2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-dpif.h"
20 #include "ofproto/ofproto-provider.h"
27 #include "byte-order.h"
28 #include "connectivity.h"
34 #include "dynamic-string.h"
35 #include "fail-open.h"
36 #include "guarded-list.h"
40 #include "mac-learning.h"
41 #include "mcast-snooping.h"
42 #include "meta-flow.h"
43 #include "multipath.h"
44 #include "netdev-vport.h"
49 #include "odp-execute.h"
52 #include "ofp-actions.h"
53 #include "ofp-parse.h"
54 #include "ofp-print.h"
55 #include "ofproto-dpif-ipfix.h"
56 #include "ofproto-dpif-mirror.h"
57 #include "ofproto-dpif-monitor.h"
58 #include "ofproto-dpif-rid.h"
59 #include "ofproto-dpif-sflow.h"
60 #include "ofproto-dpif-upcall.h"
61 #include "ofproto-dpif-xlate.h"
62 #include "poll-loop.h"
64 #include "ovs-router.h"
70 #include "unaligned.h"
72 #include "vlan-bitmap.h"
73 #include "openvswitch/vlog.h"
75 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
77 COVERAGE_DEFINE(ofproto_dpif_expired);
78 COVERAGE_DEFINE(packet_in_overflow);
87 * - Do include packets and bytes from datapath flows which have not
88 * recently been processed by a revalidator. */
89 struct ovs_mutex stats_mutex;
90 struct dpif_flow_stats stats OVS_GUARDED;
92 /* In non-NULL, will point to a new rule (for which a reference is held) to
93 * which all the stats updates should be forwarded. This exists only
94 * transitionally when flows are replaced.
96 * Protected by stats_mutex. If both 'rule->stats_mutex' and
97 * 'rule->new_rule->stats_mutex' must be held together, acquire them in that
99 struct rule_dpif *new_rule OVS_GUARDED;
101 /* If non-zero then the recirculation id that has
102 * been allocated for use with this rule.
103 * The recirculation id and associated internal flow should
104 * be freed when the rule is freed */
108 /* RULE_CAST() depends on this. */
109 BUILD_ASSERT_DECL(offsetof(struct rule_dpif, up) == 0);
111 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes,
112 long long int *used);
113 static struct rule_dpif *rule_dpif_cast(const struct rule *);
114 static void rule_expire(struct rule_dpif *);
121 * - Do include packets and bytes from datapath flows which have not
122 * recently been processed by a revalidator. */
123 struct ovs_mutex stats_mutex;
124 uint64_t packet_count OVS_GUARDED; /* Number of packets received. */
125 uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */
129 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
130 struct ofproto_dpif *ofproto; /* Owning ofproto. */
131 void *aux; /* Key supplied by ofproto's client. */
132 char *name; /* Identifier for log messages. */
135 struct ovs_list ports; /* Contains "struct ofport"s. */
136 enum port_vlan_mode vlan_mode; /* VLAN mode */
137 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
138 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
139 * NULL if all VLANs are trunked. */
140 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
141 struct bond *bond; /* Nonnull iff more than one port. */
142 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
145 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
148 static void bundle_remove(struct ofport *);
149 static void bundle_update(struct ofbundle *);
150 static void bundle_destroy(struct ofbundle *);
151 static void bundle_del_port(struct ofport_dpif *);
152 static void bundle_run(struct ofbundle *);
153 static void bundle_wait(struct ofbundle *);
154 static void bundle_flush_macs(struct ofbundle *, bool);
155 static void bundle_move(struct ofbundle *, struct ofbundle *);
157 static void stp_run(struct ofproto_dpif *ofproto);
158 static void stp_wait(struct ofproto_dpif *ofproto);
159 static int set_stp_port(struct ofport *,
160 const struct ofproto_port_stp_settings *);
162 static void rstp_run(struct ofproto_dpif *ofproto);
163 static void set_rstp_port(struct ofport *,
164 const struct ofproto_port_rstp_settings *);
167 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
171 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
172 struct ovs_list bundle_node;/* In struct ofbundle's "ports" list. */
173 struct cfm *cfm; /* Connectivity Fault Management, if any. */
174 struct bfd *bfd; /* BFD, if any. */
175 struct lldp *lldp; /* lldp, if any. */
176 bool may_enable; /* May be enabled in bonds. */
177 bool is_tunnel; /* This port is a tunnel. */
178 bool is_layer3; /* This is a layer 3 port. */
179 long long int carrier_seq; /* Carrier status changes. */
180 struct ofport_dpif *peer; /* Peer if patch port. */
183 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
184 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
185 long long int stp_state_entered;
187 /* Rapid Spanning Tree. */
188 struct rstp_port *rstp_port; /* Rapid Spanning Tree Protocol, if any. */
189 enum rstp_state rstp_state; /* Always RSTP_DISABLED if RSTP not in use. */
191 /* Queue to DSCP mapping. */
192 struct ofproto_port_queue *qdscp;
195 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
197 * This is deprecated. It is only for compatibility with broken device
198 * drivers in old versions of Linux that do not properly support VLANs when
199 * VLAN devices are not used. When broken device drivers are no longer in
200 * widespread use, we will delete these interfaces. */
201 ofp_port_t realdev_ofp_port;
205 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
207 * This is deprecated. It is only for compatibility with broken device drivers
208 * in old versions of Linux that do not properly support VLANs when VLAN
209 * devices are not used. When broken device drivers are no longer in
210 * widespread use, we will delete these interfaces. */
211 struct vlan_splinter {
212 struct hmap_node realdev_vid_node;
213 struct hmap_node vlandev_node;
214 ofp_port_t realdev_ofp_port;
215 ofp_port_t vlandev_ofp_port;
219 static void vsp_remove(struct ofport_dpif *);
220 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
222 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
225 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
228 static struct ofport_dpif *
229 ofport_dpif_cast(const struct ofport *ofport)
231 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
234 static void port_run(struct ofport_dpif *);
235 static int set_bfd(struct ofport *, const struct smap *);
236 static int set_cfm(struct ofport *, const struct cfm_settings *);
237 static int set_lldp(struct ofport *ofport_, const struct smap *cfg);
238 static void ofport_update_peer(struct ofport_dpif *);
240 /* Reasons that we might need to revalidate every datapath flow, and
241 * corresponding coverage counters.
243 * A value of 0 means that there is no need to revalidate.
245 * It would be nice to have some cleaner way to integrate with coverage
246 * counters, but with only a few reasons I guess this is good enough for
248 enum revalidate_reason {
249 REV_RECONFIGURE = 1, /* Switch configuration changed. */
250 REV_STP, /* Spanning tree protocol port status change. */
251 REV_RSTP, /* RSTP port status change. */
252 REV_BOND, /* Bonding changed. */
253 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
254 REV_FLOW_TABLE, /* Flow table changed. */
255 REV_MAC_LEARNING, /* Mac learning changed. */
256 REV_MCAST_SNOOPING, /* Multicast snooping changed. */
258 COVERAGE_DEFINE(rev_reconfigure);
259 COVERAGE_DEFINE(rev_stp);
260 COVERAGE_DEFINE(rev_rstp);
261 COVERAGE_DEFINE(rev_bond);
262 COVERAGE_DEFINE(rev_port_toggled);
263 COVERAGE_DEFINE(rev_flow_table);
264 COVERAGE_DEFINE(rev_mac_learning);
265 COVERAGE_DEFINE(rev_mcast_snooping);
267 /* All datapaths of a given type share a single dpif backer instance. */
274 struct ovs_rwlock odp_to_ofport_lock;
275 struct hmap odp_to_ofport_map OVS_GUARDED; /* Contains "struct ofport"s. */
277 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
279 enum revalidate_reason need_revalidate; /* Revalidate all flows. */
281 bool recv_set_enable; /* Enables or disables receiving packets. */
283 /* Version string of the datapath stored in OVSDB. */
284 char *dp_version_string;
286 /* Datapath feature support. */
287 struct dpif_backer_support support;
288 struct atomic_count tnl_count;
291 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
292 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
294 struct ofproto_dpif {
295 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
297 struct dpif_backer *backer;
299 ATOMIC(cls_version_t) tables_version; /* For classifier lookups. */
301 uint64_t dump_seq; /* Last read of udpif_dump_seq(). */
303 /* Special OpenFlow rules. */
304 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
305 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
306 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
309 struct netflow *netflow;
310 struct dpif_sflow *sflow;
311 struct dpif_ipfix *ipfix;
312 struct hmap bundles; /* Contains "struct ofbundle"s. */
313 struct mac_learning *ml;
314 struct mcast_snooping *ms;
315 bool has_bonded_bundles;
317 struct mbridge *mbridge;
319 struct ovs_mutex stats_mutex;
320 struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
321 * consumed in userspace. */
325 long long int stp_last_tick;
327 /* Rapid Spanning Tree. */
329 long long int rstp_last_tick;
331 /* VLAN splinters. */
332 struct ovs_mutex vsp_mutex;
333 struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
334 struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */
337 struct sset ports; /* Set of standard port names. */
338 struct sset ghost_ports; /* Ports with no datapath port. */
339 struct sset port_poll_set; /* Queued names for port_poll() reply. */
340 int port_poll_errno; /* Last errno for port_poll() reply. */
341 uint64_t change_seq; /* Connectivity status changes. */
344 struct guarded_list pins; /* Contains "struct ofputil_packet_in"s. */
345 struct seq *pins_seq; /* For notifying 'pins' reception. */
349 /* All existing ofproto_dpif instances, indexed by ->up.name. */
350 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
352 static bool ofproto_use_tnl_push_pop = true;
353 static void ofproto_unixctl_init(void);
355 static inline struct ofproto_dpif *
356 ofproto_dpif_cast(const struct ofproto *ofproto)
358 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
359 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
363 ofproto_dpif_get_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 'pin' to the queue of "packet ins" to be sent to the controller.
404 * Takes ownership of 'pin' and pin->packet. */
406 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
407 struct ofproto_packet_in *pin)
409 if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
410 COVERAGE_INC(packet_in_overflow);
411 free(CONST_CAST(void *, pin->up.packet));
415 /* Wakes up main thread for packet-in I/O. */
416 seq_change(ofproto->pins_seq);
419 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
420 * packet rather than to send the packet to the controller.
422 * This function returns false to indicate that a packet_in message
423 * for a "table-miss" should be sent to at least one controller.
424 * False otherwise. */
426 ofproto_dpif_wants_packet_in_on_miss(struct ofproto_dpif *ofproto)
428 return connmgr_wants_packet_in_on_miss(ofproto->up.connmgr);
431 /* Factory functions. */
434 init(const struct shash *iface_hints)
436 struct shash_node *node;
438 /* Make a local copy, since we don't own 'iface_hints' elements. */
439 SHASH_FOR_EACH(node, iface_hints) {
440 const struct iface_hint *orig_hint = node->data;
441 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
443 new_hint->br_name = xstrdup(orig_hint->br_name);
444 new_hint->br_type = xstrdup(orig_hint->br_type);
445 new_hint->ofp_port = orig_hint->ofp_port;
447 shash_add(&init_ofp_ports, node->name, new_hint);
450 ofproto_unixctl_init();
455 enumerate_types(struct sset *types)
457 dp_enumerate_types(types);
461 enumerate_names(const char *type, struct sset *names)
463 struct ofproto_dpif *ofproto;
466 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
467 if (strcmp(type, ofproto->up.type)) {
470 sset_add(names, ofproto->up.name);
477 del(const char *type, const char *name)
482 error = dpif_open(name, type, &dpif);
484 error = dpif_delete(dpif);
491 port_open_type(const char *datapath_type, const char *port_type)
493 return dpif_port_open_type(datapath_type, port_type);
496 /* Type functions. */
498 static void process_dpif_port_changes(struct dpif_backer *);
499 static void process_dpif_all_ports_changed(struct dpif_backer *);
500 static void process_dpif_port_change(struct dpif_backer *,
501 const char *devname);
502 static void process_dpif_port_error(struct dpif_backer *, int error);
504 static struct ofproto_dpif *
505 lookup_ofproto_dpif_by_port_name(const char *name)
507 struct ofproto_dpif *ofproto;
509 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
510 if (sset_contains(&ofproto->ports, name)) {
519 ofproto_dpif_backer_enabled(struct dpif_backer* backer)
521 return backer->recv_set_enable;
525 type_run(const char *type)
527 struct dpif_backer *backer;
529 backer = shash_find_data(&all_dpif_backers, type);
531 /* This is not necessarily a problem, since backers are only
532 * created on demand. */
537 if (dpif_run(backer->dpif)) {
538 backer->need_revalidate = REV_RECONFIGURE;
541 udpif_run(backer->udpif);
543 /* If vswitchd started with other_config:flow_restore_wait set as "true",
544 * and the configuration has now changed to "false", enable receiving
545 * packets from the datapath. */
546 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
549 backer->recv_set_enable = true;
551 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
553 VLOG_ERR("Failed to enable receiving packets in dpif.");
556 dpif_flow_flush(backer->dpif);
557 backer->need_revalidate = REV_RECONFIGURE;
560 if (backer->recv_set_enable) {
561 udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
564 dpif_poll_threads_set(backer->dpif, n_dpdk_rxqs, pmd_cpu_mask);
566 if (backer->need_revalidate) {
567 struct ofproto_dpif *ofproto;
568 struct simap_node *node;
569 struct simap tmp_backers;
571 /* Handle tunnel garbage collection. */
572 simap_init(&tmp_backers);
573 simap_swap(&backer->tnl_backers, &tmp_backers);
575 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
576 struct ofport_dpif *iter;
578 if (backer != ofproto->backer) {
582 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
583 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
586 if (!iter->is_tunnel) {
590 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
591 namebuf, sizeof namebuf);
592 node = simap_find(&tmp_backers, dp_port);
594 simap_put(&backer->tnl_backers, dp_port, node->data);
595 simap_delete(&tmp_backers, node);
596 node = simap_find(&backer->tnl_backers, dp_port);
598 node = simap_find(&backer->tnl_backers, dp_port);
600 odp_port_t odp_port = ODPP_NONE;
602 if (!dpif_port_add(backer->dpif, iter->up.netdev,
604 simap_put(&backer->tnl_backers, dp_port,
605 odp_to_u32(odp_port));
606 node = simap_find(&backer->tnl_backers, dp_port);
611 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
612 if (tnl_port_reconfigure(iter, iter->up.netdev,
614 ovs_native_tunneling_is_on(ofproto), dp_port)) {
615 backer->need_revalidate = REV_RECONFIGURE;
620 SIMAP_FOR_EACH (node, &tmp_backers) {
621 dpif_port_del(backer->dpif, u32_to_odp(node->data));
623 simap_destroy(&tmp_backers);
625 switch (backer->need_revalidate) {
626 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
627 case REV_STP: COVERAGE_INC(rev_stp); break;
628 case REV_RSTP: COVERAGE_INC(rev_rstp); break;
629 case REV_BOND: COVERAGE_INC(rev_bond); break;
630 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
631 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
632 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
633 case REV_MCAST_SNOOPING: COVERAGE_INC(rev_mcast_snooping); break;
635 backer->need_revalidate = 0;
637 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
638 struct ofport_dpif *ofport;
639 struct ofbundle *bundle;
641 if (ofproto->backer != backer) {
646 xlate_ofproto_set(ofproto, ofproto->up.name,
647 ofproto->backer->dpif, ofproto->ml,
648 ofproto->stp, ofproto->rstp, ofproto->ms,
649 ofproto->mbridge, ofproto->sflow, ofproto->ipfix,
651 ofproto->up.forward_bpdu,
652 connmgr_has_in_band(ofproto->up.connmgr),
653 &ofproto->backer->support);
655 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
656 xlate_bundle_set(ofproto, bundle, bundle->name,
657 bundle->vlan_mode, bundle->vlan,
658 bundle->trunks, bundle->use_priority_tags,
659 bundle->bond, bundle->lacp,
663 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
664 int stp_port = ofport->stp_port
665 ? stp_port_no(ofport->stp_port)
667 xlate_ofport_set(ofproto, ofport->bundle, ofport,
668 ofport->up.ofp_port, ofport->odp_port,
669 ofport->up.netdev, ofport->cfm, ofport->bfd,
670 ofport->lldp, ofport->peer, stp_port,
671 ofport->rstp_port, ofport->qdscp,
672 ofport->n_qdscp, ofport->up.pp.config,
673 ofport->up.pp.state, ofport->is_tunnel,
679 udpif_revalidate(backer->udpif);
682 process_dpif_port_changes(backer);
687 /* Check for and handle port changes in 'backer''s dpif. */
689 process_dpif_port_changes(struct dpif_backer *backer)
695 error = dpif_port_poll(backer->dpif, &devname);
701 process_dpif_all_ports_changed(backer);
705 process_dpif_port_change(backer, devname);
710 process_dpif_port_error(backer, error);
717 process_dpif_all_ports_changed(struct dpif_backer *backer)
719 struct ofproto_dpif *ofproto;
720 struct dpif_port dpif_port;
721 struct dpif_port_dump dump;
722 struct sset devnames;
725 sset_init(&devnames);
726 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
727 if (ofproto->backer == backer) {
728 struct ofport *ofport;
730 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
731 sset_add(&devnames, netdev_get_name(ofport->netdev));
735 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
736 sset_add(&devnames, dpif_port.name);
739 SSET_FOR_EACH (devname, &devnames) {
740 process_dpif_port_change(backer, devname);
742 sset_destroy(&devnames);
746 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
748 struct ofproto_dpif *ofproto;
749 struct dpif_port port;
751 /* Don't report on the datapath's device. */
752 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
756 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
757 &all_ofproto_dpifs) {
758 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
763 ofproto = lookup_ofproto_dpif_by_port_name(devname);
764 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
765 /* The port was removed. If we know the datapath,
766 * report it through poll_set(). If we don't, it may be
767 * notifying us of a removal we initiated, so ignore it.
768 * If there's a pending ENOBUFS, let it stand, since
769 * everything will be reevaluated. */
770 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
771 sset_add(&ofproto->port_poll_set, devname);
772 ofproto->port_poll_errno = 0;
774 } else if (!ofproto) {
775 /* The port was added, but we don't know with which
776 * ofproto we should associate it. Delete it. */
777 dpif_port_del(backer->dpif, port.port_no);
779 struct ofport_dpif *ofport;
781 ofport = ofport_dpif_cast(shash_find_data(
782 &ofproto->up.port_by_name, devname));
784 && ofport->odp_port != port.port_no
785 && !odp_port_to_ofport(backer, port.port_no))
787 /* 'ofport''s datapath port number has changed from
788 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
789 * structures to match. */
790 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
791 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
792 ofport->odp_port = port.port_no;
793 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
794 hash_odp_port(port.port_no));
795 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
796 backer->need_revalidate = REV_RECONFIGURE;
799 dpif_port_destroy(&port);
802 /* Propagate 'error' to all ofprotos based on 'backer'. */
804 process_dpif_port_error(struct dpif_backer *backer, int error)
806 struct ofproto_dpif *ofproto;
808 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
809 if (ofproto->backer == backer) {
810 sset_clear(&ofproto->port_poll_set);
811 ofproto->port_poll_errno = error;
817 type_wait(const char *type)
819 struct dpif_backer *backer;
821 backer = shash_find_data(&all_dpif_backers, type);
823 /* This is not necessarily a problem, since backers are only
824 * created on demand. */
828 dpif_wait(backer->dpif);
831 /* Basic life-cycle. */
833 static int add_internal_flows(struct ofproto_dpif *);
835 static struct ofproto *
838 struct ofproto_dpif *ofproto = xzalloc(sizeof *ofproto);
843 dealloc(struct ofproto *ofproto_)
845 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
850 close_dpif_backer(struct dpif_backer *backer)
852 ovs_assert(backer->refcount > 0);
854 if (--backer->refcount) {
858 udpif_destroy(backer->udpif);
860 simap_destroy(&backer->tnl_backers);
861 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
862 hmap_destroy(&backer->odp_to_ofport_map);
863 shash_find_and_delete(&all_dpif_backers, backer->type);
865 free(backer->dp_version_string);
866 dpif_close(backer->dpif);
870 /* Datapath port slated for removal from datapath. */
872 struct ovs_list list_node;
876 static bool check_variable_length_userdata(struct dpif_backer *backer);
877 static void check_support(struct dpif_backer *backer);
880 open_dpif_backer(const char *type, struct dpif_backer **backerp)
882 struct dpif_backer *backer;
883 struct dpif_port_dump port_dump;
884 struct dpif_port port;
885 struct shash_node *node;
886 struct ovs_list garbage_list;
887 struct odp_garbage *garbage;
896 backer = shash_find_data(&all_dpif_backers, type);
903 backer_name = xasprintf("ovs-%s", type);
905 /* Remove any existing datapaths, since we assume we're the only
906 * userspace controlling the datapath. */
908 dp_enumerate_names(type, &names);
909 SSET_FOR_EACH(name, &names) {
910 struct dpif *old_dpif;
912 /* Don't remove our backer if it exists. */
913 if (!strcmp(name, backer_name)) {
917 if (dpif_open(name, type, &old_dpif)) {
918 VLOG_WARN("couldn't open old datapath %s to remove it", name);
920 dpif_delete(old_dpif);
921 dpif_close(old_dpif);
924 sset_destroy(&names);
926 backer = xmalloc(sizeof *backer);
928 error = dpif_create_and_open(backer_name, type, &backer->dpif);
931 VLOG_ERR("failed to open datapath of type %s: %s", type,
932 ovs_strerror(error));
936 backer->udpif = udpif_create(backer, backer->dpif);
938 backer->type = xstrdup(type);
939 backer->refcount = 1;
940 hmap_init(&backer->odp_to_ofport_map);
941 ovs_rwlock_init(&backer->odp_to_ofport_lock);
942 backer->need_revalidate = 0;
943 simap_init(&backer->tnl_backers);
944 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
947 if (backer->recv_set_enable) {
948 dpif_flow_flush(backer->dpif);
951 /* Loop through the ports already on the datapath and remove any
952 * that we don't need anymore. */
953 list_init(&garbage_list);
954 dpif_port_dump_start(&port_dump, backer->dpif);
955 while (dpif_port_dump_next(&port_dump, &port)) {
956 node = shash_find(&init_ofp_ports, port.name);
957 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
958 garbage = xmalloc(sizeof *garbage);
959 garbage->odp_port = port.port_no;
960 list_push_front(&garbage_list, &garbage->list_node);
963 dpif_port_dump_done(&port_dump);
965 LIST_FOR_EACH_POP (garbage, list_node, &garbage_list) {
966 dpif_port_del(backer->dpif, garbage->odp_port);
970 shash_add(&all_dpif_backers, type, backer);
972 check_support(backer);
973 atomic_count_init(&backer->tnl_count, 0);
975 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
977 VLOG_ERR("failed to listen on datapath of type %s: %s",
978 type, ovs_strerror(error));
979 close_dpif_backer(backer);
983 if (backer->recv_set_enable) {
984 udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
987 /* This check fails if performed before udpif threads have been set,
988 * as the kernel module checks that the 'pid' in userspace action
990 backer->support.variable_length_userdata
991 = check_variable_length_userdata(backer);
992 backer->dp_version_string = dpif_get_dp_version(backer->dpif);
998 ovs_native_tunneling_is_on(struct ofproto_dpif *ofproto)
1000 return ofproto_use_tnl_push_pop && ofproto->backer->support.tnl_push_pop &&
1001 atomic_count_get(&ofproto->backer->tnl_count);
1004 /* Tests whether 'backer''s datapath supports recirculation. Only newer
1005 * datapaths support OVS_KEY_ATTR_RECIRC_ID in keys. We need to disable some
1006 * features on older datapaths that don't support this feature.
1008 * Returns false if 'backer' definitely does not support recirculation, true if
1009 * it seems to support recirculation or if at least the error we get is
1012 check_recirc(struct dpif_backer *backer)
1015 struct odputil_keybuf keybuf;
1018 struct odp_flow_key_parms odp_parms = {
1025 memset(&flow, 0, sizeof flow);
1029 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1030 odp_flow_key_from_flow(&odp_parms, &key);
1031 enable_recirc = dpif_probe_feature(backer->dpif, "recirculation", &key,
1034 if (enable_recirc) {
1035 VLOG_INFO("%s: Datapath supports recirculation",
1036 dpif_name(backer->dpif));
1038 VLOG_INFO("%s: Datapath does not support recirculation",
1039 dpif_name(backer->dpif));
1042 return enable_recirc;
1045 /* Tests whether 'dpif' supports unique flow ids. We can skip serializing
1046 * some flow attributes for datapaths that support this feature.
1048 * Returns true if 'dpif' supports UFID for flow operations.
1049 * Returns false if 'dpif' does not support UFID. */
1051 check_ufid(struct dpif_backer *backer)
1054 struct odputil_keybuf keybuf;
1058 struct odp_flow_key_parms odp_parms = {
1062 memset(&flow, 0, sizeof flow);
1063 flow.dl_type = htons(0x1234);
1065 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1066 odp_flow_key_from_flow(&odp_parms, &key);
1067 dpif_flow_hash(backer->dpif, key.data, key.size, &ufid);
1069 enable_ufid = dpif_probe_feature(backer->dpif, "UFID", &key, &ufid);
1072 VLOG_INFO("%s: Datapath supports unique flow ids",
1073 dpif_name(backer->dpif));
1075 VLOG_INFO("%s: Datapath does not support unique flow ids",
1076 dpif_name(backer->dpif));
1081 /* Tests whether 'backer''s datapath supports variable-length
1082 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions. We need
1083 * to disable some features on older datapaths that don't support this
1086 * Returns false if 'backer' definitely does not support variable-length
1087 * userdata, true if it seems to support them or if at least the error we get
1090 check_variable_length_userdata(struct dpif_backer *backer)
1092 struct eth_header *eth;
1093 struct ofpbuf actions;
1094 struct dpif_execute execute;
1095 struct dp_packet packet;
1099 /* Compose a userspace action that will cause an ERANGE error on older
1100 * datapaths that don't support variable-length userdata.
1102 * We really test for using userdata longer than 8 bytes, but older
1103 * datapaths accepted these, silently truncating the userdata to 8 bytes.
1104 * The same older datapaths rejected userdata shorter than 8 bytes, so we
1105 * test for that instead as a proxy for longer userdata support. */
1106 ofpbuf_init(&actions, 64);
1107 start = nl_msg_start_nested(&actions, OVS_ACTION_ATTR_USERSPACE);
1108 nl_msg_put_u32(&actions, OVS_USERSPACE_ATTR_PID,
1109 dpif_port_get_pid(backer->dpif, ODPP_NONE, 0));
1110 nl_msg_put_unspec_zero(&actions, OVS_USERSPACE_ATTR_USERDATA, 4);
1111 nl_msg_end_nested(&actions, start);
1113 /* Compose a dummy ethernet packet. */
1114 dp_packet_init(&packet, ETH_HEADER_LEN);
1115 eth = dp_packet_put_zeros(&packet, ETH_HEADER_LEN);
1116 eth->eth_type = htons(0x1234);
1118 /* Execute the actions. On older datapaths this fails with ERANGE, on
1119 * newer datapaths it succeeds. */
1120 execute.actions = actions.data;
1121 execute.actions_len = actions.size;
1122 execute.packet = &packet;
1123 execute.needs_help = false;
1124 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;
1224 error = dpif_execute(backer->dpif, &execute);
1226 dp_packet_uninit(&packet);
1227 ofpbuf_uninit(&actions);
1230 /* Masked set action is not supported. */
1231 VLOG_INFO("%s: datapath does not support masked set action feature.",
1232 dpif_name(backer->dpif));
1237 #define CHECK_FEATURE__(NAME, FIELD) \
1239 check_##NAME(struct dpif_backer *backer) \
1242 struct odputil_keybuf keybuf; \
1243 struct ofpbuf key; \
1245 struct odp_flow_key_parms odp_parms = { \
1252 memset(&flow, 0, sizeof flow); \
1255 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); \
1256 odp_flow_key_from_flow(&odp_parms, &key); \
1257 enable = dpif_probe_feature(backer->dpif, #NAME, &key, NULL); \
1260 VLOG_INFO("%s: Datapath supports "#NAME, dpif_name(backer->dpif)); \
1262 VLOG_INFO("%s: Datapath does not support "#NAME, \
1263 dpif_name(backer->dpif)); \
1268 #define CHECK_FEATURE(FIELD) CHECK_FEATURE__(FIELD, FIELD)
1270 CHECK_FEATURE(ct_state)
1271 CHECK_FEATURE(ct_zone)
1273 #undef CHECK_FEATURE
1274 #undef CHECK_FEATURE__
1277 check_support(struct dpif_backer *backer)
1279 /* This feature needs to be tested after udpif threads are set. */
1280 backer->support.variable_length_userdata = false;
1282 backer->support.odp.recirc = check_recirc(backer);
1283 backer->support.odp.max_mpls_depth = check_max_mpls_depth(backer);
1284 backer->support.masked_set_action = check_masked_set_action(backer);
1285 backer->support.ufid = check_ufid(backer);
1286 backer->support.tnl_push_pop = dpif_supports_tnl_push_pop(backer->dpif);
1288 backer->support.odp.ct_state = check_ct_state(backer);
1289 backer->support.odp.ct_zone = check_ct_zone(backer);
1293 construct(struct ofproto *ofproto_)
1295 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1296 struct shash_node *node, *next;
1299 /* Tunnel module can get used right after the udpif threads are running. */
1300 ofproto_tunnel_init();
1302 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1307 atomic_init(&ofproto->tables_version, CLS_MIN_VERSION);
1308 ofproto->netflow = NULL;
1309 ofproto->sflow = NULL;
1310 ofproto->ipfix = NULL;
1311 ofproto->stp = NULL;
1312 ofproto->rstp = NULL;
1313 ofproto->dump_seq = 0;
1314 hmap_init(&ofproto->bundles);
1315 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1317 ofproto->mbridge = mbridge_create();
1318 ofproto->has_bonded_bundles = false;
1319 ofproto->lacp_enabled = false;
1320 ovs_mutex_init_adaptive(&ofproto->stats_mutex);
1321 ovs_mutex_init(&ofproto->vsp_mutex);
1323 guarded_list_init(&ofproto->pins);
1325 hmap_init(&ofproto->vlandev_map);
1326 hmap_init(&ofproto->realdev_vid_map);
1328 sset_init(&ofproto->ports);
1329 sset_init(&ofproto->ghost_ports);
1330 sset_init(&ofproto->port_poll_set);
1331 ofproto->port_poll_errno = 0;
1332 ofproto->change_seq = 0;
1333 ofproto->pins_seq = seq_create();
1334 ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1337 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1338 struct iface_hint *iface_hint = node->data;
1340 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1341 /* Check if the datapath already has this port. */
1342 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1343 sset_add(&ofproto->ports, node->name);
1346 free(iface_hint->br_name);
1347 free(iface_hint->br_type);
1349 shash_delete(&init_ofp_ports, node);
1353 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1354 hash_string(ofproto->up.name, 0));
1355 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1357 ofproto_init_tables(ofproto_, N_TABLES);
1358 error = add_internal_flows(ofproto);
1360 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1366 add_internal_miss_flow(struct ofproto_dpif *ofproto, int id,
1367 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1373 match_init_catchall(&match);
1374 match_set_reg(&match, 0, id);
1376 error = ofproto_dpif_add_internal_flow(ofproto, &match, 0, 0, ofpacts,
1378 *rulep = error ? NULL : rule_dpif_cast(rule);
1384 add_internal_flows(struct ofproto_dpif *ofproto)
1386 struct ofpact_controller *controller;
1387 uint64_t ofpacts_stub[128 / 8];
1388 struct ofpbuf ofpacts;
1389 struct rule *unused_rulep OVS_UNUSED;
1394 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1397 controller = ofpact_put_CONTROLLER(&ofpacts);
1398 controller->max_len = UINT16_MAX;
1399 controller->controller_id = 0;
1400 controller->reason = OFPR_NO_MATCH;
1401 ofpact_pad(&ofpacts);
1403 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1404 &ofproto->miss_rule);
1409 ofpbuf_clear(&ofpacts);
1410 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1411 &ofproto->no_packet_in_rule);
1416 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1417 &ofproto->drop_frags_rule);
1422 /* Drop any run away non-recirc rule lookups. Recirc_id has to be
1423 * zero when reaching this rule.
1425 * (priority=2), recirc_id=0, actions=drop
1427 ofpbuf_clear(&ofpacts);
1428 match_init_catchall(&match);
1429 match_set_recirc_id(&match, 0);
1430 error = ofproto_dpif_add_internal_flow(ofproto, &match, 2, 0, &ofpacts,
1436 destruct(struct ofproto *ofproto_)
1438 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1439 struct ofproto_packet_in *pin;
1440 struct rule_dpif *rule;
1441 struct oftable *table;
1442 struct ovs_list pins;
1444 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1446 xlate_remove_ofproto(ofproto);
1449 /* Ensure that the upcall processing threads have no remaining references
1450 * to the ofproto or anything in it. */
1451 udpif_synchronize(ofproto->backer->udpif);
1453 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1455 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1456 CLS_FOR_EACH (rule, up.cr, &table->cls) {
1457 ofproto_rule_delete(&ofproto->up, &rule->up);
1460 ofproto_group_delete_all(&ofproto->up);
1462 guarded_list_pop_all(&ofproto->pins, &pins);
1463 LIST_FOR_EACH_POP (pin, list_node, &pins) {
1464 free(CONST_CAST(void *, pin->up.packet));
1467 guarded_list_destroy(&ofproto->pins);
1469 recirc_free_ofproto(ofproto, ofproto->up.name);
1471 mbridge_unref(ofproto->mbridge);
1473 netflow_unref(ofproto->netflow);
1474 dpif_sflow_unref(ofproto->sflow);
1475 dpif_ipfix_unref(ofproto->ipfix);
1476 hmap_destroy(&ofproto->bundles);
1477 mac_learning_unref(ofproto->ml);
1478 mcast_snooping_unref(ofproto->ms);
1480 hmap_destroy(&ofproto->vlandev_map);
1481 hmap_destroy(&ofproto->realdev_vid_map);
1483 sset_destroy(&ofproto->ports);
1484 sset_destroy(&ofproto->ghost_ports);
1485 sset_destroy(&ofproto->port_poll_set);
1487 ovs_mutex_destroy(&ofproto->stats_mutex);
1488 ovs_mutex_destroy(&ofproto->vsp_mutex);
1490 seq_destroy(ofproto->pins_seq);
1492 close_dpif_backer(ofproto->backer);
1496 run(struct ofproto *ofproto_)
1498 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1499 uint64_t new_seq, new_dump_seq;
1501 if (mbridge_need_revalidate(ofproto->mbridge)) {
1502 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1503 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1504 mac_learning_flush(ofproto->ml);
1505 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1506 mcast_snooping_mdb_flush(ofproto->ms);
1509 /* Always updates the ofproto->pins_seqno to avoid frequent wakeup during
1510 * flow restore. Even though nothing is processed during flow restore,
1511 * all queued 'pins' will be handled immediately when flow restore
1513 ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1515 /* Do not perform any periodic activity required by 'ofproto' while
1516 * waiting for flow restore to complete. */
1517 if (!ofproto_get_flow_restore_wait()) {
1518 struct ofproto_packet_in *pin;
1519 struct ovs_list pins;
1521 guarded_list_pop_all(&ofproto->pins, &pins);
1522 LIST_FOR_EACH_POP (pin, list_node, &pins) {
1523 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1524 free(CONST_CAST(void *, pin->up.packet));
1529 if (ofproto->netflow) {
1530 netflow_run(ofproto->netflow);
1532 if (ofproto->sflow) {
1533 dpif_sflow_run(ofproto->sflow);
1535 if (ofproto->ipfix) {
1536 dpif_ipfix_run(ofproto->ipfix);
1539 new_seq = seq_read(connectivity_seq_get());
1540 if (ofproto->change_seq != new_seq) {
1541 struct ofport_dpif *ofport;
1543 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1547 ofproto->change_seq = new_seq;
1549 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1550 struct ofbundle *bundle;
1552 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1559 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1560 if (mac_learning_run(ofproto->ml)) {
1561 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1563 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1565 if (mcast_snooping_run(ofproto->ms)) {
1566 ofproto->backer->need_revalidate = REV_MCAST_SNOOPING;
1569 new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
1570 if (ofproto->dump_seq != new_dump_seq) {
1571 struct rule *rule, *next_rule;
1573 /* We know stats are relatively fresh, so now is a good time to do some
1575 ofproto->dump_seq = new_dump_seq;
1577 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1579 ovs_mutex_lock(&ofproto_mutex);
1580 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
1581 &ofproto->up.expirable) {
1582 rule_expire(rule_dpif_cast(rule));
1584 ovs_mutex_unlock(&ofproto_mutex);
1586 /* All outstanding data in existing flows has been accounted, so it's a
1587 * good time to do bond rebalancing. */
1588 if (ofproto->has_bonded_bundles) {
1589 struct ofbundle *bundle;
1591 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1593 bond_rebalance(bundle->bond);
1602 wait(struct ofproto *ofproto_)
1604 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1606 if (ofproto_get_flow_restore_wait()) {
1610 if (ofproto->sflow) {
1611 dpif_sflow_wait(ofproto->sflow);
1613 if (ofproto->ipfix) {
1614 dpif_ipfix_wait(ofproto->ipfix);
1616 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1617 struct ofbundle *bundle;
1619 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1620 bundle_wait(bundle);
1623 if (ofproto->netflow) {
1624 netflow_wait(ofproto->netflow);
1626 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1627 mac_learning_wait(ofproto->ml);
1628 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1629 mcast_snooping_wait(ofproto->ms);
1631 if (ofproto->backer->need_revalidate) {
1632 /* Shouldn't happen, but if it does just go around again. */
1633 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1634 poll_immediate_wake();
1637 seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
1638 seq_wait(ofproto->pins_seq, ofproto->pins_seqno);
1642 type_get_memory_usage(const char *type, struct simap *usage)
1644 struct dpif_backer *backer;
1646 backer = shash_find_data(&all_dpif_backers, type);
1648 udpif_get_memory_usage(backer->udpif, usage);
1653 flush(struct ofproto *ofproto_)
1655 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1656 struct dpif_backer *backer = ofproto->backer;
1659 udpif_flush(backer->udpif);
1664 query_tables(struct ofproto *ofproto,
1665 struct ofputil_table_features *features,
1666 struct ofputil_table_stats *stats)
1668 strcpy(features->name, "classifier");
1673 for (i = 0; i < ofproto->n_tables; i++) {
1674 unsigned long missed, matched;
1676 atomic_read_relaxed(&ofproto->tables[i].n_matched, &matched);
1677 atomic_read_relaxed(&ofproto->tables[i].n_missed, &missed);
1679 stats[i].matched_count = matched;
1680 stats[i].lookup_count = matched + missed;
1686 set_tables_version(struct ofproto *ofproto_, cls_version_t version)
1688 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1690 atomic_store_relaxed(&ofproto->tables_version, version);
1694 static struct ofport *
1697 struct ofport_dpif *port = xzalloc(sizeof *port);
1702 port_dealloc(struct ofport *port_)
1704 struct ofport_dpif *port = ofport_dpif_cast(port_);
1709 port_construct(struct ofport *port_)
1711 struct ofport_dpif *port = ofport_dpif_cast(port_);
1712 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1713 const struct netdev *netdev = port->up.netdev;
1714 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1715 const char *dp_port_name;
1716 struct dpif_port dpif_port;
1719 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1720 port->bundle = NULL;
1724 port->may_enable = false;
1725 port->stp_port = NULL;
1726 port->stp_state = STP_DISABLED;
1727 port->rstp_port = NULL;
1728 port->rstp_state = RSTP_DISABLED;
1729 port->is_tunnel = false;
1733 port->realdev_ofp_port = 0;
1734 port->vlandev_vid = 0;
1735 port->carrier_seq = netdev_get_carrier_resets(netdev);
1736 port->is_layer3 = netdev_vport_is_layer3(netdev);
1738 if (netdev_vport_is_patch(netdev)) {
1739 /* By bailing out here, we don't submit the port to the sFlow module
1740 * to be considered for counter polling export. This is correct
1741 * because the patch port represents an interface that sFlow considers
1742 * to be "internal" to the switch as a whole, and therefore not a
1743 * candidate for counter polling. */
1744 port->odp_port = ODPP_NONE;
1745 ofport_update_peer(port);
1749 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
1750 error = dpif_port_query_by_name(ofproto->backer->dpif, dp_port_name,
1756 port->odp_port = dpif_port.port_no;
1758 if (netdev_get_tunnel_config(netdev)) {
1759 atomic_count_inc(&ofproto->backer->tnl_count);
1760 error = tnl_port_add(port, port->up.netdev, port->odp_port,
1761 ovs_native_tunneling_is_on(ofproto), dp_port_name);
1763 atomic_count_dec(&ofproto->backer->tnl_count);
1764 dpif_port_destroy(&dpif_port);
1768 port->is_tunnel = true;
1769 if (ofproto->ipfix) {
1770 dpif_ipfix_add_tunnel_port(ofproto->ipfix, port_, port->odp_port);
1773 /* Sanity-check that a mapping doesn't already exist. This
1774 * shouldn't happen for non-tunnel ports. */
1775 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1776 VLOG_ERR("port %s already has an OpenFlow port number",
1778 dpif_port_destroy(&dpif_port);
1782 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1783 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1784 hash_odp_port(port->odp_port));
1785 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1787 dpif_port_destroy(&dpif_port);
1789 if (ofproto->sflow) {
1790 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1797 port_destruct(struct ofport *port_)
1799 struct ofport_dpif *port = ofport_dpif_cast(port_);
1800 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1801 const char *devname = netdev_get_name(port->up.netdev);
1802 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1803 const char *dp_port_name;
1805 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1807 xlate_ofport_remove(port);
1810 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1812 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1813 /* The underlying device is still there, so delete it. This
1814 * happens when the ofproto is being destroyed, since the caller
1815 * assumes that removal of attached ports will happen as part of
1817 if (!port->is_tunnel) {
1818 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1823 port->peer->peer = NULL;
1827 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1828 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1829 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1830 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1833 if (port->is_tunnel) {
1834 atomic_count_dec(&ofproto->backer->tnl_count);
1837 if (port->is_tunnel && ofproto->ipfix) {
1838 dpif_ipfix_del_tunnel_port(ofproto->ipfix, port->odp_port);
1842 sset_find_and_delete(&ofproto->ports, devname);
1843 sset_find_and_delete(&ofproto->ghost_ports, devname);
1844 bundle_remove(port_);
1845 set_cfm(port_, NULL);
1846 set_bfd(port_, NULL);
1847 set_lldp(port_, NULL);
1848 if (port->stp_port) {
1849 stp_port_disable(port->stp_port);
1851 set_rstp_port(port_, NULL);
1852 if (ofproto->sflow) {
1853 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1860 port_modified(struct ofport *port_)
1862 struct ofport_dpif *port = ofport_dpif_cast(port_);
1863 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1864 const char *dp_port_name;
1865 struct netdev *netdev = port->up.netdev;
1867 if (port->bundle && port->bundle->bond) {
1868 bond_slave_set_netdev(port->bundle->bond, port, netdev);
1872 cfm_set_netdev(port->cfm, netdev);
1876 bfd_set_netdev(port->bfd, netdev);
1879 ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1880 port->lldp, &port->up.pp.hw_addr);
1882 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
1884 if (port->is_tunnel) {
1885 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1887 if (tnl_port_reconfigure(port, netdev, port->odp_port,
1888 ovs_native_tunneling_is_on(ofproto),
1890 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1894 ofport_update_peer(port);
1898 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1900 struct ofport_dpif *port = ofport_dpif_cast(port_);
1901 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1902 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1904 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1905 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1906 OFPUTIL_PC_NO_PACKET_IN)) {
1907 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1909 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1910 bundle_update(port->bundle);
1916 set_sflow(struct ofproto *ofproto_,
1917 const struct ofproto_sflow_options *sflow_options)
1919 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1920 struct dpif_sflow *ds = ofproto->sflow;
1922 if (sflow_options) {
1923 uint32_t old_probability = ds ? dpif_sflow_get_probability(ds) : 0;
1925 struct ofport_dpif *ofport;
1927 ds = ofproto->sflow = dpif_sflow_create();
1928 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1929 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1932 dpif_sflow_set_options(ds, sflow_options);
1933 if (dpif_sflow_get_probability(ds) != old_probability) {
1934 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1938 dpif_sflow_unref(ds);
1939 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1940 ofproto->sflow = NULL;
1948 struct ofproto *ofproto_,
1949 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1950 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1951 size_t n_flow_exporters_options)
1953 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1954 struct dpif_ipfix *di = ofproto->ipfix;
1955 bool has_options = bridge_exporter_options || flow_exporters_options;
1956 bool new_di = false;
1958 if (has_options && !di) {
1959 di = ofproto->ipfix = dpif_ipfix_create();
1964 /* Call set_options in any case to cleanly flush the flow
1965 * caches in the last exporters that are to be destroyed. */
1966 dpif_ipfix_set_options(
1967 di, bridge_exporter_options, flow_exporters_options,
1968 n_flow_exporters_options);
1970 /* Add tunnel ports only when a new ipfix created */
1971 if (new_di == true) {
1972 struct ofport_dpif *ofport;
1973 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1974 if (ofport->is_tunnel == true) {
1975 dpif_ipfix_add_tunnel_port(di, &ofport->up, ofport->odp_port);
1981 dpif_ipfix_unref(di);
1982 ofproto->ipfix = NULL;
1990 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1992 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1993 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1994 struct cfm *old = ofport->cfm;
1999 ofport->cfm = cfm_create(ofport->up.netdev);
2002 if (cfm_configure(ofport->cfm, s)) {
2009 cfm_unref(ofport->cfm);
2012 if (ofport->cfm != old) {
2013 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2015 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
2016 ofport->lldp, &ofport->up.pp.hw_addr);
2021 cfm_status_changed(struct ofport *ofport_)
2023 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2025 return ofport->cfm ? cfm_check_status_change(ofport->cfm) : true;
2029 get_cfm_status(const struct ofport *ofport_,
2030 struct cfm_status *status)
2032 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2036 cfm_get_status(ofport->cfm, status);
2045 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2047 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2048 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2052 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
2053 cfg, ofport->up.netdev);
2054 if (ofport->bfd != old) {
2055 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2057 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
2058 ofport->lldp, &ofport->up.pp.hw_addr);
2063 bfd_status_changed(struct ofport *ofport_)
2065 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2067 return ofport->bfd ? bfd_check_status_change(ofport->bfd) : true;
2071 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2073 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2077 bfd_get_status(ofport->bfd, smap);
2086 set_lldp(struct ofport *ofport_,
2087 const struct smap *cfg)
2089 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2093 if (!ofport->lldp) {
2094 struct ofproto_dpif *ofproto;
2096 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2097 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2098 ofport->lldp = lldp_create(ofport->up.netdev, ofport_->mtu, cfg);
2101 if (!lldp_configure(ofport->lldp, cfg)) {
2106 lldp_unref(ofport->lldp);
2107 ofport->lldp = NULL;
2110 ofproto_dpif_monitor_port_update(ofport,
2114 &ofport->up.pp.hw_addr);
2119 get_lldp_status(const struct ofport *ofport_,
2120 struct lldp_status *status OVS_UNUSED)
2122 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2124 return ofport->lldp ? true : false;
2128 set_aa(struct ofproto *ofproto OVS_UNUSED,
2129 const struct aa_settings *s)
2131 return aa_configure(s);
2135 aa_mapping_set(struct ofproto *ofproto_ OVS_UNUSED, void *aux,
2136 const struct aa_mapping_settings *s)
2138 return aa_mapping_register(aux, s);
2142 aa_mapping_unset(struct ofproto *ofproto OVS_UNUSED, void *aux)
2144 return aa_mapping_unregister(aux);
2148 aa_vlan_get_queued(struct ofproto *ofproto OVS_UNUSED, struct ovs_list *list)
2150 return aa_get_vlan_queued(list);
2154 aa_vlan_get_queue_size(struct ofproto *ofproto OVS_UNUSED)
2156 return aa_get_vlan_queue_size();
2160 /* Spanning Tree. */
2162 /* Called while rstp_mutex is held. */
2164 rstp_send_bpdu_cb(struct dp_packet *pkt, void *ofport_, void *ofproto_)
2166 struct ofproto_dpif *ofproto = ofproto_;
2167 struct ofport_dpif *ofport = ofport_;
2168 struct eth_header *eth = dp_packet_l2(pkt);
2170 netdev_get_etheraddr(ofport->up.netdev, ð->eth_src);
2171 if (eth_addr_is_zero(eth->eth_src)) {
2172 VLOG_WARN_RL(&rl, "%s port %d: cannot send RSTP BPDU on a port which "
2173 "does not have a configured source MAC address.",
2174 ofproto->up.name, ofp_to_u16(ofport->up.ofp_port));
2176 ofproto_dpif_send_packet(ofport, pkt);
2178 dp_packet_delete(pkt);
2182 send_bpdu_cb(struct dp_packet *pkt, int port_num, void *ofproto_)
2184 struct ofproto_dpif *ofproto = ofproto_;
2185 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2186 struct ofport_dpif *ofport;
2188 ofport = stp_port_get_aux(sp);
2190 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2191 ofproto->up.name, port_num);
2193 struct eth_header *eth = dp_packet_l2(pkt);
2195 netdev_get_etheraddr(ofport->up.netdev, ð->eth_src);
2196 if (eth_addr_is_zero(eth->eth_src)) {
2197 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2198 "with unknown MAC", ofproto->up.name, port_num);
2200 ofproto_dpif_send_packet(ofport, pkt);
2203 dp_packet_delete(pkt);
2206 /* Configure RSTP on 'ofproto_' using the settings defined in 's'. */
2208 set_rstp(struct ofproto *ofproto_, const struct ofproto_rstp_settings *s)
2210 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2212 /* Only revalidate flows if the configuration changed. */
2213 if (!s != !ofproto->rstp) {
2214 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2218 if (!ofproto->rstp) {
2219 ofproto->rstp = rstp_create(ofproto_->name, s->address,
2220 rstp_send_bpdu_cb, ofproto);
2221 ofproto->rstp_last_tick = time_msec();
2223 rstp_set_bridge_address(ofproto->rstp, s->address);
2224 rstp_set_bridge_priority(ofproto->rstp, s->priority);
2225 rstp_set_bridge_ageing_time(ofproto->rstp, s->ageing_time);
2226 rstp_set_bridge_force_protocol_version(ofproto->rstp,
2227 s->force_protocol_version);
2228 rstp_set_bridge_max_age(ofproto->rstp, s->bridge_max_age);
2229 rstp_set_bridge_forward_delay(ofproto->rstp, s->bridge_forward_delay);
2230 rstp_set_bridge_transmit_hold_count(ofproto->rstp,
2231 s->transmit_hold_count);
2233 struct ofport *ofport;
2234 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2235 set_rstp_port(ofport, NULL);
2237 rstp_unref(ofproto->rstp);
2238 ofproto->rstp = NULL;
2243 get_rstp_status(struct ofproto *ofproto_, struct ofproto_rstp_status *s)
2245 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2247 if (ofproto->rstp) {
2249 s->root_id = rstp_get_root_id(ofproto->rstp);
2250 s->bridge_id = rstp_get_bridge_id(ofproto->rstp);
2251 s->designated_id = rstp_get_designated_id(ofproto->rstp);
2252 s->root_path_cost = rstp_get_root_path_cost(ofproto->rstp);
2253 s->designated_port_id = rstp_get_designated_port_id(ofproto->rstp);
2254 s->bridge_port_id = rstp_get_bridge_port_id(ofproto->rstp);
2261 update_rstp_port_state(struct ofport_dpif *ofport)
2263 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2264 enum rstp_state state;
2266 /* Figure out new state. */
2267 state = ofport->rstp_port ? rstp_port_get_state(ofport->rstp_port)
2271 if (ofport->rstp_state != state) {
2272 enum ofputil_port_state of_state;
2275 VLOG_DBG("port %s: RSTP state changed from %s to %s",
2276 netdev_get_name(ofport->up.netdev),
2277 rstp_state_name(ofport->rstp_state),
2278 rstp_state_name(state));
2280 if (rstp_learn_in_state(ofport->rstp_state)
2281 != rstp_learn_in_state(state)) {
2282 /* XXX: Learning action flows should also be flushed. */
2283 if (ofport->bundle) {
2284 if (!rstp_shift_root_learned_address(ofproto->rstp)
2285 || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2286 bundle_flush_macs(ofport->bundle, false);
2290 fwd_change = rstp_forward_in_state(ofport->rstp_state)
2291 != rstp_forward_in_state(state);
2293 ofproto->backer->need_revalidate = REV_RSTP;
2294 ofport->rstp_state = state;
2296 if (fwd_change && ofport->bundle) {
2297 bundle_update(ofport->bundle);
2300 /* Update the RSTP state bits in the OpenFlow port description. */
2301 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2302 of_state |= (state == RSTP_LEARNING ? OFPUTIL_PS_STP_LEARN
2303 : state == RSTP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2304 : state == RSTP_DISCARDING ? OFPUTIL_PS_STP_LISTEN
2306 ofproto_port_set_state(&ofport->up, of_state);
2311 rstp_run(struct ofproto_dpif *ofproto)
2313 if (ofproto->rstp) {
2314 long long int now = time_msec();
2315 long long int elapsed = now - ofproto->rstp_last_tick;
2316 struct rstp_port *rp;
2317 struct ofport_dpif *ofport;
2319 /* Every second, decrease the values of the timers. */
2320 if (elapsed >= 1000) {
2321 rstp_tick_timers(ofproto->rstp);
2322 ofproto->rstp_last_tick = now;
2325 while ((ofport = rstp_get_next_changed_port_aux(ofproto->rstp, &rp))) {
2326 update_rstp_port_state(ofport);
2330 /* FIXME: This check should be done on-event (i.e., when setting
2331 * p->fdb_flush) and not periodically.
2333 while ((ofport = rstp_check_and_reset_fdb_flush(ofproto->rstp, &rp))) {
2334 if (!rstp_shift_root_learned_address(ofproto->rstp)
2335 || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2336 bundle_flush_macs(ofport->bundle, false);
2340 if (rstp_shift_root_learned_address(ofproto->rstp)) {
2341 struct ofport_dpif *old_root_aux =
2342 (struct ofport_dpif *)rstp_get_old_root_aux(ofproto->rstp);
2343 struct ofport_dpif *new_root_aux =
2344 (struct ofport_dpif *)rstp_get_new_root_aux(ofproto->rstp);
2345 if (old_root_aux != NULL && new_root_aux != NULL) {
2346 bundle_move(old_root_aux->bundle, new_root_aux->bundle);
2347 rstp_reset_root_changed(ofproto->rstp);
2353 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2355 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2357 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2359 /* Only revalidate flows if the configuration changed. */
2360 if (!s != !ofproto->stp) {
2361 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2365 if (!ofproto->stp) {
2366 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2367 send_bpdu_cb, ofproto);
2368 ofproto->stp_last_tick = time_msec();
2371 stp_set_bridge_id(ofproto->stp, s->system_id);
2372 stp_set_bridge_priority(ofproto->stp, s->priority);
2373 stp_set_hello_time(ofproto->stp, s->hello_time);
2374 stp_set_max_age(ofproto->stp, s->max_age);
2375 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2377 struct ofport *ofport;
2379 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2380 set_stp_port(ofport, NULL);
2383 stp_unref(ofproto->stp);
2384 ofproto->stp = NULL;
2391 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2393 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2397 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2398 s->designated_root = stp_get_designated_root(ofproto->stp);
2399 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2408 update_stp_port_state(struct ofport_dpif *ofport)
2410 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2411 enum stp_state state;
2413 /* Figure out new state. */
2414 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2418 if (ofport->stp_state != state) {
2419 enum ofputil_port_state of_state;
2422 VLOG_DBG("port %s: STP state changed from %s to %s",
2423 netdev_get_name(ofport->up.netdev),
2424 stp_state_name(ofport->stp_state),
2425 stp_state_name(state));
2426 if (stp_learn_in_state(ofport->stp_state)
2427 != stp_learn_in_state(state)) {
2428 /* xxx Learning action flows should also be flushed. */
2429 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2430 mac_learning_flush(ofproto->ml);
2431 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2432 mcast_snooping_mdb_flush(ofproto->ms);
2434 fwd_change = stp_forward_in_state(ofport->stp_state)
2435 != stp_forward_in_state(state);
2437 ofproto->backer->need_revalidate = REV_STP;
2438 ofport->stp_state = state;
2439 ofport->stp_state_entered = time_msec();
2441 if (fwd_change && ofport->bundle) {
2442 bundle_update(ofport->bundle);
2445 /* Update the STP state bits in the OpenFlow port description. */
2446 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2447 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2448 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2449 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2450 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2452 ofproto_port_set_state(&ofport->up, of_state);
2456 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2457 * caller is responsible for assigning STP port numbers and ensuring
2458 * there are no duplicates. */
2460 set_stp_port(struct ofport *ofport_,
2461 const struct ofproto_port_stp_settings *s)
2463 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2464 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2465 struct stp_port *sp = ofport->stp_port;
2467 if (!s || !s->enable) {
2469 ofport->stp_port = NULL;
2470 stp_port_disable(sp);
2471 update_stp_port_state(ofport);
2474 } else if (sp && stp_port_no(sp) != s->port_num
2475 && ofport == stp_port_get_aux(sp)) {
2476 /* The port-id changed, so disable the old one if it's not
2477 * already in use by another port. */
2478 stp_port_disable(sp);
2481 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2483 /* Set name before enabling the port so that debugging messages can print
2485 stp_port_set_name(sp, netdev_get_name(ofport->up.netdev));
2486 stp_port_enable(sp);
2488 stp_port_set_aux(sp, ofport);
2489 stp_port_set_priority(sp, s->priority);
2490 stp_port_set_path_cost(sp, s->path_cost);
2492 update_stp_port_state(ofport);
2498 get_stp_port_status(struct ofport *ofport_,
2499 struct ofproto_port_stp_status *s)
2501 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2502 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2503 struct stp_port *sp = ofport->stp_port;
2505 if (!ofproto->stp || !sp) {
2511 s->port_id = stp_port_get_id(sp);
2512 s->state = stp_port_get_state(sp);
2513 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2514 s->role = stp_port_get_role(sp);
2520 get_stp_port_stats(struct ofport *ofport_,
2521 struct ofproto_port_stp_stats *s)
2523 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2524 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2525 struct stp_port *sp = ofport->stp_port;
2527 if (!ofproto->stp || !sp) {
2533 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2539 stp_run(struct ofproto_dpif *ofproto)
2542 long long int now = time_msec();
2543 long long int elapsed = now - ofproto->stp_last_tick;
2544 struct stp_port *sp;
2547 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2548 ofproto->stp_last_tick = now;
2550 while (stp_get_changed_port(ofproto->stp, &sp)) {
2551 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2554 update_stp_port_state(ofport);
2558 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2559 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2560 mac_learning_flush(ofproto->ml);
2561 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2562 mcast_snooping_mdb_flush(ofproto->ms);
2568 stp_wait(struct ofproto_dpif *ofproto)
2571 poll_timer_wait(1000);
2575 /* Configures RSTP on 'ofport_' using the settings defined in 's'. The
2576 * caller is responsible for assigning RSTP port numbers and ensuring
2577 * there are no duplicates. */
2579 set_rstp_port(struct ofport *ofport_,
2580 const struct ofproto_port_rstp_settings *s)
2582 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2583 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2584 struct rstp_port *rp = ofport->rstp_port;
2586 if (!s || !s->enable) {
2588 rstp_port_set_aux(rp, NULL);
2589 rstp_port_set_state(rp, RSTP_DISABLED);
2590 rstp_port_set_mac_operational(rp, false);
2591 ofport->rstp_port = NULL;
2592 rstp_port_unref(rp);
2593 update_rstp_port_state(ofport);
2598 /* Check if need to add a new port. */
2600 rp = ofport->rstp_port = rstp_add_port(ofproto->rstp);
2603 rstp_port_set(rp, s->port_num, s->priority, s->path_cost,
2604 s->admin_edge_port, s->auto_edge,
2605 s->admin_p2p_mac_state, s->admin_port_state, s->mcheck,
2607 update_rstp_port_state(ofport);
2608 /* Synchronize operational status. */
2609 rstp_port_set_mac_operational(rp, ofport->may_enable);
2613 get_rstp_port_status(struct ofport *ofport_,
2614 struct ofproto_port_rstp_status *s)
2616 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2617 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2618 struct rstp_port *rp = ofport->rstp_port;
2620 if (!ofproto->rstp || !rp) {
2626 rstp_port_get_status(rp, &s->port_id, &s->state, &s->role,
2627 &s->designated_bridge_id, &s->designated_port_id,
2628 &s->designated_path_cost, &s->tx_count,
2629 &s->rx_count, &s->error_count, &s->uptime);
2634 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2637 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2638 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2640 if (ofport->n_qdscp != n_qdscp
2641 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2642 n_qdscp * sizeof *qdscp))) {
2643 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2644 free(ofport->qdscp);
2645 ofport->qdscp = n_qdscp
2646 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2648 ofport->n_qdscp = n_qdscp;
2656 /* Expires all MAC learning entries associated with 'bundle' and forces its
2657 * ofproto to revalidate every flow.
2659 * Normally MAC learning entries are removed only from the ofproto associated
2660 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2661 * are removed from every ofproto. When patch ports and SLB bonds are in use
2662 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2663 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2664 * with the host from which it migrated. */
2666 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2668 struct ofproto_dpif *ofproto = bundle->ofproto;
2669 struct mac_learning *ml = ofproto->ml;
2670 struct mac_entry *mac, *next_mac;
2672 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2673 ovs_rwlock_wrlock(&ml->rwlock);
2674 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2675 if (mac_entry_get_port(ml, mac) == bundle) {
2677 struct ofproto_dpif *o;
2679 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2681 struct mac_entry *e;
2683 ovs_rwlock_wrlock(&o->ml->rwlock);
2684 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2686 mac_learning_expire(o->ml, e);
2688 ovs_rwlock_unlock(&o->ml->rwlock);
2693 mac_learning_expire(ml, mac);
2696 ovs_rwlock_unlock(&ml->rwlock);
2700 bundle_move(struct ofbundle *old, struct ofbundle *new)
2702 struct ofproto_dpif *ofproto = old->ofproto;
2703 struct mac_learning *ml = ofproto->ml;
2704 struct mac_entry *mac, *next_mac;
2706 ovs_assert(new->ofproto == old->ofproto);
2708 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2709 ovs_rwlock_wrlock(&ml->rwlock);
2710 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2711 if (mac_entry_get_port(ml, mac) == old) {
2712 mac_entry_set_port(ml, mac, new);
2715 ovs_rwlock_unlock(&ml->rwlock);
2718 static struct ofbundle *
2719 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2721 struct ofbundle *bundle;
2723 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2724 &ofproto->bundles) {
2725 if (bundle->aux == aux) {
2733 bundle_update(struct ofbundle *bundle)
2735 struct ofport_dpif *port;
2737 bundle->floodable = true;
2738 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2739 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2741 || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2742 || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2743 bundle->floodable = false;
2750 bundle_del_port(struct ofport_dpif *port)
2752 struct ofbundle *bundle = port->bundle;
2754 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2756 list_remove(&port->bundle_node);
2757 port->bundle = NULL;
2760 lacp_slave_unregister(bundle->lacp, port);
2763 bond_slave_unregister(bundle->bond, port);
2766 bundle_update(bundle);
2770 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2771 struct lacp_slave_settings *lacp)
2773 struct ofport_dpif *port;
2775 port = ofp_port_to_ofport(bundle->ofproto, ofp_port);
2780 if (port->bundle != bundle) {
2781 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2783 bundle_remove(&port->up);
2786 port->bundle = bundle;
2787 list_push_back(&bundle->ports, &port->bundle_node);
2788 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2790 || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2791 || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2792 bundle->floodable = false;
2796 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2797 lacp_slave_register(bundle->lacp, port, lacp);
2804 bundle_destroy(struct ofbundle *bundle)
2806 struct ofproto_dpif *ofproto;
2807 struct ofport_dpif *port, *next_port;
2813 ofproto = bundle->ofproto;
2814 mbridge_unregister_bundle(ofproto->mbridge, bundle);
2817 xlate_bundle_remove(bundle);
2820 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2821 bundle_del_port(port);
2824 bundle_flush_macs(bundle, true);
2825 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2827 free(bundle->trunks);
2828 lacp_unref(bundle->lacp);
2829 bond_unref(bundle->bond);
2834 bundle_set(struct ofproto *ofproto_, void *aux,
2835 const struct ofproto_bundle_settings *s)
2837 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2838 bool need_flush = false;
2839 struct ofport_dpif *port;
2840 struct ofbundle *bundle;
2841 unsigned long *trunks;
2847 bundle_destroy(bundle_lookup(ofproto, aux));
2851 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2852 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2854 bundle = bundle_lookup(ofproto, aux);
2856 bundle = xmalloc(sizeof *bundle);
2858 bundle->ofproto = ofproto;
2859 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2860 hash_pointer(aux, 0));
2862 bundle->name = NULL;
2864 list_init(&bundle->ports);
2865 bundle->vlan_mode = PORT_VLAN_TRUNK;
2867 bundle->trunks = NULL;
2868 bundle->use_priority_tags = s->use_priority_tags;
2869 bundle->lacp = NULL;
2870 bundle->bond = NULL;
2872 bundle->floodable = true;
2873 mbridge_register_bundle(ofproto->mbridge, bundle);
2876 if (!bundle->name || strcmp(s->name, bundle->name)) {
2878 bundle->name = xstrdup(s->name);
2883 ofproto->lacp_enabled = true;
2884 if (!bundle->lacp) {
2885 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2886 bundle->lacp = lacp_create();
2888 lacp_configure(bundle->lacp, s->lacp);
2890 lacp_unref(bundle->lacp);
2891 bundle->lacp = NULL;
2894 /* Update set of ports. */
2896 for (i = 0; i < s->n_slaves; i++) {
2897 if (!bundle_add_port(bundle, s->slaves[i],
2898 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2902 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2903 struct ofport_dpif *next_port;
2905 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2906 for (i = 0; i < s->n_slaves; i++) {
2907 if (s->slaves[i] == port->up.ofp_port) {
2912 bundle_del_port(port);
2916 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2918 if (list_is_empty(&bundle->ports)) {
2919 bundle_destroy(bundle);
2923 /* Set VLAN tagging mode */
2924 if (s->vlan_mode != bundle->vlan_mode
2925 || s->use_priority_tags != bundle->use_priority_tags) {
2926 bundle->vlan_mode = s->vlan_mode;
2927 bundle->use_priority_tags = s->use_priority_tags;
2932 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2933 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2935 if (vlan != bundle->vlan) {
2936 bundle->vlan = vlan;
2940 /* Get trunked VLANs. */
2941 switch (s->vlan_mode) {
2942 case PORT_VLAN_ACCESS:
2946 case PORT_VLAN_TRUNK:
2947 trunks = CONST_CAST(unsigned long *, s->trunks);
2950 case PORT_VLAN_NATIVE_UNTAGGED:
2951 case PORT_VLAN_NATIVE_TAGGED:
2952 if (vlan != 0 && (!s->trunks
2953 || !bitmap_is_set(s->trunks, vlan)
2954 || bitmap_is_set(s->trunks, 0))) {
2955 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2957 trunks = bitmap_clone(s->trunks, 4096);
2959 trunks = bitmap_allocate1(4096);
2961 bitmap_set1(trunks, vlan);
2962 bitmap_set0(trunks, 0);
2964 trunks = CONST_CAST(unsigned long *, s->trunks);
2971 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2972 free(bundle->trunks);
2973 if (trunks == s->trunks) {
2974 bundle->trunks = vlan_bitmap_clone(trunks);
2976 bundle->trunks = trunks;
2981 if (trunks != s->trunks) {
2986 if (!list_is_short(&bundle->ports)) {
2987 bundle->ofproto->has_bonded_bundles = true;
2989 if (bond_reconfigure(bundle->bond, s->bond)) {
2990 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2993 bundle->bond = bond_create(s->bond, ofproto);
2994 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2997 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2998 bond_slave_register(bundle->bond, port,
2999 port->up.ofp_port, port->up.netdev);
3002 bond_unref(bundle->bond);
3003 bundle->bond = NULL;
3006 /* If we changed something that would affect MAC learning, un-learn
3007 * everything on this port and force flow revalidation. */
3009 bundle_flush_macs(bundle, false);
3016 bundle_remove(struct ofport *port_)
3018 struct ofport_dpif *port = ofport_dpif_cast(port_);
3019 struct ofbundle *bundle = port->bundle;
3022 bundle_del_port(port);
3023 if (list_is_empty(&bundle->ports)) {
3024 bundle_destroy(bundle);
3025 } else if (list_is_short(&bundle->ports)) {
3026 bond_unref(bundle->bond);
3027 bundle->bond = NULL;
3033 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
3035 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3036 struct ofport_dpif *port = port_;
3040 error = netdev_get_etheraddr(port->up.netdev, &ea);
3042 struct dp_packet packet;
3045 dp_packet_init(&packet, 0);
3046 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
3048 memcpy(packet_pdu, pdu, pdu_size);
3050 ofproto_dpif_send_packet(port, &packet);
3051 dp_packet_uninit(&packet);
3053 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
3054 "%s (%s)", port->bundle->name,
3055 netdev_get_name(port->up.netdev), ovs_strerror(error));
3060 bundle_send_learning_packets(struct ofbundle *bundle)
3062 struct ofproto_dpif *ofproto = bundle->ofproto;
3063 int error, n_packets, n_errors;
3064 struct mac_entry *e;
3066 struct ovs_list list_node;
3067 struct ofport_dpif *port;
3068 struct dp_packet *pkt;
3070 struct ovs_list packets;
3072 list_init(&packets);
3073 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
3074 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3075 if (mac_entry_get_port(ofproto->ml, e) != bundle) {
3076 pkt_node = xmalloc(sizeof *pkt_node);
3077 pkt_node->pkt = bond_compose_learning_packet(bundle->bond,
3079 (void **)&pkt_node->port);
3080 list_push_back(&packets, &pkt_node->list_node);
3083 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3085 error = n_packets = n_errors = 0;
3086 LIST_FOR_EACH_POP (pkt_node, list_node, &packets) {
3089 ret = ofproto_dpif_send_packet(pkt_node->port, pkt_node->pkt);
3090 dp_packet_delete(pkt_node->pkt);
3100 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3101 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3102 "packets, last error was: %s",
3103 bundle->name, n_errors, n_packets, ovs_strerror(error));
3105 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3106 bundle->name, n_packets);
3111 bundle_run(struct ofbundle *bundle)
3114 lacp_run(bundle->lacp, send_pdu_cb);
3117 struct ofport_dpif *port;
3119 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
3120 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
3123 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
3124 bundle->ofproto->backer->need_revalidate = REV_BOND;
3127 if (bond_should_send_learning_packets(bundle->bond)) {
3128 bundle_send_learning_packets(bundle);
3134 bundle_wait(struct ofbundle *bundle)
3137 lacp_wait(bundle->lacp);
3140 bond_wait(bundle->bond);
3147 mirror_set__(struct ofproto *ofproto_, void *aux,
3148 const struct ofproto_mirror_settings *s)
3150 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3151 struct ofbundle **srcs, **dsts;
3156 mirror_destroy(ofproto->mbridge, aux);
3160 srcs = xmalloc(s->n_srcs * sizeof *srcs);
3161 dsts = xmalloc(s->n_dsts * sizeof *dsts);
3163 for (i = 0; i < s->n_srcs; i++) {
3164 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
3167 for (i = 0; i < s->n_dsts; i++) {
3168 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
3171 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
3172 s->n_dsts, s->src_vlans,
3173 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
3180 mirror_get_stats__(struct ofproto *ofproto, void *aux,
3181 uint64_t *packets, uint64_t *bytes)
3183 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
3188 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3190 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3191 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3192 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3193 mac_learning_flush(ofproto->ml);
3195 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3200 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3202 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3203 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3204 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
3208 forward_bpdu_changed(struct ofproto *ofproto_)
3210 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3211 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3215 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3218 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3219 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3220 mac_learning_set_idle_time(ofproto->ml, idle_time);
3221 mac_learning_set_max_entries(ofproto->ml, max_entries);
3222 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3225 /* Configures multicast snooping on 'ofport' using the settings
3226 * defined in 's'. */
3228 set_mcast_snooping(struct ofproto *ofproto_,
3229 const struct ofproto_mcast_snooping_settings *s)
3231 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3233 /* Only revalidate flows if the configuration changed. */
3234 if (!s != !ofproto->ms) {
3235 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3240 ofproto->ms = mcast_snooping_create();
3243 ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3244 mcast_snooping_set_idle_time(ofproto->ms, s->idle_time);
3245 mcast_snooping_set_max_entries(ofproto->ms, s->max_entries);
3246 if (mcast_snooping_set_flood_unreg(ofproto->ms, s->flood_unreg)) {
3247 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3249 ovs_rwlock_unlock(&ofproto->ms->rwlock);
3251 mcast_snooping_unref(ofproto->ms);
3258 /* Configures multicast snooping port's flood settings on 'ofproto'. */
3260 set_mcast_snooping_port(struct ofproto *ofproto_, void *aux,
3261 const struct ofproto_mcast_snooping_port_settings *s)
3263 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3264 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3266 if (ofproto->ms && s) {
3267 ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3268 mcast_snooping_set_port_flood(ofproto->ms, bundle, s->flood);
3269 mcast_snooping_set_port_flood_reports(ofproto->ms, bundle,
3271 ovs_rwlock_unlock(&ofproto->ms->rwlock);
3279 struct ofport_dpif *
3280 ofp_port_to_ofport(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
3282 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3283 return ofport ? ofport_dpif_cast(ofport) : NULL;
3287 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3288 struct ofproto_port *ofproto_port,
3289 struct dpif_port *dpif_port)
3291 ofproto_port->name = dpif_port->name;
3292 ofproto_port->type = dpif_port->type;
3293 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3297 ofport_update_peer(struct ofport_dpif *ofport)
3299 const struct ofproto_dpif *ofproto;
3300 struct dpif_backer *backer;
3303 if (!netdev_vport_is_patch(ofport->up.netdev)) {
3307 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
3308 backer->need_revalidate = REV_RECONFIGURE;
3311 ofport->peer->peer = NULL;
3312 ofport->peer = NULL;
3315 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
3320 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3321 struct ofport *peer_ofport;
3322 struct ofport_dpif *peer;
3325 if (ofproto->backer != backer) {
3329 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
3334 peer = ofport_dpif_cast(peer_ofport);
3335 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
3336 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
3338 ofport->peer = peer;
3339 ofport->peer->peer = ofport;
3349 port_run(struct ofport_dpif *ofport)
3351 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3352 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3353 bool enable = netdev_get_carrier(ofport->up.netdev);
3354 bool cfm_enable = false;
3355 bool bfd_enable = false;
3357 ofport->carrier_seq = carrier_seq;
3360 int cfm_opup = cfm_get_opup(ofport->cfm);
3362 cfm_enable = !cfm_get_fault(ofport->cfm);
3364 if (cfm_opup >= 0) {
3365 cfm_enable = cfm_enable && cfm_opup;
3370 bfd_enable = bfd_forwarding(ofport->bfd);
3373 if (ofport->bfd || ofport->cfm) {
3374 enable = enable && (cfm_enable || bfd_enable);
3377 if (ofport->bundle) {
3378 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3379 if (carrier_changed) {
3380 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3384 if (ofport->may_enable != enable) {
3385 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3387 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3389 if (ofport->rstp_port) {
3390 rstp_port_set_mac_operational(ofport->rstp_port, enable);
3394 ofport->may_enable = enable;
3398 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3399 struct ofproto_port *ofproto_port)
3401 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3402 struct dpif_port dpif_port;
3405 if (sset_contains(&ofproto->ghost_ports, devname)) {
3406 const char *type = netdev_get_type_from_name(devname);
3408 /* We may be called before ofproto->up.port_by_name is populated with
3409 * the appropriate ofport. For this reason, we must get the name and
3410 * type from the netdev layer directly. */
3412 const struct ofport *ofport;
3414 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3415 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3416 ofproto_port->name = xstrdup(devname);
3417 ofproto_port->type = xstrdup(type);
3423 if (!sset_contains(&ofproto->ports, devname)) {
3426 error = dpif_port_query_by_name(ofproto->backer->dpif,
3427 devname, &dpif_port);
3429 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3435 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3437 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3438 const char *devname = netdev_get_name(netdev);
3439 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3440 const char *dp_port_name;
3442 if (netdev_vport_is_patch(netdev)) {
3443 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3447 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3448 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3449 odp_port_t port_no = ODPP_NONE;
3452 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3456 if (netdev_get_tunnel_config(netdev)) {
3457 simap_put(&ofproto->backer->tnl_backers,
3458 dp_port_name, odp_to_u32(port_no));
3462 if (netdev_get_tunnel_config(netdev)) {
3463 sset_add(&ofproto->ghost_ports, devname);
3465 sset_add(&ofproto->ports, devname);
3471 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3473 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3474 struct ofport_dpif *ofport = ofp_port_to_ofport(ofproto, ofp_port);
3481 sset_find_and_delete(&ofproto->ghost_ports,
3482 netdev_get_name(ofport->up.netdev));
3483 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3484 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3485 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3487 /* The caller is going to close ofport->up.netdev. If this is a
3488 * bonded port, then the bond is using that netdev, so remove it
3489 * from the bond. The client will need to reconfigure everything
3490 * after deleting ports, so then the slave will get re-added. */
3491 bundle_remove(&ofport->up);
3498 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3500 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3503 error = netdev_get_stats(ofport->up.netdev, stats);
3505 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3506 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3508 ovs_mutex_lock(&ofproto->stats_mutex);
3509 /* ofproto->stats.tx_packets represents packets that we created
3510 * internally and sent to some port (e.g. packets sent with
3511 * ofproto_dpif_send_packet()). Account for them as if they had
3512 * come from OFPP_LOCAL and got forwarded. */
3514 if (stats->rx_packets != UINT64_MAX) {
3515 stats->rx_packets += ofproto->stats.tx_packets;
3518 if (stats->rx_bytes != UINT64_MAX) {
3519 stats->rx_bytes += ofproto->stats.tx_bytes;
3522 /* ofproto->stats.rx_packets represents packets that were received on
3523 * some port and we processed internally and dropped (e.g. STP).
3524 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3526 if (stats->tx_packets != UINT64_MAX) {
3527 stats->tx_packets += ofproto->stats.rx_packets;
3530 if (stats->tx_bytes != UINT64_MAX) {
3531 stats->tx_bytes += ofproto->stats.rx_bytes;
3533 ovs_mutex_unlock(&ofproto->stats_mutex);
3540 port_get_lacp_stats(const struct ofport *ofport_, struct lacp_slave_stats *stats)
3542 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3543 if (ofport->bundle && ofport->bundle->lacp) {
3544 if (lacp_get_slave_stats(ofport->bundle->lacp, ofport, stats)) {
3551 struct port_dump_state {
3556 struct ofproto_port port;
3561 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3563 *statep = xzalloc(sizeof(struct port_dump_state));
3568 port_dump_next(const struct ofproto *ofproto_, void *state_,
3569 struct ofproto_port *port)
3571 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3572 struct port_dump_state *state = state_;
3573 const struct sset *sset;
3574 struct sset_node *node;
3576 if (state->has_port) {
3577 ofproto_port_destroy(&state->port);
3578 state->has_port = false;
3580 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3581 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3584 error = port_query_by_name(ofproto_, node->name, &state->port);
3586 *port = state->port;
3587 state->has_port = true;
3589 } else if (error != ENODEV) {
3594 if (!state->ghost) {
3595 state->ghost = true;
3598 return port_dump_next(ofproto_, state_, port);
3605 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3607 struct port_dump_state *state = state_;
3609 if (state->has_port) {
3610 ofproto_port_destroy(&state->port);
3617 port_poll(const struct ofproto *ofproto_, char **devnamep)
3619 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3621 if (ofproto->port_poll_errno) {
3622 int error = ofproto->port_poll_errno;
3623 ofproto->port_poll_errno = 0;
3627 if (sset_is_empty(&ofproto->port_poll_set)) {
3631 *devnamep = sset_pop(&ofproto->port_poll_set);
3636 port_poll_wait(const struct ofproto *ofproto_)
3638 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3639 dpif_port_poll_wait(ofproto->backer->dpif);
3643 port_is_lacp_current(const struct ofport *ofport_)
3645 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3646 return (ofport->bundle && ofport->bundle->lacp
3647 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3651 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3652 * then delete it entirely. */
3654 rule_expire(struct rule_dpif *rule)
3655 OVS_REQUIRES(ofproto_mutex)
3657 uint16_t hard_timeout, idle_timeout;
3658 long long int now = time_msec();
3661 hard_timeout = rule->up.hard_timeout;
3662 idle_timeout = rule->up.idle_timeout;
3664 /* Has 'rule' expired? */
3666 long long int modified;
3668 ovs_mutex_lock(&rule->up.mutex);
3669 modified = rule->up.modified;
3670 ovs_mutex_unlock(&rule->up.mutex);
3672 if (now > modified + hard_timeout * 1000) {
3673 reason = OFPRR_HARD_TIMEOUT;
3677 if (reason < 0 && idle_timeout) {
3680 ovs_mutex_lock(&rule->stats_mutex);
3681 used = rule->stats.used;
3682 ovs_mutex_unlock(&rule->stats_mutex);
3684 if (now > used + idle_timeout * 1000) {
3685 reason = OFPRR_IDLE_TIMEOUT;
3690 COVERAGE_INC(ofproto_dpif_expired);
3691 ofproto_rule_expire(&rule->up, reason);
3696 ofproto_dpif_execute_actions__(struct ofproto_dpif *ofproto,
3697 const struct flow *flow,
3698 struct rule_dpif *rule,
3699 const struct ofpact *ofpacts, size_t ofpacts_len,
3700 int recurse, int resubmits,
3701 struct dp_packet *packet)
3703 struct dpif_flow_stats stats;
3704 struct xlate_out xout;
3705 struct xlate_in xin;
3707 struct dpif_execute execute;
3710 ovs_assert((rule != NULL) != (ofpacts != NULL));
3712 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3715 rule_dpif_credit_stats(rule, &stats);
3718 uint64_t odp_actions_stub[1024 / 8];
3719 struct ofpbuf odp_actions = OFPBUF_STUB_INITIALIZER(odp_actions_stub);
3720 xlate_in_init(&xin, ofproto, flow, flow->in_port.ofp_port, rule,
3721 stats.tcp_flags, packet, NULL, &odp_actions);
3722 xin.ofpacts = ofpacts;
3723 xin.ofpacts_len = ofpacts_len;
3724 xin.resubmit_stats = &stats;
3725 xin.recurse = recurse;
3726 xin.resubmits = resubmits;
3727 xlate_actions(&xin, &xout);
3729 execute.actions = odp_actions.data;
3730 execute.actions_len = odp_actions.size;
3732 pkt_metadata_from_flow(&packet->md, flow);
3733 execute.packet = packet;
3734 execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
3735 execute.probe = false;
3737 /* Fix up in_port. */
3738 in_port = flow->in_port.ofp_port;
3739 if (in_port == OFPP_NONE) {
3740 in_port = OFPP_LOCAL;
3742 execute.packet->md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
3744 error = dpif_execute(ofproto->backer->dpif, &execute);
3746 xlate_out_uninit(&xout);
3747 ofpbuf_uninit(&odp_actions);
3752 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3753 * 'flow' must reflect the data in 'packet'. */
3755 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3756 const struct flow *flow,
3757 struct rule_dpif *rule,
3758 const struct ofpact *ofpacts, size_t ofpacts_len,
3759 struct dp_packet *packet)
3761 return ofproto_dpif_execute_actions__(ofproto, flow, rule, ofpacts,
3762 ofpacts_len, 0, 0, packet);
3766 rule_dpif_credit_stats(struct rule_dpif *rule,
3767 const struct dpif_flow_stats *stats)
3769 ovs_mutex_lock(&rule->stats_mutex);
3770 if (OVS_UNLIKELY(rule->new_rule)) {
3771 rule_dpif_credit_stats(rule->new_rule, stats);
3773 rule->stats.n_packets += stats->n_packets;
3774 rule->stats.n_bytes += stats->n_bytes;
3775 rule->stats.used = MAX(rule->stats.used, stats->used);
3777 ovs_mutex_unlock(&rule->stats_mutex);
3781 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
3782 OVS_REQUIRES(rule->up.mutex)
3784 return rule->up.flow_cookie;
3788 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
3789 uint16_t hard_timeout)
3791 ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
3794 /* Returns 'rule''s actions. The returned actions are RCU-protected, and can
3795 * be read until the calling thread quiesces. */
3796 const struct rule_actions *
3797 rule_dpif_get_actions(const struct rule_dpif *rule)
3799 return rule_get_actions(&rule->up);
3802 /* Sets 'rule''s recirculation id. */
3804 rule_dpif_set_recirc_id(struct rule_dpif *rule, uint32_t id)
3805 OVS_REQUIRES(rule->up.mutex)
3807 ovs_assert(!rule->recirc_id || rule->recirc_id == id);
3808 if (rule->recirc_id == id) {
3809 /* Release the new reference to the same id. */
3812 rule->recirc_id = id;
3816 /* Sets 'rule''s recirculation id. */
3818 rule_set_recirc_id(struct rule *rule_, uint32_t id)
3820 struct rule_dpif *rule = rule_dpif_cast(rule_);
3822 ovs_mutex_lock(&rule->up.mutex);
3823 rule_dpif_set_recirc_id(rule, id);
3824 ovs_mutex_unlock(&rule->up.mutex);
3828 ofproto_dpif_get_tables_version(struct ofproto_dpif *ofproto OVS_UNUSED)
3830 cls_version_t version;
3832 atomic_read_relaxed(&ofproto->tables_version, &version);
3837 /* The returned rule (if any) is valid at least until the next RCU quiescent
3838 * period. If the rule needs to stay around longer, the caller should take
3841 * 'flow' is non-const to allow for temporary modifications during the lookup.
3842 * Any changes are restored before returning. */
3843 static struct rule_dpif *
3844 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, cls_version_t version,
3845 uint8_t table_id, struct flow *flow,
3846 struct flow_wildcards *wc)
3848 struct classifier *cls = &ofproto->up.tables[table_id].cls;
3849 return rule_dpif_cast(rule_from_cls_rule(classifier_lookup(cls, version,
3853 /* Look up 'flow' in 'ofproto''s classifier version 'version', starting from
3854 * table '*table_id'. Returns the rule that was found, which may be one of the
3855 * special rules according to packet miss hadling. If 'may_packet_in' is
3856 * false, returning of the miss_rule (which issues packet ins for the
3857 * controller) is avoided. Updates 'wc', if nonnull, to reflect the fields
3858 * that were used during the lookup.
3860 * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3861 * if none is found then the table miss configuration for that table is
3862 * honored, which can result in additional lookups in other OpenFlow tables.
3863 * In this case the function updates '*table_id' to reflect the final OpenFlow
3864 * table that was searched.
3866 * If 'honor_table_miss' is false, then only one table lookup occurs, in
3869 * The rule is returned in '*rule', which is valid at least until the next
3870 * RCU quiescent period. If the '*rule' needs to stay around longer, the
3871 * caller must take a reference.
3873 * 'in_port' allows the lookup to take place as if the in port had the value
3874 * 'in_port'. This is needed for resubmit action support.
3876 * 'flow' is non-const to allow for temporary modifications during the lookup.
3877 * Any changes are restored before returning. */
3879 rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto,
3880 cls_version_t version, struct flow *flow,
3881 struct flow_wildcards *wc,
3882 const struct dpif_flow_stats *stats,
3883 uint8_t *table_id, ofp_port_t in_port,
3884 bool may_packet_in, bool honor_table_miss)
3886 ovs_be16 old_tp_src = flow->tp_src, old_tp_dst = flow->tp_dst;
3887 ofp_port_t old_in_port = flow->in_port.ofp_port;
3888 enum ofputil_table_miss miss_config;
3889 struct rule_dpif *rule;
3892 /* We always unwildcard nw_frag (for IP), so they
3893 * need not be unwildcarded here. */
3894 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3895 && ofproto->up.frag_handling != OFPC_FRAG_NX_MATCH) {
3896 if (ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3897 /* We must pretend that transport ports are unavailable. */
3898 flow->tp_src = htons(0);
3899 flow->tp_dst = htons(0);
3901 /* Must be OFPC_FRAG_DROP (we don't have OFPC_FRAG_REASM).
3902 * Use the drop_frags_rule (which cannot disappear). */
3903 rule = ofproto->drop_frags_rule;
3905 struct oftable *tbl = &ofproto->up.tables[*table_id];
3908 atomic_add_relaxed(&tbl->n_matched, stats->n_packets, &orig);
3914 /* Look up a flow with 'in_port' as the input port. Then restore the
3915 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
3916 * have surprising behavior). */
3917 flow->in_port.ofp_port = in_port;
3919 /* Our current implementation depends on n_tables == N_TABLES, and
3920 * TBL_INTERNAL being the last table. */
3921 BUILD_ASSERT_DECL(N_TABLES == TBL_INTERNAL + 1);
3923 miss_config = OFPUTIL_TABLE_MISS_CONTINUE;
3925 for (next_id = *table_id;
3926 next_id < ofproto->up.n_tables;
3927 next_id++, next_id += (next_id == TBL_INTERNAL))
3929 *table_id = next_id;
3930 rule = rule_dpif_lookup_in_table(ofproto, version, next_id, flow, wc);
3932 struct oftable *tbl = &ofproto->up.tables[next_id];
3935 atomic_add_relaxed(rule ? &tbl->n_matched : &tbl->n_missed,
3936 stats->n_packets, &orig);
3939 goto out; /* Match. */
3941 if (honor_table_miss) {
3942 miss_config = ofproto_table_get_miss_config(&ofproto->up,
3944 if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE) {
3951 rule = ofproto->no_packet_in_rule;
3952 if (may_packet_in) {
3953 if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE
3954 || miss_config == OFPUTIL_TABLE_MISS_CONTROLLER) {
3955 struct ofport_dpif *port;
3957 port = ofp_port_to_ofport(ofproto, old_in_port);
3959 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
3961 } else if (!(port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN)) {
3962 rule = ofproto->miss_rule;
3964 } else if (miss_config == OFPUTIL_TABLE_MISS_DEFAULT &&
3965 connmgr_wants_packet_in_on_miss(ofproto->up.connmgr)) {
3966 rule = ofproto->miss_rule;
3970 /* Restore port numbers, as they may have been modified above. */
3971 flow->tp_src = old_tp_src;
3972 flow->tp_dst = old_tp_dst;
3973 /* Restore the old in port. */
3974 flow->in_port.ofp_port = old_in_port;
3980 complete_operation(struct rule_dpif *rule)
3981 OVS_REQUIRES(ofproto_mutex)
3983 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3985 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3988 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
3990 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
3993 static struct rule *
3996 struct rule_dpif *rule = xzalloc(sizeof *rule);
4001 rule_dealloc(struct rule *rule_)
4003 struct rule_dpif *rule = rule_dpif_cast(rule_);
4008 rule_check(struct rule *rule)
4010 uint16_t ct_state, ct_zone;
4012 ct_state = MINIFLOW_GET_U16(rule->cr.match.flow, ct_state);
4013 ct_zone = MINIFLOW_GET_U16(rule->cr.match.flow, ct_zone);
4015 if (ct_state || ct_zone) {
4016 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->ofproto);
4017 const struct odp_support *support = &ofproto_dpif_get_support(ofproto)->odp;
4019 if ((ct_state && !support->ct_state)
4020 || (ct_zone && !support->ct_zone)) {
4021 return OFPERR_OFPBMC_BAD_FIELD;
4023 if (ct_state & CS_UNSUPPORTED_MASK) {
4024 return OFPERR_OFPBMC_BAD_MASK;
4031 rule_construct(struct rule *rule_)
4032 OVS_NO_THREAD_SAFETY_ANALYSIS
4034 struct rule_dpif *rule = rule_dpif_cast(rule_);
4037 error = rule_check(rule_);
4042 ovs_mutex_init_adaptive(&rule->stats_mutex);
4043 rule->stats.n_packets = 0;
4044 rule->stats.n_bytes = 0;
4045 rule->stats.used = rule->up.modified;
4046 rule->recirc_id = 0;
4047 rule->new_rule = NULL;
4053 rule_insert(struct rule *rule_, struct rule *old_rule_, bool forward_stats)
4054 OVS_REQUIRES(ofproto_mutex)
4056 struct rule_dpif *rule = rule_dpif_cast(rule_);
4058 if (old_rule_ && forward_stats) {
4059 struct rule_dpif *old_rule = rule_dpif_cast(old_rule_);
4061 ovs_assert(!old_rule->new_rule);
4063 /* Take a reference to the new rule, and refer all stats updates from
4064 * the old rule to the new rule. */
4065 rule_dpif_ref(rule);
4067 ovs_mutex_lock(&old_rule->stats_mutex);
4068 ovs_mutex_lock(&rule->stats_mutex);
4069 old_rule->new_rule = rule; /* Forward future stats. */
4070 rule->stats = old_rule->stats; /* Transfer stats to the new rule. */
4071 ovs_mutex_unlock(&rule->stats_mutex);
4072 ovs_mutex_unlock(&old_rule->stats_mutex);
4075 complete_operation(rule);
4079 rule_delete(struct rule *rule_)
4080 OVS_REQUIRES(ofproto_mutex)
4082 struct rule_dpif *rule = rule_dpif_cast(rule_);
4083 complete_operation(rule);
4087 rule_destruct(struct rule *rule_)
4088 OVS_NO_THREAD_SAFETY_ANALYSIS
4090 struct rule_dpif *rule = rule_dpif_cast(rule_);
4092 ovs_mutex_destroy(&rule->stats_mutex);
4093 /* Release reference to the new rule, if any. */
4094 if (rule->new_rule) {
4095 rule_dpif_unref(rule->new_rule);
4097 if (rule->recirc_id) {
4098 recirc_free_id(rule->recirc_id);
4103 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
4104 long long int *used)
4106 struct rule_dpif *rule = rule_dpif_cast(rule_);
4108 ovs_mutex_lock(&rule->stats_mutex);
4109 if (OVS_UNLIKELY(rule->new_rule)) {
4110 rule_get_stats(&rule->new_rule->up, packets, bytes, used);
4112 *packets = rule->stats.n_packets;
4113 *bytes = rule->stats.n_bytes;
4114 *used = rule->stats.used;
4116 ovs_mutex_unlock(&rule->stats_mutex);
4120 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4121 struct dp_packet *packet)
4123 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4125 ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
4129 rule_execute(struct rule *rule, const struct flow *flow,
4130 struct dp_packet *packet)
4132 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4133 dp_packet_delete(packet);
4137 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
4139 return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
4142 static struct ofgroup *
4145 struct group_dpif *group = xzalloc(sizeof *group);
4150 group_dealloc(struct ofgroup *group_)
4152 struct group_dpif *group = group_dpif_cast(group_);
4157 group_construct_stats(struct group_dpif *group)
4158 OVS_REQUIRES(group->stats_mutex)
4160 struct ofputil_bucket *bucket;
4161 const struct ovs_list *buckets;
4163 group->packet_count = 0;
4164 group->byte_count = 0;
4166 group_dpif_get_buckets(group, &buckets);
4167 LIST_FOR_EACH (bucket, list_node, buckets) {
4168 bucket->stats.packet_count = 0;
4169 bucket->stats.byte_count = 0;
4174 group_dpif_credit_stats(struct group_dpif *group,
4175 struct ofputil_bucket *bucket,
4176 const struct dpif_flow_stats *stats)
4178 ovs_mutex_lock(&group->stats_mutex);
4179 group->packet_count += stats->n_packets;
4180 group->byte_count += stats->n_bytes;
4182 bucket->stats.packet_count += stats->n_packets;
4183 bucket->stats.byte_count += stats->n_bytes;
4184 } else { /* Credit to all buckets */
4185 const struct ovs_list *buckets;
4187 group_dpif_get_buckets(group, &buckets);
4188 LIST_FOR_EACH (bucket, list_node, buckets) {
4189 bucket->stats.packet_count += stats->n_packets;
4190 bucket->stats.byte_count += stats->n_bytes;
4193 ovs_mutex_unlock(&group->stats_mutex);
4197 group_construct(struct ofgroup *group_)
4199 struct group_dpif *group = group_dpif_cast(group_);
4201 ovs_mutex_init_adaptive(&group->stats_mutex);
4202 ovs_mutex_lock(&group->stats_mutex);
4203 group_construct_stats(group);
4204 ovs_mutex_unlock(&group->stats_mutex);
4209 group_destruct(struct ofgroup *group_)
4211 struct group_dpif *group = group_dpif_cast(group_);
4212 ovs_mutex_destroy(&group->stats_mutex);
4216 group_modify(struct ofgroup *group_)
4218 struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
4220 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4226 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
4228 struct group_dpif *group = group_dpif_cast(group_);
4229 struct ofputil_bucket *bucket;
4230 const struct ovs_list *buckets;
4231 struct bucket_counter *bucket_stats;
4233 ovs_mutex_lock(&group->stats_mutex);
4234 ogs->packet_count = group->packet_count;
4235 ogs->byte_count = group->byte_count;
4237 group_dpif_get_buckets(group, &buckets);
4238 bucket_stats = ogs->bucket_stats;
4239 LIST_FOR_EACH (bucket, list_node, buckets) {
4240 bucket_stats->packet_count = bucket->stats.packet_count;
4241 bucket_stats->byte_count = bucket->stats.byte_count;
4244 ovs_mutex_unlock(&group->stats_mutex);
4249 /* If the group exists, this function increments the groups's reference count.
4251 * Make sure to call group_dpif_unref() after no longer needing to maintain
4252 * a reference to the group. */
4254 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
4255 struct group_dpif **group)
4257 struct ofgroup *ofgroup;
4260 found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
4261 *group = found ? group_dpif_cast(ofgroup) : NULL;
4267 group_dpif_get_buckets(const struct group_dpif *group,
4268 const struct ovs_list **buckets)
4270 *buckets = &group->up.buckets;
4273 enum ofp11_group_type
4274 group_dpif_get_type(const struct group_dpif *group)
4276 return group->up.type;
4280 group_dpif_get_selection_method(const struct group_dpif *group)
4282 return group->up.props.selection_method;
4285 /* Sends 'packet' out 'ofport'.
4286 * May modify 'packet'.
4287 * Returns 0 if successful, otherwise a positive errno value. */
4289 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct dp_packet *packet)
4291 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4294 error = xlate_send_packet(ofport, packet);
4296 ovs_mutex_lock(&ofproto->stats_mutex);
4297 ofproto->stats.tx_packets++;
4298 ofproto->stats.tx_bytes += dp_packet_size(packet);
4299 ovs_mutex_unlock(&ofproto->stats_mutex);
4304 group_dpif_get_selection_method_param(const struct group_dpif *group)
4306 return group->up.props.selection_method_param;
4309 const struct field_array *
4310 group_dpif_get_fields(const struct group_dpif *group)
4312 return &group->up.props.fields;
4315 /* Return the version string of the datapath that backs up
4319 get_datapath_version(const struct ofproto *ofproto_)
4321 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4323 return ofproto->backer->dp_version_string;
4327 set_frag_handling(struct ofproto *ofproto_,
4328 enum ofp_config_flags frag_handling)
4330 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4331 if (frag_handling != OFPC_FRAG_REASM) {
4332 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4340 packet_out(struct ofproto *ofproto_, struct dp_packet *packet,
4341 const struct flow *flow,
4342 const struct ofpact *ofpacts, size_t ofpacts_len)
4344 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4346 ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
4347 ofpacts_len, packet);
4354 set_netflow(struct ofproto *ofproto_,
4355 const struct netflow_options *netflow_options)
4357 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4359 if (netflow_options) {
4360 if (!ofproto->netflow) {
4361 ofproto->netflow = netflow_create();
4362 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4364 return netflow_set_options(ofproto->netflow, netflow_options);
4365 } else if (ofproto->netflow) {
4366 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4367 netflow_unref(ofproto->netflow);
4368 ofproto->netflow = NULL;
4375 get_netflow_ids(const struct ofproto *ofproto_,
4376 uint8_t *engine_type, uint8_t *engine_id)
4378 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4380 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
4383 static struct ofproto_dpif *
4384 ofproto_dpif_lookup(const char *name)
4386 struct ofproto_dpif *ofproto;
4388 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
4389 hash_string(name, 0), &all_ofproto_dpifs) {
4390 if (!strcmp(ofproto->up.name, name)) {
4398 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
4399 const char *argv[], void *aux OVS_UNUSED)
4401 struct ofproto_dpif *ofproto;
4404 ofproto = ofproto_dpif_lookup(argv[1]);
4406 unixctl_command_reply_error(conn, "no such bridge");
4409 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4410 mac_learning_flush(ofproto->ml);
4411 ovs_rwlock_unlock(&ofproto->ml->rwlock);
4413 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4414 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4415 mac_learning_flush(ofproto->ml);
4416 ovs_rwlock_unlock(&ofproto->ml->rwlock);
4420 unixctl_command_reply(conn, "table successfully flushed");
4424 ofproto_unixctl_mcast_snooping_flush(struct unixctl_conn *conn, int argc,
4425 const char *argv[], void *aux OVS_UNUSED)
4427 struct ofproto_dpif *ofproto;
4430 ofproto = ofproto_dpif_lookup(argv[1]);
4432 unixctl_command_reply_error(conn, "no such bridge");
4436 if (!mcast_snooping_enabled(ofproto->ms)) {
4437 unixctl_command_reply_error(conn, "multicast snooping is disabled");
4440 mcast_snooping_mdb_flush(ofproto->ms);
4442 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4443 if (!mcast_snooping_enabled(ofproto->ms)) {
4446 mcast_snooping_mdb_flush(ofproto->ms);
4450 unixctl_command_reply(conn, "table successfully flushed");
4453 static struct ofport_dpif *
4454 ofbundle_get_a_port(const struct ofbundle *bundle)
4456 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
4461 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4462 const char *argv[], void *aux OVS_UNUSED)
4464 struct ds ds = DS_EMPTY_INITIALIZER;
4465 const struct ofproto_dpif *ofproto;
4466 const struct mac_entry *e;
4468 ofproto = ofproto_dpif_lookup(argv[1]);
4470 unixctl_command_reply_error(conn, "no such bridge");
4474 ds_put_cstr(&ds, " port VLAN MAC Age\n");
4475 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
4476 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
4477 struct ofbundle *bundle = mac_entry_get_port(ofproto->ml, e);
4478 char name[OFP_MAX_PORT_NAME_LEN];
4480 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4482 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
4483 name, e->vlan, ETH_ADDR_ARGS(e->mac),
4484 mac_entry_age(ofproto->ml, e));
4486 ovs_rwlock_unlock(&ofproto->ml->rwlock);
4487 unixctl_command_reply(conn, ds_cstr(&ds));
4492 ofproto_unixctl_mcast_snooping_show(struct unixctl_conn *conn,
4493 int argc OVS_UNUSED,
4495 void *aux OVS_UNUSED)
4497 struct ds ds = DS_EMPTY_INITIALIZER;
4498 const struct ofproto_dpif *ofproto;
4499 const struct ofbundle *bundle;
4500 const struct mcast_group *grp;
4501 struct mcast_group_bundle *b;
4502 struct mcast_mrouter_bundle *mrouter;
4504 ofproto = ofproto_dpif_lookup(argv[1]);
4506 unixctl_command_reply_error(conn, "no such bridge");
4510 if (!mcast_snooping_enabled(ofproto->ms)) {
4511 unixctl_command_reply_error(conn, "multicast snooping is disabled");
4515 ds_put_cstr(&ds, " port VLAN GROUP Age\n");
4516 ovs_rwlock_rdlock(&ofproto->ms->rwlock);
4517 LIST_FOR_EACH (grp, group_node, &ofproto->ms->group_lru) {
4518 LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
4519 char name[OFP_MAX_PORT_NAME_LEN];
4522 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4524 ds_put_format(&ds, "%5s %4d ", name, grp->vlan);
4525 print_ipv6_mapped(&ds, &grp->addr);
4526 ds_put_format(&ds, " %3d\n",
4527 mcast_bundle_age(ofproto->ms, b));
4531 /* ports connected to multicast routers */
4532 LIST_FOR_EACH(mrouter, mrouter_node, &ofproto->ms->mrouter_lru) {
4533 char name[OFP_MAX_PORT_NAME_LEN];
4535 bundle = mrouter->port;
4536 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4538 ds_put_format(&ds, "%5s %4d querier %3d\n",
4539 name, mrouter->vlan,
4540 mcast_mrouter_age(ofproto->ms, mrouter));
4542 ovs_rwlock_unlock(&ofproto->ms->rwlock);
4543 unixctl_command_reply(conn, ds_cstr(&ds));
4548 struct xlate_out xout;
4549 struct xlate_in xin;
4550 const struct flow *key;
4553 struct flow_wildcards wc;
4554 struct ofpbuf odp_actions;
4558 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
4560 const struct rule_actions *actions;
4563 ds_put_char_multiple(result, '\t', level);
4565 ds_put_cstr(result, "No match\n");
4569 ovs_mutex_lock(&rule->up.mutex);
4570 cookie = rule->up.flow_cookie;
4571 ovs_mutex_unlock(&rule->up.mutex);
4573 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
4574 rule ? rule->up.table_id : 0, ntohll(cookie));
4575 cls_rule_format(&rule->up.cr, result);
4576 ds_put_char(result, '\n');
4578 actions = rule_dpif_get_actions(rule);
4580 ds_put_char_multiple(result, '\t', level);
4581 ds_put_cstr(result, "OpenFlow actions=");
4582 ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
4583 ds_put_char(result, '\n');
4587 trace_format_flow(struct ds *result, int level, const char *title,
4588 struct trace_ctx *trace)
4590 ds_put_char_multiple(result, '\t', level);
4591 ds_put_format(result, "%s: ", title);
4592 /* Do not report unchanged flows for resubmits. */
4593 if ((level > 0 && flow_equal(&trace->xin.flow, &trace->flow))
4594 || (level == 0 && flow_equal(&trace->xin.flow, trace->key))) {
4595 ds_put_cstr(result, "unchanged");
4597 flow_format(result, &trace->xin.flow);
4598 trace->flow = trace->xin.flow;
4600 ds_put_char(result, '\n');
4604 trace_format_regs(struct ds *result, int level, const char *title,
4605 struct trace_ctx *trace)
4609 ds_put_char_multiple(result, '\t', level);
4610 ds_put_format(result, "%s:", title);
4611 for (i = 0; i < FLOW_N_REGS; i++) {
4612 ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
4614 ds_put_char(result, '\n');
4618 trace_format_odp(struct ds *result, int level, const char *title,
4619 struct trace_ctx *trace)
4621 struct ofpbuf *odp_actions = &trace->odp_actions;
4623 ds_put_char_multiple(result, '\t', level);
4624 ds_put_format(result, "%s: ", title);
4625 format_odp_actions(result, odp_actions->data, odp_actions->size);
4626 ds_put_char(result, '\n');
4630 trace_format_megaflow(struct ds *result, int level, const char *title,
4631 struct trace_ctx *trace)
4635 ds_put_char_multiple(result, '\t', level);
4636 ds_put_format(result, "%s: ", title);
4637 match_init(&match, trace->key, &trace->wc);
4638 match_format(&match, result, OFP_DEFAULT_PRIORITY);
4639 ds_put_char(result, '\n');
4642 static void trace_report(struct xlate_in *, int recurse,
4643 const char *format, ...)
4644 OVS_PRINTF_FORMAT(3, 4);
4645 static void trace_report_valist(struct xlate_in *, int recurse,
4646 const char *format, va_list args)
4647 OVS_PRINTF_FORMAT(3, 0);
4650 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
4652 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4653 struct ds *result = trace->result;
4656 if (rule == xin->ofproto->miss_rule) {
4657 trace_report(xin, recurse,
4658 "No match, flow generates \"packet in\"s.");
4659 } else if (rule == xin->ofproto->no_packet_in_rule) {
4660 trace_report(xin, recurse, "No match, packets dropped because "
4661 "OFPPC_NO_PACKET_IN is set on in_port.");
4662 } else if (rule == xin->ofproto->drop_frags_rule) {
4663 trace_report(xin, recurse, "Packets dropped because they are IP "
4664 "fragments and the fragment handling mode is "
4669 ds_put_char(result, '\n');
4671 trace_format_flow(result, recurse, "Resubmitted flow", trace);
4672 trace_format_regs(result, recurse, "Resubmitted regs", trace);
4673 trace_format_odp(result, recurse, "Resubmitted odp", trace);
4674 trace_format_megaflow(result, recurse, "Resubmitted megaflow", trace);
4676 trace_format_rule(result, recurse, rule);
4680 trace_report_valist(struct xlate_in *xin, int recurse,
4681 const char *format, va_list args)
4683 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4684 struct ds *result = trace->result;
4686 ds_put_char_multiple(result, '\t', recurse);
4687 ds_put_format_valist(result, format, args);
4688 ds_put_char(result, '\n');
4692 trace_report(struct xlate_in *xin, int recurse, const char *format, ...)
4696 va_start(args, format);
4697 trace_report_valist(xin, recurse, format, args);
4701 /* Parses the 'argc' elements of 'argv', ignoring argv[0]. The following
4702 * forms are supported:
4704 * - [dpname] odp_flow [-generate | packet]
4705 * - bridge br_flow [-generate | packet]
4707 * On success, initializes '*ofprotop' and 'flow' and returns NULL. On failure
4708 * returns a nonnull malloced error message. */
4709 static char * OVS_WARN_UNUSED_RESULT
4710 parse_flow_and_packet(int argc, const char *argv[],
4711 struct ofproto_dpif **ofprotop, struct flow *flow,
4712 struct dp_packet **packetp)
4714 const struct dpif_backer *backer = NULL;
4715 const char *error = NULL;
4717 struct simap port_names = SIMAP_INITIALIZER(&port_names);
4718 struct dp_packet *packet;
4719 struct ofpbuf odp_key;
4720 struct ofpbuf odp_mask;
4722 ofpbuf_init(&odp_key, 0);
4723 ofpbuf_init(&odp_mask, 0);
4725 /* Handle "-generate" or a hex string as the last argument. */
4726 if (!strcmp(argv[argc - 1], "-generate")) {
4727 packet = dp_packet_new(0);
4730 error = eth_from_hex(argv[argc - 1], &packet);
4733 } else if (argc == 4) {
4734 /* The 3-argument form must end in "-generate' or a hex string. */
4740 /* odp_flow can have its in_port specified as a name instead of port no.
4741 * We do not yet know whether a given flow is a odp_flow or a br_flow.
4742 * But, to know whether a flow is odp_flow through odp_flow_from_string(),
4743 * we need to create a simap of name to port no. */
4745 const char *dp_type;
4746 if (!strncmp(argv[1], "ovs-", 4)) {
4747 dp_type = argv[1] + 4;
4751 backer = shash_find_data(&all_dpif_backers, dp_type);
4752 } else if (argc == 2) {
4753 struct shash_node *node;
4754 if (shash_count(&all_dpif_backers) == 1) {
4755 node = shash_first(&all_dpif_backers);
4756 backer = node->data;
4759 error = "Syntax error";
4762 if (backer && backer->dpif) {
4763 struct dpif_port dpif_port;
4764 struct dpif_port_dump port_dump;
4765 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
4766 simap_put(&port_names, dpif_port.name,
4767 odp_to_u32(dpif_port.port_no));
4771 /* Parse the flow and determine whether a datapath or
4772 * bridge is specified. If function odp_flow_key_from_string()
4773 * returns 0, the flow is a odp_flow. If function
4774 * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
4775 if (!odp_flow_from_string(argv[argc - 1], &port_names,
4776 &odp_key, &odp_mask)) {
4778 error = "Cannot find the datapath";
4782 if (odp_flow_key_to_flow(odp_key.data, odp_key.size, flow) == ODP_FIT_ERROR) {
4783 error = "Failed to parse datapath flow key";
4787 *ofprotop = xlate_lookup_ofproto(backer, flow,
4788 &flow->in_port.ofp_port);
4789 if (*ofprotop == NULL) {
4790 error = "Invalid datapath flow";
4794 vsp_adjust_flow(*ofprotop, flow, NULL);
4797 char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
4800 m_err = xasprintf("Bad openflow flow syntax: %s", err);
4805 error = "Must specify bridge name";
4809 *ofprotop = ofproto_dpif_lookup(argv[1]);
4811 error = "Unknown bridge name";
4817 /* Generate a packet, if requested. */
4819 if (!dp_packet_size(packet)) {
4820 flow_compose(packet, flow);
4822 /* Use the metadata from the flow and the packet argument
4823 * to reconstruct the flow. */
4824 pkt_metadata_from_flow(&packet->md, flow);
4825 flow_extract(packet, flow);
4830 if (error && !m_err) {
4831 m_err = xstrdup(error);
4834 dp_packet_delete(packet);
4838 ofpbuf_uninit(&odp_key);
4839 ofpbuf_uninit(&odp_mask);
4840 simap_destroy(&port_names);
4845 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
4846 void *aux OVS_UNUSED)
4848 struct ofproto_dpif *ofproto;
4849 struct dp_packet *packet;
4853 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4858 ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
4859 unixctl_command_reply(conn, ds_cstr(&result));
4860 ds_destroy(&result);
4861 dp_packet_delete(packet);
4863 unixctl_command_reply_error(conn, error);
4869 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
4870 const char *argv[], void *aux OVS_UNUSED)
4872 enum ofputil_protocol usable_protocols;
4873 struct ofproto_dpif *ofproto;
4874 bool enforce_consistency;
4875 struct ofpbuf ofpacts;
4876 struct dp_packet *packet;
4881 /* Three kinds of error return values! */
4887 ofpbuf_init(&ofpacts, 0);
4889 /* Parse actions. */
4890 error = ofpacts_parse_actions(argv[--argc], &ofpacts, &usable_protocols);
4892 unixctl_command_reply_error(conn, error);
4897 /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
4898 * consistency between the flow and the actions. With -consistent, we
4899 * enforce consistency even for a flow supported in OpenFlow 1.0. */
4900 if (!strcmp(argv[1], "-consistent")) {
4901 enforce_consistency = true;
4905 enforce_consistency = false;
4908 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4910 unixctl_command_reply_error(conn, error);
4915 /* Do the same checks as handle_packet_out() in ofproto.c.
4917 * We pass a 'table_id' of 0 to ofpacts_check(), which isn't
4918 * strictly correct because these actions aren't in any table, but it's OK
4919 * because it 'table_id' is used only to check goto_table instructions, but
4920 * packet-outs take a list of actions and therefore it can't include
4923 * We skip the "meter" check here because meter is an instruction, not an
4924 * action, and thus cannot appear in ofpacts. */
4925 in_port = ofp_to_u16(flow.in_port.ofp_port);
4926 if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
4927 unixctl_command_reply_error(conn, "invalid in_port");
4930 if (enforce_consistency) {
4931 retval = ofpacts_check_consistency(ofpacts.data, ofpacts.size,
4932 &flow, u16_to_ofp(ofproto->up.max_ports),
4933 0, 0, usable_protocols);
4935 retval = ofpacts_check(ofpacts.data, ofpacts.size, &flow,
4936 u16_to_ofp(ofproto->up.max_ports), 0, 0,
4942 ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
4943 unixctl_command_reply_error(conn, ds_cstr(&result));
4947 ofproto_trace(ofproto, &flow, packet,
4948 ofpacts.data, ofpacts.size, &result);
4949 unixctl_command_reply(conn, ds_cstr(&result));
4952 ds_destroy(&result);
4953 dp_packet_delete(packet);
4954 ofpbuf_uninit(&ofpacts);
4957 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
4958 * description of the results to 'ds'.
4960 * The trace follows a packet with the specified 'flow' through the flow
4961 * table. 'packet' may be nonnull to trace an actual packet, with consequent
4962 * side effects (if it is nonnull then its flow must be 'flow').
4964 * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
4965 * trace, otherwise the actions are determined by a flow table lookup. */
4967 ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
4968 const struct dp_packet *packet,
4969 const struct ofpact ofpacts[], size_t ofpacts_len,
4972 struct trace_ctx trace;
4974 ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
4975 ds_put_cstr(ds, "Flow: ");
4976 flow_format(ds, flow);
4977 ds_put_char(ds, '\n');
4979 ofpbuf_init(&trace.odp_actions, 0);
4982 trace.key = flow; /* Original flow key, used for megaflow. */
4983 trace.flow = *flow; /* May be modified by actions. */
4984 xlate_in_init(&trace.xin, ofproto, flow, flow->in_port.ofp_port, NULL,
4985 ntohs(flow->tcp_flags), packet, &trace.wc,
4986 &trace.odp_actions);
4987 trace.xin.ofpacts = ofpacts;
4988 trace.xin.ofpacts_len = ofpacts_len;
4989 trace.xin.resubmit_hook = trace_resubmit;
4990 trace.xin.report_hook = trace_report_valist;
4992 xlate_actions(&trace.xin, &trace.xout);
4994 ds_put_char(ds, '\n');
4995 trace_format_flow(ds, 0, "Final flow", &trace);
4996 trace_format_megaflow(ds, 0, "Megaflow", &trace);
4998 ds_put_cstr(ds, "Datapath actions: ");
4999 format_odp_actions(ds, trace.odp_actions.data, trace.odp_actions.size);
5001 if (trace.xout.slow) {
5002 enum slow_path_reason slow;
5004 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5005 "slow path because it:");
5007 slow = trace.xout.slow;
5009 enum slow_path_reason bit = rightmost_1bit(slow);
5011 ds_put_format(ds, "\n\t- %s.",
5012 slow_path_reason_to_explanation(bit));
5018 xlate_out_uninit(&trace.xout);
5019 ofpbuf_uninit(&trace.odp_actions);
5022 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5023 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5024 * to destroy 'ofproto_shash' and free the returned value. */
5025 static const struct shash_node **
5026 get_ofprotos(struct shash *ofproto_shash)
5028 const struct ofproto_dpif *ofproto;
5030 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5031 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5032 shash_add_nocopy(ofproto_shash, name, ofproto);
5035 return shash_sort(ofproto_shash);
5039 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5040 const char *argv[] OVS_UNUSED,
5041 void *aux OVS_UNUSED)
5043 struct ds ds = DS_EMPTY_INITIALIZER;
5044 struct shash ofproto_shash;
5045 const struct shash_node **sorted_ofprotos;
5048 shash_init(&ofproto_shash);
5049 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5050 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5051 const struct shash_node *node = sorted_ofprotos[i];
5052 ds_put_format(&ds, "%s\n", node->name);
5055 shash_destroy(&ofproto_shash);
5056 free(sorted_ofprotos);
5058 unixctl_command_reply(conn, ds_cstr(&ds));
5063 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5065 const struct shash_node **ofprotos;
5066 struct dpif_dp_stats dp_stats;
5067 struct shash ofproto_shash;
5070 dpif_get_dp_stats(backer->dpif, &dp_stats);
5072 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5073 dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
5075 shash_init(&ofproto_shash);
5076 ofprotos = get_ofprotos(&ofproto_shash);
5077 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5078 struct ofproto_dpif *ofproto = ofprotos[i]->data;
5079 const struct shash_node **ports;
5082 if (ofproto->backer != backer) {
5086 ds_put_format(ds, "\t%s:\n", ofproto->up.name);
5088 ports = shash_sort(&ofproto->up.port_by_name);
5089 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5090 const struct shash_node *node = ports[j];
5091 struct ofport *ofport = node->data;
5093 odp_port_t odp_port;
5095 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5098 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5099 if (odp_port != ODPP_NONE) {
5100 ds_put_format(ds, "%"PRIu32":", odp_port);
5102 ds_put_cstr(ds, "none:");
5105 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5108 if (!netdev_get_config(ofport->netdev, &config)) {
5109 const struct smap_node **nodes;
5112 nodes = smap_sort(&config);
5113 for (i = 0; i < smap_count(&config); i++) {
5114 const struct smap_node *node = nodes[i];
5115 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5116 node->key, node->value);
5120 smap_destroy(&config);
5122 ds_put_char(ds, ')');
5123 ds_put_char(ds, '\n');
5127 shash_destroy(&ofproto_shash);
5132 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5133 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5135 struct ds ds = DS_EMPTY_INITIALIZER;
5136 const struct shash_node **backers;
5139 backers = shash_sort(&all_dpif_backers);
5140 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5141 dpif_show_backer(backers[i]->data, &ds);
5145 unixctl_command_reply(conn, ds_cstr(&ds));
5150 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5151 int argc OVS_UNUSED, const char *argv[],
5152 void *aux OVS_UNUSED)
5154 const struct ofproto_dpif *ofproto;
5156 struct ds ds = DS_EMPTY_INITIALIZER;
5157 bool verbosity = false;
5159 struct dpif_port dpif_port;
5160 struct dpif_port_dump port_dump;
5161 struct hmap portno_names;
5163 struct dpif_flow_dump *flow_dump;
5164 struct dpif_flow_dump_thread *flow_dump_thread;
5168 ofproto = ofproto_dpif_lookup(argv[argc - 1]);
5170 unixctl_command_reply_error(conn, "no such bridge");
5174 if (argc > 2 && !strcmp(argv[1], "-m")) {
5178 hmap_init(&portno_names);
5179 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
5180 odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
5184 flow_dump = dpif_flow_dump_create(ofproto->backer->dpif, false);
5185 flow_dump_thread = dpif_flow_dump_thread_create(flow_dump);
5186 while (dpif_flow_dump_next(flow_dump_thread, &f, 1)) {
5189 if (odp_flow_key_to_flow(f.key, f.key_len, &flow) == ODP_FIT_ERROR
5190 || xlate_lookup_ofproto(ofproto->backer, &flow, NULL) != ofproto) {
5195 odp_format_ufid(&f.ufid, &ds);
5196 ds_put_cstr(&ds, " ");
5198 odp_flow_format(f.key, f.key_len, f.mask, f.mask_len,
5199 &portno_names, &ds, verbosity);
5200 ds_put_cstr(&ds, ", ");
5201 dpif_flow_stats_format(&f.stats, &ds);
5202 ds_put_cstr(&ds, ", actions:");
5203 format_odp_actions(&ds, f.actions, f.actions_len);
5204 ds_put_char(&ds, '\n');
5206 dpif_flow_dump_thread_destroy(flow_dump_thread);
5207 error = dpif_flow_dump_destroy(flow_dump);
5211 ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
5212 unixctl_command_reply_error(conn, ds_cstr(&ds));
5214 unixctl_command_reply(conn, ds_cstr(&ds));
5216 odp_portno_names_destroy(&portno_names);
5217 hmap_destroy(&portno_names);
5222 ofproto_revalidate_all_backers(void)
5224 const struct shash_node **backers;
5227 backers = shash_sort(&all_dpif_backers);
5228 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5229 struct dpif_backer *backer = backers[i]->data;
5230 backer->need_revalidate = REV_RECONFIGURE;
5236 disable_tnl_push_pop(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5237 const char *argv[], void *aux OVS_UNUSED)
5239 if (!strcasecmp(argv[1], "off")) {
5240 ofproto_use_tnl_push_pop = false;
5241 unixctl_command_reply(conn, "Tunnel push-pop off");
5242 ofproto_revalidate_all_backers();
5243 } else if (!strcasecmp(argv[1], "on")) {
5244 ofproto_use_tnl_push_pop = true;
5245 unixctl_command_reply(conn, "Tunnel push-pop on");
5246 ofproto_revalidate_all_backers();
5251 ofproto_unixctl_init(void)
5253 static bool registered;
5259 unixctl_command_register(
5261 "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
5262 1, 3, ofproto_unixctl_trace, NULL);
5263 unixctl_command_register(
5264 "ofproto/trace-packet-out",
5265 "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
5266 2, 6, ofproto_unixctl_trace_actions, NULL);
5267 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5268 ofproto_unixctl_fdb_flush, NULL);
5269 unixctl_command_register("fdb/show", "bridge", 1, 1,
5270 ofproto_unixctl_fdb_show, NULL);
5271 unixctl_command_register("mdb/flush", "[bridge]", 0, 1,
5272 ofproto_unixctl_mcast_snooping_flush, NULL);
5273 unixctl_command_register("mdb/show", "bridge", 1, 1,
5274 ofproto_unixctl_mcast_snooping_show, NULL);
5275 unixctl_command_register("dpif/dump-dps", "", 0, 0,
5276 ofproto_unixctl_dpif_dump_dps, NULL);
5277 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5279 unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
5280 ofproto_unixctl_dpif_dump_flows, NULL);
5282 unixctl_command_register("ofproto/tnl-push-pop", "[on]|[off]", 1, 1,
5283 disable_tnl_push_pop, NULL);
5286 /* Returns true if 'table' is the table used for internal rules,
5287 * false otherwise. */
5289 table_is_internal(uint8_t table_id)
5291 return table_id == TBL_INTERNAL;
5294 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5296 * This is deprecated. It is only for compatibility with broken device drivers
5297 * in old versions of Linux that do not properly support VLANs when VLAN
5298 * devices are not used. When broken device drivers are no longer in
5299 * widespread use, we will delete these interfaces. */
5302 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
5304 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5305 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5307 if (realdev_ofp_port == ofport->realdev_ofp_port
5308 && vid == ofport->vlandev_vid) {
5312 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5314 if (ofport->realdev_ofp_port) {
5317 if (realdev_ofp_port && ofport->bundle) {
5318 /* vlandevs are enslaved to their realdevs, so they are not allowed to
5319 * themselves be part of a bundle. */
5320 bundle_set(ofport_->ofproto, ofport->bundle, NULL);
5323 ofport->realdev_ofp_port = realdev_ofp_port;
5324 ofport->vlandev_vid = vid;
5326 if (realdev_ofp_port) {
5327 vsp_add(ofport, realdev_ofp_port, vid);
5334 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
5336 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
5340 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
5341 OVS_EXCLUDED(ofproto->vsp_mutex)
5343 /* hmap_is_empty is thread safe. */
5344 return !hmap_is_empty(&ofproto->realdev_vid_map);
5349 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
5350 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5351 OVS_REQUIRES(ofproto->vsp_mutex)
5353 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5354 int vid = vlan_tci_to_vid(vlan_tci);
5355 const struct vlan_splinter *vsp;
5357 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5358 hash_realdev_vid(realdev_ofp_port, vid),
5359 &ofproto->realdev_vid_map) {
5360 if (vsp->realdev_ofp_port == realdev_ofp_port
5361 && vsp->vid == vid) {
5362 return vsp->vlandev_ofp_port;
5366 return realdev_ofp_port;
5369 /* Returns the OFP port number of the Linux VLAN device that corresponds to
5370 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
5371 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
5372 * 'vlan_tci' 9, it would return the port number of eth0.9.
5374 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
5375 * function just returns its 'realdev_ofp_port' argument. */
5377 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5378 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5379 OVS_EXCLUDED(ofproto->vsp_mutex)
5383 /* hmap_is_empty is thread safe, see if we can return immediately. */
5384 if (hmap_is_empty(&ofproto->realdev_vid_map)) {
5385 return realdev_ofp_port;
5387 ovs_mutex_lock(&ofproto->vsp_mutex);
5388 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
5389 ovs_mutex_unlock(&ofproto->vsp_mutex);
5393 static struct vlan_splinter *
5394 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
5396 struct vlan_splinter *vsp;
5398 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
5399 hash_ofp_port(vlandev_ofp_port),
5400 &ofproto->vlandev_map) {
5401 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5409 /* Returns the OpenFlow port number of the "real" device underlying the Linux
5410 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
5411 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
5412 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
5413 * eth0 and store 9 in '*vid'.
5415 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
5416 * VLAN device. Unless VLAN splinters are enabled, this is what this function
5419 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5420 ofp_port_t vlandev_ofp_port, int *vid)
5421 OVS_REQUIRES(ofproto->vsp_mutex)
5423 if (!hmap_is_empty(&ofproto->vlandev_map)) {
5424 const struct vlan_splinter *vsp;
5426 vsp = vlandev_find(ofproto, vlandev_ofp_port);
5431 return vsp->realdev_ofp_port;
5437 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
5438 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
5439 * 'flow->in_port' to the "real" device backing the VLAN device, sets
5440 * 'flow->vlan_tci' to the VLAN VID, and returns true. Optionally pushes the
5441 * appropriate VLAN on 'packet' if provided. Otherwise (which is always the
5442 * case unless VLAN splinters are enabled), returns false without making any
5445 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow,
5446 struct dp_packet *packet)
5447 OVS_EXCLUDED(ofproto->vsp_mutex)
5452 /* hmap_is_empty is thread safe. */
5453 if (hmap_is_empty(&ofproto->vlandev_map)) {
5457 ovs_mutex_lock(&ofproto->vsp_mutex);
5458 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
5459 ovs_mutex_unlock(&ofproto->vsp_mutex);
5464 /* Cause the flow to be processed as if it came in on the real device with
5465 * the VLAN device's VLAN ID. */
5466 flow->in_port.ofp_port = realdev;
5467 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
5470 /* Make the packet resemble the flow, so that it gets sent to an
5471 * OpenFlow controller properly, so that it looks correct for sFlow,
5472 * and so that flow_extract() will get the correct vlan_tci if it is
5473 * called on 'packet'. */
5474 eth_push_vlan(packet, htons(ETH_TYPE_VLAN), flow->vlan_tci);
5481 vsp_remove(struct ofport_dpif *port)
5483 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5484 struct vlan_splinter *vsp;
5486 ovs_mutex_lock(&ofproto->vsp_mutex);
5487 vsp = vlandev_find(ofproto, port->up.ofp_port);
5489 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
5490 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
5493 port->realdev_ofp_port = 0;
5495 VLOG_ERR("missing vlan device record");
5497 ovs_mutex_unlock(&ofproto->vsp_mutex);
5501 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
5503 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5505 ovs_mutex_lock(&ofproto->vsp_mutex);
5506 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
5507 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
5508 == realdev_ofp_port)) {
5509 struct vlan_splinter *vsp;
5511 vsp = xmalloc(sizeof *vsp);
5512 vsp->realdev_ofp_port = realdev_ofp_port;
5513 vsp->vlandev_ofp_port = port->up.ofp_port;
5516 port->realdev_ofp_port = realdev_ofp_port;
5518 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
5519 hash_ofp_port(port->up.ofp_port));
5520 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
5521 hash_realdev_vid(realdev_ofp_port, vid));
5523 VLOG_ERR("duplicate vlan device record");
5525 ovs_mutex_unlock(&ofproto->vsp_mutex);
5529 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
5531 const struct ofport_dpif *ofport = ofp_port_to_ofport(ofproto, ofp_port);
5532 return ofport ? ofport->odp_port : ODPP_NONE;
5535 struct ofport_dpif *
5536 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
5538 struct ofport_dpif *port;
5540 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
5541 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
5542 &backer->odp_to_ofport_map) {
5543 if (port->odp_port == odp_port) {
5544 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5549 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5554 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
5556 struct ofport_dpif *port;
5558 port = odp_port_to_ofport(ofproto->backer, odp_port);
5559 if (port && &ofproto->up == port->up.ofproto) {
5560 return port->up.ofp_port;
5567 ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
5568 const struct match *match, int priority,
5569 uint16_t idle_timeout,
5570 const struct ofpbuf *ofpacts,
5571 struct rule **rulep)
5573 struct ofproto_flow_mod ofm;
5574 struct rule_dpif *rule;
5577 ofm.fm.match = *match;
5578 ofm.fm.priority = priority;
5579 ofm.fm.new_cookie = htonll(0);
5580 ofm.fm.cookie = htonll(0);
5581 ofm.fm.cookie_mask = htonll(0);
5582 ofm.fm.modify_cookie = false;
5583 ofm.fm.table_id = TBL_INTERNAL;
5584 ofm.fm.command = OFPFC_ADD;
5585 ofm.fm.idle_timeout = idle_timeout;
5586 ofm.fm.hard_timeout = 0;
5587 ofm.fm.importance = 0;
5588 ofm.fm.buffer_id = 0;
5589 ofm.fm.out_port = 0;
5590 ofm.fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5591 ofm.fm.ofpacts = ofpacts->data;
5592 ofm.fm.ofpacts_len = ofpacts->size;
5594 error = ofproto_flow_mod(&ofproto->up, &ofm);
5596 VLOG_ERR_RL(&rl, "failed to add internal flow (%s)",
5597 ofperr_to_string(error));
5602 rule = rule_dpif_lookup_in_table(ofproto,
5603 ofproto_dpif_get_tables_version(ofproto),
5604 TBL_INTERNAL, &ofm.fm.match.flow,
5615 ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
5616 struct match *match, int priority)
5618 struct ofproto_flow_mod ofm;
5621 ofm.fm.match = *match;
5622 ofm.fm.priority = priority;
5623 ofm.fm.new_cookie = htonll(0);
5624 ofm.fm.cookie = htonll(0);
5625 ofm.fm.cookie_mask = htonll(0);
5626 ofm.fm.modify_cookie = false;
5627 ofm.fm.table_id = TBL_INTERNAL;
5628 ofm.fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5629 ofm.fm.command = OFPFC_DELETE_STRICT;
5631 error = ofproto_flow_mod(&ofproto->up, &ofm);
5633 VLOG_ERR_RL(&rl, "failed to delete internal flow (%s)",
5634 ofperr_to_string(error));
5641 const struct ofproto_class ofproto_dpif_class = {
5655 NULL, /* get_memory_usage. */
5656 type_get_memory_usage,
5675 port_is_lacp_current,
5676 port_get_lacp_stats,
5677 NULL, /* rule_choose_table */
5701 aa_vlan_get_queue_size,
5708 get_stp_port_status,
5713 get_rstp_port_status,
5720 is_mirror_output_bundle,
5721 forward_bpdu_changed,
5722 set_mac_table_config,
5724 set_mcast_snooping_port,
5726 NULL, /* meter_get_features */
5727 NULL, /* meter_set */
5728 NULL, /* meter_get */
5729 NULL, /* meter_del */
5730 group_alloc, /* group_alloc */
5731 group_construct, /* group_construct */
5732 group_destruct, /* group_destruct */
5733 group_dealloc, /* group_dealloc */
5734 group_modify, /* group_modify */
5735 group_get_stats, /* group_get_stats */
5736 get_datapath_version, /* get_datapath_version */