2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-dpif.h"
20 #include "ofproto/ofproto-provider.h"
27 #include "byte-order.h"
28 #include "connectivity.h"
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"
61 #include "poll-loop.h"
67 #include "unaligned.h"
69 #include "vlan-bitmap.h"
72 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
74 COVERAGE_DEFINE(ofproto_dpif_expired);
75 COVERAGE_DEFINE(packet_in_overflow);
77 /* No bfd/cfm status change. */
78 #define NO_STATUS_CHANGE -1
87 * - Do include packets and bytes from datapath flows which have not
88 * recently been processed by a revalidator. */
89 struct ovs_mutex stats_mutex;
90 struct dpif_flow_stats stats OVS_GUARDED;
93 /* RULE_CAST() depends on this. */
94 BUILD_ASSERT_DECL(offsetof(struct rule_dpif, up) == 0);
96 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes,
98 static struct rule_dpif *rule_dpif_cast(const struct rule *);
99 static void rule_expire(struct rule_dpif *);
106 * - Do include packets and bytes from datapath flows which have not
107 * recently been processed by a revalidator. */
108 struct ovs_mutex stats_mutex;
109 uint64_t packet_count OVS_GUARDED; /* Number of packets received. */
110 uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */
111 struct bucket_counter *bucket_stats OVS_GUARDED; /* Bucket statistics. */
115 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
116 struct ofproto_dpif *ofproto; /* Owning ofproto. */
117 void *aux; /* Key supplied by ofproto's client. */
118 char *name; /* Identifier for log messages. */
121 struct list ports; /* Contains "struct ofport"s. */
122 enum port_vlan_mode vlan_mode; /* VLAN mode */
123 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
124 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
125 * NULL if all VLANs are trunked. */
126 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
127 struct bond *bond; /* Nonnull iff more than one port. */
128 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
131 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
134 static void bundle_remove(struct ofport *);
135 static void bundle_update(struct ofbundle *);
136 static void bundle_destroy(struct ofbundle *);
137 static void bundle_del_port(struct ofport_dpif *);
138 static void bundle_run(struct ofbundle *);
139 static void bundle_wait(struct ofbundle *);
141 static void stp_run(struct ofproto_dpif *ofproto);
142 static void stp_wait(struct ofproto_dpif *ofproto);
143 static int set_stp_port(struct ofport *,
144 const struct ofproto_port_stp_settings *);
147 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
151 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
152 struct list bundle_node; /* In struct ofbundle's "ports" list. */
153 struct cfm *cfm; /* Connectivity Fault Management, if any. */
154 struct bfd *bfd; /* BFD, if any. */
155 bool may_enable; /* May be enabled in bonds. */
156 bool is_tunnel; /* This port is a tunnel. */
157 bool is_layer3; /* This is a layer 3 port. */
158 long long int carrier_seq; /* Carrier status changes. */
159 struct ofport_dpif *peer; /* Peer if patch port. */
162 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
163 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
164 long long int stp_state_entered;
166 /* Queue to DSCP mapping. */
167 struct ofproto_port_queue *qdscp;
170 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
172 * This is deprecated. It is only for compatibility with broken device
173 * drivers in old versions of Linux that do not properly support VLANs when
174 * VLAN devices are not used. When broken device drivers are no longer in
175 * widespread use, we will delete these interfaces. */
176 ofp_port_t realdev_ofp_port;
180 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
182 * This is deprecated. It is only for compatibility with broken device drivers
183 * in old versions of Linux that do not properly support VLANs when VLAN
184 * devices are not used. When broken device drivers are no longer in
185 * widespread use, we will delete these interfaces. */
186 struct vlan_splinter {
187 struct hmap_node realdev_vid_node;
188 struct hmap_node vlandev_node;
189 ofp_port_t realdev_ofp_port;
190 ofp_port_t vlandev_ofp_port;
194 static void vsp_remove(struct ofport_dpif *);
195 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
197 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
200 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
203 static struct ofport_dpif *
204 ofport_dpif_cast(const struct ofport *ofport)
206 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
209 static void port_run(struct ofport_dpif *);
210 static int set_bfd(struct ofport *, const struct smap *);
211 static int set_cfm(struct ofport *, const struct cfm_settings *);
212 static void ofport_update_peer(struct ofport_dpif *);
214 struct dpif_completion {
215 struct list list_node;
216 struct ofoperation *op;
219 /* Reasons that we might need to revalidate every datapath flow, and
220 * corresponding coverage counters.
222 * A value of 0 means that there is no need to revalidate.
224 * It would be nice to have some cleaner way to integrate with coverage
225 * counters, but with only a few reasons I guess this is good enough for
227 enum revalidate_reason {
228 REV_RECONFIGURE = 1, /* Switch configuration changed. */
229 REV_STP, /* Spanning tree protocol port status change. */
230 REV_BOND, /* Bonding changed. */
231 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
232 REV_FLOW_TABLE, /* Flow table changed. */
233 REV_MAC_LEARNING, /* Mac learning changed. */
235 COVERAGE_DEFINE(rev_reconfigure);
236 COVERAGE_DEFINE(rev_stp);
237 COVERAGE_DEFINE(rev_bond);
238 COVERAGE_DEFINE(rev_port_toggled);
239 COVERAGE_DEFINE(rev_flow_table);
240 COVERAGE_DEFINE(rev_mac_learning);
242 /* Stores mapping between 'recirc_id' and 'ofproto-dpif'. */
243 struct dpif_backer_recirc_node {
244 struct hmap_node hmap_node;
245 struct ofproto_dpif *ofproto;
249 /* All datapaths of a given type share a single dpif backer instance. */
256 struct ovs_rwlock odp_to_ofport_lock;
257 struct hmap odp_to_ofport_map OVS_GUARDED; /* Contains "struct ofport"s. */
259 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
261 enum revalidate_reason need_revalidate; /* Revalidate all flows. */
263 bool recv_set_enable; /* Enables or disables receiving packets. */
266 struct recirc_id_pool *rid_pool; /* Recirculation ID pool. */
267 struct hmap recirc_map; /* Map of 'recirc_id's to 'ofproto's. */
268 struct ovs_mutex recirc_mutex; /* Protects 'recirc_map'. */
269 bool enable_recirc; /* True if the datapath supports recirculation */
271 /* True if the datapath supports variable-length
272 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.
273 * False if the datapath supports only 8-byte (or shorter) userdata. */
274 bool variable_length_userdata;
276 /* Maximum number of MPLS label stack entries that the datapath supports
278 size_t max_mpls_depth;
281 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
282 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
284 struct ofproto_dpif {
285 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
287 struct dpif_backer *backer;
289 uint64_t dump_seq; /* Last read of udpif_dump_seq(). */
291 /* Special OpenFlow rules. */
292 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
293 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
294 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
297 struct netflow *netflow;
298 struct dpif_sflow *sflow;
299 struct dpif_ipfix *ipfix;
300 struct hmap bundles; /* Contains "struct ofbundle"s. */
301 struct mac_learning *ml;
302 bool has_bonded_bundles;
304 struct mbridge *mbridge;
306 struct ovs_mutex stats_mutex;
307 struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
308 * consumed in userspace. */
312 long long int stp_last_tick;
314 /* VLAN splinters. */
315 struct ovs_mutex vsp_mutex;
316 struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
317 struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */
320 struct sset ports; /* Set of standard port names. */
321 struct sset ghost_ports; /* Ports with no datapath port. */
322 struct sset port_poll_set; /* Queued names for port_poll() reply. */
323 int port_poll_errno; /* Last errno for port_poll() reply. */
324 uint64_t change_seq; /* Connectivity status changes. */
327 struct guarded_list pins; /* Contains "struct ofputil_packet_in"s. */
328 struct seq *pins_seq; /* For notifying 'pins' reception. */
332 /* All existing ofproto_dpif instances, indexed by ->up.name. */
333 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
335 static void ofproto_dpif_unixctl_init(void);
337 static inline struct ofproto_dpif *
338 ofproto_dpif_cast(const struct ofproto *ofproto)
340 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
341 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
345 ofproto_dpif_get_max_mpls_depth(const struct ofproto_dpif *ofproto)
347 return ofproto->backer->max_mpls_depth;
351 ofproto_dpif_get_enable_recirc(const struct ofproto_dpif *ofproto)
353 return ofproto->backer->enable_recirc;
356 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
357 ofp_port_t ofp_port);
358 static void ofproto_trace(struct ofproto_dpif *, struct flow *,
359 const struct ofpbuf *packet,
360 const struct ofpact[], size_t ofpacts_len,
363 /* Global variables. */
364 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
366 /* Initial mappings of port to bridge mappings. */
367 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
369 /* Executes 'fm'. The caller retains ownership of 'fm' and everything in
372 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
373 struct ofputil_flow_mod *fm)
375 ofproto_flow_mod(&ofproto->up, fm);
378 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
379 * Takes ownership of 'pin' and pin->packet. */
381 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
382 struct ofproto_packet_in *pin)
384 if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
385 COVERAGE_INC(packet_in_overflow);
386 free(CONST_CAST(void *, pin->up.packet));
390 /* Wakes up main thread for packet-in I/O. */
391 seq_change(ofproto->pins_seq);
394 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
395 * packet rather than to send the packet to the controller.
397 * This function returns false to indicate that a packet_in message
398 * for a "table-miss" should be sent to at least one controller.
399 * False otherwise. */
401 ofproto_dpif_wants_packet_in_on_miss(struct ofproto_dpif *ofproto)
403 return connmgr_wants_packet_in_on_miss(ofproto->up.connmgr);
406 /* Factory functions. */
409 init(const struct shash *iface_hints)
411 struct shash_node *node;
413 fat_rwlock_init(&xlate_rwlock);
415 /* Make a local copy, since we don't own 'iface_hints' elements. */
416 SHASH_FOR_EACH(node, iface_hints) {
417 const struct iface_hint *orig_hint = node->data;
418 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
420 new_hint->br_name = xstrdup(orig_hint->br_name);
421 new_hint->br_type = xstrdup(orig_hint->br_type);
422 new_hint->ofp_port = orig_hint->ofp_port;
424 shash_add(&init_ofp_ports, node->name, new_hint);
429 enumerate_types(struct sset *types)
431 dp_enumerate_types(types);
435 enumerate_names(const char *type, struct sset *names)
437 struct ofproto_dpif *ofproto;
440 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
441 if (strcmp(type, ofproto->up.type)) {
444 sset_add(names, ofproto->up.name);
451 del(const char *type, const char *name)
456 error = dpif_open(name, type, &dpif);
458 error = dpif_delete(dpif);
465 port_open_type(const char *datapath_type, const char *port_type)
467 return dpif_port_open_type(datapath_type, port_type);
470 /* Type functions. */
472 static void process_dpif_port_changes(struct dpif_backer *);
473 static void process_dpif_all_ports_changed(struct dpif_backer *);
474 static void process_dpif_port_change(struct dpif_backer *,
475 const char *devname);
476 static void process_dpif_port_error(struct dpif_backer *, int error);
478 static struct ofproto_dpif *
479 lookup_ofproto_dpif_by_port_name(const char *name)
481 struct ofproto_dpif *ofproto;
483 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
484 if (sset_contains(&ofproto->ports, name)) {
493 type_run(const char *type)
495 struct dpif_backer *backer;
497 backer = shash_find_data(&all_dpif_backers, type);
499 /* This is not necessarily a problem, since backers are only
500 * created on demand. */
504 dpif_run(backer->dpif);
506 /* If vswitchd started with other_config:flow_restore_wait set as "true",
507 * and the configuration has now changed to "false", enable receiving
508 * packets from the datapath. */
509 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
512 backer->recv_set_enable = true;
514 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
516 VLOG_ERR("Failed to enable receiving packets in dpif.");
519 dpif_flow_flush(backer->dpif);
520 backer->need_revalidate = REV_RECONFIGURE;
523 if (backer->recv_set_enable) {
524 udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
527 if (backer->need_revalidate) {
528 struct ofproto_dpif *ofproto;
529 struct simap_node *node;
530 struct simap tmp_backers;
532 /* Handle tunnel garbage collection. */
533 simap_init(&tmp_backers);
534 simap_swap(&backer->tnl_backers, &tmp_backers);
536 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
537 struct ofport_dpif *iter;
539 if (backer != ofproto->backer) {
543 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
544 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
547 if (!iter->is_tunnel) {
551 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
552 namebuf, sizeof namebuf);
553 node = simap_find(&tmp_backers, dp_port);
555 simap_put(&backer->tnl_backers, dp_port, node->data);
556 simap_delete(&tmp_backers, node);
557 node = simap_find(&backer->tnl_backers, dp_port);
559 node = simap_find(&backer->tnl_backers, dp_port);
561 odp_port_t odp_port = ODPP_NONE;
563 if (!dpif_port_add(backer->dpif, iter->up.netdev,
565 simap_put(&backer->tnl_backers, dp_port,
566 odp_to_u32(odp_port));
567 node = simap_find(&backer->tnl_backers, dp_port);
572 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
573 if (tnl_port_reconfigure(iter, iter->up.netdev,
575 backer->need_revalidate = REV_RECONFIGURE;
580 SIMAP_FOR_EACH (node, &tmp_backers) {
581 dpif_port_del(backer->dpif, u32_to_odp(node->data));
583 simap_destroy(&tmp_backers);
585 switch (backer->need_revalidate) {
586 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
587 case REV_STP: COVERAGE_INC(rev_stp); break;
588 case REV_BOND: COVERAGE_INC(rev_bond); break;
589 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
590 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
591 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
593 backer->need_revalidate = 0;
595 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
596 struct ofport_dpif *ofport;
597 struct ofbundle *bundle;
599 if (ofproto->backer != backer) {
603 fat_rwlock_wrlock(&xlate_rwlock);
604 xlate_ofproto_set(ofproto, ofproto->up.name,
605 ofproto->backer->dpif, ofproto->miss_rule,
606 ofproto->no_packet_in_rule, ofproto->ml,
607 ofproto->stp, ofproto->mbridge,
608 ofproto->sflow, ofproto->ipfix,
609 ofproto->netflow, ofproto->up.frag_handling,
610 ofproto->up.forward_bpdu,
611 connmgr_has_in_band(ofproto->up.connmgr),
612 ofproto->backer->enable_recirc,
613 ofproto->backer->variable_length_userdata,
614 ofproto->backer->max_mpls_depth);
616 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
617 xlate_bundle_set(ofproto, bundle, bundle->name,
618 bundle->vlan_mode, bundle->vlan,
619 bundle->trunks, bundle->use_priority_tags,
620 bundle->bond, bundle->lacp,
624 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
625 int stp_port = ofport->stp_port
626 ? stp_port_no(ofport->stp_port)
628 xlate_ofport_set(ofproto, ofport->bundle, ofport,
629 ofport->up.ofp_port, ofport->odp_port,
630 ofport->up.netdev, ofport->cfm,
631 ofport->bfd, ofport->peer, stp_port,
632 ofport->qdscp, ofport->n_qdscp,
633 ofport->up.pp.config, ofport->up.pp.state,
634 ofport->is_tunnel, ofport->may_enable);
636 fat_rwlock_unlock(&xlate_rwlock);
639 udpif_revalidate(backer->udpif);
642 process_dpif_port_changes(backer);
647 /* Check for and handle port changes in 'backer''s dpif. */
649 process_dpif_port_changes(struct dpif_backer *backer)
655 error = dpif_port_poll(backer->dpif, &devname);
661 process_dpif_all_ports_changed(backer);
665 process_dpif_port_change(backer, devname);
670 process_dpif_port_error(backer, error);
677 process_dpif_all_ports_changed(struct dpif_backer *backer)
679 struct ofproto_dpif *ofproto;
680 struct dpif_port dpif_port;
681 struct dpif_port_dump dump;
682 struct sset devnames;
685 sset_init(&devnames);
686 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
687 if (ofproto->backer == backer) {
688 struct ofport *ofport;
690 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
691 sset_add(&devnames, netdev_get_name(ofport->netdev));
695 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
696 sset_add(&devnames, dpif_port.name);
699 SSET_FOR_EACH (devname, &devnames) {
700 process_dpif_port_change(backer, devname);
702 sset_destroy(&devnames);
706 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
708 struct ofproto_dpif *ofproto;
709 struct dpif_port port;
711 /* Don't report on the datapath's device. */
712 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
716 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
717 &all_ofproto_dpifs) {
718 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
723 ofproto = lookup_ofproto_dpif_by_port_name(devname);
724 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
725 /* The port was removed. If we know the datapath,
726 * report it through poll_set(). If we don't, it may be
727 * notifying us of a removal we initiated, so ignore it.
728 * If there's a pending ENOBUFS, let it stand, since
729 * everything will be reevaluated. */
730 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
731 sset_add(&ofproto->port_poll_set, devname);
732 ofproto->port_poll_errno = 0;
734 } else if (!ofproto) {
735 /* The port was added, but we don't know with which
736 * ofproto we should associate it. Delete it. */
737 dpif_port_del(backer->dpif, port.port_no);
739 struct ofport_dpif *ofport;
741 ofport = ofport_dpif_cast(shash_find_data(
742 &ofproto->up.port_by_name, devname));
744 && ofport->odp_port != port.port_no
745 && !odp_port_to_ofport(backer, port.port_no))
747 /* 'ofport''s datapath port number has changed from
748 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
749 * structures to match. */
750 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
751 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
752 ofport->odp_port = port.port_no;
753 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
754 hash_odp_port(port.port_no));
755 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
756 backer->need_revalidate = REV_RECONFIGURE;
759 dpif_port_destroy(&port);
762 /* Propagate 'error' to all ofprotos based on 'backer'. */
764 process_dpif_port_error(struct dpif_backer *backer, int error)
766 struct ofproto_dpif *ofproto;
768 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
769 if (ofproto->backer == backer) {
770 sset_clear(&ofproto->port_poll_set);
771 ofproto->port_poll_errno = error;
777 type_wait(const char *type)
779 struct dpif_backer *backer;
781 backer = shash_find_data(&all_dpif_backers, type);
783 /* This is not necessarily a problem, since backers are only
784 * created on demand. */
788 dpif_wait(backer->dpif);
791 /* Basic life-cycle. */
793 static int add_internal_flows(struct ofproto_dpif *);
795 static struct ofproto *
798 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
803 dealloc(struct ofproto *ofproto_)
805 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
809 /* Called when 'ofproto' is destructed. Checks for and clears any
812 dpif_backer_recirc_clear_ofproto(struct dpif_backer *backer,
813 struct ofproto_dpif *ofproto)
815 struct dpif_backer_recirc_node *node;
817 ovs_mutex_lock(&backer->recirc_mutex);
818 HMAP_FOR_EACH (node, hmap_node, &backer->recirc_map) {
819 if (node->ofproto == ofproto) {
820 VLOG_ERR("recirc_id %"PRIu32", not freed when ofproto (%s) "
821 "is destructed", node->recirc_id, ofproto->up.name);
822 hmap_remove(&backer->recirc_map, &node->hmap_node);
823 ovsrcu_postpone(free, node);
826 ovs_mutex_unlock(&backer->recirc_mutex);
830 close_dpif_backer(struct dpif_backer *backer)
832 ovs_assert(backer->refcount > 0);
834 if (--backer->refcount) {
838 udpif_destroy(backer->udpif);
840 simap_destroy(&backer->tnl_backers);
841 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
842 hmap_destroy(&backer->odp_to_ofport_map);
843 shash_find_and_delete(&all_dpif_backers, backer->type);
844 recirc_id_pool_destroy(backer->rid_pool);
845 hmap_destroy(&backer->recirc_map);
846 ovs_mutex_destroy(&backer->recirc_mutex);
848 dpif_close(backer->dpif);
852 /* Datapath port slated for removal from datapath. */
854 struct list list_node;
858 static bool check_variable_length_userdata(struct dpif_backer *backer);
859 static size_t check_max_mpls_depth(struct dpif_backer *backer);
860 static bool check_recirc(struct dpif_backer *backer);
863 open_dpif_backer(const char *type, struct dpif_backer **backerp)
865 struct dpif_backer *backer;
866 struct dpif_port_dump port_dump;
867 struct dpif_port port;
868 struct shash_node *node;
869 struct list garbage_list;
870 struct odp_garbage *garbage, *next;
877 backer = shash_find_data(&all_dpif_backers, type);
884 backer_name = xasprintf("ovs-%s", type);
886 /* Remove any existing datapaths, since we assume we're the only
887 * userspace controlling the datapath. */
889 dp_enumerate_names(type, &names);
890 SSET_FOR_EACH(name, &names) {
891 struct dpif *old_dpif;
893 /* Don't remove our backer if it exists. */
894 if (!strcmp(name, backer_name)) {
898 if (dpif_open(name, type, &old_dpif)) {
899 VLOG_WARN("couldn't open old datapath %s to remove it", name);
901 dpif_delete(old_dpif);
902 dpif_close(old_dpif);
905 sset_destroy(&names);
907 backer = xmalloc(sizeof *backer);
909 error = dpif_create_and_open(backer_name, type, &backer->dpif);
912 VLOG_ERR("failed to open datapath of type %s: %s", type,
913 ovs_strerror(error));
917 backer->udpif = udpif_create(backer, backer->dpif);
919 backer->type = xstrdup(type);
920 backer->refcount = 1;
921 hmap_init(&backer->odp_to_ofport_map);
922 ovs_rwlock_init(&backer->odp_to_ofport_lock);
923 backer->need_revalidate = 0;
924 simap_init(&backer->tnl_backers);
925 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
928 if (backer->recv_set_enable) {
929 dpif_flow_flush(backer->dpif);
932 /* Loop through the ports already on the datapath and remove any
933 * that we don't need anymore. */
934 list_init(&garbage_list);
935 dpif_port_dump_start(&port_dump, backer->dpif);
936 while (dpif_port_dump_next(&port_dump, &port)) {
937 node = shash_find(&init_ofp_ports, port.name);
938 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
939 garbage = xmalloc(sizeof *garbage);
940 garbage->odp_port = port.port_no;
941 list_push_front(&garbage_list, &garbage->list_node);
944 dpif_port_dump_done(&port_dump);
946 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
947 dpif_port_del(backer->dpif, garbage->odp_port);
948 list_remove(&garbage->list_node);
952 shash_add(&all_dpif_backers, type, backer);
954 backer->enable_recirc = check_recirc(backer);
955 backer->variable_length_userdata = check_variable_length_userdata(backer);
956 backer->max_mpls_depth = check_max_mpls_depth(backer);
957 backer->rid_pool = recirc_id_pool_create();
958 ovs_mutex_init(&backer->recirc_mutex);
959 hmap_init(&backer->recirc_map);
961 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
963 VLOG_ERR("failed to listen on datapath of type %s: %s",
964 type, ovs_strerror(error));
965 close_dpif_backer(backer);
969 if (backer->recv_set_enable) {
970 udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
976 /* Tests whether 'backer''s datapath supports recirculation. Only newer
977 * datapaths support OVS_KEY_ATTR_RECIRC_ID in keys. We need to disable some
978 * features on older datapaths that don't support this feature.
980 * Returns false if 'backer' definitely does not support recirculation, true if
981 * it seems to support recirculation or if at least the error we get is
984 check_recirc(struct dpif_backer *backer)
987 struct odputil_keybuf keybuf;
990 bool enable_recirc = false;
992 memset(&flow, 0, sizeof flow);
996 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
997 odp_flow_key_from_flow(&key, &flow, NULL, 0);
999 error = dpif_flow_put(backer->dpif, DPIF_FP_CREATE,
1000 ofpbuf_data(&key), ofpbuf_size(&key), NULL, 0, NULL,
1002 if (error && error != EEXIST) {
1003 if (error != EINVAL) {
1004 VLOG_WARN("%s: Reciculation flow probe failed (%s)",
1005 dpif_name(backer->dpif), ovs_strerror(error));
1010 error = dpif_flow_del(backer->dpif, ofpbuf_data(&key), ofpbuf_size(&key),
1013 VLOG_WARN("%s: failed to delete recirculation feature probe flow",
1014 dpif_name(backer->dpif));
1017 enable_recirc = true;
1020 if (enable_recirc) {
1021 VLOG_INFO("%s: Datapath supports recirculation",
1022 dpif_name(backer->dpif));
1024 VLOG_INFO("%s: Datapath does not support recirculation",
1025 dpif_name(backer->dpif));
1028 return enable_recirc;
1031 /* Tests whether 'backer''s datapath supports variable-length
1032 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions. We need
1033 * to disable some features on older datapaths that don't support this
1036 * Returns false if 'backer' definitely does not support variable-length
1037 * userdata, true if it seems to support them or if at least the error we get
1040 check_variable_length_userdata(struct dpif_backer *backer)
1042 struct eth_header *eth;
1043 struct ofpbuf actions;
1044 struct dpif_execute execute;
1045 struct ofpbuf packet;
1049 /* Compose a userspace action that will cause an ERANGE error on older
1050 * datapaths that don't support variable-length userdata.
1052 * We really test for using userdata longer than 8 bytes, but older
1053 * datapaths accepted these, silently truncating the userdata to 8 bytes.
1054 * The same older datapaths rejected userdata shorter than 8 bytes, so we
1055 * test for that instead as a proxy for longer userdata support. */
1056 ofpbuf_init(&actions, 64);
1057 start = nl_msg_start_nested(&actions, OVS_ACTION_ATTR_USERSPACE);
1058 nl_msg_put_u32(&actions, OVS_USERSPACE_ATTR_PID,
1059 dpif_port_get_pid(backer->dpif, ODPP_NONE, 0));
1060 nl_msg_put_unspec_zero(&actions, OVS_USERSPACE_ATTR_USERDATA, 4);
1061 nl_msg_end_nested(&actions, start);
1063 /* Compose a dummy ethernet packet. */
1064 ofpbuf_init(&packet, ETH_HEADER_LEN);
1065 eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
1066 eth->eth_type = htons(0x1234);
1068 /* Execute the actions. On older datapaths this fails with ERANGE, on
1069 * newer datapaths it succeeds. */
1070 execute.actions = ofpbuf_data(&actions);
1071 execute.actions_len = ofpbuf_size(&actions);
1072 execute.packet = &packet;
1073 execute.md = PKT_METADATA_INITIALIZER(0);
1074 execute.needs_help = false;
1076 error = dpif_execute(backer->dpif, &execute);
1078 ofpbuf_uninit(&packet);
1079 ofpbuf_uninit(&actions);
1086 /* Variable-length userdata is not supported. */
1087 VLOG_WARN("%s: datapath does not support variable-length userdata "
1088 "feature (needs Linux 3.10+ or kernel module from OVS "
1089 "1..11+). The NXAST_SAMPLE action will be ignored.",
1090 dpif_name(backer->dpif));
1094 /* Something odd happened. We're not sure whether variable-length
1095 * userdata is supported. Default to "yes". */
1096 VLOG_WARN("%s: variable-length userdata feature probe failed (%s)",
1097 dpif_name(backer->dpif), ovs_strerror(error));
1102 /* Tests the MPLS label stack depth supported by 'backer''s datapath.
1104 * Returns the number of elements in a struct flow's mpls_lse field
1105 * if the datapath supports at least that many entries in an
1107 * Otherwise returns the number of MPLS push actions supported by
1110 check_max_mpls_depth(struct dpif_backer *backer)
1115 for (n = 0; n < FLOW_MAX_MPLS_LABELS; n++) {
1116 struct odputil_keybuf keybuf;
1120 memset(&flow, 0, sizeof flow);
1121 flow.dl_type = htons(ETH_TYPE_MPLS);
1122 flow_set_mpls_bos(&flow, n, 1);
1124 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1125 odp_flow_key_from_flow(&key, &flow, NULL, 0);
1127 error = dpif_flow_put(backer->dpif, DPIF_FP_CREATE,
1128 ofpbuf_data(&key), ofpbuf_size(&key), NULL, 0, NULL, 0, NULL);
1129 if (error && error != EEXIST) {
1130 if (error != EINVAL) {
1131 VLOG_WARN("%s: MPLS stack length feature probe failed (%s)",
1132 dpif_name(backer->dpif), ovs_strerror(error));
1137 error = dpif_flow_del(backer->dpif, ofpbuf_data(&key), ofpbuf_size(&key), NULL);
1139 VLOG_WARN("%s: failed to delete MPLS feature probe flow",
1140 dpif_name(backer->dpif));
1144 VLOG_INFO("%s: MPLS label stack length probed as %d",
1145 dpif_name(backer->dpif), n);
1150 construct(struct ofproto *ofproto_)
1152 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1153 struct shash_node *node, *next;
1156 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1161 ofproto->netflow = NULL;
1162 ofproto->sflow = NULL;
1163 ofproto->ipfix = NULL;
1164 ofproto->stp = NULL;
1165 ofproto->dump_seq = 0;
1166 hmap_init(&ofproto->bundles);
1167 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1168 ofproto->mbridge = mbridge_create();
1169 ofproto->has_bonded_bundles = false;
1170 ofproto->lacp_enabled = false;
1171 ovs_mutex_init_adaptive(&ofproto->stats_mutex);
1172 ovs_mutex_init(&ofproto->vsp_mutex);
1174 guarded_list_init(&ofproto->pins);
1176 ofproto_dpif_unixctl_init();
1178 hmap_init(&ofproto->vlandev_map);
1179 hmap_init(&ofproto->realdev_vid_map);
1181 sset_init(&ofproto->ports);
1182 sset_init(&ofproto->ghost_ports);
1183 sset_init(&ofproto->port_poll_set);
1184 ofproto->port_poll_errno = 0;
1185 ofproto->change_seq = 0;
1186 ofproto->pins_seq = seq_create();
1187 ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1190 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1191 struct iface_hint *iface_hint = node->data;
1193 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1194 /* Check if the datapath already has this port. */
1195 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1196 sset_add(&ofproto->ports, node->name);
1199 free(iface_hint->br_name);
1200 free(iface_hint->br_type);
1202 shash_delete(&init_ofp_ports, node);
1206 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1207 hash_string(ofproto->up.name, 0));
1208 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1210 ofproto_init_tables(ofproto_, N_TABLES);
1211 error = add_internal_flows(ofproto);
1213 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1219 add_internal_miss_flow(struct ofproto_dpif *ofproto, int id,
1220 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1226 match_init_catchall(&match);
1227 match_set_reg(&match, 0, id);
1229 error = ofproto_dpif_add_internal_flow(ofproto, &match, 0, ofpacts, &rule);
1230 *rulep = error ? NULL : rule_dpif_cast(rule);
1236 add_internal_flows(struct ofproto_dpif *ofproto)
1238 struct ofpact_controller *controller;
1239 uint64_t ofpacts_stub[128 / 8];
1240 struct ofpbuf ofpacts;
1241 struct rule *unused_rulep OVS_UNUSED;
1242 struct ofpact_resubmit *resubmit;
1247 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1250 controller = ofpact_put_CONTROLLER(&ofpacts);
1251 controller->max_len = UINT16_MAX;
1252 controller->controller_id = 0;
1253 controller->reason = OFPR_NO_MATCH;
1254 ofpact_pad(&ofpacts);
1256 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1257 &ofproto->miss_rule);
1262 ofpbuf_clear(&ofpacts);
1263 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1264 &ofproto->no_packet_in_rule);
1269 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1270 &ofproto->drop_frags_rule);
1275 /* Drop any run away non-recirc rule lookups. Recirc_id has to be
1276 * zero when reaching this rule.
1278 * (priority=2), recirc_id=0, actions=drop
1280 ofpbuf_clear(&ofpacts);
1281 match_init_catchall(&match);
1282 match_set_recirc_id(&match, 0);
1283 error = ofproto_dpif_add_internal_flow(ofproto, &match, 2, &ofpacts,
1289 /* Continue rule lookups for not-matched recirc rules from table 0.
1291 * (priority=1), actions=resubmit(, 0)
1293 resubmit = ofpact_put_RESUBMIT(&ofpacts);
1294 resubmit->ofpact.compat = 0;
1295 resubmit->in_port = OFPP_IN_PORT;
1296 resubmit->table_id = 0;
1298 match_init_catchall(&match);
1300 error = ofproto_dpif_add_internal_flow(ofproto, &match, 1, &ofpacts,
1307 destruct(struct ofproto *ofproto_)
1309 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1310 struct rule_dpif *rule, *next_rule;
1311 struct ofproto_packet_in *pin, *next_pin;
1312 struct oftable *table;
1315 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1316 fat_rwlock_wrlock(&xlate_rwlock);
1317 xlate_remove_ofproto(ofproto);
1318 fat_rwlock_unlock(&xlate_rwlock);
1320 /* Ensure that the upcall processing threads have no remaining references
1321 * to the ofproto or anything in it. */
1322 udpif_synchronize(ofproto->backer->udpif);
1324 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1326 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1327 struct cls_cursor cursor;
1329 fat_rwlock_rdlock(&table->cls.rwlock);
1330 cls_cursor_init(&cursor, &table->cls, NULL);
1331 fat_rwlock_unlock(&table->cls.rwlock);
1332 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1333 ofproto_rule_delete(&ofproto->up, &rule->up);
1337 guarded_list_pop_all(&ofproto->pins, &pins);
1338 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1339 list_remove(&pin->list_node);
1340 free(CONST_CAST(void *, pin->up.packet));
1343 guarded_list_destroy(&ofproto->pins);
1345 dpif_backer_recirc_clear_ofproto(ofproto->backer, ofproto);
1347 mbridge_unref(ofproto->mbridge);
1349 netflow_unref(ofproto->netflow);
1350 dpif_sflow_unref(ofproto->sflow);
1351 hmap_destroy(&ofproto->bundles);
1352 mac_learning_unref(ofproto->ml);
1354 hmap_destroy(&ofproto->vlandev_map);
1355 hmap_destroy(&ofproto->realdev_vid_map);
1357 sset_destroy(&ofproto->ports);
1358 sset_destroy(&ofproto->ghost_ports);
1359 sset_destroy(&ofproto->port_poll_set);
1361 ovs_mutex_destroy(&ofproto->stats_mutex);
1362 ovs_mutex_destroy(&ofproto->vsp_mutex);
1364 seq_destroy(ofproto->pins_seq);
1366 close_dpif_backer(ofproto->backer);
1370 run(struct ofproto *ofproto_)
1372 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1373 uint64_t new_seq, new_dump_seq;
1375 if (mbridge_need_revalidate(ofproto->mbridge)) {
1376 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1377 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1378 mac_learning_flush(ofproto->ml);
1379 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1382 /* Always updates the ofproto->pins_seqno to avoid frequent wakeup during
1383 * flow restore. Even though nothing is processed during flow restore,
1384 * all queued 'pins' will be handled immediately when flow restore
1386 ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1388 /* Do not perform any periodic activity required by 'ofproto' while
1389 * waiting for flow restore to complete. */
1390 if (!ofproto_get_flow_restore_wait()) {
1391 struct ofproto_packet_in *pin, *next_pin;
1394 guarded_list_pop_all(&ofproto->pins, &pins);
1395 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1396 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1397 list_remove(&pin->list_node);
1398 free(CONST_CAST(void *, pin->up.packet));
1403 if (ofproto->netflow) {
1404 netflow_run(ofproto->netflow);
1406 if (ofproto->sflow) {
1407 dpif_sflow_run(ofproto->sflow);
1409 if (ofproto->ipfix) {
1410 dpif_ipfix_run(ofproto->ipfix);
1413 new_seq = seq_read(connectivity_seq_get());
1414 if (ofproto->change_seq != new_seq) {
1415 struct ofport_dpif *ofport;
1417 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1421 ofproto->change_seq = new_seq;
1423 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1424 struct ofbundle *bundle;
1426 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1432 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1433 if (mac_learning_run(ofproto->ml)) {
1434 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1436 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1438 new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
1439 if (ofproto->dump_seq != new_dump_seq) {
1440 struct rule *rule, *next_rule;
1442 /* We know stats are relatively fresh, so now is a good time to do some
1444 ofproto->dump_seq = new_dump_seq;
1446 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1448 ovs_mutex_lock(&ofproto_mutex);
1449 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
1450 &ofproto->up.expirable) {
1451 rule_expire(rule_dpif_cast(rule));
1453 ovs_mutex_unlock(&ofproto_mutex);
1455 /* All outstanding data in existing flows has been accounted, so it's a
1456 * good time to do bond rebalancing. */
1457 if (ofproto->has_bonded_bundles) {
1458 struct ofbundle *bundle;
1460 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1462 bond_rebalance(bundle->bond);
1472 wait(struct ofproto *ofproto_)
1474 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1476 if (ofproto_get_flow_restore_wait()) {
1480 if (ofproto->sflow) {
1481 dpif_sflow_wait(ofproto->sflow);
1483 if (ofproto->ipfix) {
1484 dpif_ipfix_wait(ofproto->ipfix);
1486 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1487 struct ofbundle *bundle;
1489 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1490 bundle_wait(bundle);
1493 if (ofproto->netflow) {
1494 netflow_wait(ofproto->netflow);
1496 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1497 mac_learning_wait(ofproto->ml);
1498 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1500 if (ofproto->backer->need_revalidate) {
1501 /* Shouldn't happen, but if it does just go around again. */
1502 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1503 poll_immediate_wake();
1506 seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
1507 seq_wait(ofproto->pins_seq, ofproto->pins_seqno);
1511 type_get_memory_usage(const char *type, struct simap *usage)
1513 struct dpif_backer *backer;
1515 backer = shash_find_data(&all_dpif_backers, type);
1517 udpif_get_memory_usage(backer->udpif, usage);
1522 flush(struct ofproto *ofproto_)
1524 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1525 struct dpif_backer *backer = ofproto->backer;
1528 udpif_flush(backer->udpif);
1533 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1534 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1536 *arp_match_ip = true;
1537 *actions = (OFPUTIL_A_OUTPUT |
1538 OFPUTIL_A_SET_VLAN_VID |
1539 OFPUTIL_A_SET_VLAN_PCP |
1540 OFPUTIL_A_STRIP_VLAN |
1541 OFPUTIL_A_SET_DL_SRC |
1542 OFPUTIL_A_SET_DL_DST |
1543 OFPUTIL_A_SET_NW_SRC |
1544 OFPUTIL_A_SET_NW_DST |
1545 OFPUTIL_A_SET_NW_TOS |
1546 OFPUTIL_A_SET_TP_SRC |
1547 OFPUTIL_A_SET_TP_DST |
1552 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1554 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1555 struct dpif_dp_stats s;
1556 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1560 strcpy(ots->name, "classifier");
1562 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1563 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes, &used);
1564 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes,
1566 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes,
1568 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1569 ots->lookup_count = htonll(n_lookup);
1570 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1573 static struct ofport *
1576 struct ofport_dpif *port = xmalloc(sizeof *port);
1581 port_dealloc(struct ofport *port_)
1583 struct ofport_dpif *port = ofport_dpif_cast(port_);
1588 port_construct(struct ofport *port_)
1590 struct ofport_dpif *port = ofport_dpif_cast(port_);
1591 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1592 const struct netdev *netdev = port->up.netdev;
1593 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1594 struct dpif_port dpif_port;
1597 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1598 port->bundle = NULL;
1601 port->may_enable = true;
1602 port->stp_port = NULL;
1603 port->stp_state = STP_DISABLED;
1604 port->is_tunnel = false;
1608 port->realdev_ofp_port = 0;
1609 port->vlandev_vid = 0;
1610 port->carrier_seq = netdev_get_carrier_resets(netdev);
1611 port->is_layer3 = netdev_vport_is_layer3(netdev);
1613 if (netdev_vport_is_patch(netdev)) {
1614 /* By bailing out here, we don't submit the port to the sFlow module
1615 * to be considered for counter polling export. This is correct
1616 * because the patch port represents an interface that sFlow considers
1617 * to be "internal" to the switch as a whole, and therefore not an
1618 * candidate for counter polling. */
1619 port->odp_port = ODPP_NONE;
1620 ofport_update_peer(port);
1624 error = dpif_port_query_by_name(ofproto->backer->dpif,
1625 netdev_vport_get_dpif_port(netdev, namebuf,
1632 port->odp_port = dpif_port.port_no;
1634 if (netdev_get_tunnel_config(netdev)) {
1635 tnl_port_add(port, port->up.netdev, port->odp_port);
1636 port->is_tunnel = true;
1638 /* Sanity-check that a mapping doesn't already exist. This
1639 * shouldn't happen for non-tunnel ports. */
1640 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1641 VLOG_ERR("port %s already has an OpenFlow port number",
1643 dpif_port_destroy(&dpif_port);
1647 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1648 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1649 hash_odp_port(port->odp_port));
1650 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1652 dpif_port_destroy(&dpif_port);
1654 if (ofproto->sflow) {
1655 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1662 port_destruct(struct ofport *port_)
1664 struct ofport_dpif *port = ofport_dpif_cast(port_);
1665 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1666 const char *devname = netdev_get_name(port->up.netdev);
1667 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1668 const char *dp_port_name;
1670 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1671 fat_rwlock_wrlock(&xlate_rwlock);
1672 xlate_ofport_remove(port);
1673 fat_rwlock_unlock(&xlate_rwlock);
1675 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1677 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1678 /* The underlying device is still there, so delete it. This
1679 * happens when the ofproto is being destroyed, since the caller
1680 * assumes that removal of attached ports will happen as part of
1682 if (!port->is_tunnel) {
1683 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1688 port->peer->peer = NULL;
1692 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1693 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1694 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1695 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1699 sset_find_and_delete(&ofproto->ports, devname);
1700 sset_find_and_delete(&ofproto->ghost_ports, devname);
1701 bundle_remove(port_);
1702 set_cfm(port_, NULL);
1703 set_bfd(port_, NULL);
1704 if (port->stp_port) {
1705 stp_port_disable(port->stp_port);
1707 if (ofproto->sflow) {
1708 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1715 port_modified(struct ofport *port_)
1717 struct ofport_dpif *port = ofport_dpif_cast(port_);
1719 if (port->bundle && port->bundle->bond) {
1720 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1724 cfm_set_netdev(port->cfm, port->up.netdev);
1728 bfd_set_netdev(port->bfd, port->up.netdev);
1731 ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1732 port->up.pp.hw_addr);
1734 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1736 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1740 ofport_update_peer(port);
1744 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1746 struct ofport_dpif *port = ofport_dpif_cast(port_);
1747 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1748 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1750 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1751 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1752 OFPUTIL_PC_NO_PACKET_IN)) {
1753 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1755 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1756 bundle_update(port->bundle);
1762 set_sflow(struct ofproto *ofproto_,
1763 const struct ofproto_sflow_options *sflow_options)
1765 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1766 struct dpif_sflow *ds = ofproto->sflow;
1768 if (sflow_options) {
1769 uint32_t old_probability = ds ? dpif_sflow_get_probability(ds) : 0;
1771 struct ofport_dpif *ofport;
1773 ds = ofproto->sflow = dpif_sflow_create();
1774 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1775 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1778 dpif_sflow_set_options(ds, sflow_options);
1779 if (dpif_sflow_get_probability(ds) != old_probability) {
1780 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1784 dpif_sflow_unref(ds);
1785 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1786 ofproto->sflow = NULL;
1794 struct ofproto *ofproto_,
1795 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1796 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1797 size_t n_flow_exporters_options)
1799 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1800 struct dpif_ipfix *di = ofproto->ipfix;
1801 bool has_options = bridge_exporter_options || flow_exporters_options;
1803 if (has_options && !di) {
1804 di = ofproto->ipfix = dpif_ipfix_create();
1808 /* Call set_options in any case to cleanly flush the flow
1809 * caches in the last exporters that are to be destroyed. */
1810 dpif_ipfix_set_options(
1811 di, bridge_exporter_options, flow_exporters_options,
1812 n_flow_exporters_options);
1815 dpif_ipfix_unref(di);
1816 ofproto->ipfix = NULL;
1824 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1826 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1827 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1828 struct cfm *old = ofport->cfm;
1833 ofport->cfm = cfm_create(ofport->up.netdev);
1836 if (cfm_configure(ofport->cfm, s)) {
1843 cfm_unref(ofport->cfm);
1846 if (ofport->cfm != old) {
1847 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1849 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1850 ofport->up.pp.hw_addr);
1855 get_cfm_status(const struct ofport *ofport_, bool force,
1856 struct ofproto_cfm_status *status)
1858 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1862 if (cfm_check_status_change(ofport->cfm) || force) {
1863 status->faults = cfm_get_fault(ofport->cfm);
1864 status->flap_count = cfm_get_flap_count(ofport->cfm);
1865 status->remote_opstate = cfm_get_opup(ofport->cfm);
1866 status->health = cfm_get_health(ofport->cfm);
1867 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1869 ret = NO_STATUS_CHANGE;
1879 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1881 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1882 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1886 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1887 cfg, ofport->up.netdev);
1888 if (ofport->bfd != old) {
1889 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1891 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1892 ofport->up.pp.hw_addr);
1897 get_bfd_status(struct ofport *ofport_, bool force, struct smap *smap)
1899 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1903 if (bfd_check_status_change(ofport->bfd) || force) {
1904 bfd_get_status(ofport->bfd, smap);
1906 ret = NO_STATUS_CHANGE;
1915 /* Spanning Tree. */
1918 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1920 struct ofproto_dpif *ofproto = ofproto_;
1921 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1922 struct ofport_dpif *ofport;
1924 ofport = stp_port_get_aux(sp);
1926 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1927 ofproto->up.name, port_num);
1929 struct eth_header *eth = ofpbuf_l2(pkt);
1931 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1932 if (eth_addr_is_zero(eth->eth_src)) {
1933 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1934 "with unknown MAC", ofproto->up.name, port_num);
1936 ofproto_dpif_send_packet(ofport, pkt);
1942 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1944 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1946 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1948 /* Only revalidate flows if the configuration changed. */
1949 if (!s != !ofproto->stp) {
1950 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1954 if (!ofproto->stp) {
1955 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1956 send_bpdu_cb, ofproto);
1957 ofproto->stp_last_tick = time_msec();
1960 stp_set_bridge_id(ofproto->stp, s->system_id);
1961 stp_set_bridge_priority(ofproto->stp, s->priority);
1962 stp_set_hello_time(ofproto->stp, s->hello_time);
1963 stp_set_max_age(ofproto->stp, s->max_age);
1964 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1966 struct ofport *ofport;
1968 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1969 set_stp_port(ofport, NULL);
1972 stp_unref(ofproto->stp);
1973 ofproto->stp = NULL;
1980 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1982 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1986 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1987 s->designated_root = stp_get_designated_root(ofproto->stp);
1988 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1997 update_stp_port_state(struct ofport_dpif *ofport)
1999 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2000 enum stp_state state;
2002 /* Figure out new state. */
2003 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2007 if (ofport->stp_state != state) {
2008 enum ofputil_port_state of_state;
2011 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2012 netdev_get_name(ofport->up.netdev),
2013 stp_state_name(ofport->stp_state),
2014 stp_state_name(state));
2015 if (stp_learn_in_state(ofport->stp_state)
2016 != stp_learn_in_state(state)) {
2017 /* xxx Learning action flows should also be flushed. */
2018 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2019 mac_learning_flush(ofproto->ml);
2020 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2022 fwd_change = stp_forward_in_state(ofport->stp_state)
2023 != stp_forward_in_state(state);
2025 ofproto->backer->need_revalidate = REV_STP;
2026 ofport->stp_state = state;
2027 ofport->stp_state_entered = time_msec();
2029 if (fwd_change && ofport->bundle) {
2030 bundle_update(ofport->bundle);
2033 /* Update the STP state bits in the OpenFlow port description. */
2034 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2035 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2036 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2037 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2038 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2040 ofproto_port_set_state(&ofport->up, of_state);
2044 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2045 * caller is responsible for assigning STP port numbers and ensuring
2046 * there are no duplicates. */
2048 set_stp_port(struct ofport *ofport_,
2049 const struct ofproto_port_stp_settings *s)
2051 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2052 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2053 struct stp_port *sp = ofport->stp_port;
2055 if (!s || !s->enable) {
2057 ofport->stp_port = NULL;
2058 stp_port_disable(sp);
2059 update_stp_port_state(ofport);
2062 } else if (sp && stp_port_no(sp) != s->port_num
2063 && ofport == stp_port_get_aux(sp)) {
2064 /* The port-id changed, so disable the old one if it's not
2065 * already in use by another port. */
2066 stp_port_disable(sp);
2069 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2070 stp_port_enable(sp);
2072 stp_port_set_aux(sp, ofport);
2073 stp_port_set_priority(sp, s->priority);
2074 stp_port_set_path_cost(sp, s->path_cost);
2076 update_stp_port_state(ofport);
2082 get_stp_port_status(struct ofport *ofport_,
2083 struct ofproto_port_stp_status *s)
2085 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2086 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2087 struct stp_port *sp = ofport->stp_port;
2089 if (!ofproto->stp || !sp) {
2095 s->port_id = stp_port_get_id(sp);
2096 s->state = stp_port_get_state(sp);
2097 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2098 s->role = stp_port_get_role(sp);
2104 get_stp_port_stats(struct ofport *ofport_,
2105 struct ofproto_port_stp_stats *s)
2107 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2108 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2109 struct stp_port *sp = ofport->stp_port;
2111 if (!ofproto->stp || !sp) {
2117 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2123 stp_run(struct ofproto_dpif *ofproto)
2126 long long int now = time_msec();
2127 long long int elapsed = now - ofproto->stp_last_tick;
2128 struct stp_port *sp;
2131 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2132 ofproto->stp_last_tick = now;
2134 while (stp_get_changed_port(ofproto->stp, &sp)) {
2135 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2138 update_stp_port_state(ofport);
2142 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2143 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2144 mac_learning_flush(ofproto->ml);
2145 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2151 stp_wait(struct ofproto_dpif *ofproto)
2154 poll_timer_wait(1000);
2159 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2162 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2163 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2165 if (ofport->n_qdscp != n_qdscp
2166 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2167 n_qdscp * sizeof *qdscp))) {
2168 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2169 free(ofport->qdscp);
2170 ofport->qdscp = n_qdscp
2171 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2173 ofport->n_qdscp = n_qdscp;
2181 /* Expires all MAC learning entries associated with 'bundle' and forces its
2182 * ofproto to revalidate every flow.
2184 * Normally MAC learning entries are removed only from the ofproto associated
2185 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2186 * are removed from every ofproto. When patch ports and SLB bonds are in use
2187 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2188 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2189 * with the host from which it migrated. */
2191 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2193 struct ofproto_dpif *ofproto = bundle->ofproto;
2194 struct mac_learning *ml = ofproto->ml;
2195 struct mac_entry *mac, *next_mac;
2197 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2198 ovs_rwlock_wrlock(&ml->rwlock);
2199 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2200 if (mac_entry_get_port(ml, mac) == bundle) {
2202 struct ofproto_dpif *o;
2204 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2206 struct mac_entry *e;
2208 ovs_rwlock_wrlock(&o->ml->rwlock);
2209 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2211 mac_learning_expire(o->ml, e);
2213 ovs_rwlock_unlock(&o->ml->rwlock);
2218 mac_learning_expire(ml, mac);
2221 ovs_rwlock_unlock(&ml->rwlock);
2224 static struct ofbundle *
2225 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2227 struct ofbundle *bundle;
2229 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2230 &ofproto->bundles) {
2231 if (bundle->aux == aux) {
2239 bundle_update(struct ofbundle *bundle)
2241 struct ofport_dpif *port;
2243 bundle->floodable = true;
2244 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2245 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2247 || !stp_forward_in_state(port->stp_state)) {
2248 bundle->floodable = false;
2255 bundle_del_port(struct ofport_dpif *port)
2257 struct ofbundle *bundle = port->bundle;
2259 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2261 list_remove(&port->bundle_node);
2262 port->bundle = NULL;
2265 lacp_slave_unregister(bundle->lacp, port);
2268 bond_slave_unregister(bundle->bond, port);
2271 bundle_update(bundle);
2275 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2276 struct lacp_slave_settings *lacp)
2278 struct ofport_dpif *port;
2280 port = get_ofp_port(bundle->ofproto, ofp_port);
2285 if (port->bundle != bundle) {
2286 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2288 bundle_remove(&port->up);
2291 port->bundle = bundle;
2292 list_push_back(&bundle->ports, &port->bundle_node);
2293 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2295 || !stp_forward_in_state(port->stp_state)) {
2296 bundle->floodable = false;
2300 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2301 lacp_slave_register(bundle->lacp, port, lacp);
2308 bundle_destroy(struct ofbundle *bundle)
2310 struct ofproto_dpif *ofproto;
2311 struct ofport_dpif *port, *next_port;
2317 ofproto = bundle->ofproto;
2318 mbridge_unregister_bundle(ofproto->mbridge, bundle);
2320 fat_rwlock_wrlock(&xlate_rwlock);
2321 xlate_bundle_remove(bundle);
2322 fat_rwlock_unlock(&xlate_rwlock);
2324 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2325 bundle_del_port(port);
2328 bundle_flush_macs(bundle, true);
2329 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2331 free(bundle->trunks);
2332 lacp_unref(bundle->lacp);
2333 bond_unref(bundle->bond);
2338 bundle_set(struct ofproto *ofproto_, void *aux,
2339 const struct ofproto_bundle_settings *s)
2341 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2342 bool need_flush = false;
2343 struct ofport_dpif *port;
2344 struct ofbundle *bundle;
2345 unsigned long *trunks;
2351 bundle_destroy(bundle_lookup(ofproto, aux));
2355 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2356 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2358 bundle = bundle_lookup(ofproto, aux);
2360 bundle = xmalloc(sizeof *bundle);
2362 bundle->ofproto = ofproto;
2363 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2364 hash_pointer(aux, 0));
2366 bundle->name = NULL;
2368 list_init(&bundle->ports);
2369 bundle->vlan_mode = PORT_VLAN_TRUNK;
2371 bundle->trunks = NULL;
2372 bundle->use_priority_tags = s->use_priority_tags;
2373 bundle->lacp = NULL;
2374 bundle->bond = NULL;
2376 bundle->floodable = true;
2377 mbridge_register_bundle(ofproto->mbridge, bundle);
2380 if (!bundle->name || strcmp(s->name, bundle->name)) {
2382 bundle->name = xstrdup(s->name);
2387 ofproto->lacp_enabled = true;
2388 if (!bundle->lacp) {
2389 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2390 bundle->lacp = lacp_create();
2392 lacp_configure(bundle->lacp, s->lacp);
2394 lacp_unref(bundle->lacp);
2395 bundle->lacp = NULL;
2398 /* Update set of ports. */
2400 for (i = 0; i < s->n_slaves; i++) {
2401 if (!bundle_add_port(bundle, s->slaves[i],
2402 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2406 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2407 struct ofport_dpif *next_port;
2409 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2410 for (i = 0; i < s->n_slaves; i++) {
2411 if (s->slaves[i] == port->up.ofp_port) {
2416 bundle_del_port(port);
2420 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2422 if (list_is_empty(&bundle->ports)) {
2423 bundle_destroy(bundle);
2427 /* Set VLAN tagging mode */
2428 if (s->vlan_mode != bundle->vlan_mode
2429 || s->use_priority_tags != bundle->use_priority_tags) {
2430 bundle->vlan_mode = s->vlan_mode;
2431 bundle->use_priority_tags = s->use_priority_tags;
2436 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2437 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2439 if (vlan != bundle->vlan) {
2440 bundle->vlan = vlan;
2444 /* Get trunked VLANs. */
2445 switch (s->vlan_mode) {
2446 case PORT_VLAN_ACCESS:
2450 case PORT_VLAN_TRUNK:
2451 trunks = CONST_CAST(unsigned long *, s->trunks);
2454 case PORT_VLAN_NATIVE_UNTAGGED:
2455 case PORT_VLAN_NATIVE_TAGGED:
2456 if (vlan != 0 && (!s->trunks
2457 || !bitmap_is_set(s->trunks, vlan)
2458 || bitmap_is_set(s->trunks, 0))) {
2459 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2461 trunks = bitmap_clone(s->trunks, 4096);
2463 trunks = bitmap_allocate1(4096);
2465 bitmap_set1(trunks, vlan);
2466 bitmap_set0(trunks, 0);
2468 trunks = CONST_CAST(unsigned long *, s->trunks);
2475 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2476 free(bundle->trunks);
2477 if (trunks == s->trunks) {
2478 bundle->trunks = vlan_bitmap_clone(trunks);
2480 bundle->trunks = trunks;
2485 if (trunks != s->trunks) {
2490 if (!list_is_short(&bundle->ports)) {
2491 bundle->ofproto->has_bonded_bundles = true;
2493 if (bond_reconfigure(bundle->bond, s->bond)) {
2494 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2497 bundle->bond = bond_create(s->bond, ofproto);
2498 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2501 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2502 bond_slave_register(bundle->bond, port,
2503 port->up.ofp_port, port->up.netdev);
2506 bond_unref(bundle->bond);
2507 bundle->bond = NULL;
2510 /* If we changed something that would affect MAC learning, un-learn
2511 * everything on this port and force flow revalidation. */
2513 bundle_flush_macs(bundle, false);
2520 bundle_remove(struct ofport *port_)
2522 struct ofport_dpif *port = ofport_dpif_cast(port_);
2523 struct ofbundle *bundle = port->bundle;
2526 bundle_del_port(port);
2527 if (list_is_empty(&bundle->ports)) {
2528 bundle_destroy(bundle);
2529 } else if (list_is_short(&bundle->ports)) {
2530 bond_unref(bundle->bond);
2531 bundle->bond = NULL;
2537 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2539 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2540 struct ofport_dpif *port = port_;
2541 uint8_t ea[ETH_ADDR_LEN];
2544 error = netdev_get_etheraddr(port->up.netdev, ea);
2546 struct ofpbuf packet;
2549 ofpbuf_init(&packet, 0);
2550 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2552 memcpy(packet_pdu, pdu, pdu_size);
2554 ofproto_dpif_send_packet(port, &packet);
2555 ofpbuf_uninit(&packet);
2557 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2558 "%s (%s)", port->bundle->name,
2559 netdev_get_name(port->up.netdev), ovs_strerror(error));
2564 bundle_send_learning_packets(struct ofbundle *bundle)
2566 struct ofproto_dpif *ofproto = bundle->ofproto;
2567 struct ofpbuf *learning_packet;
2568 int error, n_packets, n_errors;
2569 struct mac_entry *e;
2570 struct list packets;
2572 list_init(&packets);
2573 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2574 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2575 if (mac_entry_get_port(ofproto->ml, e) != bundle) {
2578 learning_packet = bond_compose_learning_packet(bundle->bond,
2581 /* Temporarily use 'frame' as a private pointer (see below). */
2582 ovs_assert(learning_packet->frame == ofpbuf_data(learning_packet));
2583 learning_packet->frame = port_void;
2584 list_push_back(&packets, &learning_packet->list_node);
2587 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2589 error = n_packets = n_errors = 0;
2590 LIST_FOR_EACH (learning_packet, list_node, &packets) {
2592 void *port_void = learning_packet->frame;
2594 /* Restore 'frame'. */
2595 learning_packet->frame = ofpbuf_data(learning_packet);
2596 ret = ofproto_dpif_send_packet(port_void, learning_packet);
2603 ofpbuf_list_delete(&packets);
2606 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2607 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2608 "packets, last error was: %s",
2609 bundle->name, n_errors, n_packets, ovs_strerror(error));
2611 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2612 bundle->name, n_packets);
2617 bundle_run(struct ofbundle *bundle)
2620 lacp_run(bundle->lacp, send_pdu_cb);
2623 struct ofport_dpif *port;
2625 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2626 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2629 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2630 bundle->ofproto->backer->need_revalidate = REV_BOND;
2633 if (bond_should_send_learning_packets(bundle->bond)) {
2634 bundle_send_learning_packets(bundle);
2640 bundle_wait(struct ofbundle *bundle)
2643 lacp_wait(bundle->lacp);
2646 bond_wait(bundle->bond);
2653 mirror_set__(struct ofproto *ofproto_, void *aux,
2654 const struct ofproto_mirror_settings *s)
2656 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2657 struct ofbundle **srcs, **dsts;
2662 mirror_destroy(ofproto->mbridge, aux);
2666 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2667 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2669 for (i = 0; i < s->n_srcs; i++) {
2670 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2673 for (i = 0; i < s->n_dsts; i++) {
2674 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2677 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2678 s->n_dsts, s->src_vlans,
2679 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2686 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2687 uint64_t *packets, uint64_t *bytes)
2689 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2694 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2696 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2697 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2698 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2699 mac_learning_flush(ofproto->ml);
2701 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2706 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2708 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2709 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2710 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2714 forward_bpdu_changed(struct ofproto *ofproto_)
2716 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2717 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2721 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2724 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2725 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2726 mac_learning_set_idle_time(ofproto->ml, idle_time);
2727 mac_learning_set_max_entries(ofproto->ml, max_entries);
2728 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2733 static struct ofport_dpif *
2734 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2736 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2737 return ofport ? ofport_dpif_cast(ofport) : NULL;
2741 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2742 struct ofproto_port *ofproto_port,
2743 struct dpif_port *dpif_port)
2745 ofproto_port->name = dpif_port->name;
2746 ofproto_port->type = dpif_port->type;
2747 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2751 ofport_update_peer(struct ofport_dpif *ofport)
2753 const struct ofproto_dpif *ofproto;
2754 struct dpif_backer *backer;
2757 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2761 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2762 backer->need_revalidate = REV_RECONFIGURE;
2765 ofport->peer->peer = NULL;
2766 ofport->peer = NULL;
2769 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2774 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2775 struct ofport *peer_ofport;
2776 struct ofport_dpif *peer;
2779 if (ofproto->backer != backer) {
2783 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2788 peer = ofport_dpif_cast(peer_ofport);
2789 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2790 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2792 ofport->peer = peer;
2793 ofport->peer->peer = ofport;
2803 port_run(struct ofport_dpif *ofport)
2805 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2806 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2807 bool enable = netdev_get_carrier(ofport->up.netdev);
2808 bool cfm_enable = false;
2809 bool bfd_enable = false;
2811 ofport->carrier_seq = carrier_seq;
2814 int cfm_opup = cfm_get_opup(ofport->cfm);
2816 cfm_enable = !cfm_get_fault(ofport->cfm);
2818 if (cfm_opup >= 0) {
2819 cfm_enable = cfm_enable && cfm_opup;
2824 bfd_enable = bfd_forwarding(ofport->bfd);
2827 if (ofport->bfd || ofport->cfm) {
2828 enable = enable && (cfm_enable || bfd_enable);
2831 if (ofport->bundle) {
2832 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2833 if (carrier_changed) {
2834 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2838 if (ofport->may_enable != enable) {
2839 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2840 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2843 ofport->may_enable = enable;
2847 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2848 struct ofproto_port *ofproto_port)
2850 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2851 struct dpif_port dpif_port;
2854 if (sset_contains(&ofproto->ghost_ports, devname)) {
2855 const char *type = netdev_get_type_from_name(devname);
2857 /* We may be called before ofproto->up.port_by_name is populated with
2858 * the appropriate ofport. For this reason, we must get the name and
2859 * type from the netdev layer directly. */
2861 const struct ofport *ofport;
2863 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2864 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2865 ofproto_port->name = xstrdup(devname);
2866 ofproto_port->type = xstrdup(type);
2872 if (!sset_contains(&ofproto->ports, devname)) {
2875 error = dpif_port_query_by_name(ofproto->backer->dpif,
2876 devname, &dpif_port);
2878 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2884 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2886 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2887 const char *devname = netdev_get_name(netdev);
2888 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2889 const char *dp_port_name;
2891 if (netdev_vport_is_patch(netdev)) {
2892 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2896 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2897 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2898 odp_port_t port_no = ODPP_NONE;
2901 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2905 if (netdev_get_tunnel_config(netdev)) {
2906 simap_put(&ofproto->backer->tnl_backers,
2907 dp_port_name, odp_to_u32(port_no));
2911 if (netdev_get_tunnel_config(netdev)) {
2912 sset_add(&ofproto->ghost_ports, devname);
2914 sset_add(&ofproto->ports, devname);
2920 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
2922 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2923 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2930 sset_find_and_delete(&ofproto->ghost_ports,
2931 netdev_get_name(ofport->up.netdev));
2932 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2933 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
2934 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
2936 /* The caller is going to close ofport->up.netdev. If this is a
2937 * bonded port, then the bond is using that netdev, so remove it
2938 * from the bond. The client will need to reconfigure everything
2939 * after deleting ports, so then the slave will get re-added. */
2940 bundle_remove(&ofport->up);
2947 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2949 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2952 error = netdev_get_stats(ofport->up.netdev, stats);
2954 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
2955 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2957 ovs_mutex_lock(&ofproto->stats_mutex);
2958 /* ofproto->stats.tx_packets represents packets that we created
2959 * internally and sent to some port (e.g. packets sent with
2960 * ofproto_dpif_send_packet()). Account for them as if they had
2961 * come from OFPP_LOCAL and got forwarded. */
2963 if (stats->rx_packets != UINT64_MAX) {
2964 stats->rx_packets += ofproto->stats.tx_packets;
2967 if (stats->rx_bytes != UINT64_MAX) {
2968 stats->rx_bytes += ofproto->stats.tx_bytes;
2971 /* ofproto->stats.rx_packets represents packets that were received on
2972 * some port and we processed internally and dropped (e.g. STP).
2973 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2975 if (stats->tx_packets != UINT64_MAX) {
2976 stats->tx_packets += ofproto->stats.rx_packets;
2979 if (stats->tx_bytes != UINT64_MAX) {
2980 stats->tx_bytes += ofproto->stats.rx_bytes;
2982 ovs_mutex_unlock(&ofproto->stats_mutex);
2988 struct port_dump_state {
2993 struct ofproto_port port;
2998 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3000 *statep = xzalloc(sizeof(struct port_dump_state));
3005 port_dump_next(const struct ofproto *ofproto_, void *state_,
3006 struct ofproto_port *port)
3008 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3009 struct port_dump_state *state = state_;
3010 const struct sset *sset;
3011 struct sset_node *node;
3013 if (state->has_port) {
3014 ofproto_port_destroy(&state->port);
3015 state->has_port = false;
3017 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3018 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3021 error = port_query_by_name(ofproto_, node->name, &state->port);
3023 *port = state->port;
3024 state->has_port = true;
3026 } else if (error != ENODEV) {
3031 if (!state->ghost) {
3032 state->ghost = true;
3035 return port_dump_next(ofproto_, state_, port);
3042 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3044 struct port_dump_state *state = state_;
3046 if (state->has_port) {
3047 ofproto_port_destroy(&state->port);
3054 port_poll(const struct ofproto *ofproto_, char **devnamep)
3056 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3058 if (ofproto->port_poll_errno) {
3059 int error = ofproto->port_poll_errno;
3060 ofproto->port_poll_errno = 0;
3064 if (sset_is_empty(&ofproto->port_poll_set)) {
3068 *devnamep = sset_pop(&ofproto->port_poll_set);
3073 port_poll_wait(const struct ofproto *ofproto_)
3075 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3076 dpif_port_poll_wait(ofproto->backer->dpif);
3080 port_is_lacp_current(const struct ofport *ofport_)
3082 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3083 return (ofport->bundle && ofport->bundle->lacp
3084 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3088 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3089 * then delete it entirely. */
3091 rule_expire(struct rule_dpif *rule)
3092 OVS_REQUIRES(ofproto_mutex)
3094 uint16_t hard_timeout, idle_timeout;
3095 long long int now = time_msec();
3098 ovs_assert(!rule->up.pending);
3100 hard_timeout = rule->up.hard_timeout;
3101 idle_timeout = rule->up.idle_timeout;
3103 /* Has 'rule' expired? */
3105 long long int modified;
3107 ovs_mutex_lock(&rule->up.mutex);
3108 modified = rule->up.modified;
3109 ovs_mutex_unlock(&rule->up.mutex);
3111 if (now > modified + hard_timeout * 1000) {
3112 reason = OFPRR_HARD_TIMEOUT;
3116 if (reason < 0 && idle_timeout) {
3119 ovs_mutex_lock(&rule->stats_mutex);
3120 used = rule->stats.used;
3121 ovs_mutex_unlock(&rule->stats_mutex);
3123 if (now > used + idle_timeout * 1000) {
3124 reason = OFPRR_IDLE_TIMEOUT;
3129 COVERAGE_INC(ofproto_dpif_expired);
3130 ofproto_rule_expire(&rule->up, reason);
3134 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3135 * 'flow' must reflect the data in 'packet'. */
3137 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3138 const struct flow *flow,
3139 struct rule_dpif *rule,
3140 const struct ofpact *ofpacts, size_t ofpacts_len,
3141 struct ofpbuf *packet)
3143 struct dpif_flow_stats stats;
3144 struct xlate_out xout;
3145 struct xlate_in xin;
3147 struct dpif_execute execute;
3150 ovs_assert((rule != NULL) != (ofpacts != NULL));
3152 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3155 rule_dpif_credit_stats(rule, &stats);
3158 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
3159 xin.ofpacts = ofpacts;
3160 xin.ofpacts_len = ofpacts_len;
3161 xin.resubmit_stats = &stats;
3162 xlate_actions(&xin, &xout);
3164 execute.actions = ofpbuf_data(&xout.odp_actions);
3165 execute.actions_len = ofpbuf_size(&xout.odp_actions);
3166 execute.packet = packet;
3167 execute.md = pkt_metadata_from_flow(flow);
3168 execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
3170 /* Fix up in_port. */
3171 in_port = flow->in_port.ofp_port;
3172 if (in_port == OFPP_NONE) {
3173 in_port = OFPP_LOCAL;
3175 execute.md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
3177 error = dpif_execute(ofproto->backer->dpif, &execute);
3179 xlate_out_uninit(&xout);
3185 rule_dpif_credit_stats(struct rule_dpif *rule,
3186 const struct dpif_flow_stats *stats)
3188 ovs_mutex_lock(&rule->stats_mutex);
3189 rule->stats.n_packets += stats->n_packets;
3190 rule->stats.n_bytes += stats->n_bytes;
3191 rule->stats.used = MAX(rule->stats.used, stats->used);
3192 ovs_mutex_unlock(&rule->stats_mutex);
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 const 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;
3245 table_id = rule_dpif_lookup_get_init_table_id(flow);
3250 verdict = rule_dpif_lookup_from_table(ofproto, flow, wc, true,
3251 &table_id, rule, take_ref);
3254 case RULE_DPIF_LOOKUP_VERDICT_MATCH:
3256 case RULE_DPIF_LOOKUP_VERDICT_CONTROLLER: {
3257 struct ofport_dpif *port;
3259 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
3261 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
3262 flow->in_port.ofp_port);
3264 config = port ? port->up.pp.config : 0;
3267 case RULE_DPIF_LOOKUP_VERDICT_DROP:
3268 config = OFPUTIL_PC_NO_PACKET_IN;
3270 case RULE_DPIF_LOOKUP_VERDICT_DEFAULT:
3271 if (!connmgr_wants_packet_in_on_miss(ofproto->up.connmgr)) {
3272 config = OFPUTIL_PC_NO_PACKET_IN;
3279 choose_miss_rule(config, ofproto->miss_rule,
3280 ofproto->no_packet_in_rule, rule, take_ref);
3284 /* The returned rule is valid at least until the next RCU quiescent period.
3285 * If the '*rule' needs to stay around longer, a non-zero 'take_ref' must be
3286 * passed in to cause a reference to be taken on it before this returns. */
3287 static struct rule_dpif *
3288 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, uint8_t table_id,
3289 const struct flow *flow, struct flow_wildcards *wc,
3292 struct classifier *cls = &ofproto->up.tables[table_id].cls;
3293 const struct cls_rule *cls_rule;
3294 struct rule_dpif *rule;
3296 fat_rwlock_rdlock(&cls->rwlock);
3297 if (ofproto->up.frag_handling != OFPC_FRAG_NX_MATCH) {
3299 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
3300 if (is_ip_any(flow)) {
3301 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
3305 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
3306 if (ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3307 /* We must pretend that transport ports are unavailable. */
3308 struct flow ofpc_normal_flow = *flow;
3309 ofpc_normal_flow.tp_src = htons(0);
3310 ofpc_normal_flow.tp_dst = htons(0);
3311 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
3313 /* Must be OFPC_FRAG_DROP (we don't have OFPC_FRAG_REASM). */
3314 cls_rule = &ofproto->drop_frags_rule->up.cr;
3317 cls_rule = classifier_lookup(cls, flow, wc);
3320 cls_rule = classifier_lookup(cls, flow, wc);
3323 rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
3325 rule_dpif_ref(rule);
3327 fat_rwlock_unlock(&cls->rwlock);
3332 /* Look up 'flow' in 'ofproto''s classifier starting from table '*table_id'.
3333 * Stores the rule that was found in '*rule', or NULL if none was found.
3334 * Updates 'wc', if nonnull, to reflect the fields that were used during the
3337 * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3338 * if none is found then the table miss configuration for that table is
3339 * honored, which can result in additional lookups in other OpenFlow tables.
3340 * In this case the function updates '*table_id' to reflect the final OpenFlow
3341 * table that was searched.
3343 * If 'honor_table_miss' is false, then only one table lookup occurs, in
3348 * - RULE_DPIF_LOOKUP_VERDICT_MATCH if a rule (in '*rule') was found.
3350 * - RULE_OFPTC_TABLE_MISS_CONTROLLER if no rule was found and either:
3351 * + 'honor_table_miss' is false
3352 * + a table miss configuration specified that the packet should be
3353 * sent to the controller in this case.
3355 * - RULE_DPIF_LOOKUP_VERDICT_DROP if no rule was found, 'honor_table_miss'
3356 * is true and a table miss configuration specified that the packet
3357 * should be dropped in this case.
3359 * - RULE_DPIF_LOOKUP_VERDICT_DEFAULT if no rule was found,
3360 * 'honor_table_miss' is true and a table miss configuration has
3361 * not been specified in this case.
3363 * The rule is returned in '*rule', which is valid at least until the next
3364 * RCU quiescent period. If the '*rule' needs to stay around longer,
3365 * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3366 * on it before this returns. */
3367 enum rule_dpif_lookup_verdict
3368 rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto,
3369 const struct flow *flow,
3370 struct flow_wildcards *wc,
3371 bool honor_table_miss,
3372 uint8_t *table_id, struct rule_dpif **rule,
3377 for (next_id = *table_id;
3378 next_id < ofproto->up.n_tables;
3379 next_id++, next_id += (next_id == TBL_INTERNAL))
3381 *table_id = next_id;
3382 *rule = rule_dpif_lookup_in_table(ofproto, *table_id, flow, wc,
3385 return RULE_DPIF_LOOKUP_VERDICT_MATCH;
3386 } else if (!honor_table_miss) {
3387 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3389 switch (ofproto_table_get_config(&ofproto->up, *table_id)) {
3390 case OFPROTO_TABLE_MISS_CONTINUE:
3393 case OFPROTO_TABLE_MISS_CONTROLLER:
3394 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3396 case OFPROTO_TABLE_MISS_DROP:
3397 return RULE_DPIF_LOOKUP_VERDICT_DROP;
3399 case OFPROTO_TABLE_MISS_DEFAULT:
3400 return RULE_DPIF_LOOKUP_VERDICT_DEFAULT;
3405 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3408 /* Given a port configuration (specified as zero if there's no port), chooses
3409 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
3412 * The rule is returned in '*rule', which is valid at least until the next
3413 * RCU quiescent period. If the '*rule' needs to stay around longer,
3414 * a reference must be taken on it (rule_dpif_ref()).
3417 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
3418 struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule,
3421 *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
3423 rule_dpif_ref(*rule);
3428 complete_operation(struct rule_dpif *rule)
3429 OVS_REQUIRES(ofproto_mutex)
3431 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3433 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3434 ofoperation_complete(rule->up.pending, 0);
3437 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
3439 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
3442 static struct rule *
3445 struct rule_dpif *rule = xmalloc(sizeof *rule);
3450 rule_dealloc(struct rule *rule_)
3452 struct rule_dpif *rule = rule_dpif_cast(rule_);
3457 rule_construct(struct rule *rule_)
3458 OVS_NO_THREAD_SAFETY_ANALYSIS
3460 struct rule_dpif *rule = rule_dpif_cast(rule_);
3461 ovs_mutex_init_adaptive(&rule->stats_mutex);
3462 rule->stats.n_packets = 0;
3463 rule->stats.n_bytes = 0;
3464 rule->stats.used = rule->up.modified;
3469 rule_insert(struct rule *rule_)
3470 OVS_REQUIRES(ofproto_mutex)
3472 struct rule_dpif *rule = rule_dpif_cast(rule_);
3473 complete_operation(rule);
3477 rule_delete(struct rule *rule_)
3478 OVS_REQUIRES(ofproto_mutex)
3480 struct rule_dpif *rule = rule_dpif_cast(rule_);
3481 complete_operation(rule);
3485 rule_destruct(struct rule *rule_)
3487 struct rule_dpif *rule = rule_dpif_cast(rule_);
3488 ovs_mutex_destroy(&rule->stats_mutex);
3492 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
3493 long long int *used)
3495 struct rule_dpif *rule = rule_dpif_cast(rule_);
3497 ovs_mutex_lock(&rule->stats_mutex);
3498 *packets = rule->stats.n_packets;
3499 *bytes = rule->stats.n_bytes;
3500 *used = rule->stats.used;
3501 ovs_mutex_unlock(&rule->stats_mutex);
3505 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
3506 struct ofpbuf *packet)
3508 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3510 ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
3514 rule_execute(struct rule *rule, const struct flow *flow,
3515 struct ofpbuf *packet)
3517 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
3518 ofpbuf_delete(packet);
3523 rule_modify_actions(struct rule *rule_, bool reset_counters)
3524 OVS_REQUIRES(ofproto_mutex)
3526 struct rule_dpif *rule = rule_dpif_cast(rule_);
3528 if (reset_counters) {
3529 ovs_mutex_lock(&rule->stats_mutex);
3530 rule->stats.n_packets = 0;
3531 rule->stats.n_bytes = 0;
3532 ovs_mutex_unlock(&rule->stats_mutex);
3535 complete_operation(rule);
3538 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
3540 return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
3543 static struct ofgroup *
3546 struct group_dpif *group = xzalloc(sizeof *group);
3551 group_dealloc(struct ofgroup *group_)
3553 struct group_dpif *group = group_dpif_cast(group_);
3558 group_construct_stats(struct group_dpif *group)
3559 OVS_REQUIRES(group->stats_mutex)
3561 group->packet_count = 0;
3562 group->byte_count = 0;
3563 if (!group->bucket_stats) {
3564 group->bucket_stats = xcalloc(group->up.n_buckets,
3565 sizeof *group->bucket_stats);
3567 memset(group->bucket_stats, 0, group->up.n_buckets *
3568 sizeof *group->bucket_stats);
3573 group_construct(struct ofgroup *group_)
3575 struct group_dpif *group = group_dpif_cast(group_);
3576 const struct ofputil_bucket *bucket;
3578 /* Prevent group chaining because our locking structure makes it hard to
3579 * implement deadlock-free. (See xlate_group_resource_check().) */
3580 LIST_FOR_EACH (bucket, list_node, &group->up.buckets) {
3581 const struct ofpact *a;
3583 OFPACT_FOR_EACH (a, bucket->ofpacts, bucket->ofpacts_len) {
3584 if (a->type == OFPACT_GROUP) {
3585 return OFPERR_OFPGMFC_CHAINING_UNSUPPORTED;
3590 ovs_mutex_init_adaptive(&group->stats_mutex);
3591 ovs_mutex_lock(&group->stats_mutex);
3592 group_construct_stats(group);
3593 ovs_mutex_unlock(&group->stats_mutex);
3598 group_destruct__(struct group_dpif *group)
3599 OVS_REQUIRES(group->stats_mutex)
3601 free(group->bucket_stats);
3602 group->bucket_stats = NULL;
3606 group_destruct(struct ofgroup *group_)
3608 struct group_dpif *group = group_dpif_cast(group_);
3609 ovs_mutex_lock(&group->stats_mutex);
3610 group_destruct__(group);
3611 ovs_mutex_unlock(&group->stats_mutex);
3612 ovs_mutex_destroy(&group->stats_mutex);
3616 group_modify(struct ofgroup *group_, struct ofgroup *victim_)
3618 struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
3619 struct group_dpif *group = group_dpif_cast(group_);
3620 struct group_dpif *victim = group_dpif_cast(victim_);
3622 ovs_mutex_lock(&group->stats_mutex);
3623 if (victim->up.n_buckets < group->up.n_buckets) {
3624 group_destruct__(group);
3626 group_construct_stats(group);
3627 ovs_mutex_unlock(&group->stats_mutex);
3629 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3635 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
3637 struct group_dpif *group = group_dpif_cast(group_);
3639 ovs_mutex_lock(&group->stats_mutex);
3640 ogs->packet_count = group->packet_count;
3641 ogs->byte_count = group->byte_count;
3642 memcpy(ogs->bucket_stats, group->bucket_stats,
3643 group->up.n_buckets * sizeof *group->bucket_stats);
3644 ovs_mutex_unlock(&group->stats_mutex);
3650 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
3651 struct group_dpif **group)
3652 OVS_TRY_RDLOCK(true, (*group)->up.rwlock)
3654 struct ofgroup *ofgroup;
3658 found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
3659 *group = found ? group_dpif_cast(ofgroup) : NULL;
3665 group_dpif_release(struct group_dpif *group)
3666 OVS_RELEASES(group->up.rwlock)
3668 ofproto_group_release(&group->up);
3672 group_dpif_get_buckets(const struct group_dpif *group,
3673 const struct list **buckets)
3675 *buckets = &group->up.buckets;
3678 enum ofp11_group_type
3679 group_dpif_get_type(const struct group_dpif *group)
3681 return group->up.type;
3684 /* Sends 'packet' out 'ofport'.
3685 * May modify 'packet'.
3686 * Returns 0 if successful, otherwise a positive errno value. */
3688 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
3690 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3693 error = xlate_send_packet(ofport, packet);
3695 ovs_mutex_lock(&ofproto->stats_mutex);
3696 ofproto->stats.tx_packets++;
3697 ofproto->stats.tx_bytes += ofpbuf_size(packet);
3698 ovs_mutex_unlock(&ofproto->stats_mutex);
3703 set_frag_handling(struct ofproto *ofproto_,
3704 enum ofp_config_flags frag_handling)
3706 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3707 if (frag_handling != OFPC_FRAG_REASM) {
3708 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3716 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
3717 const struct flow *flow,
3718 const struct ofpact *ofpacts, size_t ofpacts_len)
3720 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3722 ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
3723 ofpacts_len, packet);
3730 set_netflow(struct ofproto *ofproto_,
3731 const struct netflow_options *netflow_options)
3733 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3735 if (netflow_options) {
3736 if (!ofproto->netflow) {
3737 ofproto->netflow = netflow_create();
3738 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3740 return netflow_set_options(ofproto->netflow, netflow_options);
3741 } else if (ofproto->netflow) {
3742 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3743 netflow_unref(ofproto->netflow);
3744 ofproto->netflow = NULL;
3751 get_netflow_ids(const struct ofproto *ofproto_,
3752 uint8_t *engine_type, uint8_t *engine_id)
3754 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3756 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
3759 static struct ofproto_dpif *
3760 ofproto_dpif_lookup(const char *name)
3762 struct ofproto_dpif *ofproto;
3764 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
3765 hash_string(name, 0), &all_ofproto_dpifs) {
3766 if (!strcmp(ofproto->up.name, name)) {
3774 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
3775 const char *argv[], void *aux OVS_UNUSED)
3777 struct ofproto_dpif *ofproto;
3780 ofproto = ofproto_dpif_lookup(argv[1]);
3782 unixctl_command_reply_error(conn, "no such bridge");
3785 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3786 mac_learning_flush(ofproto->ml);
3787 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3789 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3790 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3791 mac_learning_flush(ofproto->ml);
3792 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3796 unixctl_command_reply(conn, "table successfully flushed");
3799 static struct ofport_dpif *
3800 ofbundle_get_a_port(const struct ofbundle *bundle)
3802 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
3807 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
3808 const char *argv[], void *aux OVS_UNUSED)
3810 struct ds ds = DS_EMPTY_INITIALIZER;
3811 const struct ofproto_dpif *ofproto;
3812 const struct mac_entry *e;
3814 ofproto = ofproto_dpif_lookup(argv[1]);
3816 unixctl_command_reply_error(conn, "no such bridge");
3820 ds_put_cstr(&ds, " port VLAN MAC Age\n");
3821 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
3822 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3823 struct ofbundle *bundle = mac_entry_get_port(ofproto->ml, e);
3824 char name[OFP_MAX_PORT_NAME_LEN];
3826 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
3828 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
3829 name, e->vlan, ETH_ADDR_ARGS(e->mac),
3830 mac_entry_age(ofproto->ml, e));
3832 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3833 unixctl_command_reply(conn, ds_cstr(&ds));
3838 struct xlate_out xout;
3839 struct xlate_in xin;
3840 const struct flow *key;
3842 struct flow_wildcards wc;
3847 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
3849 const struct rule_actions *actions;
3852 ds_put_char_multiple(result, '\t', level);
3854 ds_put_cstr(result, "No match\n");
3858 ovs_mutex_lock(&rule->up.mutex);
3859 cookie = rule->up.flow_cookie;
3860 ovs_mutex_unlock(&rule->up.mutex);
3862 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
3863 rule ? rule->up.table_id : 0, ntohll(cookie));
3864 cls_rule_format(&rule->up.cr, result);
3865 ds_put_char(result, '\n');
3867 actions = rule_dpif_get_actions(rule);
3869 ds_put_char_multiple(result, '\t', level);
3870 ds_put_cstr(result, "OpenFlow actions=");
3871 ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
3872 ds_put_char(result, '\n');
3876 trace_format_flow(struct ds *result, int level, const char *title,
3877 struct trace_ctx *trace)
3879 ds_put_char_multiple(result, '\t', level);
3880 ds_put_format(result, "%s: ", title);
3881 /* Do not report unchanged flows for resubmits. */
3882 if ((level > 0 && flow_equal(&trace->xin.flow, &trace->flow))
3883 || (level == 0 && flow_equal(&trace->xin.flow, trace->key))) {
3884 ds_put_cstr(result, "unchanged");
3886 flow_format(result, &trace->xin.flow);
3887 trace->flow = trace->xin.flow;
3889 ds_put_char(result, '\n');
3893 trace_format_regs(struct ds *result, int level, const char *title,
3894 struct trace_ctx *trace)
3898 ds_put_char_multiple(result, '\t', level);
3899 ds_put_format(result, "%s:", title);
3900 for (i = 0; i < FLOW_N_REGS; i++) {
3901 ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
3903 ds_put_char(result, '\n');
3907 trace_format_odp(struct ds *result, int level, const char *title,
3908 struct trace_ctx *trace)
3910 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
3912 ds_put_char_multiple(result, '\t', level);
3913 ds_put_format(result, "%s: ", title);
3914 format_odp_actions(result, ofpbuf_data(odp_actions),
3915 ofpbuf_size(odp_actions));
3916 ds_put_char(result, '\n');
3920 trace_format_megaflow(struct ds *result, int level, const char *title,
3921 struct trace_ctx *trace)
3925 ds_put_char_multiple(result, '\t', level);
3926 ds_put_format(result, "%s: ", title);
3927 flow_wildcards_or(&trace->wc, &trace->xout.wc, &trace->wc);
3928 match_init(&match, trace->key, &trace->wc);
3929 match_format(&match, result, OFP_DEFAULT_PRIORITY);
3930 ds_put_char(result, '\n');
3934 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
3936 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
3937 struct ds *result = trace->result;
3939 ds_put_char(result, '\n');
3940 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
3941 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
3942 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
3943 trace_format_megaflow(result, recurse + 1, "Resubmitted megaflow", trace);
3944 trace_format_rule(result, recurse + 1, rule);
3948 trace_report(struct xlate_in *xin, const char *s, int recurse)
3950 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
3951 struct ds *result = trace->result;
3953 ds_put_char_multiple(result, '\t', recurse);
3954 ds_put_cstr(result, s);
3955 ds_put_char(result, '\n');
3958 /* Parses the 'argc' elements of 'argv', ignoring argv[0]. The following
3959 * forms are supported:
3961 * - [dpname] odp_flow [-generate | packet]
3962 * - bridge br_flow [-generate | packet]
3964 * On success, initializes '*ofprotop' and 'flow' and returns NULL. On failure
3965 * returns a nonnull malloced error message. */
3966 static char * WARN_UNUSED_RESULT
3967 parse_flow_and_packet(int argc, const char *argv[],
3968 struct ofproto_dpif **ofprotop, struct flow *flow,
3969 struct ofpbuf **packetp)
3971 const struct dpif_backer *backer = NULL;
3972 const char *error = NULL;
3974 struct simap port_names = SIMAP_INITIALIZER(&port_names);
3975 struct ofpbuf *packet;
3976 struct ofpbuf odp_key;
3977 struct ofpbuf odp_mask;
3979 ofpbuf_init(&odp_key, 0);
3980 ofpbuf_init(&odp_mask, 0);
3982 /* Handle "-generate" or a hex string as the last argument. */
3983 if (!strcmp(argv[argc - 1], "-generate")) {
3984 packet = ofpbuf_new(0);
3987 error = eth_from_hex(argv[argc - 1], &packet);
3990 } else if (argc == 4) {
3991 /* The 3-argument form must end in "-generate' or a hex string. */
3997 /* odp_flow can have its in_port specified as a name instead of port no.
3998 * We do not yet know whether a given flow is a odp_flow or a br_flow.
3999 * But, to know whether a flow is odp_flow through odp_flow_from_string(),
4000 * we need to create a simap of name to port no. */
4002 const char *dp_type;
4003 if (!strncmp(argv[1], "ovs-", 4)) {
4004 dp_type = argv[1] + 4;
4008 backer = shash_find_data(&all_dpif_backers, dp_type);
4009 } else if (argc == 2) {
4010 struct shash_node *node;
4011 if (shash_count(&all_dpif_backers) == 1) {
4012 node = shash_first(&all_dpif_backers);
4013 backer = node->data;
4016 error = "Syntax error";
4019 if (backer && backer->dpif) {
4020 struct dpif_port dpif_port;
4021 struct dpif_port_dump port_dump;
4022 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
4023 simap_put(&port_names, dpif_port.name,
4024 odp_to_u32(dpif_port.port_no));
4028 /* Parse the flow and determine whether a datapath or
4029 * bridge is specified. If function odp_flow_key_from_string()
4030 * returns 0, the flow is a odp_flow. If function
4031 * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
4032 if (!odp_flow_from_string(argv[argc - 1], &port_names,
4033 &odp_key, &odp_mask)) {
4035 error = "Cannot find the datapath";
4039 if (xlate_receive(backer, NULL, ofpbuf_data(&odp_key),
4040 ofpbuf_size(&odp_key), flow,
4041 ofprotop, NULL, NULL, NULL, NULL)) {
4042 error = "Invalid datapath flow";
4046 char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
4049 m_err = xasprintf("Bad flow syntax: %s", err);
4054 error = "Must specify bridge name";
4058 *ofprotop = ofproto_dpif_lookup(argv[1]);
4060 error = "Unknown bridge name";
4066 /* Generate a packet, if requested. */
4068 if (!ofpbuf_size(packet)) {
4069 flow_compose(packet, flow);
4071 struct pkt_metadata md = pkt_metadata_from_flow(flow);
4073 /* Use the metadata from the flow and the packet argument
4074 * to reconstruct the flow. */
4075 flow_extract(packet, &md, flow);
4080 if (error && !m_err) {
4081 m_err = xstrdup(error);
4084 ofpbuf_delete(packet);
4088 ofpbuf_uninit(&odp_key);
4089 ofpbuf_uninit(&odp_mask);
4090 simap_destroy(&port_names);
4095 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
4096 void *aux OVS_UNUSED)
4098 struct ofproto_dpif *ofproto;
4099 struct ofpbuf *packet;
4103 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4108 ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
4109 unixctl_command_reply(conn, ds_cstr(&result));
4110 ds_destroy(&result);
4111 ofpbuf_delete(packet);
4113 unixctl_command_reply_error(conn, error);
4119 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
4120 const char *argv[], void *aux OVS_UNUSED)
4122 enum ofputil_protocol usable_protocols;
4123 struct ofproto_dpif *ofproto;
4124 bool enforce_consistency;
4125 struct ofpbuf ofpacts;
4126 struct ofpbuf *packet;
4131 /* Three kinds of error return values! */
4137 ofpbuf_init(&ofpacts, 0);
4139 /* Parse actions. */
4140 error = parse_ofpacts(argv[--argc], &ofpacts, &usable_protocols);
4142 unixctl_command_reply_error(conn, error);
4147 /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
4148 * consistency between the flow and the actions. With -consistent, we
4149 * enforce consistency even for a flow supported in OpenFlow 1.0. */
4150 if (!strcmp(argv[1], "-consistent")) {
4151 enforce_consistency = true;
4155 enforce_consistency = false;
4158 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4160 unixctl_command_reply_error(conn, error);
4165 /* Do the same checks as handle_packet_out() in ofproto.c.
4167 * We pass a 'table_id' of 0 to ofproto_check_ofpacts(), which isn't
4168 * strictly correct because these actions aren't in any table, but it's OK
4169 * because it 'table_id' is used only to check goto_table instructions, but
4170 * packet-outs take a list of actions and therefore it can't include
4173 * We skip the "meter" check here because meter is an instruction, not an
4174 * action, and thus cannot appear in ofpacts. */
4175 in_port = ofp_to_u16(flow.in_port.ofp_port);
4176 if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
4177 unixctl_command_reply_error(conn, "invalid in_port");
4180 if (enforce_consistency) {
4181 retval = ofpacts_check_consistency(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts),
4182 &flow, u16_to_ofp(ofproto->up.max_ports),
4183 0, 0, usable_protocols);
4185 retval = ofpacts_check(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &flow,
4186 u16_to_ofp(ofproto->up.max_ports), 0, 0,
4192 ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
4193 unixctl_command_reply_error(conn, ds_cstr(&result));
4197 ofproto_trace(ofproto, &flow, packet,
4198 ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &result);
4199 unixctl_command_reply(conn, ds_cstr(&result));
4202 ds_destroy(&result);
4203 ofpbuf_delete(packet);
4204 ofpbuf_uninit(&ofpacts);
4207 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
4208 * description of the results to 'ds'.
4210 * The trace follows a packet with the specified 'flow' through the flow
4211 * table. 'packet' may be nonnull to trace an actual packet, with consequent
4212 * side effects (if it is nonnull then its flow must be 'flow').
4214 * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
4215 * trace, otherwise the actions are determined by a flow table lookup. */
4217 ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
4218 const struct ofpbuf *packet,
4219 const struct ofpact ofpacts[], size_t ofpacts_len,
4222 struct rule_dpif *rule;
4223 struct trace_ctx trace;
4225 ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
4226 ds_put_cstr(ds, "Flow: ");
4227 flow_format(ds, flow);
4228 ds_put_char(ds, '\n');
4230 flow_wildcards_init_catchall(&trace.wc);
4234 rule_dpif_lookup(ofproto, flow, &trace.wc, &rule, false);
4236 trace_format_rule(ds, 0, rule);
4237 if (rule == ofproto->miss_rule) {
4238 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
4239 } else if (rule == ofproto->no_packet_in_rule) {
4240 ds_put_cstr(ds, "\nNo match, packets dropped because "
4241 "OFPPC_NO_PACKET_IN is set on in_port.\n");
4242 } else if (rule == ofproto->drop_frags_rule) {
4243 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
4244 "and the fragment handling mode is \"drop\".\n");
4248 if (rule || ofpacts) {
4250 trace.key = flow; /* Original flow key, used for megaflow. */
4251 trace.flow = *flow; /* May be modified by actions. */
4252 xlate_in_init(&trace.xin, ofproto, flow, rule, ntohs(flow->tcp_flags),
4255 trace.xin.ofpacts = ofpacts;
4256 trace.xin.ofpacts_len = ofpacts_len;
4258 trace.xin.resubmit_hook = trace_resubmit;
4259 trace.xin.report_hook = trace_report;
4261 xlate_actions(&trace.xin, &trace.xout);
4263 ds_put_char(ds, '\n');
4264 trace_format_flow(ds, 0, "Final flow", &trace);
4265 trace_format_megaflow(ds, 0, "Megaflow", &trace);
4267 ds_put_cstr(ds, "Datapath actions: ");
4268 format_odp_actions(ds, ofpbuf_data(&trace.xout.odp_actions),
4269 ofpbuf_size(&trace.xout.odp_actions));
4271 if (trace.xout.slow) {
4272 enum slow_path_reason slow;
4274 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
4275 "slow path because it:");
4277 slow = trace.xout.slow;
4279 enum slow_path_reason bit = rightmost_1bit(slow);
4281 ds_put_format(ds, "\n\t- %s.",
4282 slow_path_reason_to_explanation(bit));
4288 xlate_out_uninit(&trace.xout);
4292 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
4293 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
4294 * to destroy 'ofproto_shash' and free the returned value. */
4295 static const struct shash_node **
4296 get_ofprotos(struct shash *ofproto_shash)
4298 const struct ofproto_dpif *ofproto;
4300 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4301 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
4302 shash_add_nocopy(ofproto_shash, name, ofproto);
4305 return shash_sort(ofproto_shash);
4309 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
4310 const char *argv[] OVS_UNUSED,
4311 void *aux OVS_UNUSED)
4313 struct ds ds = DS_EMPTY_INITIALIZER;
4314 struct shash ofproto_shash;
4315 const struct shash_node **sorted_ofprotos;
4318 shash_init(&ofproto_shash);
4319 sorted_ofprotos = get_ofprotos(&ofproto_shash);
4320 for (i = 0; i < shash_count(&ofproto_shash); i++) {
4321 const struct shash_node *node = sorted_ofprotos[i];
4322 ds_put_format(&ds, "%s\n", node->name);
4325 shash_destroy(&ofproto_shash);
4326 free(sorted_ofprotos);
4328 unixctl_command_reply(conn, ds_cstr(&ds));
4333 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
4335 const struct shash_node **ofprotos;
4336 struct dpif_dp_stats dp_stats;
4337 struct shash ofproto_shash;
4340 dpif_get_dp_stats(backer->dpif, &dp_stats);
4342 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
4343 dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
4345 shash_init(&ofproto_shash);
4346 ofprotos = get_ofprotos(&ofproto_shash);
4347 for (i = 0; i < shash_count(&ofproto_shash); i++) {
4348 struct ofproto_dpif *ofproto = ofprotos[i]->data;
4349 const struct shash_node **ports;
4352 if (ofproto->backer != backer) {
4356 ds_put_format(ds, "\t%s:\n", ofproto->up.name);
4358 ports = shash_sort(&ofproto->up.port_by_name);
4359 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
4360 const struct shash_node *node = ports[j];
4361 struct ofport *ofport = node->data;
4363 odp_port_t odp_port;
4365 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
4368 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
4369 if (odp_port != ODPP_NONE) {
4370 ds_put_format(ds, "%"PRIu32":", odp_port);
4372 ds_put_cstr(ds, "none:");
4375 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
4378 if (!netdev_get_config(ofport->netdev, &config)) {
4379 const struct smap_node **nodes;
4382 nodes = smap_sort(&config);
4383 for (i = 0; i < smap_count(&config); i++) {
4384 const struct smap_node *node = nodes[i];
4385 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
4386 node->key, node->value);
4390 smap_destroy(&config);
4392 ds_put_char(ds, ')');
4393 ds_put_char(ds, '\n');
4397 shash_destroy(&ofproto_shash);
4402 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4403 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4405 struct ds ds = DS_EMPTY_INITIALIZER;
4406 const struct shash_node **backers;
4409 backers = shash_sort(&all_dpif_backers);
4410 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
4411 dpif_show_backer(backers[i]->data, &ds);
4415 unixctl_command_reply(conn, ds_cstr(&ds));
4420 ofproto_dpif_contains_flow(const struct ofproto_dpif *ofproto,
4421 const struct nlattr *key, size_t key_len)
4423 struct ofproto_dpif *ofp;
4426 xlate_receive(ofproto->backer, NULL, key, key_len, &flow, &ofp,
4427 NULL, NULL, NULL, NULL);
4428 return ofp == ofproto;
4432 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
4433 int argc OVS_UNUSED, const char *argv[],
4434 void *aux OVS_UNUSED)
4436 struct ds ds = DS_EMPTY_INITIALIZER;
4437 const struct dpif_flow_stats *stats;
4438 const struct ofproto_dpif *ofproto;
4439 struct dpif_flow_dump flow_dump;
4440 const struct nlattr *actions;
4441 const struct nlattr *mask;
4442 const struct nlattr *key;
4446 bool verbosity = false;
4447 struct dpif_port dpif_port;
4448 struct dpif_port_dump port_dump;
4449 struct hmap portno_names;
4453 ofproto = ofproto_dpif_lookup(argv[argc - 1]);
4455 unixctl_command_reply_error(conn, "no such bridge");
4459 if (argc > 2 && !strcmp(argv[1], "-m")) {
4463 hmap_init(&portno_names);
4464 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
4465 odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
4469 error = dpif_flow_dump_start(&flow_dump, ofproto->backer->dpif);
4473 dpif_flow_dump_state_init(ofproto->backer->dpif, &state);
4474 while (dpif_flow_dump_next(&flow_dump, state, &key, &key_len,
4475 &mask, &mask_len, &actions, &actions_len,
4477 if (!ofproto_dpif_contains_flow(ofproto, key, key_len)) {
4481 odp_flow_format(key, key_len, mask, mask_len, &portno_names, &ds,
4483 ds_put_cstr(&ds, ", ");
4484 dpif_flow_stats_format(stats, &ds);
4485 ds_put_cstr(&ds, ", actions:");
4486 format_odp_actions(&ds, actions, actions_len);
4487 ds_put_char(&ds, '\n');
4489 dpif_flow_dump_state_uninit(ofproto->backer->dpif, state);
4490 error = dpif_flow_dump_done(&flow_dump);
4495 ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
4496 unixctl_command_reply_error(conn, ds_cstr(&ds));
4498 unixctl_command_reply(conn, ds_cstr(&ds));
4500 odp_portno_names_destroy(&portno_names);
4501 hmap_destroy(&portno_names);
4506 ofproto_dpif_unixctl_init(void)
4508 static bool registered;
4514 unixctl_command_register(
4516 "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
4517 1, 3, ofproto_unixctl_trace, NULL);
4518 unixctl_command_register(
4519 "ofproto/trace-packet-out",
4520 "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
4521 2, 6, ofproto_unixctl_trace_actions, NULL);
4522 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
4523 ofproto_unixctl_fdb_flush, NULL);
4524 unixctl_command_register("fdb/show", "bridge", 1, 1,
4525 ofproto_unixctl_fdb_show, NULL);
4526 unixctl_command_register("dpif/dump-dps", "", 0, 0,
4527 ofproto_unixctl_dpif_dump_dps, NULL);
4528 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
4530 unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
4531 ofproto_unixctl_dpif_dump_flows, NULL);
4534 /* Returns true if 'table' is the table used for internal rules,
4535 * false otherwise. */
4537 table_is_internal(uint8_t table_id)
4539 return table_id == TBL_INTERNAL;
4542 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4544 * This is deprecated. It is only for compatibility with broken device drivers
4545 * in old versions of Linux that do not properly support VLANs when VLAN
4546 * devices are not used. When broken device drivers are no longer in
4547 * widespread use, we will delete these interfaces. */
4550 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
4552 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
4553 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
4555 if (realdev_ofp_port == ofport->realdev_ofp_port
4556 && vid == ofport->vlandev_vid) {
4560 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4562 if (ofport->realdev_ofp_port) {
4565 if (realdev_ofp_port && ofport->bundle) {
4566 /* vlandevs are enslaved to their realdevs, so they are not allowed to
4567 * themselves be part of a bundle. */
4568 bundle_set(ofport_->ofproto, ofport->bundle, NULL);
4571 ofport->realdev_ofp_port = realdev_ofp_port;
4572 ofport->vlandev_vid = vid;
4574 if (realdev_ofp_port) {
4575 vsp_add(ofport, realdev_ofp_port, vid);
4582 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
4584 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
4588 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
4589 OVS_EXCLUDED(ofproto->vsp_mutex)
4591 /* hmap_is_empty is thread safe. */
4592 return !hmap_is_empty(&ofproto->realdev_vid_map);
4597 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
4598 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
4599 OVS_REQUIRES(ofproto->vsp_mutex)
4601 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
4602 int vid = vlan_tci_to_vid(vlan_tci);
4603 const struct vlan_splinter *vsp;
4605 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
4606 hash_realdev_vid(realdev_ofp_port, vid),
4607 &ofproto->realdev_vid_map) {
4608 if (vsp->realdev_ofp_port == realdev_ofp_port
4609 && vsp->vid == vid) {
4610 return vsp->vlandev_ofp_port;
4614 return realdev_ofp_port;
4617 /* Returns the OFP port number of the Linux VLAN device that corresponds to
4618 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
4619 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
4620 * 'vlan_tci' 9, it would return the port number of eth0.9.
4622 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
4623 * function just returns its 'realdev_ofp_port' argument. */
4625 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
4626 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
4627 OVS_EXCLUDED(ofproto->vsp_mutex)
4631 /* hmap_is_empty is thread safe, see if we can return immediately. */
4632 if (hmap_is_empty(&ofproto->realdev_vid_map)) {
4633 return realdev_ofp_port;
4635 ovs_mutex_lock(&ofproto->vsp_mutex);
4636 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
4637 ovs_mutex_unlock(&ofproto->vsp_mutex);
4641 static struct vlan_splinter *
4642 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
4644 struct vlan_splinter *vsp;
4646 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
4647 hash_ofp_port(vlandev_ofp_port),
4648 &ofproto->vlandev_map) {
4649 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
4657 /* Returns the OpenFlow port number of the "real" device underlying the Linux
4658 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
4659 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
4660 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
4661 * eth0 and store 9 in '*vid'.
4663 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
4664 * VLAN device. Unless VLAN splinters are enabled, this is what this function
4667 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
4668 ofp_port_t vlandev_ofp_port, int *vid)
4669 OVS_REQUIRES(ofproto->vsp_mutex)
4671 if (!hmap_is_empty(&ofproto->vlandev_map)) {
4672 const struct vlan_splinter *vsp;
4674 vsp = vlandev_find(ofproto, vlandev_ofp_port);
4679 return vsp->realdev_ofp_port;
4685 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
4686 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
4687 * 'flow->in_port' to the "real" device backing the VLAN device, sets
4688 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
4689 * always the case unless VLAN splinters are enabled), returns false without
4690 * making any changes. */
4692 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
4693 OVS_EXCLUDED(ofproto->vsp_mutex)
4698 /* hmap_is_empty is thread safe. */
4699 if (hmap_is_empty(&ofproto->vlandev_map)) {
4703 ovs_mutex_lock(&ofproto->vsp_mutex);
4704 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
4705 ovs_mutex_unlock(&ofproto->vsp_mutex);
4710 /* Cause the flow to be processed as if it came in on the real device with
4711 * the VLAN device's VLAN ID. */
4712 flow->in_port.ofp_port = realdev;
4713 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
4718 vsp_remove(struct ofport_dpif *port)
4720 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
4721 struct vlan_splinter *vsp;
4723 ovs_mutex_lock(&ofproto->vsp_mutex);
4724 vsp = vlandev_find(ofproto, port->up.ofp_port);
4726 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
4727 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
4730 port->realdev_ofp_port = 0;
4732 VLOG_ERR("missing vlan device record");
4734 ovs_mutex_unlock(&ofproto->vsp_mutex);
4738 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
4740 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
4742 ovs_mutex_lock(&ofproto->vsp_mutex);
4743 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
4744 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
4745 == realdev_ofp_port)) {
4746 struct vlan_splinter *vsp;
4748 vsp = xmalloc(sizeof *vsp);
4749 vsp->realdev_ofp_port = realdev_ofp_port;
4750 vsp->vlandev_ofp_port = port->up.ofp_port;
4753 port->realdev_ofp_port = realdev_ofp_port;
4755 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
4756 hash_ofp_port(port->up.ofp_port));
4757 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
4758 hash_realdev_vid(realdev_ofp_port, vid));
4760 VLOG_ERR("duplicate vlan device record");
4762 ovs_mutex_unlock(&ofproto->vsp_mutex);
4766 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
4768 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
4769 return ofport ? ofport->odp_port : ODPP_NONE;
4772 struct ofport_dpif *
4773 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
4775 struct ofport_dpif *port;
4777 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
4778 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
4779 &backer->odp_to_ofport_map) {
4780 if (port->odp_port == odp_port) {
4781 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
4786 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
4791 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
4793 struct ofport_dpif *port;
4795 port = odp_port_to_ofport(ofproto->backer, odp_port);
4796 if (port && &ofproto->up == port->up.ofproto) {
4797 return port->up.ofp_port;
4803 struct ofproto_dpif *
4804 ofproto_dpif_recirc_get_ofproto(const struct dpif_backer *backer,
4807 struct dpif_backer_recirc_node *node;
4809 ovs_mutex_lock(&backer->recirc_mutex);
4810 node = CONTAINER_OF(hmap_first_with_hash(&backer->recirc_map, recirc_id),
4811 struct dpif_backer_recirc_node, hmap_node);
4812 ovs_mutex_unlock(&backer->recirc_mutex);
4814 return node ? node->ofproto : NULL;
4818 ofproto_dpif_alloc_recirc_id(struct ofproto_dpif *ofproto)
4820 struct dpif_backer *backer = ofproto->backer;
4821 uint32_t recirc_id = recirc_id_alloc(backer->rid_pool);
4824 struct dpif_backer_recirc_node *node = xmalloc(sizeof *node);
4826 node->recirc_id = recirc_id;
4827 node->ofproto = ofproto;
4829 ovs_mutex_lock(&backer->recirc_mutex);
4830 hmap_insert(&backer->recirc_map, &node->hmap_node, node->recirc_id);
4831 ovs_mutex_unlock(&backer->recirc_mutex);
4838 ofproto_dpif_free_recirc_id(struct ofproto_dpif *ofproto, uint32_t recirc_id)
4840 struct dpif_backer *backer = ofproto->backer;
4841 struct dpif_backer_recirc_node *node;
4843 ovs_mutex_lock(&backer->recirc_mutex);
4844 node = CONTAINER_OF(hmap_first_with_hash(&backer->recirc_map, recirc_id),
4845 struct dpif_backer_recirc_node, hmap_node);
4847 hmap_remove(&backer->recirc_map, &node->hmap_node);
4848 ovs_mutex_unlock(&backer->recirc_mutex);
4849 recirc_id_free(backer->rid_pool, node->recirc_id);
4851 if (node->ofproto != ofproto) {
4852 VLOG_ERR("recirc_id %"PRIu32", freed by incorrect ofproto (%s),"
4853 " expect ofproto (%s)", node->recirc_id, ofproto->up.name,
4854 node->ofproto->up.name);
4857 /* RCU postpone the free, since other threads may be referring
4858 * to 'node' at same time. */
4859 ovsrcu_postpone(free, node);
4861 ovs_mutex_unlock(&backer->recirc_mutex);
4866 ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
4867 const struct match *match, int priority,
4868 const struct ofpbuf *ofpacts,
4869 struct rule **rulep)
4871 struct ofputil_flow_mod fm;
4872 struct rule_dpif *rule;
4876 fm.priority = priority;
4877 fm.new_cookie = htonll(0);
4878 fm.cookie = htonll(0);
4879 fm.cookie_mask = htonll(0);
4880 fm.modify_cookie = false;
4881 fm.table_id = TBL_INTERNAL;
4882 fm.command = OFPFC_ADD;
4883 fm.idle_timeout = 0;
4884 fm.hard_timeout = 0;
4887 fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
4888 fm.ofpacts = ofpbuf_data(ofpacts);
4889 fm.ofpacts_len = ofpbuf_size(ofpacts);
4891 error = ofproto_flow_mod(&ofproto->up, &fm);
4893 VLOG_ERR_RL(&rl, "failed to add internal flow (%s)",
4894 ofperr_to_string(error));
4899 rule = rule_dpif_lookup_in_table(ofproto, TBL_INTERNAL, &fm.match.flow,
4900 &fm.match.wc, false);
4910 ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
4911 struct match *match, int priority)
4913 struct ofputil_flow_mod fm;
4917 fm.priority = priority;
4918 fm.new_cookie = htonll(0);
4919 fm.cookie = htonll(0);
4920 fm.cookie_mask = htonll(0);
4921 fm.modify_cookie = false;
4922 fm.table_id = TBL_INTERNAL;
4923 fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
4924 fm.command = OFPFC_DELETE_STRICT;
4926 error = ofproto_flow_mod(&ofproto->up, &fm);
4928 VLOG_ERR_RL(&rl, "failed to delete internal flow (%s)",
4929 ofperr_to_string(error));
4936 const struct ofproto_class ofproto_dpif_class = {
4950 NULL, /* get_memory_usage. */
4951 type_get_memory_usage,
4970 port_is_lacp_current,
4971 NULL, /* rule_choose_table */
4980 rule_modify_actions,
4994 get_stp_port_status,
5002 is_mirror_output_bundle,
5003 forward_bpdu_changed,
5004 set_mac_table_config,
5006 NULL, /* meter_get_features */
5007 NULL, /* meter_set */
5008 NULL, /* meter_get */
5009 NULL, /* meter_del */
5010 group_alloc, /* group_alloc */
5011 group_construct, /* group_construct */
5012 group_destruct, /* group_destruct */
5013 group_dealloc, /* group_dealloc */
5014 group_modify, /* group_modify */
5015 group_get_stats, /* group_get_stats */