2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
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.
26 #include "byte-order.h"
27 #include "classifier.h"
30 #include "dynamic-string.h"
33 #include "meta-flow.h"
36 #include "ofp-actions.h"
37 #include "ofp-errors.h"
39 #include "ofp-print.h"
42 #include "ofproto-provider.h"
43 #include "openflow/nicira-ext.h"
44 #include "openflow/openflow.h"
48 #include "poll-loop.h"
54 #include "unaligned.h"
58 VLOG_DEFINE_THIS_MODULE(ofproto);
60 COVERAGE_DEFINE(ofproto_error);
61 COVERAGE_DEFINE(ofproto_flush);
62 COVERAGE_DEFINE(ofproto_no_packet_in);
63 COVERAGE_DEFINE(ofproto_packet_out);
64 COVERAGE_DEFINE(ofproto_queue_req);
65 COVERAGE_DEFINE(ofproto_recv_openflow);
66 COVERAGE_DEFINE(ofproto_reinit_ports);
67 COVERAGE_DEFINE(ofproto_uninstallable);
68 COVERAGE_DEFINE(ofproto_update_port);
71 S_OPENFLOW, /* Processing OpenFlow commands. */
72 S_EVICT, /* Evicting flows from over-limit tables. */
73 S_FLUSH, /* Deleting all flow table rules. */
76 enum ofoperation_type {
83 /* A single OpenFlow request can execute any number of operations. The
84 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
85 * ofconn to which an error reply should be sent if necessary.
87 * ofproto initiates some operations internally. These operations are still
88 * assigned to groups but will not have an associated ofconn. */
90 struct ofproto *ofproto; /* Owning ofproto. */
91 struct list ofproto_node; /* In ofproto's "pending" list. */
92 struct list ops; /* List of "struct ofoperation"s. */
93 int n_running; /* Number of ops still pending. */
95 /* Data needed to send OpenFlow reply on failure or to send a buffered
98 * If list_is_empty(ofconn_node) then this ofopgroup never had an
99 * associated ofconn or its ofconn's connection dropped after it initiated
100 * the operation. In the latter case 'ofconn' is a wild pointer that
101 * refers to freed memory, so the 'ofconn' member must be used only if
102 * !list_is_empty(ofconn_node).
104 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
105 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
106 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
107 uint32_t buffer_id; /* Buffer id from original request. */
110 static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
111 static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
112 const struct ofp_header *,
114 static void ofopgroup_submit(struct ofopgroup *);
115 static void ofopgroup_complete(struct ofopgroup *);
117 /* A single flow table operation. */
119 struct ofopgroup *group; /* Owning group. */
120 struct list group_node; /* In ofopgroup's "ops" list. */
121 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
122 struct rule *rule; /* Rule being operated upon. */
123 enum ofoperation_type type; /* Type of operation. */
125 /* OFOPERATION_MODIFY, OFOPERATION_REPLACE: The old actions, if the actions
127 struct ofpact *ofpacts;
131 /* OFOPERATION_DELETE. */
132 enum ofp_flow_removed_reason reason; /* Reason flow was removed. */
134 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
135 uint16_t idle_timeout; /* Rule's old idle timeout. */
136 uint16_t hard_timeout; /* Rule's old hard timeout. */
137 bool send_flow_removed; /* Rule's old 'send_flow_removed'. */
138 enum ofperr error; /* 0 if no error. */
141 static struct ofoperation *ofoperation_create(struct ofopgroup *,
143 enum ofoperation_type,
144 enum ofp_flow_removed_reason);
145 static void ofoperation_destroy(struct ofoperation *);
148 static void oftable_init(struct oftable *);
149 static void oftable_destroy(struct oftable *);
151 static void oftable_set_name(struct oftable *, const char *name);
153 static void oftable_disable_eviction(struct oftable *);
154 static void oftable_enable_eviction(struct oftable *,
155 const struct mf_subfield *fields,
158 static void oftable_remove_rule(struct rule *rule) OVS_RELEASES(rule->rwlock);
159 static void oftable_remove_rule__(struct ofproto *ofproto,
160 struct classifier *cls, struct rule *rule)
161 OVS_REQ_WRLOCK(cls->rwlock) OVS_RELEASES(rule->rwlock);
162 static void oftable_insert_rule(struct rule *);
164 /* A set of rules within a single OpenFlow table (oftable) that have the same
165 * values for the oftable's eviction_fields. A rule to be evicted, when one is
166 * needed, is taken from the eviction group that contains the greatest number
169 * An oftable owns any number of eviction groups, each of which contains any
172 * Membership in an eviction group is imprecise, based on the hash of the
173 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
174 * That is, if two rules have different eviction_fields, but those
175 * eviction_fields hash to the same value, then they will belong to the same
176 * eviction_group anyway.
178 * (When eviction is not enabled on an oftable, we don't track any eviction
179 * groups, to save time and space.) */
180 struct eviction_group {
181 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
182 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
183 struct heap rules; /* Contains "struct rule"s. */
186 static bool choose_rule_to_evict(struct oftable *table, struct rule **rulep)
187 OVS_TRY_WRLOCK(true, (*rulep)->rwlock);
188 static void ofproto_evict(struct ofproto *);
189 static uint32_t rule_eviction_priority(struct rule *);
190 static void eviction_group_add_rule(struct rule *);
191 static void eviction_group_remove_rule(struct rule *);
194 static void ofport_destroy__(struct ofport *);
195 static void ofport_destroy(struct ofport *);
197 static void update_port(struct ofproto *, const char *devname);
198 static int init_ports(struct ofproto *);
199 static void reinit_ports(struct ofproto *);
202 static void ofproto_rule_destroy(struct rule *);
203 static void ofproto_rule_destroy__(struct rule *);
204 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
205 static bool rule_is_modifiable(const struct rule *);
208 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
209 struct ofputil_flow_mod *,
210 const struct ofp_header *);
211 static enum ofperr modify_flows__(struct ofproto *, struct ofconn *,
212 struct ofputil_flow_mod *,
213 const struct ofp_header *, struct list *);
214 static void delete_flow__(struct rule *rule, struct ofopgroup *,
215 enum ofp_flow_removed_reason)
216 OVS_RELEASES(rule->rwlock);
217 static bool handle_openflow(struct ofconn *, const struct ofpbuf *);
218 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
219 struct ofputil_flow_mod *,
220 const struct ofp_header *);
221 static void calc_duration(long long int start, long long int now,
222 uint32_t *sec, uint32_t *nsec);
225 static uint64_t pick_datapath_id(const struct ofproto *);
226 static uint64_t pick_fallback_dpid(void);
227 static void ofproto_destroy__(struct ofproto *);
228 static void update_mtu(struct ofproto *, struct ofport *);
229 static void meter_delete(struct ofproto *, uint32_t first, uint32_t last);
232 static void ofproto_unixctl_init(void);
234 /* All registered ofproto classes, in probe order. */
235 static const struct ofproto_class **ofproto_classes;
236 static size_t n_ofproto_classes;
237 static size_t allocated_ofproto_classes;
239 unsigned flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT;
240 unsigned n_handler_threads;
241 enum ofproto_flow_miss_model flow_miss_model = OFPROTO_HANDLE_MISS_AUTO;
243 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
244 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
246 /* Initial mappings of port to OpenFlow number mappings. */
247 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
249 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
251 /* The default value of true waits for flow restore. */
252 static bool flow_restore_wait = true;
254 /* Must be called to initialize the ofproto library.
256 * The caller may pass in 'iface_hints', which contains an shash of
257 * "iface_hint" elements indexed by the interface's name. The provider
258 * may use these hints to describe the startup configuration in order to
259 * reinitialize its state. The caller owns the provided data, so a
260 * provider will make copies of anything required. An ofproto provider
261 * will remove any existing state that is not described by the hint, and
262 * may choose to remove it all. */
264 ofproto_init(const struct shash *iface_hints)
266 struct shash_node *node;
269 ofproto_class_register(&ofproto_dpif_class);
271 /* Make a local copy, since we don't own 'iface_hints' elements. */
272 SHASH_FOR_EACH(node, iface_hints) {
273 const struct iface_hint *orig_hint = node->data;
274 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
275 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
277 new_hint->br_name = xstrdup(orig_hint->br_name);
278 new_hint->br_type = xstrdup(br_type);
279 new_hint->ofp_port = orig_hint->ofp_port;
281 shash_add(&init_ofp_ports, node->name, new_hint);
284 for (i = 0; i < n_ofproto_classes; i++) {
285 ofproto_classes[i]->init(&init_ofp_ports);
289 /* 'type' should be a normalized datapath type, as returned by
290 * ofproto_normalize_type(). Returns the corresponding ofproto_class
291 * structure, or a null pointer if there is none registered for 'type'. */
292 static const struct ofproto_class *
293 ofproto_class_find__(const char *type)
297 for (i = 0; i < n_ofproto_classes; i++) {
298 const struct ofproto_class *class = ofproto_classes[i];
303 class->enumerate_types(&types);
304 found = sset_contains(&types, type);
305 sset_destroy(&types);
311 VLOG_WARN("unknown datapath type %s", type);
315 /* Registers a new ofproto class. After successful registration, new ofprotos
316 * of that type can be created using ofproto_create(). */
318 ofproto_class_register(const struct ofproto_class *new_class)
322 for (i = 0; i < n_ofproto_classes; i++) {
323 if (ofproto_classes[i] == new_class) {
328 if (n_ofproto_classes >= allocated_ofproto_classes) {
329 ofproto_classes = x2nrealloc(ofproto_classes,
330 &allocated_ofproto_classes,
331 sizeof *ofproto_classes);
333 ofproto_classes[n_ofproto_classes++] = new_class;
337 /* Unregisters a datapath provider. 'type' must have been previously
338 * registered and not currently be in use by any ofprotos. After
339 * unregistration new datapaths of that type cannot be opened using
340 * ofproto_create(). */
342 ofproto_class_unregister(const struct ofproto_class *class)
346 for (i = 0; i < n_ofproto_classes; i++) {
347 if (ofproto_classes[i] == class) {
348 for (i++; i < n_ofproto_classes; i++) {
349 ofproto_classes[i - 1] = ofproto_classes[i];
355 VLOG_WARN("attempted to unregister an ofproto class that is not "
360 /* Clears 'types' and enumerates all registered ofproto types into it. The
361 * caller must first initialize the sset. */
363 ofproto_enumerate_types(struct sset *types)
367 for (i = 0; i < n_ofproto_classes; i++) {
368 ofproto_classes[i]->enumerate_types(types);
372 /* Returns the fully spelled out name for the given ofproto 'type'.
374 * Normalized type string can be compared with strcmp(). Unnormalized type
375 * string might be the same even if they have different spellings. */
377 ofproto_normalize_type(const char *type)
379 return type && type[0] ? type : "system";
382 /* Clears 'names' and enumerates the names of all known created ofprotos with
383 * the given 'type'. The caller must first initialize the sset. Returns 0 if
384 * successful, otherwise a positive errno value.
386 * Some kinds of datapaths might not be practically enumerable. This is not
387 * considered an error. */
389 ofproto_enumerate_names(const char *type, struct sset *names)
391 const struct ofproto_class *class = ofproto_class_find__(type);
392 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
396 ofproto_create(const char *datapath_name, const char *datapath_type,
397 struct ofproto **ofprotop)
399 const struct ofproto_class *class;
400 struct ofproto *ofproto;
406 ofproto_unixctl_init();
408 datapath_type = ofproto_normalize_type(datapath_type);
409 class = ofproto_class_find__(datapath_type);
411 VLOG_WARN("could not create datapath %s of unknown type %s",
412 datapath_name, datapath_type);
416 ofproto = class->alloc();
418 VLOG_ERR("failed to allocate datapath %s of type %s",
419 datapath_name, datapath_type);
424 memset(ofproto, 0, sizeof *ofproto);
425 ofproto->ofproto_class = class;
426 ofproto->name = xstrdup(datapath_name);
427 ofproto->type = xstrdup(datapath_type);
428 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
429 hash_string(ofproto->name, 0));
430 ofproto->datapath_id = 0;
431 ofproto->forward_bpdu = false;
432 ofproto->fallback_dpid = pick_fallback_dpid();
433 ofproto->mfr_desc = NULL;
434 ofproto->hw_desc = NULL;
435 ofproto->sw_desc = NULL;
436 ofproto->serial_desc = NULL;
437 ofproto->dp_desc = NULL;
438 ofproto->frag_handling = OFPC_FRAG_NORMAL;
439 hmap_init(&ofproto->ports);
440 shash_init(&ofproto->port_by_name);
441 simap_init(&ofproto->ofp_requests);
442 ofproto->max_ports = ofp_to_u16(OFPP_MAX);
443 ofproto->eviction_group_timer = LLONG_MIN;
444 ofproto->tables = NULL;
445 ofproto->n_tables = 0;
446 hindex_init(&ofproto->cookies);
447 list_init(&ofproto->expirable);
448 ovs_mutex_init_recursive(&ofproto->expirable_mutex);
449 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
450 ofproto->state = S_OPENFLOW;
451 list_init(&ofproto->pending);
452 ofproto->n_pending = 0;
453 hmap_init(&ofproto->deletions);
454 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
455 ofproto->first_op = ofproto->last_op = LLONG_MIN;
456 ofproto->next_op_report = LLONG_MAX;
457 ofproto->op_backoff = LLONG_MIN;
458 ofproto->vlan_bitmap = NULL;
459 ofproto->vlans_changed = false;
460 ofproto->min_mtu = INT_MAX;
462 error = ofproto->ofproto_class->construct(ofproto);
464 VLOG_ERR("failed to open datapath %s: %s",
465 datapath_name, ovs_strerror(error));
466 ofproto_destroy__(ofproto);
470 /* The "max_ports" member should have been set by ->construct(ofproto).
471 * Port 0 is not a valid OpenFlow port, so mark that as unavailable. */
472 ofproto->ofp_port_ids = bitmap_allocate(ofproto->max_ports);
473 bitmap_set1(ofproto->ofp_port_ids, 0);
475 /* Check that hidden tables, if any, are at the end. */
476 ovs_assert(ofproto->n_tables);
477 for (i = 0; i + 1 < ofproto->n_tables; i++) {
478 enum oftable_flags flags = ofproto->tables[i].flags;
479 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
481 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
484 ofproto->datapath_id = pick_datapath_id(ofproto);
487 /* Initialize meters table. */
488 if (ofproto->ofproto_class->meter_get_features) {
489 ofproto->ofproto_class->meter_get_features(ofproto,
490 &ofproto->meter_features);
492 memset(&ofproto->meter_features, 0, sizeof ofproto->meter_features);
494 ofproto->meters = xzalloc((ofproto->meter_features.max_meters + 1)
495 * sizeof(struct meter *));
501 /* Must be called (only) by an ofproto implementation in its constructor
502 * function. See the large comment on 'construct' in struct ofproto_class for
505 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
507 struct oftable *table;
509 ovs_assert(!ofproto->n_tables);
510 ovs_assert(n_tables >= 1 && n_tables <= 255);
512 ofproto->n_tables = n_tables;
513 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
514 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
519 /* To be optionally called (only) by an ofproto implementation in its
520 * constructor function. See the large comment on 'construct' in struct
521 * ofproto_class for details.
523 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
524 * will then ensure that actions passed into the ofproto implementation will
525 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
526 * function is not called, there will be no such restriction.
528 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
529 * the 'max_ports' restriction. */
531 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
533 ovs_assert(max_ports <= ofp_to_u16(OFPP_MAX));
534 ofproto->max_ports = max_ports;
538 ofproto_get_datapath_id(const struct ofproto *ofproto)
540 return ofproto->datapath_id;
544 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
546 uint64_t old_dpid = p->datapath_id;
547 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
548 if (p->datapath_id != old_dpid) {
549 /* Force all active connections to reconnect, since there is no way to
550 * notify a controller that the datapath ID has changed. */
551 ofproto_reconnect_controllers(p);
556 ofproto_set_controllers(struct ofproto *p,
557 const struct ofproto_controller *controllers,
558 size_t n_controllers, uint32_t allowed_versions)
560 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
565 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
567 connmgr_set_fail_mode(p->connmgr, fail_mode);
570 /* Drops the connections between 'ofproto' and all of its controllers, forcing
571 * them to reconnect. */
573 ofproto_reconnect_controllers(struct ofproto *ofproto)
575 connmgr_reconnect(ofproto->connmgr);
578 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
579 * in-band control should guarantee access, in the same way that in-band
580 * control guarantees access to OpenFlow controllers. */
582 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
583 const struct sockaddr_in *extras, size_t n)
585 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
588 /* Sets the OpenFlow queue used by flows set up by in-band control on
589 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
590 * flows will use the default queue. */
592 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
594 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
597 /* Sets the number of flows at which eviction from the kernel flow table
600 ofproto_set_flow_eviction_threshold(unsigned threshold)
602 flow_eviction_threshold = MAX(OFPROTO_FLOW_EVICTION_THRESHOLD_MIN,
606 /* Sets the path for handling flow misses. */
608 ofproto_set_flow_miss_model(unsigned model)
610 flow_miss_model = model;
613 /* If forward_bpdu is true, the NORMAL action will forward frames with
614 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
615 * the NORMAL action will drop these frames. */
617 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
619 bool old_val = ofproto->forward_bpdu;
620 ofproto->forward_bpdu = forward_bpdu;
621 if (old_val != ofproto->forward_bpdu) {
622 if (ofproto->ofproto_class->forward_bpdu_changed) {
623 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
628 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
629 * 'idle_time', in seconds, and the maximum number of MAC table entries to
632 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
635 if (ofproto->ofproto_class->set_mac_table_config) {
636 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
641 /* Sets number of upcall handler threads. The default is
642 * (number of online cores - 2). */
644 ofproto_set_n_handler_threads(unsigned limit)
647 n_handler_threads = limit;
649 int n_proc = sysconf(_SC_NPROCESSORS_ONLN);
650 n_handler_threads = n_proc > 2 ? n_proc - 2 : 1;
655 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
658 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
662 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
664 return connmgr_set_snoops(ofproto->connmgr, snoops);
668 ofproto_set_netflow(struct ofproto *ofproto,
669 const struct netflow_options *nf_options)
671 if (nf_options && sset_is_empty(&nf_options->collectors)) {
675 if (ofproto->ofproto_class->set_netflow) {
676 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
678 return nf_options ? EOPNOTSUPP : 0;
683 ofproto_set_sflow(struct ofproto *ofproto,
684 const struct ofproto_sflow_options *oso)
686 if (oso && sset_is_empty(&oso->targets)) {
690 if (ofproto->ofproto_class->set_sflow) {
691 return ofproto->ofproto_class->set_sflow(ofproto, oso);
693 return oso ? EOPNOTSUPP : 0;
698 ofproto_set_ipfix(struct ofproto *ofproto,
699 const struct ofproto_ipfix_bridge_exporter_options *bo,
700 const struct ofproto_ipfix_flow_exporter_options *fo,
703 if (ofproto->ofproto_class->set_ipfix) {
704 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
706 return (bo || fo) ? EOPNOTSUPP : 0;
711 ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
713 flow_restore_wait = flow_restore_wait_db;
717 ofproto_get_flow_restore_wait(void)
719 return flow_restore_wait;
723 /* Spanning Tree Protocol (STP) configuration. */
725 /* Configures STP on 'ofproto' using the settings defined in 's'. If
726 * 's' is NULL, disables STP.
728 * Returns 0 if successful, otherwise a positive errno value. */
730 ofproto_set_stp(struct ofproto *ofproto,
731 const struct ofproto_stp_settings *s)
733 return (ofproto->ofproto_class->set_stp
734 ? ofproto->ofproto_class->set_stp(ofproto, s)
738 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
739 * 'enabled' member of 's' is false, then the other members are not
742 * Returns 0 if successful, otherwise a positive errno value. */
744 ofproto_get_stp_status(struct ofproto *ofproto,
745 struct ofproto_stp_status *s)
747 return (ofproto->ofproto_class->get_stp_status
748 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
752 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
753 * in 's'. The caller is responsible for assigning STP port numbers
754 * (using the 'port_num' member in the range of 1 through 255, inclusive)
755 * and ensuring there are no duplicates. If the 's' is NULL, then STP
756 * is disabled on the port.
758 * Returns 0 if successful, otherwise a positive errno value.*/
760 ofproto_port_set_stp(struct ofproto *ofproto, ofp_port_t ofp_port,
761 const struct ofproto_port_stp_settings *s)
763 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
765 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
766 ofproto->name, ofp_port);
770 return (ofproto->ofproto_class->set_stp_port
771 ? ofproto->ofproto_class->set_stp_port(ofport, s)
775 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
776 * 's'. If the 'enabled' member in 's' is false, then the other members
777 * are not meaningful.
779 * Returns 0 if successful, otherwise a positive errno value.*/
781 ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
782 struct ofproto_port_stp_status *s)
784 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
786 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
787 "port %"PRIu16, ofproto->name, ofp_port);
791 return (ofproto->ofproto_class->get_stp_port_status
792 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
796 /* Queue DSCP configuration. */
798 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
799 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
800 * to implement QoS. Instead, it is used to implement features which require
801 * knowledge of what queues exist on a port, and some basic information about
804 * Returns 0 if successful, otherwise a positive errno value. */
806 ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port,
807 const struct ofproto_port_queue *queues,
810 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
813 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
814 ofproto->name, ofp_port);
818 return (ofproto->ofproto_class->set_queues
819 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
823 /* Connectivity Fault Management configuration. */
825 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
827 ofproto_port_clear_cfm(struct ofproto *ofproto, ofp_port_t ofp_port)
829 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
830 if (ofport && ofproto->ofproto_class->set_cfm) {
831 ofproto->ofproto_class->set_cfm(ofport, NULL);
835 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
836 * basic configuration from the configuration members in 'cfm', and the remote
837 * maintenance point ID from remote_mpid. Ignores the statistics members of
840 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
842 ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port,
843 const struct cfm_settings *s)
845 struct ofport *ofport;
848 ofport = ofproto_get_port(ofproto, ofp_port);
850 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
851 ofproto->name, ofp_port);
855 /* XXX: For configuration simplicity, we only support one remote_mpid
856 * outside of the CFM module. It's not clear if this is the correct long
857 * term solution or not. */
858 error = (ofproto->ofproto_class->set_cfm
859 ? ofproto->ofproto_class->set_cfm(ofport, s)
862 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
863 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
864 ovs_strerror(error));
868 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
869 * 'ofproto' does not have a port 'ofp_port'. */
871 ofproto_port_set_bfd(struct ofproto *ofproto, ofp_port_t ofp_port,
872 const struct smap *cfg)
874 struct ofport *ofport;
877 ofport = ofproto_get_port(ofproto, ofp_port);
879 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
880 ofproto->name, ofp_port);
884 error = (ofproto->ofproto_class->set_bfd
885 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
888 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
889 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
890 ovs_strerror(error));
894 /* Populates 'status' with key value pairs indicating the status of the BFD
895 * session on 'ofp_port'. This information is intended to be populated in the
896 * OVS database. Has no effect if 'ofp_port' is not na OpenFlow port in
899 ofproto_port_get_bfd_status(struct ofproto *ofproto, ofp_port_t ofp_port,
902 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
903 return (ofport && ofproto->ofproto_class->get_bfd_status
904 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
908 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
909 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
910 * 0 if LACP partner information is not current (generally indicating a
911 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
913 ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port)
915 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
916 return (ofport && ofproto->ofproto_class->port_is_lacp_current
917 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
923 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
924 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
925 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
926 * configuration plus, if there is more than one slave, a bonding
929 * If 'aux' is already registered then this function updates its configuration
930 * to 's'. Otherwise, this function registers a new bundle.
932 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
935 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
936 const struct ofproto_bundle_settings *s)
938 return (ofproto->ofproto_class->bundle_set
939 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
943 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
944 * If no such bundle has been registered, this has no effect. */
946 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
948 return ofproto_bundle_register(ofproto, aux, NULL);
952 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
953 * If 'aux' is already registered then this function updates its configuration
954 * to 's'. Otherwise, this function registers a new mirror. */
956 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
957 const struct ofproto_mirror_settings *s)
959 return (ofproto->ofproto_class->mirror_set
960 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
964 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
965 * If no mirror has been registered, this has no effect. */
967 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
969 return ofproto_mirror_register(ofproto, aux, NULL);
972 /* Retrieves statistics from mirror associated with client data pointer
973 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
974 * 'bytes', respectively. If a particular counters is not supported,
975 * the appropriate argument is set to UINT64_MAX. */
977 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
978 uint64_t *packets, uint64_t *bytes)
980 if (!ofproto->ofproto_class->mirror_get_stats) {
981 *packets = *bytes = UINT64_MAX;
985 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
989 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
990 * which all packets are flooded, instead of using MAC learning. If
991 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
993 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
996 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
998 return (ofproto->ofproto_class->set_flood_vlans
999 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
1003 /* Returns true if 'aux' is a registered bundle that is currently in use as the
1004 * output for a mirror. */
1006 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
1008 return (ofproto->ofproto_class->is_mirror_output_bundle
1009 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
1013 /* Configuration of OpenFlow tables. */
1015 /* Returns the number of OpenFlow tables in 'ofproto'. */
1017 ofproto_get_n_tables(const struct ofproto *ofproto)
1019 return ofproto->n_tables;
1022 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
1023 * settings from 's'. 'table_id' must be in the range 0 through the number of
1024 * OpenFlow tables in 'ofproto' minus 1, inclusive.
1026 * For read-only tables, only the name may be configured. */
1028 ofproto_configure_table(struct ofproto *ofproto, int table_id,
1029 const struct ofproto_table_settings *s)
1031 struct oftable *table;
1033 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
1034 table = &ofproto->tables[table_id];
1036 oftable_set_name(table, s->name);
1038 if (table->flags & OFTABLE_READONLY) {
1043 oftable_enable_eviction(table, s->groups, s->n_groups);
1045 oftable_disable_eviction(table);
1048 table->max_flows = s->max_flows;
1049 ovs_rwlock_rdlock(&table->cls.rwlock);
1050 if (classifier_count(&table->cls) > table->max_flows
1051 && table->eviction_fields) {
1052 /* 'table' contains more flows than allowed. We might not be able to
1053 * evict them right away because of the asynchronous nature of flow
1054 * table changes. Schedule eviction for later. */
1055 switch (ofproto->state) {
1057 ofproto->state = S_EVICT;
1061 /* We're already deleting flows, nothing more to do. */
1065 ovs_rwlock_unlock(&table->cls.rwlock);
1069 ofproto_has_snoops(const struct ofproto *ofproto)
1071 return connmgr_has_snoops(ofproto->connmgr);
1075 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1077 connmgr_get_snoops(ofproto->connmgr, snoops);
1080 /* Deletes 'rule' from 'cls' within 'ofproto'.
1082 * The 'cls' argument is redundant (it is &ofproto->tables[rule->table_id].cls)
1083 * but it allows Clang to do better checking. */
1085 ofproto_delete_rule(struct ofproto *ofproto, struct classifier *cls,
1087 OVS_REQ_WRLOCK(cls->rwlock)
1089 struct ofopgroup *group;
1091 ovs_assert(!rule->pending);
1092 ovs_assert(cls == &ofproto->tables[rule->table_id].cls);
1094 group = ofopgroup_create_unattached(ofproto);
1095 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1096 ovs_rwlock_wrlock(&rule->rwlock);
1097 oftable_remove_rule__(ofproto, cls, rule);
1098 ofproto->ofproto_class->rule_delete(rule);
1099 ofopgroup_submit(group);
1103 ofproto_flush__(struct ofproto *ofproto)
1105 struct oftable *table;
1107 if (ofproto->ofproto_class->flush) {
1108 ofproto->ofproto_class->flush(ofproto);
1111 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1112 struct rule *rule, *next_rule;
1113 struct cls_cursor cursor;
1115 if (table->flags & OFTABLE_HIDDEN) {
1119 ovs_rwlock_wrlock(&table->cls.rwlock);
1120 cls_cursor_init(&cursor, &table->cls, NULL);
1121 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1122 if (!rule->pending) {
1123 ofproto_delete_rule(ofproto, &table->cls, rule);
1126 ovs_rwlock_unlock(&table->cls.rwlock);
1131 ofproto_destroy__(struct ofproto *ofproto)
1133 struct oftable *table;
1135 ovs_assert(list_is_empty(&ofproto->pending));
1136 ovs_assert(!ofproto->n_pending);
1138 if (ofproto->meters) {
1139 meter_delete(ofproto, 1, ofproto->meter_features.max_meters);
1140 free(ofproto->meters);
1143 connmgr_destroy(ofproto->connmgr);
1145 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1146 free(ofproto->name);
1147 free(ofproto->type);
1148 free(ofproto->mfr_desc);
1149 free(ofproto->hw_desc);
1150 free(ofproto->sw_desc);
1151 free(ofproto->serial_desc);
1152 free(ofproto->dp_desc);
1153 hmap_destroy(&ofproto->ports);
1154 shash_destroy(&ofproto->port_by_name);
1155 bitmap_free(ofproto->ofp_port_ids);
1156 simap_destroy(&ofproto->ofp_requests);
1158 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1159 oftable_destroy(table);
1161 free(ofproto->tables);
1163 hmap_destroy(&ofproto->deletions);
1165 free(ofproto->vlan_bitmap);
1167 ovs_mutex_destroy(&ofproto->expirable_mutex);
1168 ofproto->ofproto_class->dealloc(ofproto);
1172 ofproto_destroy(struct ofproto *p)
1174 struct ofport *ofport, *next_ofport;
1181 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1182 ofport_destroy(ofport);
1185 p->ofproto_class->destruct(p);
1186 ofproto_destroy__(p);
1189 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1190 * kernel datapath, for example, this destroys the datapath in the kernel, and
1191 * with the netdev-based datapath, it tears down the data structures that
1192 * represent the datapath.
1194 * The datapath should not be currently open as an ofproto. */
1196 ofproto_delete(const char *name, const char *type)
1198 const struct ofproto_class *class = ofproto_class_find__(type);
1199 return (!class ? EAFNOSUPPORT
1200 : !class->del ? EACCES
1201 : class->del(type, name));
1205 process_port_change(struct ofproto *ofproto, int error, char *devname)
1207 if (error == ENOBUFS) {
1208 reinit_ports(ofproto);
1209 } else if (!error) {
1210 update_port(ofproto, devname);
1216 ofproto_type_run(const char *datapath_type)
1218 const struct ofproto_class *class;
1221 datapath_type = ofproto_normalize_type(datapath_type);
1222 class = ofproto_class_find__(datapath_type);
1224 error = class->type_run ? class->type_run(datapath_type) : 0;
1225 if (error && error != EAGAIN) {
1226 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1227 datapath_type, ovs_strerror(error));
1233 ofproto_type_run_fast(const char *datapath_type)
1235 const struct ofproto_class *class;
1238 datapath_type = ofproto_normalize_type(datapath_type);
1239 class = ofproto_class_find__(datapath_type);
1241 error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
1242 if (error && error != EAGAIN) {
1243 VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
1244 datapath_type, ovs_strerror(error));
1250 ofproto_type_wait(const char *datapath_type)
1252 const struct ofproto_class *class;
1254 datapath_type = ofproto_normalize_type(datapath_type);
1255 class = ofproto_class_find__(datapath_type);
1257 if (class->type_wait) {
1258 class->type_wait(datapath_type);
1263 ofproto_run(struct ofproto *p)
1265 struct sset changed_netdevs;
1266 const char *changed_netdev;
1267 struct ofport *ofport;
1270 error = p->ofproto_class->run(p);
1271 if (error && error != EAGAIN) {
1272 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, ovs_strerror(error));
1275 /* Restore the eviction group heap invariant occasionally. */
1276 if (p->eviction_group_timer < time_msec()) {
1279 p->eviction_group_timer = time_msec() + 1000;
1281 for (i = 0; i < p->n_tables; i++) {
1282 struct oftable *table = &p->tables[i];
1283 struct eviction_group *evg;
1284 struct cls_cursor cursor;
1287 if (!table->eviction_fields) {
1291 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
1292 heap_rebuild(&evg->rules);
1295 ovs_rwlock_rdlock(&table->cls.rwlock);
1296 cls_cursor_init(&cursor, &table->cls, NULL);
1297 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1298 if (!rule->eviction_group
1299 && (rule->idle_timeout || rule->hard_timeout)) {
1300 eviction_group_add_rule(rule);
1303 ovs_rwlock_unlock(&table->cls.rwlock);
1307 if (p->ofproto_class->port_poll) {
1310 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1311 process_port_change(p, error, devname);
1315 /* Update OpenFlow port status for any port whose netdev has changed.
1317 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1318 * destroyed, so it's not safe to update ports directly from the
1319 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1320 * need this two-phase approach. */
1321 sset_init(&changed_netdevs);
1322 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1323 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1324 if (ofport->change_seq != change_seq) {
1325 ofport->change_seq = change_seq;
1326 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1329 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1330 update_port(p, changed_netdev);
1332 sset_destroy(&changed_netdevs);
1336 connmgr_run(p->connmgr, handle_openflow);
1340 connmgr_run(p->connmgr, NULL);
1342 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1343 p->state = S_OPENFLOW;
1348 connmgr_run(p->connmgr, NULL);
1350 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1351 connmgr_flushed(p->connmgr);
1352 p->state = S_OPENFLOW;
1360 if (time_msec() >= p->next_op_report) {
1361 long long int ago = (time_msec() - p->first_op) / 1000;
1362 long long int interval = (p->last_op - p->first_op) / 1000;
1366 ds_put_format(&s, "%d flow_mods ",
1367 p->n_add + p->n_delete + p->n_modify);
1368 if (interval == ago) {
1369 ds_put_format(&s, "in the last %lld s", ago);
1370 } else if (interval) {
1371 ds_put_format(&s, "in the %lld s starting %lld s ago",
1374 ds_put_format(&s, "%lld s ago", ago);
1377 ds_put_cstr(&s, " (");
1379 ds_put_format(&s, "%d adds, ", p->n_add);
1382 ds_put_format(&s, "%d deletes, ", p->n_delete);
1385 ds_put_format(&s, "%d modifications, ", p->n_modify);
1388 ds_put_char(&s, ')');
1390 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1393 p->n_add = p->n_delete = p->n_modify = 0;
1394 p->next_op_report = LLONG_MAX;
1400 /* Performs periodic activity required by 'ofproto' that needs to be done
1401 * with the least possible latency.
1403 * It makes sense to call this function a couple of times per poll loop, to
1404 * provide a significant performance boost on some benchmarks with the
1405 * ofproto-dpif implementation. */
1407 ofproto_run_fast(struct ofproto *p)
1411 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1412 if (error && error != EAGAIN) {
1413 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1414 p->name, ovs_strerror(error));
1420 ofproto_wait(struct ofproto *p)
1422 struct ofport *ofport;
1424 p->ofproto_class->wait(p);
1425 if (p->ofproto_class->port_poll_wait) {
1426 p->ofproto_class->port_poll_wait(p);
1429 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1430 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1431 poll_immediate_wake();
1437 connmgr_wait(p->connmgr, true);
1442 connmgr_wait(p->connmgr, false);
1443 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1444 poll_immediate_wake();
1451 ofproto_is_alive(const struct ofproto *p)
1453 return connmgr_has_controllers(p->connmgr);
1456 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1457 * memory_report(). */
1459 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1461 const struct oftable *table;
1462 unsigned int n_rules;
1464 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1465 simap_increase(usage, "ops",
1466 ofproto->n_pending + hmap_count(&ofproto->deletions));
1469 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1470 ovs_rwlock_rdlock(&table->cls.rwlock);
1471 n_rules += classifier_count(&table->cls);
1472 ovs_rwlock_unlock(&table->cls.rwlock);
1474 simap_increase(usage, "rules", n_rules);
1476 if (ofproto->ofproto_class->get_memory_usage) {
1477 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1480 connmgr_get_memory_usage(ofproto->connmgr, usage);
1484 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1487 connmgr_get_controller_info(ofproto->connmgr, info);
1491 ofproto_free_ofproto_controller_info(struct shash *info)
1493 connmgr_free_controller_info(info);
1496 /* Makes a deep copy of 'old' into 'port'. */
1498 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1500 port->name = xstrdup(old->name);
1501 port->type = xstrdup(old->type);
1502 port->ofp_port = old->ofp_port;
1505 /* Frees memory allocated to members of 'ofproto_port'.
1507 * Do not call this function on an ofproto_port obtained from
1508 * ofproto_port_dump_next(): that function retains ownership of the data in the
1511 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1513 free(ofproto_port->name);
1514 free(ofproto_port->type);
1517 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1519 * This function provides no status indication. An error status for the entire
1520 * dump operation is provided when it is completed by calling
1521 * ofproto_port_dump_done().
1524 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1525 const struct ofproto *ofproto)
1527 dump->ofproto = ofproto;
1528 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1532 /* Attempts to retrieve another port from 'dump', which must have been created
1533 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1534 * 'port' and returns true. On failure, returns false.
1536 * Failure might indicate an actual error or merely that the last port has been
1537 * dumped. An error status for the entire dump operation is provided when it
1538 * is completed by calling ofproto_port_dump_done().
1540 * The ofproto owns the data stored in 'port'. It will remain valid until at
1541 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1542 * ofproto_port_dump_done(). */
1544 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1545 struct ofproto_port *port)
1547 const struct ofproto *ofproto = dump->ofproto;
1553 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1556 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1562 /* Completes port table dump operation 'dump', which must have been created
1563 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1564 * error-free, otherwise a positive errno value describing the problem. */
1566 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1568 const struct ofproto *ofproto = dump->ofproto;
1570 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1573 return dump->error == EOF ? 0 : dump->error;
1576 /* Returns the type to pass to netdev_open() when a datapath of type
1577 * 'datapath_type' has a port of type 'port_type', for a few special
1578 * cases when a netdev type differs from a port type. For example, when
1579 * using the userspace datapath, a port of type "internal" needs to be
1582 * Returns either 'type' itself or a string literal, which must not be
1585 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1587 const struct ofproto_class *class;
1589 datapath_type = ofproto_normalize_type(datapath_type);
1590 class = ofproto_class_find__(datapath_type);
1595 return (class->port_open_type
1596 ? class->port_open_type(datapath_type, port_type)
1600 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1601 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1602 * the port's OpenFlow port number.
1604 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1605 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1606 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1607 * 'ofp_portp' is non-null). */
1609 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1610 ofp_port_t *ofp_portp)
1612 ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1615 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1617 const char *netdev_name = netdev_get_name(netdev);
1619 simap_put(&ofproto->ofp_requests, netdev_name,
1620 ofp_to_u16(ofp_port));
1621 update_port(ofproto, netdev_name);
1624 struct ofproto_port ofproto_port;
1626 ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
1628 *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
1629 ofproto_port_destroy(&ofproto_port);
1634 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1635 * initializes '*port' appropriately; on failure, returns a positive errno
1638 * The caller owns the data in 'ofproto_port' and must free it with
1639 * ofproto_port_destroy() when it is no longer needed. */
1641 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1642 struct ofproto_port *port)
1646 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1648 memset(port, 0, sizeof *port);
1653 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1654 * Returns 0 if successful, otherwise a positive errno. */
1656 ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port)
1658 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1659 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1660 struct simap_node *ofp_request_node;
1663 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1664 if (ofp_request_node) {
1665 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1668 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1669 if (!error && ofport) {
1670 /* 'name' is the netdev's name and update_port() is going to close the
1671 * netdev. Just in case update_port() refers to 'name' after it
1672 * destroys 'ofport', make a copy of it around the update_port()
1674 char *devname = xstrdup(name);
1675 update_port(ofproto, devname);
1681 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1682 * performs the 'n_actions' actions in 'actions'. The new flow will not
1685 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1686 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1687 * controllers; otherwise, it will be hidden.
1689 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1691 * This is a helper function for in-band control and fail-open. */
1693 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1694 unsigned int priority,
1695 const struct ofpact *ofpacts, size_t ofpacts_len)
1697 const struct rule *rule;
1699 ovs_rwlock_rdlock(&ofproto->tables[0].cls.rwlock);
1700 rule = rule_from_cls_rule(classifier_find_match_exactly(
1701 &ofproto->tables[0].cls, match, priority));
1702 ovs_rwlock_unlock(&ofproto->tables[0].cls.rwlock);
1703 if (!rule || !ofpacts_equal(rule->ofpacts, rule->ofpacts_len,
1704 ofpacts, ofpacts_len)) {
1705 struct ofputil_flow_mod fm;
1707 memset(&fm, 0, sizeof fm);
1709 fm.priority = priority;
1710 fm.buffer_id = UINT32_MAX;
1711 fm.ofpacts = xmemdup(ofpacts, ofpacts_len);
1712 fm.ofpacts_len = ofpacts_len;
1713 add_flow(ofproto, NULL, &fm, NULL);
1718 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1719 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1720 * operation cannot be initiated now but may be retried later.
1722 * This is a helper function for in-band control and fail-open. */
1724 ofproto_flow_mod(struct ofproto *ofproto, struct ofputil_flow_mod *fm)
1726 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1729 /* Searches for a rule with matching criteria exactly equal to 'target' in
1730 * ofproto's table 0 and, if it finds one, deletes it.
1732 * This is a helper function for in-band control and fail-open. */
1734 ofproto_delete_flow(struct ofproto *ofproto,
1735 const struct match *target, unsigned int priority)
1737 struct classifier *cls = &ofproto->tables[0].cls;
1740 ovs_rwlock_rdlock(&cls->rwlock);
1741 rule = rule_from_cls_rule(classifier_find_match_exactly(cls, target,
1743 ovs_rwlock_unlock(&cls->rwlock);
1745 /* No such rule -> success. */
1747 } else if (rule->pending) {
1748 /* An operation on the rule is already pending -> failure.
1749 * Caller must retry later if it's important. */
1752 /* Initiate deletion -> success. */
1753 ovs_rwlock_wrlock(&cls->rwlock);
1754 ofproto_delete_rule(ofproto, cls, rule);
1755 ovs_rwlock_unlock(&cls->rwlock);
1762 /* Starts the process of deleting all of the flows from all of ofproto's flow
1763 * tables and then reintroducing the flows required by in-band control and
1764 * fail-open. The process will complete in a later call to ofproto_run(). */
1766 ofproto_flush_flows(struct ofproto *ofproto)
1768 COVERAGE_INC(ofproto_flush);
1769 ofproto->state = S_FLUSH;
1773 reinit_ports(struct ofproto *p)
1775 struct ofproto_port_dump dump;
1776 struct sset devnames;
1777 struct ofport *ofport;
1778 struct ofproto_port ofproto_port;
1779 const char *devname;
1781 COVERAGE_INC(ofproto_reinit_ports);
1783 sset_init(&devnames);
1784 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1785 sset_add(&devnames, netdev_get_name(ofport->netdev));
1787 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1788 sset_add(&devnames, ofproto_port.name);
1791 SSET_FOR_EACH (devname, &devnames) {
1792 update_port(p, devname);
1794 sset_destroy(&devnames);
1798 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
1802 port_idx = simap_get(&ofproto->ofp_requests, netdev_name);
1803 port_idx = port_idx ? port_idx : UINT16_MAX;
1805 if (port_idx >= ofproto->max_ports
1806 || bitmap_is_set(ofproto->ofp_port_ids, port_idx)) {
1807 uint16_t end_port_no = ofproto->alloc_port_no;
1809 /* Search for a free OpenFlow port number. We try not to
1810 * immediately reuse them to prevent problems due to old
1813 if (++ofproto->alloc_port_no >= ofproto->max_ports) {
1814 ofproto->alloc_port_no = 0;
1816 if (!bitmap_is_set(ofproto->ofp_port_ids,
1817 ofproto->alloc_port_no)) {
1818 port_idx = ofproto->alloc_port_no;
1821 if (ofproto->alloc_port_no == end_port_no) {
1826 bitmap_set1(ofproto->ofp_port_ids, port_idx);
1827 return u16_to_ofp(port_idx);
1831 dealloc_ofp_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
1833 if (ofp_to_u16(ofp_port) < ofproto->max_ports) {
1834 bitmap_set0(ofproto->ofp_port_ids, ofp_to_u16(ofp_port));
1838 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1839 * pointer if the netdev cannot be opened. On success, also fills in
1841 static struct netdev *
1842 ofport_open(struct ofproto *ofproto,
1843 struct ofproto_port *ofproto_port,
1844 struct ofputil_phy_port *pp)
1846 enum netdev_flags flags;
1847 struct netdev *netdev;
1850 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1852 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
1853 "cannot be opened (%s)",
1855 ofproto_port->name, ofproto_port->ofp_port,
1856 ofproto_port->name, ovs_strerror(error));
1860 if (ofproto_port->ofp_port == OFPP_NONE) {
1861 if (!strcmp(ofproto->name, ofproto_port->name)) {
1862 ofproto_port->ofp_port = OFPP_LOCAL;
1864 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
1865 ofproto_port->name);
1868 pp->port_no = ofproto_port->ofp_port;
1869 netdev_get_etheraddr(netdev, pp->hw_addr);
1870 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
1871 netdev_get_flags(netdev, &flags);
1872 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
1873 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
1874 netdev_get_features(netdev, &pp->curr, &pp->advertised,
1875 &pp->supported, &pp->peer);
1876 pp->curr_speed = netdev_features_to_bps(pp->curr, 0) / 1000;
1877 pp->max_speed = netdev_features_to_bps(pp->supported, 0) / 1000;
1882 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1883 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1886 ofport_equal(const struct ofputil_phy_port *a,
1887 const struct ofputil_phy_port *b)
1889 return (eth_addr_equals(a->hw_addr, b->hw_addr)
1890 && a->state == b->state
1891 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
1892 && a->curr == b->curr
1893 && a->advertised == b->advertised
1894 && a->supported == b->supported
1895 && a->peer == b->peer
1896 && a->curr_speed == b->curr_speed
1897 && a->max_speed == b->max_speed);
1900 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1901 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1902 * one with the same name or port number). */
1904 ofport_install(struct ofproto *p,
1905 struct netdev *netdev, const struct ofputil_phy_port *pp)
1907 const char *netdev_name = netdev_get_name(netdev);
1908 struct ofport *ofport;
1911 /* Create ofport. */
1912 ofport = p->ofproto_class->port_alloc();
1917 ofport->ofproto = p;
1918 ofport->netdev = netdev;
1919 ofport->change_seq = netdev_change_seq(netdev);
1921 ofport->ofp_port = pp->port_no;
1922 ofport->created = time_msec();
1924 /* Add port to 'p'. */
1925 hmap_insert(&p->ports, &ofport->hmap_node,
1926 hash_ofp_port(ofport->ofp_port));
1927 shash_add(&p->port_by_name, netdev_name, ofport);
1929 update_mtu(p, ofport);
1931 /* Let the ofproto_class initialize its private data. */
1932 error = p->ofproto_class->port_construct(ofport);
1936 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
1940 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1941 p->name, netdev_name, ovs_strerror(error));
1943 ofport_destroy__(ofport);
1945 netdev_close(netdev);
1949 /* Removes 'ofport' from 'p' and destroys it. */
1951 ofport_remove(struct ofport *ofport)
1953 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
1955 ofport_destroy(ofport);
1958 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1961 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1963 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1965 ofport_remove(port);
1969 /* Updates 'port' with new 'pp' description.
1971 * Does not handle a name or port number change. The caller must implement
1972 * such a change as a delete followed by an add. */
1974 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
1976 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
1977 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
1978 | (pp->config & OFPUTIL_PC_PORT_DOWN));
1979 port->pp.state = pp->state;
1980 port->pp.curr = pp->curr;
1981 port->pp.advertised = pp->advertised;
1982 port->pp.supported = pp->supported;
1983 port->pp.peer = pp->peer;
1984 port->pp.curr_speed = pp->curr_speed;
1985 port->pp.max_speed = pp->max_speed;
1987 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
1990 /* Update OpenFlow 'state' in 'port' and notify controller. */
1992 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
1994 if (port->pp.state != state) {
1995 port->pp.state = state;
1996 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
2002 ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port)
2004 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
2006 if (port->ofproto->ofproto_class->set_realdev) {
2007 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
2009 if (port->ofproto->ofproto_class->set_stp_port) {
2010 port->ofproto->ofproto_class->set_stp_port(port, NULL);
2012 if (port->ofproto->ofproto_class->set_cfm) {
2013 port->ofproto->ofproto_class->set_cfm(port, NULL);
2015 if (port->ofproto->ofproto_class->bundle_remove) {
2016 port->ofproto->ofproto_class->bundle_remove(port);
2022 ofport_destroy__(struct ofport *port)
2024 struct ofproto *ofproto = port->ofproto;
2025 const char *name = netdev_get_name(port->netdev);
2027 hmap_remove(&ofproto->ports, &port->hmap_node);
2028 shash_delete(&ofproto->port_by_name,
2029 shash_find(&ofproto->port_by_name, name));
2031 netdev_close(port->netdev);
2032 ofproto->ofproto_class->port_dealloc(port);
2036 ofport_destroy(struct ofport *port)
2039 dealloc_ofp_port(port->ofproto, port->ofp_port);
2040 port->ofproto->ofproto_class->port_destruct(port);
2041 ofport_destroy__(port);
2046 ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
2048 struct ofport *port;
2050 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_ofp_port(ofp_port),
2052 if (port->ofp_port == ofp_port) {
2060 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
2062 struct ofproto *ofproto = port->ofproto;
2065 if (ofproto->ofproto_class->port_get_stats) {
2066 error = ofproto->ofproto_class->port_get_stats(port, stats);
2075 update_port(struct ofproto *ofproto, const char *name)
2077 struct ofproto_port ofproto_port;
2078 struct ofputil_phy_port pp;
2079 struct netdev *netdev;
2080 struct ofport *port;
2082 COVERAGE_INC(ofproto_update_port);
2084 /* Fetch 'name''s location and properties from the datapath. */
2085 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
2086 ? ofport_open(ofproto, &ofproto_port, &pp)
2090 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
2091 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
2092 struct netdev *old_netdev = port->netdev;
2094 /* 'name' hasn't changed location. Any properties changed? */
2095 if (!ofport_equal(&port->pp, &pp)) {
2096 ofport_modified(port, &pp);
2099 update_mtu(ofproto, port);
2101 /* Install the newly opened netdev in case it has changed.
2102 * Don't close the old netdev yet in case port_modified has to
2103 * remove a retained reference to it.*/
2104 port->netdev = netdev;
2105 port->change_seq = netdev_change_seq(netdev);
2107 if (port->ofproto->ofproto_class->port_modified) {
2108 port->ofproto->ofproto_class->port_modified(port);
2111 netdev_close(old_netdev);
2113 /* If 'port' is nonnull then its name differs from 'name' and thus
2114 * we should delete it. If we think there's a port named 'name'
2115 * then its port number must be wrong now so delete it too. */
2117 ofport_remove(port);
2119 ofport_remove_with_name(ofproto, name);
2120 ofport_install(ofproto, netdev, &pp);
2123 /* Any port named 'name' is gone now. */
2124 ofport_remove_with_name(ofproto, name);
2126 ofproto_port_destroy(&ofproto_port);
2130 init_ports(struct ofproto *p)
2132 struct ofproto_port_dump dump;
2133 struct ofproto_port ofproto_port;
2134 struct shash_node *node, *next;
2136 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2137 const char *name = ofproto_port.name;
2139 if (shash_find(&p->port_by_name, name)) {
2140 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2143 struct ofputil_phy_port pp;
2144 struct netdev *netdev;
2146 /* Check if an OpenFlow port number had been requested. */
2147 node = shash_find(&init_ofp_ports, name);
2149 const struct iface_hint *iface_hint = node->data;
2150 simap_put(&p->ofp_requests, name,
2151 ofp_to_u16(iface_hint->ofp_port));
2154 netdev = ofport_open(p, &ofproto_port, &pp);
2156 ofport_install(p, netdev, &pp);
2157 if (ofp_to_u16(ofproto_port.ofp_port) < p->max_ports) {
2158 p->alloc_port_no = MAX(p->alloc_port_no,
2159 ofp_to_u16(ofproto_port.ofp_port));
2165 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2166 struct iface_hint *iface_hint = node->data;
2168 if (!strcmp(iface_hint->br_name, p->name)) {
2169 free(iface_hint->br_name);
2170 free(iface_hint->br_type);
2172 shash_delete(&init_ofp_ports, node);
2179 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2180 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2182 find_min_mtu(struct ofproto *p)
2184 struct ofport *ofport;
2187 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2188 struct netdev *netdev = ofport->netdev;
2191 /* Skip any internal ports, since that's what we're trying to
2193 if (!strcmp(netdev_get_type(netdev), "internal")) {
2197 if (netdev_get_mtu(netdev, &dev_mtu)) {
2200 if (!mtu || dev_mtu < mtu) {
2205 return mtu ? mtu: ETH_PAYLOAD_MAX;
2208 /* Update MTU of all datapath devices on 'p' to the minimum of the
2209 * non-datapath ports in event of 'port' added or changed. */
2211 update_mtu(struct ofproto *p, struct ofport *port)
2213 struct ofport *ofport;
2214 struct netdev *netdev = port->netdev;
2215 int dev_mtu, old_min;
2217 if (netdev_get_mtu(netdev, &dev_mtu)) {
2221 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2222 if (dev_mtu > p->min_mtu) {
2223 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2224 dev_mtu = p->min_mtu;
2227 port->mtu = dev_mtu;
2231 /* For non-internal port find new min mtu. */
2232 old_min = p->min_mtu;
2233 port->mtu = dev_mtu;
2234 p->min_mtu = find_min_mtu(p);
2235 if (p->min_mtu == old_min) {
2239 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2240 struct netdev *netdev = ofport->netdev;
2242 if (!strcmp(netdev_get_type(netdev), "internal")) {
2243 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2244 ofport->mtu = p->min_mtu;
2251 ofproto_rule_destroy(struct rule *rule)
2254 rule->ofproto->ofproto_class->rule_destruct(rule);
2255 ofproto_rule_destroy__(rule);
2260 ofproto_rule_destroy__(struct rule *rule)
2262 cls_rule_destroy(&rule->cr);
2263 free(rule->ofpacts);
2264 ovs_mutex_destroy(&rule->timeout_mutex);
2265 ovs_rwlock_destroy(&rule->rwlock);
2266 rule->ofproto->ofproto_class->rule_dealloc(rule);
2269 /* This function allows an ofproto implementation to destroy any rules that
2270 * remain when its ->destruct() function is called.. This function implements
2271 * steps 4.4 and 4.5 in the section titled "Rule Life Cycle" in
2272 * ofproto-provider.h.
2274 * This function should only be called from an ofproto implementation's
2275 * ->destruct() function. It is not suitable elsewhere. */
2277 ofproto_rule_delete(struct ofproto *ofproto, struct classifier *cls,
2279 OVS_REQ_WRLOCK(cls->rwlock)
2281 ofproto_delete_rule(ofproto, cls, rule);
2284 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2285 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2287 ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port)
2289 return (port == OFPP_ANY
2290 || ofpacts_output_to_port(rule->ofpacts, rule->ofpacts_len, port));
2293 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2294 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2296 ofoperation_has_out_port(const struct ofoperation *op, ofp_port_t out_port)
2298 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2303 case OFOPERATION_ADD:
2304 case OFOPERATION_DELETE:
2307 case OFOPERATION_MODIFY:
2308 case OFOPERATION_REPLACE:
2309 return ofpacts_output_to_port(op->ofpacts, op->ofpacts_len, out_port);
2315 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2316 * statistics appropriately.
2318 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2319 * with statistics for 'packet' either way.
2321 * Takes ownership of 'packet'. */
2323 rule_execute(struct rule *rule, ofp_port_t in_port, struct ofpbuf *packet)
2326 union flow_in_port in_port_;
2328 in_port_.ofp_port = in_port;
2329 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
2330 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
2333 /* Returns true if 'rule' should be hidden from the controller.
2335 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2336 * (e.g. by in-band control) and are intentionally hidden from the
2339 ofproto_rule_is_hidden(const struct rule *rule)
2341 return rule->cr.priority > UINT16_MAX;
2344 static enum oftable_flags
2345 rule_get_flags(const struct rule *rule)
2347 return rule->ofproto->tables[rule->table_id].flags;
2351 rule_is_modifiable(const struct rule *rule)
2353 return !(rule_get_flags(rule) & OFTABLE_READONLY);
2357 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2359 ofconn_send_reply(ofconn, make_echo_reply(oh));
2364 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2366 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2367 struct ofputil_switch_features features;
2368 struct ofport *port;
2374 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2376 ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2378 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2379 * if present, are always at the end.) */
2380 n_tables = ofproto->n_tables;
2381 for (i = 0; i < ofproto->n_tables; i++) {
2382 if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
2388 features.datapath_id = ofproto->datapath_id;
2389 features.n_buffers = pktbuf_capacity();
2390 features.n_tables = n_tables;
2391 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2392 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2394 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2396 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2397 features.auxiliary_id = 0;
2398 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2400 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2401 ofputil_put_switch_features_port(&port->pp, b);
2404 ofconn_send_reply(ofconn, b);
2409 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2411 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2412 struct ofp_switch_config *osc;
2413 enum ofp_config_flags flags;
2417 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2418 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2419 flags = ofproto->frag_handling;
2420 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2421 if (oh->version < OFP13_VERSION
2422 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2423 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2425 osc->flags = htons(flags);
2426 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2427 ofconn_send_reply(ofconn, buf);
2433 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2435 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2436 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2437 uint16_t flags = ntohs(osc->flags);
2439 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2440 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
2441 enum ofp_config_flags cur = ofproto->frag_handling;
2442 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2444 ovs_assert((cur & OFPC_FRAG_MASK) == cur);
2446 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2447 ofproto->frag_handling = next;
2449 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2451 ofputil_frag_handling_to_string(next));
2455 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2456 ofconn_set_invalid_ttl_to_controller(ofconn,
2457 (oh->version < OFP13_VERSION
2458 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2460 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2465 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2466 * error message code for the caller to propagate upward. Otherwise, returns
2469 * The log message mentions 'msg_type'. */
2471 reject_slave_controller(struct ofconn *ofconn)
2473 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2474 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
2475 return OFPERR_OFPBRC_EPERM;
2481 /* Finds the OFPACT_METER action, if any, in the 'ofpacts_len' bytes of
2482 * 'ofpacts'. If found, returns its meter ID; if not, returns 0.
2484 * This function relies on the order of 'ofpacts' being correct (as checked by
2485 * ofpacts_verify()). */
2487 find_meter(const struct ofpact ofpacts[], size_t ofpacts_len)
2489 const struct ofpact *a;
2491 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2492 enum ovs_instruction_type inst;
2494 inst = ovs_instruction_type_from_ofpact_type(a->type);
2495 if (a->type == OFPACT_METER) {
2496 return ofpact_get_METER(a)->meter_id;
2497 } else if (inst > OVSINST_OFPIT13_METER) {
2505 /* Checks that the 'ofpacts_len' bytes of actions in 'ofpacts' are appropriate
2506 * for a packet with the prerequisites satisfied by 'flow' in table 'table_id'.
2507 * 'flow' may be temporarily modified, but is restored at return.
2510 ofproto_check_ofpacts(struct ofproto *ofproto,
2511 const struct ofpact ofpacts[], size_t ofpacts_len,
2512 struct flow *flow, uint8_t table_id)
2517 error = ofpacts_check(ofpacts, ofpacts_len, flow,
2518 u16_to_ofp(ofproto->max_ports), table_id);
2523 mid = find_meter(ofpacts, ofpacts_len);
2524 if (mid && ofproto_get_provider_meter_id(ofproto, mid) == UINT32_MAX) {
2525 return OFPERR_OFPMMFC_INVALID_METER;
2531 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2533 struct ofproto *p = ofconn_get_ofproto(ofconn);
2534 struct ofputil_packet_out po;
2535 struct ofpbuf *payload;
2536 uint64_t ofpacts_stub[1024 / 8];
2537 struct ofpbuf ofpacts;
2539 union flow_in_port in_port_;
2542 COVERAGE_INC(ofproto_packet_out);
2544 error = reject_slave_controller(ofconn);
2549 /* Decode message. */
2550 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2551 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2553 goto exit_free_ofpacts;
2555 if (ofp_to_u16(po.in_port) >= p->max_ports
2556 && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) {
2557 error = OFPERR_OFPBRC_BAD_PORT;
2558 goto exit_free_ofpacts;
2563 if (po.buffer_id != UINT32_MAX) {
2564 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2565 if (error || !payload) {
2566 goto exit_free_ofpacts;
2569 /* Ensure that the L3 header is 32-bit aligned. */
2570 payload = ofpbuf_clone_data_with_headroom(po.packet, po.packet_len, 2);
2573 /* Verify actions against packet, then send packet if successful. */
2574 in_port_.ofp_port = po.in_port;
2575 flow_extract(payload, 0, 0, NULL, &in_port_, &flow);
2576 error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len, &flow, 0);
2578 error = p->ofproto_class->packet_out(p, payload, &flow,
2579 po.ofpacts, po.ofpacts_len);
2581 ofpbuf_delete(payload);
2584 ofpbuf_uninit(&ofpacts);
2590 update_port_config(struct ofport *port,
2591 enum ofputil_port_config config,
2592 enum ofputil_port_config mask)
2594 enum ofputil_port_config old_config = port->pp.config;
2595 enum ofputil_port_config toggle;
2597 toggle = (config ^ port->pp.config) & mask;
2598 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2599 if (config & OFPUTIL_PC_PORT_DOWN) {
2600 netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL);
2602 netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL);
2604 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2607 port->pp.config ^= toggle;
2608 if (port->pp.config != old_config) {
2609 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2614 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2616 struct ofproto *p = ofconn_get_ofproto(ofconn);
2617 struct ofputil_port_mod pm;
2618 struct ofport *port;
2621 error = reject_slave_controller(ofconn);
2626 error = ofputil_decode_port_mod(oh, &pm);
2631 port = ofproto_get_port(p, pm.port_no);
2633 return OFPERR_OFPPMFC_BAD_PORT;
2634 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2635 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2637 update_port_config(port, pm.config, pm.mask);
2639 netdev_set_advertisements(port->netdev, pm.advertise);
2646 handle_desc_stats_request(struct ofconn *ofconn,
2647 const struct ofp_header *request)
2649 static const char *default_mfr_desc = "Nicira, Inc.";
2650 static const char *default_hw_desc = "Open vSwitch";
2651 static const char *default_sw_desc = VERSION;
2652 static const char *default_serial_desc = "None";
2653 static const char *default_dp_desc = "None";
2655 struct ofproto *p = ofconn_get_ofproto(ofconn);
2656 struct ofp_desc_stats *ods;
2659 msg = ofpraw_alloc_stats_reply(request, 0);
2660 ods = ofpbuf_put_zeros(msg, sizeof *ods);
2661 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
2662 sizeof ods->mfr_desc);
2663 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
2664 sizeof ods->hw_desc);
2665 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
2666 sizeof ods->sw_desc);
2667 ovs_strlcpy(ods->serial_num,
2668 p->serial_desc ? p->serial_desc : default_serial_desc,
2669 sizeof ods->serial_num);
2670 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
2671 sizeof ods->dp_desc);
2672 ofconn_send_reply(ofconn, msg);
2678 handle_table_stats_request(struct ofconn *ofconn,
2679 const struct ofp_header *request)
2681 struct ofproto *p = ofconn_get_ofproto(ofconn);
2682 struct ofp12_table_stats *ots;
2687 /* Set up default values.
2689 * ofp12_table_stats is used as a generic structure as
2690 * it is able to hold all the fields for ofp10_table_stats
2691 * and ofp11_table_stats (and of course itself).
2693 ots = xcalloc(p->n_tables, sizeof *ots);
2694 for (i = 0; i < p->n_tables; i++) {
2695 ots[i].table_id = i;
2696 sprintf(ots[i].name, "table%zu", i);
2697 ots[i].match = htonll(OFPXMT12_MASK);
2698 ots[i].wildcards = htonll(OFPXMT12_MASK);
2699 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
2700 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
2701 ots[i].write_setfields = htonll(OFPXMT12_MASK);
2702 ots[i].apply_setfields = htonll(OFPXMT12_MASK);
2703 ots[i].metadata_match = htonll(UINT64_MAX);
2704 ots[i].metadata_write = htonll(UINT64_MAX);
2705 ots[i].instructions = htonl(OFPIT11_ALL);
2706 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
2707 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2708 ovs_rwlock_rdlock(&p->tables[i].cls.rwlock);
2709 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2710 ovs_rwlock_unlock(&p->tables[i].cls.rwlock);
2713 p->ofproto_class->get_tables(p, ots);
2715 /* Post-process the tables, dropping hidden tables. */
2716 n_tables = p->n_tables;
2717 for (i = 0; i < p->n_tables; i++) {
2718 const struct oftable *table = &p->tables[i];
2720 if (table->flags & OFTABLE_HIDDEN) {
2726 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2729 if (table->max_flows < ntohl(ots[i].max_entries)) {
2730 ots[i].max_entries = htonl(table->max_flows);
2734 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
2735 ofconn_send_reply(ofconn, msg);
2743 append_port_stat(struct ofport *port, struct list *replies)
2745 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
2747 calc_duration(port->created, time_msec(),
2748 &ops.duration_sec, &ops.duration_nsec);
2750 /* Intentionally ignore return value, since errors will set
2751 * 'stats' to all-1s, which is correct for OpenFlow, and
2752 * netdev_get_stats() will log errors. */
2753 ofproto_port_get_stats(port, &ops.stats);
2755 ofputil_append_port_stat(replies, &ops);
2759 handle_port_stats_request(struct ofconn *ofconn,
2760 const struct ofp_header *request)
2762 struct ofproto *p = ofconn_get_ofproto(ofconn);
2763 struct ofport *port;
2764 struct list replies;
2768 error = ofputil_decode_port_stats_request(request, &port_no);
2773 ofpmp_init(&replies, request);
2774 if (port_no != OFPP_ANY) {
2775 port = ofproto_get_port(p, port_no);
2777 append_port_stat(port, &replies);
2780 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2781 append_port_stat(port, &replies);
2785 ofconn_send_replies(ofconn, &replies);
2790 handle_port_desc_stats_request(struct ofconn *ofconn,
2791 const struct ofp_header *request)
2793 struct ofproto *p = ofconn_get_ofproto(ofconn);
2794 enum ofp_version version;
2795 struct ofport *port;
2796 struct list replies;
2798 ofpmp_init(&replies, request);
2800 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
2801 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2802 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
2805 ofconn_send_replies(ofconn, &replies);
2810 hash_cookie(ovs_be64 cookie)
2812 return hash_2words((OVS_FORCE uint64_t)cookie >> 32,
2813 (OVS_FORCE uint64_t)cookie);
2817 cookies_insert(struct ofproto *ofproto, struct rule *rule)
2819 hindex_insert(&ofproto->cookies, &rule->cookie_node,
2820 hash_cookie(rule->flow_cookie));
2824 cookies_remove(struct ofproto *ofproto, struct rule *rule)
2826 hindex_remove(&ofproto->cookies, &rule->cookie_node);
2830 ofproto_rule_change_cookie(struct ofproto *ofproto, struct rule *rule,
2831 ovs_be64 new_cookie)
2833 if (new_cookie != rule->flow_cookie) {
2834 cookies_remove(ofproto, rule);
2836 ovs_rwlock_wrlock(&rule->rwlock);
2837 rule->flow_cookie = new_cookie;
2838 ovs_rwlock_unlock(&rule->rwlock);
2840 cookies_insert(ofproto, rule);
2845 calc_duration(long long int start, long long int now,
2846 uint32_t *sec, uint32_t *nsec)
2848 long long int msecs = now - start;
2849 *sec = msecs / 1000;
2850 *nsec = (msecs % 1000) * (1000 * 1000);
2853 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2854 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2856 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
2858 return (table_id == 0xff || table_id < ofproto->n_tables
2860 : OFPERR_OFPBRC_BAD_TABLE_ID);
2864 static struct oftable *
2865 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
2867 struct oftable *table;
2869 for (table = &ofproto->tables[table_id];
2870 table < &ofproto->tables[ofproto->n_tables];
2872 if (!(table->flags & OFTABLE_HIDDEN)) {
2880 static struct oftable *
2881 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
2883 if (table_id == 0xff) {
2884 return next_visible_table(ofproto, 0);
2885 } else if (table_id < ofproto->n_tables) {
2886 return &ofproto->tables[table_id];
2892 static struct oftable *
2893 next_matching_table(const struct ofproto *ofproto,
2894 const struct oftable *table, uint8_t table_id)
2896 return (table_id == 0xff
2897 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
2901 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2903 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2904 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2906 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2907 * only once, for that table. (This can be used to access tables marked
2910 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2911 * entered at all. (Perhaps you should have validated TABLE_ID with
2912 * check_table_id().)
2914 * All parameters are evaluated multiple times.
2916 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2917 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2919 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2921 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2922 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2923 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2926 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2927 * to 'out_port' are included.
2929 * Hidden rules are always omitted.
2931 * Returns 0 on success, otherwise an OpenFlow error code. */
2933 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
2934 const struct match *match,
2935 ovs_be64 cookie, ovs_be64 cookie_mask,
2936 ofp_port_t out_port, struct list *rules)
2938 struct oftable *table;
2942 error = check_table_id(ofproto, table_id);
2948 cls_rule_init(&cr, match, 0);
2950 if (cookie_mask == htonll(UINT64_MAX)) {
2953 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node, hash_cookie(cookie),
2954 &ofproto->cookies) {
2955 if (table_id != rule->table_id && table_id != 0xff) {
2958 if (ofproto_rule_is_hidden(rule)) {
2961 if (cls_rule_is_loose_match(&rule->cr, &cr.match)) {
2962 if (rule->pending) {
2963 error = OFPROTO_POSTPONE;
2966 if (rule->flow_cookie == cookie /* Hash collisions possible. */
2967 && ofproto_rule_has_out_port(rule, out_port)) {
2968 list_push_back(rules, &rule->ofproto_node);
2975 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2976 struct cls_cursor cursor;
2979 ovs_rwlock_rdlock(&table->cls.rwlock);
2980 cls_cursor_init(&cursor, &table->cls, &cr);
2981 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2982 if (rule->pending) {
2983 ovs_rwlock_unlock(&table->cls.rwlock);
2984 error = OFPROTO_POSTPONE;
2987 if (!ofproto_rule_is_hidden(rule)
2988 && ofproto_rule_has_out_port(rule, out_port)
2989 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2990 list_push_back(rules, &rule->ofproto_node);
2993 ovs_rwlock_unlock(&table->cls.rwlock);
2997 cls_rule_destroy(&cr);
3001 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
3002 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
3003 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
3006 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
3007 * to 'out_port' are included.
3009 * Hidden rules are always omitted.
3011 * Returns 0 on success, otherwise an OpenFlow error code. */
3013 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
3014 const struct match *match, unsigned int priority,
3015 ovs_be64 cookie, ovs_be64 cookie_mask,
3016 ofp_port_t out_port, struct list *rules)
3018 struct oftable *table;
3022 error = check_table_id(ofproto, table_id);
3028 cls_rule_init(&cr, match, priority);
3030 if (cookie_mask == htonll(UINT64_MAX)) {
3033 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node, hash_cookie(cookie),
3034 &ofproto->cookies) {
3035 if (table_id != rule->table_id && table_id != 0xff) {
3038 if (ofproto_rule_is_hidden(rule)) {
3041 if (cls_rule_equal(&rule->cr, &cr)) {
3042 if (rule->pending) {
3043 error = OFPROTO_POSTPONE;
3046 if (rule->flow_cookie == cookie /* Hash collisions possible. */
3047 && ofproto_rule_has_out_port(rule, out_port)) {
3048 list_push_back(rules, &rule->ofproto_node);
3055 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
3058 ovs_rwlock_rdlock(&table->cls.rwlock);
3059 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls,
3061 ovs_rwlock_unlock(&table->cls.rwlock);
3063 if (rule->pending) {
3064 error = OFPROTO_POSTPONE;
3067 if (!ofproto_rule_is_hidden(rule)
3068 && ofproto_rule_has_out_port(rule, out_port)
3069 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
3070 list_push_back(rules, &rule->ofproto_node);
3076 cls_rule_destroy(&cr);
3080 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
3081 * forced into the range of a uint16_t. */
3083 age_secs(long long int age_ms)
3085 return (age_ms < 0 ? 0
3086 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
3087 : (unsigned int) age_ms / 1000);
3091 handle_flow_stats_request(struct ofconn *ofconn,
3092 const struct ofp_header *request)
3094 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3095 struct ofputil_flow_stats_request fsr;
3096 struct list replies;
3101 error = ofputil_decode_flow_stats_request(&fsr, request);
3106 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
3107 fsr.cookie, fsr.cookie_mask,
3108 fsr.out_port, &rules);
3113 ofpmp_init(&replies, request);
3114 LIST_FOR_EACH (rule, ofproto_node, &rules) {
3115 long long int now = time_msec();
3116 struct ofputil_flow_stats fs;
3118 minimatch_expand(&rule->cr.match, &fs.match);
3119 fs.priority = rule->cr.priority;
3120 fs.cookie = rule->flow_cookie;
3121 fs.table_id = rule->table_id;
3122 calc_duration(rule->created, now, &fs.duration_sec, &fs.duration_nsec);
3123 fs.idle_age = age_secs(now - rule->used);
3124 fs.hard_age = age_secs(now - rule->modified);
3125 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
3127 fs.ofpacts = rule->ofpacts;
3128 fs.ofpacts_len = rule->ofpacts_len;
3130 ovs_mutex_lock(&rule->timeout_mutex);
3131 fs.idle_timeout = rule->idle_timeout;
3132 fs.hard_timeout = rule->hard_timeout;
3133 ovs_mutex_unlock(&rule->timeout_mutex);
3136 if (rule->send_flow_removed) {
3137 fs.flags |= OFPUTIL_FF_SEND_FLOW_REM;
3138 /* FIXME: Implement OFPUTIL_FF_NO_PKT_COUNTS and
3139 OFPUTIL_FF_NO_BYT_COUNTS. */
3141 ofputil_append_flow_stats_reply(&fs, &replies);
3143 ofconn_send_replies(ofconn, &replies);
3149 flow_stats_ds(struct rule *rule, struct ds *results)
3151 uint64_t packet_count, byte_count;
3153 rule->ofproto->ofproto_class->rule_get_stats(rule,
3154 &packet_count, &byte_count);
3156 if (rule->table_id != 0) {
3157 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
3159 ds_put_format(results, "duration=%llds, ",
3160 (time_msec() - rule->created) / 1000);
3161 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3162 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3163 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3164 cls_rule_format(&rule->cr, results);
3165 ds_put_char(results, ',');
3166 ofpacts_format(rule->ofpacts, rule->ofpacts_len, results);
3167 ds_put_cstr(results, "\n");
3170 /* Adds a pretty-printed description of all flows to 'results', including
3171 * hidden flows (e.g., set up by in-band control). */
3173 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3175 struct oftable *table;
3177 OFPROTO_FOR_EACH_TABLE (table, p) {
3178 struct cls_cursor cursor;
3181 ovs_rwlock_rdlock(&table->cls.rwlock);
3182 cls_cursor_init(&cursor, &table->cls, NULL);
3183 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3184 flow_stats_ds(rule, results);
3186 ovs_rwlock_unlock(&table->cls.rwlock);
3190 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
3191 * '*engine_type' and '*engine_id', respectively. */
3193 ofproto_get_netflow_ids(const struct ofproto *ofproto,
3194 uint8_t *engine_type, uint8_t *engine_id)
3196 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
3199 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
3200 * true if the port's CFM status was successfully stored into '*status'.
3201 * Returns false if the port did not have CFM configured, in which case
3202 * '*status' is indeterminate.
3204 * The caller must provide and owns '*status', and must free 'status->rmps'. */
3206 ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port,
3207 struct ofproto_cfm_status *status)
3209 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
3211 && ofproto->ofproto_class->get_cfm_status
3212 && ofproto->ofproto_class->get_cfm_status(ofport, status));
3216 handle_aggregate_stats_request(struct ofconn *ofconn,
3217 const struct ofp_header *oh)
3219 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3220 struct ofputil_flow_stats_request request;
3221 struct ofputil_aggregate_stats stats;
3222 bool unknown_packets, unknown_bytes;
3223 struct ofpbuf *reply;
3228 error = ofputil_decode_flow_stats_request(&request, oh);
3233 error = collect_rules_loose(ofproto, request.table_id, &request.match,
3234 request.cookie, request.cookie_mask,
3235 request.out_port, &rules);
3240 memset(&stats, 0, sizeof stats);
3241 unknown_packets = unknown_bytes = false;
3242 LIST_FOR_EACH (rule, ofproto_node, &rules) {
3243 uint64_t packet_count;
3244 uint64_t byte_count;
3246 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
3249 if (packet_count == UINT64_MAX) {
3250 unknown_packets = true;
3252 stats.packet_count += packet_count;
3255 if (byte_count == UINT64_MAX) {
3256 unknown_bytes = true;
3258 stats.byte_count += byte_count;
3263 if (unknown_packets) {
3264 stats.packet_count = UINT64_MAX;
3266 if (unknown_bytes) {
3267 stats.byte_count = UINT64_MAX;
3270 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
3271 ofconn_send_reply(ofconn, reply);
3276 struct queue_stats_cbdata {
3277 struct ofport *ofport;
3278 struct list replies;
3283 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3284 const struct netdev_queue_stats *stats)
3286 struct ofputil_queue_stats oqs;
3288 oqs.port_no = cbdata->ofport->pp.port_no;
3289 oqs.queue_id = queue_id;
3290 oqs.tx_bytes = stats->tx_bytes;
3291 oqs.tx_packets = stats->tx_packets;
3292 oqs.tx_errors = stats->tx_errors;
3293 if (stats->created != LLONG_MIN) {
3294 calc_duration(stats->created, cbdata->now,
3295 &oqs.duration_sec, &oqs.duration_nsec);
3297 oqs.duration_sec = oqs.duration_nsec = UINT32_MAX;
3299 ofputil_append_queue_stat(&cbdata->replies, &oqs);
3303 handle_queue_stats_dump_cb(uint32_t queue_id,
3304 struct netdev_queue_stats *stats,
3307 struct queue_stats_cbdata *cbdata = cbdata_;
3309 put_queue_stats(cbdata, queue_id, stats);
3313 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3314 struct queue_stats_cbdata *cbdata)
3316 cbdata->ofport = port;
3317 if (queue_id == OFPQ_ALL) {
3318 netdev_dump_queue_stats(port->netdev,
3319 handle_queue_stats_dump_cb, cbdata);
3321 struct netdev_queue_stats stats;
3323 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3324 put_queue_stats(cbdata, queue_id, &stats);
3326 return OFPERR_OFPQOFC_BAD_QUEUE;
3333 handle_queue_stats_request(struct ofconn *ofconn,
3334 const struct ofp_header *rq)
3336 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3337 struct queue_stats_cbdata cbdata;
3338 struct ofport *port;
3340 struct ofputil_queue_stats_request oqsr;
3342 COVERAGE_INC(ofproto_queue_req);
3344 ofpmp_init(&cbdata.replies, rq);
3345 cbdata.now = time_msec();
3347 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3352 if (oqsr.port_no == OFPP_ANY) {
3353 error = OFPERR_OFPQOFC_BAD_QUEUE;
3354 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3355 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3360 port = ofproto_get_port(ofproto, oqsr.port_no);
3362 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3363 : OFPERR_OFPQOFC_BAD_PORT);
3366 ofconn_send_replies(ofconn, &cbdata.replies);
3368 ofpbuf_list_delete(&cbdata.replies);
3375 is_flow_deletion_pending(const struct ofproto *ofproto,
3376 const struct cls_rule *cls_rule,
3379 if (!hmap_is_empty(&ofproto->deletions)) {
3380 struct ofoperation *op;
3382 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3383 cls_rule_hash(cls_rule, table_id),
3384 &ofproto->deletions) {
3385 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
3395 evict_rule_from_table(struct ofproto *ofproto, struct oftable *table)
3400 ovs_rwlock_rdlock(&table->cls.rwlock);
3401 n_rules = classifier_count(&table->cls);
3402 ovs_rwlock_unlock(&table->cls.rwlock);
3404 if (n_rules < table->max_flows) {
3406 } else if (!choose_rule_to_evict(table, &rule)) {
3407 return OFPERR_OFPFMFC_TABLE_FULL;
3408 } else if (rule->pending) {
3409 ovs_rwlock_unlock(&rule->rwlock);
3410 return OFPROTO_POSTPONE;
3412 struct ofopgroup *group;
3414 group = ofopgroup_create_unattached(ofproto);
3415 delete_flow__(rule, group, OFPRR_EVICTION);
3416 ofopgroup_submit(group);
3422 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3423 * in which no matching flow already exists in the flow table.
3425 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3426 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3427 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3428 * initiated now but may be retried later.
3430 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
3431 * ownership remains with the caller.
3433 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3436 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
3437 struct ofputil_flow_mod *fm, const struct ofp_header *request)
3439 struct oftable *table;
3440 struct ofopgroup *group;
3446 error = check_table_id(ofproto, fm->table_id);
3452 if (fm->table_id == 0xff) {
3453 if (ofproto->ofproto_class->rule_choose_table) {
3454 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3460 ovs_assert(table_id < ofproto->n_tables);
3464 } else if (fm->table_id < ofproto->n_tables) {
3465 table_id = fm->table_id;
3467 return OFPERR_OFPBRC_BAD_TABLE_ID;
3470 table = &ofproto->tables[table_id];
3472 if (table->flags & OFTABLE_READONLY) {
3473 return OFPERR_OFPBRC_EPERM;
3476 cls_rule_init(&cr, &fm->match, fm->priority);
3478 /* Transform "add" into "modify" if there's an existing identical flow. */
3479 ovs_rwlock_rdlock(&table->cls.rwlock);
3480 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr));
3481 ovs_rwlock_unlock(&table->cls.rwlock);
3483 cls_rule_destroy(&cr);
3484 if (!rule_is_modifiable(rule)) {
3485 return OFPERR_OFPBRC_EPERM;
3486 } else if (rule->pending) {
3487 return OFPROTO_POSTPONE;
3492 list_push_back(&rules, &rule->ofproto_node);
3493 fm->modify_cookie = true;
3494 return modify_flows__(ofproto, ofconn, fm, request, &rules);
3498 /* Verify actions. */
3499 error = ofproto_check_ofpacts(ofproto, fm->ofpacts, fm->ofpacts_len,
3500 &fm->match.flow, table_id);
3502 cls_rule_destroy(&cr);
3506 /* Serialize against pending deletion. */
3507 if (is_flow_deletion_pending(ofproto, &cr, table_id)) {
3508 cls_rule_destroy(&cr);
3509 return OFPROTO_POSTPONE;
3512 /* Check for overlap, if requested. */
3513 if (fm->flags & OFPUTIL_FF_CHECK_OVERLAP) {
3516 ovs_rwlock_rdlock(&table->cls.rwlock);
3517 overlaps = classifier_rule_overlaps(&table->cls, &cr);
3518 ovs_rwlock_unlock(&table->cls.rwlock);
3521 cls_rule_destroy(&cr);
3522 return OFPERR_OFPFMFC_OVERLAP;
3526 /* If necessary, evict an existing rule to clear out space. */
3527 error = evict_rule_from_table(ofproto, table);
3529 cls_rule_destroy(&cr);
3533 /* Allocate new rule. */
3534 rule = ofproto->ofproto_class->rule_alloc();
3536 cls_rule_destroy(&cr);
3537 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
3538 ofproto->name, ovs_strerror(error));
3542 /* Initialize base state. */
3543 rule->ofproto = ofproto;
3544 cls_rule_move(&rule->cr, &cr);
3545 rule->pending = NULL;
3546 rule->flow_cookie = fm->new_cookie;
3547 rule->created = rule->modified = rule->used = time_msec();
3549 ovs_mutex_init(&rule->timeout_mutex);
3550 ovs_mutex_lock(&rule->timeout_mutex);
3551 rule->idle_timeout = fm->idle_timeout;
3552 rule->hard_timeout = fm->hard_timeout;
3553 ovs_mutex_unlock(&rule->timeout_mutex);
3555 rule->table_id = table - ofproto->tables;
3556 rule->send_flow_removed = (fm->flags & OFPUTIL_FF_SEND_FLOW_REM) != 0;
3557 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3558 rule->ofpacts_len = fm->ofpacts_len;
3559 rule->meter_id = find_meter(rule->ofpacts, rule->ofpacts_len);
3560 list_init(&rule->meter_list_node);
3561 rule->eviction_group = NULL;
3562 list_init(&rule->expirable);
3563 rule->monitor_flags = 0;
3564 rule->add_seqno = 0;
3565 rule->modify_seqno = 0;
3566 ovs_rwlock_init(&rule->rwlock);
3568 /* Construct rule, initializing derived state. */
3569 error = ofproto->ofproto_class->rule_construct(rule);
3571 ofproto_rule_destroy__(rule);
3576 oftable_insert_rule(rule);
3578 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3579 ofoperation_create(group, rule, OFOPERATION_ADD, 0);
3580 ofproto->ofproto_class->rule_insert(rule);
3581 ofopgroup_submit(group);
3586 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3588 /* Modifies the rules listed in 'rules', changing their actions to match those
3591 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3594 * Returns 0 on success, otherwise an OpenFlow error code. */
3596 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3597 struct ofputil_flow_mod *fm, const struct ofp_header *request,
3600 enum ofoperation_type type;
3601 struct ofopgroup *group;
3605 type = fm->command == OFPFC_ADD ? OFOPERATION_REPLACE : OFOPERATION_MODIFY;
3606 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3607 error = OFPERR_OFPBRC_EPERM;
3608 LIST_FOR_EACH (rule, ofproto_node, rules) {
3609 struct ofoperation *op;
3610 bool actions_changed;
3611 bool reset_counters;
3613 /* FIXME: Implement OFPFUTIL_FF_RESET_COUNTS */
3615 if (rule_is_modifiable(rule)) {
3616 /* At least one rule is modifiable, don't report EPERM error. */
3622 /* Verify actions. */
3623 error = ofpacts_check(fm->ofpacts, fm->ofpacts_len, &fm->match.flow,
3624 u16_to_ofp(ofproto->max_ports), rule->table_id);
3629 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
3630 rule->ofpacts, rule->ofpacts_len);
3632 op = ofoperation_create(group, rule, type, 0);
3634 if (fm->modify_cookie && fm->new_cookie != htonll(UINT64_MAX)) {
3635 ofproto_rule_change_cookie(ofproto, rule, fm->new_cookie);
3637 if (type == OFOPERATION_REPLACE) {
3638 ovs_mutex_lock(&rule->timeout_mutex);
3639 rule->idle_timeout = fm->idle_timeout;
3640 rule->hard_timeout = fm->hard_timeout;
3641 ovs_mutex_unlock(&rule->timeout_mutex);
3643 rule->send_flow_removed = (fm->flags
3644 & OFPUTIL_FF_SEND_FLOW_REM) != 0;
3646 if (fm->idle_timeout || fm->hard_timeout) {
3647 if (!rule->eviction_group) {
3648 eviction_group_add_rule(rule);
3651 eviction_group_remove_rule(rule);
3655 reset_counters = (fm->flags & OFPUTIL_FF_RESET_COUNTS) != 0;
3656 if (actions_changed || reset_counters) {
3657 op->ofpacts = rule->ofpacts;
3658 op->ofpacts_len = rule->ofpacts_len;
3659 op->meter_id = rule->meter_id;
3661 ovs_rwlock_wrlock(&rule->rwlock);
3662 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3663 rule->ofpacts_len = fm->ofpacts_len;
3664 ovs_rwlock_unlock(&rule->rwlock);
3666 rule->meter_id = find_meter(rule->ofpacts, rule->ofpacts_len);
3667 rule->ofproto->ofproto_class->rule_modify_actions(rule,
3670 ofoperation_complete(op, 0);
3673 ofopgroup_submit(group);
3679 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
3680 struct ofputil_flow_mod *fm, const struct ofp_header *request)
3682 if (fm->cookie_mask != htonll(0) || fm->new_cookie == htonll(UINT64_MAX)) {
3685 return add_flow(ofproto, ofconn, fm, request);
3688 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
3691 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3694 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3695 struct ofputil_flow_mod *fm,
3696 const struct ofp_header *request)
3701 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3702 fm->cookie, fm->cookie_mask,
3706 } else if (list_is_empty(&rules)) {
3707 return modify_flows_add(ofproto, ofconn, fm, request);
3709 return modify_flows__(ofproto, ofconn, fm, request, &rules);
3713 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3716 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3719 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3720 struct ofputil_flow_mod *fm,
3721 const struct ofp_header *request)
3726 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3727 fm->priority, fm->cookie, fm->cookie_mask,
3732 } else if (list_is_empty(&rules)) {
3733 return modify_flows_add(ofproto, ofconn, fm, request);
3735 return list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
3736 fm, request, &rules)
3741 /* OFPFC_DELETE implementation. */
3744 delete_flow__(struct rule *rule, struct ofopgroup *group,
3745 enum ofp_flow_removed_reason reason)
3747 struct ofproto *ofproto = rule->ofproto;
3749 ofproto_rule_send_removed(rule, reason);
3751 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
3752 oftable_remove_rule(rule);
3753 ofproto->ofproto_class->rule_delete(rule);
3756 /* Deletes the rules listed in 'rules'.
3758 * Returns 0 on success, otherwise an OpenFlow error code. */
3760 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3761 const struct ofp_header *request, struct list *rules,
3762 enum ofp_flow_removed_reason reason)
3764 struct rule *rule, *next;
3765 struct ofopgroup *group;
3767 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
3768 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
3769 ovs_rwlock_wrlock(&rule->rwlock);
3770 delete_flow__(rule, group, reason);
3772 ofopgroup_submit(group);
3777 /* Implements OFPFC_DELETE. */
3779 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3780 const struct ofputil_flow_mod *fm,
3781 const struct ofp_header *request)
3786 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3787 fm->cookie, fm->cookie_mask,
3788 fm->out_port, &rules);
3789 return (error ? error
3790 : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
3791 &rules, OFPRR_DELETE)
3795 /* Implements OFPFC_DELETE_STRICT. */
3797 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3798 const struct ofputil_flow_mod *fm,
3799 const struct ofp_header *request)
3804 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3805 fm->priority, fm->cookie, fm->cookie_mask,
3806 fm->out_port, &rules);
3807 return (error ? error
3808 : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
3815 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
3817 struct ofputil_flow_removed fr;
3819 if (ofproto_rule_is_hidden(rule) || !rule->send_flow_removed) {
3823 minimatch_expand(&rule->cr.match, &fr.match);
3824 fr.priority = rule->cr.priority;
3825 fr.cookie = rule->flow_cookie;
3827 fr.table_id = rule->table_id;
3828 calc_duration(rule->created, time_msec(),
3829 &fr.duration_sec, &fr.duration_nsec);
3830 ovs_mutex_lock(&rule->timeout_mutex);
3831 fr.idle_timeout = rule->idle_timeout;
3832 fr.hard_timeout = rule->hard_timeout;
3833 ovs_mutex_unlock(&rule->timeout_mutex);
3834 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
3837 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
3840 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3841 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3844 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
3847 * ofproto implementation ->run() functions should use this function to expire
3848 * OpenFlow flows. */
3850 ofproto_rule_expire(struct rule *rule, uint8_t reason)
3852 struct ofproto *ofproto = rule->ofproto;
3853 struct classifier *cls = &ofproto->tables[rule->table_id].cls;
3855 ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
3856 ofproto_rule_send_removed(rule, reason);
3858 ovs_rwlock_wrlock(&cls->rwlock);
3859 ofproto_delete_rule(ofproto, cls, rule);
3860 ovs_rwlock_unlock(&cls->rwlock);
3863 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
3864 * means "infinite". */
3866 reduce_timeout(uint16_t max, uint16_t *timeout)
3868 if (max && (!*timeout || *timeout > max)) {
3873 /* If 'idle_timeout' is nonzero, and 'rule' has no idle timeout or an idle
3874 * timeout greater than 'idle_timeout', lowers 'rule''s idle timeout to
3875 * 'idle_timeout' seconds. Similarly for 'hard_timeout'.
3877 * Suitable for implementing OFPACT_FIN_TIMEOUT. */
3879 ofproto_rule_reduce_timeouts(struct rule *rule,
3880 uint16_t idle_timeout, uint16_t hard_timeout)
3881 OVS_EXCLUDED(rule->ofproto->expirable_mutex, rule->timeout_mutex)
3883 if (!idle_timeout && !hard_timeout) {
3887 ovs_mutex_lock(&rule->ofproto->expirable_mutex);
3888 if (list_is_empty(&rule->expirable)) {
3889 list_insert(&rule->ofproto->expirable, &rule->expirable);
3891 ovs_mutex_unlock(&rule->ofproto->expirable_mutex);
3893 ovs_mutex_lock(&rule->timeout_mutex);
3894 reduce_timeout(idle_timeout, &rule->idle_timeout);
3895 reduce_timeout(hard_timeout, &rule->hard_timeout);
3896 ovs_mutex_unlock(&rule->timeout_mutex);
3900 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3902 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3903 struct ofputil_flow_mod fm;
3904 uint64_t ofpacts_stub[1024 / 8];
3905 struct ofpbuf ofpacts;
3909 error = reject_slave_controller(ofconn);
3914 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3915 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
3918 error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
3921 goto exit_free_ofpacts;
3924 /* Record the operation for logging a summary report. */
3925 switch (fm.command) {
3931 case OFPFC_MODIFY_STRICT:
3932 ofproto->n_modify++;
3936 case OFPFC_DELETE_STRICT:
3937 ofproto->n_delete++;
3942 if (ofproto->next_op_report == LLONG_MAX) {
3943 ofproto->first_op = now;
3944 ofproto->next_op_report = MAX(now + 10 * 1000,
3945 ofproto->op_backoff);
3946 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
3948 ofproto->last_op = now;
3951 ofpbuf_uninit(&ofpacts);
3957 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
3958 struct ofputil_flow_mod *fm, const struct ofp_header *oh)
3960 if (ofproto->n_pending >= 50) {
3961 ovs_assert(!list_is_empty(&ofproto->pending));
3962 return OFPROTO_POSTPONE;
3965 switch (fm->command) {
3967 return add_flow(ofproto, ofconn, fm, oh);
3970 return modify_flows_loose(ofproto, ofconn, fm, oh);
3972 case OFPFC_MODIFY_STRICT:
3973 return modify_flow_strict(ofproto, ofconn, fm, oh);
3976 return delete_flows_loose(ofproto, ofconn, fm, oh);
3978 case OFPFC_DELETE_STRICT:
3979 return delete_flow_strict(ofproto, ofconn, fm, oh);
3982 if (fm->command > 0xff) {
3983 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
3984 "flow_mod_table_id extension is not enabled",
3987 return OFPERR_OFPFMFC_BAD_COMMAND;
3992 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3994 struct ofputil_role_request request;
3995 struct ofputil_role_request reply;
3999 error = ofputil_decode_role_message(oh, &request);
4004 if (request.role != OFPCR12_ROLE_NOCHANGE) {
4005 if (ofconn_get_role(ofconn) != request.role
4006 && ofconn_has_pending_opgroups(ofconn)) {
4007 return OFPROTO_POSTPONE;
4010 if (request.have_generation_id
4011 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
4012 return OFPERR_OFPRRFC_STALE;
4015 ofconn_set_role(ofconn, request.role);
4018 reply.role = ofconn_get_role(ofconn);
4019 reply.have_generation_id = ofconn_get_master_election_id(
4020 ofconn, &reply.generation_id);
4021 buf = ofputil_encode_role_reply(oh, &reply);
4022 ofconn_send_reply(ofconn, buf);
4028 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
4029 const struct ofp_header *oh)
4031 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
4032 enum ofputil_protocol cur, next;
4034 cur = ofconn_get_protocol(ofconn);
4035 next = ofputil_protocol_set_tid(cur, msg->set != 0);
4036 ofconn_set_protocol(ofconn, next);
4042 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4044 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
4045 enum ofputil_protocol cur, next;
4046 enum ofputil_protocol next_base;
4048 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
4050 return OFPERR_OFPBRC_EPERM;
4053 cur = ofconn_get_protocol(ofconn);
4054 next = ofputil_protocol_set_base(cur, next_base);
4055 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
4056 /* Avoid sending async messages in surprising protocol. */
4057 return OFPROTO_POSTPONE;
4060 ofconn_set_protocol(ofconn, next);
4065 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
4066 const struct ofp_header *oh)
4068 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
4071 format = ntohl(msg->format);
4072 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
4073 return OFPERR_OFPBRC_EPERM;
4076 if (format != ofconn_get_packet_in_format(ofconn)
4077 && ofconn_has_pending_opgroups(ofconn)) {
4078 /* Avoid sending async message in surprsing packet in format. */
4079 return OFPROTO_POSTPONE;
4082 ofconn_set_packet_in_format(ofconn, format);
4087 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
4089 const struct nx_async_config *msg = ofpmsg_body(oh);
4090 uint32_t master[OAM_N_TYPES];
4091 uint32_t slave[OAM_N_TYPES];
4093 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
4094 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
4095 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
4097 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
4098 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
4099 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
4101 ofconn_set_async_config(ofconn, master, slave);
4102 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
4103 !ofconn_get_miss_send_len(ofconn)) {
4104 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
4111 handle_nxt_set_controller_id(struct ofconn *ofconn,
4112 const struct ofp_header *oh)
4114 const struct nx_controller_id *nci = ofpmsg_body(oh);
4116 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
4117 return OFPERR_NXBRC_MUST_BE_ZERO;
4120 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
4125 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4129 if (ofconn_has_pending_opgroups(ofconn)) {
4130 return OFPROTO_POSTPONE;
4133 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
4134 ? OFPRAW_OFPT10_BARRIER_REPLY
4135 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
4136 ofconn_send_reply(ofconn, buf);
4141 ofproto_compose_flow_refresh_update(const struct rule *rule,
4142 enum nx_flow_monitor_flags flags,
4145 struct ofoperation *op = rule->pending;
4146 struct ofputil_flow_update fu;
4149 if (op && op->type == OFOPERATION_ADD) {
4150 /* We'll report the final flow when the operation completes. Reporting
4151 * it now would cause a duplicate report later. */
4155 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
4156 ? NXFME_ADDED : NXFME_MODIFIED);
4158 ovs_mutex_lock(&rule->timeout_mutex);
4159 fu.idle_timeout = rule->idle_timeout;
4160 fu.hard_timeout = rule->hard_timeout;
4161 ovs_mutex_unlock(&rule->timeout_mutex);
4162 fu.table_id = rule->table_id;
4163 fu.cookie = rule->flow_cookie;
4164 minimatch_expand(&rule->cr.match, &match);
4166 fu.priority = rule->cr.priority;
4167 if (!(flags & NXFMF_ACTIONS)) {
4171 fu.ofpacts = rule->ofpacts;
4172 fu.ofpacts_len = rule->ofpacts_len;
4174 /* An operation is in progress. Use the previous version of the flow's
4175 * actions, so that when the operation commits we report the change. */
4177 case OFOPERATION_ADD:
4180 case OFOPERATION_MODIFY:
4181 case OFOPERATION_REPLACE:
4183 fu.ofpacts = op->ofpacts;
4184 fu.ofpacts_len = op->ofpacts_len;
4186 fu.ofpacts = rule->ofpacts;
4187 fu.ofpacts_len = rule->ofpacts_len;
4191 case OFOPERATION_DELETE:
4192 fu.ofpacts = rule->ofpacts;
4193 fu.ofpacts_len = rule->ofpacts_len;
4201 if (list_is_empty(msgs)) {
4202 ofputil_start_flow_update(msgs);
4204 ofputil_append_flow_update(&fu, msgs);
4208 ofmonitor_compose_refresh_updates(struct list *rules, struct list *msgs)
4212 LIST_FOR_EACH (rule, ofproto_node, rules) {
4213 enum nx_flow_monitor_flags flags = rule->monitor_flags;
4214 rule->monitor_flags = 0;
4216 ofproto_compose_flow_refresh_update(rule, flags, msgs);
4221 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
4222 struct rule *rule, uint64_t seqno,
4225 enum nx_flow_monitor_flags update;
4227 if (ofproto_rule_is_hidden(rule)) {
4232 ? ofoperation_has_out_port(rule->pending, m->out_port)
4233 : ofproto_rule_has_out_port(rule, m->out_port))) {
4238 if (rule->add_seqno > seqno) {
4239 update = NXFMF_ADD | NXFMF_MODIFY;
4240 } else if (rule->modify_seqno > seqno) {
4241 update = NXFMF_MODIFY;
4246 if (!(m->flags & update)) {
4250 update = NXFMF_INITIAL;
4253 if (!rule->monitor_flags) {
4254 list_push_back(rules, &rule->ofproto_node);
4256 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
4260 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
4264 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
4265 const struct ofoperation *op;
4266 const struct oftable *table;
4267 struct cls_rule target;
4269 cls_rule_init_from_minimatch(&target, &m->match, 0);
4270 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
4271 struct cls_cursor cursor;
4274 ovs_rwlock_rdlock(&table->cls.rwlock);
4275 cls_cursor_init(&cursor, &table->cls, &target);
4276 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4277 ovs_assert(!rule->pending); /* XXX */
4278 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4280 ovs_rwlock_unlock(&table->cls.rwlock);
4283 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
4284 struct rule *rule = op->rule;
4286 if (((m->table_id == 0xff
4287 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
4288 : m->table_id == rule->table_id))
4289 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
4290 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4293 cls_rule_destroy(&target);
4297 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
4300 if (m->flags & NXFMF_INITIAL) {
4301 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
4306 ofmonitor_collect_resume_rules(struct ofmonitor *m,
4307 uint64_t seqno, struct list *rules)
4309 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
4313 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
4315 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4316 struct ofmonitor **monitors;
4317 size_t n_monitors, allocated_monitors;
4318 struct list replies;
4325 ofpbuf_use_const(&b, oh, ntohs(oh->length));
4327 n_monitors = allocated_monitors = 0;
4329 struct ofputil_flow_monitor_request request;
4330 struct ofmonitor *m;
4333 retval = ofputil_decode_flow_monitor_request(&request, &b);
4334 if (retval == EOF) {
4336 } else if (retval) {
4341 if (request.table_id != 0xff
4342 && request.table_id >= ofproto->n_tables) {
4343 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4347 error = ofmonitor_create(&request, ofconn, &m);
4352 if (n_monitors >= allocated_monitors) {
4353 monitors = x2nrealloc(monitors, &allocated_monitors,
4356 monitors[n_monitors++] = m;
4360 for (i = 0; i < n_monitors; i++) {
4361 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
4364 ofpmp_init(&replies, oh);
4365 ofmonitor_compose_refresh_updates(&rules, &replies);
4366 ofconn_send_replies(ofconn, &replies);
4373 for (i = 0; i < n_monitors; i++) {
4374 ofmonitor_destroy(monitors[i]);
4381 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
4383 struct ofmonitor *m;
4386 id = ofputil_decode_flow_monitor_cancel(oh);
4387 m = ofmonitor_lookup(ofconn, id);
4389 return OFPERR_NXBRC_FM_BAD_ID;
4392 ofmonitor_destroy(m);
4396 /* Meters implementation.
4398 * Meter table entry, indexed by the OpenFlow meter_id.
4399 * These are always dynamically allocated to allocate enough space for
4401 * 'created' is used to compute the duration for meter stats.
4402 * 'list rules' is needed so that we can delete the dependent rules when the
4403 * meter table entry is deleted.
4404 * 'provider_meter_id' is for the provider's private use.
4407 long long int created; /* Time created. */
4408 struct list rules; /* List of "struct rule_dpif"s. */
4409 ofproto_meter_id provider_meter_id;
4410 uint16_t flags; /* Meter flags. */
4411 uint16_t n_bands; /* Number of meter bands. */
4412 struct ofputil_meter_band *bands;
4416 * This is used in instruction validation at flow set-up time,
4417 * as flows may not use non-existing meters.
4418 * This is also used by ofproto-providers to translate OpenFlow meter_ids
4419 * in METER instructions to the corresponding provider meter IDs.
4420 * Return value of UINT32_MAX signifies an invalid meter.
4423 ofproto_get_provider_meter_id(const struct ofproto * ofproto,
4424 uint32_t of_meter_id)
4426 if (of_meter_id && of_meter_id <= ofproto->meter_features.max_meters) {
4427 const struct meter *meter = ofproto->meters[of_meter_id];
4429 return meter->provider_meter_id.uint32;
4436 meter_update(struct meter *meter, const struct ofputil_meter_config *config)
4440 meter->flags = config->flags;
4441 meter->n_bands = config->n_bands;
4442 meter->bands = xmemdup(config->bands,
4443 config->n_bands * sizeof *meter->bands);
4446 static struct meter *
4447 meter_create(const struct ofputil_meter_config *config,
4448 ofproto_meter_id provider_meter_id)
4450 struct meter *meter;
4452 meter = xzalloc(sizeof *meter);
4453 meter->provider_meter_id = provider_meter_id;
4454 meter->created = time_msec();
4455 list_init(&meter->rules);
4457 meter_update(meter, config);
4463 meter_delete(struct ofproto *ofproto, uint32_t first, uint32_t last)
4466 for (mid = first; mid <= last; ++mid) {
4467 struct meter *meter = ofproto->meters[mid];
4469 ofproto->meters[mid] = NULL;
4470 ofproto->ofproto_class->meter_del(ofproto,
4471 meter->provider_meter_id);
4479 handle_add_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
4481 ofproto_meter_id provider_meter_id = { UINT32_MAX };
4482 struct meter **meterp = &ofproto->meters[mm->meter.meter_id];
4486 return OFPERR_OFPMMFC_METER_EXISTS;
4489 error = ofproto->ofproto_class->meter_set(ofproto, &provider_meter_id,
4492 ovs_assert(provider_meter_id.uint32 != UINT32_MAX);
4493 *meterp = meter_create(&mm->meter, provider_meter_id);
4499 handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
4501 struct meter *meter = ofproto->meters[mm->meter.meter_id];
4505 return OFPERR_OFPMMFC_UNKNOWN_METER;
4508 error = ofproto->ofproto_class->meter_set(ofproto,
4509 &meter->provider_meter_id,
4511 ovs_assert(meter->provider_meter_id.uint32 != UINT32_MAX);
4513 meter_update(meter, &mm->meter);
4519 handle_delete_meter(struct ofconn *ofconn, const struct ofp_header *oh,
4520 struct ofputil_meter_mod *mm)
4522 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4523 uint32_t meter_id = mm->meter.meter_id;
4524 uint32_t first, last;
4527 if (meter_id == OFPM13_ALL) {
4529 last = ofproto->meter_features.max_meters;
4531 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
4534 first = last = meter_id;
4537 /* First delete the rules that use this meter. If any of those rules are
4538 * currently being modified, postpone the whole operation until later. */
4540 for (meter_id = first; meter_id <= last; ++meter_id) {
4541 struct meter *meter = ofproto->meters[meter_id];
4542 if (meter && !list_is_empty(&meter->rules)) {
4545 LIST_FOR_EACH (rule, meter_list_node, &meter->rules) {
4546 if (rule->pending) {
4547 return OFPROTO_POSTPONE;
4549 list_push_back(&rules, &rule->ofproto_node);
4553 if (!list_is_empty(&rules)) {
4554 delete_flows__(ofproto, ofconn, oh, &rules, OFPRR_METER_DELETE);
4557 /* Delete the meters. */
4558 meter_delete(ofproto, first, last);
4564 handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4566 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4567 struct ofputil_meter_mod mm;
4568 uint64_t bands_stub[256 / 8];
4569 struct ofpbuf bands;
4573 error = reject_slave_controller(ofconn);
4578 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
4580 error = ofputil_decode_meter_mod(oh, &mm, &bands);
4582 goto exit_free_bands;
4585 meter_id = mm.meter.meter_id;
4587 if (mm.command != OFPMC13_DELETE) {
4588 /* Fails also when meters are not implemented by the provider. */
4589 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
4590 error = OFPERR_OFPMMFC_INVALID_METER;
4591 goto exit_free_bands;
4593 if (mm.meter.n_bands > ofproto->meter_features.max_bands) {
4594 error = OFPERR_OFPMMFC_OUT_OF_BANDS;
4595 goto exit_free_bands;
4599 switch (mm.command) {
4601 error = handle_add_meter(ofproto, &mm);
4604 case OFPMC13_MODIFY:
4605 error = handle_modify_meter(ofproto, &mm);
4608 case OFPMC13_DELETE:
4609 error = handle_delete_meter(ofconn, oh, &mm);
4613 error = OFPERR_OFPMMFC_BAD_COMMAND;
4618 ofpbuf_uninit(&bands);
4623 handle_meter_features_request(struct ofconn *ofconn,
4624 const struct ofp_header *request)
4626 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4627 struct ofputil_meter_features features;
4630 if (ofproto->ofproto_class->meter_get_features) {
4631 ofproto->ofproto_class->meter_get_features(ofproto, &features);
4633 memset(&features, 0, sizeof features);
4635 b = ofputil_encode_meter_features_reply(&features, request);
4637 ofconn_send_reply(ofconn, b);
4642 handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request,
4645 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4646 struct list replies;
4647 uint64_t bands_stub[256 / 8];
4648 struct ofpbuf bands;
4649 uint32_t meter_id, first, last;
4651 ofputil_decode_meter_request(request, &meter_id);
4653 if (meter_id == OFPM13_ALL) {
4655 last = ofproto->meter_features.max_meters;
4657 if (!meter_id || meter_id > ofproto->meter_features.max_meters ||
4658 !ofproto->meters[meter_id]) {
4659 return OFPERR_OFPMMFC_UNKNOWN_METER;
4661 first = last = meter_id;
4664 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
4665 ofpmp_init(&replies, request);
4667 for (meter_id = first; meter_id <= last; ++meter_id) {
4668 struct meter *meter = ofproto->meters[meter_id];
4670 continue; /* Skip non-existing meters. */
4672 if (type == OFPTYPE_METER_STATS_REQUEST) {
4673 struct ofputil_meter_stats stats;
4675 stats.meter_id = meter_id;
4677 /* Provider sets the packet and byte counts, we do the rest. */
4678 stats.flow_count = list_size(&meter->rules);
4679 calc_duration(meter->created, time_msec(),
4680 &stats.duration_sec, &stats.duration_nsec);
4681 stats.n_bands = meter->n_bands;
4682 ofpbuf_clear(&bands);
4684 = ofpbuf_put_uninit(&bands,
4685 meter->n_bands * sizeof *stats.bands);
4687 if (!ofproto->ofproto_class->meter_get(ofproto,
4688 meter->provider_meter_id,
4690 ofputil_append_meter_stats(&replies, &stats);
4692 } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */
4693 struct ofputil_meter_config config;
4695 config.meter_id = meter_id;
4696 config.flags = meter->flags;
4697 config.n_bands = meter->n_bands;
4698 config.bands = meter->bands;
4699 ofputil_append_meter_config(&replies, &config);
4703 ofconn_send_replies(ofconn, &replies);
4704 ofpbuf_uninit(&bands);
4709 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
4711 const struct ofp_header *oh = msg->data;
4715 error = ofptype_decode(&type, oh);
4721 /* OpenFlow requests. */
4722 case OFPTYPE_ECHO_REQUEST:
4723 return handle_echo_request(ofconn, oh);
4725 case OFPTYPE_FEATURES_REQUEST:
4726 return handle_features_request(ofconn, oh);
4728 case OFPTYPE_GET_CONFIG_REQUEST:
4729 return handle_get_config_request(ofconn, oh);
4731 case OFPTYPE_SET_CONFIG:
4732 return handle_set_config(ofconn, oh);
4734 case OFPTYPE_PACKET_OUT:
4735 return handle_packet_out(ofconn, oh);
4737 case OFPTYPE_PORT_MOD:
4738 return handle_port_mod(ofconn, oh);
4740 case OFPTYPE_FLOW_MOD:
4741 return handle_flow_mod(ofconn, oh);
4743 case OFPTYPE_METER_MOD:
4744 return handle_meter_mod(ofconn, oh);
4746 case OFPTYPE_BARRIER_REQUEST:
4747 return handle_barrier_request(ofconn, oh);
4749 case OFPTYPE_ROLE_REQUEST:
4750 return handle_role_request(ofconn, oh);
4752 /* OpenFlow replies. */
4753 case OFPTYPE_ECHO_REPLY:
4756 /* Nicira extension requests. */
4757 case OFPTYPE_FLOW_MOD_TABLE_ID:
4758 return handle_nxt_flow_mod_table_id(ofconn, oh);
4760 case OFPTYPE_SET_FLOW_FORMAT:
4761 return handle_nxt_set_flow_format(ofconn, oh);
4763 case OFPTYPE_SET_PACKET_IN_FORMAT:
4764 return handle_nxt_set_packet_in_format(ofconn, oh);
4766 case OFPTYPE_SET_CONTROLLER_ID:
4767 return handle_nxt_set_controller_id(ofconn, oh);
4769 case OFPTYPE_FLOW_AGE:
4770 /* Nothing to do. */
4773 case OFPTYPE_FLOW_MONITOR_CANCEL:
4774 return handle_flow_monitor_cancel(ofconn, oh);
4776 case OFPTYPE_SET_ASYNC_CONFIG:
4777 return handle_nxt_set_async_config(ofconn, oh);
4779 /* Statistics requests. */
4780 case OFPTYPE_DESC_STATS_REQUEST:
4781 return handle_desc_stats_request(ofconn, oh);
4783 case OFPTYPE_FLOW_STATS_REQUEST:
4784 return handle_flow_stats_request(ofconn, oh);
4786 case OFPTYPE_AGGREGATE_STATS_REQUEST:
4787 return handle_aggregate_stats_request(ofconn, oh);
4789 case OFPTYPE_TABLE_STATS_REQUEST:
4790 return handle_table_stats_request(ofconn, oh);
4792 case OFPTYPE_PORT_STATS_REQUEST:
4793 return handle_port_stats_request(ofconn, oh);
4795 case OFPTYPE_QUEUE_STATS_REQUEST:
4796 return handle_queue_stats_request(ofconn, oh);
4798 case OFPTYPE_PORT_DESC_STATS_REQUEST:
4799 return handle_port_desc_stats_request(ofconn, oh);
4801 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
4802 return handle_flow_monitor_request(ofconn, oh);
4804 case OFPTYPE_METER_STATS_REQUEST:
4805 case OFPTYPE_METER_CONFIG_STATS_REQUEST:
4806 return handle_meter_request(ofconn, oh, type);
4808 case OFPTYPE_METER_FEATURES_STATS_REQUEST:
4809 return handle_meter_features_request(ofconn, oh);
4811 /* FIXME: Change the following once they are implemented: */
4812 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
4813 case OFPTYPE_GET_ASYNC_REQUEST:
4814 case OFPTYPE_GROUP_STATS_REQUEST:
4815 case OFPTYPE_GROUP_DESC_STATS_REQUEST:
4816 case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
4817 case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
4818 return OFPERR_OFPBRC_BAD_TYPE;
4822 case OFPTYPE_FEATURES_REPLY:
4823 case OFPTYPE_GET_CONFIG_REPLY:
4824 case OFPTYPE_PACKET_IN:
4825 case OFPTYPE_FLOW_REMOVED:
4826 case OFPTYPE_PORT_STATUS:
4827 case OFPTYPE_BARRIER_REPLY:
4828 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
4829 case OFPTYPE_DESC_STATS_REPLY:
4830 case OFPTYPE_FLOW_STATS_REPLY:
4831 case OFPTYPE_QUEUE_STATS_REPLY:
4832 case OFPTYPE_PORT_STATS_REPLY:
4833 case OFPTYPE_TABLE_STATS_REPLY:
4834 case OFPTYPE_AGGREGATE_STATS_REPLY:
4835 case OFPTYPE_PORT_DESC_STATS_REPLY:
4836 case OFPTYPE_ROLE_REPLY:
4837 case OFPTYPE_FLOW_MONITOR_PAUSED:
4838 case OFPTYPE_FLOW_MONITOR_RESUMED:
4839 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
4840 case OFPTYPE_GET_ASYNC_REPLY:
4841 case OFPTYPE_GROUP_STATS_REPLY:
4842 case OFPTYPE_GROUP_DESC_STATS_REPLY:
4843 case OFPTYPE_GROUP_FEATURES_STATS_REPLY:
4844 case OFPTYPE_METER_STATS_REPLY:
4845 case OFPTYPE_METER_CONFIG_STATS_REPLY:
4846 case OFPTYPE_METER_FEATURES_STATS_REPLY:
4847 case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
4849 return OFPERR_OFPBRC_BAD_TYPE;
4854 handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
4856 int error = handle_openflow__(ofconn, ofp_msg);
4857 if (error && error != OFPROTO_POSTPONE) {
4858 ofconn_send_error(ofconn, ofp_msg->data, error);
4860 COVERAGE_INC(ofproto_recv_openflow);
4861 return error != OFPROTO_POSTPONE;
4864 /* Asynchronous operations. */
4866 /* Creates and returns a new ofopgroup that is not associated with any
4867 * OpenFlow connection.
4869 * The caller should add operations to the returned group with
4870 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4871 static struct ofopgroup *
4872 ofopgroup_create_unattached(struct ofproto *ofproto)
4874 struct ofopgroup *group = xzalloc(sizeof *group);
4875 group->ofproto = ofproto;
4876 list_init(&group->ofproto_node);
4877 list_init(&group->ops);
4878 list_init(&group->ofconn_node);
4882 /* Creates and returns a new ofopgroup for 'ofproto'.
4884 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
4885 * connection. The 'request' and 'buffer_id' arguments are ignored.
4887 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
4888 * If the ofopgroup eventually fails, then the error reply will include
4889 * 'request'. If the ofopgroup eventually succeeds, then the packet with
4890 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
4892 * The caller should add operations to the returned group with
4893 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4894 static struct ofopgroup *
4895 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
4896 const struct ofp_header *request, uint32_t buffer_id)
4898 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
4900 size_t request_len = ntohs(request->length);
4902 ovs_assert(ofconn_get_ofproto(ofconn) == ofproto);
4904 ofconn_add_opgroup(ofconn, &group->ofconn_node);
4905 group->ofconn = ofconn;
4906 group->request = xmemdup(request, MIN(request_len, 64));
4907 group->buffer_id = buffer_id;
4912 /* Submits 'group' for processing.
4914 * If 'group' contains no operations (e.g. none were ever added, or all of the
4915 * ones that were added completed synchronously), then it is destroyed
4916 * immediately. Otherwise it is added to the ofproto's list of pending
4919 ofopgroup_submit(struct ofopgroup *group)
4921 if (!group->n_running) {
4922 ofopgroup_complete(group);
4924 list_push_back(&group->ofproto->pending, &group->ofproto_node);
4925 group->ofproto->n_pending++;
4930 ofopgroup_complete(struct ofopgroup *group)
4932 struct ofproto *ofproto = group->ofproto;
4934 struct ofconn *abbrev_ofconn;
4935 ovs_be32 abbrev_xid;
4937 struct ofoperation *op, *next_op;
4940 ovs_assert(!group->n_running);
4943 LIST_FOR_EACH (op, group_node, &group->ops) {
4950 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
4951 LIST_FOR_EACH (op, group_node, &group->ops) {
4952 if (op->type != OFOPERATION_DELETE) {
4953 struct ofpbuf *packet;
4956 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
4960 error = rule_execute(op->rule, in_port, packet);
4967 if (!error && !list_is_empty(&group->ofconn_node)) {
4968 abbrev_ofconn = group->ofconn;
4969 abbrev_xid = group->request->xid;
4971 abbrev_ofconn = NULL;
4972 abbrev_xid = htonl(0);
4974 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
4975 struct rule *rule = op->rule;
4977 /* We generally want to report the change to active OpenFlow flow
4978 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
4980 - The operation failed.
4982 - The affected rule is not visible to controllers.
4984 - The operation's only effect was to update rule->modified. */
4986 || ofproto_rule_is_hidden(rule)
4987 || (op->type == OFOPERATION_MODIFY
4989 && rule->flow_cookie == op->flow_cookie))) {
4990 /* Check that we can just cast from ofoperation_type to
4991 * nx_flow_update_event. */
4992 enum nx_flow_update_event event_type;
4995 case OFOPERATION_ADD:
4996 case OFOPERATION_REPLACE:
4997 event_type = NXFME_ADDED;
5000 case OFOPERATION_DELETE:
5001 event_type = NXFME_DELETED;
5004 case OFOPERATION_MODIFY:
5005 event_type = NXFME_MODIFIED;
5012 ofmonitor_report(ofproto->connmgr, rule, event_type,
5013 op->reason, abbrev_ofconn, abbrev_xid);
5016 rule->pending = NULL;
5019 case OFOPERATION_ADD:
5023 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
5024 if (vid_mask == VLAN_VID_MASK) {
5025 if (ofproto->vlan_bitmap) {
5026 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
5027 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
5028 bitmap_set1(ofproto->vlan_bitmap, vid);
5029 ofproto->vlans_changed = true;
5032 ofproto->vlans_changed = true;
5036 ovs_rwlock_wrlock(&rule->rwlock);
5037 oftable_remove_rule(rule);
5038 ofproto_rule_destroy(rule);
5042 case OFOPERATION_DELETE:
5043 ovs_assert(!op->error);
5044 ofproto_rule_destroy(rule);
5048 case OFOPERATION_MODIFY:
5049 case OFOPERATION_REPLACE:
5051 long long int now = time_msec();
5053 rule->modified = now;
5054 if (op->type == OFOPERATION_REPLACE) {
5055 rule->created = rule->used = now;
5058 ofproto_rule_change_cookie(ofproto, rule, op->flow_cookie);
5059 ovs_mutex_lock(&rule->timeout_mutex);
5060 rule->idle_timeout = op->idle_timeout;
5061 rule->hard_timeout = op->hard_timeout;
5062 ovs_mutex_unlock(&rule->timeout_mutex);
5064 free(rule->ofpacts);
5066 ovs_rwlock_wrlock(&rule->rwlock);
5067 rule->ofpacts = op->ofpacts;
5068 rule->ofpacts_len = op->ofpacts_len;
5069 ovs_rwlock_unlock(&rule->rwlock);
5072 op->ofpacts_len = 0;
5074 rule->send_flow_removed = op->send_flow_removed;
5082 ofoperation_destroy(op);
5085 ofmonitor_flush(ofproto->connmgr);
5087 if (!list_is_empty(&group->ofproto_node)) {
5088 ovs_assert(ofproto->n_pending > 0);
5089 ofproto->n_pending--;
5090 list_remove(&group->ofproto_node);
5092 if (!list_is_empty(&group->ofconn_node)) {
5093 list_remove(&group->ofconn_node);
5095 ofconn_send_error(group->ofconn, group->request, error);
5097 connmgr_retry(ofproto->connmgr);
5099 free(group->request);
5103 /* Initiates a new operation on 'rule', of the specified 'type', within
5104 * 'group'. Prior to calling, 'rule' must not have any pending operation.
5106 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
5107 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
5109 * Returns the newly created ofoperation (which is also available as
5110 * rule->pending). */
5111 static struct ofoperation *
5112 ofoperation_create(struct ofopgroup *group, struct rule *rule,
5113 enum ofoperation_type type,
5114 enum ofp_flow_removed_reason reason)
5116 struct ofproto *ofproto = group->ofproto;
5117 struct ofoperation *op;
5119 ovs_assert(!rule->pending);
5121 op = rule->pending = xzalloc(sizeof *op);
5123 list_push_back(&group->ops, &op->group_node);
5126 op->reason = reason;
5127 op->flow_cookie = rule->flow_cookie;
5128 ovs_mutex_lock(&rule->timeout_mutex);
5129 op->idle_timeout = rule->idle_timeout;
5130 op->hard_timeout = rule->hard_timeout;
5131 ovs_mutex_unlock(&rule->timeout_mutex);
5132 op->send_flow_removed = rule->send_flow_removed;
5136 if (type == OFOPERATION_DELETE) {
5137 hmap_insert(&ofproto->deletions, &op->hmap_node,
5138 cls_rule_hash(&rule->cr, rule->table_id));
5145 ofoperation_destroy(struct ofoperation *op)
5147 struct ofopgroup *group = op->group;
5150 op->rule->pending = NULL;
5152 if (op->type == OFOPERATION_DELETE) {
5153 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
5155 list_remove(&op->group_node);
5160 /* Indicates that 'op' completed with status 'error', which is either 0 to
5161 * indicate success or an OpenFlow error code on failure.
5163 * If 'error' is 0, indicating success, the operation will be committed
5164 * permanently to the flow table.
5166 * If 'error' is nonzero, then generally the operation will be rolled back:
5168 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
5169 * restores the original rule. The caller must have uninitialized any
5170 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
5171 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
5172 * and 7 for the new rule, calling its ->rule_dealloc() function.
5174 * - If 'op' is a "modify flow" operation, ofproto restores the original
5177 * - 'op' must not be a "delete flow" operation. Removing a rule is not
5178 * allowed to fail. It must always succeed.
5180 * Please see the large comment in ofproto/ofproto-provider.h titled
5181 * "Asynchronous Operation Support" for more information. */
5183 ofoperation_complete(struct ofoperation *op, enum ofperr error)
5185 struct ofopgroup *group = op->group;
5187 ovs_assert(op->rule->pending == op);
5188 ovs_assert(group->n_running > 0);
5189 ovs_assert(!error || op->type != OFOPERATION_DELETE);
5192 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
5193 ofopgroup_complete(group);
5198 pick_datapath_id(const struct ofproto *ofproto)
5200 const struct ofport *port;
5202 port = ofproto_get_port(ofproto, OFPP_LOCAL);
5204 uint8_t ea[ETH_ADDR_LEN];
5207 error = netdev_get_etheraddr(port->netdev, ea);
5209 return eth_addr_to_uint64(ea);
5211 VLOG_WARN("%s: could not get MAC address for %s (%s)",
5212 ofproto->name, netdev_get_name(port->netdev),
5213 ovs_strerror(error));
5215 return ofproto->fallback_dpid;
5219 pick_fallback_dpid(void)
5221 uint8_t ea[ETH_ADDR_LEN];
5222 eth_addr_nicira_random(ea);
5223 return eth_addr_to_uint64(ea);
5226 /* Table overflow policy. */
5228 /* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep'
5229 * to NULL if the table is not configured to evict rules or if the table
5230 * contains no evictable rules. (Rules with a readlock on their evict rwlock,
5231 * or with no timeouts are not evictable.) */
5233 choose_rule_to_evict(struct oftable *table, struct rule **rulep)
5235 struct eviction_group *evg;
5238 if (!table->eviction_fields) {
5242 /* In the common case, the outer and inner loops here will each be entered
5245 * - The inner loop normally "return"s in its first iteration. If the
5246 * eviction group has any evictable rules, then it always returns in
5249 * - The outer loop only iterates more than once if the largest eviction
5250 * group has no evictable rules.
5252 * - The outer loop can exit only if table's 'max_flows' is all filled up
5253 * by unevictable rules. */
5254 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
5257 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
5258 if (!ovs_rwlock_trywrlock(&rule->rwlock)) {
5268 /* Searches 'ofproto' for tables that have more flows than their configured
5269 * maximum and that have flow eviction enabled, and evicts as many flows as
5270 * necessary and currently feasible from them.
5272 * This triggers only when an OpenFlow table has N flows in it and then the
5273 * client configures a maximum number of flows less than N. */
5275 ofproto_evict(struct ofproto *ofproto)
5277 struct ofopgroup *group;
5278 struct oftable *table;
5280 group = ofopgroup_create_unattached(ofproto);
5281 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
5282 while (table->eviction_fields) {
5286 ovs_rwlock_rdlock(&table->cls.rwlock);
5287 n_rules = classifier_count(&table->cls);
5288 ovs_rwlock_unlock(&table->cls.rwlock);
5290 if (n_rules <= table->max_flows) {
5294 if (!choose_rule_to_evict(table, &rule)) {
5298 if (rule->pending) {
5299 ovs_rwlock_unlock(&rule->rwlock);
5303 ofoperation_create(group, rule,
5304 OFOPERATION_DELETE, OFPRR_EVICTION);
5305 oftable_remove_rule(rule);
5306 ofproto->ofproto_class->rule_delete(rule);
5309 ofopgroup_submit(group);
5312 /* Eviction groups. */
5314 /* Returns the priority to use for an eviction_group that contains 'n_rules'
5315 * rules. The priority contains low-order random bits to ensure that eviction
5316 * groups with the same number of rules are prioritized randomly. */
5318 eviction_group_priority(size_t n_rules)
5320 uint16_t size = MIN(UINT16_MAX, n_rules);
5321 return (size << 16) | random_uint16();
5324 /* Updates 'evg', an eviction_group within 'table', following a change that
5325 * adds or removes rules in 'evg'. */
5327 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
5329 heap_change(&table->eviction_groups_by_size, &evg->size_node,
5330 eviction_group_priority(heap_count(&evg->rules)));
5333 /* Destroys 'evg', an eviction_group within 'table':
5335 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
5336 * rules themselves, just removes them from the eviction group.)
5338 * - Removes 'evg' from 'table'.
5342 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
5344 while (!heap_is_empty(&evg->rules)) {
5347 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
5348 rule->eviction_group = NULL;
5350 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
5351 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
5352 heap_destroy(&evg->rules);
5356 /* Removes 'rule' from its eviction group, if any. */
5358 eviction_group_remove_rule(struct rule *rule)
5360 if (rule->eviction_group) {
5361 struct oftable *table = &rule->ofproto->tables[rule->table_id];
5362 struct eviction_group *evg = rule->eviction_group;
5364 rule->eviction_group = NULL;
5365 heap_remove(&evg->rules, &rule->evg_node);
5366 if (heap_is_empty(&evg->rules)) {
5367 eviction_group_destroy(table, evg);
5369 eviction_group_resized(table, evg);
5374 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
5375 * returns the hash value. */
5377 eviction_group_hash_rule(struct rule *rule)
5379 struct oftable *table = &rule->ofproto->tables[rule->table_id];
5380 const struct mf_subfield *sf;
5384 hash = table->eviction_group_id_basis;
5385 miniflow_expand(&rule->cr.match.flow, &flow);
5386 for (sf = table->eviction_fields;
5387 sf < &table->eviction_fields[table->n_eviction_fields];
5390 if (mf_are_prereqs_ok(sf->field, &flow)) {
5391 union mf_value value;
5393 mf_get_value(sf->field, &flow, &value);
5395 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
5397 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
5398 unsigned int start = sf->ofs + sf->n_bits;
5399 bitwise_zero(&value, sf->field->n_bytes, start,
5400 sf->field->n_bytes * 8 - start);
5402 hash = hash_bytes(&value, sf->field->n_bytes, hash);
5404 hash = hash_int(hash, 0);
5411 /* Returns an eviction group within 'table' with the given 'id', creating one
5413 static struct eviction_group *
5414 eviction_group_find(struct oftable *table, uint32_t id)
5416 struct eviction_group *evg;
5418 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
5422 evg = xmalloc(sizeof *evg);
5423 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
5424 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
5425 eviction_group_priority(0));
5426 heap_init(&evg->rules);
5431 /* Returns an eviction priority for 'rule'. The return value should be
5432 * interpreted so that higher priorities make a rule more attractive candidates
5435 rule_eviction_priority(struct rule *rule)
5437 long long int hard_expiration;
5438 long long int idle_expiration;
5439 long long int expiration;
5440 uint32_t expiration_offset;
5442 /* Calculate time of expiration. */
5443 ovs_mutex_lock(&rule->timeout_mutex);
5444 hard_expiration = (rule->hard_timeout
5445 ? rule->modified + rule->hard_timeout * 1000
5447 idle_expiration = (rule->idle_timeout
5448 ? rule->used + rule->idle_timeout * 1000
5450 expiration = MIN(hard_expiration, idle_expiration);
5451 ovs_mutex_unlock(&rule->timeout_mutex);
5452 if (expiration == LLONG_MAX) {
5456 /* Calculate the time of expiration as a number of (approximate) seconds
5457 * after program startup.
5459 * This should work OK for program runs that last UINT32_MAX seconds or
5460 * less. Therefore, please restart OVS at least once every 136 years. */
5461 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
5463 /* Invert the expiration offset because we're using a max-heap. */
5464 return UINT32_MAX - expiration_offset;
5467 /* Adds 'rule' to an appropriate eviction group for its oftable's
5468 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
5469 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
5472 * The caller must ensure that 'rule' is not already in an eviction group. */
5474 eviction_group_add_rule(struct rule *rule)
5476 struct ofproto *ofproto = rule->ofproto;
5477 struct oftable *table = &ofproto->tables[rule->table_id];
5480 ovs_mutex_lock(&rule->timeout_mutex);
5481 has_timeout = rule->hard_timeout || rule->idle_timeout;
5482 ovs_mutex_unlock(&rule->timeout_mutex);
5484 if (table->eviction_fields && has_timeout) {
5485 struct eviction_group *evg;
5487 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
5489 rule->eviction_group = evg;
5490 heap_insert(&evg->rules, &rule->evg_node,
5491 rule_eviction_priority(rule));
5492 eviction_group_resized(table, evg);
5498 /* Initializes 'table'. */
5500 oftable_init(struct oftable *table)
5502 memset(table, 0, sizeof *table);
5503 classifier_init(&table->cls);
5504 table->max_flows = UINT_MAX;
5507 /* Destroys 'table', including its classifier and eviction groups.
5509 * The caller is responsible for freeing 'table' itself. */
5511 oftable_destroy(struct oftable *table)
5513 ovs_rwlock_rdlock(&table->cls.rwlock);
5514 ovs_assert(classifier_is_empty(&table->cls));
5515 ovs_rwlock_unlock(&table->cls.rwlock);
5516 oftable_disable_eviction(table);
5517 classifier_destroy(&table->cls);
5521 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
5522 * string, then 'table' will use its default name.
5524 * This only affects the name exposed for a table exposed through the OpenFlow
5525 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
5527 oftable_set_name(struct oftable *table, const char *name)
5529 if (name && name[0]) {
5530 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
5531 if (!table->name || strncmp(name, table->name, len)) {
5533 table->name = xmemdup0(name, len);
5541 /* oftables support a choice of two policies when adding a rule would cause the
5542 * number of flows in the table to exceed the configured maximum number: either
5543 * they can refuse to add the new flow or they can evict some existing flow.
5544 * This function configures the former policy on 'table'. */
5546 oftable_disable_eviction(struct oftable *table)
5548 if (table->eviction_fields) {
5549 struct eviction_group *evg, *next;
5551 HMAP_FOR_EACH_SAFE (evg, next, id_node,
5552 &table->eviction_groups_by_id) {
5553 eviction_group_destroy(table, evg);
5555 hmap_destroy(&table->eviction_groups_by_id);
5556 heap_destroy(&table->eviction_groups_by_size);
5558 free(table->eviction_fields);
5559 table->eviction_fields = NULL;
5560 table->n_eviction_fields = 0;
5564 /* oftables support a choice of two policies when adding a rule would cause the
5565 * number of flows in the table to exceed the configured maximum number: either
5566 * they can refuse to add the new flow or they can evict some existing flow.
5567 * This function configures the latter policy on 'table', with fairness based
5568 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
5569 * 'n_fields' as 0 disables fairness.) */
5571 oftable_enable_eviction(struct oftable *table,
5572 const struct mf_subfield *fields, size_t n_fields)
5574 struct cls_cursor cursor;
5577 if (table->eviction_fields
5578 && n_fields == table->n_eviction_fields
5580 || !memcmp(fields, table->eviction_fields,
5581 n_fields * sizeof *fields))) {
5586 oftable_disable_eviction(table);
5588 table->n_eviction_fields = n_fields;
5589 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
5591 table->eviction_group_id_basis = random_uint32();
5592 hmap_init(&table->eviction_groups_by_id);
5593 heap_init(&table->eviction_groups_by_size);
5595 ovs_rwlock_rdlock(&table->cls.rwlock);
5596 cls_cursor_init(&cursor, &table->cls, NULL);
5597 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
5598 eviction_group_add_rule(rule);
5600 ovs_rwlock_unlock(&table->cls.rwlock);
5603 /* Removes 'rule' from the oftable that contains it. */
5605 oftable_remove_rule__(struct ofproto *ofproto, struct classifier *cls,
5607 OVS_REQ_WRLOCK(cls->rwlock) OVS_RELEASES(rule->rwlock)
5609 classifier_remove(cls, &rule->cr);
5610 if (rule->meter_id) {
5611 list_remove(&rule->meter_list_node);
5613 cookies_remove(ofproto, rule);
5614 eviction_group_remove_rule(rule);
5615 ovs_mutex_lock(&ofproto->expirable_mutex);
5616 if (!list_is_empty(&rule->expirable)) {
5617 list_remove(&rule->expirable);
5619 ovs_mutex_unlock(&ofproto->expirable_mutex);
5620 if (!list_is_empty(&rule->meter_list_node)) {
5621 list_remove(&rule->meter_list_node);
5623 ovs_rwlock_unlock(&rule->rwlock);
5627 oftable_remove_rule(struct rule *rule)
5629 struct ofproto *ofproto = rule->ofproto;
5630 struct oftable *table = &ofproto->tables[rule->table_id];
5632 ovs_rwlock_wrlock(&table->cls.rwlock);
5633 oftable_remove_rule__(ofproto, &table->cls, rule);
5634 ovs_rwlock_unlock(&table->cls.rwlock);
5637 /* Inserts 'rule' into its oftable, which must not already contain any rule for
5638 * the same cls_rule. */
5640 oftable_insert_rule(struct rule *rule)
5642 struct ofproto *ofproto = rule->ofproto;
5643 struct oftable *table = &ofproto->tables[rule->table_id];
5646 ovs_mutex_lock(&rule->timeout_mutex);
5647 may_expire = rule->hard_timeout || rule->idle_timeout;
5648 ovs_mutex_unlock(&rule->timeout_mutex);
5651 ovs_mutex_lock(&ofproto->expirable_mutex);
5652 list_insert(&ofproto->expirable, &rule->expirable);
5653 ovs_mutex_unlock(&ofproto->expirable_mutex);
5655 cookies_insert(ofproto, rule);
5656 if (rule->meter_id) {
5657 struct meter *meter = ofproto->meters[rule->meter_id];
5658 list_insert(&meter->rules, &rule->meter_list_node);
5660 ovs_rwlock_wrlock(&table->cls.rwlock);
5661 classifier_insert(&table->cls, &rule->cr);
5662 ovs_rwlock_unlock(&table->cls.rwlock);
5663 eviction_group_add_rule(rule);
5666 /* unixctl commands. */
5669 ofproto_lookup(const char *name)
5671 struct ofproto *ofproto;
5673 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
5675 if (!strcmp(ofproto->name, name)) {
5683 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
5684 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5686 struct ofproto *ofproto;
5690 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
5691 ds_put_format(&results, "%s\n", ofproto->name);
5693 unixctl_command_reply(conn, ds_cstr(&results));
5694 ds_destroy(&results);
5698 ofproto_unixctl_init(void)
5700 static bool registered;
5706 unixctl_command_register("ofproto/list", "", 0, 0,
5707 ofproto_unixctl_list, NULL);
5710 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5712 * This is deprecated. It is only for compatibility with broken device drivers
5713 * in old versions of Linux that do not properly support VLANs when VLAN
5714 * devices are not used. When broken device drivers are no longer in
5715 * widespread use, we will delete these interfaces. */
5717 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
5718 * (exactly) by an OpenFlow rule in 'ofproto'. */
5720 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
5722 const struct oftable *oftable;
5724 free(ofproto->vlan_bitmap);
5725 ofproto->vlan_bitmap = bitmap_allocate(4096);
5726 ofproto->vlans_changed = false;
5728 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
5729 const struct cls_table *table;
5731 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
5732 if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
5733 const struct cls_rule *rule;
5735 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
5736 uint16_t vid = miniflow_get_vid(&rule->match.flow);
5737 bitmap_set1(vlan_bitmap, vid);
5738 bitmap_set1(ofproto->vlan_bitmap, vid);
5745 /* Returns true if new VLANs have come into use by the flow table since the
5746 * last call to ofproto_get_vlan_usage().
5748 * We don't track when old VLANs stop being used. */
5750 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
5752 return ofproto->vlans_changed;
5755 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
5756 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
5757 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
5758 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
5759 * then the VLAN device is un-enslaved. */
5761 ofproto_port_set_realdev(struct ofproto *ofproto, ofp_port_t vlandev_ofp_port,
5762 ofp_port_t realdev_ofp_port, int vid)
5764 struct ofport *ofport;
5767 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
5769 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
5771 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
5772 ofproto->name, vlandev_ofp_port);
5776 if (!ofproto->ofproto_class->set_realdev) {
5777 if (!vlandev_ofp_port) {
5780 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
5784 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
5786 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
5787 ofproto->name, vlandev_ofp_port,
5788 netdev_get_name(ofport->netdev), ovs_strerror(error));