2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-dpif.h"
20 #include "ofproto/ofproto-provider.h"
27 #include "byte-order.h"
28 #include "connectivity.h"
33 #include "dynamic-string.h"
34 #include "fail-open.h"
35 #include "guarded-list.h"
39 #include "mac-learning.h"
40 #include "meta-flow.h"
41 #include "multipath.h"
42 #include "netdev-vport.h"
47 #include "odp-execute.h"
50 #include "ofp-actions.h"
51 #include "ofp-parse.h"
52 #include "ofp-print.h"
53 #include "ofproto-dpif-ipfix.h"
54 #include "ofproto-dpif-mirror.h"
55 #include "ofproto-dpif-monitor.h"
56 #include "ofproto-dpif-rid.h"
57 #include "ofproto-dpif-sflow.h"
58 #include "ofproto-dpif-upcall.h"
59 #include "ofproto-dpif-xlate.h"
60 #include "poll-loop.h"
66 #include "unaligned.h"
68 #include "vlan-bitmap.h"
71 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
73 COVERAGE_DEFINE(ofproto_dpif_expired);
74 COVERAGE_DEFINE(packet_in_overflow);
76 /* Number of implemented OpenFlow tables. */
77 enum { N_TABLES = 255 };
78 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
79 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
81 /* No bfd/cfm status change. */
82 #define NO_STATUS_CHANGE -1
91 * - Do include packets and bytes from datapath flows which have not
92 * recently been processed by a revalidator. */
93 struct ovs_mutex stats_mutex;
94 struct dpif_flow_stats stats OVS_GUARDED;
97 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes,
99 static struct rule_dpif *rule_dpif_cast(const struct rule *);
100 static void rule_expire(struct rule_dpif *);
107 * - Do include packets and bytes from datapath flows which have not
108 * recently been processed by a revalidator. */
109 struct ovs_mutex stats_mutex;
110 uint64_t packet_count OVS_GUARDED; /* Number of packets received. */
111 uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */
112 struct bucket_counter *bucket_stats OVS_GUARDED; /* Bucket statistics. */
116 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
117 struct ofproto_dpif *ofproto; /* Owning ofproto. */
118 void *aux; /* Key supplied by ofproto's client. */
119 char *name; /* Identifier for log messages. */
122 struct list ports; /* Contains "struct ofport"s. */
123 enum port_vlan_mode vlan_mode; /* VLAN mode */
124 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
125 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
126 * NULL if all VLANs are trunked. */
127 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
128 struct bond *bond; /* Nonnull iff more than one port. */
129 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
132 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
135 static void bundle_remove(struct ofport *);
136 static void bundle_update(struct ofbundle *);
137 static void bundle_destroy(struct ofbundle *);
138 static void bundle_del_port(struct ofport_dpif *);
139 static void bundle_run(struct ofbundle *);
140 static void bundle_wait(struct ofbundle *);
142 static void stp_run(struct ofproto_dpif *ofproto);
143 static void stp_wait(struct ofproto_dpif *ofproto);
144 static int set_stp_port(struct ofport *,
145 const struct ofproto_port_stp_settings *);
148 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
152 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
153 struct list bundle_node; /* In struct ofbundle's "ports" list. */
154 struct cfm *cfm; /* Connectivity Fault Management, if any. */
155 struct bfd *bfd; /* BFD, if any. */
156 bool may_enable; /* May be enabled in bonds. */
157 bool is_tunnel; /* This port is a tunnel. */
158 bool is_layer3; /* This is a layer 3 port. */
159 long long int carrier_seq; /* Carrier status changes. */
160 struct ofport_dpif *peer; /* Peer if patch port. */
163 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
164 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
165 long long int stp_state_entered;
167 /* Queue to DSCP mapping. */
168 struct ofproto_port_queue *qdscp;
171 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
173 * This is deprecated. It is only for compatibility with broken device
174 * drivers in old versions of Linux that do not properly support VLANs when
175 * VLAN devices are not used. When broken device drivers are no longer in
176 * widespread use, we will delete these interfaces. */
177 ofp_port_t realdev_ofp_port;
181 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
183 * This is deprecated. It is only for compatibility with broken device drivers
184 * in old versions of Linux that do not properly support VLANs when VLAN
185 * devices are not used. When broken device drivers are no longer in
186 * widespread use, we will delete these interfaces. */
187 struct vlan_splinter {
188 struct hmap_node realdev_vid_node;
189 struct hmap_node vlandev_node;
190 ofp_port_t realdev_ofp_port;
191 ofp_port_t vlandev_ofp_port;
195 static void vsp_remove(struct ofport_dpif *);
196 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
198 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
201 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
204 static struct ofport_dpif *
205 ofport_dpif_cast(const struct ofport *ofport)
207 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
210 static void port_run(struct ofport_dpif *);
211 static int set_bfd(struct ofport *, const struct smap *);
212 static int set_cfm(struct ofport *, const struct cfm_settings *);
213 static void ofport_update_peer(struct ofport_dpif *);
215 struct dpif_completion {
216 struct list list_node;
217 struct ofoperation *op;
220 /* Reasons that we might need to revalidate every datapath flow, and
221 * corresponding coverage counters.
223 * A value of 0 means that there is no need to revalidate.
225 * It would be nice to have some cleaner way to integrate with coverage
226 * counters, but with only a few reasons I guess this is good enough for
228 enum revalidate_reason {
229 REV_RECONFIGURE = 1, /* Switch configuration changed. */
230 REV_STP, /* Spanning tree protocol port status change. */
231 REV_BOND, /* Bonding changed. */
232 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
233 REV_FLOW_TABLE, /* Flow table changed. */
234 REV_MAC_LEARNING, /* Mac learning changed. */
236 COVERAGE_DEFINE(rev_reconfigure);
237 COVERAGE_DEFINE(rev_stp);
238 COVERAGE_DEFINE(rev_bond);
239 COVERAGE_DEFINE(rev_port_toggled);
240 COVERAGE_DEFINE(rev_flow_table);
241 COVERAGE_DEFINE(rev_mac_learning);
243 /* All datapaths of a given type share a single dpif backer instance. */
250 struct ovs_rwlock odp_to_ofport_lock;
251 struct hmap odp_to_ofport_map OVS_GUARDED; /* Contains "struct ofport"s. */
253 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
255 enum revalidate_reason need_revalidate; /* Revalidate all flows. */
257 bool recv_set_enable; /* Enables or disables receiving packets. */
260 struct recirc_id_pool *rid_pool; /* Recirculation ID pool. */
261 bool enable_recirc; /* True if the datapath supports recirculation */
263 /* True if the datapath supports variable-length
264 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.
265 * False if the datapath supports only 8-byte (or shorter) userdata. */
266 bool variable_length_userdata;
268 /* Maximum number of MPLS label stack entries that the datapath supports
270 size_t max_mpls_depth;
273 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
274 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
276 struct ofproto_dpif {
277 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
279 struct dpif_backer *backer;
281 uint64_t dump_seq; /* Last read of udpif_dump_seq(). */
283 /* Special OpenFlow rules. */
284 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
285 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
286 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
289 struct netflow *netflow;
290 struct dpif_sflow *sflow;
291 struct dpif_ipfix *ipfix;
292 struct hmap bundles; /* Contains "struct ofbundle"s. */
293 struct mac_learning *ml;
294 bool has_bonded_bundles;
296 struct mbridge *mbridge;
298 struct ovs_mutex stats_mutex;
299 struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
300 * consumed in userspace. */
304 long long int stp_last_tick;
306 /* VLAN splinters. */
307 struct ovs_mutex vsp_mutex;
308 struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
309 struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */
312 struct sset ports; /* Set of standard port names. */
313 struct sset ghost_ports; /* Ports with no datapath port. */
314 struct sset port_poll_set; /* Queued names for port_poll() reply. */
315 int port_poll_errno; /* Last errno for port_poll() reply. */
316 uint64_t change_seq; /* Connectivity status changes. */
319 struct guarded_list pins; /* Contains "struct ofputil_packet_in"s. */
320 struct seq *pins_seq; /* For notifying 'pins' reception. */
324 /* All existing ofproto_dpif instances, indexed by ->up.name. */
325 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
327 static void ofproto_dpif_unixctl_init(void);
329 static inline struct ofproto_dpif *
330 ofproto_dpif_cast(const struct ofproto *ofproto)
332 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
333 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
337 ofproto_dpif_get_max_mpls_depth(const struct ofproto_dpif *ofproto)
339 return ofproto->backer->max_mpls_depth;
343 ofproto_dpif_get_enable_recirc(const struct ofproto_dpif *ofproto)
345 return ofproto->backer->enable_recirc;
348 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
349 ofp_port_t ofp_port);
350 static void ofproto_trace(struct ofproto_dpif *, struct flow *,
351 const struct ofpbuf *packet,
352 const struct ofpact[], size_t ofpacts_len,
355 /* Global variables. */
356 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
358 /* Initial mappings of port to bridge mappings. */
359 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
361 /* Executes 'fm'. The caller retains ownership of 'fm' and everything in
364 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
365 struct ofputil_flow_mod *fm)
367 ofproto_flow_mod(&ofproto->up, fm);
370 /* Resets the modified time for 'rule' or an equivalent rule. If 'rule' is not
371 * in the classifier, but an equivalent rule is, unref 'rule' and ref the new
372 * rule. Otherwise if 'rule' is no longer installed in the classifier,
375 * Returns the rule whose modified time has been reset. */
377 ofproto_dpif_refresh_rule(struct rule_dpif *rule)
379 return rule_dpif_cast(ofproto_refresh_rule(&rule->up));
382 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
383 * Takes ownership of 'pin' and pin->packet. */
385 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
386 struct ofproto_packet_in *pin)
388 if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
389 COVERAGE_INC(packet_in_overflow);
390 free(CONST_CAST(void *, pin->up.packet));
394 /* Wakes up main thread for packet-in I/O. */
395 seq_change(ofproto->pins_seq);
398 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
399 * packet rather than to send the packet to the controller.
401 * This function returns false to indicate that a packet_in message
402 * for a "table-miss" should be sent to at least one controller.
403 * False otherwise. */
405 ofproto_dpif_wants_packet_in_on_miss(struct ofproto_dpif *ofproto)
407 return connmgr_wants_packet_in_on_miss(ofproto->up.connmgr);
410 /* Factory functions. */
413 init(const struct shash *iface_hints)
415 struct shash_node *node;
417 /* Make a local copy, since we don't own 'iface_hints' elements. */
418 SHASH_FOR_EACH(node, iface_hints) {
419 const struct iface_hint *orig_hint = node->data;
420 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
422 new_hint->br_name = xstrdup(orig_hint->br_name);
423 new_hint->br_type = xstrdup(orig_hint->br_type);
424 new_hint->ofp_port = orig_hint->ofp_port;
426 shash_add(&init_ofp_ports, node->name, new_hint);
431 enumerate_types(struct sset *types)
433 dp_enumerate_types(types);
437 enumerate_names(const char *type, struct sset *names)
439 struct ofproto_dpif *ofproto;
442 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
443 if (strcmp(type, ofproto->up.type)) {
446 sset_add(names, ofproto->up.name);
453 del(const char *type, const char *name)
458 error = dpif_open(name, type, &dpif);
460 error = dpif_delete(dpif);
467 port_open_type(const char *datapath_type, const char *port_type)
469 return dpif_port_open_type(datapath_type, port_type);
472 /* Type functions. */
474 static void process_dpif_port_changes(struct dpif_backer *);
475 static void process_dpif_all_ports_changed(struct dpif_backer *);
476 static void process_dpif_port_change(struct dpif_backer *,
477 const char *devname);
478 static void process_dpif_port_error(struct dpif_backer *, int error);
480 static struct ofproto_dpif *
481 lookup_ofproto_dpif_by_port_name(const char *name)
483 struct ofproto_dpif *ofproto;
485 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
486 if (sset_contains(&ofproto->ports, name)) {
495 type_run(const char *type)
497 struct dpif_backer *backer;
499 backer = shash_find_data(&all_dpif_backers, type);
501 /* This is not necessarily a problem, since backers are only
502 * created on demand. */
506 dpif_run(backer->dpif);
508 /* If vswitchd started with other_config:flow_restore_wait set as "true",
509 * and the configuration has now changed to "false", enable receiving
510 * packets from the datapath. */
511 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
514 backer->recv_set_enable = true;
516 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
518 VLOG_ERR("Failed to enable receiving packets in dpif.");
521 dpif_flow_flush(backer->dpif);
522 backer->need_revalidate = REV_RECONFIGURE;
525 if (backer->recv_set_enable) {
526 udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
529 if (backer->need_revalidate) {
530 struct ofproto_dpif *ofproto;
531 struct simap_node *node;
532 struct simap tmp_backers;
534 /* Handle tunnel garbage collection. */
535 simap_init(&tmp_backers);
536 simap_swap(&backer->tnl_backers, &tmp_backers);
538 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
539 struct ofport_dpif *iter;
541 if (backer != ofproto->backer) {
545 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
546 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
549 if (!iter->is_tunnel) {
553 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
554 namebuf, sizeof namebuf);
555 node = simap_find(&tmp_backers, dp_port);
557 simap_put(&backer->tnl_backers, dp_port, node->data);
558 simap_delete(&tmp_backers, node);
559 node = simap_find(&backer->tnl_backers, dp_port);
561 node = simap_find(&backer->tnl_backers, dp_port);
563 odp_port_t odp_port = ODPP_NONE;
565 if (!dpif_port_add(backer->dpif, iter->up.netdev,
567 simap_put(&backer->tnl_backers, dp_port,
568 odp_to_u32(odp_port));
569 node = simap_find(&backer->tnl_backers, dp_port);
574 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
575 if (tnl_port_reconfigure(iter, iter->up.netdev,
577 backer->need_revalidate = REV_RECONFIGURE;
582 SIMAP_FOR_EACH (node, &tmp_backers) {
583 dpif_port_del(backer->dpif, u32_to_odp(node->data));
585 simap_destroy(&tmp_backers);
587 switch (backer->need_revalidate) {
588 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
589 case REV_STP: COVERAGE_INC(rev_stp); break;
590 case REV_BOND: COVERAGE_INC(rev_bond); break;
591 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
592 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
593 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
595 backer->need_revalidate = 0;
597 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
598 struct ofport_dpif *ofport;
599 struct ofbundle *bundle;
601 if (ofproto->backer != backer) {
605 ovs_rwlock_wrlock(&xlate_rwlock);
606 xlate_ofproto_set(ofproto, ofproto->up.name,
607 ofproto->backer->dpif, ofproto->miss_rule,
608 ofproto->no_packet_in_rule, ofproto->ml,
609 ofproto->stp, ofproto->mbridge,
610 ofproto->sflow, ofproto->ipfix,
611 ofproto->netflow, ofproto->up.frag_handling,
612 ofproto->up.forward_bpdu,
613 connmgr_has_in_band(ofproto->up.connmgr),
614 ofproto->backer->enable_recirc,
615 ofproto->backer->variable_length_userdata,
616 ofproto->backer->max_mpls_depth);
618 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
619 xlate_bundle_set(ofproto, bundle, bundle->name,
620 bundle->vlan_mode, bundle->vlan,
621 bundle->trunks, bundle->use_priority_tags,
622 bundle->bond, bundle->lacp,
626 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
627 int stp_port = ofport->stp_port
628 ? stp_port_no(ofport->stp_port)
630 xlate_ofport_set(ofproto, ofport->bundle, ofport,
631 ofport->up.ofp_port, ofport->odp_port,
632 ofport->up.netdev, ofport->cfm,
633 ofport->bfd, ofport->peer, stp_port,
634 ofport->qdscp, ofport->n_qdscp,
635 ofport->up.pp.config, ofport->up.pp.state,
636 ofport->is_tunnel, ofport->may_enable);
638 ovs_rwlock_unlock(&xlate_rwlock);
641 udpif_revalidate(backer->udpif);
644 process_dpif_port_changes(backer);
649 /* Check for and handle port changes in 'backer''s dpif. */
651 process_dpif_port_changes(struct dpif_backer *backer)
657 error = dpif_port_poll(backer->dpif, &devname);
663 process_dpif_all_ports_changed(backer);
667 process_dpif_port_change(backer, devname);
672 process_dpif_port_error(backer, error);
679 process_dpif_all_ports_changed(struct dpif_backer *backer)
681 struct ofproto_dpif *ofproto;
682 struct dpif_port dpif_port;
683 struct dpif_port_dump dump;
684 struct sset devnames;
687 sset_init(&devnames);
688 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
689 if (ofproto->backer == backer) {
690 struct ofport *ofport;
692 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
693 sset_add(&devnames, netdev_get_name(ofport->netdev));
697 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
698 sset_add(&devnames, dpif_port.name);
701 SSET_FOR_EACH (devname, &devnames) {
702 process_dpif_port_change(backer, devname);
704 sset_destroy(&devnames);
708 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
710 struct ofproto_dpif *ofproto;
711 struct dpif_port port;
713 /* Don't report on the datapath's device. */
714 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
718 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
719 &all_ofproto_dpifs) {
720 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
725 ofproto = lookup_ofproto_dpif_by_port_name(devname);
726 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
727 /* The port was removed. If we know the datapath,
728 * report it through poll_set(). If we don't, it may be
729 * notifying us of a removal we initiated, so ignore it.
730 * If there's a pending ENOBUFS, let it stand, since
731 * everything will be reevaluated. */
732 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
733 sset_add(&ofproto->port_poll_set, devname);
734 ofproto->port_poll_errno = 0;
736 } else if (!ofproto) {
737 /* The port was added, but we don't know with which
738 * ofproto we should associate it. Delete it. */
739 dpif_port_del(backer->dpif, port.port_no);
741 struct ofport_dpif *ofport;
743 ofport = ofport_dpif_cast(shash_find_data(
744 &ofproto->up.port_by_name, devname));
746 && ofport->odp_port != port.port_no
747 && !odp_port_to_ofport(backer, port.port_no))
749 /* 'ofport''s datapath port number has changed from
750 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
751 * structures to match. */
752 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
753 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
754 ofport->odp_port = port.port_no;
755 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
756 hash_odp_port(port.port_no));
757 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
758 backer->need_revalidate = REV_RECONFIGURE;
761 dpif_port_destroy(&port);
764 /* Propagate 'error' to all ofprotos based on 'backer'. */
766 process_dpif_port_error(struct dpif_backer *backer, int error)
768 struct ofproto_dpif *ofproto;
770 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
771 if (ofproto->backer == backer) {
772 sset_clear(&ofproto->port_poll_set);
773 ofproto->port_poll_errno = error;
779 type_wait(const char *type)
781 struct dpif_backer *backer;
783 backer = shash_find_data(&all_dpif_backers, type);
785 /* This is not necessarily a problem, since backers are only
786 * created on demand. */
790 dpif_wait(backer->dpif);
793 /* Basic life-cycle. */
795 static int add_internal_flows(struct ofproto_dpif *);
797 static struct ofproto *
800 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
805 dealloc(struct ofproto *ofproto_)
807 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
812 close_dpif_backer(struct dpif_backer *backer)
814 ovs_assert(backer->refcount > 0);
816 if (--backer->refcount) {
820 udpif_destroy(backer->udpif);
822 simap_destroy(&backer->tnl_backers);
823 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
824 hmap_destroy(&backer->odp_to_ofport_map);
825 shash_find_and_delete(&all_dpif_backers, backer->type);
826 recirc_id_pool_destroy(backer->rid_pool);
828 dpif_close(backer->dpif);
832 /* Datapath port slated for removal from datapath. */
834 struct list list_node;
838 static bool check_variable_length_userdata(struct dpif_backer *backer);
839 static size_t check_max_mpls_depth(struct dpif_backer *backer);
840 static bool check_recirc(struct dpif_backer *backer);
843 open_dpif_backer(const char *type, struct dpif_backer **backerp)
845 struct dpif_backer *backer;
846 struct dpif_port_dump port_dump;
847 struct dpif_port port;
848 struct shash_node *node;
849 struct list garbage_list;
850 struct odp_garbage *garbage, *next;
857 backer = shash_find_data(&all_dpif_backers, type);
864 backer_name = xasprintf("ovs-%s", type);
866 /* Remove any existing datapaths, since we assume we're the only
867 * userspace controlling the datapath. */
869 dp_enumerate_names(type, &names);
870 SSET_FOR_EACH(name, &names) {
871 struct dpif *old_dpif;
873 /* Don't remove our backer if it exists. */
874 if (!strcmp(name, backer_name)) {
878 if (dpif_open(name, type, &old_dpif)) {
879 VLOG_WARN("couldn't open old datapath %s to remove it", name);
881 dpif_delete(old_dpif);
882 dpif_close(old_dpif);
885 sset_destroy(&names);
887 backer = xmalloc(sizeof *backer);
889 error = dpif_create_and_open(backer_name, type, &backer->dpif);
892 VLOG_ERR("failed to open datapath of type %s: %s", type,
893 ovs_strerror(error));
897 backer->udpif = udpif_create(backer, backer->dpif);
899 backer->type = xstrdup(type);
900 backer->refcount = 1;
901 hmap_init(&backer->odp_to_ofport_map);
902 ovs_rwlock_init(&backer->odp_to_ofport_lock);
903 backer->need_revalidate = 0;
904 simap_init(&backer->tnl_backers);
905 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
908 if (backer->recv_set_enable) {
909 dpif_flow_flush(backer->dpif);
912 /* Loop through the ports already on the datapath and remove any
913 * that we don't need anymore. */
914 list_init(&garbage_list);
915 dpif_port_dump_start(&port_dump, backer->dpif);
916 while (dpif_port_dump_next(&port_dump, &port)) {
917 node = shash_find(&init_ofp_ports, port.name);
918 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
919 garbage = xmalloc(sizeof *garbage);
920 garbage->odp_port = port.port_no;
921 list_push_front(&garbage_list, &garbage->list_node);
924 dpif_port_dump_done(&port_dump);
926 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
927 dpif_port_del(backer->dpif, garbage->odp_port);
928 list_remove(&garbage->list_node);
932 shash_add(&all_dpif_backers, type, backer);
934 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
936 VLOG_ERR("failed to listen on datapath of type %s: %s",
937 type, ovs_strerror(error));
938 close_dpif_backer(backer);
941 backer->enable_recirc = check_recirc(backer);
942 backer->variable_length_userdata = check_variable_length_userdata(backer);
943 backer->max_mpls_depth = check_max_mpls_depth(backer);
944 backer->rid_pool = recirc_id_pool_create();
946 if (backer->recv_set_enable) {
947 udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
953 /* Tests whether 'backer''s datapath supports recirculation Only newer datapath
954 * supports OVS_KEY_ATTR in OVS_ACTION_ATTR_USERSPACE actions. We need to
955 * disable some features on older datapaths that don't support this feature.
957 * Returns false if 'backer' definitely does not support recirculation, true if
958 * it seems to support recirculation or if at least the error we get is
961 check_recirc(struct dpif_backer *backer)
964 struct odputil_keybuf keybuf;
967 bool enable_recirc = false;
969 memset(&flow, 0, sizeof flow);
973 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
974 odp_flow_key_from_flow(&key, &flow, NULL, 0);
976 error = dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
977 ofpbuf_data(&key), ofpbuf_size(&key), NULL, 0, NULL,
979 if (error && error != EEXIST) {
980 if (error != EINVAL) {
981 VLOG_WARN("%s: Reciculation flow probe failed (%s)",
982 dpif_name(backer->dpif), ovs_strerror(error));
987 error = dpif_flow_del(backer->dpif, ofpbuf_data(&key), ofpbuf_size(&key),
990 VLOG_WARN("%s: failed to delete recirculation feature probe flow",
991 dpif_name(backer->dpif));
994 enable_recirc = true;
998 VLOG_INFO("%s: Datapath supports recirculation",
999 dpif_name(backer->dpif));
1001 VLOG_INFO("%s: Datapath does not support recirculation",
1002 dpif_name(backer->dpif));
1005 return enable_recirc;
1008 /* Tests whether 'backer''s datapath supports variable-length
1009 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions. We need
1010 * to disable some features on older datapaths that don't support this
1013 * Returns false if 'backer' definitely does not support variable-length
1014 * userdata, true if it seems to support them or if at least the error we get
1017 check_variable_length_userdata(struct dpif_backer *backer)
1019 struct eth_header *eth;
1020 struct ofpbuf actions;
1021 struct dpif_execute execute;
1022 struct ofpbuf packet;
1026 /* Compose a userspace action that will cause an ERANGE error on older
1027 * datapaths that don't support variable-length userdata.
1029 * We really test for using userdata longer than 8 bytes, but older
1030 * datapaths accepted these, silently truncating the userdata to 8 bytes.
1031 * The same older datapaths rejected userdata shorter than 8 bytes, so we
1032 * test for that instead as a proxy for longer userdata support. */
1033 ofpbuf_init(&actions, 64);
1034 start = nl_msg_start_nested(&actions, OVS_ACTION_ATTR_USERSPACE);
1035 nl_msg_put_u32(&actions, OVS_USERSPACE_ATTR_PID,
1036 dpif_port_get_pid(backer->dpif, ODPP_NONE, 0));
1037 nl_msg_put_unspec_zero(&actions, OVS_USERSPACE_ATTR_USERDATA, 4);
1038 nl_msg_end_nested(&actions, start);
1040 /* Compose a dummy ethernet packet. */
1041 ofpbuf_init(&packet, ETH_HEADER_LEN);
1042 eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
1043 eth->eth_type = htons(0x1234);
1045 /* Execute the actions. On older datapaths this fails with ERANGE, on
1046 * newer datapaths it succeeds. */
1047 execute.actions = ofpbuf_data(&actions);
1048 execute.actions_len = ofpbuf_size(&actions);
1049 execute.packet = &packet;
1050 execute.md = PKT_METADATA_INITIALIZER(0);
1051 execute.needs_help = false;
1053 error = dpif_execute(backer->dpif, &execute);
1055 ofpbuf_uninit(&packet);
1056 ofpbuf_uninit(&actions);
1060 /* Variable-length userdata is supported.
1062 * Purge received packets to avoid processing the nonsense packet we
1063 * sent to userspace, then report success. */
1064 dpif_recv_purge(backer->dpif);
1068 /* Variable-length userdata is not supported. */
1069 VLOG_WARN("%s: datapath does not support variable-length userdata "
1070 "feature (needs Linux 3.10+ or kernel module from OVS "
1071 "1..11+). The NXAST_SAMPLE action will be ignored.",
1072 dpif_name(backer->dpif));
1076 /* Something odd happened. We're not sure whether variable-length
1077 * userdata is supported. Default to "yes". */
1078 VLOG_WARN("%s: variable-length userdata feature probe failed (%s)",
1079 dpif_name(backer->dpif), ovs_strerror(error));
1084 /* Tests the MPLS label stack depth supported by 'backer''s datapath.
1086 * Returns the number of elements in a struct flow's mpls_lse field
1087 * if the datapath supports at least that many entries in an
1089 * Otherwise returns the number of MPLS push actions supported by
1092 check_max_mpls_depth(struct dpif_backer *backer)
1097 for (n = 0; n < FLOW_MAX_MPLS_LABELS; n++) {
1098 struct odputil_keybuf keybuf;
1102 memset(&flow, 0, sizeof flow);
1103 flow.dl_type = htons(ETH_TYPE_MPLS);
1104 flow_set_mpls_bos(&flow, n, 1);
1106 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1107 odp_flow_key_from_flow(&key, &flow, NULL, 0);
1109 error = dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
1110 ofpbuf_data(&key), ofpbuf_size(&key), NULL, 0, NULL, 0, NULL);
1111 if (error && error != EEXIST) {
1112 if (error != EINVAL) {
1113 VLOG_WARN("%s: MPLS stack length feature probe failed (%s)",
1114 dpif_name(backer->dpif), ovs_strerror(error));
1119 error = dpif_flow_del(backer->dpif, ofpbuf_data(&key), ofpbuf_size(&key), NULL);
1121 VLOG_WARN("%s: failed to delete MPLS feature probe flow",
1122 dpif_name(backer->dpif));
1126 VLOG_INFO("%s: MPLS label stack length probed as %d",
1127 dpif_name(backer->dpif), n);
1132 construct(struct ofproto *ofproto_)
1134 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1135 struct shash_node *node, *next;
1138 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1143 ofproto->netflow = NULL;
1144 ofproto->sflow = NULL;
1145 ofproto->ipfix = NULL;
1146 ofproto->stp = NULL;
1147 ofproto->dump_seq = 0;
1148 hmap_init(&ofproto->bundles);
1149 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1150 ofproto->mbridge = mbridge_create();
1151 ofproto->has_bonded_bundles = false;
1152 ofproto->lacp_enabled = false;
1153 ovs_mutex_init_adaptive(&ofproto->stats_mutex);
1154 ovs_mutex_init(&ofproto->vsp_mutex);
1156 guarded_list_init(&ofproto->pins);
1158 ofproto_dpif_unixctl_init();
1160 hmap_init(&ofproto->vlandev_map);
1161 hmap_init(&ofproto->realdev_vid_map);
1163 sset_init(&ofproto->ports);
1164 sset_init(&ofproto->ghost_ports);
1165 sset_init(&ofproto->port_poll_set);
1166 ofproto->port_poll_errno = 0;
1167 ofproto->change_seq = 0;
1168 ofproto->pins_seq = seq_create();
1169 ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1172 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1173 struct iface_hint *iface_hint = node->data;
1175 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1176 /* Check if the datapath already has this port. */
1177 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1178 sset_add(&ofproto->ports, node->name);
1181 free(iface_hint->br_name);
1182 free(iface_hint->br_type);
1184 shash_delete(&init_ofp_ports, node);
1188 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1189 hash_string(ofproto->up.name, 0));
1190 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1192 ofproto_init_tables(ofproto_, N_TABLES);
1193 error = add_internal_flows(ofproto);
1195 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1201 add_internal_miss_flow(struct ofproto_dpif *ofproto, int id,
1202 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1208 match_init_catchall(&match);
1209 match_set_reg(&match, 0, id);
1211 error = ofproto_dpif_add_internal_flow(ofproto, &match, 0, ofpacts, &rule);
1212 *rulep = error ? NULL : rule_dpif_cast(rule);
1218 add_internal_flows(struct ofproto_dpif *ofproto)
1220 struct ofpact_controller *controller;
1221 uint64_t ofpacts_stub[128 / 8];
1222 struct ofpbuf ofpacts;
1223 struct rule *unused_rulep OVS_UNUSED;
1224 struct ofpact_resubmit *resubmit;
1229 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1232 controller = ofpact_put_CONTROLLER(&ofpacts);
1233 controller->max_len = UINT16_MAX;
1234 controller->controller_id = 0;
1235 controller->reason = OFPR_NO_MATCH;
1236 ofpact_pad(&ofpacts);
1238 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1239 &ofproto->miss_rule);
1244 ofpbuf_clear(&ofpacts);
1245 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1246 &ofproto->no_packet_in_rule);
1251 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1252 &ofproto->drop_frags_rule);
1257 /* Continue non-recirculation rule lookups from table 0.
1259 * (priority=2), recirc=0, actions=resubmit(, 0)
1261 resubmit = ofpact_put_RESUBMIT(&ofpacts);
1262 resubmit->ofpact.compat = 0;
1263 resubmit->in_port = OFPP_IN_PORT;
1264 resubmit->table_id = 0;
1266 match_init_catchall(&match);
1267 match_set_recirc_id(&match, 0);
1269 error = ofproto_dpif_add_internal_flow(ofproto, &match, 2, &ofpacts,
1275 /* Drop any run away recirc rule lookups. Recirc_id has to be
1276 * non-zero when reaching this rule.
1278 * (priority=1), *, actions=drop
1280 ofpbuf_clear(&ofpacts);
1281 match_init_catchall(&match);
1282 error = ofproto_dpif_add_internal_flow(ofproto, &match, 1, &ofpacts,
1289 destruct(struct ofproto *ofproto_)
1291 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1292 struct rule_dpif *rule, *next_rule;
1293 struct ofproto_packet_in *pin, *next_pin;
1294 struct oftable *table;
1297 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1298 ovs_rwlock_wrlock(&xlate_rwlock);
1299 xlate_remove_ofproto(ofproto);
1300 ovs_rwlock_unlock(&xlate_rwlock);
1302 /* Ensure that the upcall processing threads have no remaining references
1303 * to the ofproto or anything in it. */
1304 udpif_synchronize(ofproto->backer->udpif);
1306 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1308 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1309 struct cls_cursor cursor;
1311 fat_rwlock_rdlock(&table->cls.rwlock);
1312 cls_cursor_init(&cursor, &table->cls, NULL);
1313 fat_rwlock_unlock(&table->cls.rwlock);
1314 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1315 ofproto_rule_delete(&ofproto->up, &rule->up);
1319 guarded_list_pop_all(&ofproto->pins, &pins);
1320 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1321 list_remove(&pin->list_node);
1322 free(CONST_CAST(void *, pin->up.packet));
1325 guarded_list_destroy(&ofproto->pins);
1327 mbridge_unref(ofproto->mbridge);
1329 netflow_unref(ofproto->netflow);
1330 dpif_sflow_unref(ofproto->sflow);
1331 hmap_destroy(&ofproto->bundles);
1332 mac_learning_unref(ofproto->ml);
1334 hmap_destroy(&ofproto->vlandev_map);
1335 hmap_destroy(&ofproto->realdev_vid_map);
1337 sset_destroy(&ofproto->ports);
1338 sset_destroy(&ofproto->ghost_ports);
1339 sset_destroy(&ofproto->port_poll_set);
1341 ovs_mutex_destroy(&ofproto->stats_mutex);
1342 ovs_mutex_destroy(&ofproto->vsp_mutex);
1344 seq_destroy(ofproto->pins_seq);
1346 close_dpif_backer(ofproto->backer);
1350 run(struct ofproto *ofproto_)
1352 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1353 uint64_t new_seq, new_dump_seq;
1355 if (mbridge_need_revalidate(ofproto->mbridge)) {
1356 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1357 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1358 mac_learning_flush(ofproto->ml);
1359 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1362 /* Do not perform any periodic activity required by 'ofproto' while
1363 * waiting for flow restore to complete. */
1364 if (!ofproto_get_flow_restore_wait()) {
1365 struct ofproto_packet_in *pin, *next_pin;
1368 guarded_list_pop_all(&ofproto->pins, &pins);
1369 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1370 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1371 list_remove(&pin->list_node);
1372 free(CONST_CAST(void *, pin->up.packet));
1377 /* Always updates the ofproto->pins_seqno to avoid frequent wakeup during
1378 * flow restore. Even though nothing is processed during flow restore,
1379 * all queued 'pins' will be handled immediately when flow restore
1381 ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1383 if (ofproto->netflow) {
1384 netflow_run(ofproto->netflow);
1386 if (ofproto->sflow) {
1387 dpif_sflow_run(ofproto->sflow);
1389 if (ofproto->ipfix) {
1390 dpif_ipfix_run(ofproto->ipfix);
1393 new_seq = seq_read(connectivity_seq_get());
1394 if (ofproto->change_seq != new_seq) {
1395 struct ofport_dpif *ofport;
1397 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1401 ofproto->change_seq = new_seq;
1403 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1404 struct ofbundle *bundle;
1406 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1412 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1413 if (mac_learning_run(ofproto->ml)) {
1414 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1416 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1418 new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
1419 if (ofproto->dump_seq != new_dump_seq) {
1420 struct rule *rule, *next_rule;
1422 /* We know stats are relatively fresh, so now is a good time to do some
1424 ofproto->dump_seq = new_dump_seq;
1426 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1428 ovs_mutex_lock(&ofproto_mutex);
1429 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
1430 &ofproto->up.expirable) {
1431 rule_expire(rule_dpif_cast(rule));
1433 ovs_mutex_unlock(&ofproto_mutex);
1435 /* All outstanding data in existing flows has been accounted, so it's a
1436 * good time to do bond rebalancing. */
1437 if (ofproto->has_bonded_bundles) {
1438 struct ofbundle *bundle;
1440 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1442 bond_rebalance(bundle->bond);
1452 wait(struct ofproto *ofproto_)
1454 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1456 if (ofproto_get_flow_restore_wait()) {
1460 if (ofproto->sflow) {
1461 dpif_sflow_wait(ofproto->sflow);
1463 if (ofproto->ipfix) {
1464 dpif_ipfix_wait(ofproto->ipfix);
1466 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1467 struct ofbundle *bundle;
1469 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1470 bundle_wait(bundle);
1473 if (ofproto->netflow) {
1474 netflow_wait(ofproto->netflow);
1476 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1477 mac_learning_wait(ofproto->ml);
1478 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1480 if (ofproto->backer->need_revalidate) {
1481 /* Shouldn't happen, but if it does just go around again. */
1482 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1483 poll_immediate_wake();
1486 seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
1487 seq_wait(ofproto->pins_seq, ofproto->pins_seqno);
1491 type_get_memory_usage(const char *type, struct simap *usage)
1493 struct dpif_backer *backer;
1495 backer = shash_find_data(&all_dpif_backers, type);
1497 udpif_get_memory_usage(backer->udpif, usage);
1502 flush(struct ofproto *ofproto_)
1504 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1505 struct dpif_backer *backer = ofproto->backer;
1508 udpif_flush(backer->udpif);
1513 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1514 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1516 *arp_match_ip = true;
1517 *actions = (OFPUTIL_A_OUTPUT |
1518 OFPUTIL_A_SET_VLAN_VID |
1519 OFPUTIL_A_SET_VLAN_PCP |
1520 OFPUTIL_A_STRIP_VLAN |
1521 OFPUTIL_A_SET_DL_SRC |
1522 OFPUTIL_A_SET_DL_DST |
1523 OFPUTIL_A_SET_NW_SRC |
1524 OFPUTIL_A_SET_NW_DST |
1525 OFPUTIL_A_SET_NW_TOS |
1526 OFPUTIL_A_SET_TP_SRC |
1527 OFPUTIL_A_SET_TP_DST |
1532 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1534 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1535 struct dpif_dp_stats s;
1536 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1540 strcpy(ots->name, "classifier");
1542 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1543 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes, &used);
1544 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes,
1546 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes,
1548 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1549 ots->lookup_count = htonll(n_lookup);
1550 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1553 static struct ofport *
1556 struct ofport_dpif *port = xmalloc(sizeof *port);
1561 port_dealloc(struct ofport *port_)
1563 struct ofport_dpif *port = ofport_dpif_cast(port_);
1568 port_construct(struct ofport *port_)
1570 struct ofport_dpif *port = ofport_dpif_cast(port_);
1571 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1572 const struct netdev *netdev = port->up.netdev;
1573 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1574 struct dpif_port dpif_port;
1577 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1578 port->bundle = NULL;
1581 port->may_enable = true;
1582 port->stp_port = NULL;
1583 port->stp_state = STP_DISABLED;
1584 port->is_tunnel = false;
1588 port->realdev_ofp_port = 0;
1589 port->vlandev_vid = 0;
1590 port->carrier_seq = netdev_get_carrier_resets(netdev);
1591 port->is_layer3 = netdev_vport_is_layer3(netdev);
1593 if (netdev_vport_is_patch(netdev)) {
1594 /* By bailing out here, we don't submit the port to the sFlow module
1595 * to be considered for counter polling export. This is correct
1596 * because the patch port represents an interface that sFlow considers
1597 * to be "internal" to the switch as a whole, and therefore not an
1598 * candidate for counter polling. */
1599 port->odp_port = ODPP_NONE;
1600 ofport_update_peer(port);
1604 error = dpif_port_query_by_name(ofproto->backer->dpif,
1605 netdev_vport_get_dpif_port(netdev, namebuf,
1612 port->odp_port = dpif_port.port_no;
1614 if (netdev_get_tunnel_config(netdev)) {
1615 tnl_port_add(port, port->up.netdev, port->odp_port);
1616 port->is_tunnel = true;
1618 /* Sanity-check that a mapping doesn't already exist. This
1619 * shouldn't happen for non-tunnel ports. */
1620 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1621 VLOG_ERR("port %s already has an OpenFlow port number",
1623 dpif_port_destroy(&dpif_port);
1627 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1628 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1629 hash_odp_port(port->odp_port));
1630 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1632 dpif_port_destroy(&dpif_port);
1634 if (ofproto->sflow) {
1635 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1642 port_destruct(struct ofport *port_)
1644 struct ofport_dpif *port = ofport_dpif_cast(port_);
1645 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1646 const char *devname = netdev_get_name(port->up.netdev);
1647 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1648 const char *dp_port_name;
1650 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1651 ovs_rwlock_wrlock(&xlate_rwlock);
1652 xlate_ofport_remove(port);
1653 ovs_rwlock_unlock(&xlate_rwlock);
1655 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1657 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1658 /* The underlying device is still there, so delete it. This
1659 * happens when the ofproto is being destroyed, since the caller
1660 * assumes that removal of attached ports will happen as part of
1662 if (!port->is_tunnel) {
1663 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1668 port->peer->peer = NULL;
1672 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1673 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1674 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1675 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1679 sset_find_and_delete(&ofproto->ports, devname);
1680 sset_find_and_delete(&ofproto->ghost_ports, devname);
1681 bundle_remove(port_);
1682 set_cfm(port_, NULL);
1683 set_bfd(port_, NULL);
1684 if (port->stp_port) {
1685 stp_port_disable(port->stp_port);
1687 if (ofproto->sflow) {
1688 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1695 port_modified(struct ofport *port_)
1697 struct ofport_dpif *port = ofport_dpif_cast(port_);
1699 if (port->bundle && port->bundle->bond) {
1700 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1704 cfm_set_netdev(port->cfm, port->up.netdev);
1708 bfd_set_netdev(port->bfd, port->up.netdev);
1711 ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1712 port->up.pp.hw_addr);
1714 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1716 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1720 ofport_update_peer(port);
1724 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1726 struct ofport_dpif *port = ofport_dpif_cast(port_);
1727 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1728 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1730 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1731 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1732 OFPUTIL_PC_NO_PACKET_IN)) {
1733 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1735 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1736 bundle_update(port->bundle);
1742 set_sflow(struct ofproto *ofproto_,
1743 const struct ofproto_sflow_options *sflow_options)
1745 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1746 struct dpif_sflow *ds = ofproto->sflow;
1748 if (sflow_options) {
1750 struct ofport_dpif *ofport;
1752 ds = ofproto->sflow = dpif_sflow_create();
1753 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1754 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1756 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1758 dpif_sflow_set_options(ds, sflow_options);
1761 dpif_sflow_unref(ds);
1762 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1763 ofproto->sflow = NULL;
1771 struct ofproto *ofproto_,
1772 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1773 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1774 size_t n_flow_exporters_options)
1776 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1777 struct dpif_ipfix *di = ofproto->ipfix;
1778 bool has_options = bridge_exporter_options || flow_exporters_options;
1780 if (has_options && !di) {
1781 di = ofproto->ipfix = dpif_ipfix_create();
1785 /* Call set_options in any case to cleanly flush the flow
1786 * caches in the last exporters that are to be destroyed. */
1787 dpif_ipfix_set_options(
1788 di, bridge_exporter_options, flow_exporters_options,
1789 n_flow_exporters_options);
1792 dpif_ipfix_unref(di);
1793 ofproto->ipfix = NULL;
1801 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1803 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1808 struct ofproto_dpif *ofproto;
1810 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1811 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1812 ofport->cfm = cfm_create(ofport->up.netdev);
1815 if (cfm_configure(ofport->cfm, s)) {
1822 cfm_unref(ofport->cfm);
1825 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1826 ofport->up.pp.hw_addr);
1831 get_cfm_status(const struct ofport *ofport_,
1832 struct ofproto_cfm_status *status)
1834 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1838 if (cfm_check_status_change(ofport->cfm)) {
1839 status->faults = cfm_get_fault(ofport->cfm);
1840 status->flap_count = cfm_get_flap_count(ofport->cfm);
1841 status->remote_opstate = cfm_get_opup(ofport->cfm);
1842 status->health = cfm_get_health(ofport->cfm);
1843 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1845 ret = NO_STATUS_CHANGE;
1855 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1857 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1858 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1862 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1863 cfg, ofport->up.netdev);
1864 if (ofport->bfd != old) {
1865 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1867 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1868 ofport->up.pp.hw_addr);
1873 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1875 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1879 if (bfd_check_status_change(ofport->bfd)) {
1880 bfd_get_status(ofport->bfd, smap);
1882 ret = NO_STATUS_CHANGE;
1891 /* Spanning Tree. */
1894 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1896 struct ofproto_dpif *ofproto = ofproto_;
1897 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1898 struct ofport_dpif *ofport;
1900 ofport = stp_port_get_aux(sp);
1902 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1903 ofproto->up.name, port_num);
1905 struct eth_header *eth = ofpbuf_l2(pkt);
1907 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1908 if (eth_addr_is_zero(eth->eth_src)) {
1909 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1910 "with unknown MAC", ofproto->up.name, port_num);
1912 ofproto_dpif_send_packet(ofport, pkt);
1918 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1920 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1922 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1924 /* Only revalidate flows if the configuration changed. */
1925 if (!s != !ofproto->stp) {
1926 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1930 if (!ofproto->stp) {
1931 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1932 send_bpdu_cb, ofproto);
1933 ofproto->stp_last_tick = time_msec();
1936 stp_set_bridge_id(ofproto->stp, s->system_id);
1937 stp_set_bridge_priority(ofproto->stp, s->priority);
1938 stp_set_hello_time(ofproto->stp, s->hello_time);
1939 stp_set_max_age(ofproto->stp, s->max_age);
1940 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1942 struct ofport *ofport;
1944 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1945 set_stp_port(ofport, NULL);
1948 stp_unref(ofproto->stp);
1949 ofproto->stp = NULL;
1956 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1958 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1962 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1963 s->designated_root = stp_get_designated_root(ofproto->stp);
1964 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1973 update_stp_port_state(struct ofport_dpif *ofport)
1975 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1976 enum stp_state state;
1978 /* Figure out new state. */
1979 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1983 if (ofport->stp_state != state) {
1984 enum ofputil_port_state of_state;
1987 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1988 netdev_get_name(ofport->up.netdev),
1989 stp_state_name(ofport->stp_state),
1990 stp_state_name(state));
1991 if (stp_learn_in_state(ofport->stp_state)
1992 != stp_learn_in_state(state)) {
1993 /* xxx Learning action flows should also be flushed. */
1994 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1995 mac_learning_flush(ofproto->ml);
1996 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1998 fwd_change = stp_forward_in_state(ofport->stp_state)
1999 != stp_forward_in_state(state);
2001 ofproto->backer->need_revalidate = REV_STP;
2002 ofport->stp_state = state;
2003 ofport->stp_state_entered = time_msec();
2005 if (fwd_change && ofport->bundle) {
2006 bundle_update(ofport->bundle);
2009 /* Update the STP state bits in the OpenFlow port description. */
2010 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2011 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2012 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2013 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2014 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2016 ofproto_port_set_state(&ofport->up, of_state);
2020 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2021 * caller is responsible for assigning STP port numbers and ensuring
2022 * there are no duplicates. */
2024 set_stp_port(struct ofport *ofport_,
2025 const struct ofproto_port_stp_settings *s)
2027 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2028 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2029 struct stp_port *sp = ofport->stp_port;
2031 if (!s || !s->enable) {
2033 ofport->stp_port = NULL;
2034 stp_port_disable(sp);
2035 update_stp_port_state(ofport);
2038 } else if (sp && stp_port_no(sp) != s->port_num
2039 && ofport == stp_port_get_aux(sp)) {
2040 /* The port-id changed, so disable the old one if it's not
2041 * already in use by another port. */
2042 stp_port_disable(sp);
2045 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2046 stp_port_enable(sp);
2048 stp_port_set_aux(sp, ofport);
2049 stp_port_set_priority(sp, s->priority);
2050 stp_port_set_path_cost(sp, s->path_cost);
2052 update_stp_port_state(ofport);
2058 get_stp_port_status(struct ofport *ofport_,
2059 struct ofproto_port_stp_status *s)
2061 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2062 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2063 struct stp_port *sp = ofport->stp_port;
2065 if (!ofproto->stp || !sp) {
2071 s->port_id = stp_port_get_id(sp);
2072 s->state = stp_port_get_state(sp);
2073 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2074 s->role = stp_port_get_role(sp);
2080 get_stp_port_stats(struct ofport *ofport_,
2081 struct ofproto_port_stp_stats *s)
2083 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2084 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2085 struct stp_port *sp = ofport->stp_port;
2087 if (!ofproto->stp || !sp) {
2093 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2099 stp_run(struct ofproto_dpif *ofproto)
2102 long long int now = time_msec();
2103 long long int elapsed = now - ofproto->stp_last_tick;
2104 struct stp_port *sp;
2107 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2108 ofproto->stp_last_tick = now;
2110 while (stp_get_changed_port(ofproto->stp, &sp)) {
2111 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2114 update_stp_port_state(ofport);
2118 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2119 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2120 mac_learning_flush(ofproto->ml);
2121 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2127 stp_wait(struct ofproto_dpif *ofproto)
2130 poll_timer_wait(1000);
2135 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2138 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2139 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2141 if (ofport->n_qdscp != n_qdscp
2142 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2143 n_qdscp * sizeof *qdscp))) {
2144 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2145 free(ofport->qdscp);
2146 ofport->qdscp = n_qdscp
2147 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2149 ofport->n_qdscp = n_qdscp;
2157 /* Expires all MAC learning entries associated with 'bundle' and forces its
2158 * ofproto to revalidate every flow.
2160 * Normally MAC learning entries are removed only from the ofproto associated
2161 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2162 * are removed from every ofproto. When patch ports and SLB bonds are in use
2163 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2164 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2165 * with the host from which it migrated. */
2167 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2169 struct ofproto_dpif *ofproto = bundle->ofproto;
2170 struct mac_learning *ml = ofproto->ml;
2171 struct mac_entry *mac, *next_mac;
2173 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2174 ovs_rwlock_wrlock(&ml->rwlock);
2175 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2176 if (mac->port.p == bundle) {
2178 struct ofproto_dpif *o;
2180 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2182 struct mac_entry *e;
2184 ovs_rwlock_wrlock(&o->ml->rwlock);
2185 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2187 mac_learning_expire(o->ml, e);
2189 ovs_rwlock_unlock(&o->ml->rwlock);
2194 mac_learning_expire(ml, mac);
2197 ovs_rwlock_unlock(&ml->rwlock);
2200 static struct ofbundle *
2201 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2203 struct ofbundle *bundle;
2205 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2206 &ofproto->bundles) {
2207 if (bundle->aux == aux) {
2215 bundle_update(struct ofbundle *bundle)
2217 struct ofport_dpif *port;
2219 bundle->floodable = true;
2220 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2221 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2223 || !stp_forward_in_state(port->stp_state)) {
2224 bundle->floodable = false;
2231 bundle_del_port(struct ofport_dpif *port)
2233 struct ofbundle *bundle = port->bundle;
2235 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2237 list_remove(&port->bundle_node);
2238 port->bundle = NULL;
2241 lacp_slave_unregister(bundle->lacp, port);
2244 bond_slave_unregister(bundle->bond, port);
2247 bundle_update(bundle);
2251 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2252 struct lacp_slave_settings *lacp)
2254 struct ofport_dpif *port;
2256 port = get_ofp_port(bundle->ofproto, ofp_port);
2261 if (port->bundle != bundle) {
2262 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2264 bundle_remove(&port->up);
2267 port->bundle = bundle;
2268 list_push_back(&bundle->ports, &port->bundle_node);
2269 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2271 || !stp_forward_in_state(port->stp_state)) {
2272 bundle->floodable = false;
2276 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2277 lacp_slave_register(bundle->lacp, port, lacp);
2284 bundle_destroy(struct ofbundle *bundle)
2286 struct ofproto_dpif *ofproto;
2287 struct ofport_dpif *port, *next_port;
2293 ofproto = bundle->ofproto;
2294 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2296 ovs_rwlock_wrlock(&xlate_rwlock);
2297 xlate_bundle_remove(bundle);
2298 ovs_rwlock_unlock(&xlate_rwlock);
2300 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2301 bundle_del_port(port);
2304 bundle_flush_macs(bundle, true);
2305 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2307 free(bundle->trunks);
2308 lacp_unref(bundle->lacp);
2309 bond_unref(bundle->bond);
2314 bundle_set(struct ofproto *ofproto_, void *aux,
2315 const struct ofproto_bundle_settings *s)
2317 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2318 bool need_flush = false;
2319 struct ofport_dpif *port;
2320 struct ofbundle *bundle;
2321 unsigned long *trunks;
2327 bundle_destroy(bundle_lookup(ofproto, aux));
2331 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2332 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2334 bundle = bundle_lookup(ofproto, aux);
2336 bundle = xmalloc(sizeof *bundle);
2338 bundle->ofproto = ofproto;
2339 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2340 hash_pointer(aux, 0));
2342 bundle->name = NULL;
2344 list_init(&bundle->ports);
2345 bundle->vlan_mode = PORT_VLAN_TRUNK;
2347 bundle->trunks = NULL;
2348 bundle->use_priority_tags = s->use_priority_tags;
2349 bundle->lacp = NULL;
2350 bundle->bond = NULL;
2352 bundle->floodable = true;
2353 mbridge_register_bundle(ofproto->mbridge, bundle);
2356 if (!bundle->name || strcmp(s->name, bundle->name)) {
2358 bundle->name = xstrdup(s->name);
2363 ofproto->lacp_enabled = true;
2364 if (!bundle->lacp) {
2365 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2366 bundle->lacp = lacp_create();
2368 lacp_configure(bundle->lacp, s->lacp);
2370 lacp_unref(bundle->lacp);
2371 bundle->lacp = NULL;
2374 /* Update set of ports. */
2376 for (i = 0; i < s->n_slaves; i++) {
2377 if (!bundle_add_port(bundle, s->slaves[i],
2378 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2382 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2383 struct ofport_dpif *next_port;
2385 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2386 for (i = 0; i < s->n_slaves; i++) {
2387 if (s->slaves[i] == port->up.ofp_port) {
2392 bundle_del_port(port);
2396 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2398 if (list_is_empty(&bundle->ports)) {
2399 bundle_destroy(bundle);
2403 /* Set VLAN tagging mode */
2404 if (s->vlan_mode != bundle->vlan_mode
2405 || s->use_priority_tags != bundle->use_priority_tags) {
2406 bundle->vlan_mode = s->vlan_mode;
2407 bundle->use_priority_tags = s->use_priority_tags;
2412 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2413 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2415 if (vlan != bundle->vlan) {
2416 bundle->vlan = vlan;
2420 /* Get trunked VLANs. */
2421 switch (s->vlan_mode) {
2422 case PORT_VLAN_ACCESS:
2426 case PORT_VLAN_TRUNK:
2427 trunks = CONST_CAST(unsigned long *, s->trunks);
2430 case PORT_VLAN_NATIVE_UNTAGGED:
2431 case PORT_VLAN_NATIVE_TAGGED:
2432 if (vlan != 0 && (!s->trunks
2433 || !bitmap_is_set(s->trunks, vlan)
2434 || bitmap_is_set(s->trunks, 0))) {
2435 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2437 trunks = bitmap_clone(s->trunks, 4096);
2439 trunks = bitmap_allocate1(4096);
2441 bitmap_set1(trunks, vlan);
2442 bitmap_set0(trunks, 0);
2444 trunks = CONST_CAST(unsigned long *, s->trunks);
2451 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2452 free(bundle->trunks);
2453 if (trunks == s->trunks) {
2454 bundle->trunks = vlan_bitmap_clone(trunks);
2456 bundle->trunks = trunks;
2461 if (trunks != s->trunks) {
2466 if (!list_is_short(&bundle->ports)) {
2467 bundle->ofproto->has_bonded_bundles = true;
2469 if (bond_reconfigure(bundle->bond, s->bond)) {
2470 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2473 bundle->bond = bond_create(s->bond, ofproto);
2474 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2477 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2478 bond_slave_register(bundle->bond, port,
2479 port->up.ofp_port, port->up.netdev);
2482 bond_unref(bundle->bond);
2483 bundle->bond = NULL;
2486 /* If we changed something that would affect MAC learning, un-learn
2487 * everything on this port and force flow revalidation. */
2489 bundle_flush_macs(bundle, false);
2496 bundle_remove(struct ofport *port_)
2498 struct ofport_dpif *port = ofport_dpif_cast(port_);
2499 struct ofbundle *bundle = port->bundle;
2502 bundle_del_port(port);
2503 if (list_is_empty(&bundle->ports)) {
2504 bundle_destroy(bundle);
2505 } else if (list_is_short(&bundle->ports)) {
2506 bond_unref(bundle->bond);
2507 bundle->bond = NULL;
2513 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2515 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2516 struct ofport_dpif *port = port_;
2517 uint8_t ea[ETH_ADDR_LEN];
2520 error = netdev_get_etheraddr(port->up.netdev, ea);
2522 struct ofpbuf packet;
2525 ofpbuf_init(&packet, 0);
2526 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2528 memcpy(packet_pdu, pdu, pdu_size);
2530 ofproto_dpif_send_packet(port, &packet);
2531 ofpbuf_uninit(&packet);
2533 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2534 "%s (%s)", port->bundle->name,
2535 netdev_get_name(port->up.netdev), ovs_strerror(error));
2540 bundle_send_learning_packets(struct ofbundle *bundle)
2542 struct ofproto_dpif *ofproto = bundle->ofproto;
2543 struct ofpbuf *learning_packet;
2544 int error, n_packets, n_errors;
2545 struct mac_entry *e;
2546 struct list packets;
2548 list_init(&packets);
2549 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2550 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2551 if (e->port.p != bundle) {
2554 learning_packet = bond_compose_learning_packet(bundle->bond,
2557 /* Temporarily use 'frame' as a private pointer (see below). */
2558 ovs_assert(learning_packet->frame == ofpbuf_data(learning_packet));
2559 learning_packet->frame = port_void;
2560 list_push_back(&packets, &learning_packet->list_node);
2563 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2565 error = n_packets = n_errors = 0;
2566 LIST_FOR_EACH (learning_packet, list_node, &packets) {
2568 void *port_void = learning_packet->frame;
2570 /* Restore 'frame'. */
2571 learning_packet->frame = ofpbuf_data(learning_packet);
2572 ret = ofproto_dpif_send_packet(port_void, learning_packet);
2579 ofpbuf_list_delete(&packets);
2582 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2583 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2584 "packets, last error was: %s",
2585 bundle->name, n_errors, n_packets, ovs_strerror(error));
2587 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2588 bundle->name, n_packets);
2593 bundle_run(struct ofbundle *bundle)
2596 lacp_run(bundle->lacp, send_pdu_cb);
2599 struct ofport_dpif *port;
2601 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2602 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2605 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2606 bundle->ofproto->backer->need_revalidate = REV_BOND;
2609 if (bond_should_send_learning_packets(bundle->bond)) {
2610 bundle_send_learning_packets(bundle);
2616 bundle_wait(struct ofbundle *bundle)
2619 lacp_wait(bundle->lacp);
2622 bond_wait(bundle->bond);
2629 mirror_set__(struct ofproto *ofproto_, void *aux,
2630 const struct ofproto_mirror_settings *s)
2632 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2633 struct ofbundle **srcs, **dsts;
2638 mirror_destroy(ofproto->mbridge, aux);
2642 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2643 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2645 for (i = 0; i < s->n_srcs; i++) {
2646 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2649 for (i = 0; i < s->n_dsts; i++) {
2650 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2653 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2654 s->n_dsts, s->src_vlans,
2655 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2662 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2663 uint64_t *packets, uint64_t *bytes)
2665 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2670 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2672 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2673 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2674 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2675 mac_learning_flush(ofproto->ml);
2677 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2682 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2684 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2685 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2686 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2690 forward_bpdu_changed(struct ofproto *ofproto_)
2692 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2693 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2697 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2700 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2701 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2702 mac_learning_set_idle_time(ofproto->ml, idle_time);
2703 mac_learning_set_max_entries(ofproto->ml, max_entries);
2704 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2709 static struct ofport_dpif *
2710 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2712 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2713 return ofport ? ofport_dpif_cast(ofport) : NULL;
2717 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2718 struct ofproto_port *ofproto_port,
2719 struct dpif_port *dpif_port)
2721 ofproto_port->name = dpif_port->name;
2722 ofproto_port->type = dpif_port->type;
2723 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2727 ofport_update_peer(struct ofport_dpif *ofport)
2729 const struct ofproto_dpif *ofproto;
2730 struct dpif_backer *backer;
2733 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2737 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2738 backer->need_revalidate = REV_RECONFIGURE;
2741 ofport->peer->peer = NULL;
2742 ofport->peer = NULL;
2745 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2750 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2751 struct ofport *peer_ofport;
2752 struct ofport_dpif *peer;
2755 if (ofproto->backer != backer) {
2759 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2764 peer = ofport_dpif_cast(peer_ofport);
2765 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2766 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2768 ofport->peer = peer;
2769 ofport->peer->peer = ofport;
2779 port_run(struct ofport_dpif *ofport)
2781 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2782 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2783 bool enable = netdev_get_carrier(ofport->up.netdev);
2784 bool cfm_enable = false;
2785 bool bfd_enable = false;
2787 ofport->carrier_seq = carrier_seq;
2790 int cfm_opup = cfm_get_opup(ofport->cfm);
2792 cfm_enable = !cfm_get_fault(ofport->cfm);
2794 if (cfm_opup >= 0) {
2795 cfm_enable = cfm_enable && cfm_opup;
2800 bfd_enable = bfd_forwarding(ofport->bfd);
2803 if (ofport->bfd || ofport->cfm) {
2804 enable = enable && (cfm_enable || bfd_enable);
2807 if (ofport->bundle) {
2808 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2809 if (carrier_changed) {
2810 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2814 if (ofport->may_enable != enable) {
2815 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2816 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2819 ofport->may_enable = enable;
2823 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2824 struct ofproto_port *ofproto_port)
2826 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2827 struct dpif_port dpif_port;
2830 if (sset_contains(&ofproto->ghost_ports, devname)) {
2831 const char *type = netdev_get_type_from_name(devname);
2833 /* We may be called before ofproto->up.port_by_name is populated with
2834 * the appropriate ofport. For this reason, we must get the name and
2835 * type from the netdev layer directly. */
2837 const struct ofport *ofport;
2839 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2840 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2841 ofproto_port->name = xstrdup(devname);
2842 ofproto_port->type = xstrdup(type);
2848 if (!sset_contains(&ofproto->ports, devname)) {
2851 error = dpif_port_query_by_name(ofproto->backer->dpif,
2852 devname, &dpif_port);
2854 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2860 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2862 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2863 const char *devname = netdev_get_name(netdev);
2864 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2865 const char *dp_port_name;
2867 if (netdev_vport_is_patch(netdev)) {
2868 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2872 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2873 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2874 odp_port_t port_no = ODPP_NONE;
2877 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2881 if (netdev_get_tunnel_config(netdev)) {
2882 simap_put(&ofproto->backer->tnl_backers,
2883 dp_port_name, odp_to_u32(port_no));
2887 if (netdev_get_tunnel_config(netdev)) {
2888 sset_add(&ofproto->ghost_ports, devname);
2890 sset_add(&ofproto->ports, devname);
2896 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
2898 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2899 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2906 sset_find_and_delete(&ofproto->ghost_ports,
2907 netdev_get_name(ofport->up.netdev));
2908 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2909 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
2910 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
2912 /* The caller is going to close ofport->up.netdev. If this is a
2913 * bonded port, then the bond is using that netdev, so remove it
2914 * from the bond. The client will need to reconfigure everything
2915 * after deleting ports, so then the slave will get re-added. */
2916 bundle_remove(&ofport->up);
2923 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2925 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2928 error = netdev_get_stats(ofport->up.netdev, stats);
2930 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
2931 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2933 ovs_mutex_lock(&ofproto->stats_mutex);
2934 /* ofproto->stats.tx_packets represents packets that we created
2935 * internally and sent to some port (e.g. packets sent with
2936 * ofproto_dpif_send_packet()). Account for them as if they had
2937 * come from OFPP_LOCAL and got forwarded. */
2939 if (stats->rx_packets != UINT64_MAX) {
2940 stats->rx_packets += ofproto->stats.tx_packets;
2943 if (stats->rx_bytes != UINT64_MAX) {
2944 stats->rx_bytes += ofproto->stats.tx_bytes;
2947 /* ofproto->stats.rx_packets represents packets that were received on
2948 * some port and we processed internally and dropped (e.g. STP).
2949 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2951 if (stats->tx_packets != UINT64_MAX) {
2952 stats->tx_packets += ofproto->stats.rx_packets;
2955 if (stats->tx_bytes != UINT64_MAX) {
2956 stats->tx_bytes += ofproto->stats.rx_bytes;
2958 ovs_mutex_unlock(&ofproto->stats_mutex);
2964 struct port_dump_state {
2969 struct ofproto_port port;
2974 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
2976 *statep = xzalloc(sizeof(struct port_dump_state));
2981 port_dump_next(const struct ofproto *ofproto_, void *state_,
2982 struct ofproto_port *port)
2984 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2985 struct port_dump_state *state = state_;
2986 const struct sset *sset;
2987 struct sset_node *node;
2989 if (state->has_port) {
2990 ofproto_port_destroy(&state->port);
2991 state->has_port = false;
2993 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
2994 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
2997 error = port_query_by_name(ofproto_, node->name, &state->port);
2999 *port = state->port;
3000 state->has_port = true;
3002 } else if (error != ENODEV) {
3007 if (!state->ghost) {
3008 state->ghost = true;
3011 return port_dump_next(ofproto_, state_, port);
3018 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3020 struct port_dump_state *state = state_;
3022 if (state->has_port) {
3023 ofproto_port_destroy(&state->port);
3030 port_poll(const struct ofproto *ofproto_, char **devnamep)
3032 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3034 if (ofproto->port_poll_errno) {
3035 int error = ofproto->port_poll_errno;
3036 ofproto->port_poll_errno = 0;
3040 if (sset_is_empty(&ofproto->port_poll_set)) {
3044 *devnamep = sset_pop(&ofproto->port_poll_set);
3049 port_poll_wait(const struct ofproto *ofproto_)
3051 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3052 dpif_port_poll_wait(ofproto->backer->dpif);
3056 port_is_lacp_current(const struct ofport *ofport_)
3058 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3059 return (ofport->bundle && ofport->bundle->lacp
3060 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3064 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3065 * then delete it entirely. */
3067 rule_expire(struct rule_dpif *rule)
3068 OVS_REQUIRES(ofproto_mutex)
3070 uint16_t hard_timeout, idle_timeout;
3071 long long int now = time_msec();
3074 ovs_assert(!rule->up.pending);
3076 hard_timeout = rule->up.hard_timeout;
3077 idle_timeout = rule->up.idle_timeout;
3079 /* Has 'rule' expired? */
3081 long long int modified;
3083 ovs_mutex_lock(&rule->up.mutex);
3084 modified = rule->up.modified;
3085 ovs_mutex_unlock(&rule->up.mutex);
3087 if (now > modified + hard_timeout * 1000) {
3088 reason = OFPRR_HARD_TIMEOUT;
3092 if (reason < 0 && idle_timeout) {
3095 ovs_mutex_lock(&rule->stats_mutex);
3096 used = rule->stats.used;
3097 ovs_mutex_unlock(&rule->stats_mutex);
3099 if (now > used + idle_timeout * 1000) {
3100 reason = OFPRR_IDLE_TIMEOUT;
3105 COVERAGE_INC(ofproto_dpif_expired);
3106 ofproto_rule_expire(&rule->up, reason);
3110 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3111 * 'flow' must reflect the data in 'packet'. */
3113 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3114 const struct flow *flow,
3115 struct rule_dpif *rule,
3116 const struct ofpact *ofpacts, size_t ofpacts_len,
3117 struct ofpbuf *packet)
3119 struct dpif_flow_stats stats;
3120 struct xlate_out xout;
3121 struct xlate_in xin;
3123 struct dpif_execute execute;
3126 ovs_assert((rule != NULL) != (ofpacts != NULL));
3128 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3131 rule_dpif_credit_stats(rule, &stats);
3134 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
3135 xin.ofpacts = ofpacts;
3136 xin.ofpacts_len = ofpacts_len;
3137 xin.resubmit_stats = &stats;
3138 xlate_actions(&xin, &xout);
3140 in_port = flow->in_port.ofp_port;
3141 if (in_port == OFPP_NONE) {
3142 in_port = OFPP_LOCAL;
3144 execute.actions = ofpbuf_data(&xout.odp_actions);
3145 execute.actions_len = ofpbuf_size(&xout.odp_actions);
3146 execute.packet = packet;
3147 execute.md.tunnel = flow->tunnel;
3148 execute.md.skb_priority = flow->skb_priority;
3149 execute.md.pkt_mark = flow->pkt_mark;
3150 execute.md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
3151 execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
3153 error = dpif_execute(ofproto->backer->dpif, &execute);
3155 xlate_out_uninit(&xout);
3161 rule_dpif_credit_stats(struct rule_dpif *rule,
3162 const struct dpif_flow_stats *stats)
3164 ovs_mutex_lock(&rule->stats_mutex);
3165 rule->stats.n_packets += stats->n_packets;
3166 rule->stats.n_bytes += stats->n_bytes;
3167 rule->stats.used = MAX(rule->stats.used, stats->used);
3168 ovs_mutex_unlock(&rule->stats_mutex);
3172 rule_dpif_is_fail_open(const struct rule_dpif *rule)
3174 return is_fail_open_rule(&rule->up);
3178 rule_dpif_is_table_miss(const struct rule_dpif *rule)
3180 return rule_is_table_miss(&rule->up);
3184 rule_dpif_is_internal(const struct rule_dpif *rule)
3186 return rule_is_internal(&rule->up);
3190 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
3191 OVS_REQUIRES(rule->up.mutex)
3193 return rule->up.flow_cookie;
3197 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
3198 uint16_t hard_timeout)
3200 ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
3203 /* Returns 'rule''s actions. The caller owns a reference on the returned
3204 * actions and must eventually release it (with rule_actions_unref()) to avoid
3206 struct rule_actions *
3207 rule_dpif_get_actions(const struct rule_dpif *rule)
3209 return rule_get_actions(&rule->up);
3212 /* Lookup 'flow' in table 0 of 'ofproto''s classifier.
3213 * If 'wc' is non-null, sets the fields that were relevant as part of
3214 * the lookup. Returns the table_id where a match or miss occurred.
3216 * The return value will be zero unless there was a miss and
3217 * OFPTC11_TABLE_MISS_CONTINUE is in effect for the sequence of tables
3218 * where misses occur.
3220 * The rule is returned in '*rule', which is valid at least until the next
3221 * RCU quiescent period. If the '*rule' needs to stay around longer,
3222 * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3223 * on it before this returns. */
3225 rule_dpif_lookup(struct ofproto_dpif *ofproto, struct flow *flow,
3226 struct flow_wildcards *wc, struct rule_dpif **rule,
3229 enum rule_dpif_lookup_verdict verdict;
3230 enum ofputil_port_config config = 0;
3233 if (ofproto_dpif_get_enable_recirc(ofproto)) {
3234 /* Always exactly match recirc_id since datapath supports
3237 wc->masks.recirc_id = UINT32_MAX;
3240 /* Start looking up from internal table for post recirculation flows
3241 * or packets. We can also simply send all, including normal flows
3242 * or packets to the internal table. They will not match any post
3243 * recirculation rules except the 'catch all' rule that resubmit
3246 * As an optimization, we send normal flows and packets to table 0
3247 * directly, saving one table lookup. */
3248 table_id = flow->recirc_id ? TBL_INTERNAL : 0;
3253 verdict = rule_dpif_lookup_from_table(ofproto, flow, wc, true,
3254 &table_id, rule, take_ref);
3257 case RULE_DPIF_LOOKUP_VERDICT_MATCH:
3259 case RULE_DPIF_LOOKUP_VERDICT_CONTROLLER: {
3260 struct ofport_dpif *port;
3262 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
3264 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
3265 flow->in_port.ofp_port);
3267 config = port ? port->up.pp.config : 0;
3270 case RULE_DPIF_LOOKUP_VERDICT_DROP:
3271 config = OFPUTIL_PC_NO_PACKET_IN;
3273 case RULE_DPIF_LOOKUP_VERDICT_DEFAULT:
3274 if (!connmgr_wants_packet_in_on_miss(ofproto->up.connmgr)) {
3275 config = OFPUTIL_PC_NO_PACKET_IN;
3282 choose_miss_rule(config, ofproto->miss_rule,
3283 ofproto->no_packet_in_rule, rule, take_ref);
3287 /* The returned rule is valid at least until the next RCU quiescent period.
3288 * If the '*rule' needs to stay around longer, a non-zero 'take_ref' must be
3289 * passed in to cause a reference to be taken on it before this returns. */
3290 static struct rule_dpif *
3291 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, uint8_t table_id,
3292 const struct flow *flow, struct flow_wildcards *wc,
3295 struct classifier *cls = &ofproto->up.tables[table_id].cls;
3296 const struct cls_rule *cls_rule;
3297 struct rule_dpif *rule;
3299 fat_rwlock_rdlock(&cls->rwlock);
3300 if (ofproto->up.frag_handling != OFPC_FRAG_NX_MATCH) {
3302 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
3303 if (is_ip_any(flow)) {
3304 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
3308 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
3309 if (ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3310 /* We must pretend that transport ports are unavailable. */
3311 struct flow ofpc_normal_flow = *flow;
3312 ofpc_normal_flow.tp_src = htons(0);
3313 ofpc_normal_flow.tp_dst = htons(0);
3314 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
3316 /* Must be OFPC_FRAG_DROP (we don't have OFPC_FRAG_REASM). */
3317 cls_rule = &ofproto->drop_frags_rule->up.cr;
3320 cls_rule = classifier_lookup(cls, flow, wc);
3323 cls_rule = classifier_lookup(cls, flow, wc);
3326 rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
3328 rule_dpif_ref(rule);
3330 fat_rwlock_unlock(&cls->rwlock);
3335 /* Look up 'flow' in 'ofproto''s classifier starting from table '*table_id'.
3336 * Stores the rule that was found in '*rule', or NULL if none was found.
3337 * Updates 'wc', if nonnull, to reflect the fields that were used during the
3340 * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3341 * if none is found then the table miss configuration for that table is
3342 * honored, which can result in additional lookups in other OpenFlow tables.
3343 * In this case the function updates '*table_id' to reflect the final OpenFlow
3344 * table that was searched.
3346 * If 'honor_table_miss' is false, then only one table lookup occurs, in
3351 * - RULE_DPIF_LOOKUP_VERDICT_MATCH if a rule (in '*rule') was found.
3353 * - RULE_OFPTC_TABLE_MISS_CONTROLLER if no rule was found and either:
3354 * + 'honor_table_miss' is false
3355 * + a table miss configuration specified that the packet should be
3356 * sent to the controller in this case.
3358 * - RULE_DPIF_LOOKUP_VERDICT_DROP if no rule was found, 'honor_table_miss'
3359 * is true and a table miss configuration specified that the packet
3360 * should be dropped in this case.
3362 * - RULE_DPIF_LOOKUP_VERDICT_DEFAULT if no rule was found,
3363 * 'honor_table_miss' is true and a table miss configuration has
3364 * not been specified in this case.
3366 * The rule is returned in '*rule', which is valid at least until the next
3367 * RCU quiescent period. If the '*rule' needs to stay around longer,
3368 * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3369 * on it before this returns. */
3370 enum rule_dpif_lookup_verdict
3371 rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto,
3372 const struct flow *flow,
3373 struct flow_wildcards *wc,
3374 bool honor_table_miss,
3375 uint8_t *table_id, struct rule_dpif **rule,
3380 for (next_id = *table_id;
3381 next_id < ofproto->up.n_tables;
3382 next_id++, next_id += (next_id == TBL_INTERNAL))
3384 *table_id = next_id;
3385 *rule = rule_dpif_lookup_in_table(ofproto, *table_id, flow, wc,
3388 return RULE_DPIF_LOOKUP_VERDICT_MATCH;
3389 } else if (!honor_table_miss) {
3390 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3392 switch (ofproto_table_get_config(&ofproto->up, *table_id)) {
3393 case OFPROTO_TABLE_MISS_CONTINUE:
3396 case OFPROTO_TABLE_MISS_CONTROLLER:
3397 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3399 case OFPROTO_TABLE_MISS_DROP:
3400 return RULE_DPIF_LOOKUP_VERDICT_DROP;
3402 case OFPROTO_TABLE_MISS_DEFAULT:
3403 return RULE_DPIF_LOOKUP_VERDICT_DEFAULT;
3408 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3411 /* Given a port configuration (specified as zero if there's no port), chooses
3412 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
3415 * The rule is returned in '*rule', which is valid at least until the next
3416 * RCU quiescent period. If the '*rule' needs to stay around longer,
3417 * a reference must be taken on it (rule_dpif_ref()).
3420 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
3421 struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule,
3424 *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
3426 rule_dpif_ref(*rule);
3431 rule_dpif_ref(struct rule_dpif *rule)
3434 ofproto_rule_ref(&rule->up);
3439 rule_dpif_unref(struct rule_dpif *rule)
3442 ofproto_rule_unref(&rule->up);
3447 complete_operation(struct rule_dpif *rule)
3448 OVS_REQUIRES(ofproto_mutex)
3450 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3452 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3453 ofoperation_complete(rule->up.pending, 0);
3456 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
3458 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
3461 static struct rule *
3464 struct rule_dpif *rule = xmalloc(sizeof *rule);
3469 rule_dealloc(struct rule *rule_)
3471 struct rule_dpif *rule = rule_dpif_cast(rule_);
3476 rule_construct(struct rule *rule_)
3477 OVS_NO_THREAD_SAFETY_ANALYSIS
3479 struct rule_dpif *rule = rule_dpif_cast(rule_);
3480 ovs_mutex_init_adaptive(&rule->stats_mutex);
3481 rule->stats.n_packets = 0;
3482 rule->stats.n_bytes = 0;
3483 rule->stats.used = rule->up.modified;
3488 rule_insert(struct rule *rule_)
3489 OVS_REQUIRES(ofproto_mutex)
3491 struct rule_dpif *rule = rule_dpif_cast(rule_);
3492 complete_operation(rule);
3496 rule_delete(struct rule *rule_)
3497 OVS_REQUIRES(ofproto_mutex)
3499 struct rule_dpif *rule = rule_dpif_cast(rule_);
3500 complete_operation(rule);
3504 rule_destruct(struct rule *rule_)
3506 struct rule_dpif *rule = rule_dpif_cast(rule_);
3507 ovs_mutex_destroy(&rule->stats_mutex);
3511 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
3512 long long int *used)
3514 struct rule_dpif *rule = rule_dpif_cast(rule_);
3516 ovs_mutex_lock(&rule->stats_mutex);
3517 *packets = rule->stats.n_packets;
3518 *bytes = rule->stats.n_bytes;
3519 *used = rule->stats.used;
3520 ovs_mutex_unlock(&rule->stats_mutex);
3524 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
3525 struct ofpbuf *packet)
3527 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3529 ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
3533 rule_execute(struct rule *rule, const struct flow *flow,
3534 struct ofpbuf *packet)
3536 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
3537 ofpbuf_delete(packet);
3542 rule_modify_actions(struct rule *rule_, bool reset_counters)
3543 OVS_REQUIRES(ofproto_mutex)
3545 struct rule_dpif *rule = rule_dpif_cast(rule_);
3547 if (reset_counters) {
3548 ovs_mutex_lock(&rule->stats_mutex);
3549 rule->stats.n_packets = 0;
3550 rule->stats.n_bytes = 0;
3551 ovs_mutex_unlock(&rule->stats_mutex);
3554 complete_operation(rule);
3557 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
3559 return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
3562 static struct ofgroup *
3565 struct group_dpif *group = xzalloc(sizeof *group);
3570 group_dealloc(struct ofgroup *group_)
3572 struct group_dpif *group = group_dpif_cast(group_);
3577 group_construct_stats(struct group_dpif *group)
3578 OVS_REQUIRES(group->stats_mutex)
3580 group->packet_count = 0;
3581 group->byte_count = 0;
3582 if (!group->bucket_stats) {
3583 group->bucket_stats = xcalloc(group->up.n_buckets,
3584 sizeof *group->bucket_stats);
3586 memset(group->bucket_stats, 0, group->up.n_buckets *
3587 sizeof *group->bucket_stats);
3592 group_construct(struct ofgroup *group_)
3594 struct group_dpif *group = group_dpif_cast(group_);
3595 const struct ofputil_bucket *bucket;
3597 /* Prevent group chaining because our locking structure makes it hard to
3598 * implement deadlock-free. (See xlate_group_resource_check().) */
3599 LIST_FOR_EACH (bucket, list_node, &group->up.buckets) {
3600 const struct ofpact *a;
3602 OFPACT_FOR_EACH (a, bucket->ofpacts, bucket->ofpacts_len) {
3603 if (a->type == OFPACT_GROUP) {
3604 return OFPERR_OFPGMFC_CHAINING_UNSUPPORTED;
3609 ovs_mutex_init_adaptive(&group->stats_mutex);
3610 ovs_mutex_lock(&group->stats_mutex);
3611 group_construct_stats(group);
3612 ovs_mutex_unlock(&group->stats_mutex);
3617 group_destruct__(struct group_dpif *group)
3618 OVS_REQUIRES(group->stats_mutex)
3620 free(group->bucket_stats);
3621 group->bucket_stats = NULL;
3625 group_destruct(struct ofgroup *group_)
3627 struct group_dpif *group = group_dpif_cast(group_);
3628 ovs_mutex_lock(&group->stats_mutex);
3629 group_destruct__(group);
3630 ovs_mutex_unlock(&group->stats_mutex);
3631 ovs_mutex_destroy(&group->stats_mutex);
3635 group_modify(struct ofgroup *group_, struct ofgroup *victim_)
3637 struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
3638 struct group_dpif *group = group_dpif_cast(group_);
3639 struct group_dpif *victim = group_dpif_cast(victim_);
3641 ovs_mutex_lock(&group->stats_mutex);
3642 if (victim->up.n_buckets < group->up.n_buckets) {
3643 group_destruct__(group);
3645 group_construct_stats(group);
3646 ovs_mutex_unlock(&group->stats_mutex);
3648 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3654 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
3656 struct group_dpif *group = group_dpif_cast(group_);
3658 ovs_mutex_lock(&group->stats_mutex);
3659 ogs->packet_count = group->packet_count;
3660 ogs->byte_count = group->byte_count;
3661 memcpy(ogs->bucket_stats, group->bucket_stats,
3662 group->up.n_buckets * sizeof *group->bucket_stats);
3663 ovs_mutex_unlock(&group->stats_mutex);
3669 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
3670 struct group_dpif **group)
3671 OVS_TRY_RDLOCK(true, (*group)->up.rwlock)
3673 struct ofgroup *ofgroup;
3677 found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
3678 *group = found ? group_dpif_cast(ofgroup) : NULL;
3684 group_dpif_release(struct group_dpif *group)
3685 OVS_RELEASES(group->up.rwlock)
3687 ofproto_group_release(&group->up);
3691 group_dpif_get_buckets(const struct group_dpif *group,
3692 const struct list **buckets)
3694 *buckets = &group->up.buckets;
3697 enum ofp11_group_type
3698 group_dpif_get_type(const struct group_dpif *group)
3700 return group->up.type;
3703 /* Sends 'packet' out 'ofport'.
3704 * May modify 'packet'.
3705 * Returns 0 if successful, otherwise a positive errno value. */
3707 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
3709 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3712 error = xlate_send_packet(ofport, packet);
3714 ovs_mutex_lock(&ofproto->stats_mutex);
3715 ofproto->stats.tx_packets++;
3716 ofproto->stats.tx_bytes += ofpbuf_size(packet);
3717 ovs_mutex_unlock(&ofproto->stats_mutex);
3722 set_frag_handling(struct ofproto *ofproto_,
3723 enum ofp_config_flags frag_handling)
3725 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3726 if (frag_handling != OFPC_FRAG_REASM) {
3727 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3735 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
3736 const struct flow *flow,
3737 const struct ofpact *ofpacts, size_t ofpacts_len)
3739 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3741 ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
3742 ofpacts_len, packet);
3749 set_netflow(struct ofproto *ofproto_,
3750 const struct netflow_options *netflow_options)
3752 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3754 if (netflow_options) {
3755 if (!ofproto->netflow) {
3756 ofproto->netflow = netflow_create();
3757 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3759 return netflow_set_options(ofproto->netflow, netflow_options);
3760 } else if (ofproto->netflow) {
3761 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3762 netflow_unref(ofproto->netflow);
3763 ofproto->netflow = NULL;
3770 get_netflow_ids(const struct ofproto *ofproto_,
3771 uint8_t *engine_type, uint8_t *engine_id)
3773 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3775 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
3778 static struct ofproto_dpif *
3779 ofproto_dpif_lookup(const char *name)
3781 struct ofproto_dpif *ofproto;
3783 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
3784 hash_string(name, 0), &all_ofproto_dpifs) {
3785 if (!strcmp(ofproto->up.name, name)) {
3793 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
3794 const char *argv[], void *aux OVS_UNUSED)
3796 struct ofproto_dpif *ofproto;
3799 ofproto = ofproto_dpif_lookup(argv[1]);
3801 unixctl_command_reply_error(conn, "no such bridge");
3804 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3805 mac_learning_flush(ofproto->ml);
3806 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3808 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3809 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3810 mac_learning_flush(ofproto->ml);
3811 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3815 unixctl_command_reply(conn, "table successfully flushed");
3818 static struct ofport_dpif *
3819 ofbundle_get_a_port(const struct ofbundle *bundle)
3821 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
3826 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
3827 const char *argv[], void *aux OVS_UNUSED)
3829 struct ds ds = DS_EMPTY_INITIALIZER;
3830 const struct ofproto_dpif *ofproto;
3831 const struct mac_entry *e;
3833 ofproto = ofproto_dpif_lookup(argv[1]);
3835 unixctl_command_reply_error(conn, "no such bridge");
3839 ds_put_cstr(&ds, " port VLAN MAC Age\n");
3840 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
3841 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3842 struct ofbundle *bundle = e->port.p;
3843 char name[OFP_MAX_PORT_NAME_LEN];
3845 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
3847 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
3848 name, e->vlan, ETH_ADDR_ARGS(e->mac),
3849 mac_entry_age(ofproto->ml, e));
3851 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3852 unixctl_command_reply(conn, ds_cstr(&ds));
3857 struct xlate_out xout;
3858 struct xlate_in xin;
3859 const struct flow *key;
3861 struct flow_wildcards wc;
3866 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
3868 struct rule_actions *actions;
3871 ds_put_char_multiple(result, '\t', level);
3873 ds_put_cstr(result, "No match\n");
3877 ovs_mutex_lock(&rule->up.mutex);
3878 cookie = rule->up.flow_cookie;
3879 ovs_mutex_unlock(&rule->up.mutex);
3881 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
3882 rule ? rule->up.table_id : 0, ntohll(cookie));
3883 cls_rule_format(&rule->up.cr, result);
3884 ds_put_char(result, '\n');
3886 actions = rule_dpif_get_actions(rule);
3888 ds_put_char_multiple(result, '\t', level);
3889 ds_put_cstr(result, "OpenFlow actions=");
3890 ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
3891 ds_put_char(result, '\n');
3895 trace_format_flow(struct ds *result, int level, const char *title,
3896 struct trace_ctx *trace)
3898 ds_put_char_multiple(result, '\t', level);
3899 ds_put_format(result, "%s: ", title);
3900 /* Do not report unchanged flows for resubmits. */
3901 if ((level > 0 && flow_equal(&trace->xin.flow, &trace->flow))
3902 || (level == 0 && flow_equal(&trace->xin.flow, trace->key))) {
3903 ds_put_cstr(result, "unchanged");
3905 flow_format(result, &trace->xin.flow);
3906 trace->flow = trace->xin.flow;
3908 ds_put_char(result, '\n');
3912 trace_format_regs(struct ds *result, int level, const char *title,
3913 struct trace_ctx *trace)
3917 ds_put_char_multiple(result, '\t', level);
3918 ds_put_format(result, "%s:", title);
3919 for (i = 0; i < FLOW_N_REGS; i++) {
3920 ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
3922 ds_put_char(result, '\n');
3926 trace_format_odp(struct ds *result, int level, const char *title,
3927 struct trace_ctx *trace)
3929 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
3931 ds_put_char_multiple(result, '\t', level);
3932 ds_put_format(result, "%s: ", title);
3933 format_odp_actions(result, ofpbuf_data(odp_actions),
3934 ofpbuf_size(odp_actions));
3935 ds_put_char(result, '\n');
3939 trace_format_megaflow(struct ds *result, int level, const char *title,
3940 struct trace_ctx *trace)
3944 ds_put_char_multiple(result, '\t', level);
3945 ds_put_format(result, "%s: ", title);
3946 flow_wildcards_or(&trace->wc, &trace->xout.wc, &trace->wc);
3947 match_init(&match, trace->key, &trace->wc);
3948 match_format(&match, result, OFP_DEFAULT_PRIORITY);
3949 ds_put_char(result, '\n');
3953 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
3955 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
3956 struct ds *result = trace->result;
3958 ds_put_char(result, '\n');
3959 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
3960 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
3961 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
3962 trace_format_megaflow(result, recurse + 1, "Resubmitted megaflow", trace);
3963 trace_format_rule(result, recurse + 1, rule);
3967 trace_report(struct xlate_in *xin, const char *s, int recurse)
3969 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
3970 struct ds *result = trace->result;
3972 ds_put_char_multiple(result, '\t', recurse);
3973 ds_put_cstr(result, s);
3974 ds_put_char(result, '\n');
3977 /* Parses the 'argc' elements of 'argv', ignoring argv[0]. The following
3978 * forms are supported:
3980 * - [dpname] odp_flow [-generate | packet]
3981 * - bridge br_flow [-generate | packet]
3983 * On success, initializes '*ofprotop' and 'flow' and returns NULL. On failure
3984 * returns a nonnull malloced error message. */
3985 static char * WARN_UNUSED_RESULT
3986 parse_flow_and_packet(int argc, const char *argv[],
3987 struct ofproto_dpif **ofprotop, struct flow *flow,
3988 struct ofpbuf **packetp)
3990 const struct dpif_backer *backer = NULL;
3991 const char *error = NULL;
3993 struct simap port_names = SIMAP_INITIALIZER(&port_names);
3994 struct ofpbuf *packet;
3995 struct ofpbuf odp_key;
3996 struct ofpbuf odp_mask;
3998 ofpbuf_init(&odp_key, 0);
3999 ofpbuf_init(&odp_mask, 0);
4001 /* Handle "-generate" or a hex string as the last argument. */
4002 if (!strcmp(argv[argc - 1], "-generate")) {
4003 packet = ofpbuf_new(0);
4006 error = eth_from_hex(argv[argc - 1], &packet);
4009 } else if (argc == 4) {
4010 /* The 3-argument form must end in "-generate' or a hex string. */
4016 /* odp_flow can have its in_port specified as a name instead of port no.
4017 * We do not yet know whether a given flow is a odp_flow or a br_flow.
4018 * But, to know whether a flow is odp_flow through odp_flow_from_string(),
4019 * we need to create a simap of name to port no. */
4021 const char *dp_type;
4022 if (!strncmp(argv[1], "ovs-", 4)) {
4023 dp_type = argv[1] + 4;
4027 backer = shash_find_data(&all_dpif_backers, dp_type);
4028 } else if (argc == 2) {
4029 struct shash_node *node;
4030 if (shash_count(&all_dpif_backers) == 1) {
4031 node = shash_first(&all_dpif_backers);
4032 backer = node->data;
4035 error = "Syntax error";
4038 if (backer && backer->dpif) {
4039 struct dpif_port dpif_port;
4040 struct dpif_port_dump port_dump;
4041 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
4042 simap_put(&port_names, dpif_port.name,
4043 odp_to_u32(dpif_port.port_no));
4047 /* Parse the flow and determine whether a datapath or
4048 * bridge is specified. If function odp_flow_key_from_string()
4049 * returns 0, the flow is a odp_flow. If function
4050 * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
4051 if (!odp_flow_from_string(argv[argc - 1], &port_names,
4052 &odp_key, &odp_mask)) {
4054 error = "Cannot find the datapath";
4058 if (xlate_receive(backer, NULL, ofpbuf_data(&odp_key),
4059 ofpbuf_size(&odp_key), flow,
4060 ofprotop, NULL, NULL, NULL, NULL)) {
4061 error = "Invalid datapath flow";
4065 char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
4068 m_err = xasprintf("Bad flow syntax: %s", err);
4073 error = "Must specify bridge name";
4077 *ofprotop = ofproto_dpif_lookup(argv[1]);
4079 error = "Unknown bridge name";
4085 /* Generate a packet, if requested. */
4087 if (!ofpbuf_size(packet)) {
4088 flow_compose(packet, flow);
4090 struct pkt_metadata md = pkt_metadata_from_flow(flow);
4092 /* Use the metadata from the flow and the packet argument
4093 * to reconstruct the flow. */
4094 flow_extract(packet, &md, flow);
4099 if (error && !m_err) {
4100 m_err = xstrdup(error);
4103 ofpbuf_delete(packet);
4107 ofpbuf_uninit(&odp_key);
4108 ofpbuf_uninit(&odp_mask);
4109 simap_destroy(&port_names);
4114 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
4115 void *aux OVS_UNUSED)
4117 struct ofproto_dpif *ofproto;
4118 struct ofpbuf *packet;
4122 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4127 ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
4128 unixctl_command_reply(conn, ds_cstr(&result));
4129 ds_destroy(&result);
4130 ofpbuf_delete(packet);
4132 unixctl_command_reply_error(conn, error);
4138 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
4139 const char *argv[], void *aux OVS_UNUSED)
4141 enum ofputil_protocol usable_protocols;
4142 struct ofproto_dpif *ofproto;
4143 bool enforce_consistency;
4144 struct ofpbuf ofpacts;
4145 struct ofpbuf *packet;
4150 /* Three kinds of error return values! */
4156 ofpbuf_init(&ofpacts, 0);
4158 /* Parse actions. */
4159 error = parse_ofpacts(argv[--argc], &ofpacts, &usable_protocols);
4161 unixctl_command_reply_error(conn, error);
4166 /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
4167 * consistency between the flow and the actions. With -consistent, we
4168 * enforce consistency even for a flow supported in OpenFlow 1.0. */
4169 if (!strcmp(argv[1], "-consistent")) {
4170 enforce_consistency = true;
4174 enforce_consistency = false;
4177 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4179 unixctl_command_reply_error(conn, error);
4184 /* Do the same checks as handle_packet_out() in ofproto.c.
4186 * We pass a 'table_id' of 0 to ofproto_check_ofpacts(), which isn't
4187 * strictly correct because these actions aren't in any table, but it's OK
4188 * because it 'table_id' is used only to check goto_table instructions, but
4189 * packet-outs take a list of actions and therefore it can't include
4192 * We skip the "meter" check here because meter is an instruction, not an
4193 * action, and thus cannot appear in ofpacts. */
4194 in_port = ofp_to_u16(flow.in_port.ofp_port);
4195 if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
4196 unixctl_command_reply_error(conn, "invalid in_port");
4199 if (enforce_consistency) {
4200 retval = ofpacts_check_consistency(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts),
4201 &flow, u16_to_ofp(ofproto->up.max_ports),
4202 0, 0, usable_protocols);
4204 retval = ofpacts_check(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &flow,
4205 u16_to_ofp(ofproto->up.max_ports), 0, 0,
4211 ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
4212 unixctl_command_reply_error(conn, ds_cstr(&result));
4216 ofproto_trace(ofproto, &flow, packet,
4217 ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &result);
4218 unixctl_command_reply(conn, ds_cstr(&result));
4221 ds_destroy(&result);
4222 ofpbuf_delete(packet);
4223 ofpbuf_uninit(&ofpacts);
4226 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
4227 * description of the results to 'ds'.
4229 * The trace follows a packet with the specified 'flow' through the flow
4230 * table. 'packet' may be nonnull to trace an actual packet, with consequent
4231 * side effects (if it is nonnull then its flow must be 'flow').
4233 * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
4234 * trace, otherwise the actions are determined by a flow table lookup. */
4236 ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
4237 const struct ofpbuf *packet,
4238 const struct ofpact ofpacts[], size_t ofpacts_len,
4241 struct rule_dpif *rule;
4242 struct trace_ctx trace;
4244 ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
4245 ds_put_cstr(ds, "Flow: ");
4246 flow_format(ds, flow);
4247 ds_put_char(ds, '\n');
4249 flow_wildcards_init_catchall(&trace.wc);
4253 rule_dpif_lookup(ofproto, flow, &trace.wc, &rule, false);
4255 trace_format_rule(ds, 0, rule);
4256 if (rule == ofproto->miss_rule) {
4257 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
4258 } else if (rule == ofproto->no_packet_in_rule) {
4259 ds_put_cstr(ds, "\nNo match, packets dropped because "
4260 "OFPPC_NO_PACKET_IN is set on in_port.\n");
4261 } else if (rule == ofproto->drop_frags_rule) {
4262 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
4263 "and the fragment handling mode is \"drop\".\n");
4267 if (rule || ofpacts) {
4269 trace.key = flow; /* Original flow key, used for megaflow. */
4270 trace.flow = *flow; /* May be modified by actions. */
4271 xlate_in_init(&trace.xin, ofproto, flow, rule, ntohs(flow->tcp_flags),
4274 trace.xin.ofpacts = ofpacts;
4275 trace.xin.ofpacts_len = ofpacts_len;
4277 trace.xin.resubmit_hook = trace_resubmit;
4278 trace.xin.report_hook = trace_report;
4280 xlate_actions(&trace.xin, &trace.xout);
4282 ds_put_char(ds, '\n');
4283 trace_format_flow(ds, 0, "Final flow", &trace);
4284 trace_format_megaflow(ds, 0, "Megaflow", &trace);
4286 ds_put_cstr(ds, "Datapath actions: ");
4287 format_odp_actions(ds, ofpbuf_data(&trace.xout.odp_actions),
4288 ofpbuf_size(&trace.xout.odp_actions));
4290 if (trace.xout.slow) {
4291 enum slow_path_reason slow;
4293 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
4294 "slow path because it:");
4296 slow = trace.xout.slow;
4298 enum slow_path_reason bit = rightmost_1bit(slow);
4300 ds_put_format(ds, "\n\t- %s.",
4301 slow_path_reason_to_explanation(bit));
4307 xlate_out_uninit(&trace.xout);
4311 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
4312 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
4313 * to destroy 'ofproto_shash' and free the returned value. */
4314 static const struct shash_node **
4315 get_ofprotos(struct shash *ofproto_shash)
4317 const struct ofproto_dpif *ofproto;
4319 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4320 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
4321 shash_add_nocopy(ofproto_shash, name, ofproto);
4324 return shash_sort(ofproto_shash);
4328 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
4329 const char *argv[] OVS_UNUSED,
4330 void *aux OVS_UNUSED)
4332 struct ds ds = DS_EMPTY_INITIALIZER;
4333 struct shash ofproto_shash;
4334 const struct shash_node **sorted_ofprotos;
4337 shash_init(&ofproto_shash);
4338 sorted_ofprotos = get_ofprotos(&ofproto_shash);
4339 for (i = 0; i < shash_count(&ofproto_shash); i++) {
4340 const struct shash_node *node = sorted_ofprotos[i];
4341 ds_put_format(&ds, "%s\n", node->name);
4344 shash_destroy(&ofproto_shash);
4345 free(sorted_ofprotos);
4347 unixctl_command_reply(conn, ds_cstr(&ds));
4352 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
4354 const struct shash_node **ofprotos;
4355 struct dpif_dp_stats dp_stats;
4356 struct shash ofproto_shash;
4359 dpif_get_dp_stats(backer->dpif, &dp_stats);
4361 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
4362 dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
4364 shash_init(&ofproto_shash);
4365 ofprotos = get_ofprotos(&ofproto_shash);
4366 for (i = 0; i < shash_count(&ofproto_shash); i++) {
4367 struct ofproto_dpif *ofproto = ofprotos[i]->data;
4368 const struct shash_node **ports;
4371 if (ofproto->backer != backer) {
4375 ds_put_format(ds, "\t%s:\n", ofproto->up.name);
4377 ports = shash_sort(&ofproto->up.port_by_name);
4378 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
4379 const struct shash_node *node = ports[j];
4380 struct ofport *ofport = node->data;
4382 odp_port_t odp_port;
4384 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
4387 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
4388 if (odp_port != ODPP_NONE) {
4389 ds_put_format(ds, "%"PRIu32":", odp_port);
4391 ds_put_cstr(ds, "none:");
4394 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
4397 if (!netdev_get_config(ofport->netdev, &config)) {
4398 const struct smap_node **nodes;
4401 nodes = smap_sort(&config);
4402 for (i = 0; i < smap_count(&config); i++) {
4403 const struct smap_node *node = nodes[i];
4404 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
4405 node->key, node->value);
4409 smap_destroy(&config);
4411 ds_put_char(ds, ')');
4412 ds_put_char(ds, '\n');
4416 shash_destroy(&ofproto_shash);
4421 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4422 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4424 struct ds ds = DS_EMPTY_INITIALIZER;
4425 const struct shash_node **backers;
4428 backers = shash_sort(&all_dpif_backers);
4429 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
4430 dpif_show_backer(backers[i]->data, &ds);
4434 unixctl_command_reply(conn, ds_cstr(&ds));
4439 ofproto_dpif_contains_flow(const struct ofproto_dpif *ofproto,
4440 const struct nlattr *key, size_t key_len)
4442 struct ofproto_dpif *ofp;
4445 xlate_receive(ofproto->backer, NULL, key, key_len, &flow, &ofp,
4446 NULL, NULL, NULL, NULL);
4447 return ofp == ofproto;
4451 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
4452 int argc OVS_UNUSED, const char *argv[],
4453 void *aux OVS_UNUSED)
4455 struct ds ds = DS_EMPTY_INITIALIZER;
4456 const struct dpif_flow_stats *stats;
4457 const struct ofproto_dpif *ofproto;
4458 struct dpif_flow_dump flow_dump;
4459 const struct nlattr *actions;
4460 const struct nlattr *mask;
4461 const struct nlattr *key;
4465 bool verbosity = false;
4466 struct dpif_port dpif_port;
4467 struct dpif_port_dump port_dump;
4468 struct hmap portno_names;
4472 ofproto = ofproto_dpif_lookup(argv[argc - 1]);
4474 unixctl_command_reply_error(conn, "no such bridge");
4478 if (argc > 2 && !strcmp(argv[1], "-m")) {
4482 hmap_init(&portno_names);
4483 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
4484 odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
4488 error = dpif_flow_dump_start(&flow_dump, ofproto->backer->dpif);
4492 dpif_flow_dump_state_init(ofproto->backer->dpif, &state);
4493 while (dpif_flow_dump_next(&flow_dump, state, &key, &key_len,
4494 &mask, &mask_len, &actions, &actions_len,
4496 if (!ofproto_dpif_contains_flow(ofproto, key, key_len)) {
4500 odp_flow_format(key, key_len, mask, mask_len, &portno_names, &ds,
4502 ds_put_cstr(&ds, ", ");
4503 dpif_flow_stats_format(stats, &ds);
4504 ds_put_cstr(&ds, ", actions:");
4505 format_odp_actions(&ds, actions, actions_len);
4506 ds_put_char(&ds, '\n');
4508 dpif_flow_dump_state_uninit(ofproto->backer->dpif, state);
4509 error = dpif_flow_dump_done(&flow_dump);
4514 ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
4515 unixctl_command_reply_error(conn, ds_cstr(&ds));
4517 unixctl_command_reply(conn, ds_cstr(&ds));
4519 odp_portno_names_destroy(&portno_names);
4520 hmap_destroy(&portno_names);
4525 ofproto_dpif_unixctl_init(void)
4527 static bool registered;
4533 unixctl_command_register(
4535 "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
4536 1, 3, ofproto_unixctl_trace, NULL);
4537 unixctl_command_register(
4538 "ofproto/trace-packet-out",
4539 "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
4540 2, 6, ofproto_unixctl_trace_actions, NULL);
4541 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
4542 ofproto_unixctl_fdb_flush, NULL);
4543 unixctl_command_register("fdb/show", "bridge", 1, 1,
4544 ofproto_unixctl_fdb_show, NULL);
4545 unixctl_command_register("dpif/dump-dps", "", 0, 0,
4546 ofproto_unixctl_dpif_dump_dps, NULL);
4547 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
4549 unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
4550 ofproto_unixctl_dpif_dump_flows, NULL);
4554 /* Returns true if 'rule' is an internal rule, false otherwise. */
4556 rule_is_internal(const struct rule *rule)
4558 return rule->table_id == TBL_INTERNAL;
4561 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4563 * This is deprecated. It is only for compatibility with broken device drivers
4564 * in old versions of Linux that do not properly support VLANs when VLAN
4565 * devices are not used. When broken device drivers are no longer in
4566 * widespread use, we will delete these interfaces. */
4569 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
4571 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
4572 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
4574 if (realdev_ofp_port == ofport->realdev_ofp_port
4575 && vid == ofport->vlandev_vid) {
4579 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4581 if (ofport->realdev_ofp_port) {
4584 if (realdev_ofp_port && ofport->bundle) {
4585 /* vlandevs are enslaved to their realdevs, so they are not allowed to
4586 * themselves be part of a bundle. */
4587 bundle_set(ofport_->ofproto, ofport->bundle, NULL);
4590 ofport->realdev_ofp_port = realdev_ofp_port;
4591 ofport->vlandev_vid = vid;
4593 if (realdev_ofp_port) {
4594 vsp_add(ofport, realdev_ofp_port, vid);
4601 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
4603 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
4607 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
4608 OVS_EXCLUDED(ofproto->vsp_mutex)
4610 /* hmap_is_empty is thread safe. */
4611 return !hmap_is_empty(&ofproto->realdev_vid_map);
4616 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
4617 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
4618 OVS_REQUIRES(ofproto->vsp_mutex)
4620 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
4621 int vid = vlan_tci_to_vid(vlan_tci);
4622 const struct vlan_splinter *vsp;
4624 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
4625 hash_realdev_vid(realdev_ofp_port, vid),
4626 &ofproto->realdev_vid_map) {
4627 if (vsp->realdev_ofp_port == realdev_ofp_port
4628 && vsp->vid == vid) {
4629 return vsp->vlandev_ofp_port;
4633 return realdev_ofp_port;
4636 /* Returns the OFP port number of the Linux VLAN device that corresponds to
4637 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
4638 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
4639 * 'vlan_tci' 9, it would return the port number of eth0.9.
4641 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
4642 * function just returns its 'realdev_ofp_port' argument. */
4644 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
4645 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
4646 OVS_EXCLUDED(ofproto->vsp_mutex)
4650 /* hmap_is_empty is thread safe, see if we can return immediately. */
4651 if (hmap_is_empty(&ofproto->realdev_vid_map)) {
4652 return realdev_ofp_port;
4654 ovs_mutex_lock(&ofproto->vsp_mutex);
4655 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
4656 ovs_mutex_unlock(&ofproto->vsp_mutex);
4660 static struct vlan_splinter *
4661 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
4663 struct vlan_splinter *vsp;
4665 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
4666 hash_ofp_port(vlandev_ofp_port),
4667 &ofproto->vlandev_map) {
4668 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
4676 /* Returns the OpenFlow port number of the "real" device underlying the Linux
4677 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
4678 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
4679 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
4680 * eth0 and store 9 in '*vid'.
4682 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
4683 * VLAN device. Unless VLAN splinters are enabled, this is what this function
4686 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
4687 ofp_port_t vlandev_ofp_port, int *vid)
4688 OVS_REQUIRES(ofproto->vsp_mutex)
4690 if (!hmap_is_empty(&ofproto->vlandev_map)) {
4691 const struct vlan_splinter *vsp;
4693 vsp = vlandev_find(ofproto, vlandev_ofp_port);
4698 return vsp->realdev_ofp_port;
4704 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
4705 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
4706 * 'flow->in_port' to the "real" device backing the VLAN device, sets
4707 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
4708 * always the case unless VLAN splinters are enabled), returns false without
4709 * making any changes. */
4711 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
4712 OVS_EXCLUDED(ofproto->vsp_mutex)
4717 /* hmap_is_empty is thread safe. */
4718 if (hmap_is_empty(&ofproto->vlandev_map)) {
4722 ovs_mutex_lock(&ofproto->vsp_mutex);
4723 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
4724 ovs_mutex_unlock(&ofproto->vsp_mutex);
4729 /* Cause the flow to be processed as if it came in on the real device with
4730 * the VLAN device's VLAN ID. */
4731 flow->in_port.ofp_port = realdev;
4732 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
4737 vsp_remove(struct ofport_dpif *port)
4739 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
4740 struct vlan_splinter *vsp;
4742 ovs_mutex_lock(&ofproto->vsp_mutex);
4743 vsp = vlandev_find(ofproto, port->up.ofp_port);
4745 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
4746 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
4749 port->realdev_ofp_port = 0;
4751 VLOG_ERR("missing vlan device record");
4753 ovs_mutex_unlock(&ofproto->vsp_mutex);
4757 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
4759 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
4761 ovs_mutex_lock(&ofproto->vsp_mutex);
4762 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
4763 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
4764 == realdev_ofp_port)) {
4765 struct vlan_splinter *vsp;
4767 vsp = xmalloc(sizeof *vsp);
4768 vsp->realdev_ofp_port = realdev_ofp_port;
4769 vsp->vlandev_ofp_port = port->up.ofp_port;
4772 port->realdev_ofp_port = realdev_ofp_port;
4774 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
4775 hash_ofp_port(port->up.ofp_port));
4776 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
4777 hash_realdev_vid(realdev_ofp_port, vid));
4779 VLOG_ERR("duplicate vlan device record");
4781 ovs_mutex_unlock(&ofproto->vsp_mutex);
4785 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
4787 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
4788 return ofport ? ofport->odp_port : ODPP_NONE;
4791 struct ofport_dpif *
4792 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
4794 struct ofport_dpif *port;
4796 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
4797 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
4798 &backer->odp_to_ofport_map) {
4799 if (port->odp_port == odp_port) {
4800 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
4805 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
4810 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
4812 struct ofport_dpif *port;
4814 port = odp_port_to_ofport(ofproto->backer, odp_port);
4815 if (port && &ofproto->up == port->up.ofproto) {
4816 return port->up.ofp_port;
4823 ofproto_dpif_alloc_recirc_id(struct ofproto_dpif *ofproto)
4825 struct dpif_backer *backer = ofproto->backer;
4827 return recirc_id_alloc(backer->rid_pool);
4831 ofproto_dpif_free_recirc_id(struct ofproto_dpif *ofproto, uint32_t recirc_id)
4833 struct dpif_backer *backer = ofproto->backer;
4835 recirc_id_free(backer->rid_pool, recirc_id);
4839 ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
4840 struct match *match, int priority,
4841 const struct ofpbuf *ofpacts,
4842 struct rule **rulep)
4844 struct ofputil_flow_mod fm;
4845 struct rule_dpif *rule;
4849 fm.priority = priority;
4850 fm.new_cookie = htonll(0);
4851 fm.cookie = htonll(0);
4852 fm.cookie_mask = htonll(0);
4853 fm.modify_cookie = false;
4854 fm.table_id = TBL_INTERNAL;
4855 fm.command = OFPFC_ADD;
4856 fm.idle_timeout = 0;
4857 fm.hard_timeout = 0;
4860 fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
4861 fm.ofpacts = ofpbuf_data(ofpacts);
4862 fm.ofpacts_len = ofpbuf_size(ofpacts);
4864 error = ofproto_flow_mod(&ofproto->up, &fm);
4866 VLOG_ERR_RL(&rl, "failed to add internal flow (%s)",
4867 ofperr_to_string(error));
4872 rule = rule_dpif_lookup_in_table(ofproto, TBL_INTERNAL, &match->flow,
4883 ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
4884 struct match *match, int priority)
4886 struct ofputil_flow_mod fm;
4890 fm.priority = priority;
4891 fm.new_cookie = htonll(0);
4892 fm.cookie = htonll(0);
4893 fm.cookie_mask = htonll(0);
4894 fm.modify_cookie = false;
4895 fm.table_id = TBL_INTERNAL;
4896 fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
4897 fm.command = OFPFC_DELETE_STRICT;
4899 error = ofproto_flow_mod(&ofproto->up, &fm);
4901 VLOG_ERR_RL(&rl, "failed to delete internal flow (%s)",
4902 ofperr_to_string(error));
4909 const struct ofproto_class ofproto_dpif_class = {
4923 NULL, /* get_memory_usage. */
4924 type_get_memory_usage,
4943 port_is_lacp_current,
4944 NULL, /* rule_choose_table */
4953 rule_modify_actions,
4967 get_stp_port_status,
4975 is_mirror_output_bundle,
4976 forward_bpdu_changed,
4977 set_mac_table_config,
4979 NULL, /* meter_get_features */
4980 NULL, /* meter_set */
4981 NULL, /* meter_get */
4982 NULL, /* meter_del */
4983 group_alloc, /* group_alloc */
4984 group_construct, /* group_construct */
4985 group_destruct, /* group_destruct */
4986 group_dealloc, /* group_dealloc */
4987 group_modify, /* group_modify */
4988 group_get_stats, /* group_get_stats */