2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 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"
28 #include "connectivity.h"
31 #include "dynamic-string.h"
34 #include "meta-flow.h"
37 #include "ofp-actions.h"
38 #include "ofp-errors.h"
40 #include "ofp-print.h"
43 #include "ofproto-provider.h"
44 #include "openflow/nicira-ext.h"
45 #include "openflow/openflow.h"
50 #include "poll-loop.h"
58 #include "unaligned.h"
63 VLOG_DEFINE_THIS_MODULE(ofproto);
65 COVERAGE_DEFINE(ofproto_flush);
66 COVERAGE_DEFINE(ofproto_packet_out);
67 COVERAGE_DEFINE(ofproto_queue_req);
68 COVERAGE_DEFINE(ofproto_recv_openflow);
69 COVERAGE_DEFINE(ofproto_reinit_ports);
70 COVERAGE_DEFINE(ofproto_update_port);
73 S_OPENFLOW, /* Processing OpenFlow commands. */
74 S_EVICT, /* Evicting flows from over-limit tables. */
75 S_FLUSH, /* Deleting all flow table rules. */
78 enum ofoperation_type {
85 /* A single OpenFlow request can execute any number of operations. The
86 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
87 * ofconn to which an error reply should be sent if necessary.
89 * ofproto initiates some operations internally. These operations are still
90 * assigned to groups but will not have an associated ofconn. */
92 struct ofproto *ofproto; /* Owning ofproto. */
93 struct list ofproto_node; /* In ofproto's "pending" list. */
94 struct list ops; /* List of "struct ofoperation"s. */
95 int n_running; /* Number of ops still pending. */
97 /* Data needed to send OpenFlow reply on failure or to send a buffered
100 * If list_is_empty(ofconn_node) then this ofopgroup never had an
101 * associated ofconn or its ofconn's connection dropped after it initiated
102 * the operation. In the latter case 'ofconn' is a wild pointer that
103 * refers to freed memory, so the 'ofconn' member must be used only if
104 * !list_is_empty(ofconn_node).
106 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
107 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
108 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
109 uint32_t buffer_id; /* Buffer id from original request. */
112 static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
113 static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
114 const struct ofp_header *,
116 static void ofopgroup_submit(struct ofopgroup *);
117 static void ofopgroup_complete(struct ofopgroup *);
119 /* A single flow table operation. */
121 struct ofopgroup *group; /* Owning group. */
122 struct list group_node; /* In ofopgroup's "ops" list. */
123 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
124 struct rule *rule; /* Rule being operated upon. */
125 enum ofoperation_type type; /* Type of operation. */
127 /* OFOPERATION_MODIFY, OFOPERATION_REPLACE: The old actions, if the actions
129 const struct rule_actions *actions;
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 enum ofputil_flow_mod_flags flags; /* Rule's old flags. */
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_REQUIRES(ofproto_mutex);
159 static void oftable_remove_rule__(struct ofproto *, struct rule *)
160 OVS_REQUIRES(ofproto_mutex);
161 static void oftable_insert_rule(struct rule *);
163 /* A set of rules within a single OpenFlow table (oftable) that have the same
164 * values for the oftable's eviction_fields. A rule to be evicted, when one is
165 * needed, is taken from the eviction group that contains the greatest number
168 * An oftable owns any number of eviction groups, each of which contains any
171 * Membership in an eviction group is imprecise, based on the hash of the
172 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
173 * That is, if two rules have different eviction_fields, but those
174 * eviction_fields hash to the same value, then they will belong to the same
175 * eviction_group anyway.
177 * (When eviction is not enabled on an oftable, we don't track any eviction
178 * groups, to save time and space.) */
179 struct eviction_group {
180 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
181 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
182 struct heap rules; /* Contains "struct rule"s. */
185 static bool choose_rule_to_evict(struct oftable *table, struct rule **rulep);
186 static void ofproto_evict(struct ofproto *) OVS_EXCLUDED(ofproto_mutex);
187 static uint32_t rule_eviction_priority(struct ofproto *ofproto, struct rule *);
188 static void eviction_group_add_rule(struct rule *);
189 static void eviction_group_remove_rule(struct rule *);
191 /* Criteria that flow_mod and other operations use for selecting rules on
192 * which to operate. */
193 struct rule_criteria {
194 /* An OpenFlow table or 255 for all tables. */
197 /* OpenFlow matching criteria. Interpreted different in "loose" way by
198 * collect_rules_loose() and "strict" way by collect_rules_strict(), as
199 * defined in the OpenFlow spec. */
202 /* Matching criteria for the OpenFlow cookie. Consider a bit B in a rule's
203 * cookie and the corresponding bits C in 'cookie' and M in 'cookie_mask'.
204 * The rule will not be selected if M is 1 and B != C. */
206 ovs_be64 cookie_mask;
208 /* Selection based on actions within a rule:
210 * If out_port != OFPP_ANY, selects only rules that output to out_port.
211 * If out_group != OFPG_ALL, select only rules that output to out_group. */
216 static void rule_criteria_init(struct rule_criteria *, uint8_t table_id,
217 const struct match *match,
218 unsigned int priority,
219 ovs_be64 cookie, ovs_be64 cookie_mask,
220 ofp_port_t out_port, uint32_t out_group);
221 static void rule_criteria_destroy(struct rule_criteria *);
223 /* A packet that needs to be passed to rule_execute().
225 * (We can't do this immediately from ofopgroup_complete() because that holds
226 * ofproto_mutex, which rule_execute() needs released.) */
227 struct rule_execute {
228 struct list list_node; /* In struct ofproto's "rule_executes" list. */
229 struct rule *rule; /* Owns a reference to the rule. */
231 struct ofpbuf *packet; /* Owns the packet. */
234 static void run_rule_executes(struct ofproto *) OVS_EXCLUDED(ofproto_mutex);
235 static void destroy_rule_executes(struct ofproto *);
238 static void ofport_destroy__(struct ofport *) OVS_EXCLUDED(ofproto_mutex);
239 static void ofport_destroy(struct ofport *);
241 static void update_port(struct ofproto *, const char *devname);
242 static int init_ports(struct ofproto *);
243 static void reinit_ports(struct ofproto *);
245 static long long int ofport_get_usage(const struct ofproto *,
246 ofp_port_t ofp_port);
247 static void ofport_set_usage(struct ofproto *, ofp_port_t ofp_port,
248 long long int last_used);
249 static void ofport_remove_usage(struct ofproto *, ofp_port_t ofp_port);
253 * Keeps track of the currently used and recently used ofport values and is
254 * used to prevent immediate recycling of ofport values. */
255 struct ofport_usage {
256 struct hmap_node hmap_node; /* In struct ofproto's "ofport_usage" hmap. */
257 ofp_port_t ofp_port; /* OpenFlow port number. */
258 long long int last_used; /* Last time the 'ofp_port' was used. LLONG_MAX
259 represents in-use ofports. */
263 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
264 static bool rule_is_modifiable(const struct rule *rule,
265 enum ofputil_flow_mod_flags flag);
268 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
269 struct ofputil_flow_mod *,
270 const struct ofp_header *);
271 static void do_add_flow(struct ofproto *, struct ofconn *,
272 const struct ofp_header *request, uint32_t buffer_id,
274 static enum ofperr modify_flows__(struct ofproto *, struct ofconn *,
275 struct ofputil_flow_mod *,
276 const struct ofp_header *,
277 const struct rule_collection *);
278 static void delete_flow__(struct rule *rule, struct ofopgroup *,
279 enum ofp_flow_removed_reason)
280 OVS_REQUIRES(ofproto_mutex);
281 static bool ofproto_group_exists__(const struct ofproto *ofproto,
283 OVS_REQ_RDLOCK(ofproto->groups_rwlock);
284 static bool ofproto_group_exists(const struct ofproto *ofproto,
286 OVS_EXCLUDED(ofproto->groups_rwlock);
287 static enum ofperr add_group(struct ofproto *, struct ofputil_group_mod *);
288 static bool handle_openflow(struct ofconn *, const struct ofpbuf *);
289 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
290 struct ofputil_flow_mod *,
291 const struct ofp_header *)
292 OVS_EXCLUDED(ofproto_mutex);
293 static void calc_duration(long long int start, long long int now,
294 uint32_t *sec, uint32_t *nsec);
297 static uint64_t pick_datapath_id(const struct ofproto *);
298 static uint64_t pick_fallback_dpid(void);
299 static void ofproto_destroy__(struct ofproto *);
300 static void update_mtu(struct ofproto *, struct ofport *);
301 static void meter_delete(struct ofproto *, uint32_t first, uint32_t last);
304 static void ofproto_unixctl_init(void);
306 /* All registered ofproto classes, in probe order. */
307 static const struct ofproto_class **ofproto_classes;
308 static size_t n_ofproto_classes;
309 static size_t allocated_ofproto_classes;
311 /* Global lock that protects all flow table operations. */
312 struct ovs_mutex ofproto_mutex = OVS_MUTEX_INITIALIZER;
314 unsigned ofproto_flow_limit = OFPROTO_FLOW_LIMIT_DEFAULT;
315 unsigned ofproto_max_idle = OFPROTO_MAX_IDLE_DEFAULT;
317 size_t n_handlers, n_revalidators;
319 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
320 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
322 /* Initial mappings of port to OpenFlow number mappings. */
323 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
325 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
327 /* The default value of true waits for flow restore. */
328 static bool flow_restore_wait = true;
330 /* Must be called to initialize the ofproto library.
332 * The caller may pass in 'iface_hints', which contains an shash of
333 * "iface_hint" elements indexed by the interface's name. The provider
334 * may use these hints to describe the startup configuration in order to
335 * reinitialize its state. The caller owns the provided data, so a
336 * provider will make copies of anything required. An ofproto provider
337 * will remove any existing state that is not described by the hint, and
338 * may choose to remove it all. */
340 ofproto_init(const struct shash *iface_hints)
342 struct shash_node *node;
345 ofproto_class_register(&ofproto_dpif_class);
347 /* Make a local copy, since we don't own 'iface_hints' elements. */
348 SHASH_FOR_EACH(node, iface_hints) {
349 const struct iface_hint *orig_hint = node->data;
350 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
351 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
353 new_hint->br_name = xstrdup(orig_hint->br_name);
354 new_hint->br_type = xstrdup(br_type);
355 new_hint->ofp_port = orig_hint->ofp_port;
357 shash_add(&init_ofp_ports, node->name, new_hint);
360 for (i = 0; i < n_ofproto_classes; i++) {
361 ofproto_classes[i]->init(&init_ofp_ports);
365 /* 'type' should be a normalized datapath type, as returned by
366 * ofproto_normalize_type(). Returns the corresponding ofproto_class
367 * structure, or a null pointer if there is none registered for 'type'. */
368 static const struct ofproto_class *
369 ofproto_class_find__(const char *type)
373 for (i = 0; i < n_ofproto_classes; i++) {
374 const struct ofproto_class *class = ofproto_classes[i];
379 class->enumerate_types(&types);
380 found = sset_contains(&types, type);
381 sset_destroy(&types);
387 VLOG_WARN("unknown datapath type %s", type);
391 /* Registers a new ofproto class. After successful registration, new ofprotos
392 * of that type can be created using ofproto_create(). */
394 ofproto_class_register(const struct ofproto_class *new_class)
398 for (i = 0; i < n_ofproto_classes; i++) {
399 if (ofproto_classes[i] == new_class) {
404 if (n_ofproto_classes >= allocated_ofproto_classes) {
405 ofproto_classes = x2nrealloc(ofproto_classes,
406 &allocated_ofproto_classes,
407 sizeof *ofproto_classes);
409 ofproto_classes[n_ofproto_classes++] = new_class;
413 /* Unregisters a datapath provider. 'type' must have been previously
414 * registered and not currently be in use by any ofprotos. After
415 * unregistration new datapaths of that type cannot be opened using
416 * ofproto_create(). */
418 ofproto_class_unregister(const struct ofproto_class *class)
422 for (i = 0; i < n_ofproto_classes; i++) {
423 if (ofproto_classes[i] == class) {
424 for (i++; i < n_ofproto_classes; i++) {
425 ofproto_classes[i - 1] = ofproto_classes[i];
431 VLOG_WARN("attempted to unregister an ofproto class that is not "
436 /* Clears 'types' and enumerates all registered ofproto types into it. The
437 * caller must first initialize the sset. */
439 ofproto_enumerate_types(struct sset *types)
444 for (i = 0; i < n_ofproto_classes; i++) {
445 ofproto_classes[i]->enumerate_types(types);
449 /* Returns the fully spelled out name for the given ofproto 'type'.
451 * Normalized type string can be compared with strcmp(). Unnormalized type
452 * string might be the same even if they have different spellings. */
454 ofproto_normalize_type(const char *type)
456 return type && type[0] ? type : "system";
459 /* Clears 'names' and enumerates the names of all known created ofprotos with
460 * the given 'type'. The caller must first initialize the sset. Returns 0 if
461 * successful, otherwise a positive errno value.
463 * Some kinds of datapaths might not be practically enumerable. This is not
464 * considered an error. */
466 ofproto_enumerate_names(const char *type, struct sset *names)
468 const struct ofproto_class *class = ofproto_class_find__(type);
469 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
473 ofproto_create(const char *datapath_name, const char *datapath_type,
474 struct ofproto **ofprotop)
476 const struct ofproto_class *class;
477 struct ofproto *ofproto;
483 ofproto_unixctl_init();
485 datapath_type = ofproto_normalize_type(datapath_type);
486 class = ofproto_class_find__(datapath_type);
488 VLOG_WARN("could not create datapath %s of unknown type %s",
489 datapath_name, datapath_type);
493 ofproto = class->alloc();
495 VLOG_ERR("failed to allocate datapath %s of type %s",
496 datapath_name, datapath_type);
501 ovs_mutex_lock(&ofproto_mutex);
502 memset(ofproto, 0, sizeof *ofproto);
503 ofproto->ofproto_class = class;
504 ofproto->name = xstrdup(datapath_name);
505 ofproto->type = xstrdup(datapath_type);
506 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
507 hash_string(ofproto->name, 0));
508 ofproto->datapath_id = 0;
509 ofproto->forward_bpdu = false;
510 ofproto->fallback_dpid = pick_fallback_dpid();
511 ofproto->mfr_desc = NULL;
512 ofproto->hw_desc = NULL;
513 ofproto->sw_desc = NULL;
514 ofproto->serial_desc = NULL;
515 ofproto->dp_desc = NULL;
516 ofproto->frag_handling = OFPC_FRAG_NORMAL;
517 hmap_init(&ofproto->ports);
518 hmap_init(&ofproto->ofport_usage);
519 shash_init(&ofproto->port_by_name);
520 simap_init(&ofproto->ofp_requests);
521 ofproto->max_ports = ofp_to_u16(OFPP_MAX);
522 ofproto->eviction_group_timer = LLONG_MIN;
523 ofproto->tables = NULL;
524 ofproto->n_tables = 0;
525 hindex_init(&ofproto->cookies);
526 list_init(&ofproto->expirable);
527 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
528 ofproto->state = S_OPENFLOW;
529 list_init(&ofproto->pending);
530 ofproto->n_pending = 0;
531 hmap_init(&ofproto->deletions);
532 guarded_list_init(&ofproto->rule_executes);
533 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
534 ofproto->first_op = ofproto->last_op = LLONG_MIN;
535 ofproto->next_op_report = LLONG_MAX;
536 ofproto->op_backoff = LLONG_MIN;
537 ofproto->vlan_bitmap = NULL;
538 ofproto->vlans_changed = false;
539 ofproto->min_mtu = INT_MAX;
540 ovs_rwlock_init(&ofproto->groups_rwlock);
541 hmap_init(&ofproto->groups);
542 ovs_mutex_unlock(&ofproto_mutex);
543 ofproto->ogf.capabilities = OFPGFC_CHAINING | OFPGFC_SELECT_LIVENESS |
544 OFPGFC_SELECT_WEIGHT;
545 ofproto->ogf.max_groups[OFPGT11_ALL] = OFPG_MAX;
546 ofproto->ogf.max_groups[OFPGT11_SELECT] = OFPG_MAX;
547 ofproto->ogf.max_groups[OFPGT11_INDIRECT] = OFPG_MAX;
548 ofproto->ogf.max_groups[OFPGT11_FF] = OFPG_MAX;
549 ofproto->ogf.actions[0] =
550 (1 << OFPAT11_OUTPUT) |
551 (1 << OFPAT11_COPY_TTL_OUT) |
552 (1 << OFPAT11_COPY_TTL_IN) |
553 (1 << OFPAT11_SET_MPLS_TTL) |
554 (1 << OFPAT11_DEC_MPLS_TTL) |
555 (1 << OFPAT11_PUSH_VLAN) |
556 (1 << OFPAT11_POP_VLAN) |
557 (1 << OFPAT11_PUSH_MPLS) |
558 (1 << OFPAT11_POP_MPLS) |
559 (1 << OFPAT11_SET_QUEUE) |
560 (1 << OFPAT11_GROUP) |
561 (1 << OFPAT11_SET_NW_TTL) |
562 (1 << OFPAT11_DEC_NW_TTL) |
563 (1 << OFPAT12_SET_FIELD);
565 * (1 << OFPAT13_PUSH_PBB) |
566 * (1 << OFPAT13_POP_PBB) */
568 error = ofproto->ofproto_class->construct(ofproto);
570 VLOG_ERR("failed to open datapath %s: %s",
571 datapath_name, ovs_strerror(error));
572 ofproto_destroy__(ofproto);
576 /* Check that hidden tables, if any, are at the end. */
577 ovs_assert(ofproto->n_tables);
578 for (i = 0; i + 1 < ofproto->n_tables; i++) {
579 enum oftable_flags flags = ofproto->tables[i].flags;
580 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
582 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
585 ofproto->datapath_id = pick_datapath_id(ofproto);
588 /* Initialize meters table. */
589 if (ofproto->ofproto_class->meter_get_features) {
590 ofproto->ofproto_class->meter_get_features(ofproto,
591 &ofproto->meter_features);
593 memset(&ofproto->meter_features, 0, sizeof ofproto->meter_features);
595 ofproto->meters = xzalloc((ofproto->meter_features.max_meters + 1)
596 * sizeof(struct meter *));
602 /* Must be called (only) by an ofproto implementation in its constructor
603 * function. See the large comment on 'construct' in struct ofproto_class for
606 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
608 struct oftable *table;
610 ovs_assert(!ofproto->n_tables);
611 ovs_assert(n_tables >= 1 && n_tables <= 255);
613 ofproto->n_tables = n_tables;
614 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
615 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
620 /* To be optionally called (only) by an ofproto implementation in its
621 * constructor function. See the large comment on 'construct' in struct
622 * ofproto_class for details.
624 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
625 * will then ensure that actions passed into the ofproto implementation will
626 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
627 * function is not called, there will be no such restriction.
629 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
630 * the 'max_ports' restriction. */
632 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
634 ovs_assert(max_ports <= ofp_to_u16(OFPP_MAX));
635 ofproto->max_ports = max_ports;
639 ofproto_get_datapath_id(const struct ofproto *ofproto)
641 return ofproto->datapath_id;
645 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
647 uint64_t old_dpid = p->datapath_id;
648 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
649 if (p->datapath_id != old_dpid) {
650 /* Force all active connections to reconnect, since there is no way to
651 * notify a controller that the datapath ID has changed. */
652 ofproto_reconnect_controllers(p);
657 ofproto_set_controllers(struct ofproto *p,
658 const struct ofproto_controller *controllers,
659 size_t n_controllers, uint32_t allowed_versions)
661 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
666 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
668 connmgr_set_fail_mode(p->connmgr, fail_mode);
671 /* Drops the connections between 'ofproto' and all of its controllers, forcing
672 * them to reconnect. */
674 ofproto_reconnect_controllers(struct ofproto *ofproto)
676 connmgr_reconnect(ofproto->connmgr);
679 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
680 * in-band control should guarantee access, in the same way that in-band
681 * control guarantees access to OpenFlow controllers. */
683 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
684 const struct sockaddr_in *extras, size_t n)
686 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
689 /* Sets the OpenFlow queue used by flows set up by in-band control on
690 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
691 * flows will use the default queue. */
693 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
695 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
698 /* Sets the number of flows at which eviction from the kernel flow table
701 ofproto_set_flow_limit(unsigned limit)
703 ofproto_flow_limit = limit;
706 /* Sets the maximum idle time for flows in the datapath before they are
709 ofproto_set_max_idle(unsigned max_idle)
711 ofproto_max_idle = max_idle;
714 /* If forward_bpdu is true, the NORMAL action will forward frames with
715 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
716 * the NORMAL action will drop these frames. */
718 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
720 bool old_val = ofproto->forward_bpdu;
721 ofproto->forward_bpdu = forward_bpdu;
722 if (old_val != ofproto->forward_bpdu) {
723 if (ofproto->ofproto_class->forward_bpdu_changed) {
724 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
729 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
730 * 'idle_time', in seconds, and the maximum number of MAC table entries to
733 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
736 if (ofproto->ofproto_class->set_mac_table_config) {
737 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
743 ofproto_set_threads(int n_handlers_, int n_revalidators_)
745 int threads = MAX(count_cpu_cores(), 2);
747 n_revalidators = MAX(n_revalidators_, 0);
748 n_handlers = MAX(n_handlers_, 0);
750 if (!n_revalidators) {
751 n_revalidators = n_handlers
752 ? MAX(threads - (int) n_handlers, 1)
757 n_handlers = MAX(threads - (int) n_revalidators, 1);
762 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
765 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
769 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
771 return connmgr_set_snoops(ofproto->connmgr, snoops);
775 ofproto_set_netflow(struct ofproto *ofproto,
776 const struct netflow_options *nf_options)
778 if (nf_options && sset_is_empty(&nf_options->collectors)) {
782 if (ofproto->ofproto_class->set_netflow) {
783 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
785 return nf_options ? EOPNOTSUPP : 0;
790 ofproto_set_sflow(struct ofproto *ofproto,
791 const struct ofproto_sflow_options *oso)
793 if (oso && sset_is_empty(&oso->targets)) {
797 if (ofproto->ofproto_class->set_sflow) {
798 return ofproto->ofproto_class->set_sflow(ofproto, oso);
800 return oso ? EOPNOTSUPP : 0;
805 ofproto_set_ipfix(struct ofproto *ofproto,
806 const struct ofproto_ipfix_bridge_exporter_options *bo,
807 const struct ofproto_ipfix_flow_exporter_options *fo,
810 if (ofproto->ofproto_class->set_ipfix) {
811 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
813 return (bo || fo) ? EOPNOTSUPP : 0;
818 ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
820 flow_restore_wait = flow_restore_wait_db;
824 ofproto_get_flow_restore_wait(void)
826 return flow_restore_wait;
830 /* Spanning Tree Protocol (STP) configuration. */
832 /* Configures STP on 'ofproto' using the settings defined in 's'. If
833 * 's' is NULL, disables STP.
835 * Returns 0 if successful, otherwise a positive errno value. */
837 ofproto_set_stp(struct ofproto *ofproto,
838 const struct ofproto_stp_settings *s)
840 return (ofproto->ofproto_class->set_stp
841 ? ofproto->ofproto_class->set_stp(ofproto, s)
845 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
846 * 'enabled' member of 's' is false, then the other members are not
849 * Returns 0 if successful, otherwise a positive errno value. */
851 ofproto_get_stp_status(struct ofproto *ofproto,
852 struct ofproto_stp_status *s)
854 return (ofproto->ofproto_class->get_stp_status
855 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
859 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
860 * in 's'. The caller is responsible for assigning STP port numbers
861 * (using the 'port_num' member in the range of 1 through 255, inclusive)
862 * and ensuring there are no duplicates. If the 's' is NULL, then STP
863 * is disabled on the port.
865 * Returns 0 if successful, otherwise a positive errno value.*/
867 ofproto_port_set_stp(struct ofproto *ofproto, ofp_port_t ofp_port,
868 const struct ofproto_port_stp_settings *s)
870 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
872 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
873 ofproto->name, ofp_port);
877 return (ofproto->ofproto_class->set_stp_port
878 ? ofproto->ofproto_class->set_stp_port(ofport, s)
882 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
883 * 's'. If the 'enabled' member in 's' is false, then the other members
884 * are not meaningful.
886 * Returns 0 if successful, otherwise a positive errno value.*/
888 ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
889 struct ofproto_port_stp_status *s)
891 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
893 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
894 "port %"PRIu16, ofproto->name, ofp_port);
898 return (ofproto->ofproto_class->get_stp_port_status
899 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
903 /* Retrieves STP port statistics of 'ofp_port' on 'ofproto' and stores it in
904 * 's'. If the 'enabled' member in 's' is false, then the other members
905 * are not meaningful.
907 * Returns 0 if successful, otherwise a positive errno value.*/
909 ofproto_port_get_stp_stats(struct ofproto *ofproto, ofp_port_t ofp_port,
910 struct ofproto_port_stp_stats *s)
912 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
914 VLOG_WARN_RL(&rl, "%s: cannot get STP stats on nonexistent "
915 "port %"PRIu16, ofproto->name, ofp_port);
919 return (ofproto->ofproto_class->get_stp_port_stats
920 ? ofproto->ofproto_class->get_stp_port_stats(ofport, s)
924 /* Queue DSCP configuration. */
926 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
927 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
928 * to implement QoS. Instead, it is used to implement features which require
929 * knowledge of what queues exist on a port, and some basic information about
932 * Returns 0 if successful, otherwise a positive errno value. */
934 ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port,
935 const struct ofproto_port_queue *queues,
938 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
941 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
942 ofproto->name, ofp_port);
946 return (ofproto->ofproto_class->set_queues
947 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
951 /* Connectivity Fault Management configuration. */
953 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
955 ofproto_port_clear_cfm(struct ofproto *ofproto, ofp_port_t ofp_port)
957 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
958 if (ofport && ofproto->ofproto_class->set_cfm) {
959 ofproto->ofproto_class->set_cfm(ofport, NULL);
963 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
964 * basic configuration from the configuration members in 'cfm', and the remote
965 * maintenance point ID from remote_mpid. Ignores the statistics members of
968 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
970 ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port,
971 const struct cfm_settings *s)
973 struct ofport *ofport;
976 ofport = ofproto_get_port(ofproto, ofp_port);
978 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
979 ofproto->name, ofp_port);
983 /* XXX: For configuration simplicity, we only support one remote_mpid
984 * outside of the CFM module. It's not clear if this is the correct long
985 * term solution or not. */
986 error = (ofproto->ofproto_class->set_cfm
987 ? ofproto->ofproto_class->set_cfm(ofport, s)
990 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
991 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
992 ovs_strerror(error));
996 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
997 * 'ofproto' does not have a port 'ofp_port'. */
999 ofproto_port_set_bfd(struct ofproto *ofproto, ofp_port_t ofp_port,
1000 const struct smap *cfg)
1002 struct ofport *ofport;
1005 ofport = ofproto_get_port(ofproto, ofp_port);
1007 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
1008 ofproto->name, ofp_port);
1012 error = (ofproto->ofproto_class->set_bfd
1013 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
1016 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
1017 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
1018 ovs_strerror(error));
1022 /* Populates 'status' with the status of BFD on 'ofport'. Returns 0 on
1023 * success. Returns a negative number if there is no status change since
1024 * last update. Returns a positive errno otherwise. Has no effect if
1025 * 'ofp_port' is not an OpenFlow port in 'ofproto'.
1027 * The caller must provide and own '*status'. */
1029 ofproto_port_get_bfd_status(struct ofproto *ofproto, ofp_port_t ofp_port,
1030 struct smap *status)
1032 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1033 return (ofport && ofproto->ofproto_class->get_bfd_status
1034 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
1038 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
1039 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
1040 * 0 if LACP partner information is not current (generally indicating a
1041 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
1043 ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port)
1045 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1046 return (ofport && ofproto->ofproto_class->port_is_lacp_current
1047 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
1053 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
1054 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
1055 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
1056 * configuration plus, if there is more than one slave, a bonding
1059 * If 'aux' is already registered then this function updates its configuration
1060 * to 's'. Otherwise, this function registers a new bundle.
1062 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
1065 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
1066 const struct ofproto_bundle_settings *s)
1068 return (ofproto->ofproto_class->bundle_set
1069 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
1073 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
1074 * If no such bundle has been registered, this has no effect. */
1076 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
1078 return ofproto_bundle_register(ofproto, aux, NULL);
1082 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
1083 * If 'aux' is already registered then this function updates its configuration
1084 * to 's'. Otherwise, this function registers a new mirror. */
1086 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
1087 const struct ofproto_mirror_settings *s)
1089 return (ofproto->ofproto_class->mirror_set
1090 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
1094 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
1095 * If no mirror has been registered, this has no effect. */
1097 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
1099 return ofproto_mirror_register(ofproto, aux, NULL);
1102 /* Retrieves statistics from mirror associated with client data pointer
1103 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
1104 * 'bytes', respectively. If a particular counters is not supported,
1105 * the appropriate argument is set to UINT64_MAX. */
1107 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
1108 uint64_t *packets, uint64_t *bytes)
1110 if (!ofproto->ofproto_class->mirror_get_stats) {
1111 *packets = *bytes = UINT64_MAX;
1115 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
1119 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
1120 * which all packets are flooded, instead of using MAC learning. If
1121 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1123 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
1126 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
1128 return (ofproto->ofproto_class->set_flood_vlans
1129 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
1133 /* Returns true if 'aux' is a registered bundle that is currently in use as the
1134 * output for a mirror. */
1136 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
1138 return (ofproto->ofproto_class->is_mirror_output_bundle
1139 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
1143 /* Configuration of OpenFlow tables. */
1145 /* Returns the number of OpenFlow tables in 'ofproto'. */
1147 ofproto_get_n_tables(const struct ofproto *ofproto)
1149 return ofproto->n_tables;
1152 /* Returns the number of Controller visible OpenFlow tables
1153 * in 'ofproto'. This number will exclude Hidden tables.
1154 * This funtion's return value should be less or equal to that of
1155 * ofproto_get_n_tables() . */
1157 ofproto_get_n_visible_tables(const struct ofproto *ofproto)
1159 uint8_t n = ofproto->n_tables;
1161 /* Count only non-hidden tables in the number of tables. (Hidden tables,
1162 * if present, are always at the end.) */
1163 while(n && (ofproto->tables[n - 1].flags & OFTABLE_HIDDEN)) {
1170 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
1171 * settings from 's'. 'table_id' must be in the range 0 through the number of
1172 * OpenFlow tables in 'ofproto' minus 1, inclusive.
1174 * For read-only tables, only the name may be configured. */
1176 ofproto_configure_table(struct ofproto *ofproto, int table_id,
1177 const struct ofproto_table_settings *s)
1179 struct oftable *table;
1181 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
1182 table = &ofproto->tables[table_id];
1184 oftable_set_name(table, s->name);
1186 if (table->flags & OFTABLE_READONLY) {
1191 oftable_enable_eviction(table, s->groups, s->n_groups);
1193 oftable_disable_eviction(table);
1196 table->max_flows = s->max_flows;
1197 fat_rwlock_wrlock(&table->cls.rwlock);
1198 if (classifier_count(&table->cls) > table->max_flows
1199 && table->eviction_fields) {
1200 /* 'table' contains more flows than allowed. We might not be able to
1201 * evict them right away because of the asynchronous nature of flow
1202 * table changes. Schedule eviction for later. */
1203 switch (ofproto->state) {
1205 ofproto->state = S_EVICT;
1209 /* We're already deleting flows, nothing more to do. */
1214 classifier_set_prefix_fields(&table->cls,
1215 s->prefix_fields, s->n_prefix_fields);
1217 fat_rwlock_unlock(&table->cls.rwlock);
1221 ofproto_has_snoops(const struct ofproto *ofproto)
1223 return connmgr_has_snoops(ofproto->connmgr);
1227 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1229 connmgr_get_snoops(ofproto->connmgr, snoops);
1233 ofproto_rule_delete__(struct ofproto *ofproto, struct rule *rule,
1235 OVS_REQUIRES(ofproto_mutex)
1237 struct ofopgroup *group;
1239 ovs_assert(!rule->pending);
1241 group = ofopgroup_create_unattached(ofproto);
1242 delete_flow__(rule, group, reason);
1243 ofopgroup_submit(group);
1246 /* Deletes 'rule' from 'ofproto'.
1248 * Within an ofproto implementation, this function allows an ofproto
1249 * implementation to destroy any rules that remain when its ->destruct()
1250 * function is called. This function is not suitable for use elsewhere in an
1251 * ofproto implementation.
1253 * This function implements steps 4.4 and 4.5 in the section titled "Rule Life
1254 * Cycle" in ofproto-provider.h. */
1256 ofproto_rule_delete(struct ofproto *ofproto, struct rule *rule)
1257 OVS_EXCLUDED(ofproto_mutex)
1259 struct ofopgroup *group;
1261 ovs_mutex_lock(&ofproto_mutex);
1262 ovs_assert(!rule->pending);
1264 group = ofopgroup_create_unattached(ofproto);
1265 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1266 oftable_remove_rule__(ofproto, rule);
1267 ofproto->ofproto_class->rule_delete(rule);
1268 ofopgroup_submit(group);
1270 ovs_mutex_unlock(&ofproto_mutex);
1274 ofproto_flush__(struct ofproto *ofproto)
1275 OVS_EXCLUDED(ofproto_mutex)
1277 struct oftable *table;
1279 if (ofproto->ofproto_class->flush) {
1280 ofproto->ofproto_class->flush(ofproto);
1283 ovs_mutex_lock(&ofproto_mutex);
1284 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1285 struct rule *rule, *next_rule;
1286 struct cls_cursor cursor;
1288 if (table->flags & OFTABLE_HIDDEN) {
1292 fat_rwlock_rdlock(&table->cls.rwlock);
1293 cls_cursor_init(&cursor, &table->cls, NULL);
1294 fat_rwlock_unlock(&table->cls.rwlock);
1295 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1296 if (!rule->pending) {
1297 ofproto_rule_delete__(ofproto, rule, OFPRR_DELETE);
1301 ovs_mutex_unlock(&ofproto_mutex);
1304 static void delete_group(struct ofproto *ofproto, uint32_t group_id);
1307 ofproto_destroy__(struct ofproto *ofproto)
1308 OVS_EXCLUDED(ofproto_mutex)
1310 struct oftable *table;
1312 ovs_assert(list_is_empty(&ofproto->pending));
1314 destroy_rule_executes(ofproto);
1315 delete_group(ofproto, OFPG_ALL);
1317 guarded_list_destroy(&ofproto->rule_executes);
1318 ovs_rwlock_destroy(&ofproto->groups_rwlock);
1319 hmap_destroy(&ofproto->groups);
1321 connmgr_destroy(ofproto->connmgr);
1323 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1324 free(ofproto->name);
1325 free(ofproto->type);
1326 free(ofproto->mfr_desc);
1327 free(ofproto->hw_desc);
1328 free(ofproto->sw_desc);
1329 free(ofproto->serial_desc);
1330 free(ofproto->dp_desc);
1331 hmap_destroy(&ofproto->ports);
1332 hmap_destroy(&ofproto->ofport_usage);
1333 shash_destroy(&ofproto->port_by_name);
1334 simap_destroy(&ofproto->ofp_requests);
1336 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1337 oftable_destroy(table);
1339 free(ofproto->tables);
1341 hmap_destroy(&ofproto->deletions);
1343 free(ofproto->vlan_bitmap);
1345 ofproto->ofproto_class->dealloc(ofproto);
1349 ofproto_destroy(struct ofproto *p)
1350 OVS_EXCLUDED(ofproto_mutex)
1352 struct ofport *ofport, *next_ofport;
1353 struct ofport_usage *usage, *next_usage;
1360 meter_delete(p, 1, p->meter_features.max_meters);
1361 p->meter_features.max_meters = 0;
1367 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1368 ofport_destroy(ofport);
1371 HMAP_FOR_EACH_SAFE (usage, next_usage, hmap_node, &p->ofport_usage) {
1372 hmap_remove(&p->ofport_usage, &usage->hmap_node);
1376 p->ofproto_class->destruct(p);
1377 /* Destroying rules is deferred, must have 'ofproto' around for them. */
1378 ovsrcu_postpone(ofproto_destroy__, p);
1381 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1382 * kernel datapath, for example, this destroys the datapath in the kernel, and
1383 * with the netdev-based datapath, it tears down the data structures that
1384 * represent the datapath.
1386 * The datapath should not be currently open as an ofproto. */
1388 ofproto_delete(const char *name, const char *type)
1390 const struct ofproto_class *class = ofproto_class_find__(type);
1391 return (!class ? EAFNOSUPPORT
1392 : !class->del ? EACCES
1393 : class->del(type, name));
1397 process_port_change(struct ofproto *ofproto, int error, char *devname)
1399 if (error == ENOBUFS) {
1400 reinit_ports(ofproto);
1401 } else if (!error) {
1402 update_port(ofproto, devname);
1408 ofproto_type_run(const char *datapath_type)
1410 const struct ofproto_class *class;
1413 datapath_type = ofproto_normalize_type(datapath_type);
1414 class = ofproto_class_find__(datapath_type);
1416 error = class->type_run ? class->type_run(datapath_type) : 0;
1417 if (error && error != EAGAIN) {
1418 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1419 datapath_type, ovs_strerror(error));
1425 ofproto_type_wait(const char *datapath_type)
1427 const struct ofproto_class *class;
1429 datapath_type = ofproto_normalize_type(datapath_type);
1430 class = ofproto_class_find__(datapath_type);
1432 if (class->type_wait) {
1433 class->type_wait(datapath_type);
1438 any_pending_ops(const struct ofproto *p)
1439 OVS_EXCLUDED(ofproto_mutex)
1443 ovs_mutex_lock(&ofproto_mutex);
1444 b = !list_is_empty(&p->pending);
1445 ovs_mutex_unlock(&ofproto_mutex);
1451 ofproto_run(struct ofproto *p)
1456 error = p->ofproto_class->run(p);
1457 if (error && error != EAGAIN) {
1458 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, ovs_strerror(error));
1461 run_rule_executes(p);
1463 /* Restore the eviction group heap invariant occasionally. */
1464 if (p->eviction_group_timer < time_msec()) {
1467 p->eviction_group_timer = time_msec() + 1000;
1469 for (i = 0; i < p->n_tables; i++) {
1470 struct oftable *table = &p->tables[i];
1471 struct eviction_group *evg;
1472 struct cls_cursor cursor;
1475 if (!table->eviction_fields) {
1479 ovs_mutex_lock(&ofproto_mutex);
1480 fat_rwlock_rdlock(&table->cls.rwlock);
1481 cls_cursor_init(&cursor, &table->cls, NULL);
1482 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1483 if (rule->idle_timeout || rule->hard_timeout) {
1484 if (!rule->eviction_group) {
1485 eviction_group_add_rule(rule);
1487 heap_raw_change(&rule->evg_node,
1488 rule_eviction_priority(p, rule));
1492 fat_rwlock_unlock(&table->cls.rwlock);
1494 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
1495 heap_rebuild(&evg->rules);
1497 ovs_mutex_unlock(&ofproto_mutex);
1501 if (p->ofproto_class->port_poll) {
1504 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1505 process_port_change(p, error, devname);
1509 new_seq = seq_read(connectivity_seq_get());
1510 if (new_seq != p->change_seq) {
1511 struct sset devnames;
1512 const char *devname;
1513 struct ofport *ofport;
1515 /* Update OpenFlow port status for any port whose netdev has changed.
1517 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1518 * destroyed, so it's not safe to update ports directly from the
1519 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1520 * need this two-phase approach. */
1521 sset_init(&devnames);
1522 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1523 uint64_t port_change_seq;
1525 port_change_seq = netdev_get_change_seq(ofport->netdev);
1526 if (ofport->change_seq != port_change_seq) {
1527 ofport->change_seq = port_change_seq;
1528 sset_add(&devnames, netdev_get_name(ofport->netdev));
1531 SSET_FOR_EACH (devname, &devnames) {
1532 update_port(p, devname);
1534 sset_destroy(&devnames);
1536 p->change_seq = new_seq;
1541 connmgr_run(p->connmgr, handle_openflow);
1545 connmgr_run(p->connmgr, NULL);
1547 if (!any_pending_ops(p)) {
1548 p->state = S_OPENFLOW;
1553 connmgr_run(p->connmgr, NULL);
1555 if (!any_pending_ops(p)) {
1556 connmgr_flushed(p->connmgr);
1557 p->state = S_OPENFLOW;
1565 if (time_msec() >= p->next_op_report) {
1566 long long int ago = (time_msec() - p->first_op) / 1000;
1567 long long int interval = (p->last_op - p->first_op) / 1000;
1571 ds_put_format(&s, "%d flow_mods ",
1572 p->n_add + p->n_delete + p->n_modify);
1573 if (interval == ago) {
1574 ds_put_format(&s, "in the last %lld s", ago);
1575 } else if (interval) {
1576 ds_put_format(&s, "in the %lld s starting %lld s ago",
1579 ds_put_format(&s, "%lld s ago", ago);
1582 ds_put_cstr(&s, " (");
1584 ds_put_format(&s, "%d adds, ", p->n_add);
1587 ds_put_format(&s, "%d deletes, ", p->n_delete);
1590 ds_put_format(&s, "%d modifications, ", p->n_modify);
1593 ds_put_char(&s, ')');
1595 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1598 p->n_add = p->n_delete = p->n_modify = 0;
1599 p->next_op_report = LLONG_MAX;
1606 ofproto_wait(struct ofproto *p)
1608 p->ofproto_class->wait(p);
1609 if (p->ofproto_class->port_poll_wait) {
1610 p->ofproto_class->port_poll_wait(p);
1612 seq_wait(connectivity_seq_get(), p->change_seq);
1616 connmgr_wait(p->connmgr, true);
1621 connmgr_wait(p->connmgr, false);
1622 if (!any_pending_ops(p)) {
1623 poll_immediate_wake();
1630 ofproto_is_alive(const struct ofproto *p)
1632 return connmgr_has_controllers(p->connmgr);
1635 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1636 * memory_report(). */
1638 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1640 const struct oftable *table;
1641 unsigned int n_rules;
1643 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1645 ovs_mutex_lock(&ofproto_mutex);
1646 simap_increase(usage, "ops",
1647 ofproto->n_pending + hmap_count(&ofproto->deletions));
1648 ovs_mutex_unlock(&ofproto_mutex);
1651 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1652 fat_rwlock_rdlock(&table->cls.rwlock);
1653 n_rules += classifier_count(&table->cls);
1654 fat_rwlock_unlock(&table->cls.rwlock);
1656 simap_increase(usage, "rules", n_rules);
1658 if (ofproto->ofproto_class->get_memory_usage) {
1659 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1662 connmgr_get_memory_usage(ofproto->connmgr, usage);
1666 ofproto_type_get_memory_usage(const char *datapath_type, struct simap *usage)
1668 const struct ofproto_class *class;
1670 datapath_type = ofproto_normalize_type(datapath_type);
1671 class = ofproto_class_find__(datapath_type);
1673 if (class && class->type_get_memory_usage) {
1674 class->type_get_memory_usage(datapath_type, usage);
1679 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1682 connmgr_get_controller_info(ofproto->connmgr, info);
1686 ofproto_free_ofproto_controller_info(struct shash *info)
1688 connmgr_free_controller_info(info);
1691 /* Makes a deep copy of 'old' into 'port'. */
1693 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1695 port->name = xstrdup(old->name);
1696 port->type = xstrdup(old->type);
1697 port->ofp_port = old->ofp_port;
1700 /* Frees memory allocated to members of 'ofproto_port'.
1702 * Do not call this function on an ofproto_port obtained from
1703 * ofproto_port_dump_next(): that function retains ownership of the data in the
1706 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1708 free(ofproto_port->name);
1709 free(ofproto_port->type);
1712 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1714 * This function provides no status indication. An error status for the entire
1715 * dump operation is provided when it is completed by calling
1716 * ofproto_port_dump_done().
1719 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1720 const struct ofproto *ofproto)
1722 dump->ofproto = ofproto;
1723 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1727 /* Attempts to retrieve another port from 'dump', which must have been created
1728 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1729 * 'port' and returns true. On failure, returns false.
1731 * Failure might indicate an actual error or merely that the last port has been
1732 * dumped. An error status for the entire dump operation is provided when it
1733 * is completed by calling ofproto_port_dump_done().
1735 * The ofproto owns the data stored in 'port'. It will remain valid until at
1736 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1737 * ofproto_port_dump_done(). */
1739 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1740 struct ofproto_port *port)
1742 const struct ofproto *ofproto = dump->ofproto;
1748 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1751 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1757 /* Completes port table dump operation 'dump', which must have been created
1758 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1759 * error-free, otherwise a positive errno value describing the problem. */
1761 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1763 const struct ofproto *ofproto = dump->ofproto;
1765 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1768 return dump->error == EOF ? 0 : dump->error;
1771 /* Returns the type to pass to netdev_open() when a datapath of type
1772 * 'datapath_type' has a port of type 'port_type', for a few special
1773 * cases when a netdev type differs from a port type. For example, when
1774 * using the userspace datapath, a port of type "internal" needs to be
1777 * Returns either 'type' itself or a string literal, which must not be
1780 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1782 const struct ofproto_class *class;
1784 datapath_type = ofproto_normalize_type(datapath_type);
1785 class = ofproto_class_find__(datapath_type);
1790 return (class->port_open_type
1791 ? class->port_open_type(datapath_type, port_type)
1795 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1796 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1797 * the port's OpenFlow port number.
1799 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1800 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1801 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1802 * 'ofp_portp' is non-null). */
1804 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1805 ofp_port_t *ofp_portp)
1807 ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1810 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1812 const char *netdev_name = netdev_get_name(netdev);
1814 simap_put(&ofproto->ofp_requests, netdev_name,
1815 ofp_to_u16(ofp_port));
1816 update_port(ofproto, netdev_name);
1819 *ofp_portp = OFPP_NONE;
1821 struct ofproto_port ofproto_port;
1823 error = ofproto_port_query_by_name(ofproto,
1824 netdev_get_name(netdev),
1827 *ofp_portp = ofproto_port.ofp_port;
1828 ofproto_port_destroy(&ofproto_port);
1835 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1836 * initializes '*port' appropriately; on failure, returns a positive errno
1839 * The caller owns the data in 'ofproto_port' and must free it with
1840 * ofproto_port_destroy() when it is no longer needed. */
1842 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1843 struct ofproto_port *port)
1847 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1849 memset(port, 0, sizeof *port);
1854 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1855 * Returns 0 if successful, otherwise a positive errno. */
1857 ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port)
1859 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1860 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1861 struct simap_node *ofp_request_node;
1864 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1865 if (ofp_request_node) {
1866 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1869 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1870 if (!error && ofport) {
1871 /* 'name' is the netdev's name and update_port() is going to close the
1872 * netdev. Just in case update_port() refers to 'name' after it
1873 * destroys 'ofport', make a copy of it around the update_port()
1875 char *devname = xstrdup(name);
1876 update_port(ofproto, devname);
1883 flow_mod_init(struct ofputil_flow_mod *fm,
1884 const struct match *match, unsigned int priority,
1885 const struct ofpact *ofpacts, size_t ofpacts_len,
1886 enum ofp_flow_mod_command command)
1888 memset(fm, 0, sizeof *fm);
1890 fm->priority = priority;
1893 fm->modify_cookie = false;
1895 fm->command = command;
1896 fm->idle_timeout = 0;
1897 fm->hard_timeout = 0;
1898 fm->buffer_id = UINT32_MAX;
1899 fm->out_port = OFPP_ANY;
1900 fm->out_group = OFPG_ANY;
1902 fm->ofpacts = CONST_CAST(struct ofpact *, ofpacts);
1903 fm->ofpacts_len = ofpacts_len;
1907 simple_flow_mod(struct ofproto *ofproto,
1908 const struct match *match, unsigned int priority,
1909 const struct ofpact *ofpacts, size_t ofpacts_len,
1910 enum ofp_flow_mod_command command)
1912 struct ofputil_flow_mod fm;
1914 flow_mod_init(&fm, match, priority, ofpacts, ofpacts_len, command);
1916 return handle_flow_mod__(ofproto, NULL, &fm, NULL);
1919 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1920 * performs the 'n_actions' actions in 'actions'. The new flow will not
1923 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1924 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1925 * controllers; otherwise, it will be hidden.
1927 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1929 * This is a helper function for in-band control and fail-open. */
1931 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1932 unsigned int priority,
1933 const struct ofpact *ofpacts, size_t ofpacts_len)
1934 OVS_EXCLUDED(ofproto_mutex)
1936 const struct rule *rule;
1939 /* First do a cheap check whether the rule we're looking for already exists
1940 * with the actions that we want. If it does, then we're done. */
1941 fat_rwlock_rdlock(&ofproto->tables[0].cls.rwlock);
1942 rule = rule_from_cls_rule(classifier_find_match_exactly(
1943 &ofproto->tables[0].cls, match, priority));
1945 const struct rule_actions *actions = rule_get_actions(rule);
1946 must_add = !ofpacts_equal(actions->ofpacts, actions->ofpacts_len,
1947 ofpacts, ofpacts_len);
1951 fat_rwlock_unlock(&ofproto->tables[0].cls.rwlock);
1953 /* If there's no such rule or the rule doesn't have the actions we want,
1954 * fall back to a executing a full flow mod. We can't optimize this at
1955 * all because we didn't take enough locks above to ensure that the flow
1956 * table didn't already change beneath us. */
1958 simple_flow_mod(ofproto, match, priority, ofpacts, ofpacts_len,
1959 OFPFC_MODIFY_STRICT);
1963 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1964 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1965 * operation cannot be initiated now but may be retried later.
1967 * This is a helper function for in-band control and fail-open and the "learn"
1970 ofproto_flow_mod(struct ofproto *ofproto, struct ofputil_flow_mod *fm)
1971 OVS_EXCLUDED(ofproto_mutex)
1973 /* Optimize for the most common case of a repeated learn action.
1974 * If an identical flow already exists we only need to update its
1975 * 'modified' time. */
1976 if (fm->command == OFPFC_MODIFY_STRICT && fm->table_id != OFPTT_ALL
1977 && !(fm->flags & OFPUTIL_FF_RESET_COUNTS)) {
1978 struct oftable *table = &ofproto->tables[fm->table_id];
1982 fat_rwlock_rdlock(&table->cls.rwlock);
1983 rule = rule_from_cls_rule(classifier_find_match_exactly(&table->cls,
1987 /* Reading many of the rule fields and writing on 'modified'
1988 * requires the rule->mutex. Also, rule->actions may change
1989 * if rule->mutex is not held. */
1990 const struct rule_actions *actions;
1992 ovs_mutex_lock(&rule->mutex);
1993 actions = rule_get_actions(rule);
1994 if (rule->idle_timeout == fm->idle_timeout
1995 && rule->hard_timeout == fm->hard_timeout
1996 && rule->flags == (fm->flags & OFPUTIL_FF_STATE)
1997 && (!fm->modify_cookie || (fm->new_cookie == rule->flow_cookie))
1998 && ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
1999 actions->ofpacts, actions->ofpacts_len)) {
2000 /* Rule already exists and need not change, only update the
2001 modified timestamp. */
2002 rule->modified = time_msec();
2005 ovs_mutex_unlock(&rule->mutex);
2007 fat_rwlock_unlock(&table->cls.rwlock);
2014 return handle_flow_mod__(ofproto, NULL, fm, NULL);
2017 /* Resets the modified time for 'rule' or an equivalent rule. If 'rule' is not
2018 * in the classifier, but an equivalent rule is, unref 'rule' and ref the new
2019 * rule. Otherwise if 'rule' is no longer installed in the classifier,
2022 * Returns the rule whose modified time has been reset. */
2024 ofproto_refresh_rule(struct rule *rule)
2026 const struct oftable *table = &rule->ofproto->tables[rule->table_id];
2027 const struct cls_rule *cr = &rule->cr;
2030 /* do_add_flow() requires that the rule is not installed. We lock the
2031 * ofproto_mutex here so that another thread cannot add the flow before
2032 * we get a chance to add it.*/
2033 ovs_mutex_lock(&ofproto_mutex);
2035 fat_rwlock_rdlock(&table->cls.rwlock);
2036 r = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, cr));
2038 ofproto_rule_ref(r);
2040 fat_rwlock_unlock(&table->cls.rwlock);
2043 do_add_flow(rule->ofproto, NULL, NULL, 0, rule);
2044 } else if (r != rule) {
2045 ofproto_rule_unref(rule);
2048 ovs_mutex_unlock(&ofproto_mutex);
2050 /* Refresh the modified time for the rule. */
2051 ovs_mutex_lock(&rule->mutex);
2052 rule->modified = MAX(rule->modified, time_msec());
2053 ovs_mutex_unlock(&rule->mutex);
2058 /* Searches for a rule with matching criteria exactly equal to 'target' in
2059 * ofproto's table 0 and, if it finds one, deletes it.
2061 * This is a helper function for in-band control and fail-open. */
2063 ofproto_delete_flow(struct ofproto *ofproto,
2064 const struct match *target, unsigned int priority)
2065 OVS_EXCLUDED(ofproto_mutex)
2067 struct classifier *cls = &ofproto->tables[0].cls;
2070 /* First do a cheap check whether the rule we're looking for has already
2071 * been deleted. If so, then we're done. */
2072 fat_rwlock_rdlock(&cls->rwlock);
2073 rule = rule_from_cls_rule(classifier_find_match_exactly(cls, target,
2075 fat_rwlock_unlock(&cls->rwlock);
2080 /* Fall back to a executing a full flow mod. We can't optimize this at all
2081 * because we didn't take enough locks above to ensure that the flow table
2082 * didn't already change beneath us. */
2083 return simple_flow_mod(ofproto, target, priority, NULL, 0,
2084 OFPFC_DELETE_STRICT) != OFPROTO_POSTPONE;
2087 /* Starts the process of deleting all of the flows from all of ofproto's flow
2088 * tables and then reintroducing the flows required by in-band control and
2089 * fail-open. The process will complete in a later call to ofproto_run(). */
2091 ofproto_flush_flows(struct ofproto *ofproto)
2093 COVERAGE_INC(ofproto_flush);
2094 ofproto->state = S_FLUSH;
2098 reinit_ports(struct ofproto *p)
2100 struct ofproto_port_dump dump;
2101 struct sset devnames;
2102 struct ofport *ofport;
2103 struct ofproto_port ofproto_port;
2104 const char *devname;
2106 COVERAGE_INC(ofproto_reinit_ports);
2108 sset_init(&devnames);
2109 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2110 sset_add(&devnames, netdev_get_name(ofport->netdev));
2112 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2113 sset_add(&devnames, ofproto_port.name);
2116 SSET_FOR_EACH (devname, &devnames) {
2117 update_port(p, devname);
2119 sset_destroy(&devnames);
2123 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
2127 port_idx = simap_get(&ofproto->ofp_requests, netdev_name);
2128 port_idx = port_idx ? port_idx : UINT16_MAX;
2130 if (port_idx >= ofproto->max_ports
2131 || ofport_get_usage(ofproto, u16_to_ofp(port_idx)) == LLONG_MAX) {
2132 uint16_t lru_ofport = 0, end_port_no = ofproto->alloc_port_no;
2133 long long int last_used_at, lru = LLONG_MAX;
2135 /* Search for a free OpenFlow port number. We try not to
2136 * immediately reuse them to prevent problems due to old
2139 * We limit the automatically assigned port numbers to the lower half
2140 * of the port range, to reserve the upper half for assignment by
2143 if (++ofproto->alloc_port_no >= MIN(ofproto->max_ports, 32768)) {
2144 ofproto->alloc_port_no = 1;
2146 last_used_at = ofport_get_usage(ofproto,
2147 u16_to_ofp(ofproto->alloc_port_no));
2148 if (!last_used_at) {
2149 port_idx = ofproto->alloc_port_no;
2151 } else if ( last_used_at < time_msec() - 60*60*1000) {
2152 /* If the port with ofport 'ofproto->alloc_port_no' was deleted
2153 * more than an hour ago, consider it usable. */
2154 ofport_remove_usage(ofproto,
2155 u16_to_ofp(ofproto->alloc_port_no));
2156 port_idx = ofproto->alloc_port_no;
2158 } else if (last_used_at < lru) {
2160 lru_ofport = ofproto->alloc_port_no;
2163 if (ofproto->alloc_port_no == end_port_no) {
2165 port_idx = lru_ofport;
2172 ofport_set_usage(ofproto, u16_to_ofp(port_idx), LLONG_MAX);
2173 return u16_to_ofp(port_idx);
2177 dealloc_ofp_port(struct ofproto *ofproto, ofp_port_t ofp_port)
2179 if (ofp_to_u16(ofp_port) < ofproto->max_ports) {
2180 ofport_set_usage(ofproto, ofp_port, time_msec());
2184 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
2185 * pointer if the netdev cannot be opened. On success, also fills in
2187 static struct netdev *
2188 ofport_open(struct ofproto *ofproto,
2189 struct ofproto_port *ofproto_port,
2190 struct ofputil_phy_port *pp)
2192 enum netdev_flags flags;
2193 struct netdev *netdev;
2196 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
2198 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
2199 "cannot be opened (%s)",
2201 ofproto_port->name, ofproto_port->ofp_port,
2202 ofproto_port->name, ovs_strerror(error));
2206 if (ofproto_port->ofp_port == OFPP_NONE) {
2207 if (!strcmp(ofproto->name, ofproto_port->name)) {
2208 ofproto_port->ofp_port = OFPP_LOCAL;
2210 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
2211 ofproto_port->name);
2214 pp->port_no = ofproto_port->ofp_port;
2215 netdev_get_etheraddr(netdev, pp->hw_addr);
2216 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
2217 netdev_get_flags(netdev, &flags);
2218 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
2219 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
2220 netdev_get_features(netdev, &pp->curr, &pp->advertised,
2221 &pp->supported, &pp->peer);
2222 pp->curr_speed = netdev_features_to_bps(pp->curr, 0) / 1000;
2223 pp->max_speed = netdev_features_to_bps(pp->supported, 0) / 1000;
2228 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
2229 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
2232 ofport_equal(const struct ofputil_phy_port *a,
2233 const struct ofputil_phy_port *b)
2235 return (eth_addr_equals(a->hw_addr, b->hw_addr)
2236 && a->state == b->state
2237 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
2238 && a->curr == b->curr
2239 && a->advertised == b->advertised
2240 && a->supported == b->supported
2241 && a->peer == b->peer
2242 && a->curr_speed == b->curr_speed
2243 && a->max_speed == b->max_speed);
2246 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
2247 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
2248 * one with the same name or port number). */
2250 ofport_install(struct ofproto *p,
2251 struct netdev *netdev, const struct ofputil_phy_port *pp)
2253 const char *netdev_name = netdev_get_name(netdev);
2254 struct ofport *ofport;
2257 /* Create ofport. */
2258 ofport = p->ofproto_class->port_alloc();
2263 ofport->ofproto = p;
2264 ofport->netdev = netdev;
2265 ofport->change_seq = netdev_get_change_seq(netdev);
2267 ofport->ofp_port = pp->port_no;
2268 ofport->created = time_msec();
2270 /* Add port to 'p'. */
2271 hmap_insert(&p->ports, &ofport->hmap_node,
2272 hash_ofp_port(ofport->ofp_port));
2273 shash_add(&p->port_by_name, netdev_name, ofport);
2275 update_mtu(p, ofport);
2277 /* Let the ofproto_class initialize its private data. */
2278 error = p->ofproto_class->port_construct(ofport);
2282 connmgr_send_port_status(p->connmgr, NULL, pp, OFPPR_ADD);
2286 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
2287 p->name, netdev_name, ovs_strerror(error));
2289 ofport_destroy__(ofport);
2291 netdev_close(netdev);
2295 /* Removes 'ofport' from 'p' and destroys it. */
2297 ofport_remove(struct ofport *ofport)
2299 connmgr_send_port_status(ofport->ofproto->connmgr, NULL, &ofport->pp,
2301 ofport_destroy(ofport);
2304 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
2307 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
2309 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
2311 ofport_remove(port);
2315 /* Updates 'port' with new 'pp' description.
2317 * Does not handle a name or port number change. The caller must implement
2318 * such a change as a delete followed by an add. */
2320 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
2322 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
2323 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
2324 | (pp->config & OFPUTIL_PC_PORT_DOWN));
2325 port->pp.state = ((port->pp.state & ~OFPUTIL_PS_LINK_DOWN)
2326 | (pp->state & OFPUTIL_PS_LINK_DOWN));
2327 port->pp.curr = pp->curr;
2328 port->pp.advertised = pp->advertised;
2329 port->pp.supported = pp->supported;
2330 port->pp.peer = pp->peer;
2331 port->pp.curr_speed = pp->curr_speed;
2332 port->pp.max_speed = pp->max_speed;
2334 connmgr_send_port_status(port->ofproto->connmgr, NULL,
2335 &port->pp, OFPPR_MODIFY);
2338 /* Update OpenFlow 'state' in 'port' and notify controller. */
2340 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
2342 if (port->pp.state != state) {
2343 port->pp.state = state;
2344 connmgr_send_port_status(port->ofproto->connmgr, NULL,
2345 &port->pp, OFPPR_MODIFY);
2350 ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port)
2352 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
2354 if (port->ofproto->ofproto_class->set_realdev) {
2355 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
2357 if (port->ofproto->ofproto_class->set_stp_port) {
2358 port->ofproto->ofproto_class->set_stp_port(port, NULL);
2360 if (port->ofproto->ofproto_class->set_cfm) {
2361 port->ofproto->ofproto_class->set_cfm(port, NULL);
2363 if (port->ofproto->ofproto_class->bundle_remove) {
2364 port->ofproto->ofproto_class->bundle_remove(port);
2370 ofport_destroy__(struct ofport *port)
2372 struct ofproto *ofproto = port->ofproto;
2373 const char *name = netdev_get_name(port->netdev);
2375 hmap_remove(&ofproto->ports, &port->hmap_node);
2376 shash_delete(&ofproto->port_by_name,
2377 shash_find(&ofproto->port_by_name, name));
2379 netdev_close(port->netdev);
2380 ofproto->ofproto_class->port_dealloc(port);
2384 ofport_destroy(struct ofport *port)
2387 dealloc_ofp_port(port->ofproto, port->ofp_port);
2388 port->ofproto->ofproto_class->port_destruct(port);
2389 ofport_destroy__(port);
2394 ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
2396 struct ofport *port;
2398 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_ofp_port(ofp_port),
2400 if (port->ofp_port == ofp_port) {
2407 static long long int
2408 ofport_get_usage(const struct ofproto *ofproto, ofp_port_t ofp_port)
2410 struct ofport_usage *usage;
2412 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2413 &ofproto->ofport_usage) {
2414 if (usage->ofp_port == ofp_port) {
2415 return usage->last_used;
2422 ofport_set_usage(struct ofproto *ofproto, ofp_port_t ofp_port,
2423 long long int last_used)
2425 struct ofport_usage *usage;
2426 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2427 &ofproto->ofport_usage) {
2428 if (usage->ofp_port == ofp_port) {
2429 usage->last_used = last_used;
2433 ovs_assert(last_used == LLONG_MAX);
2435 usage = xmalloc(sizeof *usage);
2436 usage->ofp_port = ofp_port;
2437 usage->last_used = last_used;
2438 hmap_insert(&ofproto->ofport_usage, &usage->hmap_node,
2439 hash_ofp_port(ofp_port));
2443 ofport_remove_usage(struct ofproto *ofproto, ofp_port_t ofp_port)
2445 struct ofport_usage *usage;
2446 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2447 &ofproto->ofport_usage) {
2448 if (usage->ofp_port == ofp_port) {
2449 hmap_remove(&ofproto->ofport_usage, &usage->hmap_node);
2457 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
2459 struct ofproto *ofproto = port->ofproto;
2462 if (ofproto->ofproto_class->port_get_stats) {
2463 error = ofproto->ofproto_class->port_get_stats(port, stats);
2472 update_port(struct ofproto *ofproto, const char *name)
2474 struct ofproto_port ofproto_port;
2475 struct ofputil_phy_port pp;
2476 struct netdev *netdev;
2477 struct ofport *port;
2479 COVERAGE_INC(ofproto_update_port);
2481 /* Fetch 'name''s location and properties from the datapath. */
2482 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
2483 ? ofport_open(ofproto, &ofproto_port, &pp)
2487 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
2488 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
2489 struct netdev *old_netdev = port->netdev;
2491 /* 'name' hasn't changed location. Any properties changed? */
2492 if (!ofport_equal(&port->pp, &pp)) {
2493 ofport_modified(port, &pp);
2496 update_mtu(ofproto, port);
2498 /* Install the newly opened netdev in case it has changed.
2499 * Don't close the old netdev yet in case port_modified has to
2500 * remove a retained reference to it.*/
2501 port->netdev = netdev;
2502 port->change_seq = netdev_get_change_seq(netdev);
2504 if (port->ofproto->ofproto_class->port_modified) {
2505 port->ofproto->ofproto_class->port_modified(port);
2508 netdev_close(old_netdev);
2510 /* If 'port' is nonnull then its name differs from 'name' and thus
2511 * we should delete it. If we think there's a port named 'name'
2512 * then its port number must be wrong now so delete it too. */
2514 ofport_remove(port);
2516 ofport_remove_with_name(ofproto, name);
2517 ofport_install(ofproto, netdev, &pp);
2520 /* Any port named 'name' is gone now. */
2521 ofport_remove_with_name(ofproto, name);
2523 ofproto_port_destroy(&ofproto_port);
2527 init_ports(struct ofproto *p)
2529 struct ofproto_port_dump dump;
2530 struct ofproto_port ofproto_port;
2531 struct shash_node *node, *next;
2533 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2534 const char *name = ofproto_port.name;
2536 if (shash_find(&p->port_by_name, name)) {
2537 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2540 struct ofputil_phy_port pp;
2541 struct netdev *netdev;
2543 /* Check if an OpenFlow port number had been requested. */
2544 node = shash_find(&init_ofp_ports, name);
2546 const struct iface_hint *iface_hint = node->data;
2547 simap_put(&p->ofp_requests, name,
2548 ofp_to_u16(iface_hint->ofp_port));
2551 netdev = ofport_open(p, &ofproto_port, &pp);
2553 ofport_install(p, netdev, &pp);
2554 if (ofp_to_u16(ofproto_port.ofp_port) < p->max_ports) {
2555 p->alloc_port_no = MAX(p->alloc_port_no,
2556 ofp_to_u16(ofproto_port.ofp_port));
2562 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2563 struct iface_hint *iface_hint = node->data;
2565 if (!strcmp(iface_hint->br_name, p->name)) {
2566 free(iface_hint->br_name);
2567 free(iface_hint->br_type);
2569 shash_delete(&init_ofp_ports, node);
2576 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2577 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2579 find_min_mtu(struct ofproto *p)
2581 struct ofport *ofport;
2584 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2585 struct netdev *netdev = ofport->netdev;
2588 /* Skip any internal ports, since that's what we're trying to
2590 if (!strcmp(netdev_get_type(netdev), "internal")) {
2594 if (netdev_get_mtu(netdev, &dev_mtu)) {
2597 if (!mtu || dev_mtu < mtu) {
2602 return mtu ? mtu: ETH_PAYLOAD_MAX;
2605 /* Update MTU of all datapath devices on 'p' to the minimum of the
2606 * non-datapath ports in event of 'port' added or changed. */
2608 update_mtu(struct ofproto *p, struct ofport *port)
2610 struct ofport *ofport;
2611 struct netdev *netdev = port->netdev;
2612 int dev_mtu, old_min;
2614 if (netdev_get_mtu(netdev, &dev_mtu)) {
2618 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2619 if (dev_mtu > p->min_mtu) {
2620 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2621 dev_mtu = p->min_mtu;
2624 port->mtu = dev_mtu;
2628 /* For non-internal port find new min mtu. */
2629 old_min = p->min_mtu;
2630 port->mtu = dev_mtu;
2631 p->min_mtu = find_min_mtu(p);
2632 if (p->min_mtu == old_min) {
2636 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2637 struct netdev *netdev = ofport->netdev;
2639 if (!strcmp(netdev_get_type(netdev), "internal")) {
2640 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2641 ofport->mtu = p->min_mtu;
2648 ofproto_rule_destroy__(struct rule *rule)
2649 OVS_NO_THREAD_SAFETY_ANALYSIS
2651 cls_rule_destroy(CONST_CAST(struct cls_rule *, &rule->cr));
2652 rule_actions_destroy(rule_get_actions(rule));
2653 ovs_mutex_destroy(&rule->mutex);
2654 rule->ofproto->ofproto_class->rule_dealloc(rule);
2658 rule_destroy_cb(struct rule *rule)
2660 rule->ofproto->ofproto_class->rule_destruct(rule);
2661 ofproto_rule_destroy__(rule);
2665 ofproto_rule_ref(struct rule *rule)
2668 ovs_refcount_ref(&rule->ref_count);
2672 /* Decrements 'rule''s ref_count and schedules 'rule' to be destroyed if the
2673 * ref_count reaches 0.
2675 * Use of RCU allows short term use (between RCU quiescent periods) without
2676 * keeping a reference. A reference must be taken if the rule needs to
2677 * stay around accross the RCU quiescent periods. */
2679 ofproto_rule_unref(struct rule *rule)
2681 if (rule && ovs_refcount_unref(&rule->ref_count) == 1) {
2682 ovsrcu_postpone(rule_destroy_cb, rule);
2686 static uint32_t get_provider_meter_id(const struct ofproto *,
2687 uint32_t of_meter_id);
2689 /* Creates and returns a new 'struct rule_actions', whose actions are a copy
2690 * of from the 'ofpacts_len' bytes of 'ofpacts'. */
2691 const struct rule_actions *
2692 rule_actions_create(const struct ofproto *ofproto,
2693 const struct ofpact *ofpacts, size_t ofpacts_len)
2695 struct rule_actions *actions;
2697 actions = xmalloc(sizeof *actions + ofpacts_len);
2698 actions->ofpacts_len = ofpacts_len;
2699 actions->provider_meter_id
2700 = get_provider_meter_id(ofproto,
2701 ofpacts_get_meter(ofpacts, ofpacts_len));
2702 memcpy(actions->ofpacts, ofpacts, ofpacts_len);
2707 /* Free the actions after the RCU quiescent period is reached. */
2709 rule_actions_destroy(const struct rule_actions *actions)
2712 ovsrcu_postpone(free, CONST_CAST(struct rule_actions *, actions));
2716 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2717 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2719 ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port)
2720 OVS_REQUIRES(ofproto_mutex)
2722 if (port == OFPP_ANY) {
2725 const struct rule_actions *actions = rule_get_actions(rule);
2726 return ofpacts_output_to_port(actions->ofpacts,
2727 actions->ofpacts_len, port);
2731 /* Returns true if 'rule' has group and equals group_id. */
2733 ofproto_rule_has_out_group(const struct rule *rule, uint32_t group_id)
2734 OVS_REQUIRES(ofproto_mutex)
2736 if (group_id == OFPG_ANY) {
2739 const struct rule_actions *actions = rule_get_actions(rule);
2740 return ofpacts_output_to_group(actions->ofpacts,
2741 actions->ofpacts_len, group_id);
2745 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2746 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2748 ofoperation_has_out_port(const struct ofoperation *op, ofp_port_t out_port)
2749 OVS_REQUIRES(ofproto_mutex)
2751 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2756 case OFOPERATION_ADD:
2757 case OFOPERATION_DELETE:
2760 case OFOPERATION_MODIFY:
2761 case OFOPERATION_REPLACE:
2762 return ofpacts_output_to_port(op->actions->ofpacts,
2763 op->actions->ofpacts_len, out_port);
2770 rule_execute_destroy(struct rule_execute *e)
2772 ofproto_rule_unref(e->rule);
2773 list_remove(&e->list_node);
2777 /* Executes all "rule_execute" operations queued up in ofproto->rule_executes,
2778 * by passing them to the ofproto provider. */
2780 run_rule_executes(struct ofproto *ofproto)
2781 OVS_EXCLUDED(ofproto_mutex)
2783 struct rule_execute *e, *next;
2784 struct list executes;
2786 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2787 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2790 flow_extract(e->packet, NULL, &flow);
2791 flow.in_port.ofp_port = e->in_port;
2792 ofproto->ofproto_class->rule_execute(e->rule, &flow, e->packet);
2794 rule_execute_destroy(e);
2798 /* Destroys and discards all "rule_execute" operations queued up in
2799 * ofproto->rule_executes. */
2801 destroy_rule_executes(struct ofproto *ofproto)
2803 struct rule_execute *e, *next;
2804 struct list executes;
2806 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2807 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2808 ofpbuf_delete(e->packet);
2809 rule_execute_destroy(e);
2813 /* Returns true if 'rule' should be hidden from the controller.
2815 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2816 * (e.g. by in-band control) and are intentionally hidden from the
2819 ofproto_rule_is_hidden(const struct rule *rule)
2821 return (rule->cr.priority > UINT16_MAX);
2825 oftable_is_modifiable(const struct oftable *table,
2826 enum ofputil_flow_mod_flags flags)
2828 if (flags & OFPUTIL_FF_NO_READONLY) {
2832 return !(table->flags & OFTABLE_READONLY);
2836 rule_is_modifiable(const struct rule *rule, enum ofputil_flow_mod_flags flags)
2838 const struct oftable *rule_table;
2840 rule_table = &rule->ofproto->tables[rule->table_id];
2841 return oftable_is_modifiable(rule_table, flags);
2845 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2847 ofconn_send_reply(ofconn, make_echo_reply(oh));
2852 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2854 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2855 struct ofputil_switch_features features;
2856 struct ofport *port;
2860 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2862 ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2864 features.datapath_id = ofproto->datapath_id;
2865 features.n_buffers = pktbuf_capacity();
2866 features.n_tables = ofproto_get_n_visible_tables(ofproto);
2867 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2868 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2870 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2872 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2873 features.auxiliary_id = 0;
2874 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2876 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2877 ofputil_put_switch_features_port(&port->pp, b);
2880 ofconn_send_reply(ofconn, b);
2885 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2887 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2888 struct ofp_switch_config *osc;
2889 enum ofp_config_flags flags;
2893 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2894 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2895 flags = ofproto->frag_handling;
2896 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2897 if (oh->version < OFP13_VERSION
2898 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2899 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2901 osc->flags = htons(flags);
2902 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2903 ofconn_send_reply(ofconn, buf);
2909 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2911 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2912 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2913 uint16_t flags = ntohs(osc->flags);
2915 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2916 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
2917 enum ofp_config_flags cur = ofproto->frag_handling;
2918 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2920 ovs_assert((cur & OFPC_FRAG_MASK) == cur);
2922 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2923 ofproto->frag_handling = next;
2925 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2927 ofputil_frag_handling_to_string(next));
2931 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2932 ofconn_set_invalid_ttl_to_controller(ofconn,
2933 (oh->version < OFP13_VERSION
2934 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2936 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2941 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2942 * error message code for the caller to propagate upward. Otherwise, returns
2945 * The log message mentions 'msg_type'. */
2947 reject_slave_controller(struct ofconn *ofconn)
2949 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2950 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
2951 return OFPERR_OFPBRC_EPERM;
2957 /* Checks that the 'ofpacts_len' bytes of action in 'ofpacts' are appropriate
2960 * - If they use a meter, then 'ofproto' has that meter configured.
2962 * - If they use any groups, then 'ofproto' has that group configured.
2964 * Returns 0 if successful, otherwise an OpenFlow error. */
2966 ofproto_check_ofpacts(struct ofproto *ofproto,
2967 const struct ofpact ofpacts[], size_t ofpacts_len)
2969 const struct ofpact *a;
2972 mid = ofpacts_get_meter(ofpacts, ofpacts_len);
2973 if (mid && get_provider_meter_id(ofproto, mid) == UINT32_MAX) {
2974 return OFPERR_OFPMMFC_INVALID_METER;
2977 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2978 if (a->type == OFPACT_GROUP
2979 && !ofproto_group_exists(ofproto, ofpact_get_GROUP(a)->group_id)) {
2980 return OFPERR_OFPBAC_BAD_OUT_GROUP;
2988 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2990 struct ofproto *p = ofconn_get_ofproto(ofconn);
2991 struct ofputil_packet_out po;
2992 struct ofpbuf *payload;
2993 uint64_t ofpacts_stub[1024 / 8];
2994 struct ofpbuf ofpacts;
2998 COVERAGE_INC(ofproto_packet_out);
3000 error = reject_slave_controller(ofconn);
3005 /* Decode message. */
3006 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3007 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
3009 goto exit_free_ofpacts;
3011 if (ofp_to_u16(po.in_port) >= p->max_ports
3012 && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) {
3013 error = OFPERR_OFPBRC_BAD_PORT;
3014 goto exit_free_ofpacts;
3018 if (po.buffer_id != UINT32_MAX) {
3019 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
3020 if (error || !payload) {
3021 goto exit_free_ofpacts;
3024 /* Ensure that the L3 header is 32-bit aligned. */
3025 payload = ofpbuf_clone_data_with_headroom(po.packet, po.packet_len, 2);
3028 /* Verify actions against packet, then send packet if successful. */
3029 flow_extract(payload, NULL, &flow);
3030 flow.in_port.ofp_port = po.in_port;
3031 error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len);
3033 error = p->ofproto_class->packet_out(p, payload, &flow,
3034 po.ofpacts, po.ofpacts_len);
3036 ofpbuf_delete(payload);
3039 ofpbuf_uninit(&ofpacts);
3045 update_port_config(struct ofconn *ofconn, struct ofport *port,
3046 enum ofputil_port_config config,
3047 enum ofputil_port_config mask)
3049 enum ofputil_port_config toggle = (config ^ port->pp.config) & mask;
3051 if (toggle & OFPUTIL_PC_PORT_DOWN
3052 && (config & OFPUTIL_PC_PORT_DOWN
3053 ? netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL)
3054 : netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL))) {
3055 /* We tried to bring the port up or down, but it failed, so don't
3056 * update the "down" bit. */
3057 toggle &= ~OFPUTIL_PC_PORT_DOWN;
3061 enum ofputil_port_config old_config = port->pp.config;
3062 port->pp.config ^= toggle;
3063 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
3064 connmgr_send_port_status(port->ofproto->connmgr, ofconn, &port->pp,
3070 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3072 struct ofproto *p = ofconn_get_ofproto(ofconn);
3073 struct ofputil_port_mod pm;
3074 struct ofport *port;
3077 error = reject_slave_controller(ofconn);
3082 error = ofputil_decode_port_mod(oh, &pm);
3087 port = ofproto_get_port(p, pm.port_no);
3089 return OFPERR_OFPPMFC_BAD_PORT;
3090 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
3091 return OFPERR_OFPPMFC_BAD_HW_ADDR;
3093 update_port_config(ofconn, port, pm.config, pm.mask);
3095 netdev_set_advertisements(port->netdev, pm.advertise);
3102 handle_desc_stats_request(struct ofconn *ofconn,
3103 const struct ofp_header *request)
3105 static const char *default_mfr_desc = "Nicira, Inc.";
3106 static const char *default_hw_desc = "Open vSwitch";
3107 static const char *default_sw_desc = VERSION;
3108 static const char *default_serial_desc = "None";
3109 static const char *default_dp_desc = "None";
3111 struct ofproto *p = ofconn_get_ofproto(ofconn);
3112 struct ofp_desc_stats *ods;
3115 msg = ofpraw_alloc_stats_reply(request, 0);
3116 ods = ofpbuf_put_zeros(msg, sizeof *ods);
3117 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
3118 sizeof ods->mfr_desc);
3119 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
3120 sizeof ods->hw_desc);
3121 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
3122 sizeof ods->sw_desc);
3123 ovs_strlcpy(ods->serial_num,
3124 p->serial_desc ? p->serial_desc : default_serial_desc,
3125 sizeof ods->serial_num);
3126 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
3127 sizeof ods->dp_desc);
3128 ofconn_send_reply(ofconn, msg);
3134 handle_table_stats_request(struct ofconn *ofconn,
3135 const struct ofp_header *request)
3137 struct ofproto *p = ofconn_get_ofproto(ofconn);
3138 struct ofp12_table_stats *ots;
3143 /* Set up default values.
3145 * ofp12_table_stats is used as a generic structure as
3146 * it is able to hold all the fields for ofp10_table_stats
3147 * and ofp11_table_stats (and of course itself).
3149 ots = xcalloc(p->n_tables, sizeof *ots);
3150 for (i = 0; i < p->n_tables; i++) {
3151 ots[i].table_id = i;
3152 sprintf(ots[i].name, "table%"PRIuSIZE, i);
3153 ots[i].match = htonll(OFPXMT13_MASK);
3154 ots[i].wildcards = htonll(OFPXMT13_MASK);
3155 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
3156 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
3157 ots[i].write_setfields = htonll(OFPXMT13_MASK);
3158 ots[i].apply_setfields = htonll(OFPXMT13_MASK);
3159 ots[i].metadata_match = OVS_BE64_MAX;
3160 ots[i].metadata_write = OVS_BE64_MAX;
3161 ots[i].instructions = htonl(OFPIT11_ALL);
3162 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
3163 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
3164 fat_rwlock_rdlock(&p->tables[i].cls.rwlock);
3165 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
3166 fat_rwlock_unlock(&p->tables[i].cls.rwlock);
3169 p->ofproto_class->get_tables(p, ots);
3171 /* Post-process the tables, dropping hidden tables. */
3172 n_tables = p->n_tables;
3173 for (i = 0; i < p->n_tables; i++) {
3174 const struct oftable *table = &p->tables[i];
3176 if (table->flags & OFTABLE_HIDDEN) {
3182 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
3185 if (table->max_flows < ntohl(ots[i].max_entries)) {
3186 ots[i].max_entries = htonl(table->max_flows);
3190 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
3191 ofconn_send_reply(ofconn, msg);
3199 append_port_stat(struct ofport *port, struct list *replies)
3201 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
3203 calc_duration(port->created, time_msec(),
3204 &ops.duration_sec, &ops.duration_nsec);
3206 /* Intentionally ignore return value, since errors will set
3207 * 'stats' to all-1s, which is correct for OpenFlow, and
3208 * netdev_get_stats() will log errors. */
3209 ofproto_port_get_stats(port, &ops.stats);
3211 ofputil_append_port_stat(replies, &ops);
3215 handle_port_stats_request(struct ofconn *ofconn,
3216 const struct ofp_header *request)
3218 struct ofproto *p = ofconn_get_ofproto(ofconn);
3219 struct ofport *port;
3220 struct list replies;
3224 error = ofputil_decode_port_stats_request(request, &port_no);
3229 ofpmp_init(&replies, request);
3230 if (port_no != OFPP_ANY) {
3231 port = ofproto_get_port(p, port_no);
3233 append_port_stat(port, &replies);
3236 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3237 append_port_stat(port, &replies);
3241 ofconn_send_replies(ofconn, &replies);
3246 handle_port_desc_stats_request(struct ofconn *ofconn,
3247 const struct ofp_header *request)
3249 struct ofproto *p = ofconn_get_ofproto(ofconn);
3250 enum ofp_version version;
3251 struct ofport *port;
3252 struct list replies;
3254 ofpmp_init(&replies, request);
3256 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
3257 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3258 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
3261 ofconn_send_replies(ofconn, &replies);
3266 hash_cookie(ovs_be64 cookie)
3268 return hash_uint64((OVS_FORCE uint64_t)cookie);
3272 cookies_insert(struct ofproto *ofproto, struct rule *rule)
3273 OVS_REQUIRES(ofproto_mutex)
3275 hindex_insert(&ofproto->cookies, &rule->cookie_node,
3276 hash_cookie(rule->flow_cookie));
3280 cookies_remove(struct ofproto *ofproto, struct rule *rule)
3281 OVS_REQUIRES(ofproto_mutex)
3283 hindex_remove(&ofproto->cookies, &rule->cookie_node);
3287 ofproto_rule_change_cookie(struct ofproto *ofproto, struct rule *rule,
3288 ovs_be64 new_cookie)
3289 OVS_REQUIRES(ofproto_mutex)
3291 if (new_cookie != rule->flow_cookie) {
3292 cookies_remove(ofproto, rule);
3294 ovs_mutex_lock(&rule->mutex);
3295 rule->flow_cookie = new_cookie;
3296 ovs_mutex_unlock(&rule->mutex);
3298 cookies_insert(ofproto, rule);
3303 calc_duration(long long int start, long long int now,
3304 uint32_t *sec, uint32_t *nsec)
3306 long long int msecs = now - start;
3307 *sec = msecs / 1000;
3308 *nsec = (msecs % 1000) * (1000 * 1000);
3311 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
3312 * true if 'table_id' is OK, false otherwise. */
3314 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
3316 return table_id == OFPTT_ALL || table_id < ofproto->n_tables;
3319 static struct oftable *
3320 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
3322 struct oftable *table;
3324 for (table = &ofproto->tables[table_id];
3325 table < &ofproto->tables[ofproto->n_tables];
3327 if (!(table->flags & OFTABLE_HIDDEN)) {
3335 static struct oftable *
3336 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
3338 if (table_id == 0xff) {
3339 return next_visible_table(ofproto, 0);
3340 } else if (table_id < ofproto->n_tables) {
3341 return &ofproto->tables[table_id];
3347 static struct oftable *
3348 next_matching_table(const struct ofproto *ofproto,
3349 const struct oftable *table, uint8_t table_id)
3351 return (table_id == 0xff
3352 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
3356 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
3358 * - If TABLE_ID is 0xff, this iterates over every classifier table in
3359 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
3361 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
3362 * only once, for that table. (This can be used to access tables marked
3365 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
3366 * entered at all. (Perhaps you should have validated TABLE_ID with
3367 * check_table_id().)
3369 * All parameters are evaluated multiple times.
3371 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
3372 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
3374 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
3376 /* Initializes 'criteria' in a straightforward way based on the other
3379 * For "loose" matching, the 'priority' parameter is unimportant and may be
3382 rule_criteria_init(struct rule_criteria *criteria, uint8_t table_id,
3383 const struct match *match, unsigned int priority,
3384 ovs_be64 cookie, ovs_be64 cookie_mask,
3385 ofp_port_t out_port, uint32_t out_group)
3387 criteria->table_id = table_id;
3388 cls_rule_init(&criteria->cr, match, priority);
3389 criteria->cookie = cookie;
3390 criteria->cookie_mask = cookie_mask;
3391 criteria->out_port = out_port;
3392 criteria->out_group = out_group;
3396 rule_criteria_destroy(struct rule_criteria *criteria)
3398 cls_rule_destroy(&criteria->cr);
3402 rule_collection_init(struct rule_collection *rules)
3404 rules->rules = rules->stub;
3406 rules->capacity = ARRAY_SIZE(rules->stub);
3410 rule_collection_add(struct rule_collection *rules, struct rule *rule)
3412 if (rules->n >= rules->capacity) {
3413 size_t old_size, new_size;
3415 old_size = rules->capacity * sizeof *rules->rules;
3416 rules->capacity *= 2;
3417 new_size = rules->capacity * sizeof *rules->rules;
3419 if (rules->rules == rules->stub) {
3420 rules->rules = xmalloc(new_size);
3421 memcpy(rules->rules, rules->stub, old_size);
3423 rules->rules = xrealloc(rules->rules, new_size);
3427 rules->rules[rules->n++] = rule;
3431 rule_collection_ref(struct rule_collection *rules)
3432 OVS_REQUIRES(ofproto_mutex)
3436 for (i = 0; i < rules->n; i++) {
3437 ofproto_rule_ref(rules->rules[i]);
3442 rule_collection_unref(struct rule_collection *rules)
3446 for (i = 0; i < rules->n; i++) {
3447 ofproto_rule_unref(rules->rules[i]);
3452 rule_collection_destroy(struct rule_collection *rules)
3454 if (rules->rules != rules->stub) {
3460 collect_rule(struct rule *rule, const struct rule_criteria *c,
3461 struct rule_collection *rules)
3462 OVS_REQUIRES(ofproto_mutex)
3464 /* We ordinarily want to skip hidden rules, but there has to be a way for
3465 * code internal to OVS to modify and delete them, so if the criteria
3466 * specify a priority that can only be for a hidden flow, then allow hidden
3467 * rules to be selected. (This doesn't allow OpenFlow clients to meddle
3468 * with hidden flows because OpenFlow uses only a 16-bit field to specify
3470 if (ofproto_rule_is_hidden(rule) && c->cr.priority <= UINT16_MAX) {
3472 } else if (rule->pending) {
3473 return OFPROTO_POSTPONE;
3475 if ((c->table_id == rule->table_id || c->table_id == 0xff)
3476 && ofproto_rule_has_out_port(rule, c->out_port)
3477 && ofproto_rule_has_out_group(rule, c->out_group)
3478 && !((rule->flow_cookie ^ c->cookie) & c->cookie_mask)) {
3479 rule_collection_add(rules, rule);
3485 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3486 * on classifiers rules are done in the "loose" way required for OpenFlow
3487 * OFPFC_MODIFY and OFPFC_DELETE requests. Puts the selected rules on list
3490 * Hidden rules are always omitted.
3492 * Returns 0 on success, otherwise an OpenFlow error code. */
3494 collect_rules_loose(struct ofproto *ofproto,
3495 const struct rule_criteria *criteria,
3496 struct rule_collection *rules)
3497 OVS_REQUIRES(ofproto_mutex)
3499 struct oftable *table;
3500 enum ofperr error = 0;
3502 rule_collection_init(rules);
3504 if (!check_table_id(ofproto, criteria->table_id)) {
3505 error = OFPERR_OFPBRC_BAD_TABLE_ID;
3509 if (criteria->cookie_mask == OVS_BE64_MAX) {
3512 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
3513 hash_cookie(criteria->cookie),
3514 &ofproto->cookies) {
3515 if (cls_rule_is_loose_match(&rule->cr, &criteria->cr.match)) {
3516 error = collect_rule(rule, criteria, rules);
3523 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
3524 struct cls_cursor cursor;
3527 fat_rwlock_rdlock(&table->cls.rwlock);
3528 cls_cursor_init(&cursor, &table->cls, &criteria->cr);
3529 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3530 error = collect_rule(rule, criteria, rules);
3535 fat_rwlock_unlock(&table->cls.rwlock);
3541 rule_collection_destroy(rules);
3546 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3547 * on classifiers rules are done in the "strict" way required for OpenFlow
3548 * OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests. Puts the selected
3549 * rules on list 'rules'.
3551 * Hidden rules are always omitted.
3553 * Returns 0 on success, otherwise an OpenFlow error code. */
3555 collect_rules_strict(struct ofproto *ofproto,
3556 const struct rule_criteria *criteria,
3557 struct rule_collection *rules)
3558 OVS_REQUIRES(ofproto_mutex)
3560 struct oftable *table;
3563 rule_collection_init(rules);
3565 if (!check_table_id(ofproto, criteria->table_id)) {
3566 error = OFPERR_OFPBRC_BAD_TABLE_ID;
3570 if (criteria->cookie_mask == OVS_BE64_MAX) {
3573 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
3574 hash_cookie(criteria->cookie),
3575 &ofproto->cookies) {
3576 if (cls_rule_equal(&rule->cr, &criteria->cr)) {
3577 error = collect_rule(rule, criteria, rules);
3584 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
3587 fat_rwlock_rdlock(&table->cls.rwlock);
3588 rule = rule_from_cls_rule(classifier_find_rule_exactly(
3589 &table->cls, &criteria->cr));
3590 fat_rwlock_unlock(&table->cls.rwlock);
3592 error = collect_rule(rule, criteria, rules);
3602 rule_collection_destroy(rules);
3607 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
3608 * forced into the range of a uint16_t. */
3610 age_secs(long long int age_ms)
3612 return (age_ms < 0 ? 0
3613 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
3614 : (unsigned int) age_ms / 1000);
3618 handle_flow_stats_request(struct ofconn *ofconn,
3619 const struct ofp_header *request)
3620 OVS_EXCLUDED(ofproto_mutex)
3622 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3623 struct ofputil_flow_stats_request fsr;
3624 struct rule_criteria criteria;
3625 struct rule_collection rules;
3626 struct list replies;
3630 error = ofputil_decode_flow_stats_request(&fsr, request);
3635 rule_criteria_init(&criteria, fsr.table_id, &fsr.match, 0, fsr.cookie,
3636 fsr.cookie_mask, fsr.out_port, fsr.out_group);
3638 ovs_mutex_lock(&ofproto_mutex);
3639 error = collect_rules_loose(ofproto, &criteria, &rules);
3640 rule_criteria_destroy(&criteria);
3642 rule_collection_ref(&rules);
3644 ovs_mutex_unlock(&ofproto_mutex);
3650 ofpmp_init(&replies, request);
3651 for (i = 0; i < rules.n; i++) {
3652 struct rule *rule = rules.rules[i];
3653 long long int now = time_msec();
3654 struct ofputil_flow_stats fs;
3655 long long int created, used, modified;
3656 const struct rule_actions *actions;
3657 enum ofputil_flow_mod_flags flags;
3659 ovs_mutex_lock(&rule->mutex);
3660 fs.cookie = rule->flow_cookie;
3661 fs.idle_timeout = rule->idle_timeout;
3662 fs.hard_timeout = rule->hard_timeout;
3663 created = rule->created;
3664 modified = rule->modified;
3665 actions = rule_get_actions(rule);
3666 flags = rule->flags;
3667 ovs_mutex_unlock(&rule->mutex);
3669 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
3670 &fs.byte_count, &used);
3672 minimatch_expand(&rule->cr.match, &fs.match);
3673 fs.table_id = rule->table_id;
3674 calc_duration(created, now, &fs.duration_sec, &fs.duration_nsec);
3675 fs.priority = rule->cr.priority;
3676 fs.idle_age = age_secs(now - used);
3677 fs.hard_age = age_secs(now - modified);
3678 fs.ofpacts = actions->ofpacts;
3679 fs.ofpacts_len = actions->ofpacts_len;
3682 ofputil_append_flow_stats_reply(&fs, &replies);
3685 rule_collection_unref(&rules);
3686 rule_collection_destroy(&rules);
3688 ofconn_send_replies(ofconn, &replies);
3694 flow_stats_ds(struct rule *rule, struct ds *results)
3696 uint64_t packet_count, byte_count;
3697 const struct rule_actions *actions;
3698 long long int created, used;
3700 rule->ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
3701 &byte_count, &used);
3703 ovs_mutex_lock(&rule->mutex);
3704 actions = rule_get_actions(rule);
3705 created = rule->created;
3706 ovs_mutex_unlock(&rule->mutex);
3708 if (rule->table_id != 0) {
3709 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
3711 ds_put_format(results, "duration=%llds, ", (time_msec() - created) / 1000);
3712 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3713 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3714 cls_rule_format(&rule->cr, results);
3715 ds_put_char(results, ',');
3717 ds_put_cstr(results, "actions=");
3718 ofpacts_format(actions->ofpacts, actions->ofpacts_len, results);
3720 ds_put_cstr(results, "\n");
3723 /* Adds a pretty-printed description of all flows to 'results', including
3724 * hidden flows (e.g., set up by in-band control). */
3726 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3728 struct oftable *table;
3730 OFPROTO_FOR_EACH_TABLE (table, p) {
3731 struct cls_cursor cursor;
3734 fat_rwlock_rdlock(&table->cls.rwlock);
3735 cls_cursor_init(&cursor, &table->cls, NULL);
3736 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3737 flow_stats_ds(rule, results);
3739 fat_rwlock_unlock(&table->cls.rwlock);
3743 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
3744 * '*engine_type' and '*engine_id', respectively. */
3746 ofproto_get_netflow_ids(const struct ofproto *ofproto,
3747 uint8_t *engine_type, uint8_t *engine_id)
3749 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
3752 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'.
3753 * Returns 0 if the port's CFM status was successfully stored into
3754 * '*status'. Returns positive errno if the port did not have CFM
3755 * configured. Returns negative number if there is no status change
3756 * since last update.
3758 * The caller must provide and own '*status', and must free 'status->rmps'.
3759 * '*status' is indeterminate if the return value is non-zero. */
3761 ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port,
3762 struct ofproto_cfm_status *status)
3764 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
3765 return (ofport && ofproto->ofproto_class->get_cfm_status
3766 ? ofproto->ofproto_class->get_cfm_status(ofport, status)
3771 handle_aggregate_stats_request(struct ofconn *ofconn,
3772 const struct ofp_header *oh)
3773 OVS_EXCLUDED(ofproto_mutex)
3775 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3776 struct ofputil_flow_stats_request request;
3777 struct ofputil_aggregate_stats stats;
3778 bool unknown_packets, unknown_bytes;
3779 struct rule_criteria criteria;
3780 struct rule_collection rules;
3781 struct ofpbuf *reply;
3785 error = ofputil_decode_flow_stats_request(&request, oh);
3790 rule_criteria_init(&criteria, request.table_id, &request.match, 0,
3791 request.cookie, request.cookie_mask,
3792 request.out_port, request.out_group);
3794 ovs_mutex_lock(&ofproto_mutex);
3795 error = collect_rules_loose(ofproto, &criteria, &rules);
3796 rule_criteria_destroy(&criteria);
3798 rule_collection_ref(&rules);
3800 ovs_mutex_unlock(&ofproto_mutex);
3806 memset(&stats, 0, sizeof stats);
3807 unknown_packets = unknown_bytes = false;
3808 for (i = 0; i < rules.n; i++) {
3809 struct rule *rule = rules.rules[i];
3810 uint64_t packet_count;
3811 uint64_t byte_count;
3814 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
3815 &byte_count, &used);
3817 if (packet_count == UINT64_MAX) {
3818 unknown_packets = true;
3820 stats.packet_count += packet_count;
3823 if (byte_count == UINT64_MAX) {
3824 unknown_bytes = true;
3826 stats.byte_count += byte_count;
3831 if (unknown_packets) {
3832 stats.packet_count = UINT64_MAX;
3834 if (unknown_bytes) {
3835 stats.byte_count = UINT64_MAX;
3838 rule_collection_unref(&rules);
3839 rule_collection_destroy(&rules);
3841 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
3842 ofconn_send_reply(ofconn, reply);
3847 struct queue_stats_cbdata {
3848 struct ofport *ofport;
3849 struct list replies;
3854 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3855 const struct netdev_queue_stats *stats)
3857 struct ofputil_queue_stats oqs;
3859 oqs.port_no = cbdata->ofport->pp.port_no;
3860 oqs.queue_id = queue_id;
3861 oqs.tx_bytes = stats->tx_bytes;
3862 oqs.tx_packets = stats->tx_packets;
3863 oqs.tx_errors = stats->tx_errors;
3864 if (stats->created != LLONG_MIN) {
3865 calc_duration(stats->created, cbdata->now,
3866 &oqs.duration_sec, &oqs.duration_nsec);
3868 oqs.duration_sec = oqs.duration_nsec = UINT32_MAX;
3870 ofputil_append_queue_stat(&cbdata->replies, &oqs);
3874 handle_queue_stats_dump_cb(uint32_t queue_id,
3875 struct netdev_queue_stats *stats,
3878 struct queue_stats_cbdata *cbdata = cbdata_;
3880 put_queue_stats(cbdata, queue_id, stats);
3884 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3885 struct queue_stats_cbdata *cbdata)
3887 cbdata->ofport = port;
3888 if (queue_id == OFPQ_ALL) {
3889 netdev_dump_queue_stats(port->netdev,
3890 handle_queue_stats_dump_cb, cbdata);
3892 struct netdev_queue_stats stats;
3894 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3895 put_queue_stats(cbdata, queue_id, &stats);
3897 return OFPERR_OFPQOFC_BAD_QUEUE;
3904 handle_queue_stats_request(struct ofconn *ofconn,
3905 const struct ofp_header *rq)
3907 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3908 struct queue_stats_cbdata cbdata;
3909 struct ofport *port;
3911 struct ofputil_queue_stats_request oqsr;
3913 COVERAGE_INC(ofproto_queue_req);
3915 ofpmp_init(&cbdata.replies, rq);
3916 cbdata.now = time_msec();
3918 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3923 if (oqsr.port_no == OFPP_ANY) {
3924 error = OFPERR_OFPQOFC_BAD_QUEUE;
3925 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3926 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3931 port = ofproto_get_port(ofproto, oqsr.port_no);
3933 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3934 : OFPERR_OFPQOFC_BAD_PORT);
3937 ofconn_send_replies(ofconn, &cbdata.replies);
3939 ofpbuf_list_delete(&cbdata.replies);
3946 is_flow_deletion_pending(const struct ofproto *ofproto,
3947 const struct cls_rule *cls_rule,
3949 OVS_REQUIRES(ofproto_mutex)
3951 if (!hmap_is_empty(&ofproto->deletions)) {
3952 struct ofoperation *op;
3954 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3955 cls_rule_hash(cls_rule, table_id),
3956 &ofproto->deletions) {
3957 if (op->rule->table_id == table_id
3958 && cls_rule_equal(cls_rule, &op->rule->cr)) {
3968 should_evict_a_rule(struct oftable *table, unsigned int extra_space)
3969 OVS_REQUIRES(ofproto_mutex)
3970 OVS_NO_THREAD_SAFETY_ANALYSIS
3972 return classifier_count(&table->cls) + extra_space > table->max_flows;
3976 evict_rules_from_table(struct ofproto *ofproto, struct oftable *table,
3977 unsigned int extra_space)
3978 OVS_REQUIRES(ofproto_mutex)
3980 while (should_evict_a_rule(table, extra_space)) {
3983 if (!choose_rule_to_evict(table, &rule)) {
3984 return OFPERR_OFPFMFC_TABLE_FULL;
3985 } else if (rule->pending) {
3986 return OFPROTO_POSTPONE;
3988 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
3989 delete_flow__(rule, group, OFPRR_EVICTION);
3990 ofopgroup_submit(group);
3997 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3998 * in which no matching flow already exists in the flow table.
4000 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
4001 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
4002 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
4003 * initiated now but may be retried later.
4005 * The caller retains ownership of 'fm->ofpacts'.
4007 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4010 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
4011 struct ofputil_flow_mod *fm, const struct ofp_header *request)
4012 OVS_REQUIRES(ofproto_mutex)
4014 struct oftable *table;
4020 if (!check_table_id(ofproto, fm->table_id)) {
4021 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4026 if (fm->table_id == 0xff) {
4027 if (ofproto->ofproto_class->rule_choose_table) {
4028 error = ofproto->ofproto_class->rule_choose_table(ofproto,
4034 ovs_assert(table_id < ofproto->n_tables);
4038 } else if (fm->table_id < ofproto->n_tables) {
4039 table_id = fm->table_id;
4041 return OFPERR_OFPBRC_BAD_TABLE_ID;
4044 table = &ofproto->tables[table_id];
4046 if (!oftable_is_modifiable(table, fm->flags)) {
4047 return OFPERR_OFPBRC_EPERM;
4050 if (!(fm->flags & OFPUTIL_FF_HIDDEN_FIELDS)) {
4051 if (!match_has_default_hidden_fields(&fm->match)) {
4052 VLOG_WARN_RL(&rl, "%s: (add_flow) only internal flows can set "
4053 "non-default values to hidden fields", ofproto->name);
4054 return OFPERR_OFPBRC_EPERM;
4058 cls_rule_init(&cr, &fm->match, fm->priority);
4060 /* Transform "add" into "modify" if there's an existing identical flow. */
4061 fat_rwlock_rdlock(&table->cls.rwlock);
4062 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr));
4063 fat_rwlock_unlock(&table->cls.rwlock);
4065 cls_rule_destroy(&cr);
4066 if (!rule_is_modifiable(rule, fm->flags)) {
4067 return OFPERR_OFPBRC_EPERM;
4068 } else if (rule->pending) {
4069 return OFPROTO_POSTPONE;
4071 struct rule_collection rules;
4073 rule_collection_init(&rules);
4074 rule_collection_add(&rules, rule);
4075 fm->modify_cookie = true;
4076 error = modify_flows__(ofproto, ofconn, fm, request, &rules);
4077 rule_collection_destroy(&rules);
4083 /* Serialize against pending deletion. */
4084 if (is_flow_deletion_pending(ofproto, &cr, table_id)) {
4085 cls_rule_destroy(&cr);
4086 return OFPROTO_POSTPONE;
4089 /* Check for overlap, if requested. */
4090 if (fm->flags & OFPUTIL_FF_CHECK_OVERLAP) {
4093 fat_rwlock_rdlock(&table->cls.rwlock);
4094 overlaps = classifier_rule_overlaps(&table->cls, &cr);
4095 fat_rwlock_unlock(&table->cls.rwlock);
4098 cls_rule_destroy(&cr);
4099 return OFPERR_OFPFMFC_OVERLAP;
4103 /* If necessary, evict an existing rule to clear out space. */
4104 error = evict_rules_from_table(ofproto, table, 1);
4106 cls_rule_destroy(&cr);
4110 /* Allocate new rule. */
4111 rule = ofproto->ofproto_class->rule_alloc();
4113 cls_rule_destroy(&cr);
4114 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
4115 ofproto->name, ovs_strerror(error));
4119 /* Initialize base state. */
4120 *CONST_CAST(struct ofproto **, &rule->ofproto) = ofproto;
4121 cls_rule_move(CONST_CAST(struct cls_rule *, &rule->cr), &cr);
4122 ovs_refcount_init(&rule->ref_count);
4123 rule->pending = NULL;
4124 rule->flow_cookie = fm->new_cookie;
4125 rule->created = rule->modified = time_msec();
4127 ovs_mutex_init(&rule->mutex);
4128 ovs_mutex_lock(&rule->mutex);
4129 rule->idle_timeout = fm->idle_timeout;
4130 rule->hard_timeout = fm->hard_timeout;
4131 ovs_mutex_unlock(&rule->mutex);
4133 *CONST_CAST(uint8_t *, &rule->table_id) = table - ofproto->tables;
4134 rule->flags = fm->flags & OFPUTIL_FF_STATE;
4135 ovsrcu_set(&rule->actions,
4136 rule_actions_create(ofproto, fm->ofpacts, fm->ofpacts_len));
4137 list_init(&rule->meter_list_node);
4138 rule->eviction_group = NULL;
4139 list_init(&rule->expirable);
4140 rule->monitor_flags = 0;
4141 rule->add_seqno = 0;
4142 rule->modify_seqno = 0;
4144 /* Construct rule, initializing derived state. */
4145 error = ofproto->ofproto_class->rule_construct(rule);
4147 ofproto_rule_destroy__(rule);
4152 do_add_flow(ofproto, ofconn, request, fm->buffer_id, rule);
4158 do_add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
4159 const struct ofp_header *request, uint32_t buffer_id,
4161 OVS_REQUIRES(ofproto_mutex)
4163 struct ofopgroup *group;
4165 oftable_insert_rule(rule);
4167 group = ofopgroup_create(ofproto, ofconn, request, buffer_id);
4168 ofoperation_create(group, rule, OFOPERATION_ADD, 0);
4169 ofproto->ofproto_class->rule_insert(rule);
4170 ofopgroup_submit(group);
4173 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
4175 /* Modifies the rules listed in 'rules', changing their actions to match those
4178 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4181 * Returns 0 on success, otherwise an OpenFlow error code. */
4183 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
4184 struct ofputil_flow_mod *fm, const struct ofp_header *request,
4185 const struct rule_collection *rules)
4186 OVS_REQUIRES(ofproto_mutex)
4188 enum ofoperation_type type;
4189 struct ofopgroup *group;
4193 type = fm->command == OFPFC_ADD ? OFOPERATION_REPLACE : OFOPERATION_MODIFY;
4194 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
4195 error = OFPERR_OFPBRC_EPERM;
4196 for (i = 0; i < rules->n; i++) {
4197 struct rule *rule = rules->rules[i];
4198 const struct rule_actions *actions;
4199 struct ofoperation *op;
4200 bool actions_changed;
4201 bool reset_counters;
4203 /* FIXME: Implement OFPFUTIL_FF_RESET_COUNTS */
4205 if (rule_is_modifiable(rule, fm->flags)) {
4206 /* At least one rule is modifiable, don't report EPERM error. */
4212 actions = rule_get_actions(rule);
4213 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
4215 actions->ofpacts_len);
4217 op = ofoperation_create(group, rule, type, 0);
4219 if (fm->modify_cookie && fm->new_cookie != OVS_BE64_MAX) {
4220 ofproto_rule_change_cookie(ofproto, rule, fm->new_cookie);
4222 if (type == OFOPERATION_REPLACE) {
4223 ovs_mutex_lock(&rule->mutex);
4224 rule->idle_timeout = fm->idle_timeout;
4225 rule->hard_timeout = fm->hard_timeout;
4226 ovs_mutex_unlock(&rule->mutex);
4228 rule->flags = fm->flags & OFPUTIL_FF_STATE;
4229 if (fm->idle_timeout || fm->hard_timeout) {
4230 if (!rule->eviction_group) {
4231 eviction_group_add_rule(rule);
4234 eviction_group_remove_rule(rule);
4238 reset_counters = (fm->flags & OFPUTIL_FF_RESET_COUNTS) != 0;
4239 if (actions_changed || reset_counters) {
4240 const struct rule_actions *new_actions;
4242 op->actions = rule_get_actions(rule);
4243 new_actions = rule_actions_create(ofproto,
4244 fm->ofpacts, fm->ofpacts_len);
4246 ovsrcu_set(&rule->actions, new_actions);
4248 rule->ofproto->ofproto_class->rule_modify_actions(rule,
4251 ofoperation_complete(op, 0);
4254 ofopgroup_submit(group);
4260 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
4261 struct ofputil_flow_mod *fm, const struct ofp_header *request)
4262 OVS_REQUIRES(ofproto_mutex)
4264 if (fm->cookie_mask != htonll(0) || fm->new_cookie == OVS_BE64_MAX) {
4267 return add_flow(ofproto, ofconn, fm, request);
4270 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
4273 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4276 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
4277 struct ofputil_flow_mod *fm,
4278 const struct ofp_header *request)
4279 OVS_REQUIRES(ofproto_mutex)
4281 struct rule_criteria criteria;
4282 struct rule_collection rules;
4285 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0,
4286 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY);
4287 error = collect_rules_loose(ofproto, &criteria, &rules);
4288 rule_criteria_destroy(&criteria);
4291 error = (rules.n > 0
4292 ? modify_flows__(ofproto, ofconn, fm, request, &rules)
4293 : modify_flows_add(ofproto, ofconn, fm, request));
4296 rule_collection_destroy(&rules);
4301 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4304 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4307 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
4308 struct ofputil_flow_mod *fm,
4309 const struct ofp_header *request)
4310 OVS_REQUIRES(ofproto_mutex)
4312 struct rule_criteria criteria;
4313 struct rule_collection rules;
4316 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
4317 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY);
4318 error = collect_rules_strict(ofproto, &criteria, &rules);
4319 rule_criteria_destroy(&criteria);
4323 error = modify_flows_add(ofproto, ofconn, fm, request);
4324 } else if (rules.n == 1) {
4325 error = modify_flows__(ofproto, ofconn, fm, request, &rules);
4329 rule_collection_destroy(&rules);
4334 /* OFPFC_DELETE implementation. */
4337 delete_flow__(struct rule *rule, struct ofopgroup *group,
4338 enum ofp_flow_removed_reason reason)
4339 OVS_REQUIRES(ofproto_mutex)
4341 struct ofproto *ofproto = rule->ofproto;
4343 ofproto_rule_send_removed(rule, reason);
4345 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
4346 oftable_remove_rule(rule);
4347 ofproto->ofproto_class->rule_delete(rule);
4350 /* Deletes the rules listed in 'rules'.
4352 * Returns 0 on success, otherwise an OpenFlow error code. */
4354 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
4355 const struct ofp_header *request,
4356 const struct rule_collection *rules,
4357 enum ofp_flow_removed_reason reason)
4358 OVS_REQUIRES(ofproto_mutex)
4360 struct ofopgroup *group;
4363 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
4364 for (i = 0; i < rules->n; i++) {
4365 delete_flow__(rules->rules[i], group, reason);
4367 ofopgroup_submit(group);
4372 /* Implements OFPFC_DELETE. */
4374 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
4375 const struct ofputil_flow_mod *fm,
4376 const struct ofp_header *request)
4377 OVS_REQUIRES(ofproto_mutex)
4379 struct rule_criteria criteria;
4380 struct rule_collection rules;
4383 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0,
4384 fm->cookie, fm->cookie_mask,
4385 fm->out_port, fm->out_group);
4386 error = collect_rules_loose(ofproto, &criteria, &rules);
4387 rule_criteria_destroy(&criteria);
4389 if (!error && rules.n > 0) {
4390 error = delete_flows__(ofproto, ofconn, request, &rules, OFPRR_DELETE);
4392 rule_collection_destroy(&rules);
4397 /* Implements OFPFC_DELETE_STRICT. */
4399 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
4400 const struct ofputil_flow_mod *fm,
4401 const struct ofp_header *request)
4402 OVS_REQUIRES(ofproto_mutex)
4404 struct rule_criteria criteria;
4405 struct rule_collection rules;
4408 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
4409 fm->cookie, fm->cookie_mask,
4410 fm->out_port, fm->out_group);
4411 error = collect_rules_strict(ofproto, &criteria, &rules);
4412 rule_criteria_destroy(&criteria);
4414 if (!error && rules.n > 0) {
4415 error = delete_flows__(ofproto, ofconn, request, &rules, OFPRR_DELETE);
4417 rule_collection_destroy(&rules);
4423 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
4424 OVS_REQUIRES(ofproto_mutex)
4426 struct ofputil_flow_removed fr;
4429 if (ofproto_rule_is_hidden(rule) ||
4430 !(rule->flags & OFPUTIL_FF_SEND_FLOW_REM)) {
4434 minimatch_expand(&rule->cr.match, &fr.match);
4435 fr.priority = rule->cr.priority;
4436 fr.cookie = rule->flow_cookie;
4438 fr.table_id = rule->table_id;
4439 calc_duration(rule->created, time_msec(),
4440 &fr.duration_sec, &fr.duration_nsec);
4441 ovs_mutex_lock(&rule->mutex);
4442 fr.idle_timeout = rule->idle_timeout;
4443 fr.hard_timeout = rule->hard_timeout;
4444 ovs_mutex_unlock(&rule->mutex);
4445 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
4446 &fr.byte_count, &used);
4448 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
4451 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
4452 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
4455 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
4458 * ofproto implementation ->run() functions should use this function to expire
4459 * OpenFlow flows. */
4461 ofproto_rule_expire(struct rule *rule, uint8_t reason)
4462 OVS_REQUIRES(ofproto_mutex)
4464 struct ofproto *ofproto = rule->ofproto;
4466 ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT
4467 || reason == OFPRR_DELETE || reason == OFPRR_GROUP_DELETE);
4469 ofproto_rule_delete__(ofproto, rule, reason);
4472 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
4473 * means "infinite". */
4475 reduce_timeout(uint16_t max, uint16_t *timeout)
4477 if (max && (!*timeout || *timeout > max)) {
4482 /* If 'idle_timeout' is nonzero, and 'rule' has no idle timeout or an idle
4483 * timeout greater than 'idle_timeout', lowers 'rule''s idle timeout to
4484 * 'idle_timeout' seconds. Similarly for 'hard_timeout'.
4486 * Suitable for implementing OFPACT_FIN_TIMEOUT. */
4488 ofproto_rule_reduce_timeouts(struct rule *rule,
4489 uint16_t idle_timeout, uint16_t hard_timeout)
4490 OVS_EXCLUDED(ofproto_mutex, rule->mutex)
4492 if (!idle_timeout && !hard_timeout) {
4496 ovs_mutex_lock(&ofproto_mutex);
4497 if (list_is_empty(&rule->expirable)) {
4498 list_insert(&rule->ofproto->expirable, &rule->expirable);
4500 ovs_mutex_unlock(&ofproto_mutex);
4502 ovs_mutex_lock(&rule->mutex);
4503 reduce_timeout(idle_timeout, &rule->idle_timeout);
4504 reduce_timeout(hard_timeout, &rule->hard_timeout);
4505 ovs_mutex_unlock(&rule->mutex);
4509 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4510 OVS_EXCLUDED(ofproto_mutex)
4512 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4513 struct ofputil_flow_mod fm;
4514 uint64_t ofpacts_stub[1024 / 8];
4515 struct ofpbuf ofpacts;
4519 error = reject_slave_controller(ofconn);
4524 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
4525 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
4527 u16_to_ofp(ofproto->max_ports),
4530 error = ofproto_check_ofpacts(ofproto, fm.ofpacts, fm.ofpacts_len);
4533 error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
4536 goto exit_free_ofpacts;
4539 /* Record the operation for logging a summary report. */
4540 switch (fm.command) {
4546 case OFPFC_MODIFY_STRICT:
4547 ofproto->n_modify++;
4551 case OFPFC_DELETE_STRICT:
4552 ofproto->n_delete++;
4557 if (ofproto->next_op_report == LLONG_MAX) {
4558 ofproto->first_op = now;
4559 ofproto->next_op_report = MAX(now + 10 * 1000,
4560 ofproto->op_backoff);
4561 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
4563 ofproto->last_op = now;
4566 ofpbuf_uninit(&ofpacts);
4572 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
4573 struct ofputil_flow_mod *fm, const struct ofp_header *oh)
4574 OVS_EXCLUDED(ofproto_mutex)
4578 ovs_mutex_lock(&ofproto_mutex);
4579 if (ofproto->n_pending < 50) {
4580 switch (fm->command) {
4582 error = add_flow(ofproto, ofconn, fm, oh);
4586 error = modify_flows_loose(ofproto, ofconn, fm, oh);
4589 case OFPFC_MODIFY_STRICT:
4590 error = modify_flow_strict(ofproto, ofconn, fm, oh);
4594 error = delete_flows_loose(ofproto, ofconn, fm, oh);
4597 case OFPFC_DELETE_STRICT:
4598 error = delete_flow_strict(ofproto, ofconn, fm, oh);
4602 if (fm->command > 0xff) {
4603 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
4604 "flow_mod_table_id extension is not enabled",
4607 error = OFPERR_OFPFMFC_BAD_COMMAND;
4611 ovs_assert(!list_is_empty(&ofproto->pending));
4612 error = OFPROTO_POSTPONE;
4614 ovs_mutex_unlock(&ofproto_mutex);
4616 run_rule_executes(ofproto);
4621 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4623 struct ofputil_role_request request;
4624 struct ofputil_role_request reply;
4628 error = ofputil_decode_role_message(oh, &request);
4633 if (request.role != OFPCR12_ROLE_NOCHANGE) {
4634 if (ofconn_get_role(ofconn) != request.role
4635 && ofconn_has_pending_opgroups(ofconn)) {
4636 return OFPROTO_POSTPONE;
4639 if (request.have_generation_id
4640 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
4641 return OFPERR_OFPRRFC_STALE;
4644 ofconn_set_role(ofconn, request.role);
4647 reply.role = ofconn_get_role(ofconn);
4648 reply.have_generation_id = ofconn_get_master_election_id(
4649 ofconn, &reply.generation_id);
4650 buf = ofputil_encode_role_reply(oh, &reply);
4651 ofconn_send_reply(ofconn, buf);
4657 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
4658 const struct ofp_header *oh)
4660 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
4661 enum ofputil_protocol cur, next;
4663 cur = ofconn_get_protocol(ofconn);
4664 next = ofputil_protocol_set_tid(cur, msg->set != 0);
4665 ofconn_set_protocol(ofconn, next);
4671 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4673 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
4674 enum ofputil_protocol cur, next;
4675 enum ofputil_protocol next_base;
4677 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
4679 return OFPERR_OFPBRC_EPERM;
4682 cur = ofconn_get_protocol(ofconn);
4683 next = ofputil_protocol_set_base(cur, next_base);
4684 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
4685 /* Avoid sending async messages in surprising protocol. */
4686 return OFPROTO_POSTPONE;
4689 ofconn_set_protocol(ofconn, next);
4694 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
4695 const struct ofp_header *oh)
4697 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
4700 format = ntohl(msg->format);
4701 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
4702 return OFPERR_OFPBRC_EPERM;
4705 if (format != ofconn_get_packet_in_format(ofconn)
4706 && ofconn_has_pending_opgroups(ofconn)) {
4707 /* Avoid sending async message in surprsing packet in format. */
4708 return OFPROTO_POSTPONE;
4711 ofconn_set_packet_in_format(ofconn, format);
4716 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
4718 const struct nx_async_config *msg = ofpmsg_body(oh);
4719 uint32_t master[OAM_N_TYPES];
4720 uint32_t slave[OAM_N_TYPES];
4722 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
4723 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
4724 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
4726 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
4727 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
4728 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
4730 ofconn_set_async_config(ofconn, master, slave);
4731 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
4732 !ofconn_get_miss_send_len(ofconn)) {
4733 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
4740 handle_nxt_get_async_request(struct ofconn *ofconn, const struct ofp_header *oh)
4743 uint32_t master[OAM_N_TYPES];
4744 uint32_t slave[OAM_N_TYPES];
4745 struct nx_async_config *msg;
4747 ofconn_get_async_config(ofconn, master, slave);
4748 buf = ofpraw_alloc_reply(OFPRAW_OFPT13_GET_ASYNC_REPLY, oh, 0);
4749 msg = ofpbuf_put_zeros(buf, sizeof *msg);
4751 msg->packet_in_mask[0] = htonl(master[OAM_PACKET_IN]);
4752 msg->port_status_mask[0] = htonl(master[OAM_PORT_STATUS]);
4753 msg->flow_removed_mask[0] = htonl(master[OAM_FLOW_REMOVED]);
4755 msg->packet_in_mask[1] = htonl(slave[OAM_PACKET_IN]);
4756 msg->port_status_mask[1] = htonl(slave[OAM_PORT_STATUS]);
4757 msg->flow_removed_mask[1] = htonl(slave[OAM_FLOW_REMOVED]);
4759 ofconn_send_reply(ofconn, buf);
4765 handle_nxt_set_controller_id(struct ofconn *ofconn,
4766 const struct ofp_header *oh)
4768 const struct nx_controller_id *nci = ofpmsg_body(oh);
4770 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
4771 return OFPERR_NXBRC_MUST_BE_ZERO;
4774 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
4779 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4783 if (ofconn_has_pending_opgroups(ofconn)) {
4784 return OFPROTO_POSTPONE;
4787 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
4788 ? OFPRAW_OFPT10_BARRIER_REPLY
4789 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
4790 ofconn_send_reply(ofconn, buf);
4795 ofproto_compose_flow_refresh_update(const struct rule *rule,
4796 enum nx_flow_monitor_flags flags,
4798 OVS_REQUIRES(ofproto_mutex)
4800 struct ofoperation *op = rule->pending;
4801 const struct rule_actions *actions;
4802 struct ofputil_flow_update fu;
4805 if (op && op->type == OFOPERATION_ADD) {
4806 /* We'll report the final flow when the operation completes. Reporting
4807 * it now would cause a duplicate report later. */
4811 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
4812 ? NXFME_ADDED : NXFME_MODIFIED);
4814 ovs_mutex_lock(&rule->mutex);
4815 fu.idle_timeout = rule->idle_timeout;
4816 fu.hard_timeout = rule->hard_timeout;
4817 ovs_mutex_unlock(&rule->mutex);
4818 fu.table_id = rule->table_id;
4819 fu.cookie = rule->flow_cookie;
4820 minimatch_expand(&rule->cr.match, &match);
4822 fu.priority = rule->cr.priority;
4824 if (!(flags & NXFMF_ACTIONS)) {
4827 actions = rule_get_actions(rule);
4829 /* An operation is in progress. Use the previous version of the flow's
4830 * actions, so that when the operation commits we report the change. */
4832 case OFOPERATION_ADD:
4835 case OFOPERATION_MODIFY:
4836 case OFOPERATION_REPLACE:
4837 actions = op->actions ? op->actions : rule_get_actions(rule);
4840 case OFOPERATION_DELETE:
4841 actions = rule_get_actions(rule);
4848 fu.ofpacts = actions ? actions->ofpacts : NULL;
4849 fu.ofpacts_len = actions ? actions->ofpacts_len : 0;
4851 if (list_is_empty(msgs)) {
4852 ofputil_start_flow_update(msgs);
4854 ofputil_append_flow_update(&fu, msgs);
4858 ofmonitor_compose_refresh_updates(struct rule_collection *rules,
4860 OVS_REQUIRES(ofproto_mutex)
4864 for (i = 0; i < rules->n; i++) {
4865 struct rule *rule = rules->rules[i];
4866 enum nx_flow_monitor_flags flags = rule->monitor_flags;
4867 rule->monitor_flags = 0;
4869 ofproto_compose_flow_refresh_update(rule, flags, msgs);
4874 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
4875 struct rule *rule, uint64_t seqno,
4876 struct rule_collection *rules)
4877 OVS_REQUIRES(ofproto_mutex)
4879 enum nx_flow_monitor_flags update;
4881 if (ofproto_rule_is_hidden(rule)) {
4886 ? ofoperation_has_out_port(rule->pending, m->out_port)
4887 : ofproto_rule_has_out_port(rule, m->out_port))) {
4892 if (rule->add_seqno > seqno) {
4893 update = NXFMF_ADD | NXFMF_MODIFY;
4894 } else if (rule->modify_seqno > seqno) {
4895 update = NXFMF_MODIFY;
4900 if (!(m->flags & update)) {
4904 update = NXFMF_INITIAL;
4907 if (!rule->monitor_flags) {
4908 rule_collection_add(rules, rule);
4910 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
4914 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
4916 struct rule_collection *rules)
4917 OVS_REQUIRES(ofproto_mutex)
4919 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
4920 const struct ofoperation *op;
4921 const struct oftable *table;
4922 struct cls_rule target;
4924 cls_rule_init_from_minimatch(&target, &m->match, 0);
4925 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
4926 struct cls_cursor cursor;
4929 fat_rwlock_rdlock(&table->cls.rwlock);
4930 cls_cursor_init(&cursor, &table->cls, &target);
4931 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4932 ovs_assert(!rule->pending); /* XXX */
4933 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4935 fat_rwlock_unlock(&table->cls.rwlock);
4938 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
4939 struct rule *rule = op->rule;
4941 if (((m->table_id == 0xff
4942 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
4943 : m->table_id == rule->table_id))
4944 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
4945 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4948 cls_rule_destroy(&target);
4952 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
4953 struct rule_collection *rules)
4954 OVS_REQUIRES(ofproto_mutex)
4956 if (m->flags & NXFMF_INITIAL) {
4957 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
4962 ofmonitor_collect_resume_rules(struct ofmonitor *m,
4963 uint64_t seqno, struct rule_collection *rules)
4964 OVS_REQUIRES(ofproto_mutex)
4966 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
4970 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
4971 OVS_EXCLUDED(ofproto_mutex)
4973 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4974 struct ofmonitor **monitors;
4975 size_t n_monitors, allocated_monitors;
4976 struct rule_collection rules;
4977 struct list replies;
4983 ofpbuf_use_const(&b, oh, ntohs(oh->length));
4985 n_monitors = allocated_monitors = 0;
4987 ovs_mutex_lock(&ofproto_mutex);
4989 struct ofputil_flow_monitor_request request;
4990 struct ofmonitor *m;
4993 retval = ofputil_decode_flow_monitor_request(&request, &b);
4994 if (retval == EOF) {
4996 } else if (retval) {
5001 if (request.table_id != 0xff
5002 && request.table_id >= ofproto->n_tables) {
5003 error = OFPERR_OFPBRC_BAD_TABLE_ID;
5007 error = ofmonitor_create(&request, ofconn, &m);
5012 if (n_monitors >= allocated_monitors) {
5013 monitors = x2nrealloc(monitors, &allocated_monitors,
5016 monitors[n_monitors++] = m;
5019 rule_collection_init(&rules);
5020 for (i = 0; i < n_monitors; i++) {
5021 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
5024 ofpmp_init(&replies, oh);
5025 ofmonitor_compose_refresh_updates(&rules, &replies);
5026 ovs_mutex_unlock(&ofproto_mutex);
5028 rule_collection_destroy(&rules);
5030 ofconn_send_replies(ofconn, &replies);
5036 for (i = 0; i < n_monitors; i++) {
5037 ofmonitor_destroy(monitors[i]);
5040 ovs_mutex_unlock(&ofproto_mutex);
5046 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
5047 OVS_EXCLUDED(ofproto_mutex)
5049 struct ofmonitor *m;
5053 id = ofputil_decode_flow_monitor_cancel(oh);
5055 ovs_mutex_lock(&ofproto_mutex);
5056 m = ofmonitor_lookup(ofconn, id);
5058 ofmonitor_destroy(m);
5061 error = OFPERR_NXBRC_FM_BAD_ID;
5063 ovs_mutex_unlock(&ofproto_mutex);
5068 /* Meters implementation.
5070 * Meter table entry, indexed by the OpenFlow meter_id.
5071 * These are always dynamically allocated to allocate enough space for
5073 * 'created' is used to compute the duration for meter stats.
5074 * 'list rules' is needed so that we can delete the dependent rules when the
5075 * meter table entry is deleted.
5076 * 'provider_meter_id' is for the provider's private use.
5079 long long int created; /* Time created. */
5080 struct list rules; /* List of "struct rule_dpif"s. */
5081 ofproto_meter_id provider_meter_id;
5082 uint16_t flags; /* Meter flags. */
5083 uint16_t n_bands; /* Number of meter bands. */
5084 struct ofputil_meter_band *bands;
5088 * This is used in instruction validation at flow set-up time,
5089 * as flows may not use non-existing meters.
5090 * Return value of UINT32_MAX signifies an invalid meter.
5093 get_provider_meter_id(const struct ofproto *ofproto, uint32_t of_meter_id)
5095 if (of_meter_id && of_meter_id <= ofproto->meter_features.max_meters) {
5096 const struct meter *meter = ofproto->meters[of_meter_id];
5098 return meter->provider_meter_id.uint32;
5105 meter_update(struct meter *meter, const struct ofputil_meter_config *config)
5109 meter->flags = config->flags;
5110 meter->n_bands = config->n_bands;
5111 meter->bands = xmemdup(config->bands,
5112 config->n_bands * sizeof *meter->bands);
5115 static struct meter *
5116 meter_create(const struct ofputil_meter_config *config,
5117 ofproto_meter_id provider_meter_id)
5119 struct meter *meter;
5121 meter = xzalloc(sizeof *meter);
5122 meter->provider_meter_id = provider_meter_id;
5123 meter->created = time_msec();
5124 list_init(&meter->rules);
5126 meter_update(meter, config);
5132 meter_delete(struct ofproto *ofproto, uint32_t first, uint32_t last)
5133 OVS_REQUIRES(ofproto_mutex)
5136 for (mid = first; mid <= last; ++mid) {
5137 struct meter *meter = ofproto->meters[mid];
5139 ofproto->meters[mid] = NULL;
5140 ofproto->ofproto_class->meter_del(ofproto,
5141 meter->provider_meter_id);
5149 handle_add_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
5151 ofproto_meter_id provider_meter_id = { UINT32_MAX };
5152 struct meter **meterp = &ofproto->meters[mm->meter.meter_id];
5156 return OFPERR_OFPMMFC_METER_EXISTS;
5159 error = ofproto->ofproto_class->meter_set(ofproto, &provider_meter_id,
5162 ovs_assert(provider_meter_id.uint32 != UINT32_MAX);
5163 *meterp = meter_create(&mm->meter, provider_meter_id);
5169 handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
5171 struct meter *meter = ofproto->meters[mm->meter.meter_id];
5173 uint32_t provider_meter_id;
5176 return OFPERR_OFPMMFC_UNKNOWN_METER;
5179 provider_meter_id = meter->provider_meter_id.uint32;
5180 error = ofproto->ofproto_class->meter_set(ofproto,
5181 &meter->provider_meter_id,
5183 ovs_assert(meter->provider_meter_id.uint32 == provider_meter_id);
5185 meter_update(meter, &mm->meter);
5191 handle_delete_meter(struct ofconn *ofconn, const struct ofp_header *oh,
5192 struct ofputil_meter_mod *mm)
5193 OVS_EXCLUDED(ofproto_mutex)
5195 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5196 uint32_t meter_id = mm->meter.meter_id;
5197 struct rule_collection rules;
5198 enum ofperr error = 0;
5199 uint32_t first, last;
5201 if (meter_id == OFPM13_ALL) {
5203 last = ofproto->meter_features.max_meters;
5205 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
5208 first = last = meter_id;
5211 /* First delete the rules that use this meter. If any of those rules are
5212 * currently being modified, postpone the whole operation until later. */
5213 rule_collection_init(&rules);
5214 ovs_mutex_lock(&ofproto_mutex);
5215 for (meter_id = first; meter_id <= last; ++meter_id) {
5216 struct meter *meter = ofproto->meters[meter_id];
5217 if (meter && !list_is_empty(&meter->rules)) {
5220 LIST_FOR_EACH (rule, meter_list_node, &meter->rules) {
5221 if (rule->pending) {
5222 error = OFPROTO_POSTPONE;
5225 rule_collection_add(&rules, rule);
5230 delete_flows__(ofproto, ofconn, oh, &rules, OFPRR_METER_DELETE);
5233 /* Delete the meters. */
5234 meter_delete(ofproto, first, last);
5237 ovs_mutex_unlock(&ofproto_mutex);
5238 rule_collection_destroy(&rules);
5244 handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5246 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5247 struct ofputil_meter_mod mm;
5248 uint64_t bands_stub[256 / 8];
5249 struct ofpbuf bands;
5253 error = reject_slave_controller(ofconn);
5258 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5260 error = ofputil_decode_meter_mod(oh, &mm, &bands);
5262 goto exit_free_bands;
5265 meter_id = mm.meter.meter_id;
5267 if (mm.command != OFPMC13_DELETE) {
5268 /* Fails also when meters are not implemented by the provider. */
5269 if (meter_id == 0 || meter_id > OFPM13_MAX) {
5270 error = OFPERR_OFPMMFC_INVALID_METER;
5271 goto exit_free_bands;
5272 } else if (meter_id > ofproto->meter_features.max_meters) {
5273 error = OFPERR_OFPMMFC_OUT_OF_METERS;
5274 goto exit_free_bands;
5276 if (mm.meter.n_bands > ofproto->meter_features.max_bands) {
5277 error = OFPERR_OFPMMFC_OUT_OF_BANDS;
5278 goto exit_free_bands;
5282 switch (mm.command) {
5284 error = handle_add_meter(ofproto, &mm);
5287 case OFPMC13_MODIFY:
5288 error = handle_modify_meter(ofproto, &mm);
5291 case OFPMC13_DELETE:
5292 error = handle_delete_meter(ofconn, oh, &mm);
5296 error = OFPERR_OFPMMFC_BAD_COMMAND;
5301 ofpbuf_uninit(&bands);
5306 handle_meter_features_request(struct ofconn *ofconn,
5307 const struct ofp_header *request)
5309 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5310 struct ofputil_meter_features features;
5313 if (ofproto->ofproto_class->meter_get_features) {
5314 ofproto->ofproto_class->meter_get_features(ofproto, &features);
5316 memset(&features, 0, sizeof features);
5318 b = ofputil_encode_meter_features_reply(&features, request);
5320 ofconn_send_reply(ofconn, b);
5325 handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request,
5328 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5329 struct list replies;
5330 uint64_t bands_stub[256 / 8];
5331 struct ofpbuf bands;
5332 uint32_t meter_id, first, last;
5334 ofputil_decode_meter_request(request, &meter_id);
5336 if (meter_id == OFPM13_ALL) {
5338 last = ofproto->meter_features.max_meters;
5340 if (!meter_id || meter_id > ofproto->meter_features.max_meters ||
5341 !ofproto->meters[meter_id]) {
5342 return OFPERR_OFPMMFC_UNKNOWN_METER;
5344 first = last = meter_id;
5347 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5348 ofpmp_init(&replies, request);
5350 for (meter_id = first; meter_id <= last; ++meter_id) {
5351 struct meter *meter = ofproto->meters[meter_id];
5353 continue; /* Skip non-existing meters. */
5355 if (type == OFPTYPE_METER_STATS_REQUEST) {
5356 struct ofputil_meter_stats stats;
5358 stats.meter_id = meter_id;
5360 /* Provider sets the packet and byte counts, we do the rest. */
5361 stats.flow_count = list_size(&meter->rules);
5362 calc_duration(meter->created, time_msec(),
5363 &stats.duration_sec, &stats.duration_nsec);
5364 stats.n_bands = meter->n_bands;
5365 ofpbuf_clear(&bands);
5367 = ofpbuf_put_uninit(&bands,
5368 meter->n_bands * sizeof *stats.bands);
5370 if (!ofproto->ofproto_class->meter_get(ofproto,
5371 meter->provider_meter_id,
5373 ofputil_append_meter_stats(&replies, &stats);
5375 } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */
5376 struct ofputil_meter_config config;
5378 config.meter_id = meter_id;
5379 config.flags = meter->flags;
5380 config.n_bands = meter->n_bands;
5381 config.bands = meter->bands;
5382 ofputil_append_meter_config(&replies, &config);
5386 ofconn_send_replies(ofconn, &replies);
5387 ofpbuf_uninit(&bands);
5392 ofproto_group_lookup(const struct ofproto *ofproto, uint32_t group_id,
5393 struct ofgroup **group)
5394 OVS_TRY_RDLOCK(true, (*group)->rwlock)
5396 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5397 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
5398 hash_int(group_id, 0), &ofproto->groups) {
5399 if ((*group)->group_id == group_id) {
5400 ovs_rwlock_rdlock(&(*group)->rwlock);
5401 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5405 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5410 ofproto_group_release(struct ofgroup *group)
5411 OVS_RELEASES(group->rwlock)
5413 ovs_rwlock_unlock(&group->rwlock);
5417 ofproto_group_write_lookup(const struct ofproto *ofproto, uint32_t group_id,
5418 struct ofgroup **group)
5419 OVS_TRY_WRLOCK(true, ofproto->groups_rwlock)
5420 OVS_TRY_WRLOCK(true, (*group)->rwlock)
5422 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5423 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
5424 hash_int(group_id, 0), &ofproto->groups) {
5425 if ((*group)->group_id == group_id) {
5426 ovs_rwlock_wrlock(&(*group)->rwlock);
5430 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5435 ofproto_group_exists__(const struct ofproto *ofproto, uint32_t group_id)
5436 OVS_REQ_RDLOCK(ofproto->groups_rwlock)
5438 struct ofgroup *grp;
5440 HMAP_FOR_EACH_IN_BUCKET (grp, hmap_node,
5441 hash_int(group_id, 0), &ofproto->groups) {
5442 if (grp->group_id == group_id) {
5450 ofproto_group_exists(const struct ofproto *ofproto, uint32_t group_id)
5451 OVS_EXCLUDED(ofproto->groups_rwlock)
5455 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5456 exists = ofproto_group_exists__(ofproto, group_id);
5457 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5463 group_get_ref_count(struct ofgroup *group)
5464 OVS_EXCLUDED(ofproto_mutex)
5466 struct ofproto *ofproto = group->ofproto;
5467 struct rule_criteria criteria;
5468 struct rule_collection rules;
5473 match_init_catchall(&match);
5474 rule_criteria_init(&criteria, 0xff, &match, 0, htonll(0), htonll(0),
5475 OFPP_ANY, group->group_id);
5476 ovs_mutex_lock(&ofproto_mutex);
5477 error = collect_rules_loose(ofproto, &criteria, &rules);
5478 ovs_mutex_unlock(&ofproto_mutex);
5479 rule_criteria_destroy(&criteria);
5481 count = !error && rules.n < UINT32_MAX ? rules.n : UINT32_MAX;
5483 rule_collection_destroy(&rules);
5488 append_group_stats(struct ofgroup *group, struct list *replies)
5489 OVS_REQ_RDLOCK(group->rwlock)
5491 struct ofputil_group_stats ogs;
5492 struct ofproto *ofproto = group->ofproto;
5493 long long int now = time_msec();
5496 ogs.bucket_stats = xmalloc(group->n_buckets * sizeof *ogs.bucket_stats);
5498 /* Provider sets the packet and byte counts, we do the rest. */
5499 ogs.ref_count = group_get_ref_count(group);
5500 ogs.n_buckets = group->n_buckets;
5502 error = (ofproto->ofproto_class->group_get_stats
5503 ? ofproto->ofproto_class->group_get_stats(group, &ogs)
5506 ogs.packet_count = UINT64_MAX;
5507 ogs.byte_count = UINT64_MAX;
5508 memset(ogs.bucket_stats, 0xff,
5509 ogs.n_buckets * sizeof *ogs.bucket_stats);
5512 ogs.group_id = group->group_id;
5513 calc_duration(group->created, now, &ogs.duration_sec, &ogs.duration_nsec);
5515 ofputil_append_group_stats(replies, &ogs);
5517 free(ogs.bucket_stats);
5521 handle_group_stats_request(struct ofconn *ofconn,
5522 const struct ofp_header *request)
5524 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5525 struct list replies;
5527 struct ofgroup *group;
5530 error = ofputil_decode_group_stats_request(request, &group_id);
5535 ofpmp_init(&replies, request);
5537 if (group_id == OFPG_ALL) {
5538 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5539 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
5540 ovs_rwlock_rdlock(&group->rwlock);
5541 append_group_stats(group, &replies);
5542 ovs_rwlock_unlock(&group->rwlock);
5544 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5546 if (ofproto_group_lookup(ofproto, group_id, &group)) {
5547 append_group_stats(group, &replies);
5548 ofproto_group_release(group);
5552 ofconn_send_replies(ofconn, &replies);
5558 handle_group_desc_stats_request(struct ofconn *ofconn,
5559 const struct ofp_header *request)
5561 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5562 struct list replies;
5563 struct ofputil_group_desc gds;
5564 struct ofgroup *group;
5566 ofpmp_init(&replies, request);
5568 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5569 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
5570 gds.group_id = group->group_id;
5571 gds.type = group->type;
5572 ofputil_append_group_desc_reply(&gds, &group->buckets, &replies);
5574 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5576 ofconn_send_replies(ofconn, &replies);
5582 handle_group_features_stats_request(struct ofconn *ofconn,
5583 const struct ofp_header *request)
5585 struct ofproto *p = ofconn_get_ofproto(ofconn);
5588 msg = ofputil_encode_group_features_reply(&p->ogf, request);
5590 ofconn_send_reply(ofconn, msg);
5597 handle_queue_get_config_request(struct ofconn *ofconn,
5598 const struct ofp_header *oh)
5600 struct ofproto *p = ofconn_get_ofproto(ofconn);
5601 struct netdev_queue_dump queue_dump;
5602 struct ofport *ofport;
5603 unsigned int queue_id;
5604 struct ofpbuf *reply;
5605 struct smap details;
5609 error = ofputil_decode_queue_get_config_request(oh, &request);
5614 ofport = ofproto_get_port(p, request);
5616 return OFPERR_OFPQOFC_BAD_PORT;
5619 reply = ofputil_encode_queue_get_config_reply(oh);
5621 smap_init(&details);
5622 NETDEV_QUEUE_FOR_EACH (&queue_id, &details, &queue_dump, ofport->netdev) {
5623 struct ofputil_queue_config queue;
5625 /* None of the existing queues have compatible properties, so we
5626 * hard-code omitting min_rate and max_rate. */
5627 queue.queue_id = queue_id;
5628 queue.min_rate = UINT16_MAX;
5629 queue.max_rate = UINT16_MAX;
5630 ofputil_append_queue_get_config_reply(reply, &queue);
5632 smap_destroy(&details);
5634 ofconn_send_reply(ofconn, reply);
5639 /* Implements OFPGC11_ADD
5640 * in which no matching flow already exists in the flow table.
5642 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
5643 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
5644 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
5645 * initiated now but may be retried later.
5647 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
5648 * ownership remains with the caller.
5650 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
5653 add_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
5655 struct ofgroup *ofgroup;
5658 if (gm->group_id > OFPG_MAX) {
5659 return OFPERR_OFPGMFC_INVALID_GROUP;
5661 if (gm->type > OFPGT11_FF) {
5662 return OFPERR_OFPGMFC_BAD_TYPE;
5665 /* Allocate new group and initialize it. */
5666 ofgroup = ofproto->ofproto_class->group_alloc();
5668 VLOG_WARN_RL(&rl, "%s: failed to create group", ofproto->name);
5669 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
5672 ovs_rwlock_init(&ofgroup->rwlock);
5673 ofgroup->ofproto = ofproto;
5674 ofgroup->group_id = gm->group_id;
5675 ofgroup->type = gm->type;
5676 ofgroup->created = ofgroup->modified = time_msec();
5678 list_move(&ofgroup->buckets, &gm->buckets);
5679 ofgroup->n_buckets = list_size(&ofgroup->buckets);
5681 /* Construct called BEFORE any locks are held. */
5682 error = ofproto->ofproto_class->group_construct(ofgroup);
5687 /* We wrlock as late as possible to minimize the time we jam any other
5688 * threads: No visible state changes before acquiring the lock. */
5689 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5691 if (ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
5692 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
5696 if (ofproto_group_exists__(ofproto, gm->group_id)) {
5697 error = OFPERR_OFPGMFC_GROUP_EXISTS;
5702 /* Insert new group. */
5703 hmap_insert(&ofproto->groups, &ofgroup->hmap_node,
5704 hash_int(ofgroup->group_id, 0));
5705 ofproto->n_groups[ofgroup->type]++;
5707 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5712 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5713 ofproto->ofproto_class->group_destruct(ofgroup);
5715 ofputil_bucket_list_destroy(&ofgroup->buckets);
5716 ofproto->ofproto_class->group_dealloc(ofgroup);
5721 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
5724 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
5727 modify_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
5729 struct ofgroup *ofgroup;
5730 struct ofgroup *victim;
5733 if (gm->group_id > OFPG_MAX) {
5734 return OFPERR_OFPGMFC_INVALID_GROUP;
5737 if (gm->type > OFPGT11_FF) {
5738 return OFPERR_OFPGMFC_BAD_TYPE;
5741 victim = ofproto->ofproto_class->group_alloc();
5743 VLOG_WARN_RL(&rl, "%s: failed to allocate group", ofproto->name);
5744 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
5747 if (!ofproto_group_write_lookup(ofproto, gm->group_id, &ofgroup)) {
5748 error = OFPERR_OFPGMFC_UNKNOWN_GROUP;
5751 /* Both group's and its container's write locks held now.
5752 * Also, n_groups[] is protected by ofproto->groups_rwlock. */
5753 if (ofgroup->type != gm->type
5754 && ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
5755 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
5760 list_move(&victim->buckets, &ofgroup->buckets);
5762 ofgroup->type = gm->type;
5763 list_move(&ofgroup->buckets, &gm->buckets);
5764 ofgroup->n_buckets = list_size(&ofgroup->buckets);
5766 error = ofproto->ofproto_class->group_modify(ofgroup, victim);
5768 ofputil_bucket_list_destroy(&victim->buckets);
5769 ofproto->n_groups[victim->type]--;
5770 ofproto->n_groups[ofgroup->type]++;
5771 ofgroup->modified = time_msec();
5773 ofputil_bucket_list_destroy(&ofgroup->buckets);
5776 list_move(&ofgroup->buckets, &victim->buckets);
5780 ovs_rwlock_unlock(&ofgroup->rwlock);
5781 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5783 ofproto->ofproto_class->group_dealloc(victim);
5788 delete_group__(struct ofproto *ofproto, struct ofgroup *ofgroup)
5789 OVS_RELEASES(ofproto->groups_rwlock)
5792 struct ofputil_flow_mod fm;
5794 /* Delete all flow entries containing this group in a group action */
5795 match_init_catchall(&match);
5796 flow_mod_init(&fm, &match, 0, NULL, 0, OFPFC_DELETE);
5797 fm.out_group = ofgroup->group_id;
5798 handle_flow_mod__(ofproto, NULL, &fm, NULL);
5800 /* Must wait until existing readers are done,
5801 * while holding the container's write lock at the same time. */
5802 ovs_rwlock_wrlock(&ofgroup->rwlock);
5803 hmap_remove(&ofproto->groups, &ofgroup->hmap_node);
5804 /* No-one can find this group any more. */
5805 ofproto->n_groups[ofgroup->type]--;
5806 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5808 ofproto->ofproto_class->group_destruct(ofgroup);
5809 ofputil_bucket_list_destroy(&ofgroup->buckets);
5810 ovs_rwlock_unlock(&ofgroup->rwlock);
5811 ovs_rwlock_destroy(&ofgroup->rwlock);
5812 ofproto->ofproto_class->group_dealloc(ofgroup);
5815 /* Implements OFPGC_DELETE. */
5817 delete_group(struct ofproto *ofproto, uint32_t group_id)
5819 struct ofgroup *ofgroup;
5821 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5822 if (group_id == OFPG_ALL) {
5824 struct hmap_node *node = hmap_first(&ofproto->groups);
5828 ofgroup = CONTAINER_OF(node, struct ofgroup, hmap_node);
5829 delete_group__(ofproto, ofgroup);
5830 /* Lock for each node separately, so that we will not jam the
5831 * other threads for too long time. */
5832 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5835 HMAP_FOR_EACH_IN_BUCKET (ofgroup, hmap_node,
5836 hash_int(group_id, 0), &ofproto->groups) {
5837 if (ofgroup->group_id == group_id) {
5838 delete_group__(ofproto, ofgroup);
5843 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5847 handle_group_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5849 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5850 struct ofputil_group_mod gm;
5853 error = reject_slave_controller(ofconn);
5858 error = ofputil_decode_group_mod(oh, &gm);
5863 switch (gm.command) {
5865 return add_group(ofproto, &gm);
5867 case OFPGC11_MODIFY:
5868 return modify_group(ofproto, &gm);
5870 case OFPGC11_DELETE:
5871 delete_group(ofproto, gm.group_id);
5875 if (gm.command > OFPGC11_DELETE) {
5876 VLOG_WARN_RL(&rl, "%s: Invalid group_mod command type %d",
5877 ofproto->name, gm.command);
5879 return OFPERR_OFPGMFC_BAD_COMMAND;
5883 enum ofproto_table_config
5884 ofproto_table_get_config(const struct ofproto *ofproto, uint8_t table_id)
5887 atomic_read(&ofproto->tables[table_id].config, &value);
5888 return (enum ofproto_table_config)value;
5892 table_mod(struct ofproto *ofproto, const struct ofputil_table_mod *tm)
5894 /* Only accept currently supported configurations */
5895 if (tm->config & ~OFPTC11_TABLE_MISS_MASK) {
5896 return OFPERR_OFPTMFC_BAD_CONFIG;
5899 if (tm->table_id == OFPTT_ALL) {
5901 for (i = 0; i < ofproto->n_tables; i++) {
5902 atomic_store(&ofproto->tables[i].config,
5903 (unsigned int)tm->config);
5905 } else if (!check_table_id(ofproto, tm->table_id)) {
5906 return OFPERR_OFPTMFC_BAD_TABLE;
5908 atomic_store(&ofproto->tables[tm->table_id].config,
5909 (unsigned int)tm->config);
5916 handle_table_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5918 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5919 struct ofputil_table_mod tm;
5922 error = reject_slave_controller(ofconn);
5927 error = ofputil_decode_table_mod(oh, &tm);
5932 return table_mod(ofproto, &tm);
5936 handle_bundle_control(struct ofconn *ofconn, const struct ofp_header *oh)
5939 struct ofputil_bundle_ctrl_msg bctrl;
5941 struct ofputil_bundle_ctrl_msg reply;
5943 error = ofputil_decode_bundle_ctrl(oh, &bctrl);
5948 reply.bundle_id = bctrl.bundle_id;
5950 switch (bctrl.type) {
5951 case OFPBCT_OPEN_REQUEST:
5952 error = ofp_bundle_open(ofconn, bctrl.bundle_id, bctrl.flags);
5953 reply.type = OFPBCT_OPEN_REPLY;
5955 case OFPBCT_CLOSE_REQUEST:
5956 error = ofp_bundle_close(ofconn, bctrl.bundle_id, bctrl.flags);
5957 reply.type = OFPBCT_CLOSE_REPLY;;
5959 case OFPBCT_COMMIT_REQUEST:
5960 error = ofp_bundle_commit(ofconn, bctrl.bundle_id, bctrl.flags);
5961 reply.type = OFPBCT_COMMIT_REPLY;
5963 case OFPBCT_DISCARD_REQUEST:
5964 error = ofp_bundle_discard(ofconn, bctrl.bundle_id);
5965 reply.type = OFPBCT_DISCARD_REPLY;
5968 case OFPBCT_OPEN_REPLY:
5969 case OFPBCT_CLOSE_REPLY:
5970 case OFPBCT_COMMIT_REPLY:
5971 case OFPBCT_DISCARD_REPLY:
5972 return OFPERR_OFPBFC_BAD_TYPE;
5977 buf = ofputil_encode_bundle_ctrl_reply(oh, &reply);
5978 ofconn_send_reply(ofconn, buf);
5985 handle_bundle_add(struct ofconn *ofconn, const struct ofp_header *oh)
5988 struct ofputil_bundle_add_msg badd;
5990 error = ofputil_decode_bundle_add(oh, &badd);
5995 return ofp_bundle_add_message(ofconn, &badd);
5999 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
6000 OVS_EXCLUDED(ofproto_mutex)
6002 const struct ofp_header *oh = ofpbuf_data(msg);
6006 error = ofptype_decode(&type, oh);
6010 if (oh->version >= OFP13_VERSION && ofpmsg_is_stat_request(oh)
6011 && ofpmp_more(oh)) {
6012 /* We have no buffer implementation for multipart requests.
6013 * Report overflow for requests which consists of multiple
6015 return OFPERR_OFPBRC_MULTIPART_BUFFER_OVERFLOW;
6019 /* OpenFlow requests. */
6020 case OFPTYPE_ECHO_REQUEST:
6021 return handle_echo_request(ofconn, oh);
6023 case OFPTYPE_FEATURES_REQUEST:
6024 return handle_features_request(ofconn, oh);
6026 case OFPTYPE_GET_CONFIG_REQUEST:
6027 return handle_get_config_request(ofconn, oh);
6029 case OFPTYPE_SET_CONFIG:
6030 return handle_set_config(ofconn, oh);
6032 case OFPTYPE_PACKET_OUT:
6033 return handle_packet_out(ofconn, oh);
6035 case OFPTYPE_PORT_MOD:
6036 return handle_port_mod(ofconn, oh);
6038 case OFPTYPE_FLOW_MOD:
6039 return handle_flow_mod(ofconn, oh);
6041 case OFPTYPE_GROUP_MOD:
6042 return handle_group_mod(ofconn, oh);
6044 case OFPTYPE_TABLE_MOD:
6045 return handle_table_mod(ofconn, oh);
6047 case OFPTYPE_METER_MOD:
6048 return handle_meter_mod(ofconn, oh);
6050 case OFPTYPE_BARRIER_REQUEST:
6051 return handle_barrier_request(ofconn, oh);
6053 case OFPTYPE_ROLE_REQUEST:
6054 return handle_role_request(ofconn, oh);
6056 /* OpenFlow replies. */
6057 case OFPTYPE_ECHO_REPLY:
6060 /* Nicira extension requests. */
6061 case OFPTYPE_FLOW_MOD_TABLE_ID:
6062 return handle_nxt_flow_mod_table_id(ofconn, oh);
6064 case OFPTYPE_SET_FLOW_FORMAT:
6065 return handle_nxt_set_flow_format(ofconn, oh);
6067 case OFPTYPE_SET_PACKET_IN_FORMAT:
6068 return handle_nxt_set_packet_in_format(ofconn, oh);
6070 case OFPTYPE_SET_CONTROLLER_ID:
6071 return handle_nxt_set_controller_id(ofconn, oh);
6073 case OFPTYPE_FLOW_AGE:
6074 /* Nothing to do. */
6077 case OFPTYPE_FLOW_MONITOR_CANCEL:
6078 return handle_flow_monitor_cancel(ofconn, oh);
6080 case OFPTYPE_SET_ASYNC_CONFIG:
6081 return handle_nxt_set_async_config(ofconn, oh);
6083 case OFPTYPE_GET_ASYNC_REQUEST:
6084 return handle_nxt_get_async_request(ofconn, oh);
6086 /* Statistics requests. */
6087 case OFPTYPE_DESC_STATS_REQUEST:
6088 return handle_desc_stats_request(ofconn, oh);
6090 case OFPTYPE_FLOW_STATS_REQUEST:
6091 return handle_flow_stats_request(ofconn, oh);
6093 case OFPTYPE_AGGREGATE_STATS_REQUEST:
6094 return handle_aggregate_stats_request(ofconn, oh);
6096 case OFPTYPE_TABLE_STATS_REQUEST:
6097 return handle_table_stats_request(ofconn, oh);
6099 case OFPTYPE_PORT_STATS_REQUEST:
6100 return handle_port_stats_request(ofconn, oh);
6102 case OFPTYPE_QUEUE_STATS_REQUEST:
6103 return handle_queue_stats_request(ofconn, oh);
6105 case OFPTYPE_PORT_DESC_STATS_REQUEST:
6106 return handle_port_desc_stats_request(ofconn, oh);
6108 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
6109 return handle_flow_monitor_request(ofconn, oh);
6111 case OFPTYPE_METER_STATS_REQUEST:
6112 case OFPTYPE_METER_CONFIG_STATS_REQUEST:
6113 return handle_meter_request(ofconn, oh, type);
6115 case OFPTYPE_METER_FEATURES_STATS_REQUEST:
6116 return handle_meter_features_request(ofconn, oh);
6118 case OFPTYPE_GROUP_STATS_REQUEST:
6119 return handle_group_stats_request(ofconn, oh);
6121 case OFPTYPE_GROUP_DESC_STATS_REQUEST:
6122 return handle_group_desc_stats_request(ofconn, oh);
6124 case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
6125 return handle_group_features_stats_request(ofconn, oh);
6127 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
6128 return handle_queue_get_config_request(ofconn, oh);
6130 case OFPTYPE_BUNDLE_CONTROL:
6131 return handle_bundle_control(ofconn, oh);
6133 case OFPTYPE_BUNDLE_ADD_MESSAGE:
6134 return handle_bundle_add(ofconn, oh);
6138 case OFPTYPE_FEATURES_REPLY:
6139 case OFPTYPE_GET_CONFIG_REPLY:
6140 case OFPTYPE_PACKET_IN:
6141 case OFPTYPE_FLOW_REMOVED:
6142 case OFPTYPE_PORT_STATUS:
6143 case OFPTYPE_BARRIER_REPLY:
6144 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
6145 case OFPTYPE_DESC_STATS_REPLY:
6146 case OFPTYPE_FLOW_STATS_REPLY:
6147 case OFPTYPE_QUEUE_STATS_REPLY:
6148 case OFPTYPE_PORT_STATS_REPLY:
6149 case OFPTYPE_TABLE_STATS_REPLY:
6150 case OFPTYPE_AGGREGATE_STATS_REPLY:
6151 case OFPTYPE_PORT_DESC_STATS_REPLY:
6152 case OFPTYPE_ROLE_REPLY:
6153 case OFPTYPE_FLOW_MONITOR_PAUSED:
6154 case OFPTYPE_FLOW_MONITOR_RESUMED:
6155 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
6156 case OFPTYPE_GET_ASYNC_REPLY:
6157 case OFPTYPE_GROUP_STATS_REPLY:
6158 case OFPTYPE_GROUP_DESC_STATS_REPLY:
6159 case OFPTYPE_GROUP_FEATURES_STATS_REPLY:
6160 case OFPTYPE_METER_STATS_REPLY:
6161 case OFPTYPE_METER_CONFIG_STATS_REPLY:
6162 case OFPTYPE_METER_FEATURES_STATS_REPLY:
6163 case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
6164 case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
6165 case OFPTYPE_ROLE_STATUS:
6167 if (ofpmsg_is_stat_request(oh)) {
6168 return OFPERR_OFPBRC_BAD_STAT;
6170 return OFPERR_OFPBRC_BAD_TYPE;
6176 handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
6177 OVS_EXCLUDED(ofproto_mutex)
6179 int error = handle_openflow__(ofconn, ofp_msg);
6180 if (error && error != OFPROTO_POSTPONE) {
6181 ofconn_send_error(ofconn, ofpbuf_data(ofp_msg), error);
6183 COVERAGE_INC(ofproto_recv_openflow);
6184 return error != OFPROTO_POSTPONE;
6187 /* Asynchronous operations. */
6189 /* Creates and returns a new ofopgroup that is not associated with any
6190 * OpenFlow connection.
6192 * The caller should add operations to the returned group with
6193 * ofoperation_create() and then submit it with ofopgroup_submit(). */
6194 static struct ofopgroup *
6195 ofopgroup_create_unattached(struct ofproto *ofproto)
6196 OVS_REQUIRES(ofproto_mutex)
6198 struct ofopgroup *group = xzalloc(sizeof *group);
6199 group->ofproto = ofproto;
6200 list_init(&group->ofproto_node);
6201 list_init(&group->ops);
6202 list_init(&group->ofconn_node);
6206 /* Creates and returns a new ofopgroup for 'ofproto'.
6208 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
6209 * connection. The 'request' and 'buffer_id' arguments are ignored.
6211 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
6212 * If the ofopgroup eventually fails, then the error reply will include
6213 * 'request'. If the ofopgroup eventually succeeds, then the packet with
6214 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
6216 * The caller should add operations to the returned group with
6217 * ofoperation_create() and then submit it with ofopgroup_submit(). */
6218 static struct ofopgroup *
6219 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
6220 const struct ofp_header *request, uint32_t buffer_id)
6221 OVS_REQUIRES(ofproto_mutex)
6223 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
6225 size_t request_len = ntohs(request->length);
6227 ovs_assert(ofconn_get_ofproto(ofconn) == ofproto);
6229 ofconn_add_opgroup(ofconn, &group->ofconn_node);
6230 group->ofconn = ofconn;
6231 group->request = xmemdup(request, MIN(request_len, 64));
6232 group->buffer_id = buffer_id;
6237 /* Submits 'group' for processing.
6239 * If 'group' contains no operations (e.g. none were ever added, or all of the
6240 * ones that were added completed synchronously), then it is destroyed
6241 * immediately. Otherwise it is added to the ofproto's list of pending
6244 ofopgroup_submit(struct ofopgroup *group)
6245 OVS_REQUIRES(ofproto_mutex)
6247 if (!group->n_running) {
6248 ofopgroup_complete(group);
6250 list_push_back(&group->ofproto->pending, &group->ofproto_node);
6251 group->ofproto->n_pending++;
6256 ofopgroup_complete(struct ofopgroup *group)
6257 OVS_REQUIRES(ofproto_mutex)
6259 struct ofproto *ofproto = group->ofproto;
6261 struct ofconn *abbrev_ofconn;
6262 ovs_be32 abbrev_xid;
6264 struct ofoperation *op, *next_op;
6267 ovs_assert(!group->n_running);
6270 LIST_FOR_EACH (op, group_node, &group->ops) {
6277 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
6278 LIST_FOR_EACH (op, group_node, &group->ops) {
6279 if (op->type != OFOPERATION_DELETE) {
6280 struct ofpbuf *packet;
6283 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
6286 struct rule_execute *re;
6290 ofproto_rule_ref(op->rule);
6292 re = xmalloc(sizeof *re);
6293 re->rule = op->rule;
6294 re->in_port = in_port;
6295 re->packet = packet;
6297 if (!guarded_list_push_back(&ofproto->rule_executes,
6298 &re->list_node, 1024)) {
6299 ofproto_rule_unref(op->rule);
6300 ofpbuf_delete(re->packet);
6309 if (!error && !list_is_empty(&group->ofconn_node)) {
6310 abbrev_ofconn = group->ofconn;
6311 abbrev_xid = group->request->xid;
6313 abbrev_ofconn = NULL;
6314 abbrev_xid = htonl(0);
6316 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
6317 struct rule *rule = op->rule;
6319 /* We generally want to report the change to active OpenFlow flow
6320 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
6322 - The operation failed.
6324 - The affected rule is not visible to controllers.
6326 - The operation's only effect was to update rule->modified. */
6328 || ofproto_rule_is_hidden(rule)
6329 || (op->type == OFOPERATION_MODIFY
6331 && rule->flow_cookie == op->flow_cookie))) {
6332 /* Check that we can just cast from ofoperation_type to
6333 * nx_flow_update_event. */
6334 enum nx_flow_update_event event_type;
6337 case OFOPERATION_ADD:
6338 case OFOPERATION_REPLACE:
6339 event_type = NXFME_ADDED;
6342 case OFOPERATION_DELETE:
6343 event_type = NXFME_DELETED;
6346 case OFOPERATION_MODIFY:
6347 event_type = NXFME_MODIFIED;
6354 ofmonitor_report(ofproto->connmgr, rule, event_type,
6355 op->reason, abbrev_ofconn, abbrev_xid);
6358 rule->pending = NULL;
6361 case OFOPERATION_ADD:
6365 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
6366 if (vid_mask == VLAN_VID_MASK) {
6367 if (ofproto->vlan_bitmap) {
6368 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
6369 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
6370 bitmap_set1(ofproto->vlan_bitmap, vid);
6371 ofproto->vlans_changed = true;
6374 ofproto->vlans_changed = true;
6378 oftable_remove_rule(rule);
6379 ofproto_rule_unref(rule);
6383 case OFOPERATION_DELETE:
6384 ovs_assert(!op->error);
6385 ofproto_rule_unref(rule);
6389 case OFOPERATION_MODIFY:
6390 case OFOPERATION_REPLACE:
6392 long long int now = time_msec();
6394 ovs_mutex_lock(&rule->mutex);
6395 rule->modified = now;
6396 if (op->type == OFOPERATION_REPLACE) {
6397 rule->created = now;
6399 ovs_mutex_unlock(&rule->mutex);
6401 ofproto_rule_change_cookie(ofproto, rule, op->flow_cookie);
6402 ovs_mutex_lock(&rule->mutex);
6403 rule->idle_timeout = op->idle_timeout;
6404 rule->hard_timeout = op->hard_timeout;
6405 ovs_mutex_unlock(&rule->mutex);
6407 const struct rule_actions *old_actions;
6409 ovs_mutex_lock(&rule->mutex);
6410 old_actions = rule_get_actions(rule);
6411 ovsrcu_set(&rule->actions, op->actions);
6412 ovs_mutex_unlock(&rule->mutex);
6415 rule_actions_destroy(old_actions);
6417 rule->flags = op->flags;
6425 ofoperation_destroy(op);
6428 ofmonitor_flush(ofproto->connmgr);
6430 if (!list_is_empty(&group->ofproto_node)) {
6431 ovs_assert(ofproto->n_pending > 0);
6432 ofproto->n_pending--;
6433 list_remove(&group->ofproto_node);
6435 if (!list_is_empty(&group->ofconn_node)) {
6436 list_remove(&group->ofconn_node);
6438 ofconn_send_error(group->ofconn, group->request, error);
6440 connmgr_retry(ofproto->connmgr);
6442 free(group->request);
6446 /* Initiates a new operation on 'rule', of the specified 'type', within
6447 * 'group'. Prior to calling, 'rule' must not have any pending operation.
6449 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
6450 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
6452 * Returns the newly created ofoperation (which is also available as
6453 * rule->pending). */
6454 static struct ofoperation *
6455 ofoperation_create(struct ofopgroup *group, struct rule *rule,
6456 enum ofoperation_type type,
6457 enum ofp_flow_removed_reason reason)
6458 OVS_REQUIRES(ofproto_mutex)
6460 struct ofproto *ofproto = group->ofproto;
6461 struct ofoperation *op;
6463 ovs_assert(!rule->pending);
6465 op = rule->pending = xzalloc(sizeof *op);
6467 list_push_back(&group->ops, &op->group_node);
6470 op->reason = reason;
6471 op->flow_cookie = rule->flow_cookie;
6472 ovs_mutex_lock(&rule->mutex);
6473 op->idle_timeout = rule->idle_timeout;
6474 op->hard_timeout = rule->hard_timeout;
6475 ovs_mutex_unlock(&rule->mutex);
6476 op->flags = rule->flags;
6480 if (type == OFOPERATION_DELETE) {
6481 hmap_insert(&ofproto->deletions, &op->hmap_node,
6482 cls_rule_hash(&rule->cr, rule->table_id));
6489 ofoperation_destroy(struct ofoperation *op)
6490 OVS_REQUIRES(ofproto_mutex)
6492 struct ofopgroup *group = op->group;
6495 op->rule->pending = NULL;
6497 if (op->type == OFOPERATION_DELETE) {
6498 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
6500 list_remove(&op->group_node);
6501 rule_actions_destroy(op->actions);
6505 /* Indicates that 'op' completed with status 'error', which is either 0 to
6506 * indicate success or an OpenFlow error code on failure.
6508 * If 'error' is 0, indicating success, the operation will be committed
6509 * permanently to the flow table.
6511 * If 'error' is nonzero, then generally the operation will be rolled back:
6513 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
6514 * restores the original rule. The caller must have uninitialized any
6515 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
6516 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
6517 * and 7 for the new rule, calling its ->rule_dealloc() function.
6519 * - If 'op' is a "modify flow" operation, ofproto restores the original
6522 * - 'op' must not be a "delete flow" operation. Removing a rule is not
6523 * allowed to fail. It must always succeed.
6525 * Please see the large comment in ofproto/ofproto-provider.h titled
6526 * "Asynchronous Operation Support" for more information. */
6528 ofoperation_complete(struct ofoperation *op, enum ofperr error)
6530 struct ofopgroup *group = op->group;
6532 ovs_assert(group->n_running > 0);
6533 ovs_assert(!error || op->type != OFOPERATION_DELETE);
6536 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
6537 /* This function can be called from ->rule_construct(), in which case
6538 * ofproto_mutex is held, or it can be called from ->run(), in which
6539 * case ofproto_mutex is not held. But only in the latter case can we
6540 * arrive here, so we can safely take ofproto_mutex now. */
6541 ovs_mutex_lock(&ofproto_mutex);
6542 ovs_assert(op->rule->pending == op);
6543 ofopgroup_complete(group);
6544 ovs_mutex_unlock(&ofproto_mutex);
6549 pick_datapath_id(const struct ofproto *ofproto)
6551 const struct ofport *port;
6553 port = ofproto_get_port(ofproto, OFPP_LOCAL);
6555 uint8_t ea[ETH_ADDR_LEN];
6558 error = netdev_get_etheraddr(port->netdev, ea);
6560 return eth_addr_to_uint64(ea);
6562 VLOG_WARN("%s: could not get MAC address for %s (%s)",
6563 ofproto->name, netdev_get_name(port->netdev),
6564 ovs_strerror(error));
6566 return ofproto->fallback_dpid;
6570 pick_fallback_dpid(void)
6572 uint8_t ea[ETH_ADDR_LEN];
6573 eth_addr_nicira_random(ea);
6574 return eth_addr_to_uint64(ea);
6577 /* Table overflow policy. */
6579 /* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep'
6580 * to NULL if the table is not configured to evict rules or if the table
6581 * contains no evictable rules. (Rules with a readlock on their evict rwlock,
6582 * or with no timeouts are not evictable.) */
6584 choose_rule_to_evict(struct oftable *table, struct rule **rulep)
6585 OVS_REQUIRES(ofproto_mutex)
6587 struct eviction_group *evg;
6590 if (!table->eviction_fields) {
6594 /* In the common case, the outer and inner loops here will each be entered
6597 * - The inner loop normally "return"s in its first iteration. If the
6598 * eviction group has any evictable rules, then it always returns in
6601 * - The outer loop only iterates more than once if the largest eviction
6602 * group has no evictable rules.
6604 * - The outer loop can exit only if table's 'max_flows' is all filled up
6605 * by unevictable rules. */
6606 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
6609 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
6618 /* Searches 'ofproto' for tables that have more flows than their configured
6619 * maximum and that have flow eviction enabled, and evicts as many flows as
6620 * necessary and currently feasible from them.
6622 * This triggers only when an OpenFlow table has N flows in it and then the
6623 * client configures a maximum number of flows less than N. */
6625 ofproto_evict(struct ofproto *ofproto)
6627 struct oftable *table;
6629 ovs_mutex_lock(&ofproto_mutex);
6630 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
6631 evict_rules_from_table(ofproto, table, 0);
6633 ovs_mutex_unlock(&ofproto_mutex);
6636 /* Eviction groups. */
6638 /* Returns the priority to use for an eviction_group that contains 'n_rules'
6639 * rules. The priority contains low-order random bits to ensure that eviction
6640 * groups with the same number of rules are prioritized randomly. */
6642 eviction_group_priority(size_t n_rules)
6644 uint16_t size = MIN(UINT16_MAX, n_rules);
6645 return (size << 16) | random_uint16();
6648 /* Updates 'evg', an eviction_group within 'table', following a change that
6649 * adds or removes rules in 'evg'. */
6651 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
6652 OVS_REQUIRES(ofproto_mutex)
6654 heap_change(&table->eviction_groups_by_size, &evg->size_node,
6655 eviction_group_priority(heap_count(&evg->rules)));
6658 /* Destroys 'evg', an eviction_group within 'table':
6660 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
6661 * rules themselves, just removes them from the eviction group.)
6663 * - Removes 'evg' from 'table'.
6667 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
6668 OVS_REQUIRES(ofproto_mutex)
6670 while (!heap_is_empty(&evg->rules)) {
6673 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
6674 rule->eviction_group = NULL;
6676 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
6677 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
6678 heap_destroy(&evg->rules);
6682 /* Removes 'rule' from its eviction group, if any. */
6684 eviction_group_remove_rule(struct rule *rule)
6685 OVS_REQUIRES(ofproto_mutex)
6687 if (rule->eviction_group) {
6688 struct oftable *table = &rule->ofproto->tables[rule->table_id];
6689 struct eviction_group *evg = rule->eviction_group;
6691 rule->eviction_group = NULL;
6692 heap_remove(&evg->rules, &rule->evg_node);
6693 if (heap_is_empty(&evg->rules)) {
6694 eviction_group_destroy(table, evg);
6696 eviction_group_resized(table, evg);
6701 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
6702 * returns the hash value. */
6704 eviction_group_hash_rule(struct rule *rule)
6705 OVS_REQUIRES(ofproto_mutex)
6707 struct oftable *table = &rule->ofproto->tables[rule->table_id];
6708 const struct mf_subfield *sf;
6712 hash = table->eviction_group_id_basis;
6713 miniflow_expand(&rule->cr.match.flow, &flow);
6714 for (sf = table->eviction_fields;
6715 sf < &table->eviction_fields[table->n_eviction_fields];
6718 if (mf_are_prereqs_ok(sf->field, &flow)) {
6719 union mf_value value;
6721 mf_get_value(sf->field, &flow, &value);
6723 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
6725 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
6726 unsigned int start = sf->ofs + sf->n_bits;
6727 bitwise_zero(&value, sf->field->n_bytes, start,
6728 sf->field->n_bytes * 8 - start);
6730 hash = hash_bytes(&value, sf->field->n_bytes, hash);
6732 hash = hash_int(hash, 0);
6739 /* Returns an eviction group within 'table' with the given 'id', creating one
6741 static struct eviction_group *
6742 eviction_group_find(struct oftable *table, uint32_t id)
6743 OVS_REQUIRES(ofproto_mutex)
6745 struct eviction_group *evg;
6747 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
6751 evg = xmalloc(sizeof *evg);
6752 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
6753 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
6754 eviction_group_priority(0));
6755 heap_init(&evg->rules);
6760 /* Returns an eviction priority for 'rule'. The return value should be
6761 * interpreted so that higher priorities make a rule more attractive candidates
6763 * Called only if have a timeout. */
6765 rule_eviction_priority(struct ofproto *ofproto, struct rule *rule)
6766 OVS_REQUIRES(ofproto_mutex)
6768 long long int expiration = LLONG_MAX;
6769 long long int modified;
6770 uint32_t expiration_offset;
6772 /* 'modified' needs protection even when we hold 'ofproto_mutex'. */
6773 ovs_mutex_lock(&rule->mutex);
6774 modified = rule->modified;
6775 ovs_mutex_unlock(&rule->mutex);
6777 if (rule->hard_timeout) {
6778 expiration = modified + rule->hard_timeout * 1000;
6780 if (rule->idle_timeout) {
6781 uint64_t packets, bytes;
6783 long long int idle_expiration;
6785 ofproto->ofproto_class->rule_get_stats(rule, &packets, &bytes, &used);
6786 idle_expiration = used + rule->idle_timeout * 1000;
6787 expiration = MIN(expiration, idle_expiration);
6790 if (expiration == LLONG_MAX) {
6794 /* Calculate the time of expiration as a number of (approximate) seconds
6795 * after program startup.
6797 * This should work OK for program runs that last UINT32_MAX seconds or
6798 * less. Therefore, please restart OVS at least once every 136 years. */
6799 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
6801 /* Invert the expiration offset because we're using a max-heap. */
6802 return UINT32_MAX - expiration_offset;
6805 /* Adds 'rule' to an appropriate eviction group for its oftable's
6806 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
6807 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
6810 * The caller must ensure that 'rule' is not already in an eviction group. */
6812 eviction_group_add_rule(struct rule *rule)
6813 OVS_REQUIRES(ofproto_mutex)
6815 struct ofproto *ofproto = rule->ofproto;
6816 struct oftable *table = &ofproto->tables[rule->table_id];
6819 /* Timeouts may be modified only when holding 'ofproto_mutex'. We have it
6820 * so no additional protection is needed. */
6821 has_timeout = rule->hard_timeout || rule->idle_timeout;
6823 if (table->eviction_fields && has_timeout) {
6824 struct eviction_group *evg;
6826 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
6828 rule->eviction_group = evg;
6829 heap_insert(&evg->rules, &rule->evg_node,
6830 rule_eviction_priority(ofproto, rule));
6831 eviction_group_resized(table, evg);
6837 /* Initializes 'table'. */
6839 oftable_init(struct oftable *table)
6841 memset(table, 0, sizeof *table);
6842 classifier_init(&table->cls, flow_segment_u32s);
6843 table->max_flows = UINT_MAX;
6844 atomic_init(&table->config, (unsigned int)OFPROTO_TABLE_MISS_DEFAULT);
6847 /* Destroys 'table', including its classifier and eviction groups.
6849 * The caller is responsible for freeing 'table' itself. */
6851 oftable_destroy(struct oftable *table)
6853 fat_rwlock_rdlock(&table->cls.rwlock);
6854 ovs_assert(classifier_is_empty(&table->cls));
6855 fat_rwlock_unlock(&table->cls.rwlock);
6856 oftable_disable_eviction(table);
6857 classifier_destroy(&table->cls);
6861 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
6862 * string, then 'table' will use its default name.
6864 * This only affects the name exposed for a table exposed through the OpenFlow
6865 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
6867 oftable_set_name(struct oftable *table, const char *name)
6869 if (name && name[0]) {
6870 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
6871 if (!table->name || strncmp(name, table->name, len)) {
6873 table->name = xmemdup0(name, len);
6881 /* oftables support a choice of two policies when adding a rule would cause the
6882 * number of flows in the table to exceed the configured maximum number: either
6883 * they can refuse to add the new flow or they can evict some existing flow.
6884 * This function configures the former policy on 'table'. */
6886 oftable_disable_eviction(struct oftable *table)
6887 OVS_REQUIRES(ofproto_mutex)
6889 if (table->eviction_fields) {
6890 struct eviction_group *evg, *next;
6892 HMAP_FOR_EACH_SAFE (evg, next, id_node,
6893 &table->eviction_groups_by_id) {
6894 eviction_group_destroy(table, evg);
6896 hmap_destroy(&table->eviction_groups_by_id);
6897 heap_destroy(&table->eviction_groups_by_size);
6899 free(table->eviction_fields);
6900 table->eviction_fields = NULL;
6901 table->n_eviction_fields = 0;
6905 /* oftables support a choice of two policies when adding a rule would cause the
6906 * number of flows in the table to exceed the configured maximum number: either
6907 * they can refuse to add the new flow or they can evict some existing flow.
6908 * This function configures the latter policy on 'table', with fairness based
6909 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
6910 * 'n_fields' as 0 disables fairness.) */
6912 oftable_enable_eviction(struct oftable *table,
6913 const struct mf_subfield *fields, size_t n_fields)
6914 OVS_REQUIRES(ofproto_mutex)
6916 struct cls_cursor cursor;
6919 if (table->eviction_fields
6920 && n_fields == table->n_eviction_fields
6922 || !memcmp(fields, table->eviction_fields,
6923 n_fields * sizeof *fields))) {
6928 oftable_disable_eviction(table);
6930 table->n_eviction_fields = n_fields;
6931 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
6933 table->eviction_group_id_basis = random_uint32();
6934 hmap_init(&table->eviction_groups_by_id);
6935 heap_init(&table->eviction_groups_by_size);
6937 fat_rwlock_rdlock(&table->cls.rwlock);
6938 cls_cursor_init(&cursor, &table->cls, NULL);
6939 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
6940 eviction_group_add_rule(rule);
6942 fat_rwlock_unlock(&table->cls.rwlock);
6945 /* Removes 'rule' from the oftable that contains it. */
6947 oftable_remove_rule__(struct ofproto *ofproto, struct rule *rule)
6948 OVS_REQUIRES(ofproto_mutex)
6950 struct classifier *cls = &ofproto->tables[rule->table_id].cls;
6952 fat_rwlock_wrlock(&cls->rwlock);
6953 classifier_remove(cls, CONST_CAST(struct cls_rule *, &rule->cr));
6954 fat_rwlock_unlock(&cls->rwlock);
6956 cookies_remove(ofproto, rule);
6958 eviction_group_remove_rule(rule);
6959 if (!list_is_empty(&rule->expirable)) {
6960 list_remove(&rule->expirable);
6962 if (!list_is_empty(&rule->meter_list_node)) {
6963 list_remove(&rule->meter_list_node);
6964 list_init(&rule->meter_list_node);
6969 oftable_remove_rule(struct rule *rule)
6970 OVS_REQUIRES(ofproto_mutex)
6972 oftable_remove_rule__(rule->ofproto, rule);
6975 /* Inserts 'rule' into its oftable, which must not already contain any rule for
6976 * the same cls_rule. */
6978 oftable_insert_rule(struct rule *rule)
6979 OVS_REQUIRES(ofproto_mutex)
6981 struct ofproto *ofproto = rule->ofproto;
6982 struct oftable *table = &ofproto->tables[rule->table_id];
6983 const struct rule_actions *actions;
6986 ovs_mutex_lock(&rule->mutex);
6987 may_expire = rule->hard_timeout || rule->idle_timeout;
6988 ovs_mutex_unlock(&rule->mutex);
6991 list_insert(&ofproto->expirable, &rule->expirable);
6994 cookies_insert(ofproto, rule);
6996 actions = rule_get_actions(rule);
6997 if (actions->provider_meter_id != UINT32_MAX) {
6998 uint32_t meter_id = ofpacts_get_meter(actions->ofpacts,
6999 actions->ofpacts_len);
7000 struct meter *meter = ofproto->meters[meter_id];
7001 list_insert(&meter->rules, &rule->meter_list_node);
7003 fat_rwlock_wrlock(&table->cls.rwlock);
7004 classifier_insert(&table->cls, CONST_CAST(struct cls_rule *, &rule->cr));
7005 fat_rwlock_unlock(&table->cls.rwlock);
7006 eviction_group_add_rule(rule);
7009 /* unixctl commands. */
7012 ofproto_lookup(const char *name)
7014 struct ofproto *ofproto;
7016 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
7018 if (!strcmp(ofproto->name, name)) {
7026 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
7027 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7029 struct ofproto *ofproto;
7033 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
7034 ds_put_format(&results, "%s\n", ofproto->name);
7036 unixctl_command_reply(conn, ds_cstr(&results));
7037 ds_destroy(&results);
7041 ofproto_unixctl_init(void)
7043 static bool registered;
7049 unixctl_command_register("ofproto/list", "", 0, 0,
7050 ofproto_unixctl_list, NULL);
7053 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
7055 * This is deprecated. It is only for compatibility with broken device drivers
7056 * in old versions of Linux that do not properly support VLANs when VLAN
7057 * devices are not used. When broken device drivers are no longer in
7058 * widespread use, we will delete these interfaces. */
7060 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
7061 * (exactly) by an OpenFlow rule in 'ofproto'. */
7063 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
7066 struct cls_rule target;
7067 const struct oftable *oftable;
7069 match_init_catchall(&match);
7070 match_set_vlan_vid_masked(&match, htons(VLAN_CFI), htons(VLAN_CFI));
7071 cls_rule_init(&target, &match, 0);
7073 free(ofproto->vlan_bitmap);
7074 ofproto->vlan_bitmap = bitmap_allocate(4096);
7075 ofproto->vlans_changed = false;
7077 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
7078 struct cls_cursor cursor;
7081 fat_rwlock_rdlock(&oftable->cls.rwlock);
7082 cls_cursor_init(&cursor, &oftable->cls, &target);
7083 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
7084 if (minimask_get_vid_mask(&rule->cr.match.mask) == VLAN_VID_MASK) {
7085 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
7087 bitmap_set1(vlan_bitmap, vid);
7088 bitmap_set1(ofproto->vlan_bitmap, vid);
7091 fat_rwlock_unlock(&oftable->cls.rwlock);
7095 /* Returns true if new VLANs have come into use by the flow table since the
7096 * last call to ofproto_get_vlan_usage().
7098 * We don't track when old VLANs stop being used. */
7100 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
7102 return ofproto->vlans_changed;
7105 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
7106 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
7107 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
7108 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
7109 * then the VLAN device is un-enslaved. */
7111 ofproto_port_set_realdev(struct ofproto *ofproto, ofp_port_t vlandev_ofp_port,
7112 ofp_port_t realdev_ofp_port, int vid)
7114 struct ofport *ofport;
7117 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
7119 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
7121 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
7122 ofproto->name, vlandev_ofp_port);
7126 if (!ofproto->ofproto_class->set_realdev) {
7127 if (!vlandev_ofp_port) {
7130 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
7134 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
7136 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
7137 ofproto->name, vlandev_ofp_port,
7138 netdev_get_name(ofport->netdev), ovs_strerror(error));