2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
24 #include "byte-order.h"
25 #include "classifier.h"
28 #include "dynamic-string.h"
33 #include "ofp-print.h"
36 #include "ofproto-provider.h"
37 #include "openflow/nicira-ext.h"
38 #include "openflow/openflow.h"
42 #include "poll-loop.h"
46 #include "unaligned.h"
50 VLOG_DEFINE_THIS_MODULE(ofproto);
52 COVERAGE_DEFINE(ofproto_error);
53 COVERAGE_DEFINE(ofproto_flush);
54 COVERAGE_DEFINE(ofproto_no_packet_in);
55 COVERAGE_DEFINE(ofproto_packet_out);
56 COVERAGE_DEFINE(ofproto_queue_req);
57 COVERAGE_DEFINE(ofproto_recv_openflow);
58 COVERAGE_DEFINE(ofproto_reinit_ports);
59 COVERAGE_DEFINE(ofproto_uninstallable);
60 COVERAGE_DEFINE(ofproto_update_port);
63 S_OPENFLOW, /* Processing OpenFlow commands. */
64 S_FLUSH, /* Deleting all flow table rules. */
67 enum ofoperation_type {
73 /* A single OpenFlow request can execute any number of operations. The
74 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
75 * ofconn to which an error reply should be sent if necessary.
77 * ofproto initiates some operations internally. These operations are still
78 * assigned to groups but will not have an associated ofconn. */
80 struct ofproto *ofproto; /* Owning ofproto. */
81 struct list ofproto_node; /* In ofproto's "pending" list. */
82 struct list ops; /* List of "struct ofoperation"s. */
84 /* Data needed to send OpenFlow reply on failure or to send a buffered
87 * If list_is_empty(ofconn_node) then this ofopgroup never had an
88 * associated ofconn or its ofconn's connection dropped after it initiated
89 * the operation. In the latter case 'ofconn' is a wild pointer that
90 * refers to freed memory, so the 'ofconn' member must be used only if
91 * !list_is_empty(ofconn_node).
93 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
94 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
95 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
96 uint32_t buffer_id; /* Buffer id from original request. */
97 int error; /* 0 if no error yet, otherwise error code. */
100 static struct ofopgroup *ofopgroup_create(struct ofproto *);
101 static struct ofopgroup *ofopgroup_create_for_ofconn(struct ofconn *,
102 const struct ofp_header *,
104 static void ofopgroup_submit(struct ofopgroup *);
105 static void ofopgroup_destroy(struct ofopgroup *);
107 /* A single flow table operation. */
109 struct ofopgroup *group; /* Owning group. */
110 struct list group_node; /* In ofopgroup's "ops" list. */
111 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
112 struct rule *rule; /* Rule being operated upon. */
113 enum ofoperation_type type; /* Type of operation. */
114 int status; /* -1 if pending, otherwise 0 or error code. */
115 struct rule *victim; /* OFOPERATION_ADDING: Replaced rule. */
116 union ofp_action *actions; /* OFOPERATION_MODIFYING: Replaced actions. */
117 int n_actions; /* OFOPERATION_MODIFYING: # of old actions. */
118 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
121 static void ofoperation_create(struct ofopgroup *, struct rule *,
122 enum ofoperation_type);
123 static void ofoperation_destroy(struct ofoperation *);
125 static void ofport_destroy__(struct ofport *);
126 static void ofport_destroy(struct ofport *);
128 static uint64_t pick_datapath_id(const struct ofproto *);
129 static uint64_t pick_fallback_dpid(void);
131 static void ofproto_destroy__(struct ofproto *);
133 static void ofproto_rule_destroy__(struct rule *);
134 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
136 static void ofopgroup_destroy(struct ofopgroup *);
138 static int add_flow(struct ofproto *, struct ofconn *, struct flow_mod *,
139 const struct ofp_header *);
141 /* This return value tells handle_openflow() that processing of the current
142 * OpenFlow message must be postponed until some ongoing operations have
145 * This particular value is a good choice because it is negative (so it won't
146 * collide with any errno value or any value returned by ofp_mkerr()) and large
147 * (so it won't accidentally collide with EOF or a negative errno value). */
148 enum { OFPROTO_POSTPONE = -100000 };
150 static bool handle_openflow(struct ofconn *, struct ofpbuf *);
152 static void update_port(struct ofproto *, const char *devname);
153 static int init_ports(struct ofproto *);
154 static void reinit_ports(struct ofproto *);
156 static void ofproto_unixctl_init(void);
158 /* All registered ofproto classes, in probe order. */
159 static const struct ofproto_class **ofproto_classes;
160 static size_t n_ofproto_classes;
161 static size_t allocated_ofproto_classes;
163 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
164 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
166 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
169 ofproto_initialize(void)
175 ofproto_class_register(&ofproto_dpif_class);
179 /* 'type' should be a normalized datapath type, as returned by
180 * ofproto_normalize_type(). Returns the corresponding ofproto_class
181 * structure, or a null pointer if there is none registered for 'type'. */
182 static const struct ofproto_class *
183 ofproto_class_find__(const char *type)
187 ofproto_initialize();
188 for (i = 0; i < n_ofproto_classes; i++) {
189 const struct ofproto_class *class = ofproto_classes[i];
194 class->enumerate_types(&types);
195 found = sset_contains(&types, type);
196 sset_destroy(&types);
202 VLOG_WARN("unknown datapath type %s", type);
206 /* Registers a new ofproto class. After successful registration, new ofprotos
207 * of that type can be created using ofproto_create(). */
209 ofproto_class_register(const struct ofproto_class *new_class)
213 for (i = 0; i < n_ofproto_classes; i++) {
214 if (ofproto_classes[i] == new_class) {
219 if (n_ofproto_classes >= allocated_ofproto_classes) {
220 ofproto_classes = x2nrealloc(ofproto_classes,
221 &allocated_ofproto_classes,
222 sizeof *ofproto_classes);
224 ofproto_classes[n_ofproto_classes++] = new_class;
228 /* Unregisters a datapath provider. 'type' must have been previously
229 * registered and not currently be in use by any ofprotos. After
230 * unregistration new datapaths of that type cannot be opened using
231 * ofproto_create(). */
233 ofproto_class_unregister(const struct ofproto_class *class)
237 for (i = 0; i < n_ofproto_classes; i++) {
238 if (ofproto_classes[i] == class) {
239 for (i++; i < n_ofproto_classes; i++) {
240 ofproto_classes[i - 1] = ofproto_classes[i];
246 VLOG_WARN("attempted to unregister an ofproto class that is not "
251 /* Clears 'types' and enumerates all registered ofproto types into it. The
252 * caller must first initialize the sset. */
254 ofproto_enumerate_types(struct sset *types)
258 ofproto_initialize();
259 for (i = 0; i < n_ofproto_classes; i++) {
260 ofproto_classes[i]->enumerate_types(types);
264 /* Returns the fully spelled out name for the given ofproto 'type'.
266 * Normalized type string can be compared with strcmp(). Unnormalized type
267 * string might be the same even if they have different spellings. */
269 ofproto_normalize_type(const char *type)
271 return type && type[0] ? type : "system";
274 /* Clears 'names' and enumerates the names of all known created ofprotos with
275 * the given 'type'. The caller must first initialize the sset. Returns 0 if
276 * successful, otherwise a positive errno value.
278 * Some kinds of datapaths might not be practically enumerable. This is not
279 * considered an error. */
281 ofproto_enumerate_names(const char *type, struct sset *names)
283 const struct ofproto_class *class = ofproto_class_find__(type);
284 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
288 ofproto_create(const char *datapath_name, const char *datapath_type,
289 struct ofproto **ofprotop)
291 const struct ofproto_class *class;
292 struct ofproto *ofproto;
299 ofproto_initialize();
300 ofproto_unixctl_init();
302 datapath_type = ofproto_normalize_type(datapath_type);
303 class = ofproto_class_find__(datapath_type);
305 VLOG_WARN("could not create datapath %s of unknown type %s",
306 datapath_name, datapath_type);
310 ofproto = class->alloc();
312 VLOG_ERR("failed to allocate datapath %s of type %s",
313 datapath_name, datapath_type);
318 memset(ofproto, 0, sizeof *ofproto);
319 ofproto->ofproto_class = class;
320 ofproto->name = xstrdup(datapath_name);
321 ofproto->type = xstrdup(datapath_type);
322 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
323 hash_string(ofproto->name, 0));
324 ofproto->datapath_id = 0;
325 ofproto_set_flow_eviction_threshold(ofproto,
326 OFPROTO_FLOW_EVICTON_THRESHOLD_DEFAULT);
327 ofproto->fallback_dpid = pick_fallback_dpid();
328 ofproto->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
329 ofproto->hw_desc = xstrdup(DEFAULT_HW_DESC);
330 ofproto->sw_desc = xstrdup(DEFAULT_SW_DESC);
331 ofproto->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
332 ofproto->dp_desc = xstrdup(DEFAULT_DP_DESC);
333 hmap_init(&ofproto->ports);
334 shash_init(&ofproto->port_by_name);
335 ofproto->tables = NULL;
336 ofproto->n_tables = 0;
337 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
338 ofproto->state = S_OPENFLOW;
339 list_init(&ofproto->pending);
340 hmap_init(&ofproto->deletions);
342 error = ofproto->ofproto_class->construct(ofproto, &n_tables);
344 VLOG_ERR("failed to open datapath %s: %s",
345 datapath_name, strerror(error));
346 ofproto_destroy__(ofproto);
350 assert(n_tables >= 1 && n_tables <= 255);
351 ofproto->n_tables = n_tables;
352 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
353 for (i = 0; i < n_tables; i++) {
354 classifier_init(&ofproto->tables[i]);
357 ofproto->datapath_id = pick_datapath_id(ofproto);
358 VLOG_INFO("using datapath ID %016"PRIx64, ofproto->datapath_id);
366 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
368 uint64_t old_dpid = p->datapath_id;
369 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
370 if (p->datapath_id != old_dpid) {
371 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
373 /* Force all active connections to reconnect, since there is no way to
374 * notify a controller that the datapath ID has changed. */
375 ofproto_reconnect_controllers(p);
380 ofproto_set_controllers(struct ofproto *p,
381 const struct ofproto_controller *controllers,
382 size_t n_controllers)
384 connmgr_set_controllers(p->connmgr, controllers, n_controllers);
388 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
390 connmgr_set_fail_mode(p->connmgr, fail_mode);
393 /* Drops the connections between 'ofproto' and all of its controllers, forcing
394 * them to reconnect. */
396 ofproto_reconnect_controllers(struct ofproto *ofproto)
398 connmgr_reconnect(ofproto->connmgr);
401 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
402 * in-band control should guarantee access, in the same way that in-band
403 * control guarantees access to OpenFlow controllers. */
405 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
406 const struct sockaddr_in *extras, size_t n)
408 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
411 /* Sets the OpenFlow queue used by flows set up by in-band control on
412 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
413 * flows will use the default queue. */
415 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
417 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
420 /* Sets the number of flows at which eviction from the kernel flow table
423 ofproto_set_flow_eviction_threshold(struct ofproto *ofproto, unsigned threshold)
425 if (threshold < OFPROTO_FLOW_EVICTION_THRESHOLD_MIN) {
426 ofproto->flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_MIN;
428 ofproto->flow_eviction_threshold = threshold;
433 ofproto_set_desc(struct ofproto *p,
434 const char *mfr_desc, const char *hw_desc,
435 const char *sw_desc, const char *serial_desc,
438 struct ofp_desc_stats *ods;
441 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
442 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
443 sizeof ods->mfr_desc);
446 p->mfr_desc = xstrdup(mfr_desc);
449 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
450 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
451 sizeof ods->hw_desc);
454 p->hw_desc = xstrdup(hw_desc);
457 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
458 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
459 sizeof ods->sw_desc);
462 p->sw_desc = xstrdup(sw_desc);
465 if (strlen(serial_desc) >= sizeof ods->serial_num) {
466 VLOG_WARN("truncating serial_desc, must be less than %zu "
468 sizeof ods->serial_num);
470 free(p->serial_desc);
471 p->serial_desc = xstrdup(serial_desc);
474 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
475 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
476 sizeof ods->dp_desc);
479 p->dp_desc = xstrdup(dp_desc);
484 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
486 return connmgr_set_snoops(ofproto->connmgr, snoops);
490 ofproto_set_netflow(struct ofproto *ofproto,
491 const struct netflow_options *nf_options)
493 if (nf_options && sset_is_empty(&nf_options->collectors)) {
497 if (ofproto->ofproto_class->set_netflow) {
498 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
500 return nf_options ? EOPNOTSUPP : 0;
505 ofproto_set_sflow(struct ofproto *ofproto,
506 const struct ofproto_sflow_options *oso)
508 if (oso && sset_is_empty(&oso->targets)) {
512 if (ofproto->ofproto_class->set_sflow) {
513 return ofproto->ofproto_class->set_sflow(ofproto, oso);
515 return oso ? EOPNOTSUPP : 0;
519 /* Connectivity Fault Management configuration. */
521 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
523 ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
525 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
526 if (ofport && ofproto->ofproto_class->set_cfm) {
527 ofproto->ofproto_class->set_cfm(ofport, NULL);
531 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
532 * basic configuration from the configuration members in 'cfm', and the remote
533 * maintenance point ID from remote_mpid. Ignores the statistics members of
536 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
538 ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
539 const struct cfm_settings *s)
541 struct ofport *ofport;
544 ofport = ofproto_get_port(ofproto, ofp_port);
546 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
547 ofproto->name, ofp_port);
551 /* XXX: For configuration simplicity, we only support one remote_mpid
552 * outside of the CFM module. It's not clear if this is the correct long
553 * term solution or not. */
554 error = (ofproto->ofproto_class->set_cfm
555 ? ofproto->ofproto_class->set_cfm(ofport, s)
558 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
559 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
564 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
565 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
566 * 0 if LACP partner information is not current (generally indicating a
567 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
569 ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
571 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
572 return (ofport && ofproto->ofproto_class->port_is_lacp_current
573 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
579 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
580 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
581 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
582 * configuration plus, if there is more than one slave, a bonding
585 * If 'aux' is already registered then this function updates its configuration
586 * to 's'. Otherwise, this function registers a new bundle.
588 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
591 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
592 const struct ofproto_bundle_settings *s)
594 return (ofproto->ofproto_class->bundle_set
595 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
599 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
600 * If no such bundle has been registered, this has no effect. */
602 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
604 return ofproto_bundle_register(ofproto, aux, NULL);
608 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
609 * If 'aux' is already registered then this function updates its configuration
610 * to 's'. Otherwise, this function registers a new mirror.
612 * Mirrors affect only the treatment of packets output to the OFPP_NORMAL
615 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
616 const struct ofproto_mirror_settings *s)
618 return (ofproto->ofproto_class->mirror_set
619 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
623 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
624 * If no mirror has been registered, this has no effect. */
626 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
628 return ofproto_mirror_register(ofproto, aux, NULL);
631 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
632 * which all packets are flooded, instead of using MAC learning. If
633 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
635 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
638 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
640 return (ofproto->ofproto_class->set_flood_vlans
641 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
645 /* Returns true if 'aux' is a registered bundle that is currently in use as the
646 * output for a mirror. */
648 ofproto_is_mirror_output_bundle(struct ofproto *ofproto, void *aux)
650 return (ofproto->ofproto_class->is_mirror_output_bundle
651 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
656 ofproto_has_snoops(const struct ofproto *ofproto)
658 return connmgr_has_snoops(ofproto->connmgr);
662 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
664 connmgr_get_snoops(ofproto->connmgr, snoops);
668 ofproto_flush__(struct ofproto *ofproto)
670 struct classifier *table;
671 struct ofopgroup *group;
673 if (ofproto->ofproto_class->flush) {
674 ofproto->ofproto_class->flush(ofproto);
677 group = ofopgroup_create(ofproto);
678 for (table = ofproto->tables; table < &ofproto->tables[ofproto->n_tables];
680 struct rule *rule, *next_rule;
681 struct cls_cursor cursor;
683 cls_cursor_init(&cursor, table, NULL);
684 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
685 if (!rule->pending) {
686 ofoperation_create(group, rule, OFOPERATION_DELETE);
687 classifier_remove(table, &rule->cr);
688 ofproto->ofproto_class->rule_destruct(rule);
692 ofopgroup_submit(group);
696 ofproto_destroy__(struct ofproto *ofproto)
700 assert(list_is_empty(&ofproto->pending));
702 connmgr_destroy(ofproto->connmgr);
704 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
707 free(ofproto->mfr_desc);
708 free(ofproto->hw_desc);
709 free(ofproto->sw_desc);
710 free(ofproto->serial_desc);
711 free(ofproto->dp_desc);
712 hmap_destroy(&ofproto->ports);
713 shash_destroy(&ofproto->port_by_name);
715 for (i = 0; i < ofproto->n_tables; i++) {
716 assert(classifier_is_empty(&ofproto->tables[i]));
717 classifier_destroy(&ofproto->tables[i]);
719 free(ofproto->tables);
721 hmap_destroy(&ofproto->deletions);
723 ofproto->ofproto_class->dealloc(ofproto);
727 ofproto_destroy(struct ofproto *p)
729 struct ofport *ofport, *next_ofport;
736 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
737 ofport_destroy(ofport);
740 p->ofproto_class->destruct(p);
741 ofproto_destroy__(p);
744 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
745 * kernel datapath, for example, this destroys the datapath in the kernel, and
746 * with the netdev-based datapath, it tears down the data structures that
747 * represent the datapath.
749 * The datapath should not be currently open as an ofproto. */
751 ofproto_delete(const char *name, const char *type)
753 const struct ofproto_class *class = ofproto_class_find__(type);
754 return (!class ? EAFNOSUPPORT
755 : !class->del ? EACCES
756 : class->del(type, name));
760 process_port_change(struct ofproto *ofproto, int error, char *devname)
762 if (error == ENOBUFS) {
763 reinit_ports(ofproto);
765 update_port(ofproto, devname);
771 ofproto_run(struct ofproto *p)
773 struct ofport *ofport;
777 error = p->ofproto_class->run(p);
778 if (error == ENODEV) {
779 /* Someone destroyed the datapath behind our back. The caller
780 * better destroy us and give up, because we're just going to
781 * spin from here on out. */
782 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
783 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
788 if (p->ofproto_class->port_poll) {
789 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
790 process_port_change(p, error, devname);
794 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
795 unsigned int change_seq = netdev_change_seq(ofport->netdev);
796 if (ofport->change_seq != change_seq) {
797 ofport->change_seq = change_seq;
798 update_port(p, netdev_get_name(ofport->netdev));
805 connmgr_run(p->connmgr, handle_openflow);
809 connmgr_run(p->connmgr, NULL);
811 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
812 connmgr_flushed(p->connmgr);
813 p->state = S_OPENFLOW;
825 ofproto_wait(struct ofproto *p)
827 struct ofport *ofport;
829 p->ofproto_class->wait(p);
830 if (p->ofproto_class->port_poll_wait) {
831 p->ofproto_class->port_poll_wait(p);
834 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
835 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
836 poll_immediate_wake();
842 connmgr_wait(p->connmgr, true);
846 connmgr_wait(p->connmgr, false);
847 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
848 poll_immediate_wake();
855 ofproto_is_alive(const struct ofproto *p)
857 return connmgr_has_controllers(p->connmgr);
861 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
864 connmgr_get_controller_info(ofproto->connmgr, info);
868 ofproto_free_ofproto_controller_info(struct shash *info)
870 connmgr_free_controller_info(info);
873 /* Makes a deep copy of 'old' into 'port'. */
875 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
877 port->name = xstrdup(old->name);
878 port->type = xstrdup(old->type);
879 port->ofp_port = old->ofp_port;
882 /* Frees memory allocated to members of 'ofproto_port'.
884 * Do not call this function on an ofproto_port obtained from
885 * ofproto_port_dump_next(): that function retains ownership of the data in the
888 ofproto_port_destroy(struct ofproto_port *ofproto_port)
890 free(ofproto_port->name);
891 free(ofproto_port->type);
894 /* Initializes 'dump' to begin dumping the ports in an ofproto.
896 * This function provides no status indication. An error status for the entire
897 * dump operation is provided when it is completed by calling
898 * ofproto_port_dump_done().
901 ofproto_port_dump_start(struct ofproto_port_dump *dump,
902 const struct ofproto *ofproto)
904 dump->ofproto = ofproto;
905 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
909 /* Attempts to retrieve another port from 'dump', which must have been created
910 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
911 * 'port' and returns true. On failure, returns false.
913 * Failure might indicate an actual error or merely that the last port has been
914 * dumped. An error status for the entire dump operation is provided when it
915 * is completed by calling ofproto_port_dump_done().
917 * The ofproto owns the data stored in 'port'. It will remain valid until at
918 * least the next time 'dump' is passed to ofproto_port_dump_next() or
919 * ofproto_port_dump_done(). */
921 ofproto_port_dump_next(struct ofproto_port_dump *dump,
922 struct ofproto_port *port)
924 const struct ofproto *ofproto = dump->ofproto;
930 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
933 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
939 /* Completes port table dump operation 'dump', which must have been created
940 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
941 * error-free, otherwise a positive errno value describing the problem. */
943 ofproto_port_dump_done(struct ofproto_port_dump *dump)
945 const struct ofproto *ofproto = dump->ofproto;
947 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
950 return dump->error == EOF ? 0 : dump->error;
953 /* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
954 * and sets '*ofp_portp' to the new port's OpenFlow port number (if 'ofp_portp'
955 * is non-null). On failure, returns a positive errno value and sets
956 * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
958 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
964 error = ofproto->ofproto_class->port_add(ofproto, netdev, &ofp_port);
966 update_port(ofproto, netdev_get_name(netdev));
969 *ofp_portp = error ? OFPP_NONE : ofp_port;
974 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
975 * initializes '*port' appropriately; on failure, returns a positive errno
978 * The caller owns the data in 'ofproto_port' and must free it with
979 * ofproto_port_destroy() when it is no longer needed. */
981 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
982 struct ofproto_port *port)
986 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
988 memset(port, 0, sizeof *port);
993 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
994 * Returns 0 if successful, otherwise a positive errno. */
996 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
998 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
999 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1002 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1003 if (!error && ofport) {
1004 /* 'name' is the netdev's name and update_port() is going to close the
1005 * netdev. Just in case update_port() refers to 'name' after it
1006 * destroys 'ofport', make a copy of it around the update_port()
1008 char *devname = xstrdup(name);
1009 update_port(ofproto, devname);
1015 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1016 * performs the 'n_actions' actions in 'actions'. The new flow will not
1019 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1020 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1021 * controllers; otherwise, it will be hidden.
1023 * The caller retains ownership of 'cls_rule' and 'actions'.
1025 * This is a helper function for in-band control and fail-open. */
1027 ofproto_add_flow(struct ofproto *ofproto, const struct cls_rule *cls_rule,
1028 const union ofp_action *actions, size_t n_actions)
1030 const struct rule *rule;
1032 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1033 &ofproto->tables[0], cls_rule));
1034 if (!rule || !ofputil_actions_equal(rule->actions, rule->n_actions,
1035 actions, n_actions)) {
1038 memset(&fm, 0, sizeof fm);
1040 fm.buffer_id = UINT32_MAX;
1041 fm.actions = (union ofp_action *) actions;
1042 fm.n_actions = n_actions;
1043 add_flow(ofproto, NULL, &fm, NULL);
1047 /* Searches for a rule with matching criteria exactly equal to 'target' in
1048 * ofproto's table 0 and, if it finds one, deletes it.
1050 * This is a helper function for in-band control and fail-open. */
1052 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1056 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1057 &ofproto->tables[0], target));
1059 /* No such rule -> success. */
1061 } else if (rule->pending) {
1062 /* An operation on the rule is already pending -> failure.
1063 * Caller must retry later if it's important. */
1066 /* Initiate deletion -> success. */
1067 struct ofopgroup *group = ofopgroup_create(ofproto);
1068 ofoperation_create(group, rule, OFOPERATION_DELETE);
1069 classifier_remove(&ofproto->tables[rule->table_id], &rule->cr);
1070 rule->ofproto->ofproto_class->rule_destruct(rule);
1071 ofopgroup_submit(group);
1077 /* Starts the process of deleting all of the flows from all of ofproto's flow
1078 * tables and then reintroducing the flows required by in-band control and
1079 * fail-open. The process will complete in a later call to ofproto_run(). */
1081 ofproto_flush_flows(struct ofproto *ofproto)
1083 COVERAGE_INC(ofproto_flush);
1084 ofproto->state = S_FLUSH;
1088 reinit_ports(struct ofproto *p)
1090 struct ofproto_port_dump dump;
1091 struct sset devnames;
1092 struct ofport *ofport;
1093 struct ofproto_port ofproto_port;
1094 const char *devname;
1096 COVERAGE_INC(ofproto_reinit_ports);
1098 sset_init(&devnames);
1099 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1100 sset_add(&devnames, netdev_get_name(ofport->netdev));
1102 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1103 sset_add(&devnames, ofproto_port.name);
1106 SSET_FOR_EACH (devname, &devnames) {
1107 update_port(p, devname);
1109 sset_destroy(&devnames);
1112 /* Opens and returns a netdev for 'ofproto_port', or a null pointer if the
1113 * netdev cannot be opened. On success, also fills in 'opp'. */
1114 static struct netdev *
1115 ofport_open(const struct ofproto_port *ofproto_port, struct ofp_phy_port *opp)
1117 uint32_t curr, advertised, supported, peer;
1118 enum netdev_flags flags;
1119 struct netdev *netdev;
1122 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1124 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1125 "cannot be opened (%s)",
1126 ofproto_port->name, ofproto_port->ofp_port,
1127 ofproto_port->name, strerror(error));
1131 netdev_get_flags(netdev, &flags);
1132 netdev_get_features(netdev, &curr, &advertised, &supported, &peer);
1134 opp->port_no = htons(ofproto_port->ofp_port);
1135 netdev_get_etheraddr(netdev, opp->hw_addr);
1136 ovs_strzcpy(opp->name, ofproto_port->name, sizeof opp->name);
1137 opp->config = flags & NETDEV_UP ? 0 : htonl(OFPPC_PORT_DOWN);
1138 opp->state = netdev_get_carrier(netdev) ? 0 : htonl(OFPPS_LINK_DOWN);
1139 opp->curr = htonl(curr);
1140 opp->advertised = htonl(advertised);
1141 opp->supported = htonl(supported);
1142 opp->peer = htonl(peer);
1147 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1148 * port number, and 'config' bits other than OFPPC_PORT_DOWN are
1151 ofport_equal(const struct ofp_phy_port *a, const struct ofp_phy_port *b)
1153 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1154 return (!memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1155 && a->state == b->state
1156 && !((a->config ^ b->config) & htonl(OFPPC_PORT_DOWN))
1157 && a->curr == b->curr
1158 && a->advertised == b->advertised
1159 && a->supported == b->supported
1160 && a->peer == b->peer);
1163 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1164 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1165 * one with the same name or port number). */
1167 ofport_install(struct ofproto *p,
1168 struct netdev *netdev, const struct ofp_phy_port *opp)
1170 const char *netdev_name = netdev_get_name(netdev);
1171 struct ofport *ofport;
1174 /* Create ofport. */
1175 ofport = p->ofproto_class->port_alloc();
1180 ofport->ofproto = p;
1181 ofport->netdev = netdev;
1182 ofport->change_seq = netdev_change_seq(netdev);
1184 ofport->ofp_port = ntohs(opp->port_no);
1186 /* Add port to 'p'. */
1187 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
1188 shash_add(&p->port_by_name, netdev_name, ofport);
1190 /* Let the ofproto_class initialize its private data. */
1191 error = p->ofproto_class->port_construct(ofport);
1195 connmgr_send_port_status(p->connmgr, opp, OFPPR_ADD);
1199 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1200 p->name, netdev_name, strerror(error));
1202 ofport_destroy__(ofport);
1204 netdev_close(netdev);
1208 /* Removes 'ofport' from 'p' and destroys it. */
1210 ofport_remove(struct ofport *ofport)
1212 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->opp,
1214 ofport_destroy(ofport);
1217 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1220 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1222 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1224 ofport_remove(port);
1228 /* Updates 'port' within 'ofproto' with the new 'netdev' and 'opp'.
1230 * Does not handle a name or port number change. The caller must implement
1231 * such a change as a delete followed by an add. */
1233 ofport_modified(struct ofport *port, struct ofp_phy_port *opp)
1235 memcpy(port->opp.hw_addr, opp->hw_addr, ETH_ADDR_LEN);
1236 port->opp.config = ((port->opp.config & ~htonl(OFPPC_PORT_DOWN))
1237 | (opp->config & htonl(OFPPC_PORT_DOWN)));
1238 port->opp.state = opp->state;
1239 port->opp.curr = opp->curr;
1240 port->opp.advertised = opp->advertised;
1241 port->opp.supported = opp->supported;
1242 port->opp.peer = opp->peer;
1244 connmgr_send_port_status(port->ofproto->connmgr, &port->opp, OFPPR_MODIFY);
1248 ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
1250 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1252 if (port->ofproto->ofproto_class->set_cfm) {
1253 port->ofproto->ofproto_class->set_cfm(port, NULL);
1255 if (port->ofproto->ofproto_class->bundle_remove) {
1256 port->ofproto->ofproto_class->bundle_remove(port);
1262 ofport_destroy__(struct ofport *port)
1264 struct ofproto *ofproto = port->ofproto;
1265 const char *name = netdev_get_name(port->netdev);
1267 hmap_remove(&ofproto->ports, &port->hmap_node);
1268 shash_delete(&ofproto->port_by_name,
1269 shash_find(&ofproto->port_by_name, name));
1271 netdev_close(port->netdev);
1272 ofproto->ofproto_class->port_dealloc(port);
1276 ofport_destroy(struct ofport *port)
1279 port->ofproto->ofproto_class->port_destruct(port);
1280 ofport_destroy__(port);
1285 ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
1287 struct ofport *port;
1289 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1290 hash_int(ofp_port, 0), &ofproto->ports) {
1291 if (port->ofp_port == ofp_port) {
1299 update_port(struct ofproto *ofproto, const char *name)
1301 struct ofproto_port ofproto_port;
1302 struct ofp_phy_port opp;
1303 struct netdev *netdev;
1304 struct ofport *port;
1306 COVERAGE_INC(ofproto_update_port);
1308 /* Fetch 'name''s location and properties from the datapath. */
1309 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1310 ? ofport_open(&ofproto_port, &opp)
1313 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1314 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1315 struct netdev *old_netdev = port->netdev;
1317 /* 'name' hasn't changed location. Any properties changed? */
1318 if (!ofport_equal(&port->opp, &opp)) {
1319 ofport_modified(port, &opp);
1322 /* Install the newly opened netdev in case it has changed.
1323 * Don't close the old netdev yet in case port_modified has to
1324 * remove a retained reference to it.*/
1325 port->netdev = netdev;
1326 port->change_seq = netdev_change_seq(netdev);
1328 if (port->ofproto->ofproto_class->port_modified) {
1329 port->ofproto->ofproto_class->port_modified(port);
1332 netdev_close(old_netdev);
1334 /* If 'port' is nonnull then its name differs from 'name' and thus
1335 * we should delete it. If we think there's a port named 'name'
1336 * then its port number must be wrong now so delete it too. */
1338 ofport_remove(port);
1340 ofport_remove_with_name(ofproto, name);
1341 ofport_install(ofproto, netdev, &opp);
1344 /* Any port named 'name' is gone now. */
1345 ofport_remove_with_name(ofproto, name);
1347 ofproto_port_destroy(&ofproto_port);
1351 init_ports(struct ofproto *p)
1353 struct ofproto_port_dump dump;
1354 struct ofproto_port ofproto_port;
1356 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1357 uint16_t ofp_port = ofproto_port.ofp_port;
1358 if (ofproto_get_port(p, ofp_port)) {
1359 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1361 } else if (shash_find(&p->port_by_name, ofproto_port.name)) {
1362 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1365 struct ofp_phy_port opp;
1366 struct netdev *netdev;
1368 netdev = ofport_open(&ofproto_port, &opp);
1370 ofport_install(p, netdev, &opp);
1379 ofproto_rule_destroy__(struct rule *rule)
1381 free(rule->actions);
1382 rule->ofproto->ofproto_class->rule_dealloc(rule);
1385 /* This function allows an ofproto implementation to destroy any rules that
1386 * remain when its ->destruct() function is called. The caller must have
1387 * already uninitialized any derived members of 'rule' (step 5 described in the
1388 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
1389 * This function implements steps 6 and 7.
1391 * This function should only be called from an ofproto implementation's
1392 * ->destruct() function. It is not suitable elsewhere. */
1394 ofproto_rule_destroy(struct rule *rule)
1396 assert(!rule->pending);
1397 classifier_remove(&rule->ofproto->tables[rule->table_id], &rule->cr);
1398 ofproto_rule_destroy__(rule);
1401 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1402 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1405 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1407 const union ofp_action *oa;
1410 if (out_port == OFPP_NONE) {
1413 OFPUTIL_ACTION_FOR_EACH_UNSAFE (oa, left, rule->actions, rule->n_actions) {
1414 if (action_outputs_to_port(oa, htons(out_port))) {
1421 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1422 * statistics appropriately. 'packet' must have at least sizeof(struct
1423 * ofp_packet_in) bytes of headroom.
1425 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1426 * with statistics for 'packet' either way.
1428 * Takes ownership of 'packet'. */
1430 rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
1434 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1436 flow_extract(packet, 0, in_port, &flow);
1437 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
1440 /* Returns true if 'rule' should be hidden from the controller.
1442 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1443 * (e.g. by in-band control) and are intentionally hidden from the
1446 rule_is_hidden(const struct rule *rule)
1448 return rule->cr.priority > UINT16_MAX;
1452 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
1454 ofconn_send_reply(ofconn, make_echo_reply(oh));
1459 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
1461 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1462 struct ofp_switch_features *osf;
1464 struct ofport *port;
1468 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip, &actions);
1469 assert(actions & (1 << OFPAT_OUTPUT)); /* sanity check */
1471 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1472 osf->datapath_id = htonll(ofproto->datapath_id);
1473 osf->n_buffers = htonl(pktbuf_capacity());
1474 osf->n_tables = ofproto->n_tables;
1475 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1478 osf->capabilities |= htonl(OFPC_ARP_MATCH_IP);
1480 osf->actions = htonl(actions);
1482 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
1483 ofpbuf_put(buf, &port->opp, sizeof port->opp);
1486 ofconn_send_reply(ofconn, buf);
1491 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
1493 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1495 struct ofp_switch_config *osc;
1499 /* Figure out flags. */
1500 drop_frags = ofproto->ofproto_class->get_drop_frags(ofproto);
1501 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1504 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1505 osc->flags = htons(flags);
1506 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
1507 ofconn_send_reply(ofconn, buf);
1513 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
1515 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1516 uint16_t flags = ntohs(osc->flags);
1518 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
1519 && ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
1520 switch (flags & OFPC_FRAG_MASK) {
1521 case OFPC_FRAG_NORMAL:
1522 ofproto->ofproto_class->set_drop_frags(ofproto, false);
1524 case OFPC_FRAG_DROP:
1525 ofproto->ofproto_class->set_drop_frags(ofproto, true);
1528 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1534 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
1539 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
1540 * error message code (composed with ofp_mkerr()) for the caller to propagate
1541 * upward. Otherwise, returns 0.
1543 * The log message mentions 'msg_type'. */
1545 reject_slave_controller(struct ofconn *ofconn, const char *msg_type)
1547 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
1548 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
1549 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
1550 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
1553 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
1560 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
1562 struct ofproto *p = ofconn_get_ofproto(ofconn);
1563 struct ofp_packet_out *opo;
1564 struct ofpbuf payload, *buffer;
1565 union ofp_action *ofp_actions;
1566 struct ofpbuf request;
1568 size_t n_ofp_actions;
1572 COVERAGE_INC(ofproto_packet_out);
1574 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
1579 /* Get ofp_packet_out. */
1580 ofpbuf_use_const(&request, oh, ntohs(oh->length));
1581 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
1584 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
1585 &ofp_actions, &n_ofp_actions);
1591 if (opo->buffer_id != htonl(UINT32_MAX)) {
1592 error = ofconn_pktbuf_retrieve(ofconn, ntohl(opo->buffer_id),
1594 if (error || !buffer) {
1603 /* Send out packet. */
1604 flow_extract(&payload, 0, ntohs(opo->in_port), &flow);
1605 error = p->ofproto_class->packet_out(p, &payload, &flow,
1606 ofp_actions, n_ofp_actions);
1607 ofpbuf_delete(buffer);
1613 update_port_config(struct ofport *port, ovs_be32 config, ovs_be32 mask)
1615 ovs_be32 old_config = port->opp.config;
1617 mask &= config ^ port->opp.config;
1618 if (mask & htonl(OFPPC_PORT_DOWN)) {
1619 if (config & htonl(OFPPC_PORT_DOWN)) {
1620 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
1622 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
1626 port->opp.config ^= mask & (htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
1627 OFPPC_NO_FLOOD | OFPPC_NO_FWD |
1628 OFPPC_NO_PACKET_IN));
1629 if (port->opp.config != old_config) {
1630 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
1635 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
1637 struct ofproto *p = ofconn_get_ofproto(ofconn);
1638 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
1639 struct ofport *port;
1642 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
1647 port = ofproto_get_port(p, ntohs(opm->port_no));
1649 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
1650 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
1651 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
1653 update_port_config(port, opm->config, opm->mask);
1654 if (opm->advertise) {
1655 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
1662 handle_desc_stats_request(struct ofconn *ofconn,
1663 const struct ofp_stats_msg *request)
1665 struct ofproto *p = ofconn_get_ofproto(ofconn);
1666 struct ofp_desc_stats *ods;
1669 ods = ofputil_make_stats_reply(sizeof *ods, request, &msg);
1670 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
1671 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
1672 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
1673 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
1674 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
1675 ofconn_send_reply(ofconn, msg);
1681 handle_table_stats_request(struct ofconn *ofconn,
1682 const struct ofp_stats_msg *request)
1684 struct ofproto *p = ofconn_get_ofproto(ofconn);
1685 struct ofp_table_stats *ots;
1689 ofputil_make_stats_reply(sizeof(struct ofp_stats_msg), request, &msg);
1691 ots = ofpbuf_put_zeros(msg, sizeof *ots * p->n_tables);
1692 for (i = 0; i < p->n_tables; i++) {
1693 ots[i].table_id = i;
1694 sprintf(ots[i].name, "table%zu", i);
1695 ots[i].wildcards = htonl(OFPFW_ALL);
1696 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
1697 ots[i].active_count = htonl(classifier_count(&p->tables[i]));
1700 p->ofproto_class->get_tables(p, ots);
1702 ofconn_send_reply(ofconn, msg);
1707 append_port_stat(struct ofport *port, struct list *replies)
1709 struct netdev_stats stats;
1710 struct ofp_port_stats *ops;
1712 /* Intentionally ignore return value, since errors will set
1713 * 'stats' to all-1s, which is correct for OpenFlow, and
1714 * netdev_get_stats() will log errors. */
1715 netdev_get_stats(port->netdev, &stats);
1717 ops = ofputil_append_stats_reply(sizeof *ops, replies);
1718 ops->port_no = port->opp.port_no;
1719 memset(ops->pad, 0, sizeof ops->pad);
1720 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
1721 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
1722 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
1723 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
1724 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
1725 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
1726 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
1727 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
1728 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
1729 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
1730 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
1731 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
1735 handle_port_stats_request(struct ofconn *ofconn,
1736 const struct ofp_port_stats_request *psr)
1738 struct ofproto *p = ofconn_get_ofproto(ofconn);
1739 struct ofport *port;
1740 struct list replies;
1742 ofputil_start_stats_reply(&psr->osm, &replies);
1743 if (psr->port_no != htons(OFPP_NONE)) {
1744 port = ofproto_get_port(p, ntohs(psr->port_no));
1746 append_port_stat(port, &replies);
1749 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
1750 append_port_stat(port, &replies);
1754 ofconn_send_replies(ofconn, &replies);
1759 calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec)
1761 long long int msecs = time_msec() - start;
1762 *sec = msecs / 1000;
1763 *nsec = (msecs % 1000) * (1000 * 1000);
1766 static struct classifier *
1767 first_matching_table(struct ofproto *ofproto, uint8_t table_id)
1769 if (table_id == 0xff) {
1770 return &ofproto->tables[0];
1771 } else if (table_id < ofproto->n_tables) {
1772 return &ofproto->tables[table_id];
1774 /* It would probably be better to reply with an error but there doesn't
1775 * seem to be any appropriate value, so that might just be
1777 VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8,
1783 static struct classifier *
1784 next_matching_table(struct ofproto *ofproto,
1785 struct classifier *cls, uint8_t table_id)
1787 return (table_id == 0xff && cls != &ofproto->tables[ofproto->n_tables - 1]
1792 /* Assigns CLS to each classifier table, in turn, that matches TABLE_ID in
1795 * - If TABLE_ID is 0xff, this iterates over every classifier table in
1798 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
1799 * only once, for that table.
1801 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so ofproto logs a warning
1802 * and does not enter the loop at all.
1804 * All parameters are evaluated multiple times.
1806 #define FOR_EACH_MATCHING_TABLE(CLS, TABLE_ID, OFPROTO) \
1807 for ((CLS) = first_matching_table(OFPROTO, TABLE_ID); \
1809 (CLS) = next_matching_table(OFPROTO, CLS, TABLE_ID))
1811 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
1812 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
1813 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
1816 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
1817 * to 'out_port' are included.
1819 * Hidden rules are always omitted.
1821 * Returns 0 on success, otherwise an OpenFlow error code. */
1823 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
1824 const struct cls_rule *match, uint16_t out_port,
1827 struct classifier *cls;
1830 FOR_EACH_MATCHING_TABLE (cls, table_id, ofproto) {
1831 struct cls_cursor cursor;
1834 cls_cursor_init(&cursor, cls, match);
1835 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1836 if (rule->pending) {
1837 return OFPROTO_POSTPONE;
1839 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
1840 list_push_back(rules, &rule->ofproto_node);
1847 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
1848 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
1849 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
1852 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
1853 * to 'out_port' are included.
1855 * Hidden rules are always omitted.
1857 * Returns 0 on success, otherwise an OpenFlow error code. */
1859 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
1860 const struct cls_rule *match, uint16_t out_port,
1863 struct classifier *cls;
1866 FOR_EACH_MATCHING_TABLE (cls, table_id, ofproto) {
1869 rule = rule_from_cls_rule(classifier_find_rule_exactly(cls, match));
1871 if (rule->pending) {
1872 return OFPROTO_POSTPONE;
1874 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
1875 list_push_back(rules, &rule->ofproto_node);
1883 handle_flow_stats_request(struct ofconn *ofconn,
1884 const struct ofp_stats_msg *osm)
1886 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1887 struct flow_stats_request fsr;
1888 struct list replies;
1893 error = ofputil_decode_flow_stats_request(&fsr, &osm->header);
1898 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
1899 fsr.out_port, &rules);
1904 ofputil_start_stats_reply(osm, &replies);
1905 LIST_FOR_EACH (rule, ofproto_node, &rules) {
1906 struct ofputil_flow_stats fs;
1909 fs.cookie = rule->flow_cookie;
1910 fs.table_id = rule->table_id;
1911 calc_flow_duration__(rule->created, &fs.duration_sec,
1913 fs.idle_timeout = rule->idle_timeout;
1914 fs.hard_timeout = rule->hard_timeout;
1915 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
1917 fs.actions = rule->actions;
1918 fs.n_actions = rule->n_actions;
1919 ofputil_append_flow_stats_reply(&fs, &replies);
1921 ofconn_send_replies(ofconn, &replies);
1927 flow_stats_ds(struct rule *rule, struct ds *results)
1929 uint64_t packet_count, byte_count;
1931 rule->ofproto->ofproto_class->rule_get_stats(rule,
1932 &packet_count, &byte_count);
1934 if (rule->table_id != 0) {
1935 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
1937 ds_put_format(results, "duration=%llds, ",
1938 (time_msec() - rule->created) / 1000);
1939 ds_put_format(results, "priority=%u, ", rule->cr.priority);
1940 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
1941 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
1942 cls_rule_format(&rule->cr, results);
1943 ds_put_char(results, ',');
1944 if (rule->n_actions > 0) {
1945 ofp_print_actions(results, rule->actions, rule->n_actions);
1947 ds_put_cstr(results, "drop");
1949 ds_put_cstr(results, "\n");
1952 /* Adds a pretty-printed description of all flows to 'results', including
1953 * hidden flows (e.g., set up by in-band control). */
1955 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
1957 struct classifier *cls;
1959 for (cls = &p->tables[0]; cls < &p->tables[p->n_tables]; cls++) {
1960 struct cls_cursor cursor;
1963 cls_cursor_init(&cursor, cls, NULL);
1964 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1965 flow_stats_ds(rule, results);
1970 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
1971 * '*engine_type' and '*engine_id', respectively. */
1973 ofproto_get_netflow_ids(const struct ofproto *ofproto,
1974 uint8_t *engine_type, uint8_t *engine_id)
1976 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
1979 /* Checks the fault status of CFM for 'ofp_port' within 'ofproto'. Returns 1
1980 * if CFM is faulted (generally indiciating a connectivity problem), 0 if CFM
1981 * is not faulted, and -1 if CFM is not enabled on 'ofp_port'. */
1983 ofproto_port_get_cfm_fault(const struct ofproto *ofproto, uint16_t ofp_port)
1985 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1986 return (ofport && ofproto->ofproto_class->get_cfm_fault
1987 ? ofproto->ofproto_class->get_cfm_fault(ofport)
1992 handle_aggregate_stats_request(struct ofconn *ofconn,
1993 const struct ofp_stats_msg *osm)
1995 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1996 struct flow_stats_request request;
1997 struct ofputil_aggregate_stats stats;
1998 bool unknown_packets, unknown_bytes;
1999 struct ofpbuf *reply;
2004 error = ofputil_decode_flow_stats_request(&request, &osm->header);
2009 error = collect_rules_loose(ofproto, request.table_id, &request.match,
2010 request.out_port, &rules);
2015 memset(&stats, 0, sizeof stats);
2016 unknown_packets = unknown_bytes = false;
2017 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2018 uint64_t packet_count;
2019 uint64_t byte_count;
2021 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
2024 if (packet_count == UINT64_MAX) {
2025 unknown_packets = true;
2027 stats.packet_count += packet_count;
2030 if (byte_count == UINT64_MAX) {
2031 unknown_bytes = true;
2033 stats.byte_count += byte_count;
2038 if (unknown_packets) {
2039 stats.packet_count = UINT64_MAX;
2041 if (unknown_bytes) {
2042 stats.byte_count = UINT64_MAX;
2045 reply = ofputil_encode_aggregate_stats_reply(&stats, osm);
2046 ofconn_send_reply(ofconn, reply);
2051 struct queue_stats_cbdata {
2052 struct ofport *ofport;
2053 struct list replies;
2057 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
2058 const struct netdev_queue_stats *stats)
2060 struct ofp_queue_stats *reply;
2062 reply = ofputil_append_stats_reply(sizeof *reply, &cbdata->replies);
2063 reply->port_no = cbdata->ofport->opp.port_no;
2064 memset(reply->pad, 0, sizeof reply->pad);
2065 reply->queue_id = htonl(queue_id);
2066 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
2067 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
2068 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
2072 handle_queue_stats_dump_cb(uint32_t queue_id,
2073 struct netdev_queue_stats *stats,
2076 struct queue_stats_cbdata *cbdata = cbdata_;
2078 put_queue_stats(cbdata, queue_id, stats);
2082 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
2083 struct queue_stats_cbdata *cbdata)
2085 cbdata->ofport = port;
2086 if (queue_id == OFPQ_ALL) {
2087 netdev_dump_queue_stats(port->netdev,
2088 handle_queue_stats_dump_cb, cbdata);
2090 struct netdev_queue_stats stats;
2092 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
2093 put_queue_stats(cbdata, queue_id, &stats);
2099 handle_queue_stats_request(struct ofconn *ofconn,
2100 const struct ofp_queue_stats_request *qsr)
2102 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2103 struct queue_stats_cbdata cbdata;
2104 struct ofport *port;
2105 unsigned int port_no;
2108 COVERAGE_INC(ofproto_queue_req);
2110 ofputil_start_stats_reply(&qsr->osm, &cbdata.replies);
2112 port_no = ntohs(qsr->port_no);
2113 queue_id = ntohl(qsr->queue_id);
2114 if (port_no == OFPP_ALL) {
2115 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2116 handle_queue_stats_for_port(port, queue_id, &cbdata);
2118 } else if (port_no < OFPP_MAX) {
2119 port = ofproto_get_port(ofproto, port_no);
2121 handle_queue_stats_for_port(port, queue_id, &cbdata);
2124 ofpbuf_list_delete(&cbdata.replies);
2125 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
2127 ofconn_send_replies(ofconn, &cbdata.replies);
2133 is_flow_deletion_pending(const struct ofproto *ofproto,
2134 const struct cls_rule *cls_rule,
2137 if (!hmap_is_empty(&ofproto->deletions)) {
2138 struct ofoperation *op;
2140 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
2141 cls_rule_hash(cls_rule, table_id),
2142 &ofproto->deletions) {
2143 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
2152 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
2153 * in which no matching flow already exists in the flow table.
2155 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
2156 * ofp_actions, to the ofproto's flow table. Returns 0 on success or an
2157 * OpenFlow error code as encoded by ofp_mkerr() on failure.
2159 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2162 add_flow(struct ofproto *ofproto, struct ofconn *ofconn, struct flow_mod *fm,
2163 const struct ofp_header *request)
2165 struct classifier *table;
2166 struct ofopgroup *group;
2167 struct rule *victim;
2171 /* Check for overlap, if requested. */
2172 if (fm->flags & OFPFF_CHECK_OVERLAP) {
2173 struct classifier *cls;
2175 FOR_EACH_MATCHING_TABLE (cls, fm->table_id, ofproto) {
2176 if (classifier_rule_overlaps(cls, &fm->cr)) {
2177 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
2183 if (fm->table_id == 0xff) {
2185 if (ofproto->n_tables > 1) {
2186 error = ofproto->ofproto_class->rule_choose_table(ofproto, &fm->cr,
2191 assert(table_id < ofproto->n_tables);
2192 table = &ofproto->tables[table_id];
2194 table = &ofproto->tables[0];
2196 } else if (fm->table_id < ofproto->n_tables) {
2197 table = &ofproto->tables[fm->table_id];
2199 return ofp_mkerr_nicira(OFPET_FLOW_MOD_FAILED, NXFMFC_BAD_TABLE_ID);
2202 /* Serialize against pending deletion. */
2203 if (is_flow_deletion_pending(ofproto, &fm->cr, table - ofproto->tables)) {
2204 return OFPROTO_POSTPONE;
2207 /* Allocate new rule. */
2208 rule = ofproto->ofproto_class->rule_alloc();
2210 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
2211 ofproto->name, strerror(error));
2214 rule->ofproto = ofproto;
2216 rule->pending = NULL;
2217 rule->flow_cookie = fm->cookie;
2218 rule->created = time_msec();
2219 rule->idle_timeout = fm->idle_timeout;
2220 rule->hard_timeout = fm->hard_timeout;
2221 rule->table_id = table - ofproto->tables;
2222 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
2223 rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions);
2224 rule->n_actions = fm->n_actions;
2226 /* Insert new rule. */
2227 victim = rule_from_cls_rule(classifier_replace(table, &rule->cr));
2228 if (victim && victim->pending) {
2229 error = OFPROTO_POSTPONE;
2232 ? ofopgroup_create_for_ofconn(ofconn, request, fm->buffer_id)
2233 : ofopgroup_create(ofproto));
2234 ofoperation_create(group, rule, OFOPERATION_ADD);
2235 rule->pending->victim = victim;
2237 error = ofproto->ofproto_class->rule_construct(rule);
2239 ofoperation_destroy(rule->pending);
2241 ofopgroup_submit(group);
2244 /* Back out if an error occurred. */
2247 classifier_replace(table, &victim->cr);
2249 classifier_remove(table, &rule->cr);
2251 ofproto_rule_destroy__(rule);
2256 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
2258 /* Modifies the rules listed in 'rules', changing their actions to match those
2261 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2264 * Returns 0 on success, otherwise an OpenFlow error code. */
2266 modify_flows__(struct ofconn *ofconn, const struct flow_mod *fm,
2267 const struct ofp_header *request, struct list *rules)
2269 struct ofopgroup *group;
2272 group = ofopgroup_create_for_ofconn(ofconn, request, fm->buffer_id);
2273 LIST_FOR_EACH (rule, ofproto_node, rules) {
2274 if (!ofputil_actions_equal(fm->actions, fm->n_actions,
2275 rule->actions, rule->n_actions)) {
2276 ofoperation_create(group, rule, OFOPERATION_MODIFY);
2277 rule->pending->actions = rule->actions;
2278 rule->pending->n_actions = rule->n_actions;
2279 rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions);
2280 rule->n_actions = fm->n_actions;
2281 rule->ofproto->ofproto_class->rule_modify_actions(rule);
2283 rule->flow_cookie = fm->cookie;
2285 ofopgroup_submit(group);
2290 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
2291 * encoded by ofp_mkerr() on failure.
2293 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2296 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm,
2297 const struct ofp_header *request)
2299 struct ofproto *p = ofconn_get_ofproto(ofconn);
2303 error = collect_rules_loose(p, fm->table_id, &fm->cr, OFPP_NONE, &rules);
2304 return (error ? error
2305 : list_is_empty(&rules) ? add_flow(p, ofconn, fm, request)
2306 : modify_flows__(ofconn, fm, request, &rules));
2309 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
2310 * code as encoded by ofp_mkerr() on failure.
2312 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2315 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm,
2316 const struct ofp_header *request)
2318 struct ofproto *p = ofconn_get_ofproto(ofconn);
2322 error = collect_rules_strict(p, fm->table_id, &fm->cr, OFPP_NONE, &rules);
2323 return (error ? error
2324 : list_is_empty(&rules) ? add_flow(p, ofconn, fm, request)
2325 : list_is_singleton(&rules) ? modify_flows__(ofconn, fm, request,
2330 /* OFPFC_DELETE implementation. */
2332 /* Deletes the rules listed in 'rules'.
2334 * Returns 0 on success, otherwise an OpenFlow error code. */
2336 delete_flows__(struct ofconn *ofconn, const struct ofp_header *request,
2339 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2340 struct rule *rule, *next;
2341 struct ofopgroup *group;
2343 group = ofopgroup_create_for_ofconn(ofconn, request, UINT32_MAX);
2344 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
2345 ofproto_rule_send_removed(rule, OFPRR_DELETE);
2347 ofoperation_create(group, rule, OFOPERATION_DELETE);
2348 classifier_remove(&ofproto->tables[rule->table_id], &rule->cr);
2349 rule->ofproto->ofproto_class->rule_destruct(rule);
2351 ofopgroup_submit(group);
2356 /* Implements OFPFC_DELETE. */
2358 delete_flows_loose(struct ofconn *ofconn, const struct flow_mod *fm,
2359 const struct ofp_header *request)
2361 struct ofproto *p = ofconn_get_ofproto(ofconn);
2365 error = collect_rules_loose(p, fm->table_id, &fm->cr, fm->out_port,
2367 return (error ? error
2368 : !list_is_empty(&rules) ? delete_flows__(ofconn, request, &rules)
2372 /* Implements OFPFC_DELETE_STRICT. */
2374 delete_flow_strict(struct ofconn *ofconn, struct flow_mod *fm,
2375 const struct ofp_header *request)
2377 struct ofproto *p = ofconn_get_ofproto(ofconn);
2381 error = collect_rules_strict(p, fm->table_id, &fm->cr, fm->out_port,
2383 return (error ? error
2384 : list_is_singleton(&rules) ? delete_flows__(ofconn, request,
2390 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
2392 struct ofputil_flow_removed fr;
2394 if (rule_is_hidden(rule) || !rule->send_flow_removed) {
2399 fr.cookie = rule->flow_cookie;
2401 calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec);
2402 fr.idle_timeout = rule->idle_timeout;
2403 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
2406 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
2409 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
2410 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
2413 * ofproto implementation ->run() functions should use this function to expire
2414 * OpenFlow flows. */
2416 ofproto_rule_expire(struct rule *rule, uint8_t reason)
2418 struct ofproto *ofproto = rule->ofproto;
2419 struct ofopgroup *group;
2421 assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
2423 ofproto_rule_send_removed(rule, reason);
2425 group = ofopgroup_create(ofproto);
2426 ofoperation_create(group, rule, OFOPERATION_DELETE);
2427 classifier_remove(&ofproto->tables[rule->table_id], &rule->cr);
2428 rule->ofproto->ofproto_class->rule_destruct(rule);
2429 ofopgroup_submit(group);
2433 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2435 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2439 error = reject_slave_controller(ofconn, "flow_mod");
2444 if (list_size(&ofproto->pending) >= 50) {
2445 return OFPROTO_POSTPONE;
2448 error = ofputil_decode_flow_mod(&fm, oh,
2449 ofconn_get_flow_mod_table_id(ofconn));
2454 /* We do not support the emergency flow cache. It will hopefully get
2455 * dropped from OpenFlow in the near future. */
2456 if (fm.flags & OFPFF_EMERG) {
2457 /* There isn't a good fit for an error code, so just state that the
2458 * flow table is full. */
2459 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
2462 switch (fm.command) {
2464 return add_flow(ofproto, ofconn, &fm, oh);
2467 return modify_flows_loose(ofconn, &fm, oh);
2469 case OFPFC_MODIFY_STRICT:
2470 return modify_flow_strict(ofconn, &fm, oh);
2473 return delete_flows_loose(ofconn, &fm, oh);
2475 case OFPFC_DELETE_STRICT:
2476 return delete_flow_strict(ofconn, &fm, oh);
2479 if (fm.command > 0xff) {
2480 VLOG_WARN_RL(&rl, "flow_mod has explicit table_id but "
2481 "flow_mod_table_id extension is not enabled");
2483 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
2488 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
2490 struct nx_role_request *nrr = (struct nx_role_request *) oh;
2491 struct nx_role_request *reply;
2495 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY) {
2496 VLOG_WARN_RL(&rl, "ignoring role request on service connection");
2497 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2500 role = ntohl(nrr->role);
2501 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
2502 && role != NX_ROLE_SLAVE) {
2503 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
2505 /* There's no good error code for this. */
2506 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
2509 if (ofconn_get_role(ofconn) != role
2510 && ofconn_has_pending_opgroups(ofconn)) {
2511 return OFPROTO_POSTPONE;
2514 ofconn_set_role(ofconn, role);
2516 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
2517 reply->role = htonl(role);
2518 ofconn_send_reply(ofconn, buf);
2524 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
2525 const struct ofp_header *oh)
2527 const struct nxt_flow_mod_table_id *msg
2528 = (const struct nxt_flow_mod_table_id *) oh;
2530 ofconn_set_flow_mod_table_id(ofconn, msg->set != 0);
2535 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
2537 const struct nxt_set_flow_format *msg
2538 = (const struct nxt_set_flow_format *) oh;
2541 format = ntohl(msg->format);
2542 if (format != NXFF_OPENFLOW10 && format != NXFF_NXM) {
2543 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2546 if (format != ofconn_get_flow_format(ofconn)
2547 && ofconn_has_pending_opgroups(ofconn)) {
2548 /* Avoid sending async messages in surprising flow format. */
2549 return OFPROTO_POSTPONE;
2552 ofconn_set_flow_format(ofconn, format);
2557 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
2559 struct ofp_header *ob;
2562 if (ofconn_has_pending_opgroups(ofconn)) {
2563 return OFPROTO_POSTPONE;
2566 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
2567 ofconn_send_reply(ofconn, buf);
2572 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
2574 const struct ofp_header *oh = msg->data;
2575 const struct ofputil_msg_type *type;
2578 error = ofputil_decode_msg_type(oh, &type);
2583 switch (ofputil_msg_type_code(type)) {
2584 /* OpenFlow requests. */
2585 case OFPUTIL_OFPT_ECHO_REQUEST:
2586 return handle_echo_request(ofconn, oh);
2588 case OFPUTIL_OFPT_FEATURES_REQUEST:
2589 return handle_features_request(ofconn, oh);
2591 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
2592 return handle_get_config_request(ofconn, oh);
2594 case OFPUTIL_OFPT_SET_CONFIG:
2595 return handle_set_config(ofconn, msg->data);
2597 case OFPUTIL_OFPT_PACKET_OUT:
2598 return handle_packet_out(ofconn, oh);
2600 case OFPUTIL_OFPT_PORT_MOD:
2601 return handle_port_mod(ofconn, oh);
2603 case OFPUTIL_OFPT_FLOW_MOD:
2604 return handle_flow_mod(ofconn, oh);
2606 case OFPUTIL_OFPT_BARRIER_REQUEST:
2607 return handle_barrier_request(ofconn, oh);
2609 /* OpenFlow replies. */
2610 case OFPUTIL_OFPT_ECHO_REPLY:
2613 /* Nicira extension requests. */
2614 case OFPUTIL_NXT_ROLE_REQUEST:
2615 return handle_role_request(ofconn, oh);
2617 case OFPUTIL_NXT_FLOW_MOD_TABLE_ID:
2618 return handle_nxt_flow_mod_table_id(ofconn, oh);
2620 case OFPUTIL_NXT_SET_FLOW_FORMAT:
2621 return handle_nxt_set_flow_format(ofconn, oh);
2623 case OFPUTIL_NXT_FLOW_MOD:
2624 return handle_flow_mod(ofconn, oh);
2626 /* Statistics requests. */
2627 case OFPUTIL_OFPST_DESC_REQUEST:
2628 return handle_desc_stats_request(ofconn, msg->data);
2630 case OFPUTIL_OFPST_FLOW_REQUEST:
2631 case OFPUTIL_NXST_FLOW_REQUEST:
2632 return handle_flow_stats_request(ofconn, msg->data);
2634 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
2635 case OFPUTIL_NXST_AGGREGATE_REQUEST:
2636 return handle_aggregate_stats_request(ofconn, msg->data);
2638 case OFPUTIL_OFPST_TABLE_REQUEST:
2639 return handle_table_stats_request(ofconn, msg->data);
2641 case OFPUTIL_OFPST_PORT_REQUEST:
2642 return handle_port_stats_request(ofconn, msg->data);
2644 case OFPUTIL_OFPST_QUEUE_REQUEST:
2645 return handle_queue_stats_request(ofconn, msg->data);
2647 case OFPUTIL_MSG_INVALID:
2648 case OFPUTIL_OFPT_HELLO:
2649 case OFPUTIL_OFPT_ERROR:
2650 case OFPUTIL_OFPT_FEATURES_REPLY:
2651 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
2652 case OFPUTIL_OFPT_PACKET_IN:
2653 case OFPUTIL_OFPT_FLOW_REMOVED:
2654 case OFPUTIL_OFPT_PORT_STATUS:
2655 case OFPUTIL_OFPT_BARRIER_REPLY:
2656 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
2657 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
2658 case OFPUTIL_OFPST_DESC_REPLY:
2659 case OFPUTIL_OFPST_FLOW_REPLY:
2660 case OFPUTIL_OFPST_QUEUE_REPLY:
2661 case OFPUTIL_OFPST_PORT_REPLY:
2662 case OFPUTIL_OFPST_TABLE_REPLY:
2663 case OFPUTIL_OFPST_AGGREGATE_REPLY:
2664 case OFPUTIL_NXT_ROLE_REPLY:
2665 case OFPUTIL_NXT_FLOW_REMOVED:
2666 case OFPUTIL_NXST_FLOW_REPLY:
2667 case OFPUTIL_NXST_AGGREGATE_REPLY:
2669 if (VLOG_IS_WARN_ENABLED()) {
2670 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
2671 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
2674 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
2675 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2677 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
2683 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
2685 int error = handle_openflow__(ofconn, ofp_msg);
2686 if (error && error != OFPROTO_POSTPONE) {
2687 ofconn_send_error(ofconn, ofp_msg->data, error);
2689 COVERAGE_INC(ofproto_recv_openflow);
2690 return error != OFPROTO_POSTPONE;
2693 /* Asynchronous operations. */
2695 /* Creates and returns a new ofopgroup that is not associated with any
2696 * OpenFlow connection.
2698 * The caller should add operations to the returned group with
2699 * ofoperation_create() and then submit it with ofopgroup_submit(). */
2700 static struct ofopgroup *
2701 ofopgroup_create(struct ofproto *ofproto)
2703 struct ofopgroup *group = xzalloc(sizeof *group);
2704 group->ofproto = ofproto;
2705 list_init(&group->ofproto_node);
2706 list_init(&group->ops);
2707 list_init(&group->ofconn_node);
2711 /* Creates and returns a new ofopgroup that is associated with 'ofconn'. If
2712 * the ofopgroup eventually fails, then the error reply will include 'request'.
2713 * If the ofopgroup eventually succeeds, then the packet with buffer id
2714 * 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
2716 * The caller should add operations to the returned group with
2717 * ofoperation_create() and then submit it with ofopgroup_submit(). */
2718 static struct ofopgroup *
2719 ofopgroup_create_for_ofconn(struct ofconn *ofconn,
2720 const struct ofp_header *request,
2723 struct ofopgroup *group = ofopgroup_create(ofconn_get_ofproto(ofconn));
2724 size_t request_len = ntohs(request->length);
2726 ofconn_add_opgroup(ofconn, &group->ofconn_node);
2727 group->ofconn = ofconn;
2728 group->request = xmemdup(request, MIN(request_len, 64));
2729 group->buffer_id = buffer_id;
2734 /* Submits 'group' for processing.
2736 * If 'group' contains no operations (e.g. none were ever added, or all of the
2737 * ones that were added completed synchronously), then it is destroyed
2738 * immediately. Otherwise it is added to the ofproto's list of pending
2741 ofopgroup_submit(struct ofopgroup *group)
2743 if (list_is_empty(&group->ops)) {
2744 ofopgroup_destroy(group);
2746 list_push_back(&group->ofproto->pending, &group->ofproto_node);
2751 ofopgroup_destroy(struct ofopgroup *group)
2753 assert(list_is_empty(&group->ops));
2754 if (!list_is_empty(&group->ofproto_node)) {
2755 list_remove(&group->ofproto_node);
2757 if (!list_is_empty(&group->ofconn_node)) {
2758 list_remove(&group->ofconn_node);
2760 ofconn_send_error(group->ofconn, group->request, group->error);
2762 connmgr_retry(group->ofproto->connmgr);
2764 free(group->request);
2768 /* Initiates a new operation on 'rule', of the specified 'type', within
2769 * 'group'. Prior to calling, 'rule' must not have any pending operation. */
2771 ofoperation_create(struct ofopgroup *group, struct rule *rule,
2772 enum ofoperation_type type)
2774 struct ofoperation *op;
2776 assert(!rule->pending);
2778 op = rule->pending = xzalloc(sizeof *op);
2780 list_push_back(&group->ops, &op->group_node);
2784 op->flow_cookie = rule->flow_cookie;
2786 if (type == OFOPERATION_DELETE) {
2787 hmap_insert(&op->group->ofproto->deletions, &op->hmap_node,
2788 cls_rule_hash(&rule->cr, rule->table_id));
2793 ofoperation_destroy(struct ofoperation *op)
2795 struct ofopgroup *group = op->group;
2798 op->rule->pending = NULL;
2800 if (op->type == OFOPERATION_DELETE) {
2801 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
2803 list_remove(&op->group_node);
2807 if (list_is_empty(&group->ops) && !list_is_empty(&group->ofproto_node)) {
2808 ofopgroup_destroy(group);
2812 /* Indicates that 'op' completed with status 'error', which is either 0 to
2813 * indicate success or an OpenFlow error code (constructed with
2814 * e.g. ofp_mkerr()).
2816 * If 'op' is a "delete flow" operation, 'error' must be 0. That is, flow
2817 * deletions are not allowed to fail.
2819 * Please see the large comment in ofproto/ofproto-provider.h titled
2820 * "Asynchronous Operation Support" for more information. */
2822 ofoperation_complete(struct ofoperation *op, int error)
2824 struct ofopgroup *group = op->group;
2825 struct rule *rule = op->rule;
2826 struct classifier *table = &rule->ofproto->tables[rule->table_id];
2828 assert(rule->pending == op);
2829 assert(op->status < 0);
2834 && op->type != OFOPERATION_DELETE
2836 && group->buffer_id != UINT32_MAX
2837 && list_is_singleton(&op->group_node)) {
2838 struct ofpbuf *packet;
2841 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
2845 error = rule_execute(rule, in_port, packet);
2848 if (!group->error) {
2849 group->error = error;
2853 case OFOPERATION_ADD:
2856 ofproto_rule_destroy__(op->victim);
2860 classifier_replace(table, &op->victim->cr);
2863 classifier_remove(table, &rule->cr);
2865 ofproto_rule_destroy__(rule);
2870 case OFOPERATION_DELETE:
2872 ofproto_rule_destroy__(rule);
2876 case OFOPERATION_MODIFY:
2878 free(rule->actions);
2879 rule->actions = op->actions;
2880 rule->n_actions = op->n_actions;
2888 ofoperation_destroy(op);
2892 ofoperation_get_victim(struct ofoperation *op)
2894 assert(op->type == OFOPERATION_ADD);
2899 pick_datapath_id(const struct ofproto *ofproto)
2901 const struct ofport *port;
2903 port = ofproto_get_port(ofproto, OFPP_LOCAL);
2905 uint8_t ea[ETH_ADDR_LEN];
2908 error = netdev_get_etheraddr(port->netdev, ea);
2910 return eth_addr_to_uint64(ea);
2912 VLOG_WARN("could not get MAC address for %s (%s)",
2913 netdev_get_name(port->netdev), strerror(error));
2915 return ofproto->fallback_dpid;
2919 pick_fallback_dpid(void)
2921 uint8_t ea[ETH_ADDR_LEN];
2922 eth_addr_nicira_random(ea);
2923 return eth_addr_to_uint64(ea);
2926 /* unixctl commands. */
2929 ofproto_lookup(const char *name)
2931 struct ofproto *ofproto;
2933 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
2935 if (!strcmp(ofproto->name, name)) {
2943 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
2944 void *aux OVS_UNUSED)
2946 struct ofproto *ofproto;
2950 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
2951 ds_put_format(&results, "%s\n", ofproto->name);
2953 unixctl_command_reply(conn, 200, ds_cstr(&results));
2954 ds_destroy(&results);
2958 ofproto_unixctl_init(void)
2960 static bool registered;
2966 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);