2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
26 #include "byte-order.h"
27 #include "classifier.h"
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"
49 #include "poll-loop.h"
57 #include "unaligned.h"
61 VLOG_DEFINE_THIS_MODULE(ofproto);
63 COVERAGE_DEFINE(ofproto_flush);
64 COVERAGE_DEFINE(ofproto_packet_out);
65 COVERAGE_DEFINE(ofproto_queue_req);
66 COVERAGE_DEFINE(ofproto_recv_openflow);
67 COVERAGE_DEFINE(ofproto_reinit_ports);
68 COVERAGE_DEFINE(ofproto_update_port);
71 S_OPENFLOW, /* Processing OpenFlow commands. */
72 S_EVICT, /* Evicting flows from over-limit tables. */
73 S_FLUSH, /* Deleting all flow table rules. */
76 enum ofoperation_type {
83 /* A single OpenFlow request can execute any number of operations. The
84 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
85 * ofconn to which an error reply should be sent if necessary.
87 * ofproto initiates some operations internally. These operations are still
88 * assigned to groups but will not have an associated ofconn. */
90 struct ofproto *ofproto; /* Owning ofproto. */
91 struct list ofproto_node; /* In ofproto's "pending" list. */
92 struct list ops; /* List of "struct ofoperation"s. */
93 int n_running; /* Number of ops still pending. */
95 /* Data needed to send OpenFlow reply on failure or to send a buffered
98 * If list_is_empty(ofconn_node) then this ofopgroup never had an
99 * associated ofconn or its ofconn's connection dropped after it initiated
100 * the operation. In the latter case 'ofconn' is a wild pointer that
101 * refers to freed memory, so the 'ofconn' member must be used only if
102 * !list_is_empty(ofconn_node).
104 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
105 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
106 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
107 uint32_t buffer_id; /* Buffer id from original request. */
110 static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
111 static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
112 const struct ofp_header *,
114 static void ofopgroup_submit(struct ofopgroup *);
115 static void ofopgroup_complete(struct ofopgroup *);
117 /* A single flow table operation. */
119 struct ofopgroup *group; /* Owning group. */
120 struct list group_node; /* In ofopgroup's "ops" list. */
121 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
122 struct rule *rule; /* Rule being operated upon. */
123 enum ofoperation_type type; /* Type of operation. */
125 /* OFOPERATION_MODIFY, OFOPERATION_REPLACE: The old actions, if the actions
127 struct rule_actions *actions;
129 /* OFOPERATION_DELETE. */
130 enum ofp_flow_removed_reason reason; /* Reason flow was removed. */
132 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
133 uint16_t idle_timeout; /* Rule's old idle timeout. */
134 uint16_t hard_timeout; /* Rule's old hard timeout. */
135 enum ofputil_flow_mod_flags flags; /* Rule's old flags. */
136 enum ofperr error; /* 0 if no error. */
139 static struct ofoperation *ofoperation_create(struct ofopgroup *,
141 enum ofoperation_type,
142 enum ofp_flow_removed_reason);
143 static void ofoperation_destroy(struct ofoperation *);
146 static void oftable_init(struct oftable *);
147 static void oftable_destroy(struct oftable *);
149 static void oftable_set_name(struct oftable *, const char *name);
151 static void oftable_disable_eviction(struct oftable *);
152 static void oftable_enable_eviction(struct oftable *,
153 const struct mf_subfield *fields,
156 static void oftable_remove_rule(struct rule *rule) OVS_REQUIRES(ofproto_mutex);
157 static void oftable_remove_rule__(struct ofproto *, struct rule *)
158 OVS_REQUIRES(ofproto_mutex);
159 static void oftable_insert_rule(struct rule *);
161 /* A set of rules within a single OpenFlow table (oftable) that have the same
162 * values for the oftable's eviction_fields. A rule to be evicted, when one is
163 * needed, is taken from the eviction group that contains the greatest number
166 * An oftable owns any number of eviction groups, each of which contains any
169 * Membership in an eviction group is imprecise, based on the hash of the
170 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
171 * That is, if two rules have different eviction_fields, but those
172 * eviction_fields hash to the same value, then they will belong to the same
173 * eviction_group anyway.
175 * (When eviction is not enabled on an oftable, we don't track any eviction
176 * groups, to save time and space.) */
177 struct eviction_group {
178 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
179 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
180 struct heap rules; /* Contains "struct rule"s. */
183 static bool choose_rule_to_evict(struct oftable *table, struct rule **rulep);
184 static void ofproto_evict(struct ofproto *) OVS_EXCLUDED(ofproto_mutex);
185 static uint32_t rule_eviction_priority(struct rule *);
186 static void eviction_group_add_rule(struct rule *);
187 static void eviction_group_remove_rule(struct rule *);
189 /* Criteria that flow_mod and other operations use for selecting rules on
190 * which to operate. */
191 struct rule_criteria {
192 /* An OpenFlow table or 255 for all tables. */
195 /* OpenFlow matching criteria. Interpreted different in "loose" way by
196 * collect_rules_loose() and "strict" way by collect_rules_strict(), as
197 * defined in the OpenFlow spec. */
200 /* Matching criteria for the OpenFlow cookie. Consider a bit B in a rule's
201 * cookie and the corresponding bits C in 'cookie' and M in 'cookie_mask'.
202 * The rule will not be selected if M is 1 and B != C. */
204 ovs_be64 cookie_mask;
206 /* Selection based on actions within a rule:
208 * If out_port != OFPP_ANY, selects only rules that output to out_port.
209 * If out_group != OFPG_ALL, select only rules that output to out_group. */
214 static void rule_criteria_init(struct rule_criteria *, uint8_t table_id,
215 const struct match *match,
216 unsigned int priority,
217 ovs_be64 cookie, ovs_be64 cookie_mask,
218 ofp_port_t out_port, uint32_t out_group);
219 static void rule_criteria_destroy(struct rule_criteria *);
221 /* A packet that needs to be passed to rule_execute().
223 * (We can't do this immediately from ofopgroup_complete() because that holds
224 * ofproto_mutex, which rule_execute() needs released.) */
225 struct rule_execute {
226 struct list list_node; /* In struct ofproto's "rule_executes" list. */
227 struct rule *rule; /* Owns a reference to the rule. */
229 struct ofpbuf *packet; /* Owns the packet. */
232 static void run_rule_executes(struct ofproto *) OVS_EXCLUDED(ofproto_mutex);
233 static void destroy_rule_executes(struct ofproto *);
236 static void ofport_destroy__(struct ofport *) OVS_EXCLUDED(ofproto_mutex);
237 static void ofport_destroy(struct ofport *);
239 static void update_port(struct ofproto *, const char *devname);
240 static int init_ports(struct ofproto *);
241 static void reinit_ports(struct ofproto *);
243 static long long int ofport_get_usage(const struct ofproto *,
244 ofp_port_t ofp_port);
245 static void ofport_set_usage(struct ofproto *, ofp_port_t ofp_port,
246 long long int last_used);
247 static void ofport_remove_usage(struct ofproto *, ofp_port_t ofp_port);
251 * Keeps track of the currently used and recently used ofport values and is
252 * used to prevent immediate recycling of ofport values. */
253 struct ofport_usage {
254 struct hmap_node hmap_node; /* In struct ofproto's "ofport_usage" hmap. */
255 ofp_port_t ofp_port; /* OpenFlow port number. */
256 long long int last_used; /* Last time the 'ofp_port' was used. LLONG_MAX
257 represents in-use ofports. */
261 static void ofproto_rule_destroy__(struct rule *);
262 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
263 static bool rule_is_modifiable(const struct rule *);
266 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
267 struct ofputil_flow_mod *,
268 const struct ofp_header *);
269 static enum ofperr modify_flows__(struct ofproto *, struct ofconn *,
270 struct ofputil_flow_mod *,
271 const struct ofp_header *,
272 const struct rule_collection *);
273 static void delete_flow__(struct rule *rule, struct ofopgroup *,
274 enum ofp_flow_removed_reason)
275 OVS_REQUIRES(ofproto_mutex);
276 static bool ofproto_group_exists__(const struct ofproto *ofproto,
278 OVS_REQ_RDLOCK(ofproto->groups_rwlock);
279 static bool ofproto_group_exists(const struct ofproto *ofproto,
281 OVS_EXCLUDED(ofproto->groups_rwlock);
282 static enum ofperr add_group(struct ofproto *, struct ofputil_group_mod *);
283 static bool handle_openflow(struct ofconn *, const struct ofpbuf *);
284 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
285 struct ofputil_flow_mod *,
286 const struct ofp_header *)
287 OVS_EXCLUDED(ofproto_mutex);
288 static void calc_duration(long long int start, long long int now,
289 uint32_t *sec, uint32_t *nsec);
292 static uint64_t pick_datapath_id(const struct ofproto *);
293 static uint64_t pick_fallback_dpid(void);
294 static void ofproto_destroy__(struct ofproto *);
295 static void update_mtu(struct ofproto *, struct ofport *);
296 static void meter_delete(struct ofproto *, uint32_t first, uint32_t last);
299 static void ofproto_unixctl_init(void);
301 /* All registered ofproto classes, in probe order. */
302 static const struct ofproto_class **ofproto_classes;
303 static size_t n_ofproto_classes;
304 static size_t allocated_ofproto_classes;
306 /* Global lock that protects all flow table operations. */
307 struct ovs_mutex ofproto_mutex = OVS_MUTEX_INITIALIZER;
309 unsigned flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT;
310 unsigned n_handler_threads;
311 enum ofproto_flow_miss_model flow_miss_model = OFPROTO_HANDLE_MISS_AUTO;
313 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
314 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
316 /* Initial mappings of port to OpenFlow number mappings. */
317 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
319 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
321 /* The default value of true waits for flow restore. */
322 static bool flow_restore_wait = true;
324 /* Must be called to initialize the ofproto library.
326 * The caller may pass in 'iface_hints', which contains an shash of
327 * "iface_hint" elements indexed by the interface's name. The provider
328 * may use these hints to describe the startup configuration in order to
329 * reinitialize its state. The caller owns the provided data, so a
330 * provider will make copies of anything required. An ofproto provider
331 * will remove any existing state that is not described by the hint, and
332 * may choose to remove it all. */
334 ofproto_init(const struct shash *iface_hints)
336 struct shash_node *node;
339 ofproto_class_register(&ofproto_dpif_class);
341 /* Make a local copy, since we don't own 'iface_hints' elements. */
342 SHASH_FOR_EACH(node, iface_hints) {
343 const struct iface_hint *orig_hint = node->data;
344 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
345 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
347 new_hint->br_name = xstrdup(orig_hint->br_name);
348 new_hint->br_type = xstrdup(br_type);
349 new_hint->ofp_port = orig_hint->ofp_port;
351 shash_add(&init_ofp_ports, node->name, new_hint);
354 for (i = 0; i < n_ofproto_classes; i++) {
355 ofproto_classes[i]->init(&init_ofp_ports);
359 /* 'type' should be a normalized datapath type, as returned by
360 * ofproto_normalize_type(). Returns the corresponding ofproto_class
361 * structure, or a null pointer if there is none registered for 'type'. */
362 static const struct ofproto_class *
363 ofproto_class_find__(const char *type)
367 for (i = 0; i < n_ofproto_classes; i++) {
368 const struct ofproto_class *class = ofproto_classes[i];
373 class->enumerate_types(&types);
374 found = sset_contains(&types, type);
375 sset_destroy(&types);
381 VLOG_WARN("unknown datapath type %s", type);
385 /* Registers a new ofproto class. After successful registration, new ofprotos
386 * of that type can be created using ofproto_create(). */
388 ofproto_class_register(const struct ofproto_class *new_class)
392 for (i = 0; i < n_ofproto_classes; i++) {
393 if (ofproto_classes[i] == new_class) {
398 if (n_ofproto_classes >= allocated_ofproto_classes) {
399 ofproto_classes = x2nrealloc(ofproto_classes,
400 &allocated_ofproto_classes,
401 sizeof *ofproto_classes);
403 ofproto_classes[n_ofproto_classes++] = new_class;
407 /* Unregisters a datapath provider. 'type' must have been previously
408 * registered and not currently be in use by any ofprotos. After
409 * unregistration new datapaths of that type cannot be opened using
410 * ofproto_create(). */
412 ofproto_class_unregister(const struct ofproto_class *class)
416 for (i = 0; i < n_ofproto_classes; i++) {
417 if (ofproto_classes[i] == class) {
418 for (i++; i < n_ofproto_classes; i++) {
419 ofproto_classes[i - 1] = ofproto_classes[i];
425 VLOG_WARN("attempted to unregister an ofproto class that is not "
430 /* Clears 'types' and enumerates all registered ofproto types into it. The
431 * caller must first initialize the sset. */
433 ofproto_enumerate_types(struct sset *types)
438 for (i = 0; i < n_ofproto_classes; i++) {
439 ofproto_classes[i]->enumerate_types(types);
443 /* Returns the fully spelled out name for the given ofproto 'type'.
445 * Normalized type string can be compared with strcmp(). Unnormalized type
446 * string might be the same even if they have different spellings. */
448 ofproto_normalize_type(const char *type)
450 return type && type[0] ? type : "system";
453 /* Clears 'names' and enumerates the names of all known created ofprotos with
454 * the given 'type'. The caller must first initialize the sset. Returns 0 if
455 * successful, otherwise a positive errno value.
457 * Some kinds of datapaths might not be practically enumerable. This is not
458 * considered an error. */
460 ofproto_enumerate_names(const char *type, struct sset *names)
462 const struct ofproto_class *class = ofproto_class_find__(type);
463 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
467 ofproto_create(const char *datapath_name, const char *datapath_type,
468 struct ofproto **ofprotop)
470 const struct ofproto_class *class;
471 struct ofproto *ofproto;
477 ofproto_unixctl_init();
479 datapath_type = ofproto_normalize_type(datapath_type);
480 class = ofproto_class_find__(datapath_type);
482 VLOG_WARN("could not create datapath %s of unknown type %s",
483 datapath_name, datapath_type);
487 ofproto = class->alloc();
489 VLOG_ERR("failed to allocate datapath %s of type %s",
490 datapath_name, datapath_type);
495 ovs_mutex_lock(&ofproto_mutex);
496 memset(ofproto, 0, sizeof *ofproto);
497 ofproto->ofproto_class = class;
498 ofproto->name = xstrdup(datapath_name);
499 ofproto->type = xstrdup(datapath_type);
500 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
501 hash_string(ofproto->name, 0));
502 ofproto->datapath_id = 0;
503 ofproto->forward_bpdu = false;
504 ofproto->fallback_dpid = pick_fallback_dpid();
505 ofproto->mfr_desc = NULL;
506 ofproto->hw_desc = NULL;
507 ofproto->sw_desc = NULL;
508 ofproto->serial_desc = NULL;
509 ofproto->dp_desc = NULL;
510 ofproto->frag_handling = OFPC_FRAG_NORMAL;
511 hmap_init(&ofproto->ports);
512 hmap_init(&ofproto->ofport_usage);
513 shash_init(&ofproto->port_by_name);
514 simap_init(&ofproto->ofp_requests);
515 ofproto->max_ports = ofp_to_u16(OFPP_MAX);
516 ofproto->eviction_group_timer = LLONG_MIN;
517 ofproto->tables = NULL;
518 ofproto->n_tables = 0;
519 hindex_init(&ofproto->cookies);
520 list_init(&ofproto->expirable);
521 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
522 ofproto->state = S_OPENFLOW;
523 list_init(&ofproto->pending);
524 ofproto->n_pending = 0;
525 hmap_init(&ofproto->deletions);
526 guarded_list_init(&ofproto->rule_executes);
527 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
528 ofproto->first_op = ofproto->last_op = LLONG_MIN;
529 ofproto->next_op_report = LLONG_MAX;
530 ofproto->op_backoff = LLONG_MIN;
531 ofproto->vlan_bitmap = NULL;
532 ofproto->vlans_changed = false;
533 ofproto->min_mtu = INT_MAX;
534 ovs_rwlock_init(&ofproto->groups_rwlock);
535 hmap_init(&ofproto->groups);
536 ovs_mutex_unlock(&ofproto_mutex);
537 ofproto->ogf.capabilities = OFPGFC_CHAINING | OFPGFC_SELECT_LIVENESS |
538 OFPGFC_SELECT_WEIGHT;
539 ofproto->ogf.max_groups[OFPGT11_ALL] = OFPG_MAX;
540 ofproto->ogf.max_groups[OFPGT11_SELECT] = OFPG_MAX;
541 ofproto->ogf.max_groups[OFPGT11_INDIRECT] = OFPG_MAX;
542 ofproto->ogf.max_groups[OFPGT11_FF] = OFPG_MAX;
543 ofproto->ogf.actions[0] =
544 (1 << OFPAT11_OUTPUT) |
545 (1 << OFPAT11_COPY_TTL_OUT) |
546 (1 << OFPAT11_COPY_TTL_IN) |
547 (1 << OFPAT11_SET_MPLS_TTL) |
548 (1 << OFPAT11_DEC_MPLS_TTL) |
549 (1 << OFPAT11_PUSH_VLAN) |
550 (1 << OFPAT11_POP_VLAN) |
551 (1 << OFPAT11_PUSH_MPLS) |
552 (1 << OFPAT11_POP_MPLS) |
553 (1 << OFPAT11_SET_QUEUE) |
554 (1 << OFPAT11_GROUP) |
555 (1 << OFPAT11_SET_NW_TTL) |
556 (1 << OFPAT11_DEC_NW_TTL) |
557 (1 << OFPAT12_SET_FIELD);
559 * (1 << OFPAT13_PUSH_PBB) |
560 * (1 << OFPAT13_POP_PBB) */
562 error = ofproto->ofproto_class->construct(ofproto);
564 VLOG_ERR("failed to open datapath %s: %s",
565 datapath_name, ovs_strerror(error));
566 ofproto_destroy__(ofproto);
570 /* Check that hidden tables, if any, are at the end. */
571 ovs_assert(ofproto->n_tables);
572 for (i = 0; i + 1 < ofproto->n_tables; i++) {
573 enum oftable_flags flags = ofproto->tables[i].flags;
574 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
576 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
579 ofproto->datapath_id = pick_datapath_id(ofproto);
582 /* Initialize meters table. */
583 if (ofproto->ofproto_class->meter_get_features) {
584 ofproto->ofproto_class->meter_get_features(ofproto,
585 &ofproto->meter_features);
587 memset(&ofproto->meter_features, 0, sizeof ofproto->meter_features);
589 ofproto->meters = xzalloc((ofproto->meter_features.max_meters + 1)
590 * sizeof(struct meter *));
596 /* Must be called (only) by an ofproto implementation in its constructor
597 * function. See the large comment on 'construct' in struct ofproto_class for
600 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
602 struct oftable *table;
604 ovs_assert(!ofproto->n_tables);
605 ovs_assert(n_tables >= 1 && n_tables <= 255);
607 ofproto->n_tables = n_tables;
608 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
609 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
614 /* To be optionally called (only) by an ofproto implementation in its
615 * constructor function. See the large comment on 'construct' in struct
616 * ofproto_class for details.
618 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
619 * will then ensure that actions passed into the ofproto implementation will
620 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
621 * function is not called, there will be no such restriction.
623 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
624 * the 'max_ports' restriction. */
626 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
628 ovs_assert(max_ports <= ofp_to_u16(OFPP_MAX));
629 ofproto->max_ports = max_ports;
633 ofproto_get_datapath_id(const struct ofproto *ofproto)
635 return ofproto->datapath_id;
639 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
641 uint64_t old_dpid = p->datapath_id;
642 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
643 if (p->datapath_id != old_dpid) {
644 /* Force all active connections to reconnect, since there is no way to
645 * notify a controller that the datapath ID has changed. */
646 ofproto_reconnect_controllers(p);
651 ofproto_set_controllers(struct ofproto *p,
652 const struct ofproto_controller *controllers,
653 size_t n_controllers, uint32_t allowed_versions)
655 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
660 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
662 connmgr_set_fail_mode(p->connmgr, fail_mode);
665 /* Drops the connections between 'ofproto' and all of its controllers, forcing
666 * them to reconnect. */
668 ofproto_reconnect_controllers(struct ofproto *ofproto)
670 connmgr_reconnect(ofproto->connmgr);
673 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
674 * in-band control should guarantee access, in the same way that in-band
675 * control guarantees access to OpenFlow controllers. */
677 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
678 const struct sockaddr_in *extras, size_t n)
680 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
683 /* Sets the OpenFlow queue used by flows set up by in-band control on
684 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
685 * flows will use the default queue. */
687 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
689 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
692 /* Sets the number of flows at which eviction from the kernel flow table
695 ofproto_set_flow_eviction_threshold(unsigned threshold)
697 flow_eviction_threshold = MAX(OFPROTO_FLOW_EVICTION_THRESHOLD_MIN,
701 /* Sets the path for handling flow misses. */
703 ofproto_set_flow_miss_model(unsigned model)
705 flow_miss_model = model;
708 /* If forward_bpdu is true, the NORMAL action will forward frames with
709 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
710 * the NORMAL action will drop these frames. */
712 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
714 bool old_val = ofproto->forward_bpdu;
715 ofproto->forward_bpdu = forward_bpdu;
716 if (old_val != ofproto->forward_bpdu) {
717 if (ofproto->ofproto_class->forward_bpdu_changed) {
718 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
723 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
724 * 'idle_time', in seconds, and the maximum number of MAC table entries to
727 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
730 if (ofproto->ofproto_class->set_mac_table_config) {
731 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
736 /* Sets number of upcall handler threads. The default is
737 * (number of online cores - 2). */
739 ofproto_set_n_handler_threads(unsigned limit)
742 n_handler_threads = limit;
744 int n_proc = count_cpu_cores();
745 n_handler_threads = n_proc > 2 ? n_proc - 2 : 1;
750 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
753 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
757 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
759 return connmgr_set_snoops(ofproto->connmgr, snoops);
763 ofproto_set_netflow(struct ofproto *ofproto,
764 const struct netflow_options *nf_options)
766 if (nf_options && sset_is_empty(&nf_options->collectors)) {
770 if (ofproto->ofproto_class->set_netflow) {
771 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
773 return nf_options ? EOPNOTSUPP : 0;
778 ofproto_set_sflow(struct ofproto *ofproto,
779 const struct ofproto_sflow_options *oso)
781 if (oso && sset_is_empty(&oso->targets)) {
785 if (ofproto->ofproto_class->set_sflow) {
786 return ofproto->ofproto_class->set_sflow(ofproto, oso);
788 return oso ? EOPNOTSUPP : 0;
793 ofproto_set_ipfix(struct ofproto *ofproto,
794 const struct ofproto_ipfix_bridge_exporter_options *bo,
795 const struct ofproto_ipfix_flow_exporter_options *fo,
798 if (ofproto->ofproto_class->set_ipfix) {
799 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
801 return (bo || fo) ? EOPNOTSUPP : 0;
806 ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
808 flow_restore_wait = flow_restore_wait_db;
812 ofproto_get_flow_restore_wait(void)
814 return flow_restore_wait;
818 /* Spanning Tree Protocol (STP) configuration. */
820 /* Configures STP on 'ofproto' using the settings defined in 's'. If
821 * 's' is NULL, disables STP.
823 * Returns 0 if successful, otherwise a positive errno value. */
825 ofproto_set_stp(struct ofproto *ofproto,
826 const struct ofproto_stp_settings *s)
828 return (ofproto->ofproto_class->set_stp
829 ? ofproto->ofproto_class->set_stp(ofproto, s)
833 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
834 * 'enabled' member of 's' is false, then the other members are not
837 * Returns 0 if successful, otherwise a positive errno value. */
839 ofproto_get_stp_status(struct ofproto *ofproto,
840 struct ofproto_stp_status *s)
842 return (ofproto->ofproto_class->get_stp_status
843 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
847 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
848 * in 's'. The caller is responsible for assigning STP port numbers
849 * (using the 'port_num' member in the range of 1 through 255, inclusive)
850 * and ensuring there are no duplicates. If the 's' is NULL, then STP
851 * is disabled on the port.
853 * Returns 0 if successful, otherwise a positive errno value.*/
855 ofproto_port_set_stp(struct ofproto *ofproto, ofp_port_t ofp_port,
856 const struct ofproto_port_stp_settings *s)
858 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
860 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
861 ofproto->name, ofp_port);
865 return (ofproto->ofproto_class->set_stp_port
866 ? ofproto->ofproto_class->set_stp_port(ofport, s)
870 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
871 * 's'. If the 'enabled' member in 's' is false, then the other members
872 * are not meaningful.
874 * Returns 0 if successful, otherwise a positive errno value.*/
876 ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
877 struct ofproto_port_stp_status *s)
879 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
881 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
882 "port %"PRIu16, ofproto->name, ofp_port);
886 return (ofproto->ofproto_class->get_stp_port_status
887 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
891 /* Retrieves STP port statistics of 'ofp_port' on 'ofproto' and stores it in
892 * 's'. If the 'enabled' member in 's' is false, then the other members
893 * are not meaningful.
895 * Returns 0 if successful, otherwise a positive errno value.*/
897 ofproto_port_get_stp_stats(struct ofproto *ofproto, ofp_port_t ofp_port,
898 struct ofproto_port_stp_stats *s)
900 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
902 VLOG_WARN_RL(&rl, "%s: cannot get STP stats on nonexistent "
903 "port %"PRIu16, ofproto->name, ofp_port);
907 return (ofproto->ofproto_class->get_stp_port_stats
908 ? ofproto->ofproto_class->get_stp_port_stats(ofport, s)
912 /* Queue DSCP configuration. */
914 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
915 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
916 * to implement QoS. Instead, it is used to implement features which require
917 * knowledge of what queues exist on a port, and some basic information about
920 * Returns 0 if successful, otherwise a positive errno value. */
922 ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port,
923 const struct ofproto_port_queue *queues,
926 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
929 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
930 ofproto->name, ofp_port);
934 return (ofproto->ofproto_class->set_queues
935 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
939 /* Connectivity Fault Management configuration. */
941 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
943 ofproto_port_clear_cfm(struct ofproto *ofproto, ofp_port_t ofp_port)
945 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
946 if (ofport && ofproto->ofproto_class->set_cfm) {
947 ofproto->ofproto_class->set_cfm(ofport, NULL);
951 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
952 * basic configuration from the configuration members in 'cfm', and the remote
953 * maintenance point ID from remote_mpid. Ignores the statistics members of
956 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
958 ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port,
959 const struct cfm_settings *s)
961 struct ofport *ofport;
964 ofport = ofproto_get_port(ofproto, ofp_port);
966 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
967 ofproto->name, ofp_port);
971 /* XXX: For configuration simplicity, we only support one remote_mpid
972 * outside of the CFM module. It's not clear if this is the correct long
973 * term solution or not. */
974 error = (ofproto->ofproto_class->set_cfm
975 ? ofproto->ofproto_class->set_cfm(ofport, s)
978 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
979 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
980 ovs_strerror(error));
984 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
985 * 'ofproto' does not have a port 'ofp_port'. */
987 ofproto_port_set_bfd(struct ofproto *ofproto, ofp_port_t ofp_port,
988 const struct smap *cfg)
990 struct ofport *ofport;
993 ofport = ofproto_get_port(ofproto, ofp_port);
995 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
996 ofproto->name, ofp_port);
1000 error = (ofproto->ofproto_class->set_bfd
1001 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
1004 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
1005 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
1006 ovs_strerror(error));
1010 /* Populates 'status' with key value pairs indicating the status of the BFD
1011 * session on 'ofp_port'. This information is intended to be populated in the
1012 * OVS database. Has no effect if 'ofp_port' is not na OpenFlow port in
1015 ofproto_port_get_bfd_status(struct ofproto *ofproto, ofp_port_t ofp_port,
1016 struct smap *status)
1018 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1019 return (ofport && ofproto->ofproto_class->get_bfd_status
1020 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
1024 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
1025 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
1026 * 0 if LACP partner information is not current (generally indicating a
1027 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
1029 ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port)
1031 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1032 return (ofport && ofproto->ofproto_class->port_is_lacp_current
1033 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
1039 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
1040 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
1041 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
1042 * configuration plus, if there is more than one slave, a bonding
1045 * If 'aux' is already registered then this function updates its configuration
1046 * to 's'. Otherwise, this function registers a new bundle.
1048 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
1051 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
1052 const struct ofproto_bundle_settings *s)
1054 return (ofproto->ofproto_class->bundle_set
1055 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
1059 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
1060 * If no such bundle has been registered, this has no effect. */
1062 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
1064 return ofproto_bundle_register(ofproto, aux, NULL);
1068 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
1069 * If 'aux' is already registered then this function updates its configuration
1070 * to 's'. Otherwise, this function registers a new mirror. */
1072 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
1073 const struct ofproto_mirror_settings *s)
1075 return (ofproto->ofproto_class->mirror_set
1076 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
1080 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
1081 * If no mirror has been registered, this has no effect. */
1083 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
1085 return ofproto_mirror_register(ofproto, aux, NULL);
1088 /* Retrieves statistics from mirror associated with client data pointer
1089 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
1090 * 'bytes', respectively. If a particular counters is not supported,
1091 * the appropriate argument is set to UINT64_MAX. */
1093 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
1094 uint64_t *packets, uint64_t *bytes)
1096 if (!ofproto->ofproto_class->mirror_get_stats) {
1097 *packets = *bytes = UINT64_MAX;
1101 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
1105 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
1106 * which all packets are flooded, instead of using MAC learning. If
1107 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1109 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
1112 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
1114 return (ofproto->ofproto_class->set_flood_vlans
1115 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
1119 /* Returns true if 'aux' is a registered bundle that is currently in use as the
1120 * output for a mirror. */
1122 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
1124 return (ofproto->ofproto_class->is_mirror_output_bundle
1125 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
1129 /* Configuration of OpenFlow tables. */
1131 /* Returns the number of OpenFlow tables in 'ofproto'. */
1133 ofproto_get_n_tables(const struct ofproto *ofproto)
1135 return ofproto->n_tables;
1138 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
1139 * settings from 's'. 'table_id' must be in the range 0 through the number of
1140 * OpenFlow tables in 'ofproto' minus 1, inclusive.
1142 * For read-only tables, only the name may be configured. */
1144 ofproto_configure_table(struct ofproto *ofproto, int table_id,
1145 const struct ofproto_table_settings *s)
1147 struct oftable *table;
1149 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
1150 table = &ofproto->tables[table_id];
1152 oftable_set_name(table, s->name);
1154 if (table->flags & OFTABLE_READONLY) {
1159 oftable_enable_eviction(table, s->groups, s->n_groups);
1161 oftable_disable_eviction(table);
1164 table->max_flows = s->max_flows;
1165 ovs_rwlock_wrlock(&table->cls.rwlock);
1166 if (classifier_count(&table->cls) > table->max_flows
1167 && table->eviction_fields) {
1168 /* 'table' contains more flows than allowed. We might not be able to
1169 * evict them right away because of the asynchronous nature of flow
1170 * table changes. Schedule eviction for later. */
1171 switch (ofproto->state) {
1173 ofproto->state = S_EVICT;
1177 /* We're already deleting flows, nothing more to do. */
1182 classifier_set_prefix_fields(&table->cls,
1183 s->prefix_fields, s->n_prefix_fields);
1185 ovs_rwlock_unlock(&table->cls.rwlock);
1189 ofproto_has_snoops(const struct ofproto *ofproto)
1191 return connmgr_has_snoops(ofproto->connmgr);
1195 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1197 connmgr_get_snoops(ofproto->connmgr, snoops);
1201 ofproto_rule_delete__(struct ofproto *ofproto, struct rule *rule,
1203 OVS_REQUIRES(ofproto_mutex)
1205 struct ofopgroup *group;
1207 ovs_assert(!rule->pending);
1209 group = ofopgroup_create_unattached(ofproto);
1210 delete_flow__(rule, group, reason);
1211 ofopgroup_submit(group);
1214 /* Deletes 'rule' from 'cls' within 'ofproto'.
1216 * Within an ofproto implementation, this function allows an ofproto
1217 * implementation to destroy any rules that remain when its ->destruct()
1218 * function is called. This function is not suitable for use elsewhere in an
1219 * ofproto implementation.
1221 * This function implements steps 4.4 and 4.5 in the section titled "Rule Life
1222 * Cycle" in ofproto-provider.h. */
1224 ofproto_rule_delete(struct ofproto *ofproto, struct rule *rule)
1225 OVS_EXCLUDED(ofproto_mutex)
1227 struct ofopgroup *group;
1229 ovs_mutex_lock(&ofproto_mutex);
1230 ovs_assert(!rule->pending);
1232 group = ofopgroup_create_unattached(ofproto);
1233 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1234 oftable_remove_rule__(ofproto, rule);
1235 ofproto->ofproto_class->rule_delete(rule);
1236 ofopgroup_submit(group);
1238 ovs_mutex_unlock(&ofproto_mutex);
1242 ofproto_flush__(struct ofproto *ofproto)
1243 OVS_EXCLUDED(ofproto_mutex)
1245 struct oftable *table;
1247 if (ofproto->ofproto_class->flush) {
1248 ofproto->ofproto_class->flush(ofproto);
1251 ovs_mutex_lock(&ofproto_mutex);
1252 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1253 struct rule *rule, *next_rule;
1254 struct cls_cursor cursor;
1256 if (table->flags & OFTABLE_HIDDEN) {
1260 ovs_rwlock_rdlock(&table->cls.rwlock);
1261 cls_cursor_init(&cursor, &table->cls, NULL);
1262 ovs_rwlock_unlock(&table->cls.rwlock);
1263 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1264 if (!rule->pending) {
1265 ofproto_rule_delete__(ofproto, rule, OFPRR_DELETE);
1269 ovs_mutex_unlock(&ofproto_mutex);
1272 static void delete_group(struct ofproto *ofproto, uint32_t group_id);
1275 ofproto_destroy__(struct ofproto *ofproto)
1276 OVS_EXCLUDED(ofproto_mutex)
1278 struct oftable *table;
1280 ovs_assert(list_is_empty(&ofproto->pending));
1282 destroy_rule_executes(ofproto);
1283 guarded_list_destroy(&ofproto->rule_executes);
1285 delete_group(ofproto, OFPG_ALL);
1286 ovs_rwlock_destroy(&ofproto->groups_rwlock);
1287 hmap_destroy(&ofproto->groups);
1289 connmgr_destroy(ofproto->connmgr);
1291 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1292 free(ofproto->name);
1293 free(ofproto->type);
1294 free(ofproto->mfr_desc);
1295 free(ofproto->hw_desc);
1296 free(ofproto->sw_desc);
1297 free(ofproto->serial_desc);
1298 free(ofproto->dp_desc);
1299 hmap_destroy(&ofproto->ports);
1300 hmap_destroy(&ofproto->ofport_usage);
1301 shash_destroy(&ofproto->port_by_name);
1302 simap_destroy(&ofproto->ofp_requests);
1304 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1305 oftable_destroy(table);
1307 free(ofproto->tables);
1309 hmap_destroy(&ofproto->deletions);
1311 free(ofproto->vlan_bitmap);
1313 ofproto->ofproto_class->dealloc(ofproto);
1317 ofproto_destroy(struct ofproto *p)
1318 OVS_EXCLUDED(ofproto_mutex)
1320 struct ofport *ofport, *next_ofport;
1321 struct ofport_usage *usage, *next_usage;
1328 meter_delete(p, 1, p->meter_features.max_meters);
1329 p->meter_features.max_meters = 0;
1335 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1336 ofport_destroy(ofport);
1339 HMAP_FOR_EACH_SAFE (usage, next_usage, hmap_node, &p->ofport_usage) {
1340 hmap_remove(&p->ofport_usage, &usage->hmap_node);
1344 p->ofproto_class->destruct(p);
1345 ofproto_destroy__(p);
1348 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1349 * kernel datapath, for example, this destroys the datapath in the kernel, and
1350 * with the netdev-based datapath, it tears down the data structures that
1351 * represent the datapath.
1353 * The datapath should not be currently open as an ofproto. */
1355 ofproto_delete(const char *name, const char *type)
1357 const struct ofproto_class *class = ofproto_class_find__(type);
1358 return (!class ? EAFNOSUPPORT
1359 : !class->del ? EACCES
1360 : class->del(type, name));
1364 process_port_change(struct ofproto *ofproto, int error, char *devname)
1366 if (error == ENOBUFS) {
1367 reinit_ports(ofproto);
1368 } else if (!error) {
1369 update_port(ofproto, devname);
1375 ofproto_type_run(const char *datapath_type)
1377 const struct ofproto_class *class;
1380 datapath_type = ofproto_normalize_type(datapath_type);
1381 class = ofproto_class_find__(datapath_type);
1383 error = class->type_run ? class->type_run(datapath_type) : 0;
1384 if (error && error != EAGAIN) {
1385 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1386 datapath_type, ovs_strerror(error));
1392 ofproto_type_run_fast(const char *datapath_type)
1394 const struct ofproto_class *class;
1397 datapath_type = ofproto_normalize_type(datapath_type);
1398 class = ofproto_class_find__(datapath_type);
1400 error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
1401 if (error && error != EAGAIN) {
1402 VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
1403 datapath_type, ovs_strerror(error));
1409 ofproto_type_wait(const char *datapath_type)
1411 const struct ofproto_class *class;
1413 datapath_type = ofproto_normalize_type(datapath_type);
1414 class = ofproto_class_find__(datapath_type);
1416 if (class->type_wait) {
1417 class->type_wait(datapath_type);
1422 any_pending_ops(const struct ofproto *p)
1423 OVS_EXCLUDED(ofproto_mutex)
1427 ovs_mutex_lock(&ofproto_mutex);
1428 b = !list_is_empty(&p->pending);
1429 ovs_mutex_unlock(&ofproto_mutex);
1435 ofproto_run(struct ofproto *p)
1440 error = p->ofproto_class->run(p);
1441 if (error && error != EAGAIN) {
1442 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, ovs_strerror(error));
1445 run_rule_executes(p);
1447 /* Restore the eviction group heap invariant occasionally. */
1448 if (p->eviction_group_timer < time_msec()) {
1451 p->eviction_group_timer = time_msec() + 1000;
1453 for (i = 0; i < p->n_tables; i++) {
1454 struct oftable *table = &p->tables[i];
1455 struct eviction_group *evg;
1456 struct cls_cursor cursor;
1459 if (!table->eviction_fields) {
1463 ovs_mutex_lock(&ofproto_mutex);
1464 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
1465 heap_rebuild(&evg->rules);
1468 ovs_rwlock_rdlock(&table->cls.rwlock);
1469 cls_cursor_init(&cursor, &table->cls, NULL);
1470 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1471 if (!rule->eviction_group
1472 && (rule->idle_timeout || rule->hard_timeout)) {
1473 eviction_group_add_rule(rule);
1476 ovs_rwlock_unlock(&table->cls.rwlock);
1477 ovs_mutex_unlock(&ofproto_mutex);
1481 if (p->ofproto_class->port_poll) {
1484 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1485 process_port_change(p, error, devname);
1489 new_seq = seq_read(connectivity_seq_get());
1490 if (new_seq != p->change_seq) {
1491 struct sset devnames;
1492 const char *devname;
1493 struct ofport *ofport;
1495 /* Update OpenFlow port status for any port whose netdev has changed.
1497 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1498 * destroyed, so it's not safe to update ports directly from the
1499 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1500 * need this two-phase approach. */
1501 sset_init(&devnames);
1502 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1503 sset_add(&devnames, netdev_get_name(ofport->netdev));
1505 SSET_FOR_EACH (devname, &devnames) {
1506 update_port(p, devname);
1508 sset_destroy(&devnames);
1510 p->change_seq = new_seq;
1515 connmgr_run(p->connmgr, handle_openflow);
1519 connmgr_run(p->connmgr, NULL);
1521 if (!any_pending_ops(p)) {
1522 p->state = S_OPENFLOW;
1527 connmgr_run(p->connmgr, NULL);
1529 if (!any_pending_ops(p)) {
1530 connmgr_flushed(p->connmgr);
1531 p->state = S_OPENFLOW;
1539 if (time_msec() >= p->next_op_report) {
1540 long long int ago = (time_msec() - p->first_op) / 1000;
1541 long long int interval = (p->last_op - p->first_op) / 1000;
1545 ds_put_format(&s, "%d flow_mods ",
1546 p->n_add + p->n_delete + p->n_modify);
1547 if (interval == ago) {
1548 ds_put_format(&s, "in the last %lld s", ago);
1549 } else if (interval) {
1550 ds_put_format(&s, "in the %lld s starting %lld s ago",
1553 ds_put_format(&s, "%lld s ago", ago);
1556 ds_put_cstr(&s, " (");
1558 ds_put_format(&s, "%d adds, ", p->n_add);
1561 ds_put_format(&s, "%d deletes, ", p->n_delete);
1564 ds_put_format(&s, "%d modifications, ", p->n_modify);
1567 ds_put_char(&s, ')');
1569 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1572 p->n_add = p->n_delete = p->n_modify = 0;
1573 p->next_op_report = LLONG_MAX;
1579 /* Performs periodic activity required by 'ofproto' that needs to be done
1580 * with the least possible latency.
1582 * It makes sense to call this function a couple of times per poll loop, to
1583 * provide a significant performance boost on some benchmarks with the
1584 * ofproto-dpif implementation. */
1586 ofproto_run_fast(struct ofproto *p)
1590 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1591 if (error && error != EAGAIN) {
1592 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1593 p->name, ovs_strerror(error));
1599 ofproto_wait(struct ofproto *p)
1601 p->ofproto_class->wait(p);
1602 if (p->ofproto_class->port_poll_wait) {
1603 p->ofproto_class->port_poll_wait(p);
1605 seq_wait(connectivity_seq_get(), p->change_seq);
1609 connmgr_wait(p->connmgr, true);
1614 connmgr_wait(p->connmgr, false);
1615 if (!any_pending_ops(p)) {
1616 poll_immediate_wake();
1623 ofproto_is_alive(const struct ofproto *p)
1625 return connmgr_has_controllers(p->connmgr);
1628 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1629 * memory_report(). */
1631 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1633 const struct oftable *table;
1634 unsigned int n_rules;
1636 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1638 ovs_mutex_lock(&ofproto_mutex);
1639 simap_increase(usage, "ops",
1640 ofproto->n_pending + hmap_count(&ofproto->deletions));
1641 ovs_mutex_unlock(&ofproto_mutex);
1644 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1645 ovs_rwlock_rdlock(&table->cls.rwlock);
1646 n_rules += classifier_count(&table->cls);
1647 ovs_rwlock_unlock(&table->cls.rwlock);
1649 simap_increase(usage, "rules", n_rules);
1651 if (ofproto->ofproto_class->get_memory_usage) {
1652 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1655 connmgr_get_memory_usage(ofproto->connmgr, usage);
1659 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1662 connmgr_get_controller_info(ofproto->connmgr, info);
1666 ofproto_free_ofproto_controller_info(struct shash *info)
1668 connmgr_free_controller_info(info);
1671 /* Makes a deep copy of 'old' into 'port'. */
1673 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1675 port->name = xstrdup(old->name);
1676 port->type = xstrdup(old->type);
1677 port->ofp_port = old->ofp_port;
1680 /* Frees memory allocated to members of 'ofproto_port'.
1682 * Do not call this function on an ofproto_port obtained from
1683 * ofproto_port_dump_next(): that function retains ownership of the data in the
1686 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1688 free(ofproto_port->name);
1689 free(ofproto_port->type);
1692 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1694 * This function provides no status indication. An error status for the entire
1695 * dump operation is provided when it is completed by calling
1696 * ofproto_port_dump_done().
1699 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1700 const struct ofproto *ofproto)
1702 dump->ofproto = ofproto;
1703 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1707 /* Attempts to retrieve another port from 'dump', which must have been created
1708 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1709 * 'port' and returns true. On failure, returns false.
1711 * Failure might indicate an actual error or merely that the last port has been
1712 * dumped. An error status for the entire dump operation is provided when it
1713 * is completed by calling ofproto_port_dump_done().
1715 * The ofproto owns the data stored in 'port'. It will remain valid until at
1716 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1717 * ofproto_port_dump_done(). */
1719 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1720 struct ofproto_port *port)
1722 const struct ofproto *ofproto = dump->ofproto;
1728 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1731 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1737 /* Completes port table dump operation 'dump', which must have been created
1738 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1739 * error-free, otherwise a positive errno value describing the problem. */
1741 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1743 const struct ofproto *ofproto = dump->ofproto;
1745 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1748 return dump->error == EOF ? 0 : dump->error;
1751 /* Returns the type to pass to netdev_open() when a datapath of type
1752 * 'datapath_type' has a port of type 'port_type', for a few special
1753 * cases when a netdev type differs from a port type. For example, when
1754 * using the userspace datapath, a port of type "internal" needs to be
1757 * Returns either 'type' itself or a string literal, which must not be
1760 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1762 const struct ofproto_class *class;
1764 datapath_type = ofproto_normalize_type(datapath_type);
1765 class = ofproto_class_find__(datapath_type);
1770 return (class->port_open_type
1771 ? class->port_open_type(datapath_type, port_type)
1775 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1776 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1777 * the port's OpenFlow port number.
1779 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1780 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1781 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1782 * 'ofp_portp' is non-null). */
1784 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1785 ofp_port_t *ofp_portp)
1787 ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1790 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1792 const char *netdev_name = netdev_get_name(netdev);
1794 simap_put(&ofproto->ofp_requests, netdev_name,
1795 ofp_to_u16(ofp_port));
1796 update_port(ofproto, netdev_name);
1799 *ofp_portp = OFPP_NONE;
1801 struct ofproto_port ofproto_port;
1803 error = ofproto_port_query_by_name(ofproto,
1804 netdev_get_name(netdev),
1807 *ofp_portp = ofproto_port.ofp_port;
1808 ofproto_port_destroy(&ofproto_port);
1815 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1816 * initializes '*port' appropriately; on failure, returns a positive errno
1819 * The caller owns the data in 'ofproto_port' and must free it with
1820 * ofproto_port_destroy() when it is no longer needed. */
1822 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1823 struct ofproto_port *port)
1827 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1829 memset(port, 0, sizeof *port);
1834 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1835 * Returns 0 if successful, otherwise a positive errno. */
1837 ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port)
1839 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1840 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1841 struct simap_node *ofp_request_node;
1844 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1845 if (ofp_request_node) {
1846 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1849 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1850 if (!error && ofport) {
1851 /* 'name' is the netdev's name and update_port() is going to close the
1852 * netdev. Just in case update_port() refers to 'name' after it
1853 * destroys 'ofport', make a copy of it around the update_port()
1855 char *devname = xstrdup(name);
1856 update_port(ofproto, devname);
1863 flow_mod_init(struct ofputil_flow_mod *fm,
1864 const struct match *match, unsigned int priority,
1865 const struct ofpact *ofpacts, size_t ofpacts_len,
1866 enum ofp_flow_mod_command command)
1868 memset(fm, 0, sizeof *fm);
1870 fm->priority = priority;
1873 fm->modify_cookie = false;
1875 fm->command = command;
1876 fm->idle_timeout = 0;
1877 fm->hard_timeout = 0;
1878 fm->buffer_id = UINT32_MAX;
1879 fm->out_port = OFPP_ANY;
1880 fm->out_group = OFPG_ANY;
1882 fm->ofpacts = CONST_CAST(struct ofpact *, ofpacts);
1883 fm->ofpacts_len = ofpacts_len;
1887 simple_flow_mod(struct ofproto *ofproto,
1888 const struct match *match, unsigned int priority,
1889 const struct ofpact *ofpacts, size_t ofpacts_len,
1890 enum ofp_flow_mod_command command)
1892 struct ofputil_flow_mod fm;
1894 flow_mod_init(&fm, match, priority, ofpacts, ofpacts_len, command);
1896 return handle_flow_mod__(ofproto, NULL, &fm, NULL);
1899 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1900 * performs the 'n_actions' actions in 'actions'. The new flow will not
1903 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1904 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1905 * controllers; otherwise, it will be hidden.
1907 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1909 * This is a helper function for in-band control and fail-open. */
1911 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1912 unsigned int priority,
1913 const struct ofpact *ofpacts, size_t ofpacts_len)
1914 OVS_EXCLUDED(ofproto_mutex)
1916 const struct rule *rule;
1919 /* First do a cheap check whether the rule we're looking for already exists
1920 * with the actions that we want. If it does, then we're done. */
1921 ovs_rwlock_rdlock(&ofproto->tables[0].cls.rwlock);
1922 rule = rule_from_cls_rule(classifier_find_match_exactly(
1923 &ofproto->tables[0].cls, match, priority));
1925 ovs_mutex_lock(&rule->mutex);
1926 must_add = !ofpacts_equal(rule->actions->ofpacts,
1927 rule->actions->ofpacts_len,
1928 ofpacts, ofpacts_len);
1929 ovs_mutex_unlock(&rule->mutex);
1933 ovs_rwlock_unlock(&ofproto->tables[0].cls.rwlock);
1935 /* If there's no such rule or the rule doesn't have the actions we want,
1936 * fall back to a executing a full flow mod. We can't optimize this at
1937 * all because we didn't take enough locks above to ensure that the flow
1938 * table didn't already change beneath us. */
1940 simple_flow_mod(ofproto, match, priority, ofpacts, ofpacts_len,
1941 OFPFC_MODIFY_STRICT);
1945 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1946 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1947 * operation cannot be initiated now but may be retried later.
1949 * This is a helper function for in-band control and fail-open and the "learn"
1952 ofproto_flow_mod(struct ofproto *ofproto, struct ofputil_flow_mod *fm)
1953 OVS_EXCLUDED(ofproto_mutex)
1955 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1958 /* Searches for a rule with matching criteria exactly equal to 'target' in
1959 * ofproto's table 0 and, if it finds one, deletes it.
1961 * This is a helper function for in-band control and fail-open. */
1963 ofproto_delete_flow(struct ofproto *ofproto,
1964 const struct match *target, unsigned int priority)
1965 OVS_EXCLUDED(ofproto_mutex)
1967 struct classifier *cls = &ofproto->tables[0].cls;
1970 /* First do a cheap check whether the rule we're looking for has already
1971 * been deleted. If so, then we're done. */
1972 ovs_rwlock_rdlock(&cls->rwlock);
1973 rule = rule_from_cls_rule(classifier_find_match_exactly(cls, target,
1975 ovs_rwlock_unlock(&cls->rwlock);
1980 /* Fall back to a executing a full flow mod. We can't optimize this at all
1981 * because we didn't take enough locks above to ensure that the flow table
1982 * didn't already change beneath us. */
1983 return simple_flow_mod(ofproto, target, priority, NULL, 0,
1984 OFPFC_DELETE_STRICT) != OFPROTO_POSTPONE;
1987 /* Starts the process of deleting all of the flows from all of ofproto's flow
1988 * tables and then reintroducing the flows required by in-band control and
1989 * fail-open. The process will complete in a later call to ofproto_run(). */
1991 ofproto_flush_flows(struct ofproto *ofproto)
1993 COVERAGE_INC(ofproto_flush);
1994 ofproto->state = S_FLUSH;
1998 reinit_ports(struct ofproto *p)
2000 struct ofproto_port_dump dump;
2001 struct sset devnames;
2002 struct ofport *ofport;
2003 struct ofproto_port ofproto_port;
2004 const char *devname;
2006 COVERAGE_INC(ofproto_reinit_ports);
2008 sset_init(&devnames);
2009 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2010 sset_add(&devnames, netdev_get_name(ofport->netdev));
2012 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2013 sset_add(&devnames, ofproto_port.name);
2016 SSET_FOR_EACH (devname, &devnames) {
2017 update_port(p, devname);
2019 sset_destroy(&devnames);
2023 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
2027 port_idx = simap_get(&ofproto->ofp_requests, netdev_name);
2028 port_idx = port_idx ? port_idx : UINT16_MAX;
2030 if (port_idx >= ofproto->max_ports
2031 || ofport_get_usage(ofproto, u16_to_ofp(port_idx)) == LLONG_MAX) {
2032 uint16_t lru_ofport = 0, end_port_no = ofproto->alloc_port_no;
2033 long long int last_used_at, lru = LLONG_MAX;
2035 /* Search for a free OpenFlow port number. We try not to
2036 * immediately reuse them to prevent problems due to old
2039 * We limit the automatically assigned port numbers to the lower half
2040 * of the port range, to reserve the upper half for assignment by
2043 if (++ofproto->alloc_port_no >= MIN(ofproto->max_ports, 32768)) {
2044 ofproto->alloc_port_no = 1;
2046 last_used_at = ofport_get_usage(ofproto,
2047 u16_to_ofp(ofproto->alloc_port_no));
2048 if (!last_used_at) {
2049 port_idx = ofproto->alloc_port_no;
2051 } else if ( last_used_at < time_msec() - 60*60*1000) {
2052 /* If the port with ofport 'ofproto->alloc_port_no' was deleted
2053 * more than an hour ago, consider it usable. */
2054 ofport_remove_usage(ofproto,
2055 u16_to_ofp(ofproto->alloc_port_no));
2056 port_idx = ofproto->alloc_port_no;
2058 } else if (last_used_at < lru) {
2060 lru_ofport = ofproto->alloc_port_no;
2063 if (ofproto->alloc_port_no == end_port_no) {
2065 port_idx = lru_ofport;
2072 ofport_set_usage(ofproto, u16_to_ofp(port_idx), LLONG_MAX);
2073 return u16_to_ofp(port_idx);
2077 dealloc_ofp_port(struct ofproto *ofproto, ofp_port_t ofp_port)
2079 if (ofp_to_u16(ofp_port) < ofproto->max_ports) {
2080 ofport_set_usage(ofproto, ofp_port, time_msec());
2084 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
2085 * pointer if the netdev cannot be opened. On success, also fills in
2087 static struct netdev *
2088 ofport_open(struct ofproto *ofproto,
2089 struct ofproto_port *ofproto_port,
2090 struct ofputil_phy_port *pp)
2092 enum netdev_flags flags;
2093 struct netdev *netdev;
2096 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
2098 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
2099 "cannot be opened (%s)",
2101 ofproto_port->name, ofproto_port->ofp_port,
2102 ofproto_port->name, ovs_strerror(error));
2106 if (ofproto_port->ofp_port == OFPP_NONE) {
2107 if (!strcmp(ofproto->name, ofproto_port->name)) {
2108 ofproto_port->ofp_port = OFPP_LOCAL;
2110 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
2111 ofproto_port->name);
2114 pp->port_no = ofproto_port->ofp_port;
2115 netdev_get_etheraddr(netdev, pp->hw_addr);
2116 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
2117 netdev_get_flags(netdev, &flags);
2118 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
2119 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
2120 netdev_get_features(netdev, &pp->curr, &pp->advertised,
2121 &pp->supported, &pp->peer);
2122 pp->curr_speed = netdev_features_to_bps(pp->curr, 0) / 1000;
2123 pp->max_speed = netdev_features_to_bps(pp->supported, 0) / 1000;
2128 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
2129 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
2132 ofport_equal(const struct ofputil_phy_port *a,
2133 const struct ofputil_phy_port *b)
2135 return (eth_addr_equals(a->hw_addr, b->hw_addr)
2136 && a->state == b->state
2137 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
2138 && a->curr == b->curr
2139 && a->advertised == b->advertised
2140 && a->supported == b->supported
2141 && a->peer == b->peer
2142 && a->curr_speed == b->curr_speed
2143 && a->max_speed == b->max_speed);
2146 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
2147 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
2148 * one with the same name or port number). */
2150 ofport_install(struct ofproto *p,
2151 struct netdev *netdev, const struct ofputil_phy_port *pp)
2153 const char *netdev_name = netdev_get_name(netdev);
2154 struct ofport *ofport;
2157 /* Create ofport. */
2158 ofport = p->ofproto_class->port_alloc();
2163 ofport->ofproto = p;
2164 ofport->netdev = netdev;
2166 ofport->ofp_port = pp->port_no;
2167 ofport->created = time_msec();
2169 /* Add port to 'p'. */
2170 hmap_insert(&p->ports, &ofport->hmap_node,
2171 hash_ofp_port(ofport->ofp_port));
2172 shash_add(&p->port_by_name, netdev_name, ofport);
2174 update_mtu(p, ofport);
2176 /* Let the ofproto_class initialize its private data. */
2177 error = p->ofproto_class->port_construct(ofport);
2181 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
2185 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
2186 p->name, netdev_name, ovs_strerror(error));
2188 ofport_destroy__(ofport);
2190 netdev_close(netdev);
2194 /* Removes 'ofport' from 'p' and destroys it. */
2196 ofport_remove(struct ofport *ofport)
2198 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
2200 ofport_destroy(ofport);
2203 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
2206 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
2208 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
2210 ofport_remove(port);
2214 /* Updates 'port' with new 'pp' description.
2216 * Does not handle a name or port number change. The caller must implement
2217 * such a change as a delete followed by an add. */
2219 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
2221 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
2222 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
2223 | (pp->config & OFPUTIL_PC_PORT_DOWN));
2224 port->pp.state = pp->state;
2225 port->pp.curr = pp->curr;
2226 port->pp.advertised = pp->advertised;
2227 port->pp.supported = pp->supported;
2228 port->pp.peer = pp->peer;
2229 port->pp.curr_speed = pp->curr_speed;
2230 port->pp.max_speed = pp->max_speed;
2232 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
2235 /* Update OpenFlow 'state' in 'port' and notify controller. */
2237 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
2239 if (port->pp.state != state) {
2240 port->pp.state = state;
2241 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
2247 ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port)
2249 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
2251 if (port->ofproto->ofproto_class->set_realdev) {
2252 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
2254 if (port->ofproto->ofproto_class->set_stp_port) {
2255 port->ofproto->ofproto_class->set_stp_port(port, NULL);
2257 if (port->ofproto->ofproto_class->set_cfm) {
2258 port->ofproto->ofproto_class->set_cfm(port, NULL);
2260 if (port->ofproto->ofproto_class->bundle_remove) {
2261 port->ofproto->ofproto_class->bundle_remove(port);
2267 ofport_destroy__(struct ofport *port)
2269 struct ofproto *ofproto = port->ofproto;
2270 const char *name = netdev_get_name(port->netdev);
2272 hmap_remove(&ofproto->ports, &port->hmap_node);
2273 shash_delete(&ofproto->port_by_name,
2274 shash_find(&ofproto->port_by_name, name));
2276 netdev_close(port->netdev);
2277 ofproto->ofproto_class->port_dealloc(port);
2281 ofport_destroy(struct ofport *port)
2284 dealloc_ofp_port(port->ofproto, port->ofp_port);
2285 port->ofproto->ofproto_class->port_destruct(port);
2286 ofport_destroy__(port);
2291 ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
2293 struct ofport *port;
2295 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_ofp_port(ofp_port),
2297 if (port->ofp_port == ofp_port) {
2304 static long long int
2305 ofport_get_usage(const struct ofproto *ofproto, ofp_port_t ofp_port)
2307 struct ofport_usage *usage;
2309 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2310 &ofproto->ofport_usage) {
2311 if (usage->ofp_port == ofp_port) {
2312 return usage->last_used;
2319 ofport_set_usage(struct ofproto *ofproto, ofp_port_t ofp_port,
2320 long long int last_used)
2322 struct ofport_usage *usage;
2323 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2324 &ofproto->ofport_usage) {
2325 if (usage->ofp_port == ofp_port) {
2326 usage->last_used = last_used;
2330 ovs_assert(last_used == LLONG_MAX);
2332 usage = xmalloc(sizeof *usage);
2333 usage->ofp_port = ofp_port;
2334 usage->last_used = last_used;
2335 hmap_insert(&ofproto->ofport_usage, &usage->hmap_node,
2336 hash_ofp_port(ofp_port));
2340 ofport_remove_usage(struct ofproto *ofproto, ofp_port_t ofp_port)
2342 struct ofport_usage *usage;
2343 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2344 &ofproto->ofport_usage) {
2345 if (usage->ofp_port == ofp_port) {
2346 hmap_remove(&ofproto->ofport_usage, &usage->hmap_node);
2354 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
2356 struct ofproto *ofproto = port->ofproto;
2359 if (ofproto->ofproto_class->port_get_stats) {
2360 error = ofproto->ofproto_class->port_get_stats(port, stats);
2369 update_port(struct ofproto *ofproto, const char *name)
2371 struct ofproto_port ofproto_port;
2372 struct ofputil_phy_port pp;
2373 struct netdev *netdev;
2374 struct ofport *port;
2376 COVERAGE_INC(ofproto_update_port);
2378 /* Fetch 'name''s location and properties from the datapath. */
2379 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
2380 ? ofport_open(ofproto, &ofproto_port, &pp)
2384 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
2385 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
2386 struct netdev *old_netdev = port->netdev;
2388 /* 'name' hasn't changed location. Any properties changed? */
2389 if (!ofport_equal(&port->pp, &pp)) {
2390 ofport_modified(port, &pp);
2393 update_mtu(ofproto, port);
2395 /* Install the newly opened netdev in case it has changed.
2396 * Don't close the old netdev yet in case port_modified has to
2397 * remove a retained reference to it.*/
2398 port->netdev = netdev;
2400 if (port->ofproto->ofproto_class->port_modified) {
2401 port->ofproto->ofproto_class->port_modified(port);
2404 netdev_close(old_netdev);
2406 /* If 'port' is nonnull then its name differs from 'name' and thus
2407 * we should delete it. If we think there's a port named 'name'
2408 * then its port number must be wrong now so delete it too. */
2410 ofport_remove(port);
2412 ofport_remove_with_name(ofproto, name);
2413 ofport_install(ofproto, netdev, &pp);
2416 /* Any port named 'name' is gone now. */
2417 ofport_remove_with_name(ofproto, name);
2419 ofproto_port_destroy(&ofproto_port);
2423 init_ports(struct ofproto *p)
2425 struct ofproto_port_dump dump;
2426 struct ofproto_port ofproto_port;
2427 struct shash_node *node, *next;
2429 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2430 const char *name = ofproto_port.name;
2432 if (shash_find(&p->port_by_name, name)) {
2433 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2436 struct ofputil_phy_port pp;
2437 struct netdev *netdev;
2439 /* Check if an OpenFlow port number had been requested. */
2440 node = shash_find(&init_ofp_ports, name);
2442 const struct iface_hint *iface_hint = node->data;
2443 simap_put(&p->ofp_requests, name,
2444 ofp_to_u16(iface_hint->ofp_port));
2447 netdev = ofport_open(p, &ofproto_port, &pp);
2449 ofport_install(p, netdev, &pp);
2450 if (ofp_to_u16(ofproto_port.ofp_port) < p->max_ports) {
2451 p->alloc_port_no = MAX(p->alloc_port_no,
2452 ofp_to_u16(ofproto_port.ofp_port));
2458 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2459 struct iface_hint *iface_hint = node->data;
2461 if (!strcmp(iface_hint->br_name, p->name)) {
2462 free(iface_hint->br_name);
2463 free(iface_hint->br_type);
2465 shash_delete(&init_ofp_ports, node);
2472 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2473 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2475 find_min_mtu(struct ofproto *p)
2477 struct ofport *ofport;
2480 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2481 struct netdev *netdev = ofport->netdev;
2484 /* Skip any internal ports, since that's what we're trying to
2486 if (!strcmp(netdev_get_type(netdev), "internal")) {
2490 if (netdev_get_mtu(netdev, &dev_mtu)) {
2493 if (!mtu || dev_mtu < mtu) {
2498 return mtu ? mtu: ETH_PAYLOAD_MAX;
2501 /* Update MTU of all datapath devices on 'p' to the minimum of the
2502 * non-datapath ports in event of 'port' added or changed. */
2504 update_mtu(struct ofproto *p, struct ofport *port)
2506 struct ofport *ofport;
2507 struct netdev *netdev = port->netdev;
2508 int dev_mtu, old_min;
2510 if (netdev_get_mtu(netdev, &dev_mtu)) {
2514 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2515 if (dev_mtu > p->min_mtu) {
2516 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2517 dev_mtu = p->min_mtu;
2520 port->mtu = dev_mtu;
2524 /* For non-internal port find new min mtu. */
2525 old_min = p->min_mtu;
2526 port->mtu = dev_mtu;
2527 p->min_mtu = find_min_mtu(p);
2528 if (p->min_mtu == old_min) {
2532 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2533 struct netdev *netdev = ofport->netdev;
2535 if (!strcmp(netdev_get_type(netdev), "internal")) {
2536 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2537 ofport->mtu = p->min_mtu;
2544 ofproto_rule_ref(struct rule *rule)
2549 atomic_add(&rule->ref_count, 1, &orig);
2550 ovs_assert(orig != 0);
2555 ofproto_rule_unref(struct rule *rule)
2560 atomic_sub(&rule->ref_count, 1, &orig);
2562 rule->ofproto->ofproto_class->rule_destruct(rule);
2563 ofproto_rule_destroy__(rule);
2565 ovs_assert(orig != 0);
2570 struct rule_actions *
2571 rule_get_actions(const struct rule *rule)
2572 OVS_EXCLUDED(rule->mutex)
2574 struct rule_actions *actions;
2576 ovs_mutex_lock(&rule->mutex);
2577 actions = rule_get_actions__(rule);
2578 ovs_mutex_unlock(&rule->mutex);
2583 struct rule_actions *
2584 rule_get_actions__(const struct rule *rule)
2585 OVS_REQUIRES(rule->mutex)
2587 rule_actions_ref(rule->actions);
2588 return rule->actions;
2592 ofproto_rule_destroy__(struct rule *rule)
2593 OVS_NO_THREAD_SAFETY_ANALYSIS
2595 cls_rule_destroy(CONST_CAST(struct cls_rule *, &rule->cr));
2596 rule_actions_unref(rule->actions);
2597 ovs_mutex_destroy(&rule->mutex);
2598 rule->ofproto->ofproto_class->rule_dealloc(rule);
2601 static uint32_t get_provider_meter_id(const struct ofproto *,
2602 uint32_t of_meter_id);
2604 /* Creates and returns a new 'struct rule_actions', with a ref_count of 1,
2605 * whose actions are a copy of from the 'ofpacts_len' bytes of 'ofpacts'. */
2606 struct rule_actions *
2607 rule_actions_create(const struct ofproto *ofproto,
2608 const struct ofpact *ofpacts, size_t ofpacts_len)
2610 struct rule_actions *actions;
2612 actions = xmalloc(sizeof *actions);
2613 atomic_init(&actions->ref_count, 1);
2614 actions->ofpacts = xmemdup(ofpacts, ofpacts_len);
2615 actions->ofpacts_len = ofpacts_len;
2616 actions->provider_meter_id
2617 = get_provider_meter_id(ofproto,
2618 ofpacts_get_meter(ofpacts, ofpacts_len));
2623 /* Increments 'actions''s ref_count. */
2625 rule_actions_ref(struct rule_actions *actions)
2630 atomic_add(&actions->ref_count, 1, &orig);
2631 ovs_assert(orig != 0);
2635 /* Decrements 'actions''s ref_count and frees 'actions' if the ref_count
2638 rule_actions_unref(struct rule_actions *actions)
2643 atomic_sub(&actions->ref_count, 1, &orig);
2645 free(actions->ofpacts);
2648 ovs_assert(orig != 0);
2653 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2654 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2656 ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port)
2657 OVS_REQUIRES(ofproto_mutex)
2659 return (port == OFPP_ANY
2660 || ofpacts_output_to_port(rule->actions->ofpacts,
2661 rule->actions->ofpacts_len, port));
2664 /* Returns true if 'rule' has group and equals group_id. */
2666 ofproto_rule_has_out_group(const struct rule *rule, uint32_t group_id)
2667 OVS_REQUIRES(ofproto_mutex)
2669 return (group_id == OFPG11_ANY
2670 || ofpacts_output_to_group(rule->actions->ofpacts,
2671 rule->actions->ofpacts_len, group_id));
2674 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2675 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2677 ofoperation_has_out_port(const struct ofoperation *op, ofp_port_t out_port)
2678 OVS_REQUIRES(ofproto_mutex)
2680 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2685 case OFOPERATION_ADD:
2686 case OFOPERATION_DELETE:
2689 case OFOPERATION_MODIFY:
2690 case OFOPERATION_REPLACE:
2691 return ofpacts_output_to_port(op->actions->ofpacts,
2692 op->actions->ofpacts_len, out_port);
2699 rule_execute_destroy(struct rule_execute *e)
2701 ofproto_rule_unref(e->rule);
2702 list_remove(&e->list_node);
2706 /* Executes all "rule_execute" operations queued up in ofproto->rule_executes,
2707 * by passing them to the ofproto provider. */
2709 run_rule_executes(struct ofproto *ofproto)
2710 OVS_EXCLUDED(ofproto_mutex)
2712 struct rule_execute *e, *next;
2713 struct list executes;
2715 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2716 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2717 union flow_in_port in_port_;
2720 in_port_.ofp_port = e->in_port;
2721 flow_extract(e->packet, 0, 0, NULL, &in_port_, &flow);
2722 ofproto->ofproto_class->rule_execute(e->rule, &flow, e->packet);
2724 rule_execute_destroy(e);
2728 /* Destroys and discards all "rule_execute" operations queued up in
2729 * ofproto->rule_executes. */
2731 destroy_rule_executes(struct ofproto *ofproto)
2733 struct rule_execute *e, *next;
2734 struct list executes;
2736 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2737 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2738 ofpbuf_delete(e->packet);
2739 rule_execute_destroy(e);
2743 /* Returns true if 'rule' should be hidden from the controller.
2745 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2746 * (e.g. by in-band control) and are intentionally hidden from the
2749 ofproto_rule_is_hidden(const struct rule *rule)
2751 return rule->cr.priority > UINT16_MAX;
2754 static enum oftable_flags
2755 rule_get_flags(const struct rule *rule)
2757 return rule->ofproto->tables[rule->table_id].flags;
2761 rule_is_modifiable(const struct rule *rule)
2763 return !(rule_get_flags(rule) & OFTABLE_READONLY);
2767 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2769 ofconn_send_reply(ofconn, make_echo_reply(oh));
2774 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2776 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2777 struct ofputil_switch_features features;
2778 struct ofport *port;
2784 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2786 ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2788 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2789 * if present, are always at the end.) */
2790 n_tables = ofproto->n_tables;
2791 for (i = 0; i < ofproto->n_tables; i++) {
2792 if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
2798 features.datapath_id = ofproto->datapath_id;
2799 features.n_buffers = pktbuf_capacity();
2800 features.n_tables = n_tables;
2801 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2802 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2804 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2806 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2807 features.auxiliary_id = 0;
2808 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2810 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2811 ofputil_put_switch_features_port(&port->pp, b);
2814 ofconn_send_reply(ofconn, b);
2819 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2821 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2822 struct ofp_switch_config *osc;
2823 enum ofp_config_flags flags;
2827 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2828 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2829 flags = ofproto->frag_handling;
2830 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2831 if (oh->version < OFP13_VERSION
2832 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2833 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2835 osc->flags = htons(flags);
2836 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2837 ofconn_send_reply(ofconn, buf);
2843 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2845 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2846 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2847 uint16_t flags = ntohs(osc->flags);
2849 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2850 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
2851 enum ofp_config_flags cur = ofproto->frag_handling;
2852 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2854 ovs_assert((cur & OFPC_FRAG_MASK) == cur);
2856 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2857 ofproto->frag_handling = next;
2859 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2861 ofputil_frag_handling_to_string(next));
2865 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2866 ofconn_set_invalid_ttl_to_controller(ofconn,
2867 (oh->version < OFP13_VERSION
2868 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2870 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2875 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2876 * error message code for the caller to propagate upward. Otherwise, returns
2879 * The log message mentions 'msg_type'. */
2881 reject_slave_controller(struct ofconn *ofconn)
2883 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2884 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
2885 return OFPERR_OFPBRC_EPERM;
2891 /* Checks that the 'ofpacts_len' bytes of action in 'ofpacts' are appropriate
2894 * - If they use a meter, then 'ofproto' has that meter configured.
2896 * - If they use any groups, then 'ofproto' has that group configured.
2898 * Returns 0 if successful, otherwise an OpenFlow error. */
2900 ofproto_check_ofpacts(struct ofproto *ofproto,
2901 const struct ofpact ofpacts[], size_t ofpacts_len)
2903 const struct ofpact *a;
2906 mid = ofpacts_get_meter(ofpacts, ofpacts_len);
2907 if (mid && get_provider_meter_id(ofproto, mid) == UINT32_MAX) {
2908 return OFPERR_OFPMMFC_INVALID_METER;
2911 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2912 if (a->type == OFPACT_GROUP
2913 && !ofproto_group_exists(ofproto, ofpact_get_GROUP(a)->group_id)) {
2914 return OFPERR_OFPBAC_BAD_OUT_GROUP;
2922 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2924 struct ofproto *p = ofconn_get_ofproto(ofconn);
2925 struct ofputil_packet_out po;
2926 struct ofpbuf *payload;
2927 uint64_t ofpacts_stub[1024 / 8];
2928 struct ofpbuf ofpacts;
2930 union flow_in_port in_port_;
2933 COVERAGE_INC(ofproto_packet_out);
2935 error = reject_slave_controller(ofconn);
2940 /* Decode message. */
2941 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2942 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2944 goto exit_free_ofpacts;
2946 if (ofp_to_u16(po.in_port) >= p->max_ports
2947 && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) {
2948 error = OFPERR_OFPBRC_BAD_PORT;
2949 goto exit_free_ofpacts;
2953 if (po.buffer_id != UINT32_MAX) {
2954 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2955 if (error || !payload) {
2956 goto exit_free_ofpacts;
2959 /* Ensure that the L3 header is 32-bit aligned. */
2960 payload = ofpbuf_clone_data_with_headroom(po.packet, po.packet_len, 2);
2963 /* Verify actions against packet, then send packet if successful. */
2964 in_port_.ofp_port = po.in_port;
2965 flow_extract(payload, 0, 0, NULL, &in_port_, &flow);
2966 error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len);
2968 error = p->ofproto_class->packet_out(p, payload, &flow,
2969 po.ofpacts, po.ofpacts_len);
2971 ofpbuf_delete(payload);
2974 ofpbuf_uninit(&ofpacts);
2980 update_port_config(struct ofport *port,
2981 enum ofputil_port_config config,
2982 enum ofputil_port_config mask)
2984 enum ofputil_port_config old_config = port->pp.config;
2985 enum ofputil_port_config toggle;
2987 toggle = (config ^ port->pp.config) & mask;
2988 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2989 if (config & OFPUTIL_PC_PORT_DOWN) {
2990 netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL);
2992 netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL);
2994 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2997 port->pp.config ^= toggle;
2998 if (port->pp.config != old_config) {
2999 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
3004 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3006 struct ofproto *p = ofconn_get_ofproto(ofconn);
3007 struct ofputil_port_mod pm;
3008 struct ofport *port;
3011 error = reject_slave_controller(ofconn);
3016 error = ofputil_decode_port_mod(oh, &pm);
3021 port = ofproto_get_port(p, pm.port_no);
3023 return OFPERR_OFPPMFC_BAD_PORT;
3024 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
3025 return OFPERR_OFPPMFC_BAD_HW_ADDR;
3027 update_port_config(port, pm.config, pm.mask);
3029 netdev_set_advertisements(port->netdev, pm.advertise);
3036 handle_desc_stats_request(struct ofconn *ofconn,
3037 const struct ofp_header *request)
3039 static const char *default_mfr_desc = "Nicira, Inc.";
3040 static const char *default_hw_desc = "Open vSwitch";
3041 static const char *default_sw_desc = VERSION;
3042 static const char *default_serial_desc = "None";
3043 static const char *default_dp_desc = "None";
3045 struct ofproto *p = ofconn_get_ofproto(ofconn);
3046 struct ofp_desc_stats *ods;
3049 msg = ofpraw_alloc_stats_reply(request, 0);
3050 ods = ofpbuf_put_zeros(msg, sizeof *ods);
3051 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
3052 sizeof ods->mfr_desc);
3053 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
3054 sizeof ods->hw_desc);
3055 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
3056 sizeof ods->sw_desc);
3057 ovs_strlcpy(ods->serial_num,
3058 p->serial_desc ? p->serial_desc : default_serial_desc,
3059 sizeof ods->serial_num);
3060 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
3061 sizeof ods->dp_desc);
3062 ofconn_send_reply(ofconn, msg);
3068 handle_table_stats_request(struct ofconn *ofconn,
3069 const struct ofp_header *request)
3071 struct ofproto *p = ofconn_get_ofproto(ofconn);
3072 struct ofp12_table_stats *ots;
3077 /* Set up default values.
3079 * ofp12_table_stats is used as a generic structure as
3080 * it is able to hold all the fields for ofp10_table_stats
3081 * and ofp11_table_stats (and of course itself).
3083 ots = xcalloc(p->n_tables, sizeof *ots);
3084 for (i = 0; i < p->n_tables; i++) {
3085 ots[i].table_id = i;
3086 sprintf(ots[i].name, "table%"PRIuSIZE, i);
3087 ots[i].match = htonll(OFPXMT13_MASK);
3088 ots[i].wildcards = htonll(OFPXMT13_MASK);
3089 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
3090 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
3091 ots[i].write_setfields = htonll(OFPXMT13_MASK);
3092 ots[i].apply_setfields = htonll(OFPXMT13_MASK);
3093 ots[i].metadata_match = OVS_BE64_MAX;
3094 ots[i].metadata_write = OVS_BE64_MAX;
3095 ots[i].instructions = htonl(OFPIT11_ALL);
3096 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
3097 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
3098 ovs_rwlock_rdlock(&p->tables[i].cls.rwlock);
3099 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
3100 ovs_rwlock_unlock(&p->tables[i].cls.rwlock);
3103 p->ofproto_class->get_tables(p, ots);
3105 /* Post-process the tables, dropping hidden tables. */
3106 n_tables = p->n_tables;
3107 for (i = 0; i < p->n_tables; i++) {
3108 const struct oftable *table = &p->tables[i];
3110 if (table->flags & OFTABLE_HIDDEN) {
3116 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
3119 if (table->max_flows < ntohl(ots[i].max_entries)) {
3120 ots[i].max_entries = htonl(table->max_flows);
3124 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
3125 ofconn_send_reply(ofconn, msg);
3133 append_port_stat(struct ofport *port, struct list *replies)
3135 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
3137 calc_duration(port->created, time_msec(),
3138 &ops.duration_sec, &ops.duration_nsec);
3140 /* Intentionally ignore return value, since errors will set
3141 * 'stats' to all-1s, which is correct for OpenFlow, and
3142 * netdev_get_stats() will log errors. */
3143 ofproto_port_get_stats(port, &ops.stats);
3145 ofputil_append_port_stat(replies, &ops);
3149 handle_port_stats_request(struct ofconn *ofconn,
3150 const struct ofp_header *request)
3152 struct ofproto *p = ofconn_get_ofproto(ofconn);
3153 struct ofport *port;
3154 struct list replies;
3158 error = ofputil_decode_port_stats_request(request, &port_no);
3163 ofpmp_init(&replies, request);
3164 if (port_no != OFPP_ANY) {
3165 port = ofproto_get_port(p, port_no);
3167 append_port_stat(port, &replies);
3170 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3171 append_port_stat(port, &replies);
3175 ofconn_send_replies(ofconn, &replies);
3180 handle_port_desc_stats_request(struct ofconn *ofconn,
3181 const struct ofp_header *request)
3183 struct ofproto *p = ofconn_get_ofproto(ofconn);
3184 enum ofp_version version;
3185 struct ofport *port;
3186 struct list replies;
3188 ofpmp_init(&replies, request);
3190 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
3191 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3192 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
3195 ofconn_send_replies(ofconn, &replies);
3200 hash_cookie(ovs_be64 cookie)
3202 return hash_2words((OVS_FORCE uint64_t)cookie >> 32,
3203 (OVS_FORCE uint64_t)cookie);
3207 cookies_insert(struct ofproto *ofproto, struct rule *rule)
3208 OVS_REQUIRES(ofproto_mutex)
3210 hindex_insert(&ofproto->cookies, &rule->cookie_node,
3211 hash_cookie(rule->flow_cookie));
3215 cookies_remove(struct ofproto *ofproto, struct rule *rule)
3216 OVS_REQUIRES(ofproto_mutex)
3218 hindex_remove(&ofproto->cookies, &rule->cookie_node);
3222 ofproto_rule_change_cookie(struct ofproto *ofproto, struct rule *rule,
3223 ovs_be64 new_cookie)
3224 OVS_REQUIRES(ofproto_mutex)
3226 if (new_cookie != rule->flow_cookie) {
3227 cookies_remove(ofproto, rule);
3229 ovs_mutex_lock(&rule->mutex);
3230 rule->flow_cookie = new_cookie;
3231 ovs_mutex_unlock(&rule->mutex);
3233 cookies_insert(ofproto, rule);
3238 calc_duration(long long int start, long long int now,
3239 uint32_t *sec, uint32_t *nsec)
3241 long long int msecs = now - start;
3242 *sec = msecs / 1000;
3243 *nsec = (msecs % 1000) * (1000 * 1000);
3246 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
3247 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
3249 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
3251 return (table_id == 0xff || table_id < ofproto->n_tables
3253 : OFPERR_OFPBRC_BAD_TABLE_ID);
3257 static struct oftable *
3258 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
3260 struct oftable *table;
3262 for (table = &ofproto->tables[table_id];
3263 table < &ofproto->tables[ofproto->n_tables];
3265 if (!(table->flags & OFTABLE_HIDDEN)) {
3273 static struct oftable *
3274 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
3276 if (table_id == 0xff) {
3277 return next_visible_table(ofproto, 0);
3278 } else if (table_id < ofproto->n_tables) {
3279 return &ofproto->tables[table_id];
3285 static struct oftable *
3286 next_matching_table(const struct ofproto *ofproto,
3287 const struct oftable *table, uint8_t table_id)
3289 return (table_id == 0xff
3290 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
3294 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
3296 * - If TABLE_ID is 0xff, this iterates over every classifier table in
3297 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
3299 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
3300 * only once, for that table. (This can be used to access tables marked
3303 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
3304 * entered at all. (Perhaps you should have validated TABLE_ID with
3305 * check_table_id().)
3307 * All parameters are evaluated multiple times.
3309 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
3310 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
3312 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
3314 /* Initializes 'criteria' in a straightforward way based on the other
3317 * For "loose" matching, the 'priority' parameter is unimportant and may be
3320 rule_criteria_init(struct rule_criteria *criteria, uint8_t table_id,
3321 const struct match *match, unsigned int priority,
3322 ovs_be64 cookie, ovs_be64 cookie_mask,
3323 ofp_port_t out_port, uint32_t out_group)
3325 criteria->table_id = table_id;
3326 cls_rule_init(&criteria->cr, match, priority);
3327 criteria->cookie = cookie;
3328 criteria->cookie_mask = cookie_mask;
3329 criteria->out_port = out_port;
3330 criteria->out_group = out_group;
3334 rule_criteria_destroy(struct rule_criteria *criteria)
3336 cls_rule_destroy(&criteria->cr);
3340 rule_collection_init(struct rule_collection *rules)
3342 rules->rules = rules->stub;
3344 rules->capacity = ARRAY_SIZE(rules->stub);
3348 rule_collection_add(struct rule_collection *rules, struct rule *rule)
3350 if (rules->n >= rules->capacity) {
3351 size_t old_size, new_size;
3353 old_size = rules->capacity * sizeof *rules->rules;
3354 rules->capacity *= 2;
3355 new_size = rules->capacity * sizeof *rules->rules;
3357 if (rules->rules == rules->stub) {
3358 rules->rules = xmalloc(new_size);
3359 memcpy(rules->rules, rules->stub, old_size);
3361 rules->rules = xrealloc(rules->rules, new_size);
3365 rules->rules[rules->n++] = rule;
3369 rule_collection_ref(struct rule_collection *rules)
3370 OVS_REQUIRES(ofproto_mutex)
3374 for (i = 0; i < rules->n; i++) {
3375 ofproto_rule_ref(rules->rules[i]);
3380 rule_collection_unref(struct rule_collection *rules)
3384 for (i = 0; i < rules->n; i++) {
3385 ofproto_rule_unref(rules->rules[i]);
3390 rule_collection_destroy(struct rule_collection *rules)
3392 if (rules->rules != rules->stub) {
3398 collect_rule(struct rule *rule, const struct rule_criteria *c,
3399 struct rule_collection *rules)
3400 OVS_REQUIRES(ofproto_mutex)
3402 /* We ordinarily want to skip hidden rules, but there has to be a way for
3403 * code internal to OVS to modify and delete them, so if the criteria
3404 * specify a priority that can only be for a hidden flow, then allow hidden
3405 * rules to be selected. (This doesn't allow OpenFlow clients to meddle
3406 * with hidden flows because OpenFlow uses only a 16-bit field to specify
3408 if (ofproto_rule_is_hidden(rule) && c->cr.priority <= UINT16_MAX) {
3410 } else if (rule->pending) {
3411 return OFPROTO_POSTPONE;
3413 if ((c->table_id == rule->table_id || c->table_id == 0xff)
3414 && ofproto_rule_has_out_port(rule, c->out_port)
3415 && ofproto_rule_has_out_group(rule, c->out_group)
3416 && !((rule->flow_cookie ^ c->cookie) & c->cookie_mask)) {
3417 rule_collection_add(rules, rule);
3423 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3424 * on classifiers rules are done in the "loose" way required for OpenFlow
3425 * OFPFC_MODIFY and OFPFC_DELETE requests. Puts the selected rules on list
3428 * Hidden rules are always omitted.
3430 * Returns 0 on success, otherwise an OpenFlow error code. */
3432 collect_rules_loose(struct ofproto *ofproto,
3433 const struct rule_criteria *criteria,
3434 struct rule_collection *rules)
3435 OVS_REQUIRES(ofproto_mutex)
3437 struct oftable *table;
3440 rule_collection_init(rules);
3442 error = check_table_id(ofproto, criteria->table_id);
3447 if (criteria->cookie_mask == OVS_BE64_MAX) {
3450 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
3451 hash_cookie(criteria->cookie),
3452 &ofproto->cookies) {
3453 if (cls_rule_is_loose_match(&rule->cr, &criteria->cr.match)) {
3454 error = collect_rule(rule, criteria, rules);
3461 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
3462 struct cls_cursor cursor;
3465 ovs_rwlock_rdlock(&table->cls.rwlock);
3466 cls_cursor_init(&cursor, &table->cls, &criteria->cr);
3467 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3468 error = collect_rule(rule, criteria, rules);
3473 ovs_rwlock_unlock(&table->cls.rwlock);
3479 rule_collection_destroy(rules);
3484 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3485 * on classifiers rules are done in the "strict" way required for OpenFlow
3486 * OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests. Puts the selected
3487 * rules on list 'rules'.
3489 * Hidden rules are always omitted.
3491 * Returns 0 on success, otherwise an OpenFlow error code. */
3493 collect_rules_strict(struct ofproto *ofproto,
3494 const struct rule_criteria *criteria,
3495 struct rule_collection *rules)
3496 OVS_REQUIRES(ofproto_mutex)
3498 struct oftable *table;
3501 rule_collection_init(rules);
3503 error = check_table_id(ofproto, criteria->table_id);
3508 if (criteria->cookie_mask == OVS_BE64_MAX) {
3511 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
3512 hash_cookie(criteria->cookie),
3513 &ofproto->cookies) {
3514 if (cls_rule_equal(&rule->cr, &criteria->cr)) {
3515 error = collect_rule(rule, criteria, rules);
3522 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
3525 ovs_rwlock_rdlock(&table->cls.rwlock);
3526 rule = rule_from_cls_rule(classifier_find_rule_exactly(
3527 &table->cls, &criteria->cr));
3528 ovs_rwlock_unlock(&table->cls.rwlock);
3530 error = collect_rule(rule, criteria, rules);
3540 rule_collection_destroy(rules);
3545 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
3546 * forced into the range of a uint16_t. */
3548 age_secs(long long int age_ms)
3550 return (age_ms < 0 ? 0
3551 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
3552 : (unsigned int) age_ms / 1000);
3556 handle_flow_stats_request(struct ofconn *ofconn,
3557 const struct ofp_header *request)
3558 OVS_EXCLUDED(ofproto_mutex)
3560 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3561 struct ofputil_flow_stats_request fsr;
3562 struct rule_criteria criteria;
3563 struct rule_collection rules;
3564 struct list replies;
3568 error = ofputil_decode_flow_stats_request(&fsr, request);
3573 rule_criteria_init(&criteria, fsr.table_id, &fsr.match, 0, fsr.cookie,
3574 fsr.cookie_mask, fsr.out_port, fsr.out_group);
3576 ovs_mutex_lock(&ofproto_mutex);
3577 error = collect_rules_loose(ofproto, &criteria, &rules);
3578 rule_criteria_destroy(&criteria);
3580 rule_collection_ref(&rules);
3582 ovs_mutex_unlock(&ofproto_mutex);
3588 ofpmp_init(&replies, request);
3589 for (i = 0; i < rules.n; i++) {
3590 struct rule *rule = rules.rules[i];
3591 long long int now = time_msec();
3592 struct ofputil_flow_stats fs;
3593 long long int created, used, modified;
3594 struct rule_actions *actions;
3595 enum ofputil_flow_mod_flags flags;
3597 ovs_mutex_lock(&rule->mutex);
3598 fs.cookie = rule->flow_cookie;
3599 fs.idle_timeout = rule->idle_timeout;
3600 fs.hard_timeout = rule->hard_timeout;
3601 created = rule->created;
3603 modified = rule->modified;
3604 actions = rule_get_actions__(rule);
3605 flags = rule->flags;
3606 ovs_mutex_unlock(&rule->mutex);
3608 minimatch_expand(&rule->cr.match, &fs.match);
3609 fs.table_id = rule->table_id;
3610 calc_duration(created, now, &fs.duration_sec, &fs.duration_nsec);
3611 fs.priority = rule->cr.priority;
3612 fs.idle_age = age_secs(now - used);
3613 fs.hard_age = age_secs(now - modified);
3614 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
3616 fs.ofpacts = actions->ofpacts;
3617 fs.ofpacts_len = actions->ofpacts_len;
3620 ofputil_append_flow_stats_reply(&fs, &replies);
3622 rule_actions_unref(actions);
3625 rule_collection_unref(&rules);
3626 rule_collection_destroy(&rules);
3628 ofconn_send_replies(ofconn, &replies);
3634 flow_stats_ds(struct rule *rule, struct ds *results)
3636 uint64_t packet_count, byte_count;
3637 struct rule_actions *actions;
3638 long long int created;
3640 rule->ofproto->ofproto_class->rule_get_stats(rule,
3641 &packet_count, &byte_count);
3643 ovs_mutex_lock(&rule->mutex);
3644 actions = rule_get_actions__(rule);
3645 created = rule->created;
3646 ovs_mutex_unlock(&rule->mutex);
3648 if (rule->table_id != 0) {
3649 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
3651 ds_put_format(results, "duration=%llds, ", (time_msec() - created) / 1000);
3652 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3653 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3654 cls_rule_format(&rule->cr, results);
3655 ds_put_char(results, ',');
3657 ds_put_cstr(results, "actions=");
3658 ofpacts_format(actions->ofpacts, actions->ofpacts_len, results);
3660 ds_put_cstr(results, "\n");
3662 rule_actions_unref(actions);
3665 /* Adds a pretty-printed description of all flows to 'results', including
3666 * hidden flows (e.g., set up by in-band control). */
3668 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3670 struct oftable *table;
3672 OFPROTO_FOR_EACH_TABLE (table, p) {
3673 struct cls_cursor cursor;
3676 ovs_rwlock_rdlock(&table->cls.rwlock);
3677 cls_cursor_init(&cursor, &table->cls, NULL);
3678 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3679 flow_stats_ds(rule, results);
3681 ovs_rwlock_unlock(&table->cls.rwlock);
3685 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
3686 * '*engine_type' and '*engine_id', respectively. */
3688 ofproto_get_netflow_ids(const struct ofproto *ofproto,
3689 uint8_t *engine_type, uint8_t *engine_id)
3691 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
3694 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
3695 * true if the port's CFM status was successfully stored into '*status'.
3696 * Returns false if the port did not have CFM configured, in which case
3697 * '*status' is indeterminate.
3699 * The caller must provide and owns '*status', and must free 'status->rmps'. */
3701 ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port,
3702 struct ofproto_cfm_status *status)
3704 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
3706 && ofproto->ofproto_class->get_cfm_status
3707 && ofproto->ofproto_class->get_cfm_status(ofport, status));
3711 handle_aggregate_stats_request(struct ofconn *ofconn,
3712 const struct ofp_header *oh)
3713 OVS_EXCLUDED(ofproto_mutex)
3715 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3716 struct ofputil_flow_stats_request request;
3717 struct ofputil_aggregate_stats stats;
3718 bool unknown_packets, unknown_bytes;
3719 struct rule_criteria criteria;
3720 struct rule_collection rules;
3721 struct ofpbuf *reply;
3725 error = ofputil_decode_flow_stats_request(&request, oh);
3730 rule_criteria_init(&criteria, request.table_id, &request.match, 0,
3731 request.cookie, request.cookie_mask,
3732 request.out_port, request.out_group);
3734 ovs_mutex_lock(&ofproto_mutex);
3735 error = collect_rules_loose(ofproto, &criteria, &rules);
3736 rule_criteria_destroy(&criteria);
3738 rule_collection_ref(&rules);
3740 ovs_mutex_unlock(&ofproto_mutex);
3746 memset(&stats, 0, sizeof stats);
3747 unknown_packets = unknown_bytes = false;
3748 for (i = 0; i < rules.n; i++) {
3749 struct rule *rule = rules.rules[i];
3750 uint64_t packet_count;
3751 uint64_t byte_count;
3753 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
3756 if (packet_count == UINT64_MAX) {
3757 unknown_packets = true;
3759 stats.packet_count += packet_count;
3762 if (byte_count == UINT64_MAX) {
3763 unknown_bytes = true;
3765 stats.byte_count += byte_count;
3770 if (unknown_packets) {
3771 stats.packet_count = UINT64_MAX;
3773 if (unknown_bytes) {
3774 stats.byte_count = UINT64_MAX;
3777 rule_collection_unref(&rules);
3778 rule_collection_destroy(&rules);
3780 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
3781 ofconn_send_reply(ofconn, reply);
3786 struct queue_stats_cbdata {
3787 struct ofport *ofport;
3788 struct list replies;
3793 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3794 const struct netdev_queue_stats *stats)
3796 struct ofputil_queue_stats oqs;
3798 oqs.port_no = cbdata->ofport->pp.port_no;
3799 oqs.queue_id = queue_id;
3800 oqs.tx_bytes = stats->tx_bytes;
3801 oqs.tx_packets = stats->tx_packets;
3802 oqs.tx_errors = stats->tx_errors;
3803 if (stats->created != LLONG_MIN) {
3804 calc_duration(stats->created, cbdata->now,
3805 &oqs.duration_sec, &oqs.duration_nsec);
3807 oqs.duration_sec = oqs.duration_nsec = UINT32_MAX;
3809 ofputil_append_queue_stat(&cbdata->replies, &oqs);
3813 handle_queue_stats_dump_cb(uint32_t queue_id,
3814 struct netdev_queue_stats *stats,
3817 struct queue_stats_cbdata *cbdata = cbdata_;
3819 put_queue_stats(cbdata, queue_id, stats);
3823 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3824 struct queue_stats_cbdata *cbdata)
3826 cbdata->ofport = port;
3827 if (queue_id == OFPQ_ALL) {
3828 netdev_dump_queue_stats(port->netdev,
3829 handle_queue_stats_dump_cb, cbdata);
3831 struct netdev_queue_stats stats;
3833 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3834 put_queue_stats(cbdata, queue_id, &stats);
3836 return OFPERR_OFPQOFC_BAD_QUEUE;
3843 handle_queue_stats_request(struct ofconn *ofconn,
3844 const struct ofp_header *rq)
3846 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3847 struct queue_stats_cbdata cbdata;
3848 struct ofport *port;
3850 struct ofputil_queue_stats_request oqsr;
3852 COVERAGE_INC(ofproto_queue_req);
3854 ofpmp_init(&cbdata.replies, rq);
3855 cbdata.now = time_msec();
3857 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3862 if (oqsr.port_no == OFPP_ANY) {
3863 error = OFPERR_OFPQOFC_BAD_QUEUE;
3864 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3865 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3870 port = ofproto_get_port(ofproto, oqsr.port_no);
3872 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3873 : OFPERR_OFPQOFC_BAD_PORT);
3876 ofconn_send_replies(ofconn, &cbdata.replies);
3878 ofpbuf_list_delete(&cbdata.replies);
3885 is_flow_deletion_pending(const struct ofproto *ofproto,
3886 const struct cls_rule *cls_rule,
3888 OVS_REQUIRES(ofproto_mutex)
3890 if (!hmap_is_empty(&ofproto->deletions)) {
3891 struct ofoperation *op;
3893 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3894 cls_rule_hash(cls_rule, table_id),
3895 &ofproto->deletions) {
3896 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
3906 should_evict_a_rule(struct oftable *table, unsigned int extra_space)
3907 OVS_REQUIRES(ofproto_mutex)
3908 OVS_NO_THREAD_SAFETY_ANALYSIS
3910 return classifier_count(&table->cls) + extra_space > table->max_flows;
3914 evict_rules_from_table(struct ofproto *ofproto, struct oftable *table,
3915 unsigned int extra_space)
3916 OVS_REQUIRES(ofproto_mutex)
3918 while (should_evict_a_rule(table, extra_space)) {
3921 if (!choose_rule_to_evict(table, &rule)) {
3922 return OFPERR_OFPFMFC_TABLE_FULL;
3923 } else if (rule->pending) {
3924 return OFPROTO_POSTPONE;
3926 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
3927 delete_flow__(rule, group, OFPRR_EVICTION);
3928 ofopgroup_submit(group);
3935 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3936 * in which no matching flow already exists in the flow table.
3938 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3939 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3940 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3941 * initiated now but may be retried later.
3943 * The caller retains ownership of 'fm->ofpacts'.
3945 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3948 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
3949 struct ofputil_flow_mod *fm, const struct ofp_header *request)
3950 OVS_REQUIRES(ofproto_mutex)
3952 struct oftable *table;
3953 struct ofopgroup *group;
3959 error = check_table_id(ofproto, fm->table_id);
3965 if (fm->table_id == 0xff) {
3966 if (ofproto->ofproto_class->rule_choose_table) {
3967 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3973 ovs_assert(table_id < ofproto->n_tables);
3977 } else if (fm->table_id < ofproto->n_tables) {
3978 table_id = fm->table_id;
3980 return OFPERR_OFPBRC_BAD_TABLE_ID;
3983 table = &ofproto->tables[table_id];
3985 if (table->flags & OFTABLE_READONLY) {
3986 return OFPERR_OFPBRC_EPERM;
3989 cls_rule_init(&cr, &fm->match, fm->priority);
3991 /* Transform "add" into "modify" if there's an existing identical flow. */
3992 ovs_rwlock_rdlock(&table->cls.rwlock);
3993 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr));
3994 ovs_rwlock_unlock(&table->cls.rwlock);
3996 cls_rule_destroy(&cr);
3997 if (!rule_is_modifiable(rule)) {
3998 return OFPERR_OFPBRC_EPERM;
3999 } else if (rule->pending) {
4000 return OFPROTO_POSTPONE;
4002 struct rule_collection rules;
4004 rule_collection_init(&rules);
4005 rule_collection_add(&rules, rule);
4006 fm->modify_cookie = true;
4007 error = modify_flows__(ofproto, ofconn, fm, request, &rules);
4008 rule_collection_destroy(&rules);
4014 /* Serialize against pending deletion. */
4015 if (is_flow_deletion_pending(ofproto, &cr, table_id)) {
4016 cls_rule_destroy(&cr);
4017 return OFPROTO_POSTPONE;
4020 /* Check for overlap, if requested. */
4021 if (fm->flags & OFPUTIL_FF_CHECK_OVERLAP) {
4024 ovs_rwlock_rdlock(&table->cls.rwlock);
4025 overlaps = classifier_rule_overlaps(&table->cls, &cr);
4026 ovs_rwlock_unlock(&table->cls.rwlock);
4029 cls_rule_destroy(&cr);
4030 return OFPERR_OFPFMFC_OVERLAP;
4034 /* If necessary, evict an existing rule to clear out space. */
4035 error = evict_rules_from_table(ofproto, table, 1);
4037 cls_rule_destroy(&cr);
4041 /* Allocate new rule. */
4042 rule = ofproto->ofproto_class->rule_alloc();
4044 cls_rule_destroy(&cr);
4045 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
4046 ofproto->name, ovs_strerror(error));
4050 /* Initialize base state. */
4051 *CONST_CAST(struct ofproto **, &rule->ofproto) = ofproto;
4052 cls_rule_move(CONST_CAST(struct cls_rule *, &rule->cr), &cr);
4053 atomic_init(&rule->ref_count, 1);
4054 rule->pending = NULL;
4055 rule->flow_cookie = fm->new_cookie;
4056 rule->created = rule->modified = rule->used = time_msec();
4058 ovs_mutex_init(&rule->mutex);
4059 ovs_mutex_lock(&rule->mutex);
4060 rule->idle_timeout = fm->idle_timeout;
4061 rule->hard_timeout = fm->hard_timeout;
4062 ovs_mutex_unlock(&rule->mutex);
4064 *CONST_CAST(uint8_t *, &rule->table_id) = table - ofproto->tables;
4065 rule->flags = fm->flags & OFPUTIL_FF_STATE;
4066 rule->actions = rule_actions_create(ofproto, fm->ofpacts, fm->ofpacts_len);
4067 list_init(&rule->meter_list_node);
4068 rule->eviction_group = NULL;
4069 list_init(&rule->expirable);
4070 rule->monitor_flags = 0;
4071 rule->add_seqno = 0;
4072 rule->modify_seqno = 0;
4074 /* Construct rule, initializing derived state. */
4075 error = ofproto->ofproto_class->rule_construct(rule);
4077 ofproto_rule_destroy__(rule);
4082 oftable_insert_rule(rule);
4084 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
4085 ofoperation_create(group, rule, OFOPERATION_ADD, 0);
4086 ofproto->ofproto_class->rule_insert(rule);
4087 ofopgroup_submit(group);
4092 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
4094 /* Modifies the rules listed in 'rules', changing their actions to match those
4097 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4100 * Returns 0 on success, otherwise an OpenFlow error code. */
4102 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
4103 struct ofputil_flow_mod *fm, const struct ofp_header *request,
4104 const struct rule_collection *rules)
4105 OVS_REQUIRES(ofproto_mutex)
4107 enum ofoperation_type type;
4108 struct ofopgroup *group;
4112 type = fm->command == OFPFC_ADD ? OFOPERATION_REPLACE : OFOPERATION_MODIFY;
4113 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
4114 error = OFPERR_OFPBRC_EPERM;
4115 for (i = 0; i < rules->n; i++) {
4116 struct rule *rule = rules->rules[i];
4117 struct ofoperation *op;
4118 bool actions_changed;
4119 bool reset_counters;
4121 /* FIXME: Implement OFPFUTIL_FF_RESET_COUNTS */
4123 if (rule_is_modifiable(rule)) {
4124 /* At least one rule is modifiable, don't report EPERM error. */
4130 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
4131 rule->actions->ofpacts,
4132 rule->actions->ofpacts_len);
4134 op = ofoperation_create(group, rule, type, 0);
4136 if (fm->modify_cookie && fm->new_cookie != OVS_BE64_MAX) {
4137 ofproto_rule_change_cookie(ofproto, rule, fm->new_cookie);
4139 if (type == OFOPERATION_REPLACE) {
4140 ovs_mutex_lock(&rule->mutex);
4141 rule->idle_timeout = fm->idle_timeout;
4142 rule->hard_timeout = fm->hard_timeout;
4143 ovs_mutex_unlock(&rule->mutex);
4145 rule->flags = fm->flags & OFPUTIL_FF_STATE;
4146 if (fm->idle_timeout || fm->hard_timeout) {
4147 if (!rule->eviction_group) {
4148 eviction_group_add_rule(rule);
4151 eviction_group_remove_rule(rule);
4155 reset_counters = (fm->flags & OFPUTIL_FF_RESET_COUNTS) != 0;
4156 if (actions_changed || reset_counters) {
4157 struct rule_actions *new_actions;
4159 op->actions = rule->actions;
4160 new_actions = rule_actions_create(ofproto,
4161 fm->ofpacts, fm->ofpacts_len);
4163 ovs_mutex_lock(&rule->mutex);
4164 rule->actions = new_actions;
4165 ovs_mutex_unlock(&rule->mutex);
4167 rule->ofproto->ofproto_class->rule_modify_actions(rule,
4170 ofoperation_complete(op, 0);
4173 ofopgroup_submit(group);
4179 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
4180 struct ofputil_flow_mod *fm, const struct ofp_header *request)
4181 OVS_REQUIRES(ofproto_mutex)
4183 if (fm->cookie_mask != htonll(0) || fm->new_cookie == OVS_BE64_MAX) {
4186 return add_flow(ofproto, ofconn, fm, request);
4189 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
4192 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4195 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
4196 struct ofputil_flow_mod *fm,
4197 const struct ofp_header *request)
4198 OVS_REQUIRES(ofproto_mutex)
4200 struct rule_criteria criteria;
4201 struct rule_collection rules;
4204 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0,
4205 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY);
4206 error = collect_rules_loose(ofproto, &criteria, &rules);
4207 rule_criteria_destroy(&criteria);
4210 error = (rules.n > 0
4211 ? modify_flows__(ofproto, ofconn, fm, request, &rules)
4212 : modify_flows_add(ofproto, ofconn, fm, request));
4215 rule_collection_destroy(&rules);
4220 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4223 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4226 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
4227 struct ofputil_flow_mod *fm,
4228 const struct ofp_header *request)
4229 OVS_REQUIRES(ofproto_mutex)
4231 struct rule_criteria criteria;
4232 struct rule_collection rules;
4235 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
4236 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY);
4237 error = collect_rules_strict(ofproto, &criteria, &rules);
4238 rule_criteria_destroy(&criteria);
4242 error = modify_flows_add(ofproto, ofconn, fm, request);
4243 } else if (rules.n == 1) {
4244 error = modify_flows__(ofproto, ofconn, fm, request, &rules);
4248 rule_collection_destroy(&rules);
4253 /* OFPFC_DELETE implementation. */
4256 delete_flow__(struct rule *rule, struct ofopgroup *group,
4257 enum ofp_flow_removed_reason reason)
4258 OVS_REQUIRES(ofproto_mutex)
4260 struct ofproto *ofproto = rule->ofproto;
4262 ofproto_rule_send_removed(rule, reason);
4264 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
4265 oftable_remove_rule(rule);
4266 ofproto->ofproto_class->rule_delete(rule);
4269 /* Deletes the rules listed in 'rules'.
4271 * Returns 0 on success, otherwise an OpenFlow error code. */
4273 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
4274 const struct ofp_header *request,
4275 const struct rule_collection *rules,
4276 enum ofp_flow_removed_reason reason)
4277 OVS_REQUIRES(ofproto_mutex)
4279 struct ofopgroup *group;
4282 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
4283 for (i = 0; i < rules->n; i++) {
4284 delete_flow__(rules->rules[i], group, reason);
4286 ofopgroup_submit(group);
4291 /* Implements OFPFC_DELETE. */
4293 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
4294 const struct ofputil_flow_mod *fm,
4295 const struct ofp_header *request)
4296 OVS_REQUIRES(ofproto_mutex)
4298 struct rule_criteria criteria;
4299 struct rule_collection rules;
4302 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0,
4303 fm->cookie, fm->cookie_mask,
4304 fm->out_port, fm->out_group);
4305 error = collect_rules_loose(ofproto, &criteria, &rules);
4306 rule_criteria_destroy(&criteria);
4308 if (!error && rules.n > 0) {
4309 error = delete_flows__(ofproto, ofconn, request, &rules, OFPRR_DELETE);
4311 rule_collection_destroy(&rules);
4316 /* Implements OFPFC_DELETE_STRICT. */
4318 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
4319 const struct ofputil_flow_mod *fm,
4320 const struct ofp_header *request)
4321 OVS_REQUIRES(ofproto_mutex)
4323 struct rule_criteria criteria;
4324 struct rule_collection rules;
4327 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
4328 fm->cookie, fm->cookie_mask,
4329 fm->out_port, fm->out_group);
4330 error = collect_rules_strict(ofproto, &criteria, &rules);
4331 rule_criteria_destroy(&criteria);
4333 if (!error && rules.n > 0) {
4334 error = delete_flows__(ofproto, ofconn, request, &rules, OFPRR_DELETE);
4336 rule_collection_destroy(&rules);
4342 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
4343 OVS_REQUIRES(ofproto_mutex)
4345 struct ofputil_flow_removed fr;
4347 if (ofproto_rule_is_hidden(rule) ||
4348 !(rule->flags & OFPUTIL_FF_SEND_FLOW_REM)) {
4352 minimatch_expand(&rule->cr.match, &fr.match);
4353 fr.priority = rule->cr.priority;
4354 fr.cookie = rule->flow_cookie;
4356 fr.table_id = rule->table_id;
4357 calc_duration(rule->created, time_msec(),
4358 &fr.duration_sec, &fr.duration_nsec);
4359 ovs_mutex_lock(&rule->mutex);
4360 fr.idle_timeout = rule->idle_timeout;
4361 fr.hard_timeout = rule->hard_timeout;
4362 ovs_mutex_unlock(&rule->mutex);
4363 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
4366 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
4369 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
4370 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
4373 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
4376 * ofproto implementation ->run() functions should use this function to expire
4377 * OpenFlow flows. */
4379 ofproto_rule_expire(struct rule *rule, uint8_t reason)
4380 OVS_REQUIRES(ofproto_mutex)
4382 struct ofproto *ofproto = rule->ofproto;
4384 ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT
4385 || reason == OFPRR_DELETE || reason == OFPRR_GROUP_DELETE);
4387 ofproto_rule_delete__(ofproto, rule, reason);
4390 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
4391 * means "infinite". */
4393 reduce_timeout(uint16_t max, uint16_t *timeout)
4395 if (max && (!*timeout || *timeout > max)) {
4400 /* If 'idle_timeout' is nonzero, and 'rule' has no idle timeout or an idle
4401 * timeout greater than 'idle_timeout', lowers 'rule''s idle timeout to
4402 * 'idle_timeout' seconds. Similarly for 'hard_timeout'.
4404 * Suitable for implementing OFPACT_FIN_TIMEOUT. */
4406 ofproto_rule_reduce_timeouts(struct rule *rule,
4407 uint16_t idle_timeout, uint16_t hard_timeout)
4408 OVS_EXCLUDED(ofproto_mutex, rule->mutex)
4410 if (!idle_timeout && !hard_timeout) {
4414 ovs_mutex_lock(&ofproto_mutex);
4415 if (list_is_empty(&rule->expirable)) {
4416 list_insert(&rule->ofproto->expirable, &rule->expirable);
4418 ovs_mutex_unlock(&ofproto_mutex);
4420 ovs_mutex_lock(&rule->mutex);
4421 reduce_timeout(idle_timeout, &rule->idle_timeout);
4422 reduce_timeout(hard_timeout, &rule->hard_timeout);
4423 ovs_mutex_unlock(&rule->mutex);
4427 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4428 OVS_EXCLUDED(ofproto_mutex)
4430 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4431 struct ofputil_flow_mod fm;
4432 uint64_t ofpacts_stub[1024 / 8];
4433 struct ofpbuf ofpacts;
4437 error = reject_slave_controller(ofconn);
4442 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
4443 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
4445 u16_to_ofp(ofproto->max_ports),
4448 error = ofproto_check_ofpacts(ofproto, fm.ofpacts, fm.ofpacts_len);
4451 error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
4454 goto exit_free_ofpacts;
4457 /* Record the operation for logging a summary report. */
4458 switch (fm.command) {
4464 case OFPFC_MODIFY_STRICT:
4465 ofproto->n_modify++;
4469 case OFPFC_DELETE_STRICT:
4470 ofproto->n_delete++;
4475 if (ofproto->next_op_report == LLONG_MAX) {
4476 ofproto->first_op = now;
4477 ofproto->next_op_report = MAX(now + 10 * 1000,
4478 ofproto->op_backoff);
4479 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
4481 ofproto->last_op = now;
4484 ofpbuf_uninit(&ofpacts);
4490 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
4491 struct ofputil_flow_mod *fm, const struct ofp_header *oh)
4492 OVS_EXCLUDED(ofproto_mutex)
4496 ovs_mutex_lock(&ofproto_mutex);
4497 if (ofproto->n_pending < 50) {
4498 switch (fm->command) {
4500 error = add_flow(ofproto, ofconn, fm, oh);
4504 error = modify_flows_loose(ofproto, ofconn, fm, oh);
4507 case OFPFC_MODIFY_STRICT:
4508 error = modify_flow_strict(ofproto, ofconn, fm, oh);
4512 error = delete_flows_loose(ofproto, ofconn, fm, oh);
4515 case OFPFC_DELETE_STRICT:
4516 error = delete_flow_strict(ofproto, ofconn, fm, oh);
4520 if (fm->command > 0xff) {
4521 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
4522 "flow_mod_table_id extension is not enabled",
4525 error = OFPERR_OFPFMFC_BAD_COMMAND;
4529 ovs_assert(!list_is_empty(&ofproto->pending));
4530 error = OFPROTO_POSTPONE;
4532 ovs_mutex_unlock(&ofproto_mutex);
4534 run_rule_executes(ofproto);
4539 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4541 struct ofputil_role_request request;
4542 struct ofputil_role_request reply;
4546 error = ofputil_decode_role_message(oh, &request);
4551 if (request.role != OFPCR12_ROLE_NOCHANGE) {
4552 if (ofconn_get_role(ofconn) != request.role
4553 && ofconn_has_pending_opgroups(ofconn)) {
4554 return OFPROTO_POSTPONE;
4557 if (request.have_generation_id
4558 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
4559 return OFPERR_OFPRRFC_STALE;
4562 ofconn_set_role(ofconn, request.role);
4565 reply.role = ofconn_get_role(ofconn);
4566 reply.have_generation_id = ofconn_get_master_election_id(
4567 ofconn, &reply.generation_id);
4568 buf = ofputil_encode_role_reply(oh, &reply);
4569 ofconn_send_reply(ofconn, buf);
4575 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
4576 const struct ofp_header *oh)
4578 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
4579 enum ofputil_protocol cur, next;
4581 cur = ofconn_get_protocol(ofconn);
4582 next = ofputil_protocol_set_tid(cur, msg->set != 0);
4583 ofconn_set_protocol(ofconn, next);
4589 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4591 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
4592 enum ofputil_protocol cur, next;
4593 enum ofputil_protocol next_base;
4595 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
4597 return OFPERR_OFPBRC_EPERM;
4600 cur = ofconn_get_protocol(ofconn);
4601 next = ofputil_protocol_set_base(cur, next_base);
4602 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
4603 /* Avoid sending async messages in surprising protocol. */
4604 return OFPROTO_POSTPONE;
4607 ofconn_set_protocol(ofconn, next);
4612 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
4613 const struct ofp_header *oh)
4615 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
4618 format = ntohl(msg->format);
4619 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
4620 return OFPERR_OFPBRC_EPERM;
4623 if (format != ofconn_get_packet_in_format(ofconn)
4624 && ofconn_has_pending_opgroups(ofconn)) {
4625 /* Avoid sending async message in surprsing packet in format. */
4626 return OFPROTO_POSTPONE;
4629 ofconn_set_packet_in_format(ofconn, format);
4634 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
4636 const struct nx_async_config *msg = ofpmsg_body(oh);
4637 uint32_t master[OAM_N_TYPES];
4638 uint32_t slave[OAM_N_TYPES];
4640 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
4641 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
4642 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
4644 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
4645 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
4646 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
4648 ofconn_set_async_config(ofconn, master, slave);
4649 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
4650 !ofconn_get_miss_send_len(ofconn)) {
4651 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
4658 handle_nxt_get_async_request(struct ofconn *ofconn, const struct ofp_header *oh)
4661 uint32_t master[OAM_N_TYPES];
4662 uint32_t slave[OAM_N_TYPES];
4663 struct nx_async_config *msg;
4665 ofconn_get_async_config(ofconn, master, slave);
4666 buf = ofpraw_alloc_reply(OFPRAW_OFPT13_GET_ASYNC_REPLY, oh, 0);
4667 msg = ofpbuf_put_zeros(buf, sizeof *msg);
4669 msg->packet_in_mask[0] = htonl(master[OAM_PACKET_IN]);
4670 msg->port_status_mask[0] = htonl(master[OAM_PORT_STATUS]);
4671 msg->flow_removed_mask[0] = htonl(master[OAM_FLOW_REMOVED]);
4673 msg->packet_in_mask[1] = htonl(slave[OAM_PACKET_IN]);
4674 msg->port_status_mask[1] = htonl(slave[OAM_PORT_STATUS]);
4675 msg->flow_removed_mask[1] = htonl(slave[OAM_FLOW_REMOVED]);
4677 ofconn_send_reply(ofconn, buf);
4683 handle_nxt_set_controller_id(struct ofconn *ofconn,
4684 const struct ofp_header *oh)
4686 const struct nx_controller_id *nci = ofpmsg_body(oh);
4688 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
4689 return OFPERR_NXBRC_MUST_BE_ZERO;
4692 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
4697 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4701 if (ofconn_has_pending_opgroups(ofconn)) {
4702 return OFPROTO_POSTPONE;
4705 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
4706 ? OFPRAW_OFPT10_BARRIER_REPLY
4707 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
4708 ofconn_send_reply(ofconn, buf);
4713 ofproto_compose_flow_refresh_update(const struct rule *rule,
4714 enum nx_flow_monitor_flags flags,
4716 OVS_REQUIRES(ofproto_mutex)
4718 struct ofoperation *op = rule->pending;
4719 const struct rule_actions *actions;
4720 struct ofputil_flow_update fu;
4723 if (op && op->type == OFOPERATION_ADD) {
4724 /* We'll report the final flow when the operation completes. Reporting
4725 * it now would cause a duplicate report later. */
4729 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
4730 ? NXFME_ADDED : NXFME_MODIFIED);
4732 ovs_mutex_lock(&rule->mutex);
4733 fu.idle_timeout = rule->idle_timeout;
4734 fu.hard_timeout = rule->hard_timeout;
4735 ovs_mutex_unlock(&rule->mutex);
4736 fu.table_id = rule->table_id;
4737 fu.cookie = rule->flow_cookie;
4738 minimatch_expand(&rule->cr.match, &match);
4740 fu.priority = rule->cr.priority;
4742 if (!(flags & NXFMF_ACTIONS)) {
4745 actions = rule->actions;
4747 /* An operation is in progress. Use the previous version of the flow's
4748 * actions, so that when the operation commits we report the change. */
4750 case OFOPERATION_ADD:
4753 case OFOPERATION_MODIFY:
4754 case OFOPERATION_REPLACE:
4755 actions = op->actions ? op->actions : rule->actions;
4758 case OFOPERATION_DELETE:
4759 actions = rule->actions;
4766 fu.ofpacts = actions ? actions->ofpacts : NULL;
4767 fu.ofpacts_len = actions ? actions->ofpacts_len : 0;
4769 if (list_is_empty(msgs)) {
4770 ofputil_start_flow_update(msgs);
4772 ofputil_append_flow_update(&fu, msgs);
4776 ofmonitor_compose_refresh_updates(struct rule_collection *rules,
4778 OVS_REQUIRES(ofproto_mutex)
4782 for (i = 0; i < rules->n; i++) {
4783 struct rule *rule = rules->rules[i];
4784 enum nx_flow_monitor_flags flags = rule->monitor_flags;
4785 rule->monitor_flags = 0;
4787 ofproto_compose_flow_refresh_update(rule, flags, msgs);
4792 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
4793 struct rule *rule, uint64_t seqno,
4794 struct rule_collection *rules)
4795 OVS_REQUIRES(ofproto_mutex)
4797 enum nx_flow_monitor_flags update;
4799 if (ofproto_rule_is_hidden(rule)) {
4804 ? ofoperation_has_out_port(rule->pending, m->out_port)
4805 : ofproto_rule_has_out_port(rule, m->out_port))) {
4810 if (rule->add_seqno > seqno) {
4811 update = NXFMF_ADD | NXFMF_MODIFY;
4812 } else if (rule->modify_seqno > seqno) {
4813 update = NXFMF_MODIFY;
4818 if (!(m->flags & update)) {
4822 update = NXFMF_INITIAL;
4825 if (!rule->monitor_flags) {
4826 rule_collection_add(rules, rule);
4828 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
4832 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
4834 struct rule_collection *rules)
4835 OVS_REQUIRES(ofproto_mutex)
4837 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
4838 const struct ofoperation *op;
4839 const struct oftable *table;
4840 struct cls_rule target;
4842 cls_rule_init_from_minimatch(&target, &m->match, 0);
4843 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
4844 struct cls_cursor cursor;
4847 ovs_rwlock_rdlock(&table->cls.rwlock);
4848 cls_cursor_init(&cursor, &table->cls, &target);
4849 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4850 ovs_assert(!rule->pending); /* XXX */
4851 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4853 ovs_rwlock_unlock(&table->cls.rwlock);
4856 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
4857 struct rule *rule = op->rule;
4859 if (((m->table_id == 0xff
4860 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
4861 : m->table_id == rule->table_id))
4862 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
4863 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4866 cls_rule_destroy(&target);
4870 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
4871 struct rule_collection *rules)
4872 OVS_REQUIRES(ofproto_mutex)
4874 if (m->flags & NXFMF_INITIAL) {
4875 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
4880 ofmonitor_collect_resume_rules(struct ofmonitor *m,
4881 uint64_t seqno, struct rule_collection *rules)
4882 OVS_REQUIRES(ofproto_mutex)
4884 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
4888 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
4889 OVS_EXCLUDED(ofproto_mutex)
4891 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4892 struct ofmonitor **monitors;
4893 size_t n_monitors, allocated_monitors;
4894 struct rule_collection rules;
4895 struct list replies;
4901 ofpbuf_use_const(&b, oh, ntohs(oh->length));
4903 n_monitors = allocated_monitors = 0;
4905 ovs_mutex_lock(&ofproto_mutex);
4907 struct ofputil_flow_monitor_request request;
4908 struct ofmonitor *m;
4911 retval = ofputil_decode_flow_monitor_request(&request, &b);
4912 if (retval == EOF) {
4914 } else if (retval) {
4919 if (request.table_id != 0xff
4920 && request.table_id >= ofproto->n_tables) {
4921 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4925 error = ofmonitor_create(&request, ofconn, &m);
4930 if (n_monitors >= allocated_monitors) {
4931 monitors = x2nrealloc(monitors, &allocated_monitors,
4934 monitors[n_monitors++] = m;
4937 rule_collection_init(&rules);
4938 for (i = 0; i < n_monitors; i++) {
4939 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
4942 ofpmp_init(&replies, oh);
4943 ofmonitor_compose_refresh_updates(&rules, &replies);
4944 ovs_mutex_unlock(&ofproto_mutex);
4946 rule_collection_destroy(&rules);
4948 ofconn_send_replies(ofconn, &replies);
4954 for (i = 0; i < n_monitors; i++) {
4955 ofmonitor_destroy(monitors[i]);
4958 ovs_mutex_unlock(&ofproto_mutex);
4964 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
4965 OVS_EXCLUDED(ofproto_mutex)
4967 struct ofmonitor *m;
4971 id = ofputil_decode_flow_monitor_cancel(oh);
4973 ovs_mutex_lock(&ofproto_mutex);
4974 m = ofmonitor_lookup(ofconn, id);
4976 ofmonitor_destroy(m);
4979 error = OFPERR_NXBRC_FM_BAD_ID;
4981 ovs_mutex_unlock(&ofproto_mutex);
4986 /* Meters implementation.
4988 * Meter table entry, indexed by the OpenFlow meter_id.
4989 * These are always dynamically allocated to allocate enough space for
4991 * 'created' is used to compute the duration for meter stats.
4992 * 'list rules' is needed so that we can delete the dependent rules when the
4993 * meter table entry is deleted.
4994 * 'provider_meter_id' is for the provider's private use.
4997 long long int created; /* Time created. */
4998 struct list rules; /* List of "struct rule_dpif"s. */
4999 ofproto_meter_id provider_meter_id;
5000 uint16_t flags; /* Meter flags. */
5001 uint16_t n_bands; /* Number of meter bands. */
5002 struct ofputil_meter_band *bands;
5006 * This is used in instruction validation at flow set-up time,
5007 * as flows may not use non-existing meters.
5008 * Return value of UINT32_MAX signifies an invalid meter.
5011 get_provider_meter_id(const struct ofproto *ofproto, uint32_t of_meter_id)
5013 if (of_meter_id && of_meter_id <= ofproto->meter_features.max_meters) {
5014 const struct meter *meter = ofproto->meters[of_meter_id];
5016 return meter->provider_meter_id.uint32;
5023 meter_update(struct meter *meter, const struct ofputil_meter_config *config)
5027 meter->flags = config->flags;
5028 meter->n_bands = config->n_bands;
5029 meter->bands = xmemdup(config->bands,
5030 config->n_bands * sizeof *meter->bands);
5033 static struct meter *
5034 meter_create(const struct ofputil_meter_config *config,
5035 ofproto_meter_id provider_meter_id)
5037 struct meter *meter;
5039 meter = xzalloc(sizeof *meter);
5040 meter->provider_meter_id = provider_meter_id;
5041 meter->created = time_msec();
5042 list_init(&meter->rules);
5044 meter_update(meter, config);
5050 meter_delete(struct ofproto *ofproto, uint32_t first, uint32_t last)
5051 OVS_REQUIRES(ofproto_mutex)
5054 for (mid = first; mid <= last; ++mid) {
5055 struct meter *meter = ofproto->meters[mid];
5057 ofproto->meters[mid] = NULL;
5058 ofproto->ofproto_class->meter_del(ofproto,
5059 meter->provider_meter_id);
5067 handle_add_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
5069 ofproto_meter_id provider_meter_id = { UINT32_MAX };
5070 struct meter **meterp = &ofproto->meters[mm->meter.meter_id];
5074 return OFPERR_OFPMMFC_METER_EXISTS;
5077 error = ofproto->ofproto_class->meter_set(ofproto, &provider_meter_id,
5080 ovs_assert(provider_meter_id.uint32 != UINT32_MAX);
5081 *meterp = meter_create(&mm->meter, provider_meter_id);
5087 handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
5089 struct meter *meter = ofproto->meters[mm->meter.meter_id];
5091 uint32_t provider_meter_id;
5094 return OFPERR_OFPMMFC_UNKNOWN_METER;
5097 provider_meter_id = meter->provider_meter_id.uint32;
5098 error = ofproto->ofproto_class->meter_set(ofproto,
5099 &meter->provider_meter_id,
5101 ovs_assert(meter->provider_meter_id.uint32 == provider_meter_id);
5103 meter_update(meter, &mm->meter);
5109 handle_delete_meter(struct ofconn *ofconn, const struct ofp_header *oh,
5110 struct ofputil_meter_mod *mm)
5111 OVS_EXCLUDED(ofproto_mutex)
5113 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5114 uint32_t meter_id = mm->meter.meter_id;
5115 struct rule_collection rules;
5116 enum ofperr error = 0;
5117 uint32_t first, last;
5119 if (meter_id == OFPM13_ALL) {
5121 last = ofproto->meter_features.max_meters;
5123 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
5126 first = last = meter_id;
5129 /* First delete the rules that use this meter. If any of those rules are
5130 * currently being modified, postpone the whole operation until later. */
5131 rule_collection_init(&rules);
5132 ovs_mutex_lock(&ofproto_mutex);
5133 for (meter_id = first; meter_id <= last; ++meter_id) {
5134 struct meter *meter = ofproto->meters[meter_id];
5135 if (meter && !list_is_empty(&meter->rules)) {
5138 LIST_FOR_EACH (rule, meter_list_node, &meter->rules) {
5139 if (rule->pending) {
5140 error = OFPROTO_POSTPONE;
5143 rule_collection_add(&rules, rule);
5148 delete_flows__(ofproto, ofconn, oh, &rules, OFPRR_METER_DELETE);
5151 /* Delete the meters. */
5152 meter_delete(ofproto, first, last);
5155 ovs_mutex_unlock(&ofproto_mutex);
5156 rule_collection_destroy(&rules);
5162 handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5164 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5165 struct ofputil_meter_mod mm;
5166 uint64_t bands_stub[256 / 8];
5167 struct ofpbuf bands;
5171 error = reject_slave_controller(ofconn);
5176 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5178 error = ofputil_decode_meter_mod(oh, &mm, &bands);
5180 goto exit_free_bands;
5183 meter_id = mm.meter.meter_id;
5185 if (mm.command != OFPMC13_DELETE) {
5186 /* Fails also when meters are not implemented by the provider. */
5187 if (meter_id == 0 || meter_id > OFPM13_MAX) {
5188 error = OFPERR_OFPMMFC_INVALID_METER;
5189 goto exit_free_bands;
5190 } else if (meter_id > ofproto->meter_features.max_meters) {
5191 error = OFPERR_OFPMMFC_OUT_OF_METERS;
5192 goto exit_free_bands;
5194 if (mm.meter.n_bands > ofproto->meter_features.max_bands) {
5195 error = OFPERR_OFPMMFC_OUT_OF_BANDS;
5196 goto exit_free_bands;
5200 switch (mm.command) {
5202 error = handle_add_meter(ofproto, &mm);
5205 case OFPMC13_MODIFY:
5206 error = handle_modify_meter(ofproto, &mm);
5209 case OFPMC13_DELETE:
5210 error = handle_delete_meter(ofconn, oh, &mm);
5214 error = OFPERR_OFPMMFC_BAD_COMMAND;
5219 ofpbuf_uninit(&bands);
5224 handle_meter_features_request(struct ofconn *ofconn,
5225 const struct ofp_header *request)
5227 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5228 struct ofputil_meter_features features;
5231 if (ofproto->ofproto_class->meter_get_features) {
5232 ofproto->ofproto_class->meter_get_features(ofproto, &features);
5234 memset(&features, 0, sizeof features);
5236 b = ofputil_encode_meter_features_reply(&features, request);
5238 ofconn_send_reply(ofconn, b);
5243 handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request,
5246 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5247 struct list replies;
5248 uint64_t bands_stub[256 / 8];
5249 struct ofpbuf bands;
5250 uint32_t meter_id, first, last;
5252 ofputil_decode_meter_request(request, &meter_id);
5254 if (meter_id == OFPM13_ALL) {
5256 last = ofproto->meter_features.max_meters;
5258 if (!meter_id || meter_id > ofproto->meter_features.max_meters ||
5259 !ofproto->meters[meter_id]) {
5260 return OFPERR_OFPMMFC_UNKNOWN_METER;
5262 first = last = meter_id;
5265 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5266 ofpmp_init(&replies, request);
5268 for (meter_id = first; meter_id <= last; ++meter_id) {
5269 struct meter *meter = ofproto->meters[meter_id];
5271 continue; /* Skip non-existing meters. */
5273 if (type == OFPTYPE_METER_STATS_REQUEST) {
5274 struct ofputil_meter_stats stats;
5276 stats.meter_id = meter_id;
5278 /* Provider sets the packet and byte counts, we do the rest. */
5279 stats.flow_count = list_size(&meter->rules);
5280 calc_duration(meter->created, time_msec(),
5281 &stats.duration_sec, &stats.duration_nsec);
5282 stats.n_bands = meter->n_bands;
5283 ofpbuf_clear(&bands);
5285 = ofpbuf_put_uninit(&bands,
5286 meter->n_bands * sizeof *stats.bands);
5288 if (!ofproto->ofproto_class->meter_get(ofproto,
5289 meter->provider_meter_id,
5291 ofputil_append_meter_stats(&replies, &stats);
5293 } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */
5294 struct ofputil_meter_config config;
5296 config.meter_id = meter_id;
5297 config.flags = meter->flags;
5298 config.n_bands = meter->n_bands;
5299 config.bands = meter->bands;
5300 ofputil_append_meter_config(&replies, &config);
5304 ofconn_send_replies(ofconn, &replies);
5305 ofpbuf_uninit(&bands);
5310 ofproto_group_lookup(const struct ofproto *ofproto, uint32_t group_id,
5311 struct ofgroup **group)
5312 OVS_TRY_RDLOCK(true, (*group)->rwlock)
5314 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5315 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
5316 hash_int(group_id, 0), &ofproto->groups) {
5317 if ((*group)->group_id == group_id) {
5318 ovs_rwlock_rdlock(&(*group)->rwlock);
5319 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5323 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5328 ofproto_group_release(struct ofgroup *group)
5329 OVS_RELEASES(group->rwlock)
5331 ovs_rwlock_unlock(&group->rwlock);
5335 ofproto_group_write_lookup(const struct ofproto *ofproto, uint32_t group_id,
5336 struct ofgroup **group)
5337 OVS_TRY_WRLOCK(true, ofproto->groups_rwlock)
5338 OVS_TRY_WRLOCK(true, (*group)->rwlock)
5340 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5341 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
5342 hash_int(group_id, 0), &ofproto->groups) {
5343 if ((*group)->group_id == group_id) {
5344 ovs_rwlock_wrlock(&(*group)->rwlock);
5348 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5353 ofproto_group_exists__(const struct ofproto *ofproto, uint32_t group_id)
5354 OVS_REQ_RDLOCK(ofproto->groups_rwlock)
5356 struct ofgroup *grp;
5358 HMAP_FOR_EACH_IN_BUCKET (grp, hmap_node,
5359 hash_int(group_id, 0), &ofproto->groups) {
5360 if (grp->group_id == group_id) {
5368 ofproto_group_exists(const struct ofproto *ofproto, uint32_t group_id)
5369 OVS_EXCLUDED(ofproto->groups_rwlock)
5373 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5374 exists = ofproto_group_exists__(ofproto, group_id);
5375 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5381 group_get_ref_count(struct ofgroup *group)
5382 OVS_EXCLUDED(ofproto_mutex)
5384 struct ofproto *ofproto = group->ofproto;
5385 struct rule_criteria criteria;
5386 struct rule_collection rules;
5391 match_init_catchall(&match);
5392 rule_criteria_init(&criteria, 0xff, &match, 0, htonll(0), htonll(0),
5393 OFPP_ANY, group->group_id);
5394 ovs_mutex_lock(&ofproto_mutex);
5395 error = collect_rules_loose(ofproto, &criteria, &rules);
5396 ovs_mutex_unlock(&ofproto_mutex);
5397 rule_criteria_destroy(&criteria);
5399 count = !error && rules.n < UINT32_MAX ? rules.n : UINT32_MAX;
5401 rule_collection_destroy(&rules);
5406 append_group_stats(struct ofgroup *group, struct list *replies)
5407 OVS_REQ_RDLOCK(group->rwlock)
5409 struct ofputil_group_stats ogs;
5410 struct ofproto *ofproto = group->ofproto;
5411 long long int now = time_msec();
5414 ogs.bucket_stats = xmalloc(group->n_buckets * sizeof *ogs.bucket_stats);
5416 /* Provider sets the packet and byte counts, we do the rest. */
5417 ogs.ref_count = group_get_ref_count(group);
5418 ogs.n_buckets = group->n_buckets;
5420 error = (ofproto->ofproto_class->group_get_stats
5421 ? ofproto->ofproto_class->group_get_stats(group, &ogs)
5424 ogs.packet_count = UINT64_MAX;
5425 ogs.byte_count = UINT64_MAX;
5426 memset(ogs.bucket_stats, 0xff,
5427 ogs.n_buckets * sizeof *ogs.bucket_stats);
5430 ogs.group_id = group->group_id;
5431 calc_duration(group->created, now, &ogs.duration_sec, &ogs.duration_nsec);
5433 ofputil_append_group_stats(replies, &ogs);
5435 free(ogs.bucket_stats);
5439 handle_group_stats_request(struct ofconn *ofconn,
5440 const struct ofp_header *request)
5442 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5443 struct list replies;
5445 struct ofgroup *group;
5448 error = ofputil_decode_group_stats_request(request, &group_id);
5453 ofpmp_init(&replies, request);
5455 if (group_id == OFPG_ALL) {
5456 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5457 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
5458 ovs_rwlock_rdlock(&group->rwlock);
5459 append_group_stats(group, &replies);
5460 ovs_rwlock_unlock(&group->rwlock);
5462 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5464 if (ofproto_group_lookup(ofproto, group_id, &group)) {
5465 append_group_stats(group, &replies);
5466 ofproto_group_release(group);
5470 ofconn_send_replies(ofconn, &replies);
5476 handle_group_desc_stats_request(struct ofconn *ofconn,
5477 const struct ofp_header *request)
5479 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5480 struct list replies;
5481 struct ofputil_group_desc gds;
5482 struct ofgroup *group;
5484 ofpmp_init(&replies, request);
5486 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5487 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
5488 gds.group_id = group->group_id;
5489 gds.type = group->type;
5490 ofputil_append_group_desc_reply(&gds, &group->buckets, &replies);
5492 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5494 ofconn_send_replies(ofconn, &replies);
5500 handle_group_features_stats_request(struct ofconn *ofconn,
5501 const struct ofp_header *request)
5503 struct ofproto *p = ofconn_get_ofproto(ofconn);
5506 msg = ofputil_encode_group_features_reply(&p->ogf, request);
5508 ofconn_send_reply(ofconn, msg);
5515 handle_queue_get_config_request(struct ofconn *ofconn,
5516 const struct ofp_header *oh)
5518 struct ofproto *p = ofconn_get_ofproto(ofconn);
5519 struct netdev_queue_dump queue_dump;
5520 struct ofport *ofport;
5521 unsigned int queue_id;
5522 struct ofpbuf *reply;
5523 struct smap details;
5527 error = ofputil_decode_queue_get_config_request(oh, &request);
5532 ofport = ofproto_get_port(p, request);
5534 return OFPERR_OFPQOFC_BAD_PORT;
5537 reply = ofputil_encode_queue_get_config_reply(oh);
5539 smap_init(&details);
5540 NETDEV_QUEUE_FOR_EACH (&queue_id, &details, &queue_dump, ofport->netdev) {
5541 struct ofputil_queue_config queue;
5543 /* None of the existing queues have compatible properties, so we
5544 * hard-code omitting min_rate and max_rate. */
5545 queue.queue_id = queue_id;
5546 queue.min_rate = UINT16_MAX;
5547 queue.max_rate = UINT16_MAX;
5548 ofputil_append_queue_get_config_reply(reply, &queue);
5550 smap_destroy(&details);
5552 ofconn_send_reply(ofconn, reply);
5557 /* Implements OFPGC11_ADD
5558 * in which no matching flow already exists in the flow table.
5560 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
5561 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
5562 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
5563 * initiated now but may be retried later.
5565 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
5566 * ownership remains with the caller.
5568 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
5571 add_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
5573 struct ofgroup *ofgroup;
5576 if (gm->group_id > OFPG_MAX) {
5577 return OFPERR_OFPGMFC_INVALID_GROUP;
5579 if (gm->type > OFPGT11_FF) {
5580 return OFPERR_OFPGMFC_BAD_TYPE;
5583 /* Allocate new group and initialize it. */
5584 ofgroup = ofproto->ofproto_class->group_alloc();
5586 VLOG_WARN_RL(&rl, "%s: failed to create group", ofproto->name);
5587 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
5590 ovs_rwlock_init(&ofgroup->rwlock);
5591 ofgroup->ofproto = ofproto;
5592 ofgroup->group_id = gm->group_id;
5593 ofgroup->type = gm->type;
5594 ofgroup->created = ofgroup->modified = time_msec();
5596 list_move(&ofgroup->buckets, &gm->buckets);
5597 ofgroup->n_buckets = list_size(&ofgroup->buckets);
5599 /* Construct called BEFORE any locks are held. */
5600 error = ofproto->ofproto_class->group_construct(ofgroup);
5605 /* We wrlock as late as possible to minimize the time we jam any other
5606 * threads: No visible state changes before acquiring the lock. */
5607 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5609 if (ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
5610 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
5614 if (ofproto_group_exists__(ofproto, gm->group_id)) {
5615 error = OFPERR_OFPGMFC_GROUP_EXISTS;
5620 /* Insert new group. */
5621 hmap_insert(&ofproto->groups, &ofgroup->hmap_node,
5622 hash_int(ofgroup->group_id, 0));
5623 ofproto->n_groups[ofgroup->type]++;
5625 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5630 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5631 ofproto->ofproto_class->group_destruct(ofgroup);
5633 ofputil_bucket_list_destroy(&ofgroup->buckets);
5634 ofproto->ofproto_class->group_dealloc(ofgroup);
5639 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
5642 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
5645 modify_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
5647 struct ofgroup *ofgroup;
5648 struct ofgroup *victim;
5651 if (gm->group_id > OFPG_MAX) {
5652 return OFPERR_OFPGMFC_INVALID_GROUP;
5655 if (gm->type > OFPGT11_FF) {
5656 return OFPERR_OFPGMFC_BAD_TYPE;
5659 victim = ofproto->ofproto_class->group_alloc();
5661 VLOG_WARN_RL(&rl, "%s: failed to allocate group", ofproto->name);
5662 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
5665 if (!ofproto_group_write_lookup(ofproto, gm->group_id, &ofgroup)) {
5666 error = OFPERR_OFPGMFC_UNKNOWN_GROUP;
5669 /* Both group's and its container's write locks held now.
5670 * Also, n_groups[] is protected by ofproto->groups_rwlock. */
5671 if (ofgroup->type != gm->type
5672 && ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
5673 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
5678 list_move(&victim->buckets, &ofgroup->buckets);
5680 ofgroup->type = gm->type;
5681 list_move(&ofgroup->buckets, &gm->buckets);
5682 ofgroup->n_buckets = list_size(&ofgroup->buckets);
5684 error = ofproto->ofproto_class->group_modify(ofgroup, victim);
5686 ofputil_bucket_list_destroy(&victim->buckets);
5687 ofproto->n_groups[victim->type]--;
5688 ofproto->n_groups[ofgroup->type]++;
5689 ofgroup->modified = time_msec();
5691 ofputil_bucket_list_destroy(&ofgroup->buckets);
5694 list_move(&ofgroup->buckets, &victim->buckets);
5698 ovs_rwlock_unlock(&ofgroup->rwlock);
5699 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5701 ofproto->ofproto_class->group_dealloc(victim);
5706 delete_group__(struct ofproto *ofproto, struct ofgroup *ofgroup)
5707 OVS_RELEASES(ofproto->groups_rwlock)
5710 struct ofputil_flow_mod fm;
5712 /* Delete all flow entries containing this group in a group action */
5713 match_init_catchall(&match);
5714 flow_mod_init(&fm, &match, 0, NULL, 0, OFPFC_DELETE);
5715 fm.out_group = ofgroup->group_id;
5716 handle_flow_mod__(ofproto, NULL, &fm, NULL);
5718 /* Must wait until existing readers are done,
5719 * while holding the container's write lock at the same time. */
5720 ovs_rwlock_wrlock(&ofgroup->rwlock);
5721 hmap_remove(&ofproto->groups, &ofgroup->hmap_node);
5722 /* No-one can find this group any more. */
5723 ofproto->n_groups[ofgroup->type]--;
5724 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5726 ofproto->ofproto_class->group_destruct(ofgroup);
5727 ofputil_bucket_list_destroy(&ofgroup->buckets);
5728 ovs_rwlock_unlock(&ofgroup->rwlock);
5729 ovs_rwlock_destroy(&ofgroup->rwlock);
5730 ofproto->ofproto_class->group_dealloc(ofgroup);
5733 /* Implements OFPGC_DELETE. */
5735 delete_group(struct ofproto *ofproto, uint32_t group_id)
5737 struct ofgroup *ofgroup;
5739 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5740 if (group_id == OFPG_ALL) {
5742 struct hmap_node *node = hmap_first(&ofproto->groups);
5746 ofgroup = CONTAINER_OF(node, struct ofgroup, hmap_node);
5747 delete_group__(ofproto, ofgroup);
5748 /* Lock for each node separately, so that we will not jam the
5749 * other threads for too long time. */
5750 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5753 HMAP_FOR_EACH_IN_BUCKET (ofgroup, hmap_node,
5754 hash_int(group_id, 0), &ofproto->groups) {
5755 if (ofgroup->group_id == group_id) {
5756 delete_group__(ofproto, ofgroup);
5761 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5765 handle_group_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5767 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5768 struct ofputil_group_mod gm;
5771 error = reject_slave_controller(ofconn);
5776 error = ofputil_decode_group_mod(oh, &gm);
5781 switch (gm.command) {
5783 return add_group(ofproto, &gm);
5785 case OFPGC11_MODIFY:
5786 return modify_group(ofproto, &gm);
5788 case OFPGC11_DELETE:
5789 delete_group(ofproto, gm.group_id);
5793 if (gm.command > OFPGC11_DELETE) {
5794 VLOG_WARN_RL(&rl, "%s: Invalid group_mod command type %d",
5795 ofproto->name, gm.command);
5797 return OFPERR_OFPGMFC_BAD_COMMAND;
5802 handle_table_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5804 struct ofputil_table_mod tm;
5807 error = reject_slave_controller(ofconn);
5812 error = ofputil_decode_table_mod(oh, &tm);
5817 /* XXX Actual table mod support is not implemented yet. */
5822 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
5823 OVS_EXCLUDED(ofproto_mutex)
5825 const struct ofp_header *oh = msg->data;
5829 error = ofptype_decode(&type, oh);
5833 if (oh->version >= OFP13_VERSION && ofpmsg_is_stat_request(oh)
5834 && ofpmp_more(oh)) {
5835 /* We have no buffer implementation for multipart requests.
5836 * Report overflow for requests which consists of multiple
5838 return OFPERR_OFPBRC_MULTIPART_BUFFER_OVERFLOW;
5842 /* OpenFlow requests. */
5843 case OFPTYPE_ECHO_REQUEST:
5844 return handle_echo_request(ofconn, oh);
5846 case OFPTYPE_FEATURES_REQUEST:
5847 return handle_features_request(ofconn, oh);
5849 case OFPTYPE_GET_CONFIG_REQUEST:
5850 return handle_get_config_request(ofconn, oh);
5852 case OFPTYPE_SET_CONFIG:
5853 return handle_set_config(ofconn, oh);
5855 case OFPTYPE_PACKET_OUT:
5856 return handle_packet_out(ofconn, oh);
5858 case OFPTYPE_PORT_MOD:
5859 return handle_port_mod(ofconn, oh);
5861 case OFPTYPE_FLOW_MOD:
5862 return handle_flow_mod(ofconn, oh);
5864 case OFPTYPE_GROUP_MOD:
5865 return handle_group_mod(ofconn, oh);
5867 case OFPTYPE_TABLE_MOD:
5868 return handle_table_mod(ofconn, oh);
5870 case OFPTYPE_METER_MOD:
5871 return handle_meter_mod(ofconn, oh);
5873 case OFPTYPE_BARRIER_REQUEST:
5874 return handle_barrier_request(ofconn, oh);
5876 case OFPTYPE_ROLE_REQUEST:
5877 return handle_role_request(ofconn, oh);
5879 /* OpenFlow replies. */
5880 case OFPTYPE_ECHO_REPLY:
5883 /* Nicira extension requests. */
5884 case OFPTYPE_FLOW_MOD_TABLE_ID:
5885 return handle_nxt_flow_mod_table_id(ofconn, oh);
5887 case OFPTYPE_SET_FLOW_FORMAT:
5888 return handle_nxt_set_flow_format(ofconn, oh);
5890 case OFPTYPE_SET_PACKET_IN_FORMAT:
5891 return handle_nxt_set_packet_in_format(ofconn, oh);
5893 case OFPTYPE_SET_CONTROLLER_ID:
5894 return handle_nxt_set_controller_id(ofconn, oh);
5896 case OFPTYPE_FLOW_AGE:
5897 /* Nothing to do. */
5900 case OFPTYPE_FLOW_MONITOR_CANCEL:
5901 return handle_flow_monitor_cancel(ofconn, oh);
5903 case OFPTYPE_SET_ASYNC_CONFIG:
5904 return handle_nxt_set_async_config(ofconn, oh);
5906 case OFPTYPE_GET_ASYNC_REQUEST:
5907 return handle_nxt_get_async_request(ofconn, oh);
5909 /* Statistics requests. */
5910 case OFPTYPE_DESC_STATS_REQUEST:
5911 return handle_desc_stats_request(ofconn, oh);
5913 case OFPTYPE_FLOW_STATS_REQUEST:
5914 return handle_flow_stats_request(ofconn, oh);
5916 case OFPTYPE_AGGREGATE_STATS_REQUEST:
5917 return handle_aggregate_stats_request(ofconn, oh);
5919 case OFPTYPE_TABLE_STATS_REQUEST:
5920 return handle_table_stats_request(ofconn, oh);
5922 case OFPTYPE_PORT_STATS_REQUEST:
5923 return handle_port_stats_request(ofconn, oh);
5925 case OFPTYPE_QUEUE_STATS_REQUEST:
5926 return handle_queue_stats_request(ofconn, oh);
5928 case OFPTYPE_PORT_DESC_STATS_REQUEST:
5929 return handle_port_desc_stats_request(ofconn, oh);
5931 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
5932 return handle_flow_monitor_request(ofconn, oh);
5934 case OFPTYPE_METER_STATS_REQUEST:
5935 case OFPTYPE_METER_CONFIG_STATS_REQUEST:
5936 return handle_meter_request(ofconn, oh, type);
5938 case OFPTYPE_METER_FEATURES_STATS_REQUEST:
5939 return handle_meter_features_request(ofconn, oh);
5941 case OFPTYPE_GROUP_STATS_REQUEST:
5942 return handle_group_stats_request(ofconn, oh);
5944 case OFPTYPE_GROUP_DESC_STATS_REQUEST:
5945 return handle_group_desc_stats_request(ofconn, oh);
5947 case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
5948 return handle_group_features_stats_request(ofconn, oh);
5950 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
5951 return handle_queue_get_config_request(ofconn, oh);
5955 case OFPTYPE_FEATURES_REPLY:
5956 case OFPTYPE_GET_CONFIG_REPLY:
5957 case OFPTYPE_PACKET_IN:
5958 case OFPTYPE_FLOW_REMOVED:
5959 case OFPTYPE_PORT_STATUS:
5960 case OFPTYPE_BARRIER_REPLY:
5961 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
5962 case OFPTYPE_DESC_STATS_REPLY:
5963 case OFPTYPE_FLOW_STATS_REPLY:
5964 case OFPTYPE_QUEUE_STATS_REPLY:
5965 case OFPTYPE_PORT_STATS_REPLY:
5966 case OFPTYPE_TABLE_STATS_REPLY:
5967 case OFPTYPE_AGGREGATE_STATS_REPLY:
5968 case OFPTYPE_PORT_DESC_STATS_REPLY:
5969 case OFPTYPE_ROLE_REPLY:
5970 case OFPTYPE_FLOW_MONITOR_PAUSED:
5971 case OFPTYPE_FLOW_MONITOR_RESUMED:
5972 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
5973 case OFPTYPE_GET_ASYNC_REPLY:
5974 case OFPTYPE_GROUP_STATS_REPLY:
5975 case OFPTYPE_GROUP_DESC_STATS_REPLY:
5976 case OFPTYPE_GROUP_FEATURES_STATS_REPLY:
5977 case OFPTYPE_METER_STATS_REPLY:
5978 case OFPTYPE_METER_CONFIG_STATS_REPLY:
5979 case OFPTYPE_METER_FEATURES_STATS_REPLY:
5980 case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
5981 case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
5982 case OFPTYPE_ROLE_STATUS:
5984 if (ofpmsg_is_stat_request(oh)) {
5985 return OFPERR_OFPBRC_BAD_STAT;
5987 return OFPERR_OFPBRC_BAD_TYPE;
5993 handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
5994 OVS_EXCLUDED(ofproto_mutex)
5996 int error = handle_openflow__(ofconn, ofp_msg);
5997 if (error && error != OFPROTO_POSTPONE) {
5998 ofconn_send_error(ofconn, ofp_msg->data, error);
6000 COVERAGE_INC(ofproto_recv_openflow);
6001 return error != OFPROTO_POSTPONE;
6004 /* Asynchronous operations. */
6006 /* Creates and returns a new ofopgroup that is not associated with any
6007 * OpenFlow connection.
6009 * The caller should add operations to the returned group with
6010 * ofoperation_create() and then submit it with ofopgroup_submit(). */
6011 static struct ofopgroup *
6012 ofopgroup_create_unattached(struct ofproto *ofproto)
6013 OVS_REQUIRES(ofproto_mutex)
6015 struct ofopgroup *group = xzalloc(sizeof *group);
6016 group->ofproto = ofproto;
6017 list_init(&group->ofproto_node);
6018 list_init(&group->ops);
6019 list_init(&group->ofconn_node);
6023 /* Creates and returns a new ofopgroup for 'ofproto'.
6025 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
6026 * connection. The 'request' and 'buffer_id' arguments are ignored.
6028 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
6029 * If the ofopgroup eventually fails, then the error reply will include
6030 * 'request'. If the ofopgroup eventually succeeds, then the packet with
6031 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
6033 * The caller should add operations to the returned group with
6034 * ofoperation_create() and then submit it with ofopgroup_submit(). */
6035 static struct ofopgroup *
6036 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
6037 const struct ofp_header *request, uint32_t buffer_id)
6038 OVS_REQUIRES(ofproto_mutex)
6040 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
6042 size_t request_len = ntohs(request->length);
6044 ovs_assert(ofconn_get_ofproto(ofconn) == ofproto);
6046 ofconn_add_opgroup(ofconn, &group->ofconn_node);
6047 group->ofconn = ofconn;
6048 group->request = xmemdup(request, MIN(request_len, 64));
6049 group->buffer_id = buffer_id;
6054 /* Submits 'group' for processing.
6056 * If 'group' contains no operations (e.g. none were ever added, or all of the
6057 * ones that were added completed synchronously), then it is destroyed
6058 * immediately. Otherwise it is added to the ofproto's list of pending
6061 ofopgroup_submit(struct ofopgroup *group)
6062 OVS_REQUIRES(ofproto_mutex)
6064 if (!group->n_running) {
6065 ofopgroup_complete(group);
6067 list_push_back(&group->ofproto->pending, &group->ofproto_node);
6068 group->ofproto->n_pending++;
6073 ofopgroup_complete(struct ofopgroup *group)
6074 OVS_REQUIRES(ofproto_mutex)
6076 struct ofproto *ofproto = group->ofproto;
6078 struct ofconn *abbrev_ofconn;
6079 ovs_be32 abbrev_xid;
6081 struct ofoperation *op, *next_op;
6084 ovs_assert(!group->n_running);
6087 LIST_FOR_EACH (op, group_node, &group->ops) {
6094 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
6095 LIST_FOR_EACH (op, group_node, &group->ops) {
6096 if (op->type != OFOPERATION_DELETE) {
6097 struct ofpbuf *packet;
6100 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
6103 struct rule_execute *re;
6107 ofproto_rule_ref(op->rule);
6109 re = xmalloc(sizeof *re);
6110 re->rule = op->rule;
6111 re->in_port = in_port;
6112 re->packet = packet;
6114 if (!guarded_list_push_back(&ofproto->rule_executes,
6115 &re->list_node, 1024)) {
6116 ofproto_rule_unref(op->rule);
6117 ofpbuf_delete(re->packet);
6126 if (!error && !list_is_empty(&group->ofconn_node)) {
6127 abbrev_ofconn = group->ofconn;
6128 abbrev_xid = group->request->xid;
6130 abbrev_ofconn = NULL;
6131 abbrev_xid = htonl(0);
6133 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
6134 struct rule *rule = op->rule;
6136 /* We generally want to report the change to active OpenFlow flow
6137 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
6139 - The operation failed.
6141 - The affected rule is not visible to controllers.
6143 - The operation's only effect was to update rule->modified. */
6145 || ofproto_rule_is_hidden(rule)
6146 || (op->type == OFOPERATION_MODIFY
6148 && rule->flow_cookie == op->flow_cookie))) {
6149 /* Check that we can just cast from ofoperation_type to
6150 * nx_flow_update_event. */
6151 enum nx_flow_update_event event_type;
6154 case OFOPERATION_ADD:
6155 case OFOPERATION_REPLACE:
6156 event_type = NXFME_ADDED;
6159 case OFOPERATION_DELETE:
6160 event_type = NXFME_DELETED;
6163 case OFOPERATION_MODIFY:
6164 event_type = NXFME_MODIFIED;
6171 ofmonitor_report(ofproto->connmgr, rule, event_type,
6172 op->reason, abbrev_ofconn, abbrev_xid);
6175 rule->pending = NULL;
6178 case OFOPERATION_ADD:
6182 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
6183 if (vid_mask == VLAN_VID_MASK) {
6184 if (ofproto->vlan_bitmap) {
6185 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
6186 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
6187 bitmap_set1(ofproto->vlan_bitmap, vid);
6188 ofproto->vlans_changed = true;
6191 ofproto->vlans_changed = true;
6195 oftable_remove_rule(rule);
6196 ofproto_rule_unref(rule);
6200 case OFOPERATION_DELETE:
6201 ovs_assert(!op->error);
6202 ofproto_rule_unref(rule);
6206 case OFOPERATION_MODIFY:
6207 case OFOPERATION_REPLACE:
6209 long long int now = time_msec();
6211 rule->modified = now;
6212 if (op->type == OFOPERATION_REPLACE) {
6213 rule->created = rule->used = now;
6216 ofproto_rule_change_cookie(ofproto, rule, op->flow_cookie);
6217 ovs_mutex_lock(&rule->mutex);
6218 rule->idle_timeout = op->idle_timeout;
6219 rule->hard_timeout = op->hard_timeout;
6220 ovs_mutex_unlock(&rule->mutex);
6222 struct rule_actions *old_actions;
6224 ovs_mutex_lock(&rule->mutex);
6225 old_actions = rule->actions;
6226 rule->actions = op->actions;
6227 ovs_mutex_unlock(&rule->mutex);
6230 rule_actions_unref(old_actions);
6232 rule->flags = op->flags;
6240 ofoperation_destroy(op);
6243 ofmonitor_flush(ofproto->connmgr);
6245 if (!list_is_empty(&group->ofproto_node)) {
6246 ovs_assert(ofproto->n_pending > 0);
6247 ofproto->n_pending--;
6248 list_remove(&group->ofproto_node);
6250 if (!list_is_empty(&group->ofconn_node)) {
6251 list_remove(&group->ofconn_node);
6253 ofconn_send_error(group->ofconn, group->request, error);
6255 connmgr_retry(ofproto->connmgr);
6257 free(group->request);
6261 /* Initiates a new operation on 'rule', of the specified 'type', within
6262 * 'group'. Prior to calling, 'rule' must not have any pending operation.
6264 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
6265 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
6267 * Returns the newly created ofoperation (which is also available as
6268 * rule->pending). */
6269 static struct ofoperation *
6270 ofoperation_create(struct ofopgroup *group, struct rule *rule,
6271 enum ofoperation_type type,
6272 enum ofp_flow_removed_reason reason)
6273 OVS_REQUIRES(ofproto_mutex)
6275 struct ofproto *ofproto = group->ofproto;
6276 struct ofoperation *op;
6278 ovs_assert(!rule->pending);
6280 op = rule->pending = xzalloc(sizeof *op);
6282 list_push_back(&group->ops, &op->group_node);
6285 op->reason = reason;
6286 op->flow_cookie = rule->flow_cookie;
6287 ovs_mutex_lock(&rule->mutex);
6288 op->idle_timeout = rule->idle_timeout;
6289 op->hard_timeout = rule->hard_timeout;
6290 ovs_mutex_unlock(&rule->mutex);
6291 op->flags = rule->flags;
6295 if (type == OFOPERATION_DELETE) {
6296 hmap_insert(&ofproto->deletions, &op->hmap_node,
6297 cls_rule_hash(&rule->cr, rule->table_id));
6304 ofoperation_destroy(struct ofoperation *op)
6305 OVS_REQUIRES(ofproto_mutex)
6307 struct ofopgroup *group = op->group;
6310 op->rule->pending = NULL;
6312 if (op->type == OFOPERATION_DELETE) {
6313 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
6315 list_remove(&op->group_node);
6316 rule_actions_unref(op->actions);
6320 /* Indicates that 'op' completed with status 'error', which is either 0 to
6321 * indicate success or an OpenFlow error code on failure.
6323 * If 'error' is 0, indicating success, the operation will be committed
6324 * permanently to the flow table.
6326 * If 'error' is nonzero, then generally the operation will be rolled back:
6328 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
6329 * restores the original rule. The caller must have uninitialized any
6330 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
6331 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
6332 * and 7 for the new rule, calling its ->rule_dealloc() function.
6334 * - If 'op' is a "modify flow" operation, ofproto restores the original
6337 * - 'op' must not be a "delete flow" operation. Removing a rule is not
6338 * allowed to fail. It must always succeed.
6340 * Please see the large comment in ofproto/ofproto-provider.h titled
6341 * "Asynchronous Operation Support" for more information. */
6343 ofoperation_complete(struct ofoperation *op, enum ofperr error)
6345 struct ofopgroup *group = op->group;
6347 ovs_assert(group->n_running > 0);
6348 ovs_assert(!error || op->type != OFOPERATION_DELETE);
6351 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
6352 /* This function can be called from ->rule_construct(), in which case
6353 * ofproto_mutex is held, or it can be called from ->run(), in which
6354 * case ofproto_mutex is not held. But only in the latter case can we
6355 * arrive here, so we can safely take ofproto_mutex now. */
6356 ovs_mutex_lock(&ofproto_mutex);
6357 ovs_assert(op->rule->pending == op);
6358 ofopgroup_complete(group);
6359 ovs_mutex_unlock(&ofproto_mutex);
6364 pick_datapath_id(const struct ofproto *ofproto)
6366 const struct ofport *port;
6368 port = ofproto_get_port(ofproto, OFPP_LOCAL);
6370 uint8_t ea[ETH_ADDR_LEN];
6373 error = netdev_get_etheraddr(port->netdev, ea);
6375 return eth_addr_to_uint64(ea);
6377 VLOG_WARN("%s: could not get MAC address for %s (%s)",
6378 ofproto->name, netdev_get_name(port->netdev),
6379 ovs_strerror(error));
6381 return ofproto->fallback_dpid;
6385 pick_fallback_dpid(void)
6387 uint8_t ea[ETH_ADDR_LEN];
6388 eth_addr_nicira_random(ea);
6389 return eth_addr_to_uint64(ea);
6392 /* Table overflow policy. */
6394 /* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep'
6395 * to NULL if the table is not configured to evict rules or if the table
6396 * contains no evictable rules. (Rules with a readlock on their evict rwlock,
6397 * or with no timeouts are not evictable.) */
6399 choose_rule_to_evict(struct oftable *table, struct rule **rulep)
6400 OVS_REQUIRES(ofproto_mutex)
6402 struct eviction_group *evg;
6405 if (!table->eviction_fields) {
6409 /* In the common case, the outer and inner loops here will each be entered
6412 * - The inner loop normally "return"s in its first iteration. If the
6413 * eviction group has any evictable rules, then it always returns in
6416 * - The outer loop only iterates more than once if the largest eviction
6417 * group has no evictable rules.
6419 * - The outer loop can exit only if table's 'max_flows' is all filled up
6420 * by unevictable rules. */
6421 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
6424 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
6433 /* Searches 'ofproto' for tables that have more flows than their configured
6434 * maximum and that have flow eviction enabled, and evicts as many flows as
6435 * necessary and currently feasible from them.
6437 * This triggers only when an OpenFlow table has N flows in it and then the
6438 * client configures a maximum number of flows less than N. */
6440 ofproto_evict(struct ofproto *ofproto)
6442 struct oftable *table;
6444 ovs_mutex_lock(&ofproto_mutex);
6445 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
6446 evict_rules_from_table(ofproto, table, 0);
6448 ovs_mutex_unlock(&ofproto_mutex);
6451 /* Eviction groups. */
6453 /* Returns the priority to use for an eviction_group that contains 'n_rules'
6454 * rules. The priority contains low-order random bits to ensure that eviction
6455 * groups with the same number of rules are prioritized randomly. */
6457 eviction_group_priority(size_t n_rules)
6459 uint16_t size = MIN(UINT16_MAX, n_rules);
6460 return (size << 16) | random_uint16();
6463 /* Updates 'evg', an eviction_group within 'table', following a change that
6464 * adds or removes rules in 'evg'. */
6466 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
6467 OVS_REQUIRES(ofproto_mutex)
6469 heap_change(&table->eviction_groups_by_size, &evg->size_node,
6470 eviction_group_priority(heap_count(&evg->rules)));
6473 /* Destroys 'evg', an eviction_group within 'table':
6475 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
6476 * rules themselves, just removes them from the eviction group.)
6478 * - Removes 'evg' from 'table'.
6482 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
6483 OVS_REQUIRES(ofproto_mutex)
6485 while (!heap_is_empty(&evg->rules)) {
6488 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
6489 rule->eviction_group = NULL;
6491 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
6492 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
6493 heap_destroy(&evg->rules);
6497 /* Removes 'rule' from its eviction group, if any. */
6499 eviction_group_remove_rule(struct rule *rule)
6500 OVS_REQUIRES(ofproto_mutex)
6502 if (rule->eviction_group) {
6503 struct oftable *table = &rule->ofproto->tables[rule->table_id];
6504 struct eviction_group *evg = rule->eviction_group;
6506 rule->eviction_group = NULL;
6507 heap_remove(&evg->rules, &rule->evg_node);
6508 if (heap_is_empty(&evg->rules)) {
6509 eviction_group_destroy(table, evg);
6511 eviction_group_resized(table, evg);
6516 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
6517 * returns the hash value. */
6519 eviction_group_hash_rule(struct rule *rule)
6520 OVS_REQUIRES(ofproto_mutex)
6522 struct oftable *table = &rule->ofproto->tables[rule->table_id];
6523 const struct mf_subfield *sf;
6527 hash = table->eviction_group_id_basis;
6528 miniflow_expand(&rule->cr.match.flow, &flow);
6529 for (sf = table->eviction_fields;
6530 sf < &table->eviction_fields[table->n_eviction_fields];
6533 if (mf_are_prereqs_ok(sf->field, &flow)) {
6534 union mf_value value;
6536 mf_get_value(sf->field, &flow, &value);
6538 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
6540 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
6541 unsigned int start = sf->ofs + sf->n_bits;
6542 bitwise_zero(&value, sf->field->n_bytes, start,
6543 sf->field->n_bytes * 8 - start);
6545 hash = hash_bytes(&value, sf->field->n_bytes, hash);
6547 hash = hash_int(hash, 0);
6554 /* Returns an eviction group within 'table' with the given 'id', creating one
6556 static struct eviction_group *
6557 eviction_group_find(struct oftable *table, uint32_t id)
6558 OVS_REQUIRES(ofproto_mutex)
6560 struct eviction_group *evg;
6562 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
6566 evg = xmalloc(sizeof *evg);
6567 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
6568 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
6569 eviction_group_priority(0));
6570 heap_init(&evg->rules);
6575 /* Returns an eviction priority for 'rule'. The return value should be
6576 * interpreted so that higher priorities make a rule more attractive candidates
6579 rule_eviction_priority(struct rule *rule)
6580 OVS_REQUIRES(ofproto_mutex)
6582 long long int hard_expiration;
6583 long long int idle_expiration;
6584 long long int expiration;
6585 uint32_t expiration_offset;
6587 /* Calculate time of expiration. */
6588 ovs_mutex_lock(&rule->mutex);
6589 hard_expiration = (rule->hard_timeout
6590 ? rule->modified + rule->hard_timeout * 1000
6592 idle_expiration = (rule->idle_timeout
6593 ? rule->used + rule->idle_timeout * 1000
6595 expiration = MIN(hard_expiration, idle_expiration);
6596 ovs_mutex_unlock(&rule->mutex);
6597 if (expiration == LLONG_MAX) {
6601 /* Calculate the time of expiration as a number of (approximate) seconds
6602 * after program startup.
6604 * This should work OK for program runs that last UINT32_MAX seconds or
6605 * less. Therefore, please restart OVS at least once every 136 years. */
6606 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
6608 /* Invert the expiration offset because we're using a max-heap. */
6609 return UINT32_MAX - expiration_offset;
6612 /* Adds 'rule' to an appropriate eviction group for its oftable's
6613 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
6614 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
6617 * The caller must ensure that 'rule' is not already in an eviction group. */
6619 eviction_group_add_rule(struct rule *rule)
6620 OVS_REQUIRES(ofproto_mutex)
6622 struct ofproto *ofproto = rule->ofproto;
6623 struct oftable *table = &ofproto->tables[rule->table_id];
6626 ovs_mutex_lock(&rule->mutex);
6627 has_timeout = rule->hard_timeout || rule->idle_timeout;
6628 ovs_mutex_unlock(&rule->mutex);
6630 if (table->eviction_fields && has_timeout) {
6631 struct eviction_group *evg;
6633 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
6635 rule->eviction_group = evg;
6636 heap_insert(&evg->rules, &rule->evg_node,
6637 rule_eviction_priority(rule));
6638 eviction_group_resized(table, evg);
6644 /* Initializes 'table'. */
6646 oftable_init(struct oftable *table)
6648 memset(table, 0, sizeof *table);
6649 classifier_init(&table->cls, flow_segment_u32s);
6650 table->max_flows = UINT_MAX;
6653 /* Destroys 'table', including its classifier and eviction groups.
6655 * The caller is responsible for freeing 'table' itself. */
6657 oftable_destroy(struct oftable *table)
6659 ovs_rwlock_rdlock(&table->cls.rwlock);
6660 ovs_assert(classifier_is_empty(&table->cls));
6661 ovs_rwlock_unlock(&table->cls.rwlock);
6662 oftable_disable_eviction(table);
6663 classifier_destroy(&table->cls);
6667 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
6668 * string, then 'table' will use its default name.
6670 * This only affects the name exposed for a table exposed through the OpenFlow
6671 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
6673 oftable_set_name(struct oftable *table, const char *name)
6675 if (name && name[0]) {
6676 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
6677 if (!table->name || strncmp(name, table->name, len)) {
6679 table->name = xmemdup0(name, len);
6687 /* oftables support a choice of two policies when adding a rule would cause the
6688 * number of flows in the table to exceed the configured maximum number: either
6689 * they can refuse to add the new flow or they can evict some existing flow.
6690 * This function configures the former policy on 'table'. */
6692 oftable_disable_eviction(struct oftable *table)
6693 OVS_REQUIRES(ofproto_mutex)
6695 if (table->eviction_fields) {
6696 struct eviction_group *evg, *next;
6698 HMAP_FOR_EACH_SAFE (evg, next, id_node,
6699 &table->eviction_groups_by_id) {
6700 eviction_group_destroy(table, evg);
6702 hmap_destroy(&table->eviction_groups_by_id);
6703 heap_destroy(&table->eviction_groups_by_size);
6705 free(table->eviction_fields);
6706 table->eviction_fields = NULL;
6707 table->n_eviction_fields = 0;
6711 /* oftables support a choice of two policies when adding a rule would cause the
6712 * number of flows in the table to exceed the configured maximum number: either
6713 * they can refuse to add the new flow or they can evict some existing flow.
6714 * This function configures the latter policy on 'table', with fairness based
6715 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
6716 * 'n_fields' as 0 disables fairness.) */
6718 oftable_enable_eviction(struct oftable *table,
6719 const struct mf_subfield *fields, size_t n_fields)
6720 OVS_REQUIRES(ofproto_mutex)
6722 struct cls_cursor cursor;
6725 if (table->eviction_fields
6726 && n_fields == table->n_eviction_fields
6728 || !memcmp(fields, table->eviction_fields,
6729 n_fields * sizeof *fields))) {
6734 oftable_disable_eviction(table);
6736 table->n_eviction_fields = n_fields;
6737 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
6739 table->eviction_group_id_basis = random_uint32();
6740 hmap_init(&table->eviction_groups_by_id);
6741 heap_init(&table->eviction_groups_by_size);
6743 ovs_rwlock_rdlock(&table->cls.rwlock);
6744 cls_cursor_init(&cursor, &table->cls, NULL);
6745 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
6746 eviction_group_add_rule(rule);
6748 ovs_rwlock_unlock(&table->cls.rwlock);
6751 /* Removes 'rule' from the oftable that contains it. */
6753 oftable_remove_rule__(struct ofproto *ofproto, struct rule *rule)
6754 OVS_REQUIRES(ofproto_mutex)
6756 struct classifier *cls = &ofproto->tables[rule->table_id].cls;
6758 ovs_rwlock_wrlock(&cls->rwlock);
6759 classifier_remove(cls, CONST_CAST(struct cls_rule *, &rule->cr));
6760 ovs_rwlock_unlock(&cls->rwlock);
6762 cookies_remove(ofproto, rule);
6764 eviction_group_remove_rule(rule);
6765 if (!list_is_empty(&rule->expirable)) {
6766 list_remove(&rule->expirable);
6768 if (!list_is_empty(&rule->meter_list_node)) {
6769 list_remove(&rule->meter_list_node);
6770 list_init(&rule->meter_list_node);
6775 oftable_remove_rule(struct rule *rule)
6776 OVS_REQUIRES(ofproto_mutex)
6778 oftable_remove_rule__(rule->ofproto, rule);
6781 /* Inserts 'rule' into its oftable, which must not already contain any rule for
6782 * the same cls_rule. */
6784 oftable_insert_rule(struct rule *rule)
6785 OVS_REQUIRES(ofproto_mutex)
6787 struct ofproto *ofproto = rule->ofproto;
6788 struct oftable *table = &ofproto->tables[rule->table_id];
6791 ovs_mutex_lock(&rule->mutex);
6792 may_expire = rule->hard_timeout || rule->idle_timeout;
6793 ovs_mutex_unlock(&rule->mutex);
6796 list_insert(&ofproto->expirable, &rule->expirable);
6799 cookies_insert(ofproto, rule);
6801 if (rule->actions->provider_meter_id != UINT32_MAX) {
6802 uint32_t meter_id = ofpacts_get_meter(rule->actions->ofpacts,
6803 rule->actions->ofpacts_len);
6804 struct meter *meter = ofproto->meters[meter_id];
6805 list_insert(&meter->rules, &rule->meter_list_node);
6807 ovs_rwlock_wrlock(&table->cls.rwlock);
6808 classifier_insert(&table->cls, CONST_CAST(struct cls_rule *, &rule->cr));
6809 ovs_rwlock_unlock(&table->cls.rwlock);
6810 eviction_group_add_rule(rule);
6813 /* unixctl commands. */
6816 ofproto_lookup(const char *name)
6818 struct ofproto *ofproto;
6820 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
6822 if (!strcmp(ofproto->name, name)) {
6830 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
6831 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6833 struct ofproto *ofproto;
6837 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
6838 ds_put_format(&results, "%s\n", ofproto->name);
6840 unixctl_command_reply(conn, ds_cstr(&results));
6841 ds_destroy(&results);
6845 ofproto_unixctl_init(void)
6847 static bool registered;
6853 unixctl_command_register("ofproto/list", "", 0, 0,
6854 ofproto_unixctl_list, NULL);
6857 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6859 * This is deprecated. It is only for compatibility with broken device drivers
6860 * in old versions of Linux that do not properly support VLANs when VLAN
6861 * devices are not used. When broken device drivers are no longer in
6862 * widespread use, we will delete these interfaces. */
6864 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
6865 * (exactly) by an OpenFlow rule in 'ofproto'. */
6867 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
6869 const struct oftable *oftable;
6871 free(ofproto->vlan_bitmap);
6872 ofproto->vlan_bitmap = bitmap_allocate(4096);
6873 ofproto->vlans_changed = false;
6875 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
6876 const struct cls_subtable *table;
6878 ovs_rwlock_rdlock(&oftable->cls.rwlock);
6879 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.subtables) {
6880 if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
6881 const struct cls_rule *rule;
6883 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
6884 uint16_t vid = miniflow_get_vid(&rule->match.flow);
6885 bitmap_set1(vlan_bitmap, vid);
6886 bitmap_set1(ofproto->vlan_bitmap, vid);
6890 ovs_rwlock_unlock(&oftable->cls.rwlock);
6894 /* Returns true if new VLANs have come into use by the flow table since the
6895 * last call to ofproto_get_vlan_usage().
6897 * We don't track when old VLANs stop being used. */
6899 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
6901 return ofproto->vlans_changed;
6904 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
6905 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
6906 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
6907 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
6908 * then the VLAN device is un-enslaved. */
6910 ofproto_port_set_realdev(struct ofproto *ofproto, ofp_port_t vlandev_ofp_port,
6911 ofp_port_t realdev_ofp_port, int vid)
6913 struct ofport *ofport;
6916 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
6918 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
6920 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
6921 ofproto->name, vlandev_ofp_port);
6925 if (!ofproto->ofproto_class->set_realdev) {
6926 if (!vlandev_ofp_port) {
6929 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
6933 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
6935 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
6936 ofproto->name, vlandev_ofp_port,
6937 netdev_get_name(ofport->netdev), ovs_strerror(error));