2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
26 #include "byte-order.h"
27 #include "classifier.h"
30 #include "dynamic-string.h"
33 #include "meta-flow.h"
36 #include "ofp-actions.h"
37 #include "ofp-errors.h"
39 #include "ofp-print.h"
42 #include "ofproto-provider.h"
43 #include "openflow/nicira-ext.h"
44 #include "openflow/openflow.h"
48 #include "poll-loop.h"
55 #include "unaligned.h"
59 VLOG_DEFINE_THIS_MODULE(ofproto);
61 COVERAGE_DEFINE(ofproto_flush);
62 COVERAGE_DEFINE(ofproto_packet_out);
63 COVERAGE_DEFINE(ofproto_queue_req);
64 COVERAGE_DEFINE(ofproto_recv_openflow);
65 COVERAGE_DEFINE(ofproto_reinit_ports);
66 COVERAGE_DEFINE(ofproto_update_port);
69 S_OPENFLOW, /* Processing OpenFlow commands. */
70 S_EVICT, /* Evicting flows from over-limit tables. */
71 S_FLUSH, /* Deleting all flow table rules. */
74 enum ofoperation_type {
81 /* A single OpenFlow request can execute any number of operations. The
82 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
83 * ofconn to which an error reply should be sent if necessary.
85 * ofproto initiates some operations internally. These operations are still
86 * assigned to groups but will not have an associated ofconn. */
88 struct ofproto *ofproto; /* Owning ofproto. */
89 struct list ofproto_node; /* In ofproto's "pending" list. */
90 struct list ops; /* List of "struct ofoperation"s. */
91 int n_running; /* Number of ops still pending. */
93 /* Data needed to send OpenFlow reply on failure or to send a buffered
96 * If list_is_empty(ofconn_node) then this ofopgroup never had an
97 * associated ofconn or its ofconn's connection dropped after it initiated
98 * the operation. In the latter case 'ofconn' is a wild pointer that
99 * refers to freed memory, so the 'ofconn' member must be used only if
100 * !list_is_empty(ofconn_node).
102 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
103 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
104 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
105 uint32_t buffer_id; /* Buffer id from original request. */
108 static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
109 static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
110 const struct ofp_header *,
112 static void ofopgroup_submit(struct ofopgroup *);
113 static void ofopgroup_complete(struct ofopgroup *);
115 /* A single flow table operation. */
117 struct ofopgroup *group; /* Owning group. */
118 struct list group_node; /* In ofopgroup's "ops" list. */
119 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
120 struct rule *rule; /* Rule being operated upon. */
121 enum ofoperation_type type; /* Type of operation. */
123 /* OFOPERATION_MODIFY, OFOPERATION_REPLACE: The old actions, if the actions
125 struct rule_actions *actions;
127 /* OFOPERATION_DELETE. */
128 enum ofp_flow_removed_reason reason; /* Reason flow was removed. */
130 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
131 uint16_t idle_timeout; /* Rule's old idle timeout. */
132 uint16_t hard_timeout; /* Rule's old hard timeout. */
133 enum ofputil_flow_mod_flags flags; /* Rule's old flags. */
134 enum ofperr error; /* 0 if no error. */
137 static struct ofoperation *ofoperation_create(struct ofopgroup *,
139 enum ofoperation_type,
140 enum ofp_flow_removed_reason);
141 static void ofoperation_destroy(struct ofoperation *);
144 static void oftable_init(struct oftable *);
145 static void oftable_destroy(struct oftable *);
147 static void oftable_set_name(struct oftable *, const char *name);
149 static void oftable_disable_eviction(struct oftable *);
150 static void oftable_enable_eviction(struct oftable *,
151 const struct mf_subfield *fields,
154 static void oftable_remove_rule(struct rule *rule) OVS_REQUIRES(ofproto_mutex);
155 static void oftable_remove_rule__(struct ofproto *, struct rule *)
156 OVS_REQUIRES(ofproto_mutex);
157 static void oftable_insert_rule(struct rule *);
159 /* A set of rules within a single OpenFlow table (oftable) that have the same
160 * values for the oftable's eviction_fields. A rule to be evicted, when one is
161 * needed, is taken from the eviction group that contains the greatest number
164 * An oftable owns any number of eviction groups, each of which contains any
167 * Membership in an eviction group is imprecise, based on the hash of the
168 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
169 * That is, if two rules have different eviction_fields, but those
170 * eviction_fields hash to the same value, then they will belong to the same
171 * eviction_group anyway.
173 * (When eviction is not enabled on an oftable, we don't track any eviction
174 * groups, to save time and space.) */
175 struct eviction_group {
176 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
177 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
178 struct heap rules; /* Contains "struct rule"s. */
181 static bool choose_rule_to_evict(struct oftable *table, struct rule **rulep);
182 static void ofproto_evict(struct ofproto *) OVS_EXCLUDED(ofproto_mutex);
183 static uint32_t rule_eviction_priority(struct rule *);
184 static void eviction_group_add_rule(struct rule *);
185 static void eviction_group_remove_rule(struct rule *);
187 /* Criteria that flow_mod and other operations use for selecting rules on
188 * which to operate. */
189 struct rule_criteria {
190 /* An OpenFlow table or 255 for all tables. */
193 /* OpenFlow matching criteria. Interpreted different in "loose" way by
194 * collect_rules_loose() and "strict" way by collect_rules_strict(), as
195 * defined in the OpenFlow spec. */
198 /* Matching criteria for the OpenFlow cookie. Consider a bit B in a rule's
199 * cookie and the corresponding bits C in 'cookie' and M in 'cookie_mask'.
200 * The rule will not be selected if M is 1 and B != C. */
202 ovs_be64 cookie_mask;
204 /* Selection based on actions within a rule:
206 * If out_port != OFPP_ANY, selects only rules that output to out_port.
207 * If out_group != OFPG_ALL, select only rules that output to out_group. */
212 static void rule_criteria_init(struct rule_criteria *, uint8_t table_id,
213 const struct match *match,
214 unsigned int priority,
215 ovs_be64 cookie, ovs_be64 cookie_mask,
216 ofp_port_t out_port, uint32_t out_group);
217 static void rule_criteria_destroy(struct rule_criteria *);
219 /* A packet that needs to be passed to rule_execute().
221 * (We can't do this immediately from ofopgroup_complete() because that holds
222 * ofproto_mutex, which rule_execute() needs released.) */
223 struct rule_execute {
224 struct list list_node; /* In struct ofproto's "rule_executes" list. */
225 struct rule *rule; /* Owns a reference to the rule. */
227 struct ofpbuf *packet; /* Owns the packet. */
230 static void run_rule_executes(struct ofproto *) OVS_EXCLUDED(ofproto_mutex);
231 static void destroy_rule_executes(struct ofproto *);
234 static void ofport_destroy__(struct ofport *) OVS_EXCLUDED(ofproto_mutex);
235 static void ofport_destroy(struct ofport *);
237 static void update_port(struct ofproto *, const char *devname);
238 static int init_ports(struct ofproto *);
239 static void reinit_ports(struct ofproto *);
241 static long long int ofport_get_usage(const struct ofproto *,
242 ofp_port_t ofp_port);
243 static void ofport_set_usage(struct ofproto *, ofp_port_t ofp_port,
244 long long int last_used);
245 static void ofport_remove_usage(struct ofproto *, ofp_port_t ofp_port);
249 * Keeps track of the currently used and recently used ofport values and is
250 * used to prevent immediate recycling of ofport values. */
251 struct ofport_usage {
252 struct hmap_node hmap_node; /* In struct ofproto's "ofport_usage" hmap. */
253 ofp_port_t ofp_port; /* OpenFlow port number. */
254 long long int last_used; /* Last time the 'ofp_port' was used. LLONG_MAX
255 represents in-use ofports. */
259 static void ofproto_rule_destroy__(struct rule *);
260 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
261 static bool rule_is_modifiable(const struct rule *);
264 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
265 struct ofputil_flow_mod *,
266 const struct ofp_header *);
267 static enum ofperr modify_flows__(struct ofproto *, struct ofconn *,
268 struct ofputil_flow_mod *,
269 const struct ofp_header *,
270 const struct rule_collection *);
271 static void delete_flow__(struct rule *rule, struct ofopgroup *,
272 enum ofp_flow_removed_reason)
273 OVS_REQUIRES(ofproto_mutex);
274 static bool ofproto_group_exists__(const struct ofproto *ofproto,
276 OVS_REQ_RDLOCK(ofproto->groups_rwlock);
277 static bool ofproto_group_exists(const struct ofproto *ofproto,
279 OVS_EXCLUDED(ofproto->groups_rwlock);
280 static enum ofperr add_group(struct ofproto *, struct ofputil_group_mod *);
281 static bool handle_openflow(struct ofconn *, const struct ofpbuf *);
282 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
283 struct ofputil_flow_mod *,
284 const struct ofp_header *)
285 OVS_EXCLUDED(ofproto_mutex);
286 static void calc_duration(long long int start, long long int now,
287 uint32_t *sec, uint32_t *nsec);
290 static uint64_t pick_datapath_id(const struct ofproto *);
291 static uint64_t pick_fallback_dpid(void);
292 static void ofproto_destroy__(struct ofproto *);
293 static void update_mtu(struct ofproto *, struct ofport *);
294 static void meter_delete(struct ofproto *, uint32_t first, uint32_t last);
297 static void ofproto_unixctl_init(void);
299 /* All registered ofproto classes, in probe order. */
300 static const struct ofproto_class **ofproto_classes;
301 static size_t n_ofproto_classes;
302 static size_t allocated_ofproto_classes;
304 /* Global lock that protects all flow table operations. */
305 struct ovs_mutex ofproto_mutex = OVS_MUTEX_INITIALIZER;
307 unsigned flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT;
308 unsigned n_handler_threads;
309 enum ofproto_flow_miss_model flow_miss_model = OFPROTO_HANDLE_MISS_AUTO;
311 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
312 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
314 /* Initial mappings of port to OpenFlow number mappings. */
315 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
317 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
319 /* The default value of true waits for flow restore. */
320 static bool flow_restore_wait = true;
322 /* Must be called to initialize the ofproto library.
324 * The caller may pass in 'iface_hints', which contains an shash of
325 * "iface_hint" elements indexed by the interface's name. The provider
326 * may use these hints to describe the startup configuration in order to
327 * reinitialize its state. The caller owns the provided data, so a
328 * provider will make copies of anything required. An ofproto provider
329 * will remove any existing state that is not described by the hint, and
330 * may choose to remove it all. */
332 ofproto_init(const struct shash *iface_hints)
334 struct shash_node *node;
337 ofproto_class_register(&ofproto_dpif_class);
339 /* Make a local copy, since we don't own 'iface_hints' elements. */
340 SHASH_FOR_EACH(node, iface_hints) {
341 const struct iface_hint *orig_hint = node->data;
342 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
343 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
345 new_hint->br_name = xstrdup(orig_hint->br_name);
346 new_hint->br_type = xstrdup(br_type);
347 new_hint->ofp_port = orig_hint->ofp_port;
349 shash_add(&init_ofp_ports, node->name, new_hint);
352 for (i = 0; i < n_ofproto_classes; i++) {
353 ofproto_classes[i]->init(&init_ofp_ports);
357 /* 'type' should be a normalized datapath type, as returned by
358 * ofproto_normalize_type(). Returns the corresponding ofproto_class
359 * structure, or a null pointer if there is none registered for 'type'. */
360 static const struct ofproto_class *
361 ofproto_class_find__(const char *type)
365 for (i = 0; i < n_ofproto_classes; i++) {
366 const struct ofproto_class *class = ofproto_classes[i];
371 class->enumerate_types(&types);
372 found = sset_contains(&types, type);
373 sset_destroy(&types);
379 VLOG_WARN("unknown datapath type %s", type);
383 /* Registers a new ofproto class. After successful registration, new ofprotos
384 * of that type can be created using ofproto_create(). */
386 ofproto_class_register(const struct ofproto_class *new_class)
390 for (i = 0; i < n_ofproto_classes; i++) {
391 if (ofproto_classes[i] == new_class) {
396 if (n_ofproto_classes >= allocated_ofproto_classes) {
397 ofproto_classes = x2nrealloc(ofproto_classes,
398 &allocated_ofproto_classes,
399 sizeof *ofproto_classes);
401 ofproto_classes[n_ofproto_classes++] = new_class;
405 /* Unregisters a datapath provider. 'type' must have been previously
406 * registered and not currently be in use by any ofprotos. After
407 * unregistration new datapaths of that type cannot be opened using
408 * ofproto_create(). */
410 ofproto_class_unregister(const struct ofproto_class *class)
414 for (i = 0; i < n_ofproto_classes; i++) {
415 if (ofproto_classes[i] == class) {
416 for (i++; i < n_ofproto_classes; i++) {
417 ofproto_classes[i - 1] = ofproto_classes[i];
423 VLOG_WARN("attempted to unregister an ofproto class that is not "
428 /* Clears 'types' and enumerates all registered ofproto types into it. The
429 * caller must first initialize the sset. */
431 ofproto_enumerate_types(struct sset *types)
435 for (i = 0; i < n_ofproto_classes; i++) {
436 ofproto_classes[i]->enumerate_types(types);
440 /* Returns the fully spelled out name for the given ofproto 'type'.
442 * Normalized type string can be compared with strcmp(). Unnormalized type
443 * string might be the same even if they have different spellings. */
445 ofproto_normalize_type(const char *type)
447 return type && type[0] ? type : "system";
450 /* Clears 'names' and enumerates the names of all known created ofprotos with
451 * the given 'type'. The caller must first initialize the sset. Returns 0 if
452 * successful, otherwise a positive errno value.
454 * Some kinds of datapaths might not be practically enumerable. This is not
455 * considered an error. */
457 ofproto_enumerate_names(const char *type, struct sset *names)
459 const struct ofproto_class *class = ofproto_class_find__(type);
460 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
464 ofproto_create(const char *datapath_name, const char *datapath_type,
465 struct ofproto **ofprotop)
467 const struct ofproto_class *class;
468 struct ofproto *ofproto;
474 ofproto_unixctl_init();
476 datapath_type = ofproto_normalize_type(datapath_type);
477 class = ofproto_class_find__(datapath_type);
479 VLOG_WARN("could not create datapath %s of unknown type %s",
480 datapath_name, datapath_type);
484 ofproto = class->alloc();
486 VLOG_ERR("failed to allocate datapath %s of type %s",
487 datapath_name, datapath_type);
492 ovs_mutex_lock(&ofproto_mutex);
493 memset(ofproto, 0, sizeof *ofproto);
494 ofproto->ofproto_class = class;
495 ofproto->name = xstrdup(datapath_name);
496 ofproto->type = xstrdup(datapath_type);
497 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
498 hash_string(ofproto->name, 0));
499 ofproto->datapath_id = 0;
500 ofproto->forward_bpdu = false;
501 ofproto->fallback_dpid = pick_fallback_dpid();
502 ofproto->mfr_desc = NULL;
503 ofproto->hw_desc = NULL;
504 ofproto->sw_desc = NULL;
505 ofproto->serial_desc = NULL;
506 ofproto->dp_desc = NULL;
507 ofproto->frag_handling = OFPC_FRAG_NORMAL;
508 hmap_init(&ofproto->ports);
509 hmap_init(&ofproto->ofport_usage);
510 shash_init(&ofproto->port_by_name);
511 simap_init(&ofproto->ofp_requests);
512 ofproto->max_ports = ofp_to_u16(OFPP_MAX);
513 ofproto->eviction_group_timer = LLONG_MIN;
514 ofproto->tables = NULL;
515 ofproto->n_tables = 0;
516 hindex_init(&ofproto->cookies);
517 list_init(&ofproto->expirable);
518 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
519 ofproto->state = S_OPENFLOW;
520 list_init(&ofproto->pending);
521 ofproto->n_pending = 0;
522 hmap_init(&ofproto->deletions);
523 guarded_list_init(&ofproto->rule_executes);
524 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
525 ofproto->first_op = ofproto->last_op = LLONG_MIN;
526 ofproto->next_op_report = LLONG_MAX;
527 ofproto->op_backoff = LLONG_MIN;
528 ofproto->vlan_bitmap = NULL;
529 ofproto->vlans_changed = false;
530 ofproto->min_mtu = INT_MAX;
531 ovs_rwlock_init(&ofproto->groups_rwlock);
532 hmap_init(&ofproto->groups);
533 ovs_mutex_unlock(&ofproto_mutex);
534 ofproto->ogf.capabilities = OFPGFC_CHAINING | OFPGFC_SELECT_LIVENESS |
535 OFPGFC_SELECT_WEIGHT;
536 ofproto->ogf.max_groups[OFPGT11_ALL] = OFPG_MAX;
537 ofproto->ogf.max_groups[OFPGT11_SELECT] = OFPG_MAX;
538 ofproto->ogf.max_groups[OFPGT11_INDIRECT] = OFPG_MAX;
539 ofproto->ogf.max_groups[OFPGT11_FF] = OFPG_MAX;
540 ofproto->ogf.actions[0] =
541 (1 << OFPAT11_OUTPUT) |
542 (1 << OFPAT11_COPY_TTL_OUT) |
543 (1 << OFPAT11_COPY_TTL_IN) |
544 (1 << OFPAT11_SET_MPLS_TTL) |
545 (1 << OFPAT11_DEC_MPLS_TTL) |
546 (1 << OFPAT11_PUSH_VLAN) |
547 (1 << OFPAT11_POP_VLAN) |
548 (1 << OFPAT11_PUSH_MPLS) |
549 (1 << OFPAT11_POP_MPLS) |
550 (1 << OFPAT11_SET_QUEUE) |
551 (1 << OFPAT11_GROUP) |
552 (1 << OFPAT11_SET_NW_TTL) |
553 (1 << OFPAT11_DEC_NW_TTL) |
554 (1 << OFPAT12_SET_FIELD);
556 * (1 << OFPAT13_PUSH_PBB) |
557 * (1 << OFPAT13_POP_PBB) */
559 error = ofproto->ofproto_class->construct(ofproto);
561 VLOG_ERR("failed to open datapath %s: %s",
562 datapath_name, ovs_strerror(error));
563 ofproto_destroy__(ofproto);
567 /* Check that hidden tables, if any, are at the end. */
568 ovs_assert(ofproto->n_tables);
569 for (i = 0; i + 1 < ofproto->n_tables; i++) {
570 enum oftable_flags flags = ofproto->tables[i].flags;
571 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
573 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
576 ofproto->datapath_id = pick_datapath_id(ofproto);
579 /* Initialize meters table. */
580 if (ofproto->ofproto_class->meter_get_features) {
581 ofproto->ofproto_class->meter_get_features(ofproto,
582 &ofproto->meter_features);
584 memset(&ofproto->meter_features, 0, sizeof ofproto->meter_features);
586 ofproto->meters = xzalloc((ofproto->meter_features.max_meters + 1)
587 * sizeof(struct meter *));
593 /* Must be called (only) by an ofproto implementation in its constructor
594 * function. See the large comment on 'construct' in struct ofproto_class for
597 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
599 struct oftable *table;
601 ovs_assert(!ofproto->n_tables);
602 ovs_assert(n_tables >= 1 && n_tables <= 255);
604 ofproto->n_tables = n_tables;
605 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
606 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
611 /* To be optionally called (only) by an ofproto implementation in its
612 * constructor function. See the large comment on 'construct' in struct
613 * ofproto_class for details.
615 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
616 * will then ensure that actions passed into the ofproto implementation will
617 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
618 * function is not called, there will be no such restriction.
620 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
621 * the 'max_ports' restriction. */
623 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
625 ovs_assert(max_ports <= ofp_to_u16(OFPP_MAX));
626 ofproto->max_ports = max_ports;
630 ofproto_get_datapath_id(const struct ofproto *ofproto)
632 return ofproto->datapath_id;
636 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
638 uint64_t old_dpid = p->datapath_id;
639 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
640 if (p->datapath_id != old_dpid) {
641 /* Force all active connections to reconnect, since there is no way to
642 * notify a controller that the datapath ID has changed. */
643 ofproto_reconnect_controllers(p);
648 ofproto_set_controllers(struct ofproto *p,
649 const struct ofproto_controller *controllers,
650 size_t n_controllers, uint32_t allowed_versions)
652 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
657 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
659 connmgr_set_fail_mode(p->connmgr, fail_mode);
662 /* Drops the connections between 'ofproto' and all of its controllers, forcing
663 * them to reconnect. */
665 ofproto_reconnect_controllers(struct ofproto *ofproto)
667 connmgr_reconnect(ofproto->connmgr);
670 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
671 * in-band control should guarantee access, in the same way that in-band
672 * control guarantees access to OpenFlow controllers. */
674 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
675 const struct sockaddr_in *extras, size_t n)
677 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
680 /* Sets the OpenFlow queue used by flows set up by in-band control on
681 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
682 * flows will use the default queue. */
684 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
686 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
689 /* Sets the number of flows at which eviction from the kernel flow table
692 ofproto_set_flow_eviction_threshold(unsigned threshold)
694 flow_eviction_threshold = MAX(OFPROTO_FLOW_EVICTION_THRESHOLD_MIN,
698 /* Sets the path for handling flow misses. */
700 ofproto_set_flow_miss_model(unsigned model)
702 flow_miss_model = model;
705 /* If forward_bpdu is true, the NORMAL action will forward frames with
706 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
707 * the NORMAL action will drop these frames. */
709 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
711 bool old_val = ofproto->forward_bpdu;
712 ofproto->forward_bpdu = forward_bpdu;
713 if (old_val != ofproto->forward_bpdu) {
714 if (ofproto->ofproto_class->forward_bpdu_changed) {
715 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
720 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
721 * 'idle_time', in seconds, and the maximum number of MAC table entries to
724 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
727 if (ofproto->ofproto_class->set_mac_table_config) {
728 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
733 /* Sets number of upcall handler threads. The default is
734 * (number of online cores - 2). */
736 ofproto_set_n_handler_threads(unsigned limit)
739 n_handler_threads = limit;
741 int n_proc = sysconf(_SC_NPROCESSORS_ONLN);
742 n_handler_threads = n_proc > 2 ? n_proc - 2 : 1;
747 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
750 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
754 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
756 return connmgr_set_snoops(ofproto->connmgr, snoops);
760 ofproto_set_netflow(struct ofproto *ofproto,
761 const struct netflow_options *nf_options)
763 if (nf_options && sset_is_empty(&nf_options->collectors)) {
767 if (ofproto->ofproto_class->set_netflow) {
768 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
770 return nf_options ? EOPNOTSUPP : 0;
775 ofproto_set_sflow(struct ofproto *ofproto,
776 const struct ofproto_sflow_options *oso)
778 if (oso && sset_is_empty(&oso->targets)) {
782 if (ofproto->ofproto_class->set_sflow) {
783 return ofproto->ofproto_class->set_sflow(ofproto, oso);
785 return oso ? EOPNOTSUPP : 0;
790 ofproto_set_ipfix(struct ofproto *ofproto,
791 const struct ofproto_ipfix_bridge_exporter_options *bo,
792 const struct ofproto_ipfix_flow_exporter_options *fo,
795 if (ofproto->ofproto_class->set_ipfix) {
796 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
798 return (bo || fo) ? EOPNOTSUPP : 0;
803 ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
805 flow_restore_wait = flow_restore_wait_db;
809 ofproto_get_flow_restore_wait(void)
811 return flow_restore_wait;
815 /* Spanning Tree Protocol (STP) configuration. */
817 /* Configures STP on 'ofproto' using the settings defined in 's'. If
818 * 's' is NULL, disables STP.
820 * Returns 0 if successful, otherwise a positive errno value. */
822 ofproto_set_stp(struct ofproto *ofproto,
823 const struct ofproto_stp_settings *s)
825 return (ofproto->ofproto_class->set_stp
826 ? ofproto->ofproto_class->set_stp(ofproto, s)
830 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
831 * 'enabled' member of 's' is false, then the other members are not
834 * Returns 0 if successful, otherwise a positive errno value. */
836 ofproto_get_stp_status(struct ofproto *ofproto,
837 struct ofproto_stp_status *s)
839 return (ofproto->ofproto_class->get_stp_status
840 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
844 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
845 * in 's'. The caller is responsible for assigning STP port numbers
846 * (using the 'port_num' member in the range of 1 through 255, inclusive)
847 * and ensuring there are no duplicates. If the 's' is NULL, then STP
848 * is disabled on the port.
850 * Returns 0 if successful, otherwise a positive errno value.*/
852 ofproto_port_set_stp(struct ofproto *ofproto, ofp_port_t ofp_port,
853 const struct ofproto_port_stp_settings *s)
855 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
857 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
858 ofproto->name, ofp_port);
862 return (ofproto->ofproto_class->set_stp_port
863 ? ofproto->ofproto_class->set_stp_port(ofport, s)
867 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
868 * 's'. If the 'enabled' member in 's' is false, then the other members
869 * are not meaningful.
871 * Returns 0 if successful, otherwise a positive errno value.*/
873 ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
874 struct ofproto_port_stp_status *s)
876 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
878 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
879 "port %"PRIu16, ofproto->name, ofp_port);
883 return (ofproto->ofproto_class->get_stp_port_status
884 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
888 /* Queue DSCP configuration. */
890 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
891 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
892 * to implement QoS. Instead, it is used to implement features which require
893 * knowledge of what queues exist on a port, and some basic information about
896 * Returns 0 if successful, otherwise a positive errno value. */
898 ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port,
899 const struct ofproto_port_queue *queues,
902 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
905 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
906 ofproto->name, ofp_port);
910 return (ofproto->ofproto_class->set_queues
911 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
915 /* Connectivity Fault Management configuration. */
917 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
919 ofproto_port_clear_cfm(struct ofproto *ofproto, ofp_port_t ofp_port)
921 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
922 if (ofport && ofproto->ofproto_class->set_cfm) {
923 ofproto->ofproto_class->set_cfm(ofport, NULL);
927 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
928 * basic configuration from the configuration members in 'cfm', and the remote
929 * maintenance point ID from remote_mpid. Ignores the statistics members of
932 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
934 ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port,
935 const struct cfm_settings *s)
937 struct ofport *ofport;
940 ofport = ofproto_get_port(ofproto, ofp_port);
942 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
943 ofproto->name, ofp_port);
947 /* XXX: For configuration simplicity, we only support one remote_mpid
948 * outside of the CFM module. It's not clear if this is the correct long
949 * term solution or not. */
950 error = (ofproto->ofproto_class->set_cfm
951 ? ofproto->ofproto_class->set_cfm(ofport, s)
954 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
955 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
956 ovs_strerror(error));
960 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
961 * 'ofproto' does not have a port 'ofp_port'. */
963 ofproto_port_set_bfd(struct ofproto *ofproto, ofp_port_t ofp_port,
964 const struct smap *cfg)
966 struct ofport *ofport;
969 ofport = ofproto_get_port(ofproto, ofp_port);
971 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
972 ofproto->name, ofp_port);
976 error = (ofproto->ofproto_class->set_bfd
977 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
980 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
981 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
982 ovs_strerror(error));
986 /* Populates 'status' with key value pairs indicating the status of the BFD
987 * session on 'ofp_port'. This information is intended to be populated in the
988 * OVS database. Has no effect if 'ofp_port' is not na OpenFlow port in
991 ofproto_port_get_bfd_status(struct ofproto *ofproto, ofp_port_t ofp_port,
994 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
995 return (ofport && ofproto->ofproto_class->get_bfd_status
996 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
1000 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
1001 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
1002 * 0 if LACP partner information is not current (generally indicating a
1003 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
1005 ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port)
1007 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1008 return (ofport && ofproto->ofproto_class->port_is_lacp_current
1009 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
1015 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
1016 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
1017 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
1018 * configuration plus, if there is more than one slave, a bonding
1021 * If 'aux' is already registered then this function updates its configuration
1022 * to 's'. Otherwise, this function registers a new bundle.
1024 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
1027 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
1028 const struct ofproto_bundle_settings *s)
1030 return (ofproto->ofproto_class->bundle_set
1031 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
1035 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
1036 * If no such bundle has been registered, this has no effect. */
1038 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
1040 return ofproto_bundle_register(ofproto, aux, NULL);
1044 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
1045 * If 'aux' is already registered then this function updates its configuration
1046 * to 's'. Otherwise, this function registers a new mirror. */
1048 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
1049 const struct ofproto_mirror_settings *s)
1051 return (ofproto->ofproto_class->mirror_set
1052 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
1056 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
1057 * If no mirror has been registered, this has no effect. */
1059 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
1061 return ofproto_mirror_register(ofproto, aux, NULL);
1064 /* Retrieves statistics from mirror associated with client data pointer
1065 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
1066 * 'bytes', respectively. If a particular counters is not supported,
1067 * the appropriate argument is set to UINT64_MAX. */
1069 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
1070 uint64_t *packets, uint64_t *bytes)
1072 if (!ofproto->ofproto_class->mirror_get_stats) {
1073 *packets = *bytes = UINT64_MAX;
1077 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
1081 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
1082 * which all packets are flooded, instead of using MAC learning. If
1083 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1085 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
1088 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
1090 return (ofproto->ofproto_class->set_flood_vlans
1091 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
1095 /* Returns true if 'aux' is a registered bundle that is currently in use as the
1096 * output for a mirror. */
1098 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
1100 return (ofproto->ofproto_class->is_mirror_output_bundle
1101 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
1105 /* Configuration of OpenFlow tables. */
1107 /* Returns the number of OpenFlow tables in 'ofproto'. */
1109 ofproto_get_n_tables(const struct ofproto *ofproto)
1111 return ofproto->n_tables;
1114 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
1115 * settings from 's'. 'table_id' must be in the range 0 through the number of
1116 * OpenFlow tables in 'ofproto' minus 1, inclusive.
1118 * For read-only tables, only the name may be configured. */
1120 ofproto_configure_table(struct ofproto *ofproto, int table_id,
1121 const struct ofproto_table_settings *s)
1123 struct oftable *table;
1125 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
1126 table = &ofproto->tables[table_id];
1128 oftable_set_name(table, s->name);
1130 if (table->flags & OFTABLE_READONLY) {
1135 oftable_enable_eviction(table, s->groups, s->n_groups);
1137 oftable_disable_eviction(table);
1140 table->max_flows = s->max_flows;
1141 ovs_rwlock_rdlock(&table->cls.rwlock);
1142 if (classifier_count(&table->cls) > table->max_flows
1143 && table->eviction_fields) {
1144 /* 'table' contains more flows than allowed. We might not be able to
1145 * evict them right away because of the asynchronous nature of flow
1146 * table changes. Schedule eviction for later. */
1147 switch (ofproto->state) {
1149 ofproto->state = S_EVICT;
1153 /* We're already deleting flows, nothing more to do. */
1157 ovs_rwlock_unlock(&table->cls.rwlock);
1161 ofproto_has_snoops(const struct ofproto *ofproto)
1163 return connmgr_has_snoops(ofproto->connmgr);
1167 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1169 connmgr_get_snoops(ofproto->connmgr, snoops);
1173 ofproto_rule_delete__(struct ofproto *ofproto, struct rule *rule,
1175 OVS_REQUIRES(ofproto_mutex)
1177 struct ofopgroup *group;
1179 ovs_assert(!rule->pending);
1181 group = ofopgroup_create_unattached(ofproto);
1182 delete_flow__(rule, group, reason);
1183 ofopgroup_submit(group);
1186 /* Deletes 'rule' from 'cls' within 'ofproto'.
1188 * Within an ofproto implementation, this function allows an ofproto
1189 * implementation to destroy any rules that remain when its ->destruct()
1190 * function is called. This function is not suitable for use elsewhere in an
1191 * ofproto implementation.
1193 * This function implements steps 4.4 and 4.5 in the section titled "Rule Life
1194 * Cycle" in ofproto-provider.h. */
1196 ofproto_rule_delete(struct ofproto *ofproto, struct rule *rule)
1197 OVS_EXCLUDED(ofproto_mutex)
1199 struct ofopgroup *group;
1201 ovs_mutex_lock(&ofproto_mutex);
1202 ovs_assert(!rule->pending);
1204 group = ofopgroup_create_unattached(ofproto);
1205 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1206 oftable_remove_rule__(ofproto, rule);
1207 ofproto->ofproto_class->rule_delete(rule);
1208 ofopgroup_submit(group);
1210 ovs_mutex_unlock(&ofproto_mutex);
1214 ofproto_flush__(struct ofproto *ofproto)
1215 OVS_EXCLUDED(ofproto_mutex)
1217 struct oftable *table;
1219 if (ofproto->ofproto_class->flush) {
1220 ofproto->ofproto_class->flush(ofproto);
1223 ovs_mutex_lock(&ofproto_mutex);
1224 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1225 struct rule *rule, *next_rule;
1226 struct cls_cursor cursor;
1228 if (table->flags & OFTABLE_HIDDEN) {
1232 ovs_rwlock_rdlock(&table->cls.rwlock);
1233 cls_cursor_init(&cursor, &table->cls, NULL);
1234 ovs_rwlock_unlock(&table->cls.rwlock);
1235 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1236 if (!rule->pending) {
1237 ofproto_rule_delete__(ofproto, rule, OFPRR_DELETE);
1241 ovs_mutex_unlock(&ofproto_mutex);
1244 static void delete_group(struct ofproto *ofproto, uint32_t group_id);
1247 ofproto_destroy__(struct ofproto *ofproto)
1248 OVS_EXCLUDED(ofproto_mutex)
1250 struct oftable *table;
1252 ovs_assert(list_is_empty(&ofproto->pending));
1254 destroy_rule_executes(ofproto);
1255 guarded_list_destroy(&ofproto->rule_executes);
1257 delete_group(ofproto, OFPG_ALL);
1258 ovs_rwlock_destroy(&ofproto->groups_rwlock);
1259 hmap_destroy(&ofproto->groups);
1261 connmgr_destroy(ofproto->connmgr);
1263 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1264 free(ofproto->name);
1265 free(ofproto->type);
1266 free(ofproto->mfr_desc);
1267 free(ofproto->hw_desc);
1268 free(ofproto->sw_desc);
1269 free(ofproto->serial_desc);
1270 free(ofproto->dp_desc);
1271 hmap_destroy(&ofproto->ports);
1272 hmap_destroy(&ofproto->ofport_usage);
1273 shash_destroy(&ofproto->port_by_name);
1274 simap_destroy(&ofproto->ofp_requests);
1276 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1277 oftable_destroy(table);
1279 free(ofproto->tables);
1281 hmap_destroy(&ofproto->deletions);
1283 free(ofproto->vlan_bitmap);
1285 ofproto->ofproto_class->dealloc(ofproto);
1289 ofproto_destroy(struct ofproto *p)
1290 OVS_EXCLUDED(ofproto_mutex)
1292 struct ofport *ofport, *next_ofport;
1293 struct ofport_usage *usage, *next_usage;
1300 meter_delete(p, 1, p->meter_features.max_meters);
1301 p->meter_features.max_meters = 0;
1307 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1308 ofport_destroy(ofport);
1311 HMAP_FOR_EACH_SAFE (usage, next_usage, hmap_node, &p->ofport_usage) {
1312 hmap_remove(&p->ofport_usage, &usage->hmap_node);
1316 p->ofproto_class->destruct(p);
1317 ofproto_destroy__(p);
1320 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1321 * kernel datapath, for example, this destroys the datapath in the kernel, and
1322 * with the netdev-based datapath, it tears down the data structures that
1323 * represent the datapath.
1325 * The datapath should not be currently open as an ofproto. */
1327 ofproto_delete(const char *name, const char *type)
1329 const struct ofproto_class *class = ofproto_class_find__(type);
1330 return (!class ? EAFNOSUPPORT
1331 : !class->del ? EACCES
1332 : class->del(type, name));
1336 process_port_change(struct ofproto *ofproto, int error, char *devname)
1338 if (error == ENOBUFS) {
1339 reinit_ports(ofproto);
1340 } else if (!error) {
1341 update_port(ofproto, devname);
1347 ofproto_type_run(const char *datapath_type)
1349 const struct ofproto_class *class;
1352 datapath_type = ofproto_normalize_type(datapath_type);
1353 class = ofproto_class_find__(datapath_type);
1355 error = class->type_run ? class->type_run(datapath_type) : 0;
1356 if (error && error != EAGAIN) {
1357 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1358 datapath_type, ovs_strerror(error));
1364 ofproto_type_run_fast(const char *datapath_type)
1366 const struct ofproto_class *class;
1369 datapath_type = ofproto_normalize_type(datapath_type);
1370 class = ofproto_class_find__(datapath_type);
1372 error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
1373 if (error && error != EAGAIN) {
1374 VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
1375 datapath_type, ovs_strerror(error));
1381 ofproto_type_wait(const char *datapath_type)
1383 const struct ofproto_class *class;
1385 datapath_type = ofproto_normalize_type(datapath_type);
1386 class = ofproto_class_find__(datapath_type);
1388 if (class->type_wait) {
1389 class->type_wait(datapath_type);
1394 any_pending_ops(const struct ofproto *p)
1395 OVS_EXCLUDED(ofproto_mutex)
1399 ovs_mutex_lock(&ofproto_mutex);
1400 b = !list_is_empty(&p->pending);
1401 ovs_mutex_unlock(&ofproto_mutex);
1407 ofproto_run(struct ofproto *p)
1409 struct sset changed_netdevs;
1410 const char *changed_netdev;
1411 struct ofport *ofport;
1414 error = p->ofproto_class->run(p);
1415 if (error && error != EAGAIN) {
1416 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, ovs_strerror(error));
1419 run_rule_executes(p);
1421 /* Restore the eviction group heap invariant occasionally. */
1422 if (p->eviction_group_timer < time_msec()) {
1425 p->eviction_group_timer = time_msec() + 1000;
1427 for (i = 0; i < p->n_tables; i++) {
1428 struct oftable *table = &p->tables[i];
1429 struct eviction_group *evg;
1430 struct cls_cursor cursor;
1433 if (!table->eviction_fields) {
1437 ovs_mutex_lock(&ofproto_mutex);
1438 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
1439 heap_rebuild(&evg->rules);
1442 ovs_rwlock_rdlock(&table->cls.rwlock);
1443 cls_cursor_init(&cursor, &table->cls, NULL);
1444 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1445 if (!rule->eviction_group
1446 && (rule->idle_timeout || rule->hard_timeout)) {
1447 eviction_group_add_rule(rule);
1450 ovs_rwlock_unlock(&table->cls.rwlock);
1451 ovs_mutex_unlock(&ofproto_mutex);
1455 if (p->ofproto_class->port_poll) {
1458 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1459 process_port_change(p, error, devname);
1463 /* Update OpenFlow port status for any port whose netdev has changed.
1465 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1466 * destroyed, so it's not safe to update ports directly from the
1467 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1468 * need this two-phase approach. */
1469 sset_init(&changed_netdevs);
1470 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1471 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1472 if (ofport->change_seq != change_seq) {
1473 ofport->change_seq = change_seq;
1474 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1477 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1478 update_port(p, changed_netdev);
1480 sset_destroy(&changed_netdevs);
1484 connmgr_run(p->connmgr, handle_openflow);
1488 connmgr_run(p->connmgr, NULL);
1490 if (!any_pending_ops(p)) {
1491 p->state = S_OPENFLOW;
1496 connmgr_run(p->connmgr, NULL);
1498 if (!any_pending_ops(p)) {
1499 connmgr_flushed(p->connmgr);
1500 p->state = S_OPENFLOW;
1508 if (time_msec() >= p->next_op_report) {
1509 long long int ago = (time_msec() - p->first_op) / 1000;
1510 long long int interval = (p->last_op - p->first_op) / 1000;
1514 ds_put_format(&s, "%d flow_mods ",
1515 p->n_add + p->n_delete + p->n_modify);
1516 if (interval == ago) {
1517 ds_put_format(&s, "in the last %lld s", ago);
1518 } else if (interval) {
1519 ds_put_format(&s, "in the %lld s starting %lld s ago",
1522 ds_put_format(&s, "%lld s ago", ago);
1525 ds_put_cstr(&s, " (");
1527 ds_put_format(&s, "%d adds, ", p->n_add);
1530 ds_put_format(&s, "%d deletes, ", p->n_delete);
1533 ds_put_format(&s, "%d modifications, ", p->n_modify);
1536 ds_put_char(&s, ')');
1538 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1541 p->n_add = p->n_delete = p->n_modify = 0;
1542 p->next_op_report = LLONG_MAX;
1548 /* Performs periodic activity required by 'ofproto' that needs to be done
1549 * with the least possible latency.
1551 * It makes sense to call this function a couple of times per poll loop, to
1552 * provide a significant performance boost on some benchmarks with the
1553 * ofproto-dpif implementation. */
1555 ofproto_run_fast(struct ofproto *p)
1559 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1560 if (error && error != EAGAIN) {
1561 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1562 p->name, ovs_strerror(error));
1568 ofproto_wait(struct ofproto *p)
1570 struct ofport *ofport;
1572 p->ofproto_class->wait(p);
1573 if (p->ofproto_class->port_poll_wait) {
1574 p->ofproto_class->port_poll_wait(p);
1577 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1578 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1579 poll_immediate_wake();
1585 connmgr_wait(p->connmgr, true);
1590 connmgr_wait(p->connmgr, false);
1591 if (!any_pending_ops(p)) {
1592 poll_immediate_wake();
1599 ofproto_is_alive(const struct ofproto *p)
1601 return connmgr_has_controllers(p->connmgr);
1604 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1605 * memory_report(). */
1607 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1609 const struct oftable *table;
1610 unsigned int n_rules;
1612 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1614 ovs_mutex_lock(&ofproto_mutex);
1615 simap_increase(usage, "ops",
1616 ofproto->n_pending + hmap_count(&ofproto->deletions));
1617 ovs_mutex_unlock(&ofproto_mutex);
1620 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1621 ovs_rwlock_rdlock(&table->cls.rwlock);
1622 n_rules += classifier_count(&table->cls);
1623 ovs_rwlock_unlock(&table->cls.rwlock);
1625 simap_increase(usage, "rules", n_rules);
1627 if (ofproto->ofproto_class->get_memory_usage) {
1628 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1631 connmgr_get_memory_usage(ofproto->connmgr, usage);
1635 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1638 connmgr_get_controller_info(ofproto->connmgr, info);
1642 ofproto_free_ofproto_controller_info(struct shash *info)
1644 connmgr_free_controller_info(info);
1647 /* Makes a deep copy of 'old' into 'port'. */
1649 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1651 port->name = xstrdup(old->name);
1652 port->type = xstrdup(old->type);
1653 port->ofp_port = old->ofp_port;
1656 /* Frees memory allocated to members of 'ofproto_port'.
1658 * Do not call this function on an ofproto_port obtained from
1659 * ofproto_port_dump_next(): that function retains ownership of the data in the
1662 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1664 free(ofproto_port->name);
1665 free(ofproto_port->type);
1668 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1670 * This function provides no status indication. An error status for the entire
1671 * dump operation is provided when it is completed by calling
1672 * ofproto_port_dump_done().
1675 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1676 const struct ofproto *ofproto)
1678 dump->ofproto = ofproto;
1679 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1683 /* Attempts to retrieve another port from 'dump', which must have been created
1684 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1685 * 'port' and returns true. On failure, returns false.
1687 * Failure might indicate an actual error or merely that the last port has been
1688 * dumped. An error status for the entire dump operation is provided when it
1689 * is completed by calling ofproto_port_dump_done().
1691 * The ofproto owns the data stored in 'port'. It will remain valid until at
1692 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1693 * ofproto_port_dump_done(). */
1695 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1696 struct ofproto_port *port)
1698 const struct ofproto *ofproto = dump->ofproto;
1704 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1707 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1713 /* Completes port table dump operation 'dump', which must have been created
1714 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1715 * error-free, otherwise a positive errno value describing the problem. */
1717 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1719 const struct ofproto *ofproto = dump->ofproto;
1721 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1724 return dump->error == EOF ? 0 : dump->error;
1727 /* Returns the type to pass to netdev_open() when a datapath of type
1728 * 'datapath_type' has a port of type 'port_type', for a few special
1729 * cases when a netdev type differs from a port type. For example, when
1730 * using the userspace datapath, a port of type "internal" needs to be
1733 * Returns either 'type' itself or a string literal, which must not be
1736 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1738 const struct ofproto_class *class;
1740 datapath_type = ofproto_normalize_type(datapath_type);
1741 class = ofproto_class_find__(datapath_type);
1746 return (class->port_open_type
1747 ? class->port_open_type(datapath_type, port_type)
1751 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1752 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1753 * the port's OpenFlow port number.
1755 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1756 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1757 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1758 * 'ofp_portp' is non-null). */
1760 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1761 ofp_port_t *ofp_portp)
1763 ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1766 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1768 const char *netdev_name = netdev_get_name(netdev);
1770 simap_put(&ofproto->ofp_requests, netdev_name,
1771 ofp_to_u16(ofp_port));
1772 update_port(ofproto, netdev_name);
1775 struct ofproto_port ofproto_port;
1777 ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
1779 *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
1780 ofproto_port_destroy(&ofproto_port);
1785 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1786 * initializes '*port' appropriately; on failure, returns a positive errno
1789 * The caller owns the data in 'ofproto_port' and must free it with
1790 * ofproto_port_destroy() when it is no longer needed. */
1792 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1793 struct ofproto_port *port)
1797 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1799 memset(port, 0, sizeof *port);
1804 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1805 * Returns 0 if successful, otherwise a positive errno. */
1807 ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port)
1809 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1810 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1811 struct simap_node *ofp_request_node;
1814 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1815 if (ofp_request_node) {
1816 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1819 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1820 if (!error && ofport) {
1821 /* 'name' is the netdev's name and update_port() is going to close the
1822 * netdev. Just in case update_port() refers to 'name' after it
1823 * destroys 'ofport', make a copy of it around the update_port()
1825 char *devname = xstrdup(name);
1826 update_port(ofproto, devname);
1833 flow_mod_init(struct ofputil_flow_mod *fm,
1834 const struct match *match, unsigned int priority,
1835 const struct ofpact *ofpacts, size_t ofpacts_len,
1836 enum ofp_flow_mod_command command)
1838 memset(fm, 0, sizeof *fm);
1840 fm->priority = priority;
1843 fm->modify_cookie = false;
1845 fm->command = command;
1846 fm->idle_timeout = 0;
1847 fm->hard_timeout = 0;
1848 fm->buffer_id = UINT32_MAX;
1849 fm->out_port = OFPP_ANY;
1850 fm->out_group = OFPG_ANY;
1852 fm->ofpacts = CONST_CAST(struct ofpact *, ofpacts);
1853 fm->ofpacts_len = ofpacts_len;
1857 simple_flow_mod(struct ofproto *ofproto,
1858 const struct match *match, unsigned int priority,
1859 const struct ofpact *ofpacts, size_t ofpacts_len,
1860 enum ofp_flow_mod_command command)
1862 struct ofputil_flow_mod fm;
1864 flow_mod_init(&fm, match, priority, ofpacts, ofpacts_len, command);
1866 return handle_flow_mod__(ofproto, NULL, &fm, NULL);
1869 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1870 * performs the 'n_actions' actions in 'actions'. The new flow will not
1873 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1874 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1875 * controllers; otherwise, it will be hidden.
1877 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1879 * This is a helper function for in-band control and fail-open. */
1881 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1882 unsigned int priority,
1883 const struct ofpact *ofpacts, size_t ofpacts_len)
1884 OVS_EXCLUDED(ofproto_mutex)
1886 const struct rule *rule;
1889 /* First do a cheap check whether the rule we're looking for already exists
1890 * with the actions that we want. If it does, then we're done. */
1891 ovs_rwlock_rdlock(&ofproto->tables[0].cls.rwlock);
1892 rule = rule_from_cls_rule(classifier_find_match_exactly(
1893 &ofproto->tables[0].cls, match, priority));
1895 ovs_mutex_lock(&rule->mutex);
1896 must_add = !ofpacts_equal(rule->actions->ofpacts,
1897 rule->actions->ofpacts_len,
1898 ofpacts, ofpacts_len);
1899 ovs_mutex_unlock(&rule->mutex);
1903 ovs_rwlock_unlock(&ofproto->tables[0].cls.rwlock);
1905 /* If there's no such rule or the rule doesn't have the actions we want,
1906 * fall back to a executing a full flow mod. We can't optimize this at
1907 * all because we didn't take enough locks above to ensure that the flow
1908 * table didn't already change beneath us. */
1910 simple_flow_mod(ofproto, match, priority, ofpacts, ofpacts_len,
1911 OFPFC_MODIFY_STRICT);
1915 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1916 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1917 * operation cannot be initiated now but may be retried later.
1919 * This is a helper function for in-band control and fail-open and the "learn"
1922 ofproto_flow_mod(struct ofproto *ofproto, struct ofputil_flow_mod *fm)
1923 OVS_EXCLUDED(ofproto_mutex)
1925 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1928 /* Searches for a rule with matching criteria exactly equal to 'target' in
1929 * ofproto's table 0 and, if it finds one, deletes it.
1931 * This is a helper function for in-band control and fail-open. */
1933 ofproto_delete_flow(struct ofproto *ofproto,
1934 const struct match *target, unsigned int priority)
1935 OVS_EXCLUDED(ofproto_mutex)
1937 struct classifier *cls = &ofproto->tables[0].cls;
1940 /* First do a cheap check whether the rule we're looking for has already
1941 * been deleted. If so, then we're done. */
1942 ovs_rwlock_rdlock(&cls->rwlock);
1943 rule = rule_from_cls_rule(classifier_find_match_exactly(cls, target,
1945 ovs_rwlock_unlock(&cls->rwlock);
1950 /* Fall back to a executing a full flow mod. We can't optimize this at all
1951 * because we didn't take enough locks above to ensure that the flow table
1952 * didn't already change beneath us. */
1953 return simple_flow_mod(ofproto, target, priority, NULL, 0,
1954 OFPFC_DELETE_STRICT) != OFPROTO_POSTPONE;
1957 /* Starts the process of deleting all of the flows from all of ofproto's flow
1958 * tables and then reintroducing the flows required by in-band control and
1959 * fail-open. The process will complete in a later call to ofproto_run(). */
1961 ofproto_flush_flows(struct ofproto *ofproto)
1963 COVERAGE_INC(ofproto_flush);
1964 ofproto->state = S_FLUSH;
1968 reinit_ports(struct ofproto *p)
1970 struct ofproto_port_dump dump;
1971 struct sset devnames;
1972 struct ofport *ofport;
1973 struct ofproto_port ofproto_port;
1974 const char *devname;
1976 COVERAGE_INC(ofproto_reinit_ports);
1978 sset_init(&devnames);
1979 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1980 sset_add(&devnames, netdev_get_name(ofport->netdev));
1982 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1983 sset_add(&devnames, ofproto_port.name);
1986 SSET_FOR_EACH (devname, &devnames) {
1987 update_port(p, devname);
1989 sset_destroy(&devnames);
1993 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
1997 port_idx = simap_get(&ofproto->ofp_requests, netdev_name);
1998 port_idx = port_idx ? port_idx : UINT16_MAX;
2000 if (port_idx >= ofproto->max_ports
2001 || ofport_get_usage(ofproto, u16_to_ofp(port_idx)) == LLONG_MAX) {
2002 uint16_t lru_ofport = 0, end_port_no = ofproto->alloc_port_no;
2003 long long int last_used_at, lru = LLONG_MAX;
2005 /* Search for a free OpenFlow port number. We try not to
2006 * immediately reuse them to prevent problems due to old
2009 if (++ofproto->alloc_port_no >= ofproto->max_ports) {
2010 ofproto->alloc_port_no = 1;
2012 last_used_at = ofport_get_usage(ofproto,
2013 u16_to_ofp(ofproto->alloc_port_no));
2014 if (!last_used_at) {
2015 port_idx = ofproto->alloc_port_no;
2017 } else if ( last_used_at < time_msec() - 60*60*1000) {
2018 /* If the port with ofport 'ofproto->alloc_port_no' was deleted
2019 * more than an hour ago, consider it usable. */
2020 ofport_remove_usage(ofproto,
2021 u16_to_ofp(ofproto->alloc_port_no));
2022 port_idx = ofproto->alloc_port_no;
2024 } else if (last_used_at < lru) {
2026 lru_ofport = ofproto->alloc_port_no;
2029 if (ofproto->alloc_port_no == end_port_no) {
2031 port_idx = lru_ofport;
2038 ofport_set_usage(ofproto, u16_to_ofp(port_idx), LLONG_MAX);
2039 return u16_to_ofp(port_idx);
2043 dealloc_ofp_port(struct ofproto *ofproto, ofp_port_t ofp_port)
2045 if (ofp_to_u16(ofp_port) < ofproto->max_ports) {
2046 ofport_set_usage(ofproto, ofp_port, time_msec());
2050 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
2051 * pointer if the netdev cannot be opened. On success, also fills in
2053 static struct netdev *
2054 ofport_open(struct ofproto *ofproto,
2055 struct ofproto_port *ofproto_port,
2056 struct ofputil_phy_port *pp)
2058 enum netdev_flags flags;
2059 struct netdev *netdev;
2062 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
2064 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
2065 "cannot be opened (%s)",
2067 ofproto_port->name, ofproto_port->ofp_port,
2068 ofproto_port->name, ovs_strerror(error));
2072 if (ofproto_port->ofp_port == OFPP_NONE) {
2073 if (!strcmp(ofproto->name, ofproto_port->name)) {
2074 ofproto_port->ofp_port = OFPP_LOCAL;
2076 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
2077 ofproto_port->name);
2080 pp->port_no = ofproto_port->ofp_port;
2081 netdev_get_etheraddr(netdev, pp->hw_addr);
2082 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
2083 netdev_get_flags(netdev, &flags);
2084 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
2085 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
2086 netdev_get_features(netdev, &pp->curr, &pp->advertised,
2087 &pp->supported, &pp->peer);
2088 pp->curr_speed = netdev_features_to_bps(pp->curr, 0) / 1000;
2089 pp->max_speed = netdev_features_to_bps(pp->supported, 0) / 1000;
2094 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
2095 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
2098 ofport_equal(const struct ofputil_phy_port *a,
2099 const struct ofputil_phy_port *b)
2101 return (eth_addr_equals(a->hw_addr, b->hw_addr)
2102 && a->state == b->state
2103 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
2104 && a->curr == b->curr
2105 && a->advertised == b->advertised
2106 && a->supported == b->supported
2107 && a->peer == b->peer
2108 && a->curr_speed == b->curr_speed
2109 && a->max_speed == b->max_speed);
2112 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
2113 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
2114 * one with the same name or port number). */
2116 ofport_install(struct ofproto *p,
2117 struct netdev *netdev, const struct ofputil_phy_port *pp)
2119 const char *netdev_name = netdev_get_name(netdev);
2120 struct ofport *ofport;
2123 /* Create ofport. */
2124 ofport = p->ofproto_class->port_alloc();
2129 ofport->ofproto = p;
2130 ofport->netdev = netdev;
2131 ofport->change_seq = netdev_change_seq(netdev);
2133 ofport->ofp_port = pp->port_no;
2134 ofport->created = time_msec();
2136 /* Add port to 'p'. */
2137 hmap_insert(&p->ports, &ofport->hmap_node,
2138 hash_ofp_port(ofport->ofp_port));
2139 shash_add(&p->port_by_name, netdev_name, ofport);
2141 update_mtu(p, ofport);
2143 /* Let the ofproto_class initialize its private data. */
2144 error = p->ofproto_class->port_construct(ofport);
2148 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
2152 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
2153 p->name, netdev_name, ovs_strerror(error));
2155 ofport_destroy__(ofport);
2157 netdev_close(netdev);
2161 /* Removes 'ofport' from 'p' and destroys it. */
2163 ofport_remove(struct ofport *ofport)
2165 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
2167 ofport_destroy(ofport);
2170 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
2173 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
2175 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
2177 ofport_remove(port);
2181 /* Updates 'port' with new 'pp' description.
2183 * Does not handle a name or port number change. The caller must implement
2184 * such a change as a delete followed by an add. */
2186 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
2188 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
2189 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
2190 | (pp->config & OFPUTIL_PC_PORT_DOWN));
2191 port->pp.state = pp->state;
2192 port->pp.curr = pp->curr;
2193 port->pp.advertised = pp->advertised;
2194 port->pp.supported = pp->supported;
2195 port->pp.peer = pp->peer;
2196 port->pp.curr_speed = pp->curr_speed;
2197 port->pp.max_speed = pp->max_speed;
2199 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
2202 /* Update OpenFlow 'state' in 'port' and notify controller. */
2204 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
2206 if (port->pp.state != state) {
2207 port->pp.state = state;
2208 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
2214 ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port)
2216 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
2218 if (port->ofproto->ofproto_class->set_realdev) {
2219 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
2221 if (port->ofproto->ofproto_class->set_stp_port) {
2222 port->ofproto->ofproto_class->set_stp_port(port, NULL);
2224 if (port->ofproto->ofproto_class->set_cfm) {
2225 port->ofproto->ofproto_class->set_cfm(port, NULL);
2227 if (port->ofproto->ofproto_class->bundle_remove) {
2228 port->ofproto->ofproto_class->bundle_remove(port);
2234 ofport_destroy__(struct ofport *port)
2236 struct ofproto *ofproto = port->ofproto;
2237 const char *name = netdev_get_name(port->netdev);
2239 hmap_remove(&ofproto->ports, &port->hmap_node);
2240 shash_delete(&ofproto->port_by_name,
2241 shash_find(&ofproto->port_by_name, name));
2243 netdev_close(port->netdev);
2244 ofproto->ofproto_class->port_dealloc(port);
2248 ofport_destroy(struct ofport *port)
2251 dealloc_ofp_port(port->ofproto, port->ofp_port);
2252 port->ofproto->ofproto_class->port_destruct(port);
2253 ofport_destroy__(port);
2258 ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
2260 struct ofport *port;
2262 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_ofp_port(ofp_port),
2264 if (port->ofp_port == ofp_port) {
2271 static long long int
2272 ofport_get_usage(const struct ofproto *ofproto, ofp_port_t ofp_port)
2274 struct ofport_usage *usage;
2276 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2277 &ofproto->ofport_usage) {
2278 if (usage->ofp_port == ofp_port) {
2279 return usage->last_used;
2286 ofport_set_usage(struct ofproto *ofproto, ofp_port_t ofp_port,
2287 long long int last_used)
2289 struct ofport_usage *usage;
2290 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2291 &ofproto->ofport_usage) {
2292 if (usage->ofp_port == ofp_port) {
2293 usage->last_used = last_used;
2297 ovs_assert(last_used == LLONG_MAX);
2299 usage = xmalloc(sizeof *usage);
2300 usage->ofp_port = ofp_port;
2301 usage->last_used = last_used;
2302 hmap_insert(&ofproto->ofport_usage, &usage->hmap_node,
2303 hash_ofp_port(ofp_port));
2307 ofport_remove_usage(struct ofproto *ofproto, ofp_port_t ofp_port)
2309 struct ofport_usage *usage;
2310 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2311 &ofproto->ofport_usage) {
2312 if (usage->ofp_port == ofp_port) {
2313 hmap_remove(&ofproto->ofport_usage, &usage->hmap_node);
2321 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
2323 struct ofproto *ofproto = port->ofproto;
2326 if (ofproto->ofproto_class->port_get_stats) {
2327 error = ofproto->ofproto_class->port_get_stats(port, stats);
2336 update_port(struct ofproto *ofproto, const char *name)
2338 struct ofproto_port ofproto_port;
2339 struct ofputil_phy_port pp;
2340 struct netdev *netdev;
2341 struct ofport *port;
2343 COVERAGE_INC(ofproto_update_port);
2345 /* Fetch 'name''s location and properties from the datapath. */
2346 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
2347 ? ofport_open(ofproto, &ofproto_port, &pp)
2351 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
2352 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
2353 struct netdev *old_netdev = port->netdev;
2355 /* 'name' hasn't changed location. Any properties changed? */
2356 if (!ofport_equal(&port->pp, &pp)) {
2357 ofport_modified(port, &pp);
2360 update_mtu(ofproto, port);
2362 /* Install the newly opened netdev in case it has changed.
2363 * Don't close the old netdev yet in case port_modified has to
2364 * remove a retained reference to it.*/
2365 port->netdev = netdev;
2366 port->change_seq = netdev_change_seq(netdev);
2368 if (port->ofproto->ofproto_class->port_modified) {
2369 port->ofproto->ofproto_class->port_modified(port);
2372 netdev_close(old_netdev);
2374 /* If 'port' is nonnull then its name differs from 'name' and thus
2375 * we should delete it. If we think there's a port named 'name'
2376 * then its port number must be wrong now so delete it too. */
2378 ofport_remove(port);
2380 ofport_remove_with_name(ofproto, name);
2381 ofport_install(ofproto, netdev, &pp);
2384 /* Any port named 'name' is gone now. */
2385 ofport_remove_with_name(ofproto, name);
2387 ofproto_port_destroy(&ofproto_port);
2391 init_ports(struct ofproto *p)
2393 struct ofproto_port_dump dump;
2394 struct ofproto_port ofproto_port;
2395 struct shash_node *node, *next;
2397 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2398 const char *name = ofproto_port.name;
2400 if (shash_find(&p->port_by_name, name)) {
2401 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2404 struct ofputil_phy_port pp;
2405 struct netdev *netdev;
2407 /* Check if an OpenFlow port number had been requested. */
2408 node = shash_find(&init_ofp_ports, name);
2410 const struct iface_hint *iface_hint = node->data;
2411 simap_put(&p->ofp_requests, name,
2412 ofp_to_u16(iface_hint->ofp_port));
2415 netdev = ofport_open(p, &ofproto_port, &pp);
2417 ofport_install(p, netdev, &pp);
2418 if (ofp_to_u16(ofproto_port.ofp_port) < p->max_ports) {
2419 p->alloc_port_no = MAX(p->alloc_port_no,
2420 ofp_to_u16(ofproto_port.ofp_port));
2426 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2427 struct iface_hint *iface_hint = node->data;
2429 if (!strcmp(iface_hint->br_name, p->name)) {
2430 free(iface_hint->br_name);
2431 free(iface_hint->br_type);
2433 shash_delete(&init_ofp_ports, node);
2440 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2441 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2443 find_min_mtu(struct ofproto *p)
2445 struct ofport *ofport;
2448 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2449 struct netdev *netdev = ofport->netdev;
2452 /* Skip any internal ports, since that's what we're trying to
2454 if (!strcmp(netdev_get_type(netdev), "internal")) {
2458 if (netdev_get_mtu(netdev, &dev_mtu)) {
2461 if (!mtu || dev_mtu < mtu) {
2466 return mtu ? mtu: ETH_PAYLOAD_MAX;
2469 /* Update MTU of all datapath devices on 'p' to the minimum of the
2470 * non-datapath ports in event of 'port' added or changed. */
2472 update_mtu(struct ofproto *p, struct ofport *port)
2474 struct ofport *ofport;
2475 struct netdev *netdev = port->netdev;
2476 int dev_mtu, old_min;
2478 if (netdev_get_mtu(netdev, &dev_mtu)) {
2482 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2483 if (dev_mtu > p->min_mtu) {
2484 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2485 dev_mtu = p->min_mtu;
2488 port->mtu = dev_mtu;
2492 /* For non-internal port find new min mtu. */
2493 old_min = p->min_mtu;
2494 port->mtu = dev_mtu;
2495 p->min_mtu = find_min_mtu(p);
2496 if (p->min_mtu == old_min) {
2500 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2501 struct netdev *netdev = ofport->netdev;
2503 if (!strcmp(netdev_get_type(netdev), "internal")) {
2504 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2505 ofport->mtu = p->min_mtu;
2512 ofproto_rule_ref(struct rule *rule)
2517 atomic_add(&rule->ref_count, 1, &orig);
2518 ovs_assert(orig != 0);
2523 ofproto_rule_unref(struct rule *rule)
2528 atomic_sub(&rule->ref_count, 1, &orig);
2530 rule->ofproto->ofproto_class->rule_destruct(rule);
2531 ofproto_rule_destroy__(rule);
2533 ovs_assert(orig != 0);
2538 struct rule_actions *
2539 rule_get_actions(const struct rule *rule)
2540 OVS_EXCLUDED(rule->mutex)
2542 struct rule_actions *actions;
2544 ovs_mutex_lock(&rule->mutex);
2545 actions = rule_get_actions__(rule);
2546 ovs_mutex_unlock(&rule->mutex);
2551 struct rule_actions *
2552 rule_get_actions__(const struct rule *rule)
2553 OVS_REQUIRES(rule->mutex)
2555 rule_actions_ref(rule->actions);
2556 return rule->actions;
2560 ofproto_rule_destroy__(struct rule *rule)
2561 OVS_NO_THREAD_SAFETY_ANALYSIS
2563 cls_rule_destroy(CONST_CAST(struct cls_rule *, &rule->cr));
2564 rule_actions_unref(rule->actions);
2565 ovs_mutex_destroy(&rule->mutex);
2566 rule->ofproto->ofproto_class->rule_dealloc(rule);
2569 static uint32_t get_provider_meter_id(const struct ofproto *,
2570 uint32_t of_meter_id);
2572 /* Creates and returns a new 'struct rule_actions', with a ref_count of 1,
2573 * whose actions are a copy of from the 'ofpacts_len' bytes of 'ofpacts'. */
2574 struct rule_actions *
2575 rule_actions_create(const struct ofproto *ofproto,
2576 const struct ofpact *ofpacts, size_t ofpacts_len)
2578 struct rule_actions *actions;
2580 actions = xmalloc(sizeof *actions);
2581 atomic_init(&actions->ref_count, 1);
2582 actions->ofpacts = xmemdup(ofpacts, ofpacts_len);
2583 actions->ofpacts_len = ofpacts_len;
2584 actions->provider_meter_id
2585 = get_provider_meter_id(ofproto,
2586 ofpacts_get_meter(ofpacts, ofpacts_len));
2591 /* Increments 'actions''s ref_count. */
2593 rule_actions_ref(struct rule_actions *actions)
2598 atomic_add(&actions->ref_count, 1, &orig);
2599 ovs_assert(orig != 0);
2603 /* Decrements 'actions''s ref_count and frees 'actions' if the ref_count
2606 rule_actions_unref(struct rule_actions *actions)
2611 atomic_sub(&actions->ref_count, 1, &orig);
2613 free(actions->ofpacts);
2616 ovs_assert(orig != 0);
2621 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2622 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2624 ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port)
2625 OVS_REQUIRES(ofproto_mutex)
2627 return (port == OFPP_ANY
2628 || ofpacts_output_to_port(rule->actions->ofpacts,
2629 rule->actions->ofpacts_len, port));
2632 /* Returns true if 'rule' has group and equals group_id. */
2634 ofproto_rule_has_out_group(const struct rule *rule, uint32_t group_id)
2635 OVS_REQUIRES(ofproto_mutex)
2637 return (group_id == OFPG11_ANY
2638 || ofpacts_output_to_group(rule->actions->ofpacts,
2639 rule->actions->ofpacts_len, group_id));
2642 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2643 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2645 ofoperation_has_out_port(const struct ofoperation *op, ofp_port_t out_port)
2646 OVS_REQUIRES(ofproto_mutex)
2648 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2653 case OFOPERATION_ADD:
2654 case OFOPERATION_DELETE:
2657 case OFOPERATION_MODIFY:
2658 case OFOPERATION_REPLACE:
2659 return ofpacts_output_to_port(op->actions->ofpacts,
2660 op->actions->ofpacts_len, out_port);
2667 rule_execute_destroy(struct rule_execute *e)
2669 ofproto_rule_unref(e->rule);
2670 list_remove(&e->list_node);
2674 /* Executes all "rule_execute" operations queued up in ofproto->rule_executes,
2675 * by passing them to the ofproto provider. */
2677 run_rule_executes(struct ofproto *ofproto)
2678 OVS_EXCLUDED(ofproto_mutex)
2680 struct rule_execute *e, *next;
2681 struct list executes;
2683 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2684 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2685 union flow_in_port in_port_;
2688 in_port_.ofp_port = e->in_port;
2689 flow_extract(e->packet, 0, 0, NULL, &in_port_, &flow);
2690 ofproto->ofproto_class->rule_execute(e->rule, &flow, e->packet);
2692 rule_execute_destroy(e);
2696 /* Destroys and discards all "rule_execute" operations queued up in
2697 * ofproto->rule_executes. */
2699 destroy_rule_executes(struct ofproto *ofproto)
2701 struct rule_execute *e, *next;
2702 struct list executes;
2704 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2705 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2706 ofpbuf_delete(e->packet);
2707 rule_execute_destroy(e);
2711 /* Returns true if 'rule' should be hidden from the controller.
2713 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2714 * (e.g. by in-band control) and are intentionally hidden from the
2717 ofproto_rule_is_hidden(const struct rule *rule)
2719 return rule->cr.priority > UINT16_MAX;
2722 static enum oftable_flags
2723 rule_get_flags(const struct rule *rule)
2725 return rule->ofproto->tables[rule->table_id].flags;
2729 rule_is_modifiable(const struct rule *rule)
2731 return !(rule_get_flags(rule) & OFTABLE_READONLY);
2735 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2737 ofconn_send_reply(ofconn, make_echo_reply(oh));
2742 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2744 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2745 struct ofputil_switch_features features;
2746 struct ofport *port;
2752 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2754 ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2756 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2757 * if present, are always at the end.) */
2758 n_tables = ofproto->n_tables;
2759 for (i = 0; i < ofproto->n_tables; i++) {
2760 if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
2766 features.datapath_id = ofproto->datapath_id;
2767 features.n_buffers = pktbuf_capacity();
2768 features.n_tables = n_tables;
2769 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2770 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2772 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2774 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2775 features.auxiliary_id = 0;
2776 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2778 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2779 ofputil_put_switch_features_port(&port->pp, b);
2782 ofconn_send_reply(ofconn, b);
2787 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2789 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2790 struct ofp_switch_config *osc;
2791 enum ofp_config_flags flags;
2795 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2796 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2797 flags = ofproto->frag_handling;
2798 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2799 if (oh->version < OFP13_VERSION
2800 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2801 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2803 osc->flags = htons(flags);
2804 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2805 ofconn_send_reply(ofconn, buf);
2811 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2813 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2814 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2815 uint16_t flags = ntohs(osc->flags);
2817 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2818 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
2819 enum ofp_config_flags cur = ofproto->frag_handling;
2820 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2822 ovs_assert((cur & OFPC_FRAG_MASK) == cur);
2824 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2825 ofproto->frag_handling = next;
2827 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2829 ofputil_frag_handling_to_string(next));
2833 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2834 ofconn_set_invalid_ttl_to_controller(ofconn,
2835 (oh->version < OFP13_VERSION
2836 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2838 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2843 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2844 * error message code for the caller to propagate upward. Otherwise, returns
2847 * The log message mentions 'msg_type'. */
2849 reject_slave_controller(struct ofconn *ofconn)
2851 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2852 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
2853 return OFPERR_OFPBRC_EPERM;
2859 /* Checks that the 'ofpacts_len' bytes of action in 'ofpacts' are appropriate
2862 * - If they use a meter, then 'ofproto' has that meter configured.
2864 * - If they use any groups, then 'ofproto' has that group configured.
2866 * Returns 0 if successful, otherwise an OpenFlow error. */
2868 ofproto_check_ofpacts(struct ofproto *ofproto,
2869 const struct ofpact ofpacts[], size_t ofpacts_len)
2871 const struct ofpact *a;
2874 mid = ofpacts_get_meter(ofpacts, ofpacts_len);
2875 if (mid && get_provider_meter_id(ofproto, mid) == UINT32_MAX) {
2876 return OFPERR_OFPMMFC_INVALID_METER;
2879 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2880 if (a->type == OFPACT_GROUP
2881 && !ofproto_group_exists(ofproto, ofpact_get_GROUP(a)->group_id)) {
2882 return OFPERR_OFPBAC_BAD_OUT_GROUP;
2890 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2892 struct ofproto *p = ofconn_get_ofproto(ofconn);
2893 struct ofputil_packet_out po;
2894 struct ofpbuf *payload;
2895 uint64_t ofpacts_stub[1024 / 8];
2896 struct ofpbuf ofpacts;
2898 union flow_in_port in_port_;
2901 COVERAGE_INC(ofproto_packet_out);
2903 error = reject_slave_controller(ofconn);
2908 /* Decode message. */
2909 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2910 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2912 goto exit_free_ofpacts;
2914 if (ofp_to_u16(po.in_port) >= p->max_ports
2915 && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) {
2916 error = OFPERR_OFPBRC_BAD_PORT;
2917 goto exit_free_ofpacts;
2921 if (po.buffer_id != UINT32_MAX) {
2922 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2923 if (error || !payload) {
2924 goto exit_free_ofpacts;
2927 /* Ensure that the L3 header is 32-bit aligned. */
2928 payload = ofpbuf_clone_data_with_headroom(po.packet, po.packet_len, 2);
2931 /* Verify actions against packet, then send packet if successful. */
2932 in_port_.ofp_port = po.in_port;
2933 flow_extract(payload, 0, 0, NULL, &in_port_, &flow);
2934 error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len);
2936 error = p->ofproto_class->packet_out(p, payload, &flow,
2937 po.ofpacts, po.ofpacts_len);
2939 ofpbuf_delete(payload);
2942 ofpbuf_uninit(&ofpacts);
2948 update_port_config(struct ofport *port,
2949 enum ofputil_port_config config,
2950 enum ofputil_port_config mask)
2952 enum ofputil_port_config old_config = port->pp.config;
2953 enum ofputil_port_config toggle;
2955 toggle = (config ^ port->pp.config) & mask;
2956 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2957 if (config & OFPUTIL_PC_PORT_DOWN) {
2958 netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL);
2960 netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL);
2962 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2965 port->pp.config ^= toggle;
2966 if (port->pp.config != old_config) {
2967 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2972 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2974 struct ofproto *p = ofconn_get_ofproto(ofconn);
2975 struct ofputil_port_mod pm;
2976 struct ofport *port;
2979 error = reject_slave_controller(ofconn);
2984 error = ofputil_decode_port_mod(oh, &pm);
2989 port = ofproto_get_port(p, pm.port_no);
2991 return OFPERR_OFPPMFC_BAD_PORT;
2992 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2993 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2995 update_port_config(port, pm.config, pm.mask);
2997 netdev_set_advertisements(port->netdev, pm.advertise);
3004 handle_desc_stats_request(struct ofconn *ofconn,
3005 const struct ofp_header *request)
3007 static const char *default_mfr_desc = "Nicira, Inc.";
3008 static const char *default_hw_desc = "Open vSwitch";
3009 static const char *default_sw_desc = VERSION;
3010 static const char *default_serial_desc = "None";
3011 static const char *default_dp_desc = "None";
3013 struct ofproto *p = ofconn_get_ofproto(ofconn);
3014 struct ofp_desc_stats *ods;
3017 msg = ofpraw_alloc_stats_reply(request, 0);
3018 ods = ofpbuf_put_zeros(msg, sizeof *ods);
3019 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
3020 sizeof ods->mfr_desc);
3021 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
3022 sizeof ods->hw_desc);
3023 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
3024 sizeof ods->sw_desc);
3025 ovs_strlcpy(ods->serial_num,
3026 p->serial_desc ? p->serial_desc : default_serial_desc,
3027 sizeof ods->serial_num);
3028 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
3029 sizeof ods->dp_desc);
3030 ofconn_send_reply(ofconn, msg);
3036 handle_table_stats_request(struct ofconn *ofconn,
3037 const struct ofp_header *request)
3039 struct ofproto *p = ofconn_get_ofproto(ofconn);
3040 struct ofp12_table_stats *ots;
3045 /* Set up default values.
3047 * ofp12_table_stats is used as a generic structure as
3048 * it is able to hold all the fields for ofp10_table_stats
3049 * and ofp11_table_stats (and of course itself).
3051 ots = xcalloc(p->n_tables, sizeof *ots);
3052 for (i = 0; i < p->n_tables; i++) {
3053 ots[i].table_id = i;
3054 sprintf(ots[i].name, "table%zu", i);
3055 ots[i].match = htonll(OFPXMT13_MASK);
3056 ots[i].wildcards = htonll(OFPXMT13_MASK);
3057 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
3058 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
3059 ots[i].write_setfields = htonll(OFPXMT13_MASK);
3060 ots[i].apply_setfields = htonll(OFPXMT13_MASK);
3061 ots[i].metadata_match = OVS_BE64_MAX;
3062 ots[i].metadata_write = OVS_BE64_MAX;
3063 ots[i].instructions = htonl(OFPIT11_ALL);
3064 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
3065 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
3066 ovs_rwlock_rdlock(&p->tables[i].cls.rwlock);
3067 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
3068 ovs_rwlock_unlock(&p->tables[i].cls.rwlock);
3071 p->ofproto_class->get_tables(p, ots);
3073 /* Post-process the tables, dropping hidden tables. */
3074 n_tables = p->n_tables;
3075 for (i = 0; i < p->n_tables; i++) {
3076 const struct oftable *table = &p->tables[i];
3078 if (table->flags & OFTABLE_HIDDEN) {
3084 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
3087 if (table->max_flows < ntohl(ots[i].max_entries)) {
3088 ots[i].max_entries = htonl(table->max_flows);
3092 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
3093 ofconn_send_reply(ofconn, msg);
3101 append_port_stat(struct ofport *port, struct list *replies)
3103 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
3105 calc_duration(port->created, time_msec(),
3106 &ops.duration_sec, &ops.duration_nsec);
3108 /* Intentionally ignore return value, since errors will set
3109 * 'stats' to all-1s, which is correct for OpenFlow, and
3110 * netdev_get_stats() will log errors. */
3111 ofproto_port_get_stats(port, &ops.stats);
3113 ofputil_append_port_stat(replies, &ops);
3117 handle_port_stats_request(struct ofconn *ofconn,
3118 const struct ofp_header *request)
3120 struct ofproto *p = ofconn_get_ofproto(ofconn);
3121 struct ofport *port;
3122 struct list replies;
3126 error = ofputil_decode_port_stats_request(request, &port_no);
3131 ofpmp_init(&replies, request);
3132 if (port_no != OFPP_ANY) {
3133 port = ofproto_get_port(p, port_no);
3135 append_port_stat(port, &replies);
3138 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3139 append_port_stat(port, &replies);
3143 ofconn_send_replies(ofconn, &replies);
3148 handle_port_desc_stats_request(struct ofconn *ofconn,
3149 const struct ofp_header *request)
3151 struct ofproto *p = ofconn_get_ofproto(ofconn);
3152 enum ofp_version version;
3153 struct ofport *port;
3154 struct list replies;
3156 ofpmp_init(&replies, request);
3158 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
3159 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3160 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
3163 ofconn_send_replies(ofconn, &replies);
3168 hash_cookie(ovs_be64 cookie)
3170 return hash_2words((OVS_FORCE uint64_t)cookie >> 32,
3171 (OVS_FORCE uint64_t)cookie);
3175 cookies_insert(struct ofproto *ofproto, struct rule *rule)
3176 OVS_REQUIRES(ofproto_mutex)
3178 hindex_insert(&ofproto->cookies, &rule->cookie_node,
3179 hash_cookie(rule->flow_cookie));
3183 cookies_remove(struct ofproto *ofproto, struct rule *rule)
3184 OVS_REQUIRES(ofproto_mutex)
3186 hindex_remove(&ofproto->cookies, &rule->cookie_node);
3190 ofproto_rule_change_cookie(struct ofproto *ofproto, struct rule *rule,
3191 ovs_be64 new_cookie)
3192 OVS_REQUIRES(ofproto_mutex)
3194 if (new_cookie != rule->flow_cookie) {
3195 cookies_remove(ofproto, rule);
3197 ovs_mutex_lock(&rule->mutex);
3198 rule->flow_cookie = new_cookie;
3199 ovs_mutex_unlock(&rule->mutex);
3201 cookies_insert(ofproto, rule);
3206 calc_duration(long long int start, long long int now,
3207 uint32_t *sec, uint32_t *nsec)
3209 long long int msecs = now - start;
3210 *sec = msecs / 1000;
3211 *nsec = (msecs % 1000) * (1000 * 1000);
3214 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
3215 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
3217 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
3219 return (table_id == 0xff || table_id < ofproto->n_tables
3221 : OFPERR_OFPBRC_BAD_TABLE_ID);
3225 static struct oftable *
3226 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
3228 struct oftable *table;
3230 for (table = &ofproto->tables[table_id];
3231 table < &ofproto->tables[ofproto->n_tables];
3233 if (!(table->flags & OFTABLE_HIDDEN)) {
3241 static struct oftable *
3242 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
3244 if (table_id == 0xff) {
3245 return next_visible_table(ofproto, 0);
3246 } else if (table_id < ofproto->n_tables) {
3247 return &ofproto->tables[table_id];
3253 static struct oftable *
3254 next_matching_table(const struct ofproto *ofproto,
3255 const struct oftable *table, uint8_t table_id)
3257 return (table_id == 0xff
3258 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
3262 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
3264 * - If TABLE_ID is 0xff, this iterates over every classifier table in
3265 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
3267 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
3268 * only once, for that table. (This can be used to access tables marked
3271 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
3272 * entered at all. (Perhaps you should have validated TABLE_ID with
3273 * check_table_id().)
3275 * All parameters are evaluated multiple times.
3277 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
3278 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
3280 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
3282 /* Initializes 'criteria' in a straightforward way based on the other
3285 * For "loose" matching, the 'priority' parameter is unimportant and may be
3288 rule_criteria_init(struct rule_criteria *criteria, uint8_t table_id,
3289 const struct match *match, unsigned int priority,
3290 ovs_be64 cookie, ovs_be64 cookie_mask,
3291 ofp_port_t out_port, uint32_t out_group)
3293 criteria->table_id = table_id;
3294 cls_rule_init(&criteria->cr, match, priority);
3295 criteria->cookie = cookie;
3296 criteria->cookie_mask = cookie_mask;
3297 criteria->out_port = out_port;
3298 criteria->out_group = out_group;
3302 rule_criteria_destroy(struct rule_criteria *criteria)
3304 cls_rule_destroy(&criteria->cr);
3308 rule_collection_init(struct rule_collection *rules)
3310 rules->rules = rules->stub;
3312 rules->capacity = ARRAY_SIZE(rules->stub);
3316 rule_collection_add(struct rule_collection *rules, struct rule *rule)
3318 if (rules->n >= rules->capacity) {
3319 size_t old_size, new_size;
3321 old_size = rules->capacity * sizeof *rules->rules;
3322 rules->capacity *= 2;
3323 new_size = rules->capacity * sizeof *rules->rules;
3325 if (rules->rules == rules->stub) {
3326 rules->rules = xmalloc(new_size);
3327 memcpy(rules->rules, rules->stub, old_size);
3329 rules->rules = xrealloc(rules->rules, new_size);
3333 rules->rules[rules->n++] = rule;
3337 rule_collection_ref(struct rule_collection *rules)
3338 OVS_REQUIRES(ofproto_mutex)
3342 for (i = 0; i < rules->n; i++) {
3343 ofproto_rule_ref(rules->rules[i]);
3348 rule_collection_unref(struct rule_collection *rules)
3352 for (i = 0; i < rules->n; i++) {
3353 ofproto_rule_unref(rules->rules[i]);
3358 rule_collection_destroy(struct rule_collection *rules)
3360 if (rules->rules != rules->stub) {
3366 collect_rule(struct rule *rule, const struct rule_criteria *c,
3367 struct rule_collection *rules)
3368 OVS_REQUIRES(ofproto_mutex)
3370 /* We ordinarily want to skip hidden rules, but there has to be a way for
3371 * code internal to OVS to modify and delete them, so if the criteria
3372 * specify a priority that can only be for a hidden flow, then allow hidden
3373 * rules to be selected. (This doesn't allow OpenFlow clients to meddle
3374 * with hidden flows because OpenFlow uses only a 16-bit field to specify
3376 if (ofproto_rule_is_hidden(rule) && c->cr.priority <= UINT16_MAX) {
3378 } else if (rule->pending) {
3379 return OFPROTO_POSTPONE;
3381 if ((c->table_id == rule->table_id || c->table_id == 0xff)
3382 && ofproto_rule_has_out_port(rule, c->out_port)
3383 && ofproto_rule_has_out_group(rule, c->out_group)
3384 && !((rule->flow_cookie ^ c->cookie) & c->cookie_mask)) {
3385 rule_collection_add(rules, rule);
3391 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3392 * on classifiers rules are done in the "loose" way required for OpenFlow
3393 * OFPFC_MODIFY and OFPFC_DELETE requests. Puts the selected rules on list
3396 * Hidden rules are always omitted.
3398 * Returns 0 on success, otherwise an OpenFlow error code. */
3400 collect_rules_loose(struct ofproto *ofproto,
3401 const struct rule_criteria *criteria,
3402 struct rule_collection *rules)
3403 OVS_REQUIRES(ofproto_mutex)
3405 struct oftable *table;
3408 rule_collection_init(rules);
3410 error = check_table_id(ofproto, criteria->table_id);
3415 if (criteria->cookie_mask == OVS_BE64_MAX) {
3418 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
3419 hash_cookie(criteria->cookie),
3420 &ofproto->cookies) {
3421 if (cls_rule_is_loose_match(&rule->cr, &criteria->cr.match)) {
3422 error = collect_rule(rule, criteria, rules);
3429 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
3430 struct cls_cursor cursor;
3433 ovs_rwlock_rdlock(&table->cls.rwlock);
3434 cls_cursor_init(&cursor, &table->cls, &criteria->cr);
3435 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3436 error = collect_rule(rule, criteria, rules);
3441 ovs_rwlock_unlock(&table->cls.rwlock);
3447 rule_collection_destroy(rules);
3452 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3453 * on classifiers rules are done in the "strict" way required for OpenFlow
3454 * OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests. Puts the selected
3455 * rules on list 'rules'.
3457 * Hidden rules are always omitted.
3459 * Returns 0 on success, otherwise an OpenFlow error code. */
3461 collect_rules_strict(struct ofproto *ofproto,
3462 const struct rule_criteria *criteria,
3463 struct rule_collection *rules)
3464 OVS_REQUIRES(ofproto_mutex)
3466 struct oftable *table;
3469 rule_collection_init(rules);
3471 error = check_table_id(ofproto, criteria->table_id);
3476 if (criteria->cookie_mask == OVS_BE64_MAX) {
3479 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
3480 hash_cookie(criteria->cookie),
3481 &ofproto->cookies) {
3482 if (cls_rule_equal(&rule->cr, &criteria->cr)) {
3483 error = collect_rule(rule, criteria, rules);
3490 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
3493 ovs_rwlock_rdlock(&table->cls.rwlock);
3494 rule = rule_from_cls_rule(classifier_find_rule_exactly(
3495 &table->cls, &criteria->cr));
3496 ovs_rwlock_unlock(&table->cls.rwlock);
3498 error = collect_rule(rule, criteria, rules);
3508 rule_collection_destroy(rules);
3513 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
3514 * forced into the range of a uint16_t. */
3516 age_secs(long long int age_ms)
3518 return (age_ms < 0 ? 0
3519 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
3520 : (unsigned int) age_ms / 1000);
3524 handle_flow_stats_request(struct ofconn *ofconn,
3525 const struct ofp_header *request)
3526 OVS_EXCLUDED(ofproto_mutex)
3528 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3529 struct ofputil_flow_stats_request fsr;
3530 struct rule_criteria criteria;
3531 struct rule_collection rules;
3532 struct list replies;
3536 error = ofputil_decode_flow_stats_request(&fsr, request);
3541 rule_criteria_init(&criteria, fsr.table_id, &fsr.match, 0, fsr.cookie,
3542 fsr.cookie_mask, fsr.out_port, fsr.out_group);
3544 ovs_mutex_lock(&ofproto_mutex);
3545 error = collect_rules_loose(ofproto, &criteria, &rules);
3546 rule_criteria_destroy(&criteria);
3548 rule_collection_ref(&rules);
3550 ovs_mutex_unlock(&ofproto_mutex);
3556 ofpmp_init(&replies, request);
3557 for (i = 0; i < rules.n; i++) {
3558 struct rule *rule = rules.rules[i];
3559 long long int now = time_msec();
3560 struct ofputil_flow_stats fs;
3561 long long int created, used, modified;
3562 struct rule_actions *actions;
3563 enum ofputil_flow_mod_flags flags;
3565 ovs_mutex_lock(&rule->mutex);
3566 fs.cookie = rule->flow_cookie;
3567 fs.idle_timeout = rule->idle_timeout;
3568 fs.hard_timeout = rule->hard_timeout;
3569 created = rule->created;
3571 modified = rule->modified;
3572 actions = rule_get_actions__(rule);
3573 flags = rule->flags;
3574 ovs_mutex_unlock(&rule->mutex);
3576 minimatch_expand(&rule->cr.match, &fs.match);
3577 fs.table_id = rule->table_id;
3578 calc_duration(created, now, &fs.duration_sec, &fs.duration_nsec);
3579 fs.priority = rule->cr.priority;
3580 fs.idle_age = age_secs(now - used);
3581 fs.hard_age = age_secs(now - modified);
3582 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
3584 fs.ofpacts = actions->ofpacts;
3585 fs.ofpacts_len = actions->ofpacts_len;
3588 ofputil_append_flow_stats_reply(&fs, &replies);
3590 rule_actions_unref(actions);
3593 rule_collection_unref(&rules);
3594 rule_collection_destroy(&rules);
3596 ofconn_send_replies(ofconn, &replies);
3602 flow_stats_ds(struct rule *rule, struct ds *results)
3604 uint64_t packet_count, byte_count;
3605 struct rule_actions *actions;
3606 long long int created;
3608 rule->ofproto->ofproto_class->rule_get_stats(rule,
3609 &packet_count, &byte_count);
3611 ovs_mutex_lock(&rule->mutex);
3612 actions = rule_get_actions__(rule);
3613 created = rule->created;
3614 ovs_mutex_unlock(&rule->mutex);
3616 if (rule->table_id != 0) {
3617 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
3619 ds_put_format(results, "duration=%llds, ", (time_msec() - created) / 1000);
3620 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3621 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3622 cls_rule_format(&rule->cr, results);
3623 ds_put_char(results, ',');
3625 ds_put_cstr(results, "actions=");
3626 ofpacts_format(actions->ofpacts, actions->ofpacts_len, results);
3628 ds_put_cstr(results, "\n");
3630 rule_actions_unref(actions);
3633 /* Adds a pretty-printed description of all flows to 'results', including
3634 * hidden flows (e.g., set up by in-band control). */
3636 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3638 struct oftable *table;
3640 OFPROTO_FOR_EACH_TABLE (table, p) {
3641 struct cls_cursor cursor;
3644 ovs_rwlock_rdlock(&table->cls.rwlock);
3645 cls_cursor_init(&cursor, &table->cls, NULL);
3646 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3647 flow_stats_ds(rule, results);
3649 ovs_rwlock_unlock(&table->cls.rwlock);
3653 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
3654 * '*engine_type' and '*engine_id', respectively. */
3656 ofproto_get_netflow_ids(const struct ofproto *ofproto,
3657 uint8_t *engine_type, uint8_t *engine_id)
3659 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
3662 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
3663 * true if the port's CFM status was successfully stored into '*status'.
3664 * Returns false if the port did not have CFM configured, in which case
3665 * '*status' is indeterminate.
3667 * The caller must provide and owns '*status', and must free 'status->rmps'. */
3669 ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port,
3670 struct ofproto_cfm_status *status)
3672 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
3674 && ofproto->ofproto_class->get_cfm_status
3675 && ofproto->ofproto_class->get_cfm_status(ofport, status));
3679 handle_aggregate_stats_request(struct ofconn *ofconn,
3680 const struct ofp_header *oh)
3681 OVS_EXCLUDED(ofproto_mutex)
3683 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3684 struct ofputil_flow_stats_request request;
3685 struct ofputil_aggregate_stats stats;
3686 bool unknown_packets, unknown_bytes;
3687 struct rule_criteria criteria;
3688 struct rule_collection rules;
3689 struct ofpbuf *reply;
3693 error = ofputil_decode_flow_stats_request(&request, oh);
3698 rule_criteria_init(&criteria, request.table_id, &request.match, 0,
3699 request.cookie, request.cookie_mask,
3700 request.out_port, request.out_group);
3702 ovs_mutex_lock(&ofproto_mutex);
3703 error = collect_rules_loose(ofproto, &criteria, &rules);
3704 rule_criteria_destroy(&criteria);
3706 rule_collection_ref(&rules);
3708 ovs_mutex_unlock(&ofproto_mutex);
3714 memset(&stats, 0, sizeof stats);
3715 unknown_packets = unknown_bytes = false;
3716 for (i = 0; i < rules.n; i++) {
3717 struct rule *rule = rules.rules[i];
3718 uint64_t packet_count;
3719 uint64_t byte_count;
3721 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
3724 if (packet_count == UINT64_MAX) {
3725 unknown_packets = true;
3727 stats.packet_count += packet_count;
3730 if (byte_count == UINT64_MAX) {
3731 unknown_bytes = true;
3733 stats.byte_count += byte_count;
3738 if (unknown_packets) {
3739 stats.packet_count = UINT64_MAX;
3741 if (unknown_bytes) {
3742 stats.byte_count = UINT64_MAX;
3745 rule_collection_unref(&rules);
3746 rule_collection_destroy(&rules);
3748 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
3749 ofconn_send_reply(ofconn, reply);
3754 struct queue_stats_cbdata {
3755 struct ofport *ofport;
3756 struct list replies;
3761 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3762 const struct netdev_queue_stats *stats)
3764 struct ofputil_queue_stats oqs;
3766 oqs.port_no = cbdata->ofport->pp.port_no;
3767 oqs.queue_id = queue_id;
3768 oqs.tx_bytes = stats->tx_bytes;
3769 oqs.tx_packets = stats->tx_packets;
3770 oqs.tx_errors = stats->tx_errors;
3771 if (stats->created != LLONG_MIN) {
3772 calc_duration(stats->created, cbdata->now,
3773 &oqs.duration_sec, &oqs.duration_nsec);
3775 oqs.duration_sec = oqs.duration_nsec = UINT32_MAX;
3777 ofputil_append_queue_stat(&cbdata->replies, &oqs);
3781 handle_queue_stats_dump_cb(uint32_t queue_id,
3782 struct netdev_queue_stats *stats,
3785 struct queue_stats_cbdata *cbdata = cbdata_;
3787 put_queue_stats(cbdata, queue_id, stats);
3791 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3792 struct queue_stats_cbdata *cbdata)
3794 cbdata->ofport = port;
3795 if (queue_id == OFPQ_ALL) {
3796 netdev_dump_queue_stats(port->netdev,
3797 handle_queue_stats_dump_cb, cbdata);
3799 struct netdev_queue_stats stats;
3801 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3802 put_queue_stats(cbdata, queue_id, &stats);
3804 return OFPERR_OFPQOFC_BAD_QUEUE;
3811 handle_queue_stats_request(struct ofconn *ofconn,
3812 const struct ofp_header *rq)
3814 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3815 struct queue_stats_cbdata cbdata;
3816 struct ofport *port;
3818 struct ofputil_queue_stats_request oqsr;
3820 COVERAGE_INC(ofproto_queue_req);
3822 ofpmp_init(&cbdata.replies, rq);
3823 cbdata.now = time_msec();
3825 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3830 if (oqsr.port_no == OFPP_ANY) {
3831 error = OFPERR_OFPQOFC_BAD_QUEUE;
3832 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3833 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3838 port = ofproto_get_port(ofproto, oqsr.port_no);
3840 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3841 : OFPERR_OFPQOFC_BAD_PORT);
3844 ofconn_send_replies(ofconn, &cbdata.replies);
3846 ofpbuf_list_delete(&cbdata.replies);
3853 is_flow_deletion_pending(const struct ofproto *ofproto,
3854 const struct cls_rule *cls_rule,
3856 OVS_REQUIRES(ofproto_mutex)
3858 if (!hmap_is_empty(&ofproto->deletions)) {
3859 struct ofoperation *op;
3861 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3862 cls_rule_hash(cls_rule, table_id),
3863 &ofproto->deletions) {
3864 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
3874 should_evict_a_rule(struct oftable *table, unsigned int extra_space)
3875 OVS_REQUIRES(ofproto_mutex)
3876 OVS_NO_THREAD_SAFETY_ANALYSIS
3878 return classifier_count(&table->cls) + extra_space > table->max_flows;
3882 evict_rules_from_table(struct ofproto *ofproto, struct oftable *table,
3883 unsigned int extra_space)
3884 OVS_REQUIRES(ofproto_mutex)
3886 while (should_evict_a_rule(table, extra_space)) {
3889 if (!choose_rule_to_evict(table, &rule)) {
3890 return OFPERR_OFPFMFC_TABLE_FULL;
3891 } else if (rule->pending) {
3892 return OFPROTO_POSTPONE;
3894 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
3895 delete_flow__(rule, group, OFPRR_EVICTION);
3896 ofopgroup_submit(group);
3903 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3904 * in which no matching flow already exists in the flow table.
3906 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3907 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3908 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3909 * initiated now but may be retried later.
3911 * The caller retains ownership of 'fm->ofpacts'.
3913 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3916 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
3917 struct ofputil_flow_mod *fm, const struct ofp_header *request)
3918 OVS_REQUIRES(ofproto_mutex)
3920 struct oftable *table;
3921 struct ofopgroup *group;
3927 error = check_table_id(ofproto, fm->table_id);
3933 if (fm->table_id == 0xff) {
3934 if (ofproto->ofproto_class->rule_choose_table) {
3935 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3941 ovs_assert(table_id < ofproto->n_tables);
3945 } else if (fm->table_id < ofproto->n_tables) {
3946 table_id = fm->table_id;
3948 return OFPERR_OFPBRC_BAD_TABLE_ID;
3951 table = &ofproto->tables[table_id];
3953 if (table->flags & OFTABLE_READONLY) {
3954 return OFPERR_OFPBRC_EPERM;
3957 cls_rule_init(&cr, &fm->match, fm->priority);
3959 /* Transform "add" into "modify" if there's an existing identical flow. */
3960 ovs_rwlock_rdlock(&table->cls.rwlock);
3961 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr));
3962 ovs_rwlock_unlock(&table->cls.rwlock);
3964 cls_rule_destroy(&cr);
3965 if (!rule_is_modifiable(rule)) {
3966 return OFPERR_OFPBRC_EPERM;
3967 } else if (rule->pending) {
3968 return OFPROTO_POSTPONE;
3970 struct rule_collection rules;
3972 rule_collection_init(&rules);
3973 rule_collection_add(&rules, rule);
3974 fm->modify_cookie = true;
3975 error = modify_flows__(ofproto, ofconn, fm, request, &rules);
3976 rule_collection_destroy(&rules);
3982 /* Serialize against pending deletion. */
3983 if (is_flow_deletion_pending(ofproto, &cr, table_id)) {
3984 cls_rule_destroy(&cr);
3985 return OFPROTO_POSTPONE;
3988 /* Check for overlap, if requested. */
3989 if (fm->flags & OFPUTIL_FF_CHECK_OVERLAP) {
3992 ovs_rwlock_rdlock(&table->cls.rwlock);
3993 overlaps = classifier_rule_overlaps(&table->cls, &cr);
3994 ovs_rwlock_unlock(&table->cls.rwlock);
3997 cls_rule_destroy(&cr);
3998 return OFPERR_OFPFMFC_OVERLAP;
4002 /* If necessary, evict an existing rule to clear out space. */
4003 error = evict_rules_from_table(ofproto, table, 1);
4005 cls_rule_destroy(&cr);
4009 /* Allocate new rule. */
4010 rule = ofproto->ofproto_class->rule_alloc();
4012 cls_rule_destroy(&cr);
4013 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
4014 ofproto->name, ovs_strerror(error));
4018 /* Initialize base state. */
4019 *CONST_CAST(struct ofproto **, &rule->ofproto) = ofproto;
4020 cls_rule_move(CONST_CAST(struct cls_rule *, &rule->cr), &cr);
4021 atomic_init(&rule->ref_count, 1);
4022 rule->pending = NULL;
4023 rule->flow_cookie = fm->new_cookie;
4024 rule->created = rule->modified = rule->used = time_msec();
4026 ovs_mutex_init(&rule->mutex);
4027 ovs_mutex_lock(&rule->mutex);
4028 rule->idle_timeout = fm->idle_timeout;
4029 rule->hard_timeout = fm->hard_timeout;
4030 ovs_mutex_unlock(&rule->mutex);
4032 *CONST_CAST(uint8_t *, &rule->table_id) = table - ofproto->tables;
4033 rule->flags = fm->flags & OFPUTIL_FF_STATE;
4034 rule->actions = rule_actions_create(ofproto, fm->ofpacts, fm->ofpacts_len);
4035 list_init(&rule->meter_list_node);
4036 rule->eviction_group = NULL;
4037 list_init(&rule->expirable);
4038 rule->monitor_flags = 0;
4039 rule->add_seqno = 0;
4040 rule->modify_seqno = 0;
4042 /* Construct rule, initializing derived state. */
4043 error = ofproto->ofproto_class->rule_construct(rule);
4045 ofproto_rule_destroy__(rule);
4050 oftable_insert_rule(rule);
4052 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
4053 ofoperation_create(group, rule, OFOPERATION_ADD, 0);
4054 ofproto->ofproto_class->rule_insert(rule);
4055 ofopgroup_submit(group);
4060 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
4062 /* Modifies the rules listed in 'rules', changing their actions to match those
4065 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4068 * Returns 0 on success, otherwise an OpenFlow error code. */
4070 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
4071 struct ofputil_flow_mod *fm, const struct ofp_header *request,
4072 const struct rule_collection *rules)
4073 OVS_REQUIRES(ofproto_mutex)
4075 enum ofoperation_type type;
4076 struct ofopgroup *group;
4080 type = fm->command == OFPFC_ADD ? OFOPERATION_REPLACE : OFOPERATION_MODIFY;
4081 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
4082 error = OFPERR_OFPBRC_EPERM;
4083 for (i = 0; i < rules->n; i++) {
4084 struct rule *rule = rules->rules[i];
4085 struct ofoperation *op;
4086 bool actions_changed;
4087 bool reset_counters;
4089 /* FIXME: Implement OFPFUTIL_FF_RESET_COUNTS */
4091 if (rule_is_modifiable(rule)) {
4092 /* At least one rule is modifiable, don't report EPERM error. */
4098 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
4099 rule->actions->ofpacts,
4100 rule->actions->ofpacts_len);
4102 op = ofoperation_create(group, rule, type, 0);
4104 if (fm->modify_cookie && fm->new_cookie != OVS_BE64_MAX) {
4105 ofproto_rule_change_cookie(ofproto, rule, fm->new_cookie);
4107 if (type == OFOPERATION_REPLACE) {
4108 ovs_mutex_lock(&rule->mutex);
4109 rule->idle_timeout = fm->idle_timeout;
4110 rule->hard_timeout = fm->hard_timeout;
4111 ovs_mutex_unlock(&rule->mutex);
4113 rule->flags = fm->flags & OFPUTIL_FF_STATE;
4114 if (fm->idle_timeout || fm->hard_timeout) {
4115 if (!rule->eviction_group) {
4116 eviction_group_add_rule(rule);
4119 eviction_group_remove_rule(rule);
4123 reset_counters = (fm->flags & OFPUTIL_FF_RESET_COUNTS) != 0;
4124 if (actions_changed || reset_counters) {
4125 struct rule_actions *new_actions;
4127 op->actions = rule->actions;
4128 new_actions = rule_actions_create(ofproto,
4129 fm->ofpacts, fm->ofpacts_len);
4131 ovs_mutex_lock(&rule->mutex);
4132 rule->actions = new_actions;
4133 ovs_mutex_unlock(&rule->mutex);
4135 rule->ofproto->ofproto_class->rule_modify_actions(rule,
4138 ofoperation_complete(op, 0);
4141 ofopgroup_submit(group);
4147 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
4148 struct ofputil_flow_mod *fm, const struct ofp_header *request)
4149 OVS_REQUIRES(ofproto_mutex)
4151 if (fm->cookie_mask != htonll(0) || fm->new_cookie == OVS_BE64_MAX) {
4154 return add_flow(ofproto, ofconn, fm, request);
4157 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
4160 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4163 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
4164 struct ofputil_flow_mod *fm,
4165 const struct ofp_header *request)
4166 OVS_REQUIRES(ofproto_mutex)
4168 struct rule_criteria criteria;
4169 struct rule_collection rules;
4172 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0,
4173 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY);
4174 error = collect_rules_loose(ofproto, &criteria, &rules);
4175 rule_criteria_destroy(&criteria);
4178 error = (rules.n > 0
4179 ? modify_flows__(ofproto, ofconn, fm, request, &rules)
4180 : modify_flows_add(ofproto, ofconn, fm, request));
4183 rule_collection_destroy(&rules);
4188 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4191 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4194 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
4195 struct ofputil_flow_mod *fm,
4196 const struct ofp_header *request)
4197 OVS_REQUIRES(ofproto_mutex)
4199 struct rule_criteria criteria;
4200 struct rule_collection rules;
4203 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
4204 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY);
4205 error = collect_rules_strict(ofproto, &criteria, &rules);
4206 rule_criteria_destroy(&criteria);
4210 error = modify_flows_add(ofproto, ofconn, fm, request);
4211 } else if (rules.n == 1) {
4212 error = modify_flows__(ofproto, ofconn, fm, request, &rules);
4216 rule_collection_destroy(&rules);
4221 /* OFPFC_DELETE implementation. */
4224 delete_flow__(struct rule *rule, struct ofopgroup *group,
4225 enum ofp_flow_removed_reason reason)
4226 OVS_REQUIRES(ofproto_mutex)
4228 struct ofproto *ofproto = rule->ofproto;
4230 ofproto_rule_send_removed(rule, reason);
4232 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
4233 oftable_remove_rule(rule);
4234 ofproto->ofproto_class->rule_delete(rule);
4237 /* Deletes the rules listed in 'rules'.
4239 * Returns 0 on success, otherwise an OpenFlow error code. */
4241 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
4242 const struct ofp_header *request,
4243 const struct rule_collection *rules,
4244 enum ofp_flow_removed_reason reason)
4245 OVS_REQUIRES(ofproto_mutex)
4247 struct ofopgroup *group;
4250 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
4251 for (i = 0; i < rules->n; i++) {
4252 delete_flow__(rules->rules[i], group, reason);
4254 ofopgroup_submit(group);
4259 /* Implements OFPFC_DELETE. */
4261 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
4262 const struct ofputil_flow_mod *fm,
4263 const struct ofp_header *request)
4264 OVS_REQUIRES(ofproto_mutex)
4266 struct rule_criteria criteria;
4267 struct rule_collection rules;
4270 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0,
4271 fm->cookie, fm->cookie_mask,
4272 fm->out_port, fm->out_group);
4273 error = collect_rules_loose(ofproto, &criteria, &rules);
4274 rule_criteria_destroy(&criteria);
4276 if (!error && rules.n > 0) {
4277 error = delete_flows__(ofproto, ofconn, request, &rules, OFPRR_DELETE);
4279 rule_collection_destroy(&rules);
4284 /* Implements OFPFC_DELETE_STRICT. */
4286 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
4287 const struct ofputil_flow_mod *fm,
4288 const struct ofp_header *request)
4289 OVS_REQUIRES(ofproto_mutex)
4291 struct rule_criteria criteria;
4292 struct rule_collection rules;
4295 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
4296 fm->cookie, fm->cookie_mask,
4297 fm->out_port, fm->out_group);
4298 error = collect_rules_strict(ofproto, &criteria, &rules);
4299 rule_criteria_destroy(&criteria);
4301 if (!error && rules.n > 0) {
4302 error = delete_flows__(ofproto, ofconn, request, &rules, OFPRR_DELETE);
4304 rule_collection_destroy(&rules);
4310 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
4311 OVS_REQUIRES(ofproto_mutex)
4313 struct ofputil_flow_removed fr;
4315 if (ofproto_rule_is_hidden(rule) ||
4316 !(rule->flags & OFPUTIL_FF_SEND_FLOW_REM)) {
4320 minimatch_expand(&rule->cr.match, &fr.match);
4321 fr.priority = rule->cr.priority;
4322 fr.cookie = rule->flow_cookie;
4324 fr.table_id = rule->table_id;
4325 calc_duration(rule->created, time_msec(),
4326 &fr.duration_sec, &fr.duration_nsec);
4327 ovs_mutex_lock(&rule->mutex);
4328 fr.idle_timeout = rule->idle_timeout;
4329 fr.hard_timeout = rule->hard_timeout;
4330 ovs_mutex_unlock(&rule->mutex);
4331 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
4334 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
4337 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
4338 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
4341 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
4344 * ofproto implementation ->run() functions should use this function to expire
4345 * OpenFlow flows. */
4347 ofproto_rule_expire(struct rule *rule, uint8_t reason)
4348 OVS_REQUIRES(ofproto_mutex)
4350 struct ofproto *ofproto = rule->ofproto;
4352 ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT
4353 || reason == OFPRR_DELETE || reason == OFPRR_GROUP_DELETE);
4355 ofproto_rule_delete__(ofproto, rule, reason);
4358 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
4359 * means "infinite". */
4361 reduce_timeout(uint16_t max, uint16_t *timeout)
4363 if (max && (!*timeout || *timeout > max)) {
4368 /* If 'idle_timeout' is nonzero, and 'rule' has no idle timeout or an idle
4369 * timeout greater than 'idle_timeout', lowers 'rule''s idle timeout to
4370 * 'idle_timeout' seconds. Similarly for 'hard_timeout'.
4372 * Suitable for implementing OFPACT_FIN_TIMEOUT. */
4374 ofproto_rule_reduce_timeouts(struct rule *rule,
4375 uint16_t idle_timeout, uint16_t hard_timeout)
4376 OVS_EXCLUDED(ofproto_mutex, rule->mutex)
4378 if (!idle_timeout && !hard_timeout) {
4382 ovs_mutex_lock(&ofproto_mutex);
4383 if (list_is_empty(&rule->expirable)) {
4384 list_insert(&rule->ofproto->expirable, &rule->expirable);
4386 ovs_mutex_unlock(&ofproto_mutex);
4388 ovs_mutex_lock(&rule->mutex);
4389 reduce_timeout(idle_timeout, &rule->idle_timeout);
4390 reduce_timeout(hard_timeout, &rule->hard_timeout);
4391 ovs_mutex_unlock(&rule->mutex);
4395 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4396 OVS_EXCLUDED(ofproto_mutex)
4398 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4399 struct ofputil_flow_mod fm;
4400 uint64_t ofpacts_stub[1024 / 8];
4401 struct ofpbuf ofpacts;
4405 error = reject_slave_controller(ofconn);
4410 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
4411 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
4413 u16_to_ofp(ofproto->max_ports),
4416 error = ofproto_check_ofpacts(ofproto, fm.ofpacts, fm.ofpacts_len);
4419 error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
4422 goto exit_free_ofpacts;
4425 /* Record the operation for logging a summary report. */
4426 switch (fm.command) {
4432 case OFPFC_MODIFY_STRICT:
4433 ofproto->n_modify++;
4437 case OFPFC_DELETE_STRICT:
4438 ofproto->n_delete++;
4443 if (ofproto->next_op_report == LLONG_MAX) {
4444 ofproto->first_op = now;
4445 ofproto->next_op_report = MAX(now + 10 * 1000,
4446 ofproto->op_backoff);
4447 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
4449 ofproto->last_op = now;
4452 ofpbuf_uninit(&ofpacts);
4458 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
4459 struct ofputil_flow_mod *fm, const struct ofp_header *oh)
4460 OVS_EXCLUDED(ofproto_mutex)
4464 ovs_mutex_lock(&ofproto_mutex);
4465 if (ofproto->n_pending < 50) {
4466 switch (fm->command) {
4468 error = add_flow(ofproto, ofconn, fm, oh);
4472 error = modify_flows_loose(ofproto, ofconn, fm, oh);
4475 case OFPFC_MODIFY_STRICT:
4476 error = modify_flow_strict(ofproto, ofconn, fm, oh);
4480 error = delete_flows_loose(ofproto, ofconn, fm, oh);
4483 case OFPFC_DELETE_STRICT:
4484 error = delete_flow_strict(ofproto, ofconn, fm, oh);
4488 if (fm->command > 0xff) {
4489 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
4490 "flow_mod_table_id extension is not enabled",
4493 error = OFPERR_OFPFMFC_BAD_COMMAND;
4497 ovs_assert(!list_is_empty(&ofproto->pending));
4498 error = OFPROTO_POSTPONE;
4500 ovs_mutex_unlock(&ofproto_mutex);
4502 run_rule_executes(ofproto);
4507 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4509 struct ofputil_role_request request;
4510 struct ofputil_role_request reply;
4514 error = ofputil_decode_role_message(oh, &request);
4519 if (request.role != OFPCR12_ROLE_NOCHANGE) {
4520 if (ofconn_get_role(ofconn) != request.role
4521 && ofconn_has_pending_opgroups(ofconn)) {
4522 return OFPROTO_POSTPONE;
4525 if (request.have_generation_id
4526 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
4527 return OFPERR_OFPRRFC_STALE;
4530 ofconn_set_role(ofconn, request.role);
4533 reply.role = ofconn_get_role(ofconn);
4534 reply.have_generation_id = ofconn_get_master_election_id(
4535 ofconn, &reply.generation_id);
4536 buf = ofputil_encode_role_reply(oh, &reply);
4537 ofconn_send_reply(ofconn, buf);
4543 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
4544 const struct ofp_header *oh)
4546 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
4547 enum ofputil_protocol cur, next;
4549 cur = ofconn_get_protocol(ofconn);
4550 next = ofputil_protocol_set_tid(cur, msg->set != 0);
4551 ofconn_set_protocol(ofconn, next);
4557 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4559 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
4560 enum ofputil_protocol cur, next;
4561 enum ofputil_protocol next_base;
4563 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
4565 return OFPERR_OFPBRC_EPERM;
4568 cur = ofconn_get_protocol(ofconn);
4569 next = ofputil_protocol_set_base(cur, next_base);
4570 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
4571 /* Avoid sending async messages in surprising protocol. */
4572 return OFPROTO_POSTPONE;
4575 ofconn_set_protocol(ofconn, next);
4580 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
4581 const struct ofp_header *oh)
4583 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
4586 format = ntohl(msg->format);
4587 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
4588 return OFPERR_OFPBRC_EPERM;
4591 if (format != ofconn_get_packet_in_format(ofconn)
4592 && ofconn_has_pending_opgroups(ofconn)) {
4593 /* Avoid sending async message in surprsing packet in format. */
4594 return OFPROTO_POSTPONE;
4597 ofconn_set_packet_in_format(ofconn, format);
4602 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
4604 const struct nx_async_config *msg = ofpmsg_body(oh);
4605 uint32_t master[OAM_N_TYPES];
4606 uint32_t slave[OAM_N_TYPES];
4608 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
4609 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
4610 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
4612 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
4613 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
4614 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
4616 ofconn_set_async_config(ofconn, master, slave);
4617 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
4618 !ofconn_get_miss_send_len(ofconn)) {
4619 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
4626 handle_nxt_get_async_request(struct ofconn *ofconn, const struct ofp_header *oh)
4629 uint32_t master[OAM_N_TYPES];
4630 uint32_t slave[OAM_N_TYPES];
4631 struct nx_async_config *msg;
4633 ofconn_get_async_config(ofconn, master, slave);
4634 buf = ofpraw_alloc_reply(OFPRAW_OFPT13_GET_ASYNC_REPLY, oh, 0);
4635 msg = ofpbuf_put_zeros(buf, sizeof *msg);
4637 msg->packet_in_mask[0] = htonl(master[OAM_PACKET_IN]);
4638 msg->port_status_mask[0] = htonl(master[OAM_PORT_STATUS]);
4639 msg->flow_removed_mask[0] = htonl(master[OAM_FLOW_REMOVED]);
4641 msg->packet_in_mask[1] = htonl(slave[OAM_PACKET_IN]);
4642 msg->port_status_mask[1] = htonl(slave[OAM_PORT_STATUS]);
4643 msg->flow_removed_mask[1] = htonl(slave[OAM_FLOW_REMOVED]);
4645 ofconn_send_reply(ofconn, buf);
4651 handle_nxt_set_controller_id(struct ofconn *ofconn,
4652 const struct ofp_header *oh)
4654 const struct nx_controller_id *nci = ofpmsg_body(oh);
4656 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
4657 return OFPERR_NXBRC_MUST_BE_ZERO;
4660 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
4665 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4669 if (ofconn_has_pending_opgroups(ofconn)) {
4670 return OFPROTO_POSTPONE;
4673 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
4674 ? OFPRAW_OFPT10_BARRIER_REPLY
4675 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
4676 ofconn_send_reply(ofconn, buf);
4681 ofproto_compose_flow_refresh_update(const struct rule *rule,
4682 enum nx_flow_monitor_flags flags,
4684 OVS_REQUIRES(ofproto_mutex)
4686 struct ofoperation *op = rule->pending;
4687 const struct rule_actions *actions;
4688 struct ofputil_flow_update fu;
4691 if (op && op->type == OFOPERATION_ADD) {
4692 /* We'll report the final flow when the operation completes. Reporting
4693 * it now would cause a duplicate report later. */
4697 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
4698 ? NXFME_ADDED : NXFME_MODIFIED);
4700 ovs_mutex_lock(&rule->mutex);
4701 fu.idle_timeout = rule->idle_timeout;
4702 fu.hard_timeout = rule->hard_timeout;
4703 ovs_mutex_unlock(&rule->mutex);
4704 fu.table_id = rule->table_id;
4705 fu.cookie = rule->flow_cookie;
4706 minimatch_expand(&rule->cr.match, &match);
4708 fu.priority = rule->cr.priority;
4710 if (!(flags & NXFMF_ACTIONS)) {
4713 actions = rule->actions;
4715 /* An operation is in progress. Use the previous version of the flow's
4716 * actions, so that when the operation commits we report the change. */
4718 case OFOPERATION_ADD:
4721 case OFOPERATION_MODIFY:
4722 case OFOPERATION_REPLACE:
4723 actions = op->actions ? op->actions : rule->actions;
4726 case OFOPERATION_DELETE:
4727 actions = rule->actions;
4734 fu.ofpacts = actions ? actions->ofpacts : NULL;
4735 fu.ofpacts_len = actions ? actions->ofpacts_len : 0;
4737 if (list_is_empty(msgs)) {
4738 ofputil_start_flow_update(msgs);
4740 ofputil_append_flow_update(&fu, msgs);
4744 ofmonitor_compose_refresh_updates(struct rule_collection *rules,
4746 OVS_REQUIRES(ofproto_mutex)
4750 for (i = 0; i < rules->n; i++) {
4751 struct rule *rule = rules->rules[i];
4752 enum nx_flow_monitor_flags flags = rule->monitor_flags;
4753 rule->monitor_flags = 0;
4755 ofproto_compose_flow_refresh_update(rule, flags, msgs);
4760 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
4761 struct rule *rule, uint64_t seqno,
4762 struct rule_collection *rules)
4763 OVS_REQUIRES(ofproto_mutex)
4765 enum nx_flow_monitor_flags update;
4767 if (ofproto_rule_is_hidden(rule)) {
4772 ? ofoperation_has_out_port(rule->pending, m->out_port)
4773 : ofproto_rule_has_out_port(rule, m->out_port))) {
4778 if (rule->add_seqno > seqno) {
4779 update = NXFMF_ADD | NXFMF_MODIFY;
4780 } else if (rule->modify_seqno > seqno) {
4781 update = NXFMF_MODIFY;
4786 if (!(m->flags & update)) {
4790 update = NXFMF_INITIAL;
4793 if (!rule->monitor_flags) {
4794 rule_collection_add(rules, rule);
4796 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
4800 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
4802 struct rule_collection *rules)
4803 OVS_REQUIRES(ofproto_mutex)
4805 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
4806 const struct ofoperation *op;
4807 const struct oftable *table;
4808 struct cls_rule target;
4810 cls_rule_init_from_minimatch(&target, &m->match, 0);
4811 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
4812 struct cls_cursor cursor;
4815 ovs_rwlock_rdlock(&table->cls.rwlock);
4816 cls_cursor_init(&cursor, &table->cls, &target);
4817 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4818 ovs_assert(!rule->pending); /* XXX */
4819 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4821 ovs_rwlock_unlock(&table->cls.rwlock);
4824 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
4825 struct rule *rule = op->rule;
4827 if (((m->table_id == 0xff
4828 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
4829 : m->table_id == rule->table_id))
4830 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
4831 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4834 cls_rule_destroy(&target);
4838 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
4839 struct rule_collection *rules)
4840 OVS_REQUIRES(ofproto_mutex)
4842 if (m->flags & NXFMF_INITIAL) {
4843 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
4848 ofmonitor_collect_resume_rules(struct ofmonitor *m,
4849 uint64_t seqno, struct rule_collection *rules)
4850 OVS_REQUIRES(ofproto_mutex)
4852 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
4856 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
4857 OVS_EXCLUDED(ofproto_mutex)
4859 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4860 struct ofmonitor **monitors;
4861 size_t n_monitors, allocated_monitors;
4862 struct rule_collection rules;
4863 struct list replies;
4869 ofpbuf_use_const(&b, oh, ntohs(oh->length));
4871 n_monitors = allocated_monitors = 0;
4873 ovs_mutex_lock(&ofproto_mutex);
4875 struct ofputil_flow_monitor_request request;
4876 struct ofmonitor *m;
4879 retval = ofputil_decode_flow_monitor_request(&request, &b);
4880 if (retval == EOF) {
4882 } else if (retval) {
4887 if (request.table_id != 0xff
4888 && request.table_id >= ofproto->n_tables) {
4889 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4893 error = ofmonitor_create(&request, ofconn, &m);
4898 if (n_monitors >= allocated_monitors) {
4899 monitors = x2nrealloc(monitors, &allocated_monitors,
4902 monitors[n_monitors++] = m;
4905 rule_collection_init(&rules);
4906 for (i = 0; i < n_monitors; i++) {
4907 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
4910 ofpmp_init(&replies, oh);
4911 ofmonitor_compose_refresh_updates(&rules, &replies);
4912 ovs_mutex_unlock(&ofproto_mutex);
4914 rule_collection_destroy(&rules);
4916 ofconn_send_replies(ofconn, &replies);
4922 for (i = 0; i < n_monitors; i++) {
4923 ofmonitor_destroy(monitors[i]);
4926 ovs_mutex_unlock(&ofproto_mutex);
4932 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
4933 OVS_EXCLUDED(ofproto_mutex)
4935 struct ofmonitor *m;
4939 id = ofputil_decode_flow_monitor_cancel(oh);
4941 ovs_mutex_lock(&ofproto_mutex);
4942 m = ofmonitor_lookup(ofconn, id);
4944 ofmonitor_destroy(m);
4947 error = OFPERR_NXBRC_FM_BAD_ID;
4949 ovs_mutex_unlock(&ofproto_mutex);
4954 /* Meters implementation.
4956 * Meter table entry, indexed by the OpenFlow meter_id.
4957 * These are always dynamically allocated to allocate enough space for
4959 * 'created' is used to compute the duration for meter stats.
4960 * 'list rules' is needed so that we can delete the dependent rules when the
4961 * meter table entry is deleted.
4962 * 'provider_meter_id' is for the provider's private use.
4965 long long int created; /* Time created. */
4966 struct list rules; /* List of "struct rule_dpif"s. */
4967 ofproto_meter_id provider_meter_id;
4968 uint16_t flags; /* Meter flags. */
4969 uint16_t n_bands; /* Number of meter bands. */
4970 struct ofputil_meter_band *bands;
4974 * This is used in instruction validation at flow set-up time,
4975 * as flows may not use non-existing meters.
4976 * Return value of UINT32_MAX signifies an invalid meter.
4979 get_provider_meter_id(const struct ofproto *ofproto, uint32_t of_meter_id)
4981 if (of_meter_id && of_meter_id <= ofproto->meter_features.max_meters) {
4982 const struct meter *meter = ofproto->meters[of_meter_id];
4984 return meter->provider_meter_id.uint32;
4991 meter_update(struct meter *meter, const struct ofputil_meter_config *config)
4995 meter->flags = config->flags;
4996 meter->n_bands = config->n_bands;
4997 meter->bands = xmemdup(config->bands,
4998 config->n_bands * sizeof *meter->bands);
5001 static struct meter *
5002 meter_create(const struct ofputil_meter_config *config,
5003 ofproto_meter_id provider_meter_id)
5005 struct meter *meter;
5007 meter = xzalloc(sizeof *meter);
5008 meter->provider_meter_id = provider_meter_id;
5009 meter->created = time_msec();
5010 list_init(&meter->rules);
5012 meter_update(meter, config);
5018 meter_delete(struct ofproto *ofproto, uint32_t first, uint32_t last)
5019 OVS_REQUIRES(ofproto_mutex)
5022 for (mid = first; mid <= last; ++mid) {
5023 struct meter *meter = ofproto->meters[mid];
5025 ofproto->meters[mid] = NULL;
5026 ofproto->ofproto_class->meter_del(ofproto,
5027 meter->provider_meter_id);
5035 handle_add_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
5037 ofproto_meter_id provider_meter_id = { UINT32_MAX };
5038 struct meter **meterp = &ofproto->meters[mm->meter.meter_id];
5042 return OFPERR_OFPMMFC_METER_EXISTS;
5045 error = ofproto->ofproto_class->meter_set(ofproto, &provider_meter_id,
5048 ovs_assert(provider_meter_id.uint32 != UINT32_MAX);
5049 *meterp = meter_create(&mm->meter, provider_meter_id);
5055 handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
5057 struct meter *meter = ofproto->meters[mm->meter.meter_id];
5059 uint32_t provider_meter_id;
5062 return OFPERR_OFPMMFC_UNKNOWN_METER;
5065 provider_meter_id = meter->provider_meter_id.uint32;
5066 error = ofproto->ofproto_class->meter_set(ofproto,
5067 &meter->provider_meter_id,
5069 ovs_assert(meter->provider_meter_id.uint32 == provider_meter_id);
5071 meter_update(meter, &mm->meter);
5077 handle_delete_meter(struct ofconn *ofconn, const struct ofp_header *oh,
5078 struct ofputil_meter_mod *mm)
5079 OVS_EXCLUDED(ofproto_mutex)
5081 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5082 uint32_t meter_id = mm->meter.meter_id;
5083 struct rule_collection rules;
5084 enum ofperr error = 0;
5085 uint32_t first, last;
5087 if (meter_id == OFPM13_ALL) {
5089 last = ofproto->meter_features.max_meters;
5091 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
5094 first = last = meter_id;
5097 /* First delete the rules that use this meter. If any of those rules are
5098 * currently being modified, postpone the whole operation until later. */
5099 rule_collection_init(&rules);
5100 ovs_mutex_lock(&ofproto_mutex);
5101 for (meter_id = first; meter_id <= last; ++meter_id) {
5102 struct meter *meter = ofproto->meters[meter_id];
5103 if (meter && !list_is_empty(&meter->rules)) {
5106 LIST_FOR_EACH (rule, meter_list_node, &meter->rules) {
5107 if (rule->pending) {
5108 error = OFPROTO_POSTPONE;
5111 rule_collection_add(&rules, rule);
5116 delete_flows__(ofproto, ofconn, oh, &rules, OFPRR_METER_DELETE);
5119 /* Delete the meters. */
5120 meter_delete(ofproto, first, last);
5123 ovs_mutex_unlock(&ofproto_mutex);
5124 rule_collection_destroy(&rules);
5130 handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5132 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5133 struct ofputil_meter_mod mm;
5134 uint64_t bands_stub[256 / 8];
5135 struct ofpbuf bands;
5139 error = reject_slave_controller(ofconn);
5144 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5146 error = ofputil_decode_meter_mod(oh, &mm, &bands);
5148 goto exit_free_bands;
5151 meter_id = mm.meter.meter_id;
5153 if (mm.command != OFPMC13_DELETE) {
5154 /* Fails also when meters are not implemented by the provider. */
5155 if (meter_id == 0 || meter_id > OFPM13_MAX) {
5156 error = OFPERR_OFPMMFC_INVALID_METER;
5157 goto exit_free_bands;
5158 } else if (meter_id > ofproto->meter_features.max_meters) {
5159 error = OFPERR_OFPMMFC_OUT_OF_METERS;
5160 goto exit_free_bands;
5162 if (mm.meter.n_bands > ofproto->meter_features.max_bands) {
5163 error = OFPERR_OFPMMFC_OUT_OF_BANDS;
5164 goto exit_free_bands;
5168 switch (mm.command) {
5170 error = handle_add_meter(ofproto, &mm);
5173 case OFPMC13_MODIFY:
5174 error = handle_modify_meter(ofproto, &mm);
5177 case OFPMC13_DELETE:
5178 error = handle_delete_meter(ofconn, oh, &mm);
5182 error = OFPERR_OFPMMFC_BAD_COMMAND;
5187 ofpbuf_uninit(&bands);
5192 handle_meter_features_request(struct ofconn *ofconn,
5193 const struct ofp_header *request)
5195 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5196 struct ofputil_meter_features features;
5199 if (ofproto->ofproto_class->meter_get_features) {
5200 ofproto->ofproto_class->meter_get_features(ofproto, &features);
5202 memset(&features, 0, sizeof features);
5204 b = ofputil_encode_meter_features_reply(&features, request);
5206 ofconn_send_reply(ofconn, b);
5211 handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request,
5214 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5215 struct list replies;
5216 uint64_t bands_stub[256 / 8];
5217 struct ofpbuf bands;
5218 uint32_t meter_id, first, last;
5220 ofputil_decode_meter_request(request, &meter_id);
5222 if (meter_id == OFPM13_ALL) {
5224 last = ofproto->meter_features.max_meters;
5226 if (!meter_id || meter_id > ofproto->meter_features.max_meters ||
5227 !ofproto->meters[meter_id]) {
5228 return OFPERR_OFPMMFC_UNKNOWN_METER;
5230 first = last = meter_id;
5233 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5234 ofpmp_init(&replies, request);
5236 for (meter_id = first; meter_id <= last; ++meter_id) {
5237 struct meter *meter = ofproto->meters[meter_id];
5239 continue; /* Skip non-existing meters. */
5241 if (type == OFPTYPE_METER_STATS_REQUEST) {
5242 struct ofputil_meter_stats stats;
5244 stats.meter_id = meter_id;
5246 /* Provider sets the packet and byte counts, we do the rest. */
5247 stats.flow_count = list_size(&meter->rules);
5248 calc_duration(meter->created, time_msec(),
5249 &stats.duration_sec, &stats.duration_nsec);
5250 stats.n_bands = meter->n_bands;
5251 ofpbuf_clear(&bands);
5253 = ofpbuf_put_uninit(&bands,
5254 meter->n_bands * sizeof *stats.bands);
5256 if (!ofproto->ofproto_class->meter_get(ofproto,
5257 meter->provider_meter_id,
5259 ofputil_append_meter_stats(&replies, &stats);
5261 } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */
5262 struct ofputil_meter_config config;
5264 config.meter_id = meter_id;
5265 config.flags = meter->flags;
5266 config.n_bands = meter->n_bands;
5267 config.bands = meter->bands;
5268 ofputil_append_meter_config(&replies, &config);
5272 ofconn_send_replies(ofconn, &replies);
5273 ofpbuf_uninit(&bands);
5278 ofproto_group_lookup(const struct ofproto *ofproto, uint32_t group_id,
5279 struct ofgroup **group)
5280 OVS_TRY_RDLOCK(true, (*group)->rwlock)
5282 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5283 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
5284 hash_int(group_id, 0), &ofproto->groups) {
5285 if ((*group)->group_id == group_id) {
5286 ovs_rwlock_rdlock(&(*group)->rwlock);
5287 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5291 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5296 ofproto_group_release(struct ofgroup *group)
5297 OVS_RELEASES(group->rwlock)
5299 ovs_rwlock_unlock(&group->rwlock);
5303 ofproto_group_write_lookup(const struct ofproto *ofproto, uint32_t group_id,
5304 struct ofgroup **group)
5305 OVS_TRY_WRLOCK(true, ofproto->groups_rwlock)
5306 OVS_TRY_WRLOCK(true, (*group)->rwlock)
5308 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5309 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
5310 hash_int(group_id, 0), &ofproto->groups) {
5311 if ((*group)->group_id == group_id) {
5312 ovs_rwlock_wrlock(&(*group)->rwlock);
5316 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5321 ofproto_group_exists__(const struct ofproto *ofproto, uint32_t group_id)
5322 OVS_REQ_RDLOCK(ofproto->groups_rwlock)
5324 struct ofgroup *grp;
5326 HMAP_FOR_EACH_IN_BUCKET (grp, hmap_node,
5327 hash_int(group_id, 0), &ofproto->groups) {
5328 if (grp->group_id == group_id) {
5336 ofproto_group_exists(const struct ofproto *ofproto, uint32_t group_id)
5337 OVS_EXCLUDED(ofproto->groups_rwlock)
5341 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5342 exists = ofproto_group_exists__(ofproto, group_id);
5343 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5349 group_get_ref_count(struct ofgroup *group)
5350 OVS_EXCLUDED(ofproto_mutex)
5352 struct ofproto *ofproto = group->ofproto;
5353 struct rule_criteria criteria;
5354 struct rule_collection rules;
5359 match_init_catchall(&match);
5360 rule_criteria_init(&criteria, 0xff, &match, 0, htonll(0), htonll(0),
5361 OFPP_ANY, group->group_id);
5362 ovs_mutex_lock(&ofproto_mutex);
5363 error = collect_rules_loose(ofproto, &criteria, &rules);
5364 ovs_mutex_unlock(&ofproto_mutex);
5365 rule_criteria_destroy(&criteria);
5367 count = !error && rules.n < UINT32_MAX ? rules.n : UINT32_MAX;
5369 rule_collection_destroy(&rules);
5374 append_group_stats(struct ofgroup *group, struct list *replies)
5375 OVS_REQ_RDLOCK(group->rwlock)
5377 struct ofputil_group_stats ogs;
5378 struct ofproto *ofproto = group->ofproto;
5379 long long int now = time_msec();
5382 ogs.bucket_stats = xmalloc(group->n_buckets * sizeof *ogs.bucket_stats);
5384 /* Provider sets the packet and byte counts, we do the rest. */
5385 ogs.ref_count = group_get_ref_count(group);
5386 ogs.n_buckets = group->n_buckets;
5388 error = (ofproto->ofproto_class->group_get_stats
5389 ? ofproto->ofproto_class->group_get_stats(group, &ogs)
5392 ogs.packet_count = UINT64_MAX;
5393 ogs.byte_count = UINT64_MAX;
5394 memset(ogs.bucket_stats, 0xff,
5395 ogs.n_buckets * sizeof *ogs.bucket_stats);
5398 ogs.group_id = group->group_id;
5399 calc_duration(group->created, now, &ogs.duration_sec, &ogs.duration_nsec);
5401 ofputil_append_group_stats(replies, &ogs);
5403 free(ogs.bucket_stats);
5407 handle_group_stats_request(struct ofconn *ofconn,
5408 const struct ofp_header *request)
5410 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5411 struct list replies;
5413 struct ofgroup *group;
5416 error = ofputil_decode_group_stats_request(request, &group_id);
5421 ofpmp_init(&replies, request);
5423 if (group_id == OFPG_ALL) {
5424 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5425 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
5426 ovs_rwlock_rdlock(&group->rwlock);
5427 append_group_stats(group, &replies);
5428 ovs_rwlock_unlock(&group->rwlock);
5430 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5432 if (ofproto_group_lookup(ofproto, group_id, &group)) {
5433 append_group_stats(group, &replies);
5434 ofproto_group_release(group);
5438 ofconn_send_replies(ofconn, &replies);
5444 handle_group_desc_stats_request(struct ofconn *ofconn,
5445 const struct ofp_header *request)
5447 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5448 struct list replies;
5449 struct ofputil_group_desc gds;
5450 struct ofgroup *group;
5452 ofpmp_init(&replies, request);
5454 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5455 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
5456 gds.group_id = group->group_id;
5457 gds.type = group->type;
5458 ofputil_append_group_desc_reply(&gds, &group->buckets, &replies);
5460 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5462 ofconn_send_replies(ofconn, &replies);
5468 handle_group_features_stats_request(struct ofconn *ofconn,
5469 const struct ofp_header *request)
5471 struct ofproto *p = ofconn_get_ofproto(ofconn);
5474 msg = ofputil_encode_group_features_reply(&p->ogf, request);
5476 ofconn_send_reply(ofconn, msg);
5483 handle_queue_get_config_request(struct ofconn *ofconn,
5484 const struct ofp_header *oh)
5486 struct ofproto *p = ofconn_get_ofproto(ofconn);
5487 struct netdev_queue_dump queue_dump;
5488 struct ofport *ofport;
5489 unsigned int queue_id;
5490 struct ofpbuf *reply;
5491 struct smap details;
5495 error = ofputil_decode_queue_get_config_request(oh, &request);
5500 ofport = ofproto_get_port(p, request);
5502 return OFPERR_OFPQOFC_BAD_PORT;
5505 reply = ofputil_encode_queue_get_config_reply(oh);
5507 smap_init(&details);
5508 NETDEV_QUEUE_FOR_EACH (&queue_id, &details, &queue_dump, ofport->netdev) {
5509 struct ofputil_queue_config queue;
5511 /* None of the existing queues have compatible properties, so we
5512 * hard-code omitting min_rate and max_rate. */
5513 queue.queue_id = queue_id;
5514 queue.min_rate = UINT16_MAX;
5515 queue.max_rate = UINT16_MAX;
5516 ofputil_append_queue_get_config_reply(reply, &queue);
5518 smap_destroy(&details);
5520 ofconn_send_reply(ofconn, reply);
5525 /* Implements OFPGC11_ADD
5526 * in which no matching flow already exists in the flow table.
5528 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
5529 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
5530 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
5531 * initiated now but may be retried later.
5533 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
5534 * ownership remains with the caller.
5536 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
5539 add_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
5541 struct ofgroup *ofgroup;
5544 if (gm->group_id > OFPG_MAX) {
5545 return OFPERR_OFPGMFC_INVALID_GROUP;
5547 if (gm->type > OFPGT11_FF) {
5548 return OFPERR_OFPGMFC_BAD_TYPE;
5551 /* Allocate new group and initialize it. */
5552 ofgroup = ofproto->ofproto_class->group_alloc();
5554 VLOG_WARN_RL(&rl, "%s: failed to create group", ofproto->name);
5555 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
5558 ovs_rwlock_init(&ofgroup->rwlock);
5559 ofgroup->ofproto = ofproto;
5560 ofgroup->group_id = gm->group_id;
5561 ofgroup->type = gm->type;
5562 ofgroup->created = ofgroup->modified = time_msec();
5564 list_move(&ofgroup->buckets, &gm->buckets);
5565 ofgroup->n_buckets = list_size(&ofgroup->buckets);
5567 /* Construct called BEFORE any locks are held. */
5568 error = ofproto->ofproto_class->group_construct(ofgroup);
5573 /* We wrlock as late as possible to minimize the time we jam any other
5574 * threads: No visible state changes before acquiring the lock. */
5575 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5577 if (ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
5578 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
5582 if (ofproto_group_exists__(ofproto, gm->group_id)) {
5583 error = OFPERR_OFPGMFC_GROUP_EXISTS;
5588 /* Insert new group. */
5589 hmap_insert(&ofproto->groups, &ofgroup->hmap_node,
5590 hash_int(ofgroup->group_id, 0));
5591 ofproto->n_groups[ofgroup->type]++;
5593 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5598 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5599 ofproto->ofproto_class->group_destruct(ofgroup);
5601 ofputil_bucket_list_destroy(&ofgroup->buckets);
5602 ofproto->ofproto_class->group_dealloc(ofgroup);
5607 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
5610 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
5613 modify_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
5615 struct ofgroup *ofgroup;
5616 struct ofgroup *victim;
5619 if (gm->group_id > OFPG_MAX) {
5620 return OFPERR_OFPGMFC_INVALID_GROUP;
5623 if (gm->type > OFPGT11_FF) {
5624 return OFPERR_OFPGMFC_BAD_TYPE;
5627 victim = ofproto->ofproto_class->group_alloc();
5629 VLOG_WARN_RL(&rl, "%s: failed to allocate group", ofproto->name);
5630 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
5633 if (!ofproto_group_write_lookup(ofproto, gm->group_id, &ofgroup)) {
5634 error = OFPERR_OFPGMFC_UNKNOWN_GROUP;
5637 /* Both group's and its container's write locks held now.
5638 * Also, n_groups[] is protected by ofproto->groups_rwlock. */
5639 if (ofgroup->type != gm->type
5640 && ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
5641 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
5646 list_move(&victim->buckets, &ofgroup->buckets);
5648 ofgroup->type = gm->type;
5649 list_move(&ofgroup->buckets, &gm->buckets);
5650 ofgroup->n_buckets = list_size(&ofgroup->buckets);
5652 error = ofproto->ofproto_class->group_modify(ofgroup, victim);
5654 ofputil_bucket_list_destroy(&victim->buckets);
5655 ofproto->n_groups[victim->type]--;
5656 ofproto->n_groups[ofgroup->type]++;
5657 ofgroup->modified = time_msec();
5659 ofputil_bucket_list_destroy(&ofgroup->buckets);
5662 list_move(&ofgroup->buckets, &victim->buckets);
5666 ovs_rwlock_unlock(&ofgroup->rwlock);
5667 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5669 ofproto->ofproto_class->group_dealloc(victim);
5674 delete_group__(struct ofproto *ofproto, struct ofgroup *ofgroup)
5675 OVS_RELEASES(ofproto->groups_rwlock)
5678 struct ofputil_flow_mod fm;
5680 /* Delete all flow entries containing this group in a group action */
5681 match_init_catchall(&match);
5682 flow_mod_init(&fm, &match, 0, NULL, 0, OFPFC_DELETE);
5683 fm.out_group = ofgroup->group_id;
5684 handle_flow_mod__(ofproto, NULL, &fm, NULL);
5686 /* Must wait until existing readers are done,
5687 * while holding the container's write lock at the same time. */
5688 ovs_rwlock_wrlock(&ofgroup->rwlock);
5689 hmap_remove(&ofproto->groups, &ofgroup->hmap_node);
5690 /* No-one can find this group any more. */
5691 ofproto->n_groups[ofgroup->type]--;
5692 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5694 ofproto->ofproto_class->group_destruct(ofgroup);
5695 ofputil_bucket_list_destroy(&ofgroup->buckets);
5696 ovs_rwlock_unlock(&ofgroup->rwlock);
5697 ovs_rwlock_destroy(&ofgroup->rwlock);
5698 ofproto->ofproto_class->group_dealloc(ofgroup);
5701 /* Implements OFPGC_DELETE. */
5703 delete_group(struct ofproto *ofproto, uint32_t group_id)
5705 struct ofgroup *ofgroup;
5707 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5708 if (group_id == OFPG_ALL) {
5710 struct hmap_node *node = hmap_first(&ofproto->groups);
5714 ofgroup = CONTAINER_OF(node, struct ofgroup, hmap_node);
5715 delete_group__(ofproto, ofgroup);
5716 /* Lock for each node separately, so that we will not jam the
5717 * other threads for too long time. */
5718 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5721 HMAP_FOR_EACH_IN_BUCKET (ofgroup, hmap_node,
5722 hash_int(group_id, 0), &ofproto->groups) {
5723 if (ofgroup->group_id == group_id) {
5724 delete_group__(ofproto, ofgroup);
5729 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5733 handle_group_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5735 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5736 struct ofputil_group_mod gm;
5739 error = reject_slave_controller(ofconn);
5744 error = ofputil_decode_group_mod(oh, &gm);
5749 switch (gm.command) {
5751 return add_group(ofproto, &gm);
5753 case OFPGC11_MODIFY:
5754 return modify_group(ofproto, &gm);
5756 case OFPGC11_DELETE:
5757 delete_group(ofproto, gm.group_id);
5761 if (gm.command > OFPGC11_DELETE) {
5762 VLOG_WARN_RL(&rl, "%s: Invalid group_mod command type %d",
5763 ofproto->name, gm.command);
5765 return OFPERR_OFPGMFC_BAD_COMMAND;
5770 handle_table_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5772 struct ofputil_table_mod tm;
5775 error = reject_slave_controller(ofconn);
5780 error = ofputil_decode_table_mod(oh, &tm);
5785 /* XXX Actual table mod support is not implemented yet. */
5790 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
5791 OVS_EXCLUDED(ofproto_mutex)
5793 const struct ofp_header *oh = msg->data;
5797 error = ofptype_decode(&type, oh);
5801 if (oh->version >= OFP13_VERSION && ofpmsg_is_stat_request(oh)
5802 && ofpmp_more(oh)) {
5803 /* We have no buffer implementation for multipart requests.
5804 * Report overflow for requests which consists of multiple
5806 return OFPERR_OFPBRC_MULTIPART_BUFFER_OVERFLOW;
5810 /* OpenFlow requests. */
5811 case OFPTYPE_ECHO_REQUEST:
5812 return handle_echo_request(ofconn, oh);
5814 case OFPTYPE_FEATURES_REQUEST:
5815 return handle_features_request(ofconn, oh);
5817 case OFPTYPE_GET_CONFIG_REQUEST:
5818 return handle_get_config_request(ofconn, oh);
5820 case OFPTYPE_SET_CONFIG:
5821 return handle_set_config(ofconn, oh);
5823 case OFPTYPE_PACKET_OUT:
5824 return handle_packet_out(ofconn, oh);
5826 case OFPTYPE_PORT_MOD:
5827 return handle_port_mod(ofconn, oh);
5829 case OFPTYPE_FLOW_MOD:
5830 return handle_flow_mod(ofconn, oh);
5832 case OFPTYPE_GROUP_MOD:
5833 return handle_group_mod(ofconn, oh);
5835 case OFPTYPE_TABLE_MOD:
5836 return handle_table_mod(ofconn, oh);
5838 case OFPTYPE_METER_MOD:
5839 return handle_meter_mod(ofconn, oh);
5841 case OFPTYPE_BARRIER_REQUEST:
5842 return handle_barrier_request(ofconn, oh);
5844 case OFPTYPE_ROLE_REQUEST:
5845 return handle_role_request(ofconn, oh);
5847 /* OpenFlow replies. */
5848 case OFPTYPE_ECHO_REPLY:
5851 /* Nicira extension requests. */
5852 case OFPTYPE_FLOW_MOD_TABLE_ID:
5853 return handle_nxt_flow_mod_table_id(ofconn, oh);
5855 case OFPTYPE_SET_FLOW_FORMAT:
5856 return handle_nxt_set_flow_format(ofconn, oh);
5858 case OFPTYPE_SET_PACKET_IN_FORMAT:
5859 return handle_nxt_set_packet_in_format(ofconn, oh);
5861 case OFPTYPE_SET_CONTROLLER_ID:
5862 return handle_nxt_set_controller_id(ofconn, oh);
5864 case OFPTYPE_FLOW_AGE:
5865 /* Nothing to do. */
5868 case OFPTYPE_FLOW_MONITOR_CANCEL:
5869 return handle_flow_monitor_cancel(ofconn, oh);
5871 case OFPTYPE_SET_ASYNC_CONFIG:
5872 return handle_nxt_set_async_config(ofconn, oh);
5874 case OFPTYPE_GET_ASYNC_REQUEST:
5875 return handle_nxt_get_async_request(ofconn, oh);
5877 /* Statistics requests. */
5878 case OFPTYPE_DESC_STATS_REQUEST:
5879 return handle_desc_stats_request(ofconn, oh);
5881 case OFPTYPE_FLOW_STATS_REQUEST:
5882 return handle_flow_stats_request(ofconn, oh);
5884 case OFPTYPE_AGGREGATE_STATS_REQUEST:
5885 return handle_aggregate_stats_request(ofconn, oh);
5887 case OFPTYPE_TABLE_STATS_REQUEST:
5888 return handle_table_stats_request(ofconn, oh);
5890 case OFPTYPE_PORT_STATS_REQUEST:
5891 return handle_port_stats_request(ofconn, oh);
5893 case OFPTYPE_QUEUE_STATS_REQUEST:
5894 return handle_queue_stats_request(ofconn, oh);
5896 case OFPTYPE_PORT_DESC_STATS_REQUEST:
5897 return handle_port_desc_stats_request(ofconn, oh);
5899 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
5900 return handle_flow_monitor_request(ofconn, oh);
5902 case OFPTYPE_METER_STATS_REQUEST:
5903 case OFPTYPE_METER_CONFIG_STATS_REQUEST:
5904 return handle_meter_request(ofconn, oh, type);
5906 case OFPTYPE_METER_FEATURES_STATS_REQUEST:
5907 return handle_meter_features_request(ofconn, oh);
5909 case OFPTYPE_GROUP_STATS_REQUEST:
5910 return handle_group_stats_request(ofconn, oh);
5912 case OFPTYPE_GROUP_DESC_STATS_REQUEST:
5913 return handle_group_desc_stats_request(ofconn, oh);
5915 case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
5916 return handle_group_features_stats_request(ofconn, oh);
5918 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
5919 return handle_queue_get_config_request(ofconn, oh);
5923 case OFPTYPE_FEATURES_REPLY:
5924 case OFPTYPE_GET_CONFIG_REPLY:
5925 case OFPTYPE_PACKET_IN:
5926 case OFPTYPE_FLOW_REMOVED:
5927 case OFPTYPE_PORT_STATUS:
5928 case OFPTYPE_BARRIER_REPLY:
5929 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
5930 case OFPTYPE_DESC_STATS_REPLY:
5931 case OFPTYPE_FLOW_STATS_REPLY:
5932 case OFPTYPE_QUEUE_STATS_REPLY:
5933 case OFPTYPE_PORT_STATS_REPLY:
5934 case OFPTYPE_TABLE_STATS_REPLY:
5935 case OFPTYPE_AGGREGATE_STATS_REPLY:
5936 case OFPTYPE_PORT_DESC_STATS_REPLY:
5937 case OFPTYPE_ROLE_REPLY:
5938 case OFPTYPE_FLOW_MONITOR_PAUSED:
5939 case OFPTYPE_FLOW_MONITOR_RESUMED:
5940 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
5941 case OFPTYPE_GET_ASYNC_REPLY:
5942 case OFPTYPE_GROUP_STATS_REPLY:
5943 case OFPTYPE_GROUP_DESC_STATS_REPLY:
5944 case OFPTYPE_GROUP_FEATURES_STATS_REPLY:
5945 case OFPTYPE_METER_STATS_REPLY:
5946 case OFPTYPE_METER_CONFIG_STATS_REPLY:
5947 case OFPTYPE_METER_FEATURES_STATS_REPLY:
5948 case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
5949 case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
5950 case OFPTYPE_ROLE_STATUS:
5952 if (ofpmsg_is_stat_request(oh)) {
5953 return OFPERR_OFPBRC_BAD_STAT;
5955 return OFPERR_OFPBRC_BAD_TYPE;
5961 handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
5962 OVS_EXCLUDED(ofproto_mutex)
5964 int error = handle_openflow__(ofconn, ofp_msg);
5965 if (error && error != OFPROTO_POSTPONE) {
5966 ofconn_send_error(ofconn, ofp_msg->data, error);
5968 COVERAGE_INC(ofproto_recv_openflow);
5969 return error != OFPROTO_POSTPONE;
5972 /* Asynchronous operations. */
5974 /* Creates and returns a new ofopgroup that is not associated with any
5975 * OpenFlow connection.
5977 * The caller should add operations to the returned group with
5978 * ofoperation_create() and then submit it with ofopgroup_submit(). */
5979 static struct ofopgroup *
5980 ofopgroup_create_unattached(struct ofproto *ofproto)
5981 OVS_REQUIRES(ofproto_mutex)
5983 struct ofopgroup *group = xzalloc(sizeof *group);
5984 group->ofproto = ofproto;
5985 list_init(&group->ofproto_node);
5986 list_init(&group->ops);
5987 list_init(&group->ofconn_node);
5991 /* Creates and returns a new ofopgroup for 'ofproto'.
5993 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
5994 * connection. The 'request' and 'buffer_id' arguments are ignored.
5996 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
5997 * If the ofopgroup eventually fails, then the error reply will include
5998 * 'request'. If the ofopgroup eventually succeeds, then the packet with
5999 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
6001 * The caller should add operations to the returned group with
6002 * ofoperation_create() and then submit it with ofopgroup_submit(). */
6003 static struct ofopgroup *
6004 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
6005 const struct ofp_header *request, uint32_t buffer_id)
6006 OVS_REQUIRES(ofproto_mutex)
6008 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
6010 size_t request_len = ntohs(request->length);
6012 ovs_assert(ofconn_get_ofproto(ofconn) == ofproto);
6014 ofconn_add_opgroup(ofconn, &group->ofconn_node);
6015 group->ofconn = ofconn;
6016 group->request = xmemdup(request, MIN(request_len, 64));
6017 group->buffer_id = buffer_id;
6022 /* Submits 'group' for processing.
6024 * If 'group' contains no operations (e.g. none were ever added, or all of the
6025 * ones that were added completed synchronously), then it is destroyed
6026 * immediately. Otherwise it is added to the ofproto's list of pending
6029 ofopgroup_submit(struct ofopgroup *group)
6030 OVS_REQUIRES(ofproto_mutex)
6032 if (!group->n_running) {
6033 ofopgroup_complete(group);
6035 list_push_back(&group->ofproto->pending, &group->ofproto_node);
6036 group->ofproto->n_pending++;
6041 ofopgroup_complete(struct ofopgroup *group)
6042 OVS_REQUIRES(ofproto_mutex)
6044 struct ofproto *ofproto = group->ofproto;
6046 struct ofconn *abbrev_ofconn;
6047 ovs_be32 abbrev_xid;
6049 struct ofoperation *op, *next_op;
6052 ovs_assert(!group->n_running);
6055 LIST_FOR_EACH (op, group_node, &group->ops) {
6062 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
6063 LIST_FOR_EACH (op, group_node, &group->ops) {
6064 if (op->type != OFOPERATION_DELETE) {
6065 struct ofpbuf *packet;
6068 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
6071 struct rule_execute *re;
6075 ofproto_rule_ref(op->rule);
6077 re = xmalloc(sizeof *re);
6078 re->rule = op->rule;
6079 re->in_port = in_port;
6080 re->packet = packet;
6082 if (!guarded_list_push_back(&ofproto->rule_executes,
6083 &re->list_node, 1024)) {
6084 ofproto_rule_unref(op->rule);
6085 ofpbuf_delete(re->packet);
6094 if (!error && !list_is_empty(&group->ofconn_node)) {
6095 abbrev_ofconn = group->ofconn;
6096 abbrev_xid = group->request->xid;
6098 abbrev_ofconn = NULL;
6099 abbrev_xid = htonl(0);
6101 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
6102 struct rule *rule = op->rule;
6104 /* We generally want to report the change to active OpenFlow flow
6105 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
6107 - The operation failed.
6109 - The affected rule is not visible to controllers.
6111 - The operation's only effect was to update rule->modified. */
6113 || ofproto_rule_is_hidden(rule)
6114 || (op->type == OFOPERATION_MODIFY
6116 && rule->flow_cookie == op->flow_cookie))) {
6117 /* Check that we can just cast from ofoperation_type to
6118 * nx_flow_update_event. */
6119 enum nx_flow_update_event event_type;
6122 case OFOPERATION_ADD:
6123 case OFOPERATION_REPLACE:
6124 event_type = NXFME_ADDED;
6127 case OFOPERATION_DELETE:
6128 event_type = NXFME_DELETED;
6131 case OFOPERATION_MODIFY:
6132 event_type = NXFME_MODIFIED;
6139 ofmonitor_report(ofproto->connmgr, rule, event_type,
6140 op->reason, abbrev_ofconn, abbrev_xid);
6143 rule->pending = NULL;
6146 case OFOPERATION_ADD:
6150 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
6151 if (vid_mask == VLAN_VID_MASK) {
6152 if (ofproto->vlan_bitmap) {
6153 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
6154 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
6155 bitmap_set1(ofproto->vlan_bitmap, vid);
6156 ofproto->vlans_changed = true;
6159 ofproto->vlans_changed = true;
6163 oftable_remove_rule(rule);
6164 ofproto_rule_unref(rule);
6168 case OFOPERATION_DELETE:
6169 ovs_assert(!op->error);
6170 ofproto_rule_unref(rule);
6174 case OFOPERATION_MODIFY:
6175 case OFOPERATION_REPLACE:
6177 long long int now = time_msec();
6179 rule->modified = now;
6180 if (op->type == OFOPERATION_REPLACE) {
6181 rule->created = rule->used = now;
6184 ofproto_rule_change_cookie(ofproto, rule, op->flow_cookie);
6185 ovs_mutex_lock(&rule->mutex);
6186 rule->idle_timeout = op->idle_timeout;
6187 rule->hard_timeout = op->hard_timeout;
6188 ovs_mutex_unlock(&rule->mutex);
6190 struct rule_actions *old_actions;
6192 ovs_mutex_lock(&rule->mutex);
6193 old_actions = rule->actions;
6194 rule->actions = op->actions;
6195 ovs_mutex_unlock(&rule->mutex);
6198 rule_actions_unref(old_actions);
6200 rule->flags = op->flags;
6208 ofoperation_destroy(op);
6211 ofmonitor_flush(ofproto->connmgr);
6213 if (!list_is_empty(&group->ofproto_node)) {
6214 ovs_assert(ofproto->n_pending > 0);
6215 ofproto->n_pending--;
6216 list_remove(&group->ofproto_node);
6218 if (!list_is_empty(&group->ofconn_node)) {
6219 list_remove(&group->ofconn_node);
6221 ofconn_send_error(group->ofconn, group->request, error);
6223 connmgr_retry(ofproto->connmgr);
6225 free(group->request);
6229 /* Initiates a new operation on 'rule', of the specified 'type', within
6230 * 'group'. Prior to calling, 'rule' must not have any pending operation.
6232 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
6233 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
6235 * Returns the newly created ofoperation (which is also available as
6236 * rule->pending). */
6237 static struct ofoperation *
6238 ofoperation_create(struct ofopgroup *group, struct rule *rule,
6239 enum ofoperation_type type,
6240 enum ofp_flow_removed_reason reason)
6241 OVS_REQUIRES(ofproto_mutex)
6243 struct ofproto *ofproto = group->ofproto;
6244 struct ofoperation *op;
6246 ovs_assert(!rule->pending);
6248 op = rule->pending = xzalloc(sizeof *op);
6250 list_push_back(&group->ops, &op->group_node);
6253 op->reason = reason;
6254 op->flow_cookie = rule->flow_cookie;
6255 ovs_mutex_lock(&rule->mutex);
6256 op->idle_timeout = rule->idle_timeout;
6257 op->hard_timeout = rule->hard_timeout;
6258 ovs_mutex_unlock(&rule->mutex);
6259 op->flags = rule->flags;
6263 if (type == OFOPERATION_DELETE) {
6264 hmap_insert(&ofproto->deletions, &op->hmap_node,
6265 cls_rule_hash(&rule->cr, rule->table_id));
6272 ofoperation_destroy(struct ofoperation *op)
6273 OVS_REQUIRES(ofproto_mutex)
6275 struct ofopgroup *group = op->group;
6278 op->rule->pending = NULL;
6280 if (op->type == OFOPERATION_DELETE) {
6281 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
6283 list_remove(&op->group_node);
6284 rule_actions_unref(op->actions);
6288 /* Indicates that 'op' completed with status 'error', which is either 0 to
6289 * indicate success or an OpenFlow error code on failure.
6291 * If 'error' is 0, indicating success, the operation will be committed
6292 * permanently to the flow table.
6294 * If 'error' is nonzero, then generally the operation will be rolled back:
6296 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
6297 * restores the original rule. The caller must have uninitialized any
6298 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
6299 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
6300 * and 7 for the new rule, calling its ->rule_dealloc() function.
6302 * - If 'op' is a "modify flow" operation, ofproto restores the original
6305 * - 'op' must not be a "delete flow" operation. Removing a rule is not
6306 * allowed to fail. It must always succeed.
6308 * Please see the large comment in ofproto/ofproto-provider.h titled
6309 * "Asynchronous Operation Support" for more information. */
6311 ofoperation_complete(struct ofoperation *op, enum ofperr error)
6313 struct ofopgroup *group = op->group;
6315 ovs_assert(group->n_running > 0);
6316 ovs_assert(!error || op->type != OFOPERATION_DELETE);
6319 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
6320 /* This function can be called from ->rule_construct(), in which case
6321 * ofproto_mutex is held, or it can be called from ->run(), in which
6322 * case ofproto_mutex is not held. But only in the latter case can we
6323 * arrive here, so we can safely take ofproto_mutex now. */
6324 ovs_mutex_lock(&ofproto_mutex);
6325 ovs_assert(op->rule->pending == op);
6326 ofopgroup_complete(group);
6327 ovs_mutex_unlock(&ofproto_mutex);
6332 pick_datapath_id(const struct ofproto *ofproto)
6334 const struct ofport *port;
6336 port = ofproto_get_port(ofproto, OFPP_LOCAL);
6338 uint8_t ea[ETH_ADDR_LEN];
6341 error = netdev_get_etheraddr(port->netdev, ea);
6343 return eth_addr_to_uint64(ea);
6345 VLOG_WARN("%s: could not get MAC address for %s (%s)",
6346 ofproto->name, netdev_get_name(port->netdev),
6347 ovs_strerror(error));
6349 return ofproto->fallback_dpid;
6353 pick_fallback_dpid(void)
6355 uint8_t ea[ETH_ADDR_LEN];
6356 eth_addr_nicira_random(ea);
6357 return eth_addr_to_uint64(ea);
6360 /* Table overflow policy. */
6362 /* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep'
6363 * to NULL if the table is not configured to evict rules or if the table
6364 * contains no evictable rules. (Rules with a readlock on their evict rwlock,
6365 * or with no timeouts are not evictable.) */
6367 choose_rule_to_evict(struct oftable *table, struct rule **rulep)
6368 OVS_REQUIRES(ofproto_mutex)
6370 struct eviction_group *evg;
6373 if (!table->eviction_fields) {
6377 /* In the common case, the outer and inner loops here will each be entered
6380 * - The inner loop normally "return"s in its first iteration. If the
6381 * eviction group has any evictable rules, then it always returns in
6384 * - The outer loop only iterates more than once if the largest eviction
6385 * group has no evictable rules.
6387 * - The outer loop can exit only if table's 'max_flows' is all filled up
6388 * by unevictable rules. */
6389 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
6392 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
6401 /* Searches 'ofproto' for tables that have more flows than their configured
6402 * maximum and that have flow eviction enabled, and evicts as many flows as
6403 * necessary and currently feasible from them.
6405 * This triggers only when an OpenFlow table has N flows in it and then the
6406 * client configures a maximum number of flows less than N. */
6408 ofproto_evict(struct ofproto *ofproto)
6410 struct oftable *table;
6412 ovs_mutex_lock(&ofproto_mutex);
6413 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
6414 evict_rules_from_table(ofproto, table, 0);
6416 ovs_mutex_unlock(&ofproto_mutex);
6419 /* Eviction groups. */
6421 /* Returns the priority to use for an eviction_group that contains 'n_rules'
6422 * rules. The priority contains low-order random bits to ensure that eviction
6423 * groups with the same number of rules are prioritized randomly. */
6425 eviction_group_priority(size_t n_rules)
6427 uint16_t size = MIN(UINT16_MAX, n_rules);
6428 return (size << 16) | random_uint16();
6431 /* Updates 'evg', an eviction_group within 'table', following a change that
6432 * adds or removes rules in 'evg'. */
6434 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
6435 OVS_REQUIRES(ofproto_mutex)
6437 heap_change(&table->eviction_groups_by_size, &evg->size_node,
6438 eviction_group_priority(heap_count(&evg->rules)));
6441 /* Destroys 'evg', an eviction_group within 'table':
6443 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
6444 * rules themselves, just removes them from the eviction group.)
6446 * - Removes 'evg' from 'table'.
6450 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
6451 OVS_REQUIRES(ofproto_mutex)
6453 while (!heap_is_empty(&evg->rules)) {
6456 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
6457 rule->eviction_group = NULL;
6459 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
6460 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
6461 heap_destroy(&evg->rules);
6465 /* Removes 'rule' from its eviction group, if any. */
6467 eviction_group_remove_rule(struct rule *rule)
6468 OVS_REQUIRES(ofproto_mutex)
6470 if (rule->eviction_group) {
6471 struct oftable *table = &rule->ofproto->tables[rule->table_id];
6472 struct eviction_group *evg = rule->eviction_group;
6474 rule->eviction_group = NULL;
6475 heap_remove(&evg->rules, &rule->evg_node);
6476 if (heap_is_empty(&evg->rules)) {
6477 eviction_group_destroy(table, evg);
6479 eviction_group_resized(table, evg);
6484 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
6485 * returns the hash value. */
6487 eviction_group_hash_rule(struct rule *rule)
6488 OVS_REQUIRES(ofproto_mutex)
6490 struct oftable *table = &rule->ofproto->tables[rule->table_id];
6491 const struct mf_subfield *sf;
6495 hash = table->eviction_group_id_basis;
6496 miniflow_expand(&rule->cr.match.flow, &flow);
6497 for (sf = table->eviction_fields;
6498 sf < &table->eviction_fields[table->n_eviction_fields];
6501 if (mf_are_prereqs_ok(sf->field, &flow)) {
6502 union mf_value value;
6504 mf_get_value(sf->field, &flow, &value);
6506 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
6508 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
6509 unsigned int start = sf->ofs + sf->n_bits;
6510 bitwise_zero(&value, sf->field->n_bytes, start,
6511 sf->field->n_bytes * 8 - start);
6513 hash = hash_bytes(&value, sf->field->n_bytes, hash);
6515 hash = hash_int(hash, 0);
6522 /* Returns an eviction group within 'table' with the given 'id', creating one
6524 static struct eviction_group *
6525 eviction_group_find(struct oftable *table, uint32_t id)
6526 OVS_REQUIRES(ofproto_mutex)
6528 struct eviction_group *evg;
6530 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
6534 evg = xmalloc(sizeof *evg);
6535 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
6536 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
6537 eviction_group_priority(0));
6538 heap_init(&evg->rules);
6543 /* Returns an eviction priority for 'rule'. The return value should be
6544 * interpreted so that higher priorities make a rule more attractive candidates
6547 rule_eviction_priority(struct rule *rule)
6548 OVS_REQUIRES(ofproto_mutex)
6550 long long int hard_expiration;
6551 long long int idle_expiration;
6552 long long int expiration;
6553 uint32_t expiration_offset;
6555 /* Calculate time of expiration. */
6556 ovs_mutex_lock(&rule->mutex);
6557 hard_expiration = (rule->hard_timeout
6558 ? rule->modified + rule->hard_timeout * 1000
6560 idle_expiration = (rule->idle_timeout
6561 ? rule->used + rule->idle_timeout * 1000
6563 expiration = MIN(hard_expiration, idle_expiration);
6564 ovs_mutex_unlock(&rule->mutex);
6565 if (expiration == LLONG_MAX) {
6569 /* Calculate the time of expiration as a number of (approximate) seconds
6570 * after program startup.
6572 * This should work OK for program runs that last UINT32_MAX seconds or
6573 * less. Therefore, please restart OVS at least once every 136 years. */
6574 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
6576 /* Invert the expiration offset because we're using a max-heap. */
6577 return UINT32_MAX - expiration_offset;
6580 /* Adds 'rule' to an appropriate eviction group for its oftable's
6581 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
6582 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
6585 * The caller must ensure that 'rule' is not already in an eviction group. */
6587 eviction_group_add_rule(struct rule *rule)
6588 OVS_REQUIRES(ofproto_mutex)
6590 struct ofproto *ofproto = rule->ofproto;
6591 struct oftable *table = &ofproto->tables[rule->table_id];
6594 ovs_mutex_lock(&rule->mutex);
6595 has_timeout = rule->hard_timeout || rule->idle_timeout;
6596 ovs_mutex_unlock(&rule->mutex);
6598 if (table->eviction_fields && has_timeout) {
6599 struct eviction_group *evg;
6601 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
6603 rule->eviction_group = evg;
6604 heap_insert(&evg->rules, &rule->evg_node,
6605 rule_eviction_priority(rule));
6606 eviction_group_resized(table, evg);
6612 /* Initializes 'table'. */
6614 oftable_init(struct oftable *table)
6616 memset(table, 0, sizeof *table);
6617 classifier_init(&table->cls);
6618 table->max_flows = UINT_MAX;
6621 /* Destroys 'table', including its classifier and eviction groups.
6623 * The caller is responsible for freeing 'table' itself. */
6625 oftable_destroy(struct oftable *table)
6627 ovs_rwlock_rdlock(&table->cls.rwlock);
6628 ovs_assert(classifier_is_empty(&table->cls));
6629 ovs_rwlock_unlock(&table->cls.rwlock);
6630 oftable_disable_eviction(table);
6631 classifier_destroy(&table->cls);
6635 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
6636 * string, then 'table' will use its default name.
6638 * This only affects the name exposed for a table exposed through the OpenFlow
6639 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
6641 oftable_set_name(struct oftable *table, const char *name)
6643 if (name && name[0]) {
6644 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
6645 if (!table->name || strncmp(name, table->name, len)) {
6647 table->name = xmemdup0(name, len);
6655 /* oftables support a choice of two policies when adding a rule would cause the
6656 * number of flows in the table to exceed the configured maximum number: either
6657 * they can refuse to add the new flow or they can evict some existing flow.
6658 * This function configures the former policy on 'table'. */
6660 oftable_disable_eviction(struct oftable *table)
6661 OVS_REQUIRES(ofproto_mutex)
6663 if (table->eviction_fields) {
6664 struct eviction_group *evg, *next;
6666 HMAP_FOR_EACH_SAFE (evg, next, id_node,
6667 &table->eviction_groups_by_id) {
6668 eviction_group_destroy(table, evg);
6670 hmap_destroy(&table->eviction_groups_by_id);
6671 heap_destroy(&table->eviction_groups_by_size);
6673 free(table->eviction_fields);
6674 table->eviction_fields = NULL;
6675 table->n_eviction_fields = 0;
6679 /* oftables support a choice of two policies when adding a rule would cause the
6680 * number of flows in the table to exceed the configured maximum number: either
6681 * they can refuse to add the new flow or they can evict some existing flow.
6682 * This function configures the latter policy on 'table', with fairness based
6683 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
6684 * 'n_fields' as 0 disables fairness.) */
6686 oftable_enable_eviction(struct oftable *table,
6687 const struct mf_subfield *fields, size_t n_fields)
6688 OVS_REQUIRES(ofproto_mutex)
6690 struct cls_cursor cursor;
6693 if (table->eviction_fields
6694 && n_fields == table->n_eviction_fields
6696 || !memcmp(fields, table->eviction_fields,
6697 n_fields * sizeof *fields))) {
6702 oftable_disable_eviction(table);
6704 table->n_eviction_fields = n_fields;
6705 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
6707 table->eviction_group_id_basis = random_uint32();
6708 hmap_init(&table->eviction_groups_by_id);
6709 heap_init(&table->eviction_groups_by_size);
6711 ovs_rwlock_rdlock(&table->cls.rwlock);
6712 cls_cursor_init(&cursor, &table->cls, NULL);
6713 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
6714 eviction_group_add_rule(rule);
6716 ovs_rwlock_unlock(&table->cls.rwlock);
6719 /* Removes 'rule' from the oftable that contains it. */
6721 oftable_remove_rule__(struct ofproto *ofproto, struct rule *rule)
6722 OVS_REQUIRES(ofproto_mutex)
6724 struct classifier *cls = &ofproto->tables[rule->table_id].cls;
6726 ovs_rwlock_wrlock(&cls->rwlock);
6727 classifier_remove(cls, CONST_CAST(struct cls_rule *, &rule->cr));
6728 ovs_rwlock_unlock(&cls->rwlock);
6730 cookies_remove(ofproto, rule);
6732 eviction_group_remove_rule(rule);
6733 if (!list_is_empty(&rule->expirable)) {
6734 list_remove(&rule->expirable);
6736 if (!list_is_empty(&rule->meter_list_node)) {
6737 list_remove(&rule->meter_list_node);
6738 list_init(&rule->meter_list_node);
6743 oftable_remove_rule(struct rule *rule)
6744 OVS_REQUIRES(ofproto_mutex)
6746 oftable_remove_rule__(rule->ofproto, rule);
6749 /* Inserts 'rule' into its oftable, which must not already contain any rule for
6750 * the same cls_rule. */
6752 oftable_insert_rule(struct rule *rule)
6753 OVS_REQUIRES(ofproto_mutex)
6755 struct ofproto *ofproto = rule->ofproto;
6756 struct oftable *table = &ofproto->tables[rule->table_id];
6759 ovs_mutex_lock(&rule->mutex);
6760 may_expire = rule->hard_timeout || rule->idle_timeout;
6761 ovs_mutex_unlock(&rule->mutex);
6764 list_insert(&ofproto->expirable, &rule->expirable);
6767 cookies_insert(ofproto, rule);
6769 if (rule->actions->provider_meter_id != UINT32_MAX) {
6770 uint32_t meter_id = ofpacts_get_meter(rule->actions->ofpacts,
6771 rule->actions->ofpacts_len);
6772 struct meter *meter = ofproto->meters[meter_id];
6773 list_insert(&meter->rules, &rule->meter_list_node);
6775 ovs_rwlock_wrlock(&table->cls.rwlock);
6776 classifier_insert(&table->cls, CONST_CAST(struct cls_rule *, &rule->cr));
6777 ovs_rwlock_unlock(&table->cls.rwlock);
6778 eviction_group_add_rule(rule);
6781 /* unixctl commands. */
6784 ofproto_lookup(const char *name)
6786 struct ofproto *ofproto;
6788 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
6790 if (!strcmp(ofproto->name, name)) {
6798 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
6799 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6801 struct ofproto *ofproto;
6805 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
6806 ds_put_format(&results, "%s\n", ofproto->name);
6808 unixctl_command_reply(conn, ds_cstr(&results));
6809 ds_destroy(&results);
6813 ofproto_unixctl_init(void)
6815 static bool registered;
6821 unixctl_command_register("ofproto/list", "", 0, 0,
6822 ofproto_unixctl_list, NULL);
6825 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6827 * This is deprecated. It is only for compatibility with broken device drivers
6828 * in old versions of Linux that do not properly support VLANs when VLAN
6829 * devices are not used. When broken device drivers are no longer in
6830 * widespread use, we will delete these interfaces. */
6832 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
6833 * (exactly) by an OpenFlow rule in 'ofproto'. */
6835 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
6837 const struct oftable *oftable;
6839 free(ofproto->vlan_bitmap);
6840 ofproto->vlan_bitmap = bitmap_allocate(4096);
6841 ofproto->vlans_changed = false;
6843 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
6844 const struct cls_subtable *table;
6846 ovs_rwlock_rdlock(&oftable->cls.rwlock);
6847 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.subtables) {
6848 if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
6849 const struct cls_rule *rule;
6851 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
6852 uint16_t vid = miniflow_get_vid(&rule->match.flow);
6853 bitmap_set1(vlan_bitmap, vid);
6854 bitmap_set1(ofproto->vlan_bitmap, vid);
6858 ovs_rwlock_unlock(&oftable->cls.rwlock);
6862 /* Returns true if new VLANs have come into use by the flow table since the
6863 * last call to ofproto_get_vlan_usage().
6865 * We don't track when old VLANs stop being used. */
6867 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
6869 return ofproto->vlans_changed;
6872 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
6873 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
6874 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
6875 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
6876 * then the VLAN device is un-enslaved. */
6878 ofproto_port_set_realdev(struct ofproto *ofproto, ofp_port_t vlandev_ofp_port,
6879 ofp_port_t realdev_ofp_port, int vid)
6881 struct ofport *ofport;
6884 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
6886 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
6888 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
6889 ofproto->name, vlandev_ofp_port);
6893 if (!ofproto->ofproto_class->set_realdev) {
6894 if (!vlandev_ofp_port) {
6897 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
6901 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
6903 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
6904 ofproto->name, vlandev_ofp_port,
6905 netdev_get_name(ofport->netdev), ovs_strerror(error));