpif_cast(rule_);
3955 complete_operation(rule);
3959 rule_destruct(struct rule *rule_)
3961 struct rule_dpif *rule = rule_dpif_cast(rule_);
3963 ovs_mutex_destroy(&rule->stats_mutex);
3964 if (rule->recirc_id) {
3965 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3967 ofproto_dpif_free_recirc_id(ofproto, rule->recirc_id);
3972 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
3973 long long int *used)
3975 struct rule_dpif *rule = rule_dpif_cast(rule_);
3977 ovs_mutex_lock(&rule->stats_mutex);
3978 *packets = rule->stats.n_packets;
3979 *bytes = rule->stats.n_bytes;
3980 *used = rule->stats.used;
3981 ovs_mutex_unlock(&rule->stats_mutex);
3985 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
3986 struct dp_packet *packet)
3988 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3990 ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
3994 rule_execute(struct rule *rule, const struct flow *flow,
3995 struct dp_packet *packet)
3997 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
3998 dp_packet_delete(packet);
4003 rule_modify_actions(struct rule *rule_, bool reset_counters)
4004 OVS_REQUIRES(ofproto_mutex)
4006 struct rule_dpif *rule = rule_dpif_cast(rule_);
4008 if (reset_counters) {
4009 ovs_mutex_lock(&rule->stats_mutex);
4010 rule->stats.n_packets = 0;
4011 rule->stats.n_bytes = 0;
4012 ovs_mutex_unlock(&rule->stats_mutex);
4015 complete_operation(rule);
4018 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
4020 return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
4023 static struct ofgroup *
4026 struct group_dpif *group = xzalloc(sizeof *group);
4031 group_dealloc(struct ofgroup *group_)
4033 struct group_dpif *group = group_dpif_cast(group_);
4038 group_construct_stats(struct group_dpif *group)
4039 OVS_REQUIRES(group->stats_mutex)
4041 struct ofputil_bucket *bucket;
4042 const struct ovs_list *buckets;
4044 group->packet_count = 0;
4045 group->byte_count = 0;
4047 group_dpif_get_buckets(group, &buckets);
4048 LIST_FOR_EACH (bucket, list_node, buckets) {
4049 bucket->stats.packet_count = 0;
4050 bucket->stats.byte_count = 0;
4055 group_dpif_credit_stats(struct group_dpif *group,
4056 struct ofputil_bucket *bucket,
4057 const struct dpif_flow_stats *stats)
4059 ovs_mutex_lock(&group->stats_mutex);
4060 group->packet_count += stats->n_packets;
4061 group->byte_count += stats->n_bytes;
4063 bucket->stats.packet_count += stats->n_packets;
4064 bucket->stats.byte_count += stats->n_bytes;
4065 } else { /* Credit to all buckets */
4066 const struct ovs_list *buckets;
4068 group_dpif_get_buckets(group, &buckets);
4069 LIST_FOR_EACH (bucket, list_node, buckets) {
4070 bucket->stats.packet_count += stats->n_packets;
4071 bucket->stats.byte_count += stats->n_bytes;
4074 ovs_mutex_unlock(&group->stats_mutex);
4078 group_construct(struct ofgroup *group_)
4080 struct group_dpif *group = group_dpif_cast(group_);
4081 const struct ofputil_bucket *bucket;
4083 /* Prevent group chaining because our locking structure makes it hard to
4084 * implement deadlock-free. (See xlate_group_resource_check().) */
4085 LIST_FOR_EACH (bucket, list_node, &group->up.buckets) {
4086 const struct ofpact *a;
4088 OFPACT_FOR_EACH (a, bucket->ofpacts, bucket->ofpacts_len) {
4089 if (a->type == OFPACT_GROUP) {
4090 return OFPERR_OFPGMFC_CHAINING_UNSUPPORTED;
4095 ovs_mutex_init_adaptive(&group->stats_mutex);
4096 ovs_mutex_lock(&group->stats_mutex);
4097 group_construct_stats(group);
4098 ovs_mutex_unlock(&group->stats_mutex);
4103 group_destruct(struct ofgroup *group_)
4105 struct group_dpif *group = group_dpif_cast(group_);
4106 ovs_mutex_destroy(&group->stats_mutex);
4110 group_modify(struct ofgroup *group_)
4112 struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
4114 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4120 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
4122 struct group_dpif *group = group_dpif_cast(group_);
4123 struct ofputil_bucket *bucket;
4124 const struct ovs_list *buckets;
4125 struct bucket_counter *bucket_stats;
4127 ovs_mutex_lock(&group->stats_mutex);
4128 ogs->packet_count = group->packet_count;
4129 ogs->byte_count = group->byte_count;
4131 group_dpif_get_buckets(group, &buckets);
4132 bucket_stats = ogs->bucket_stats;
4133 LIST_FOR_EACH (bucket, list_node, buckets) {
4134 bucket_stats->packet_count = bucket->stats.packet_count;
4135 bucket_stats->byte_count = bucket->stats.byte_count;
4138 ovs_mutex_unlock(&group->stats_mutex);
4143 /* If the group exists, this function increments the groups's reference count.
4145 * Make sure to call group_dpif_unref() after no longer needing to maintain
4146 * a reference to the group. */
4148 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
4149 struct group_dpif **group)
4151 struct ofgroup *ofgroup;
4154 found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
4155 *group = found ? group_dpif_cast(ofgroup) : NULL;
4161 group_dpif_get_buckets(const struct group_dpif *group,
4162 const struct ovs_list **buckets)
4164 *buckets = &group->up.buckets;
4167 enum ofp11_group_type
4168 group_dpif_get_type(const struct group_dpif *group)
4170 return group->up.type;
4173 /* Sends 'packet' out 'ofport'.
4174 * May modify 'packet'.
4175 * Returns 0 if successful, otherwise a positive errno value. */
4177 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct dp_packet *packet)
4179 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4182 error = xlate_send_packet(ofport, packet);
4184 ovs_mutex_lock(&ofproto->stats_mutex);
4185 ofproto->stats.tx_packets++;
4186 ofproto->stats.tx_bytes += dp_packet_size(packet);
4187 ovs_mutex_unlock(&ofproto->stats_mutex);
4191 /* Return the version string of the datapath that backs up
4195 get_datapath_version(const struct ofproto *ofproto_)
4197 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4199 return ofproto->backer->dp_version_string;
4203 set_frag_handling(struct ofproto *ofproto_,
4204 enum ofp_config_flags frag_handling)
4206 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4207 if (frag_handling != OFPC_FRAG_REASM) {
4208 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4216 packet_out(struct ofproto *ofproto_, struct dp_packet *packet,
4217 const struct flow *flow,
4218 const struct ofpact *ofpacts, size_t ofpacts_len)
4220 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4222 ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
4223 ofpacts_len, packet);
4230 set_netflow(struct ofproto *ofproto_,
4231 const struct netflow_options *netflow_options)
4233 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4235 if (netflow_options) {
4236 if (!ofproto->netflow) {
4237 ofproto->netflow = netflow_create();
4238 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4240 return netflow_set_options(ofproto->netflow, netflow_options);
4241 } else if (ofproto->netflow) {
4242 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4243 netflow_unref(ofproto->netflow);
4244 ofproto->netflow = NULL;
4251 get_netflow_ids(const struct ofproto *ofproto_,
4252 uint8_t *engine_type, uint8_t *engine_id)
4254 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4256 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
4259 static struct ofproto_dpif *
4260 ofproto_dpif_lookup(const char *name)
4262 struct ofproto_dpif *ofproto;
4264 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
4265 hash_string(name, 0), &all_ofproto_dpifs) {
4266 if (!strcmp(ofproto->up.name, name)) {
4274 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
4275 const char *argv[], void *aux OVS_UNUSED)
4277 struct ofproto_dpif *ofproto;
4280 ofproto = ofproto_dpif_lookup(argv[1]);
4282 unixctl_command_reply_error(conn, "no such bridge");
4285 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4286 mac_learning_flush(ofproto->ml);
4287 ovs_rwlock_unlock(&ofproto->ml->rwlock);
4289 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4290 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4291 mac_learning_flush(ofproto->ml);
4292 ovs_rwlock_unlock(&ofproto->ml->rwlock);
4296 unixctl_command_reply(conn, "table successfully flushed");
4300 ofproto_unixctl_mcast_snooping_flush(struct unixctl_conn *conn, int argc,
4301 const char *argv[], void *aux OVS_UNUSED)
4303 struct ofproto_dpif *ofproto;
4306 ofproto = ofproto_dpif_lookup(argv[1]);
4308 unixctl_command_reply_error(conn, "no such bridge");
4312 if (!mcast_snooping_enabled(ofproto->ms)) {
4313 unixctl_command_reply_error(conn, "multicast snooping is disabled");
4316 mcast_snooping_mdb_flush(ofproto->ms);
4318 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4319 if (!mcast_snooping_enabled(ofproto->ms)) {
4322 mcast_snooping_mdb_flush(ofproto->ms);
4326 unixctl_command_reply(conn, "table successfully flushed");
4329 static struct ofport_dpif *
4330 ofbundle_get_a_port(const struct ofbundle *bundle)
4332 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
4337 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4338 const char *argv[], void *aux OVS_UNUSED)
4340 struct ds ds = DS_EMPTY_INITIALIZER;
4341 const struct ofproto_dpif *ofproto;
4342 const struct mac_entry *e;
4344 ofproto = ofproto_dpif_lookup(argv[1]);
4346 unixctl_command_reply_error(conn, "no such bridge");
4350 ds_put_cstr(&ds, " port VLAN MAC Age\n");
4351 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
4352 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
4353 struct ofbundle *bundle = mac_entry_get_port(ofproto->ml, e);
4354 char name[OFP_MAX_PORT_NAME_LEN];
4356 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4358 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
4359 name, e->vlan, ETH_ADDR_ARGS(e->mac),
4360 mac_entry_age(ofproto->ml, e));
4362 ovs_rwlock_unlock(&ofproto->ml->rwlock);
4363 unixctl_command_reply(conn, ds_cstr(&ds));
4368 ofproto_unixctl_mcast_snooping_show(struct unixctl_conn *conn,
4369 int argc OVS_UNUSED,
4371 void *aux OVS_UNUSED)
4373 struct ds ds = DS_EMPTY_INITIALIZER;
4374 const struct ofproto_dpif *ofproto;
4375 const struct ofbundle *bundle;
4376 const struct mcast_group *grp;
4377 struct mcast_group_bundle *b;
4378 struct mcast_mrouter_bundle *mrouter;
4380 ofproto = ofproto_dpif_lookup(argv[1]);
4382 unixctl_command_reply_error(conn, "no such bridge");
4386 if (!mcast_snooping_enabled(ofproto->ms)) {
4387 unixctl_command_reply_error(conn, "multicast snooping is disabled");
4391 ds_put_cstr(&ds, " port VLAN GROUP Age\n");
4392 ovs_rwlock_rdlock(&ofproto->ms->rwlock);
4393 LIST_FOR_EACH (grp, group_node, &ofproto->ms->group_lru) {
4394 LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
4395 char name[OFP_MAX_PORT_NAME_LEN];
4398 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4400 ds_put_format(&ds, "%5s %4d "IP_FMT" %3d\n",
4401 name, grp->vlan, IP_ARGS(grp->ip4),
4402 mcast_bundle_age(ofproto->ms, b));
4406 /* ports connected to multicast routers */
4407 LIST_FOR_EACH(mrouter, mrouter_node, &ofproto->ms->mrouter_lru) {
4408 char name[OFP_MAX_PORT_NAME_LEN];
4410 bundle = mrouter->port;
4411 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4413 ds_put_format(&ds, "%5s %4d querier %3d\n",
4414 name, mrouter->vlan,
4415 mcast_mrouter_age(ofproto->ms, mrouter));
4417 ovs_rwlock_unlock(&ofproto->ms->rwlock);
4418 unixctl_command_reply(conn, ds_cstr(&ds));
4423 struct xlate_out xout;
4424 struct xlate_in xin;
4425 const struct flow *key;
4427 struct flow_wildcards wc;
4432 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
4434 const struct rule_actions *actions;
4437 ds_put_char_multiple(result, '\t', level);
4439 ds_put_cstr(result, "No match\n");
4443 ovs_mutex_lock(&rule->up.mutex);
4444 cookie = rule->up.flow_cookie;
4445 ovs_mutex_unlock(&rule->up.mutex);
4447 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
4448 rule ? rule->up.table_id : 0, ntohll(cookie));
4449 cls_rule_format(&rule->up.cr, result);
4450 ds_put_char(result, '\n');
4452 actions = rule_dpif_get_actions(rule);
4454 ds_put_char_multiple(result, '\t', level);
4455 ds_put_cstr(result, "OpenFlow actions=");
4456 ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
4457 ds_put_char(result, '\n');
4461 trace_format_flow(struct ds *result, int level, const char *title,
4462 struct trace_ctx *trace)
4464 ds_put_char_multiple(result, '\t', level);
4465 ds_put_format(result, "%s: ", title);
4466 /* Do not report unchanged flows for resubmits. */
4467 if ((level > 0 && flow_equal(&trace->xin.flow, &trace->flow))
4468 || (level == 0 && flow_equal(&trace->xin.flow, trace->key))) {
4469 ds_put_cstr(result, "unchanged");
4471 flow_format(result, &trace->xin.flow);
4472 trace->flow = trace->xin.flow;
4474 ds_put_char(result, '\n');
4478 trace_format_regs(struct ds *result, int level, const char *title,
4479 struct trace_ctx *trace)
4483 ds_put_char_multiple(result, '\t', level);
4484 ds_put_format(result, "%s:", title);
4485 for (i = 0; i < FLOW_N_REGS; i++) {
4486 ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
4488 ds_put_char(result, '\n');
4492 trace_format_odp(struct ds *result, int level, const char *title,
4493 struct trace_ctx *trace)
4495 struct ofpbuf *odp_actions = trace->xout.odp_actions;
4497 ds_put_char_multiple(result, '\t', level);
4498 ds_put_format(result, "%s: ", title);
4499 format_odp_actions(result, odp_actions->data, odp_actions->size);
4500 ds_put_char(result, '\n');
4504 trace_format_megaflow(struct ds *result, int level, const char *title,
4505 struct trace_ctx *trace)
4509 ds_put_char_multiple(result, '\t', level);
4510 ds_put_format(result, "%s: ", title);
4511 flow_wildcards_or(&trace->wc, &trace->xout.wc, &trace->wc);
4512 match_init(&match, trace->key, &trace->wc);
4513 match_format(&match, result, OFP_DEFAULT_PRIORITY);
4514 ds_put_char(result, '\n');
4517 static void trace_report(struct xlate_in *xin, const char *s, int recurse);
4520 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
4522 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4523 struct ds *result = trace->result;
4526 if (rule == xin->ofproto->miss_rule) {
4527 trace_report(xin, "No match, flow generates \"packet in\"s.",
4529 } else if (rule == xin->ofproto->no_packet_in_rule) {
4530 trace_report(xin, "No match, packets dropped because "
4531 "OFPPC_NO_PACKET_IN is set on in_port.", recurse);
4532 } else if (rule == xin->ofproto->drop_frags_rule) {
4533 trace_report(xin, "Packets dropped because they are IP "
4534 "fragments and the fragment handling mode is "
4535 "\"drop\".", recurse);
4539 ds_put_char(result, '\n');
4541 trace_format_flow(result, recurse, "Resubmitted flow", trace);
4542 trace_format_regs(result, recurse, "Resubmitted regs", trace);
4543 trace_format_odp(result, recurse, "Resubmitted odp", trace);
4544 trace_format_megaflow(result, recurse, "Resubmitted megaflow", trace);
4546 trace_format_rule(result, recurse, rule);
4550 trace_report(struct xlate_in *xin, const char *s, int recurse)
4552 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4553 struct ds *result = trace->result;
4555 ds_put_char_multiple(result, '\t', recurse);
4556 ds_put_cstr(result, s);
4557 ds_put_char(result, '\n');
4560 /* Parses the 'argc' elements of 'argv', ignoring argv[0]. The following
4561 * forms are supported:
4563 * - [dpname] odp_flow [-generate | packet]
4564 * - bridge br_flow [-generate | packet]
4566 * On success, initializes '*ofprotop' and 'flow' and returns NULL. On failure
4567 * returns a nonnull malloced error message. */
4568 static char * OVS_WARN_UNUSED_RESULT
4569 parse_flow_and_packet(int argc, const char *argv[],
4570 struct ofproto_dpif **ofprotop, struct flow *flow,
4571 struct dp_packet **packetp)
4573 const struct dpif_backer *backer = NULL;
4574 const char *error = NULL;
4576 struct simap port_names = SIMAP_INITIALIZER(&port_names);
4577 struct dp_packet *packet;
4578 struct ofpbuf odp_key;
4579 struct ofpbuf odp_mask;
4581 ofpbuf_init(&odp_key, 0);
4582 ofpbuf_init(&odp_mask, 0);
4584 /* Handle "-generate" or a hex string as the last argument. */
4585 if (!strcmp(argv[argc - 1], "-generate")) {
4586 packet = dp_packet_new(0);
4589 error = eth_from_hex(argv[argc - 1], &packet);
4592 } else if (argc == 4) {
4593 /* The 3-argument form must end in "-generate' or a hex string. */
4599 /* odp_flow can have its in_port specified as a name instead of port no.
4600 * We do not yet know whether a given flow is a odp_flow or a br_flow.
4601 * But, to know whether a flow is odp_flow through odp_flow_from_string(),
4602 * we need to create a simap of name to port no. */
4604 const char *dp_type;
4605 if (!strncmp(argv[1], "ovs-", 4)) {
4606 dp_type = argv[1] + 4;
4610 backer = shash_find_data(&all_dpif_backers, dp_type);
4611 } else if (argc == 2) {
4612 struct shash_node *node;
4613 if (shash_count(&all_dpif_backers) == 1) {
4614 node = shash_first(&all_dpif_backers);
4615 backer = node->data;
4618 error = "Syntax error";
4621 if (backer && backer->dpif) {
4622 struct dpif_port dpif_port;
4623 struct dpif_port_dump port_dump;
4624 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
4625 simap_put(&port_names, dpif_port.name,
4626 odp_to_u32(dpif_port.port_no));
4630 /* Parse the flow and determine whether a datapath or
4631 * bridge is specified. If function odp_flow_key_from_string()
4632 * returns 0, the flow is a odp_flow. If function
4633 * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
4634 if (!odp_flow_from_string(argv[argc - 1], &port_names,
4635 &odp_key, &odp_mask)) {
4637 error = "Cannot find the datapath";
4641 if (odp_flow_key_to_flow(odp_key.data, odp_key.size, flow) == ODP_FIT_ERROR) {
4642 error = "Failed to parse flow key";
4646 *ofprotop = xlate_lookup_ofproto(backer, flow,
4647 &flow->in_port.ofp_port);
4648 if (*ofprotop == NULL) {
4649 error = "Invalid datapath flow";
4653 vsp_adjust_flow(*ofprotop, flow, NULL);
4656 char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
4659 m_err = xasprintf("Bad flow syntax: %s", err);
4664 error = "Must specify bridge name";
4668 *ofprotop = ofproto_dpif_lookup(argv[1]);
4670 error = "Unknown bridge name";
4676 /* Generate a packet, if requested. */
4678 if (!dp_packet_size(packet)) {
4679 flow_compose(packet, flow);
4681 /* Use the metadata from the flow and the packet argument
4682 * to reconstruct the flow. */
4683 pkt_metadata_from_flow(&packet->md, flow);
4684 flow_extract(packet, flow);
4689 if (error && !m_err) {
4690 m_err = xstrdup(error);
4693 dp_packet_delete(packet);
4697 ofpbuf_uninit(&odp_key);
4698 ofpbuf_uninit(&odp_mask);
4699 simap_destroy(&port_names);
4704 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
4705 void *aux OVS_UNUSED)
4707 struct ofproto_dpif *ofproto;
4708 struct dp_packet *packet;
4712 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4717 ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
4718 unixctl_command_reply(conn, ds_cstr(&result));
4719 ds_destroy(&result);
4720 dp_packet_delete(packet);
4722 unixctl_command_reply_error(conn, error);
4728 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
4729 const char *argv[], void *aux OVS_UNUSED)
4731 enum ofputil_protocol usable_protocols;
4732 struct ofproto_dpif *ofproto;
4733 bool enforce_consistency;
4734 struct ofpbuf ofpacts;
4735 struct dp_packet *packet;
4740 /* Three kinds of error return values! */
4746 ofpbuf_init(&ofpacts, 0);
4748 /* Parse actions. */
4749 error = ofpacts_parse_actions(argv[--argc], &ofpacts, &usable_protocols);
4751 unixctl_command_reply_error(conn, error);
4756 /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
4757 * consistency between the flow and the actions. With -consistent, we
4758 * enforce consistency even for a flow supported in OpenFlow 1.0. */
4759 if (!strcmp(argv[1], "-consistent")) {
4760 enforce_consistency = true;
4764 enforce_consistency = false;
4767 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4769 unixctl_command_reply_error(conn, error);
4774 /* Do the same checks as handle_packet_out() in ofproto.c.
4776 * We pass a 'table_id' of 0 to ofpacts_check(), which isn't
4777 * strictly correct because these actions aren't in any table, but it's OK
4778 * because it 'table_id' is used only to check goto_table instructions, but
4779 * packet-outs take a list of actions and therefore it can't include
4782 * We skip the "meter" check here because meter is an instruction, not an
4783 * action, and thus cannot appear in ofpacts. */
4784 in_port = ofp_to_u16(flow.in_port.ofp_port);
4785 if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
4786 unixctl_command_reply_error(conn, "invalid in_port");
4789 if (enforce_consistency) {
4790 retval = ofpacts_check_consistency(ofpacts.data, ofpacts.size,
4791 &flow, u16_to_ofp(ofproto->up.max_ports),
4792 0, 0, usable_protocols);
4794 retval = ofpacts_check(ofpacts.data, ofpacts.size, &flow,
4795 u16_to_ofp(ofproto->up.max_ports), 0, 0,
4801 ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
4802 unixctl_command_reply_error(conn, ds_cstr(&result));
4806 ofproto_trace(ofproto, &flow, packet,
4807 ofpacts.data, ofpacts.size, &result);
4808 unixctl_command_reply(conn, ds_cstr(&result));
4811 ds_destroy(&result);
4812 dp_packet_delete(packet);
4813 ofpbuf_uninit(&ofpacts);
4816 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
4817 * description of the results to 'ds'.
4819 * The trace follows a packet with the specified 'flow' through the flow
4820 * table. 'packet' may be nonnull to trace an actual packet, with consequent
4821 * side effects (if it is nonnull then its flow must be 'flow').
4823 * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
4824 * trace, otherwise the actions are determined by a flow table lookup. */
4826 ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
4827 const struct dp_packet *packet,
4828 const struct ofpact ofpacts[], size_t ofpacts_len,
4831 struct trace_ctx trace;
4833 ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
4834 ds_put_cstr(ds, "Flow: ");
4835 flow_format(ds, flow);
4836 ds_put_char(ds, '\n');
4838 flow_wildcards_init_catchall(&trace.wc);
4841 trace.key = flow; /* Original flow key, used for megaflow. */
4842 trace.flow = *flow; /* May be modified by actions. */
4843 xlate_in_init(&trace.xin, ofproto, flow, flow->in_port.ofp_port, NULL,
4844 ntohs(flow->tcp_flags), packet);
4845 trace.xin.ofpacts = ofpacts;
4846 trace.xin.ofpacts_len = ofpacts_len;
4847 trace.xin.resubmit_hook = trace_resubmit;
4848 trace.xin.report_hook = trace_report;
4850 xlate_actions(&trace.xin, &trace.xout);
4852 ds_put_char(ds, '\n');
4853 trace_format_flow(ds, 0, "Final flow", &trace);
4854 trace_format_megaflow(ds, 0, "Megaflow", &trace);
4856 ds_put_cstr(ds, "Datapath actions: ");
4857 format_odp_actions(ds, trace.xout.odp_actions->data,
4858 trace.xout.odp_actions->size);
4860 if (trace.xout.slow) {
4861 enum slow_path_reason slow;
4863 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
4864 "slow path because it:");
4866 slow = trace.xout.slow;
4868 enum slow_path_reason bit = rightmost_1bit(slow);
4870 ds_put_format(ds, "\n\t- %s.",
4871 slow_path_reason_to_explanation(bit));
4877 xlate_out_uninit(&trace.xout);
4880 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
4881 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
4882 * to destroy 'ofproto_shash' and free the returned value. */
4883 static const struct shash_node **
4884 get_ofprotos(struct shash *ofproto_shash)
4886 const struct ofproto_dpif *ofproto;
4888 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4889 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
4890 shash_add_nocopy(ofproto_shash, name, ofproto);
4893 return shash_sort(ofproto_shash);
4897 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
4898 const char *argv[] OVS_UNUSED,
4899 void *aux OVS_UNUSED)
4901 struct ds ds = DS_EMPTY_INITIALIZER;
4902 struct shash ofproto_shash;
4903 const struct shash_node **sorted_ofprotos;
4906 shash_init(&ofproto_shash);
4907 sorted_ofprotos = get_ofprotos(&ofproto_shash);
4908 for (i = 0; i < shash_count(&ofproto_shash); i++) {
4909 const struct shash_node *node = sorted_ofprotos[i];
4910 ds_put_format(&ds, "%s\n", node->name);
4913 shash_destroy(&ofproto_shash);
4914 free(sorted_ofprotos);
4916 unixctl_command_reply(conn, ds_cstr(&ds));
4921 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
4923 const struct shash_node **ofprotos;
4924 struct dpif_dp_stats dp_stats;
4925 struct shash ofproto_shash;
4928 dpif_get_dp_stats(backer->dpif, &dp_stats);
4930 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
4931 dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
4933 shash_init(&ofproto_shash);
4934 ofprotos = get_ofprotos(&ofproto_shash);
4935 for (i = 0; i < shash_count(&ofproto_shash); i++) {
4936 struct ofproto_dpif *ofproto = ofprotos[i]->data;
4937 const struct shash_node **ports;
4940 if (ofproto->backer != backer) {
4944 ds_put_format(ds, "\t%s:\n", ofproto->up.name);
4946 ports = shash_sort(&ofproto->up.port_by_name);
4947 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
4948 const struct shash_node *node = ports[j];
4949 struct ofport *ofport = node->data;
4951 odp_port_t odp_port;
4953 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
4956 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
4957 if (odp_port != ODPP_NONE) {
4958 ds_put_format(ds, "%"PRIu32":", odp_port);
4960 ds_put_cstr(ds, "none:");
4963 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
4966 if (!netdev_get_config(ofport->netdev, &config)) {
4967 const struct smap_node **nodes;
4970 nodes = smap_sort(&config);
4971 for (i = 0; i < smap_count(&config); i++) {
4972 const struct smap_node *node = nodes[i];
4973 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
4974 node->key, node->value);
4978 smap_destroy(&config);
4980 ds_put_char(ds, ')');
4981 ds_put_char(ds, '\n');
4985 shash_destroy(&ofproto_shash);
4990 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4991 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4993 struct ds ds = DS_EMPTY_INITIALIZER;
4994 const struct shash_node **backers;
4997 backers = shash_sort(&all_dpif_backers);
4998 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
4999 dpif_show_backer(backers[i]->data, &ds);
5003 unixctl_command_reply(conn, ds_cstr(&ds));
5008 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5009 int argc OVS_UNUSED, const char *argv[],
5010 void *aux OVS_UNUSED)
5012 const struct ofproto_dpif *ofproto;
5014 struct ds ds = DS_EMPTY_INITIALIZER;
5015 bool verbosity = false;
5017 struct dpif_port dpif_port;
5018 struct dpif_port_dump port_dump;
5019 struct hmap portno_names;
5021 struct dpif_flow_dump *flow_dump;
5022 struct dpif_flow_dump_thread *flow_dump_thread;
5026 ofproto = ofproto_dpif_lookup(argv[argc - 1]);
5028 unixctl_command_reply_error(conn, "no such bridge");
5032 if (argc > 2 && !strcmp(argv[1], "-m")) {
5036 hmap_init(&portno_names);
5037 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
5038 odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
5042 flow_dump = dpif_flow_dump_create(ofproto->backer->dpif, false);
5043 flow_dump_thread = dpif_flow_dump_thread_create(flow_dump);
5044 while (dpif_flow_dump_next(flow_dump_thread, &f, 1)) {
5047 if (odp_flow_key_to_flow(f.key, f.key_len, &flow) == ODP_FIT_ERROR
5048 || xlate_lookup_ofproto(ofproto->backer, &flow, NULL) != ofproto) {
5053 odp_format_ufid(&f.ufid, &ds);
5054 ds_put_cstr(&ds, " ");
5056 odp_flow_format(f.key, f.key_len, f.mask, f.mask_len,
5057 &portno_names, &ds, verbosity);
5058 ds_put_cstr(&ds, ", ");
5059 dpif_flow_stats_format(&f.stats, &ds);
5060 ds_put_cstr(&ds, ", actions:");
5061 format_odp_actions(&ds, f.actions, f.actions_len);
5062 ds_put_char(&ds, '\n');
5064 dpif_flow_dump_thread_destroy(flow_dump_thread);
5065 error = dpif_flow_dump_destroy(flow_dump);
5069 ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
5070 unixctl_command_reply_error(conn, ds_cstr(&ds));
5072 unixctl_command_reply(conn, ds_cstr(&ds));
5074 odp_portno_names_destroy(&portno_names);
5075 hmap_destroy(&portno_names);
5080 ofproto_revalidate_all_backers(void)
5082 const struct shash_node **backers;
5085 backers = shash_sort(&all_dpif_backers);
5086 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5087 struct dpif_backer *backer = backers[i]->data;
5088 backer->need_revalidate = REV_RECONFIGURE;
5094 disable_tnl_push_pop(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5095 const char *argv[], void *aux OVS_UNUSED)
5097 if (!strcasecmp(argv[1], "off")) {
5098 ofproto_use_tnl_push_pop = false;
5099 unixctl_command_reply(conn, "Tunnel push-pop off");
5100 ofproto_revalidate_all_backers();
5101 } else if (!strcasecmp(argv[1], "on")) {
5102 ofproto_use_tnl_push_pop = true;
5103 unixctl_command_reply(conn, "Tunnel push-pop on");
5104 ofproto_revalidate_all_backers();
5109 ofproto_unixctl_init(void)
5111 static bool registered;
5117 unixctl_command_register(
5119 "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
5120 1, 3, ofproto_unixctl_trace, NULL);
5121 unixctl_command_register(
5122 "ofproto/trace-packet-out",
5123 "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
5124 2, 6, ofproto_unixctl_trace_actions, NULL);
5125 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5126 ofproto_unixctl_fdb_flush, NULL);
5127 unixctl_command_register("fdb/show", "bridge", 1, 1,
5128 ofproto_unixctl_fdb_show, NULL);
5129 unixctl_command_register("mdb/flush", "[bridge]", 0, 1,
5130 ofproto_unixctl_mcast_snooping_flush, NULL);
5131 unixctl_command_register("mdb/show", "bridge", 1, 1,
5132 ofproto_unixctl_mcast_snooping_show, NULL);
5133 unixctl_command_register("dpif/dump-dps", "", 0, 0,
5134 ofproto_unixctl_dpif_dump_dps, NULL);
5135 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5137 unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
5138 ofproto_unixctl_dpif_dump_flows, NULL);
5140 unixctl_command_register("ofproto/tnl-push-pop", "[on]|[off]", 1, 1,
5141 disable_tnl_push_pop, NULL);
5144 /* Returns true if 'table' is the table used for internal rules,
5145 * false otherwise. */
5147 table_is_internal(uint8_t table_id)
5149 return table_id == TBL_INTERNAL;
5152 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5154 * This is deprecated. It is only for compatibility with broken device drivers
5155 * in old versions of Linux that do not properly support VLANs when VLAN
5156 * devices are not used. When broken device drivers are no longer in
5157 * widespread use, we will delete these interfaces. */
5160 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
5162 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5163 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5165 if (realdev_ofp_port == ofport->realdev_ofp_port
5166 && vid == ofport->vlandev_vid) {
5170 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5172 if (ofport->realdev_ofp_port) {
5175 if (realdev_ofp_port && ofport->bundle) {
5176 /* vlandevs are enslaved to their realdevs, so they are not allowed to
5177 * themselves be part of a bundle. */
5178 bundle_set(ofport_->ofproto, ofport->bundle, NULL);
5181 ofport->realdev_ofp_port = realdev_ofp_port;
5182 ofport->vlandev_vid = vid;
5184 if (realdev_ofp_port) {
5185 vsp_add(ofport, realdev_ofp_port, vid);
5192 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
5194 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
5198 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
5199 OVS_EXCLUDED(ofproto->vsp_mutex)
5201 /* hmap_is_empty is thread safe. */
5202 return !hmap_is_empty(&ofproto->realdev_vid_map);
5207 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
5208 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5209 OVS_REQUIRES(ofproto->vsp_mutex)
5211 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5212 int vid = vlan_tci_to_vid(vlan_tci);
5213 const struct vlan_splinter *vsp;
5215 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5216 hash_realdev_vid(realdev_ofp_port, vid),
5217 &ofproto->realdev_vid_map) {
5218 if (vsp->realdev_ofp_port == realdev_ofp_port
5219 && vsp->vid == vid) {
5220 return vsp->vlandev_ofp_port;
5224 return realdev_ofp_port;
5227 /* Returns the OFP port number of the Linux VLAN device that corresponds to
5228 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
5229 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
5230 * 'vlan_tci' 9, it would return the port number of eth0.9.
5232 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
5233 * function just returns its 'realdev_ofp_port' argument. */
5235 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5236 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5237 OVS_EXCLUDED(ofproto->vsp_mutex)
5241 /* hmap_is_empty is thread safe, see if we can return immediately. */
5242 if (hmap_is_empty(&ofproto->realdev_vid_map)) {
5243 return realdev_ofp_port;
5245 ovs_mutex_lock(&ofproto->vsp_mutex);
5246 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
5247 ovs_mutex_unlock(&ofproto->vsp_mutex);
5251 static struct vlan_splinter *
5252 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
5254 struct vlan_splinter *vsp;
5256 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
5257 hash_ofp_port(vlandev_ofp_port),
5258 &ofproto->vlandev_map) {
5259 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5267 /* Returns the OpenFlow port number of the "real" device underlying the Linux
5268 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
5269 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
5270 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
5271 * eth0 and store 9 in '*vid'.
5273 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
5274 * VLAN device. Unless VLAN splinters are enabled, this is what this function
5277 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5278 ofp_port_t vlandev_ofp_port, int *vid)
5279 OVS_REQUIRES(ofproto->vsp_mutex)
5281 if (!hmap_is_empty(&ofproto->vlandev_map)) {
5282 const struct vlan_splinter *vsp;
5284 vsp = vlandev_find(ofproto, vlandev_ofp_port);
5289 return vsp->realdev_ofp_port;
5295 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
5296 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
5297 * 'flow->in_port' to the "real" device backing the VLAN device, sets
5298 * 'flow->vlan_tci' to the VLAN VID, and returns true. Optionally pushes the
5299 * appropriate VLAN on 'packet' if provided. Otherwise (which is always the
5300 * case unless VLAN splinters are enabled), returns false without making any
5303 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow,
5304 struct dp_packet *packet)
5305 OVS_EXCLUDED(ofproto->vsp_mutex)
5310 /* hmap_is_empty is thread safe. */
5311 if (hmap_is_empty(&ofproto->vlandev_map)) {
5315 ovs_mutex_lock(&ofproto->vsp_mutex);
5316 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
5317 ovs_mutex_unlock(&ofproto->vsp_mutex);
5322 /* Cause the flow to be processed as if it came in on the real device with
5323 * the VLAN device's VLAN ID. */
5324 flow->in_port.ofp_port = realdev;
5325 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
5328 /* Make the packet resemble the flow, so that it gets sent to an
5329 * OpenFlow controller properly, so that it looks correct for sFlow,
5330 * and so that flow_extract() will get the correct vlan_tci if it is
5331 * called on 'packet'. */
5332 eth_push_vlan(packet, htons(ETH_TYPE_VLAN), flow->vlan_tci);
5339 vsp_remove(struct ofport_dpif *port)
5341 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5342 struct vlan_splinter *vsp;
5344 ovs_mutex_lock(&ofproto->vsp_mutex);
5345 vsp = vlandev_find(ofproto, port->up.ofp_port);
5347 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
5348 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
5351 port->realdev_ofp_port = 0;
5353 VLOG_ERR("missing vlan device record");
5355 ovs_mutex_unlock(&ofproto->vsp_mutex);
5359 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
5361 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5363 ovs_mutex_lock(&ofproto->vsp_mutex);
5364 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
5365 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
5366 == realdev_ofp_port)) {
5367 struct vlan_splinter *vsp;
5369 vsp = xmalloc(sizeof *vsp);
5370 vsp->realdev_ofp_port = realdev_ofp_port;
5371 vsp->vlandev_ofp_port = port->up.ofp_port;
5374 port->realdev_ofp_port = realdev_ofp_port;
5376 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
5377 hash_ofp_port(port->up.ofp_port));
5378 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
5379 hash_realdev_vid(realdev_ofp_port, vid));
5381 VLOG_ERR("duplicate vlan device record");
5383 ovs_mutex_unlock(&ofproto->vsp_mutex);
5387 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
5389 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
5390 return ofport ? ofport->odp_port : ODPP_NONE;
5393 struct ofport_dpif *
5394 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
5396 struct ofport_dpif *port;
5398 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
5399 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
5400 &backer->odp_to_ofport_map) {
5401 if (port->odp_port == odp_port) {
5402 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5407 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5412 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
5414 struct ofport_dpif *port;
5416 port = odp_port_to_ofport(ofproto->backer, odp_port);
5417 if (port && &ofproto->up == port->up.ofproto) {
5418 return port->up.ofp_port;
5424 struct ofproto_dpif *
5425 ofproto_dpif_recirc_get_ofproto(const struct dpif_backer *backer,
5428 struct dpif_backer_recirc_node *node;
5430 node = CONTAINER_OF(cmap_find(&backer->recirc_map, recirc_id),
5431 struct dpif_backer_recirc_node, cmap_node);
5433 return node ? node->ofproto : NULL;
5437 ofproto_dpif_alloc_recirc_id(struct ofproto_dpif *ofproto)
5439 struct dpif_backer *backer = ofproto->backer;
5440 uint32_t recirc_id = recirc_id_alloc(backer->rid_pool);
5443 struct dpif_backer_recirc_node *node = xmalloc(sizeof *node);
5445 node->recirc_id = recirc_id;
5446 node->ofproto = ofproto;
5448 ovs_mutex_lock(&backer->recirc_mutex);
5449 cmap_insert(&backer->recirc_map, &node->cmap_node, node->recirc_id);
5450 ovs_mutex_unlock(&backer->recirc_mutex);
5457 ofproto_dpif_free_recirc_id(struct ofproto_dpif *ofproto, uint32_t recirc_id)
5459 struct dpif_backer *backer = ofproto->backer;
5460 struct dpif_backer_recirc_node *node;
5462 node = CONTAINER_OF(cmap_find(&backer->recirc_map, recirc_id),
5463 struct dpif_backer_recirc_node, cmap_node);
5465 ovs_mutex_lock(&backer->recirc_mutex);
5466 cmap_remove(&backer->recirc_map, &node->cmap_node, node->recirc_id);
5467 ovs_mutex_unlock(&backer->recirc_mutex);
5468 recirc_id_free(backer->rid_pool, node->recirc_id);
5470 /* 'recirc_id' should never be freed by non-owning 'ofproto'. */
5471 ovs_assert(node->ofproto == ofproto);
5473 /* RCU postpone the free, since other threads may be referring
5474 * to 'node' at same time. */
5475 ovsrcu_postpone(free, node);
5480 ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
5481 const struct match *match, int priority,
5482 uint16_t idle_timeout,
5483 const struct ofpbuf *ofpacts,
5484 struct rule **rulep)
5486 struct ofputil_flow_mod fm;
5487 struct rule_dpif *rule;
5491 fm.priority = priority;
5492 fm.new_cookie = htonll(0);
5493 fm.cookie = htonll(0);
5494 fm.cookie_mask = htonll(0);
5495 fm.modify_cookie = false;
5496 fm.table_id = TBL_INTERNAL;
5497 fm.command = OFPFC_ADD;
5498 fm.idle_timeout = idle_timeout;
5499 fm.hard_timeout = 0;
5503 fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5504 fm.ofpacts = ofpacts->data;
5505 fm.ofpacts_len = ofpacts->size;
5507 error = ofproto_flow_mod(&ofproto->up, &fm);
5509 VLOG_ERR_RL(&rl, "failed to add internal flow (%s)",
5510 ofperr_to_string(error));
5515 rule = rule_dpif_lookup_in_table(ofproto, TBL_INTERNAL, &fm.match.flow,
5516 &fm.match.wc, false);
5526 ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
5527 struct match *match, int priority)
5529 struct ofputil_flow_mod fm;
5533 fm.priority = priority;
5534 fm.new_cookie = htonll(0);
5535 fm.cookie = htonll(0);
5536 fm.cookie_mask = htonll(0);
5537 fm.modify_cookie = false;
5538 fm.table_id = TBL_INTERNAL;
5539 fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5540 fm.command = OFPFC_DELETE_STRICT;
5542 error = ofproto_flow_mod(&ofproto->up, &fm);
5544 VLOG_ERR_RL(&rl, "failed to delete internal flow (%s)",
5545 ofperr_to_string(error));
5552 const struct ofproto_class ofproto_dpif_class = {
5566 NULL, /* get_memory_usage. */
5567 type_get_memory_usage,
5585 port_is_lacp_current,
5586 port_get_lacp_stats,
5587 NULL, /* rule_choose_table */
5596 NULL, /* rule_premodify_actions */
5597 rule_modify_actions,
5613 get_stp_port_status,
5618 get_rstp_port_status,
5625 is_mirror_output_bundle,
5626 forward_bpdu_changed,
5627 set_mac_table_config,
5629 set_mcast_snooping_port,
5631 NULL, /* meter_get_features */
5632 NULL, /* meter_set */
5633 NULL, /* meter_get */
5634 NULL, /* meter_del */
5635 group_alloc, /* group_alloc */
5636 group_construct, /* group_construct */
5637 group_destruct, /* group_destruct */
5638 group_dealloc, /* group_dealloc */
5639 group_modify, /* group_modify */
5640 group_get_stats, /* group_get_stats */
5641 get_datapath_version, /* get_datapath_version */