2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-dpif.h"
20 #include "ofproto/ofproto-provider.h"
27 #include "byte-order.h"
28 #include "connectivity.h"
33 #include "dynamic-string.h"
34 #include "fail-open.h"
35 #include "guarded-list.h"
39 #include "mac-learning.h"
40 #include "meta-flow.h"
41 #include "multipath.h"
42 #include "netdev-vport.h"
47 #include "odp-execute.h"
50 #include "ofp-actions.h"
51 #include "ofp-parse.h"
52 #include "ofp-print.h"
53 #include "ofproto-dpif-ipfix.h"
54 #include "ofproto-dpif-mirror.h"
55 #include "ofproto-dpif-monitor.h"
56 #include "ofproto-dpif-rid.h"
57 #include "ofproto-dpif-sflow.h"
58 #include "ofproto-dpif-upcall.h"
59 #include "ofproto-dpif-xlate.h"
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 /* Make a local copy, since we don't own 'iface_hints' elements. */
414 SHASH_FOR_EACH(node, iface_hints) {
415 const struct iface_hint *orig_hint = node->data;
416 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
418 new_hint->br_name = xstrdup(orig_hint->br_name);
419 new_hint->br_type = xstrdup(orig_hint->br_type);
420 new_hint->ofp_port = orig_hint->ofp_port;
422 shash_add(&init_ofp_ports, node->name, new_hint);
427 enumerate_types(struct sset *types)
429 dp_enumerate_types(types);
433 enumerate_names(const char *type, struct sset *names)
435 struct ofproto_dpif *ofproto;
438 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
439 if (strcmp(type, ofproto->up.type)) {
442 sset_add(names, ofproto->up.name);
449 del(const char *type, const char *name)
454 error = dpif_open(name, type, &dpif);
456 error = dpif_delete(dpif);
463 port_open_type(const char *datapath_type, const char *port_type)
465 return dpif_port_open_type(datapath_type, port_type);
468 /* Type functions. */
470 static void process_dpif_port_changes(struct dpif_backer *);
471 static void process_dpif_all_ports_changed(struct dpif_backer *);
472 static void process_dpif_port_change(struct dpif_backer *,
473 const char *devname);
474 static void process_dpif_port_error(struct dpif_backer *, int error);
476 static struct ofproto_dpif *
477 lookup_ofproto_dpif_by_port_name(const char *name)
479 struct ofproto_dpif *ofproto;
481 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
482 if (sset_contains(&ofproto->ports, name)) {
491 type_run(const char *type)
493 struct dpif_backer *backer;
495 backer = shash_find_data(&all_dpif_backers, type);
497 /* This is not necessarily a problem, since backers are only
498 * created on demand. */
502 dpif_run(backer->dpif);
504 /* If vswitchd started with other_config:flow_restore_wait set as "true",
505 * and the configuration has now changed to "false", enable receiving
506 * packets from the datapath. */
507 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
510 backer->recv_set_enable = true;
512 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
514 VLOG_ERR("Failed to enable receiving packets in dpif.");
517 dpif_flow_flush(backer->dpif);
518 backer->need_revalidate = REV_RECONFIGURE;
521 if (backer->recv_set_enable) {
522 udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
525 if (backer->need_revalidate) {
526 struct ofproto_dpif *ofproto;
527 struct simap_node *node;
528 struct simap tmp_backers;
530 /* Handle tunnel garbage collection. */
531 simap_init(&tmp_backers);
532 simap_swap(&backer->tnl_backers, &tmp_backers);
534 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
535 struct ofport_dpif *iter;
537 if (backer != ofproto->backer) {
541 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
542 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
545 if (!iter->is_tunnel) {
549 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
550 namebuf, sizeof namebuf);
551 node = simap_find(&tmp_backers, dp_port);
553 simap_put(&backer->tnl_backers, dp_port, node->data);
554 simap_delete(&tmp_backers, node);
555 node = simap_find(&backer->tnl_backers, dp_port);
557 node = simap_find(&backer->tnl_backers, dp_port);
559 odp_port_t odp_port = ODPP_NONE;
561 if (!dpif_port_add(backer->dpif, iter->up.netdev,
563 simap_put(&backer->tnl_backers, dp_port,
564 odp_to_u32(odp_port));
565 node = simap_find(&backer->tnl_backers, dp_port);
570 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
571 if (tnl_port_reconfigure(iter, iter->up.netdev,
573 backer->need_revalidate = REV_RECONFIGURE;
578 SIMAP_FOR_EACH (node, &tmp_backers) {
579 dpif_port_del(backer->dpif, u32_to_odp(node->data));
581 simap_destroy(&tmp_backers);
583 switch (backer->need_revalidate) {
584 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
585 case REV_STP: COVERAGE_INC(rev_stp); break;
586 case REV_BOND: COVERAGE_INC(rev_bond); break;
587 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
588 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
589 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
591 backer->need_revalidate = 0;
593 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
594 struct ofport_dpif *ofport;
595 struct ofbundle *bundle;
597 if (ofproto->backer != backer) {
601 ovs_rwlock_wrlock(&xlate_rwlock);
602 xlate_ofproto_set(ofproto, ofproto->up.name,
603 ofproto->backer->dpif, ofproto->miss_rule,
604 ofproto->no_packet_in_rule, ofproto->ml,
605 ofproto->stp, ofproto->mbridge,
606 ofproto->sflow, ofproto->ipfix,
607 ofproto->netflow, ofproto->up.frag_handling,
608 ofproto->up.forward_bpdu,
609 connmgr_has_in_band(ofproto->up.connmgr),
610 ofproto->backer->enable_recirc,
611 ofproto->backer->variable_length_userdata,
612 ofproto->backer->max_mpls_depth);
614 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
615 xlate_bundle_set(ofproto, bundle, bundle->name,
616 bundle->vlan_mode, bundle->vlan,
617 bundle->trunks, bundle->use_priority_tags,
618 bundle->bond, bundle->lacp,
622 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
623 int stp_port = ofport->stp_port
624 ? stp_port_no(ofport->stp_port)
626 xlate_ofport_set(ofproto, ofport->bundle, ofport,
627 ofport->up.ofp_port, ofport->odp_port,
628 ofport->up.netdev, ofport->cfm,
629 ofport->bfd, ofport->peer, stp_port,
630 ofport->qdscp, ofport->n_qdscp,
631 ofport->up.pp.config, ofport->up.pp.state,
632 ofport->is_tunnel, ofport->may_enable);
634 ovs_rwlock_unlock(&xlate_rwlock);
637 udpif_revalidate(backer->udpif);
640 process_dpif_port_changes(backer);
645 /* Check for and handle port changes in 'backer''s dpif. */
647 process_dpif_port_changes(struct dpif_backer *backer)
653 error = dpif_port_poll(backer->dpif, &devname);
659 process_dpif_all_ports_changed(backer);
663 process_dpif_port_change(backer, devname);
668 process_dpif_port_error(backer, error);
675 process_dpif_all_ports_changed(struct dpif_backer *backer)
677 struct ofproto_dpif *ofproto;
678 struct dpif_port dpif_port;
679 struct dpif_port_dump dump;
680 struct sset devnames;
683 sset_init(&devnames);
684 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
685 if (ofproto->backer == backer) {
686 struct ofport *ofport;
688 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
689 sset_add(&devnames, netdev_get_name(ofport->netdev));
693 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
694 sset_add(&devnames, dpif_port.name);
697 SSET_FOR_EACH (devname, &devnames) {
698 process_dpif_port_change(backer, devname);
700 sset_destroy(&devnames);
704 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
706 struct ofproto_dpif *ofproto;
707 struct dpif_port port;
709 /* Don't report on the datapath's device. */
710 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
714 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
715 &all_ofproto_dpifs) {
716 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
721 ofproto = lookup_ofproto_dpif_by_port_name(devname);
722 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
723 /* The port was removed. If we know the datapath,
724 * report it through poll_set(). If we don't, it may be
725 * notifying us of a removal we initiated, so ignore it.
726 * If there's a pending ENOBUFS, let it stand, since
727 * everything will be reevaluated. */
728 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
729 sset_add(&ofproto->port_poll_set, devname);
730 ofproto->port_poll_errno = 0;
732 } else if (!ofproto) {
733 /* The port was added, but we don't know with which
734 * ofproto we should associate it. Delete it. */
735 dpif_port_del(backer->dpif, port.port_no);
737 struct ofport_dpif *ofport;
739 ofport = ofport_dpif_cast(shash_find_data(
740 &ofproto->up.port_by_name, devname));
742 && ofport->odp_port != port.port_no
743 && !odp_port_to_ofport(backer, port.port_no))
745 /* 'ofport''s datapath port number has changed from
746 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
747 * structures to match. */
748 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
749 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
750 ofport->odp_port = port.port_no;
751 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
752 hash_odp_port(port.port_no));
753 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
754 backer->need_revalidate = REV_RECONFIGURE;
757 dpif_port_destroy(&port);
760 /* Propagate 'error' to all ofprotos based on 'backer'. */
762 process_dpif_port_error(struct dpif_backer *backer, int error)
764 struct ofproto_dpif *ofproto;
766 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
767 if (ofproto->backer == backer) {
768 sset_clear(&ofproto->port_poll_set);
769 ofproto->port_poll_errno = error;
775 type_wait(const char *type)
777 struct dpif_backer *backer;
779 backer = shash_find_data(&all_dpif_backers, type);
781 /* This is not necessarily a problem, since backers are only
782 * created on demand. */
786 dpif_wait(backer->dpif);
789 /* Basic life-cycle. */
791 static int add_internal_flows(struct ofproto_dpif *);
793 static struct ofproto *
796 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
801 dealloc(struct ofproto *ofproto_)
803 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
807 /* Called when 'ofproto' is destructed. Checks for and clears any
810 dpif_backer_recirc_clear_ofproto(struct dpif_backer *backer,
811 struct ofproto_dpif *ofproto)
813 struct dpif_backer_recirc_node *node;
815 ovs_mutex_lock(&backer->recirc_mutex);
816 HMAP_FOR_EACH (node, hmap_node, &backer->recirc_map) {
817 if (node->ofproto == ofproto) {
818 VLOG_ERR("recirc_id %"PRIu32", not freed when ofproto (%s) "
819 "is destructed", node->recirc_id, ofproto->up.name);
820 hmap_remove(&backer->recirc_map, &node->hmap_node);
821 ovsrcu_postpone(free, node);
824 ovs_mutex_unlock(&backer->recirc_mutex);
828 close_dpif_backer(struct dpif_backer *backer)
830 ovs_assert(backer->refcount > 0);
832 if (--backer->refcount) {
836 udpif_destroy(backer->udpif);
838 simap_destroy(&backer->tnl_backers);
839 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
840 hmap_destroy(&backer->odp_to_ofport_map);
841 shash_find_and_delete(&all_dpif_backers, backer->type);
842 recirc_id_pool_destroy(backer->rid_pool);
843 hmap_destroy(&backer->recirc_map);
844 ovs_mutex_destroy(&backer->recirc_mutex);
846 dpif_close(backer->dpif);
850 /* Datapath port slated for removal from datapath. */
852 struct list list_node;
856 static bool check_variable_length_userdata(struct dpif_backer *backer);
857 static size_t check_max_mpls_depth(struct dpif_backer *backer);
858 static bool check_recirc(struct dpif_backer *backer);
861 open_dpif_backer(const char *type, struct dpif_backer **backerp)
863 struct dpif_backer *backer;
864 struct dpif_port_dump port_dump;
865 struct dpif_port port;
866 struct shash_node *node;
867 struct list garbage_list;
868 struct odp_garbage *garbage, *next;
875 backer = shash_find_data(&all_dpif_backers, type);
882 backer_name = xasprintf("ovs-%s", type);
884 /* Remove any existing datapaths, since we assume we're the only
885 * userspace controlling the datapath. */
887 dp_enumerate_names(type, &names);
888 SSET_FOR_EACH(name, &names) {
889 struct dpif *old_dpif;
891 /* Don't remove our backer if it exists. */
892 if (!strcmp(name, backer_name)) {
896 if (dpif_open(name, type, &old_dpif)) {
897 VLOG_WARN("couldn't open old datapath %s to remove it", name);
899 dpif_delete(old_dpif);
900 dpif_close(old_dpif);
903 sset_destroy(&names);
905 backer = xmalloc(sizeof *backer);
907 error = dpif_create_and_open(backer_name, type, &backer->dpif);
910 VLOG_ERR("failed to open datapath of type %s: %s", type,
911 ovs_strerror(error));
915 backer->udpif = udpif_create(backer, backer->dpif);
917 backer->type = xstrdup(type);
918 backer->refcount = 1;
919 hmap_init(&backer->odp_to_ofport_map);
920 ovs_rwlock_init(&backer->odp_to_ofport_lock);
921 backer->need_revalidate = 0;
922 simap_init(&backer->tnl_backers);
923 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
926 if (backer->recv_set_enable) {
927 dpif_flow_flush(backer->dpif);
930 /* Loop through the ports already on the datapath and remove any
931 * that we don't need anymore. */
932 list_init(&garbage_list);
933 dpif_port_dump_start(&port_dump, backer->dpif);
934 while (dpif_port_dump_next(&port_dump, &port)) {
935 node = shash_find(&init_ofp_ports, port.name);
936 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
937 garbage = xmalloc(sizeof *garbage);
938 garbage->odp_port = port.port_no;
939 list_push_front(&garbage_list, &garbage->list_node);
942 dpif_port_dump_done(&port_dump);
944 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
945 dpif_port_del(backer->dpif, garbage->odp_port);
946 list_remove(&garbage->list_node);
950 shash_add(&all_dpif_backers, type, backer);
952 backer->enable_recirc = check_recirc(backer);
953 backer->variable_length_userdata = check_variable_length_userdata(backer);
954 backer->max_mpls_depth = check_max_mpls_depth(backer);
955 backer->rid_pool = recirc_id_pool_create();
956 ovs_mutex_init(&backer->recirc_mutex);
957 hmap_init(&backer->recirc_map);
959 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
961 VLOG_ERR("failed to listen on datapath of type %s: %s",
962 type, ovs_strerror(error));
963 close_dpif_backer(backer);
967 if (backer->recv_set_enable) {
968 udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
974 /* Tests whether 'backer''s datapath supports recirculation. Only newer
975 * datapaths support OVS_KEY_ATTR_RECIRC_ID in keys. We need to disable some
976 * features on older datapaths that don't support this feature.
978 * Returns false if 'backer' definitely does not support recirculation, true if
979 * it seems to support recirculation or if at least the error we get is
982 check_recirc(struct dpif_backer *backer)
985 struct odputil_keybuf keybuf;
988 bool enable_recirc = false;
990 memset(&flow, 0, sizeof flow);
994 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
995 odp_flow_key_from_flow(&key, &flow, NULL, 0);
997 error = dpif_flow_put(backer->dpif, DPIF_FP_CREATE,
998 ofpbuf_data(&key), ofpbuf_size(&key), NULL, 0, NULL,
1000 if (error && error != EEXIST) {
1001 if (error != EINVAL) {
1002 VLOG_WARN("%s: Reciculation flow probe failed (%s)",
1003 dpif_name(backer->dpif), ovs_strerror(error));
1008 error = dpif_flow_del(backer->dpif, ofpbuf_data(&key), ofpbuf_size(&key),
1011 VLOG_WARN("%s: failed to delete recirculation feature probe flow",
1012 dpif_name(backer->dpif));
1015 enable_recirc = true;
1018 if (enable_recirc) {
1019 VLOG_INFO("%s: Datapath supports recirculation",
1020 dpif_name(backer->dpif));
1022 VLOG_INFO("%s: Datapath does not support recirculation",
1023 dpif_name(backer->dpif));
1026 return enable_recirc;
1029 /* Tests whether 'backer''s datapath supports variable-length
1030 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions. We need
1031 * to disable some features on older datapaths that don't support this
1034 * Returns false if 'backer' definitely does not support variable-length
1035 * userdata, true if it seems to support them or if at least the error we get
1038 check_variable_length_userdata(struct dpif_backer *backer)
1040 struct eth_header *eth;
1041 struct ofpbuf actions;
1042 struct dpif_execute execute;
1043 struct ofpbuf packet;
1047 /* Compose a userspace action that will cause an ERANGE error on older
1048 * datapaths that don't support variable-length userdata.
1050 * We really test for using userdata longer than 8 bytes, but older
1051 * datapaths accepted these, silently truncating the userdata to 8 bytes.
1052 * The same older datapaths rejected userdata shorter than 8 bytes, so we
1053 * test for that instead as a proxy for longer userdata support. */
1054 ofpbuf_init(&actions, 64);
1055 start = nl_msg_start_nested(&actions, OVS_ACTION_ATTR_USERSPACE);
1056 nl_msg_put_u32(&actions, OVS_USERSPACE_ATTR_PID,
1057 dpif_port_get_pid(backer->dpif, ODPP_NONE, 0));
1058 nl_msg_put_unspec_zero(&actions, OVS_USERSPACE_ATTR_USERDATA, 4);
1059 nl_msg_end_nested(&actions, start);
1061 /* Compose a dummy ethernet packet. */
1062 ofpbuf_init(&packet, ETH_HEADER_LEN);
1063 eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
1064 eth->eth_type = htons(0x1234);
1066 /* Execute the actions. On older datapaths this fails with ERANGE, on
1067 * newer datapaths it succeeds. */
1068 execute.actions = ofpbuf_data(&actions);
1069 execute.actions_len = ofpbuf_size(&actions);
1070 execute.packet = &packet;
1071 execute.md = PKT_METADATA_INITIALIZER(0);
1072 execute.needs_help = false;
1074 error = dpif_execute(backer->dpif, &execute);
1076 ofpbuf_uninit(&packet);
1077 ofpbuf_uninit(&actions);
1084 /* Variable-length userdata is not supported. */
1085 VLOG_WARN("%s: datapath does not support variable-length userdata "
1086 "feature (needs Linux 3.10+ or kernel module from OVS "
1087 "1..11+). The NXAST_SAMPLE action will be ignored.",
1088 dpif_name(backer->dpif));
1092 /* Something odd happened. We're not sure whether variable-length
1093 * userdata is supported. Default to "yes". */
1094 VLOG_WARN("%s: variable-length userdata feature probe failed (%s)",
1095 dpif_name(backer->dpif), ovs_strerror(error));
1100 /* Tests the MPLS label stack depth supported by 'backer''s datapath.
1102 * Returns the number of elements in a struct flow's mpls_lse field
1103 * if the datapath supports at least that many entries in an
1105 * Otherwise returns the number of MPLS push actions supported by
1108 check_max_mpls_depth(struct dpif_backer *backer)
1113 for (n = 0; n < FLOW_MAX_MPLS_LABELS; n++) {
1114 struct odputil_keybuf keybuf;
1118 memset(&flow, 0, sizeof flow);
1119 flow.dl_type = htons(ETH_TYPE_MPLS);
1120 flow_set_mpls_bos(&flow, n, 1);
1122 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1123 odp_flow_key_from_flow(&key, &flow, NULL, 0);
1125 error = dpif_flow_put(backer->dpif, DPIF_FP_CREATE,
1126 ofpbuf_data(&key), ofpbuf_size(&key), NULL, 0, NULL, 0, NULL);
1127 if (error && error != EEXIST) {
1128 if (error != EINVAL) {
1129 VLOG_WARN("%s: MPLS stack length feature probe failed (%s)",
1130 dpif_name(backer->dpif), ovs_strerror(error));
1135 error = dpif_flow_del(backer->dpif, ofpbuf_data(&key), ofpbuf_size(&key), NULL);
1137 VLOG_WARN("%s: failed to delete MPLS feature probe flow",
1138 dpif_name(backer->dpif));
1142 VLOG_INFO("%s: MPLS label stack length probed as %d",
1143 dpif_name(backer->dpif), n);
1148 construct(struct ofproto *ofproto_)
1150 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1151 struct shash_node *node, *next;
1154 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1159 ofproto->netflow = NULL;
1160 ofproto->sflow = NULL;
1161 ofproto->ipfix = NULL;
1162 ofproto->stp = NULL;
1163 ofproto->dump_seq = 0;
1164 hmap_init(&ofproto->bundles);
1165 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1166 ofproto->mbridge = mbridge_create();
1167 ofproto->has_bonded_bundles = false;
1168 ofproto->lacp_enabled = false;
1169 ovs_mutex_init_adaptive(&ofproto->stats_mutex);
1170 ovs_mutex_init(&ofproto->vsp_mutex);
1172 guarded_list_init(&ofproto->pins);
1174 ofproto_dpif_unixctl_init();
1176 hmap_init(&ofproto->vlandev_map);
1177 hmap_init(&ofproto->realdev_vid_map);
1179 sset_init(&ofproto->ports);
1180 sset_init(&ofproto->ghost_ports);
1181 sset_init(&ofproto->port_poll_set);
1182 ofproto->port_poll_errno = 0;
1183 ofproto->change_seq = 0;
1184 ofproto->pins_seq = seq_create();
1185 ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1188 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1189 struct iface_hint *iface_hint = node->data;
1191 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1192 /* Check if the datapath already has this port. */
1193 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1194 sset_add(&ofproto->ports, node->name);
1197 free(iface_hint->br_name);
1198 free(iface_hint->br_type);
1200 shash_delete(&init_ofp_ports, node);
1204 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1205 hash_string(ofproto->up.name, 0));
1206 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1208 ofproto_init_tables(ofproto_, N_TABLES);
1209 error = add_internal_flows(ofproto);
1211 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1217 add_internal_miss_flow(struct ofproto_dpif *ofproto, int id,
1218 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1224 match_init_catchall(&match);
1225 match_set_reg(&match, 0, id);
1227 error = ofproto_dpif_add_internal_flow(ofproto, &match, 0, ofpacts, &rule);
1228 *rulep = error ? NULL : rule_dpif_cast(rule);
1234 add_internal_flows(struct ofproto_dpif *ofproto)
1236 struct ofpact_controller *controller;
1237 uint64_t ofpacts_stub[128 / 8];
1238 struct ofpbuf ofpacts;
1239 struct rule *unused_rulep OVS_UNUSED;
1240 struct ofpact_resubmit *resubmit;
1245 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1248 controller = ofpact_put_CONTROLLER(&ofpacts);
1249 controller->max_len = UINT16_MAX;
1250 controller->controller_id = 0;
1251 controller->reason = OFPR_NO_MATCH;
1252 ofpact_pad(&ofpacts);
1254 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1255 &ofproto->miss_rule);
1260 ofpbuf_clear(&ofpacts);
1261 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1262 &ofproto->no_packet_in_rule);
1267 error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1268 &ofproto->drop_frags_rule);
1273 /* Drop any run away non-recirc rule lookups. Recirc_id has to be
1274 * zero when reaching this rule.
1276 * (priority=2), recirc_id=0, actions=drop
1278 ofpbuf_clear(&ofpacts);
1279 match_init_catchall(&match);
1280 match_set_recirc_id(&match, 0);
1281 error = ofproto_dpif_add_internal_flow(ofproto, &match, 2, &ofpacts,
1287 /* Continue rule lookups for not-matched recirc rules from table 0.
1289 * (priority=1), actions=resubmit(, 0)
1291 resubmit = ofpact_put_RESUBMIT(&ofpacts);
1292 resubmit->ofpact.compat = 0;
1293 resubmit->in_port = OFPP_IN_PORT;
1294 resubmit->table_id = 0;
1296 match_init_catchall(&match);
1298 error = ofproto_dpif_add_internal_flow(ofproto, &match, 1, &ofpacts,
1305 destruct(struct ofproto *ofproto_)
1307 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1308 struct rule_dpif *rule, *next_rule;
1309 struct ofproto_packet_in *pin, *next_pin;
1310 struct oftable *table;
1313 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1314 ovs_rwlock_wrlock(&xlate_rwlock);
1315 xlate_remove_ofproto(ofproto);
1316 ovs_rwlock_unlock(&xlate_rwlock);
1318 /* Ensure that the upcall processing threads have no remaining references
1319 * to the ofproto or anything in it. */
1320 udpif_synchronize(ofproto->backer->udpif);
1322 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1324 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1325 struct cls_cursor cursor;
1327 fat_rwlock_rdlock(&table->cls.rwlock);
1328 cls_cursor_init(&cursor, &table->cls, NULL);
1329 fat_rwlock_unlock(&table->cls.rwlock);
1330 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1331 ofproto_rule_delete(&ofproto->up, &rule->up);
1335 guarded_list_pop_all(&ofproto->pins, &pins);
1336 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1337 list_remove(&pin->list_node);
1338 free(CONST_CAST(void *, pin->up.packet));
1341 guarded_list_destroy(&ofproto->pins);
1343 dpif_backer_recirc_clear_ofproto(ofproto->backer, ofproto);
1345 mbridge_unref(ofproto->mbridge);
1347 netflow_unref(ofproto->netflow);
1348 dpif_sflow_unref(ofproto->sflow);
1349 hmap_destroy(&ofproto->bundles);
1350 mac_learning_unref(ofproto->ml);
1352 hmap_destroy(&ofproto->vlandev_map);
1353 hmap_destroy(&ofproto->realdev_vid_map);
1355 sset_destroy(&ofproto->ports);
1356 sset_destroy(&ofproto->ghost_ports);
1357 sset_destroy(&ofproto->port_poll_set);
1359 ovs_mutex_destroy(&ofproto->stats_mutex);
1360 ovs_mutex_destroy(&ofproto->vsp_mutex);
1362 seq_destroy(ofproto->pins_seq);
1364 close_dpif_backer(ofproto->backer);
1368 run(struct ofproto *ofproto_)
1370 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1371 uint64_t new_seq, new_dump_seq;
1373 if (mbridge_need_revalidate(ofproto->mbridge)) {
1374 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1375 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1376 mac_learning_flush(ofproto->ml);
1377 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1380 /* Always updates the ofproto->pins_seqno to avoid frequent wakeup during
1381 * flow restore. Even though nothing is processed during flow restore,
1382 * all queued 'pins' will be handled immediately when flow restore
1384 ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1386 /* Do not perform any periodic activity required by 'ofproto' while
1387 * waiting for flow restore to complete. */
1388 if (!ofproto_get_flow_restore_wait()) {
1389 struct ofproto_packet_in *pin, *next_pin;
1392 guarded_list_pop_all(&ofproto->pins, &pins);
1393 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1394 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1395 list_remove(&pin->list_node);
1396 free(CONST_CAST(void *, pin->up.packet));
1401 if (ofproto->netflow) {
1402 netflow_run(ofproto->netflow);
1404 if (ofproto->sflow) {
1405 dpif_sflow_run(ofproto->sflow);
1407 if (ofproto->ipfix) {
1408 dpif_ipfix_run(ofproto->ipfix);
1411 new_seq = seq_read(connectivity_seq_get());
1412 if (ofproto->change_seq != new_seq) {
1413 struct ofport_dpif *ofport;
1415 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1419 ofproto->change_seq = new_seq;
1421 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1422 struct ofbundle *bundle;
1424 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1430 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1431 if (mac_learning_run(ofproto->ml)) {
1432 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1434 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1436 new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
1437 if (ofproto->dump_seq != new_dump_seq) {
1438 struct rule *rule, *next_rule;
1440 /* We know stats are relatively fresh, so now is a good time to do some
1442 ofproto->dump_seq = new_dump_seq;
1444 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1446 ovs_mutex_lock(&ofproto_mutex);
1447 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
1448 &ofproto->up.expirable) {
1449 rule_expire(rule_dpif_cast(rule));
1451 ovs_mutex_unlock(&ofproto_mutex);
1453 /* All outstanding data in existing flows has been accounted, so it's a
1454 * good time to do bond rebalancing. */
1455 if (ofproto->has_bonded_bundles) {
1456 struct ofbundle *bundle;
1458 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1460 bond_rebalance(bundle->bond);
1470 wait(struct ofproto *ofproto_)
1472 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1474 if (ofproto_get_flow_restore_wait()) {
1478 if (ofproto->sflow) {
1479 dpif_sflow_wait(ofproto->sflow);
1481 if (ofproto->ipfix) {
1482 dpif_ipfix_wait(ofproto->ipfix);
1484 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1485 struct ofbundle *bundle;
1487 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1488 bundle_wait(bundle);
1491 if (ofproto->netflow) {
1492 netflow_wait(ofproto->netflow);
1494 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1495 mac_learning_wait(ofproto->ml);
1496 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1498 if (ofproto->backer->need_revalidate) {
1499 /* Shouldn't happen, but if it does just go around again. */
1500 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1501 poll_immediate_wake();
1504 seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
1505 seq_wait(ofproto->pins_seq, ofproto->pins_seqno);
1509 type_get_memory_usage(const char *type, struct simap *usage)
1511 struct dpif_backer *backer;
1513 backer = shash_find_data(&all_dpif_backers, type);
1515 udpif_get_memory_usage(backer->udpif, usage);
1520 flush(struct ofproto *ofproto_)
1522 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1523 struct dpif_backer *backer = ofproto->backer;
1526 udpif_flush(backer->udpif);
1531 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1532 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1534 *arp_match_ip = true;
1535 *actions = (OFPUTIL_A_OUTPUT |
1536 OFPUTIL_A_SET_VLAN_VID |
1537 OFPUTIL_A_SET_VLAN_PCP |
1538 OFPUTIL_A_STRIP_VLAN |
1539 OFPUTIL_A_SET_DL_SRC |
1540 OFPUTIL_A_SET_DL_DST |
1541 OFPUTIL_A_SET_NW_SRC |
1542 OFPUTIL_A_SET_NW_DST |
1543 OFPUTIL_A_SET_NW_TOS |
1544 OFPUTIL_A_SET_TP_SRC |
1545 OFPUTIL_A_SET_TP_DST |
1550 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1552 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1553 struct dpif_dp_stats s;
1554 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1558 strcpy(ots->name, "classifier");
1560 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1561 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes, &used);
1562 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes,
1564 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes,
1566 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1567 ots->lookup_count = htonll(n_lookup);
1568 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1571 static struct ofport *
1574 struct ofport_dpif *port = xmalloc(sizeof *port);
1579 port_dealloc(struct ofport *port_)
1581 struct ofport_dpif *port = ofport_dpif_cast(port_);
1586 port_construct(struct ofport *port_)
1588 struct ofport_dpif *port = ofport_dpif_cast(port_);
1589 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1590 const struct netdev *netdev = port->up.netdev;
1591 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1592 struct dpif_port dpif_port;
1595 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1596 port->bundle = NULL;
1599 port->may_enable = true;
1600 port->stp_port = NULL;
1601 port->stp_state = STP_DISABLED;
1602 port->is_tunnel = false;
1606 port->realdev_ofp_port = 0;
1607 port->vlandev_vid = 0;
1608 port->carrier_seq = netdev_get_carrier_resets(netdev);
1609 port->is_layer3 = netdev_vport_is_layer3(netdev);
1611 if (netdev_vport_is_patch(netdev)) {
1612 /* By bailing out here, we don't submit the port to the sFlow module
1613 * to be considered for counter polling export. This is correct
1614 * because the patch port represents an interface that sFlow considers
1615 * to be "internal" to the switch as a whole, and therefore not an
1616 * candidate for counter polling. */
1617 port->odp_port = ODPP_NONE;
1618 ofport_update_peer(port);
1622 error = dpif_port_query_by_name(ofproto->backer->dpif,
1623 netdev_vport_get_dpif_port(netdev, namebuf,
1630 port->odp_port = dpif_port.port_no;
1632 if (netdev_get_tunnel_config(netdev)) {
1633 tnl_port_add(port, port->up.netdev, port->odp_port);
1634 port->is_tunnel = true;
1636 /* Sanity-check that a mapping doesn't already exist. This
1637 * shouldn't happen for non-tunnel ports. */
1638 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1639 VLOG_ERR("port %s already has an OpenFlow port number",
1641 dpif_port_destroy(&dpif_port);
1645 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1646 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1647 hash_odp_port(port->odp_port));
1648 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1650 dpif_port_destroy(&dpif_port);
1652 if (ofproto->sflow) {
1653 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1660 port_destruct(struct ofport *port_)
1662 struct ofport_dpif *port = ofport_dpif_cast(port_);
1663 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1664 const char *devname = netdev_get_name(port->up.netdev);
1665 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1666 const char *dp_port_name;
1668 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1669 ovs_rwlock_wrlock(&xlate_rwlock);
1670 xlate_ofport_remove(port);
1671 ovs_rwlock_unlock(&xlate_rwlock);
1673 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1675 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1676 /* The underlying device is still there, so delete it. This
1677 * happens when the ofproto is being destroyed, since the caller
1678 * assumes that removal of attached ports will happen as part of
1680 if (!port->is_tunnel) {
1681 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1686 port->peer->peer = NULL;
1690 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1691 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1692 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1693 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1697 sset_find_and_delete(&ofproto->ports, devname);
1698 sset_find_and_delete(&ofproto->ghost_ports, devname);
1699 bundle_remove(port_);
1700 set_cfm(port_, NULL);
1701 set_bfd(port_, NULL);
1702 if (port->stp_port) {
1703 stp_port_disable(port->stp_port);
1705 if (ofproto->sflow) {
1706 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1713 port_modified(struct ofport *port_)
1715 struct ofport_dpif *port = ofport_dpif_cast(port_);
1717 if (port->bundle && port->bundle->bond) {
1718 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1722 cfm_set_netdev(port->cfm, port->up.netdev);
1726 bfd_set_netdev(port->bfd, port->up.netdev);
1729 ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1730 port->up.pp.hw_addr);
1732 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1734 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1738 ofport_update_peer(port);
1742 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1744 struct ofport_dpif *port = ofport_dpif_cast(port_);
1745 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1746 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1748 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1749 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1750 OFPUTIL_PC_NO_PACKET_IN)) {
1751 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1753 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1754 bundle_update(port->bundle);
1760 set_sflow(struct ofproto *ofproto_,
1761 const struct ofproto_sflow_options *sflow_options)
1763 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1764 struct dpif_sflow *ds = ofproto->sflow;
1766 if (sflow_options) {
1768 struct ofport_dpif *ofport;
1770 ds = ofproto->sflow = dpif_sflow_create();
1771 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1772 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1774 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1776 dpif_sflow_set_options(ds, sflow_options);
1779 dpif_sflow_unref(ds);
1780 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1781 ofproto->sflow = NULL;
1789 struct ofproto *ofproto_,
1790 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1791 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1792 size_t n_flow_exporters_options)
1794 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1795 struct dpif_ipfix *di = ofproto->ipfix;
1796 bool has_options = bridge_exporter_options || flow_exporters_options;
1798 if (has_options && !di) {
1799 di = ofproto->ipfix = dpif_ipfix_create();
1803 /* Call set_options in any case to cleanly flush the flow
1804 * caches in the last exporters that are to be destroyed. */
1805 dpif_ipfix_set_options(
1806 di, bridge_exporter_options, flow_exporters_options,
1807 n_flow_exporters_options);
1810 dpif_ipfix_unref(di);
1811 ofproto->ipfix = NULL;
1819 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1821 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1826 struct ofproto_dpif *ofproto;
1828 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1829 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1830 ofport->cfm = cfm_create(ofport->up.netdev);
1833 if (cfm_configure(ofport->cfm, s)) {
1840 cfm_unref(ofport->cfm);
1843 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1844 ofport->up.pp.hw_addr);
1849 get_cfm_status(const struct ofport *ofport_, bool force,
1850 struct ofproto_cfm_status *status)
1852 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1856 if (cfm_check_status_change(ofport->cfm) || force) {
1857 status->faults = cfm_get_fault(ofport->cfm);
1858 status->flap_count = cfm_get_flap_count(ofport->cfm);
1859 status->remote_opstate = cfm_get_opup(ofport->cfm);
1860 status->health = cfm_get_health(ofport->cfm);
1861 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1863 ret = NO_STATUS_CHANGE;
1873 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1875 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1876 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1880 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1881 cfg, ofport->up.netdev);
1882 if (ofport->bfd != old) {
1883 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1885 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1886 ofport->up.pp.hw_addr);
1891 get_bfd_status(struct ofport *ofport_, bool force, struct smap *smap)
1893 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1897 if (bfd_check_status_change(ofport->bfd) || force) {
1898 bfd_get_status(ofport->bfd, smap);
1900 ret = NO_STATUS_CHANGE;
1909 /* Spanning Tree. */
1912 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1914 struct ofproto_dpif *ofproto = ofproto_;
1915 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1916 struct ofport_dpif *ofport;
1918 ofport = stp_port_get_aux(sp);
1920 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1921 ofproto->up.name, port_num);
1923 struct eth_header *eth = ofpbuf_l2(pkt);
1925 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1926 if (eth_addr_is_zero(eth->eth_src)) {
1927 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1928 "with unknown MAC", ofproto->up.name, port_num);
1930 ofproto_dpif_send_packet(ofport, pkt);
1936 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1938 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1940 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1942 /* Only revalidate flows if the configuration changed. */
1943 if (!s != !ofproto->stp) {
1944 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1948 if (!ofproto->stp) {
1949 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1950 send_bpdu_cb, ofproto);
1951 ofproto->stp_last_tick = time_msec();
1954 stp_set_bridge_id(ofproto->stp, s->system_id);
1955 stp_set_bridge_priority(ofproto->stp, s->priority);
1956 stp_set_hello_time(ofproto->stp, s->hello_time);
1957 stp_set_max_age(ofproto->stp, s->max_age);
1958 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1960 struct ofport *ofport;
1962 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1963 set_stp_port(ofport, NULL);
1966 stp_unref(ofproto->stp);
1967 ofproto->stp = NULL;
1974 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1976 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1980 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1981 s->designated_root = stp_get_designated_root(ofproto->stp);
1982 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1991 update_stp_port_state(struct ofport_dpif *ofport)
1993 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1994 enum stp_state state;
1996 /* Figure out new state. */
1997 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2001 if (ofport->stp_state != state) {
2002 enum ofputil_port_state of_state;
2005 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2006 netdev_get_name(ofport->up.netdev),
2007 stp_state_name(ofport->stp_state),
2008 stp_state_name(state));
2009 if (stp_learn_in_state(ofport->stp_state)
2010 != stp_learn_in_state(state)) {
2011 /* xxx Learning action flows should also be flushed. */
2012 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2013 mac_learning_flush(ofproto->ml);
2014 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2016 fwd_change = stp_forward_in_state(ofport->stp_state)
2017 != stp_forward_in_state(state);
2019 ofproto->backer->need_revalidate = REV_STP;
2020 ofport->stp_state = state;
2021 ofport->stp_state_entered = time_msec();
2023 if (fwd_change && ofport->bundle) {
2024 bundle_update(ofport->bundle);
2027 /* Update the STP state bits in the OpenFlow port description. */
2028 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2029 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2030 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2031 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2032 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2034 ofproto_port_set_state(&ofport->up, of_state);
2038 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2039 * caller is responsible for assigning STP port numbers and ensuring
2040 * there are no duplicates. */
2042 set_stp_port(struct ofport *ofport_,
2043 const struct ofproto_port_stp_settings *s)
2045 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2046 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2047 struct stp_port *sp = ofport->stp_port;
2049 if (!s || !s->enable) {
2051 ofport->stp_port = NULL;
2052 stp_port_disable(sp);
2053 update_stp_port_state(ofport);
2056 } else if (sp && stp_port_no(sp) != s->port_num
2057 && ofport == stp_port_get_aux(sp)) {
2058 /* The port-id changed, so disable the old one if it's not
2059 * already in use by another port. */
2060 stp_port_disable(sp);
2063 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2064 stp_port_enable(sp);
2066 stp_port_set_aux(sp, ofport);
2067 stp_port_set_priority(sp, s->priority);
2068 stp_port_set_path_cost(sp, s->path_cost);
2070 update_stp_port_state(ofport);
2076 get_stp_port_status(struct ofport *ofport_,
2077 struct ofproto_port_stp_status *s)
2079 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2080 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2081 struct stp_port *sp = ofport->stp_port;
2083 if (!ofproto->stp || !sp) {
2089 s->port_id = stp_port_get_id(sp);
2090 s->state = stp_port_get_state(sp);
2091 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2092 s->role = stp_port_get_role(sp);
2098 get_stp_port_stats(struct ofport *ofport_,
2099 struct ofproto_port_stp_stats *s)
2101 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2102 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2103 struct stp_port *sp = ofport->stp_port;
2105 if (!ofproto->stp || !sp) {
2111 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2117 stp_run(struct ofproto_dpif *ofproto)
2120 long long int now = time_msec();
2121 long long int elapsed = now - ofproto->stp_last_tick;
2122 struct stp_port *sp;
2125 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2126 ofproto->stp_last_tick = now;
2128 while (stp_get_changed_port(ofproto->stp, &sp)) {
2129 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2132 update_stp_port_state(ofport);
2136 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2137 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2138 mac_learning_flush(ofproto->ml);
2139 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2145 stp_wait(struct ofproto_dpif *ofproto)
2148 poll_timer_wait(1000);
2153 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2156 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2157 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2159 if (ofport->n_qdscp != n_qdscp
2160 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2161 n_qdscp * sizeof *qdscp))) {
2162 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2163 free(ofport->qdscp);
2164 ofport->qdscp = n_qdscp
2165 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2167 ofport->n_qdscp = n_qdscp;
2175 /* Expires all MAC learning entries associated with 'bundle' and forces its
2176 * ofproto to revalidate every flow.
2178 * Normally MAC learning entries are removed only from the ofproto associated
2179 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2180 * are removed from every ofproto. When patch ports and SLB bonds are in use
2181 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2182 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2183 * with the host from which it migrated. */
2185 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2187 struct ofproto_dpif *ofproto = bundle->ofproto;
2188 struct mac_learning *ml = ofproto->ml;
2189 struct mac_entry *mac, *next_mac;
2191 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2192 ovs_rwlock_wrlock(&ml->rwlock);
2193 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2194 if (mac->port.p == bundle) {
2196 struct ofproto_dpif *o;
2198 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2200 struct mac_entry *e;
2202 ovs_rwlock_wrlock(&o->ml->rwlock);
2203 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2205 mac_learning_expire(o->ml, e);
2207 ovs_rwlock_unlock(&o->ml->rwlock);
2212 mac_learning_expire(ml, mac);
2215 ovs_rwlock_unlock(&ml->rwlock);
2218 static struct ofbundle *
2219 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2221 struct ofbundle *bundle;
2223 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2224 &ofproto->bundles) {
2225 if (bundle->aux == aux) {
2233 bundle_update(struct ofbundle *bundle)
2235 struct ofport_dpif *port;
2237 bundle->floodable = true;
2238 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2239 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2241 || !stp_forward_in_state(port->stp_state)) {
2242 bundle->floodable = false;
2249 bundle_del_port(struct ofport_dpif *port)
2251 struct ofbundle *bundle = port->bundle;
2253 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2255 list_remove(&port->bundle_node);
2256 port->bundle = NULL;
2259 lacp_slave_unregister(bundle->lacp, port);
2262 bond_slave_unregister(bundle->bond, port);
2265 bundle_update(bundle);
2269 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2270 struct lacp_slave_settings *lacp)
2272 struct ofport_dpif *port;
2274 port = get_ofp_port(bundle->ofproto, ofp_port);
2279 if (port->bundle != bundle) {
2280 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2282 bundle_remove(&port->up);
2285 port->bundle = bundle;
2286 list_push_back(&bundle->ports, &port->bundle_node);
2287 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2289 || !stp_forward_in_state(port->stp_state)) {
2290 bundle->floodable = false;
2294 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2295 lacp_slave_register(bundle->lacp, port, lacp);
2302 bundle_destroy(struct ofbundle *bundle)
2304 struct ofproto_dpif *ofproto;
2305 struct ofport_dpif *port, *next_port;
2311 ofproto = bundle->ofproto;
2312 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2314 ovs_rwlock_wrlock(&xlate_rwlock);
2315 xlate_bundle_remove(bundle);
2316 ovs_rwlock_unlock(&xlate_rwlock);
2318 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2319 bundle_del_port(port);
2322 bundle_flush_macs(bundle, true);
2323 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2325 free(bundle->trunks);
2326 lacp_unref(bundle->lacp);
2327 bond_unref(bundle->bond);
2332 bundle_set(struct ofproto *ofproto_, void *aux,
2333 const struct ofproto_bundle_settings *s)
2335 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2336 bool need_flush = false;
2337 struct ofport_dpif *port;
2338 struct ofbundle *bundle;
2339 unsigned long *trunks;
2345 bundle_destroy(bundle_lookup(ofproto, aux));
2349 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2350 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2352 bundle = bundle_lookup(ofproto, aux);
2354 bundle = xmalloc(sizeof *bundle);
2356 bundle->ofproto = ofproto;
2357 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2358 hash_pointer(aux, 0));
2360 bundle->name = NULL;
2362 list_init(&bundle->ports);
2363 bundle->vlan_mode = PORT_VLAN_TRUNK;
2365 bundle->trunks = NULL;
2366 bundle->use_priority_tags = s->use_priority_tags;
2367 bundle->lacp = NULL;
2368 bundle->bond = NULL;
2370 bundle->floodable = true;
2371 mbridge_register_bundle(ofproto->mbridge, bundle);
2374 if (!bundle->name || strcmp(s->name, bundle->name)) {
2376 bundle->name = xstrdup(s->name);
2381 ofproto->lacp_enabled = true;
2382 if (!bundle->lacp) {
2383 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2384 bundle->lacp = lacp_create();
2386 lacp_configure(bundle->lacp, s->lacp);
2388 lacp_unref(bundle->lacp);
2389 bundle->lacp = NULL;
2392 /* Update set of ports. */
2394 for (i = 0; i < s->n_slaves; i++) {
2395 if (!bundle_add_port(bundle, s->slaves[i],
2396 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2400 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2401 struct ofport_dpif *next_port;
2403 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2404 for (i = 0; i < s->n_slaves; i++) {
2405 if (s->slaves[i] == port->up.ofp_port) {
2410 bundle_del_port(port);
2414 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2416 if (list_is_empty(&bundle->ports)) {
2417 bundle_destroy(bundle);
2421 /* Set VLAN tagging mode */
2422 if (s->vlan_mode != bundle->vlan_mode
2423 || s->use_priority_tags != bundle->use_priority_tags) {
2424 bundle->vlan_mode = s->vlan_mode;
2425 bundle->use_priority_tags = s->use_priority_tags;
2430 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2431 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2433 if (vlan != bundle->vlan) {
2434 bundle->vlan = vlan;
2438 /* Get trunked VLANs. */
2439 switch (s->vlan_mode) {
2440 case PORT_VLAN_ACCESS:
2444 case PORT_VLAN_TRUNK:
2445 trunks = CONST_CAST(unsigned long *, s->trunks);
2448 case PORT_VLAN_NATIVE_UNTAGGED:
2449 case PORT_VLAN_NATIVE_TAGGED:
2450 if (vlan != 0 && (!s->trunks
2451 || !bitmap_is_set(s->trunks, vlan)
2452 || bitmap_is_set(s->trunks, 0))) {
2453 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2455 trunks = bitmap_clone(s->trunks, 4096);
2457 trunks = bitmap_allocate1(4096);
2459 bitmap_set1(trunks, vlan);
2460 bitmap_set0(trunks, 0);
2462 trunks = CONST_CAST(unsigned long *, s->trunks);
2469 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2470 free(bundle->trunks);
2471 if (trunks == s->trunks) {
2472 bundle->trunks = vlan_bitmap_clone(trunks);
2474 bundle->trunks = trunks;
2479 if (trunks != s->trunks) {
2484 if (!list_is_short(&bundle->ports)) {
2485 bundle->ofproto->has_bonded_bundles = true;
2487 if (bond_reconfigure(bundle->bond, s->bond)) {
2488 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2491 bundle->bond = bond_create(s->bond, ofproto);
2492 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2495 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2496 bond_slave_register(bundle->bond, port,
2497 port->up.ofp_port, port->up.netdev);
2500 bond_unref(bundle->bond);
2501 bundle->bond = NULL;
2504 /* If we changed something that would affect MAC learning, un-learn
2505 * everything on this port and force flow revalidation. */
2507 bundle_flush_macs(bundle, false);
2514 bundle_remove(struct ofport *port_)
2516 struct ofport_dpif *port = ofport_dpif_cast(port_);
2517 struct ofbundle *bundle = port->bundle;
2520 bundle_del_port(port);
2521 if (list_is_empty(&bundle->ports)) {
2522 bundle_destroy(bundle);
2523 } else if (list_is_short(&bundle->ports)) {
2524 bond_unref(bundle->bond);
2525 bundle->bond = NULL;
2531 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2533 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2534 struct ofport_dpif *port = port_;
2535 uint8_t ea[ETH_ADDR_LEN];
2538 error = netdev_get_etheraddr(port->up.netdev, ea);
2540 struct ofpbuf packet;
2543 ofpbuf_init(&packet, 0);
2544 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2546 memcpy(packet_pdu, pdu, pdu_size);
2548 ofproto_dpif_send_packet(port, &packet);
2549 ofpbuf_uninit(&packet);
2551 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2552 "%s (%s)", port->bundle->name,
2553 netdev_get_name(port->up.netdev), ovs_strerror(error));
2558 bundle_send_learning_packets(struct ofbundle *bundle)
2560 struct ofproto_dpif *ofproto = bundle->ofproto;
2561 struct ofpbuf *learning_packet;
2562 int error, n_packets, n_errors;
2563 struct mac_entry *e;
2564 struct list packets;
2566 list_init(&packets);
2567 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2568 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2569 if (e->port.p != bundle) {
2572 learning_packet = bond_compose_learning_packet(bundle->bond,
2575 /* Temporarily use 'frame' as a private pointer (see below). */
2576 ovs_assert(learning_packet->frame == ofpbuf_data(learning_packet));
2577 learning_packet->frame = port_void;
2578 list_push_back(&packets, &learning_packet->list_node);
2581 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2583 error = n_packets = n_errors = 0;
2584 LIST_FOR_EACH (learning_packet, list_node, &packets) {
2586 void *port_void = learning_packet->frame;
2588 /* Restore 'frame'. */
2589 learning_packet->frame = ofpbuf_data(learning_packet);
2590 ret = ofproto_dpif_send_packet(port_void, learning_packet);
2597 ofpbuf_list_delete(&packets);
2600 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2601 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2602 "packets, last error was: %s",
2603 bundle->name, n_errors, n_packets, ovs_strerror(error));
2605 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2606 bundle->name, n_packets);
2611 bundle_run(struct ofbundle *bundle)
2614 lacp_run(bundle->lacp, send_pdu_cb);
2617 struct ofport_dpif *port;
2619 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2620 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2623 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2624 bundle->ofproto->backer->need_revalidate = REV_BOND;
2627 if (bond_should_send_learning_packets(bundle->bond)) {
2628 bundle_send_learning_packets(bundle);
2634 bundle_wait(struct ofbundle *bundle)
2637 lacp_wait(bundle->lacp);
2640 bond_wait(bundle->bond);
2647 mirror_set__(struct ofproto *ofproto_, void *aux,
2648 const struct ofproto_mirror_settings *s)
2650 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2651 struct ofbundle **srcs, **dsts;
2656 mirror_destroy(ofproto->mbridge, aux);
2660 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2661 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2663 for (i = 0; i < s->n_srcs; i++) {
2664 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2667 for (i = 0; i < s->n_dsts; i++) {
2668 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2671 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2672 s->n_dsts, s->src_vlans,
2673 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2680 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2681 uint64_t *packets, uint64_t *bytes)
2683 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2688 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2690 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2691 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2692 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2693 mac_learning_flush(ofproto->ml);
2695 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2700 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2702 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2703 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2704 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2708 forward_bpdu_changed(struct ofproto *ofproto_)
2710 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2711 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2715 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2718 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2719 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2720 mac_learning_set_idle_time(ofproto->ml, idle_time);
2721 mac_learning_set_max_entries(ofproto->ml, max_entries);
2722 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2727 static struct ofport_dpif *
2728 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2730 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2731 return ofport ? ofport_dpif_cast(ofport) : NULL;
2735 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2736 struct ofproto_port *ofproto_port,
2737 struct dpif_port *dpif_port)
2739 ofproto_port->name = dpif_port->name;
2740 ofproto_port->type = dpif_port->type;
2741 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2745 ofport_update_peer(struct ofport_dpif *ofport)
2747 const struct ofproto_dpif *ofproto;
2748 struct dpif_backer *backer;
2751 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2755 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2756 backer->need_revalidate = REV_RECONFIGURE;
2759 ofport->peer->peer = NULL;
2760 ofport->peer = NULL;
2763 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2768 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2769 struct ofport *peer_ofport;
2770 struct ofport_dpif *peer;
2773 if (ofproto->backer != backer) {
2777 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2782 peer = ofport_dpif_cast(peer_ofport);
2783 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2784 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2786 ofport->peer = peer;
2787 ofport->peer->peer = ofport;
2797 port_run(struct ofport_dpif *ofport)
2799 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2800 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2801 bool enable = netdev_get_carrier(ofport->up.netdev);
2802 bool cfm_enable = false;
2803 bool bfd_enable = false;
2805 ofport->carrier_seq = carrier_seq;
2808 int cfm_opup = cfm_get_opup(ofport->cfm);
2810 cfm_enable = !cfm_get_fault(ofport->cfm);
2812 if (cfm_opup >= 0) {
2813 cfm_enable = cfm_enable && cfm_opup;
2818 bfd_enable = bfd_forwarding(ofport->bfd);
2821 if (ofport->bfd || ofport->cfm) {
2822 enable = enable && (cfm_enable || bfd_enable);
2825 if (ofport->bundle) {
2826 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2827 if (carrier_changed) {
2828 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2832 if (ofport->may_enable != enable) {
2833 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2834 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2837 ofport->may_enable = enable;
2841 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2842 struct ofproto_port *ofproto_port)
2844 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2845 struct dpif_port dpif_port;
2848 if (sset_contains(&ofproto->ghost_ports, devname)) {
2849 const char *type = netdev_get_type_from_name(devname);
2851 /* We may be called before ofproto->up.port_by_name is populated with
2852 * the appropriate ofport. For this reason, we must get the name and
2853 * type from the netdev layer directly. */
2855 const struct ofport *ofport;
2857 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2858 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2859 ofproto_port->name = xstrdup(devname);
2860 ofproto_port->type = xstrdup(type);
2866 if (!sset_contains(&ofproto->ports, devname)) {
2869 error = dpif_port_query_by_name(ofproto->backer->dpif,
2870 devname, &dpif_port);
2872 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2878 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2880 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2881 const char *devname = netdev_get_name(netdev);
2882 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2883 const char *dp_port_name;
2885 if (netdev_vport_is_patch(netdev)) {
2886 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2890 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2891 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2892 odp_port_t port_no = ODPP_NONE;
2895 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2899 if (netdev_get_tunnel_config(netdev)) {
2900 simap_put(&ofproto->backer->tnl_backers,
2901 dp_port_name, odp_to_u32(port_no));
2905 if (netdev_get_tunnel_config(netdev)) {
2906 sset_add(&ofproto->ghost_ports, devname);
2908 sset_add(&ofproto->ports, devname);
2914 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
2916 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2917 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2924 sset_find_and_delete(&ofproto->ghost_ports,
2925 netdev_get_name(ofport->up.netdev));
2926 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2927 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
2928 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
2930 /* The caller is going to close ofport->up.netdev. If this is a
2931 * bonded port, then the bond is using that netdev, so remove it
2932 * from the bond. The client will need to reconfigure everything
2933 * after deleting ports, so then the slave will get re-added. */
2934 bundle_remove(&ofport->up);
2941 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2943 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2946 error = netdev_get_stats(ofport->up.netdev, stats);
2948 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
2949 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2951 ovs_mutex_lock(&ofproto->stats_mutex);
2952 /* ofproto->stats.tx_packets represents packets that we created
2953 * internally and sent to some port (e.g. packets sent with
2954 * ofproto_dpif_send_packet()). Account for them as if they had
2955 * come from OFPP_LOCAL and got forwarded. */
2957 if (stats->rx_packets != UINT64_MAX) {
2958 stats->rx_packets += ofproto->stats.tx_packets;
2961 if (stats->rx_bytes != UINT64_MAX) {
2962 stats->rx_bytes += ofproto->stats.tx_bytes;
2965 /* ofproto->stats.rx_packets represents packets that were received on
2966 * some port and we processed internally and dropped (e.g. STP).
2967 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2969 if (stats->tx_packets != UINT64_MAX) {
2970 stats->tx_packets += ofproto->stats.rx_packets;
2973 if (stats->tx_bytes != UINT64_MAX) {
2974 stats->tx_bytes += ofproto->stats.rx_bytes;
2976 ovs_mutex_unlock(&ofproto->stats_mutex);
2982 struct port_dump_state {
2987 struct ofproto_port port;
2992 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
2994 *statep = xzalloc(sizeof(struct port_dump_state));
2999 port_dump_next(const struct ofproto *ofproto_, void *state_,
3000 struct ofproto_port *port)
3002 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3003 struct port_dump_state *state = state_;
3004 const struct sset *sset;
3005 struct sset_node *node;
3007 if (state->has_port) {
3008 ofproto_port_destroy(&state->port);
3009 state->has_port = false;
3011 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3012 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3015 error = port_query_by_name(ofproto_, node->name, &state->port);
3017 *port = state->port;
3018 state->has_port = true;
3020 } else if (error != ENODEV) {
3025 if (!state->ghost) {
3026 state->ghost = true;
3029 return port_dump_next(ofproto_, state_, port);
3036 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3038 struct port_dump_state *state = state_;
3040 if (state->has_port) {
3041 ofproto_port_destroy(&state->port);
3048 port_poll(const struct ofproto *ofproto_, char **devnamep)
3050 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3052 if (ofproto->port_poll_errno) {
3053 int error = ofproto->port_poll_errno;
3054 ofproto->port_poll_errno = 0;
3058 if (sset_is_empty(&ofproto->port_poll_set)) {
3062 *devnamep = sset_pop(&ofproto->port_poll_set);
3067 port_poll_wait(const struct ofproto *ofproto_)
3069 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3070 dpif_port_poll_wait(ofproto->backer->dpif);
3074 port_is_lacp_current(const struct ofport *ofport_)
3076 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3077 return (ofport->bundle && ofport->bundle->lacp
3078 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3082 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3083 * then delete it entirely. */
3085 rule_expire(struct rule_dpif *rule)
3086 OVS_REQUIRES(ofproto_mutex)
3088 uint16_t hard_timeout, idle_timeout;
3089 long long int now = time_msec();
3092 ovs_assert(!rule->up.pending);
3094 hard_timeout = rule->up.hard_timeout;
3095 idle_timeout = rule->up.idle_timeout;
3097 /* Has 'rule' expired? */
3099 long long int modified;
3101 ovs_mutex_lock(&rule->up.mutex);
3102 modified = rule->up.modified;
3103 ovs_mutex_unlock(&rule->up.mutex);
3105 if (now > modified + hard_timeout * 1000) {
3106 reason = OFPRR_HARD_TIMEOUT;
3110 if (reason < 0 && idle_timeout) {
3113 ovs_mutex_lock(&rule->stats_mutex);
3114 used = rule->stats.used;
3115 ovs_mutex_unlock(&rule->stats_mutex);
3117 if (now > used + idle_timeout * 1000) {
3118 reason = OFPRR_IDLE_TIMEOUT;
3123 COVERAGE_INC(ofproto_dpif_expired);
3124 ofproto_rule_expire(&rule->up, reason);
3128 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3129 * 'flow' must reflect the data in 'packet'. */
3131 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3132 const struct flow *flow,
3133 struct rule_dpif *rule,
3134 const struct ofpact *ofpacts, size_t ofpacts_len,
3135 struct ofpbuf *packet)
3137 struct dpif_flow_stats stats;
3138 struct xlate_out xout;
3139 struct xlate_in xin;
3141 struct dpif_execute execute;
3144 ovs_assert((rule != NULL) != (ofpacts != NULL));
3146 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3149 rule_dpif_credit_stats(rule, &stats);
3152 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
3153 xin.ofpacts = ofpacts;
3154 xin.ofpacts_len = ofpacts_len;
3155 xin.resubmit_stats = &stats;
3156 xlate_actions(&xin, &xout);
3158 execute.actions = ofpbuf_data(&xout.odp_actions);
3159 execute.actions_len = ofpbuf_size(&xout.odp_actions);
3160 execute.packet = packet;
3161 execute.md = pkt_metadata_from_flow(flow);
3162 execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
3164 /* Fix up in_port. */
3165 in_port = flow->in_port.ofp_port;
3166 if (in_port == OFPP_NONE) {
3167 in_port = OFPP_LOCAL;
3169 execute.md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
3171 error = dpif_execute(ofproto->backer->dpif, &execute);
3173 xlate_out_uninit(&xout);
3179 rule_dpif_credit_stats(struct rule_dpif *rule,
3180 const struct dpif_flow_stats *stats)
3182 ovs_mutex_lock(&rule->stats_mutex);
3183 rule->stats.n_packets += stats->n_packets;
3184 rule->stats.n_bytes += stats->n_bytes;
3185 rule->stats.used = MAX(rule->stats.used, stats->used);
3186 ovs_mutex_unlock(&rule->stats_mutex);
3190 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
3191 OVS_REQUIRES(rule->up.mutex)
3193 return rule->up.flow_cookie;
3197 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
3198 uint16_t hard_timeout)
3200 ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
3203 /* Returns 'rule''s actions. The caller owns a reference on the returned
3204 * actions and must eventually release it (with rule_actions_unref()) to avoid
3206 const struct rule_actions *
3207 rule_dpif_get_actions(const struct rule_dpif *rule)
3209 return rule_get_actions(&rule->up);
3212 /* Lookup 'flow' in table 0 of 'ofproto''s classifier.
3213 * If 'wc' is non-null, sets the fields that were relevant as part of
3214 * the lookup. Returns the table_id where a match or miss occurred.
3216 * The return value will be zero unless there was a miss and
3217 * OFPTC11_TABLE_MISS_CONTINUE is in effect for the sequence of tables
3218 * where misses occur.
3220 * The rule is returned in '*rule', which is valid at least until the next
3221 * RCU quiescent period. If the '*rule' needs to stay around longer,
3222 * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3223 * on it before this returns. */
3225 rule_dpif_lookup(struct ofproto_dpif *ofproto, struct flow *flow,
3226 struct flow_wildcards *wc, struct rule_dpif **rule,
3229 enum rule_dpif_lookup_verdict verdict;
3230 enum ofputil_port_config config = 0;
3233 if (ofproto_dpif_get_enable_recirc(ofproto)) {
3234 /* Always exactly match recirc_id since datapath supports
3237 wc->masks.recirc_id = UINT32_MAX;
3239 table_id = rule_dpif_lookup_get_init_table_id(flow);
3244 verdict = rule_dpif_lookup_from_table(ofproto, flow, wc, true,
3245 &table_id, rule, take_ref);
3248 case RULE_DPIF_LOOKUP_VERDICT_MATCH:
3250 case RULE_DPIF_LOOKUP_VERDICT_CONTROLLER: {
3251 struct ofport_dpif *port;
3253 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
3255 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
3256 flow->in_port.ofp_port);
3258 config = port ? port->up.pp.config : 0;
3261 case RULE_DPIF_LOOKUP_VERDICT_DROP:
3262 config = OFPUTIL_PC_NO_PACKET_IN;
3264 case RULE_DPIF_LOOKUP_VERDICT_DEFAULT:
3265 if (!connmgr_wants_packet_in_on_miss(ofproto->up.connmgr)) {
3266 config = OFPUTIL_PC_NO_PACKET_IN;
3273 choose_miss_rule(config, ofproto->miss_rule,
3274 ofproto->no_packet_in_rule, rule, take_ref);
3278 /* The returned rule is valid at least until the next RCU quiescent period.
3279 * If the '*rule' needs to stay around longer, a non-zero 'take_ref' must be
3280 * passed in to cause a reference to be taken on it before this returns. */
3281 static struct rule_dpif *
3282 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, uint8_t table_id,
3283 const struct flow *flow, struct flow_wildcards *wc,
3286 struct classifier *cls = &ofproto->up.tables[table_id].cls;
3287 const struct cls_rule *cls_rule;
3288 struct rule_dpif *rule;
3290 fat_rwlock_rdlock(&cls->rwlock);
3291 if (ofproto->up.frag_handling != OFPC_FRAG_NX_MATCH) {
3293 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
3294 if (is_ip_any(flow)) {
3295 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
3299 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
3300 if (ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3301 /* We must pretend that transport ports are unavailable. */
3302 struct flow ofpc_normal_flow = *flow;
3303 ofpc_normal_flow.tp_src = htons(0);
3304 ofpc_normal_flow.tp_dst = htons(0);
3305 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
3307 /* Must be OFPC_FRAG_DROP (we don't have OFPC_FRAG_REASM). */
3308 cls_rule = &ofproto->drop_frags_rule->up.cr;
3311 cls_rule = classifier_lookup(cls, flow, wc);
3314 cls_rule = classifier_lookup(cls, flow, wc);
3317 rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
3319 rule_dpif_ref(rule);
3321 fat_rwlock_unlock(&cls->rwlock);
3326 /* Look up 'flow' in 'ofproto''s classifier starting from table '*table_id'.
3327 * Stores the rule that was found in '*rule', or NULL if none was found.
3328 * Updates 'wc', if nonnull, to reflect the fields that were used during the
3331 * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3332 * if none is found then the table miss configuration for that table is
3333 * honored, which can result in additional lookups in other OpenFlow tables.
3334 * In this case the function updates '*table_id' to reflect the final OpenFlow
3335 * table that was searched.
3337 * If 'honor_table_miss' is false, then only one table lookup occurs, in
3342 * - RULE_DPIF_LOOKUP_VERDICT_MATCH if a rule (in '*rule') was found.
3344 * - RULE_OFPTC_TABLE_MISS_CONTROLLER if no rule was found and either:
3345 * + 'honor_table_miss' is false
3346 * + a table miss configuration specified that the packet should be
3347 * sent to the controller in this case.
3349 * - RULE_DPIF_LOOKUP_VERDICT_DROP if no rule was found, 'honor_table_miss'
3350 * is true and a table miss configuration specified that the packet
3351 * should be dropped in this case.
3353 * - RULE_DPIF_LOOKUP_VERDICT_DEFAULT if no rule was found,
3354 * 'honor_table_miss' is true and a table miss configuration has
3355 * not been specified in this case.
3357 * The rule is returned in '*rule', which is valid at least until the next
3358 * RCU quiescent period. If the '*rule' needs to stay around longer,
3359 * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3360 * on it before this returns. */
3361 enum rule_dpif_lookup_verdict
3362 rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto,
3363 const struct flow *flow,
3364 struct flow_wildcards *wc,
3365 bool honor_table_miss,
3366 uint8_t *table_id, struct rule_dpif **rule,
3371 for (next_id = *table_id;
3372 next_id < ofproto->up.n_tables;
3373 next_id++, next_id += (next_id == TBL_INTERNAL))
3375 *table_id = next_id;
3376 *rule = rule_dpif_lookup_in_table(ofproto, *table_id, flow, wc,
3379 return RULE_DPIF_LOOKUP_VERDICT_MATCH;
3380 } else if (!honor_table_miss) {
3381 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3383 switch (ofproto_table_get_config(&ofproto->up, *table_id)) {
3384 case OFPROTO_TABLE_MISS_CONTINUE:
3387 case OFPROTO_TABLE_MISS_CONTROLLER:
3388 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3390 case OFPROTO_TABLE_MISS_DROP:
3391 return RULE_DPIF_LOOKUP_VERDICT_DROP;
3393 case OFPROTO_TABLE_MISS_DEFAULT:
3394 return RULE_DPIF_LOOKUP_VERDICT_DEFAULT;
3399 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3402 /* Given a port configuration (specified as zero if there's no port), chooses
3403 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
3406 * The rule is returned in '*rule', which is valid at least until the next
3407 * RCU quiescent period. If the '*rule' needs to stay around longer,
3408 * a reference must be taken on it (rule_dpif_ref()).
3411 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
3412 struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule,
3415 *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
3417 rule_dpif_ref(*rule);
3422 complete_operation(struct rule_dpif *rule)
3423 OVS_REQUIRES(ofproto_mutex)
3425 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3427 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3428 ofoperation_complete(rule->up.pending, 0);
3431 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
3433 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
3436 static struct rule *
3439 struct rule_dpif *rule = xmalloc(sizeof *rule);
3444 rule_dealloc(struct rule *rule_)
3446 struct rule_dpif *rule = rule_dpif_cast(rule_);
3451 rule_construct(struct rule *rule_)
3452 OVS_NO_THREAD_SAFETY_ANALYSIS
3454 struct rule_dpif *rule = rule_dpif_cast(rule_);
3455 ovs_mutex_init_adaptive(&rule->stats_mutex);
3456 rule->stats.n_packets = 0;
3457 rule->stats.n_bytes = 0;
3458 rule->stats.used = rule->up.modified;
3463 rule_insert(struct rule *rule_)
3464 OVS_REQUIRES(ofproto_mutex)
3466 struct rule_dpif *rule = rule_dpif_cast(rule_);
3467 complete_operation(rule);
3471 rule_delete(struct rule *rule_)
3472 OVS_REQUIRES(ofproto_mutex)
3474 struct rule_dpif *rule = rule_dpif_cast(rule_);
3475 complete_operation(rule);
3479 rule_destruct(struct rule *rule_)
3481 struct rule_dpif *rule = rule_dpif_cast(rule_);
3482 ovs_mutex_destroy(&rule->stats_mutex);
3486 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
3487 long long int *used)
3489 struct rule_dpif *rule = rule_dpif_cast(rule_);
3491 ovs_mutex_lock(&rule->stats_mutex);
3492 *packets = rule->stats.n_packets;
3493 *bytes = rule->stats.n_bytes;
3494 *used = rule->stats.used;
3495 ovs_mutex_unlock(&rule->stats_mutex);
3499 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
3500 struct ofpbuf *packet)
3502 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3504 ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
3508 rule_execute(struct rule *rule, const struct flow *flow,
3509 struct ofpbuf *packet)
3511 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
3512 ofpbuf_delete(packet);
3517 rule_modify_actions(struct rule *rule_, bool reset_counters)
3518 OVS_REQUIRES(ofproto_mutex)
3520 struct rule_dpif *rule = rule_dpif_cast(rule_);
3522 if (reset_counters) {
3523 ovs_mutex_lock(&rule->stats_mutex);
3524 rule->stats.n_packets = 0;
3525 rule->stats.n_bytes = 0;
3526 ovs_mutex_unlock(&rule->stats_mutex);
3529 complete_operation(rule);
3532 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
3534 return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
3537 static struct ofgroup *
3540 struct group_dpif *group = xzalloc(sizeof *group);
3545 group_dealloc(struct ofgroup *group_)
3547 struct group_dpif *group = group_dpif_cast(group_);
3552 group_construct_stats(struct group_dpif *group)
3553 OVS_REQUIRES(group->stats_mutex)
3555 group->packet_count = 0;
3556 group->byte_count = 0;
3557 if (!group->bucket_stats) {
3558 group->bucket_stats = xcalloc(group->up.n_buckets,
3559 sizeof *group->bucket_stats);
3561 memset(group->bucket_stats, 0, group->up.n_buckets *
3562 sizeof *group->bucket_stats);
3567 group_construct(struct ofgroup *group_)
3569 struct group_dpif *group = group_dpif_cast(group_);
3570 const struct ofputil_bucket *bucket;
3572 /* Prevent group chaining because our locking structure makes it hard to
3573 * implement deadlock-free. (See xlate_group_resource_check().) */
3574 LIST_FOR_EACH (bucket, list_node, &group->up.buckets) {
3575 const struct ofpact *a;
3577 OFPACT_FOR_EACH (a, bucket->ofpacts, bucket->ofpacts_len) {
3578 if (a->type == OFPACT_GROUP) {
3579 return OFPERR_OFPGMFC_CHAINING_UNSUPPORTED;
3584 ovs_mutex_init_adaptive(&group->stats_mutex);
3585 ovs_mutex_lock(&group->stats_mutex);
3586 group_construct_stats(group);
3587 ovs_mutex_unlock(&group->stats_mutex);
3592 group_destruct__(struct group_dpif *group)
3593 OVS_REQUIRES(group->stats_mutex)
3595 free(group->bucket_stats);
3596 group->bucket_stats = NULL;
3600 group_destruct(struct ofgroup *group_)
3602 struct group_dpif *group = group_dpif_cast(group_);
3603 ovs_mutex_lock(&group->stats_mutex);
3604 group_destruct__(group);
3605 ovs_mutex_unlock(&group->stats_mutex);
3606 ovs_mutex_destroy(&group->stats_mutex);
3610 group_modify(struct ofgroup *group_, struct ofgroup *victim_)
3612 struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
3613 struct group_dpif *group = group_dpif_cast(group_);
3614 struct group_dpif *victim = group_dpif_cast(victim_);
3616 ovs_mutex_lock(&group->stats_mutex);
3617 if (victim->up.n_buckets < group->up.n_buckets) {
3618 group_destruct__(group);
3620 group_construct_stats(group);
3621 ovs_mutex_unlock(&group->stats_mutex);
3623 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3629 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
3631 struct group_dpif *group = group_dpif_cast(group_);
3633 ovs_mutex_lock(&group->stats_mutex);
3634 ogs->packet_count = group->packet_count;
3635 ogs->byte_count = group->byte_count;
3636 memcpy(ogs->bucket_stats, group->bucket_stats,
3637 group->up.n_buckets * sizeof *group->bucket_stats);
3638 ovs_mutex_unlock(&group->stats_mutex);
3644 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
3645 struct group_dpif **group)
3646 OVS_TRY_RDLOCK(true, (*group)->up.rwlock)
3648 struct ofgroup *ofgroup;
3652 found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
3653 *group = found ? group_dpif_cast(ofgroup) : NULL;
3659 group_dpif_release(struct group_dpif *group)
3660 OVS_RELEASES(group->up.rwlock)
3662 ofproto_group_release(&group->up);
3666 group_dpif_get_buckets(const struct group_dpif *group,
3667 const struct list **buckets)
3669 *buckets = &group->up.buckets;
3672 enum ofp11_group_type
3673 group_dpif_get_type(const struct group_dpif *group)
3675 return group->up.type;
3678 /* Sends 'packet' out 'ofport'.
3679 * May modify 'packet'.
3680 * Returns 0 if successful, otherwise a positive errno value. */
3682 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
3684 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3687 error = xlate_send_packet(ofport, packet);
3689 ovs_mutex_lock(&ofproto->stats_mutex);
3690 ofproto->stats.tx_packets++;
3691 ofproto->stats.tx_bytes += ofpbuf_size(packet);
3692 ovs_mutex_unlock(&ofproto->stats_mutex);
3697 set_frag_handling(struct ofproto *ofproto_,
3698 enum ofp_config_flags frag_handling)
3700 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3701 if (frag_handling != OFPC_FRAG_REASM) {
3702 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3710 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
3711 const struct flow *flow,
3712 const struct ofpact *ofpacts, size_t ofpacts_len)
3714 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3716 ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
3717 ofpacts_len, packet);
3724 set_netflow(struct ofproto *ofproto_,
3725 const struct netflow_options *netflow_options)
3727 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3729 if (netflow_options) {
3730 if (!ofproto->netflow) {
3731 ofproto->netflow = netflow_create();
3732 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3734 return netflow_set_options(ofproto->netflow, netflow_options);
3735 } else if (ofproto->netflow) {
3736 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3737 netflow_unref(ofproto->netflow);
3738 ofproto->netflow = NULL;
3745 get_netflow_ids(const struct ofproto *ofproto_,
3746 uint8_t *engine_type, uint8_t *engine_id)
3748 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3750 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
3753 static struct ofproto_dpif *
3754 ofproto_dpif_lookup(const char *name)
3756 struct ofproto_dpif *ofproto;
3758 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
3759 hash_string(name, 0), &all_ofproto_dpifs) {
3760 if (!strcmp(ofproto->up.name, name)) {
3768 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
3769 const char *argv[], void *aux OVS_UNUSED)
3771 struct ofproto_dpif *ofproto;
3774 ofproto = ofproto_dpif_lookup(argv[1]);
3776 unixctl_command_reply_error(conn, "no such bridge");
3779 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3780 mac_learning_flush(ofproto->ml);
3781 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3783 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3784 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3785 mac_learning_flush(ofproto->ml);
3786 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3790 unixctl_command_reply(conn, "table successfully flushed");
3793 static struct ofport_dpif *
3794 ofbundle_get_a_port(const struct ofbundle *bundle)
3796 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
3801 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
3802 const char *argv[], void *aux OVS_UNUSED)
3804 struct ds ds = DS_EMPTY_INITIALIZER;
3805 const struct ofproto_dpif *ofproto;
3806 const struct mac_entry *e;
3808 ofproto = ofproto_dpif_lookup(argv[1]);
3810 unixctl_command_reply_error(conn, "no such bridge");
3814 ds_put_cstr(&ds, " port VLAN MAC Age\n");
3815 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
3816 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3817 struct ofbundle *bundle = e->port.p;
3818 char name[OFP_MAX_PORT_NAME_LEN];
3820 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
3822 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
3823 name, e->vlan, ETH_ADDR_ARGS(e->mac),
3824 mac_entry_age(ofproto->ml, e));
3826 ovs_rwlock_unlock(&ofproto->ml->rwlock);
3827 unixctl_command_reply(conn, ds_cstr(&ds));
3832 struct xlate_out xout;
3833 struct xlate_in xin;
3834 const struct flow *key;
3836 struct flow_wildcards wc;
3841 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
3843 const struct rule_actions *actions;
3846 ds_put_char_multiple(result, '\t', level);
3848 ds_put_cstr(result, "No match\n");
3852 ovs_mutex_lock(&rule->up.mutex);
3853 cookie = rule->up.flow_cookie;
3854 ovs_mutex_unlock(&rule->up.mutex);
3856 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
3857 rule ? rule->up.table_id : 0, ntohll(cookie));
3858 cls_rule_format(&rule->up.cr, result);
3859 ds_put_char(result, '\n');
3861 actions = rule_dpif_get_actions(rule);
3863 ds_put_char_multiple(result, '\t', level);
3864 ds_put_cstr(result, "OpenFlow actions=");
3865 ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
3866 ds_put_char(result, '\n');
3870 trace_format_flow(struct ds *result, int level, const char *title,
3871 struct trace_ctx *trace)
3873 ds_put_char_multiple(result, '\t', level);
3874 ds_put_format(result, "%s: ", title);
3875 /* Do not report unchanged flows for resubmits. */
3876 if ((level > 0 && flow_equal(&trace->xin.flow, &trace->flow))
3877 || (level == 0 && flow_equal(&trace->xin.flow, trace->key))) {
3878 ds_put_cstr(result, "unchanged");
3880 flow_format(result, &trace->xin.flow);
3881 trace->flow = trace->xin.flow;
3883 ds_put_char(result, '\n');
3887 trace_format_regs(struct ds *result, int level, const char *title,
3888 struct trace_ctx *trace)
3892 ds_put_char_multiple(result, '\t', level);
3893 ds_put_format(result, "%s:", title);
3894 for (i = 0; i < FLOW_N_REGS; i++) {
3895 ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
3897 ds_put_char(result, '\n');
3901 trace_format_odp(struct ds *result, int level, const char *title,
3902 struct trace_ctx *trace)
3904 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
3906 ds_put_char_multiple(result, '\t', level);
3907 ds_put_format(result, "%s: ", title);
3908 format_odp_actions(result, ofpbuf_data(odp_actions),
3909 ofpbuf_size(odp_actions));
3910 ds_put_char(result, '\n');
3914 trace_format_megaflow(struct ds *result, int level, const char *title,
3915 struct trace_ctx *trace)
3919 ds_put_char_multiple(result, '\t', level);
3920 ds_put_format(result, "%s: ", title);
3921 flow_wildcards_or(&trace->wc, &trace->xout.wc, &trace->wc);
3922 match_init(&match, trace->key, &trace->wc);
3923 match_format(&match, result, OFP_DEFAULT_PRIORITY);
3924 ds_put_char(result, '\n');
3928 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
3930 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
3931 struct ds *result = trace->result;
3933 ds_put_char(result, '\n');
3934 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
3935 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
3936 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
3937 trace_format_megaflow(result, recurse + 1, "Resubmitted megaflow", trace);
3938 trace_format_rule(result, recurse + 1, rule);
3942 trace_report(struct xlate_in *xin, const char *s, int recurse)
3944 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
3945 struct ds *result = trace->result;
3947 ds_put_char_multiple(result, '\t', recurse);
3948 ds_put_cstr(result, s);
3949 ds_put_char(result, '\n');
3952 /* Parses the 'argc' elements of 'argv', ignoring argv[0]. The following
3953 * forms are supported:
3955 * - [dpname] odp_flow [-generate | packet]
3956 * - bridge br_flow [-generate | packet]
3958 * On success, initializes '*ofprotop' and 'flow' and returns NULL. On failure
3959 * returns a nonnull malloced error message. */
3960 static char * WARN_UNUSED_RESULT
3961 parse_flow_and_packet(int argc, const char *argv[],
3962 struct ofproto_dpif **ofprotop, struct flow *flow,
3963 struct ofpbuf **packetp)
3965 const struct dpif_backer *backer = NULL;
3966 const char *error = NULL;
3968 struct simap port_names = SIMAP_INITIALIZER(&port_names);
3969 struct ofpbuf *packet;
3970 struct ofpbuf odp_key;
3971 struct ofpbuf odp_mask;
3973 ofpbuf_init(&odp_key, 0);
3974 ofpbuf_init(&odp_mask, 0);
3976 /* Handle "-generate" or a hex string as the last argument. */
3977 if (!strcmp(argv[argc - 1], "-generate")) {
3978 packet = ofpbuf_new(0);
3981 error = eth_from_hex(argv[argc - 1], &packet);
3984 } else if (argc == 4) {
3985 /* The 3-argument form must end in "-generate' or a hex string. */
3991 /* odp_flow can have its in_port specified as a name instead of port no.
3992 * We do not yet know whether a given flow is a odp_flow or a br_flow.
3993 * But, to know whether a flow is odp_flow through odp_flow_from_string(),
3994 * we need to create a simap of name to port no. */
3996 const char *dp_type;
3997 if (!strncmp(argv[1], "ovs-", 4)) {
3998 dp_type = argv[1] + 4;
4002 backer = shash_find_data(&all_dpif_backers, dp_type);
4003 } else if (argc == 2) {
4004 struct shash_node *node;
4005 if (shash_count(&all_dpif_backers) == 1) {
4006 node = shash_first(&all_dpif_backers);
4007 backer = node->data;
4010 error = "Syntax error";
4013 if (backer && backer->dpif) {
4014 struct dpif_port dpif_port;
4015 struct dpif_port_dump port_dump;
4016 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
4017 simap_put(&port_names, dpif_port.name,
4018 odp_to_u32(dpif_port.port_no));
4022 /* Parse the flow and determine whether a datapath or
4023 * bridge is specified. If function odp_flow_key_from_string()
4024 * returns 0, the flow is a odp_flow. If function
4025 * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
4026 if (!odp_flow_from_string(argv[argc - 1], &port_names,
4027 &odp_key, &odp_mask)) {
4029 error = "Cannot find the datapath";
4033 if (xlate_receive(backer, NULL, ofpbuf_data(&odp_key),
4034 ofpbuf_size(&odp_key), flow,
4035 ofprotop, NULL, NULL, NULL, NULL)) {
4036 error = "Invalid datapath flow";
4040 char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
4043 m_err = xasprintf("Bad flow syntax: %s", err);
4048 error = "Must specify bridge name";
4052 *ofprotop = ofproto_dpif_lookup(argv[1]);
4054 error = "Unknown bridge name";
4060 /* Generate a packet, if requested. */
4062 if (!ofpbuf_size(packet)) {
4063 flow_compose(packet, flow);
4065 struct pkt_metadata md = pkt_metadata_from_flow(flow);
4067 /* Use the metadata from the flow and the packet argument
4068 * to reconstruct the flow. */
4069 flow_extract(packet, &md, flow);
4074 if (error && !m_err) {
4075 m_err = xstrdup(error);
4078 ofpbuf_delete(packet);
4082 ofpbuf_uninit(&odp_key);
4083 ofpbuf_uninit(&odp_mask);
4084 simap_destroy(&port_names);
4089 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
4090 void *aux OVS_UNUSED)
4092 struct ofproto_dpif *ofproto;
4093 struct ofpbuf *packet;
4097 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4102 ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
4103 unixctl_command_reply(conn, ds_cstr(&result));
4104 ds_destroy(&result);
4105 ofpbuf_delete(packet);
4107 unixctl_command_reply_error(conn, error);
4113 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
4114 const char *argv[], void *aux OVS_UNUSED)
4116 enum ofputil_protocol usable_protocols;
4117 struct ofproto_dpif *ofproto;
4118 bool enforce_consistency;
4119 struct ofpbuf ofpacts;
4120 struct ofpbuf *packet;
4125 /* Three kinds of error return values! */
4131 ofpbuf_init(&ofpacts, 0);
4133 /* Parse actions. */
4134 error = parse_ofpacts(argv[--argc], &ofpacts, &usable_protocols);
4136 unixctl_command_reply_error(conn, error);
4141 /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
4142 * consistency between the flow and the actions. With -consistent, we
4143 * enforce consistency even for a flow supported in OpenFlow 1.0. */
4144 if (!strcmp(argv[1], "-consistent")) {
4145 enforce_consistency = true;
4149 enforce_consistency = false;
4152 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4154 unixctl_command_reply_error(conn, error);
4159 /* Do the same checks as handle_packet_out() in ofproto.c.
4161 * We pass a 'table_id' of 0 to ofproto_check_ofpacts(), which isn't
4162 * strictly correct because these actions aren't in any table, but it's OK
4163 * because it 'table_id' is used only to check goto_table instructions, but
4164 * packet-outs take a list of actions and therefore it can't include
4167 * We skip the "meter" check here because meter is an instruction, not an
4168 * action, and thus cannot appear in ofpacts. */
4169 in_port = ofp_to_u16(flow.in_port.ofp_port);
4170 if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
4171 unixctl_command_reply_error(conn, "invalid in_port");
4174 if (enforce_consistency) {
4175 retval = ofpacts_check_consistency(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts),
4176 &flow, u16_to_ofp(ofproto->up.max_ports),
4177 0, 0, usable_protocols);
4179 retval = ofpacts_check(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &flow,
4180 u16_to_ofp(ofproto->up.max_ports), 0, 0,
4186 ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
4187 unixctl_command_reply_error(conn, ds_cstr(&result));
4191 ofproto_trace(ofproto, &flow, packet,
4192 ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &result);
4193 unixctl_command_reply(conn, ds_cstr(&result));
4196 ds_destroy(&result);
4197 ofpbuf_delete(packet);
4198 ofpbuf_uninit(&ofpacts);
4201 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
4202 * description of the results to 'ds'.
4204 * The trace follows a packet with the specified 'flow' through the flow
4205 * table. 'packet' may be nonnull to trace an actual packet, with consequent
4206 * side effects (if it is nonnull then its flow must be 'flow').
4208 * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
4209 * trace, otherwise the actions are determined by a flow table lookup. */
4211 ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
4212 const struct ofpbuf *packet,
4213 const struct ofpact ofpacts[], size_t ofpacts_len,
4216 struct rule_dpif *rule;
4217 struct trace_ctx trace;
4219 ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
4220 ds_put_cstr(ds, "Flow: ");
4221 flow_format(ds, flow);
4222 ds_put_char(ds, '\n');
4224 flow_wildcards_init_catchall(&trace.wc);
4228 rule_dpif_lookup(ofproto, flow, &trace.wc, &rule, false);
4230 trace_format_rule(ds, 0, rule);
4231 if (rule == ofproto->miss_rule) {
4232 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
4233 } else if (rule == ofproto->no_packet_in_rule) {
4234 ds_put_cstr(ds, "\nNo match, packets dropped because "
4235 "OFPPC_NO_PACKET_IN is set on in_port.\n");
4236 } else if (rule == ofproto->drop_frags_rule) {
4237 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
4238 "and the fragment handling mode is \"drop\".\n");
4242 if (rule || ofpacts) {
4244 trace.key = flow; /* Original flow key, used for megaflow. */
4245 trace.flow = *flow; /* May be modified by actions. */
4246 xlate_in_init(&trace.xin, ofproto, flow, rule, ntohs(flow->tcp_flags),
4249 trace.xin.ofpacts = ofpacts;
4250 trace.xin.ofpacts_len = ofpacts_len;
4252 trace.xin.resubmit_hook = trace_resubmit;
4253 trace.xin.report_hook = trace_report;
4255 xlate_actions(&trace.xin, &trace.xout);
4257 ds_put_char(ds, '\n');
4258 trace_format_flow(ds, 0, "Final flow", &trace);
4259 trace_format_megaflow(ds, 0, "Megaflow", &trace);
4261 ds_put_cstr(ds, "Datapath actions: ");
4262 format_odp_actions(ds, ofpbuf_data(&trace.xout.odp_actions),
4263 ofpbuf_size(&trace.xout.odp_actions));
4265 if (trace.xout.slow) {
4266 enum slow_path_reason slow;
4268 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
4269 "slow path because it:");
4271 slow = trace.xout.slow;
4273 enum slow_path_reason bit = rightmost_1bit(slow);
4275 ds_put_format(ds, "\n\t- %s.",
4276 slow_path_reason_to_explanation(bit));
4282 xlate_out_uninit(&trace.xout);
4286 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
4287 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
4288 * to destroy 'ofproto_shash' and free the returned value. */
4289 static const struct shash_node **
4290 get_ofprotos(struct shash *ofproto_shash)
4292 const struct ofproto_dpif *ofproto;
4294 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4295 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
4296 shash_add_nocopy(ofproto_shash, name, ofproto);
4299 return shash_sort(ofproto_shash);
4303 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
4304 const char *argv[] OVS_UNUSED,
4305 void *aux OVS_UNUSED)
4307 struct ds ds = DS_EMPTY_INITIALIZER;
4308 struct shash ofproto_shash;
4309 const struct shash_node **sorted_ofprotos;
4312 shash_init(&ofproto_shash);
4313 sorted_ofprotos = get_ofprotos(&ofproto_shash);
4314 for (i = 0; i < shash_count(&ofproto_shash); i++) {
4315 const struct shash_node *node = sorted_ofprotos[i];
4316 ds_put_format(&ds, "%s\n", node->name);
4319 shash_destroy(&ofproto_shash);
4320 free(sorted_ofprotos);
4322 unixctl_command_reply(conn, ds_cstr(&ds));
4327 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
4329 const struct shash_node **ofprotos;
4330 struct dpif_dp_stats dp_stats;
4331 struct shash ofproto_shash;
4334 dpif_get_dp_stats(backer->dpif, &dp_stats);
4336 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
4337 dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
4339 shash_init(&ofproto_shash);
4340 ofprotos = get_ofprotos(&ofproto_shash);
4341 for (i = 0; i < shash_count(&ofproto_shash); i++) {
4342 struct ofproto_dpif *ofproto = ofprotos[i]->data;
4343 const struct shash_node **ports;
4346 if (ofproto->backer != backer) {
4350 ds_put_format(ds, "\t%s:\n", ofproto->up.name);
4352 ports = shash_sort(&ofproto->up.port_by_name);
4353 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
4354 const struct shash_node *node = ports[j];
4355 struct ofport *ofport = node->data;
4357 odp_port_t odp_port;
4359 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
4362 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
4363 if (odp_port != ODPP_NONE) {
4364 ds_put_format(ds, "%"PRIu32":", odp_port);
4366 ds_put_cstr(ds, "none:");
4369 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
4372 if (!netdev_get_config(ofport->netdev, &config)) {
4373 const struct smap_node **nodes;
4376 nodes = smap_sort(&config);
4377 for (i = 0; i < smap_count(&config); i++) {
4378 const struct smap_node *node = nodes[i];
4379 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
4380 node->key, node->value);
4384 smap_destroy(&config);
4386 ds_put_char(ds, ')');
4387 ds_put_char(ds, '\n');
4391 shash_destroy(&ofproto_shash);
4396 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4397 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4399 struct ds ds = DS_EMPTY_INITIALIZER;
4400 const struct shash_node **backers;
4403 backers = shash_sort(&all_dpif_backers);
4404 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
4405 dpif_show_backer(backers[i]->data, &ds);
4409 unixctl_command_reply(conn, ds_cstr(&ds));
4414 ofproto_dpif_contains_flow(const struct ofproto_dpif *ofproto,
4415 const struct nlattr *key, size_t key_len)
4417 struct ofproto_dpif *ofp;
4420 xlate_receive(ofproto->backer, NULL, key, key_len, &flow, &ofp,
4421 NULL, NULL, NULL, NULL);
4422 return ofp == ofproto;
4426 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
4427 int argc OVS_UNUSED, const char *argv[],
4428 void *aux OVS_UNUSED)
4430 struct ds ds = DS_EMPTY_INITIALIZER;
4431 const struct dpif_flow_stats *stats;
4432 const struct ofproto_dpif *ofproto;
4433 struct dpif_flow_dump flow_dump;
4434 const struct nlattr *actions;
4435 const struct nlattr *mask;
4436 const struct nlattr *key;
4440 bool verbosity = false;
4441 struct dpif_port dpif_port;
4442 struct dpif_port_dump port_dump;
4443 struct hmap portno_names;
4447 ofproto = ofproto_dpif_lookup(argv[argc - 1]);
4449 unixctl_command_reply_error(conn, "no such bridge");
4453 if (argc > 2 && !strcmp(argv[1], "-m")) {
4457 hmap_init(&portno_names);
4458 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
4459 odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
4463 error = dpif_flow_dump_start(&flow_dump, ofproto->backer->dpif);
4467 dpif_flow_dump_state_init(ofproto->backer->dpif, &state);
4468 while (dpif_flow_dump_next(&flow_dump, state, &key, &key_len,
4469 &mask, &mask_len, &actions, &actions_len,
4471 if (!ofproto_dpif_contains_flow(ofproto, key, key_len)) {
4475 odp_flow_format(key, key_len, mask, mask_len, &portno_names, &ds,
4477 ds_put_cstr(&ds, ", ");
4478 dpif_flow_stats_format(stats, &ds);
4479 ds_put_cstr(&ds, ", actions:");
4480 format_odp_actions(&ds, actions, actions_len);
4481 ds_put_char(&ds, '\n');
4483 dpif_flow_dump_state_uninit(ofproto->backer->dpif, state);
4484 error = dpif_flow_dump_done(&flow_dump);
4489 ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
4490 unixctl_command_reply_error(conn, ds_cstr(&ds));
4492 unixctl_command_reply(conn, ds_cstr(&ds));
4494 odp_portno_names_destroy(&portno_names);
4495 hmap_destroy(&portno_names);
4500 ofproto_dpif_unixctl_init(void)
4502 static bool registered;
4508 unixctl_command_register(
4510 "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
4511 1, 3, ofproto_unixctl_trace, NULL);
4512 unixctl_command_register(
4513 "ofproto/trace-packet-out",
4514 "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
4515 2, 6, ofproto_unixctl_trace_actions, NULL);
4516 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
4517 ofproto_unixctl_fdb_flush, NULL);
4518 unixctl_command_register("fdb/show", "bridge", 1, 1,
4519 ofproto_unixctl_fdb_show, NULL);
4520 unixctl_command_register("dpif/dump-dps", "", 0, 0,
4521 ofproto_unixctl_dpif_dump_dps, NULL);
4522 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
4524 unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
4525 ofproto_unixctl_dpif_dump_flows, NULL);
4528 /* Returns true if 'table' is the table used for internal rules,
4529 * false otherwise. */
4531 table_is_internal(uint8_t table_id)
4533 return table_id == TBL_INTERNAL;
4536 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4538 * This is deprecated. It is only for compatibility with broken device drivers
4539 * in old versions of Linux that do not properly support VLANs when VLAN
4540 * devices are not used. When broken device drivers are no longer in
4541 * widespread use, we will delete these interfaces. */
4544 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
4546 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
4547 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
4549 if (realdev_ofp_port == ofport->realdev_ofp_port
4550 && vid == ofport->vlandev_vid) {
4554 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4556 if (ofport->realdev_ofp_port) {
4559 if (realdev_ofp_port && ofport->bundle) {
4560 /* vlandevs are enslaved to their realdevs, so they are not allowed to
4561 * themselves be part of a bundle. */
4562 bundle_set(ofport_->ofproto, ofport->bundle, NULL);
4565 ofport->realdev_ofp_port = realdev_ofp_port;
4566 ofport->vlandev_vid = vid;
4568 if (realdev_ofp_port) {
4569 vsp_add(ofport, realdev_ofp_port, vid);
4576 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
4578 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
4582 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
4583 OVS_EXCLUDED(ofproto->vsp_mutex)
4585 /* hmap_is_empty is thread safe. */
4586 return !hmap_is_empty(&ofproto->realdev_vid_map);
4591 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
4592 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
4593 OVS_REQUIRES(ofproto->vsp_mutex)
4595 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
4596 int vid = vlan_tci_to_vid(vlan_tci);
4597 const struct vlan_splinter *vsp;
4599 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
4600 hash_realdev_vid(realdev_ofp_port, vid),
4601 &ofproto->realdev_vid_map) {
4602 if (vsp->realdev_ofp_port == realdev_ofp_port
4603 && vsp->vid == vid) {
4604 return vsp->vlandev_ofp_port;
4608 return realdev_ofp_port;
4611 /* Returns the OFP port number of the Linux VLAN device that corresponds to
4612 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
4613 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
4614 * 'vlan_tci' 9, it would return the port number of eth0.9.
4616 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
4617 * function just returns its 'realdev_ofp_port' argument. */
4619 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
4620 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
4621 OVS_EXCLUDED(ofproto->vsp_mutex)
4625 /* hmap_is_empty is thread safe, see if we can return immediately. */
4626 if (hmap_is_empty(&ofproto->realdev_vid_map)) {
4627 return realdev_ofp_port;
4629 ovs_mutex_lock(&ofproto->vsp_mutex);
4630 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
4631 ovs_mutex_unlock(&ofproto->vsp_mutex);
4635 static struct vlan_splinter *
4636 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
4638 struct vlan_splinter *vsp;
4640 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
4641 hash_ofp_port(vlandev_ofp_port),
4642 &ofproto->vlandev_map) {
4643 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
4651 /* Returns the OpenFlow port number of the "real" device underlying the Linux
4652 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
4653 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
4654 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
4655 * eth0 and store 9 in '*vid'.
4657 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
4658 * VLAN device. Unless VLAN splinters are enabled, this is what this function
4661 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
4662 ofp_port_t vlandev_ofp_port, int *vid)
4663 OVS_REQUIRES(ofproto->vsp_mutex)
4665 if (!hmap_is_empty(&ofproto->vlandev_map)) {
4666 const struct vlan_splinter *vsp;
4668 vsp = vlandev_find(ofproto, vlandev_ofp_port);
4673 return vsp->realdev_ofp_port;
4679 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
4680 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
4681 * 'flow->in_port' to the "real" device backing the VLAN device, sets
4682 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
4683 * always the case unless VLAN splinters are enabled), returns false without
4684 * making any changes. */
4686 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
4687 OVS_EXCLUDED(ofproto->vsp_mutex)
4692 /* hmap_is_empty is thread safe. */
4693 if (hmap_is_empty(&ofproto->vlandev_map)) {
4697 ovs_mutex_lock(&ofproto->vsp_mutex);
4698 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
4699 ovs_mutex_unlock(&ofproto->vsp_mutex);
4704 /* Cause the flow to be processed as if it came in on the real device with
4705 * the VLAN device's VLAN ID. */
4706 flow->in_port.ofp_port = realdev;
4707 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
4712 vsp_remove(struct ofport_dpif *port)
4714 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
4715 struct vlan_splinter *vsp;
4717 ovs_mutex_lock(&ofproto->vsp_mutex);
4718 vsp = vlandev_find(ofproto, port->up.ofp_port);
4720 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
4721 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
4724 port->realdev_ofp_port = 0;
4726 VLOG_ERR("missing vlan device record");
4728 ovs_mutex_unlock(&ofproto->vsp_mutex);
4732 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
4734 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
4736 ovs_mutex_lock(&ofproto->vsp_mutex);
4737 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
4738 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
4739 == realdev_ofp_port)) {
4740 struct vlan_splinter *vsp;
4742 vsp = xmalloc(sizeof *vsp);
4743 vsp->realdev_ofp_port = realdev_ofp_port;
4744 vsp->vlandev_ofp_port = port->up.ofp_port;
4747 port->realdev_ofp_port = realdev_ofp_port;
4749 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
4750 hash_ofp_port(port->up.ofp_port));
4751 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
4752 hash_realdev_vid(realdev_ofp_port, vid));
4754 VLOG_ERR("duplicate vlan device record");
4756 ovs_mutex_unlock(&ofproto->vsp_mutex);
4760 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
4762 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
4763 return ofport ? ofport->odp_port : ODPP_NONE;
4766 struct ofport_dpif *
4767 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
4769 struct ofport_dpif *port;
4771 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
4772 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
4773 &backer->odp_to_ofport_map) {
4774 if (port->odp_port == odp_port) {
4775 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
4780 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
4785 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
4787 struct ofport_dpif *port;
4789 port = odp_port_to_ofport(ofproto->backer, odp_port);
4790 if (port && &ofproto->up == port->up.ofproto) {
4791 return port->up.ofp_port;
4797 struct ofproto_dpif *
4798 ofproto_dpif_recirc_get_ofproto(const struct dpif_backer *backer,
4801 struct dpif_backer_recirc_node *node;
4803 ovs_mutex_lock(&backer->recirc_mutex);
4804 node = CONTAINER_OF(hmap_first_with_hash(&backer->recirc_map, recirc_id),
4805 struct dpif_backer_recirc_node, hmap_node);
4806 ovs_mutex_unlock(&backer->recirc_mutex);
4808 return node ? node->ofproto : NULL;
4812 ofproto_dpif_alloc_recirc_id(struct ofproto_dpif *ofproto)
4814 struct dpif_backer *backer = ofproto->backer;
4815 uint32_t recirc_id = recirc_id_alloc(backer->rid_pool);
4818 struct dpif_backer_recirc_node *node = xmalloc(sizeof *node);
4820 node->recirc_id = recirc_id;
4821 node->ofproto = ofproto;
4823 ovs_mutex_lock(&backer->recirc_mutex);
4824 hmap_insert(&backer->recirc_map, &node->hmap_node, node->recirc_id);
4825 ovs_mutex_unlock(&backer->recirc_mutex);
4832 ofproto_dpif_free_recirc_id(struct ofproto_dpif *ofproto, uint32_t recirc_id)
4834 struct dpif_backer *backer = ofproto->backer;
4835 struct dpif_backer_recirc_node *node;
4837 ovs_mutex_lock(&backer->recirc_mutex);
4838 node = CONTAINER_OF(hmap_first_with_hash(&backer->recirc_map, recirc_id),
4839 struct dpif_backer_recirc_node, hmap_node);
4841 hmap_remove(&backer->recirc_map, &node->hmap_node);
4842 ovs_mutex_unlock(&backer->recirc_mutex);
4843 recirc_id_free(backer->rid_pool, node->recirc_id);
4845 if (node->ofproto != ofproto) {
4846 VLOG_ERR("recirc_id %"PRIu32", freed by incorrect ofproto (%s),"
4847 " expect ofproto (%s)", node->recirc_id, ofproto->up.name,
4848 node->ofproto->up.name);
4851 /* RCU postpone the free, since other threads may be referring
4852 * to 'node' at same time. */
4853 ovsrcu_postpone(free, node);
4855 ovs_mutex_unlock(&backer->recirc_mutex);
4860 ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
4861 const struct match *match, int priority,
4862 const struct ofpbuf *ofpacts,
4863 struct rule **rulep)
4865 struct ofputil_flow_mod fm;
4866 struct rule_dpif *rule;
4870 fm.priority = priority;
4871 fm.new_cookie = htonll(0);
4872 fm.cookie = htonll(0);
4873 fm.cookie_mask = htonll(0);
4874 fm.modify_cookie = false;
4875 fm.table_id = TBL_INTERNAL;
4876 fm.command = OFPFC_ADD;
4877 fm.idle_timeout = 0;
4878 fm.hard_timeout = 0;
4881 fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
4882 fm.ofpacts = ofpbuf_data(ofpacts);
4883 fm.ofpacts_len = ofpbuf_size(ofpacts);
4885 error = ofproto_flow_mod(&ofproto->up, &fm);
4887 VLOG_ERR_RL(&rl, "failed to add internal flow (%s)",
4888 ofperr_to_string(error));
4893 rule = rule_dpif_lookup_in_table(ofproto, TBL_INTERNAL, &fm.match.flow,
4894 &fm.match.wc, false);
4904 ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
4905 struct match *match, int priority)
4907 struct ofputil_flow_mod fm;
4911 fm.priority = priority;
4912 fm.new_cookie = htonll(0);
4913 fm.cookie = htonll(0);
4914 fm.cookie_mask = htonll(0);
4915 fm.modify_cookie = false;
4916 fm.table_id = TBL_INTERNAL;
4917 fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
4918 fm.command = OFPFC_DELETE_STRICT;
4920 error = ofproto_flow_mod(&ofproto->up, &fm);
4922 VLOG_ERR_RL(&rl, "failed to delete internal flow (%s)",
4923 ofperr_to_string(error));
4930 const struct ofproto_class ofproto_dpif_class = {
4944 NULL, /* get_memory_usage. */
4945 type_get_memory_usage,
4964 port_is_lacp_current,
4965 NULL, /* rule_choose_table */
4974 rule_modify_actions,
4988 get_stp_port_status,
4996 is_mirror_output_bundle,
4997 forward_bpdu_changed,
4998 set_mac_table_config,
5000 NULL, /* meter_get_features */
5001 NULL, /* meter_set */
5002 NULL, /* meter_get */
5003 NULL, /* meter_del */
5004 group_alloc, /* group_alloc */
5005 group_construct, /* group_construct */
5006 group_destruct, /* group_destruct */
5007 group_dealloc, /* group_dealloc */
5008 group_modify, /* group_modify */
5009 group_get_stats, /* group_get_stats */