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)
1272 CHECK_FEATURE(ct_mark)
1273 CHECK_FEATURE__(ct_label, ct_label.u64.lo)
1275 #undef CHECK_FEATURE
1276 #undef CHECK_FEATURE__
1279 check_support(struct dpif_backer *backer)
1281 /* This feature needs to be tested after udpif threads are set. */
1282 backer->support.variable_length_userdata = false;
1284 backer->support.odp.recirc = check_recirc(backer);
1285 backer->support.odp.max_mpls_depth = check_max_mpls_depth(backer);
1286 backer->support.masked_set_action = check_masked_set_action(backer);
1287 backer->support.ufid = check_ufid(backer);
1288 backer->support.tnl_push_pop = dpif_supports_tnl_push_pop(backer->dpif);
1290 backer->support.odp.ct_state = check_ct_state(backer);
1291 backer->support.odp.ct_zone = check_ct_zone(backer);
1292 backer->support.odp.ct_mark = check_ct_mark(backer);
1293 backer->support.odp.ct_label = check_ct_label(backer);
1297 construct(struct ofproto *ofproto_)
1299 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1300 struct shash_node *node, *next;
1303 /* Tunnel module can get used right after the udpif threads are running. */
1304 ofproto_tunnel_init();
1306 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1311 atomic_init(&ofproto->tables_version, CLS_MIN_VERSION);
1312 ofproto->netflow = NULL;
1313 ofproto->sflow = NULL;
1314 ofproto->ipfix = NULL;
1315 ofproto->stp = NULL;
1316 ofproto->rstp = NULL;
1317 ofproto->dump_seq = 0;
1318 hmap_init(&ofproto->bundles);
1319 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1321 ofproto->mbridge = mbridge_create();
1322 ofproto->has_bonded_bundles = false;
1323 ofproto->lacp_enabled = false;
1324 ovs_mutex_init_adaptive(&ofproto->stats_mutex);
1325 ovs_mutex_init(&ofproto->vsp_mutex);
1327 guarded_list_init(&ofproto->pins);
1329 hmap_init(&ofproto->vlandev_map);
1330 hmap_init(&ofproto->realdev_vid_map);
1332 sset_init(&ofproto->ports);
1333 sset_init(&ofproto->ghost_ports);
1334 sset_init(&ofproto->port_poll_set);
1335 ofproto->port_poll_errno = 0;
1336 ofproto->change_seq = 0;
1337 ofproto->pins_seq = seq_create();
1338 ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1341 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1342 struct iface_hint *iface_hint = node->data;
1344 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1345 /* Check if the datapath already has this port. */
1346 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1347 sset_add(&ofproto->ports, node->name);
1350 free(iface_hint->br_name);
1351 free(iface_hint->br_type);
1353 shash_delete(&init_ofp_ports, node);
1357 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1358 hash_string(ofproto->up.name, 0));
1359 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1361 ofproto_init_tables(ofproto_, N_TABLES);
1362 error = add_internal_flows(ofproto);
1364 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1370 add_internal_miss_flow(struct ofproto_dpif *ofproto, int id,
1371 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1377 match_init_catchall(&match);
1378 match_set_reg(&match, 0, id);
1380 error = ofproto_dpif_add_internal_flow(ofproto, &match, 0, 0, ofpacts,
1382 *rulep = error ? NULL : rule_dpif_cast(rule);
1388 add_internal_flows(struct ofproto_dpif *ofproto)
1390 struct ofpact_controller *controller;
1391 uint64_t ofpacts_stub[128 / 8];
1392 struct ofpbuf ofpacts;
1393 struct rule *unused_rulep OVS_UNUSED;
1398 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1401 controller = ofpact_put_CONTROLLER(&ofpacts);
1402 controller->max_len = UINT16_MAX;
1403 controller->controller_id = 0;
1404 controller->reason = OFPR_NO_MATCH;
1405 ofpact_pad(&ofpacts);
1407 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1408 &ofproto->miss_rule);
1413 ofpbuf_clear(&ofpacts);
1414 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1415 &ofproto->no_packet_in_rule);
1420 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1421 &ofproto->drop_frags_rule);
1426 /* Drop any run away non-recirc rule lookups. Recirc_id has to be
1427 * zero when reaching this rule.
1429 * (priority=2), recirc_id=0, actions=drop
1431 ofpbuf_clear(&ofpacts);
1432 match_init_catchall(&match);
1433 match_set_recirc_id(&match, 0);
1434 error = ofproto_dpif_add_internal_flow(ofproto, &match, 2, 0, &ofpacts,
1440 destruct(struct ofproto *ofproto_)
1442 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1443 struct ofproto_packet_in *pin;
1444 struct rule_dpif *rule;
1445 struct oftable *table;
1446 struct ovs_list pins;
1448 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1450 xlate_remove_ofproto(ofproto);
1453 /* Ensure that the upcall processing threads have no remaining references
1454 * to the ofproto or anything in it. */
1455 udpif_synchronize(ofproto->backer->udpif);
1457 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1459 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1460 CLS_FOR_EACH (rule, up.cr, &table->cls) {
1461 ofproto_rule_delete(&ofproto->up, &rule->up);
1464 ofproto_group_delete_all(&ofproto->up);
1466 guarded_list_pop_all(&ofproto->pins, &pins);
1467 LIST_FOR_EACH_POP (pin, list_node, &pins) {
1468 free(CONST_CAST(void *, pin->up.packet));
1471 guarded_list_destroy(&ofproto->pins);
1473 recirc_free_ofproto(ofproto, ofproto->up.name);
1475 mbridge_unref(ofproto->mbridge);
1477 netflow_unref(ofproto->netflow);
1478 dpif_sflow_unref(ofproto->sflow);
1479 dpif_ipfix_unref(ofproto->ipfix);
1480 hmap_destroy(&ofproto->bundles);
1481 mac_learning_unref(ofproto->ml);
1482 mcast_snooping_unref(ofproto->ms);
1484 hmap_destroy(&ofproto->vlandev_map);
1485 hmap_destroy(&ofproto->realdev_vid_map);
1487 sset_destroy(&ofproto->ports);
1488 sset_destroy(&ofproto->ghost_ports);
1489 sset_destroy(&ofproto->port_poll_set);
1491 ovs_mutex_destroy(&ofproto->stats_mutex);
1492 ovs_mutex_destroy(&ofproto->vsp_mutex);
1494 seq_destroy(ofproto->pins_seq);
1496 close_dpif_backer(ofproto->backer);
1500 run(struct ofproto *ofproto_)
1502 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1503 uint64_t new_seq, new_dump_seq;
1505 if (mbridge_need_revalidate(ofproto->mbridge)) {
1506 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1507 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1508 mac_learning_flush(ofproto->ml);
1509 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1510 mcast_snooping_mdb_flush(ofproto->ms);
1513 /* Always updates the ofproto->pins_seqno to avoid frequent wakeup during
1514 * flow restore. Even though nothing is processed during flow restore,
1515 * all queued 'pins' will be handled immediately when flow restore
1517 ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1519 /* Do not perform any periodic activity required by 'ofproto' while
1520 * waiting for flow restore to complete. */
1521 if (!ofproto_get_flow_restore_wait()) {
1522 struct ofproto_packet_in *pin;
1523 struct ovs_list pins;
1525 guarded_list_pop_all(&ofproto->pins, &pins);
1526 LIST_FOR_EACH_POP (pin, list_node, &pins) {
1527 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1528 free(CONST_CAST(void *, pin->up.packet));
1533 if (ofproto->netflow) {
1534 netflow_run(ofproto->netflow);
1536 if (ofproto->sflow) {
1537 dpif_sflow_run(ofproto->sflow);
1539 if (ofproto->ipfix) {
1540 dpif_ipfix_run(ofproto->ipfix);
1543 new_seq = seq_read(connectivity_seq_get());
1544 if (ofproto->change_seq != new_seq) {
1545 struct ofport_dpif *ofport;
1547 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1551 ofproto->change_seq = new_seq;
1553 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1554 struct ofbundle *bundle;
1556 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1563 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1564 if (mac_learning_run(ofproto->ml)) {
1565 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1567 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1569 if (mcast_snooping_run(ofproto->ms)) {
1570 ofproto->backer->need_revalidate = REV_MCAST_SNOOPING;
1573 new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
1574 if (ofproto->dump_seq != new_dump_seq) {
1575 struct rule *rule, *next_rule;
1577 /* We know stats are relatively fresh, so now is a good time to do some
1579 ofproto->dump_seq = new_dump_seq;
1581 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1583 ovs_mutex_lock(&ofproto_mutex);
1584 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
1585 &ofproto->up.expirable) {
1586 rule_expire(rule_dpif_cast(rule));
1588 ovs_mutex_unlock(&ofproto_mutex);
1590 /* All outstanding data in existing flows has been accounted, so it's a
1591 * good time to do bond rebalancing. */
1592 if (ofproto->has_bonded_bundles) {
1593 struct ofbundle *bundle;
1595 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1597 bond_rebalance(bundle->bond);
1606 wait(struct ofproto *ofproto_)
1608 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1610 if (ofproto_get_flow_restore_wait()) {
1614 if (ofproto->sflow) {
1615 dpif_sflow_wait(ofproto->sflow);
1617 if (ofproto->ipfix) {
1618 dpif_ipfix_wait(ofproto->ipfix);
1620 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1621 struct ofbundle *bundle;
1623 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1624 bundle_wait(bundle);
1627 if (ofproto->netflow) {
1628 netflow_wait(ofproto->netflow);
1630 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1631 mac_learning_wait(ofproto->ml);
1632 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1633 mcast_snooping_wait(ofproto->ms);
1635 if (ofproto->backer->need_revalidate) {
1636 /* Shouldn't happen, but if it does just go around again. */
1637 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1638 poll_immediate_wake();
1641 seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
1642 seq_wait(ofproto->pins_seq, ofproto->pins_seqno);
1646 type_get_memory_usage(const char *type, struct simap *usage)
1648 struct dpif_backer *backer;
1650 backer = shash_find_data(&all_dpif_backers, type);
1652 udpif_get_memory_usage(backer->udpif, usage);
1657 flush(struct ofproto *ofproto_)
1659 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1660 struct dpif_backer *backer = ofproto->backer;
1663 udpif_flush(backer->udpif);
1668 query_tables(struct ofproto *ofproto,
1669 struct ofputil_table_features *features,
1670 struct ofputil_table_stats *stats)
1672 strcpy(features->name, "classifier");
1677 for (i = 0; i < ofproto->n_tables; i++) {
1678 unsigned long missed, matched;
1680 atomic_read_relaxed(&ofproto->tables[i].n_matched, &matched);
1681 atomic_read_relaxed(&ofproto->tables[i].n_missed, &missed);
1683 stats[i].matched_count = matched;
1684 stats[i].lookup_count = matched + missed;
1690 set_tables_version(struct ofproto *ofproto_, cls_version_t version)
1692 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1694 atomic_store_relaxed(&ofproto->tables_version, version);
1698 static struct ofport *
1701 struct ofport_dpif *port = xzalloc(sizeof *port);
1706 port_dealloc(struct ofport *port_)
1708 struct ofport_dpif *port = ofport_dpif_cast(port_);
1713 port_construct(struct ofport *port_)
1715 struct ofport_dpif *port = ofport_dpif_cast(port_);
1716 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1717 const struct netdev *netdev = port->up.netdev;
1718 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1719 const char *dp_port_name;
1720 struct dpif_port dpif_port;
1723 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1724 port->bundle = NULL;
1728 port->may_enable = false;
1729 port->stp_port = NULL;
1730 port->stp_state = STP_DISABLED;
1731 port->rstp_port = NULL;
1732 port->rstp_state = RSTP_DISABLED;
1733 port->is_tunnel = false;
1737 port->realdev_ofp_port = 0;
1738 port->vlandev_vid = 0;
1739 port->carrier_seq = netdev_get_carrier_resets(netdev);
1740 port->is_layer3 = netdev_vport_is_layer3(netdev);
1742 if (netdev_vport_is_patch(netdev)) {
1743 /* By bailing out here, we don't submit the port to the sFlow module
1744 * to be considered for counter polling export. This is correct
1745 * because the patch port represents an interface that sFlow considers
1746 * to be "internal" to the switch as a whole, and therefore not a
1747 * candidate for counter polling. */
1748 port->odp_port = ODPP_NONE;
1749 ofport_update_peer(port);
1753 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
1754 error = dpif_port_query_by_name(ofproto->backer->dpif, dp_port_name,
1760 port->odp_port = dpif_port.port_no;
1762 if (netdev_get_tunnel_config(netdev)) {
1763 atomic_count_inc(&ofproto->backer->tnl_count);
1764 error = tnl_port_add(port, port->up.netdev, port->odp_port,
1765 ovs_native_tunneling_is_on(ofproto), dp_port_name);
1767 atomic_count_dec(&ofproto->backer->tnl_count);
1768 dpif_port_destroy(&dpif_port);
1772 port->is_tunnel = true;
1773 if (ofproto->ipfix) {
1774 dpif_ipfix_add_tunnel_port(ofproto->ipfix, port_, port->odp_port);
1777 /* Sanity-check that a mapping doesn't already exist. This
1778 * shouldn't happen for non-tunnel ports. */
1779 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1780 VLOG_ERR("port %s already has an OpenFlow port number",
1782 dpif_port_destroy(&dpif_port);
1786 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1787 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1788 hash_odp_port(port->odp_port));
1789 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1791 dpif_port_destroy(&dpif_port);
1793 if (ofproto->sflow) {
1794 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1801 port_destruct(struct ofport *port_)
1803 struct ofport_dpif *port = ofport_dpif_cast(port_);
1804 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1805 const char *devname = netdev_get_name(port->up.netdev);
1806 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1807 const char *dp_port_name;
1809 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1811 xlate_ofport_remove(port);
1814 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1816 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1817 /* The underlying device is still there, so delete it. This
1818 * happens when the ofproto is being destroyed, since the caller
1819 * assumes that removal of attached ports will happen as part of
1821 if (!port->is_tunnel) {
1822 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1827 port->peer->peer = NULL;
1831 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1832 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1833 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1834 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1837 if (port->is_tunnel) {
1838 atomic_count_dec(&ofproto->backer->tnl_count);
1841 if (port->is_tunnel && ofproto->ipfix) {
1842 dpif_ipfix_del_tunnel_port(ofproto->ipfix, port->odp_port);
1846 sset_find_and_delete(&ofproto->ports, devname);
1847 sset_find_and_delete(&ofproto->ghost_ports, devname);
1848 bundle_remove(port_);
1849 set_cfm(port_, NULL);
1850 set_bfd(port_, NULL);
1851 set_lldp(port_, NULL);
1852 if (port->stp_port) {
1853 stp_port_disable(port->stp_port);
1855 set_rstp_port(port_, NULL);
1856 if (ofproto->sflow) {
1857 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1864 port_modified(struct ofport *port_)
1866 struct ofport_dpif *port = ofport_dpif_cast(port_);
1867 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1868 const char *dp_port_name;
1869 struct netdev *netdev = port->up.netdev;
1871 if (port->bundle && port->bundle->bond) {
1872 bond_slave_set_netdev(port->bundle->bond, port, netdev);
1876 cfm_set_netdev(port->cfm, netdev);
1880 bfd_set_netdev(port->bfd, netdev);
1883 ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1884 port->lldp, &port->up.pp.hw_addr);
1886 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
1888 if (port->is_tunnel) {
1889 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1891 if (tnl_port_reconfigure(port, netdev, port->odp_port,
1892 ovs_native_tunneling_is_on(ofproto),
1894 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1898 ofport_update_peer(port);
1902 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1904 struct ofport_dpif *port = ofport_dpif_cast(port_);
1905 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1906 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1908 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1909 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1910 OFPUTIL_PC_NO_PACKET_IN)) {
1911 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1913 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1914 bundle_update(port->bundle);
1920 set_sflow(struct ofproto *ofproto_,
1921 const struct ofproto_sflow_options *sflow_options)
1923 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1924 struct dpif_sflow *ds = ofproto->sflow;
1926 if (sflow_options) {
1927 uint32_t old_probability = ds ? dpif_sflow_get_probability(ds) : 0;
1929 struct ofport_dpif *ofport;
1931 ds = ofproto->sflow = dpif_sflow_create();
1932 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1933 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1936 dpif_sflow_set_options(ds, sflow_options);
1937 if (dpif_sflow_get_probability(ds) != old_probability) {
1938 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1942 dpif_sflow_unref(ds);
1943 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1944 ofproto->sflow = NULL;
1952 struct ofproto *ofproto_,
1953 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1954 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1955 size_t n_flow_exporters_options)
1957 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1958 struct dpif_ipfix *di = ofproto->ipfix;
1959 bool has_options = bridge_exporter_options || flow_exporters_options;
1960 bool new_di = false;
1962 if (has_options && !di) {
1963 di = ofproto->ipfix = dpif_ipfix_create();
1968 /* Call set_options in any case to cleanly flush the flow
1969 * caches in the last exporters that are to be destroyed. */
1970 dpif_ipfix_set_options(
1971 di, bridge_exporter_options, flow_exporters_options,
1972 n_flow_exporters_options);
1974 /* Add tunnel ports only when a new ipfix created */
1975 if (new_di == true) {
1976 struct ofport_dpif *ofport;
1977 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1978 if (ofport->is_tunnel == true) {
1979 dpif_ipfix_add_tunnel_port(di, &ofport->up, ofport->odp_port);
1985 dpif_ipfix_unref(di);
1986 ofproto->ipfix = NULL;
1994 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1996 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1997 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1998 struct cfm *old = ofport->cfm;
2003 ofport->cfm = cfm_create(ofport->up.netdev);
2006 if (cfm_configure(ofport->cfm, s)) {
2013 cfm_unref(ofport->cfm);
2016 if (ofport->cfm != old) {
2017 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2019 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
2020 ofport->lldp, &ofport->up.pp.hw_addr);
2025 cfm_status_changed(struct ofport *ofport_)
2027 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2029 return ofport->cfm ? cfm_check_status_change(ofport->cfm) : true;
2033 get_cfm_status(const struct ofport *ofport_,
2034 struct cfm_status *status)
2036 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2040 cfm_get_status(ofport->cfm, status);
2049 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2051 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2052 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2056 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
2057 cfg, ofport->up.netdev);
2058 if (ofport->bfd != old) {
2059 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2061 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
2062 ofport->lldp, &ofport->up.pp.hw_addr);
2067 bfd_status_changed(struct ofport *ofport_)
2069 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2071 return ofport->bfd ? bfd_check_status_change(ofport->bfd) : true;
2075 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2077 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2081 bfd_get_status(ofport->bfd, smap);
2090 set_lldp(struct ofport *ofport_,
2091 const struct smap *cfg)
2093 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2097 if (!ofport->lldp) {
2098 struct ofproto_dpif *ofproto;
2100 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2101 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2102 ofport->lldp = lldp_create(ofport->up.netdev, ofport_->mtu, cfg);
2105 if (!lldp_configure(ofport->lldp, cfg)) {
2110 lldp_unref(ofport->lldp);
2111 ofport->lldp = NULL;
2114 ofproto_dpif_monitor_port_update(ofport,
2118 &ofport->up.pp.hw_addr);
2123 get_lldp_status(const struct ofport *ofport_,
2124 struct lldp_status *status OVS_UNUSED)
2126 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2128 return ofport->lldp ? true : false;
2132 set_aa(struct ofproto *ofproto OVS_UNUSED,
2133 const struct aa_settings *s)
2135 return aa_configure(s);
2139 aa_mapping_set(struct ofproto *ofproto_ OVS_UNUSED, void *aux,
2140 const struct aa_mapping_settings *s)
2142 return aa_mapping_register(aux, s);
2146 aa_mapping_unset(struct ofproto *ofproto OVS_UNUSED, void *aux)
2148 return aa_mapping_unregister(aux);
2152 aa_vlan_get_queued(struct ofproto *ofproto OVS_UNUSED, struct ovs_list *list)
2154 return aa_get_vlan_queued(list);
2158 aa_vlan_get_queue_size(struct ofproto *ofproto OVS_UNUSED)
2160 return aa_get_vlan_queue_size();
2164 /* Spanning Tree. */
2166 /* Called while rstp_mutex is held. */
2168 rstp_send_bpdu_cb(struct dp_packet *pkt, void *ofport_, void *ofproto_)
2170 struct ofproto_dpif *ofproto = ofproto_;
2171 struct ofport_dpif *ofport = ofport_;
2172 struct eth_header *eth = dp_packet_l2(pkt);
2174 netdev_get_etheraddr(ofport->up.netdev, ð->eth_src);
2175 if (eth_addr_is_zero(eth->eth_src)) {
2176 VLOG_WARN_RL(&rl, "%s port %d: cannot send RSTP BPDU on a port which "
2177 "does not have a configured source MAC address.",
2178 ofproto->up.name, ofp_to_u16(ofport->up.ofp_port));
2180 ofproto_dpif_send_packet(ofport, pkt);
2182 dp_packet_delete(pkt);
2186 send_bpdu_cb(struct dp_packet *pkt, int port_num, void *ofproto_)
2188 struct ofproto_dpif *ofproto = ofproto_;
2189 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2190 struct ofport_dpif *ofport;
2192 ofport = stp_port_get_aux(sp);
2194 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2195 ofproto->up.name, port_num);
2197 struct eth_header *eth = dp_packet_l2(pkt);
2199 netdev_get_etheraddr(ofport->up.netdev, ð->eth_src);
2200 if (eth_addr_is_zero(eth->eth_src)) {
2201 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2202 "with unknown MAC", ofproto->up.name, port_num);
2204 ofproto_dpif_send_packet(ofport, pkt);
2207 dp_packet_delete(pkt);
2210 /* Configure RSTP on 'ofproto_' using the settings defined in 's'. */
2212 set_rstp(struct ofproto *ofproto_, const struct ofproto_rstp_settings *s)
2214 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2216 /* Only revalidate flows if the configuration changed. */
2217 if (!s != !ofproto->rstp) {
2218 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2222 if (!ofproto->rstp) {
2223 ofproto->rstp = rstp_create(ofproto_->name, s->address,
2224 rstp_send_bpdu_cb, ofproto);
2225 ofproto->rstp_last_tick = time_msec();
2227 rstp_set_bridge_address(ofproto->rstp, s->address);
2228 rstp_set_bridge_priority(ofproto->rstp, s->priority);
2229 rstp_set_bridge_ageing_time(ofproto->rstp, s->ageing_time);
2230 rstp_set_bridge_force_protocol_version(ofproto->rstp,
2231 s->force_protocol_version);
2232 rstp_set_bridge_max_age(ofproto->rstp, s->bridge_max_age);
2233 rstp_set_bridge_forward_delay(ofproto->rstp, s->bridge_forward_delay);
2234 rstp_set_bridge_transmit_hold_count(ofproto->rstp,
2235 s->transmit_hold_count);
2237 struct ofport *ofport;
2238 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2239 set_rstp_port(ofport, NULL);
2241 rstp_unref(ofproto->rstp);
2242 ofproto->rstp = NULL;
2247 get_rstp_status(struct ofproto *ofproto_, struct ofproto_rstp_status *s)
2249 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2251 if (ofproto->rstp) {
2253 s->root_id = rstp_get_root_id(ofproto->rstp);
2254 s->bridge_id = rstp_get_bridge_id(ofproto->rstp);
2255 s->designated_id = rstp_get_designated_id(ofproto->rstp);
2256 s->root_path_cost = rstp_get_root_path_cost(ofproto->rstp);
2257 s->designated_port_id = rstp_get_designated_port_id(ofproto->rstp);
2258 s->bridge_port_id = rstp_get_bridge_port_id(ofproto->rstp);
2265 update_rstp_port_state(struct ofport_dpif *ofport)
2267 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2268 enum rstp_state state;
2270 /* Figure out new state. */
2271 state = ofport->rstp_port ? rstp_port_get_state(ofport->rstp_port)
2275 if (ofport->rstp_state != state) {
2276 enum ofputil_port_state of_state;
2279 VLOG_DBG("port %s: RSTP state changed from %s to %s",
2280 netdev_get_name(ofport->up.netdev),
2281 rstp_state_name(ofport->rstp_state),
2282 rstp_state_name(state));
2284 if (rstp_learn_in_state(ofport->rstp_state)
2285 != rstp_learn_in_state(state)) {
2286 /* XXX: Learning action flows should also be flushed. */
2287 if (ofport->bundle) {
2288 if (!rstp_shift_root_learned_address(ofproto->rstp)
2289 || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2290 bundle_flush_macs(ofport->bundle, false);
2294 fwd_change = rstp_forward_in_state(ofport->rstp_state)
2295 != rstp_forward_in_state(state);
2297 ofproto->backer->need_revalidate = REV_RSTP;
2298 ofport->rstp_state = state;
2300 if (fwd_change && ofport->bundle) {
2301 bundle_update(ofport->bundle);
2304 /* Update the RSTP state bits in the OpenFlow port description. */
2305 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2306 of_state |= (state == RSTP_LEARNING ? OFPUTIL_PS_STP_LEARN
2307 : state == RSTP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2308 : state == RSTP_DISCARDING ? OFPUTIL_PS_STP_LISTEN
2310 ofproto_port_set_state(&ofport->up, of_state);
2315 rstp_run(struct ofproto_dpif *ofproto)
2317 if (ofproto->rstp) {
2318 long long int now = time_msec();
2319 long long int elapsed = now - ofproto->rstp_last_tick;
2320 struct rstp_port *rp;
2321 struct ofport_dpif *ofport;
2323 /* Every second, decrease the values of the timers. */
2324 if (elapsed >= 1000) {
2325 rstp_tick_timers(ofproto->rstp);
2326 ofproto->rstp_last_tick = now;
2329 while ((ofport = rstp_get_next_changed_port_aux(ofproto->rstp, &rp))) {
2330 update_rstp_port_state(ofport);
2334 /* FIXME: This check should be done on-event (i.e., when setting
2335 * p->fdb_flush) and not periodically.
2337 while ((ofport = rstp_check_and_reset_fdb_flush(ofproto->rstp, &rp))) {
2338 if (!rstp_shift_root_learned_address(ofproto->rstp)
2339 || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2340 bundle_flush_macs(ofport->bundle, false);
2344 if (rstp_shift_root_learned_address(ofproto->rstp)) {
2345 struct ofport_dpif *old_root_aux =
2346 (struct ofport_dpif *)rstp_get_old_root_aux(ofproto->rstp);
2347 struct ofport_dpif *new_root_aux =
2348 (struct ofport_dpif *)rstp_get_new_root_aux(ofproto->rstp);
2349 if (old_root_aux != NULL && new_root_aux != NULL) {
2350 bundle_move(old_root_aux->bundle, new_root_aux->bundle);
2351 rstp_reset_root_changed(ofproto->rstp);
2357 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2359 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2361 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2363 /* Only revalidate flows if the configuration changed. */
2364 if (!s != !ofproto->stp) {
2365 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2369 if (!ofproto->stp) {
2370 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2371 send_bpdu_cb, ofproto);
2372 ofproto->stp_last_tick = time_msec();
2375 stp_set_bridge_id(ofproto->stp, s->system_id);
2376 stp_set_bridge_priority(ofproto->stp, s->priority);
2377 stp_set_hello_time(ofproto->stp, s->hello_time);
2378 stp_set_max_age(ofproto->stp, s->max_age);
2379 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2381 struct ofport *ofport;
2383 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2384 set_stp_port(ofport, NULL);
2387 stp_unref(ofproto->stp);
2388 ofproto->stp = NULL;
2395 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2397 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2401 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2402 s->designated_root = stp_get_designated_root(ofproto->stp);
2403 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2412 update_stp_port_state(struct ofport_dpif *ofport)
2414 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2415 enum stp_state state;
2417 /* Figure out new state. */
2418 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2422 if (ofport->stp_state != state) {
2423 enum ofputil_port_state of_state;
2426 VLOG_DBG("port %s: STP state changed from %s to %s",
2427 netdev_get_name(ofport->up.netdev),
2428 stp_state_name(ofport->stp_state),
2429 stp_state_name(state));
2430 if (stp_learn_in_state(ofport->stp_state)
2431 != stp_learn_in_state(state)) {
2432 /* xxx Learning action flows should also be flushed. */
2433 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2434 mac_learning_flush(ofproto->ml);
2435 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2436 mcast_snooping_mdb_flush(ofproto->ms);
2438 fwd_change = stp_forward_in_state(ofport->stp_state)
2439 != stp_forward_in_state(state);
2441 ofproto->backer->need_revalidate = REV_STP;
2442 ofport->stp_state = state;
2443 ofport->stp_state_entered = time_msec();
2445 if (fwd_change && ofport->bundle) {
2446 bundle_update(ofport->bundle);
2449 /* Update the STP state bits in the OpenFlow port description. */
2450 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2451 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2452 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2453 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2454 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2456 ofproto_port_set_state(&ofport->up, of_state);
2460 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2461 * caller is responsible for assigning STP port numbers and ensuring
2462 * there are no duplicates. */
2464 set_stp_port(struct ofport *ofport_,
2465 const struct ofproto_port_stp_settings *s)
2467 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2468 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2469 struct stp_port *sp = ofport->stp_port;
2471 if (!s || !s->enable) {
2473 ofport->stp_port = NULL;
2474 stp_port_disable(sp);
2475 update_stp_port_state(ofport);
2478 } else if (sp && stp_port_no(sp) != s->port_num
2479 && ofport == stp_port_get_aux(sp)) {
2480 /* The port-id changed, so disable the old one if it's not
2481 * already in use by another port. */
2482 stp_port_disable(sp);
2485 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2487 /* Set name before enabling the port so that debugging messages can print
2489 stp_port_set_name(sp, netdev_get_name(ofport->up.netdev));
2490 stp_port_enable(sp);
2492 stp_port_set_aux(sp, ofport);
2493 stp_port_set_priority(sp, s->priority);
2494 stp_port_set_path_cost(sp, s->path_cost);
2496 update_stp_port_state(ofport);
2502 get_stp_port_status(struct ofport *ofport_,
2503 struct ofproto_port_stp_status *s)
2505 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2506 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2507 struct stp_port *sp = ofport->stp_port;
2509 if (!ofproto->stp || !sp) {
2515 s->port_id = stp_port_get_id(sp);
2516 s->state = stp_port_get_state(sp);
2517 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2518 s->role = stp_port_get_role(sp);
2524 get_stp_port_stats(struct ofport *ofport_,
2525 struct ofproto_port_stp_stats *s)
2527 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2528 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2529 struct stp_port *sp = ofport->stp_port;
2531 if (!ofproto->stp || !sp) {
2537 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2543 stp_run(struct ofproto_dpif *ofproto)
2546 long long int now = time_msec();
2547 long long int elapsed = now - ofproto->stp_last_tick;
2548 struct stp_port *sp;
2551 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2552 ofproto->stp_last_tick = now;
2554 while (stp_get_changed_port(ofproto->stp, &sp)) {
2555 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2558 update_stp_port_state(ofport);
2562 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2563 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2564 mac_learning_flush(ofproto->ml);
2565 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2566 mcast_snooping_mdb_flush(ofproto->ms);
2572 stp_wait(struct ofproto_dpif *ofproto)
2575 poll_timer_wait(1000);
2579 /* Configures RSTP on 'ofport_' using the settings defined in 's'. The
2580 * caller is responsible for assigning RSTP port numbers and ensuring
2581 * there are no duplicates. */
2583 set_rstp_port(struct ofport *ofport_,
2584 const struct ofproto_port_rstp_settings *s)
2586 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2587 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2588 struct rstp_port *rp = ofport->rstp_port;
2590 if (!s || !s->enable) {
2592 rstp_port_set_aux(rp, NULL);
2593 rstp_port_set_state(rp, RSTP_DISABLED);
2594 rstp_port_set_mac_operational(rp, false);
2595 ofport->rstp_port = NULL;
2596 rstp_port_unref(rp);
2597 update_rstp_port_state(ofport);
2602 /* Check if need to add a new port. */
2604 rp = ofport->rstp_port = rstp_add_port(ofproto->rstp);
2607 rstp_port_set(rp, s->port_num, s->priority, s->path_cost,
2608 s->admin_edge_port, s->auto_edge,
2609 s->admin_p2p_mac_state, s->admin_port_state, s->mcheck,
2611 update_rstp_port_state(ofport);
2612 /* Synchronize operational status. */
2613 rstp_port_set_mac_operational(rp, ofport->may_enable);
2617 get_rstp_port_status(struct ofport *ofport_,
2618 struct ofproto_port_rstp_status *s)
2620 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2621 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2622 struct rstp_port *rp = ofport->rstp_port;
2624 if (!ofproto->rstp || !rp) {
2630 rstp_port_get_status(rp, &s->port_id, &s->state, &s->role,
2631 &s->designated_bridge_id, &s->designated_port_id,
2632 &s->designated_path_cost, &s->tx_count,
2633 &s->rx_count, &s->error_count, &s->uptime);
2638 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2641 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2642 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2644 if (ofport->n_qdscp != n_qdscp
2645 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2646 n_qdscp * sizeof *qdscp))) {
2647 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2648 free(ofport->qdscp);
2649 ofport->qdscp = n_qdscp
2650 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2652 ofport->n_qdscp = n_qdscp;
2660 /* Expires all MAC learning entries associated with 'bundle' and forces its
2661 * ofproto to revalidate every flow.
2663 * Normally MAC learning entries are removed only from the ofproto associated
2664 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2665 * are removed from every ofproto. When patch ports and SLB bonds are in use
2666 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2667 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2668 * with the host from which it migrated. */
2670 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2672 struct ofproto_dpif *ofproto = bundle->ofproto;
2673 struct mac_learning *ml = ofproto->ml;
2674 struct mac_entry *mac, *next_mac;
2676 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2677 ovs_rwlock_wrlock(&ml->rwlock);
2678 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2679 if (mac_entry_get_port(ml, mac) == bundle) {
2681 struct ofproto_dpif *o;
2683 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2685 struct mac_entry *e;
2687 ovs_rwlock_wrlock(&o->ml->rwlock);
2688 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2690 mac_learning_expire(o->ml, e);
2692 ovs_rwlock_unlock(&o->ml->rwlock);
2697 mac_learning_expire(ml, mac);
2700 ovs_rwlock_unlock(&ml->rwlock);
2704 bundle_move(struct ofbundle *old, struct ofbundle *new)
2706 struct ofproto_dpif *ofproto = old->ofproto;
2707 struct mac_learning *ml = ofproto->ml;
2708 struct mac_entry *mac, *next_mac;
2710 ovs_assert(new->ofproto == old->ofproto);
2712 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2713 ovs_rwlock_wrlock(&ml->rwlock);
2714 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2715 if (mac_entry_get_port(ml, mac) == old) {
2716 mac_entry_set_port(ml, mac, new);
2719 ovs_rwlock_unlock(&ml->rwlock);
2722 static struct ofbundle *
2723 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2725 struct ofbundle *bundle;
2727 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2728 &ofproto->bundles) {
2729 if (bundle->aux == aux) {
2737 bundle_update(struct ofbundle *bundle)
2739 struct ofport_dpif *port;
2741 bundle->floodable = true;
2742 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2743 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2745 || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2746 || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2747 bundle->floodable = false;
2754 bundle_del_port(struct ofport_dpif *port)
2756 struct ofbundle *bundle = port->bundle;
2758 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2760 list_remove(&port->bundle_node);
2761 port->bundle = NULL;
2764 lacp_slave_unregister(bundle->lacp, port);
2767 bond_slave_unregister(bundle->bond, port);
2770 bundle_update(bundle);
2774 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2775 struct lacp_slave_settings *lacp)
2777 struct ofport_dpif *port;
2779 port = ofp_port_to_ofport(bundle->ofproto, ofp_port);
2784 if (port->bundle != bundle) {
2785 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2787 bundle_remove(&port->up);
2790 port->bundle = bundle;
2791 list_push_back(&bundle->ports, &port->bundle_node);
2792 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2794 || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2795 || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2796 bundle->floodable = false;
2800 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2801 lacp_slave_register(bundle->lacp, port, lacp);
2808 bundle_destroy(struct ofbundle *bundle)
2810 struct ofproto_dpif *ofproto;
2811 struct ofport_dpif *port, *next_port;
2817 ofproto = bundle->ofproto;
2818 mbridge_unregister_bundle(ofproto->mbridge, bundle);
2821 xlate_bundle_remove(bundle);
2824 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2825 bundle_del_port(port);
2828 bundle_flush_macs(bundle, true);
2829 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2831 free(bundle->trunks);
2832 lacp_unref(bundle->lacp);
2833 bond_unref(bundle->bond);
2838 bundle_set(struct ofproto *ofproto_, void *aux,
2839 const struct ofproto_bundle_settings *s)
2841 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2842 bool need_flush = false;
2843 struct ofport_dpif *port;
2844 struct ofbundle *bundle;
2845 unsigned long *trunks;
2851 bundle_destroy(bundle_lookup(ofproto, aux));
2855 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2856 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2858 bundle = bundle_lookup(ofproto, aux);
2860 bundle = xmalloc(sizeof *bundle);
2862 bundle->ofproto = ofproto;
2863 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2864 hash_pointer(aux, 0));
2866 bundle->name = NULL;
2868 list_init(&bundle->ports);
2869 bundle->vlan_mode = PORT_VLAN_TRUNK;
2871 bundle->trunks = NULL;
2872 bundle->use_priority_tags = s->use_priority_tags;
2873 bundle->lacp = NULL;
2874 bundle->bond = NULL;
2876 bundle->floodable = true;
2877 mbridge_register_bundle(ofproto->mbridge, bundle);
2880 if (!bundle->name || strcmp(s->name, bundle->name)) {
2882 bundle->name = xstrdup(s->name);
2887 ofproto->lacp_enabled = true;
2888 if (!bundle->lacp) {
2889 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2890 bundle->lacp = lacp_create();
2892 lacp_configure(bundle->lacp, s->lacp);
2894 lacp_unref(bundle->lacp);
2895 bundle->lacp = NULL;
2898 /* Update set of ports. */
2900 for (i = 0; i < s->n_slaves; i++) {
2901 if (!bundle_add_port(bundle, s->slaves[i],
2902 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2906 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2907 struct ofport_dpif *next_port;
2909 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2910 for (i = 0; i < s->n_slaves; i++) {
2911 if (s->slaves[i] == port->up.ofp_port) {
2916 bundle_del_port(port);
2920 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2922 if (list_is_empty(&bundle->ports)) {
2923 bundle_destroy(bundle);
2927 /* Set VLAN tagging mode */
2928 if (s->vlan_mode != bundle->vlan_mode
2929 || s->use_priority_tags != bundle->use_priority_tags) {
2930 bundle->vlan_mode = s->vlan_mode;
2931 bundle->use_priority_tags = s->use_priority_tags;
2936 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2937 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2939 if (vlan != bundle->vlan) {
2940 bundle->vlan = vlan;
2944 /* Get trunked VLANs. */
2945 switch (s->vlan_mode) {
2946 case PORT_VLAN_ACCESS:
2950 case PORT_VLAN_TRUNK:
2951 trunks = CONST_CAST(unsigned long *, s->trunks);
2954 case PORT_VLAN_NATIVE_UNTAGGED:
2955 case PORT_VLAN_NATIVE_TAGGED:
2956 if (vlan != 0 && (!s->trunks
2957 || !bitmap_is_set(s->trunks, vlan)
2958 || bitmap_is_set(s->trunks, 0))) {
2959 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2961 trunks = bitmap_clone(s->trunks, 4096);
2963 trunks = bitmap_allocate1(4096);
2965 bitmap_set1(trunks, vlan);
2966 bitmap_set0(trunks, 0);
2968 trunks = CONST_CAST(unsigned long *, s->trunks);
2975 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2976 free(bundle->trunks);
2977 if (trunks == s->trunks) {
2978 bundle->trunks = vlan_bitmap_clone(trunks);
2980 bundle->trunks = trunks;
2985 if (trunks != s->trunks) {
2990 if (!list_is_short(&bundle->ports)) {
2991 bundle->ofproto->has_bonded_bundles = true;
2993 if (bond_reconfigure(bundle->bond, s->bond)) {
2994 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2997 bundle->bond = bond_create(s->bond, ofproto);
2998 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3001 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
3002 bond_slave_register(bundle->bond, port,
3003 port->up.ofp_port, port->up.netdev);
3006 bond_unref(bundle->bond);
3007 bundle->bond = NULL;
3010 /* If we changed something that would affect MAC learning, un-learn
3011 * everything on this port and force flow revalidation. */
3013 bundle_flush_macs(bundle, false);
3020 bundle_remove(struct ofport *port_)
3022 struct ofport_dpif *port = ofport_dpif_cast(port_);
3023 struct ofbundle *bundle = port->bundle;
3026 bundle_del_port(port);
3027 if (list_is_empty(&bundle->ports)) {
3028 bundle_destroy(bundle);
3029 } else if (list_is_short(&bundle->ports)) {
3030 bond_unref(bundle->bond);
3031 bundle->bond = NULL;
3037 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
3039 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3040 struct ofport_dpif *port = port_;
3044 error = netdev_get_etheraddr(port->up.netdev, &ea);
3046 struct dp_packet packet;
3049 dp_packet_init(&packet, 0);
3050 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
3052 memcpy(packet_pdu, pdu, pdu_size);
3054 ofproto_dpif_send_packet(port, &packet);
3055 dp_packet_uninit(&packet);
3057 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
3058 "%s (%s)", port->bundle->name,
3059 netdev_get_name(port->up.netdev), ovs_strerror(error));
3064 bundle_send_learning_packets(struct ofbundle *bundle)
3066 struct ofproto_dpif *ofproto = bundle->ofproto;
3067 int error, n_packets, n_errors;
3068 struct mac_entry *e;
3070 struct ovs_list list_node;
3071 struct ofport_dpif *port;
3072 struct dp_packet *pkt;
3074 struct ovs_list packets;
3076 list_init(&packets);
3077 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
3078 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3079 if (mac_entry_get_port(ofproto->ml, e) != bundle) {
3080 pkt_node = xmalloc(sizeof *pkt_node);
3081 pkt_node->pkt = bond_compose_learning_packet(bundle->bond,
3083 (void **)&pkt_node->port);
3084 list_push_back(&packets, &pkt_node->list_node);
3087 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3089 error = n_packets = n_errors = 0;
3090 LIST_FOR_EACH_POP (pkt_node, list_node, &packets) {
3093 ret = ofproto_dpif_send_packet(pkt_node->port, pkt_node->pkt);
3094 dp_packet_delete(pkt_node->pkt);
3104 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3105 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3106 "packets, last error was: %s",
3107 bundle->name, n_errors, n_packets, ovs_strerror(error));
3109 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3110 bundle->name, n_packets);
3115 bundle_run(struct ofbundle *bundle)
3118 lacp_run(bundle->lacp, send_pdu_cb);
3121 struct ofport_dpif *port;
3123 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
3124 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
3127 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
3128 bundle->ofproto->backer->need_revalidate = REV_BOND;
3131 if (bond_should_send_learning_packets(bundle->bond)) {
3132 bundle_send_learning_packets(bundle);
3138 bundle_wait(struct ofbundle *bundle)
3141 lacp_wait(bundle->lacp);
3144 bond_wait(bundle->bond);
3151 mirror_set__(struct ofproto *ofproto_, void *aux,
3152 const struct ofproto_mirror_settings *s)
3154 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3155 struct ofbundle **srcs, **dsts;
3160 mirror_destroy(ofproto->mbridge, aux);
3164 srcs = xmalloc(s->n_srcs * sizeof *srcs);
3165 dsts = xmalloc(s->n_dsts * sizeof *dsts);
3167 for (i = 0; i < s->n_srcs; i++) {
3168 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
3171 for (i = 0; i < s->n_dsts; i++) {
3172 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
3175 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
3176 s->n_dsts, s->src_vlans,
3177 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
3184 mirror_get_stats__(struct ofproto *ofproto, void *aux,
3185 uint64_t *packets, uint64_t *bytes)
3187 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
3192 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3194 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3195 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3196 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3197 mac_learning_flush(ofproto->ml);
3199 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3204 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3206 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3207 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3208 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
3212 forward_bpdu_changed(struct ofproto *ofproto_)
3214 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3215 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3219 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3222 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3223 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3224 mac_learning_set_idle_time(ofproto->ml, idle_time);
3225 mac_learning_set_max_entries(ofproto->ml, max_entries);
3226 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3229 /* Configures multicast snooping on 'ofport' using the settings
3230 * defined in 's'. */
3232 set_mcast_snooping(struct ofproto *ofproto_,
3233 const struct ofproto_mcast_snooping_settings *s)
3235 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3237 /* Only revalidate flows if the configuration changed. */
3238 if (!s != !ofproto->ms) {
3239 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3244 ofproto->ms = mcast_snooping_create();
3247 ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3248 mcast_snooping_set_idle_time(ofproto->ms, s->idle_time);
3249 mcast_snooping_set_max_entries(ofproto->ms, s->max_entries);
3250 if (mcast_snooping_set_flood_unreg(ofproto->ms, s->flood_unreg)) {
3251 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3253 ovs_rwlock_unlock(&ofproto->ms->rwlock);
3255 mcast_snooping_unref(ofproto->ms);
3262 /* Configures multicast snooping port's flood settings on 'ofproto'. */
3264 set_mcast_snooping_port(struct ofproto *ofproto_, void *aux,
3265 const struct ofproto_mcast_snooping_port_settings *s)
3267 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3268 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3270 if (ofproto->ms && s) {
3271 ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3272 mcast_snooping_set_port_flood(ofproto->ms, bundle, s->flood);
3273 mcast_snooping_set_port_flood_reports(ofproto->ms, bundle,
3275 ovs_rwlock_unlock(&ofproto->ms->rwlock);
3283 struct ofport_dpif *
3284 ofp_port_to_ofport(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
3286 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3287 return ofport ? ofport_dpif_cast(ofport) : NULL;
3291 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3292 struct ofproto_port *ofproto_port,
3293 struct dpif_port *dpif_port)
3295 ofproto_port->name = dpif_port->name;
3296 ofproto_port->type = dpif_port->type;
3297 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3301 ofport_update_peer(struct ofport_dpif *ofport)
3303 const struct ofproto_dpif *ofproto;
3304 struct dpif_backer *backer;
3307 if (!netdev_vport_is_patch(ofport->up.netdev)) {
3311 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
3312 backer->need_revalidate = REV_RECONFIGURE;
3315 ofport->peer->peer = NULL;
3316 ofport->peer = NULL;
3319 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
3324 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3325 struct ofport *peer_ofport;
3326 struct ofport_dpif *peer;
3329 if (ofproto->backer != backer) {
3333 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
3338 peer = ofport_dpif_cast(peer_ofport);
3339 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
3340 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
3342 ofport->peer = peer;
3343 ofport->peer->peer = ofport;
3353 port_run(struct ofport_dpif *ofport)
3355 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3356 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3357 bool enable = netdev_get_carrier(ofport->up.netdev);
3358 bool cfm_enable = false;
3359 bool bfd_enable = false;
3361 ofport->carrier_seq = carrier_seq;
3364 int cfm_opup = cfm_get_opup(ofport->cfm);
3366 cfm_enable = !cfm_get_fault(ofport->cfm);
3368 if (cfm_opup >= 0) {
3369 cfm_enable = cfm_enable && cfm_opup;
3374 bfd_enable = bfd_forwarding(ofport->bfd);
3377 if (ofport->bfd || ofport->cfm) {
3378 enable = enable && (cfm_enable || bfd_enable);
3381 if (ofport->bundle) {
3382 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3383 if (carrier_changed) {
3384 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3388 if (ofport->may_enable != enable) {
3389 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3391 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3393 if (ofport->rstp_port) {
3394 rstp_port_set_mac_operational(ofport->rstp_port, enable);
3398 ofport->may_enable = enable;
3402 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3403 struct ofproto_port *ofproto_port)
3405 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3406 struct dpif_port dpif_port;
3409 if (sset_contains(&ofproto->ghost_ports, devname)) {
3410 const char *type = netdev_get_type_from_name(devname);
3412 /* We may be called before ofproto->up.port_by_name is populated with
3413 * the appropriate ofport. For this reason, we must get the name and
3414 * type from the netdev layer directly. */
3416 const struct ofport *ofport;
3418 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3419 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3420 ofproto_port->name = xstrdup(devname);
3421 ofproto_port->type = xstrdup(type);
3427 if (!sset_contains(&ofproto->ports, devname)) {
3430 error = dpif_port_query_by_name(ofproto->backer->dpif,
3431 devname, &dpif_port);
3433 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3439 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3441 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3442 const char *devname = netdev_get_name(netdev);
3443 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3444 const char *dp_port_name;
3446 if (netdev_vport_is_patch(netdev)) {
3447 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3451 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3452 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3453 odp_port_t port_no = ODPP_NONE;
3456 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3460 if (netdev_get_tunnel_config(netdev)) {
3461 simap_put(&ofproto->backer->tnl_backers,
3462 dp_port_name, odp_to_u32(port_no));
3466 if (netdev_get_tunnel_config(netdev)) {
3467 sset_add(&ofproto->ghost_ports, devname);
3469 sset_add(&ofproto->ports, devname);
3475 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3477 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3478 struct ofport_dpif *ofport = ofp_port_to_ofport(ofproto, ofp_port);
3485 sset_find_and_delete(&ofproto->ghost_ports,
3486 netdev_get_name(ofport->up.netdev));
3487 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3488 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3489 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3491 /* The caller is going to close ofport->up.netdev. If this is a
3492 * bonded port, then the bond is using that netdev, so remove it
3493 * from the bond. The client will need to reconfigure everything
3494 * after deleting ports, so then the slave will get re-added. */
3495 bundle_remove(&ofport->up);
3502 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3504 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3507 error = netdev_get_stats(ofport->up.netdev, stats);
3509 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3510 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3512 ovs_mutex_lock(&ofproto->stats_mutex);
3513 /* ofproto->stats.tx_packets represents packets that we created
3514 * internally and sent to some port (e.g. packets sent with
3515 * ofproto_dpif_send_packet()). Account for them as if they had
3516 * come from OFPP_LOCAL and got forwarded. */
3518 if (stats->rx_packets != UINT64_MAX) {
3519 stats->rx_packets += ofproto->stats.tx_packets;
3522 if (stats->rx_bytes != UINT64_MAX) {
3523 stats->rx_bytes += ofproto->stats.tx_bytes;
3526 /* ofproto->stats.rx_packets represents packets that were received on
3527 * some port and we processed internally and dropped (e.g. STP).
3528 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3530 if (stats->tx_packets != UINT64_MAX) {
3531 stats->tx_packets += ofproto->stats.rx_packets;
3534 if (stats->tx_bytes != UINT64_MAX) {
3535 stats->tx_bytes += ofproto->stats.rx_bytes;
3537 ovs_mutex_unlock(&ofproto->stats_mutex);
3544 port_get_lacp_stats(const struct ofport *ofport_, struct lacp_slave_stats *stats)
3546 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3547 if (ofport->bundle && ofport->bundle->lacp) {
3548 if (lacp_get_slave_stats(ofport->bundle->lacp, ofport, stats)) {
3555 struct port_dump_state {
3560 struct ofproto_port port;
3565 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3567 *statep = xzalloc(sizeof(struct port_dump_state));
3572 port_dump_next(const struct ofproto *ofproto_, void *state_,
3573 struct ofproto_port *port)
3575 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3576 struct port_dump_state *state = state_;
3577 const struct sset *sset;
3578 struct sset_node *node;
3580 if (state->has_port) {
3581 ofproto_port_destroy(&state->port);
3582 state->has_port = false;
3584 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3585 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3588 error = port_query_by_name(ofproto_, node->name, &state->port);
3590 *port = state->port;
3591 state->has_port = true;
3593 } else if (error != ENODEV) {
3598 if (!state->ghost) {
3599 state->ghost = true;
3602 return port_dump_next(ofproto_, state_, port);
3609 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3611 struct port_dump_state *state = state_;
3613 if (state->has_port) {
3614 ofproto_port_destroy(&state->port);
3621 port_poll(const struct ofproto *ofproto_, char **devnamep)
3623 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3625 if (ofproto->port_poll_errno) {
3626 int error = ofproto->port_poll_errno;
3627 ofproto->port_poll_errno = 0;
3631 if (sset_is_empty(&ofproto->port_poll_set)) {
3635 *devnamep = sset_pop(&ofproto->port_poll_set);
3640 port_poll_wait(const struct ofproto *ofproto_)
3642 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3643 dpif_port_poll_wait(ofproto->backer->dpif);
3647 port_is_lacp_current(const struct ofport *ofport_)
3649 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3650 return (ofport->bundle && ofport->bundle->lacp
3651 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3655 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3656 * then delete it entirely. */
3658 rule_expire(struct rule_dpif *rule)
3659 OVS_REQUIRES(ofproto_mutex)
3661 uint16_t hard_timeout, idle_timeout;
3662 long long int now = time_msec();
3665 hard_timeout = rule->up.hard_timeout;
3666 idle_timeout = rule->up.idle_timeout;
3668 /* Has 'rule' expired? */
3670 long long int modified;
3672 ovs_mutex_lock(&rule->up.mutex);
3673 modified = rule->up.modified;
3674 ovs_mutex_unlock(&rule->up.mutex);
3676 if (now > modified + hard_timeout * 1000) {
3677 reason = OFPRR_HARD_TIMEOUT;
3681 if (reason < 0 && idle_timeout) {
3684 ovs_mutex_lock(&rule->stats_mutex);
3685 used = rule->stats.used;
3686 ovs_mutex_unlock(&rule->stats_mutex);
3688 if (now > used + idle_timeout * 1000) {
3689 reason = OFPRR_IDLE_TIMEOUT;
3694 COVERAGE_INC(ofproto_dpif_expired);
3695 ofproto_rule_expire(&rule->up, reason);
3700 ofproto_dpif_execute_actions__(struct ofproto_dpif *ofproto,
3701 const struct flow *flow,
3702 struct rule_dpif *rule,
3703 const struct ofpact *ofpacts, size_t ofpacts_len,
3704 int recurse, int resubmits,
3705 struct dp_packet *packet)
3707 struct dpif_flow_stats stats;
3708 struct xlate_out xout;
3709 struct xlate_in xin;
3711 struct dpif_execute execute;
3714 ovs_assert((rule != NULL) != (ofpacts != NULL));
3716 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3719 rule_dpif_credit_stats(rule, &stats);
3722 uint64_t odp_actions_stub[1024 / 8];
3723 struct ofpbuf odp_actions = OFPBUF_STUB_INITIALIZER(odp_actions_stub);
3724 xlate_in_init(&xin, ofproto, flow, flow->in_port.ofp_port, rule,
3725 stats.tcp_flags, packet, NULL, &odp_actions);
3726 xin.ofpacts = ofpacts;
3727 xin.ofpacts_len = ofpacts_len;
3728 xin.resubmit_stats = &stats;
3729 xin.recurse = recurse;
3730 xin.resubmits = resubmits;
3731 xlate_actions(&xin, &xout);
3733 execute.actions = odp_actions.data;
3734 execute.actions_len = odp_actions.size;
3736 pkt_metadata_from_flow(&packet->md, flow);
3737 execute.packet = packet;
3738 execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
3739 execute.probe = false;
3741 /* Fix up in_port. */
3742 in_port = flow->in_port.ofp_port;
3743 if (in_port == OFPP_NONE) {
3744 in_port = OFPP_LOCAL;
3746 execute.packet->md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
3748 error = dpif_execute(ofproto->backer->dpif, &execute);
3750 xlate_out_uninit(&xout);
3751 ofpbuf_uninit(&odp_actions);
3756 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3757 * 'flow' must reflect the data in 'packet'. */
3759 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3760 const struct flow *flow,
3761 struct rule_dpif *rule,
3762 const struct ofpact *ofpacts, size_t ofpacts_len,
3763 struct dp_packet *packet)
3765 return ofproto_dpif_execute_actions__(ofproto, flow, rule, ofpacts,
3766 ofpacts_len, 0, 0, packet);
3770 rule_dpif_credit_stats(struct rule_dpif *rule,
3771 const struct dpif_flow_stats *stats)
3773 ovs_mutex_lock(&rule->stats_mutex);
3774 if (OVS_UNLIKELY(rule->new_rule)) {
3775 rule_dpif_credit_stats(rule->new_rule, stats);
3777 rule->stats.n_packets += stats->n_packets;
3778 rule->stats.n_bytes += stats->n_bytes;
3779 rule->stats.used = MAX(rule->stats.used, stats->used);
3781 ovs_mutex_unlock(&rule->stats_mutex);
3785 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
3786 OVS_REQUIRES(rule->up.mutex)
3788 return rule->up.flow_cookie;
3792 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
3793 uint16_t hard_timeout)
3795 ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
3798 /* Returns 'rule''s actions. The returned actions are RCU-protected, and can
3799 * be read until the calling thread quiesces. */
3800 const struct rule_actions *
3801 rule_dpif_get_actions(const struct rule_dpif *rule)
3803 return rule_get_actions(&rule->up);
3806 /* Sets 'rule''s recirculation id. */
3808 rule_dpif_set_recirc_id(struct rule_dpif *rule, uint32_t id)
3809 OVS_REQUIRES(rule->up.mutex)
3811 ovs_assert(!rule->recirc_id || rule->recirc_id == id);
3812 if (rule->recirc_id == id) {
3813 /* Release the new reference to the same id. */
3816 rule->recirc_id = id;
3820 /* Sets 'rule''s recirculation id. */
3822 rule_set_recirc_id(struct rule *rule_, uint32_t id)
3824 struct rule_dpif *rule = rule_dpif_cast(rule_);
3826 ovs_mutex_lock(&rule->up.mutex);
3827 rule_dpif_set_recirc_id(rule, id);
3828 ovs_mutex_unlock(&rule->up.mutex);
3832 ofproto_dpif_get_tables_version(struct ofproto_dpif *ofproto OVS_UNUSED)
3834 cls_version_t version;
3836 atomic_read_relaxed(&ofproto->tables_version, &version);
3841 /* The returned rule (if any) is valid at least until the next RCU quiescent
3842 * period. If the rule needs to stay around longer, the caller should take
3845 * 'flow' is non-const to allow for temporary modifications during the lookup.
3846 * Any changes are restored before returning. */
3847 static struct rule_dpif *
3848 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, cls_version_t version,
3849 uint8_t table_id, struct flow *flow,
3850 struct flow_wildcards *wc)
3852 struct classifier *cls = &ofproto->up.tables[table_id].cls;
3853 return rule_dpif_cast(rule_from_cls_rule(classifier_lookup(cls, version,
3857 /* Look up 'flow' in 'ofproto''s classifier version 'version', starting from
3858 * table '*table_id'. Returns the rule that was found, which may be one of the
3859 * special rules according to packet miss hadling. If 'may_packet_in' is
3860 * false, returning of the miss_rule (which issues packet ins for the
3861 * controller) is avoided. Updates 'wc', if nonnull, to reflect the fields
3862 * that were used during the lookup.
3864 * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3865 * if none is found then the table miss configuration for that table is
3866 * honored, which can result in additional lookups in other OpenFlow tables.
3867 * In this case the function updates '*table_id' to reflect the final OpenFlow
3868 * table that was searched.
3870 * If 'honor_table_miss' is false, then only one table lookup occurs, in
3873 * The rule is returned in '*rule', which is valid at least until the next
3874 * RCU quiescent period. If the '*rule' needs to stay around longer, the
3875 * caller must take a reference.
3877 * 'in_port' allows the lookup to take place as if the in port had the value
3878 * 'in_port'. This is needed for resubmit action support.
3880 * 'flow' is non-const to allow for temporary modifications during the lookup.
3881 * Any changes are restored before returning. */
3883 rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto,
3884 cls_version_t version, struct flow *flow,
3885 struct flow_wildcards *wc,
3886 const struct dpif_flow_stats *stats,
3887 uint8_t *table_id, ofp_port_t in_port,
3888 bool may_packet_in, bool honor_table_miss)
3890 ovs_be16 old_tp_src = flow->tp_src, old_tp_dst = flow->tp_dst;
3891 ofp_port_t old_in_port = flow->in_port.ofp_port;
3892 enum ofputil_table_miss miss_config;
3893 struct rule_dpif *rule;
3896 /* We always unwildcard nw_frag (for IP), so they
3897 * need not be unwildcarded here. */
3898 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3899 && ofproto->up.frag_handling != OFPC_FRAG_NX_MATCH) {
3900 if (ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3901 /* We must pretend that transport ports are unavailable. */
3902 flow->tp_src = htons(0);
3903 flow->tp_dst = htons(0);
3905 /* Must be OFPC_FRAG_DROP (we don't have OFPC_FRAG_REASM).
3906 * Use the drop_frags_rule (which cannot disappear). */
3907 rule = ofproto->drop_frags_rule;
3909 struct oftable *tbl = &ofproto->up.tables[*table_id];
3912 atomic_add_relaxed(&tbl->n_matched, stats->n_packets, &orig);
3918 /* Look up a flow with 'in_port' as the input port. Then restore the
3919 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
3920 * have surprising behavior). */
3921 flow->in_port.ofp_port = in_port;
3923 /* Our current implementation depends on n_tables == N_TABLES, and
3924 * TBL_INTERNAL being the last table. */
3925 BUILD_ASSERT_DECL(N_TABLES == TBL_INTERNAL + 1);
3927 miss_config = OFPUTIL_TABLE_MISS_CONTINUE;
3929 for (next_id = *table_id;
3930 next_id < ofproto->up.n_tables;
3931 next_id++, next_id += (next_id == TBL_INTERNAL))
3933 *table_id = next_id;
3934 rule = rule_dpif_lookup_in_table(ofproto, version, next_id, flow, wc);
3936 struct oftable *tbl = &ofproto->up.tables[next_id];
3939 atomic_add_relaxed(rule ? &tbl->n_matched : &tbl->n_missed,
3940 stats->n_packets, &orig);
3943 goto out; /* Match. */
3945 if (honor_table_miss) {
3946 miss_config = ofproto_table_get_miss_config(&ofproto->up,
3948 if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE) {
3955 rule = ofproto->no_packet_in_rule;
3956 if (may_packet_in) {
3957 if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE
3958 || miss_config == OFPUTIL_TABLE_MISS_CONTROLLER) {
3959 struct ofport_dpif *port;
3961 port = ofp_port_to_ofport(ofproto, old_in_port);
3963 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
3965 } else if (!(port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN)) {
3966 rule = ofproto->miss_rule;
3968 } else if (miss_config == OFPUTIL_TABLE_MISS_DEFAULT &&
3969 connmgr_wants_packet_in_on_miss(ofproto->up.connmgr)) {
3970 rule = ofproto->miss_rule;
3974 /* Restore port numbers, as they may have been modified above. */
3975 flow->tp_src = old_tp_src;
3976 flow->tp_dst = old_tp_dst;
3977 /* Restore the old in port. */
3978 flow->in_port.ofp_port = old_in_port;
3984 complete_operation(struct rule_dpif *rule)
3985 OVS_REQUIRES(ofproto_mutex)
3987 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3989 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3992 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
3994 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
3997 static struct rule *
4000 struct rule_dpif *rule = xzalloc(sizeof *rule);
4005 rule_dealloc(struct rule *rule_)
4007 struct rule_dpif *rule = rule_dpif_cast(rule_);
4012 rule_check(struct rule *rule)
4014 uint16_t ct_state, ct_zone;
4015 const ovs_u128 *labelp;
4016 ovs_u128 ct_label = { { 0, 0 } };
4019 ct_state = MINIFLOW_GET_U16(rule->cr.match.flow, ct_state);
4020 ct_zone = MINIFLOW_GET_U16(rule->cr.match.flow, ct_zone);
4021 ct_mark = MINIFLOW_GET_U32(rule->cr.match.flow, ct_mark);
4022 labelp = MINIFLOW_GET_U128_PTR(rule->cr.match.flow, ct_label);
4027 if (ct_state || ct_zone || ct_mark
4028 || !ovs_u128_is_zero(&ct_label)) {
4029 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->ofproto);
4030 const struct odp_support *support = &ofproto_dpif_get_support(ofproto)->odp;
4032 if ((ct_state && !support->ct_state)
4033 || (ct_zone && !support->ct_zone)
4034 || (ct_mark && !support->ct_mark)
4035 || (!ovs_u128_is_zero(&ct_label) && !support->ct_label)) {
4036 return OFPERR_OFPBMC_BAD_FIELD;
4038 if (ct_state & CS_UNSUPPORTED_MASK) {
4039 return OFPERR_OFPBMC_BAD_MASK;
4046 rule_construct(struct rule *rule_)
4047 OVS_NO_THREAD_SAFETY_ANALYSIS
4049 struct rule_dpif *rule = rule_dpif_cast(rule_);
4052 error = rule_check(rule_);
4057 ovs_mutex_init_adaptive(&rule->stats_mutex);
4058 rule->stats.n_packets = 0;
4059 rule->stats.n_bytes = 0;
4060 rule->stats.used = rule->up.modified;
4061 rule->recirc_id = 0;
4062 rule->new_rule = NULL;
4068 rule_insert(struct rule *rule_, struct rule *old_rule_, bool forward_stats)
4069 OVS_REQUIRES(ofproto_mutex)
4071 struct rule_dpif *rule = rule_dpif_cast(rule_);
4073 if (old_rule_ && forward_stats) {
4074 struct rule_dpif *old_rule = rule_dpif_cast(old_rule_);
4076 ovs_assert(!old_rule->new_rule);
4078 /* Take a reference to the new rule, and refer all stats updates from
4079 * the old rule to the new rule. */
4080 rule_dpif_ref(rule);
4082 ovs_mutex_lock(&old_rule->stats_mutex);
4083 ovs_mutex_lock(&rule->stats_mutex);
4084 old_rule->new_rule = rule; /* Forward future stats. */
4085 rule->stats = old_rule->stats; /* Transfer stats to the new rule. */
4086 ovs_mutex_unlock(&rule->stats_mutex);
4087 ovs_mutex_unlock(&old_rule->stats_mutex);
4090 complete_operation(rule);
4094 rule_delete(struct rule *rule_)
4095 OVS_REQUIRES(ofproto_mutex)
4097 struct rule_dpif *rule = rule_dpif_cast(rule_);
4098 complete_operation(rule);
4102 rule_destruct(struct rule *rule_)
4103 OVS_NO_THREAD_SAFETY_ANALYSIS
4105 struct rule_dpif *rule = rule_dpif_cast(rule_);
4107 ovs_mutex_destroy(&rule->stats_mutex);
4108 /* Release reference to the new rule, if any. */
4109 if (rule->new_rule) {
4110 rule_dpif_unref(rule->new_rule);
4112 if (rule->recirc_id) {
4113 recirc_free_id(rule->recirc_id);
4118 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
4119 long long int *used)
4121 struct rule_dpif *rule = rule_dpif_cast(rule_);
4123 ovs_mutex_lock(&rule->stats_mutex);
4124 if (OVS_UNLIKELY(rule->new_rule)) {
4125 rule_get_stats(&rule->new_rule->up, packets, bytes, used);
4127 *packets = rule->stats.n_packets;
4128 *bytes = rule->stats.n_bytes;
4129 *used = rule->stats.used;
4131 ovs_mutex_unlock(&rule->stats_mutex);
4135 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4136 struct dp_packet *packet)
4138 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4140 ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
4144 rule_execute(struct rule *rule, const struct flow *flow,
4145 struct dp_packet *packet)
4147 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4148 dp_packet_delete(packet);
4152 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
4154 return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
4157 static struct ofgroup *
4160 struct group_dpif *group = xzalloc(sizeof *group);
4165 group_dealloc(struct ofgroup *group_)
4167 struct group_dpif *group = group_dpif_cast(group_);
4172 group_construct_stats(struct group_dpif *group)
4173 OVS_REQUIRES(group->stats_mutex)
4175 struct ofputil_bucket *bucket;
4176 const struct ovs_list *buckets;
4178 group->packet_count = 0;
4179 group->byte_count = 0;
4181 group_dpif_get_buckets(group, &buckets);
4182 LIST_FOR_EACH (bucket, list_node, buckets) {
4183 bucket->stats.packet_count = 0;
4184 bucket->stats.byte_count = 0;
4189 group_dpif_credit_stats(struct group_dpif *group,
4190 struct ofputil_bucket *bucket,
4191 const struct dpif_flow_stats *stats)
4193 ovs_mutex_lock(&group->stats_mutex);
4194 group->packet_count += stats->n_packets;
4195 group->byte_count += stats->n_bytes;
4197 bucket->stats.packet_count += stats->n_packets;
4198 bucket->stats.byte_count += stats->n_bytes;
4199 } else { /* Credit to all buckets */
4200 const struct ovs_list *buckets;
4202 group_dpif_get_buckets(group, &buckets);
4203 LIST_FOR_EACH (bucket, list_node, buckets) {
4204 bucket->stats.packet_count += stats->n_packets;
4205 bucket->stats.byte_count += stats->n_bytes;
4208 ovs_mutex_unlock(&group->stats_mutex);
4212 group_construct(struct ofgroup *group_)
4214 struct group_dpif *group = group_dpif_cast(group_);
4216 ovs_mutex_init_adaptive(&group->stats_mutex);
4217 ovs_mutex_lock(&group->stats_mutex);
4218 group_construct_stats(group);
4219 ovs_mutex_unlock(&group->stats_mutex);
4224 group_destruct(struct ofgroup *group_)
4226 struct group_dpif *group = group_dpif_cast(group_);
4227 ovs_mutex_destroy(&group->stats_mutex);
4231 group_modify(struct ofgroup *group_)
4233 struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
4235 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4241 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
4243 struct group_dpif *group = group_dpif_cast(group_);
4244 struct ofputil_bucket *bucket;
4245 const struct ovs_list *buckets;
4246 struct bucket_counter *bucket_stats;
4248 ovs_mutex_lock(&group->stats_mutex);
4249 ogs->packet_count = group->packet_count;
4250 ogs->byte_count = group->byte_count;
4252 group_dpif_get_buckets(group, &buckets);
4253 bucket_stats = ogs->bucket_stats;
4254 LIST_FOR_EACH (bucket, list_node, buckets) {
4255 bucket_stats->packet_count = bucket->stats.packet_count;
4256 bucket_stats->byte_count = bucket->stats.byte_count;
4259 ovs_mutex_unlock(&group->stats_mutex);
4264 /* If the group exists, this function increments the groups's reference count.
4266 * Make sure to call group_dpif_unref() after no longer needing to maintain
4267 * a reference to the group. */
4269 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
4270 struct group_dpif **group)
4272 struct ofgroup *ofgroup;
4275 found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
4276 *group = found ? group_dpif_cast(ofgroup) : NULL;
4282 group_dpif_get_buckets(const struct group_dpif *group,
4283 const struct ovs_list **buckets)
4285 *buckets = &group->up.buckets;
4288 enum ofp11_group_type
4289 group_dpif_get_type(const struct group_dpif *group)
4291 return group->up.type;
4295 group_dpif_get_selection_method(const struct group_dpif *group)
4297 return group->up.props.selection_method;
4300 /* Sends 'packet' out 'ofport'.
4301 * May modify 'packet'.
4302 * Returns 0 if successful, otherwise a positive errno value. */
4304 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct dp_packet *packet)
4306 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4309 error = xlate_send_packet(ofport, packet);
4311 ovs_mutex_lock(&ofproto->stats_mutex);
4312 ofproto->stats.tx_packets++;
4313 ofproto->stats.tx_bytes += dp_packet_size(packet);
4314 ovs_mutex_unlock(&ofproto->stats_mutex);
4319 group_dpif_get_selection_method_param(const struct group_dpif *group)
4321 return group->up.props.selection_method_param;
4324 const struct field_array *
4325 group_dpif_get_fields(const struct group_dpif *group)
4327 return &group->up.props.fields;
4330 /* Return the version string of the datapath that backs up
4334 get_datapath_version(const struct ofproto *ofproto_)
4336 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4338 return ofproto->backer->dp_version_string;
4342 set_frag_handling(struct ofproto *ofproto_,
4343 enum ofp_config_flags frag_handling)
4345 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4346 if (frag_handling != OFPC_FRAG_REASM) {
4347 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4355 packet_out(struct ofproto *ofproto_, struct dp_packet *packet,
4356 const struct flow *flow,
4357 const struct ofpact *ofpacts, size_t ofpacts_len)
4359 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4361 ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
4362 ofpacts_len, packet);
4369 set_netflow(struct ofproto *ofproto_,
4370 const struct netflow_options *netflow_options)
4372 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4374 if (netflow_options) {
4375 if (!ofproto->netflow) {
4376 ofproto->netflow = netflow_create();
4377 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4379 return netflow_set_options(ofproto->netflow, netflow_options);
4380 } else if (ofproto->netflow) {
4381 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4382 netflow_unref(ofproto->netflow);
4383 ofproto->netflow = NULL;
4390 get_netflow_ids(const struct ofproto *ofproto_,
4391 uint8_t *engine_type, uint8_t *engine_id)
4393 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4395 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
4398 static struct ofproto_dpif *
4399 ofproto_dpif_lookup(const char *name)
4401 struct ofproto_dpif *ofproto;
4403 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
4404 hash_string(name, 0), &all_ofproto_dpifs) {
4405 if (!strcmp(ofproto->up.name, name)) {
4413 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
4414 const char *argv[], void *aux OVS_UNUSED)
4416 struct ofproto_dpif *ofproto;
4419 ofproto = ofproto_dpif_lookup(argv[1]);
4421 unixctl_command_reply_error(conn, "no such bridge");
4424 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4425 mac_learning_flush(ofproto->ml);
4426 ovs_rwlock_unlock(&ofproto->ml->rwlock);
4428 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4429 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4430 mac_learning_flush(ofproto->ml);
4431 ovs_rwlock_unlock(&ofproto->ml->rwlock);
4435 unixctl_command_reply(conn, "table successfully flushed");
4439 ofproto_unixctl_mcast_snooping_flush(struct unixctl_conn *conn, int argc,
4440 const char *argv[], void *aux OVS_UNUSED)
4442 struct ofproto_dpif *ofproto;
4445 ofproto = ofproto_dpif_lookup(argv[1]);
4447 unixctl_command_reply_error(conn, "no such bridge");
4451 if (!mcast_snooping_enabled(ofproto->ms)) {
4452 unixctl_command_reply_error(conn, "multicast snooping is disabled");
4455 mcast_snooping_mdb_flush(ofproto->ms);
4457 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4458 if (!mcast_snooping_enabled(ofproto->ms)) {
4461 mcast_snooping_mdb_flush(ofproto->ms);
4465 unixctl_command_reply(conn, "table successfully flushed");
4468 static struct ofport_dpif *
4469 ofbundle_get_a_port(const struct ofbundle *bundle)
4471 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
4476 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4477 const char *argv[], void *aux OVS_UNUSED)
4479 struct ds ds = DS_EMPTY_INITIALIZER;
4480 const struct ofproto_dpif *ofproto;
4481 const struct mac_entry *e;
4483 ofproto = ofproto_dpif_lookup(argv[1]);
4485 unixctl_command_reply_error(conn, "no such bridge");
4489 ds_put_cstr(&ds, " port VLAN MAC Age\n");
4490 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
4491 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
4492 struct ofbundle *bundle = mac_entry_get_port(ofproto->ml, e);
4493 char name[OFP_MAX_PORT_NAME_LEN];
4495 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4497 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
4498 name, e->vlan, ETH_ADDR_ARGS(e->mac),
4499 mac_entry_age(ofproto->ml, e));
4501 ovs_rwlock_unlock(&ofproto->ml->rwlock);
4502 unixctl_command_reply(conn, ds_cstr(&ds));
4507 ofproto_unixctl_mcast_snooping_show(struct unixctl_conn *conn,
4508 int argc OVS_UNUSED,
4510 void *aux OVS_UNUSED)
4512 struct ds ds = DS_EMPTY_INITIALIZER;
4513 const struct ofproto_dpif *ofproto;
4514 const struct ofbundle *bundle;
4515 const struct mcast_group *grp;
4516 struct mcast_group_bundle *b;
4517 struct mcast_mrouter_bundle *mrouter;
4519 ofproto = ofproto_dpif_lookup(argv[1]);
4521 unixctl_command_reply_error(conn, "no such bridge");
4525 if (!mcast_snooping_enabled(ofproto->ms)) {
4526 unixctl_command_reply_error(conn, "multicast snooping is disabled");
4530 ds_put_cstr(&ds, " port VLAN GROUP Age\n");
4531 ovs_rwlock_rdlock(&ofproto->ms->rwlock);
4532 LIST_FOR_EACH (grp, group_node, &ofproto->ms->group_lru) {
4533 LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
4534 char name[OFP_MAX_PORT_NAME_LEN];
4537 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4539 ds_put_format(&ds, "%5s %4d ", name, grp->vlan);
4540 print_ipv6_mapped(&ds, &grp->addr);
4541 ds_put_format(&ds, " %3d\n",
4542 mcast_bundle_age(ofproto->ms, b));
4546 /* ports connected to multicast routers */
4547 LIST_FOR_EACH(mrouter, mrouter_node, &ofproto->ms->mrouter_lru) {
4548 char name[OFP_MAX_PORT_NAME_LEN];
4550 bundle = mrouter->port;
4551 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4553 ds_put_format(&ds, "%5s %4d querier %3d\n",
4554 name, mrouter->vlan,
4555 mcast_mrouter_age(ofproto->ms, mrouter));
4557 ovs_rwlock_unlock(&ofproto->ms->rwlock);
4558 unixctl_command_reply(conn, ds_cstr(&ds));
4563 struct xlate_out xout;
4564 struct xlate_in xin;
4565 const struct flow *key;
4568 struct flow_wildcards wc;
4569 struct ofpbuf odp_actions;
4573 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
4575 const struct rule_actions *actions;
4578 ds_put_char_multiple(result, '\t', level);
4580 ds_put_cstr(result, "No match\n");
4584 ovs_mutex_lock(&rule->up.mutex);
4585 cookie = rule->up.flow_cookie;
4586 ovs_mutex_unlock(&rule->up.mutex);
4588 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
4589 rule ? rule->up.table_id : 0, ntohll(cookie));
4590 cls_rule_format(&rule->up.cr, result);
4591 ds_put_char(result, '\n');
4593 actions = rule_dpif_get_actions(rule);
4595 ds_put_char_multiple(result, '\t', level);
4596 ds_put_cstr(result, "OpenFlow actions=");
4597 ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
4598 ds_put_char(result, '\n');
4602 trace_format_flow(struct ds *result, int level, const char *title,
4603 struct trace_ctx *trace)
4605 ds_put_char_multiple(result, '\t', level);
4606 ds_put_format(result, "%s: ", title);
4607 /* Do not report unchanged flows for resubmits. */
4608 if ((level > 0 && flow_equal(&trace->xin.flow, &trace->flow))
4609 || (level == 0 && flow_equal(&trace->xin.flow, trace->key))) {
4610 ds_put_cstr(result, "unchanged");
4612 flow_format(result, &trace->xin.flow);
4613 trace->flow = trace->xin.flow;
4615 ds_put_char(result, '\n');
4619 trace_format_regs(struct ds *result, int level, const char *title,
4620 struct trace_ctx *trace)
4624 ds_put_char_multiple(result, '\t', level);
4625 ds_put_format(result, "%s:", title);
4626 for (i = 0; i < FLOW_N_REGS; i++) {
4627 ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
4629 ds_put_char(result, '\n');
4633 trace_format_odp(struct ds *result, int level, const char *title,
4634 struct trace_ctx *trace)
4636 struct ofpbuf *odp_actions = &trace->odp_actions;
4638 ds_put_char_multiple(result, '\t', level);
4639 ds_put_format(result, "%s: ", title);
4640 format_odp_actions(result, odp_actions->data, odp_actions->size);
4641 ds_put_char(result, '\n');
4645 trace_format_megaflow(struct ds *result, int level, const char *title,
4646 struct trace_ctx *trace)
4650 ds_put_char_multiple(result, '\t', level);
4651 ds_put_format(result, "%s: ", title);
4652 match_init(&match, trace->key, &trace->wc);
4653 match_format(&match, result, OFP_DEFAULT_PRIORITY);
4654 ds_put_char(result, '\n');
4657 static void trace_report(struct xlate_in *, int recurse,
4658 const char *format, ...)
4659 OVS_PRINTF_FORMAT(3, 4);
4660 static void trace_report_valist(struct xlate_in *, int recurse,
4661 const char *format, va_list args)
4662 OVS_PRINTF_FORMAT(3, 0);
4665 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
4667 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4668 struct ds *result = trace->result;
4671 if (rule == xin->ofproto->miss_rule) {
4672 trace_report(xin, recurse,
4673 "No match, flow generates \"packet in\"s.");
4674 } else if (rule == xin->ofproto->no_packet_in_rule) {
4675 trace_report(xin, recurse, "No match, packets dropped because "
4676 "OFPPC_NO_PACKET_IN is set on in_port.");
4677 } else if (rule == xin->ofproto->drop_frags_rule) {
4678 trace_report(xin, recurse, "Packets dropped because they are IP "
4679 "fragments and the fragment handling mode is "
4684 ds_put_char(result, '\n');
4686 trace_format_flow(result, recurse, "Resubmitted flow", trace);
4687 trace_format_regs(result, recurse, "Resubmitted regs", trace);
4688 trace_format_odp(result, recurse, "Resubmitted odp", trace);
4689 trace_format_megaflow(result, recurse, "Resubmitted megaflow", trace);
4691 trace_format_rule(result, recurse, rule);
4695 trace_report_valist(struct xlate_in *xin, int recurse,
4696 const char *format, va_list args)
4698 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4699 struct ds *result = trace->result;
4701 ds_put_char_multiple(result, '\t', recurse);
4702 ds_put_format_valist(result, format, args);
4703 ds_put_char(result, '\n');
4707 trace_report(struct xlate_in *xin, int recurse, const char *format, ...)
4711 va_start(args, format);
4712 trace_report_valist(xin, recurse, format, args);
4716 /* Parses the 'argc' elements of 'argv', ignoring argv[0]. The following
4717 * forms are supported:
4719 * - [dpname] odp_flow [-generate | packet]
4720 * - bridge br_flow [-generate | packet]
4722 * On success, initializes '*ofprotop' and 'flow' and returns NULL. On failure
4723 * returns a nonnull malloced error message. */
4724 static char * OVS_WARN_UNUSED_RESULT
4725 parse_flow_and_packet(int argc, const char *argv[],
4726 struct ofproto_dpif **ofprotop, struct flow *flow,
4727 struct dp_packet **packetp)
4729 const struct dpif_backer *backer = NULL;
4730 const char *error = NULL;
4732 struct simap port_names = SIMAP_INITIALIZER(&port_names);
4733 struct dp_packet *packet;
4734 struct ofpbuf odp_key;
4735 struct ofpbuf odp_mask;
4737 ofpbuf_init(&odp_key, 0);
4738 ofpbuf_init(&odp_mask, 0);
4740 /* Handle "-generate" or a hex string as the last argument. */
4741 if (!strcmp(argv[argc - 1], "-generate")) {
4742 packet = dp_packet_new(0);
4745 error = eth_from_hex(argv[argc - 1], &packet);
4748 } else if (argc == 4) {
4749 /* The 3-argument form must end in "-generate' or a hex string. */
4755 /* odp_flow can have its in_port specified as a name instead of port no.
4756 * We do not yet know whether a given flow is a odp_flow or a br_flow.
4757 * But, to know whether a flow is odp_flow through odp_flow_from_string(),
4758 * we need to create a simap of name to port no. */
4760 const char *dp_type;
4761 if (!strncmp(argv[1], "ovs-", 4)) {
4762 dp_type = argv[1] + 4;
4766 backer = shash_find_data(&all_dpif_backers, dp_type);
4767 } else if (argc == 2) {
4768 struct shash_node *node;
4769 if (shash_count(&all_dpif_backers) == 1) {
4770 node = shash_first(&all_dpif_backers);
4771 backer = node->data;
4774 error = "Syntax error";
4777 if (backer && backer->dpif) {
4778 struct dpif_port dpif_port;
4779 struct dpif_port_dump port_dump;
4780 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
4781 simap_put(&port_names, dpif_port.name,
4782 odp_to_u32(dpif_port.port_no));
4786 /* Parse the flow and determine whether a datapath or
4787 * bridge is specified. If function odp_flow_key_from_string()
4788 * returns 0, the flow is a odp_flow. If function
4789 * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
4790 if (!odp_flow_from_string(argv[argc - 1], &port_names,
4791 &odp_key, &odp_mask)) {
4793 error = "Cannot find the datapath";
4797 if (odp_flow_key_to_flow(odp_key.data, odp_key.size, flow) == ODP_FIT_ERROR) {
4798 error = "Failed to parse datapath flow key";
4802 *ofprotop = xlate_lookup_ofproto(backer, flow,
4803 &flow->in_port.ofp_port);
4804 if (*ofprotop == NULL) {
4805 error = "Invalid datapath flow";
4809 vsp_adjust_flow(*ofprotop, flow, NULL);
4812 char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
4815 m_err = xasprintf("Bad openflow flow syntax: %s", err);
4820 error = "Must specify bridge name";
4824 *ofprotop = ofproto_dpif_lookup(argv[1]);
4826 error = "Unknown bridge name";
4832 /* Generate a packet, if requested. */
4834 if (!dp_packet_size(packet)) {
4835 flow_compose(packet, flow);
4837 /* Use the metadata from the flow and the packet argument
4838 * to reconstruct the flow. */
4839 pkt_metadata_from_flow(&packet->md, flow);
4840 flow_extract(packet, flow);
4845 if (error && !m_err) {
4846 m_err = xstrdup(error);
4849 dp_packet_delete(packet);
4853 ofpbuf_uninit(&odp_key);
4854 ofpbuf_uninit(&odp_mask);
4855 simap_destroy(&port_names);
4860 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
4861 void *aux OVS_UNUSED)
4863 struct ofproto_dpif *ofproto;
4864 struct dp_packet *packet;
4868 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4873 ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
4874 unixctl_command_reply(conn, ds_cstr(&result));
4875 ds_destroy(&result);
4876 dp_packet_delete(packet);
4878 unixctl_command_reply_error(conn, error);
4884 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
4885 const char *argv[], void *aux OVS_UNUSED)
4887 enum ofputil_protocol usable_protocols;
4888 struct ofproto_dpif *ofproto;
4889 bool enforce_consistency;
4890 struct ofpbuf ofpacts;
4891 struct dp_packet *packet;
4896 /* Three kinds of error return values! */
4902 ofpbuf_init(&ofpacts, 0);
4904 /* Parse actions. */
4905 error = ofpacts_parse_actions(argv[--argc], &ofpacts, &usable_protocols);
4907 unixctl_command_reply_error(conn, error);
4912 /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
4913 * consistency between the flow and the actions. With -consistent, we
4914 * enforce consistency even for a flow supported in OpenFlow 1.0. */
4915 if (!strcmp(argv[1], "-consistent")) {
4916 enforce_consistency = true;
4920 enforce_consistency = false;
4923 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4925 unixctl_command_reply_error(conn, error);
4930 /* Do the same checks as handle_packet_out() in ofproto.c.
4932 * We pass a 'table_id' of 0 to ofpacts_check(), which isn't
4933 * strictly correct because these actions aren't in any table, but it's OK
4934 * because it 'table_id' is used only to check goto_table instructions, but
4935 * packet-outs take a list of actions and therefore it can't include
4938 * We skip the "meter" check here because meter is an instruction, not an
4939 * action, and thus cannot appear in ofpacts. */
4940 in_port = ofp_to_u16(flow.in_port.ofp_port);
4941 if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
4942 unixctl_command_reply_error(conn, "invalid in_port");
4945 if (enforce_consistency) {
4946 retval = ofpacts_check_consistency(ofpacts.data, ofpacts.size,
4947 &flow, u16_to_ofp(ofproto->up.max_ports),
4948 0, 0, usable_protocols);
4950 retval = ofpacts_check(ofpacts.data, ofpacts.size, &flow,
4951 u16_to_ofp(ofproto->up.max_ports), 0, 0,
4957 ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
4958 unixctl_command_reply_error(conn, ds_cstr(&result));
4962 ofproto_trace(ofproto, &flow, packet,
4963 ofpacts.data, ofpacts.size, &result);
4964 unixctl_command_reply(conn, ds_cstr(&result));
4967 ds_destroy(&result);
4968 dp_packet_delete(packet);
4969 ofpbuf_uninit(&ofpacts);
4972 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
4973 * description of the results to 'ds'.
4975 * The trace follows a packet with the specified 'flow' through the flow
4976 * table. 'packet' may be nonnull to trace an actual packet, with consequent
4977 * side effects (if it is nonnull then its flow must be 'flow').
4979 * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
4980 * trace, otherwise the actions are determined by a flow table lookup. */
4982 ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
4983 const struct dp_packet *packet,
4984 const struct ofpact ofpacts[], size_t ofpacts_len,
4987 struct trace_ctx trace;
4989 ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
4990 ds_put_cstr(ds, "Flow: ");
4991 flow_format(ds, flow);
4992 ds_put_char(ds, '\n');
4994 ofpbuf_init(&trace.odp_actions, 0);
4997 trace.key = flow; /* Original flow key, used for megaflow. */
4998 trace.flow = *flow; /* May be modified by actions. */
4999 xlate_in_init(&trace.xin, ofproto, flow, flow->in_port.ofp_port, NULL,
5000 ntohs(flow->tcp_flags), packet, &trace.wc,
5001 &trace.odp_actions);
5002 trace.xin.ofpacts = ofpacts;
5003 trace.xin.ofpacts_len = ofpacts_len;
5004 trace.xin.resubmit_hook = trace_resubmit;
5005 trace.xin.report_hook = trace_report_valist;
5007 xlate_actions(&trace.xin, &trace.xout);
5009 ds_put_char(ds, '\n');
5010 trace_format_flow(ds, 0, "Final flow", &trace);
5011 trace_format_megaflow(ds, 0, "Megaflow", &trace);
5013 ds_put_cstr(ds, "Datapath actions: ");
5014 format_odp_actions(ds, trace.odp_actions.data, trace.odp_actions.size);
5016 if (trace.xout.slow) {
5017 enum slow_path_reason slow;
5019 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5020 "slow path because it:");
5022 slow = trace.xout.slow;
5024 enum slow_path_reason bit = rightmost_1bit(slow);
5026 ds_put_format(ds, "\n\t- %s.",
5027 slow_path_reason_to_explanation(bit));
5033 xlate_out_uninit(&trace.xout);
5034 ofpbuf_uninit(&trace.odp_actions);
5037 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5038 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5039 * to destroy 'ofproto_shash' and free the returned value. */
5040 static const struct shash_node **
5041 get_ofprotos(struct shash *ofproto_shash)
5043 const struct ofproto_dpif *ofproto;
5045 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5046 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5047 shash_add_nocopy(ofproto_shash, name, ofproto);
5050 return shash_sort(ofproto_shash);
5054 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5055 const char *argv[] OVS_UNUSED,
5056 void *aux OVS_UNUSED)
5058 struct ds ds = DS_EMPTY_INITIALIZER;
5059 struct shash ofproto_shash;
5060 const struct shash_node **sorted_ofprotos;
5063 shash_init(&ofproto_shash);
5064 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5065 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5066 const struct shash_node *node = sorted_ofprotos[i];
5067 ds_put_format(&ds, "%s\n", node->name);
5070 shash_destroy(&ofproto_shash);
5071 free(sorted_ofprotos);
5073 unixctl_command_reply(conn, ds_cstr(&ds));
5078 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5080 const struct shash_node **ofprotos;
5081 struct dpif_dp_stats dp_stats;
5082 struct shash ofproto_shash;
5085 dpif_get_dp_stats(backer->dpif, &dp_stats);
5087 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5088 dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
5090 shash_init(&ofproto_shash);
5091 ofprotos = get_ofprotos(&ofproto_shash);
5092 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5093 struct ofproto_dpif *ofproto = ofprotos[i]->data;
5094 const struct shash_node **ports;
5097 if (ofproto->backer != backer) {
5101 ds_put_format(ds, "\t%s:\n", ofproto->up.name);
5103 ports = shash_sort(&ofproto->up.port_by_name);
5104 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5105 const struct shash_node *node = ports[j];
5106 struct ofport *ofport = node->data;
5108 odp_port_t odp_port;
5110 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5113 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5114 if (odp_port != ODPP_NONE) {
5115 ds_put_format(ds, "%"PRIu32":", odp_port);
5117 ds_put_cstr(ds, "none:");
5120 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5123 if (!netdev_get_config(ofport->netdev, &config)) {
5124 const struct smap_node **nodes;
5127 nodes = smap_sort(&config);
5128 for (i = 0; i < smap_count(&config); i++) {
5129 const struct smap_node *node = nodes[i];
5130 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5131 node->key, node->value);
5135 smap_destroy(&config);
5137 ds_put_char(ds, ')');
5138 ds_put_char(ds, '\n');
5142 shash_destroy(&ofproto_shash);
5147 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5148 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5150 struct ds ds = DS_EMPTY_INITIALIZER;
5151 const struct shash_node **backers;
5154 backers = shash_sort(&all_dpif_backers);
5155 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5156 dpif_show_backer(backers[i]->data, &ds);
5160 unixctl_command_reply(conn, ds_cstr(&ds));
5165 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5166 int argc OVS_UNUSED, const char *argv[],
5167 void *aux OVS_UNUSED)
5169 const struct ofproto_dpif *ofproto;
5171 struct ds ds = DS_EMPTY_INITIALIZER;
5172 bool verbosity = false;
5174 struct dpif_port dpif_port;
5175 struct dpif_port_dump port_dump;
5176 struct hmap portno_names;
5178 struct dpif_flow_dump *flow_dump;
5179 struct dpif_flow_dump_thread *flow_dump_thread;
5183 ofproto = ofproto_dpif_lookup(argv[argc - 1]);
5185 unixctl_command_reply_error(conn, "no such bridge");
5189 if (argc > 2 && !strcmp(argv[1], "-m")) {
5193 hmap_init(&portno_names);
5194 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
5195 odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
5199 flow_dump = dpif_flow_dump_create(ofproto->backer->dpif, false);
5200 flow_dump_thread = dpif_flow_dump_thread_create(flow_dump);
5201 while (dpif_flow_dump_next(flow_dump_thread, &f, 1)) {
5204 if (odp_flow_key_to_flow(f.key, f.key_len, &flow) == ODP_FIT_ERROR
5205 || xlate_lookup_ofproto(ofproto->backer, &flow, NULL) != ofproto) {
5210 odp_format_ufid(&f.ufid, &ds);
5211 ds_put_cstr(&ds, " ");
5213 odp_flow_format(f.key, f.key_len, f.mask, f.mask_len,
5214 &portno_names, &ds, verbosity);
5215 ds_put_cstr(&ds, ", ");
5216 dpif_flow_stats_format(&f.stats, &ds);
5217 ds_put_cstr(&ds, ", actions:");
5218 format_odp_actions(&ds, f.actions, f.actions_len);
5219 ds_put_char(&ds, '\n');
5221 dpif_flow_dump_thread_destroy(flow_dump_thread);
5222 error = dpif_flow_dump_destroy(flow_dump);
5226 ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
5227 unixctl_command_reply_error(conn, ds_cstr(&ds));
5229 unixctl_command_reply(conn, ds_cstr(&ds));
5231 odp_portno_names_destroy(&portno_names);
5232 hmap_destroy(&portno_names);
5237 ofproto_revalidate_all_backers(void)
5239 const struct shash_node **backers;
5242 backers = shash_sort(&all_dpif_backers);
5243 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5244 struct dpif_backer *backer = backers[i]->data;
5245 backer->need_revalidate = REV_RECONFIGURE;
5251 disable_tnl_push_pop(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5252 const char *argv[], void *aux OVS_UNUSED)
5254 if (!strcasecmp(argv[1], "off")) {
5255 ofproto_use_tnl_push_pop = false;
5256 unixctl_command_reply(conn, "Tunnel push-pop off");
5257 ofproto_revalidate_all_backers();
5258 } else if (!strcasecmp(argv[1], "on")) {
5259 ofproto_use_tnl_push_pop = true;
5260 unixctl_command_reply(conn, "Tunnel push-pop on");
5261 ofproto_revalidate_all_backers();
5266 ofproto_unixctl_init(void)
5268 static bool registered;
5274 unixctl_command_register(
5276 "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
5277 1, 3, ofproto_unixctl_trace, NULL);
5278 unixctl_command_register(
5279 "ofproto/trace-packet-out",
5280 "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
5281 2, 6, ofproto_unixctl_trace_actions, NULL);
5282 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5283 ofproto_unixctl_fdb_flush, NULL);
5284 unixctl_command_register("fdb/show", "bridge", 1, 1,
5285 ofproto_unixctl_fdb_show, NULL);
5286 unixctl_command_register("mdb/flush", "[bridge]", 0, 1,
5287 ofproto_unixctl_mcast_snooping_flush, NULL);
5288 unixctl_command_register("mdb/show", "bridge", 1, 1,
5289 ofproto_unixctl_mcast_snooping_show, NULL);
5290 unixctl_command_register("dpif/dump-dps", "", 0, 0,
5291 ofproto_unixctl_dpif_dump_dps, NULL);
5292 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5294 unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
5295 ofproto_unixctl_dpif_dump_flows, NULL);
5297 unixctl_command_register("ofproto/tnl-push-pop", "[on]|[off]", 1, 1,
5298 disable_tnl_push_pop, NULL);
5301 /* Returns true if 'table' is the table used for internal rules,
5302 * false otherwise. */
5304 table_is_internal(uint8_t table_id)
5306 return table_id == TBL_INTERNAL;
5309 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5311 * This is deprecated. It is only for compatibility with broken device drivers
5312 * in old versions of Linux that do not properly support VLANs when VLAN
5313 * devices are not used. When broken device drivers are no longer in
5314 * widespread use, we will delete these interfaces. */
5317 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
5319 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5320 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5322 if (realdev_ofp_port == ofport->realdev_ofp_port
5323 && vid == ofport->vlandev_vid) {
5327 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5329 if (ofport->realdev_ofp_port) {
5332 if (realdev_ofp_port && ofport->bundle) {
5333 /* vlandevs are enslaved to their realdevs, so they are not allowed to
5334 * themselves be part of a bundle. */
5335 bundle_set(ofport_->ofproto, ofport->bundle, NULL);
5338 ofport->realdev_ofp_port = realdev_ofp_port;
5339 ofport->vlandev_vid = vid;
5341 if (realdev_ofp_port) {
5342 vsp_add(ofport, realdev_ofp_port, vid);
5349 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
5351 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
5355 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
5356 OVS_EXCLUDED(ofproto->vsp_mutex)
5358 /* hmap_is_empty is thread safe. */
5359 return !hmap_is_empty(&ofproto->realdev_vid_map);
5364 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
5365 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5366 OVS_REQUIRES(ofproto->vsp_mutex)
5368 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5369 int vid = vlan_tci_to_vid(vlan_tci);
5370 const struct vlan_splinter *vsp;
5372 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5373 hash_realdev_vid(realdev_ofp_port, vid),
5374 &ofproto->realdev_vid_map) {
5375 if (vsp->realdev_ofp_port == realdev_ofp_port
5376 && vsp->vid == vid) {
5377 return vsp->vlandev_ofp_port;
5381 return realdev_ofp_port;
5384 /* Returns the OFP port number of the Linux VLAN device that corresponds to
5385 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
5386 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
5387 * 'vlan_tci' 9, it would return the port number of eth0.9.
5389 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
5390 * function just returns its 'realdev_ofp_port' argument. */
5392 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5393 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5394 OVS_EXCLUDED(ofproto->vsp_mutex)
5398 /* hmap_is_empty is thread safe, see if we can return immediately. */
5399 if (hmap_is_empty(&ofproto->realdev_vid_map)) {
5400 return realdev_ofp_port;
5402 ovs_mutex_lock(&ofproto->vsp_mutex);
5403 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
5404 ovs_mutex_unlock(&ofproto->vsp_mutex);
5408 static struct vlan_splinter *
5409 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
5411 struct vlan_splinter *vsp;
5413 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
5414 hash_ofp_port(vlandev_ofp_port),
5415 &ofproto->vlandev_map) {
5416 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5424 /* Returns the OpenFlow port number of the "real" device underlying the Linux
5425 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
5426 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
5427 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
5428 * eth0 and store 9 in '*vid'.
5430 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
5431 * VLAN device. Unless VLAN splinters are enabled, this is what this function
5434 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5435 ofp_port_t vlandev_ofp_port, int *vid)
5436 OVS_REQUIRES(ofproto->vsp_mutex)
5438 if (!hmap_is_empty(&ofproto->vlandev_map)) {
5439 const struct vlan_splinter *vsp;
5441 vsp = vlandev_find(ofproto, vlandev_ofp_port);
5446 return vsp->realdev_ofp_port;
5452 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
5453 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
5454 * 'flow->in_port' to the "real" device backing the VLAN device, sets
5455 * 'flow->vlan_tci' to the VLAN VID, and returns true. Optionally pushes the
5456 * appropriate VLAN on 'packet' if provided. Otherwise (which is always the
5457 * case unless VLAN splinters are enabled), returns false without making any
5460 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow,
5461 struct dp_packet *packet)
5462 OVS_EXCLUDED(ofproto->vsp_mutex)
5467 /* hmap_is_empty is thread safe. */
5468 if (hmap_is_empty(&ofproto->vlandev_map)) {
5472 ovs_mutex_lock(&ofproto->vsp_mutex);
5473 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
5474 ovs_mutex_unlock(&ofproto->vsp_mutex);
5479 /* Cause the flow to be processed as if it came in on the real device with
5480 * the VLAN device's VLAN ID. */
5481 flow->in_port.ofp_port = realdev;
5482 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
5485 /* Make the packet resemble the flow, so that it gets sent to an
5486 * OpenFlow controller properly, so that it looks correct for sFlow,
5487 * and so that flow_extract() will get the correct vlan_tci if it is
5488 * called on 'packet'. */
5489 eth_push_vlan(packet, htons(ETH_TYPE_VLAN), flow->vlan_tci);
5496 vsp_remove(struct ofport_dpif *port)
5498 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5499 struct vlan_splinter *vsp;
5501 ovs_mutex_lock(&ofproto->vsp_mutex);
5502 vsp = vlandev_find(ofproto, port->up.ofp_port);
5504 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
5505 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
5508 port->realdev_ofp_port = 0;
5510 VLOG_ERR("missing vlan device record");
5512 ovs_mutex_unlock(&ofproto->vsp_mutex);
5516 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
5518 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5520 ovs_mutex_lock(&ofproto->vsp_mutex);
5521 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
5522 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
5523 == realdev_ofp_port)) {
5524 struct vlan_splinter *vsp;
5526 vsp = xmalloc(sizeof *vsp);
5527 vsp->realdev_ofp_port = realdev_ofp_port;
5528 vsp->vlandev_ofp_port = port->up.ofp_port;
5531 port->realdev_ofp_port = realdev_ofp_port;
5533 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
5534 hash_ofp_port(port->up.ofp_port));
5535 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
5536 hash_realdev_vid(realdev_ofp_port, vid));
5538 VLOG_ERR("duplicate vlan device record");
5540 ovs_mutex_unlock(&ofproto->vsp_mutex);
5544 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
5546 const struct ofport_dpif *ofport = ofp_port_to_ofport(ofproto, ofp_port);
5547 return ofport ? ofport->odp_port : ODPP_NONE;
5550 struct ofport_dpif *
5551 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
5553 struct ofport_dpif *port;
5555 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
5556 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
5557 &backer->odp_to_ofport_map) {
5558 if (port->odp_port == odp_port) {
5559 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5564 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5569 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
5571 struct ofport_dpif *port;
5573 port = odp_port_to_ofport(ofproto->backer, odp_port);
5574 if (port && &ofproto->up == port->up.ofproto) {
5575 return port->up.ofp_port;
5582 ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
5583 const struct match *match, int priority,
5584 uint16_t idle_timeout,
5585 const struct ofpbuf *ofpacts,
5586 struct rule **rulep)
5588 struct ofproto_flow_mod ofm;
5589 struct rule_dpif *rule;
5592 ofm.fm.match = *match;
5593 ofm.fm.priority = priority;
5594 ofm.fm.new_cookie = htonll(0);
5595 ofm.fm.cookie = htonll(0);
5596 ofm.fm.cookie_mask = htonll(0);
5597 ofm.fm.modify_cookie = false;
5598 ofm.fm.table_id = TBL_INTERNAL;
5599 ofm.fm.command = OFPFC_ADD;
5600 ofm.fm.idle_timeout = idle_timeout;
5601 ofm.fm.hard_timeout = 0;
5602 ofm.fm.importance = 0;
5603 ofm.fm.buffer_id = 0;
5604 ofm.fm.out_port = 0;
5605 ofm.fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5606 ofm.fm.ofpacts = ofpacts->data;
5607 ofm.fm.ofpacts_len = ofpacts->size;
5609 error = ofproto_flow_mod(&ofproto->up, &ofm);
5611 VLOG_ERR_RL(&rl, "failed to add internal flow (%s)",
5612 ofperr_to_string(error));
5617 rule = rule_dpif_lookup_in_table(ofproto,
5618 ofproto_dpif_get_tables_version(ofproto),
5619 TBL_INTERNAL, &ofm.fm.match.flow,
5630 ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
5631 struct match *match, int priority)
5633 struct ofproto_flow_mod ofm;
5636 ofm.fm.match = *match;
5637 ofm.fm.priority = priority;
5638 ofm.fm.new_cookie = htonll(0);
5639 ofm.fm.cookie = htonll(0);
5640 ofm.fm.cookie_mask = htonll(0);
5641 ofm.fm.modify_cookie = false;
5642 ofm.fm.table_id = TBL_INTERNAL;
5643 ofm.fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5644 ofm.fm.command = OFPFC_DELETE_STRICT;
5646 error = ofproto_flow_mod(&ofproto->up, &ofm);
5648 VLOG_ERR_RL(&rl, "failed to delete internal flow (%s)",
5649 ofperr_to_string(error));
5656 const struct ofproto_class ofproto_dpif_class = {
5670 NULL, /* get_memory_usage. */
5671 type_get_memory_usage,
5690 port_is_lacp_current,
5691 port_get_lacp_stats,
5692 NULL, /* rule_choose_table */
5716 aa_vlan_get_queue_size,
5723 get_stp_port_status,
5728 get_rstp_port_status,
5735 is_mirror_output_bundle,
5736 forward_bpdu_changed,
5737 set_mac_table_config,
5739 set_mcast_snooping_port,
5741 NULL, /* meter_get_features */
5742 NULL, /* meter_set */
5743 NULL, /* meter_get */
5744 NULL, /* meter_del */
5745 group_alloc, /* group_alloc */
5746 group_construct, /* group_construct */
5747 group_destruct, /* group_destruct */
5748 group_dealloc, /* group_dealloc */
5749 group_modify, /* group_modify */
5750 group_get_stats, /* group_get_stats */
5751 get_datapath_version, /* get_datapath_version */