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);
535 error = ofproto->ofproto_class->construct(ofproto);
537 VLOG_ERR("failed to open datapath %s: %s",
538 datapath_name, ovs_strerror(error));
539 ofproto_destroy__(ofproto);
543 /* Check that hidden tables, if any, are at the end. */
544 ovs_assert(ofproto->n_tables);
545 for (i = 0; i + 1 < ofproto->n_tables; i++) {
546 enum oftable_flags flags = ofproto->tables[i].flags;
547 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
549 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
552 ofproto->datapath_id = pick_datapath_id(ofproto);
555 /* Initialize meters table. */
556 if (ofproto->ofproto_class->meter_get_features) {
557 ofproto->ofproto_class->meter_get_features(ofproto,
558 &ofproto->meter_features);
560 memset(&ofproto->meter_features, 0, sizeof ofproto->meter_features);
562 ofproto->meters = xzalloc((ofproto->meter_features.max_meters + 1)
563 * sizeof(struct meter *));
569 /* Must be called (only) by an ofproto implementation in its constructor
570 * function. See the large comment on 'construct' in struct ofproto_class for
573 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
575 struct oftable *table;
577 ovs_assert(!ofproto->n_tables);
578 ovs_assert(n_tables >= 1 && n_tables <= 255);
580 ofproto->n_tables = n_tables;
581 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
582 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
587 /* To be optionally called (only) by an ofproto implementation in its
588 * constructor function. See the large comment on 'construct' in struct
589 * ofproto_class for details.
591 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
592 * will then ensure that actions passed into the ofproto implementation will
593 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
594 * function is not called, there will be no such restriction.
596 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
597 * the 'max_ports' restriction. */
599 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
601 ovs_assert(max_ports <= ofp_to_u16(OFPP_MAX));
602 ofproto->max_ports = max_ports;
606 ofproto_get_datapath_id(const struct ofproto *ofproto)
608 return ofproto->datapath_id;
612 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
614 uint64_t old_dpid = p->datapath_id;
615 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
616 if (p->datapath_id != old_dpid) {
617 /* Force all active connections to reconnect, since there is no way to
618 * notify a controller that the datapath ID has changed. */
619 ofproto_reconnect_controllers(p);
624 ofproto_set_controllers(struct ofproto *p,
625 const struct ofproto_controller *controllers,
626 size_t n_controllers, uint32_t allowed_versions)
628 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
633 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
635 connmgr_set_fail_mode(p->connmgr, fail_mode);
638 /* Drops the connections between 'ofproto' and all of its controllers, forcing
639 * them to reconnect. */
641 ofproto_reconnect_controllers(struct ofproto *ofproto)
643 connmgr_reconnect(ofproto->connmgr);
646 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
647 * in-band control should guarantee access, in the same way that in-band
648 * control guarantees access to OpenFlow controllers. */
650 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
651 const struct sockaddr_in *extras, size_t n)
653 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
656 /* Sets the OpenFlow queue used by flows set up by in-band control on
657 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
658 * flows will use the default queue. */
660 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
662 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
665 /* Sets the number of flows at which eviction from the kernel flow table
668 ofproto_set_flow_eviction_threshold(unsigned threshold)
670 flow_eviction_threshold = MAX(OFPROTO_FLOW_EVICTION_THRESHOLD_MIN,
674 /* Sets the path for handling flow misses. */
676 ofproto_set_flow_miss_model(unsigned model)
678 flow_miss_model = model;
681 /* If forward_bpdu is true, the NORMAL action will forward frames with
682 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
683 * the NORMAL action will drop these frames. */
685 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
687 bool old_val = ofproto->forward_bpdu;
688 ofproto->forward_bpdu = forward_bpdu;
689 if (old_val != ofproto->forward_bpdu) {
690 if (ofproto->ofproto_class->forward_bpdu_changed) {
691 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
696 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
697 * 'idle_time', in seconds, and the maximum number of MAC table entries to
700 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
703 if (ofproto->ofproto_class->set_mac_table_config) {
704 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
709 /* Sets number of upcall handler threads. The default is
710 * (number of online cores - 2). */
712 ofproto_set_n_handler_threads(unsigned limit)
715 n_handler_threads = limit;
717 int n_proc = sysconf(_SC_NPROCESSORS_ONLN);
718 n_handler_threads = n_proc > 2 ? n_proc - 2 : 1;
723 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
726 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
730 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
732 return connmgr_set_snoops(ofproto->connmgr, snoops);
736 ofproto_set_netflow(struct ofproto *ofproto,
737 const struct netflow_options *nf_options)
739 if (nf_options && sset_is_empty(&nf_options->collectors)) {
743 if (ofproto->ofproto_class->set_netflow) {
744 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
746 return nf_options ? EOPNOTSUPP : 0;
751 ofproto_set_sflow(struct ofproto *ofproto,
752 const struct ofproto_sflow_options *oso)
754 if (oso && sset_is_empty(&oso->targets)) {
758 if (ofproto->ofproto_class->set_sflow) {
759 return ofproto->ofproto_class->set_sflow(ofproto, oso);
761 return oso ? EOPNOTSUPP : 0;
766 ofproto_set_ipfix(struct ofproto *ofproto,
767 const struct ofproto_ipfix_bridge_exporter_options *bo,
768 const struct ofproto_ipfix_flow_exporter_options *fo,
771 if (ofproto->ofproto_class->set_ipfix) {
772 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
774 return (bo || fo) ? EOPNOTSUPP : 0;
779 ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
781 flow_restore_wait = flow_restore_wait_db;
785 ofproto_get_flow_restore_wait(void)
787 return flow_restore_wait;
791 /* Spanning Tree Protocol (STP) configuration. */
793 /* Configures STP on 'ofproto' using the settings defined in 's'. If
794 * 's' is NULL, disables STP.
796 * Returns 0 if successful, otherwise a positive errno value. */
798 ofproto_set_stp(struct ofproto *ofproto,
799 const struct ofproto_stp_settings *s)
801 return (ofproto->ofproto_class->set_stp
802 ? ofproto->ofproto_class->set_stp(ofproto, s)
806 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
807 * 'enabled' member of 's' is false, then the other members are not
810 * Returns 0 if successful, otherwise a positive errno value. */
812 ofproto_get_stp_status(struct ofproto *ofproto,
813 struct ofproto_stp_status *s)
815 return (ofproto->ofproto_class->get_stp_status
816 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
820 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
821 * in 's'. The caller is responsible for assigning STP port numbers
822 * (using the 'port_num' member in the range of 1 through 255, inclusive)
823 * and ensuring there are no duplicates. If the 's' is NULL, then STP
824 * is disabled on the port.
826 * Returns 0 if successful, otherwise a positive errno value.*/
828 ofproto_port_set_stp(struct ofproto *ofproto, ofp_port_t ofp_port,
829 const struct ofproto_port_stp_settings *s)
831 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
833 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
834 ofproto->name, ofp_port);
838 return (ofproto->ofproto_class->set_stp_port
839 ? ofproto->ofproto_class->set_stp_port(ofport, s)
843 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
844 * 's'. If the 'enabled' member in 's' is false, then the other members
845 * are not meaningful.
847 * Returns 0 if successful, otherwise a positive errno value.*/
849 ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
850 struct ofproto_port_stp_status *s)
852 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
854 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
855 "port %"PRIu16, ofproto->name, ofp_port);
859 return (ofproto->ofproto_class->get_stp_port_status
860 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
864 /* Queue DSCP configuration. */
866 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
867 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
868 * to implement QoS. Instead, it is used to implement features which require
869 * knowledge of what queues exist on a port, and some basic information about
872 * Returns 0 if successful, otherwise a positive errno value. */
874 ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port,
875 const struct ofproto_port_queue *queues,
878 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
881 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
882 ofproto->name, ofp_port);
886 return (ofproto->ofproto_class->set_queues
887 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
891 /* Connectivity Fault Management configuration. */
893 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
895 ofproto_port_clear_cfm(struct ofproto *ofproto, ofp_port_t ofp_port)
897 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
898 if (ofport && ofproto->ofproto_class->set_cfm) {
899 ofproto->ofproto_class->set_cfm(ofport, NULL);
903 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
904 * basic configuration from the configuration members in 'cfm', and the remote
905 * maintenance point ID from remote_mpid. Ignores the statistics members of
908 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
910 ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port,
911 const struct cfm_settings *s)
913 struct ofport *ofport;
916 ofport = ofproto_get_port(ofproto, ofp_port);
918 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
919 ofproto->name, ofp_port);
923 /* XXX: For configuration simplicity, we only support one remote_mpid
924 * outside of the CFM module. It's not clear if this is the correct long
925 * term solution or not. */
926 error = (ofproto->ofproto_class->set_cfm
927 ? ofproto->ofproto_class->set_cfm(ofport, s)
930 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
931 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
932 ovs_strerror(error));
936 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
937 * 'ofproto' does not have a port 'ofp_port'. */
939 ofproto_port_set_bfd(struct ofproto *ofproto, ofp_port_t ofp_port,
940 const struct smap *cfg)
942 struct ofport *ofport;
945 ofport = ofproto_get_port(ofproto, ofp_port);
947 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
948 ofproto->name, ofp_port);
952 error = (ofproto->ofproto_class->set_bfd
953 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
956 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
957 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
958 ovs_strerror(error));
962 /* Populates 'status' with key value pairs indicating the status of the BFD
963 * session on 'ofp_port'. This information is intended to be populated in the
964 * OVS database. Has no effect if 'ofp_port' is not na OpenFlow port in
967 ofproto_port_get_bfd_status(struct ofproto *ofproto, ofp_port_t ofp_port,
970 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
971 return (ofport && ofproto->ofproto_class->get_bfd_status
972 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
976 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
977 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
978 * 0 if LACP partner information is not current (generally indicating a
979 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
981 ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port)
983 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
984 return (ofport && ofproto->ofproto_class->port_is_lacp_current
985 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
991 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
992 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
993 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
994 * configuration plus, if there is more than one slave, a bonding
997 * If 'aux' is already registered then this function updates its configuration
998 * to 's'. Otherwise, this function registers a new bundle.
1000 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
1003 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
1004 const struct ofproto_bundle_settings *s)
1006 return (ofproto->ofproto_class->bundle_set
1007 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
1011 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
1012 * If no such bundle has been registered, this has no effect. */
1014 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
1016 return ofproto_bundle_register(ofproto, aux, NULL);
1020 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
1021 * If 'aux' is already registered then this function updates its configuration
1022 * to 's'. Otherwise, this function registers a new mirror. */
1024 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
1025 const struct ofproto_mirror_settings *s)
1027 return (ofproto->ofproto_class->mirror_set
1028 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
1032 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
1033 * If no mirror has been registered, this has no effect. */
1035 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
1037 return ofproto_mirror_register(ofproto, aux, NULL);
1040 /* Retrieves statistics from mirror associated with client data pointer
1041 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
1042 * 'bytes', respectively. If a particular counters is not supported,
1043 * the appropriate argument is set to UINT64_MAX. */
1045 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
1046 uint64_t *packets, uint64_t *bytes)
1048 if (!ofproto->ofproto_class->mirror_get_stats) {
1049 *packets = *bytes = UINT64_MAX;
1053 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
1057 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
1058 * which all packets are flooded, instead of using MAC learning. If
1059 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1061 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
1064 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
1066 return (ofproto->ofproto_class->set_flood_vlans
1067 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
1071 /* Returns true if 'aux' is a registered bundle that is currently in use as the
1072 * output for a mirror. */
1074 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
1076 return (ofproto->ofproto_class->is_mirror_output_bundle
1077 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
1081 /* Configuration of OpenFlow tables. */
1083 /* Returns the number of OpenFlow tables in 'ofproto'. */
1085 ofproto_get_n_tables(const struct ofproto *ofproto)
1087 return ofproto->n_tables;
1090 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
1091 * settings from 's'. 'table_id' must be in the range 0 through the number of
1092 * OpenFlow tables in 'ofproto' minus 1, inclusive.
1094 * For read-only tables, only the name may be configured. */
1096 ofproto_configure_table(struct ofproto *ofproto, int table_id,
1097 const struct ofproto_table_settings *s)
1099 struct oftable *table;
1101 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
1102 table = &ofproto->tables[table_id];
1104 oftable_set_name(table, s->name);
1106 if (table->flags & OFTABLE_READONLY) {
1111 oftable_enable_eviction(table, s->groups, s->n_groups);
1113 oftable_disable_eviction(table);
1116 table->max_flows = s->max_flows;
1117 ovs_rwlock_rdlock(&table->cls.rwlock);
1118 if (classifier_count(&table->cls) > table->max_flows
1119 && table->eviction_fields) {
1120 /* 'table' contains more flows than allowed. We might not be able to
1121 * evict them right away because of the asynchronous nature of flow
1122 * table changes. Schedule eviction for later. */
1123 switch (ofproto->state) {
1125 ofproto->state = S_EVICT;
1129 /* We're already deleting flows, nothing more to do. */
1133 ovs_rwlock_unlock(&table->cls.rwlock);
1137 ofproto_has_snoops(const struct ofproto *ofproto)
1139 return connmgr_has_snoops(ofproto->connmgr);
1143 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1145 connmgr_get_snoops(ofproto->connmgr, snoops);
1149 ofproto_rule_delete__(struct ofproto *ofproto, struct rule *rule,
1151 OVS_REQUIRES(ofproto_mutex)
1153 struct ofopgroup *group;
1155 ovs_assert(!rule->pending);
1157 group = ofopgroup_create_unattached(ofproto);
1158 delete_flow__(rule, group, reason);
1159 ofopgroup_submit(group);
1162 /* Deletes 'rule' from 'cls' within 'ofproto'.
1164 * Within an ofproto implementation, this function allows an ofproto
1165 * implementation to destroy any rules that remain when its ->destruct()
1166 * function is called. This function is not suitable for use elsewhere in an
1167 * ofproto implementation.
1169 * This function implements steps 4.4 and 4.5 in the section titled "Rule Life
1170 * Cycle" in ofproto-provider.h. */
1172 ofproto_rule_delete(struct ofproto *ofproto, struct rule *rule)
1173 OVS_EXCLUDED(ofproto_mutex)
1175 struct ofopgroup *group;
1177 ovs_mutex_lock(&ofproto_mutex);
1178 ovs_assert(!rule->pending);
1180 group = ofopgroup_create_unattached(ofproto);
1181 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1182 oftable_remove_rule__(ofproto, rule);
1183 ofproto->ofproto_class->rule_delete(rule);
1184 ofopgroup_submit(group);
1186 ovs_mutex_unlock(&ofproto_mutex);
1190 ofproto_flush__(struct ofproto *ofproto)
1191 OVS_EXCLUDED(ofproto_mutex)
1193 struct oftable *table;
1195 if (ofproto->ofproto_class->flush) {
1196 ofproto->ofproto_class->flush(ofproto);
1199 ovs_mutex_lock(&ofproto_mutex);
1200 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1201 struct rule *rule, *next_rule;
1202 struct cls_cursor cursor;
1204 if (table->flags & OFTABLE_HIDDEN) {
1208 ovs_rwlock_rdlock(&table->cls.rwlock);
1209 cls_cursor_init(&cursor, &table->cls, NULL);
1210 ovs_rwlock_unlock(&table->cls.rwlock);
1211 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1212 if (!rule->pending) {
1213 ofproto_rule_delete__(ofproto, rule, OFPRR_DELETE);
1217 ovs_mutex_unlock(&ofproto_mutex);
1220 static void delete_group(struct ofproto *ofproto, uint32_t group_id);
1223 ofproto_destroy__(struct ofproto *ofproto)
1224 OVS_EXCLUDED(ofproto_mutex)
1226 struct oftable *table;
1228 ovs_assert(list_is_empty(&ofproto->pending));
1230 destroy_rule_executes(ofproto);
1231 guarded_list_destroy(&ofproto->rule_executes);
1233 delete_group(ofproto, OFPG_ALL);
1234 ovs_rwlock_destroy(&ofproto->groups_rwlock);
1235 hmap_destroy(&ofproto->groups);
1237 connmgr_destroy(ofproto->connmgr);
1239 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1240 free(ofproto->name);
1241 free(ofproto->type);
1242 free(ofproto->mfr_desc);
1243 free(ofproto->hw_desc);
1244 free(ofproto->sw_desc);
1245 free(ofproto->serial_desc);
1246 free(ofproto->dp_desc);
1247 hmap_destroy(&ofproto->ports);
1248 hmap_destroy(&ofproto->ofport_usage);
1249 shash_destroy(&ofproto->port_by_name);
1250 simap_destroy(&ofproto->ofp_requests);
1252 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1253 oftable_destroy(table);
1255 free(ofproto->tables);
1257 hmap_destroy(&ofproto->deletions);
1259 free(ofproto->vlan_bitmap);
1261 ofproto->ofproto_class->dealloc(ofproto);
1265 ofproto_destroy(struct ofproto *p)
1266 OVS_EXCLUDED(ofproto_mutex)
1268 struct ofport *ofport, *next_ofport;
1269 struct ofport_usage *usage, *next_usage;
1276 meter_delete(p, 1, p->meter_features.max_meters);
1277 p->meter_features.max_meters = 0;
1283 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1284 ofport_destroy(ofport);
1287 HMAP_FOR_EACH_SAFE (usage, next_usage, hmap_node, &p->ofport_usage) {
1288 hmap_remove(&p->ofport_usage, &usage->hmap_node);
1292 p->ofproto_class->destruct(p);
1293 ofproto_destroy__(p);
1296 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1297 * kernel datapath, for example, this destroys the datapath in the kernel, and
1298 * with the netdev-based datapath, it tears down the data structures that
1299 * represent the datapath.
1301 * The datapath should not be currently open as an ofproto. */
1303 ofproto_delete(const char *name, const char *type)
1305 const struct ofproto_class *class = ofproto_class_find__(type);
1306 return (!class ? EAFNOSUPPORT
1307 : !class->del ? EACCES
1308 : class->del(type, name));
1312 process_port_change(struct ofproto *ofproto, int error, char *devname)
1314 if (error == ENOBUFS) {
1315 reinit_ports(ofproto);
1316 } else if (!error) {
1317 update_port(ofproto, devname);
1323 ofproto_type_run(const char *datapath_type)
1325 const struct ofproto_class *class;
1328 datapath_type = ofproto_normalize_type(datapath_type);
1329 class = ofproto_class_find__(datapath_type);
1331 error = class->type_run ? class->type_run(datapath_type) : 0;
1332 if (error && error != EAGAIN) {
1333 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1334 datapath_type, ovs_strerror(error));
1340 ofproto_type_run_fast(const char *datapath_type)
1342 const struct ofproto_class *class;
1345 datapath_type = ofproto_normalize_type(datapath_type);
1346 class = ofproto_class_find__(datapath_type);
1348 error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
1349 if (error && error != EAGAIN) {
1350 VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
1351 datapath_type, ovs_strerror(error));
1357 ofproto_type_wait(const char *datapath_type)
1359 const struct ofproto_class *class;
1361 datapath_type = ofproto_normalize_type(datapath_type);
1362 class = ofproto_class_find__(datapath_type);
1364 if (class->type_wait) {
1365 class->type_wait(datapath_type);
1370 any_pending_ops(const struct ofproto *p)
1371 OVS_EXCLUDED(ofproto_mutex)
1375 ovs_mutex_lock(&ofproto_mutex);
1376 b = !list_is_empty(&p->pending);
1377 ovs_mutex_unlock(&ofproto_mutex);
1383 ofproto_run(struct ofproto *p)
1385 struct sset changed_netdevs;
1386 const char *changed_netdev;
1387 struct ofport *ofport;
1390 error = p->ofproto_class->run(p);
1391 if (error && error != EAGAIN) {
1392 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, ovs_strerror(error));
1395 run_rule_executes(p);
1397 /* Restore the eviction group heap invariant occasionally. */
1398 if (p->eviction_group_timer < time_msec()) {
1401 p->eviction_group_timer = time_msec() + 1000;
1403 for (i = 0; i < p->n_tables; i++) {
1404 struct oftable *table = &p->tables[i];
1405 struct eviction_group *evg;
1406 struct cls_cursor cursor;
1409 if (!table->eviction_fields) {
1413 ovs_mutex_lock(&ofproto_mutex);
1414 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
1415 heap_rebuild(&evg->rules);
1418 ovs_rwlock_rdlock(&table->cls.rwlock);
1419 cls_cursor_init(&cursor, &table->cls, NULL);
1420 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1421 if (!rule->eviction_group
1422 && (rule->idle_timeout || rule->hard_timeout)) {
1423 eviction_group_add_rule(rule);
1426 ovs_rwlock_unlock(&table->cls.rwlock);
1427 ovs_mutex_unlock(&ofproto_mutex);
1431 if (p->ofproto_class->port_poll) {
1434 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1435 process_port_change(p, error, devname);
1439 /* Update OpenFlow port status for any port whose netdev has changed.
1441 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1442 * destroyed, so it's not safe to update ports directly from the
1443 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1444 * need this two-phase approach. */
1445 sset_init(&changed_netdevs);
1446 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1447 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1448 if (ofport->change_seq != change_seq) {
1449 ofport->change_seq = change_seq;
1450 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1453 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1454 update_port(p, changed_netdev);
1456 sset_destroy(&changed_netdevs);
1460 connmgr_run(p->connmgr, handle_openflow);
1464 connmgr_run(p->connmgr, NULL);
1466 if (!any_pending_ops(p)) {
1467 p->state = S_OPENFLOW;
1472 connmgr_run(p->connmgr, NULL);
1474 if (!any_pending_ops(p)) {
1475 connmgr_flushed(p->connmgr);
1476 p->state = S_OPENFLOW;
1484 if (time_msec() >= p->next_op_report) {
1485 long long int ago = (time_msec() - p->first_op) / 1000;
1486 long long int interval = (p->last_op - p->first_op) / 1000;
1490 ds_put_format(&s, "%d flow_mods ",
1491 p->n_add + p->n_delete + p->n_modify);
1492 if (interval == ago) {
1493 ds_put_format(&s, "in the last %lld s", ago);
1494 } else if (interval) {
1495 ds_put_format(&s, "in the %lld s starting %lld s ago",
1498 ds_put_format(&s, "%lld s ago", ago);
1501 ds_put_cstr(&s, " (");
1503 ds_put_format(&s, "%d adds, ", p->n_add);
1506 ds_put_format(&s, "%d deletes, ", p->n_delete);
1509 ds_put_format(&s, "%d modifications, ", p->n_modify);
1512 ds_put_char(&s, ')');
1514 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1517 p->n_add = p->n_delete = p->n_modify = 0;
1518 p->next_op_report = LLONG_MAX;
1524 /* Performs periodic activity required by 'ofproto' that needs to be done
1525 * with the least possible latency.
1527 * It makes sense to call this function a couple of times per poll loop, to
1528 * provide a significant performance boost on some benchmarks with the
1529 * ofproto-dpif implementation. */
1531 ofproto_run_fast(struct ofproto *p)
1535 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1536 if (error && error != EAGAIN) {
1537 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1538 p->name, ovs_strerror(error));
1544 ofproto_wait(struct ofproto *p)
1546 struct ofport *ofport;
1548 p->ofproto_class->wait(p);
1549 if (p->ofproto_class->port_poll_wait) {
1550 p->ofproto_class->port_poll_wait(p);
1553 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1554 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1555 poll_immediate_wake();
1561 connmgr_wait(p->connmgr, true);
1566 connmgr_wait(p->connmgr, false);
1567 if (!any_pending_ops(p)) {
1568 poll_immediate_wake();
1575 ofproto_is_alive(const struct ofproto *p)
1577 return connmgr_has_controllers(p->connmgr);
1580 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1581 * memory_report(). */
1583 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1585 const struct oftable *table;
1586 unsigned int n_rules;
1588 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1590 ovs_mutex_lock(&ofproto_mutex);
1591 simap_increase(usage, "ops",
1592 ofproto->n_pending + hmap_count(&ofproto->deletions));
1593 ovs_mutex_unlock(&ofproto_mutex);
1596 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1597 ovs_rwlock_rdlock(&table->cls.rwlock);
1598 n_rules += classifier_count(&table->cls);
1599 ovs_rwlock_unlock(&table->cls.rwlock);
1601 simap_increase(usage, "rules", n_rules);
1603 if (ofproto->ofproto_class->get_memory_usage) {
1604 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1607 connmgr_get_memory_usage(ofproto->connmgr, usage);
1611 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1614 connmgr_get_controller_info(ofproto->connmgr, info);
1618 ofproto_free_ofproto_controller_info(struct shash *info)
1620 connmgr_free_controller_info(info);
1623 /* Makes a deep copy of 'old' into 'port'. */
1625 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1627 port->name = xstrdup(old->name);
1628 port->type = xstrdup(old->type);
1629 port->ofp_port = old->ofp_port;
1632 /* Frees memory allocated to members of 'ofproto_port'.
1634 * Do not call this function on an ofproto_port obtained from
1635 * ofproto_port_dump_next(): that function retains ownership of the data in the
1638 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1640 free(ofproto_port->name);
1641 free(ofproto_port->type);
1644 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1646 * This function provides no status indication. An error status for the entire
1647 * dump operation is provided when it is completed by calling
1648 * ofproto_port_dump_done().
1651 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1652 const struct ofproto *ofproto)
1654 dump->ofproto = ofproto;
1655 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1659 /* Attempts to retrieve another port from 'dump', which must have been created
1660 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1661 * 'port' and returns true. On failure, returns false.
1663 * Failure might indicate an actual error or merely that the last port has been
1664 * dumped. An error status for the entire dump operation is provided when it
1665 * is completed by calling ofproto_port_dump_done().
1667 * The ofproto owns the data stored in 'port'. It will remain valid until at
1668 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1669 * ofproto_port_dump_done(). */
1671 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1672 struct ofproto_port *port)
1674 const struct ofproto *ofproto = dump->ofproto;
1680 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1683 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1689 /* Completes port table dump operation 'dump', which must have been created
1690 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1691 * error-free, otherwise a positive errno value describing the problem. */
1693 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1695 const struct ofproto *ofproto = dump->ofproto;
1697 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1700 return dump->error == EOF ? 0 : dump->error;
1703 /* Returns the type to pass to netdev_open() when a datapath of type
1704 * 'datapath_type' has a port of type 'port_type', for a few special
1705 * cases when a netdev type differs from a port type. For example, when
1706 * using the userspace datapath, a port of type "internal" needs to be
1709 * Returns either 'type' itself or a string literal, which must not be
1712 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1714 const struct ofproto_class *class;
1716 datapath_type = ofproto_normalize_type(datapath_type);
1717 class = ofproto_class_find__(datapath_type);
1722 return (class->port_open_type
1723 ? class->port_open_type(datapath_type, port_type)
1727 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1728 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1729 * the port's OpenFlow port number.
1731 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1732 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1733 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1734 * 'ofp_portp' is non-null). */
1736 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1737 ofp_port_t *ofp_portp)
1739 ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1742 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1744 const char *netdev_name = netdev_get_name(netdev);
1746 simap_put(&ofproto->ofp_requests, netdev_name,
1747 ofp_to_u16(ofp_port));
1748 update_port(ofproto, netdev_name);
1751 struct ofproto_port ofproto_port;
1753 ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
1755 *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
1756 ofproto_port_destroy(&ofproto_port);
1761 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1762 * initializes '*port' appropriately; on failure, returns a positive errno
1765 * The caller owns the data in 'ofproto_port' and must free it with
1766 * ofproto_port_destroy() when it is no longer needed. */
1768 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1769 struct ofproto_port *port)
1773 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1775 memset(port, 0, sizeof *port);
1780 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1781 * Returns 0 if successful, otherwise a positive errno. */
1783 ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port)
1785 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1786 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1787 struct simap_node *ofp_request_node;
1790 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1791 if (ofp_request_node) {
1792 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1795 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1796 if (!error && ofport) {
1797 /* 'name' is the netdev's name and update_port() is going to close the
1798 * netdev. Just in case update_port() refers to 'name' after it
1799 * destroys 'ofport', make a copy of it around the update_port()
1801 char *devname = xstrdup(name);
1802 update_port(ofproto, devname);
1809 flow_mod_init(struct ofputil_flow_mod *fm,
1810 const struct match *match, unsigned int priority,
1811 const struct ofpact *ofpacts, size_t ofpacts_len,
1812 enum ofp_flow_mod_command command)
1814 memset(fm, 0, sizeof *fm);
1816 fm->priority = priority;
1819 fm->modify_cookie = false;
1821 fm->command = command;
1822 fm->idle_timeout = 0;
1823 fm->hard_timeout = 0;
1824 fm->buffer_id = UINT32_MAX;
1825 fm->out_port = OFPP_ANY;
1826 fm->out_group = OFPG_ANY;
1828 fm->ofpacts = CONST_CAST(struct ofpact *, ofpacts);
1829 fm->ofpacts_len = ofpacts_len;
1833 simple_flow_mod(struct ofproto *ofproto,
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 struct ofputil_flow_mod fm;
1840 flow_mod_init(&fm, match, priority, ofpacts, ofpacts_len, command);
1842 return handle_flow_mod__(ofproto, NULL, &fm, NULL);
1845 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1846 * performs the 'n_actions' actions in 'actions'. The new flow will not
1849 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1850 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1851 * controllers; otherwise, it will be hidden.
1853 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1855 * This is a helper function for in-band control and fail-open. */
1857 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1858 unsigned int priority,
1859 const struct ofpact *ofpacts, size_t ofpacts_len)
1860 OVS_EXCLUDED(ofproto_mutex)
1862 const struct rule *rule;
1865 /* First do a cheap check whether the rule we're looking for already exists
1866 * with the actions that we want. If it does, then we're done. */
1867 ovs_rwlock_rdlock(&ofproto->tables[0].cls.rwlock);
1868 rule = rule_from_cls_rule(classifier_find_match_exactly(
1869 &ofproto->tables[0].cls, match, priority));
1871 ovs_mutex_lock(&rule->mutex);
1872 must_add = !ofpacts_equal(rule->actions->ofpacts,
1873 rule->actions->ofpacts_len,
1874 ofpacts, ofpacts_len);
1875 ovs_mutex_unlock(&rule->mutex);
1879 ovs_rwlock_unlock(&ofproto->tables[0].cls.rwlock);
1881 /* If there's no such rule or the rule doesn't have the actions we want,
1882 * fall back to a executing a full flow mod. We can't optimize this at
1883 * all because we didn't take enough locks above to ensure that the flow
1884 * table didn't already change beneath us. */
1886 simple_flow_mod(ofproto, match, priority, ofpacts, ofpacts_len,
1887 OFPFC_MODIFY_STRICT);
1891 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1892 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1893 * operation cannot be initiated now but may be retried later.
1895 * This is a helper function for in-band control and fail-open and the "learn"
1898 ofproto_flow_mod(struct ofproto *ofproto, struct ofputil_flow_mod *fm)
1899 OVS_EXCLUDED(ofproto_mutex)
1901 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1904 /* Searches for a rule with matching criteria exactly equal to 'target' in
1905 * ofproto's table 0 and, if it finds one, deletes it.
1907 * This is a helper function for in-band control and fail-open. */
1909 ofproto_delete_flow(struct ofproto *ofproto,
1910 const struct match *target, unsigned int priority)
1911 OVS_EXCLUDED(ofproto_mutex)
1913 struct classifier *cls = &ofproto->tables[0].cls;
1916 /* First do a cheap check whether the rule we're looking for has already
1917 * been deleted. If so, then we're done. */
1918 ovs_rwlock_rdlock(&cls->rwlock);
1919 rule = rule_from_cls_rule(classifier_find_match_exactly(cls, target,
1921 ovs_rwlock_unlock(&cls->rwlock);
1926 /* Fall back to a executing a full flow mod. We can't optimize this at all
1927 * because we didn't take enough locks above to ensure that the flow table
1928 * didn't already change beneath us. */
1929 return simple_flow_mod(ofproto, target, priority, NULL, 0,
1930 OFPFC_DELETE_STRICT) != OFPROTO_POSTPONE;
1933 /* Starts the process of deleting all of the flows from all of ofproto's flow
1934 * tables and then reintroducing the flows required by in-band control and
1935 * fail-open. The process will complete in a later call to ofproto_run(). */
1937 ofproto_flush_flows(struct ofproto *ofproto)
1939 COVERAGE_INC(ofproto_flush);
1940 ofproto->state = S_FLUSH;
1944 reinit_ports(struct ofproto *p)
1946 struct ofproto_port_dump dump;
1947 struct sset devnames;
1948 struct ofport *ofport;
1949 struct ofproto_port ofproto_port;
1950 const char *devname;
1952 COVERAGE_INC(ofproto_reinit_ports);
1954 sset_init(&devnames);
1955 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1956 sset_add(&devnames, netdev_get_name(ofport->netdev));
1958 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1959 sset_add(&devnames, ofproto_port.name);
1962 SSET_FOR_EACH (devname, &devnames) {
1963 update_port(p, devname);
1965 sset_destroy(&devnames);
1969 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
1973 port_idx = simap_get(&ofproto->ofp_requests, netdev_name);
1974 port_idx = port_idx ? port_idx : UINT16_MAX;
1976 if (port_idx >= ofproto->max_ports
1977 || ofport_get_usage(ofproto, u16_to_ofp(port_idx)) == LLONG_MAX) {
1978 uint16_t lru_ofport = 0, end_port_no = ofproto->alloc_port_no;
1979 long long int last_used_at, lru = LLONG_MAX;
1981 /* Search for a free OpenFlow port number. We try not to
1982 * immediately reuse them to prevent problems due to old
1985 if (++ofproto->alloc_port_no >= ofproto->max_ports) {
1986 ofproto->alloc_port_no = 1;
1988 last_used_at = ofport_get_usage(ofproto,
1989 u16_to_ofp(ofproto->alloc_port_no));
1990 if (!last_used_at) {
1991 port_idx = ofproto->alloc_port_no;
1993 } else if ( last_used_at < time_msec() - 60*60*1000) {
1994 /* If the port with ofport 'ofproto->alloc_port_no' was deleted
1995 * more than an hour ago, consider it usable. */
1996 ofport_remove_usage(ofproto,
1997 u16_to_ofp(ofproto->alloc_port_no));
1998 port_idx = ofproto->alloc_port_no;
2000 } else if (last_used_at < lru) {
2002 lru_ofport = ofproto->alloc_port_no;
2005 if (ofproto->alloc_port_no == end_port_no) {
2007 port_idx = lru_ofport;
2014 ofport_set_usage(ofproto, u16_to_ofp(port_idx), LLONG_MAX);
2015 return u16_to_ofp(port_idx);
2019 dealloc_ofp_port(struct ofproto *ofproto, ofp_port_t ofp_port)
2021 if (ofp_to_u16(ofp_port) < ofproto->max_ports) {
2022 ofport_set_usage(ofproto, ofp_port, time_msec());
2026 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
2027 * pointer if the netdev cannot be opened. On success, also fills in
2029 static struct netdev *
2030 ofport_open(struct ofproto *ofproto,
2031 struct ofproto_port *ofproto_port,
2032 struct ofputil_phy_port *pp)
2034 enum netdev_flags flags;
2035 struct netdev *netdev;
2038 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
2040 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
2041 "cannot be opened (%s)",
2043 ofproto_port->name, ofproto_port->ofp_port,
2044 ofproto_port->name, ovs_strerror(error));
2048 if (ofproto_port->ofp_port == OFPP_NONE) {
2049 if (!strcmp(ofproto->name, ofproto_port->name)) {
2050 ofproto_port->ofp_port = OFPP_LOCAL;
2052 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
2053 ofproto_port->name);
2056 pp->port_no = ofproto_port->ofp_port;
2057 netdev_get_etheraddr(netdev, pp->hw_addr);
2058 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
2059 netdev_get_flags(netdev, &flags);
2060 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
2061 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
2062 netdev_get_features(netdev, &pp->curr, &pp->advertised,
2063 &pp->supported, &pp->peer);
2064 pp->curr_speed = netdev_features_to_bps(pp->curr, 0) / 1000;
2065 pp->max_speed = netdev_features_to_bps(pp->supported, 0) / 1000;
2070 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
2071 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
2074 ofport_equal(const struct ofputil_phy_port *a,
2075 const struct ofputil_phy_port *b)
2077 return (eth_addr_equals(a->hw_addr, b->hw_addr)
2078 && a->state == b->state
2079 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
2080 && a->curr == b->curr
2081 && a->advertised == b->advertised
2082 && a->supported == b->supported
2083 && a->peer == b->peer
2084 && a->curr_speed == b->curr_speed
2085 && a->max_speed == b->max_speed);
2088 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
2089 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
2090 * one with the same name or port number). */
2092 ofport_install(struct ofproto *p,
2093 struct netdev *netdev, const struct ofputil_phy_port *pp)
2095 const char *netdev_name = netdev_get_name(netdev);
2096 struct ofport *ofport;
2099 /* Create ofport. */
2100 ofport = p->ofproto_class->port_alloc();
2105 ofport->ofproto = p;
2106 ofport->netdev = netdev;
2107 ofport->change_seq = netdev_change_seq(netdev);
2109 ofport->ofp_port = pp->port_no;
2110 ofport->created = time_msec();
2112 /* Add port to 'p'. */
2113 hmap_insert(&p->ports, &ofport->hmap_node,
2114 hash_ofp_port(ofport->ofp_port));
2115 shash_add(&p->port_by_name, netdev_name, ofport);
2117 update_mtu(p, ofport);
2119 /* Let the ofproto_class initialize its private data. */
2120 error = p->ofproto_class->port_construct(ofport);
2124 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
2128 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
2129 p->name, netdev_name, ovs_strerror(error));
2131 ofport_destroy__(ofport);
2133 netdev_close(netdev);
2137 /* Removes 'ofport' from 'p' and destroys it. */
2139 ofport_remove(struct ofport *ofport)
2141 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
2143 ofport_destroy(ofport);
2146 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
2149 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
2151 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
2153 ofport_remove(port);
2157 /* Updates 'port' with new 'pp' description.
2159 * Does not handle a name or port number change. The caller must implement
2160 * such a change as a delete followed by an add. */
2162 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
2164 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
2165 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
2166 | (pp->config & OFPUTIL_PC_PORT_DOWN));
2167 port->pp.state = pp->state;
2168 port->pp.curr = pp->curr;
2169 port->pp.advertised = pp->advertised;
2170 port->pp.supported = pp->supported;
2171 port->pp.peer = pp->peer;
2172 port->pp.curr_speed = pp->curr_speed;
2173 port->pp.max_speed = pp->max_speed;
2175 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
2178 /* Update OpenFlow 'state' in 'port' and notify controller. */
2180 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
2182 if (port->pp.state != state) {
2183 port->pp.state = state;
2184 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
2190 ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port)
2192 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
2194 if (port->ofproto->ofproto_class->set_realdev) {
2195 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
2197 if (port->ofproto->ofproto_class->set_stp_port) {
2198 port->ofproto->ofproto_class->set_stp_port(port, NULL);
2200 if (port->ofproto->ofproto_class->set_cfm) {
2201 port->ofproto->ofproto_class->set_cfm(port, NULL);
2203 if (port->ofproto->ofproto_class->bundle_remove) {
2204 port->ofproto->ofproto_class->bundle_remove(port);
2210 ofport_destroy__(struct ofport *port)
2212 struct ofproto *ofproto = port->ofproto;
2213 const char *name = netdev_get_name(port->netdev);
2215 hmap_remove(&ofproto->ports, &port->hmap_node);
2216 shash_delete(&ofproto->port_by_name,
2217 shash_find(&ofproto->port_by_name, name));
2219 netdev_close(port->netdev);
2220 ofproto->ofproto_class->port_dealloc(port);
2224 ofport_destroy(struct ofport *port)
2227 dealloc_ofp_port(port->ofproto, port->ofp_port);
2228 port->ofproto->ofproto_class->port_destruct(port);
2229 ofport_destroy__(port);
2234 ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
2236 struct ofport *port;
2238 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_ofp_port(ofp_port),
2240 if (port->ofp_port == ofp_port) {
2247 static long long int
2248 ofport_get_usage(const struct ofproto *ofproto, ofp_port_t ofp_port)
2250 struct ofport_usage *usage;
2252 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2253 &ofproto->ofport_usage) {
2254 if (usage->ofp_port == ofp_port) {
2255 return usage->last_used;
2262 ofport_set_usage(struct ofproto *ofproto, ofp_port_t ofp_port,
2263 long long int last_used)
2265 struct ofport_usage *usage;
2266 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2267 &ofproto->ofport_usage) {
2268 if (usage->ofp_port == ofp_port) {
2269 usage->last_used = last_used;
2273 ovs_assert(last_used == LLONG_MAX);
2275 usage = xmalloc(sizeof *usage);
2276 usage->ofp_port = ofp_port;
2277 usage->last_used = last_used;
2278 hmap_insert(&ofproto->ofport_usage, &usage->hmap_node,
2279 hash_ofp_port(ofp_port));
2283 ofport_remove_usage(struct ofproto *ofproto, ofp_port_t ofp_port)
2285 struct ofport_usage *usage;
2286 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2287 &ofproto->ofport_usage) {
2288 if (usage->ofp_port == ofp_port) {
2289 hmap_remove(&ofproto->ofport_usage, &usage->hmap_node);
2297 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
2299 struct ofproto *ofproto = port->ofproto;
2302 if (ofproto->ofproto_class->port_get_stats) {
2303 error = ofproto->ofproto_class->port_get_stats(port, stats);
2312 update_port(struct ofproto *ofproto, const char *name)
2314 struct ofproto_port ofproto_port;
2315 struct ofputil_phy_port pp;
2316 struct netdev *netdev;
2317 struct ofport *port;
2319 COVERAGE_INC(ofproto_update_port);
2321 /* Fetch 'name''s location and properties from the datapath. */
2322 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
2323 ? ofport_open(ofproto, &ofproto_port, &pp)
2327 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
2328 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
2329 struct netdev *old_netdev = port->netdev;
2331 /* 'name' hasn't changed location. Any properties changed? */
2332 if (!ofport_equal(&port->pp, &pp)) {
2333 ofport_modified(port, &pp);
2336 update_mtu(ofproto, port);
2338 /* Install the newly opened netdev in case it has changed.
2339 * Don't close the old netdev yet in case port_modified has to
2340 * remove a retained reference to it.*/
2341 port->netdev = netdev;
2342 port->change_seq = netdev_change_seq(netdev);
2344 if (port->ofproto->ofproto_class->port_modified) {
2345 port->ofproto->ofproto_class->port_modified(port);
2348 netdev_close(old_netdev);
2350 /* If 'port' is nonnull then its name differs from 'name' and thus
2351 * we should delete it. If we think there's a port named 'name'
2352 * then its port number must be wrong now so delete it too. */
2354 ofport_remove(port);
2356 ofport_remove_with_name(ofproto, name);
2357 ofport_install(ofproto, netdev, &pp);
2360 /* Any port named 'name' is gone now. */
2361 ofport_remove_with_name(ofproto, name);
2363 ofproto_port_destroy(&ofproto_port);
2367 init_ports(struct ofproto *p)
2369 struct ofproto_port_dump dump;
2370 struct ofproto_port ofproto_port;
2371 struct shash_node *node, *next;
2373 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2374 const char *name = ofproto_port.name;
2376 if (shash_find(&p->port_by_name, name)) {
2377 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2380 struct ofputil_phy_port pp;
2381 struct netdev *netdev;
2383 /* Check if an OpenFlow port number had been requested. */
2384 node = shash_find(&init_ofp_ports, name);
2386 const struct iface_hint *iface_hint = node->data;
2387 simap_put(&p->ofp_requests, name,
2388 ofp_to_u16(iface_hint->ofp_port));
2391 netdev = ofport_open(p, &ofproto_port, &pp);
2393 ofport_install(p, netdev, &pp);
2394 if (ofp_to_u16(ofproto_port.ofp_port) < p->max_ports) {
2395 p->alloc_port_no = MAX(p->alloc_port_no,
2396 ofp_to_u16(ofproto_port.ofp_port));
2402 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2403 struct iface_hint *iface_hint = node->data;
2405 if (!strcmp(iface_hint->br_name, p->name)) {
2406 free(iface_hint->br_name);
2407 free(iface_hint->br_type);
2409 shash_delete(&init_ofp_ports, node);
2416 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2417 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2419 find_min_mtu(struct ofproto *p)
2421 struct ofport *ofport;
2424 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2425 struct netdev *netdev = ofport->netdev;
2428 /* Skip any internal ports, since that's what we're trying to
2430 if (!strcmp(netdev_get_type(netdev), "internal")) {
2434 if (netdev_get_mtu(netdev, &dev_mtu)) {
2437 if (!mtu || dev_mtu < mtu) {
2442 return mtu ? mtu: ETH_PAYLOAD_MAX;
2445 /* Update MTU of all datapath devices on 'p' to the minimum of the
2446 * non-datapath ports in event of 'port' added or changed. */
2448 update_mtu(struct ofproto *p, struct ofport *port)
2450 struct ofport *ofport;
2451 struct netdev *netdev = port->netdev;
2452 int dev_mtu, old_min;
2454 if (netdev_get_mtu(netdev, &dev_mtu)) {
2458 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2459 if (dev_mtu > p->min_mtu) {
2460 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2461 dev_mtu = p->min_mtu;
2464 port->mtu = dev_mtu;
2468 /* For non-internal port find new min mtu. */
2469 old_min = p->min_mtu;
2470 port->mtu = dev_mtu;
2471 p->min_mtu = find_min_mtu(p);
2472 if (p->min_mtu == old_min) {
2476 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2477 struct netdev *netdev = ofport->netdev;
2479 if (!strcmp(netdev_get_type(netdev), "internal")) {
2480 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2481 ofport->mtu = p->min_mtu;
2488 ofproto_rule_ref(struct rule *rule)
2493 atomic_add(&rule->ref_count, 1, &orig);
2494 ovs_assert(orig != 0);
2499 ofproto_rule_unref(struct rule *rule)
2504 atomic_sub(&rule->ref_count, 1, &orig);
2506 rule->ofproto->ofproto_class->rule_destruct(rule);
2507 ofproto_rule_destroy__(rule);
2509 ovs_assert(orig != 0);
2514 struct rule_actions *
2515 rule_get_actions(const struct rule *rule)
2516 OVS_EXCLUDED(rule->mutex)
2518 struct rule_actions *actions;
2520 ovs_mutex_lock(&rule->mutex);
2521 actions = rule_get_actions__(rule);
2522 ovs_mutex_unlock(&rule->mutex);
2527 struct rule_actions *
2528 rule_get_actions__(const struct rule *rule)
2529 OVS_REQUIRES(rule->mutex)
2531 rule_actions_ref(rule->actions);
2532 return rule->actions;
2536 ofproto_rule_destroy__(struct rule *rule)
2537 OVS_NO_THREAD_SAFETY_ANALYSIS
2539 cls_rule_destroy(CONST_CAST(struct cls_rule *, &rule->cr));
2540 rule_actions_unref(rule->actions);
2541 ovs_mutex_destroy(&rule->mutex);
2542 rule->ofproto->ofproto_class->rule_dealloc(rule);
2545 static uint32_t get_provider_meter_id(const struct ofproto *,
2546 uint32_t of_meter_id);
2548 /* Creates and returns a new 'struct rule_actions', with a ref_count of 1,
2549 * whose actions are a copy of from the 'ofpacts_len' bytes of 'ofpacts'. */
2550 struct rule_actions *
2551 rule_actions_create(const struct ofproto *ofproto,
2552 const struct ofpact *ofpacts, size_t ofpacts_len)
2554 struct rule_actions *actions;
2556 actions = xmalloc(sizeof *actions);
2557 atomic_init(&actions->ref_count, 1);
2558 actions->ofpacts = xmemdup(ofpacts, ofpacts_len);
2559 actions->ofpacts_len = ofpacts_len;
2560 actions->provider_meter_id
2561 = get_provider_meter_id(ofproto,
2562 ofpacts_get_meter(ofpacts, ofpacts_len));
2567 /* Increments 'actions''s ref_count. */
2569 rule_actions_ref(struct rule_actions *actions)
2574 atomic_add(&actions->ref_count, 1, &orig);
2575 ovs_assert(orig != 0);
2579 /* Decrements 'actions''s ref_count and frees 'actions' if the ref_count
2582 rule_actions_unref(struct rule_actions *actions)
2587 atomic_sub(&actions->ref_count, 1, &orig);
2589 free(actions->ofpacts);
2592 ovs_assert(orig != 0);
2597 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2598 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2600 ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port)
2601 OVS_REQUIRES(ofproto_mutex)
2603 return (port == OFPP_ANY
2604 || ofpacts_output_to_port(rule->actions->ofpacts,
2605 rule->actions->ofpacts_len, port));
2608 /* Returns true if 'rule' has group and equals group_id. */
2610 ofproto_rule_has_out_group(const struct rule *rule, uint32_t group_id)
2611 OVS_REQUIRES(ofproto_mutex)
2613 return (group_id == OFPG11_ANY
2614 || ofpacts_output_to_group(rule->actions->ofpacts,
2615 rule->actions->ofpacts_len, group_id));
2618 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2619 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2621 ofoperation_has_out_port(const struct ofoperation *op, ofp_port_t out_port)
2622 OVS_REQUIRES(ofproto_mutex)
2624 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2629 case OFOPERATION_ADD:
2630 case OFOPERATION_DELETE:
2633 case OFOPERATION_MODIFY:
2634 case OFOPERATION_REPLACE:
2635 return ofpacts_output_to_port(op->actions->ofpacts,
2636 op->actions->ofpacts_len, out_port);
2643 rule_execute_destroy(struct rule_execute *e)
2645 ofproto_rule_unref(e->rule);
2646 list_remove(&e->list_node);
2650 /* Executes all "rule_execute" operations queued up in ofproto->rule_executes,
2651 * by passing them to the ofproto provider. */
2653 run_rule_executes(struct ofproto *ofproto)
2654 OVS_EXCLUDED(ofproto_mutex)
2656 struct rule_execute *e, *next;
2657 struct list executes;
2659 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2660 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2661 union flow_in_port in_port_;
2664 in_port_.ofp_port = e->in_port;
2665 flow_extract(e->packet, 0, 0, NULL, &in_port_, &flow);
2666 ofproto->ofproto_class->rule_execute(e->rule, &flow, e->packet);
2668 rule_execute_destroy(e);
2672 /* Destroys and discards all "rule_execute" operations queued up in
2673 * ofproto->rule_executes. */
2675 destroy_rule_executes(struct ofproto *ofproto)
2677 struct rule_execute *e, *next;
2678 struct list executes;
2680 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2681 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2682 ofpbuf_delete(e->packet);
2683 rule_execute_destroy(e);
2687 /* Returns true if 'rule' should be hidden from the controller.
2689 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2690 * (e.g. by in-band control) and are intentionally hidden from the
2693 ofproto_rule_is_hidden(const struct rule *rule)
2695 return rule->cr.priority > UINT16_MAX;
2698 static enum oftable_flags
2699 rule_get_flags(const struct rule *rule)
2701 return rule->ofproto->tables[rule->table_id].flags;
2705 rule_is_modifiable(const struct rule *rule)
2707 return !(rule_get_flags(rule) & OFTABLE_READONLY);
2711 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2713 ofconn_send_reply(ofconn, make_echo_reply(oh));
2718 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2720 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2721 struct ofputil_switch_features features;
2722 struct ofport *port;
2728 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2730 ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2732 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2733 * if present, are always at the end.) */
2734 n_tables = ofproto->n_tables;
2735 for (i = 0; i < ofproto->n_tables; i++) {
2736 if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
2742 features.datapath_id = ofproto->datapath_id;
2743 features.n_buffers = pktbuf_capacity();
2744 features.n_tables = n_tables;
2745 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2746 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2748 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2750 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2751 features.auxiliary_id = 0;
2752 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2754 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2755 ofputil_put_switch_features_port(&port->pp, b);
2758 ofconn_send_reply(ofconn, b);
2763 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2765 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2766 struct ofp_switch_config *osc;
2767 enum ofp_config_flags flags;
2771 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2772 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2773 flags = ofproto->frag_handling;
2774 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2775 if (oh->version < OFP13_VERSION
2776 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2777 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2779 osc->flags = htons(flags);
2780 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2781 ofconn_send_reply(ofconn, buf);
2787 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2789 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2790 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2791 uint16_t flags = ntohs(osc->flags);
2793 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2794 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
2795 enum ofp_config_flags cur = ofproto->frag_handling;
2796 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2798 ovs_assert((cur & OFPC_FRAG_MASK) == cur);
2800 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2801 ofproto->frag_handling = next;
2803 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2805 ofputil_frag_handling_to_string(next));
2809 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2810 ofconn_set_invalid_ttl_to_controller(ofconn,
2811 (oh->version < OFP13_VERSION
2812 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2814 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2819 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2820 * error message code for the caller to propagate upward. Otherwise, returns
2823 * The log message mentions 'msg_type'. */
2825 reject_slave_controller(struct ofconn *ofconn)
2827 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2828 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
2829 return OFPERR_OFPBRC_EPERM;
2835 /* Checks that the 'ofpacts_len' bytes of action in 'ofpacts' are appropriate
2838 * - If they use a meter, then 'ofproto' has that meter configured.
2840 * - If they use any groups, then 'ofproto' has that group configured.
2842 * Returns 0 if successful, otherwise an OpenFlow error. */
2844 ofproto_check_ofpacts(struct ofproto *ofproto,
2845 const struct ofpact ofpacts[], size_t ofpacts_len)
2847 const struct ofpact *a;
2850 mid = ofpacts_get_meter(ofpacts, ofpacts_len);
2851 if (mid && get_provider_meter_id(ofproto, mid) == UINT32_MAX) {
2852 return OFPERR_OFPMMFC_INVALID_METER;
2855 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2856 if (a->type == OFPACT_GROUP
2857 && !ofproto_group_exists(ofproto, ofpact_get_GROUP(a)->group_id)) {
2858 return OFPERR_OFPBAC_BAD_OUT_GROUP;
2866 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2868 struct ofproto *p = ofconn_get_ofproto(ofconn);
2869 struct ofputil_packet_out po;
2870 struct ofpbuf *payload;
2871 uint64_t ofpacts_stub[1024 / 8];
2872 struct ofpbuf ofpacts;
2874 union flow_in_port in_port_;
2877 COVERAGE_INC(ofproto_packet_out);
2879 error = reject_slave_controller(ofconn);
2884 /* Decode message. */
2885 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2886 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2888 goto exit_free_ofpacts;
2890 if (ofp_to_u16(po.in_port) >= p->max_ports
2891 && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) {
2892 error = OFPERR_OFPBRC_BAD_PORT;
2893 goto exit_free_ofpacts;
2898 if (po.buffer_id != UINT32_MAX) {
2899 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2900 if (error || !payload) {
2901 goto exit_free_ofpacts;
2904 /* Ensure that the L3 header is 32-bit aligned. */
2905 payload = ofpbuf_clone_data_with_headroom(po.packet, po.packet_len, 2);
2908 /* Verify actions against packet, then send packet if successful. */
2909 in_port_.ofp_port = po.in_port;
2910 flow_extract(payload, 0, 0, NULL, &in_port_, &flow);
2911 error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len);
2913 error = p->ofproto_class->packet_out(p, payload, &flow,
2914 po.ofpacts, po.ofpacts_len);
2916 ofpbuf_delete(payload);
2919 ofpbuf_uninit(&ofpacts);
2925 update_port_config(struct ofport *port,
2926 enum ofputil_port_config config,
2927 enum ofputil_port_config mask)
2929 enum ofputil_port_config old_config = port->pp.config;
2930 enum ofputil_port_config toggle;
2932 toggle = (config ^ port->pp.config) & mask;
2933 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2934 if (config & OFPUTIL_PC_PORT_DOWN) {
2935 netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL);
2937 netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL);
2939 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2942 port->pp.config ^= toggle;
2943 if (port->pp.config != old_config) {
2944 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2949 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2951 struct ofproto *p = ofconn_get_ofproto(ofconn);
2952 struct ofputil_port_mod pm;
2953 struct ofport *port;
2956 error = reject_slave_controller(ofconn);
2961 error = ofputil_decode_port_mod(oh, &pm);
2966 port = ofproto_get_port(p, pm.port_no);
2968 return OFPERR_OFPPMFC_BAD_PORT;
2969 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2970 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2972 update_port_config(port, pm.config, pm.mask);
2974 netdev_set_advertisements(port->netdev, pm.advertise);
2981 handle_desc_stats_request(struct ofconn *ofconn,
2982 const struct ofp_header *request)
2984 static const char *default_mfr_desc = "Nicira, Inc.";
2985 static const char *default_hw_desc = "Open vSwitch";
2986 static const char *default_sw_desc = VERSION;
2987 static const char *default_serial_desc = "None";
2988 static const char *default_dp_desc = "None";
2990 struct ofproto *p = ofconn_get_ofproto(ofconn);
2991 struct ofp_desc_stats *ods;
2994 msg = ofpraw_alloc_stats_reply(request, 0);
2995 ods = ofpbuf_put_zeros(msg, sizeof *ods);
2996 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
2997 sizeof ods->mfr_desc);
2998 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
2999 sizeof ods->hw_desc);
3000 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
3001 sizeof ods->sw_desc);
3002 ovs_strlcpy(ods->serial_num,
3003 p->serial_desc ? p->serial_desc : default_serial_desc,
3004 sizeof ods->serial_num);
3005 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
3006 sizeof ods->dp_desc);
3007 ofconn_send_reply(ofconn, msg);
3013 handle_table_stats_request(struct ofconn *ofconn,
3014 const struct ofp_header *request)
3016 struct ofproto *p = ofconn_get_ofproto(ofconn);
3017 struct ofp12_table_stats *ots;
3022 /* Set up default values.
3024 * ofp12_table_stats is used as a generic structure as
3025 * it is able to hold all the fields for ofp10_table_stats
3026 * and ofp11_table_stats (and of course itself).
3028 ots = xcalloc(p->n_tables, sizeof *ots);
3029 for (i = 0; i < p->n_tables; i++) {
3030 ots[i].table_id = i;
3031 sprintf(ots[i].name, "table%zu", i);
3032 ots[i].match = htonll(OFPXMT13_MASK);
3033 ots[i].wildcards = htonll(OFPXMT13_MASK);
3034 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
3035 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
3036 ots[i].write_setfields = htonll(OFPXMT13_MASK);
3037 ots[i].apply_setfields = htonll(OFPXMT13_MASK);
3038 ots[i].metadata_match = OVS_BE64_MAX;
3039 ots[i].metadata_write = OVS_BE64_MAX;
3040 ots[i].instructions = htonl(OFPIT11_ALL);
3041 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
3042 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
3043 ovs_rwlock_rdlock(&p->tables[i].cls.rwlock);
3044 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
3045 ovs_rwlock_unlock(&p->tables[i].cls.rwlock);
3048 p->ofproto_class->get_tables(p, ots);
3050 /* Post-process the tables, dropping hidden tables. */
3051 n_tables = p->n_tables;
3052 for (i = 0; i < p->n_tables; i++) {
3053 const struct oftable *table = &p->tables[i];
3055 if (table->flags & OFTABLE_HIDDEN) {
3061 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
3064 if (table->max_flows < ntohl(ots[i].max_entries)) {
3065 ots[i].max_entries = htonl(table->max_flows);
3069 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
3070 ofconn_send_reply(ofconn, msg);
3078 append_port_stat(struct ofport *port, struct list *replies)
3080 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
3082 calc_duration(port->created, time_msec(),
3083 &ops.duration_sec, &ops.duration_nsec);
3085 /* Intentionally ignore return value, since errors will set
3086 * 'stats' to all-1s, which is correct for OpenFlow, and
3087 * netdev_get_stats() will log errors. */
3088 ofproto_port_get_stats(port, &ops.stats);
3090 ofputil_append_port_stat(replies, &ops);
3094 handle_port_stats_request(struct ofconn *ofconn,
3095 const struct ofp_header *request)
3097 struct ofproto *p = ofconn_get_ofproto(ofconn);
3098 struct ofport *port;
3099 struct list replies;
3103 error = ofputil_decode_port_stats_request(request, &port_no);
3108 ofpmp_init(&replies, request);
3109 if (port_no != OFPP_ANY) {
3110 port = ofproto_get_port(p, port_no);
3112 append_port_stat(port, &replies);
3115 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3116 append_port_stat(port, &replies);
3120 ofconn_send_replies(ofconn, &replies);
3125 handle_port_desc_stats_request(struct ofconn *ofconn,
3126 const struct ofp_header *request)
3128 struct ofproto *p = ofconn_get_ofproto(ofconn);
3129 enum ofp_version version;
3130 struct ofport *port;
3131 struct list replies;
3133 ofpmp_init(&replies, request);
3135 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
3136 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3137 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
3140 ofconn_send_replies(ofconn, &replies);
3145 hash_cookie(ovs_be64 cookie)
3147 return hash_2words((OVS_FORCE uint64_t)cookie >> 32,
3148 (OVS_FORCE uint64_t)cookie);
3152 cookies_insert(struct ofproto *ofproto, struct rule *rule)
3153 OVS_REQUIRES(ofproto_mutex)
3155 hindex_insert(&ofproto->cookies, &rule->cookie_node,
3156 hash_cookie(rule->flow_cookie));
3160 cookies_remove(struct ofproto *ofproto, struct rule *rule)
3161 OVS_REQUIRES(ofproto_mutex)
3163 hindex_remove(&ofproto->cookies, &rule->cookie_node);
3167 ofproto_rule_change_cookie(struct ofproto *ofproto, struct rule *rule,
3168 ovs_be64 new_cookie)
3169 OVS_REQUIRES(ofproto_mutex)
3171 if (new_cookie != rule->flow_cookie) {
3172 cookies_remove(ofproto, rule);
3174 ovs_mutex_lock(&rule->mutex);
3175 rule->flow_cookie = new_cookie;
3176 ovs_mutex_unlock(&rule->mutex);
3178 cookies_insert(ofproto, rule);
3183 calc_duration(long long int start, long long int now,
3184 uint32_t *sec, uint32_t *nsec)
3186 long long int msecs = now - start;
3187 *sec = msecs / 1000;
3188 *nsec = (msecs % 1000) * (1000 * 1000);
3191 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
3192 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
3194 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
3196 return (table_id == 0xff || table_id < ofproto->n_tables
3198 : OFPERR_OFPBRC_BAD_TABLE_ID);
3202 static struct oftable *
3203 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
3205 struct oftable *table;
3207 for (table = &ofproto->tables[table_id];
3208 table < &ofproto->tables[ofproto->n_tables];
3210 if (!(table->flags & OFTABLE_HIDDEN)) {
3218 static struct oftable *
3219 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
3221 if (table_id == 0xff) {
3222 return next_visible_table(ofproto, 0);
3223 } else if (table_id < ofproto->n_tables) {
3224 return &ofproto->tables[table_id];
3230 static struct oftable *
3231 next_matching_table(const struct ofproto *ofproto,
3232 const struct oftable *table, uint8_t table_id)
3234 return (table_id == 0xff
3235 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
3239 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
3241 * - If TABLE_ID is 0xff, this iterates over every classifier table in
3242 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
3244 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
3245 * only once, for that table. (This can be used to access tables marked
3248 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
3249 * entered at all. (Perhaps you should have validated TABLE_ID with
3250 * check_table_id().)
3252 * All parameters are evaluated multiple times.
3254 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
3255 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
3257 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
3259 /* Initializes 'criteria' in a straightforward way based on the other
3262 * For "loose" matching, the 'priority' parameter is unimportant and may be
3265 rule_criteria_init(struct rule_criteria *criteria, uint8_t table_id,
3266 const struct match *match, unsigned int priority,
3267 ovs_be64 cookie, ovs_be64 cookie_mask,
3268 ofp_port_t out_port, uint32_t out_group)
3270 criteria->table_id = table_id;
3271 cls_rule_init(&criteria->cr, match, priority);
3272 criteria->cookie = cookie;
3273 criteria->cookie_mask = cookie_mask;
3274 criteria->out_port = out_port;
3275 criteria->out_group = out_group;
3279 rule_criteria_destroy(struct rule_criteria *criteria)
3281 cls_rule_destroy(&criteria->cr);
3285 rule_collection_init(struct rule_collection *rules)
3287 rules->rules = rules->stub;
3289 rules->capacity = ARRAY_SIZE(rules->stub);
3293 rule_collection_add(struct rule_collection *rules, struct rule *rule)
3295 if (rules->n >= rules->capacity) {
3296 size_t old_size, new_size;
3298 old_size = rules->capacity * sizeof *rules->rules;
3299 rules->capacity *= 2;
3300 new_size = rules->capacity * sizeof *rules->rules;
3302 if (rules->rules == rules->stub) {
3303 rules->rules = xmalloc(new_size);
3304 memcpy(rules->rules, rules->stub, old_size);
3306 rules->rules = xrealloc(rules->rules, new_size);
3310 rules->rules[rules->n++] = rule;
3314 rule_collection_ref(struct rule_collection *rules)
3315 OVS_REQUIRES(ofproto_mutex)
3319 for (i = 0; i < rules->n; i++) {
3320 ofproto_rule_ref(rules->rules[i]);
3325 rule_collection_unref(struct rule_collection *rules)
3329 for (i = 0; i < rules->n; i++) {
3330 ofproto_rule_unref(rules->rules[i]);
3335 rule_collection_destroy(struct rule_collection *rules)
3337 if (rules->rules != rules->stub) {
3343 collect_rule(struct rule *rule, const struct rule_criteria *c,
3344 struct rule_collection *rules)
3345 OVS_REQUIRES(ofproto_mutex)
3347 /* We ordinarily want to skip hidden rules, but there has to be a way for
3348 * code internal to OVS to modify and delete them, so if the criteria
3349 * specify a priority that can only be for a hidden flow, then allow hidden
3350 * rules to be selected. (This doesn't allow OpenFlow clients to meddle
3351 * with hidden flows because OpenFlow uses only a 16-bit field to specify
3353 if (ofproto_rule_is_hidden(rule) && c->cr.priority <= UINT16_MAX) {
3355 } else if (rule->pending) {
3356 return OFPROTO_POSTPONE;
3358 if ((c->table_id == rule->table_id || c->table_id == 0xff)
3359 && ofproto_rule_has_out_port(rule, c->out_port)
3360 && ofproto_rule_has_out_group(rule, c->out_group)
3361 && !((rule->flow_cookie ^ c->cookie) & c->cookie_mask)) {
3362 rule_collection_add(rules, rule);
3368 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3369 * on classifiers rules are done in the "loose" way required for OpenFlow
3370 * OFPFC_MODIFY and OFPFC_DELETE requests. Puts the selected rules on list
3373 * Hidden rules are always omitted.
3375 * Returns 0 on success, otherwise an OpenFlow error code. */
3377 collect_rules_loose(struct ofproto *ofproto,
3378 const struct rule_criteria *criteria,
3379 struct rule_collection *rules)
3380 OVS_REQUIRES(ofproto_mutex)
3382 struct oftable *table;
3385 rule_collection_init(rules);
3387 error = check_table_id(ofproto, criteria->table_id);
3392 if (criteria->cookie_mask == OVS_BE64_MAX) {
3395 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
3396 hash_cookie(criteria->cookie),
3397 &ofproto->cookies) {
3398 if (cls_rule_is_loose_match(&rule->cr, &criteria->cr.match)) {
3399 error = collect_rule(rule, criteria, rules);
3406 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
3407 struct cls_cursor cursor;
3410 ovs_rwlock_rdlock(&table->cls.rwlock);
3411 cls_cursor_init(&cursor, &table->cls, &criteria->cr);
3412 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3413 error = collect_rule(rule, criteria, rules);
3418 ovs_rwlock_unlock(&table->cls.rwlock);
3424 rule_collection_destroy(rules);
3429 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3430 * on classifiers rules are done in the "strict" way required for OpenFlow
3431 * OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests. Puts the selected
3432 * rules on list 'rules'.
3434 * Hidden rules are always omitted.
3436 * Returns 0 on success, otherwise an OpenFlow error code. */
3438 collect_rules_strict(struct ofproto *ofproto,
3439 const struct rule_criteria *criteria,
3440 struct rule_collection *rules)
3441 OVS_REQUIRES(ofproto_mutex)
3443 struct oftable *table;
3446 rule_collection_init(rules);
3448 error = check_table_id(ofproto, criteria->table_id);
3453 if (criteria->cookie_mask == OVS_BE64_MAX) {
3456 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
3457 hash_cookie(criteria->cookie),
3458 &ofproto->cookies) {
3459 if (cls_rule_equal(&rule->cr, &criteria->cr)) {
3460 error = collect_rule(rule, criteria, rules);
3467 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
3470 ovs_rwlock_rdlock(&table->cls.rwlock);
3471 rule = rule_from_cls_rule(classifier_find_rule_exactly(
3472 &table->cls, &criteria->cr));
3473 ovs_rwlock_unlock(&table->cls.rwlock);
3475 error = collect_rule(rule, criteria, rules);
3485 rule_collection_destroy(rules);
3490 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
3491 * forced into the range of a uint16_t. */
3493 age_secs(long long int age_ms)
3495 return (age_ms < 0 ? 0
3496 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
3497 : (unsigned int) age_ms / 1000);
3501 handle_flow_stats_request(struct ofconn *ofconn,
3502 const struct ofp_header *request)
3503 OVS_EXCLUDED(ofproto_mutex)
3505 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3506 struct ofputil_flow_stats_request fsr;
3507 struct rule_criteria criteria;
3508 struct rule_collection rules;
3509 struct list replies;
3513 error = ofputil_decode_flow_stats_request(&fsr, request);
3518 rule_criteria_init(&criteria, fsr.table_id, &fsr.match, 0, fsr.cookie,
3519 fsr.cookie_mask, fsr.out_port, fsr.out_group);
3521 ovs_mutex_lock(&ofproto_mutex);
3522 error = collect_rules_loose(ofproto, &criteria, &rules);
3523 rule_criteria_destroy(&criteria);
3525 rule_collection_ref(&rules);
3527 ovs_mutex_unlock(&ofproto_mutex);
3533 ofpmp_init(&replies, request);
3534 for (i = 0; i < rules.n; i++) {
3535 struct rule *rule = rules.rules[i];
3536 long long int now = time_msec();
3537 struct ofputil_flow_stats fs;
3538 long long int created, used, modified;
3539 struct rule_actions *actions;
3540 enum ofputil_flow_mod_flags flags;
3542 ovs_mutex_lock(&rule->mutex);
3543 fs.cookie = rule->flow_cookie;
3544 fs.idle_timeout = rule->idle_timeout;
3545 fs.hard_timeout = rule->hard_timeout;
3546 created = rule->created;
3548 modified = rule->modified;
3549 actions = rule_get_actions__(rule);
3550 flags = rule->flags;
3551 ovs_mutex_unlock(&rule->mutex);
3553 minimatch_expand(&rule->cr.match, &fs.match);
3554 fs.table_id = rule->table_id;
3555 calc_duration(created, now, &fs.duration_sec, &fs.duration_nsec);
3556 fs.priority = rule->cr.priority;
3557 fs.idle_age = age_secs(now - used);
3558 fs.hard_age = age_secs(now - modified);
3559 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
3561 fs.ofpacts = actions->ofpacts;
3562 fs.ofpacts_len = actions->ofpacts_len;
3565 ofputil_append_flow_stats_reply(&fs, &replies);
3567 rule_actions_unref(actions);
3570 rule_collection_unref(&rules);
3571 rule_collection_destroy(&rules);
3573 ofconn_send_replies(ofconn, &replies);
3579 flow_stats_ds(struct rule *rule, struct ds *results)
3581 uint64_t packet_count, byte_count;
3582 struct rule_actions *actions;
3583 long long int created;
3585 rule->ofproto->ofproto_class->rule_get_stats(rule,
3586 &packet_count, &byte_count);
3588 ovs_mutex_lock(&rule->mutex);
3589 actions = rule_get_actions__(rule);
3590 created = rule->created;
3591 ovs_mutex_unlock(&rule->mutex);
3593 if (rule->table_id != 0) {
3594 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
3596 ds_put_format(results, "duration=%llds, ", (time_msec() - created) / 1000);
3597 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3598 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3599 cls_rule_format(&rule->cr, results);
3600 ds_put_char(results, ',');
3602 ds_put_cstr(results, "actions=");
3603 ofpacts_format(actions->ofpacts, actions->ofpacts_len, results);
3605 ds_put_cstr(results, "\n");
3607 rule_actions_unref(actions);
3610 /* Adds a pretty-printed description of all flows to 'results', including
3611 * hidden flows (e.g., set up by in-band control). */
3613 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3615 struct oftable *table;
3617 OFPROTO_FOR_EACH_TABLE (table, p) {
3618 struct cls_cursor cursor;
3621 ovs_rwlock_rdlock(&table->cls.rwlock);
3622 cls_cursor_init(&cursor, &table->cls, NULL);
3623 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3624 flow_stats_ds(rule, results);
3626 ovs_rwlock_unlock(&table->cls.rwlock);
3630 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
3631 * '*engine_type' and '*engine_id', respectively. */
3633 ofproto_get_netflow_ids(const struct ofproto *ofproto,
3634 uint8_t *engine_type, uint8_t *engine_id)
3636 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
3639 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
3640 * true if the port's CFM status was successfully stored into '*status'.
3641 * Returns false if the port did not have CFM configured, in which case
3642 * '*status' is indeterminate.
3644 * The caller must provide and owns '*status', and must free 'status->rmps'. */
3646 ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port,
3647 struct ofproto_cfm_status *status)
3649 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
3651 && ofproto->ofproto_class->get_cfm_status
3652 && ofproto->ofproto_class->get_cfm_status(ofport, status));
3656 handle_aggregate_stats_request(struct ofconn *ofconn,
3657 const struct ofp_header *oh)
3658 OVS_EXCLUDED(ofproto_mutex)
3660 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3661 struct ofputil_flow_stats_request request;
3662 struct ofputil_aggregate_stats stats;
3663 bool unknown_packets, unknown_bytes;
3664 struct rule_criteria criteria;
3665 struct rule_collection rules;
3666 struct ofpbuf *reply;
3670 error = ofputil_decode_flow_stats_request(&request, oh);
3675 rule_criteria_init(&criteria, request.table_id, &request.match, 0,
3676 request.cookie, request.cookie_mask,
3677 request.out_port, request.out_group);
3679 ovs_mutex_lock(&ofproto_mutex);
3680 error = collect_rules_loose(ofproto, &criteria, &rules);
3681 rule_criteria_destroy(&criteria);
3683 rule_collection_ref(&rules);
3685 ovs_mutex_unlock(&ofproto_mutex);
3691 memset(&stats, 0, sizeof stats);
3692 unknown_packets = unknown_bytes = false;
3693 for (i = 0; i < rules.n; i++) {
3694 struct rule *rule = rules.rules[i];
3695 uint64_t packet_count;
3696 uint64_t byte_count;
3698 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
3701 if (packet_count == UINT64_MAX) {
3702 unknown_packets = true;
3704 stats.packet_count += packet_count;
3707 if (byte_count == UINT64_MAX) {
3708 unknown_bytes = true;
3710 stats.byte_count += byte_count;
3715 if (unknown_packets) {
3716 stats.packet_count = UINT64_MAX;
3718 if (unknown_bytes) {
3719 stats.byte_count = UINT64_MAX;
3722 rule_collection_unref(&rules);
3723 rule_collection_destroy(&rules);
3725 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
3726 ofconn_send_reply(ofconn, reply);
3731 struct queue_stats_cbdata {
3732 struct ofport *ofport;
3733 struct list replies;
3738 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3739 const struct netdev_queue_stats *stats)
3741 struct ofputil_queue_stats oqs;
3743 oqs.port_no = cbdata->ofport->pp.port_no;
3744 oqs.queue_id = queue_id;
3745 oqs.tx_bytes = stats->tx_bytes;
3746 oqs.tx_packets = stats->tx_packets;
3747 oqs.tx_errors = stats->tx_errors;
3748 if (stats->created != LLONG_MIN) {
3749 calc_duration(stats->created, cbdata->now,
3750 &oqs.duration_sec, &oqs.duration_nsec);
3752 oqs.duration_sec = oqs.duration_nsec = UINT32_MAX;
3754 ofputil_append_queue_stat(&cbdata->replies, &oqs);
3758 handle_queue_stats_dump_cb(uint32_t queue_id,
3759 struct netdev_queue_stats *stats,
3762 struct queue_stats_cbdata *cbdata = cbdata_;
3764 put_queue_stats(cbdata, queue_id, stats);
3768 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3769 struct queue_stats_cbdata *cbdata)
3771 cbdata->ofport = port;
3772 if (queue_id == OFPQ_ALL) {
3773 netdev_dump_queue_stats(port->netdev,
3774 handle_queue_stats_dump_cb, cbdata);
3776 struct netdev_queue_stats stats;
3778 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3779 put_queue_stats(cbdata, queue_id, &stats);
3781 return OFPERR_OFPQOFC_BAD_QUEUE;
3788 handle_queue_stats_request(struct ofconn *ofconn,
3789 const struct ofp_header *rq)
3791 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3792 struct queue_stats_cbdata cbdata;
3793 struct ofport *port;
3795 struct ofputil_queue_stats_request oqsr;
3797 COVERAGE_INC(ofproto_queue_req);
3799 ofpmp_init(&cbdata.replies, rq);
3800 cbdata.now = time_msec();
3802 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3807 if (oqsr.port_no == OFPP_ANY) {
3808 error = OFPERR_OFPQOFC_BAD_QUEUE;
3809 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3810 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3815 port = ofproto_get_port(ofproto, oqsr.port_no);
3817 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3818 : OFPERR_OFPQOFC_BAD_PORT);
3821 ofconn_send_replies(ofconn, &cbdata.replies);
3823 ofpbuf_list_delete(&cbdata.replies);
3830 is_flow_deletion_pending(const struct ofproto *ofproto,
3831 const struct cls_rule *cls_rule,
3833 OVS_REQUIRES(ofproto_mutex)
3835 if (!hmap_is_empty(&ofproto->deletions)) {
3836 struct ofoperation *op;
3838 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3839 cls_rule_hash(cls_rule, table_id),
3840 &ofproto->deletions) {
3841 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
3851 should_evict_a_rule(struct oftable *table, unsigned int extra_space)
3852 OVS_REQUIRES(ofproto_mutex)
3853 OVS_NO_THREAD_SAFETY_ANALYSIS
3855 return classifier_count(&table->cls) + extra_space > table->max_flows;
3859 evict_rules_from_table(struct ofproto *ofproto, struct oftable *table,
3860 unsigned int extra_space)
3861 OVS_REQUIRES(ofproto_mutex)
3863 while (should_evict_a_rule(table, extra_space)) {
3866 if (!choose_rule_to_evict(table, &rule)) {
3867 return OFPERR_OFPFMFC_TABLE_FULL;
3868 } else if (rule->pending) {
3869 return OFPROTO_POSTPONE;
3871 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
3872 delete_flow__(rule, group, OFPRR_EVICTION);
3873 ofopgroup_submit(group);
3880 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3881 * in which no matching flow already exists in the flow table.
3883 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3884 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3885 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3886 * initiated now but may be retried later.
3888 * The caller retains ownership of 'fm->ofpacts'.
3890 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3893 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
3894 struct ofputil_flow_mod *fm, const struct ofp_header *request)
3895 OVS_REQUIRES(ofproto_mutex)
3897 struct oftable *table;
3898 struct ofopgroup *group;
3904 error = check_table_id(ofproto, fm->table_id);
3910 if (fm->table_id == 0xff) {
3911 if (ofproto->ofproto_class->rule_choose_table) {
3912 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3918 ovs_assert(table_id < ofproto->n_tables);
3922 } else if (fm->table_id < ofproto->n_tables) {
3923 table_id = fm->table_id;
3925 return OFPERR_OFPBRC_BAD_TABLE_ID;
3928 table = &ofproto->tables[table_id];
3930 if (table->flags & OFTABLE_READONLY) {
3931 return OFPERR_OFPBRC_EPERM;
3934 cls_rule_init(&cr, &fm->match, fm->priority);
3936 /* Transform "add" into "modify" if there's an existing identical flow. */
3937 ovs_rwlock_rdlock(&table->cls.rwlock);
3938 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr));
3939 ovs_rwlock_unlock(&table->cls.rwlock);
3941 cls_rule_destroy(&cr);
3942 if (!rule_is_modifiable(rule)) {
3943 return OFPERR_OFPBRC_EPERM;
3944 } else if (rule->pending) {
3945 return OFPROTO_POSTPONE;
3947 struct rule_collection rules;
3949 rule_collection_init(&rules);
3950 rule_collection_add(&rules, rule);
3951 fm->modify_cookie = true;
3952 error = modify_flows__(ofproto, ofconn, fm, request, &rules);
3953 rule_collection_destroy(&rules);
3959 /* Serialize against pending deletion. */
3960 if (is_flow_deletion_pending(ofproto, &cr, table_id)) {
3961 cls_rule_destroy(&cr);
3962 return OFPROTO_POSTPONE;
3965 /* Check for overlap, if requested. */
3966 if (fm->flags & OFPUTIL_FF_CHECK_OVERLAP) {
3969 ovs_rwlock_rdlock(&table->cls.rwlock);
3970 overlaps = classifier_rule_overlaps(&table->cls, &cr);
3971 ovs_rwlock_unlock(&table->cls.rwlock);
3974 cls_rule_destroy(&cr);
3975 return OFPERR_OFPFMFC_OVERLAP;
3979 /* If necessary, evict an existing rule to clear out space. */
3980 error = evict_rules_from_table(ofproto, table, 1);
3982 cls_rule_destroy(&cr);
3986 /* Allocate new rule. */
3987 rule = ofproto->ofproto_class->rule_alloc();
3989 cls_rule_destroy(&cr);
3990 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
3991 ofproto->name, ovs_strerror(error));
3995 /* Initialize base state. */
3996 *CONST_CAST(struct ofproto **, &rule->ofproto) = ofproto;
3997 cls_rule_move(CONST_CAST(struct cls_rule *, &rule->cr), &cr);
3998 atomic_init(&rule->ref_count, 1);
3999 rule->pending = NULL;
4000 rule->flow_cookie = fm->new_cookie;
4001 rule->created = rule->modified = rule->used = time_msec();
4003 ovs_mutex_init(&rule->mutex);
4004 ovs_mutex_lock(&rule->mutex);
4005 rule->idle_timeout = fm->idle_timeout;
4006 rule->hard_timeout = fm->hard_timeout;
4007 ovs_mutex_unlock(&rule->mutex);
4009 *CONST_CAST(uint8_t *, &rule->table_id) = table - ofproto->tables;
4010 rule->flags = fm->flags & OFPUTIL_FF_STATE;
4011 rule->actions = rule_actions_create(ofproto, fm->ofpacts, fm->ofpacts_len);
4012 list_init(&rule->meter_list_node);
4013 rule->eviction_group = NULL;
4014 list_init(&rule->expirable);
4015 rule->monitor_flags = 0;
4016 rule->add_seqno = 0;
4017 rule->modify_seqno = 0;
4019 /* Construct rule, initializing derived state. */
4020 error = ofproto->ofproto_class->rule_construct(rule);
4022 ofproto_rule_destroy__(rule);
4027 oftable_insert_rule(rule);
4029 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
4030 ofoperation_create(group, rule, OFOPERATION_ADD, 0);
4031 ofproto->ofproto_class->rule_insert(rule);
4032 ofopgroup_submit(group);
4037 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
4039 /* Modifies the rules listed in 'rules', changing their actions to match those
4042 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4045 * Returns 0 on success, otherwise an OpenFlow error code. */
4047 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
4048 struct ofputil_flow_mod *fm, const struct ofp_header *request,
4049 const struct rule_collection *rules)
4050 OVS_REQUIRES(ofproto_mutex)
4052 enum ofoperation_type type;
4053 struct ofopgroup *group;
4057 type = fm->command == OFPFC_ADD ? OFOPERATION_REPLACE : OFOPERATION_MODIFY;
4058 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
4059 error = OFPERR_OFPBRC_EPERM;
4060 for (i = 0; i < rules->n; i++) {
4061 struct rule *rule = rules->rules[i];
4062 struct ofoperation *op;
4063 bool actions_changed;
4064 bool reset_counters;
4066 /* FIXME: Implement OFPFUTIL_FF_RESET_COUNTS */
4068 if (rule_is_modifiable(rule)) {
4069 /* At least one rule is modifiable, don't report EPERM error. */
4075 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
4076 rule->actions->ofpacts,
4077 rule->actions->ofpacts_len);
4079 op = ofoperation_create(group, rule, type, 0);
4081 if (fm->modify_cookie && fm->new_cookie != OVS_BE64_MAX) {
4082 ofproto_rule_change_cookie(ofproto, rule, fm->new_cookie);
4084 if (type == OFOPERATION_REPLACE) {
4085 ovs_mutex_lock(&rule->mutex);
4086 rule->idle_timeout = fm->idle_timeout;
4087 rule->hard_timeout = fm->hard_timeout;
4088 ovs_mutex_unlock(&rule->mutex);
4090 rule->flags = fm->flags & OFPUTIL_FF_STATE;
4091 if (fm->idle_timeout || fm->hard_timeout) {
4092 if (!rule->eviction_group) {
4093 eviction_group_add_rule(rule);
4096 eviction_group_remove_rule(rule);
4100 reset_counters = (fm->flags & OFPUTIL_FF_RESET_COUNTS) != 0;
4101 if (actions_changed || reset_counters) {
4102 struct rule_actions *new_actions;
4104 op->actions = rule->actions;
4105 new_actions = rule_actions_create(ofproto,
4106 fm->ofpacts, fm->ofpacts_len);
4108 ovs_mutex_lock(&rule->mutex);
4109 rule->actions = new_actions;
4110 ovs_mutex_unlock(&rule->mutex);
4112 rule->ofproto->ofproto_class->rule_modify_actions(rule,
4115 ofoperation_complete(op, 0);
4118 ofopgroup_submit(group);
4124 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
4125 struct ofputil_flow_mod *fm, const struct ofp_header *request)
4126 OVS_REQUIRES(ofproto_mutex)
4128 if (fm->cookie_mask != htonll(0) || fm->new_cookie == OVS_BE64_MAX) {
4131 return add_flow(ofproto, ofconn, fm, request);
4134 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
4137 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4140 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
4141 struct ofputil_flow_mod *fm,
4142 const struct ofp_header *request)
4143 OVS_REQUIRES(ofproto_mutex)
4145 struct rule_criteria criteria;
4146 struct rule_collection rules;
4149 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0,
4150 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY);
4151 error = collect_rules_loose(ofproto, &criteria, &rules);
4152 rule_criteria_destroy(&criteria);
4155 error = (rules.n > 0
4156 ? modify_flows__(ofproto, ofconn, fm, request, &rules)
4157 : modify_flows_add(ofproto, ofconn, fm, request));
4160 rule_collection_destroy(&rules);
4165 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4168 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4171 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
4172 struct ofputil_flow_mod *fm,
4173 const struct ofp_header *request)
4174 OVS_REQUIRES(ofproto_mutex)
4176 struct rule_criteria criteria;
4177 struct rule_collection rules;
4180 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
4181 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY);
4182 error = collect_rules_strict(ofproto, &criteria, &rules);
4183 rule_criteria_destroy(&criteria);
4187 error = modify_flows_add(ofproto, ofconn, fm, request);
4188 } else if (rules.n == 1) {
4189 error = modify_flows__(ofproto, ofconn, fm, request, &rules);
4193 rule_collection_destroy(&rules);
4198 /* OFPFC_DELETE implementation. */
4201 delete_flow__(struct rule *rule, struct ofopgroup *group,
4202 enum ofp_flow_removed_reason reason)
4203 OVS_REQUIRES(ofproto_mutex)
4205 struct ofproto *ofproto = rule->ofproto;
4207 ofproto_rule_send_removed(rule, reason);
4209 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
4210 oftable_remove_rule(rule);
4211 ofproto->ofproto_class->rule_delete(rule);
4214 /* Deletes the rules listed in 'rules'.
4216 * Returns 0 on success, otherwise an OpenFlow error code. */
4218 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
4219 const struct ofp_header *request,
4220 const struct rule_collection *rules,
4221 enum ofp_flow_removed_reason reason)
4222 OVS_REQUIRES(ofproto_mutex)
4224 struct ofopgroup *group;
4227 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
4228 for (i = 0; i < rules->n; i++) {
4229 delete_flow__(rules->rules[i], group, reason);
4231 ofopgroup_submit(group);
4236 /* Implements OFPFC_DELETE. */
4238 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
4239 const struct ofputil_flow_mod *fm,
4240 const struct ofp_header *request)
4241 OVS_REQUIRES(ofproto_mutex)
4243 struct rule_criteria criteria;
4244 struct rule_collection rules;
4247 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0,
4248 fm->cookie, fm->cookie_mask,
4249 fm->out_port, fm->out_group);
4250 error = collect_rules_loose(ofproto, &criteria, &rules);
4251 rule_criteria_destroy(&criteria);
4253 if (!error && rules.n > 0) {
4254 error = delete_flows__(ofproto, ofconn, request, &rules, OFPRR_DELETE);
4256 rule_collection_destroy(&rules);
4261 /* Implements OFPFC_DELETE_STRICT. */
4263 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
4264 const struct ofputil_flow_mod *fm,
4265 const struct ofp_header *request)
4266 OVS_REQUIRES(ofproto_mutex)
4268 struct rule_criteria criteria;
4269 struct rule_collection rules;
4272 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
4273 fm->cookie, fm->cookie_mask,
4274 fm->out_port, fm->out_group);
4275 error = collect_rules_strict(ofproto, &criteria, &rules);
4276 rule_criteria_destroy(&criteria);
4278 if (!error && rules.n > 0) {
4279 error = delete_flows__(ofproto, ofconn, request, &rules, OFPRR_DELETE);
4281 rule_collection_destroy(&rules);
4287 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
4288 OVS_REQUIRES(ofproto_mutex)
4290 struct ofputil_flow_removed fr;
4292 if (ofproto_rule_is_hidden(rule) ||
4293 !(rule->flags & OFPUTIL_FF_SEND_FLOW_REM)) {
4297 minimatch_expand(&rule->cr.match, &fr.match);
4298 fr.priority = rule->cr.priority;
4299 fr.cookie = rule->flow_cookie;
4301 fr.table_id = rule->table_id;
4302 calc_duration(rule->created, time_msec(),
4303 &fr.duration_sec, &fr.duration_nsec);
4304 ovs_mutex_lock(&rule->mutex);
4305 fr.idle_timeout = rule->idle_timeout;
4306 fr.hard_timeout = rule->hard_timeout;
4307 ovs_mutex_unlock(&rule->mutex);
4308 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
4311 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
4314 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
4315 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
4318 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
4321 * ofproto implementation ->run() functions should use this function to expire
4322 * OpenFlow flows. */
4324 ofproto_rule_expire(struct rule *rule, uint8_t reason)
4325 OVS_REQUIRES(ofproto_mutex)
4327 struct ofproto *ofproto = rule->ofproto;
4329 ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT
4330 || reason == OFPRR_DELETE || reason == OFPRR_GROUP_DELETE);
4332 ofproto_rule_delete__(ofproto, rule, reason);
4335 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
4336 * means "infinite". */
4338 reduce_timeout(uint16_t max, uint16_t *timeout)
4340 if (max && (!*timeout || *timeout > max)) {
4345 /* If 'idle_timeout' is nonzero, and 'rule' has no idle timeout or an idle
4346 * timeout greater than 'idle_timeout', lowers 'rule''s idle timeout to
4347 * 'idle_timeout' seconds. Similarly for 'hard_timeout'.
4349 * Suitable for implementing OFPACT_FIN_TIMEOUT. */
4351 ofproto_rule_reduce_timeouts(struct rule *rule,
4352 uint16_t idle_timeout, uint16_t hard_timeout)
4353 OVS_EXCLUDED(ofproto_mutex, rule->mutex)
4355 if (!idle_timeout && !hard_timeout) {
4359 ovs_mutex_lock(&ofproto_mutex);
4360 if (list_is_empty(&rule->expirable)) {
4361 list_insert(&rule->ofproto->expirable, &rule->expirable);
4363 ovs_mutex_unlock(&ofproto_mutex);
4365 ovs_mutex_lock(&rule->mutex);
4366 reduce_timeout(idle_timeout, &rule->idle_timeout);
4367 reduce_timeout(hard_timeout, &rule->hard_timeout);
4368 ovs_mutex_unlock(&rule->mutex);
4372 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4373 OVS_EXCLUDED(ofproto_mutex)
4375 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4376 struct ofputil_flow_mod fm;
4377 uint64_t ofpacts_stub[1024 / 8];
4378 struct ofpbuf ofpacts;
4382 error = reject_slave_controller(ofconn);
4387 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
4388 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
4390 u16_to_ofp(ofproto->max_ports),
4393 error = ofproto_check_ofpacts(ofproto, fm.ofpacts, fm.ofpacts_len);
4396 error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
4399 goto exit_free_ofpacts;
4402 /* Record the operation for logging a summary report. */
4403 switch (fm.command) {
4409 case OFPFC_MODIFY_STRICT:
4410 ofproto->n_modify++;
4414 case OFPFC_DELETE_STRICT:
4415 ofproto->n_delete++;
4420 if (ofproto->next_op_report == LLONG_MAX) {
4421 ofproto->first_op = now;
4422 ofproto->next_op_report = MAX(now + 10 * 1000,
4423 ofproto->op_backoff);
4424 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
4426 ofproto->last_op = now;
4429 ofpbuf_uninit(&ofpacts);
4435 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
4436 struct ofputil_flow_mod *fm, const struct ofp_header *oh)
4437 OVS_EXCLUDED(ofproto_mutex)
4441 ovs_mutex_lock(&ofproto_mutex);
4442 if (ofproto->n_pending < 50) {
4443 switch (fm->command) {
4445 error = add_flow(ofproto, ofconn, fm, oh);
4449 error = modify_flows_loose(ofproto, ofconn, fm, oh);
4452 case OFPFC_MODIFY_STRICT:
4453 error = modify_flow_strict(ofproto, ofconn, fm, oh);
4457 error = delete_flows_loose(ofproto, ofconn, fm, oh);
4460 case OFPFC_DELETE_STRICT:
4461 error = delete_flow_strict(ofproto, ofconn, fm, oh);
4465 if (fm->command > 0xff) {
4466 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
4467 "flow_mod_table_id extension is not enabled",
4470 error = OFPERR_OFPFMFC_BAD_COMMAND;
4474 ovs_assert(!list_is_empty(&ofproto->pending));
4475 error = OFPROTO_POSTPONE;
4477 ovs_mutex_unlock(&ofproto_mutex);
4479 run_rule_executes(ofproto);
4484 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4486 struct ofputil_role_request request;
4487 struct ofputil_role_request reply;
4491 error = ofputil_decode_role_message(oh, &request);
4496 if (request.role != OFPCR12_ROLE_NOCHANGE) {
4497 if (ofconn_get_role(ofconn) != request.role
4498 && ofconn_has_pending_opgroups(ofconn)) {
4499 return OFPROTO_POSTPONE;
4502 if (request.have_generation_id
4503 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
4504 return OFPERR_OFPRRFC_STALE;
4507 ofconn_set_role(ofconn, request.role);
4510 reply.role = ofconn_get_role(ofconn);
4511 reply.have_generation_id = ofconn_get_master_election_id(
4512 ofconn, &reply.generation_id);
4513 buf = ofputil_encode_role_reply(oh, &reply);
4514 ofconn_send_reply(ofconn, buf);
4520 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
4521 const struct ofp_header *oh)
4523 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
4524 enum ofputil_protocol cur, next;
4526 cur = ofconn_get_protocol(ofconn);
4527 next = ofputil_protocol_set_tid(cur, msg->set != 0);
4528 ofconn_set_protocol(ofconn, next);
4534 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4536 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
4537 enum ofputil_protocol cur, next;
4538 enum ofputil_protocol next_base;
4540 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
4542 return OFPERR_OFPBRC_EPERM;
4545 cur = ofconn_get_protocol(ofconn);
4546 next = ofputil_protocol_set_base(cur, next_base);
4547 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
4548 /* Avoid sending async messages in surprising protocol. */
4549 return OFPROTO_POSTPONE;
4552 ofconn_set_protocol(ofconn, next);
4557 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
4558 const struct ofp_header *oh)
4560 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
4563 format = ntohl(msg->format);
4564 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
4565 return OFPERR_OFPBRC_EPERM;
4568 if (format != ofconn_get_packet_in_format(ofconn)
4569 && ofconn_has_pending_opgroups(ofconn)) {
4570 /* Avoid sending async message in surprsing packet in format. */
4571 return OFPROTO_POSTPONE;
4574 ofconn_set_packet_in_format(ofconn, format);
4579 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
4581 const struct nx_async_config *msg = ofpmsg_body(oh);
4582 uint32_t master[OAM_N_TYPES];
4583 uint32_t slave[OAM_N_TYPES];
4585 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
4586 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
4587 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
4589 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
4590 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
4591 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
4593 ofconn_set_async_config(ofconn, master, slave);
4594 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
4595 !ofconn_get_miss_send_len(ofconn)) {
4596 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
4603 handle_nxt_get_async_request(struct ofconn *ofconn, const struct ofp_header *oh)
4606 uint32_t master[OAM_N_TYPES];
4607 uint32_t slave[OAM_N_TYPES];
4608 struct nx_async_config *msg;
4610 ofconn_get_async_config(ofconn, master, slave);
4611 buf = ofpraw_alloc_reply(OFPRAW_OFPT13_GET_ASYNC_REPLY, oh, 0);
4612 msg = ofpbuf_put_zeros(buf, sizeof *msg);
4614 msg->packet_in_mask[0] = htonl(master[OAM_PACKET_IN]);
4615 msg->port_status_mask[0] = htonl(master[OAM_PORT_STATUS]);
4616 msg->flow_removed_mask[0] = htonl(master[OAM_FLOW_REMOVED]);
4618 msg->packet_in_mask[1] = htonl(slave[OAM_PACKET_IN]);
4619 msg->port_status_mask[1] = htonl(slave[OAM_PORT_STATUS]);
4620 msg->flow_removed_mask[1] = htonl(slave[OAM_FLOW_REMOVED]);
4622 ofconn_send_reply(ofconn, buf);
4628 handle_nxt_set_controller_id(struct ofconn *ofconn,
4629 const struct ofp_header *oh)
4631 const struct nx_controller_id *nci = ofpmsg_body(oh);
4633 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
4634 return OFPERR_NXBRC_MUST_BE_ZERO;
4637 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
4642 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4646 if (ofconn_has_pending_opgroups(ofconn)) {
4647 return OFPROTO_POSTPONE;
4650 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
4651 ? OFPRAW_OFPT10_BARRIER_REPLY
4652 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
4653 ofconn_send_reply(ofconn, buf);
4658 ofproto_compose_flow_refresh_update(const struct rule *rule,
4659 enum nx_flow_monitor_flags flags,
4661 OVS_REQUIRES(ofproto_mutex)
4663 struct ofoperation *op = rule->pending;
4664 const struct rule_actions *actions;
4665 struct ofputil_flow_update fu;
4668 if (op && op->type == OFOPERATION_ADD) {
4669 /* We'll report the final flow when the operation completes. Reporting
4670 * it now would cause a duplicate report later. */
4674 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
4675 ? NXFME_ADDED : NXFME_MODIFIED);
4677 ovs_mutex_lock(&rule->mutex);
4678 fu.idle_timeout = rule->idle_timeout;
4679 fu.hard_timeout = rule->hard_timeout;
4680 ovs_mutex_unlock(&rule->mutex);
4681 fu.table_id = rule->table_id;
4682 fu.cookie = rule->flow_cookie;
4683 minimatch_expand(&rule->cr.match, &match);
4685 fu.priority = rule->cr.priority;
4687 if (!(flags & NXFMF_ACTIONS)) {
4690 actions = rule->actions;
4692 /* An operation is in progress. Use the previous version of the flow's
4693 * actions, so that when the operation commits we report the change. */
4695 case OFOPERATION_ADD:
4698 case OFOPERATION_MODIFY:
4699 case OFOPERATION_REPLACE:
4700 actions = op->actions ? op->actions : rule->actions;
4703 case OFOPERATION_DELETE:
4704 actions = rule->actions;
4711 fu.ofpacts = actions ? actions->ofpacts : NULL;
4712 fu.ofpacts_len = actions ? actions->ofpacts_len : 0;
4714 if (list_is_empty(msgs)) {
4715 ofputil_start_flow_update(msgs);
4717 ofputil_append_flow_update(&fu, msgs);
4721 ofmonitor_compose_refresh_updates(struct rule_collection *rules,
4723 OVS_REQUIRES(ofproto_mutex)
4727 for (i = 0; i < rules->n; i++) {
4728 struct rule *rule = rules->rules[i];
4729 enum nx_flow_monitor_flags flags = rule->monitor_flags;
4730 rule->monitor_flags = 0;
4732 ofproto_compose_flow_refresh_update(rule, flags, msgs);
4737 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
4738 struct rule *rule, uint64_t seqno,
4739 struct rule_collection *rules)
4740 OVS_REQUIRES(ofproto_mutex)
4742 enum nx_flow_monitor_flags update;
4744 if (ofproto_rule_is_hidden(rule)) {
4749 ? ofoperation_has_out_port(rule->pending, m->out_port)
4750 : ofproto_rule_has_out_port(rule, m->out_port))) {
4755 if (rule->add_seqno > seqno) {
4756 update = NXFMF_ADD | NXFMF_MODIFY;
4757 } else if (rule->modify_seqno > seqno) {
4758 update = NXFMF_MODIFY;
4763 if (!(m->flags & update)) {
4767 update = NXFMF_INITIAL;
4770 if (!rule->monitor_flags) {
4771 rule_collection_add(rules, rule);
4773 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
4777 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
4779 struct rule_collection *rules)
4780 OVS_REQUIRES(ofproto_mutex)
4782 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
4783 const struct ofoperation *op;
4784 const struct oftable *table;
4785 struct cls_rule target;
4787 cls_rule_init_from_minimatch(&target, &m->match, 0);
4788 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
4789 struct cls_cursor cursor;
4792 ovs_rwlock_rdlock(&table->cls.rwlock);
4793 cls_cursor_init(&cursor, &table->cls, &target);
4794 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4795 ovs_assert(!rule->pending); /* XXX */
4796 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4798 ovs_rwlock_unlock(&table->cls.rwlock);
4801 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
4802 struct rule *rule = op->rule;
4804 if (((m->table_id == 0xff
4805 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
4806 : m->table_id == rule->table_id))
4807 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
4808 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4811 cls_rule_destroy(&target);
4815 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
4816 struct rule_collection *rules)
4817 OVS_REQUIRES(ofproto_mutex)
4819 if (m->flags & NXFMF_INITIAL) {
4820 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
4825 ofmonitor_collect_resume_rules(struct ofmonitor *m,
4826 uint64_t seqno, struct rule_collection *rules)
4827 OVS_REQUIRES(ofproto_mutex)
4829 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
4833 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
4834 OVS_EXCLUDED(ofproto_mutex)
4836 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4837 struct ofmonitor **monitors;
4838 size_t n_monitors, allocated_monitors;
4839 struct rule_collection rules;
4840 struct list replies;
4846 ofpbuf_use_const(&b, oh, ntohs(oh->length));
4848 n_monitors = allocated_monitors = 0;
4850 ovs_mutex_lock(&ofproto_mutex);
4852 struct ofputil_flow_monitor_request request;
4853 struct ofmonitor *m;
4856 retval = ofputil_decode_flow_monitor_request(&request, &b);
4857 if (retval == EOF) {
4859 } else if (retval) {
4864 if (request.table_id != 0xff
4865 && request.table_id >= ofproto->n_tables) {
4866 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4870 error = ofmonitor_create(&request, ofconn, &m);
4875 if (n_monitors >= allocated_monitors) {
4876 monitors = x2nrealloc(monitors, &allocated_monitors,
4879 monitors[n_monitors++] = m;
4882 rule_collection_init(&rules);
4883 for (i = 0; i < n_monitors; i++) {
4884 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
4887 ofpmp_init(&replies, oh);
4888 ofmonitor_compose_refresh_updates(&rules, &replies);
4889 ovs_mutex_unlock(&ofproto_mutex);
4891 rule_collection_destroy(&rules);
4893 ofconn_send_replies(ofconn, &replies);
4899 for (i = 0; i < n_monitors; i++) {
4900 ofmonitor_destroy(monitors[i]);
4903 ovs_mutex_unlock(&ofproto_mutex);
4909 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
4910 OVS_EXCLUDED(ofproto_mutex)
4912 struct ofmonitor *m;
4916 id = ofputil_decode_flow_monitor_cancel(oh);
4918 ovs_mutex_lock(&ofproto_mutex);
4919 m = ofmonitor_lookup(ofconn, id);
4921 ofmonitor_destroy(m);
4924 error = OFPERR_NXBRC_FM_BAD_ID;
4926 ovs_mutex_unlock(&ofproto_mutex);
4931 /* Meters implementation.
4933 * Meter table entry, indexed by the OpenFlow meter_id.
4934 * These are always dynamically allocated to allocate enough space for
4936 * 'created' is used to compute the duration for meter stats.
4937 * 'list rules' is needed so that we can delete the dependent rules when the
4938 * meter table entry is deleted.
4939 * 'provider_meter_id' is for the provider's private use.
4942 long long int created; /* Time created. */
4943 struct list rules; /* List of "struct rule_dpif"s. */
4944 ofproto_meter_id provider_meter_id;
4945 uint16_t flags; /* Meter flags. */
4946 uint16_t n_bands; /* Number of meter bands. */
4947 struct ofputil_meter_band *bands;
4951 * This is used in instruction validation at flow set-up time,
4952 * as flows may not use non-existing meters.
4953 * Return value of UINT32_MAX signifies an invalid meter.
4956 get_provider_meter_id(const struct ofproto *ofproto, uint32_t of_meter_id)
4958 if (of_meter_id && of_meter_id <= ofproto->meter_features.max_meters) {
4959 const struct meter *meter = ofproto->meters[of_meter_id];
4961 return meter->provider_meter_id.uint32;
4968 meter_update(struct meter *meter, const struct ofputil_meter_config *config)
4972 meter->flags = config->flags;
4973 meter->n_bands = config->n_bands;
4974 meter->bands = xmemdup(config->bands,
4975 config->n_bands * sizeof *meter->bands);
4978 static struct meter *
4979 meter_create(const struct ofputil_meter_config *config,
4980 ofproto_meter_id provider_meter_id)
4982 struct meter *meter;
4984 meter = xzalloc(sizeof *meter);
4985 meter->provider_meter_id = provider_meter_id;
4986 meter->created = time_msec();
4987 list_init(&meter->rules);
4989 meter_update(meter, config);
4995 meter_delete(struct ofproto *ofproto, uint32_t first, uint32_t last)
4996 OVS_REQUIRES(ofproto_mutex)
4999 for (mid = first; mid <= last; ++mid) {
5000 struct meter *meter = ofproto->meters[mid];
5002 ofproto->meters[mid] = NULL;
5003 ofproto->ofproto_class->meter_del(ofproto,
5004 meter->provider_meter_id);
5012 handle_add_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
5014 ofproto_meter_id provider_meter_id = { UINT32_MAX };
5015 struct meter **meterp = &ofproto->meters[mm->meter.meter_id];
5019 return OFPERR_OFPMMFC_METER_EXISTS;
5022 error = ofproto->ofproto_class->meter_set(ofproto, &provider_meter_id,
5025 ovs_assert(provider_meter_id.uint32 != UINT32_MAX);
5026 *meterp = meter_create(&mm->meter, provider_meter_id);
5032 handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
5034 struct meter *meter = ofproto->meters[mm->meter.meter_id];
5036 uint32_t provider_meter_id;
5039 return OFPERR_OFPMMFC_UNKNOWN_METER;
5042 provider_meter_id = meter->provider_meter_id.uint32;
5043 error = ofproto->ofproto_class->meter_set(ofproto,
5044 &meter->provider_meter_id,
5046 ovs_assert(meter->provider_meter_id.uint32 == provider_meter_id);
5048 meter_update(meter, &mm->meter);
5054 handle_delete_meter(struct ofconn *ofconn, const struct ofp_header *oh,
5055 struct ofputil_meter_mod *mm)
5056 OVS_EXCLUDED(ofproto_mutex)
5058 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5059 uint32_t meter_id = mm->meter.meter_id;
5060 struct rule_collection rules;
5061 enum ofperr error = 0;
5062 uint32_t first, last;
5064 if (meter_id == OFPM13_ALL) {
5066 last = ofproto->meter_features.max_meters;
5068 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
5071 first = last = meter_id;
5074 /* First delete the rules that use this meter. If any of those rules are
5075 * currently being modified, postpone the whole operation until later. */
5076 rule_collection_init(&rules);
5077 ovs_mutex_lock(&ofproto_mutex);
5078 for (meter_id = first; meter_id <= last; ++meter_id) {
5079 struct meter *meter = ofproto->meters[meter_id];
5080 if (meter && !list_is_empty(&meter->rules)) {
5083 LIST_FOR_EACH (rule, meter_list_node, &meter->rules) {
5084 if (rule->pending) {
5085 error = OFPROTO_POSTPONE;
5088 rule_collection_add(&rules, rule);
5093 delete_flows__(ofproto, ofconn, oh, &rules, OFPRR_METER_DELETE);
5096 /* Delete the meters. */
5097 meter_delete(ofproto, first, last);
5100 ovs_mutex_unlock(&ofproto_mutex);
5101 rule_collection_destroy(&rules);
5107 handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5109 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5110 struct ofputil_meter_mod mm;
5111 uint64_t bands_stub[256 / 8];
5112 struct ofpbuf bands;
5116 error = reject_slave_controller(ofconn);
5121 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5123 error = ofputil_decode_meter_mod(oh, &mm, &bands);
5125 goto exit_free_bands;
5128 meter_id = mm.meter.meter_id;
5130 if (mm.command != OFPMC13_DELETE) {
5131 /* Fails also when meters are not implemented by the provider. */
5132 if (meter_id == 0 || meter_id > OFPM13_MAX) {
5133 error = OFPERR_OFPMMFC_INVALID_METER;
5134 goto exit_free_bands;
5135 } else if (meter_id > ofproto->meter_features.max_meters) {
5136 error = OFPERR_OFPMMFC_OUT_OF_METERS;
5137 goto exit_free_bands;
5139 if (mm.meter.n_bands > ofproto->meter_features.max_bands) {
5140 error = OFPERR_OFPMMFC_OUT_OF_BANDS;
5141 goto exit_free_bands;
5145 switch (mm.command) {
5147 error = handle_add_meter(ofproto, &mm);
5150 case OFPMC13_MODIFY:
5151 error = handle_modify_meter(ofproto, &mm);
5154 case OFPMC13_DELETE:
5155 error = handle_delete_meter(ofconn, oh, &mm);
5159 error = OFPERR_OFPMMFC_BAD_COMMAND;
5164 ofpbuf_uninit(&bands);
5169 handle_meter_features_request(struct ofconn *ofconn,
5170 const struct ofp_header *request)
5172 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5173 struct ofputil_meter_features features;
5176 if (ofproto->ofproto_class->meter_get_features) {
5177 ofproto->ofproto_class->meter_get_features(ofproto, &features);
5179 memset(&features, 0, sizeof features);
5181 b = ofputil_encode_meter_features_reply(&features, request);
5183 ofconn_send_reply(ofconn, b);
5188 handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request,
5191 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5192 struct list replies;
5193 uint64_t bands_stub[256 / 8];
5194 struct ofpbuf bands;
5195 uint32_t meter_id, first, last;
5197 ofputil_decode_meter_request(request, &meter_id);
5199 if (meter_id == OFPM13_ALL) {
5201 last = ofproto->meter_features.max_meters;
5203 if (!meter_id || meter_id > ofproto->meter_features.max_meters ||
5204 !ofproto->meters[meter_id]) {
5205 return OFPERR_OFPMMFC_UNKNOWN_METER;
5207 first = last = meter_id;
5210 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5211 ofpmp_init(&replies, request);
5213 for (meter_id = first; meter_id <= last; ++meter_id) {
5214 struct meter *meter = ofproto->meters[meter_id];
5216 continue; /* Skip non-existing meters. */
5218 if (type == OFPTYPE_METER_STATS_REQUEST) {
5219 struct ofputil_meter_stats stats;
5221 stats.meter_id = meter_id;
5223 /* Provider sets the packet and byte counts, we do the rest. */
5224 stats.flow_count = list_size(&meter->rules);
5225 calc_duration(meter->created, time_msec(),
5226 &stats.duration_sec, &stats.duration_nsec);
5227 stats.n_bands = meter->n_bands;
5228 ofpbuf_clear(&bands);
5230 = ofpbuf_put_uninit(&bands,
5231 meter->n_bands * sizeof *stats.bands);
5233 if (!ofproto->ofproto_class->meter_get(ofproto,
5234 meter->provider_meter_id,
5236 ofputil_append_meter_stats(&replies, &stats);
5238 } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */
5239 struct ofputil_meter_config config;
5241 config.meter_id = meter_id;
5242 config.flags = meter->flags;
5243 config.n_bands = meter->n_bands;
5244 config.bands = meter->bands;
5245 ofputil_append_meter_config(&replies, &config);
5249 ofconn_send_replies(ofconn, &replies);
5250 ofpbuf_uninit(&bands);
5255 ofproto_group_lookup(const struct ofproto *ofproto, uint32_t group_id,
5256 struct ofgroup **group)
5257 OVS_TRY_RDLOCK(true, (*group)->rwlock)
5259 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5260 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
5261 hash_int(group_id, 0), &ofproto->groups) {
5262 if ((*group)->group_id == group_id) {
5263 ovs_rwlock_rdlock(&(*group)->rwlock);
5264 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5268 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5273 ofproto_group_release(struct ofgroup *group)
5274 OVS_RELEASES(group->rwlock)
5276 ovs_rwlock_unlock(&group->rwlock);
5280 ofproto_group_write_lookup(const struct ofproto *ofproto, uint32_t group_id,
5281 struct ofgroup **group)
5282 OVS_TRY_WRLOCK(true, ofproto->groups_rwlock)
5283 OVS_TRY_WRLOCK(true, (*group)->rwlock)
5285 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5286 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
5287 hash_int(group_id, 0), &ofproto->groups) {
5288 if ((*group)->group_id == group_id) {
5289 ovs_rwlock_wrlock(&(*group)->rwlock);
5293 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5298 ofproto_group_exists__(const struct ofproto *ofproto, uint32_t group_id)
5299 OVS_REQ_RDLOCK(ofproto->groups_rwlock)
5301 struct ofgroup *grp;
5303 HMAP_FOR_EACH_IN_BUCKET (grp, hmap_node,
5304 hash_int(group_id, 0), &ofproto->groups) {
5305 if (grp->group_id == group_id) {
5313 ofproto_group_exists(const struct ofproto *ofproto, uint32_t group_id)
5314 OVS_EXCLUDED(ofproto->groups_rwlock)
5318 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5319 exists = ofproto_group_exists__(ofproto, group_id);
5320 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5326 group_get_ref_count(struct ofgroup *group)
5327 OVS_EXCLUDED(ofproto_mutex)
5329 struct ofproto *ofproto = group->ofproto;
5330 struct rule_criteria criteria;
5331 struct rule_collection rules;
5336 match_init_catchall(&match);
5337 rule_criteria_init(&criteria, 0xff, &match, 0, htonll(0), htonll(0),
5338 OFPP_ANY, group->group_id);
5339 ovs_mutex_lock(&ofproto_mutex);
5340 error = collect_rules_loose(ofproto, &criteria, &rules);
5341 ovs_mutex_unlock(&ofproto_mutex);
5342 rule_criteria_destroy(&criteria);
5344 count = !error && rules.n < UINT32_MAX ? rules.n : UINT32_MAX;
5346 rule_collection_destroy(&rules);
5351 append_group_stats(struct ofgroup *group, struct list *replies)
5352 OVS_REQ_RDLOCK(group->rwlock)
5354 struct ofputil_group_stats ogs;
5355 struct ofproto *ofproto = group->ofproto;
5356 long long int now = time_msec();
5359 ogs.bucket_stats = xmalloc(group->n_buckets * sizeof *ogs.bucket_stats);
5361 /* Provider sets the packet and byte counts, we do the rest. */
5362 ogs.ref_count = group_get_ref_count(group);
5363 ogs.n_buckets = group->n_buckets;
5365 error = (ofproto->ofproto_class->group_get_stats
5366 ? ofproto->ofproto_class->group_get_stats(group, &ogs)
5369 ogs.packet_count = UINT64_MAX;
5370 ogs.byte_count = UINT64_MAX;
5371 memset(ogs.bucket_stats, 0xff,
5372 ogs.n_buckets * sizeof *ogs.bucket_stats);
5375 ogs.group_id = group->group_id;
5376 calc_duration(group->created, now, &ogs.duration_sec, &ogs.duration_nsec);
5378 ofputil_append_group_stats(replies, &ogs);
5380 free(ogs.bucket_stats);
5384 handle_group_stats_request(struct ofconn *ofconn,
5385 const struct ofp_header *request)
5387 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5388 struct list replies;
5390 struct ofgroup *group;
5393 error = ofputil_decode_group_stats_request(request, &group_id);
5398 ofpmp_init(&replies, request);
5400 if (group_id == OFPG_ALL) {
5401 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5402 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
5403 ovs_rwlock_rdlock(&group->rwlock);
5404 append_group_stats(group, &replies);
5405 ovs_rwlock_unlock(&group->rwlock);
5407 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5409 if (ofproto_group_lookup(ofproto, group_id, &group)) {
5410 append_group_stats(group, &replies);
5411 ofproto_group_release(group);
5415 ofconn_send_replies(ofconn, &replies);
5421 handle_group_desc_stats_request(struct ofconn *ofconn,
5422 const struct ofp_header *request)
5424 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5425 struct list replies;
5426 struct ofputil_group_desc gds;
5427 struct ofgroup *group;
5429 ofpmp_init(&replies, request);
5431 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5432 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
5433 gds.group_id = group->group_id;
5434 gds.type = group->type;
5435 ofputil_append_group_desc_reply(&gds, &group->buckets, &replies);
5437 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5439 ofconn_send_replies(ofconn, &replies);
5445 handle_group_features_stats_request(struct ofconn *ofconn,
5446 const struct ofp_header *request)
5448 struct ofproto *p = ofconn_get_ofproto(ofconn);
5451 msg = ofputil_encode_group_features_reply(&p->ogf, request);
5453 ofconn_send_reply(ofconn, msg);
5460 handle_queue_get_config_request(struct ofconn *ofconn,
5461 const struct ofp_header *oh)
5463 struct ofproto *p = ofconn_get_ofproto(ofconn);
5464 struct netdev_queue_dump queue_dump;
5465 struct ofport *ofport;
5466 unsigned int queue_id;
5467 struct ofpbuf *reply;
5468 struct smap details;
5472 error = ofputil_decode_queue_get_config_request(oh, &request);
5477 ofport = ofproto_get_port(p, request);
5479 return OFPERR_OFPQOFC_BAD_PORT;
5482 reply = ofputil_encode_queue_get_config_reply(oh);
5484 smap_init(&details);
5485 NETDEV_QUEUE_FOR_EACH (&queue_id, &details, &queue_dump, ofport->netdev) {
5486 struct ofputil_queue_config queue;
5488 /* None of the existing queues have compatible properties, so we
5489 * hard-code omitting min_rate and max_rate. */
5490 queue.queue_id = queue_id;
5491 queue.min_rate = UINT16_MAX;
5492 queue.max_rate = UINT16_MAX;
5493 ofputil_append_queue_get_config_reply(reply, &queue);
5495 smap_destroy(&details);
5497 ofconn_send_reply(ofconn, reply);
5502 /* Implements OFPGC11_ADD
5503 * in which no matching flow already exists in the flow table.
5505 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
5506 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
5507 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
5508 * initiated now but may be retried later.
5510 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
5511 * ownership remains with the caller.
5513 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
5516 add_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
5518 struct ofgroup *ofgroup;
5521 if (gm->group_id > OFPG_MAX) {
5522 return OFPERR_OFPGMFC_INVALID_GROUP;
5524 if (gm->type > OFPGT11_FF) {
5525 return OFPERR_OFPGMFC_BAD_TYPE;
5528 /* Allocate new group and initialize it. */
5529 ofgroup = ofproto->ofproto_class->group_alloc();
5531 VLOG_WARN_RL(&rl, "%s: failed to create group", ofproto->name);
5532 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
5535 ovs_rwlock_init(&ofgroup->rwlock);
5536 ofgroup->ofproto = ofproto;
5537 ofgroup->group_id = gm->group_id;
5538 ofgroup->type = gm->type;
5539 ofgroup->created = ofgroup->modified = time_msec();
5541 list_move(&ofgroup->buckets, &gm->buckets);
5542 ofgroup->n_buckets = list_size(&ofgroup->buckets);
5544 /* Construct called BEFORE any locks are held. */
5545 error = ofproto->ofproto_class->group_construct(ofgroup);
5550 /* We wrlock as late as possible to minimize the time we jam any other
5551 * threads: No visible state changes before acquiring the lock. */
5552 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5554 if (ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
5555 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
5559 if (ofproto_group_exists__(ofproto, gm->group_id)) {
5560 error = OFPERR_OFPGMFC_GROUP_EXISTS;
5565 /* Insert new group. */
5566 hmap_insert(&ofproto->groups, &ofgroup->hmap_node,
5567 hash_int(ofgroup->group_id, 0));
5568 ofproto->n_groups[ofgroup->type]++;
5570 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5575 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5576 ofproto->ofproto_class->group_destruct(ofgroup);
5578 ofputil_bucket_list_destroy(&ofgroup->buckets);
5579 ofproto->ofproto_class->group_dealloc(ofgroup);
5584 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
5587 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
5590 modify_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
5592 struct ofgroup *ofgroup;
5593 struct ofgroup *victim;
5596 if (gm->group_id > OFPG_MAX) {
5597 return OFPERR_OFPGMFC_INVALID_GROUP;
5600 if (gm->type > OFPGT11_FF) {
5601 return OFPERR_OFPGMFC_BAD_TYPE;
5604 victim = ofproto->ofproto_class->group_alloc();
5606 VLOG_WARN_RL(&rl, "%s: failed to allocate group", ofproto->name);
5607 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
5610 if (!ofproto_group_write_lookup(ofproto, gm->group_id, &ofgroup)) {
5611 error = OFPERR_OFPGMFC_UNKNOWN_GROUP;
5614 /* Both group's and its container's write locks held now.
5615 * Also, n_groups[] is protected by ofproto->groups_rwlock. */
5616 if (ofgroup->type != gm->type
5617 && ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
5618 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
5623 list_move(&victim->buckets, &ofgroup->buckets);
5625 ofgroup->type = gm->type;
5626 list_move(&ofgroup->buckets, &gm->buckets);
5627 ofgroup->n_buckets = list_size(&ofgroup->buckets);
5629 error = ofproto->ofproto_class->group_modify(ofgroup, victim);
5631 ofputil_bucket_list_destroy(&victim->buckets);
5632 ofproto->n_groups[victim->type]--;
5633 ofproto->n_groups[ofgroup->type]++;
5634 ofgroup->modified = time_msec();
5636 ofputil_bucket_list_destroy(&ofgroup->buckets);
5639 list_move(&ofgroup->buckets, &victim->buckets);
5643 ovs_rwlock_unlock(&ofgroup->rwlock);
5644 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5646 ofproto->ofproto_class->group_dealloc(victim);
5651 delete_group__(struct ofproto *ofproto, struct ofgroup *ofgroup)
5652 OVS_RELEASES(ofproto->groups_rwlock)
5655 struct ofputil_flow_mod fm;
5657 /* Delete all flow entries containing this group in a group action */
5658 match_init_catchall(&match);
5659 flow_mod_init(&fm, &match, 0, NULL, 0, OFPFC_DELETE);
5660 fm.out_group = ofgroup->group_id;
5661 handle_flow_mod__(ofproto, NULL, &fm, NULL);
5663 /* Must wait until existing readers are done,
5664 * while holding the container's write lock at the same time. */
5665 ovs_rwlock_wrlock(&ofgroup->rwlock);
5666 hmap_remove(&ofproto->groups, &ofgroup->hmap_node);
5667 /* No-one can find this group any more. */
5668 ofproto->n_groups[ofgroup->type]--;
5669 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5671 ofproto->ofproto_class->group_destruct(ofgroup);
5672 ofputil_bucket_list_destroy(&ofgroup->buckets);
5673 ovs_rwlock_unlock(&ofgroup->rwlock);
5674 ovs_rwlock_destroy(&ofgroup->rwlock);
5675 ofproto->ofproto_class->group_dealloc(ofgroup);
5678 /* Implements OFPGC_DELETE. */
5680 delete_group(struct ofproto *ofproto, uint32_t group_id)
5682 struct ofgroup *ofgroup;
5684 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5685 if (group_id == OFPG_ALL) {
5687 struct hmap_node *node = hmap_first(&ofproto->groups);
5691 ofgroup = CONTAINER_OF(node, struct ofgroup, hmap_node);
5692 delete_group__(ofproto, ofgroup);
5693 /* Lock for each node separately, so that we will not jam the
5694 * other threads for too long time. */
5695 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5698 HMAP_FOR_EACH_IN_BUCKET (ofgroup, hmap_node,
5699 hash_int(group_id, 0), &ofproto->groups) {
5700 if (ofgroup->group_id == group_id) {
5701 delete_group__(ofproto, ofgroup);
5706 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5710 handle_group_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5712 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5713 struct ofputil_group_mod gm;
5716 error = reject_slave_controller(ofconn);
5721 error = ofputil_decode_group_mod(oh, &gm);
5726 switch (gm.command) {
5728 return add_group(ofproto, &gm);
5730 case OFPGC11_MODIFY:
5731 return modify_group(ofproto, &gm);
5733 case OFPGC11_DELETE:
5734 delete_group(ofproto, gm.group_id);
5738 if (gm.command > OFPGC11_DELETE) {
5739 VLOG_WARN_RL(&rl, "%s: Invalid group_mod command type %d",
5740 ofproto->name, gm.command);
5742 return OFPERR_OFPGMFC_BAD_COMMAND;
5747 handle_table_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5749 struct ofputil_table_mod tm;
5752 error = reject_slave_controller(ofconn);
5757 error = ofputil_decode_table_mod(oh, &tm);
5762 /* XXX Actual table mod support is not implemented yet. */
5767 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
5768 OVS_EXCLUDED(ofproto_mutex)
5770 const struct ofp_header *oh = msg->data;
5774 error = ofptype_decode(&type, oh);
5778 if (oh->version >= OFP13_VERSION && ofpmsg_is_stat_request(oh)
5779 && ofpmp_more(oh)) {
5780 /* We have no buffer implementation for multipart requests.
5781 * Report overflow for requests which consists of multiple
5783 return OFPERR_OFPBRC_MULTIPART_BUFFER_OVERFLOW;
5787 /* OpenFlow requests. */
5788 case OFPTYPE_ECHO_REQUEST:
5789 return handle_echo_request(ofconn, oh);
5791 case OFPTYPE_FEATURES_REQUEST:
5792 return handle_features_request(ofconn, oh);
5794 case OFPTYPE_GET_CONFIG_REQUEST:
5795 return handle_get_config_request(ofconn, oh);
5797 case OFPTYPE_SET_CONFIG:
5798 return handle_set_config(ofconn, oh);
5800 case OFPTYPE_PACKET_OUT:
5801 return handle_packet_out(ofconn, oh);
5803 case OFPTYPE_PORT_MOD:
5804 return handle_port_mod(ofconn, oh);
5806 case OFPTYPE_FLOW_MOD:
5807 return handle_flow_mod(ofconn, oh);
5809 case OFPTYPE_GROUP_MOD:
5810 return handle_group_mod(ofconn, oh);
5812 case OFPTYPE_TABLE_MOD:
5813 return handle_table_mod(ofconn, oh);
5815 case OFPTYPE_METER_MOD:
5816 return handle_meter_mod(ofconn, oh);
5818 case OFPTYPE_BARRIER_REQUEST:
5819 return handle_barrier_request(ofconn, oh);
5821 case OFPTYPE_ROLE_REQUEST:
5822 return handle_role_request(ofconn, oh);
5824 /* OpenFlow replies. */
5825 case OFPTYPE_ECHO_REPLY:
5828 /* Nicira extension requests. */
5829 case OFPTYPE_FLOW_MOD_TABLE_ID:
5830 return handle_nxt_flow_mod_table_id(ofconn, oh);
5832 case OFPTYPE_SET_FLOW_FORMAT:
5833 return handle_nxt_set_flow_format(ofconn, oh);
5835 case OFPTYPE_SET_PACKET_IN_FORMAT:
5836 return handle_nxt_set_packet_in_format(ofconn, oh);
5838 case OFPTYPE_SET_CONTROLLER_ID:
5839 return handle_nxt_set_controller_id(ofconn, oh);
5841 case OFPTYPE_FLOW_AGE:
5842 /* Nothing to do. */
5845 case OFPTYPE_FLOW_MONITOR_CANCEL:
5846 return handle_flow_monitor_cancel(ofconn, oh);
5848 case OFPTYPE_SET_ASYNC_CONFIG:
5849 return handle_nxt_set_async_config(ofconn, oh);
5851 case OFPTYPE_GET_ASYNC_REQUEST:
5852 return handle_nxt_get_async_request(ofconn, oh);
5854 /* Statistics requests. */
5855 case OFPTYPE_DESC_STATS_REQUEST:
5856 return handle_desc_stats_request(ofconn, oh);
5858 case OFPTYPE_FLOW_STATS_REQUEST:
5859 return handle_flow_stats_request(ofconn, oh);
5861 case OFPTYPE_AGGREGATE_STATS_REQUEST:
5862 return handle_aggregate_stats_request(ofconn, oh);
5864 case OFPTYPE_TABLE_STATS_REQUEST:
5865 return handle_table_stats_request(ofconn, oh);
5867 case OFPTYPE_PORT_STATS_REQUEST:
5868 return handle_port_stats_request(ofconn, oh);
5870 case OFPTYPE_QUEUE_STATS_REQUEST:
5871 return handle_queue_stats_request(ofconn, oh);
5873 case OFPTYPE_PORT_DESC_STATS_REQUEST:
5874 return handle_port_desc_stats_request(ofconn, oh);
5876 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
5877 return handle_flow_monitor_request(ofconn, oh);
5879 case OFPTYPE_METER_STATS_REQUEST:
5880 case OFPTYPE_METER_CONFIG_STATS_REQUEST:
5881 return handle_meter_request(ofconn, oh, type);
5883 case OFPTYPE_METER_FEATURES_STATS_REQUEST:
5884 return handle_meter_features_request(ofconn, oh);
5886 case OFPTYPE_GROUP_STATS_REQUEST:
5887 return handle_group_stats_request(ofconn, oh);
5889 case OFPTYPE_GROUP_DESC_STATS_REQUEST:
5890 return handle_group_desc_stats_request(ofconn, oh);
5892 case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
5893 return handle_group_features_stats_request(ofconn, oh);
5895 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
5896 return handle_queue_get_config_request(ofconn, oh);
5900 case OFPTYPE_FEATURES_REPLY:
5901 case OFPTYPE_GET_CONFIG_REPLY:
5902 case OFPTYPE_PACKET_IN:
5903 case OFPTYPE_FLOW_REMOVED:
5904 case OFPTYPE_PORT_STATUS:
5905 case OFPTYPE_BARRIER_REPLY:
5906 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
5907 case OFPTYPE_DESC_STATS_REPLY:
5908 case OFPTYPE_FLOW_STATS_REPLY:
5909 case OFPTYPE_QUEUE_STATS_REPLY:
5910 case OFPTYPE_PORT_STATS_REPLY:
5911 case OFPTYPE_TABLE_STATS_REPLY:
5912 case OFPTYPE_AGGREGATE_STATS_REPLY:
5913 case OFPTYPE_PORT_DESC_STATS_REPLY:
5914 case OFPTYPE_ROLE_REPLY:
5915 case OFPTYPE_FLOW_MONITOR_PAUSED:
5916 case OFPTYPE_FLOW_MONITOR_RESUMED:
5917 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
5918 case OFPTYPE_GET_ASYNC_REPLY:
5919 case OFPTYPE_GROUP_STATS_REPLY:
5920 case OFPTYPE_GROUP_DESC_STATS_REPLY:
5921 case OFPTYPE_GROUP_FEATURES_STATS_REPLY:
5922 case OFPTYPE_METER_STATS_REPLY:
5923 case OFPTYPE_METER_CONFIG_STATS_REPLY:
5924 case OFPTYPE_METER_FEATURES_STATS_REPLY:
5925 case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
5926 case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
5927 case OFPTYPE_ROLE_STATUS:
5929 if (ofpmsg_is_stat_request(oh)) {
5930 return OFPERR_OFPBRC_BAD_STAT;
5932 return OFPERR_OFPBRC_BAD_TYPE;
5938 handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
5939 OVS_EXCLUDED(ofproto_mutex)
5941 int error = handle_openflow__(ofconn, ofp_msg);
5942 if (error && error != OFPROTO_POSTPONE) {
5943 ofconn_send_error(ofconn, ofp_msg->data, error);
5945 COVERAGE_INC(ofproto_recv_openflow);
5946 return error != OFPROTO_POSTPONE;
5949 /* Asynchronous operations. */
5951 /* Creates and returns a new ofopgroup that is not associated with any
5952 * OpenFlow connection.
5954 * The caller should add operations to the returned group with
5955 * ofoperation_create() and then submit it with ofopgroup_submit(). */
5956 static struct ofopgroup *
5957 ofopgroup_create_unattached(struct ofproto *ofproto)
5958 OVS_REQUIRES(ofproto_mutex)
5960 struct ofopgroup *group = xzalloc(sizeof *group);
5961 group->ofproto = ofproto;
5962 list_init(&group->ofproto_node);
5963 list_init(&group->ops);
5964 list_init(&group->ofconn_node);
5968 /* Creates and returns a new ofopgroup for 'ofproto'.
5970 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
5971 * connection. The 'request' and 'buffer_id' arguments are ignored.
5973 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
5974 * If the ofopgroup eventually fails, then the error reply will include
5975 * 'request'. If the ofopgroup eventually succeeds, then the packet with
5976 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
5978 * The caller should add operations to the returned group with
5979 * ofoperation_create() and then submit it with ofopgroup_submit(). */
5980 static struct ofopgroup *
5981 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
5982 const struct ofp_header *request, uint32_t buffer_id)
5983 OVS_REQUIRES(ofproto_mutex)
5985 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
5987 size_t request_len = ntohs(request->length);
5989 ovs_assert(ofconn_get_ofproto(ofconn) == ofproto);
5991 ofconn_add_opgroup(ofconn, &group->ofconn_node);
5992 group->ofconn = ofconn;
5993 group->request = xmemdup(request, MIN(request_len, 64));
5994 group->buffer_id = buffer_id;
5999 /* Submits 'group' for processing.
6001 * If 'group' contains no operations (e.g. none were ever added, or all of the
6002 * ones that were added completed synchronously), then it is destroyed
6003 * immediately. Otherwise it is added to the ofproto's list of pending
6006 ofopgroup_submit(struct ofopgroup *group)
6007 OVS_REQUIRES(ofproto_mutex)
6009 if (!group->n_running) {
6010 ofopgroup_complete(group);
6012 list_push_back(&group->ofproto->pending, &group->ofproto_node);
6013 group->ofproto->n_pending++;
6018 ofopgroup_complete(struct ofopgroup *group)
6019 OVS_REQUIRES(ofproto_mutex)
6021 struct ofproto *ofproto = group->ofproto;
6023 struct ofconn *abbrev_ofconn;
6024 ovs_be32 abbrev_xid;
6026 struct ofoperation *op, *next_op;
6029 ovs_assert(!group->n_running);
6032 LIST_FOR_EACH (op, group_node, &group->ops) {
6039 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
6040 LIST_FOR_EACH (op, group_node, &group->ops) {
6041 if (op->type != OFOPERATION_DELETE) {
6042 struct ofpbuf *packet;
6045 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
6048 struct rule_execute *re;
6052 ofproto_rule_ref(op->rule);
6054 re = xmalloc(sizeof *re);
6055 re->rule = op->rule;
6056 re->in_port = in_port;
6057 re->packet = packet;
6059 if (!guarded_list_push_back(&ofproto->rule_executes,
6060 &re->list_node, 1024)) {
6061 ofproto_rule_unref(op->rule);
6062 ofpbuf_delete(re->packet);
6071 if (!error && !list_is_empty(&group->ofconn_node)) {
6072 abbrev_ofconn = group->ofconn;
6073 abbrev_xid = group->request->xid;
6075 abbrev_ofconn = NULL;
6076 abbrev_xid = htonl(0);
6078 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
6079 struct rule *rule = op->rule;
6081 /* We generally want to report the change to active OpenFlow flow
6082 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
6084 - The operation failed.
6086 - The affected rule is not visible to controllers.
6088 - The operation's only effect was to update rule->modified. */
6090 || ofproto_rule_is_hidden(rule)
6091 || (op->type == OFOPERATION_MODIFY
6093 && rule->flow_cookie == op->flow_cookie))) {
6094 /* Check that we can just cast from ofoperation_type to
6095 * nx_flow_update_event. */
6096 enum nx_flow_update_event event_type;
6099 case OFOPERATION_ADD:
6100 case OFOPERATION_REPLACE:
6101 event_type = NXFME_ADDED;
6104 case OFOPERATION_DELETE:
6105 event_type = NXFME_DELETED;
6108 case OFOPERATION_MODIFY:
6109 event_type = NXFME_MODIFIED;
6116 ofmonitor_report(ofproto->connmgr, rule, event_type,
6117 op->reason, abbrev_ofconn, abbrev_xid);
6120 rule->pending = NULL;
6123 case OFOPERATION_ADD:
6127 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
6128 if (vid_mask == VLAN_VID_MASK) {
6129 if (ofproto->vlan_bitmap) {
6130 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
6131 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
6132 bitmap_set1(ofproto->vlan_bitmap, vid);
6133 ofproto->vlans_changed = true;
6136 ofproto->vlans_changed = true;
6140 oftable_remove_rule(rule);
6141 ofproto_rule_unref(rule);
6145 case OFOPERATION_DELETE:
6146 ovs_assert(!op->error);
6147 ofproto_rule_unref(rule);
6151 case OFOPERATION_MODIFY:
6152 case OFOPERATION_REPLACE:
6154 long long int now = time_msec();
6156 rule->modified = now;
6157 if (op->type == OFOPERATION_REPLACE) {
6158 rule->created = rule->used = now;
6161 ofproto_rule_change_cookie(ofproto, rule, op->flow_cookie);
6162 ovs_mutex_lock(&rule->mutex);
6163 rule->idle_timeout = op->idle_timeout;
6164 rule->hard_timeout = op->hard_timeout;
6165 ovs_mutex_unlock(&rule->mutex);
6167 struct rule_actions *old_actions;
6169 ovs_mutex_lock(&rule->mutex);
6170 old_actions = rule->actions;
6171 rule->actions = op->actions;
6172 ovs_mutex_unlock(&rule->mutex);
6175 rule_actions_unref(old_actions);
6177 rule->flags = op->flags;
6185 ofoperation_destroy(op);
6188 ofmonitor_flush(ofproto->connmgr);
6190 if (!list_is_empty(&group->ofproto_node)) {
6191 ovs_assert(ofproto->n_pending > 0);
6192 ofproto->n_pending--;
6193 list_remove(&group->ofproto_node);
6195 if (!list_is_empty(&group->ofconn_node)) {
6196 list_remove(&group->ofconn_node);
6198 ofconn_send_error(group->ofconn, group->request, error);
6200 connmgr_retry(ofproto->connmgr);
6202 free(group->request);
6206 /* Initiates a new operation on 'rule', of the specified 'type', within
6207 * 'group'. Prior to calling, 'rule' must not have any pending operation.
6209 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
6210 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
6212 * Returns the newly created ofoperation (which is also available as
6213 * rule->pending). */
6214 static struct ofoperation *
6215 ofoperation_create(struct ofopgroup *group, struct rule *rule,
6216 enum ofoperation_type type,
6217 enum ofp_flow_removed_reason reason)
6218 OVS_REQUIRES(ofproto_mutex)
6220 struct ofproto *ofproto = group->ofproto;
6221 struct ofoperation *op;
6223 ovs_assert(!rule->pending);
6225 op = rule->pending = xzalloc(sizeof *op);
6227 list_push_back(&group->ops, &op->group_node);
6230 op->reason = reason;
6231 op->flow_cookie = rule->flow_cookie;
6232 ovs_mutex_lock(&rule->mutex);
6233 op->idle_timeout = rule->idle_timeout;
6234 op->hard_timeout = rule->hard_timeout;
6235 ovs_mutex_unlock(&rule->mutex);
6236 op->flags = rule->flags;
6240 if (type == OFOPERATION_DELETE) {
6241 hmap_insert(&ofproto->deletions, &op->hmap_node,
6242 cls_rule_hash(&rule->cr, rule->table_id));
6249 ofoperation_destroy(struct ofoperation *op)
6250 OVS_REQUIRES(ofproto_mutex)
6252 struct ofopgroup *group = op->group;
6255 op->rule->pending = NULL;
6257 if (op->type == OFOPERATION_DELETE) {
6258 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
6260 list_remove(&op->group_node);
6261 rule_actions_unref(op->actions);
6265 /* Indicates that 'op' completed with status 'error', which is either 0 to
6266 * indicate success or an OpenFlow error code on failure.
6268 * If 'error' is 0, indicating success, the operation will be committed
6269 * permanently to the flow table.
6271 * If 'error' is nonzero, then generally the operation will be rolled back:
6273 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
6274 * restores the original rule. The caller must have uninitialized any
6275 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
6276 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
6277 * and 7 for the new rule, calling its ->rule_dealloc() function.
6279 * - If 'op' is a "modify flow" operation, ofproto restores the original
6282 * - 'op' must not be a "delete flow" operation. Removing a rule is not
6283 * allowed to fail. It must always succeed.
6285 * Please see the large comment in ofproto/ofproto-provider.h titled
6286 * "Asynchronous Operation Support" for more information. */
6288 ofoperation_complete(struct ofoperation *op, enum ofperr error)
6290 struct ofopgroup *group = op->group;
6292 ovs_assert(group->n_running > 0);
6293 ovs_assert(!error || op->type != OFOPERATION_DELETE);
6296 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
6297 /* This function can be called from ->rule_construct(), in which case
6298 * ofproto_mutex is held, or it can be called from ->run(), in which
6299 * case ofproto_mutex is not held. But only in the latter case can we
6300 * arrive here, so we can safely take ofproto_mutex now. */
6301 ovs_mutex_lock(&ofproto_mutex);
6302 ovs_assert(op->rule->pending == op);
6303 ofopgroup_complete(group);
6304 ovs_mutex_unlock(&ofproto_mutex);
6309 pick_datapath_id(const struct ofproto *ofproto)
6311 const struct ofport *port;
6313 port = ofproto_get_port(ofproto, OFPP_LOCAL);
6315 uint8_t ea[ETH_ADDR_LEN];
6318 error = netdev_get_etheraddr(port->netdev, ea);
6320 return eth_addr_to_uint64(ea);
6322 VLOG_WARN("%s: could not get MAC address for %s (%s)",
6323 ofproto->name, netdev_get_name(port->netdev),
6324 ovs_strerror(error));
6326 return ofproto->fallback_dpid;
6330 pick_fallback_dpid(void)
6332 uint8_t ea[ETH_ADDR_LEN];
6333 eth_addr_nicira_random(ea);
6334 return eth_addr_to_uint64(ea);
6337 /* Table overflow policy. */
6339 /* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep'
6340 * to NULL if the table is not configured to evict rules or if the table
6341 * contains no evictable rules. (Rules with a readlock on their evict rwlock,
6342 * or with no timeouts are not evictable.) */
6344 choose_rule_to_evict(struct oftable *table, struct rule **rulep)
6345 OVS_REQUIRES(ofproto_mutex)
6347 struct eviction_group *evg;
6350 if (!table->eviction_fields) {
6354 /* In the common case, the outer and inner loops here will each be entered
6357 * - The inner loop normally "return"s in its first iteration. If the
6358 * eviction group has any evictable rules, then it always returns in
6361 * - The outer loop only iterates more than once if the largest eviction
6362 * group has no evictable rules.
6364 * - The outer loop can exit only if table's 'max_flows' is all filled up
6365 * by unevictable rules. */
6366 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
6369 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
6378 /* Searches 'ofproto' for tables that have more flows than their configured
6379 * maximum and that have flow eviction enabled, and evicts as many flows as
6380 * necessary and currently feasible from them.
6382 * This triggers only when an OpenFlow table has N flows in it and then the
6383 * client configures a maximum number of flows less than N. */
6385 ofproto_evict(struct ofproto *ofproto)
6387 struct oftable *table;
6389 ovs_mutex_lock(&ofproto_mutex);
6390 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
6391 evict_rules_from_table(ofproto, table, 0);
6393 ovs_mutex_unlock(&ofproto_mutex);
6396 /* Eviction groups. */
6398 /* Returns the priority to use for an eviction_group that contains 'n_rules'
6399 * rules. The priority contains low-order random bits to ensure that eviction
6400 * groups with the same number of rules are prioritized randomly. */
6402 eviction_group_priority(size_t n_rules)
6404 uint16_t size = MIN(UINT16_MAX, n_rules);
6405 return (size << 16) | random_uint16();
6408 /* Updates 'evg', an eviction_group within 'table', following a change that
6409 * adds or removes rules in 'evg'. */
6411 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
6412 OVS_REQUIRES(ofproto_mutex)
6414 heap_change(&table->eviction_groups_by_size, &evg->size_node,
6415 eviction_group_priority(heap_count(&evg->rules)));
6418 /* Destroys 'evg', an eviction_group within 'table':
6420 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
6421 * rules themselves, just removes them from the eviction group.)
6423 * - Removes 'evg' from 'table'.
6427 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
6428 OVS_REQUIRES(ofproto_mutex)
6430 while (!heap_is_empty(&evg->rules)) {
6433 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
6434 rule->eviction_group = NULL;
6436 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
6437 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
6438 heap_destroy(&evg->rules);
6442 /* Removes 'rule' from its eviction group, if any. */
6444 eviction_group_remove_rule(struct rule *rule)
6445 OVS_REQUIRES(ofproto_mutex)
6447 if (rule->eviction_group) {
6448 struct oftable *table = &rule->ofproto->tables[rule->table_id];
6449 struct eviction_group *evg = rule->eviction_group;
6451 rule->eviction_group = NULL;
6452 heap_remove(&evg->rules, &rule->evg_node);
6453 if (heap_is_empty(&evg->rules)) {
6454 eviction_group_destroy(table, evg);
6456 eviction_group_resized(table, evg);
6461 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
6462 * returns the hash value. */
6464 eviction_group_hash_rule(struct rule *rule)
6465 OVS_REQUIRES(ofproto_mutex)
6467 struct oftable *table = &rule->ofproto->tables[rule->table_id];
6468 const struct mf_subfield *sf;
6472 hash = table->eviction_group_id_basis;
6473 miniflow_expand(&rule->cr.match.flow, &flow);
6474 for (sf = table->eviction_fields;
6475 sf < &table->eviction_fields[table->n_eviction_fields];
6478 if (mf_are_prereqs_ok(sf->field, &flow)) {
6479 union mf_value value;
6481 mf_get_value(sf->field, &flow, &value);
6483 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
6485 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
6486 unsigned int start = sf->ofs + sf->n_bits;
6487 bitwise_zero(&value, sf->field->n_bytes, start,
6488 sf->field->n_bytes * 8 - start);
6490 hash = hash_bytes(&value, sf->field->n_bytes, hash);
6492 hash = hash_int(hash, 0);
6499 /* Returns an eviction group within 'table' with the given 'id', creating one
6501 static struct eviction_group *
6502 eviction_group_find(struct oftable *table, uint32_t id)
6503 OVS_REQUIRES(ofproto_mutex)
6505 struct eviction_group *evg;
6507 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
6511 evg = xmalloc(sizeof *evg);
6512 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
6513 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
6514 eviction_group_priority(0));
6515 heap_init(&evg->rules);
6520 /* Returns an eviction priority for 'rule'. The return value should be
6521 * interpreted so that higher priorities make a rule more attractive candidates
6524 rule_eviction_priority(struct rule *rule)
6525 OVS_REQUIRES(ofproto_mutex)
6527 long long int hard_expiration;
6528 long long int idle_expiration;
6529 long long int expiration;
6530 uint32_t expiration_offset;
6532 /* Calculate time of expiration. */
6533 ovs_mutex_lock(&rule->mutex);
6534 hard_expiration = (rule->hard_timeout
6535 ? rule->modified + rule->hard_timeout * 1000
6537 idle_expiration = (rule->idle_timeout
6538 ? rule->used + rule->idle_timeout * 1000
6540 expiration = MIN(hard_expiration, idle_expiration);
6541 ovs_mutex_unlock(&rule->mutex);
6542 if (expiration == LLONG_MAX) {
6546 /* Calculate the time of expiration as a number of (approximate) seconds
6547 * after program startup.
6549 * This should work OK for program runs that last UINT32_MAX seconds or
6550 * less. Therefore, please restart OVS at least once every 136 years. */
6551 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
6553 /* Invert the expiration offset because we're using a max-heap. */
6554 return UINT32_MAX - expiration_offset;
6557 /* Adds 'rule' to an appropriate eviction group for its oftable's
6558 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
6559 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
6562 * The caller must ensure that 'rule' is not already in an eviction group. */
6564 eviction_group_add_rule(struct rule *rule)
6565 OVS_REQUIRES(ofproto_mutex)
6567 struct ofproto *ofproto = rule->ofproto;
6568 struct oftable *table = &ofproto->tables[rule->table_id];
6571 ovs_mutex_lock(&rule->mutex);
6572 has_timeout = rule->hard_timeout || rule->idle_timeout;
6573 ovs_mutex_unlock(&rule->mutex);
6575 if (table->eviction_fields && has_timeout) {
6576 struct eviction_group *evg;
6578 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
6580 rule->eviction_group = evg;
6581 heap_insert(&evg->rules, &rule->evg_node,
6582 rule_eviction_priority(rule));
6583 eviction_group_resized(table, evg);
6589 /* Initializes 'table'. */
6591 oftable_init(struct oftable *table)
6593 memset(table, 0, sizeof *table);
6594 classifier_init(&table->cls);
6595 table->max_flows = UINT_MAX;
6598 /* Destroys 'table', including its classifier and eviction groups.
6600 * The caller is responsible for freeing 'table' itself. */
6602 oftable_destroy(struct oftable *table)
6604 ovs_rwlock_rdlock(&table->cls.rwlock);
6605 ovs_assert(classifier_is_empty(&table->cls));
6606 ovs_rwlock_unlock(&table->cls.rwlock);
6607 oftable_disable_eviction(table);
6608 classifier_destroy(&table->cls);
6612 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
6613 * string, then 'table' will use its default name.
6615 * This only affects the name exposed for a table exposed through the OpenFlow
6616 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
6618 oftable_set_name(struct oftable *table, const char *name)
6620 if (name && name[0]) {
6621 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
6622 if (!table->name || strncmp(name, table->name, len)) {
6624 table->name = xmemdup0(name, len);
6632 /* oftables support a choice of two policies when adding a rule would cause the
6633 * number of flows in the table to exceed the configured maximum number: either
6634 * they can refuse to add the new flow or they can evict some existing flow.
6635 * This function configures the former policy on 'table'. */
6637 oftable_disable_eviction(struct oftable *table)
6638 OVS_REQUIRES(ofproto_mutex)
6640 if (table->eviction_fields) {
6641 struct eviction_group *evg, *next;
6643 HMAP_FOR_EACH_SAFE (evg, next, id_node,
6644 &table->eviction_groups_by_id) {
6645 eviction_group_destroy(table, evg);
6647 hmap_destroy(&table->eviction_groups_by_id);
6648 heap_destroy(&table->eviction_groups_by_size);
6650 free(table->eviction_fields);
6651 table->eviction_fields = NULL;
6652 table->n_eviction_fields = 0;
6656 /* oftables support a choice of two policies when adding a rule would cause the
6657 * number of flows in the table to exceed the configured maximum number: either
6658 * they can refuse to add the new flow or they can evict some existing flow.
6659 * This function configures the latter policy on 'table', with fairness based
6660 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
6661 * 'n_fields' as 0 disables fairness.) */
6663 oftable_enable_eviction(struct oftable *table,
6664 const struct mf_subfield *fields, size_t n_fields)
6665 OVS_REQUIRES(ofproto_mutex)
6667 struct cls_cursor cursor;
6670 if (table->eviction_fields
6671 && n_fields == table->n_eviction_fields
6673 || !memcmp(fields, table->eviction_fields,
6674 n_fields * sizeof *fields))) {
6679 oftable_disable_eviction(table);
6681 table->n_eviction_fields = n_fields;
6682 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
6684 table->eviction_group_id_basis = random_uint32();
6685 hmap_init(&table->eviction_groups_by_id);
6686 heap_init(&table->eviction_groups_by_size);
6688 ovs_rwlock_rdlock(&table->cls.rwlock);
6689 cls_cursor_init(&cursor, &table->cls, NULL);
6690 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
6691 eviction_group_add_rule(rule);
6693 ovs_rwlock_unlock(&table->cls.rwlock);
6696 /* Removes 'rule' from the oftable that contains it. */
6698 oftable_remove_rule__(struct ofproto *ofproto, struct rule *rule)
6699 OVS_REQUIRES(ofproto_mutex)
6701 struct classifier *cls = &ofproto->tables[rule->table_id].cls;
6703 ovs_rwlock_wrlock(&cls->rwlock);
6704 classifier_remove(cls, CONST_CAST(struct cls_rule *, &rule->cr));
6705 ovs_rwlock_unlock(&cls->rwlock);
6707 cookies_remove(ofproto, rule);
6709 eviction_group_remove_rule(rule);
6710 if (!list_is_empty(&rule->expirable)) {
6711 list_remove(&rule->expirable);
6713 if (!list_is_empty(&rule->meter_list_node)) {
6714 list_remove(&rule->meter_list_node);
6715 list_init(&rule->meter_list_node);
6720 oftable_remove_rule(struct rule *rule)
6721 OVS_REQUIRES(ofproto_mutex)
6723 oftable_remove_rule__(rule->ofproto, rule);
6726 /* Inserts 'rule' into its oftable, which must not already contain any rule for
6727 * the same cls_rule. */
6729 oftable_insert_rule(struct rule *rule)
6730 OVS_REQUIRES(ofproto_mutex)
6732 struct ofproto *ofproto = rule->ofproto;
6733 struct oftable *table = &ofproto->tables[rule->table_id];
6736 ovs_mutex_lock(&rule->mutex);
6737 may_expire = rule->hard_timeout || rule->idle_timeout;
6738 ovs_mutex_unlock(&rule->mutex);
6741 list_insert(&ofproto->expirable, &rule->expirable);
6744 cookies_insert(ofproto, rule);
6746 if (rule->actions->provider_meter_id != UINT32_MAX) {
6747 uint32_t meter_id = ofpacts_get_meter(rule->actions->ofpacts,
6748 rule->actions->ofpacts_len);
6749 struct meter *meter = ofproto->meters[meter_id];
6750 list_insert(&meter->rules, &rule->meter_list_node);
6752 ovs_rwlock_wrlock(&table->cls.rwlock);
6753 classifier_insert(&table->cls, CONST_CAST(struct cls_rule *, &rule->cr));
6754 ovs_rwlock_unlock(&table->cls.rwlock);
6755 eviction_group_add_rule(rule);
6758 /* unixctl commands. */
6761 ofproto_lookup(const char *name)
6763 struct ofproto *ofproto;
6765 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
6767 if (!strcmp(ofproto->name, name)) {
6775 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
6776 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6778 struct ofproto *ofproto;
6782 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
6783 ds_put_format(&results, "%s\n", ofproto->name);
6785 unixctl_command_reply(conn, ds_cstr(&results));
6786 ds_destroy(&results);
6790 ofproto_unixctl_init(void)
6792 static bool registered;
6798 unixctl_command_register("ofproto/list", "", 0, 0,
6799 ofproto_unixctl_list, NULL);
6802 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6804 * This is deprecated. It is only for compatibility with broken device drivers
6805 * in old versions of Linux that do not properly support VLANs when VLAN
6806 * devices are not used. When broken device drivers are no longer in
6807 * widespread use, we will delete these interfaces. */
6809 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
6810 * (exactly) by an OpenFlow rule in 'ofproto'. */
6812 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
6814 const struct oftable *oftable;
6816 free(ofproto->vlan_bitmap);
6817 ofproto->vlan_bitmap = bitmap_allocate(4096);
6818 ofproto->vlans_changed = false;
6820 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
6821 const struct cls_subtable *table;
6823 ovs_rwlock_rdlock(&oftable->cls.rwlock);
6824 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.subtables) {
6825 if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
6826 const struct cls_rule *rule;
6828 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
6829 uint16_t vid = miniflow_get_vid(&rule->match.flow);
6830 bitmap_set1(vlan_bitmap, vid);
6831 bitmap_set1(ofproto->vlan_bitmap, vid);
6835 ovs_rwlock_unlock(&oftable->cls.rwlock);
6839 /* Returns true if new VLANs have come into use by the flow table since the
6840 * last call to ofproto_get_vlan_usage().
6842 * We don't track when old VLANs stop being used. */
6844 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
6846 return ofproto->vlans_changed;
6849 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
6850 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
6851 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
6852 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
6853 * then the VLAN device is un-enslaved. */
6855 ofproto_port_set_realdev(struct ofproto *ofproto, ofp_port_t vlandev_ofp_port,
6856 ofp_port_t realdev_ofp_port, int vid)
6858 struct ofport *ofport;
6861 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
6863 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
6865 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
6866 ofproto->name, vlandev_ofp_port);
6870 if (!ofproto->ofproto_class->set_realdev) {
6871 if (!vlandev_ofp_port) {
6874 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
6878 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
6880 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
6881 ofproto->name, vlandev_ofp_port,
6882 netdev_get_name(ofport->netdev), ovs_strerror(error));