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;
1354 const bool enable_recirc = ofproto_dpif_get_enable_recirc(ofproto);
1356 if (mbridge_need_revalidate(ofproto->mbridge)) {
1357 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1358 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1359 mac_learning_flush(ofproto->ml);
1360 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1363 /* Do not perform any periodic activity required by 'ofproto' while
1364 * waiting for flow restore to complete. */
1365 if (!ofproto_get_flow_restore_wait()) {
1366 struct ofproto_packet_in *pin, *next_pin;
1369 guarded_list_pop_all(&ofproto->pins, &pins);
1370 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1371 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1372 list_remove(&pin->list_node);
1373 free(CONST_CAST(void *, pin->up.packet));
1378 /* Always updates the ofproto->pins_seqno to avoid frequent wakeup during
1379 * flow restore. Even though nothing is processed during flow restore,
1380 * all queued 'pins' will be handled immediately when flow restore
1382 ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1384 if (ofproto->netflow) {
1385 netflow_run(ofproto->netflow);
1387 if (ofproto->sflow) {
1388 dpif_sflow_run(ofproto->sflow);
1390 if (ofproto->ipfix) {
1391 dpif_ipfix_run(ofproto->ipfix);
1394 new_seq = seq_read(connectivity_seq_get());
1395 if (ofproto->change_seq != new_seq) {
1396 struct ofport_dpif *ofport;
1398 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1402 ofproto->change_seq = new_seq;
1404 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1405 struct ofbundle *bundle;
1407 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1413 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1414 if (mac_learning_run(ofproto->ml)) {
1415 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1417 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1419 new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
1420 if (ofproto->dump_seq != new_dump_seq) {
1421 struct rule *rule, *next_rule;
1423 /* We know stats are relatively fresh, so now is a good time to do some
1425 ofproto->dump_seq = new_dump_seq;
1427 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1429 ovs_mutex_lock(&ofproto_mutex);
1430 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
1431 &ofproto->up.expirable) {
1432 rule_expire(rule_dpif_cast(rule));
1434 ovs_mutex_unlock(&ofproto_mutex);
1436 /* All outstanding data in existing flows has been accounted, so it's a
1437 * good time to do bond rebalancing. */
1438 if (enable_recirc && ofproto->has_bonded_bundles) {
1439 struct ofbundle *bundle;
1441 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1442 struct bond *bond = bundle->bond;
1444 if (bond && bond_may_recirc(bond, NULL, NULL)) {
1445 bond_recirculation_account(bond);
1446 if (bond_rebalance(bundle->bond)) {
1447 bond_update_post_recirc_rules(bond, true);
1458 wait(struct ofproto *ofproto_)
1460 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1462 if (ofproto_get_flow_restore_wait()) {
1466 if (ofproto->sflow) {
1467 dpif_sflow_wait(ofproto->sflow);
1469 if (ofproto->ipfix) {
1470 dpif_ipfix_wait(ofproto->ipfix);
1472 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1473 struct ofbundle *bundle;
1475 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1476 bundle_wait(bundle);
1479 if (ofproto->netflow) {
1480 netflow_wait(ofproto->netflow);
1482 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1483 mac_learning_wait(ofproto->ml);
1484 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1486 if (ofproto->backer->need_revalidate) {
1487 /* Shouldn't happen, but if it does just go around again. */
1488 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1489 poll_immediate_wake();
1492 seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
1493 seq_wait(ofproto->pins_seq, ofproto->pins_seqno);
1497 type_get_memory_usage(const char *type, struct simap *usage)
1499 struct dpif_backer *backer;
1501 backer = shash_find_data(&all_dpif_backers, type);
1503 udpif_get_memory_usage(backer->udpif, usage);
1508 flush(struct ofproto *ofproto_)
1510 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1511 struct dpif_backer *backer = ofproto->backer;
1514 udpif_flush(backer->udpif);
1519 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1520 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1522 *arp_match_ip = true;
1523 *actions = (OFPUTIL_A_OUTPUT |
1524 OFPUTIL_A_SET_VLAN_VID |
1525 OFPUTIL_A_SET_VLAN_PCP |
1526 OFPUTIL_A_STRIP_VLAN |
1527 OFPUTIL_A_SET_DL_SRC |
1528 OFPUTIL_A_SET_DL_DST |
1529 OFPUTIL_A_SET_NW_SRC |
1530 OFPUTIL_A_SET_NW_DST |
1531 OFPUTIL_A_SET_NW_TOS |
1532 OFPUTIL_A_SET_TP_SRC |
1533 OFPUTIL_A_SET_TP_DST |
1538 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1540 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1541 struct dpif_dp_stats s;
1542 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1546 strcpy(ots->name, "classifier");
1548 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1549 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes, &used);
1550 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes,
1552 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes,
1554 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1555 ots->lookup_count = htonll(n_lookup);
1556 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1559 static struct ofport *
1562 struct ofport_dpif *port = xmalloc(sizeof *port);
1567 port_dealloc(struct ofport *port_)
1569 struct ofport_dpif *port = ofport_dpif_cast(port_);
1574 port_construct(struct ofport *port_)
1576 struct ofport_dpif *port = ofport_dpif_cast(port_);
1577 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1578 const struct netdev *netdev = port->up.netdev;
1579 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1580 struct dpif_port dpif_port;
1583 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1584 port->bundle = NULL;
1587 port->may_enable = true;
1588 port->stp_port = NULL;
1589 port->stp_state = STP_DISABLED;
1590 port->is_tunnel = false;
1594 port->realdev_ofp_port = 0;
1595 port->vlandev_vid = 0;
1596 port->carrier_seq = netdev_get_carrier_resets(netdev);
1597 port->is_layer3 = netdev_vport_is_layer3(netdev);
1599 if (netdev_vport_is_patch(netdev)) {
1600 /* By bailing out here, we don't submit the port to the sFlow module
1601 * to be considered for counter polling export. This is correct
1602 * because the patch port represents an interface that sFlow considers
1603 * to be "internal" to the switch as a whole, and therefore not an
1604 * candidate for counter polling. */
1605 port->odp_port = ODPP_NONE;
1606 ofport_update_peer(port);
1610 error = dpif_port_query_by_name(ofproto->backer->dpif,
1611 netdev_vport_get_dpif_port(netdev, namebuf,
1618 port->odp_port = dpif_port.port_no;
1620 if (netdev_get_tunnel_config(netdev)) {
1621 tnl_port_add(port, port->up.netdev, port->odp_port);
1622 port->is_tunnel = true;
1624 /* Sanity-check that a mapping doesn't already exist. This
1625 * shouldn't happen for non-tunnel ports. */
1626 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1627 VLOG_ERR("port %s already has an OpenFlow port number",
1629 dpif_port_destroy(&dpif_port);
1633 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1634 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1635 hash_odp_port(port->odp_port));
1636 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1638 dpif_port_destroy(&dpif_port);
1640 if (ofproto->sflow) {
1641 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1648 port_destruct(struct ofport *port_)
1650 struct ofport_dpif *port = ofport_dpif_cast(port_);
1651 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1652 const char *devname = netdev_get_name(port->up.netdev);
1653 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1654 const char *dp_port_name;
1656 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1657 ovs_rwlock_wrlock(&xlate_rwlock);
1658 xlate_ofport_remove(port);
1659 ovs_rwlock_unlock(&xlate_rwlock);
1661 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1663 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1664 /* The underlying device is still there, so delete it. This
1665 * happens when the ofproto is being destroyed, since the caller
1666 * assumes that removal of attached ports will happen as part of
1668 if (!port->is_tunnel) {
1669 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1674 port->peer->peer = NULL;
1678 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1679 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1680 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1681 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1685 sset_find_and_delete(&ofproto->ports, devname);
1686 sset_find_and_delete(&ofproto->ghost_ports, devname);
1687 bundle_remove(port_);
1688 set_cfm(port_, NULL);
1689 set_bfd(port_, NULL);
1690 if (port->stp_port) {
1691 stp_port_disable(port->stp_port);
1693 if (ofproto->sflow) {
1694 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1701 port_modified(struct ofport *port_)
1703 struct ofport_dpif *port = ofport_dpif_cast(port_);
1705 if (port->bundle && port->bundle->bond) {
1706 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1710 cfm_set_netdev(port->cfm, port->up.netdev);
1714 bfd_set_netdev(port->bfd, port->up.netdev);
1717 ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1718 port->up.pp.hw_addr);
1720 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1722 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1726 ofport_update_peer(port);
1730 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1732 struct ofport_dpif *port = ofport_dpif_cast(port_);
1733 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1734 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1736 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1737 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1738 OFPUTIL_PC_NO_PACKET_IN)) {
1739 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1741 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1742 bundle_update(port->bundle);
1748 set_sflow(struct ofproto *ofproto_,
1749 const struct ofproto_sflow_options *sflow_options)
1751 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1752 struct dpif_sflow *ds = ofproto->sflow;
1754 if (sflow_options) {
1756 struct ofport_dpif *ofport;
1758 ds = ofproto->sflow = dpif_sflow_create();
1759 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1760 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1762 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1764 dpif_sflow_set_options(ds, sflow_options);
1767 dpif_sflow_unref(ds);
1768 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1769 ofproto->sflow = NULL;
1777 struct ofproto *ofproto_,
1778 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1779 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1780 size_t n_flow_exporters_options)
1782 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1783 struct dpif_ipfix *di = ofproto->ipfix;
1784 bool has_options = bridge_exporter_options || flow_exporters_options;
1786 if (has_options && !di) {
1787 di = ofproto->ipfix = dpif_ipfix_create();
1791 /* Call set_options in any case to cleanly flush the flow
1792 * caches in the last exporters that are to be destroyed. */
1793 dpif_ipfix_set_options(
1794 di, bridge_exporter_options, flow_exporters_options,
1795 n_flow_exporters_options);
1798 dpif_ipfix_unref(di);
1799 ofproto->ipfix = NULL;
1807 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1809 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1814 struct ofproto_dpif *ofproto;
1816 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1817 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1818 ofport->cfm = cfm_create(ofport->up.netdev);
1821 if (cfm_configure(ofport->cfm, s)) {
1828 cfm_unref(ofport->cfm);
1831 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1832 ofport->up.pp.hw_addr);
1837 get_cfm_status(const struct ofport *ofport_,
1838 struct ofproto_cfm_status *status)
1840 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1844 if (cfm_check_status_change(ofport->cfm)) {
1845 status->faults = cfm_get_fault(ofport->cfm);
1846 status->flap_count = cfm_get_flap_count(ofport->cfm);
1847 status->remote_opstate = cfm_get_opup(ofport->cfm);
1848 status->health = cfm_get_health(ofport->cfm);
1849 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1851 ret = NO_STATUS_CHANGE;
1861 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1863 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1864 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1868 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1869 cfg, ofport->up.netdev);
1870 if (ofport->bfd != old) {
1871 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1873 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1874 ofport->up.pp.hw_addr);
1879 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1881 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1885 if (bfd_check_status_change(ofport->bfd)) {
1886 bfd_get_status(ofport->bfd, smap);
1888 ret = NO_STATUS_CHANGE;
1897 /* Spanning Tree. */
1900 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1902 struct ofproto_dpif *ofproto = ofproto_;
1903 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1904 struct ofport_dpif *ofport;
1906 ofport = stp_port_get_aux(sp);
1908 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1909 ofproto->up.name, port_num);
1911 struct eth_header *eth = ofpbuf_l2(pkt);
1913 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1914 if (eth_addr_is_zero(eth->eth_src)) {
1915 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1916 "with unknown MAC", ofproto->up.name, port_num);
1918 ofproto_dpif_send_packet(ofport, pkt);
1924 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1926 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1928 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1930 /* Only revalidate flows if the configuration changed. */
1931 if (!s != !ofproto->stp) {
1932 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1936 if (!ofproto->stp) {
1937 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1938 send_bpdu_cb, ofproto);
1939 ofproto->stp_last_tick = time_msec();
1942 stp_set_bridge_id(ofproto->stp, s->system_id);
1943 stp_set_bridge_priority(ofproto->stp, s->priority);
1944 stp_set_hello_time(ofproto->stp, s->hello_time);
1945 stp_set_max_age(ofproto->stp, s->max_age);
1946 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1948 struct ofport *ofport;
1950 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1951 set_stp_port(ofport, NULL);
1954 stp_unref(ofproto->stp);
1955 ofproto->stp = NULL;
1962 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1964 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1968 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1969 s->designated_root = stp_get_designated_root(ofproto->stp);
1970 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1979 update_stp_port_state(struct ofport_dpif *ofport)
1981 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1982 enum stp_state state;
1984 /* Figure out new state. */
1985 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1989 if (ofport->stp_state != state) {
1990 enum ofputil_port_state of_state;
1993 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1994 netdev_get_name(ofport->up.netdev),
1995 stp_state_name(ofport->stp_state),
1996 stp_state_name(state));
1997 if (stp_learn_in_state(ofport->stp_state)
1998 != stp_learn_in_state(state)) {
1999 /* xxx Learning action flows should also be flushed. */
2000 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2001 mac_learning_flush(ofproto->ml);
2002 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2004 fwd_change = stp_forward_in_state(ofport->stp_state)
2005 != stp_forward_in_state(state);
2007 ofproto->backer->need_revalidate = REV_STP;
2008 ofport->stp_state = state;
2009 ofport->stp_state_entered = time_msec();
2011 if (fwd_change && ofport->bundle) {
2012 bundle_update(ofport->bundle);
2015 /* Update the STP state bits in the OpenFlow port description. */
2016 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2017 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2018 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2019 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2020 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2022 ofproto_port_set_state(&ofport->up, of_state);
2026 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2027 * caller is responsible for assigning STP port numbers and ensuring
2028 * there are no duplicates. */
2030 set_stp_port(struct ofport *ofport_,
2031 const struct ofproto_port_stp_settings *s)
2033 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2034 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2035 struct stp_port *sp = ofport->stp_port;
2037 if (!s || !s->enable) {
2039 ofport->stp_port = NULL;
2040 stp_port_disable(sp);
2041 update_stp_port_state(ofport);
2044 } else if (sp && stp_port_no(sp) != s->port_num
2045 && ofport == stp_port_get_aux(sp)) {
2046 /* The port-id changed, so disable the old one if it's not
2047 * already in use by another port. */
2048 stp_port_disable(sp);
2051 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2052 stp_port_enable(sp);
2054 stp_port_set_aux(sp, ofport);
2055 stp_port_set_priority(sp, s->priority);
2056 stp_port_set_path_cost(sp, s->path_cost);
2058 update_stp_port_state(ofport);
2064 get_stp_port_status(struct ofport *ofport_,
2065 struct ofproto_port_stp_status *s)
2067 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2068 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2069 struct stp_port *sp = ofport->stp_port;
2071 if (!ofproto->stp || !sp) {
2077 s->port_id = stp_port_get_id(sp);
2078 s->state = stp_port_get_state(sp);
2079 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2080 s->role = stp_port_get_role(sp);
2086 get_stp_port_stats(struct ofport *ofport_,
2087 struct ofproto_port_stp_stats *s)
2089 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2090 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2091 struct stp_port *sp = ofport->stp_port;
2093 if (!ofproto->stp || !sp) {
2099 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2105 stp_run(struct ofproto_dpif *ofproto)
2108 long long int now = time_msec();
2109 long long int elapsed = now - ofproto->stp_last_tick;
2110 struct stp_port *sp;
2113 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2114 ofproto->stp_last_tick = now;
2116 while (stp_get_changed_port(ofproto->stp, &sp)) {
2117 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2120 update_stp_port_state(ofport);
2124 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2125 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2126 mac_learning_flush(ofproto->ml);
2127 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2133 stp_wait(struct ofproto_dpif *ofproto)
2136 poll_timer_wait(1000);
2141 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2144 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2145 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2147 if (ofport->n_qdscp != n_qdscp
2148 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2149 n_qdscp * sizeof *qdscp))) {
2150 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2151 free(ofport->qdscp);
2152 ofport->qdscp = n_qdscp
2153 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2155 ofport->n_qdscp = n_qdscp;
2163 /* Expires all MAC learning entries associated with 'bundle' and forces its
2164 * ofproto to revalidate every flow.
2166 * Normally MAC learning entries are removed only from the ofproto associated
2167 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2168 * are removed from every ofproto. When patch ports and SLB bonds are in use
2169 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2170 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2171 * with the host from which it migrated. */
2173 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2175 struct ofproto_dpif *ofproto = bundle->ofproto;
2176 struct mac_learning *ml = ofproto->ml;
2177 struct mac_entry *mac, *next_mac;
2179 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2180 ovs_rwlock_wrlock(&ml->rwlock);
2181 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2182 if (mac->port.p == bundle) {
2184 struct ofproto_dpif *o;
2186 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2188 struct mac_entry *e;
2190 ovs_rwlock_wrlock(&o->ml->rwlock);
2191 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2193 mac_learning_expire(o->ml, e);
2195 ovs_rwlock_unlock(&o->ml->rwlock);
2200 mac_learning_expire(ml, mac);
2203 ovs_rwlock_unlock(&ml->rwlock);
2206 static struct ofbundle *
2207 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2209 struct ofbundle *bundle;
2211 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2212 &ofproto->bundles) {
2213 if (bundle->aux == aux) {
2221 bundle_update(struct ofbundle *bundle)
2223 struct ofport_dpif *port;
2225 bundle->floodable = true;
2226 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2227 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2229 || !stp_forward_in_state(port->stp_state)) {
2230 bundle->floodable = false;
2237 bundle_del_port(struct ofport_dpif *port)
2239 struct ofbundle *bundle = port->bundle;
2241 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2243 list_remove(&port->bundle_node);
2244 port->bundle = NULL;
2247 lacp_slave_unregister(bundle->lacp, port);
2250 bond_slave_unregister(bundle->bond, port);
2253 bundle_update(bundle);
2257 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2258 struct lacp_slave_settings *lacp)
2260 struct ofport_dpif *port;
2262 port = get_ofp_port(bundle->ofproto, ofp_port);
2267 if (port->bundle != bundle) {
2268 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2270 bundle_remove(&port->up);
2273 port->bundle = bundle;
2274 list_push_back(&bundle->ports, &port->bundle_node);
2275 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2277 || !stp_forward_in_state(port->stp_state)) {
2278 bundle->floodable = false;
2282 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2283 lacp_slave_register(bundle->lacp, port, lacp);
2290 bundle_destroy(struct ofbundle *bundle)
2292 struct ofproto_dpif *ofproto;
2293 struct ofport_dpif *port, *next_port;
2299 ofproto = bundle->ofproto;
2300 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2302 ovs_rwlock_wrlock(&xlate_rwlock);
2303 xlate_bundle_remove(bundle);
2304 ovs_rwlock_unlock(&xlate_rwlock);
2306 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2307 bundle_del_port(port);
2310 bundle_flush_macs(bundle, true);
2311 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2313 free(bundle->trunks);
2314 lacp_unref(bundle->lacp);
2315 bond_unref(bundle->bond);
2320 bundle_set(struct ofproto *ofproto_, void *aux,
2321 const struct ofproto_bundle_settings *s)
2323 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2324 bool need_flush = false;
2325 struct ofport_dpif *port;
2326 struct ofbundle *bundle;
2327 unsigned long *trunks;
2333 bundle_destroy(bundle_lookup(ofproto, aux));
2337 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2338 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2340 bundle = bundle_lookup(ofproto, aux);
2342 bundle = xmalloc(sizeof *bundle);
2344 bundle->ofproto = ofproto;
2345 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2346 hash_pointer(aux, 0));
2348 bundle->name = NULL;
2350 list_init(&bundle->ports);
2351 bundle->vlan_mode = PORT_VLAN_TRUNK;
2353 bundle->trunks = NULL;
2354 bundle->use_priority_tags = s->use_priority_tags;
2355 bundle->lacp = NULL;
2356 bundle->bond = NULL;
2358 bundle->floodable = true;
2359 mbridge_register_bundle(ofproto->mbridge, bundle);
2362 if (!bundle->name || strcmp(s->name, bundle->name)) {
2364 bundle->name = xstrdup(s->name);
2369 ofproto->lacp_enabled = true;
2370 if (!bundle->lacp) {
2371 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2372 bundle->lacp = lacp_create();
2374 lacp_configure(bundle->lacp, s->lacp);
2376 lacp_unref(bundle->lacp);
2377 bundle->lacp = NULL;
2380 /* Update set of ports. */
2382 for (i = 0; i < s->n_slaves; i++) {
2383 if (!bundle_add_port(bundle, s->slaves[i],
2384 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2388 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2389 struct ofport_dpif *next_port;
2391 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2392 for (i = 0; i < s->n_slaves; i++) {
2393 if (s->slaves[i] == port->up.ofp_port) {
2398 bundle_del_port(port);
2402 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2404 if (list_is_empty(&bundle->ports)) {
2405 bundle_destroy(bundle);
2409 /* Set VLAN tagging mode */
2410 if (s->vlan_mode != bundle->vlan_mode
2411 || s->use_priority_tags != bundle->use_priority_tags) {
2412 bundle->vlan_mode = s->vlan_mode;
2413 bundle->use_priority_tags = s->use_priority_tags;
2418 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2419 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2421 if (vlan != bundle->vlan) {
2422 bundle->vlan = vlan;
2426 /* Get trunked VLANs. */
2427 switch (s->vlan_mode) {
2428 case PORT_VLAN_ACCESS:
2432 case PORT_VLAN_TRUNK:
2433 trunks = CONST_CAST(unsigned long *, s->trunks);
2436 case PORT_VLAN_NATIVE_UNTAGGED:
2437 case PORT_VLAN_NATIVE_TAGGED:
2438 if (vlan != 0 && (!s->trunks
2439 || !bitmap_is_set(s->trunks, vlan)
2440 || bitmap_is_set(s->trunks, 0))) {
2441 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2443 trunks = bitmap_clone(s->trunks, 4096);
2445 trunks = bitmap_allocate1(4096);
2447 bitmap_set1(trunks, vlan);
2448 bitmap_set0(trunks, 0);
2450 trunks = CONST_CAST(unsigned long *, s->trunks);
2457 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2458 free(bundle->trunks);
2459 if (trunks == s->trunks) {
2460 bundle->trunks = vlan_bitmap_clone(trunks);
2462 bundle->trunks = trunks;
2467 if (trunks != s->trunks) {
2472 if (!list_is_short(&bundle->ports)) {
2473 bundle->ofproto->has_bonded_bundles = true;
2475 if (bond_reconfigure(bundle->bond, s->bond)) {
2476 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2479 bundle->bond = bond_create(s->bond, ofproto);
2480 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2483 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2484 bond_slave_register(bundle->bond, port,
2485 port->up.ofp_port, port->up.netdev);
2488 bond_unref(bundle->bond);
2489 bundle->bond = NULL;
2492 /* If we changed something that would affect MAC learning, un-learn
2493 * everything on this port and force flow revalidation. */
2495 bundle_flush_macs(bundle, false);
2502 bundle_remove(struct ofport *port_)
2504 struct ofport_dpif *port = ofport_dpif_cast(port_);
2505 struct ofbundle *bundle = port->bundle;
2508 bundle_del_port(port);
2509 if (list_is_empty(&bundle->ports)) {
2510 bundle_destroy(bundle);
2511 } else if (list_is_short(&bundle->ports)) {
2512 bond_unref(bundle->bond);
2513 bundle->bond = NULL;
2519 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2521 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2522 struct ofport_dpif *port = port_;
2523 uint8_t ea[ETH_ADDR_LEN];
2526 error = netdev_get_etheraddr(port->up.netdev, ea);
2528 struct ofpbuf packet;
2531 ofpbuf_init(&packet, 0);
2532 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2534 memcpy(packet_pdu, pdu, pdu_size);
2536 ofproto_dpif_send_packet(port, &packet);
2537 ofpbuf_uninit(&packet);
2539 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2540 "%s (%s)", port->bundle->name,
2541 netdev_get_name(port->up.netdev), ovs_strerror(error));
2546 bundle_send_learning_packets(struct ofbundle *bundle)
2548 struct ofproto_dpif *ofproto = bundle->ofproto;
2549 struct ofpbuf *learning_packet;
2550 int error, n_packets, n_errors;
2551 struct mac_entry *e;
2552 struct list packets;
2554 list_init(&packets);
2555 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2556 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2557 if (e->port.p != bundle) {
2560 learning_packet = bond_compose_learning_packet(bundle->bond,
2563 /* Temporarily use 'frame' as a private pointer (see below). */
2564 ovs_assert(learning_packet->frame == ofpbuf_data(learning_packet));
2565 learning_packet->frame = port_void;
2566 list_push_back(&packets, &learning_packet->list_node);
2569 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2571 error = n_packets = n_errors = 0;
2572 LIST_FOR_EACH (learning_packet, list_node, &packets) {
2574 void *port_void = learning_packet->frame;
2576 /* Restore 'frame'. */
2577 learning_packet->frame = ofpbuf_data(learning_packet);
2578 ret = ofproto_dpif_send_packet(port_void, learning_packet);
2585 ofpbuf_list_delete(&packets);
2588 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2589 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2590 "packets, last error was: %s",
2591 bundle->name, n_errors, n_packets, ovs_strerror(error));
2593 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2594 bundle->name, n_packets);
2599 bundle_run(struct ofbundle *bundle)
2602 lacp_run(bundle->lacp, send_pdu_cb);
2605 struct ofport_dpif *port;
2607 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2608 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2611 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2612 bundle->ofproto->backer->need_revalidate = REV_BOND;
2615 if (bond_should_send_learning_packets(bundle->bond)) {
2616 bundle_send_learning_packets(bundle);
2622 bundle_wait(struct ofbundle *bundle)
2625 lacp_wait(bundle->lacp);
2628 bond_wait(bundle->bond);
2635 mirror_set__(struct ofproto *ofproto_, void *aux,
2636 const struct ofproto_mirror_settings *s)
2638 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2639 struct ofbundle **srcs, **dsts;
2644 mirror_destroy(ofproto->mbridge, aux);
2648 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2649 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2651 for (i = 0; i < s->n_srcs; i++) {
2652 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2655 for (i = 0; i < s->n_dsts; i++) {
2656 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2659 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2660 s->n_dsts, s->src_vlans,
2661 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2668 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2669 uint64_t *packets, uint64_t *bytes)
2671 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2676 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2678 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2679 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2680 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2681 mac_learning_flush(ofproto->ml);
2683 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2688 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2690 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2691 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2692 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2696 forward_bpdu_changed(struct ofproto *ofproto_)
2698 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2699 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2703 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2706 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2707 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2708 mac_learning_set_idle_time(ofproto->ml, idle_time);
2709 mac_learning_set_max_entries(ofproto->ml, max_entries);
2710 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2715 static struct ofport_dpif *
2716 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2718 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2719 return ofport ? ofport_dpif_cast(ofport) : NULL;
2723 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2724 struct ofproto_port *ofproto_port,
2725 struct dpif_port *dpif_port)
2727 ofproto_port->name = dpif_port->name;
2728 ofproto_port->type = dpif_port->type;
2729 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2733 ofport_update_peer(struct ofport_dpif *ofport)
2735 const struct ofproto_dpif *ofproto;
2736 struct dpif_backer *backer;
2739 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2743 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2744 backer->need_revalidate = REV_RECONFIGURE;
2747 ofport->peer->peer = NULL;
2748 ofport->peer = NULL;
2751 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2756 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2757 struct ofport *peer_ofport;
2758 struct ofport_dpif *peer;
2761 if (ofproto->backer != backer) {
2765 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2770 peer = ofport_dpif_cast(peer_ofport);
2771 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2772 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2774 ofport->peer = peer;
2775 ofport->peer->peer = ofport;
2785 port_run(struct ofport_dpif *ofport)
2787 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2788 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2789 bool enable = netdev_get_carrier(ofport->up.netdev);
2790 bool cfm_enable = false;
2791 bool bfd_enable = false;
2793 ofport->carrier_seq = carrier_seq;
2796 int cfm_opup = cfm_get_opup(ofport->cfm);
2798 cfm_enable = !cfm_get_fault(ofport->cfm);
2800 if (cfm_opup >= 0) {
2801 cfm_enable = cfm_enable && cfm_opup;
2806 bfd_enable = bfd_forwarding(ofport->bfd);
2809 if (ofport->bfd || ofport->cfm) {
2810 enable = enable && (cfm_enable || bfd_enable);
2813 if (ofport->bundle) {
2814 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2815 if (carrier_changed) {
2816 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2820 if (ofport->may_enable != enable) {
2821 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2822 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2825 ofport->may_enable = enable;
2829 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2830 struct ofproto_port *ofproto_port)
2832 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2833 struct dpif_port dpif_port;
2836 if (sset_contains(&ofproto->ghost_ports, devname)) {
2837 const char *type = netdev_get_type_from_name(devname);
2839 /* We may be called before ofproto->up.port_by_name is populated with
2840 * the appropriate ofport. For this reason, we must get the name and
2841 * type from the netdev layer directly. */
2843 const struct ofport *ofport;
2845 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2846 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2847 ofproto_port->name = xstrdup(devname);
2848 ofproto_port->type = xstrdup(type);
2854 if (!sset_contains(&ofproto->ports, devname)) {
2857 error = dpif_port_query_by_name(ofproto->backer->dpif,
2858 devname, &dpif_port);
2860 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2866 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2868 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2869 const char *devname = netdev_get_name(netdev);
2870 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2871 const char *dp_port_name;
2873 if (netdev_vport_is_patch(netdev)) {
2874 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2878 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2879 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2880 odp_port_t port_no = ODPP_NONE;
2883 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2887 if (netdev_get_tunnel_config(netdev)) {
2888 simap_put(&ofproto->backer->tnl_backers,
2889 dp_port_name, odp_to_u32(port_no));
2893 if (netdev_get_tunnel_config(netdev)) {
2894 sset_add(&ofproto->ghost_ports, devname);
2896 sset_add(&ofproto->ports, devname);
2902 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
2904 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2905 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2912 sset_find_and_delete(&ofproto->ghost_ports,
2913 netdev_get_name(ofport->up.netdev));
2914 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2915 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
2916 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
2918 /* The caller is going to close ofport->up.netdev. If this is a
2919 * bonded port, then the bond is using that netdev, so remove it
2920 * from the bond. The client will need to reconfigure everything
2921 * after deleting ports, so then the slave will get re-added. */
2922 bundle_remove(&ofport->up);
2929 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2931 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2934 error = netdev_get_stats(ofport->up.netdev, stats);
2936 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
2937 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2939 ovs_mutex_lock(&ofproto->stats_mutex);
2940 /* ofproto->stats.tx_packets represents packets that we created
2941 * internally and sent to some port (e.g. packets sent with
2942 * ofproto_dpif_send_packet()). Account for them as if they had
2943 * come from OFPP_LOCAL and got forwarded. */
2945 if (stats->rx_packets != UINT64_MAX) {
2946 stats->rx_packets += ofproto->stats.tx_packets;
2949 if (stats->rx_bytes != UINT64_MAX) {
2950 stats->rx_bytes += ofproto->stats.tx_bytes;
2953 /* ofproto->stats.rx_packets represents packets that were received on
2954 * some port and we processed internally and dropped (e.g. STP).
2955 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2957 if (stats->tx_packets != UINT64_MAX) {
2958 stats->tx_packets += ofproto->stats.rx_packets;
2961 if (stats->tx_bytes != UINT64_MAX) {
2962 stats->tx_bytes += ofproto->stats.rx_bytes;
2964 ovs_mutex_unlock(&ofproto->stats_mutex);
2970 struct port_dump_state {
2975 struct ofproto_port port;
2980 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
2982 *statep = xzalloc(sizeof(struct port_dump_state));
2987 port_dump_next(const struct ofproto *ofproto_, void *state_,
2988 struct ofproto_port *port)
2990 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2991 struct port_dump_state *state = state_;
2992 const struct sset *sset;
2993 struct sset_node *node;
2995 if (state->has_port) {
2996 ofproto_port_destroy(&state->port);
2997 state->has_port = false;
2999 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3000 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3003 error = port_query_by_name(ofproto_, node->name, &state->port);
3005 *port = state->port;
3006 state->has_port = true;
3008 } else if (error != ENODEV) {
3013 if (!state->ghost) {
3014 state->ghost = true;
3017 return port_dump_next(ofproto_, state_, port);
3024 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3026 struct port_dump_state *state = state_;
3028 if (state->has_port) {
3029 ofproto_port_destroy(&state->port);
3036 port_poll(const struct ofproto *ofproto_, char **devnamep)
3038 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3040 if (ofproto->port_poll_errno) {
3041 int error = ofproto->port_poll_errno;
3042 ofproto->port_poll_errno = 0;
3046 if (sset_is_empty(&ofproto->port_poll_set)) {
3050 *devnamep = sset_pop(&ofproto->port_poll_set);
3055 port_poll_wait(const struct ofproto *ofproto_)
3057 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3058 dpif_port_poll_wait(ofproto->backer->dpif);
3062 port_is_lacp_current(const struct ofport *ofport_)
3064 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3065 return (ofport->bundle && ofport->bundle->lacp
3066 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3070 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3071 * then delete it entirely. */
3073 rule_expire(struct rule_dpif *rule)
3074 OVS_REQUIRES(ofproto_mutex)
3076 uint16_t hard_timeout, idle_timeout;
3077 long long int now = time_msec();
3080 ovs_assert(!rule->up.pending);
3082 hard_timeout = rule->up.hard_timeout;
3083 idle_timeout = rule->up.idle_timeout;
3085 /* Has 'rule' expired? */
3087 long long int modified;
3089 ovs_mutex_lock(&rule->up.mutex);
3090 modified = rule->up.modified;
3091 ovs_mutex_unlock(&rule->up.mutex);
3093 if (now > modified + hard_timeout * 1000) {
3094 reason = OFPRR_HARD_TIMEOUT;
3098 if (reason < 0 && idle_timeout) {
3101 ovs_mutex_lock(&rule->stats_mutex);
3102 used = rule->stats.used;
3103 ovs_mutex_unlock(&rule->stats_mutex);
3105 if (now > used + idle_timeout * 1000) {
3106 reason = OFPRR_IDLE_TIMEOUT;
3111 COVERAGE_INC(ofproto_dpif_expired);
3112 ofproto_rule_expire(&rule->up, reason);
3116 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3117 * 'flow' must reflect the data in 'packet'. */
3119 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3120 const struct flow *flow,
3121 struct rule_dpif *rule,
3122 const struct ofpact *ofpacts, size_t ofpacts_len,
3123 struct ofpbuf *packet)
3125 struct dpif_flow_stats stats;
3126 struct xlate_out xout;
3127 struct xlate_in xin;
3129 struct dpif_execute execute;
3132 ovs_assert((rule != NULL) != (ofpacts != NULL));
3134 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3137 rule_dpif_credit_stats(rule, &stats);
3140 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
3141 xin.ofpacts = ofpacts;
3142 xin.ofpacts_len = ofpacts_len;
3143 xin.resubmit_stats = &stats;
3144 xlate_actions(&xin, &xout);
3146 in_port = flow->in_port.ofp_port;
3147 if (in_port == OFPP_NONE) {
3148 in_port = OFPP_LOCAL;
3150 execute.actions = ofpbuf_data(&xout.odp_actions);
3151 execute.actions_len = ofpbuf_size(&xout.odp_actions);
3152 execute.packet = packet;
3153 execute.md.tunnel = flow->tunnel;
3154 execute.md.skb_priority = flow->skb_priority;
3155 execute.md.pkt_mark = flow->pkt_mark;
3156 execute.md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
3157 execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
3159 error = dpif_execute(ofproto->backer->dpif, &execute);
3161 xlate_out_uninit(&xout);
3167 rule_dpif_credit_stats(struct rule_dpif *rule,
3168 const struct dpif_flow_stats *stats)
3170 ovs_mutex_lock(&rule->stats_mutex);
3171 rule->stats.n_packets += stats->n_packets;
3172 rule->stats.n_bytes += stats->n_bytes;
3173 rule->stats.used = MAX(rule->stats.used, stats->used);
3174 ovs_mutex_unlock(&rule->stats_mutex);
3178 rule_dpif_is_fail_open(const struct rule_dpif *rule)
3180 return is_fail_open_rule(&rule->up);
3184 rule_dpif_is_table_miss(const struct rule_dpif *rule)
3186 return rule_is_table_miss(&rule->up);
3190 rule_dpif_is_internal(const struct rule_dpif *rule)
3192 return rule_is_internal(&rule->up);
3196 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
3197 OVS_REQUIRES(rule->up.mutex)
3199 return rule->up.flow_cookie;
3203 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
3204 uint16_t hard_timeout)
3206 ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
3209 /* Returns 'rule''s actions. The caller owns a reference on the returned
3210 * actions and must eventually release it (with rule_actions_unref()) to avoid
3212 struct rule_actions *
3213 rule_dpif_get_actions(const struct rule_dpif *rule)
3215 return rule_get_actions(&rule->up);
3218 /* Lookup 'flow' in table 0 of 'ofproto''s classifier.
3219 * If 'wc' is non-null, sets the fields that were relevant as part of
3220 * the lookup. Returns the table_id where a match or miss occurred.
3222 * The return value will be zero unless there was a miss and
3223 * OFPTC11_TABLE_MISS_CONTINUE is in effect for the sequence of tables
3224 * where misses occur.
3226 * The rule is returned in '*rule', which is valid at least until the next
3227 * RCU quiescent period. If the '*rule' needs to stay around longer,
3228 * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3229 * on it before this returns. */
3231 rule_dpif_lookup(struct ofproto_dpif *ofproto, struct flow *flow,
3232 struct flow_wildcards *wc, struct rule_dpif **rule,
3235 enum rule_dpif_lookup_verdict verdict;
3236 enum ofputil_port_config config = 0;
3239 if (ofproto_dpif_get_enable_recirc(ofproto)) {
3240 /* Always exactly match recirc_id since datapath supports
3243 wc->masks.recirc_id = UINT32_MAX;
3246 /* Start looking up from internal table for post recirculation flows
3247 * or packets. We can also simply send all, including normal flows
3248 * or packets to the internal table. They will not match any post
3249 * recirculation rules except the 'catch all' rule that resubmit
3252 * As an optimization, we send normal flows and packets to table 0
3253 * directly, saving one table lookup. */
3254 table_id = flow->recirc_id ? TBL_INTERNAL : 0;
3259 verdict = rule_dpif_lookup_from_table(ofproto, flow, wc, true,
3260 &table_id, rule, take_ref);
3263 case RULE_DPIF_LOOKUP_VERDICT_MATCH:
3265 case RULE_DPIF_LOOKUP_VERDICT_CONTROLLER: {
3266 struct ofport_dpif *port;
3268 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
3270 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
3271 flow->in_port.ofp_port);
3273 config = port ? port->up.pp.config : 0;
3276 case RULE_DPIF_LOOKUP_VERDICT_DROP:
3277 config = OFPUTIL_PC_NO_PACKET_IN;
3279 case RULE_DPIF_LOOKUP_VERDICT_DEFAULT:
3280 if (!connmgr_wants_packet_in_on_miss(ofproto->up.connmgr)) {
3281 config = OFPUTIL_PC_NO_PACKET_IN;
3288 choose_miss_rule(config, ofproto->miss_rule,
3289 ofproto->no_packet_in_rule, rule, take_ref);
3293 /* The returned rule is valid at least until the next RCU quiescent period.
3294 * If the '*rule' needs to stay around longer, a non-zero 'take_ref' must be
3295 * passed in to cause a reference to be taken on it before this returns. */
3296 static struct rule_dpif *
3297 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, uint8_t table_id,
3298 const struct flow *flow, struct flow_wildcards *wc,
3301 struct classifier *cls = &ofproto->up.tables[table_id].cls;
3302 const struct cls_rule *cls_rule;
3303 struct rule_dpif *rule;
3305 fat_rwlock_rdlock(&cls->rwlock);
3306 if (ofproto->up.frag_handling != OFPC_FRAG_NX_MATCH) {
3308 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
3309 if (is_ip_any(flow)) {
3310 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
3314 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
3315 if (ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3316 /* We must pretend that transport ports are unavailable. */
3317 struct flow ofpc_normal_flow = *flow;
3318 ofpc_normal_flow.tp_src = htons(0);
3319 ofpc_normal_flow.tp_dst = htons(0);
3320 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
3322 /* Must be OFPC_FRAG_DROP (we don't have OFPC_FRAG_REASM). */
3323 cls_rule = &ofproto->drop_frags_rule->up.cr;
3326 cls_rule = classifier_lookup(cls, flow, wc);
3329 cls_rule = classifier_lookup(cls, flow, wc);
3332 rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
3334 rule_dpif_ref(rule);
3336 fat_rwlock_unlock(&cls->rwlock);
3341 /* Look up 'flow' in 'ofproto''s classifier starting from table '*table_id'.
3342 * Stores the rule that was found in '*rule', or NULL if none was found.
3343 * Updates 'wc', if nonnull, to reflect the fields that were used during the
3346 * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3347 * if none is found then the table miss configuration for that table is
3348 * honored, which can result in additional lookups in other OpenFlow tables.
3349 * In this case the function updates '*table_id' to reflect the final OpenFlow
3350 * table that was searched.
3352 * If 'honor_table_miss' is false, then only one table lookup occurs, in
3357 * - RULE_DPIF_LOOKUP_VERDICT_MATCH if a rule (in '*rule') was found.
3359 * - RULE_OFPTC_TABLE_MISS_CONTROLLER if no rule was found and either:
3360 * + 'honor_table_miss' is false
3361 * + a table miss configuration specified that the packet should be
3362 * sent to the controller in this case.
3364 * - RULE_DPIF_LOOKUP_VERDICT_DROP if no rule was found, 'honor_table_miss'
3365 * is true and a table miss configuration specified that the packet
3366 * should be dropped in this case.
3368 * - RULE_DPIF_LOOKUP_VERDICT_DEFAULT if no rule was found,
3369 * 'honor_table_miss' is true and a table miss configuration has
3370 * not been specified in this case.
3372 * The rule is returned in '*rule', which is valid at least until the next
3373 * RCU quiescent period. If the '*rule' needs to stay around longer,
3374 * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3375 * on it before this returns. */
3376 enum rule_dpif_lookup_verdict
3377 rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto,
3378 const struct flow *flow,
3379 struct flow_wildcards *wc,
3380 bool honor_table_miss,
3381 uint8_t *table_id, struct rule_dpif **rule,
3386 for (next_id = *table_id;
3387 next_id < ofproto->up.n_tables;
3388 next_id++, next_id += (next_id == TBL_INTERNAL))
3390 *table_id = next_id;
3391 *rule = rule_dpif_lookup_in_table(ofproto, *table_id, flow, wc,
3394 return RULE_DPIF_LOOKUP_VERDICT_MATCH;
3395 } else if (!honor_table_miss) {
3396 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3398 switch (ofproto_table_get_config(&ofproto->up, *table_id)) {
3399 case OFPROTO_TABLE_MISS_CONTINUE:
3402 case OFPROTO_TABLE_MISS_CONTROLLER:
3403 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3405 case OFPROTO_TABLE_MISS_DROP:
3406 return RULE_DPIF_LOOKUP_VERDICT_DROP;
3408 case OFPROTO_TABLE_MISS_DEFAULT:
3409 return RULE_DPIF_LOOKUP_VERDICT_DEFAULT;
3414 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3417 /* Given a port configuration (specified as zero if there's no port), chooses
3418 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
3421 * The rule is returned in '*rule', which is valid at least until the next
3422 * RCU quiescent period. If the '*rule' needs to stay around longer,
3423 * a reference must be taken on it (rule_dpif_ref()).
3426 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
3427 struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule,
3430 *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
3432 rule_dpif_ref(*rule);
3437 rule_dpif_ref(struct rule_dpif *rule)
3440 ofproto_rule_ref(&rule->up);
3445 rule_dpif_unref(struct rule_dpif *rule)
3448 ofproto_rule_unref(&rule->up);
3453 complete_operation(struct rule_dpif *rule)
3454 OVS_REQUIRES(ofproto_mutex)
3456 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3458 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3459 ofoperation_complete(rule->up.pending, 0);
3462 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
3464 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
3467 static struct rule *
3470 struct rule_dpif *rule = xmalloc(sizeof *rule);
3475 rule_dealloc(struct rule *rule_)
3477 struct rule_dpif *rule = rule_dpif_cast(rule_);
3482 rule_construct(struct rule *rule_)
3483 OVS_NO_THREAD_SAFETY_ANALYSIS
3485 struct rule_dpif *rule = rule_dpif_cast(rule_);
3486 ovs_mutex_init_adaptive(&rule->stats_mutex);
3487 rule->stats.n_packets = 0;
3488 rule->stats.n_bytes = 0;
3489 rule->stats.used = rule->up.modified;
3494 rule_insert(struct rule *rule_)
3495 OVS_REQUIRES(ofproto_mutex)
3497 struct rule_dpif *rule = rule_dpif_cast(rule_);
3498 complete_operation(rule);
3502 rule_delete(struct rule *rule_)
3503 OVS_REQUIRES(ofproto_mutex)
3505 struct rule_dpif *rule = rule_dpif_cast(rule_);
3506 complete_operation(rule);
3510 rule_destruct(struct rule *rule_)
3512 struct rule_dpif *rule = rule_dpif_cast(rule_);
3513 ovs_mutex_destroy(&rule->stats_mutex);
3517 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
3518 long long int *used)
3520 struct rule_dpif *rule = rule_dpif_cast(rule_);
3522 ovs_mutex_lock(&rule->stats_mutex);
3523 *packets = rule->stats.n_packets;
3524 *bytes = rule->stats.n_bytes;
3525 *used = rule->stats.used;
3526 ovs_mutex_unlock(&rule->stats_mutex);
3530 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
3531 struct ofpbuf *packet)
3533 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3535 ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
3539 rule_execute(struct rule *rule, const struct flow *flow,
3540 struct ofpbuf *packet)
3542 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
3543 ofpbuf_delete(packet);
3548 rule_modify_actions(struct rule *rule_, bool reset_counters)
3549 OVS_REQUIRES(ofproto_mutex)
3551 struct rule_dpif *rule = rule_dpif_cast(rule_);
3553 if (reset_counters) {
3554 ovs_mutex_lock(&rule->stats_mutex);
3555 rule->stats.n_packets = 0;
3556 rule->stats.n_bytes = 0;
3557 ovs_mutex_unlock(&rule->stats_mutex);
3560 complete_operation(rule);
3563 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
3565 return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
3568 static struct ofgroup *
3571 struct group_dpif *group = xzalloc(sizeof *group);
3576 group_dealloc(struct ofgroup *group_)
3578 struct group_dpif *group = group_dpif_cast(group_);
3583 group_construct_stats(struct group_dpif *group)
3584 OVS_REQUIRES(group->stats_mutex)
3586 group->packet_count = 0;
3587 group->byte_count = 0;
3588 if (!group->bucket_stats) {
3589 group->bucket_stats = xcalloc(group->up.n_buckets,
3590 sizeof *group->bucket_stats);
3592 memset(group->bucket_stats, 0, group->up.n_buckets *
3593 sizeof *group->bucket_stats);
3598 group_construct(struct ofgroup *group_)
3600 struct group_dpif *group = group_dpif_cast(group_);
3601 const struct ofputil_bucket *bucket;
3603 /* Prevent group chaining because our locking structure makes it hard to
3604 * implement deadlock-free. (See xlate_group_resource_check().) */
3605 LIST_FOR_EACH (bucket, list_node, &group->up.buckets) {
3606 const struct ofpact *a;
3608 OFPACT_FOR_EACH (a, bucket->ofpacts, bucket->ofpacts_len) {
3609 if (a->type == OFPACT_GROUP) {
3610 return OFPERR_OFPGMFC_CHAINING_UNSUPPORTED;
3615 ovs_mutex_init_adaptive(&group->stats_mutex);
3616 ovs_mutex_lock(&group->stats_mutex);
3617 group_construct_stats(group);
3618 ovs_mutex_unlock(&group->stats_mutex);
3623 group_destruct__(struct group_dpif *group)
3624 OVS_REQUIRES(group->stats_mutex)
3626 free(group->bucket_stats);
3627 group->bucket_stats = NULL;
3631 group_destruct(struct ofgroup *group_)
3633 struct group_dpif *group = group_dpif_cast(group_);
3634 ovs_mutex_lock(&group->stats_mutex);
3635 group_destruct__(group);
3636 ovs_mutex_unlock(&group->stats_mutex);
3637 ovs_mutex_destroy(&group->stats_mutex);
3641 group_modify(struct ofgroup *group_, struct ofgroup *victim_)
3643 struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
3644 struct group_dpif *group = group_dpif_cast(group_);
3645 struct group_dpif *victim = group_dpif_cast(victim_);
3647 ovs_mutex_lock(&group->stats_mutex);
3648 if (victim->up.n_buckets < group->up.n_buckets) {
3649 group_destruct__(group);
3651 group_construct_stats(group);
3652 ovs_mutex_unlock(&group->stats_mutex);
3654 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3660 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
3662 struct group_dpif *group = group_dpif_cast(group_);
3664 ovs_mutex_lock(&group->stats_mutex);
3665 ogs->packet_count = group->packet_count;
3666 ogs->byte_count = group->byte_count;
3667 memcpy(ogs->bucket_stats, group->bucket_stats,
3668 group->up.n_buckets * sizeof *group->bucket_stats);
3669 ovs_mutex_unlock(&group->stats_mutex);
3675 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
3676 struct group_dpif **group)
3677 OVS_TRY_RDLOCK(true, (*group)->up.rwlock)
3679 struct ofgroup *ofgroup;
3683 found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
3684 *group = found ? group_dpif_cast(ofgroup) : NULL;
3690 group_dpif_release(struct group_dpif *group)
3691 OVS_RELEASES(group->up.rwlock)
3693 ofproto_group_release(&group->up);
3697 group_dpif_get_buckets(const struct group_dpif *group,
3698 const struct list **buckets)
3700 *buckets = &group->up.buckets;
3703 enum ofp11_group_type
3704 group_dpif_get_type(const struct group_dpif *group)
3706 return group->up.type;
3709 /* Sends 'packet' out 'ofport'.
3710 * May modify 'packet'.
3711 * Returns 0 if successful, otherwise a positive errno value. */
3713 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
3715 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3718 error = xlate_send_packet(ofport, packet);
3720 ovs_mutex_lock(&ofproto->stats_mutex);
3721 ofproto->stats.tx_packets++;
3722 ofproto->stats.tx_bytes += ofpbuf_size(packet);
3723 ovs_mutex_unlock(&ofproto->stats_mutex);
3728 set_frag_handling(struct ofproto *ofproto_,
3729 enum ofp_config_flags frag_handling)
3731 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3732 if (frag_handling != OFPC_FRAG_REASM) {
3733 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3741 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
3742 const struct flow *flow,
3743 const struct ofpact *ofpacts, size_t ofpacts_len)
3745 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3747 ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
3748 ofpacts_len, packet);
3755 set_netflow(struct ofproto *ofproto_,
3756 const struct netflow_options *netflow_options)
3758 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3760 if (netflow_options) {
3761 if (!ofproto->netflow) {
3762 ofproto->netflow = netflow_create();
3763 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3765 return netflow_set_options(ofproto->netflow, netflow_options);
3766 } else if (ofproto->netflow) {
3767 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3768 netflow_unref(ofproto->netflow);
3769 ofproto->netflow = NULL;
3776 get_netflow_ids(const struct ofproto *ofproto_,
3777 uint8_t *engine_type, uint8_t *engine_id)
3779 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3781 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
3784 static struct ofproto_dpif *
3785 ofproto_dpif_lookup(const char *name)
3787 struct ofproto_dpif *ofproto;
3789 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
3790 hash_string(name, 0), &all_ofproto_dpifs) {
3791 if (!strcmp(ofproto->up.name, name)) {
3799 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
3800 const char *argv[], void *aux OVS_UNUSED)
3802 struct ofproto_dpif *ofproto;
3805 ofproto = ofproto_dpif_lookup(argv[1]);
3807 unixctl_command_reply_error(conn, "no such bridge");
3810 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3811 mac_learning_flush(ofproto->ml);
3812 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3814 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3815 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3816 mac_learning_flush(ofproto->ml);
3817 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3821 unixctl_command_reply(conn, "table successfully flushed");
3824 static struct ofport_dpif *
3825 ofbundle_get_a_port(const struct ofbundle *bundle)
3827 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
3832 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
3833 const char *argv[], void *aux OVS_UNUSED)
3835 struct ds ds = DS_EMPTY_INITIALIZER;
3836 const struct ofproto_dpif *ofproto;
3837 const struct mac_entry *e;
3839 ofproto = ofproto_dpif_lookup(argv[1]);
3841 unixctl_command_reply_error(conn, "no such bridge");
3845 ds_put_cstr(&ds, " port VLAN MAC Age\n");
3846 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
3847 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3848 struct ofbundle *bundle = e->port.p;
3849 char name[OFP_MAX_PORT_NAME_LEN];
3851 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
3853 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
3854 name, e->vlan, ETH_ADDR_ARGS(e->mac),
3855 mac_entry_age(ofproto->ml, e));
3857 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3858 unixctl_command_reply(conn, ds_cstr(&ds));
3863 struct xlate_out xout;
3864 struct xlate_in xin;
3865 const struct flow *key;
3867 struct flow_wildcards wc;
3872 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
3874 struct rule_actions *actions;
3877 ds_put_char_multiple(result, '\t', level);
3879 ds_put_cstr(result, "No match\n");
3883 ovs_mutex_lock(&rule->up.mutex);
3884 cookie = rule->up.flow_cookie;
3885 ovs_mutex_unlock(&rule->up.mutex);
3887 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
3888 rule ? rule->up.table_id : 0, ntohll(cookie));
3889 cls_rule_format(&rule->up.cr, result);
3890 ds_put_char(result, '\n');
3892 actions = rule_dpif_get_actions(rule);
3894 ds_put_char_multiple(result, '\t', level);
3895 ds_put_cstr(result, "OpenFlow actions=");
3896 ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
3897 ds_put_char(result, '\n');
3901 trace_format_flow(struct ds *result, int level, const char *title,
3902 struct trace_ctx *trace)
3904 ds_put_char_multiple(result, '\t', level);
3905 ds_put_format(result, "%s: ", title);
3906 /* Do not report unchanged flows for resubmits. */
3907 if ((level > 0 && flow_equal(&trace->xin.flow, &trace->flow))
3908 || (level == 0 && flow_equal(&trace->xin.flow, trace->key))) {
3909 ds_put_cstr(result, "unchanged");
3911 flow_format(result, &trace->xin.flow);
3912 trace->flow = trace->xin.flow;
3914 ds_put_char(result, '\n');
3918 trace_format_regs(struct ds *result, int level, const char *title,
3919 struct trace_ctx *trace)
3923 ds_put_char_multiple(result, '\t', level);
3924 ds_put_format(result, "%s:", title);
3925 for (i = 0; i < FLOW_N_REGS; i++) {
3926 ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
3928 ds_put_char(result, '\n');
3932 trace_format_odp(struct ds *result, int level, const char *title,
3933 struct trace_ctx *trace)
3935 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
3937 ds_put_char_multiple(result, '\t', level);
3938 ds_put_format(result, "%s: ", title);
3939 format_odp_actions(result, ofpbuf_data(odp_actions),
3940 ofpbuf_size(odp_actions));
3941 ds_put_char(result, '\n');
3945 trace_format_megaflow(struct ds *result, int level, const char *title,
3946 struct trace_ctx *trace)
3950 ds_put_char_multiple(result, '\t', level);
3951 ds_put_format(result, "%s: ", title);
3952 flow_wildcards_or(&trace->wc, &trace->xout.wc, &trace->wc);
3953 match_init(&match, trace->key, &trace->wc);
3954 match_format(&match, result, OFP_DEFAULT_PRIORITY);
3955 ds_put_char(result, '\n');
3959 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
3961 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
3962 struct ds *result = trace->result;
3964 ds_put_char(result, '\n');
3965 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
3966 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
3967 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
3968 trace_format_megaflow(result, recurse + 1, "Resubmitted megaflow", trace);
3969 trace_format_rule(result, recurse + 1, rule);
3973 trace_report(struct xlate_in *xin, const char *s, int recurse)
3975 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
3976 struct ds *result = trace->result;
3978 ds_put_char_multiple(result, '\t', recurse);
3979 ds_put_cstr(result, s);
3980 ds_put_char(result, '\n');
3983 /* Parses the 'argc' elements of 'argv', ignoring argv[0]. The following
3984 * forms are supported:
3986 * - [dpname] odp_flow [-generate | packet]
3987 * - bridge br_flow [-generate | packet]
3989 * On success, initializes '*ofprotop' and 'flow' and returns NULL. On failure
3990 * returns a nonnull malloced error message. */
3991 static char * WARN_UNUSED_RESULT
3992 parse_flow_and_packet(int argc, const char *argv[],
3993 struct ofproto_dpif **ofprotop, struct flow *flow,
3994 struct ofpbuf **packetp)
3996 const struct dpif_backer *backer = NULL;
3997 const char *error = NULL;
3999 struct simap port_names = SIMAP_INITIALIZER(&port_names);
4000 struct ofpbuf *packet;
4001 struct ofpbuf odp_key;
4002 struct ofpbuf odp_mask;
4004 ofpbuf_init(&odp_key, 0);
4005 ofpbuf_init(&odp_mask, 0);
4007 /* Handle "-generate" or a hex string as the last argument. */
4008 if (!strcmp(argv[argc - 1], "-generate")) {
4009 packet = ofpbuf_new(0);
4012 error = eth_from_hex(argv[argc - 1], &packet);
4015 } else if (argc == 4) {
4016 /* The 3-argument form must end in "-generate' or a hex string. */
4022 /* odp_flow can have its in_port specified as a name instead of port no.
4023 * We do not yet know whether a given flow is a odp_flow or a br_flow.
4024 * But, to know whether a flow is odp_flow through odp_flow_from_string(),
4025 * we need to create a simap of name to port no. */
4027 const char *dp_type;
4028 if (!strncmp(argv[1], "ovs-", 4)) {
4029 dp_type = argv[1] + 4;
4033 backer = shash_find_data(&all_dpif_backers, dp_type);
4034 } else if (argc == 2) {
4035 struct shash_node *node;
4036 if (shash_count(&all_dpif_backers) == 1) {
4037 node = shash_first(&all_dpif_backers);
4038 backer = node->data;
4041 error = "Syntax error";
4044 if (backer && backer->dpif) {
4045 struct dpif_port dpif_port;
4046 struct dpif_port_dump port_dump;
4047 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
4048 simap_put(&port_names, dpif_port.name,
4049 odp_to_u32(dpif_port.port_no));
4053 /* Parse the flow and determine whether a datapath or
4054 * bridge is specified. If function odp_flow_key_from_string()
4055 * returns 0, the flow is a odp_flow. If function
4056 * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
4057 if (!odp_flow_from_string(argv[argc - 1], &port_names,
4058 &odp_key, &odp_mask)) {
4060 error = "Cannot find the datapath";
4064 if (xlate_receive(backer, NULL, ofpbuf_data(&odp_key),
4065 ofpbuf_size(&odp_key), flow,
4066 ofprotop, NULL, NULL, NULL, NULL)) {
4067 error = "Invalid datapath flow";
4071 char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
4074 m_err = xasprintf("Bad flow syntax: %s", err);
4079 error = "Must specify bridge name";
4083 *ofprotop = ofproto_dpif_lookup(argv[1]);
4085 error = "Unknown bridge name";
4091 /* Generate a packet, if requested. */
4093 if (!ofpbuf_size(packet)) {
4094 flow_compose(packet, flow);
4096 struct pkt_metadata md = pkt_metadata_from_flow(flow);
4098 /* Use the metadata from the flow and the packet argument
4099 * to reconstruct the flow. */
4100 flow_extract(packet, &md, flow);
4105 if (error && !m_err) {
4106 m_err = xstrdup(error);
4109 ofpbuf_delete(packet);
4113 ofpbuf_uninit(&odp_key);
4114 ofpbuf_uninit(&odp_mask);
4115 simap_destroy(&port_names);
4120 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
4121 void *aux OVS_UNUSED)
4123 struct ofproto_dpif *ofproto;
4124 struct ofpbuf *packet;
4128 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4133 ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
4134 unixctl_command_reply(conn, ds_cstr(&result));
4135 ds_destroy(&result);
4136 ofpbuf_delete(packet);
4138 unixctl_command_reply_error(conn, error);
4144 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
4145 const char *argv[], void *aux OVS_UNUSED)
4147 enum ofputil_protocol usable_protocols;
4148 struct ofproto_dpif *ofproto;
4149 bool enforce_consistency;
4150 struct ofpbuf ofpacts;
4151 struct ofpbuf *packet;
4156 /* Three kinds of error return values! */
4162 ofpbuf_init(&ofpacts, 0);
4164 /* Parse actions. */
4165 error = parse_ofpacts(argv[--argc], &ofpacts, &usable_protocols);
4167 unixctl_command_reply_error(conn, error);
4172 /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
4173 * consistency between the flow and the actions. With -consistent, we
4174 * enforce consistency even for a flow supported in OpenFlow 1.0. */
4175 if (!strcmp(argv[1], "-consistent")) {
4176 enforce_consistency = true;
4180 enforce_consistency = false;
4183 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4185 unixctl_command_reply_error(conn, error);
4190 /* Do the same checks as handle_packet_out() in ofproto.c.
4192 * We pass a 'table_id' of 0 to ofproto_check_ofpacts(), which isn't
4193 * strictly correct because these actions aren't in any table, but it's OK
4194 * because it 'table_id' is used only to check goto_table instructions, but
4195 * packet-outs take a list of actions and therefore it can't include
4198 * We skip the "meter" check here because meter is an instruction, not an
4199 * action, and thus cannot appear in ofpacts. */
4200 in_port = ofp_to_u16(flow.in_port.ofp_port);
4201 if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
4202 unixctl_command_reply_error(conn, "invalid in_port");
4205 if (enforce_consistency) {
4206 retval = ofpacts_check_consistency(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts),
4207 &flow, u16_to_ofp(ofproto->up.max_ports),
4208 0, 0, usable_protocols);
4210 retval = ofpacts_check(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &flow,
4211 u16_to_ofp(ofproto->up.max_ports), 0, 0,
4217 ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
4218 unixctl_command_reply_error(conn, ds_cstr(&result));
4222 ofproto_trace(ofproto, &flow, packet,
4223 ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &result);
4224 unixctl_command_reply(conn, ds_cstr(&result));
4227 ds_destroy(&result);
4228 ofpbuf_delete(packet);
4229 ofpbuf_uninit(&ofpacts);
4232 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
4233 * description of the results to 'ds'.
4235 * The trace follows a packet with the specified 'flow' through the flow
4236 * table. 'packet' may be nonnull to trace an actual packet, with consequent
4237 * side effects (if it is nonnull then its flow must be 'flow').
4239 * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
4240 * trace, otherwise the actions are determined by a flow table lookup. */
4242 ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
4243 const struct ofpbuf *packet,
4244 const struct ofpact ofpacts[], size_t ofpacts_len,
4247 struct rule_dpif *rule;
4248 struct trace_ctx trace;
4250 ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
4251 ds_put_cstr(ds, "Flow: ");
4252 flow_format(ds, flow);
4253 ds_put_char(ds, '\n');
4255 flow_wildcards_init_catchall(&trace.wc);
4259 rule_dpif_lookup(ofproto, flow, &trace.wc, &rule, false);
4261 trace_format_rule(ds, 0, rule);
4262 if (rule == ofproto->miss_rule) {
4263 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
4264 } else if (rule == ofproto->no_packet_in_rule) {
4265 ds_put_cstr(ds, "\nNo match, packets dropped because "
4266 "OFPPC_NO_PACKET_IN is set on in_port.\n");
4267 } else if (rule == ofproto->drop_frags_rule) {
4268 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
4269 "and the fragment handling mode is \"drop\".\n");
4273 if (rule || ofpacts) {
4275 trace.key = flow; /* Original flow key, used for megaflow. */
4276 trace.flow = *flow; /* May be modified by actions. */
4277 xlate_in_init(&trace.xin, ofproto, flow, rule, ntohs(flow->tcp_flags),
4280 trace.xin.ofpacts = ofpacts;
4281 trace.xin.ofpacts_len = ofpacts_len;
4283 trace.xin.resubmit_hook = trace_resubmit;
4284 trace.xin.report_hook = trace_report;
4286 xlate_actions(&trace.xin, &trace.xout);
4288 ds_put_char(ds, '\n');
4289 trace_format_flow(ds, 0, "Final flow", &trace);
4290 trace_format_megaflow(ds, 0, "Megaflow", &trace);
4292 ds_put_cstr(ds, "Datapath actions: ");
4293 format_odp_actions(ds, ofpbuf_data(&trace.xout.odp_actions),
4294 ofpbuf_size(&trace.xout.odp_actions));
4296 if (trace.xout.slow) {
4297 enum slow_path_reason slow;
4299 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
4300 "slow path because it:");
4302 slow = trace.xout.slow;
4304 enum slow_path_reason bit = rightmost_1bit(slow);
4306 ds_put_format(ds, "\n\t- %s.",
4307 slow_path_reason_to_explanation(bit));
4313 xlate_out_uninit(&trace.xout);
4317 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
4318 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
4319 * to destroy 'ofproto_shash' and free the returned value. */
4320 static const struct shash_node **
4321 get_ofprotos(struct shash *ofproto_shash)
4323 const struct ofproto_dpif *ofproto;
4325 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4326 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
4327 shash_add_nocopy(ofproto_shash, name, ofproto);
4330 return shash_sort(ofproto_shash);
4334 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
4335 const char *argv[] OVS_UNUSED,
4336 void *aux OVS_UNUSED)
4338 struct ds ds = DS_EMPTY_INITIALIZER;
4339 struct shash ofproto_shash;
4340 const struct shash_node **sorted_ofprotos;
4343 shash_init(&ofproto_shash);
4344 sorted_ofprotos = get_ofprotos(&ofproto_shash);
4345 for (i = 0; i < shash_count(&ofproto_shash); i++) {
4346 const struct shash_node *node = sorted_ofprotos[i];
4347 ds_put_format(&ds, "%s\n", node->name);
4350 shash_destroy(&ofproto_shash);
4351 free(sorted_ofprotos);
4353 unixctl_command_reply(conn, ds_cstr(&ds));
4358 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
4360 const struct shash_node **ofprotos;
4361 struct dpif_dp_stats dp_stats;
4362 struct shash ofproto_shash;
4365 dpif_get_dp_stats(backer->dpif, &dp_stats);
4367 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
4368 dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
4370 shash_init(&ofproto_shash);
4371 ofprotos = get_ofprotos(&ofproto_shash);
4372 for (i = 0; i < shash_count(&ofproto_shash); i++) {
4373 struct ofproto_dpif *ofproto = ofprotos[i]->data;
4374 const struct shash_node **ports;
4377 if (ofproto->backer != backer) {
4381 ds_put_format(ds, "\t%s:\n", ofproto->up.name);
4383 ports = shash_sort(&ofproto->up.port_by_name);
4384 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
4385 const struct shash_node *node = ports[j];
4386 struct ofport *ofport = node->data;
4388 odp_port_t odp_port;
4390 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
4393 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
4394 if (odp_port != ODPP_NONE) {
4395 ds_put_format(ds, "%"PRIu32":", odp_port);
4397 ds_put_cstr(ds, "none:");
4400 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
4403 if (!netdev_get_config(ofport->netdev, &config)) {
4404 const struct smap_node **nodes;
4407 nodes = smap_sort(&config);
4408 for (i = 0; i < smap_count(&config); i++) {
4409 const struct smap_node *node = nodes[i];
4410 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
4411 node->key, node->value);
4415 smap_destroy(&config);
4417 ds_put_char(ds, ')');
4418 ds_put_char(ds, '\n');
4422 shash_destroy(&ofproto_shash);
4427 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4428 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4430 struct ds ds = DS_EMPTY_INITIALIZER;
4431 const struct shash_node **backers;
4434 backers = shash_sort(&all_dpif_backers);
4435 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
4436 dpif_show_backer(backers[i]->data, &ds);
4440 unixctl_command_reply(conn, ds_cstr(&ds));
4445 ofproto_dpif_contains_flow(const struct ofproto_dpif *ofproto,
4446 const struct nlattr *key, size_t key_len)
4448 struct ofproto_dpif *ofp;
4451 xlate_receive(ofproto->backer, NULL, key, key_len, &flow, &ofp,
4452 NULL, NULL, NULL, NULL);
4453 return ofp == ofproto;
4457 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
4458 int argc OVS_UNUSED, const char *argv[],
4459 void *aux OVS_UNUSED)
4461 struct ds ds = DS_EMPTY_INITIALIZER;
4462 const struct dpif_flow_stats *stats;
4463 const struct ofproto_dpif *ofproto;
4464 struct dpif_flow_dump flow_dump;
4465 const struct nlattr *actions;
4466 const struct nlattr *mask;
4467 const struct nlattr *key;
4471 bool verbosity = false;
4472 struct dpif_port dpif_port;
4473 struct dpif_port_dump port_dump;
4474 struct hmap portno_names;
4478 ofproto = ofproto_dpif_lookup(argv[argc - 1]);
4480 unixctl_command_reply_error(conn, "no such bridge");
4484 if (argc > 2 && !strcmp(argv[1], "-m")) {
4488 hmap_init(&portno_names);
4489 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
4490 odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
4494 error = dpif_flow_dump_start(&flow_dump, ofproto->backer->dpif);
4498 dpif_flow_dump_state_init(ofproto->backer->dpif, &state);
4499 while (dpif_flow_dump_next(&flow_dump, state, &key, &key_len,
4500 &mask, &mask_len, &actions, &actions_len,
4502 if (!ofproto_dpif_contains_flow(ofproto, key, key_len)) {
4506 odp_flow_format(key, key_len, mask, mask_len, &portno_names, &ds,
4508 ds_put_cstr(&ds, ", ");
4509 dpif_flow_stats_format(stats, &ds);
4510 ds_put_cstr(&ds, ", actions:");
4511 format_odp_actions(&ds, actions, actions_len);
4512 ds_put_char(&ds, '\n');
4514 dpif_flow_dump_state_uninit(ofproto->backer->dpif, state);
4515 error = dpif_flow_dump_done(&flow_dump);
4520 ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
4521 unixctl_command_reply_error(conn, ds_cstr(&ds));
4523 unixctl_command_reply(conn, ds_cstr(&ds));
4525 odp_portno_names_destroy(&portno_names);
4526 hmap_destroy(&portno_names);
4531 ofproto_dpif_unixctl_init(void)
4533 static bool registered;
4539 unixctl_command_register(
4541 "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
4542 1, 3, ofproto_unixctl_trace, NULL);
4543 unixctl_command_register(
4544 "ofproto/trace-packet-out",
4545 "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
4546 2, 6, ofproto_unixctl_trace_actions, NULL);
4547 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
4548 ofproto_unixctl_fdb_flush, NULL);
4549 unixctl_command_register("fdb/show", "bridge", 1, 1,
4550 ofproto_unixctl_fdb_show, NULL);
4551 unixctl_command_register("dpif/dump-dps", "", 0, 0,
4552 ofproto_unixctl_dpif_dump_dps, NULL);
4553 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
4555 unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
4556 ofproto_unixctl_dpif_dump_flows, NULL);
4560 /* Returns true if 'rule' is an internal rule, false otherwise. */
4562 rule_is_internal(const struct rule *rule)
4564 return rule->table_id == TBL_INTERNAL;
4567 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4569 * This is deprecated. It is only for compatibility with broken device drivers
4570 * in old versions of Linux that do not properly support VLANs when VLAN
4571 * devices are not used. When broken device drivers are no longer in
4572 * widespread use, we will delete these interfaces. */
4575 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
4577 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
4578 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
4580 if (realdev_ofp_port == ofport->realdev_ofp_port
4581 && vid == ofport->vlandev_vid) {
4585 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4587 if (ofport->realdev_ofp_port) {
4590 if (realdev_ofp_port && ofport->bundle) {
4591 /* vlandevs are enslaved to their realdevs, so they are not allowed to
4592 * themselves be part of a bundle. */
4593 bundle_set(ofport_->ofproto, ofport->bundle, NULL);
4596 ofport->realdev_ofp_port = realdev_ofp_port;
4597 ofport->vlandev_vid = vid;
4599 if (realdev_ofp_port) {
4600 vsp_add(ofport, realdev_ofp_port, vid);
4607 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
4609 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
4613 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
4614 OVS_EXCLUDED(ofproto->vsp_mutex)
4616 /* hmap_is_empty is thread safe. */
4617 return !hmap_is_empty(&ofproto->realdev_vid_map);
4622 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
4623 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
4624 OVS_REQUIRES(ofproto->vsp_mutex)
4626 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
4627 int vid = vlan_tci_to_vid(vlan_tci);
4628 const struct vlan_splinter *vsp;
4630 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
4631 hash_realdev_vid(realdev_ofp_port, vid),
4632 &ofproto->realdev_vid_map) {
4633 if (vsp->realdev_ofp_port == realdev_ofp_port
4634 && vsp->vid == vid) {
4635 return vsp->vlandev_ofp_port;
4639 return realdev_ofp_port;
4642 /* Returns the OFP port number of the Linux VLAN device that corresponds to
4643 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
4644 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
4645 * 'vlan_tci' 9, it would return the port number of eth0.9.
4647 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
4648 * function just returns its 'realdev_ofp_port' argument. */
4650 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
4651 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
4652 OVS_EXCLUDED(ofproto->vsp_mutex)
4656 /* hmap_is_empty is thread safe, see if we can return immediately. */
4657 if (hmap_is_empty(&ofproto->realdev_vid_map)) {
4658 return realdev_ofp_port;
4660 ovs_mutex_lock(&ofproto->vsp_mutex);
4661 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
4662 ovs_mutex_unlock(&ofproto->vsp_mutex);
4666 static struct vlan_splinter *
4667 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
4669 struct vlan_splinter *vsp;
4671 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
4672 hash_ofp_port(vlandev_ofp_port),
4673 &ofproto->vlandev_map) {
4674 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
4682 /* Returns the OpenFlow port number of the "real" device underlying the Linux
4683 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
4684 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
4685 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
4686 * eth0 and store 9 in '*vid'.
4688 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
4689 * VLAN device. Unless VLAN splinters are enabled, this is what this function
4692 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
4693 ofp_port_t vlandev_ofp_port, int *vid)
4694 OVS_REQUIRES(ofproto->vsp_mutex)
4696 if (!hmap_is_empty(&ofproto->vlandev_map)) {
4697 const struct vlan_splinter *vsp;
4699 vsp = vlandev_find(ofproto, vlandev_ofp_port);
4704 return vsp->realdev_ofp_port;
4710 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
4711 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
4712 * 'flow->in_port' to the "real" device backing the VLAN device, sets
4713 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
4714 * always the case unless VLAN splinters are enabled), returns false without
4715 * making any changes. */
4717 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
4718 OVS_EXCLUDED(ofproto->vsp_mutex)
4723 /* hmap_is_empty is thread safe. */
4724 if (hmap_is_empty(&ofproto->vlandev_map)) {
4728 ovs_mutex_lock(&ofproto->vsp_mutex);
4729 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
4730 ovs_mutex_unlock(&ofproto->vsp_mutex);
4735 /* Cause the flow to be processed as if it came in on the real device with
4736 * the VLAN device's VLAN ID. */
4737 flow->in_port.ofp_port = realdev;
4738 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
4743 vsp_remove(struct ofport_dpif *port)
4745 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
4746 struct vlan_splinter *vsp;
4748 ovs_mutex_lock(&ofproto->vsp_mutex);
4749 vsp = vlandev_find(ofproto, port->up.ofp_port);
4751 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
4752 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
4755 port->realdev_ofp_port = 0;
4757 VLOG_ERR("missing vlan device record");
4759 ovs_mutex_unlock(&ofproto->vsp_mutex);
4763 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
4765 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
4767 ovs_mutex_lock(&ofproto->vsp_mutex);
4768 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
4769 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
4770 == realdev_ofp_port)) {
4771 struct vlan_splinter *vsp;
4773 vsp = xmalloc(sizeof *vsp);
4774 vsp->realdev_ofp_port = realdev_ofp_port;
4775 vsp->vlandev_ofp_port = port->up.ofp_port;
4778 port->realdev_ofp_port = realdev_ofp_port;
4780 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
4781 hash_ofp_port(port->up.ofp_port));
4782 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
4783 hash_realdev_vid(realdev_ofp_port, vid));
4785 VLOG_ERR("duplicate vlan device record");
4787 ovs_mutex_unlock(&ofproto->vsp_mutex);
4791 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
4793 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
4794 return ofport ? ofport->odp_port : ODPP_NONE;
4797 struct ofport_dpif *
4798 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
4800 struct ofport_dpif *port;
4802 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
4803 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
4804 &backer->odp_to_ofport_map) {
4805 if (port->odp_port == odp_port) {
4806 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
4811 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
4816 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
4818 struct ofport_dpif *port;
4820 port = odp_port_to_ofport(ofproto->backer, odp_port);
4821 if (port && &ofproto->up == port->up.ofproto) {
4822 return port->up.ofp_port;
4829 ofproto_dpif_alloc_recirc_id(struct ofproto_dpif *ofproto)
4831 struct dpif_backer *backer = ofproto->backer;
4833 return recirc_id_alloc(backer->rid_pool);
4837 ofproto_dpif_free_recirc_id(struct ofproto_dpif *ofproto, uint32_t recirc_id)
4839 struct dpif_backer *backer = ofproto->backer;
4841 recirc_id_free(backer->rid_pool, recirc_id);
4845 ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
4846 struct match *match, int priority,
4847 const struct ofpbuf *ofpacts,
4848 struct rule **rulep)
4850 struct ofputil_flow_mod fm;
4851 struct rule_dpif *rule;
4855 fm.priority = priority;
4856 fm.new_cookie = htonll(0);
4857 fm.cookie = htonll(0);
4858 fm.cookie_mask = htonll(0);
4859 fm.modify_cookie = false;
4860 fm.table_id = TBL_INTERNAL;
4861 fm.command = OFPFC_ADD;
4862 fm.idle_timeout = 0;
4863 fm.hard_timeout = 0;
4866 fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
4867 fm.ofpacts = ofpbuf_data(ofpacts);
4868 fm.ofpacts_len = ofpbuf_size(ofpacts);
4870 error = ofproto_flow_mod(&ofproto->up, &fm);
4872 VLOG_ERR_RL(&rl, "failed to add internal flow (%s)",
4873 ofperr_to_string(error));
4878 rule = rule_dpif_lookup_in_table(ofproto, TBL_INTERNAL, &match->flow,
4889 ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
4890 struct match *match, int priority)
4892 struct ofputil_flow_mod fm;
4896 fm.priority = priority;
4897 fm.new_cookie = htonll(0);
4898 fm.cookie = htonll(0);
4899 fm.cookie_mask = htonll(0);
4900 fm.modify_cookie = false;
4901 fm.table_id = TBL_INTERNAL;
4902 fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
4903 fm.command = OFPFC_DELETE_STRICT;
4905 error = ofproto_flow_mod(&ofproto->up, &fm);
4907 VLOG_ERR_RL(&rl, "failed to delete internal flow (%s)",
4908 ofperr_to_string(error));
4915 const struct ofproto_class ofproto_dpif_class = {
4929 NULL, /* get_memory_usage. */
4930 type_get_memory_usage,
4949 port_is_lacp_current,
4950 NULL, /* rule_choose_table */
4959 rule_modify_actions,
4973 get_stp_port_status,
4981 is_mirror_output_bundle,
4982 forward_bpdu_changed,
4983 set_mac_table_config,
4985 NULL, /* meter_get_features */
4986 NULL, /* meter_set */
4987 NULL, /* meter_get */
4988 NULL, /* meter_del */
4989 group_alloc, /* group_alloc */
4990 group_construct, /* group_construct */
4991 group_destruct, /* group_destruct */
4992 group_dealloc, /* group_dealloc */
4993 group_modify, /* group_modify */
4994 group_get_stats, /* group_get_stats */