2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-dpif.h"
20 #include "ofproto/ofproto-provider.h"
27 #include "byte-order.h"
28 #include "connectivity.h"
34 #include "dynamic-string.h"
35 #include "fail-open.h"
36 #include "guarded-list.h"
40 #include "mac-learning.h"
41 #include "mcast-snooping.h"
42 #include "meta-flow.h"
43 #include "multipath.h"
44 #include "netdev-vport.h"
49 #include "odp-execute.h"
52 #include "ofp-actions.h"
53 #include "ofp-parse.h"
54 #include "ofp-print.h"
55 #include "ofproto-dpif-ipfix.h"
56 #include "ofproto-dpif-mirror.h"
57 #include "ofproto-dpif-monitor.h"
58 #include "ofproto-dpif-rid.h"
59 #include "ofproto-dpif-sflow.h"
60 #include "ofproto-dpif-upcall.h"
61 #include "ofproto-dpif-xlate.h"
62 #include "poll-loop.h"
64 #include "ovs-router.h"
70 #include "unaligned.h"
72 #include "vlan-bitmap.h"
73 #include "openvswitch/vlog.h"
75 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
77 COVERAGE_DEFINE(ofproto_dpif_expired);
78 COVERAGE_DEFINE(packet_in_overflow);
87 * - Do include packets and bytes from datapath flows which have not
88 * recently been processed by a revalidator. */
89 struct ovs_mutex stats_mutex;
90 struct dpif_flow_stats stats OVS_GUARDED;
92 /* In non-NULL, will point to a new rule (for which a reference is held) to
93 * which all the stats updates should be forwarded. This exists only
94 * transitionally when flows are replaced.
96 * Protected by stats_mutex. If both 'rule->stats_mutex' and
97 * 'rule->new_rule->stats_mutex' must be held together, acquire them in that
99 struct rule_dpif *new_rule OVS_GUARDED;
101 /* If non-zero then the recirculation id that has
102 * been allocated for use with this rule.
103 * The recirculation id and associated internal flow should
104 * be freed when the rule is freed */
108 /* RULE_CAST() depends on this. */
109 BUILD_ASSERT_DECL(offsetof(struct rule_dpif, up) == 0);
111 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes,
112 long long int *used);
113 static struct rule_dpif *rule_dpif_cast(const struct rule *);
114 static void rule_expire(struct rule_dpif *);
121 * - Do include packets and bytes from datapath flows which have not
122 * recently been processed by a revalidator. */
123 struct ovs_mutex stats_mutex;
124 uint64_t packet_count OVS_GUARDED; /* Number of packets received. */
125 uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */
129 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
130 struct ofproto_dpif *ofproto; /* Owning ofproto. */
131 void *aux; /* Key supplied by ofproto's client. */
132 char *name; /* Identifier for log messages. */
135 struct ovs_list ports; /* Contains "struct ofport"s. */
136 enum port_vlan_mode vlan_mode; /* VLAN mode */
137 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
138 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
139 * NULL if all VLANs are trunked. */
140 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
141 struct bond *bond; /* Nonnull iff more than one port. */
142 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
145 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
148 static void bundle_remove(struct ofport *);
149 static void bundle_update(struct ofbundle *);
150 static void bundle_destroy(struct ofbundle *);
151 static void bundle_del_port(struct ofport_dpif *);
152 static void bundle_run(struct ofbundle *);
153 static void bundle_wait(struct ofbundle *);
154 static void bundle_flush_macs(struct ofbundle *, bool);
155 static void bundle_move(struct ofbundle *, struct ofbundle *);
157 static void stp_run(struct ofproto_dpif *ofproto);
158 static void stp_wait(struct ofproto_dpif *ofproto);
159 static int set_stp_port(struct ofport *,
160 const struct ofproto_port_stp_settings *);
162 static void rstp_run(struct ofproto_dpif *ofproto);
163 static void set_rstp_port(struct ofport *,
164 const struct ofproto_port_rstp_settings *);
167 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
171 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
172 struct ovs_list bundle_node;/* In struct ofbundle's "ports" list. */
173 struct cfm *cfm; /* Connectivity Fault Management, if any. */
174 struct bfd *bfd; /* BFD, if any. */
175 struct lldp *lldp; /* lldp, if any. */
176 bool may_enable; /* May be enabled in bonds. */
177 bool is_tunnel; /* This port is a tunnel. */
178 bool is_layer3; /* This is a layer 3 port. */
179 long long int carrier_seq; /* Carrier status changes. */
180 struct ofport_dpif *peer; /* Peer if patch port. */
183 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
184 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
185 long long int stp_state_entered;
187 /* Rapid Spanning Tree. */
188 struct rstp_port *rstp_port; /* Rapid Spanning Tree Protocol, if any. */
189 enum rstp_state rstp_state; /* Always RSTP_DISABLED if RSTP not in use. */
191 /* Queue to DSCP mapping. */
192 struct ofproto_port_queue *qdscp;
195 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
197 * This is deprecated. It is only for compatibility with broken device
198 * drivers in old versions of Linux that do not properly support VLANs when
199 * VLAN devices are not used. When broken device drivers are no longer in
200 * widespread use, we will delete these interfaces. */
201 ofp_port_t realdev_ofp_port;
205 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
207 * This is deprecated. It is only for compatibility with broken device drivers
208 * in old versions of Linux that do not properly support VLANs when VLAN
209 * devices are not used. When broken device drivers are no longer in
210 * widespread use, we will delete these interfaces. */
211 struct vlan_splinter {
212 struct hmap_node realdev_vid_node;
213 struct hmap_node vlandev_node;
214 ofp_port_t realdev_ofp_port;
215 ofp_port_t vlandev_ofp_port;
219 static void vsp_remove(struct ofport_dpif *);
220 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
222 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
225 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
228 static struct ofport_dpif *
229 ofport_dpif_cast(const struct ofport *ofport)
231 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
234 static void port_run(struct ofport_dpif *);
235 static int set_bfd(struct ofport *, const struct smap *);
236 static int set_cfm(struct ofport *, const struct cfm_settings *);
237 static int set_lldp(struct ofport *ofport_, const struct smap *cfg);
238 static void ofport_update_peer(struct ofport_dpif *);
240 /* Reasons that we might need to revalidate every datapath flow, and
241 * corresponding coverage counters.
243 * A value of 0 means that there is no need to revalidate.
245 * It would be nice to have some cleaner way to integrate with coverage
246 * counters, but with only a few reasons I guess this is good enough for
248 enum revalidate_reason {
249 REV_RECONFIGURE = 1, /* Switch configuration changed. */
250 REV_STP, /* Spanning tree protocol port status change. */
251 REV_RSTP, /* RSTP port status change. */
252 REV_BOND, /* Bonding changed. */
253 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
254 REV_FLOW_TABLE, /* Flow table changed. */
255 REV_MAC_LEARNING, /* Mac learning changed. */
256 REV_MCAST_SNOOPING, /* Multicast snooping changed. */
258 COVERAGE_DEFINE(rev_reconfigure);
259 COVERAGE_DEFINE(rev_stp);
260 COVERAGE_DEFINE(rev_rstp);
261 COVERAGE_DEFINE(rev_bond);
262 COVERAGE_DEFINE(rev_port_toggled);
263 COVERAGE_DEFINE(rev_flow_table);
264 COVERAGE_DEFINE(rev_mac_learning);
265 COVERAGE_DEFINE(rev_mcast_snooping);
267 /* All datapaths of a given type share a single dpif backer instance. */
274 struct ovs_rwlock odp_to_ofport_lock;
275 struct hmap odp_to_ofport_map OVS_GUARDED; /* Contains "struct ofport"s. */
277 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
279 enum revalidate_reason need_revalidate; /* Revalidate all flows. */
281 bool recv_set_enable; /* Enables or disables receiving packets. */
283 /* Version string of the datapath stored in OVSDB. */
284 char *dp_version_string;
286 /* Datapath feature support. */
287 struct dpif_backer_support support;
288 struct atomic_count tnl_count;
291 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
292 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
294 struct ofproto_dpif {
295 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
297 struct dpif_backer *backer;
299 ATOMIC(cls_version_t) tables_version; /* For classifier lookups. */
301 uint64_t dump_seq; /* Last read of udpif_dump_seq(). */
303 /* Special OpenFlow rules. */
304 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
305 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
306 struct rule_dpif *drop_frags_rule; /* Used in OFPUTIL_FRAG_DROP mode. */
309 struct netflow *netflow;
310 struct dpif_sflow *sflow;
311 struct dpif_ipfix *ipfix;
312 struct hmap bundles; /* Contains "struct ofbundle"s. */
313 struct mac_learning *ml;
314 struct mcast_snooping *ms;
315 bool has_bonded_bundles;
317 struct mbridge *mbridge;
319 struct ovs_mutex stats_mutex;
320 struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
321 * consumed in userspace. */
325 long long int stp_last_tick;
327 /* Rapid Spanning Tree. */
329 long long int rstp_last_tick;
331 /* VLAN splinters. */
332 struct ovs_mutex vsp_mutex;
333 struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
334 struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */
337 struct sset ports; /* Set of standard port names. */
338 struct sset ghost_ports; /* Ports with no datapath port. */
339 struct sset port_poll_set; /* Queued names for port_poll() reply. */
340 int port_poll_errno; /* Last errno for port_poll() reply. */
341 uint64_t change_seq; /* Connectivity status changes. */
344 struct guarded_list ams; /* Contains "struct ofproto_async_msgs"s. */
345 struct seq *ams_seq; /* For notifying 'ams' reception. */
349 /* All existing ofproto_dpif instances, indexed by ->up.name. */
350 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
352 static bool ofproto_use_tnl_push_pop = true;
353 static void ofproto_unixctl_init(void);
355 static inline struct ofproto_dpif *
356 ofproto_dpif_cast(const struct ofproto *ofproto)
358 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
359 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
363 ofproto_dpif_get_enable_ufid(const struct dpif_backer *backer)
365 return backer->support.ufid;
368 struct dpif_backer_support *
369 ofproto_dpif_get_support(const struct ofproto_dpif *ofproto)
371 return &ofproto->backer->support;
374 static void ofproto_trace(struct ofproto_dpif *, struct flow *,
375 const struct dp_packet *packet,
376 const struct ofpact[], size_t ofpacts_len,
379 /* Global variables. */
380 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
382 /* Initial mappings of port to bridge mappings. */
383 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
385 /* Executes 'fm'. The caller retains ownership of 'fm' and everything in
388 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
389 const struct ofputil_flow_mod *fm)
391 struct ofproto_flow_mod ofm;
393 /* Multiple threads may do this for the same 'fm' at the same time.
394 * Allocate ofproto_flow_mod with execution context from stack.
396 * Note: This copy could be avoided by making ofproto_flow_mod more
397 * complex, but that may not be desireable, and a learn action is not that
398 * fast to begin with. */
400 ofproto_flow_mod(&ofproto->up, &ofm);
403 /* Appends 'am' to the queue of asynchronous messages to be sent to the
404 * controller. Takes ownership of 'am' and any data it points to. */
406 ofproto_dpif_send_async_msg(struct ofproto_dpif *ofproto,
407 struct ofproto_async_msg *am)
409 if (!guarded_list_push_back(&ofproto->ams, &am->list_node, 1024)) {
410 COVERAGE_INC(packet_in_overflow);
411 ofproto_async_msg_free(am);
414 /* Wakes up main thread for packet-in I/O. */
415 seq_change(ofproto->ams_seq);
418 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
419 * packet rather than to send the packet to the controller.
421 * This function returns false to indicate that a packet_in message
422 * for a "table-miss" should be sent to at least one controller.
423 * False otherwise. */
425 ofproto_dpif_wants_packet_in_on_miss(struct ofproto_dpif *ofproto)
427 return connmgr_wants_packet_in_on_miss(ofproto->up.connmgr);
430 /* Factory functions. */
433 init(const struct shash *iface_hints)
435 struct shash_node *node;
437 /* Make a local copy, since we don't own 'iface_hints' elements. */
438 SHASH_FOR_EACH(node, iface_hints) {
439 const struct iface_hint *orig_hint = node->data;
440 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
442 new_hint->br_name = xstrdup(orig_hint->br_name);
443 new_hint->br_type = xstrdup(orig_hint->br_type);
444 new_hint->ofp_port = orig_hint->ofp_port;
446 shash_add(&init_ofp_ports, node->name, new_hint);
449 ofproto_unixctl_init();
454 enumerate_types(struct sset *types)
456 dp_enumerate_types(types);
460 enumerate_names(const char *type, struct sset *names)
462 struct ofproto_dpif *ofproto;
465 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
466 if (strcmp(type, ofproto->up.type)) {
469 sset_add(names, ofproto->up.name);
476 del(const char *type, const char *name)
481 error = dpif_open(name, type, &dpif);
483 error = dpif_delete(dpif);
490 port_open_type(const char *datapath_type, const char *port_type)
492 return dpif_port_open_type(datapath_type, port_type);
495 /* Type functions. */
497 static void process_dpif_port_changes(struct dpif_backer *);
498 static void process_dpif_all_ports_changed(struct dpif_backer *);
499 static void process_dpif_port_change(struct dpif_backer *,
500 const char *devname);
501 static void process_dpif_port_error(struct dpif_backer *, int error);
503 static struct ofproto_dpif *
504 lookup_ofproto_dpif_by_port_name(const char *name)
506 struct ofproto_dpif *ofproto;
508 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
509 if (sset_contains(&ofproto->ports, name)) {
518 ofproto_dpif_backer_enabled(struct dpif_backer* backer)
520 return backer->recv_set_enable;
524 type_run(const char *type)
526 struct dpif_backer *backer;
528 backer = shash_find_data(&all_dpif_backers, type);
530 /* This is not necessarily a problem, since backers are only
531 * created on demand. */
536 if (dpif_run(backer->dpif)) {
537 backer->need_revalidate = REV_RECONFIGURE;
540 udpif_run(backer->udpif);
542 /* If vswitchd started with other_config:flow_restore_wait set as "true",
543 * and the configuration has now changed to "false", enable receiving
544 * packets from the datapath. */
545 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
548 backer->recv_set_enable = true;
550 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
552 VLOG_ERR("Failed to enable receiving packets in dpif.");
555 dpif_flow_flush(backer->dpif);
556 backer->need_revalidate = REV_RECONFIGURE;
559 if (backer->recv_set_enable) {
560 udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
563 dpif_poll_threads_set(backer->dpif, n_dpdk_rxqs, pmd_cpu_mask);
565 if (backer->need_revalidate) {
566 struct ofproto_dpif *ofproto;
567 struct simap_node *node;
568 struct simap tmp_backers;
570 /* Handle tunnel garbage collection. */
571 simap_init(&tmp_backers);
572 simap_swap(&backer->tnl_backers, &tmp_backers);
574 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
575 struct ofport_dpif *iter;
577 if (backer != ofproto->backer) {
581 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
582 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
585 if (!iter->is_tunnel) {
589 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
590 namebuf, sizeof namebuf);
591 node = simap_find(&tmp_backers, dp_port);
593 simap_put(&backer->tnl_backers, dp_port, node->data);
594 simap_delete(&tmp_backers, node);
595 node = simap_find(&backer->tnl_backers, dp_port);
597 node = simap_find(&backer->tnl_backers, dp_port);
599 odp_port_t odp_port = ODPP_NONE;
601 if (!dpif_port_add(backer->dpif, iter->up.netdev,
603 simap_put(&backer->tnl_backers, dp_port,
604 odp_to_u32(odp_port));
605 node = simap_find(&backer->tnl_backers, dp_port);
610 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
611 if (tnl_port_reconfigure(iter, iter->up.netdev,
613 ovs_native_tunneling_is_on(ofproto), dp_port)) {
614 backer->need_revalidate = REV_RECONFIGURE;
619 SIMAP_FOR_EACH (node, &tmp_backers) {
620 dpif_port_del(backer->dpif, u32_to_odp(node->data));
622 simap_destroy(&tmp_backers);
624 switch (backer->need_revalidate) {
625 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
626 case REV_STP: COVERAGE_INC(rev_stp); break;
627 case REV_RSTP: COVERAGE_INC(rev_rstp); break;
628 case REV_BOND: COVERAGE_INC(rev_bond); break;
629 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
630 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
631 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
632 case REV_MCAST_SNOOPING: COVERAGE_INC(rev_mcast_snooping); break;
634 backer->need_revalidate = 0;
636 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
637 struct ofport_dpif *ofport;
638 struct ofbundle *bundle;
640 if (ofproto->backer != backer) {
645 xlate_ofproto_set(ofproto, ofproto->up.name,
646 ofproto->backer->dpif, ofproto->ml,
647 ofproto->stp, ofproto->rstp, ofproto->ms,
648 ofproto->mbridge, ofproto->sflow, ofproto->ipfix,
650 ofproto->up.forward_bpdu,
651 connmgr_has_in_band(ofproto->up.connmgr),
652 &ofproto->backer->support);
654 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
655 xlate_bundle_set(ofproto, bundle, bundle->name,
656 bundle->vlan_mode, bundle->vlan,
657 bundle->trunks, bundle->use_priority_tags,
658 bundle->bond, bundle->lacp,
662 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
663 int stp_port = ofport->stp_port
664 ? stp_port_no(ofport->stp_port)
666 xlate_ofport_set(ofproto, ofport->bundle, ofport,
667 ofport->up.ofp_port, ofport->odp_port,
668 ofport->up.netdev, ofport->cfm, ofport->bfd,
669 ofport->lldp, ofport->peer, stp_port,
670 ofport->rstp_port, ofport->qdscp,
671 ofport->n_qdscp, ofport->up.pp.config,
672 ofport->up.pp.state, ofport->is_tunnel,
678 udpif_revalidate(backer->udpif);
681 process_dpif_port_changes(backer);
686 /* Check for and handle port changes in 'backer''s dpif. */
688 process_dpif_port_changes(struct dpif_backer *backer)
694 error = dpif_port_poll(backer->dpif, &devname);
700 process_dpif_all_ports_changed(backer);
704 process_dpif_port_change(backer, devname);
709 process_dpif_port_error(backer, error);
716 process_dpif_all_ports_changed(struct dpif_backer *backer)
718 struct ofproto_dpif *ofproto;
719 struct dpif_port dpif_port;
720 struct dpif_port_dump dump;
721 struct sset devnames;
724 sset_init(&devnames);
725 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
726 if (ofproto->backer == backer) {
727 struct ofport *ofport;
729 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
730 sset_add(&devnames, netdev_get_name(ofport->netdev));
734 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
735 sset_add(&devnames, dpif_port.name);
738 SSET_FOR_EACH (devname, &devnames) {
739 process_dpif_port_change(backer, devname);
741 sset_destroy(&devnames);
745 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
747 struct ofproto_dpif *ofproto;
748 struct dpif_port port;
750 /* Don't report on the datapath's device. */
751 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
755 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
756 &all_ofproto_dpifs) {
757 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
762 ofproto = lookup_ofproto_dpif_by_port_name(devname);
763 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
764 /* The port was removed. If we know the datapath,
765 * report it through poll_set(). If we don't, it may be
766 * notifying us of a removal we initiated, so ignore it.
767 * If there's a pending ENOBUFS, let it stand, since
768 * everything will be reevaluated. */
769 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
770 sset_add(&ofproto->port_poll_set, devname);
771 ofproto->port_poll_errno = 0;
773 } else if (!ofproto) {
774 /* The port was added, but we don't know with which
775 * ofproto we should associate it. Delete it. */
776 dpif_port_del(backer->dpif, port.port_no);
778 struct ofport_dpif *ofport;
780 ofport = ofport_dpif_cast(shash_find_data(
781 &ofproto->up.port_by_name, devname));
783 && ofport->odp_port != port.port_no
784 && !odp_port_to_ofport(backer, port.port_no))
786 /* 'ofport''s datapath port number has changed from
787 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
788 * structures to match. */
789 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
790 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
791 ofport->odp_port = port.port_no;
792 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
793 hash_odp_port(port.port_no));
794 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
795 backer->need_revalidate = REV_RECONFIGURE;
798 dpif_port_destroy(&port);
801 /* Propagate 'error' to all ofprotos based on 'backer'. */
803 process_dpif_port_error(struct dpif_backer *backer, int error)
805 struct ofproto_dpif *ofproto;
807 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
808 if (ofproto->backer == backer) {
809 sset_clear(&ofproto->port_poll_set);
810 ofproto->port_poll_errno = error;
816 type_wait(const char *type)
818 struct dpif_backer *backer;
820 backer = shash_find_data(&all_dpif_backers, type);
822 /* This is not necessarily a problem, since backers are only
823 * created on demand. */
827 dpif_wait(backer->dpif);
830 /* Basic life-cycle. */
832 static int add_internal_flows(struct ofproto_dpif *);
834 static struct ofproto *
837 struct ofproto_dpif *ofproto = xzalloc(sizeof *ofproto);
842 dealloc(struct ofproto *ofproto_)
844 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
849 close_dpif_backer(struct dpif_backer *backer)
851 ovs_assert(backer->refcount > 0);
853 if (--backer->refcount) {
857 udpif_destroy(backer->udpif);
859 simap_destroy(&backer->tnl_backers);
860 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
861 hmap_destroy(&backer->odp_to_ofport_map);
862 shash_find_and_delete(&all_dpif_backers, backer->type);
864 free(backer->dp_version_string);
865 dpif_close(backer->dpif);
869 /* Datapath port slated for removal from datapath. */
871 struct ovs_list list_node;
875 static bool check_variable_length_userdata(struct dpif_backer *backer);
876 static void check_support(struct dpif_backer *backer);
879 open_dpif_backer(const char *type, struct dpif_backer **backerp)
881 struct dpif_backer *backer;
882 struct dpif_port_dump port_dump;
883 struct dpif_port port;
884 struct shash_node *node;
885 struct ovs_list garbage_list;
886 struct odp_garbage *garbage;
895 backer = shash_find_data(&all_dpif_backers, type);
902 backer_name = xasprintf("ovs-%s", type);
904 /* Remove any existing datapaths, since we assume we're the only
905 * userspace controlling the datapath. */
907 dp_enumerate_names(type, &names);
908 SSET_FOR_EACH(name, &names) {
909 struct dpif *old_dpif;
911 /* Don't remove our backer if it exists. */
912 if (!strcmp(name, backer_name)) {
916 if (dpif_open(name, type, &old_dpif)) {
917 VLOG_WARN("couldn't open old datapath %s to remove it", name);
919 dpif_delete(old_dpif);
920 dpif_close(old_dpif);
923 sset_destroy(&names);
925 backer = xmalloc(sizeof *backer);
927 error = dpif_create_and_open(backer_name, type, &backer->dpif);
930 VLOG_ERR("failed to open datapath of type %s: %s", type,
931 ovs_strerror(error));
935 backer->udpif = udpif_create(backer, backer->dpif);
937 backer->type = xstrdup(type);
938 backer->refcount = 1;
939 hmap_init(&backer->odp_to_ofport_map);
940 ovs_rwlock_init(&backer->odp_to_ofport_lock);
941 backer->need_revalidate = 0;
942 simap_init(&backer->tnl_backers);
943 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
946 if (backer->recv_set_enable) {
947 dpif_flow_flush(backer->dpif);
950 /* Loop through the ports already on the datapath and remove any
951 * that we don't need anymore. */
952 list_init(&garbage_list);
953 dpif_port_dump_start(&port_dump, backer->dpif);
954 while (dpif_port_dump_next(&port_dump, &port)) {
955 node = shash_find(&init_ofp_ports, port.name);
956 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
957 garbage = xmalloc(sizeof *garbage);
958 garbage->odp_port = port.port_no;
959 list_push_front(&garbage_list, &garbage->list_node);
962 dpif_port_dump_done(&port_dump);
964 LIST_FOR_EACH_POP (garbage, list_node, &garbage_list) {
965 dpif_port_del(backer->dpif, garbage->odp_port);
969 shash_add(&all_dpif_backers, type, backer);
971 check_support(backer);
972 atomic_count_init(&backer->tnl_count, 0);
974 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
976 VLOG_ERR("failed to listen on datapath of type %s: %s",
977 type, ovs_strerror(error));
978 close_dpif_backer(backer);
982 if (backer->recv_set_enable) {
983 udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
986 /* This check fails if performed before udpif threads have been set,
987 * as the kernel module checks that the 'pid' in userspace action
989 backer->support.variable_length_userdata
990 = check_variable_length_userdata(backer);
991 backer->dp_version_string = dpif_get_dp_version(backer->dpif);
997 ovs_native_tunneling_is_on(struct ofproto_dpif *ofproto)
999 return ofproto_use_tnl_push_pop && ofproto->backer->support.tnl_push_pop &&
1000 atomic_count_get(&ofproto->backer->tnl_count);
1003 /* Tests whether 'backer''s datapath supports recirculation. Only newer
1004 * datapaths support OVS_KEY_ATTR_RECIRC_ID in keys. We need to disable some
1005 * features on older datapaths that don't support this feature.
1007 * Returns false if 'backer' definitely does not support recirculation, true if
1008 * it seems to support recirculation or if at least the error we get is
1011 check_recirc(struct dpif_backer *backer)
1014 struct odputil_keybuf keybuf;
1017 struct odp_flow_key_parms odp_parms = {
1024 memset(&flow, 0, sizeof flow);
1028 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1029 odp_flow_key_from_flow(&odp_parms, &key);
1030 enable_recirc = dpif_probe_feature(backer->dpif, "recirculation", &key,
1033 if (enable_recirc) {
1034 VLOG_INFO("%s: Datapath supports recirculation",
1035 dpif_name(backer->dpif));
1037 VLOG_INFO("%s: Datapath does not support recirculation",
1038 dpif_name(backer->dpif));
1041 return enable_recirc;
1044 /* Tests whether 'dpif' supports unique flow ids. We can skip serializing
1045 * some flow attributes for datapaths that support this feature.
1047 * Returns true if 'dpif' supports UFID for flow operations.
1048 * Returns false if 'dpif' does not support UFID. */
1050 check_ufid(struct dpif_backer *backer)
1053 struct odputil_keybuf keybuf;
1057 struct odp_flow_key_parms odp_parms = {
1061 memset(&flow, 0, sizeof flow);
1062 flow.dl_type = htons(0x1234);
1064 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1065 odp_flow_key_from_flow(&odp_parms, &key);
1066 dpif_flow_hash(backer->dpif, key.data, key.size, &ufid);
1068 enable_ufid = dpif_probe_feature(backer->dpif, "UFID", &key, &ufid);
1071 VLOG_INFO("%s: Datapath supports unique flow ids",
1072 dpif_name(backer->dpif));
1074 VLOG_INFO("%s: Datapath does not support unique flow ids",
1075 dpif_name(backer->dpif));
1080 /* Tests whether 'backer''s datapath supports variable-length
1081 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions. We need
1082 * to disable some features on older datapaths that don't support this
1085 * Returns false if 'backer' definitely does not support variable-length
1086 * userdata, true if it seems to support them or if at least the error we get
1089 check_variable_length_userdata(struct dpif_backer *backer)
1091 struct eth_header *eth;
1092 struct ofpbuf actions;
1093 struct dpif_execute execute;
1094 struct dp_packet packet;
1098 /* Compose a userspace action that will cause an ERANGE error on older
1099 * datapaths that don't support variable-length userdata.
1101 * We really test for using userdata longer than 8 bytes, but older
1102 * datapaths accepted these, silently truncating the userdata to 8 bytes.
1103 * The same older datapaths rejected userdata shorter than 8 bytes, so we
1104 * test for that instead as a proxy for longer userdata support. */
1105 ofpbuf_init(&actions, 64);
1106 start = nl_msg_start_nested(&actions, OVS_ACTION_ATTR_USERSPACE);
1107 nl_msg_put_u32(&actions, OVS_USERSPACE_ATTR_PID,
1108 dpif_port_get_pid(backer->dpif, ODPP_NONE, 0));
1109 nl_msg_put_unspec_zero(&actions, OVS_USERSPACE_ATTR_USERDATA, 4);
1110 nl_msg_end_nested(&actions, start);
1112 /* Compose a dummy ethernet packet. */
1113 dp_packet_init(&packet, ETH_HEADER_LEN);
1114 eth = dp_packet_put_zeros(&packet, ETH_HEADER_LEN);
1115 eth->eth_type = htons(0x1234);
1117 /* Execute the actions. On older datapaths this fails with ERANGE, on
1118 * newer datapaths it succeeds. */
1119 execute.actions = actions.data;
1120 execute.actions_len = actions.size;
1121 execute.packet = &packet;
1122 execute.needs_help = false;
1123 execute.probe = true;
1126 error = dpif_execute(backer->dpif, &execute);
1128 dp_packet_uninit(&packet);
1129 ofpbuf_uninit(&actions);
1136 /* Variable-length userdata is not supported. */
1137 VLOG_WARN("%s: datapath does not support variable-length userdata "
1138 "feature (needs Linux 3.10+ or kernel module from OVS "
1139 "1..11+). The NXAST_SAMPLE action will be ignored.",
1140 dpif_name(backer->dpif));
1144 /* Something odd happened. We're not sure whether variable-length
1145 * userdata is supported. Default to "yes". */
1146 VLOG_WARN("%s: variable-length userdata feature probe failed (%s)",
1147 dpif_name(backer->dpif), ovs_strerror(error));
1152 /* Tests the MPLS label stack depth supported by 'backer''s datapath.
1154 * Returns the number of elements in a struct flow's mpls_lse field
1155 * if the datapath supports at least that many entries in an
1157 * Otherwise returns the number of MPLS push actions supported by
1160 check_max_mpls_depth(struct dpif_backer *backer)
1165 for (n = 0; n < FLOW_MAX_MPLS_LABELS; n++) {
1166 struct odputil_keybuf keybuf;
1168 struct odp_flow_key_parms odp_parms = {
1172 memset(&flow, 0, sizeof flow);
1173 flow.dl_type = htons(ETH_TYPE_MPLS);
1174 flow_set_mpls_bos(&flow, n, 1);
1176 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1177 odp_flow_key_from_flow(&odp_parms, &key);
1178 if (!dpif_probe_feature(backer->dpif, "MPLS", &key, NULL)) {
1183 VLOG_INFO("%s: MPLS label stack length probed as %d",
1184 dpif_name(backer->dpif), n);
1188 /* Tests whether 'backer''s datapath supports masked data in
1189 * OVS_ACTION_ATTR_SET actions. We need to disable some features on older
1190 * datapaths that don't support this feature. */
1192 check_masked_set_action(struct dpif_backer *backer)
1194 struct eth_header *eth;
1195 struct ofpbuf actions;
1196 struct dpif_execute execute;
1197 struct dp_packet packet;
1199 struct ovs_key_ethernet key, mask;
1201 /* Compose a set action that will cause an EINVAL error on older
1202 * datapaths that don't support masked set actions.
1203 * Avoid using a full mask, as it could be translated to a non-masked
1204 * set action instead. */
1205 ofpbuf_init(&actions, 64);
1206 memset(&key, 0x53, sizeof key);
1207 memset(&mask, 0x7f, sizeof mask);
1208 commit_masked_set_action(&actions, OVS_KEY_ATTR_ETHERNET, &key, &mask,
1211 /* Compose a dummy ethernet packet. */
1212 dp_packet_init(&packet, ETH_HEADER_LEN);
1213 eth = dp_packet_put_zeros(&packet, ETH_HEADER_LEN);
1214 eth->eth_type = htons(0x1234);
1216 /* Execute the actions. On older datapaths this fails with EINVAL, on
1217 * newer datapaths it succeeds. */
1218 execute.actions = actions.data;
1219 execute.actions_len = actions.size;
1220 execute.packet = &packet;
1221 execute.needs_help = false;
1222 execute.probe = true;
1225 error = dpif_execute(backer->dpif, &execute);
1227 dp_packet_uninit(&packet);
1228 ofpbuf_uninit(&actions);
1231 /* Masked set action is not supported. */
1232 VLOG_INFO("%s: datapath does not support masked set action feature.",
1233 dpif_name(backer->dpif));
1238 #define CHECK_FEATURE__(NAME, SUPPORT, FIELD, VALUE) \
1240 check_##NAME(struct dpif_backer *backer) \
1243 struct odputil_keybuf keybuf; \
1244 struct ofpbuf key; \
1246 struct odp_flow_key_parms odp_parms = { \
1253 memset(&flow, 0, sizeof flow); \
1254 flow.FIELD = VALUE; \
1256 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); \
1257 odp_flow_key_from_flow(&odp_parms, &key); \
1258 enable = dpif_probe_feature(backer->dpif, #NAME, &key, NULL); \
1261 VLOG_INFO("%s: Datapath supports "#NAME, dpif_name(backer->dpif)); \
1263 VLOG_INFO("%s: Datapath does not support "#NAME, \
1264 dpif_name(backer->dpif)); \
1269 #define CHECK_FEATURE(FIELD) CHECK_FEATURE__(FIELD, FIELD, FIELD, 1)
1271 CHECK_FEATURE(ct_state)
1272 CHECK_FEATURE(ct_zone)
1273 CHECK_FEATURE(ct_mark)
1274 CHECK_FEATURE__(ct_label, ct_label, ct_label.u64.lo, 1)
1275 CHECK_FEATURE__(ct_state_nat, ct_state, ct_state, CS_TRACKED|CS_SRC_NAT)
1277 #undef CHECK_FEATURE
1278 #undef CHECK_FEATURE__
1281 check_support(struct dpif_backer *backer)
1283 /* This feature needs to be tested after udpif threads are set. */
1284 backer->support.variable_length_userdata = false;
1286 backer->support.odp.recirc = check_recirc(backer);
1287 backer->support.odp.max_mpls_depth = check_max_mpls_depth(backer);
1288 backer->support.masked_set_action = check_masked_set_action(backer);
1289 backer->support.ufid = check_ufid(backer);
1290 backer->support.tnl_push_pop = dpif_supports_tnl_push_pop(backer->dpif);
1292 backer->support.odp.ct_state = check_ct_state(backer);
1293 backer->support.odp.ct_zone = check_ct_zone(backer);
1294 backer->support.odp.ct_mark = check_ct_mark(backer);
1295 backer->support.odp.ct_label = check_ct_label(backer);
1297 backer->support.odp.ct_state_nat = check_ct_state_nat(backer);
1301 construct(struct ofproto *ofproto_)
1303 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1304 struct shash_node *node, *next;
1307 /* Tunnel module can get used right after the udpif threads are running. */
1308 ofproto_tunnel_init();
1310 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1315 atomic_init(&ofproto->tables_version, CLS_MIN_VERSION);
1316 ofproto->netflow = NULL;
1317 ofproto->sflow = NULL;
1318 ofproto->ipfix = NULL;
1319 ofproto->stp = NULL;
1320 ofproto->rstp = NULL;
1321 ofproto->dump_seq = 0;
1322 hmap_init(&ofproto->bundles);
1323 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1325 ofproto->mbridge = mbridge_create();
1326 ofproto->has_bonded_bundles = false;
1327 ofproto->lacp_enabled = false;
1328 ovs_mutex_init_adaptive(&ofproto->stats_mutex);
1329 ovs_mutex_init(&ofproto->vsp_mutex);
1331 guarded_list_init(&ofproto->ams);
1333 hmap_init(&ofproto->vlandev_map);
1334 hmap_init(&ofproto->realdev_vid_map);
1336 sset_init(&ofproto->ports);
1337 sset_init(&ofproto->ghost_ports);
1338 sset_init(&ofproto->port_poll_set);
1339 ofproto->port_poll_errno = 0;
1340 ofproto->change_seq = 0;
1341 ofproto->ams_seq = seq_create();
1342 ofproto->ams_seqno = seq_read(ofproto->ams_seq);
1345 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1346 struct iface_hint *iface_hint = node->data;
1348 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1349 /* Check if the datapath already has this port. */
1350 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1351 sset_add(&ofproto->ports, node->name);
1354 free(iface_hint->br_name);
1355 free(iface_hint->br_type);
1357 shash_delete(&init_ofp_ports, node);
1361 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1362 hash_string(ofproto->up.name, 0));
1363 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1365 ofproto_init_tables(ofproto_, N_TABLES);
1366 error = add_internal_flows(ofproto);
1368 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1374 add_internal_miss_flow(struct ofproto_dpif *ofproto, int id,
1375 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1381 match_init_catchall(&match);
1382 match_set_reg(&match, 0, id);
1384 error = ofproto_dpif_add_internal_flow(ofproto, &match, 0, 0, ofpacts,
1386 *rulep = error ? NULL : rule_dpif_cast(rule);
1392 add_internal_flows(struct ofproto_dpif *ofproto)
1394 struct ofpact_controller *controller;
1395 uint64_t ofpacts_stub[128 / 8];
1396 struct ofpbuf ofpacts;
1397 struct rule *unused_rulep OVS_UNUSED;
1402 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1405 controller = ofpact_put_CONTROLLER(&ofpacts);
1406 controller->max_len = UINT16_MAX;
1407 controller->controller_id = 0;
1408 controller->reason = OFPR_IMPLICIT_MISS;
1410 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1411 &ofproto->miss_rule);
1416 ofpbuf_clear(&ofpacts);
1417 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1418 &ofproto->no_packet_in_rule);
1423 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1424 &ofproto->drop_frags_rule);
1429 /* Drop any run away non-recirc rule lookups. Recirc_id has to be
1430 * zero when reaching this rule.
1432 * (priority=2), recirc_id=0, actions=drop
1434 ofpbuf_clear(&ofpacts);
1435 match_init_catchall(&match);
1436 match_set_recirc_id(&match, 0);
1437 error = ofproto_dpif_add_internal_flow(ofproto, &match, 2, 0, &ofpacts,
1443 destruct(struct ofproto *ofproto_)
1445 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1446 struct ofproto_async_msg *am;
1447 struct rule_dpif *rule;
1448 struct oftable *table;
1449 struct ovs_list ams;
1451 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1453 xlate_remove_ofproto(ofproto);
1456 /* Ensure that the upcall processing threads have no remaining references
1457 * to the ofproto or anything in it. */
1458 udpif_synchronize(ofproto->backer->udpif);
1460 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1462 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1463 CLS_FOR_EACH (rule, up.cr, &table->cls) {
1464 ofproto_rule_delete(&ofproto->up, &rule->up);
1467 ofproto_group_delete_all(&ofproto->up);
1469 guarded_list_pop_all(&ofproto->ams, &ams);
1470 LIST_FOR_EACH_POP (am, list_node, &ams) {
1471 ofproto_async_msg_free(am);
1473 guarded_list_destroy(&ofproto->ams);
1475 recirc_free_ofproto(ofproto, ofproto->up.name);
1477 mbridge_unref(ofproto->mbridge);
1479 netflow_unref(ofproto->netflow);
1480 dpif_sflow_unref(ofproto->sflow);
1481 dpif_ipfix_unref(ofproto->ipfix);
1482 hmap_destroy(&ofproto->bundles);
1483 mac_learning_unref(ofproto->ml);
1484 mcast_snooping_unref(ofproto->ms);
1486 hmap_destroy(&ofproto->vlandev_map);
1487 hmap_destroy(&ofproto->realdev_vid_map);
1489 sset_destroy(&ofproto->ports);
1490 sset_destroy(&ofproto->ghost_ports);
1491 sset_destroy(&ofproto->port_poll_set);
1493 ovs_mutex_destroy(&ofproto->stats_mutex);
1494 ovs_mutex_destroy(&ofproto->vsp_mutex);
1496 seq_destroy(ofproto->ams_seq);
1498 close_dpif_backer(ofproto->backer);
1502 run(struct ofproto *ofproto_)
1504 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1505 uint64_t new_seq, new_dump_seq;
1507 if (mbridge_need_revalidate(ofproto->mbridge)) {
1508 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1509 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1510 mac_learning_flush(ofproto->ml);
1511 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1512 mcast_snooping_mdb_flush(ofproto->ms);
1515 /* Always updates the ofproto->ams_seqno to avoid frequent wakeup during
1516 * flow restore. Even though nothing is processed during flow restore,
1517 * all queued 'ams' will be handled immediately when flow restore
1519 ofproto->ams_seqno = seq_read(ofproto->ams_seq);
1521 /* Do not perform any periodic activity required by 'ofproto' while
1522 * waiting for flow restore to complete. */
1523 if (!ofproto_get_flow_restore_wait()) {
1524 struct ofproto_async_msg *am;
1525 struct ovs_list ams;
1527 guarded_list_pop_all(&ofproto->ams, &ams);
1528 LIST_FOR_EACH_POP (am, list_node, &ams) {
1529 connmgr_send_async_msg(ofproto->up.connmgr, am);
1530 ofproto_async_msg_free(am);
1534 if (ofproto->netflow) {
1535 netflow_run(ofproto->netflow);
1537 if (ofproto->sflow) {
1538 dpif_sflow_run(ofproto->sflow);
1540 if (ofproto->ipfix) {
1541 dpif_ipfix_run(ofproto->ipfix);
1544 new_seq = seq_read(connectivity_seq_get());
1545 if (ofproto->change_seq != new_seq) {
1546 struct ofport_dpif *ofport;
1548 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1552 ofproto->change_seq = new_seq;
1554 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1555 struct ofbundle *bundle;
1557 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1564 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1565 if (mac_learning_run(ofproto->ml)) {
1566 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1568 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1570 if (mcast_snooping_run(ofproto->ms)) {
1571 ofproto->backer->need_revalidate = REV_MCAST_SNOOPING;
1574 new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
1575 if (ofproto->dump_seq != new_dump_seq) {
1576 struct rule *rule, *next_rule;
1578 /* We know stats are relatively fresh, so now is a good time to do some
1580 ofproto->dump_seq = new_dump_seq;
1582 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1584 ovs_mutex_lock(&ofproto_mutex);
1585 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
1586 &ofproto->up.expirable) {
1587 rule_expire(rule_dpif_cast(rule));
1589 ovs_mutex_unlock(&ofproto_mutex);
1591 /* All outstanding data in existing flows has been accounted, so it's a
1592 * good time to do bond rebalancing. */
1593 if (ofproto->has_bonded_bundles) {
1594 struct ofbundle *bundle;
1596 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1598 bond_rebalance(bundle->bond);
1607 wait(struct ofproto *ofproto_)
1609 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1611 if (ofproto_get_flow_restore_wait()) {
1615 if (ofproto->sflow) {
1616 dpif_sflow_wait(ofproto->sflow);
1618 if (ofproto->ipfix) {
1619 dpif_ipfix_wait(ofproto->ipfix);
1621 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1622 struct ofbundle *bundle;
1624 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1625 bundle_wait(bundle);
1628 if (ofproto->netflow) {
1629 netflow_wait(ofproto->netflow);
1631 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1632 mac_learning_wait(ofproto->ml);
1633 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1634 mcast_snooping_wait(ofproto->ms);
1636 if (ofproto->backer->need_revalidate) {
1637 poll_immediate_wake();
1640 seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
1641 seq_wait(ofproto->ams_seq, ofproto->ams_seqno);
1645 type_get_memory_usage(const char *type, struct simap *usage)
1647 struct dpif_backer *backer;
1649 backer = shash_find_data(&all_dpif_backers, type);
1651 udpif_get_memory_usage(backer->udpif, usage);
1656 flush(struct ofproto *ofproto_)
1658 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1659 struct dpif_backer *backer = ofproto->backer;
1662 udpif_flush(backer->udpif);
1667 query_tables(struct ofproto *ofproto,
1668 struct ofputil_table_features *features,
1669 struct ofputil_table_stats *stats)
1671 strcpy(features->name, "classifier");
1676 for (i = 0; i < ofproto->n_tables; i++) {
1677 unsigned long missed, matched;
1679 atomic_read_relaxed(&ofproto->tables[i].n_matched, &matched);
1680 atomic_read_relaxed(&ofproto->tables[i].n_missed, &missed);
1682 stats[i].matched_count = matched;
1683 stats[i].lookup_count = matched + missed;
1689 set_tables_version(struct ofproto *ofproto_, cls_version_t version)
1691 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1693 atomic_store_relaxed(&ofproto->tables_version, version);
1697 static struct ofport *
1700 struct ofport_dpif *port = xzalloc(sizeof *port);
1705 port_dealloc(struct ofport *port_)
1707 struct ofport_dpif *port = ofport_dpif_cast(port_);
1712 port_construct(struct ofport *port_)
1714 struct ofport_dpif *port = ofport_dpif_cast(port_);
1715 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1716 const struct netdev *netdev = port->up.netdev;
1717 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1718 const char *dp_port_name;
1719 struct dpif_port dpif_port;
1722 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1723 port->bundle = NULL;
1727 port->may_enable = false;
1728 port->stp_port = NULL;
1729 port->stp_state = STP_DISABLED;
1730 port->rstp_port = NULL;
1731 port->rstp_state = RSTP_DISABLED;
1732 port->is_tunnel = false;
1736 port->realdev_ofp_port = 0;
1737 port->vlandev_vid = 0;
1738 port->carrier_seq = netdev_get_carrier_resets(netdev);
1739 port->is_layer3 = netdev_vport_is_layer3(netdev);
1741 if (netdev_vport_is_patch(netdev)) {
1742 /* By bailing out here, we don't submit the port to the sFlow module
1743 * to be considered for counter polling export. This is correct
1744 * because the patch port represents an interface that sFlow considers
1745 * to be "internal" to the switch as a whole, and therefore not a
1746 * candidate for counter polling. */
1747 port->odp_port = ODPP_NONE;
1748 ofport_update_peer(port);
1752 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
1753 error = dpif_port_query_by_name(ofproto->backer->dpif, dp_port_name,
1759 port->odp_port = dpif_port.port_no;
1761 if (netdev_get_tunnel_config(netdev)) {
1762 atomic_count_inc(&ofproto->backer->tnl_count);
1763 error = tnl_port_add(port, port->up.netdev, port->odp_port,
1764 ovs_native_tunneling_is_on(ofproto), dp_port_name);
1766 atomic_count_dec(&ofproto->backer->tnl_count);
1767 dpif_port_destroy(&dpif_port);
1771 port->is_tunnel = true;
1772 if (ofproto->ipfix) {
1773 dpif_ipfix_add_tunnel_port(ofproto->ipfix, port_, port->odp_port);
1776 /* Sanity-check that a mapping doesn't already exist. This
1777 * shouldn't happen for non-tunnel ports. */
1778 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1779 VLOG_ERR("port %s already has an OpenFlow port number",
1781 dpif_port_destroy(&dpif_port);
1785 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1786 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1787 hash_odp_port(port->odp_port));
1788 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1790 dpif_port_destroy(&dpif_port);
1792 if (ofproto->sflow) {
1793 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1800 port_destruct(struct ofport *port_)
1802 struct ofport_dpif *port = ofport_dpif_cast(port_);
1803 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1804 const char *devname = netdev_get_name(port->up.netdev);
1805 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1806 const char *dp_port_name;
1808 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1810 xlate_ofport_remove(port);
1813 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1815 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1816 /* The underlying device is still there, so delete it. This
1817 * happens when the ofproto is being destroyed, since the caller
1818 * assumes that removal of attached ports will happen as part of
1820 if (!port->is_tunnel) {
1821 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1826 port->peer->peer = NULL;
1830 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1831 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1832 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1833 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1836 if (port->is_tunnel) {
1837 atomic_count_dec(&ofproto->backer->tnl_count);
1840 if (port->is_tunnel && ofproto->ipfix) {
1841 dpif_ipfix_del_tunnel_port(ofproto->ipfix, port->odp_port);
1845 sset_find_and_delete(&ofproto->ports, devname);
1846 sset_find_and_delete(&ofproto->ghost_ports, devname);
1847 bundle_remove(port_);
1848 set_cfm(port_, NULL);
1849 set_bfd(port_, NULL);
1850 set_lldp(port_, NULL);
1851 if (port->stp_port) {
1852 stp_port_disable(port->stp_port);
1854 set_rstp_port(port_, NULL);
1855 if (ofproto->sflow) {
1856 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1863 port_modified(struct ofport *port_)
1865 struct ofport_dpif *port = ofport_dpif_cast(port_);
1866 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1867 const char *dp_port_name;
1868 struct netdev *netdev = port->up.netdev;
1870 if (port->bundle && port->bundle->bond) {
1871 bond_slave_set_netdev(port->bundle->bond, port, netdev);
1875 cfm_set_netdev(port->cfm, netdev);
1879 bfd_set_netdev(port->bfd, netdev);
1882 ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1883 port->lldp, &port->up.pp.hw_addr);
1885 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
1887 if (port->is_tunnel) {
1888 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1890 if (tnl_port_reconfigure(port, netdev, port->odp_port,
1891 ovs_native_tunneling_is_on(ofproto),
1893 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1897 ofport_update_peer(port);
1901 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1903 struct ofport_dpif *port = ofport_dpif_cast(port_);
1904 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1905 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1907 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1908 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1909 OFPUTIL_PC_NO_PACKET_IN)) {
1910 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1912 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1913 bundle_update(port->bundle);
1919 set_sflow(struct ofproto *ofproto_,
1920 const struct ofproto_sflow_options *sflow_options)
1922 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1923 struct dpif_sflow *ds = ofproto->sflow;
1925 if (sflow_options) {
1926 uint32_t old_probability = ds ? dpif_sflow_get_probability(ds) : 0;
1928 struct ofport_dpif *ofport;
1930 ds = ofproto->sflow = dpif_sflow_create();
1931 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1932 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1935 dpif_sflow_set_options(ds, sflow_options);
1936 if (dpif_sflow_get_probability(ds) != old_probability) {
1937 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1941 dpif_sflow_unref(ds);
1942 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1943 ofproto->sflow = NULL;
1951 struct ofproto *ofproto_,
1952 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1953 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1954 size_t n_flow_exporters_options)
1956 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1957 struct dpif_ipfix *di = ofproto->ipfix;
1958 bool has_options = bridge_exporter_options || flow_exporters_options;
1959 bool new_di = false;
1961 if (has_options && !di) {
1962 di = ofproto->ipfix = dpif_ipfix_create();
1967 /* Call set_options in any case to cleanly flush the flow
1968 * caches in the last exporters that are to be destroyed. */
1969 dpif_ipfix_set_options(
1970 di, bridge_exporter_options, flow_exporters_options,
1971 n_flow_exporters_options);
1973 /* Add tunnel ports only when a new ipfix created */
1974 if (new_di == true) {
1975 struct ofport_dpif *ofport;
1976 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1977 if (ofport->is_tunnel == true) {
1978 dpif_ipfix_add_tunnel_port(di, &ofport->up, ofport->odp_port);
1984 dpif_ipfix_unref(di);
1985 ofproto->ipfix = NULL;
1993 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1995 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1996 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1997 struct cfm *old = ofport->cfm;
2002 ofport->cfm = cfm_create(ofport->up.netdev);
2005 if (cfm_configure(ofport->cfm, s)) {
2012 cfm_unref(ofport->cfm);
2015 if (ofport->cfm != old) {
2016 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2018 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
2019 ofport->lldp, &ofport->up.pp.hw_addr);
2024 cfm_status_changed(struct ofport *ofport_)
2026 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2028 return ofport->cfm ? cfm_check_status_change(ofport->cfm) : true;
2032 get_cfm_status(const struct ofport *ofport_,
2033 struct cfm_status *status)
2035 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2039 cfm_get_status(ofport->cfm, status);
2048 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2050 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2051 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2055 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
2056 cfg, ofport->up.netdev);
2057 if (ofport->bfd != old) {
2058 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2060 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
2061 ofport->lldp, &ofport->up.pp.hw_addr);
2066 bfd_status_changed(struct ofport *ofport_)
2068 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2070 return ofport->bfd ? bfd_check_status_change(ofport->bfd) : true;
2074 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2076 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2080 bfd_get_status(ofport->bfd, smap);
2089 set_lldp(struct ofport *ofport_,
2090 const struct smap *cfg)
2092 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2096 if (!ofport->lldp) {
2097 struct ofproto_dpif *ofproto;
2099 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2100 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2101 ofport->lldp = lldp_create(ofport->up.netdev, ofport_->mtu, cfg);
2104 if (!lldp_configure(ofport->lldp, cfg)) {
2109 lldp_unref(ofport->lldp);
2110 ofport->lldp = NULL;
2113 ofproto_dpif_monitor_port_update(ofport,
2117 &ofport->up.pp.hw_addr);
2122 get_lldp_status(const struct ofport *ofport_,
2123 struct lldp_status *status OVS_UNUSED)
2125 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2127 return ofport->lldp ? true : false;
2131 set_aa(struct ofproto *ofproto OVS_UNUSED,
2132 const struct aa_settings *s)
2134 return aa_configure(s);
2138 aa_mapping_set(struct ofproto *ofproto_ OVS_UNUSED, void *aux,
2139 const struct aa_mapping_settings *s)
2141 return aa_mapping_register(aux, s);
2145 aa_mapping_unset(struct ofproto *ofproto OVS_UNUSED, void *aux)
2147 return aa_mapping_unregister(aux);
2151 aa_vlan_get_queued(struct ofproto *ofproto OVS_UNUSED, struct ovs_list *list)
2153 return aa_get_vlan_queued(list);
2157 aa_vlan_get_queue_size(struct ofproto *ofproto OVS_UNUSED)
2159 return aa_get_vlan_queue_size();
2163 /* Spanning Tree. */
2165 /* Called while rstp_mutex is held. */
2167 rstp_send_bpdu_cb(struct dp_packet *pkt, void *ofport_, void *ofproto_)
2169 struct ofproto_dpif *ofproto = ofproto_;
2170 struct ofport_dpif *ofport = ofport_;
2171 struct eth_header *eth = dp_packet_l2(pkt);
2173 netdev_get_etheraddr(ofport->up.netdev, ð->eth_src);
2174 if (eth_addr_is_zero(eth->eth_src)) {
2175 VLOG_WARN_RL(&rl, "%s port %d: cannot send RSTP BPDU on a port which "
2176 "does not have a configured source MAC address.",
2177 ofproto->up.name, ofp_to_u16(ofport->up.ofp_port));
2179 ofproto_dpif_send_packet(ofport, pkt);
2181 dp_packet_delete(pkt);
2185 send_bpdu_cb(struct dp_packet *pkt, int port_num, void *ofproto_)
2187 struct ofproto_dpif *ofproto = ofproto_;
2188 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2189 struct ofport_dpif *ofport;
2191 ofport = stp_port_get_aux(sp);
2193 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2194 ofproto->up.name, port_num);
2196 struct eth_header *eth = dp_packet_l2(pkt);
2198 netdev_get_etheraddr(ofport->up.netdev, ð->eth_src);
2199 if (eth_addr_is_zero(eth->eth_src)) {
2200 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2201 "with unknown MAC", ofproto->up.name, port_num);
2203 ofproto_dpif_send_packet(ofport, pkt);
2206 dp_packet_delete(pkt);
2209 /* Configure RSTP on 'ofproto_' using the settings defined in 's'. */
2211 set_rstp(struct ofproto *ofproto_, const struct ofproto_rstp_settings *s)
2213 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2215 /* Only revalidate flows if the configuration changed. */
2216 if (!s != !ofproto->rstp) {
2217 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2221 if (!ofproto->rstp) {
2222 ofproto->rstp = rstp_create(ofproto_->name, s->address,
2223 rstp_send_bpdu_cb, ofproto);
2224 ofproto->rstp_last_tick = time_msec();
2226 rstp_set_bridge_address(ofproto->rstp, s->address);
2227 rstp_set_bridge_priority(ofproto->rstp, s->priority);
2228 rstp_set_bridge_ageing_time(ofproto->rstp, s->ageing_time);
2229 rstp_set_bridge_force_protocol_version(ofproto->rstp,
2230 s->force_protocol_version);
2231 rstp_set_bridge_max_age(ofproto->rstp, s->bridge_max_age);
2232 rstp_set_bridge_forward_delay(ofproto->rstp, s->bridge_forward_delay);
2233 rstp_set_bridge_transmit_hold_count(ofproto->rstp,
2234 s->transmit_hold_count);
2236 struct ofport *ofport;
2237 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2238 set_rstp_port(ofport, NULL);
2240 rstp_unref(ofproto->rstp);
2241 ofproto->rstp = NULL;
2246 get_rstp_status(struct ofproto *ofproto_, struct ofproto_rstp_status *s)
2248 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2250 if (ofproto->rstp) {
2252 s->root_id = rstp_get_root_id(ofproto->rstp);
2253 s->bridge_id = rstp_get_bridge_id(ofproto->rstp);
2254 s->designated_id = rstp_get_designated_id(ofproto->rstp);
2255 s->root_path_cost = rstp_get_root_path_cost(ofproto->rstp);
2256 s->designated_port_id = rstp_get_designated_port_id(ofproto->rstp);
2257 s->bridge_port_id = rstp_get_bridge_port_id(ofproto->rstp);
2264 update_rstp_port_state(struct ofport_dpif *ofport)
2266 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2267 enum rstp_state state;
2269 /* Figure out new state. */
2270 state = ofport->rstp_port ? rstp_port_get_state(ofport->rstp_port)
2274 if (ofport->rstp_state != state) {
2275 enum ofputil_port_state of_state;
2278 VLOG_DBG("port %s: RSTP state changed from %s to %s",
2279 netdev_get_name(ofport->up.netdev),
2280 rstp_state_name(ofport->rstp_state),
2281 rstp_state_name(state));
2283 if (rstp_learn_in_state(ofport->rstp_state)
2284 != rstp_learn_in_state(state)) {
2285 /* XXX: Learning action flows should also be flushed. */
2286 if (ofport->bundle) {
2287 if (!rstp_shift_root_learned_address(ofproto->rstp)
2288 || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2289 bundle_flush_macs(ofport->bundle, false);
2293 fwd_change = rstp_forward_in_state(ofport->rstp_state)
2294 != rstp_forward_in_state(state);
2296 ofproto->backer->need_revalidate = REV_RSTP;
2297 ofport->rstp_state = state;
2299 if (fwd_change && ofport->bundle) {
2300 bundle_update(ofport->bundle);
2303 /* Update the RSTP state bits in the OpenFlow port description. */
2304 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2305 of_state |= (state == RSTP_LEARNING ? OFPUTIL_PS_STP_LEARN
2306 : state == RSTP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2307 : state == RSTP_DISCARDING ? OFPUTIL_PS_STP_LISTEN
2309 ofproto_port_set_state(&ofport->up, of_state);
2314 rstp_run(struct ofproto_dpif *ofproto)
2316 if (ofproto->rstp) {
2317 long long int now = time_msec();
2318 long long int elapsed = now - ofproto->rstp_last_tick;
2319 struct rstp_port *rp;
2320 struct ofport_dpif *ofport;
2322 /* Every second, decrease the values of the timers. */
2323 if (elapsed >= 1000) {
2324 rstp_tick_timers(ofproto->rstp);
2325 ofproto->rstp_last_tick = now;
2328 while ((ofport = rstp_get_next_changed_port_aux(ofproto->rstp, &rp))) {
2329 update_rstp_port_state(ofport);
2333 /* FIXME: This check should be done on-event (i.e., when setting
2334 * p->fdb_flush) and not periodically.
2336 while ((ofport = rstp_check_and_reset_fdb_flush(ofproto->rstp, &rp))) {
2337 if (!rstp_shift_root_learned_address(ofproto->rstp)
2338 || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2339 bundle_flush_macs(ofport->bundle, false);
2343 if (rstp_shift_root_learned_address(ofproto->rstp)) {
2344 struct ofport_dpif *old_root_aux =
2345 (struct ofport_dpif *)rstp_get_old_root_aux(ofproto->rstp);
2346 struct ofport_dpif *new_root_aux =
2347 (struct ofport_dpif *)rstp_get_new_root_aux(ofproto->rstp);
2348 if (old_root_aux != NULL && new_root_aux != NULL) {
2349 bundle_move(old_root_aux->bundle, new_root_aux->bundle);
2350 rstp_reset_root_changed(ofproto->rstp);
2356 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2358 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2360 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2362 /* Only revalidate flows if the configuration changed. */
2363 if (!s != !ofproto->stp) {
2364 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2368 if (!ofproto->stp) {
2369 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2370 send_bpdu_cb, ofproto);
2371 ofproto->stp_last_tick = time_msec();
2374 stp_set_bridge_id(ofproto->stp, s->system_id);
2375 stp_set_bridge_priority(ofproto->stp, s->priority);
2376 stp_set_hello_time(ofproto->stp, s->hello_time);
2377 stp_set_max_age(ofproto->stp, s->max_age);
2378 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2380 struct ofport *ofport;
2382 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2383 set_stp_port(ofport, NULL);
2386 stp_unref(ofproto->stp);
2387 ofproto->stp = NULL;
2394 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2396 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2400 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2401 s->designated_root = stp_get_designated_root(ofproto->stp);
2402 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2411 update_stp_port_state(struct ofport_dpif *ofport)
2413 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2414 enum stp_state state;
2416 /* Figure out new state. */
2417 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2421 if (ofport->stp_state != state) {
2422 enum ofputil_port_state of_state;
2425 VLOG_DBG("port %s: STP state changed from %s to %s",
2426 netdev_get_name(ofport->up.netdev),
2427 stp_state_name(ofport->stp_state),
2428 stp_state_name(state));
2429 if (stp_learn_in_state(ofport->stp_state)
2430 != stp_learn_in_state(state)) {
2431 /* xxx Learning action flows should also be flushed. */
2432 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2433 mac_learning_flush(ofproto->ml);
2434 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2435 mcast_snooping_mdb_flush(ofproto->ms);
2437 fwd_change = stp_forward_in_state(ofport->stp_state)
2438 != stp_forward_in_state(state);
2440 ofproto->backer->need_revalidate = REV_STP;
2441 ofport->stp_state = state;
2442 ofport->stp_state_entered = time_msec();
2444 if (fwd_change && ofport->bundle) {
2445 bundle_update(ofport->bundle);
2448 /* Update the STP state bits in the OpenFlow port description. */
2449 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2450 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2451 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2452 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2453 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2455 ofproto_port_set_state(&ofport->up, of_state);
2459 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2460 * caller is responsible for assigning STP port numbers and ensuring
2461 * there are no duplicates. */
2463 set_stp_port(struct ofport *ofport_,
2464 const struct ofproto_port_stp_settings *s)
2466 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2467 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2468 struct stp_port *sp = ofport->stp_port;
2470 if (!s || !s->enable) {
2472 ofport->stp_port = NULL;
2473 stp_port_disable(sp);
2474 update_stp_port_state(ofport);
2477 } else if (sp && stp_port_no(sp) != s->port_num
2478 && ofport == stp_port_get_aux(sp)) {
2479 /* The port-id changed, so disable the old one if it's not
2480 * already in use by another port. */
2481 stp_port_disable(sp);
2484 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2486 /* Set name before enabling the port so that debugging messages can print
2488 stp_port_set_name(sp, netdev_get_name(ofport->up.netdev));
2489 stp_port_enable(sp);
2491 stp_port_set_aux(sp, ofport);
2492 stp_port_set_priority(sp, s->priority);
2493 stp_port_set_path_cost(sp, s->path_cost);
2495 update_stp_port_state(ofport);
2501 get_stp_port_status(struct ofport *ofport_,
2502 struct ofproto_port_stp_status *s)
2504 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2505 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2506 struct stp_port *sp = ofport->stp_port;
2508 if (!ofproto->stp || !sp) {
2514 s->port_id = stp_port_get_id(sp);
2515 s->state = stp_port_get_state(sp);
2516 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2517 s->role = stp_port_get_role(sp);
2523 get_stp_port_stats(struct ofport *ofport_,
2524 struct ofproto_port_stp_stats *s)
2526 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2527 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2528 struct stp_port *sp = ofport->stp_port;
2530 if (!ofproto->stp || !sp) {
2536 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2542 stp_run(struct ofproto_dpif *ofproto)
2545 long long int now = time_msec();
2546 long long int elapsed = now - ofproto->stp_last_tick;
2547 struct stp_port *sp;
2550 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2551 ofproto->stp_last_tick = now;
2553 while (stp_get_changed_port(ofproto->stp, &sp)) {
2554 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2557 update_stp_port_state(ofport);
2561 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2562 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2563 mac_learning_flush(ofproto->ml);
2564 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2565 mcast_snooping_mdb_flush(ofproto->ms);
2571 stp_wait(struct ofproto_dpif *ofproto)
2574 poll_timer_wait(1000);
2578 /* Configures RSTP on 'ofport_' using the settings defined in 's'. The
2579 * caller is responsible for assigning RSTP port numbers and ensuring
2580 * there are no duplicates. */
2582 set_rstp_port(struct ofport *ofport_,
2583 const struct ofproto_port_rstp_settings *s)
2585 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2586 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2587 struct rstp_port *rp = ofport->rstp_port;
2589 if (!s || !s->enable) {
2591 rstp_port_set_aux(rp, NULL);
2592 rstp_port_set_state(rp, RSTP_DISABLED);
2593 rstp_port_set_mac_operational(rp, false);
2594 ofport->rstp_port = NULL;
2595 rstp_port_unref(rp);
2596 update_rstp_port_state(ofport);
2601 /* Check if need to add a new port. */
2603 rp = ofport->rstp_port = rstp_add_port(ofproto->rstp);
2606 rstp_port_set(rp, s->port_num, s->priority, s->path_cost,
2607 s->admin_edge_port, s->auto_edge,
2608 s->admin_p2p_mac_state, s->admin_port_state, s->mcheck,
2610 update_rstp_port_state(ofport);
2611 /* Synchronize operational status. */
2612 rstp_port_set_mac_operational(rp, ofport->may_enable);
2616 get_rstp_port_status(struct ofport *ofport_,
2617 struct ofproto_port_rstp_status *s)
2619 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2620 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2621 struct rstp_port *rp = ofport->rstp_port;
2623 if (!ofproto->rstp || !rp) {
2629 rstp_port_get_status(rp, &s->port_id, &s->state, &s->role,
2630 &s->designated_bridge_id, &s->designated_port_id,
2631 &s->designated_path_cost, &s->tx_count,
2632 &s->rx_count, &s->error_count, &s->uptime);
2637 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2640 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2641 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2643 if (ofport->n_qdscp != n_qdscp
2644 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2645 n_qdscp * sizeof *qdscp))) {
2646 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2647 free(ofport->qdscp);
2648 ofport->qdscp = n_qdscp
2649 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2651 ofport->n_qdscp = n_qdscp;
2659 /* Expires all MAC learning entries associated with 'bundle' and forces its
2660 * ofproto to revalidate every flow.
2662 * Normally MAC learning entries are removed only from the ofproto associated
2663 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2664 * are removed from every ofproto. When patch ports and SLB bonds are in use
2665 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2666 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2667 * with the host from which it migrated. */
2669 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2671 struct ofproto_dpif *ofproto = bundle->ofproto;
2672 struct mac_learning *ml = ofproto->ml;
2673 struct mac_entry *mac, *next_mac;
2675 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2676 ovs_rwlock_wrlock(&ml->rwlock);
2677 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2678 if (mac_entry_get_port(ml, mac) == bundle) {
2680 struct ofproto_dpif *o;
2682 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2684 struct mac_entry *e;
2686 ovs_rwlock_wrlock(&o->ml->rwlock);
2687 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2689 mac_learning_expire(o->ml, e);
2691 ovs_rwlock_unlock(&o->ml->rwlock);
2696 mac_learning_expire(ml, mac);
2699 ovs_rwlock_unlock(&ml->rwlock);
2703 bundle_move(struct ofbundle *old, struct ofbundle *new)
2705 struct ofproto_dpif *ofproto = old->ofproto;
2706 struct mac_learning *ml = ofproto->ml;
2707 struct mac_entry *mac, *next_mac;
2709 ovs_assert(new->ofproto == old->ofproto);
2711 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2712 ovs_rwlock_wrlock(&ml->rwlock);
2713 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2714 if (mac_entry_get_port(ml, mac) == old) {
2715 mac_entry_set_port(ml, mac, new);
2718 ovs_rwlock_unlock(&ml->rwlock);
2721 static struct ofbundle *
2722 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2724 struct ofbundle *bundle;
2726 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2727 &ofproto->bundles) {
2728 if (bundle->aux == aux) {
2736 bundle_update(struct ofbundle *bundle)
2738 struct ofport_dpif *port;
2740 bundle->floodable = true;
2741 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2742 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2744 || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2745 || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2746 bundle->floodable = false;
2753 bundle_del_port(struct ofport_dpif *port)
2755 struct ofbundle *bundle = port->bundle;
2757 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2759 list_remove(&port->bundle_node);
2760 port->bundle = NULL;
2763 lacp_slave_unregister(bundle->lacp, port);
2766 bond_slave_unregister(bundle->bond, port);
2769 bundle_update(bundle);
2773 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2774 struct lacp_slave_settings *lacp)
2776 struct ofport_dpif *port;
2778 port = ofp_port_to_ofport(bundle->ofproto, ofp_port);
2783 if (port->bundle != bundle) {
2784 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2786 bundle_remove(&port->up);
2789 port->bundle = bundle;
2790 list_push_back(&bundle->ports, &port->bundle_node);
2791 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2793 || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2794 || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2795 bundle->floodable = false;
2799 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2800 lacp_slave_register(bundle->lacp, port, lacp);
2807 bundle_destroy(struct ofbundle *bundle)
2809 struct ofproto_dpif *ofproto;
2810 struct ofport_dpif *port, *next_port;
2816 ofproto = bundle->ofproto;
2817 mbridge_unregister_bundle(ofproto->mbridge, bundle);
2820 xlate_bundle_remove(bundle);
2823 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2824 bundle_del_port(port);
2827 bundle_flush_macs(bundle, true);
2828 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2830 free(bundle->trunks);
2831 lacp_unref(bundle->lacp);
2832 bond_unref(bundle->bond);
2837 bundle_set(struct ofproto *ofproto_, void *aux,
2838 const struct ofproto_bundle_settings *s)
2840 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2841 bool need_flush = false;
2842 struct ofport_dpif *port;
2843 struct ofbundle *bundle;
2844 unsigned long *trunks;
2850 bundle_destroy(bundle_lookup(ofproto, aux));
2854 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2855 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2857 bundle = bundle_lookup(ofproto, aux);
2859 bundle = xmalloc(sizeof *bundle);
2861 bundle->ofproto = ofproto;
2862 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2863 hash_pointer(aux, 0));
2865 bundle->name = NULL;
2867 list_init(&bundle->ports);
2868 bundle->vlan_mode = PORT_VLAN_TRUNK;
2870 bundle->trunks = NULL;
2871 bundle->use_priority_tags = s->use_priority_tags;
2872 bundle->lacp = NULL;
2873 bundle->bond = NULL;
2875 bundle->floodable = true;
2876 mbridge_register_bundle(ofproto->mbridge, bundle);
2879 if (!bundle->name || strcmp(s->name, bundle->name)) {
2881 bundle->name = xstrdup(s->name);
2886 ofproto->lacp_enabled = true;
2887 if (!bundle->lacp) {
2888 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2889 bundle->lacp = lacp_create();
2891 lacp_configure(bundle->lacp, s->lacp);
2893 lacp_unref(bundle->lacp);
2894 bundle->lacp = NULL;
2897 /* Update set of ports. */
2899 for (i = 0; i < s->n_slaves; i++) {
2900 if (!bundle_add_port(bundle, s->slaves[i],
2901 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2905 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2906 struct ofport_dpif *next_port;
2908 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2909 for (i = 0; i < s->n_slaves; i++) {
2910 if (s->slaves[i] == port->up.ofp_port) {
2915 bundle_del_port(port);
2919 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2921 if (list_is_empty(&bundle->ports)) {
2922 bundle_destroy(bundle);
2926 /* Set VLAN tagging mode */
2927 if (s->vlan_mode != bundle->vlan_mode
2928 || s->use_priority_tags != bundle->use_priority_tags) {
2929 bundle->vlan_mode = s->vlan_mode;
2930 bundle->use_priority_tags = s->use_priority_tags;
2935 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2936 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2938 if (vlan != bundle->vlan) {
2939 bundle->vlan = vlan;
2943 /* Get trunked VLANs. */
2944 switch (s->vlan_mode) {
2945 case PORT_VLAN_ACCESS:
2949 case PORT_VLAN_TRUNK:
2950 trunks = CONST_CAST(unsigned long *, s->trunks);
2953 case PORT_VLAN_NATIVE_UNTAGGED:
2954 case PORT_VLAN_NATIVE_TAGGED:
2955 if (vlan != 0 && (!s->trunks
2956 || !bitmap_is_set(s->trunks, vlan)
2957 || bitmap_is_set(s->trunks, 0))) {
2958 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2960 trunks = bitmap_clone(s->trunks, 4096);
2962 trunks = bitmap_allocate1(4096);
2964 bitmap_set1(trunks, vlan);
2965 bitmap_set0(trunks, 0);
2967 trunks = CONST_CAST(unsigned long *, s->trunks);
2974 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2975 free(bundle->trunks);
2976 if (trunks == s->trunks) {
2977 bundle->trunks = vlan_bitmap_clone(trunks);
2979 bundle->trunks = trunks;
2984 if (trunks != s->trunks) {
2989 if (!list_is_short(&bundle->ports)) {
2990 bundle->ofproto->has_bonded_bundles = true;
2992 if (bond_reconfigure(bundle->bond, s->bond)) {
2993 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2996 bundle->bond = bond_create(s->bond, ofproto);
2997 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3000 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
3001 bond_slave_register(bundle->bond, port,
3002 port->up.ofp_port, port->up.netdev);
3005 bond_unref(bundle->bond);
3006 bundle->bond = NULL;
3009 /* If we changed something that would affect MAC learning, un-learn
3010 * everything on this port and force flow revalidation. */
3012 bundle_flush_macs(bundle, false);
3019 bundle_remove(struct ofport *port_)
3021 struct ofport_dpif *port = ofport_dpif_cast(port_);
3022 struct ofbundle *bundle = port->bundle;
3025 bundle_del_port(port);
3026 if (list_is_empty(&bundle->ports)) {
3027 bundle_destroy(bundle);
3028 } else if (list_is_short(&bundle->ports)) {
3029 bond_unref(bundle->bond);
3030 bundle->bond = NULL;
3036 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
3038 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3039 struct ofport_dpif *port = port_;
3043 error = netdev_get_etheraddr(port->up.netdev, &ea);
3045 struct dp_packet packet;
3048 dp_packet_init(&packet, 0);
3049 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
3051 memcpy(packet_pdu, pdu, pdu_size);
3053 ofproto_dpif_send_packet(port, &packet);
3054 dp_packet_uninit(&packet);
3056 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
3057 "%s (%s)", port->bundle->name,
3058 netdev_get_name(port->up.netdev), ovs_strerror(error));
3063 bundle_send_learning_packets(struct ofbundle *bundle)
3065 struct ofproto_dpif *ofproto = bundle->ofproto;
3066 int error, n_packets, n_errors;
3067 struct mac_entry *e;
3069 struct ovs_list list_node;
3070 struct ofport_dpif *port;
3071 struct dp_packet *pkt;
3073 struct ovs_list packets;
3075 list_init(&packets);
3076 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
3077 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3078 if (mac_entry_get_port(ofproto->ml, e) != bundle) {
3079 pkt_node = xmalloc(sizeof *pkt_node);
3080 pkt_node->pkt = bond_compose_learning_packet(bundle->bond,
3082 (void **)&pkt_node->port);
3083 list_push_back(&packets, &pkt_node->list_node);
3086 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3088 error = n_packets = n_errors = 0;
3089 LIST_FOR_EACH_POP (pkt_node, list_node, &packets) {
3092 ret = ofproto_dpif_send_packet(pkt_node->port, pkt_node->pkt);
3093 dp_packet_delete(pkt_node->pkt);
3103 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3104 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3105 "packets, last error was: %s",
3106 bundle->name, n_errors, n_packets, ovs_strerror(error));
3108 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3109 bundle->name, n_packets);
3114 bundle_run(struct ofbundle *bundle)
3117 lacp_run(bundle->lacp, send_pdu_cb);
3120 struct ofport_dpif *port;
3122 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
3123 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
3126 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
3127 bundle->ofproto->backer->need_revalidate = REV_BOND;
3130 if (bond_should_send_learning_packets(bundle->bond)) {
3131 bundle_send_learning_packets(bundle);
3137 bundle_wait(struct ofbundle *bundle)
3140 lacp_wait(bundle->lacp);
3143 bond_wait(bundle->bond);
3150 mirror_set__(struct ofproto *ofproto_, void *aux,
3151 const struct ofproto_mirror_settings *s)
3153 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3154 struct ofbundle **srcs, **dsts;
3159 mirror_destroy(ofproto->mbridge, aux);
3163 srcs = xmalloc(s->n_srcs * sizeof *srcs);
3164 dsts = xmalloc(s->n_dsts * sizeof *dsts);
3166 for (i = 0; i < s->n_srcs; i++) {
3167 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
3170 for (i = 0; i < s->n_dsts; i++) {
3171 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
3174 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
3175 s->n_dsts, s->src_vlans,
3176 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
3183 mirror_get_stats__(struct ofproto *ofproto, void *aux,
3184 uint64_t *packets, uint64_t *bytes)
3186 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
3191 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3193 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3194 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3195 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3196 mac_learning_flush(ofproto->ml);
3198 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3203 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3205 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3206 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3207 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
3211 forward_bpdu_changed(struct ofproto *ofproto_)
3213 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3214 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3218 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3221 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3222 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3223 mac_learning_set_idle_time(ofproto->ml, idle_time);
3224 mac_learning_set_max_entries(ofproto->ml, max_entries);
3225 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3228 /* Configures multicast snooping on 'ofport' using the settings
3229 * defined in 's'. */
3231 set_mcast_snooping(struct ofproto *ofproto_,
3232 const struct ofproto_mcast_snooping_settings *s)
3234 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3236 /* Only revalidate flows if the configuration changed. */
3237 if (!s != !ofproto->ms) {
3238 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3243 ofproto->ms = mcast_snooping_create();
3246 ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3247 mcast_snooping_set_idle_time(ofproto->ms, s->idle_time);
3248 mcast_snooping_set_max_entries(ofproto->ms, s->max_entries);
3249 if (mcast_snooping_set_flood_unreg(ofproto->ms, s->flood_unreg)) {
3250 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3252 ovs_rwlock_unlock(&ofproto->ms->rwlock);
3254 mcast_snooping_unref(ofproto->ms);
3261 /* Configures multicast snooping port's flood settings on 'ofproto'. */
3263 set_mcast_snooping_port(struct ofproto *ofproto_, void *aux,
3264 const struct ofproto_mcast_snooping_port_settings *s)
3266 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3267 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3269 if (ofproto->ms && s) {
3270 ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3271 mcast_snooping_set_port_flood(ofproto->ms, bundle, s->flood);
3272 mcast_snooping_set_port_flood_reports(ofproto->ms, bundle,
3274 ovs_rwlock_unlock(&ofproto->ms->rwlock);
3282 struct ofport_dpif *
3283 ofp_port_to_ofport(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
3285 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3286 return ofport ? ofport_dpif_cast(ofport) : NULL;
3290 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3291 struct ofproto_port *ofproto_port,
3292 struct dpif_port *dpif_port)
3294 ofproto_port->name = dpif_port->name;
3295 ofproto_port->type = dpif_port->type;
3296 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3300 ofport_update_peer(struct ofport_dpif *ofport)
3302 const struct ofproto_dpif *ofproto;
3303 struct dpif_backer *backer;
3306 if (!netdev_vport_is_patch(ofport->up.netdev)) {
3310 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
3311 backer->need_revalidate = REV_RECONFIGURE;
3314 ofport->peer->peer = NULL;
3315 ofport->peer = NULL;
3318 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
3323 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3324 struct ofport *peer_ofport;
3325 struct ofport_dpif *peer;
3328 if (ofproto->backer != backer) {
3332 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
3337 peer = ofport_dpif_cast(peer_ofport);
3338 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
3339 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
3341 ofport->peer = peer;
3342 ofport->peer->peer = ofport;
3352 port_run(struct ofport_dpif *ofport)
3354 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3355 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3356 bool enable = netdev_get_carrier(ofport->up.netdev);
3357 bool cfm_enable = false;
3358 bool bfd_enable = false;
3360 ofport->carrier_seq = carrier_seq;
3363 int cfm_opup = cfm_get_opup(ofport->cfm);
3365 cfm_enable = !cfm_get_fault(ofport->cfm);
3367 if (cfm_opup >= 0) {
3368 cfm_enable = cfm_enable && cfm_opup;
3373 bfd_enable = bfd_forwarding(ofport->bfd);
3376 if (ofport->bfd || ofport->cfm) {
3377 enable = enable && (cfm_enable || bfd_enable);
3380 if (ofport->bundle) {
3381 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3382 if (carrier_changed) {
3383 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3387 if (ofport->may_enable != enable) {
3388 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3390 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3392 if (ofport->rstp_port) {
3393 rstp_port_set_mac_operational(ofport->rstp_port, enable);
3397 ofport->may_enable = enable;
3401 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3402 struct ofproto_port *ofproto_port)
3404 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3405 struct dpif_port dpif_port;
3408 if (sset_contains(&ofproto->ghost_ports, devname)) {
3409 const char *type = netdev_get_type_from_name(devname);
3411 /* We may be called before ofproto->up.port_by_name is populated with
3412 * the appropriate ofport. For this reason, we must get the name and
3413 * type from the netdev layer directly. */
3415 const struct ofport *ofport;
3417 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3418 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3419 ofproto_port->name = xstrdup(devname);
3420 ofproto_port->type = xstrdup(type);
3426 if (!sset_contains(&ofproto->ports, devname)) {
3429 error = dpif_port_query_by_name(ofproto->backer->dpif,
3430 devname, &dpif_port);
3432 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3438 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3440 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3441 const char *devname = netdev_get_name(netdev);
3442 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3443 const char *dp_port_name;
3445 if (netdev_vport_is_patch(netdev)) {
3446 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3450 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3451 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3452 odp_port_t port_no = ODPP_NONE;
3455 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3459 if (netdev_get_tunnel_config(netdev)) {
3460 simap_put(&ofproto->backer->tnl_backers,
3461 dp_port_name, odp_to_u32(port_no));
3465 if (netdev_get_tunnel_config(netdev)) {
3466 sset_add(&ofproto->ghost_ports, devname);
3468 sset_add(&ofproto->ports, devname);
3474 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3476 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3477 struct ofport_dpif *ofport = ofp_port_to_ofport(ofproto, ofp_port);
3484 sset_find_and_delete(&ofproto->ghost_ports,
3485 netdev_get_name(ofport->up.netdev));
3486 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3487 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3488 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3490 /* The caller is going to close ofport->up.netdev. If this is a
3491 * bonded port, then the bond is using that netdev, so remove it
3492 * from the bond. The client will need to reconfigure everything
3493 * after deleting ports, so then the slave will get re-added. */
3494 bundle_remove(&ofport->up);
3501 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3503 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3506 error = netdev_get_stats(ofport->up.netdev, stats);
3508 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3509 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3511 ovs_mutex_lock(&ofproto->stats_mutex);
3512 /* ofproto->stats.tx_packets represents packets that we created
3513 * internally and sent to some port (e.g. packets sent with
3514 * ofproto_dpif_send_packet()). Account for them as if they had
3515 * come from OFPP_LOCAL and got forwarded. */
3517 if (stats->rx_packets != UINT64_MAX) {
3518 stats->rx_packets += ofproto->stats.tx_packets;
3521 if (stats->rx_bytes != UINT64_MAX) {
3522 stats->rx_bytes += ofproto->stats.tx_bytes;
3525 /* ofproto->stats.rx_packets represents packets that were received on
3526 * some port and we processed internally and dropped (e.g. STP).
3527 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3529 if (stats->tx_packets != UINT64_MAX) {
3530 stats->tx_packets += ofproto->stats.rx_packets;
3533 if (stats->tx_bytes != UINT64_MAX) {
3534 stats->tx_bytes += ofproto->stats.rx_bytes;
3536 ovs_mutex_unlock(&ofproto->stats_mutex);
3543 port_get_lacp_stats(const struct ofport *ofport_, struct lacp_slave_stats *stats)
3545 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3546 if (ofport->bundle && ofport->bundle->lacp) {
3547 if (lacp_get_slave_stats(ofport->bundle->lacp, ofport, stats)) {
3554 struct port_dump_state {
3559 struct ofproto_port port;
3564 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3566 *statep = xzalloc(sizeof(struct port_dump_state));
3571 port_dump_next(const struct ofproto *ofproto_, void *state_,
3572 struct ofproto_port *port)
3574 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3575 struct port_dump_state *state = state_;
3576 const struct sset *sset;
3577 struct sset_node *node;
3579 if (state->has_port) {
3580 ofproto_port_destroy(&state->port);
3581 state->has_port = false;
3583 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3584 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3587 error = port_query_by_name(ofproto_, node->name, &state->port);
3589 *port = state->port;
3590 state->has_port = true;
3592 } else if (error != ENODEV) {
3597 if (!state->ghost) {
3598 state->ghost = true;
3601 return port_dump_next(ofproto_, state_, port);
3608 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3610 struct port_dump_state *state = state_;
3612 if (state->has_port) {
3613 ofproto_port_destroy(&state->port);
3620 port_poll(const struct ofproto *ofproto_, char **devnamep)
3622 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3624 if (ofproto->port_poll_errno) {
3625 int error = ofproto->port_poll_errno;
3626 ofproto->port_poll_errno = 0;
3630 if (sset_is_empty(&ofproto->port_poll_set)) {
3634 *devnamep = sset_pop(&ofproto->port_poll_set);
3639 port_poll_wait(const struct ofproto *ofproto_)
3641 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3642 dpif_port_poll_wait(ofproto->backer->dpif);
3646 port_is_lacp_current(const struct ofport *ofport_)
3648 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3649 return (ofport->bundle && ofport->bundle->lacp
3650 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3654 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3655 * then delete it entirely. */
3657 rule_expire(struct rule_dpif *rule)
3658 OVS_REQUIRES(ofproto_mutex)
3660 uint16_t hard_timeout, idle_timeout;
3661 long long int now = time_msec();
3664 hard_timeout = rule->up.hard_timeout;
3665 idle_timeout = rule->up.idle_timeout;
3667 /* Has 'rule' expired? */
3669 long long int modified;
3671 ovs_mutex_lock(&rule->up.mutex);
3672 modified = rule->up.modified;
3673 ovs_mutex_unlock(&rule->up.mutex);
3675 if (now > modified + hard_timeout * 1000) {
3676 reason = OFPRR_HARD_TIMEOUT;
3680 if (reason < 0 && idle_timeout) {
3683 ovs_mutex_lock(&rule->stats_mutex);
3684 used = rule->stats.used;
3685 ovs_mutex_unlock(&rule->stats_mutex);
3687 if (now > used + idle_timeout * 1000) {
3688 reason = OFPRR_IDLE_TIMEOUT;
3693 COVERAGE_INC(ofproto_dpif_expired);
3694 ofproto_rule_expire(&rule->up, reason);
3699 ofproto_dpif_execute_actions__(struct ofproto_dpif *ofproto,
3700 const struct flow *flow,
3701 struct rule_dpif *rule,
3702 const struct ofpact *ofpacts, size_t ofpacts_len,
3703 int recurse, int resubmits,
3704 struct dp_packet *packet)
3706 struct dpif_flow_stats stats;
3707 struct xlate_out xout;
3708 struct xlate_in xin;
3710 struct dpif_execute execute;
3713 ovs_assert((rule != NULL) != (ofpacts != NULL));
3715 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3718 rule_dpif_credit_stats(rule, &stats);
3721 uint64_t odp_actions_stub[1024 / 8];
3722 struct ofpbuf odp_actions = OFPBUF_STUB_INITIALIZER(odp_actions_stub);
3723 xlate_in_init(&xin, ofproto, flow, flow->in_port.ofp_port, rule,
3724 stats.tcp_flags, packet, NULL, &odp_actions);
3725 xin.ofpacts = ofpacts;
3726 xin.ofpacts_len = ofpacts_len;
3727 xin.resubmit_stats = &stats;
3728 xin.recurse = recurse;
3729 xin.resubmits = resubmits;
3730 if (xlate_actions(&xin, &xout) != XLATE_OK) {
3735 execute.actions = odp_actions.data;
3736 execute.actions_len = odp_actions.size;
3738 pkt_metadata_from_flow(&packet->md, flow);
3739 execute.packet = packet;
3740 execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
3741 execute.probe = false;
3744 /* Fix up in_port. */
3745 in_port = flow->in_port.ofp_port;
3746 if (in_port == OFPP_NONE) {
3747 in_port = OFPP_LOCAL;
3749 execute.packet->md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
3751 error = dpif_execute(ofproto->backer->dpif, &execute);
3753 xlate_out_uninit(&xout);
3754 ofpbuf_uninit(&odp_actions);
3759 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3760 * 'flow' must reflect the data in 'packet'. */
3762 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3763 const struct flow *flow,
3764 struct rule_dpif *rule,
3765 const struct ofpact *ofpacts, size_t ofpacts_len,
3766 struct dp_packet *packet)
3768 return ofproto_dpif_execute_actions__(ofproto, flow, rule, ofpacts,
3769 ofpacts_len, 0, 0, packet);
3773 rule_dpif_credit_stats(struct rule_dpif *rule,
3774 const struct dpif_flow_stats *stats)
3776 ovs_mutex_lock(&rule->stats_mutex);
3777 if (OVS_UNLIKELY(rule->new_rule)) {
3778 rule_dpif_credit_stats(rule->new_rule, stats);
3780 rule->stats.n_packets += stats->n_packets;
3781 rule->stats.n_bytes += stats->n_bytes;
3782 rule->stats.used = MAX(rule->stats.used, stats->used);
3784 ovs_mutex_unlock(&rule->stats_mutex);
3788 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
3789 OVS_REQUIRES(rule->up.mutex)
3791 return rule->up.flow_cookie;
3795 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
3796 uint16_t hard_timeout)
3798 ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
3801 /* Returns 'rule''s actions. The returned actions are RCU-protected, and can
3802 * be read until the calling thread quiesces. */
3803 const struct rule_actions *
3804 rule_dpif_get_actions(const struct rule_dpif *rule)
3806 return rule_get_actions(&rule->up);
3809 /* Sets 'rule''s recirculation id. */
3811 rule_dpif_set_recirc_id(struct rule_dpif *rule, uint32_t id)
3812 OVS_REQUIRES(rule->up.mutex)
3814 ovs_assert(!rule->recirc_id || rule->recirc_id == id);
3815 if (rule->recirc_id == id) {
3816 /* Release the new reference to the same id. */
3819 rule->recirc_id = id;
3823 /* Sets 'rule''s recirculation id. */
3825 rule_set_recirc_id(struct rule *rule_, uint32_t id)
3827 struct rule_dpif *rule = rule_dpif_cast(rule_);
3829 ovs_mutex_lock(&rule->up.mutex);
3830 rule_dpif_set_recirc_id(rule, id);
3831 ovs_mutex_unlock(&rule->up.mutex);
3835 ofproto_dpif_get_tables_version(struct ofproto_dpif *ofproto OVS_UNUSED)
3837 cls_version_t version;
3839 atomic_read_relaxed(&ofproto->tables_version, &version);
3844 /* The returned rule (if any) is valid at least until the next RCU quiescent
3845 * period. If the rule needs to stay around longer, the caller should take
3848 * 'flow' is non-const to allow for temporary modifications during the lookup.
3849 * Any changes are restored before returning. */
3850 static struct rule_dpif *
3851 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, cls_version_t version,
3852 uint8_t table_id, struct flow *flow,
3853 struct flow_wildcards *wc)
3855 struct classifier *cls = &ofproto->up.tables[table_id].cls;
3856 return rule_dpif_cast(rule_from_cls_rule(classifier_lookup(cls, version,
3860 /* Look up 'flow' in 'ofproto''s classifier version 'version', starting from
3861 * table '*table_id'. Returns the rule that was found, which may be one of the
3862 * special rules according to packet miss hadling. If 'may_packet_in' is
3863 * false, returning of the miss_rule (which issues packet ins for the
3864 * controller) is avoided. Updates 'wc', if nonnull, to reflect the fields
3865 * that were used during the lookup.
3867 * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3868 * if none is found then the table miss configuration for that table is
3869 * honored, which can result in additional lookups in other OpenFlow tables.
3870 * In this case the function updates '*table_id' to reflect the final OpenFlow
3871 * table that was searched.
3873 * If 'honor_table_miss' is false, then only one table lookup occurs, in
3876 * The rule is returned in '*rule', which is valid at least until the next
3877 * RCU quiescent period. If the '*rule' needs to stay around longer, the
3878 * caller must take a reference.
3880 * 'in_port' allows the lookup to take place as if the in port had the value
3881 * 'in_port'. This is needed for resubmit action support.
3883 * 'flow' is non-const to allow for temporary modifications during the lookup.
3884 * Any changes are restored before returning. */
3886 rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto,
3887 cls_version_t version, struct flow *flow,
3888 struct flow_wildcards *wc,
3889 const struct dpif_flow_stats *stats,
3890 uint8_t *table_id, ofp_port_t in_port,
3891 bool may_packet_in, bool honor_table_miss)
3893 ovs_be16 old_tp_src = flow->tp_src, old_tp_dst = flow->tp_dst;
3894 ofp_port_t old_in_port = flow->in_port.ofp_port;
3895 enum ofputil_table_miss miss_config;
3896 struct rule_dpif *rule;
3899 /* We always unwildcard nw_frag (for IP), so they
3900 * need not be unwildcarded here. */
3901 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3902 && ofproto->up.frag_handling != OFPUTIL_FRAG_NX_MATCH) {
3903 if (ofproto->up.frag_handling == OFPUTIL_FRAG_NORMAL) {
3904 /* We must pretend that transport ports are unavailable. */
3905 flow->tp_src = htons(0);
3906 flow->tp_dst = htons(0);
3908 /* Must be OFPUTIL_FRAG_DROP (we don't have OFPUTIL_FRAG_REASM).
3909 * Use the drop_frags_rule (which cannot disappear). */
3910 rule = ofproto->drop_frags_rule;
3912 struct oftable *tbl = &ofproto->up.tables[*table_id];
3915 atomic_add_relaxed(&tbl->n_matched, stats->n_packets, &orig);
3921 /* Look up a flow with 'in_port' as the input port. Then restore the
3922 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
3923 * have surprising behavior). */
3924 flow->in_port.ofp_port = in_port;
3926 /* Our current implementation depends on n_tables == N_TABLES, and
3927 * TBL_INTERNAL being the last table. */
3928 BUILD_ASSERT_DECL(N_TABLES == TBL_INTERNAL + 1);
3930 miss_config = OFPUTIL_TABLE_MISS_CONTINUE;
3932 for (next_id = *table_id;
3933 next_id < ofproto->up.n_tables;
3934 next_id++, next_id += (next_id == TBL_INTERNAL))
3936 *table_id = next_id;
3937 rule = rule_dpif_lookup_in_table(ofproto, version, next_id, flow, wc);
3939 struct oftable *tbl = &ofproto->up.tables[next_id];
3942 atomic_add_relaxed(rule ? &tbl->n_matched : &tbl->n_missed,
3943 stats->n_packets, &orig);
3946 goto out; /* Match. */
3948 if (honor_table_miss) {
3949 miss_config = ofproto_table_get_miss_config(&ofproto->up,
3951 if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE) {
3958 rule = ofproto->no_packet_in_rule;
3959 if (may_packet_in) {
3960 if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE
3961 || miss_config == OFPUTIL_TABLE_MISS_CONTROLLER) {
3962 struct ofport_dpif *port;
3964 port = ofp_port_to_ofport(ofproto, old_in_port);
3966 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
3968 } else if (!(port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN)) {
3969 rule = ofproto->miss_rule;
3971 } else if (miss_config == OFPUTIL_TABLE_MISS_DEFAULT &&
3972 connmgr_wants_packet_in_on_miss(ofproto->up.connmgr)) {
3973 rule = ofproto->miss_rule;
3977 /* Restore port numbers, as they may have been modified above. */
3978 flow->tp_src = old_tp_src;
3979 flow->tp_dst = old_tp_dst;
3980 /* Restore the old in port. */
3981 flow->in_port.ofp_port = old_in_port;
3987 complete_operation(struct rule_dpif *rule)
3988 OVS_REQUIRES(ofproto_mutex)
3990 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3992 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3995 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
3997 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4000 static struct rule *
4003 struct rule_dpif *rule = xzalloc(sizeof *rule);
4008 rule_dealloc(struct rule *rule_)
4010 struct rule_dpif *rule = rule_dpif_cast(rule_);
4015 check_mask(struct ofproto_dpif *ofproto, const struct miniflow *flow)
4017 const struct odp_support *support;
4018 uint16_t ct_state, ct_zone;
4022 support = &ofproto_dpif_get_support(ofproto)->odp;
4023 ct_state = MINIFLOW_GET_U16(flow, ct_state);
4024 if (support->ct_state && support->ct_zone && support->ct_mark
4025 && support->ct_label && support->ct_state_nat) {
4026 return ct_state & CS_UNSUPPORTED_MASK ? OFPERR_OFPBMC_BAD_MASK : 0;
4029 ct_zone = MINIFLOW_GET_U16(flow, ct_zone);
4030 ct_mark = MINIFLOW_GET_U32(flow, ct_mark);
4031 ct_label = MINIFLOW_GET_U128(flow, ct_label);
4033 if ((ct_state && !support->ct_state)
4034 || (ct_state & CS_UNSUPPORTED_MASK)
4035 || ((ct_state & (CS_SRC_NAT | CS_DST_NAT)) && !support->ct_state_nat)
4036 || (ct_zone && !support->ct_zone)
4037 || (ct_mark && !support->ct_mark)
4038 || (!ovs_u128_is_zero(&ct_label) && !support->ct_label)) {
4039 return OFPERR_OFPBMC_BAD_MASK;
4046 check_actions(const struct ofproto_dpif *ofproto,
4047 const struct rule_actions *const actions)
4049 const struct ofpact *ofpact;
4051 OFPACT_FOR_EACH (ofpact, actions->ofpacts, actions->ofpacts_len) {
4052 const struct odp_support *support;
4053 const struct ofpact_conntrack *ct;
4054 const struct ofpact *a;
4056 if (ofpact->type != OFPACT_CT) {
4060 ct = CONTAINER_OF(ofpact, struct ofpact_conntrack, ofpact);
4061 support = &ofproto_dpif_get_support(ofproto)->odp;
4063 if (!support->ct_state) {
4064 return OFPERR_OFPBAC_BAD_TYPE;
4066 if ((ct->zone_imm || ct->zone_src.field) && !support->ct_zone) {
4067 return OFPERR_OFPBAC_BAD_ARGUMENT;
4070 OFPACT_FOR_EACH(a, ct->actions, ofpact_ct_get_action_len(ct)) {
4071 const struct mf_field *dst = ofpact_get_mf_dst(a);
4073 if (a->type == OFPACT_NAT && !support->ct_state_nat) {
4074 /* The backer doesn't seem to support the NAT bits in
4075 * 'ct_state': assume that it doesn't support the NAT
4077 return OFPERR_OFPBAC_BAD_TYPE;
4079 if (dst && ((dst->id == MFF_CT_MARK && !support->ct_mark)
4080 || (dst->id == MFF_CT_LABEL && !support->ct_label))) {
4081 return OFPERR_OFPBAC_BAD_SET_ARGUMENT;
4090 rule_check(struct rule *rule)
4092 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->ofproto);
4095 err = check_mask(ofproto, &rule->cr.match.mask->masks);
4099 return check_actions(ofproto, rule->actions);
4103 rule_construct(struct rule *rule_)
4104 OVS_NO_THREAD_SAFETY_ANALYSIS
4106 struct rule_dpif *rule = rule_dpif_cast(rule_);
4109 error = rule_check(rule_);
4114 ovs_mutex_init_adaptive(&rule->stats_mutex);
4115 rule->stats.n_packets = 0;
4116 rule->stats.n_bytes = 0;
4117 rule->stats.used = rule->up.modified;
4118 rule->recirc_id = 0;
4119 rule->new_rule = NULL;
4125 rule_insert(struct rule *rule_, struct rule *old_rule_, bool forward_stats)
4126 OVS_REQUIRES(ofproto_mutex)
4128 struct rule_dpif *rule = rule_dpif_cast(rule_);
4130 if (old_rule_ && forward_stats) {
4131 struct rule_dpif *old_rule = rule_dpif_cast(old_rule_);
4133 ovs_assert(!old_rule->new_rule);
4135 /* Take a reference to the new rule, and refer all stats updates from
4136 * the old rule to the new rule. */
4137 rule_dpif_ref(rule);
4139 ovs_mutex_lock(&old_rule->stats_mutex);
4140 ovs_mutex_lock(&rule->stats_mutex);
4141 old_rule->new_rule = rule; /* Forward future stats. */
4142 rule->stats = old_rule->stats; /* Transfer stats to the new rule. */
4143 ovs_mutex_unlock(&rule->stats_mutex);
4144 ovs_mutex_unlock(&old_rule->stats_mutex);
4147 complete_operation(rule);
4151 rule_delete(struct rule *rule_)
4152 OVS_REQUIRES(ofproto_mutex)
4154 struct rule_dpif *rule = rule_dpif_cast(rule_);
4155 complete_operation(rule);
4159 rule_destruct(struct rule *rule_)
4160 OVS_NO_THREAD_SAFETY_ANALYSIS
4162 struct rule_dpif *rule = rule_dpif_cast(rule_);
4164 ovs_mutex_destroy(&rule->stats_mutex);
4165 /* Release reference to the new rule, if any. */
4166 if (rule->new_rule) {
4167 rule_dpif_unref(rule->new_rule);
4169 if (rule->recirc_id) {
4170 recirc_free_id(rule->recirc_id);
4175 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
4176 long long int *used)
4178 struct rule_dpif *rule = rule_dpif_cast(rule_);
4180 ovs_mutex_lock(&rule->stats_mutex);
4181 if (OVS_UNLIKELY(rule->new_rule)) {
4182 rule_get_stats(&rule->new_rule->up, packets, bytes, used);
4184 *packets = rule->stats.n_packets;
4185 *bytes = rule->stats.n_bytes;
4186 *used = rule->stats.used;
4188 ovs_mutex_unlock(&rule->stats_mutex);
4192 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4193 struct dp_packet *packet)
4195 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4197 ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
4201 rule_execute(struct rule *rule, const struct flow *flow,
4202 struct dp_packet *packet)
4204 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4205 dp_packet_delete(packet);
4209 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
4211 return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
4214 static struct ofgroup *
4217 struct group_dpif *group = xzalloc(sizeof *group);
4222 group_dealloc(struct ofgroup *group_)
4224 struct group_dpif *group = group_dpif_cast(group_);
4229 group_construct_stats(struct group_dpif *group)
4230 OVS_REQUIRES(group->stats_mutex)
4232 struct ofputil_bucket *bucket;
4233 const struct ovs_list *buckets;
4235 group->packet_count = 0;
4236 group->byte_count = 0;
4238 group_dpif_get_buckets(group, &buckets);
4239 LIST_FOR_EACH (bucket, list_node, buckets) {
4240 bucket->stats.packet_count = 0;
4241 bucket->stats.byte_count = 0;
4246 group_dpif_credit_stats(struct group_dpif *group,
4247 struct ofputil_bucket *bucket,
4248 const struct dpif_flow_stats *stats)
4250 ovs_mutex_lock(&group->stats_mutex);
4251 group->packet_count += stats->n_packets;
4252 group->byte_count += stats->n_bytes;
4254 bucket->stats.packet_count += stats->n_packets;
4255 bucket->stats.byte_count += stats->n_bytes;
4256 } else { /* Credit to all buckets */
4257 const struct ovs_list *buckets;
4259 group_dpif_get_buckets(group, &buckets);
4260 LIST_FOR_EACH (bucket, list_node, buckets) {
4261 bucket->stats.packet_count += stats->n_packets;
4262 bucket->stats.byte_count += stats->n_bytes;
4265 ovs_mutex_unlock(&group->stats_mutex);
4269 group_construct(struct ofgroup *group_)
4271 struct group_dpif *group = group_dpif_cast(group_);
4273 ovs_mutex_init_adaptive(&group->stats_mutex);
4274 ovs_mutex_lock(&group->stats_mutex);
4275 group_construct_stats(group);
4276 ovs_mutex_unlock(&group->stats_mutex);
4281 group_destruct(struct ofgroup *group_)
4283 struct group_dpif *group = group_dpif_cast(group_);
4284 ovs_mutex_destroy(&group->stats_mutex);
4288 group_modify(struct ofgroup *group_)
4290 struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
4292 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4298 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
4300 struct group_dpif *group = group_dpif_cast(group_);
4301 struct ofputil_bucket *bucket;
4302 const struct ovs_list *buckets;
4303 struct bucket_counter *bucket_stats;
4305 ovs_mutex_lock(&group->stats_mutex);
4306 ogs->packet_count = group->packet_count;
4307 ogs->byte_count = group->byte_count;
4309 group_dpif_get_buckets(group, &buckets);
4310 bucket_stats = ogs->bucket_stats;
4311 LIST_FOR_EACH (bucket, list_node, buckets) {
4312 bucket_stats->packet_count = bucket->stats.packet_count;
4313 bucket_stats->byte_count = bucket->stats.byte_count;
4316 ovs_mutex_unlock(&group->stats_mutex);
4321 /* If the group exists, this function increments the groups's reference count.
4323 * Make sure to call group_dpif_unref() after no longer needing to maintain
4324 * a reference to the group. */
4326 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
4327 struct group_dpif **group)
4329 struct ofgroup *ofgroup;
4332 found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
4333 *group = found ? group_dpif_cast(ofgroup) : NULL;
4339 group_dpif_get_buckets(const struct group_dpif *group,
4340 const struct ovs_list **buckets)
4342 *buckets = &group->up.buckets;
4345 enum ofp11_group_type
4346 group_dpif_get_type(const struct group_dpif *group)
4348 return group->up.type;
4352 group_dpif_get_selection_method(const struct group_dpif *group)
4354 return group->up.props.selection_method;
4357 /* Sends 'packet' out 'ofport'.
4358 * May modify 'packet'.
4359 * Returns 0 if successful, otherwise a positive errno value. */
4361 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct dp_packet *packet)
4363 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4366 error = xlate_send_packet(ofport, packet);
4368 ovs_mutex_lock(&ofproto->stats_mutex);
4369 ofproto->stats.tx_packets++;
4370 ofproto->stats.tx_bytes += dp_packet_size(packet);
4371 ovs_mutex_unlock(&ofproto->stats_mutex);
4376 group_dpif_get_selection_method_param(const struct group_dpif *group)
4378 return group->up.props.selection_method_param;
4381 const struct field_array *
4382 group_dpif_get_fields(const struct group_dpif *group)
4384 return &group->up.props.fields;
4387 /* Return the version string of the datapath that backs up
4391 get_datapath_version(const struct ofproto *ofproto_)
4393 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4395 return ofproto->backer->dp_version_string;
4399 set_frag_handling(struct ofproto *ofproto_,
4400 enum ofputil_frag_handling frag_handling)
4402 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4403 if (frag_handling != OFPUTIL_FRAG_REASM) {
4404 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4412 packet_out(struct ofproto *ofproto_, struct dp_packet *packet,
4413 const struct flow *flow,
4414 const struct ofpact *ofpacts, size_t ofpacts_len)
4416 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4418 ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
4419 ofpacts_len, packet);
4426 set_netflow(struct ofproto *ofproto_,
4427 const struct netflow_options *netflow_options)
4429 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4431 if (netflow_options) {
4432 if (!ofproto->netflow) {
4433 ofproto->netflow = netflow_create();
4434 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4436 return netflow_set_options(ofproto->netflow, netflow_options);
4437 } else if (ofproto->netflow) {
4438 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4439 netflow_unref(ofproto->netflow);
4440 ofproto->netflow = NULL;
4447 get_netflow_ids(const struct ofproto *ofproto_,
4448 uint8_t *engine_type, uint8_t *engine_id)
4450 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4452 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
4455 static struct ofproto_dpif *
4456 ofproto_dpif_lookup(const char *name)
4458 struct ofproto_dpif *ofproto;
4460 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
4461 hash_string(name, 0), &all_ofproto_dpifs) {
4462 if (!strcmp(ofproto->up.name, name)) {
4470 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
4471 const char *argv[], void *aux OVS_UNUSED)
4473 struct ofproto_dpif *ofproto;
4476 ofproto = ofproto_dpif_lookup(argv[1]);
4478 unixctl_command_reply_error(conn, "no such bridge");
4481 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4482 mac_learning_flush(ofproto->ml);
4483 ovs_rwlock_unlock(&ofproto->ml->rwlock);
4485 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4486 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4487 mac_learning_flush(ofproto->ml);
4488 ovs_rwlock_unlock(&ofproto->ml->rwlock);
4492 unixctl_command_reply(conn, "table successfully flushed");
4496 ofproto_unixctl_mcast_snooping_flush(struct unixctl_conn *conn, int argc,
4497 const char *argv[], void *aux OVS_UNUSED)
4499 struct ofproto_dpif *ofproto;
4502 ofproto = ofproto_dpif_lookup(argv[1]);
4504 unixctl_command_reply_error(conn, "no such bridge");
4508 if (!mcast_snooping_enabled(ofproto->ms)) {
4509 unixctl_command_reply_error(conn, "multicast snooping is disabled");
4512 mcast_snooping_mdb_flush(ofproto->ms);
4514 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4515 if (!mcast_snooping_enabled(ofproto->ms)) {
4518 mcast_snooping_mdb_flush(ofproto->ms);
4522 unixctl_command_reply(conn, "table successfully flushed");
4525 static struct ofport_dpif *
4526 ofbundle_get_a_port(const struct ofbundle *bundle)
4528 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
4533 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4534 const char *argv[], void *aux OVS_UNUSED)
4536 struct ds ds = DS_EMPTY_INITIALIZER;
4537 const struct ofproto_dpif *ofproto;
4538 const struct mac_entry *e;
4540 ofproto = ofproto_dpif_lookup(argv[1]);
4542 unixctl_command_reply_error(conn, "no such bridge");
4546 ds_put_cstr(&ds, " port VLAN MAC Age\n");
4547 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
4548 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
4549 struct ofbundle *bundle = mac_entry_get_port(ofproto->ml, e);
4550 char name[OFP_MAX_PORT_NAME_LEN];
4552 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4554 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
4555 name, e->vlan, ETH_ADDR_ARGS(e->mac),
4556 mac_entry_age(ofproto->ml, e));
4558 ovs_rwlock_unlock(&ofproto->ml->rwlock);
4559 unixctl_command_reply(conn, ds_cstr(&ds));
4564 ofproto_unixctl_mcast_snooping_show(struct unixctl_conn *conn,
4565 int argc OVS_UNUSED,
4567 void *aux OVS_UNUSED)
4569 struct ds ds = DS_EMPTY_INITIALIZER;
4570 const struct ofproto_dpif *ofproto;
4571 const struct ofbundle *bundle;
4572 const struct mcast_group *grp;
4573 struct mcast_group_bundle *b;
4574 struct mcast_mrouter_bundle *mrouter;
4576 ofproto = ofproto_dpif_lookup(argv[1]);
4578 unixctl_command_reply_error(conn, "no such bridge");
4582 if (!mcast_snooping_enabled(ofproto->ms)) {
4583 unixctl_command_reply_error(conn, "multicast snooping is disabled");
4587 ds_put_cstr(&ds, " port VLAN GROUP Age\n");
4588 ovs_rwlock_rdlock(&ofproto->ms->rwlock);
4589 LIST_FOR_EACH (grp, group_node, &ofproto->ms->group_lru) {
4590 LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
4591 char name[OFP_MAX_PORT_NAME_LEN];
4594 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4596 ds_put_format(&ds, "%5s %4d ", name, grp->vlan);
4597 ipv6_format_mapped(&grp->addr, &ds);
4598 ds_put_format(&ds, " %3d\n",
4599 mcast_bundle_age(ofproto->ms, b));
4603 /* ports connected to multicast routers */
4604 LIST_FOR_EACH(mrouter, mrouter_node, &ofproto->ms->mrouter_lru) {
4605 char name[OFP_MAX_PORT_NAME_LEN];
4607 bundle = mrouter->port;
4608 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4610 ds_put_format(&ds, "%5s %4d querier %3d\n",
4611 name, mrouter->vlan,
4612 mcast_mrouter_age(ofproto->ms, mrouter));
4614 ovs_rwlock_unlock(&ofproto->ms->rwlock);
4615 unixctl_command_reply(conn, ds_cstr(&ds));
4620 struct xlate_out xout;
4621 struct xlate_in xin;
4622 const struct flow *key;
4625 struct flow_wildcards wc;
4626 struct ofpbuf odp_actions;
4630 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
4632 const struct rule_actions *actions;
4635 ds_put_char_multiple(result, '\t', level);
4637 ds_put_cstr(result, "No match\n");
4641 ovs_mutex_lock(&rule->up.mutex);
4642 cookie = rule->up.flow_cookie;
4643 ovs_mutex_unlock(&rule->up.mutex);
4645 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
4646 rule ? rule->up.table_id : 0, ntohll(cookie));
4647 cls_rule_format(&rule->up.cr, result);
4648 ds_put_char(result, '\n');
4650 actions = rule_dpif_get_actions(rule);
4652 ds_put_char_multiple(result, '\t', level);
4653 ds_put_cstr(result, "OpenFlow actions=");
4654 ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
4655 ds_put_char(result, '\n');
4659 trace_format_flow(struct ds *result, int level, const char *title,
4660 struct trace_ctx *trace)
4662 ds_put_char_multiple(result, '\t', level);
4663 ds_put_format(result, "%s: ", title);
4664 /* Do not report unchanged flows for resubmits. */
4665 if ((level > 0 && flow_equal(&trace->xin.flow, &trace->flow))
4666 || (level == 0 && flow_equal(&trace->xin.flow, trace->key))) {
4667 ds_put_cstr(result, "unchanged");
4669 flow_format(result, &trace->xin.flow);
4670 trace->flow = trace->xin.flow;
4672 ds_put_char(result, '\n');
4676 trace_format_regs(struct ds *result, int level, const char *title,
4677 struct trace_ctx *trace)
4681 ds_put_char_multiple(result, '\t', level);
4682 ds_put_format(result, "%s:", title);
4683 for (i = 0; i < FLOW_N_REGS; i++) {
4684 ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
4686 ds_put_char(result, '\n');
4690 trace_format_odp(struct ds *result, int level, const char *title,
4691 struct trace_ctx *trace)
4693 struct ofpbuf *odp_actions = &trace->odp_actions;
4695 ds_put_char_multiple(result, '\t', level);
4696 ds_put_format(result, "%s: ", title);
4697 format_odp_actions(result, odp_actions->data, odp_actions->size);
4698 ds_put_char(result, '\n');
4702 trace_format_megaflow(struct ds *result, int level, const char *title,
4703 struct trace_ctx *trace)
4707 ds_put_char_multiple(result, '\t', level);
4708 ds_put_format(result, "%s: ", title);
4709 match_init(&match, trace->key, &trace->wc);
4710 match_format(&match, result, OFP_DEFAULT_PRIORITY);
4711 ds_put_char(result, '\n');
4714 static void trace_report(struct xlate_in *, int recurse,
4715 const char *format, ...)
4716 OVS_PRINTF_FORMAT(3, 4);
4717 static void trace_report_valist(struct xlate_in *, int recurse,
4718 const char *format, va_list args)
4719 OVS_PRINTF_FORMAT(3, 0);
4722 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
4724 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4725 struct ds *result = trace->result;
4728 if (rule == xin->ofproto->miss_rule) {
4729 trace_report(xin, recurse,
4730 "No match, flow generates \"packet in\"s.");
4731 } else if (rule == xin->ofproto->no_packet_in_rule) {
4732 trace_report(xin, recurse, "No match, packets dropped because "
4733 "OFPPC_NO_PACKET_IN is set on in_port.");
4734 } else if (rule == xin->ofproto->drop_frags_rule) {
4735 trace_report(xin, recurse, "Packets dropped because they are IP "
4736 "fragments and the fragment handling mode is "
4741 ds_put_char(result, '\n');
4743 trace_format_flow(result, recurse, "Resubmitted flow", trace);
4744 trace_format_regs(result, recurse, "Resubmitted regs", trace);
4745 trace_format_odp(result, recurse, "Resubmitted odp", trace);
4746 trace_format_megaflow(result, recurse, "Resubmitted megaflow", trace);
4748 trace_format_rule(result, recurse, rule);
4752 trace_report_valist(struct xlate_in *xin, int recurse,
4753 const char *format, va_list args)
4755 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4756 struct ds *result = trace->result;
4758 ds_put_char_multiple(result, '\t', recurse);
4759 ds_put_format_valist(result, format, args);
4760 ds_put_char(result, '\n');
4764 trace_report(struct xlate_in *xin, int recurse, const char *format, ...)
4768 va_start(args, format);
4769 trace_report_valist(xin, recurse, format, args);
4773 /* Parses the 'argc' elements of 'argv', ignoring argv[0]. The following
4774 * forms are supported:
4776 * - [dpname] odp_flow [-generate | packet]
4777 * - bridge br_flow [-generate | packet]
4779 * On success, initializes '*ofprotop' and 'flow' and returns NULL. On failure
4780 * returns a nonnull malloced error message. */
4781 static char * OVS_WARN_UNUSED_RESULT
4782 parse_flow_and_packet(int argc, const char *argv[],
4783 struct ofproto_dpif **ofprotop, struct flow *flow,
4784 struct dp_packet **packetp)
4786 const struct dpif_backer *backer = NULL;
4787 const char *error = NULL;
4789 struct simap port_names = SIMAP_INITIALIZER(&port_names);
4790 struct dp_packet *packet;
4791 struct ofpbuf odp_key;
4792 struct ofpbuf odp_mask;
4794 ofpbuf_init(&odp_key, 0);
4795 ofpbuf_init(&odp_mask, 0);
4797 /* Handle "-generate" or a hex string as the last argument. */
4798 if (!strcmp(argv[argc - 1], "-generate")) {
4799 packet = dp_packet_new(0);
4802 error = eth_from_hex(argv[argc - 1], &packet);
4805 } else if (argc == 4) {
4806 /* The 3-argument form must end in "-generate' or a hex string. */
4812 /* odp_flow can have its in_port specified as a name instead of port no.
4813 * We do not yet know whether a given flow is a odp_flow or a br_flow.
4814 * But, to know whether a flow is odp_flow through odp_flow_from_string(),
4815 * we need to create a simap of name to port no. */
4817 const char *dp_type;
4818 if (!strncmp(argv[1], "ovs-", 4)) {
4819 dp_type = argv[1] + 4;
4823 backer = shash_find_data(&all_dpif_backers, dp_type);
4824 } else if (argc == 2) {
4825 struct shash_node *node;
4826 if (shash_count(&all_dpif_backers) == 1) {
4827 node = shash_first(&all_dpif_backers);
4828 backer = node->data;
4831 error = "Syntax error";
4834 if (backer && backer->dpif) {
4835 struct dpif_port dpif_port;
4836 struct dpif_port_dump port_dump;
4837 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
4838 simap_put(&port_names, dpif_port.name,
4839 odp_to_u32(dpif_port.port_no));
4843 /* Parse the flow and determine whether a datapath or
4844 * bridge is specified. If function odp_flow_key_from_string()
4845 * returns 0, the flow is a odp_flow. If function
4846 * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
4847 if (!odp_flow_from_string(argv[argc - 1], &port_names,
4848 &odp_key, &odp_mask)) {
4850 error = "Cannot find the datapath";
4854 if (odp_flow_key_to_flow(odp_key.data, odp_key.size, flow) == ODP_FIT_ERROR) {
4855 error = "Failed to parse datapath flow key";
4859 *ofprotop = xlate_lookup_ofproto(backer, flow,
4860 &flow->in_port.ofp_port);
4861 if (*ofprotop == NULL) {
4862 error = "Invalid datapath flow";
4866 vsp_adjust_flow(*ofprotop, flow, NULL);
4869 char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
4872 m_err = xasprintf("Bad openflow flow syntax: %s", err);
4877 error = "Must specify bridge name";
4881 *ofprotop = ofproto_dpif_lookup(argv[1]);
4883 error = "Unknown bridge name";
4889 /* Generate a packet, if requested. */
4891 if (!dp_packet_size(packet)) {
4892 flow_compose(packet, flow);
4894 /* Use the metadata from the flow and the packet argument
4895 * to reconstruct the flow. */
4896 pkt_metadata_from_flow(&packet->md, flow);
4897 flow_extract(packet, flow);
4902 if (error && !m_err) {
4903 m_err = xstrdup(error);
4906 dp_packet_delete(packet);
4910 ofpbuf_uninit(&odp_key);
4911 ofpbuf_uninit(&odp_mask);
4912 simap_destroy(&port_names);
4917 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
4918 void *aux OVS_UNUSED)
4920 struct ofproto_dpif *ofproto;
4921 struct dp_packet *packet;
4925 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4930 ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
4931 unixctl_command_reply(conn, ds_cstr(&result));
4932 ds_destroy(&result);
4933 dp_packet_delete(packet);
4935 unixctl_command_reply_error(conn, error);
4941 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
4942 const char *argv[], void *aux OVS_UNUSED)
4944 enum ofputil_protocol usable_protocols;
4945 struct ofproto_dpif *ofproto;
4946 bool enforce_consistency;
4947 struct ofpbuf ofpacts;
4948 struct dp_packet *packet;
4953 /* Three kinds of error return values! */
4959 ofpbuf_init(&ofpacts, 0);
4961 /* Parse actions. */
4962 error = ofpacts_parse_actions(argv[--argc], &ofpacts, &usable_protocols);
4964 unixctl_command_reply_error(conn, error);
4969 /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
4970 * consistency between the flow and the actions. With -consistent, we
4971 * enforce consistency even for a flow supported in OpenFlow 1.0. */
4972 if (!strcmp(argv[1], "-consistent")) {
4973 enforce_consistency = true;
4977 enforce_consistency = false;
4980 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4982 unixctl_command_reply_error(conn, error);
4987 /* Do the same checks as handle_packet_out() in ofproto.c.
4989 * We pass a 'table_id' of 0 to ofpacts_check(), which isn't
4990 * strictly correct because these actions aren't in any table, but it's OK
4991 * because it 'table_id' is used only to check goto_table instructions, but
4992 * packet-outs take a list of actions and therefore it can't include
4995 * We skip the "meter" check here because meter is an instruction, not an
4996 * action, and thus cannot appear in ofpacts. */
4997 in_port = ofp_to_u16(flow.in_port.ofp_port);
4998 if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
4999 unixctl_command_reply_error(conn, "invalid in_port");
5002 if (enforce_consistency) {
5003 retval = ofpacts_check_consistency(ofpacts.data, ofpacts.size, &flow,
5004 u16_to_ofp(ofproto->up.max_ports),
5005 0, ofproto->up.n_tables,
5008 retval = ofpacts_check(ofpacts.data, ofpacts.size, &flow,
5009 u16_to_ofp(ofproto->up.max_ports), 0,
5010 ofproto->up.n_tables, &usable_protocols);
5013 retval = ofproto_check_ofpacts(&ofproto->up, ofpacts.data,
5019 ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
5020 unixctl_command_reply_error(conn, ds_cstr(&result));
5024 ofproto_trace(ofproto, &flow, packet,
5025 ofpacts.data, ofpacts.size, &result);
5026 unixctl_command_reply(conn, ds_cstr(&result));
5029 ds_destroy(&result);
5030 dp_packet_delete(packet);
5031 ofpbuf_uninit(&ofpacts);
5034 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
5035 * description of the results to 'ds'.
5037 * The trace follows a packet with the specified 'flow' through the flow
5038 * table. 'packet' may be nonnull to trace an actual packet, with consequent
5039 * side effects (if it is nonnull then its flow must be 'flow').
5041 * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
5042 * trace, otherwise the actions are determined by a flow table lookup. */
5044 ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
5045 const struct dp_packet *packet,
5046 const struct ofpact ofpacts[], size_t ofpacts_len,
5049 struct trace_ctx trace;
5050 enum xlate_error error;
5052 ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
5053 ds_put_cstr(ds, "Flow: ");
5054 flow_format(ds, flow);
5055 ds_put_char(ds, '\n');
5057 ofpbuf_init(&trace.odp_actions, 0);
5060 trace.key = flow; /* Original flow key, used for megaflow. */
5061 trace.flow = *flow; /* May be modified by actions. */
5062 xlate_in_init(&trace.xin, ofproto, flow, flow->in_port.ofp_port, NULL,
5063 ntohs(flow->tcp_flags), packet, &trace.wc,
5064 &trace.odp_actions);
5065 trace.xin.ofpacts = ofpacts;
5066 trace.xin.ofpacts_len = ofpacts_len;
5067 trace.xin.resubmit_hook = trace_resubmit;
5068 trace.xin.report_hook = trace_report_valist;
5070 error = xlate_actions(&trace.xin, &trace.xout);
5071 ds_put_char(ds, '\n');
5072 trace_format_flow(ds, 0, "Final flow", &trace);
5073 trace_format_megaflow(ds, 0, "Megaflow", &trace);
5075 ds_put_cstr(ds, "Datapath actions: ");
5076 format_odp_actions(ds, trace.odp_actions.data, trace.odp_actions.size);
5078 if (error != XLATE_OK) {
5079 ds_put_format(ds, "\nTranslation failed (%s), packet is dropped.\n",
5080 xlate_strerror(error));
5081 } else if (trace.xout.slow) {
5082 enum slow_path_reason slow;
5084 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5085 "slow path because it:");
5087 slow = trace.xout.slow;
5089 enum slow_path_reason bit = rightmost_1bit(slow);
5091 ds_put_format(ds, "\n\t- %s.",
5092 slow_path_reason_to_explanation(bit));
5098 xlate_out_uninit(&trace.xout);
5099 ofpbuf_uninit(&trace.odp_actions);
5102 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5103 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5104 * to destroy 'ofproto_shash' and free the returned value. */
5105 static const struct shash_node **
5106 get_ofprotos(struct shash *ofproto_shash)
5108 const struct ofproto_dpif *ofproto;
5110 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5111 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5112 shash_add_nocopy(ofproto_shash, name, ofproto);
5115 return shash_sort(ofproto_shash);
5119 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5120 const char *argv[] OVS_UNUSED,
5121 void *aux OVS_UNUSED)
5123 struct ds ds = DS_EMPTY_INITIALIZER;
5124 struct shash ofproto_shash;
5125 const struct shash_node **sorted_ofprotos;
5128 shash_init(&ofproto_shash);
5129 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5130 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5131 const struct shash_node *node = sorted_ofprotos[i];
5132 ds_put_format(&ds, "%s\n", node->name);
5135 shash_destroy(&ofproto_shash);
5136 free(sorted_ofprotos);
5138 unixctl_command_reply(conn, ds_cstr(&ds));
5143 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5145 const struct shash_node **ofprotos;
5146 struct dpif_dp_stats dp_stats;
5147 struct shash ofproto_shash;
5150 dpif_get_dp_stats(backer->dpif, &dp_stats);
5152 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5153 dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
5155 shash_init(&ofproto_shash);
5156 ofprotos = get_ofprotos(&ofproto_shash);
5157 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5158 struct ofproto_dpif *ofproto = ofprotos[i]->data;
5159 const struct shash_node **ports;
5162 if (ofproto->backer != backer) {
5166 ds_put_format(ds, "\t%s:\n", ofproto->up.name);
5168 ports = shash_sort(&ofproto->up.port_by_name);
5169 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5170 const struct shash_node *node = ports[j];
5171 struct ofport *ofport = node->data;
5173 odp_port_t odp_port;
5175 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5178 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5179 if (odp_port != ODPP_NONE) {
5180 ds_put_format(ds, "%"PRIu32":", odp_port);
5182 ds_put_cstr(ds, "none:");
5185 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5188 if (!netdev_get_config(ofport->netdev, &config)) {
5189 const struct smap_node **nodes;
5192 nodes = smap_sort(&config);
5193 for (i = 0; i < smap_count(&config); i++) {
5194 const struct smap_node *node = nodes[i];
5195 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5196 node->key, node->value);
5200 smap_destroy(&config);
5202 ds_put_char(ds, ')');
5203 ds_put_char(ds, '\n');
5207 shash_destroy(&ofproto_shash);
5212 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5213 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5215 struct ds ds = DS_EMPTY_INITIALIZER;
5216 const struct shash_node **backers;
5219 backers = shash_sort(&all_dpif_backers);
5220 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5221 dpif_show_backer(backers[i]->data, &ds);
5225 unixctl_command_reply(conn, ds_cstr(&ds));
5230 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5231 int argc OVS_UNUSED, const char *argv[],
5232 void *aux OVS_UNUSED)
5234 const struct ofproto_dpif *ofproto;
5236 struct ds ds = DS_EMPTY_INITIALIZER;
5237 bool verbosity = false;
5239 struct dpif_port dpif_port;
5240 struct dpif_port_dump port_dump;
5241 struct hmap portno_names;
5243 struct dpif_flow_dump *flow_dump;
5244 struct dpif_flow_dump_thread *flow_dump_thread;
5248 ofproto = ofproto_dpif_lookup(argv[argc - 1]);
5250 unixctl_command_reply_error(conn, "no such bridge");
5254 if (argc > 2 && !strcmp(argv[1], "-m")) {
5258 hmap_init(&portno_names);
5259 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
5260 odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
5264 flow_dump = dpif_flow_dump_create(ofproto->backer->dpif, false);
5265 flow_dump_thread = dpif_flow_dump_thread_create(flow_dump);
5266 while (dpif_flow_dump_next(flow_dump_thread, &f, 1)) {
5269 if (odp_flow_key_to_flow(f.key, f.key_len, &flow) == ODP_FIT_ERROR
5270 || xlate_lookup_ofproto(ofproto->backer, &flow, NULL) != ofproto) {
5275 odp_format_ufid(&f.ufid, &ds);
5276 ds_put_cstr(&ds, " ");
5278 odp_flow_format(f.key, f.key_len, f.mask, f.mask_len,
5279 &portno_names, &ds, verbosity);
5280 ds_put_cstr(&ds, ", ");
5281 dpif_flow_stats_format(&f.stats, &ds);
5282 ds_put_cstr(&ds, ", actions:");
5283 format_odp_actions(&ds, f.actions, f.actions_len);
5284 ds_put_char(&ds, '\n');
5286 dpif_flow_dump_thread_destroy(flow_dump_thread);
5287 error = dpif_flow_dump_destroy(flow_dump);
5291 ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
5292 unixctl_command_reply_error(conn, ds_cstr(&ds));
5294 unixctl_command_reply(conn, ds_cstr(&ds));
5296 odp_portno_names_destroy(&portno_names);
5297 hmap_destroy(&portno_names);
5302 ofproto_revalidate_all_backers(void)
5304 const struct shash_node **backers;
5307 backers = shash_sort(&all_dpif_backers);
5308 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5309 struct dpif_backer *backer = backers[i]->data;
5310 backer->need_revalidate = REV_RECONFIGURE;
5316 disable_tnl_push_pop(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5317 const char *argv[], void *aux OVS_UNUSED)
5319 if (!strcasecmp(argv[1], "off")) {
5320 ofproto_use_tnl_push_pop = false;
5321 unixctl_command_reply(conn, "Tunnel push-pop off");
5322 ofproto_revalidate_all_backers();
5323 } else if (!strcasecmp(argv[1], "on")) {
5324 ofproto_use_tnl_push_pop = true;
5325 unixctl_command_reply(conn, "Tunnel push-pop on");
5326 ofproto_revalidate_all_backers();
5328 unixctl_command_reply_error(conn, "Invalid argument");
5333 ofproto_unixctl_init(void)
5335 static bool registered;
5341 unixctl_command_register(
5343 "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
5344 1, 3, ofproto_unixctl_trace, NULL);
5345 unixctl_command_register(
5346 "ofproto/trace-packet-out",
5347 "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
5348 2, 6, ofproto_unixctl_trace_actions, NULL);
5349 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5350 ofproto_unixctl_fdb_flush, NULL);
5351 unixctl_command_register("fdb/show", "bridge", 1, 1,
5352 ofproto_unixctl_fdb_show, NULL);
5353 unixctl_command_register("mdb/flush", "[bridge]", 0, 1,
5354 ofproto_unixctl_mcast_snooping_flush, NULL);
5355 unixctl_command_register("mdb/show", "bridge", 1, 1,
5356 ofproto_unixctl_mcast_snooping_show, NULL);
5357 unixctl_command_register("dpif/dump-dps", "", 0, 0,
5358 ofproto_unixctl_dpif_dump_dps, NULL);
5359 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5361 unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
5362 ofproto_unixctl_dpif_dump_flows, NULL);
5364 unixctl_command_register("ofproto/tnl-push-pop", "[on]|[off]", 1, 1,
5365 disable_tnl_push_pop, NULL);
5368 /* Returns true if 'table' is the table used for internal rules,
5369 * false otherwise. */
5371 table_is_internal(uint8_t table_id)
5373 return table_id == TBL_INTERNAL;
5376 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5378 * This is deprecated. It is only for compatibility with broken device drivers
5379 * in old versions of Linux that do not properly support VLANs when VLAN
5380 * devices are not used. When broken device drivers are no longer in
5381 * widespread use, we will delete these interfaces. */
5384 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
5386 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5387 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5389 if (realdev_ofp_port == ofport->realdev_ofp_port
5390 && vid == ofport->vlandev_vid) {
5394 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5396 if (ofport->realdev_ofp_port) {
5399 if (realdev_ofp_port && ofport->bundle) {
5400 /* vlandevs are enslaved to their realdevs, so they are not allowed to
5401 * themselves be part of a bundle. */
5402 bundle_set(ofport_->ofproto, ofport->bundle, NULL);
5405 ofport->realdev_ofp_port = realdev_ofp_port;
5406 ofport->vlandev_vid = vid;
5408 if (realdev_ofp_port) {
5409 vsp_add(ofport, realdev_ofp_port, vid);
5416 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
5418 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
5422 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
5423 OVS_EXCLUDED(ofproto->vsp_mutex)
5425 /* hmap_is_empty is thread safe. */
5426 return !hmap_is_empty(&ofproto->realdev_vid_map);
5431 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
5432 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5433 OVS_REQUIRES(ofproto->vsp_mutex)
5435 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5436 int vid = vlan_tci_to_vid(vlan_tci);
5437 const struct vlan_splinter *vsp;
5439 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5440 hash_realdev_vid(realdev_ofp_port, vid),
5441 &ofproto->realdev_vid_map) {
5442 if (vsp->realdev_ofp_port == realdev_ofp_port
5443 && vsp->vid == vid) {
5444 return vsp->vlandev_ofp_port;
5448 return realdev_ofp_port;
5451 /* Returns the OFP port number of the Linux VLAN device that corresponds to
5452 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
5453 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
5454 * 'vlan_tci' 9, it would return the port number of eth0.9.
5456 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
5457 * function just returns its 'realdev_ofp_port' argument. */
5459 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5460 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5461 OVS_EXCLUDED(ofproto->vsp_mutex)
5465 /* hmap_is_empty is thread safe, see if we can return immediately. */
5466 if (hmap_is_empty(&ofproto->realdev_vid_map)) {
5467 return realdev_ofp_port;
5469 ovs_mutex_lock(&ofproto->vsp_mutex);
5470 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
5471 ovs_mutex_unlock(&ofproto->vsp_mutex);
5475 static struct vlan_splinter *
5476 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
5478 struct vlan_splinter *vsp;
5480 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
5481 hash_ofp_port(vlandev_ofp_port),
5482 &ofproto->vlandev_map) {
5483 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5491 /* Returns the OpenFlow port number of the "real" device underlying the Linux
5492 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
5493 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
5494 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
5495 * eth0 and store 9 in '*vid'.
5497 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
5498 * VLAN device. Unless VLAN splinters are enabled, this is what this function
5501 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5502 ofp_port_t vlandev_ofp_port, int *vid)
5503 OVS_REQUIRES(ofproto->vsp_mutex)
5505 if (!hmap_is_empty(&ofproto->vlandev_map)) {
5506 const struct vlan_splinter *vsp;
5508 vsp = vlandev_find(ofproto, vlandev_ofp_port);
5513 return vsp->realdev_ofp_port;
5519 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
5520 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
5521 * 'flow->in_port' to the "real" device backing the VLAN device, sets
5522 * 'flow->vlan_tci' to the VLAN VID, and returns true. Optionally pushes the
5523 * appropriate VLAN on 'packet' if provided. Otherwise (which is always the
5524 * case unless VLAN splinters are enabled), returns false without making any
5527 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow,
5528 struct dp_packet *packet)
5529 OVS_EXCLUDED(ofproto->vsp_mutex)
5534 /* hmap_is_empty is thread safe. */
5535 if (hmap_is_empty(&ofproto->vlandev_map)) {
5539 ovs_mutex_lock(&ofproto->vsp_mutex);
5540 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
5541 ovs_mutex_unlock(&ofproto->vsp_mutex);
5546 /* Cause the flow to be processed as if it came in on the real device with
5547 * the VLAN device's VLAN ID. */
5548 flow->in_port.ofp_port = realdev;
5549 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
5552 /* Make the packet resemble the flow, so that it gets sent to an
5553 * OpenFlow controller properly, so that it looks correct for sFlow,
5554 * and so that flow_extract() will get the correct vlan_tci if it is
5555 * called on 'packet'. */
5556 eth_push_vlan(packet, htons(ETH_TYPE_VLAN), flow->vlan_tci);
5563 vsp_remove(struct ofport_dpif *port)
5565 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5566 struct vlan_splinter *vsp;
5568 ovs_mutex_lock(&ofproto->vsp_mutex);
5569 vsp = vlandev_find(ofproto, port->up.ofp_port);
5571 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
5572 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
5575 port->realdev_ofp_port = 0;
5577 VLOG_ERR("missing vlan device record");
5579 ovs_mutex_unlock(&ofproto->vsp_mutex);
5583 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
5585 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5587 ovs_mutex_lock(&ofproto->vsp_mutex);
5588 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
5589 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
5590 == realdev_ofp_port)) {
5591 struct vlan_splinter *vsp;
5593 vsp = xmalloc(sizeof *vsp);
5594 vsp->realdev_ofp_port = realdev_ofp_port;
5595 vsp->vlandev_ofp_port = port->up.ofp_port;
5598 port->realdev_ofp_port = realdev_ofp_port;
5600 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
5601 hash_ofp_port(port->up.ofp_port));
5602 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
5603 hash_realdev_vid(realdev_ofp_port, vid));
5605 VLOG_ERR("duplicate vlan device record");
5607 ovs_mutex_unlock(&ofproto->vsp_mutex);
5611 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
5613 const struct ofport_dpif *ofport = ofp_port_to_ofport(ofproto, ofp_port);
5614 return ofport ? ofport->odp_port : ODPP_NONE;
5617 struct ofport_dpif *
5618 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
5620 struct ofport_dpif *port;
5622 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
5623 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
5624 &backer->odp_to_ofport_map) {
5625 if (port->odp_port == odp_port) {
5626 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5631 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5636 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
5638 struct ofport_dpif *port;
5640 port = odp_port_to_ofport(ofproto->backer, odp_port);
5641 if (port && &ofproto->up == port->up.ofproto) {
5642 return port->up.ofp_port;
5649 ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
5650 const struct match *match, int priority,
5651 uint16_t idle_timeout,
5652 const struct ofpbuf *ofpacts,
5653 struct rule **rulep)
5655 struct ofproto_flow_mod ofm;
5656 struct rule_dpif *rule;
5659 ofm.fm = (struct ofputil_flow_mod) {
5661 .priority = priority,
5662 .table_id = TBL_INTERNAL,
5663 .command = OFPFC_ADD,
5664 .idle_timeout = idle_timeout,
5665 .flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY,
5666 .ofpacts = ofpacts->data,
5667 .ofpacts_len = ofpacts->size,
5668 .delete_reason = OVS_OFPRR_NONE,
5671 error = ofproto_flow_mod(&ofproto->up, &ofm);
5673 VLOG_ERR_RL(&rl, "failed to add internal flow (%s)",
5674 ofperr_to_string(error));
5679 rule = rule_dpif_lookup_in_table(ofproto,
5680 ofproto_dpif_get_tables_version(ofproto),
5681 TBL_INTERNAL, &ofm.fm.match.flow,
5692 ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
5693 struct match *match, int priority)
5695 struct ofproto_flow_mod ofm;
5698 ofm.fm = (struct ofputil_flow_mod) {
5700 .priority = priority,
5701 .table_id = TBL_INTERNAL,
5702 .flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY,
5703 .command = OFPFC_DELETE_STRICT,
5706 error = ofproto_flow_mod(&ofproto->up, &ofm);
5708 VLOG_ERR_RL(&rl, "failed to delete internal flow (%s)",
5709 ofperr_to_string(error));
5716 const struct ofproto_class ofproto_dpif_class = {
5730 NULL, /* get_memory_usage. */
5731 type_get_memory_usage,
5750 port_is_lacp_current,
5751 port_get_lacp_stats,
5752 NULL, /* rule_choose_table */
5776 aa_vlan_get_queue_size,
5783 get_stp_port_status,
5788 get_rstp_port_status,
5795 is_mirror_output_bundle,
5796 forward_bpdu_changed,
5797 set_mac_table_config,
5799 set_mcast_snooping_port,
5801 NULL, /* meter_get_features */
5802 NULL, /* meter_set */
5803 NULL, /* meter_get */
5804 NULL, /* meter_del */
5805 group_alloc, /* group_alloc */
5806 group_construct, /* group_construct */
5807 group_destruct, /* group_destruct */
5808 group_dealloc, /* group_dealloc */
5809 group_modify, /* group_modify */
5810 group_get_stats, /* group_get_stats */
5811 get_datapath_version, /* get_datapath_version */