2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
26 #include "byte-order.h"
27 #include "classifier.h"
30 #include "dynamic-string.h"
33 #include "meta-flow.h"
36 #include "ofp-actions.h"
37 #include "ofp-errors.h"
39 #include "ofp-print.h"
42 #include "ofproto-provider.h"
43 #include "openflow/nicira-ext.h"
44 #include "openflow/openflow.h"
48 #include "poll-loop.h"
55 #include "unaligned.h"
59 VLOG_DEFINE_THIS_MODULE(ofproto);
61 COVERAGE_DEFINE(ofproto_flush);
62 COVERAGE_DEFINE(ofproto_packet_out);
63 COVERAGE_DEFINE(ofproto_queue_req);
64 COVERAGE_DEFINE(ofproto_recv_openflow);
65 COVERAGE_DEFINE(ofproto_reinit_ports);
66 COVERAGE_DEFINE(ofproto_update_port);
69 S_OPENFLOW, /* Processing OpenFlow commands. */
70 S_EVICT, /* Evicting flows from over-limit tables. */
71 S_FLUSH, /* Deleting all flow table rules. */
74 enum ofoperation_type {
81 /* A single OpenFlow request can execute any number of operations. The
82 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
83 * ofconn to which an error reply should be sent if necessary.
85 * ofproto initiates some operations internally. These operations are still
86 * assigned to groups but will not have an associated ofconn. */
88 struct ofproto *ofproto; /* Owning ofproto. */
89 struct list ofproto_node; /* In ofproto's "pending" list. */
90 struct list ops; /* List of "struct ofoperation"s. */
91 int n_running; /* Number of ops still pending. */
93 /* Data needed to send OpenFlow reply on failure or to send a buffered
96 * If list_is_empty(ofconn_node) then this ofopgroup never had an
97 * associated ofconn or its ofconn's connection dropped after it initiated
98 * the operation. In the latter case 'ofconn' is a wild pointer that
99 * refers to freed memory, so the 'ofconn' member must be used only if
100 * !list_is_empty(ofconn_node).
102 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
103 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
104 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
105 uint32_t buffer_id; /* Buffer id from original request. */
108 static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
109 static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
110 const struct ofp_header *,
112 static void ofopgroup_submit(struct ofopgroup *);
113 static void ofopgroup_complete(struct ofopgroup *);
115 /* A single flow table operation. */
117 struct ofopgroup *group; /* Owning group. */
118 struct list group_node; /* In ofopgroup's "ops" list. */
119 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
120 struct rule *rule; /* Rule being operated upon. */
121 enum ofoperation_type type; /* Type of operation. */
123 /* OFOPERATION_MODIFY, OFOPERATION_REPLACE: The old actions, if the actions
125 struct rule_actions *actions;
127 /* OFOPERATION_DELETE. */
128 enum ofp_flow_removed_reason reason; /* Reason flow was removed. */
130 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
131 uint16_t idle_timeout; /* Rule's old idle timeout. */
132 uint16_t hard_timeout; /* Rule's old hard timeout. */
133 enum ofputil_flow_mod_flags flags; /* Rule's old flags. */
134 enum ofperr error; /* 0 if no error. */
137 static struct ofoperation *ofoperation_create(struct ofopgroup *,
139 enum ofoperation_type,
140 enum ofp_flow_removed_reason);
141 static void ofoperation_destroy(struct ofoperation *);
144 static void oftable_init(struct oftable *);
145 static void oftable_destroy(struct oftable *);
147 static void oftable_set_name(struct oftable *, const char *name);
149 static void oftable_disable_eviction(struct oftable *);
150 static void oftable_enable_eviction(struct oftable *,
151 const struct mf_subfield *fields,
154 static void oftable_remove_rule(struct rule *rule) OVS_REQUIRES(ofproto_mutex);
155 static void oftable_remove_rule__(struct ofproto *, struct rule *)
156 OVS_REQUIRES(ofproto_mutex);
157 static void oftable_insert_rule(struct rule *);
159 /* A set of rules within a single OpenFlow table (oftable) that have the same
160 * values for the oftable's eviction_fields. A rule to be evicted, when one is
161 * needed, is taken from the eviction group that contains the greatest number
164 * An oftable owns any number of eviction groups, each of which contains any
167 * Membership in an eviction group is imprecise, based on the hash of the
168 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
169 * That is, if two rules have different eviction_fields, but those
170 * eviction_fields hash to the same value, then they will belong to the same
171 * eviction_group anyway.
173 * (When eviction is not enabled on an oftable, we don't track any eviction
174 * groups, to save time and space.) */
175 struct eviction_group {
176 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
177 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
178 struct heap rules; /* Contains "struct rule"s. */
181 static bool choose_rule_to_evict(struct oftable *table, struct rule **rulep);
182 static void ofproto_evict(struct ofproto *) OVS_EXCLUDED(ofproto_mutex);
183 static uint32_t rule_eviction_priority(struct rule *);
184 static void eviction_group_add_rule(struct rule *);
185 static void eviction_group_remove_rule(struct rule *);
187 /* Criteria that flow_mod and other operations use for selecting rules on
188 * which to operate. */
189 struct rule_criteria {
190 /* An OpenFlow table or 255 for all tables. */
193 /* OpenFlow matching criteria. Interpreted different in "loose" way by
194 * collect_rules_loose() and "strict" way by collect_rules_strict(), as
195 * defined in the OpenFlow spec. */
198 /* Matching criteria for the OpenFlow cookie. Consider a bit B in a rule's
199 * cookie and the corresponding bits C in 'cookie' and M in 'cookie_mask'.
200 * The rule will not be selected if M is 1 and B != C. */
202 ovs_be64 cookie_mask;
204 /* Selection based on actions within a rule:
206 * If out_port != OFPP_ANY, selects only rules that output to out_port.
207 * If out_group != OFPG_ALL, select only rules that output to out_group. */
212 static void rule_criteria_init(struct rule_criteria *, uint8_t table_id,
213 const struct match *match,
214 unsigned int priority,
215 ovs_be64 cookie, ovs_be64 cookie_mask,
216 ofp_port_t out_port, uint32_t out_group);
217 static void rule_criteria_destroy(struct rule_criteria *);
219 /* A packet that needs to be passed to rule_execute().
221 * (We can't do this immediately from ofopgroup_complete() because that holds
222 * ofproto_mutex, which rule_execute() needs released.) */
223 struct rule_execute {
224 struct list list_node; /* In struct ofproto's "rule_executes" list. */
225 struct rule *rule; /* Owns a reference to the rule. */
227 struct ofpbuf *packet; /* Owns the packet. */
230 static void run_rule_executes(struct ofproto *) OVS_EXCLUDED(ofproto_mutex);
231 static void destroy_rule_executes(struct ofproto *);
234 static void ofport_destroy__(struct ofport *) OVS_EXCLUDED(ofproto_mutex);
235 static void ofport_destroy(struct ofport *);
237 static void update_port(struct ofproto *, const char *devname);
238 static int init_ports(struct ofproto *);
239 static void reinit_ports(struct ofproto *);
241 static long long int ofport_get_usage(const struct ofproto *,
242 ofp_port_t ofp_port);
243 static void ofport_set_usage(struct ofproto *, ofp_port_t ofp_port,
244 long long int last_used);
245 static void ofport_remove_usage(struct ofproto *, ofp_port_t ofp_port);
249 * Keeps track of the currently used and recently used ofport values and is
250 * used to prevent immediate recycling of ofport values. */
251 struct ofport_usage {
252 struct hmap_node hmap_node; /* In struct ofproto's "ofport_usage" hmap. */
253 ofp_port_t ofp_port; /* OpenFlow port number. */
254 long long int last_used; /* Last time the 'ofp_port' was used. LLONG_MAX
255 represents in-use ofports. */
259 static void ofproto_rule_destroy__(struct rule *);
260 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
261 static bool rule_is_modifiable(const struct rule *);
264 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
265 struct ofputil_flow_mod *,
266 const struct ofp_header *);
267 static enum ofperr modify_flows__(struct ofproto *, struct ofconn *,
268 struct ofputil_flow_mod *,
269 const struct ofp_header *,
270 const struct rule_collection *);
271 static void delete_flow__(struct rule *rule, struct ofopgroup *,
272 enum ofp_flow_removed_reason)
273 OVS_REQUIRES(ofproto_mutex);
274 static bool ofproto_group_exists__(const struct ofproto *ofproto,
276 OVS_REQ_RDLOCK(ofproto->groups_rwlock);
277 static bool ofproto_group_exists(const struct ofproto *ofproto,
279 OVS_EXCLUDED(ofproto->groups_rwlock);
280 static enum ofperr add_group(struct ofproto *, struct ofputil_group_mod *);
281 static bool handle_openflow(struct ofconn *, const struct ofpbuf *);
282 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
283 struct ofputil_flow_mod *,
284 const struct ofp_header *)
285 OVS_EXCLUDED(ofproto_mutex);
286 static void calc_duration(long long int start, long long int now,
287 uint32_t *sec, uint32_t *nsec);
290 static uint64_t pick_datapath_id(const struct ofproto *);
291 static uint64_t pick_fallback_dpid(void);
292 static void ofproto_destroy__(struct ofproto *);
293 static void update_mtu(struct ofproto *, struct ofport *);
294 static void meter_delete(struct ofproto *, uint32_t first, uint32_t last);
297 static void ofproto_unixctl_init(void);
299 /* All registered ofproto classes, in probe order. */
300 static const struct ofproto_class **ofproto_classes;
301 static size_t n_ofproto_classes;
302 static size_t allocated_ofproto_classes;
304 /* Global lock that protects all flow table operations. */
305 struct ovs_mutex ofproto_mutex = OVS_MUTEX_INITIALIZER;
307 unsigned flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT;
308 unsigned n_handler_threads;
309 enum ofproto_flow_miss_model flow_miss_model = OFPROTO_HANDLE_MISS_AUTO;
311 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
312 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
314 /* Initial mappings of port to OpenFlow number mappings. */
315 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
317 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
319 /* The default value of true waits for flow restore. */
320 static bool flow_restore_wait = true;
322 /* Must be called to initialize the ofproto library.
324 * The caller may pass in 'iface_hints', which contains an shash of
325 * "iface_hint" elements indexed by the interface's name. The provider
326 * may use these hints to describe the startup configuration in order to
327 * reinitialize its state. The caller owns the provided data, so a
328 * provider will make copies of anything required. An ofproto provider
329 * will remove any existing state that is not described by the hint, and
330 * may choose to remove it all. */
332 ofproto_init(const struct shash *iface_hints)
334 struct shash_node *node;
337 ofproto_class_register(&ofproto_dpif_class);
339 /* Make a local copy, since we don't own 'iface_hints' elements. */
340 SHASH_FOR_EACH(node, iface_hints) {
341 const struct iface_hint *orig_hint = node->data;
342 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
343 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
345 new_hint->br_name = xstrdup(orig_hint->br_name);
346 new_hint->br_type = xstrdup(br_type);
347 new_hint->ofp_port = orig_hint->ofp_port;
349 shash_add(&init_ofp_ports, node->name, new_hint);
352 for (i = 0; i < n_ofproto_classes; i++) {
353 ofproto_classes[i]->init(&init_ofp_ports);
357 /* 'type' should be a normalized datapath type, as returned by
358 * ofproto_normalize_type(). Returns the corresponding ofproto_class
359 * structure, or a null pointer if there is none registered for 'type'. */
360 static const struct ofproto_class *
361 ofproto_class_find__(const char *type)
365 for (i = 0; i < n_ofproto_classes; i++) {
366 const struct ofproto_class *class = ofproto_classes[i];
371 class->enumerate_types(&types);
372 found = sset_contains(&types, type);
373 sset_destroy(&types);
379 VLOG_WARN("unknown datapath type %s", type);
383 /* Registers a new ofproto class. After successful registration, new ofprotos
384 * of that type can be created using ofproto_create(). */
386 ofproto_class_register(const struct ofproto_class *new_class)
390 for (i = 0; i < n_ofproto_classes; i++) {
391 if (ofproto_classes[i] == new_class) {
396 if (n_ofproto_classes >= allocated_ofproto_classes) {
397 ofproto_classes = x2nrealloc(ofproto_classes,
398 &allocated_ofproto_classes,
399 sizeof *ofproto_classes);
401 ofproto_classes[n_ofproto_classes++] = new_class;
405 /* Unregisters a datapath provider. 'type' must have been previously
406 * registered and not currently be in use by any ofprotos. After
407 * unregistration new datapaths of that type cannot be opened using
408 * ofproto_create(). */
410 ofproto_class_unregister(const struct ofproto_class *class)
414 for (i = 0; i < n_ofproto_classes; i++) {
415 if (ofproto_classes[i] == class) {
416 for (i++; i < n_ofproto_classes; i++) {
417 ofproto_classes[i - 1] = ofproto_classes[i];
423 VLOG_WARN("attempted to unregister an ofproto class that is not "
428 /* Clears 'types' and enumerates all registered ofproto types into it. The
429 * caller must first initialize the sset. */
431 ofproto_enumerate_types(struct sset *types)
435 for (i = 0; i < n_ofproto_classes; i++) {
436 ofproto_classes[i]->enumerate_types(types);
440 /* Returns the fully spelled out name for the given ofproto 'type'.
442 * Normalized type string can be compared with strcmp(). Unnormalized type
443 * string might be the same even if they have different spellings. */
445 ofproto_normalize_type(const char *type)
447 return type && type[0] ? type : "system";
450 /* Clears 'names' and enumerates the names of all known created ofprotos with
451 * the given 'type'. The caller must first initialize the sset. Returns 0 if
452 * successful, otherwise a positive errno value.
454 * Some kinds of datapaths might not be practically enumerable. This is not
455 * considered an error. */
457 ofproto_enumerate_names(const char *type, struct sset *names)
459 const struct ofproto_class *class = ofproto_class_find__(type);
460 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
464 ofproto_create(const char *datapath_name, const char *datapath_type,
465 struct ofproto **ofprotop)
467 const struct ofproto_class *class;
468 struct ofproto *ofproto;
474 ofproto_unixctl_init();
476 datapath_type = ofproto_normalize_type(datapath_type);
477 class = ofproto_class_find__(datapath_type);
479 VLOG_WARN("could not create datapath %s of unknown type %s",
480 datapath_name, datapath_type);
484 ofproto = class->alloc();
486 VLOG_ERR("failed to allocate datapath %s of type %s",
487 datapath_name, datapath_type);
492 ovs_mutex_lock(&ofproto_mutex);
493 memset(ofproto, 0, sizeof *ofproto);
494 ofproto->ofproto_class = class;
495 ofproto->name = xstrdup(datapath_name);
496 ofproto->type = xstrdup(datapath_type);
497 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
498 hash_string(ofproto->name, 0));
499 ofproto->datapath_id = 0;
500 ofproto->forward_bpdu = false;
501 ofproto->fallback_dpid = pick_fallback_dpid();
502 ofproto->mfr_desc = NULL;
503 ofproto->hw_desc = NULL;
504 ofproto->sw_desc = NULL;
505 ofproto->serial_desc = NULL;
506 ofproto->dp_desc = NULL;
507 ofproto->frag_handling = OFPC_FRAG_NORMAL;
508 hmap_init(&ofproto->ports);
509 hmap_init(&ofproto->ofport_usage);
510 shash_init(&ofproto->port_by_name);
511 simap_init(&ofproto->ofp_requests);
512 ofproto->max_ports = ofp_to_u16(OFPP_MAX);
513 ofproto->eviction_group_timer = LLONG_MIN;
514 ofproto->tables = NULL;
515 ofproto->n_tables = 0;
516 hindex_init(&ofproto->cookies);
517 list_init(&ofproto->expirable);
518 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
519 ofproto->state = S_OPENFLOW;
520 list_init(&ofproto->pending);
521 ofproto->n_pending = 0;
522 hmap_init(&ofproto->deletions);
523 guarded_list_init(&ofproto->rule_executes);
524 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
525 ofproto->first_op = ofproto->last_op = LLONG_MIN;
526 ofproto->next_op_report = LLONG_MAX;
527 ofproto->op_backoff = LLONG_MIN;
528 ofproto->vlan_bitmap = NULL;
529 ofproto->vlans_changed = false;
530 ofproto->min_mtu = INT_MAX;
531 ovs_rwlock_init(&ofproto->groups_rwlock);
532 hmap_init(&ofproto->groups);
533 ovs_mutex_unlock(&ofproto_mutex);
534 ofproto->ogf.capabilities = OFPGFC_CHAINING;
535 ofproto->ogf.max_groups[OFPGT11_ALL] = OFPG_MAX;
536 ofproto->ogf.max_groups[OFPGT11_INDIRECT] = OFPG_MAX;
537 ofproto->ogf.max_groups[OFPGT11_FF] = OFPG_MAX;
538 ofproto->ogf.actions[0] =
539 #define OFPAT11_ACTION(ENUM, STRUCT, EXTENSIBLE, NAME) (1 << ENUM) |
540 #include "ofp-util.def"
543 error = ofproto->ofproto_class->construct(ofproto);
545 VLOG_ERR("failed to open datapath %s: %s",
546 datapath_name, ovs_strerror(error));
547 ofproto_destroy__(ofproto);
551 /* Check that hidden tables, if any, are at the end. */
552 ovs_assert(ofproto->n_tables);
553 for (i = 0; i + 1 < ofproto->n_tables; i++) {
554 enum oftable_flags flags = ofproto->tables[i].flags;
555 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
557 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
560 ofproto->datapath_id = pick_datapath_id(ofproto);
563 /* Initialize meters table. */
564 if (ofproto->ofproto_class->meter_get_features) {
565 ofproto->ofproto_class->meter_get_features(ofproto,
566 &ofproto->meter_features);
568 memset(&ofproto->meter_features, 0, sizeof ofproto->meter_features);
570 ofproto->meters = xzalloc((ofproto->meter_features.max_meters + 1)
571 * sizeof(struct meter *));
577 /* Must be called (only) by an ofproto implementation in its constructor
578 * function. See the large comment on 'construct' in struct ofproto_class for
581 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
583 struct oftable *table;
585 ovs_assert(!ofproto->n_tables);
586 ovs_assert(n_tables >= 1 && n_tables <= 255);
588 ofproto->n_tables = n_tables;
589 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
590 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
595 /* To be optionally called (only) by an ofproto implementation in its
596 * constructor function. See the large comment on 'construct' in struct
597 * ofproto_class for details.
599 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
600 * will then ensure that actions passed into the ofproto implementation will
601 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
602 * function is not called, there will be no such restriction.
604 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
605 * the 'max_ports' restriction. */
607 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
609 ovs_assert(max_ports <= ofp_to_u16(OFPP_MAX));
610 ofproto->max_ports = max_ports;
614 ofproto_get_datapath_id(const struct ofproto *ofproto)
616 return ofproto->datapath_id;
620 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
622 uint64_t old_dpid = p->datapath_id;
623 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
624 if (p->datapath_id != old_dpid) {
625 /* Force all active connections to reconnect, since there is no way to
626 * notify a controller that the datapath ID has changed. */
627 ofproto_reconnect_controllers(p);
632 ofproto_set_controllers(struct ofproto *p,
633 const struct ofproto_controller *controllers,
634 size_t n_controllers, uint32_t allowed_versions)
636 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
641 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
643 connmgr_set_fail_mode(p->connmgr, fail_mode);
646 /* Drops the connections between 'ofproto' and all of its controllers, forcing
647 * them to reconnect. */
649 ofproto_reconnect_controllers(struct ofproto *ofproto)
651 connmgr_reconnect(ofproto->connmgr);
654 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
655 * in-band control should guarantee access, in the same way that in-band
656 * control guarantees access to OpenFlow controllers. */
658 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
659 const struct sockaddr_in *extras, size_t n)
661 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
664 /* Sets the OpenFlow queue used by flows set up by in-band control on
665 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
666 * flows will use the default queue. */
668 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
670 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
673 /* Sets the number of flows at which eviction from the kernel flow table
676 ofproto_set_flow_eviction_threshold(unsigned threshold)
678 flow_eviction_threshold = MAX(OFPROTO_FLOW_EVICTION_THRESHOLD_MIN,
682 /* Sets the path for handling flow misses. */
684 ofproto_set_flow_miss_model(unsigned model)
686 flow_miss_model = model;
689 /* If forward_bpdu is true, the NORMAL action will forward frames with
690 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
691 * the NORMAL action will drop these frames. */
693 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
695 bool old_val = ofproto->forward_bpdu;
696 ofproto->forward_bpdu = forward_bpdu;
697 if (old_val != ofproto->forward_bpdu) {
698 if (ofproto->ofproto_class->forward_bpdu_changed) {
699 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
704 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
705 * 'idle_time', in seconds, and the maximum number of MAC table entries to
708 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
711 if (ofproto->ofproto_class->set_mac_table_config) {
712 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
717 /* Sets number of upcall handler threads. The default is
718 * (number of online cores - 2). */
720 ofproto_set_n_handler_threads(unsigned limit)
723 n_handler_threads = limit;
725 int n_proc = sysconf(_SC_NPROCESSORS_ONLN);
726 n_handler_threads = n_proc > 2 ? n_proc - 2 : 1;
731 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
734 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
738 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
740 return connmgr_set_snoops(ofproto->connmgr, snoops);
744 ofproto_set_netflow(struct ofproto *ofproto,
745 const struct netflow_options *nf_options)
747 if (nf_options && sset_is_empty(&nf_options->collectors)) {
751 if (ofproto->ofproto_class->set_netflow) {
752 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
754 return nf_options ? EOPNOTSUPP : 0;
759 ofproto_set_sflow(struct ofproto *ofproto,
760 const struct ofproto_sflow_options *oso)
762 if (oso && sset_is_empty(&oso->targets)) {
766 if (ofproto->ofproto_class->set_sflow) {
767 return ofproto->ofproto_class->set_sflow(ofproto, oso);
769 return oso ? EOPNOTSUPP : 0;
774 ofproto_set_ipfix(struct ofproto *ofproto,
775 const struct ofproto_ipfix_bridge_exporter_options *bo,
776 const struct ofproto_ipfix_flow_exporter_options *fo,
779 if (ofproto->ofproto_class->set_ipfix) {
780 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
782 return (bo || fo) ? EOPNOTSUPP : 0;
787 ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
789 flow_restore_wait = flow_restore_wait_db;
793 ofproto_get_flow_restore_wait(void)
795 return flow_restore_wait;
799 /* Spanning Tree Protocol (STP) configuration. */
801 /* Configures STP on 'ofproto' using the settings defined in 's'. If
802 * 's' is NULL, disables STP.
804 * Returns 0 if successful, otherwise a positive errno value. */
806 ofproto_set_stp(struct ofproto *ofproto,
807 const struct ofproto_stp_settings *s)
809 return (ofproto->ofproto_class->set_stp
810 ? ofproto->ofproto_class->set_stp(ofproto, s)
814 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
815 * 'enabled' member of 's' is false, then the other members are not
818 * Returns 0 if successful, otherwise a positive errno value. */
820 ofproto_get_stp_status(struct ofproto *ofproto,
821 struct ofproto_stp_status *s)
823 return (ofproto->ofproto_class->get_stp_status
824 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
828 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
829 * in 's'. The caller is responsible for assigning STP port numbers
830 * (using the 'port_num' member in the range of 1 through 255, inclusive)
831 * and ensuring there are no duplicates. If the 's' is NULL, then STP
832 * is disabled on the port.
834 * Returns 0 if successful, otherwise a positive errno value.*/
836 ofproto_port_set_stp(struct ofproto *ofproto, ofp_port_t ofp_port,
837 const struct ofproto_port_stp_settings *s)
839 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
841 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
842 ofproto->name, ofp_port);
846 return (ofproto->ofproto_class->set_stp_port
847 ? ofproto->ofproto_class->set_stp_port(ofport, s)
851 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
852 * 's'. If the 'enabled' member in 's' is false, then the other members
853 * are not meaningful.
855 * Returns 0 if successful, otherwise a positive errno value.*/
857 ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
858 struct ofproto_port_stp_status *s)
860 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
862 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
863 "port %"PRIu16, ofproto->name, ofp_port);
867 return (ofproto->ofproto_class->get_stp_port_status
868 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
872 /* Queue DSCP configuration. */
874 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
875 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
876 * to implement QoS. Instead, it is used to implement features which require
877 * knowledge of what queues exist on a port, and some basic information about
880 * Returns 0 if successful, otherwise a positive errno value. */
882 ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port,
883 const struct ofproto_port_queue *queues,
886 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
889 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
890 ofproto->name, ofp_port);
894 return (ofproto->ofproto_class->set_queues
895 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
899 /* Connectivity Fault Management configuration. */
901 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
903 ofproto_port_clear_cfm(struct ofproto *ofproto, ofp_port_t ofp_port)
905 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
906 if (ofport && ofproto->ofproto_class->set_cfm) {
907 ofproto->ofproto_class->set_cfm(ofport, NULL);
911 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
912 * basic configuration from the configuration members in 'cfm', and the remote
913 * maintenance point ID from remote_mpid. Ignores the statistics members of
916 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
918 ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port,
919 const struct cfm_settings *s)
921 struct ofport *ofport;
924 ofport = ofproto_get_port(ofproto, ofp_port);
926 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
927 ofproto->name, ofp_port);
931 /* XXX: For configuration simplicity, we only support one remote_mpid
932 * outside of the CFM module. It's not clear if this is the correct long
933 * term solution or not. */
934 error = (ofproto->ofproto_class->set_cfm
935 ? ofproto->ofproto_class->set_cfm(ofport, s)
938 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
939 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
940 ovs_strerror(error));
944 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
945 * 'ofproto' does not have a port 'ofp_port'. */
947 ofproto_port_set_bfd(struct ofproto *ofproto, ofp_port_t ofp_port,
948 const struct smap *cfg)
950 struct ofport *ofport;
953 ofport = ofproto_get_port(ofproto, ofp_port);
955 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
956 ofproto->name, ofp_port);
960 error = (ofproto->ofproto_class->set_bfd
961 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
964 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
965 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
966 ovs_strerror(error));
970 /* Populates 'status' with key value pairs indicating the status of the BFD
971 * session on 'ofp_port'. This information is intended to be populated in the
972 * OVS database. Has no effect if 'ofp_port' is not na OpenFlow port in
975 ofproto_port_get_bfd_status(struct ofproto *ofproto, ofp_port_t ofp_port,
978 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
979 return (ofport && ofproto->ofproto_class->get_bfd_status
980 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
984 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
985 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
986 * 0 if LACP partner information is not current (generally indicating a
987 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
989 ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port)
991 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
992 return (ofport && ofproto->ofproto_class->port_is_lacp_current
993 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
999 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
1000 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
1001 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
1002 * configuration plus, if there is more than one slave, a bonding
1005 * If 'aux' is already registered then this function updates its configuration
1006 * to 's'. Otherwise, this function registers a new bundle.
1008 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
1011 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
1012 const struct ofproto_bundle_settings *s)
1014 return (ofproto->ofproto_class->bundle_set
1015 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
1019 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
1020 * If no such bundle has been registered, this has no effect. */
1022 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
1024 return ofproto_bundle_register(ofproto, aux, NULL);
1028 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
1029 * If 'aux' is already registered then this function updates its configuration
1030 * to 's'. Otherwise, this function registers a new mirror. */
1032 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
1033 const struct ofproto_mirror_settings *s)
1035 return (ofproto->ofproto_class->mirror_set
1036 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
1040 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
1041 * If no mirror has been registered, this has no effect. */
1043 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
1045 return ofproto_mirror_register(ofproto, aux, NULL);
1048 /* Retrieves statistics from mirror associated with client data pointer
1049 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
1050 * 'bytes', respectively. If a particular counters is not supported,
1051 * the appropriate argument is set to UINT64_MAX. */
1053 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
1054 uint64_t *packets, uint64_t *bytes)
1056 if (!ofproto->ofproto_class->mirror_get_stats) {
1057 *packets = *bytes = UINT64_MAX;
1061 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
1065 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
1066 * which all packets are flooded, instead of using MAC learning. If
1067 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1069 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
1072 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
1074 return (ofproto->ofproto_class->set_flood_vlans
1075 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
1079 /* Returns true if 'aux' is a registered bundle that is currently in use as the
1080 * output for a mirror. */
1082 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
1084 return (ofproto->ofproto_class->is_mirror_output_bundle
1085 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
1089 /* Configuration of OpenFlow tables. */
1091 /* Returns the number of OpenFlow tables in 'ofproto'. */
1093 ofproto_get_n_tables(const struct ofproto *ofproto)
1095 return ofproto->n_tables;
1098 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
1099 * settings from 's'. 'table_id' must be in the range 0 through the number of
1100 * OpenFlow tables in 'ofproto' minus 1, inclusive.
1102 * For read-only tables, only the name may be configured. */
1104 ofproto_configure_table(struct ofproto *ofproto, int table_id,
1105 const struct ofproto_table_settings *s)
1107 struct oftable *table;
1109 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
1110 table = &ofproto->tables[table_id];
1112 oftable_set_name(table, s->name);
1114 if (table->flags & OFTABLE_READONLY) {
1119 oftable_enable_eviction(table, s->groups, s->n_groups);
1121 oftable_disable_eviction(table);
1124 table->max_flows = s->max_flows;
1125 ovs_rwlock_rdlock(&table->cls.rwlock);
1126 if (classifier_count(&table->cls) > table->max_flows
1127 && table->eviction_fields) {
1128 /* 'table' contains more flows than allowed. We might not be able to
1129 * evict them right away because of the asynchronous nature of flow
1130 * table changes. Schedule eviction for later. */
1131 switch (ofproto->state) {
1133 ofproto->state = S_EVICT;
1137 /* We're already deleting flows, nothing more to do. */
1141 ovs_rwlock_unlock(&table->cls.rwlock);
1145 ofproto_has_snoops(const struct ofproto *ofproto)
1147 return connmgr_has_snoops(ofproto->connmgr);
1151 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1153 connmgr_get_snoops(ofproto->connmgr, snoops);
1157 ofproto_rule_delete__(struct ofproto *ofproto, struct rule *rule,
1159 OVS_REQUIRES(ofproto_mutex)
1161 struct ofopgroup *group;
1163 ovs_assert(!rule->pending);
1165 group = ofopgroup_create_unattached(ofproto);
1166 delete_flow__(rule, group, reason);
1167 ofopgroup_submit(group);
1170 /* Deletes 'rule' from 'cls' within 'ofproto'.
1172 * Within an ofproto implementation, this function allows an ofproto
1173 * implementation to destroy any rules that remain when its ->destruct()
1174 * function is called. This function is not suitable for use elsewhere in an
1175 * ofproto implementation.
1177 * This function implements steps 4.4 and 4.5 in the section titled "Rule Life
1178 * Cycle" in ofproto-provider.h. */
1180 ofproto_rule_delete(struct ofproto *ofproto, struct rule *rule)
1181 OVS_EXCLUDED(ofproto_mutex)
1183 struct ofopgroup *group;
1185 ovs_mutex_lock(&ofproto_mutex);
1186 ovs_assert(!rule->pending);
1188 group = ofopgroup_create_unattached(ofproto);
1189 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1190 oftable_remove_rule__(ofproto, rule);
1191 ofproto->ofproto_class->rule_delete(rule);
1192 ofopgroup_submit(group);
1194 ovs_mutex_unlock(&ofproto_mutex);
1198 ofproto_flush__(struct ofproto *ofproto)
1199 OVS_EXCLUDED(ofproto_mutex)
1201 struct oftable *table;
1203 if (ofproto->ofproto_class->flush) {
1204 ofproto->ofproto_class->flush(ofproto);
1207 ovs_mutex_lock(&ofproto_mutex);
1208 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1209 struct rule *rule, *next_rule;
1210 struct cls_cursor cursor;
1212 if (table->flags & OFTABLE_HIDDEN) {
1216 ovs_rwlock_rdlock(&table->cls.rwlock);
1217 cls_cursor_init(&cursor, &table->cls, NULL);
1218 ovs_rwlock_unlock(&table->cls.rwlock);
1219 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1220 if (!rule->pending) {
1221 ofproto_rule_delete__(ofproto, rule, OFPRR_DELETE);
1225 ovs_mutex_unlock(&ofproto_mutex);
1228 static void delete_group(struct ofproto *ofproto, uint32_t group_id);
1231 ofproto_destroy__(struct ofproto *ofproto)
1232 OVS_EXCLUDED(ofproto_mutex)
1234 struct oftable *table;
1236 ovs_assert(list_is_empty(&ofproto->pending));
1238 destroy_rule_executes(ofproto);
1239 guarded_list_destroy(&ofproto->rule_executes);
1241 delete_group(ofproto, OFPG_ALL);
1242 ovs_rwlock_destroy(&ofproto->groups_rwlock);
1243 hmap_destroy(&ofproto->groups);
1245 connmgr_destroy(ofproto->connmgr);
1247 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1248 free(ofproto->name);
1249 free(ofproto->type);
1250 free(ofproto->mfr_desc);
1251 free(ofproto->hw_desc);
1252 free(ofproto->sw_desc);
1253 free(ofproto->serial_desc);
1254 free(ofproto->dp_desc);
1255 hmap_destroy(&ofproto->ports);
1256 hmap_destroy(&ofproto->ofport_usage);
1257 shash_destroy(&ofproto->port_by_name);
1258 simap_destroy(&ofproto->ofp_requests);
1260 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1261 oftable_destroy(table);
1263 free(ofproto->tables);
1265 hmap_destroy(&ofproto->deletions);
1267 free(ofproto->vlan_bitmap);
1269 ofproto->ofproto_class->dealloc(ofproto);
1273 ofproto_destroy(struct ofproto *p)
1274 OVS_EXCLUDED(ofproto_mutex)
1276 struct ofport *ofport, *next_ofport;
1277 struct ofport_usage *usage, *next_usage;
1284 meter_delete(p, 1, p->meter_features.max_meters);
1285 p->meter_features.max_meters = 0;
1291 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1292 ofport_destroy(ofport);
1295 HMAP_FOR_EACH_SAFE (usage, next_usage, hmap_node, &p->ofport_usage) {
1296 hmap_remove(&p->ofport_usage, &usage->hmap_node);
1300 p->ofproto_class->destruct(p);
1301 ofproto_destroy__(p);
1304 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1305 * kernel datapath, for example, this destroys the datapath in the kernel, and
1306 * with the netdev-based datapath, it tears down the data structures that
1307 * represent the datapath.
1309 * The datapath should not be currently open as an ofproto. */
1311 ofproto_delete(const char *name, const char *type)
1313 const struct ofproto_class *class = ofproto_class_find__(type);
1314 return (!class ? EAFNOSUPPORT
1315 : !class->del ? EACCES
1316 : class->del(type, name));
1320 process_port_change(struct ofproto *ofproto, int error, char *devname)
1322 if (error == ENOBUFS) {
1323 reinit_ports(ofproto);
1324 } else if (!error) {
1325 update_port(ofproto, devname);
1331 ofproto_type_run(const char *datapath_type)
1333 const struct ofproto_class *class;
1336 datapath_type = ofproto_normalize_type(datapath_type);
1337 class = ofproto_class_find__(datapath_type);
1339 error = class->type_run ? class->type_run(datapath_type) : 0;
1340 if (error && error != EAGAIN) {
1341 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1342 datapath_type, ovs_strerror(error));
1348 ofproto_type_run_fast(const char *datapath_type)
1350 const struct ofproto_class *class;
1353 datapath_type = ofproto_normalize_type(datapath_type);
1354 class = ofproto_class_find__(datapath_type);
1356 error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
1357 if (error && error != EAGAIN) {
1358 VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
1359 datapath_type, ovs_strerror(error));
1365 ofproto_type_wait(const char *datapath_type)
1367 const struct ofproto_class *class;
1369 datapath_type = ofproto_normalize_type(datapath_type);
1370 class = ofproto_class_find__(datapath_type);
1372 if (class->type_wait) {
1373 class->type_wait(datapath_type);
1378 any_pending_ops(const struct ofproto *p)
1379 OVS_EXCLUDED(ofproto_mutex)
1383 ovs_mutex_lock(&ofproto_mutex);
1384 b = !list_is_empty(&p->pending);
1385 ovs_mutex_unlock(&ofproto_mutex);
1391 ofproto_run(struct ofproto *p)
1393 struct sset changed_netdevs;
1394 const char *changed_netdev;
1395 struct ofport *ofport;
1398 error = p->ofproto_class->run(p);
1399 if (error && error != EAGAIN) {
1400 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, ovs_strerror(error));
1403 run_rule_executes(p);
1405 /* Restore the eviction group heap invariant occasionally. */
1406 if (p->eviction_group_timer < time_msec()) {
1409 p->eviction_group_timer = time_msec() + 1000;
1411 for (i = 0; i < p->n_tables; i++) {
1412 struct oftable *table = &p->tables[i];
1413 struct eviction_group *evg;
1414 struct cls_cursor cursor;
1417 if (!table->eviction_fields) {
1421 ovs_mutex_lock(&ofproto_mutex);
1422 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
1423 heap_rebuild(&evg->rules);
1426 ovs_rwlock_rdlock(&table->cls.rwlock);
1427 cls_cursor_init(&cursor, &table->cls, NULL);
1428 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1429 if (!rule->eviction_group
1430 && (rule->idle_timeout || rule->hard_timeout)) {
1431 eviction_group_add_rule(rule);
1434 ovs_rwlock_unlock(&table->cls.rwlock);
1435 ovs_mutex_unlock(&ofproto_mutex);
1439 if (p->ofproto_class->port_poll) {
1442 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1443 process_port_change(p, error, devname);
1447 /* Update OpenFlow port status for any port whose netdev has changed.
1449 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1450 * destroyed, so it's not safe to update ports directly from the
1451 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1452 * need this two-phase approach. */
1453 sset_init(&changed_netdevs);
1454 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1455 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1456 if (ofport->change_seq != change_seq) {
1457 ofport->change_seq = change_seq;
1458 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1461 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1462 update_port(p, changed_netdev);
1464 sset_destroy(&changed_netdevs);
1468 connmgr_run(p->connmgr, handle_openflow);
1472 connmgr_run(p->connmgr, NULL);
1474 if (!any_pending_ops(p)) {
1475 p->state = S_OPENFLOW;
1480 connmgr_run(p->connmgr, NULL);
1482 if (!any_pending_ops(p)) {
1483 connmgr_flushed(p->connmgr);
1484 p->state = S_OPENFLOW;
1492 if (time_msec() >= p->next_op_report) {
1493 long long int ago = (time_msec() - p->first_op) / 1000;
1494 long long int interval = (p->last_op - p->first_op) / 1000;
1498 ds_put_format(&s, "%d flow_mods ",
1499 p->n_add + p->n_delete + p->n_modify);
1500 if (interval == ago) {
1501 ds_put_format(&s, "in the last %lld s", ago);
1502 } else if (interval) {
1503 ds_put_format(&s, "in the %lld s starting %lld s ago",
1506 ds_put_format(&s, "%lld s ago", ago);
1509 ds_put_cstr(&s, " (");
1511 ds_put_format(&s, "%d adds, ", p->n_add);
1514 ds_put_format(&s, "%d deletes, ", p->n_delete);
1517 ds_put_format(&s, "%d modifications, ", p->n_modify);
1520 ds_put_char(&s, ')');
1522 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1525 p->n_add = p->n_delete = p->n_modify = 0;
1526 p->next_op_report = LLONG_MAX;
1532 /* Performs periodic activity required by 'ofproto' that needs to be done
1533 * with the least possible latency.
1535 * It makes sense to call this function a couple of times per poll loop, to
1536 * provide a significant performance boost on some benchmarks with the
1537 * ofproto-dpif implementation. */
1539 ofproto_run_fast(struct ofproto *p)
1543 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1544 if (error && error != EAGAIN) {
1545 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1546 p->name, ovs_strerror(error));
1552 ofproto_wait(struct ofproto *p)
1554 struct ofport *ofport;
1556 p->ofproto_class->wait(p);
1557 if (p->ofproto_class->port_poll_wait) {
1558 p->ofproto_class->port_poll_wait(p);
1561 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1562 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1563 poll_immediate_wake();
1569 connmgr_wait(p->connmgr, true);
1574 connmgr_wait(p->connmgr, false);
1575 if (!any_pending_ops(p)) {
1576 poll_immediate_wake();
1583 ofproto_is_alive(const struct ofproto *p)
1585 return connmgr_has_controllers(p->connmgr);
1588 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1589 * memory_report(). */
1591 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1593 const struct oftable *table;
1594 unsigned int n_rules;
1596 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1598 ovs_mutex_lock(&ofproto_mutex);
1599 simap_increase(usage, "ops",
1600 ofproto->n_pending + hmap_count(&ofproto->deletions));
1601 ovs_mutex_unlock(&ofproto_mutex);
1604 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1605 ovs_rwlock_rdlock(&table->cls.rwlock);
1606 n_rules += classifier_count(&table->cls);
1607 ovs_rwlock_unlock(&table->cls.rwlock);
1609 simap_increase(usage, "rules", n_rules);
1611 if (ofproto->ofproto_class->get_memory_usage) {
1612 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1615 connmgr_get_memory_usage(ofproto->connmgr, usage);
1619 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1622 connmgr_get_controller_info(ofproto->connmgr, info);
1626 ofproto_free_ofproto_controller_info(struct shash *info)
1628 connmgr_free_controller_info(info);
1631 /* Makes a deep copy of 'old' into 'port'. */
1633 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1635 port->name = xstrdup(old->name);
1636 port->type = xstrdup(old->type);
1637 port->ofp_port = old->ofp_port;
1640 /* Frees memory allocated to members of 'ofproto_port'.
1642 * Do not call this function on an ofproto_port obtained from
1643 * ofproto_port_dump_next(): that function retains ownership of the data in the
1646 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1648 free(ofproto_port->name);
1649 free(ofproto_port->type);
1652 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1654 * This function provides no status indication. An error status for the entire
1655 * dump operation is provided when it is completed by calling
1656 * ofproto_port_dump_done().
1659 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1660 const struct ofproto *ofproto)
1662 dump->ofproto = ofproto;
1663 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1667 /* Attempts to retrieve another port from 'dump', which must have been created
1668 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1669 * 'port' and returns true. On failure, returns false.
1671 * Failure might indicate an actual error or merely that the last port has been
1672 * dumped. An error status for the entire dump operation is provided when it
1673 * is completed by calling ofproto_port_dump_done().
1675 * The ofproto owns the data stored in 'port'. It will remain valid until at
1676 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1677 * ofproto_port_dump_done(). */
1679 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1680 struct ofproto_port *port)
1682 const struct ofproto *ofproto = dump->ofproto;
1688 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1691 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1697 /* Completes port table dump operation 'dump', which must have been created
1698 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1699 * error-free, otherwise a positive errno value describing the problem. */
1701 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1703 const struct ofproto *ofproto = dump->ofproto;
1705 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1708 return dump->error == EOF ? 0 : dump->error;
1711 /* Returns the type to pass to netdev_open() when a datapath of type
1712 * 'datapath_type' has a port of type 'port_type', for a few special
1713 * cases when a netdev type differs from a port type. For example, when
1714 * using the userspace datapath, a port of type "internal" needs to be
1717 * Returns either 'type' itself or a string literal, which must not be
1720 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1722 const struct ofproto_class *class;
1724 datapath_type = ofproto_normalize_type(datapath_type);
1725 class = ofproto_class_find__(datapath_type);
1730 return (class->port_open_type
1731 ? class->port_open_type(datapath_type, port_type)
1735 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1736 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1737 * the port's OpenFlow port number.
1739 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1740 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1741 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1742 * 'ofp_portp' is non-null). */
1744 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1745 ofp_port_t *ofp_portp)
1747 ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1750 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1752 const char *netdev_name = netdev_get_name(netdev);
1754 simap_put(&ofproto->ofp_requests, netdev_name,
1755 ofp_to_u16(ofp_port));
1756 update_port(ofproto, netdev_name);
1759 struct ofproto_port ofproto_port;
1761 ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
1763 *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
1764 ofproto_port_destroy(&ofproto_port);
1769 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1770 * initializes '*port' appropriately; on failure, returns a positive errno
1773 * The caller owns the data in 'ofproto_port' and must free it with
1774 * ofproto_port_destroy() when it is no longer needed. */
1776 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1777 struct ofproto_port *port)
1781 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1783 memset(port, 0, sizeof *port);
1788 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1789 * Returns 0 if successful, otherwise a positive errno. */
1791 ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port)
1793 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1794 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1795 struct simap_node *ofp_request_node;
1798 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1799 if (ofp_request_node) {
1800 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1803 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1804 if (!error && ofport) {
1805 /* 'name' is the netdev's name and update_port() is going to close the
1806 * netdev. Just in case update_port() refers to 'name' after it
1807 * destroys 'ofport', make a copy of it around the update_port()
1809 char *devname = xstrdup(name);
1810 update_port(ofproto, devname);
1817 flow_mod_init(struct ofputil_flow_mod *fm,
1818 const struct match *match, unsigned int priority,
1819 const struct ofpact *ofpacts, size_t ofpacts_len,
1820 enum ofp_flow_mod_command command)
1822 memset(fm, 0, sizeof *fm);
1824 fm->priority = priority;
1827 fm->modify_cookie = false;
1829 fm->command = command;
1830 fm->idle_timeout = 0;
1831 fm->hard_timeout = 0;
1832 fm->buffer_id = UINT32_MAX;
1833 fm->out_port = OFPP_ANY;
1834 fm->out_group = OFPG_ANY;
1836 fm->ofpacts = CONST_CAST(struct ofpact *, ofpacts);
1837 fm->ofpacts_len = ofpacts_len;
1841 simple_flow_mod(struct ofproto *ofproto,
1842 const struct match *match, unsigned int priority,
1843 const struct ofpact *ofpacts, size_t ofpacts_len,
1844 enum ofp_flow_mod_command command)
1846 struct ofputil_flow_mod fm;
1848 flow_mod_init(&fm, match, priority, ofpacts, ofpacts_len, command);
1850 return handle_flow_mod__(ofproto, NULL, &fm, NULL);
1853 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1854 * performs the 'n_actions' actions in 'actions'. The new flow will not
1857 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1858 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1859 * controllers; otherwise, it will be hidden.
1861 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1863 * This is a helper function for in-band control and fail-open. */
1865 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1866 unsigned int priority,
1867 const struct ofpact *ofpacts, size_t ofpacts_len)
1868 OVS_EXCLUDED(ofproto_mutex)
1870 const struct rule *rule;
1873 /* First do a cheap check whether the rule we're looking for already exists
1874 * with the actions that we want. If it does, then we're done. */
1875 ovs_rwlock_rdlock(&ofproto->tables[0].cls.rwlock);
1876 rule = rule_from_cls_rule(classifier_find_match_exactly(
1877 &ofproto->tables[0].cls, match, priority));
1879 ovs_mutex_lock(&rule->mutex);
1880 must_add = !ofpacts_equal(rule->actions->ofpacts,
1881 rule->actions->ofpacts_len,
1882 ofpacts, ofpacts_len);
1883 ovs_mutex_unlock(&rule->mutex);
1887 ovs_rwlock_unlock(&ofproto->tables[0].cls.rwlock);
1889 /* If there's no such rule or the rule doesn't have the actions we want,
1890 * fall back to a executing a full flow mod. We can't optimize this at
1891 * all because we didn't take enough locks above to ensure that the flow
1892 * table didn't already change beneath us. */
1894 simple_flow_mod(ofproto, match, priority, ofpacts, ofpacts_len,
1895 OFPFC_MODIFY_STRICT);
1899 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1900 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1901 * operation cannot be initiated now but may be retried later.
1903 * This is a helper function for in-band control and fail-open and the "learn"
1906 ofproto_flow_mod(struct ofproto *ofproto, struct ofputil_flow_mod *fm)
1907 OVS_EXCLUDED(ofproto_mutex)
1909 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1912 /* Searches for a rule with matching criteria exactly equal to 'target' in
1913 * ofproto's table 0 and, if it finds one, deletes it.
1915 * This is a helper function for in-band control and fail-open. */
1917 ofproto_delete_flow(struct ofproto *ofproto,
1918 const struct match *target, unsigned int priority)
1919 OVS_EXCLUDED(ofproto_mutex)
1921 struct classifier *cls = &ofproto->tables[0].cls;
1924 /* First do a cheap check whether the rule we're looking for has already
1925 * been deleted. If so, then we're done. */
1926 ovs_rwlock_rdlock(&cls->rwlock);
1927 rule = rule_from_cls_rule(classifier_find_match_exactly(cls, target,
1929 ovs_rwlock_unlock(&cls->rwlock);
1934 /* Fall back to a executing a full flow mod. We can't optimize this at all
1935 * because we didn't take enough locks above to ensure that the flow table
1936 * didn't already change beneath us. */
1937 return simple_flow_mod(ofproto, target, priority, NULL, 0,
1938 OFPFC_DELETE_STRICT) != OFPROTO_POSTPONE;
1941 /* Starts the process of deleting all of the flows from all of ofproto's flow
1942 * tables and then reintroducing the flows required by in-band control and
1943 * fail-open. The process will complete in a later call to ofproto_run(). */
1945 ofproto_flush_flows(struct ofproto *ofproto)
1947 COVERAGE_INC(ofproto_flush);
1948 ofproto->state = S_FLUSH;
1952 reinit_ports(struct ofproto *p)
1954 struct ofproto_port_dump dump;
1955 struct sset devnames;
1956 struct ofport *ofport;
1957 struct ofproto_port ofproto_port;
1958 const char *devname;
1960 COVERAGE_INC(ofproto_reinit_ports);
1962 sset_init(&devnames);
1963 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1964 sset_add(&devnames, netdev_get_name(ofport->netdev));
1966 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1967 sset_add(&devnames, ofproto_port.name);
1970 SSET_FOR_EACH (devname, &devnames) {
1971 update_port(p, devname);
1973 sset_destroy(&devnames);
1977 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
1981 port_idx = simap_get(&ofproto->ofp_requests, netdev_name);
1982 port_idx = port_idx ? port_idx : UINT16_MAX;
1984 if (port_idx >= ofproto->max_ports
1985 || ofport_get_usage(ofproto, u16_to_ofp(port_idx)) == LLONG_MAX) {
1986 uint16_t lru_ofport = 0, end_port_no = ofproto->alloc_port_no;
1987 long long int last_used_at, lru = LLONG_MAX;
1989 /* Search for a free OpenFlow port number. We try not to
1990 * immediately reuse them to prevent problems due to old
1993 if (++ofproto->alloc_port_no >= ofproto->max_ports) {
1994 ofproto->alloc_port_no = 1;
1996 last_used_at = ofport_get_usage(ofproto,
1997 u16_to_ofp(ofproto->alloc_port_no));
1998 if (!last_used_at) {
1999 port_idx = ofproto->alloc_port_no;
2001 } else if ( last_used_at < time_msec() - 60*60*1000) {
2002 /* If the port with ofport 'ofproto->alloc_port_no' was deleted
2003 * more than an hour ago, consider it usable. */
2004 ofport_remove_usage(ofproto,
2005 u16_to_ofp(ofproto->alloc_port_no));
2006 port_idx = ofproto->alloc_port_no;
2008 } else if (last_used_at < lru) {
2010 lru_ofport = ofproto->alloc_port_no;
2013 if (ofproto->alloc_port_no == end_port_no) {
2015 port_idx = lru_ofport;
2022 ofport_set_usage(ofproto, u16_to_ofp(port_idx), LLONG_MAX);
2023 return u16_to_ofp(port_idx);
2027 dealloc_ofp_port(struct ofproto *ofproto, ofp_port_t ofp_port)
2029 if (ofp_to_u16(ofp_port) < ofproto->max_ports) {
2030 ofport_set_usage(ofproto, ofp_port, time_msec());
2034 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
2035 * pointer if the netdev cannot be opened. On success, also fills in
2037 static struct netdev *
2038 ofport_open(struct ofproto *ofproto,
2039 struct ofproto_port *ofproto_port,
2040 struct ofputil_phy_port *pp)
2042 enum netdev_flags flags;
2043 struct netdev *netdev;
2046 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
2048 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
2049 "cannot be opened (%s)",
2051 ofproto_port->name, ofproto_port->ofp_port,
2052 ofproto_port->name, ovs_strerror(error));
2056 if (ofproto_port->ofp_port == OFPP_NONE) {
2057 if (!strcmp(ofproto->name, ofproto_port->name)) {
2058 ofproto_port->ofp_port = OFPP_LOCAL;
2060 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
2061 ofproto_port->name);
2064 pp->port_no = ofproto_port->ofp_port;
2065 netdev_get_etheraddr(netdev, pp->hw_addr);
2066 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
2067 netdev_get_flags(netdev, &flags);
2068 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
2069 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
2070 netdev_get_features(netdev, &pp->curr, &pp->advertised,
2071 &pp->supported, &pp->peer);
2072 pp->curr_speed = netdev_features_to_bps(pp->curr, 0) / 1000;
2073 pp->max_speed = netdev_features_to_bps(pp->supported, 0) / 1000;
2078 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
2079 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
2082 ofport_equal(const struct ofputil_phy_port *a,
2083 const struct ofputil_phy_port *b)
2085 return (eth_addr_equals(a->hw_addr, b->hw_addr)
2086 && a->state == b->state
2087 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
2088 && a->curr == b->curr
2089 && a->advertised == b->advertised
2090 && a->supported == b->supported
2091 && a->peer == b->peer
2092 && a->curr_speed == b->curr_speed
2093 && a->max_speed == b->max_speed);
2096 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
2097 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
2098 * one with the same name or port number). */
2100 ofport_install(struct ofproto *p,
2101 struct netdev *netdev, const struct ofputil_phy_port *pp)
2103 const char *netdev_name = netdev_get_name(netdev);
2104 struct ofport *ofport;
2107 /* Create ofport. */
2108 ofport = p->ofproto_class->port_alloc();
2113 ofport->ofproto = p;
2114 ofport->netdev = netdev;
2115 ofport->change_seq = netdev_change_seq(netdev);
2117 ofport->ofp_port = pp->port_no;
2118 ofport->created = time_msec();
2120 /* Add port to 'p'. */
2121 hmap_insert(&p->ports, &ofport->hmap_node,
2122 hash_ofp_port(ofport->ofp_port));
2123 shash_add(&p->port_by_name, netdev_name, ofport);
2125 update_mtu(p, ofport);
2127 /* Let the ofproto_class initialize its private data. */
2128 error = p->ofproto_class->port_construct(ofport);
2132 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
2136 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
2137 p->name, netdev_name, ovs_strerror(error));
2139 ofport_destroy__(ofport);
2141 netdev_close(netdev);
2145 /* Removes 'ofport' from 'p' and destroys it. */
2147 ofport_remove(struct ofport *ofport)
2149 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
2151 ofport_destroy(ofport);
2154 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
2157 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
2159 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
2161 ofport_remove(port);
2165 /* Updates 'port' with new 'pp' description.
2167 * Does not handle a name or port number change. The caller must implement
2168 * such a change as a delete followed by an add. */
2170 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
2172 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
2173 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
2174 | (pp->config & OFPUTIL_PC_PORT_DOWN));
2175 port->pp.state = pp->state;
2176 port->pp.curr = pp->curr;
2177 port->pp.advertised = pp->advertised;
2178 port->pp.supported = pp->supported;
2179 port->pp.peer = pp->peer;
2180 port->pp.curr_speed = pp->curr_speed;
2181 port->pp.max_speed = pp->max_speed;
2183 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
2186 /* Update OpenFlow 'state' in 'port' and notify controller. */
2188 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
2190 if (port->pp.state != state) {
2191 port->pp.state = state;
2192 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
2198 ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port)
2200 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
2202 if (port->ofproto->ofproto_class->set_realdev) {
2203 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
2205 if (port->ofproto->ofproto_class->set_stp_port) {
2206 port->ofproto->ofproto_class->set_stp_port(port, NULL);
2208 if (port->ofproto->ofproto_class->set_cfm) {
2209 port->ofproto->ofproto_class->set_cfm(port, NULL);
2211 if (port->ofproto->ofproto_class->bundle_remove) {
2212 port->ofproto->ofproto_class->bundle_remove(port);
2218 ofport_destroy__(struct ofport *port)
2220 struct ofproto *ofproto = port->ofproto;
2221 const char *name = netdev_get_name(port->netdev);
2223 hmap_remove(&ofproto->ports, &port->hmap_node);
2224 shash_delete(&ofproto->port_by_name,
2225 shash_find(&ofproto->port_by_name, name));
2227 netdev_close(port->netdev);
2228 ofproto->ofproto_class->port_dealloc(port);
2232 ofport_destroy(struct ofport *port)
2235 dealloc_ofp_port(port->ofproto, port->ofp_port);
2236 port->ofproto->ofproto_class->port_destruct(port);
2237 ofport_destroy__(port);
2242 ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
2244 struct ofport *port;
2246 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_ofp_port(ofp_port),
2248 if (port->ofp_port == ofp_port) {
2255 static long long int
2256 ofport_get_usage(const struct ofproto *ofproto, ofp_port_t ofp_port)
2258 struct ofport_usage *usage;
2260 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2261 &ofproto->ofport_usage) {
2262 if (usage->ofp_port == ofp_port) {
2263 return usage->last_used;
2270 ofport_set_usage(struct ofproto *ofproto, ofp_port_t ofp_port,
2271 long long int last_used)
2273 struct ofport_usage *usage;
2274 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2275 &ofproto->ofport_usage) {
2276 if (usage->ofp_port == ofp_port) {
2277 usage->last_used = last_used;
2281 ovs_assert(last_used == LLONG_MAX);
2283 usage = xmalloc(sizeof *usage);
2284 usage->ofp_port = ofp_port;
2285 usage->last_used = last_used;
2286 hmap_insert(&ofproto->ofport_usage, &usage->hmap_node,
2287 hash_ofp_port(ofp_port));
2291 ofport_remove_usage(struct ofproto *ofproto, ofp_port_t ofp_port)
2293 struct ofport_usage *usage;
2294 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2295 &ofproto->ofport_usage) {
2296 if (usage->ofp_port == ofp_port) {
2297 hmap_remove(&ofproto->ofport_usage, &usage->hmap_node);
2305 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
2307 struct ofproto *ofproto = port->ofproto;
2310 if (ofproto->ofproto_class->port_get_stats) {
2311 error = ofproto->ofproto_class->port_get_stats(port, stats);
2320 update_port(struct ofproto *ofproto, const char *name)
2322 struct ofproto_port ofproto_port;
2323 struct ofputil_phy_port pp;
2324 struct netdev *netdev;
2325 struct ofport *port;
2327 COVERAGE_INC(ofproto_update_port);
2329 /* Fetch 'name''s location and properties from the datapath. */
2330 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
2331 ? ofport_open(ofproto, &ofproto_port, &pp)
2335 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
2336 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
2337 struct netdev *old_netdev = port->netdev;
2339 /* 'name' hasn't changed location. Any properties changed? */
2340 if (!ofport_equal(&port->pp, &pp)) {
2341 ofport_modified(port, &pp);
2344 update_mtu(ofproto, port);
2346 /* Install the newly opened netdev in case it has changed.
2347 * Don't close the old netdev yet in case port_modified has to
2348 * remove a retained reference to it.*/
2349 port->netdev = netdev;
2350 port->change_seq = netdev_change_seq(netdev);
2352 if (port->ofproto->ofproto_class->port_modified) {
2353 port->ofproto->ofproto_class->port_modified(port);
2356 netdev_close(old_netdev);
2358 /* If 'port' is nonnull then its name differs from 'name' and thus
2359 * we should delete it. If we think there's a port named 'name'
2360 * then its port number must be wrong now so delete it too. */
2362 ofport_remove(port);
2364 ofport_remove_with_name(ofproto, name);
2365 ofport_install(ofproto, netdev, &pp);
2368 /* Any port named 'name' is gone now. */
2369 ofport_remove_with_name(ofproto, name);
2371 ofproto_port_destroy(&ofproto_port);
2375 init_ports(struct ofproto *p)
2377 struct ofproto_port_dump dump;
2378 struct ofproto_port ofproto_port;
2379 struct shash_node *node, *next;
2381 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2382 const char *name = ofproto_port.name;
2384 if (shash_find(&p->port_by_name, name)) {
2385 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2388 struct ofputil_phy_port pp;
2389 struct netdev *netdev;
2391 /* Check if an OpenFlow port number had been requested. */
2392 node = shash_find(&init_ofp_ports, name);
2394 const struct iface_hint *iface_hint = node->data;
2395 simap_put(&p->ofp_requests, name,
2396 ofp_to_u16(iface_hint->ofp_port));
2399 netdev = ofport_open(p, &ofproto_port, &pp);
2401 ofport_install(p, netdev, &pp);
2402 if (ofp_to_u16(ofproto_port.ofp_port) < p->max_ports) {
2403 p->alloc_port_no = MAX(p->alloc_port_no,
2404 ofp_to_u16(ofproto_port.ofp_port));
2410 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2411 struct iface_hint *iface_hint = node->data;
2413 if (!strcmp(iface_hint->br_name, p->name)) {
2414 free(iface_hint->br_name);
2415 free(iface_hint->br_type);
2417 shash_delete(&init_ofp_ports, node);
2424 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2425 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2427 find_min_mtu(struct ofproto *p)
2429 struct ofport *ofport;
2432 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2433 struct netdev *netdev = ofport->netdev;
2436 /* Skip any internal ports, since that's what we're trying to
2438 if (!strcmp(netdev_get_type(netdev), "internal")) {
2442 if (netdev_get_mtu(netdev, &dev_mtu)) {
2445 if (!mtu || dev_mtu < mtu) {
2450 return mtu ? mtu: ETH_PAYLOAD_MAX;
2453 /* Update MTU of all datapath devices on 'p' to the minimum of the
2454 * non-datapath ports in event of 'port' added or changed. */
2456 update_mtu(struct ofproto *p, struct ofport *port)
2458 struct ofport *ofport;
2459 struct netdev *netdev = port->netdev;
2460 int dev_mtu, old_min;
2462 if (netdev_get_mtu(netdev, &dev_mtu)) {
2466 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2467 if (dev_mtu > p->min_mtu) {
2468 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2469 dev_mtu = p->min_mtu;
2472 port->mtu = dev_mtu;
2476 /* For non-internal port find new min mtu. */
2477 old_min = p->min_mtu;
2478 port->mtu = dev_mtu;
2479 p->min_mtu = find_min_mtu(p);
2480 if (p->min_mtu == old_min) {
2484 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2485 struct netdev *netdev = ofport->netdev;
2487 if (!strcmp(netdev_get_type(netdev), "internal")) {
2488 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2489 ofport->mtu = p->min_mtu;
2496 ofproto_rule_ref(struct rule *rule)
2501 atomic_add(&rule->ref_count, 1, &orig);
2502 ovs_assert(orig != 0);
2507 ofproto_rule_unref(struct rule *rule)
2512 atomic_sub(&rule->ref_count, 1, &orig);
2514 rule->ofproto->ofproto_class->rule_destruct(rule);
2515 ofproto_rule_destroy__(rule);
2517 ovs_assert(orig != 0);
2522 struct rule_actions *
2523 rule_get_actions(const struct rule *rule)
2524 OVS_EXCLUDED(rule->mutex)
2526 struct rule_actions *actions;
2528 ovs_mutex_lock(&rule->mutex);
2529 actions = rule_get_actions__(rule);
2530 ovs_mutex_unlock(&rule->mutex);
2535 struct rule_actions *
2536 rule_get_actions__(const struct rule *rule)
2537 OVS_REQUIRES(rule->mutex)
2539 rule_actions_ref(rule->actions);
2540 return rule->actions;
2544 ofproto_rule_destroy__(struct rule *rule)
2545 OVS_NO_THREAD_SAFETY_ANALYSIS
2547 cls_rule_destroy(CONST_CAST(struct cls_rule *, &rule->cr));
2548 rule_actions_unref(rule->actions);
2549 ovs_mutex_destroy(&rule->mutex);
2550 rule->ofproto->ofproto_class->rule_dealloc(rule);
2553 static uint32_t get_provider_meter_id(const struct ofproto *,
2554 uint32_t of_meter_id);
2556 /* Creates and returns a new 'struct rule_actions', with a ref_count of 1,
2557 * whose actions are a copy of from the 'ofpacts_len' bytes of 'ofpacts'. */
2558 struct rule_actions *
2559 rule_actions_create(const struct ofproto *ofproto,
2560 const struct ofpact *ofpacts, size_t ofpacts_len)
2562 struct rule_actions *actions;
2564 actions = xmalloc(sizeof *actions);
2565 atomic_init(&actions->ref_count, 1);
2566 actions->ofpacts = xmemdup(ofpacts, ofpacts_len);
2567 actions->ofpacts_len = ofpacts_len;
2568 actions->provider_meter_id
2569 = get_provider_meter_id(ofproto,
2570 ofpacts_get_meter(ofpacts, ofpacts_len));
2575 /* Increments 'actions''s ref_count. */
2577 rule_actions_ref(struct rule_actions *actions)
2582 atomic_add(&actions->ref_count, 1, &orig);
2583 ovs_assert(orig != 0);
2587 /* Decrements 'actions''s ref_count and frees 'actions' if the ref_count
2590 rule_actions_unref(struct rule_actions *actions)
2595 atomic_sub(&actions->ref_count, 1, &orig);
2597 free(actions->ofpacts);
2600 ovs_assert(orig != 0);
2605 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2606 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2608 ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port)
2609 OVS_REQUIRES(ofproto_mutex)
2611 return (port == OFPP_ANY
2612 || ofpacts_output_to_port(rule->actions->ofpacts,
2613 rule->actions->ofpacts_len, port));
2616 /* Returns true if 'rule' has group and equals group_id. */
2618 ofproto_rule_has_out_group(const struct rule *rule, uint32_t group_id)
2619 OVS_REQUIRES(ofproto_mutex)
2621 return (group_id == OFPG11_ANY
2622 || ofpacts_output_to_group(rule->actions->ofpacts,
2623 rule->actions->ofpacts_len, group_id));
2626 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2627 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2629 ofoperation_has_out_port(const struct ofoperation *op, ofp_port_t out_port)
2630 OVS_REQUIRES(ofproto_mutex)
2632 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2637 case OFOPERATION_ADD:
2638 case OFOPERATION_DELETE:
2641 case OFOPERATION_MODIFY:
2642 case OFOPERATION_REPLACE:
2643 return ofpacts_output_to_port(op->actions->ofpacts,
2644 op->actions->ofpacts_len, out_port);
2651 rule_execute_destroy(struct rule_execute *e)
2653 ofproto_rule_unref(e->rule);
2654 list_remove(&e->list_node);
2658 /* Executes all "rule_execute" operations queued up in ofproto->rule_executes,
2659 * by passing them to the ofproto provider. */
2661 run_rule_executes(struct ofproto *ofproto)
2662 OVS_EXCLUDED(ofproto_mutex)
2664 struct rule_execute *e, *next;
2665 struct list executes;
2667 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2668 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2669 union flow_in_port in_port_;
2672 in_port_.ofp_port = e->in_port;
2673 flow_extract(e->packet, 0, 0, NULL, &in_port_, &flow);
2674 ofproto->ofproto_class->rule_execute(e->rule, &flow, e->packet);
2676 rule_execute_destroy(e);
2680 /* Destroys and discards all "rule_execute" operations queued up in
2681 * ofproto->rule_executes. */
2683 destroy_rule_executes(struct ofproto *ofproto)
2685 struct rule_execute *e, *next;
2686 struct list executes;
2688 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2689 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2690 ofpbuf_delete(e->packet);
2691 rule_execute_destroy(e);
2695 /* Returns true if 'rule' should be hidden from the controller.
2697 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2698 * (e.g. by in-band control) and are intentionally hidden from the
2701 ofproto_rule_is_hidden(const struct rule *rule)
2703 return rule->cr.priority > UINT16_MAX;
2706 static enum oftable_flags
2707 rule_get_flags(const struct rule *rule)
2709 return rule->ofproto->tables[rule->table_id].flags;
2713 rule_is_modifiable(const struct rule *rule)
2715 return !(rule_get_flags(rule) & OFTABLE_READONLY);
2719 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2721 ofconn_send_reply(ofconn, make_echo_reply(oh));
2726 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2728 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2729 struct ofputil_switch_features features;
2730 struct ofport *port;
2736 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2738 ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2740 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2741 * if present, are always at the end.) */
2742 n_tables = ofproto->n_tables;
2743 for (i = 0; i < ofproto->n_tables; i++) {
2744 if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
2750 features.datapath_id = ofproto->datapath_id;
2751 features.n_buffers = pktbuf_capacity();
2752 features.n_tables = n_tables;
2753 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2754 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2756 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2758 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2759 features.auxiliary_id = 0;
2760 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2762 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2763 ofputil_put_switch_features_port(&port->pp, b);
2766 ofconn_send_reply(ofconn, b);
2771 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2773 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2774 struct ofp_switch_config *osc;
2775 enum ofp_config_flags flags;
2779 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2780 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2781 flags = ofproto->frag_handling;
2782 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2783 if (oh->version < OFP13_VERSION
2784 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2785 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2787 osc->flags = htons(flags);
2788 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2789 ofconn_send_reply(ofconn, buf);
2795 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2797 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2798 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2799 uint16_t flags = ntohs(osc->flags);
2801 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2802 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
2803 enum ofp_config_flags cur = ofproto->frag_handling;
2804 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2806 ovs_assert((cur & OFPC_FRAG_MASK) == cur);
2808 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2809 ofproto->frag_handling = next;
2811 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2813 ofputil_frag_handling_to_string(next));
2817 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2818 ofconn_set_invalid_ttl_to_controller(ofconn,
2819 (oh->version < OFP13_VERSION
2820 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2822 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2827 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2828 * error message code for the caller to propagate upward. Otherwise, returns
2831 * The log message mentions 'msg_type'. */
2833 reject_slave_controller(struct ofconn *ofconn)
2835 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2836 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
2837 return OFPERR_OFPBRC_EPERM;
2843 /* Checks that the 'ofpacts_len' bytes of action in 'ofpacts' are appropriate
2846 * - If they use a meter, then 'ofproto' has that meter configured.
2848 * - If they use any groups, then 'ofproto' has that group configured.
2850 * Returns 0 if successful, otherwise an OpenFlow error. */
2852 ofproto_check_ofpacts(struct ofproto *ofproto,
2853 const struct ofpact ofpacts[], size_t ofpacts_len)
2855 const struct ofpact *a;
2858 mid = ofpacts_get_meter(ofpacts, ofpacts_len);
2859 if (mid && get_provider_meter_id(ofproto, mid) == UINT32_MAX) {
2860 return OFPERR_OFPMMFC_INVALID_METER;
2863 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2864 if (a->type == OFPACT_GROUP
2865 && !ofproto_group_exists(ofproto, ofpact_get_GROUP(a)->group_id)) {
2866 return OFPERR_OFPBAC_BAD_OUT_GROUP;
2874 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2876 struct ofproto *p = ofconn_get_ofproto(ofconn);
2877 struct ofputil_packet_out po;
2878 struct ofpbuf *payload;
2879 uint64_t ofpacts_stub[1024 / 8];
2880 struct ofpbuf ofpacts;
2882 union flow_in_port in_port_;
2885 COVERAGE_INC(ofproto_packet_out);
2887 error = reject_slave_controller(ofconn);
2892 /* Decode message. */
2893 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2894 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2896 goto exit_free_ofpacts;
2898 if (ofp_to_u16(po.in_port) >= p->max_ports
2899 && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) {
2900 error = OFPERR_OFPBRC_BAD_PORT;
2901 goto exit_free_ofpacts;
2906 if (po.buffer_id != UINT32_MAX) {
2907 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2908 if (error || !payload) {
2909 goto exit_free_ofpacts;
2912 /* Ensure that the L3 header is 32-bit aligned. */
2913 payload = ofpbuf_clone_data_with_headroom(po.packet, po.packet_len, 2);
2916 /* Verify actions against packet, then send packet if successful. */
2917 in_port_.ofp_port = po.in_port;
2918 flow_extract(payload, 0, 0, NULL, &in_port_, &flow);
2919 error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len);
2921 error = p->ofproto_class->packet_out(p, payload, &flow,
2922 po.ofpacts, po.ofpacts_len);
2924 ofpbuf_delete(payload);
2927 ofpbuf_uninit(&ofpacts);
2933 update_port_config(struct ofport *port,
2934 enum ofputil_port_config config,
2935 enum ofputil_port_config mask)
2937 enum ofputil_port_config old_config = port->pp.config;
2938 enum ofputil_port_config toggle;
2940 toggle = (config ^ port->pp.config) & mask;
2941 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2942 if (config & OFPUTIL_PC_PORT_DOWN) {
2943 netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL);
2945 netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL);
2947 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2950 port->pp.config ^= toggle;
2951 if (port->pp.config != old_config) {
2952 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2957 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2959 struct ofproto *p = ofconn_get_ofproto(ofconn);
2960 struct ofputil_port_mod pm;
2961 struct ofport *port;
2964 error = reject_slave_controller(ofconn);
2969 error = ofputil_decode_port_mod(oh, &pm);
2974 port = ofproto_get_port(p, pm.port_no);
2976 return OFPERR_OFPPMFC_BAD_PORT;
2977 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2978 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2980 update_port_config(port, pm.config, pm.mask);
2982 netdev_set_advertisements(port->netdev, pm.advertise);
2989 handle_desc_stats_request(struct ofconn *ofconn,
2990 const struct ofp_header *request)
2992 static const char *default_mfr_desc = "Nicira, Inc.";
2993 static const char *default_hw_desc = "Open vSwitch";
2994 static const char *default_sw_desc = VERSION;
2995 static const char *default_serial_desc = "None";
2996 static const char *default_dp_desc = "None";
2998 struct ofproto *p = ofconn_get_ofproto(ofconn);
2999 struct ofp_desc_stats *ods;
3002 msg = ofpraw_alloc_stats_reply(request, 0);
3003 ods = ofpbuf_put_zeros(msg, sizeof *ods);
3004 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
3005 sizeof ods->mfr_desc);
3006 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
3007 sizeof ods->hw_desc);
3008 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
3009 sizeof ods->sw_desc);
3010 ovs_strlcpy(ods->serial_num,
3011 p->serial_desc ? p->serial_desc : default_serial_desc,
3012 sizeof ods->serial_num);
3013 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
3014 sizeof ods->dp_desc);
3015 ofconn_send_reply(ofconn, msg);
3021 handle_table_stats_request(struct ofconn *ofconn,
3022 const struct ofp_header *request)
3024 struct ofproto *p = ofconn_get_ofproto(ofconn);
3025 struct ofp12_table_stats *ots;
3030 /* Set up default values.
3032 * ofp12_table_stats is used as a generic structure as
3033 * it is able to hold all the fields for ofp10_table_stats
3034 * and ofp11_table_stats (and of course itself).
3036 ots = xcalloc(p->n_tables, sizeof *ots);
3037 for (i = 0; i < p->n_tables; i++) {
3038 ots[i].table_id = i;
3039 sprintf(ots[i].name, "table%zu", i);
3040 ots[i].match = htonll(OFPXMT13_MASK);
3041 ots[i].wildcards = htonll(OFPXMT13_MASK);
3042 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
3043 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
3044 ots[i].write_setfields = htonll(OFPXMT13_MASK);
3045 ots[i].apply_setfields = htonll(OFPXMT13_MASK);
3046 ots[i].metadata_match = OVS_BE64_MAX;
3047 ots[i].metadata_write = OVS_BE64_MAX;
3048 ots[i].instructions = htonl(OFPIT11_ALL);
3049 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
3050 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
3051 ovs_rwlock_rdlock(&p->tables[i].cls.rwlock);
3052 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
3053 ovs_rwlock_unlock(&p->tables[i].cls.rwlock);
3056 p->ofproto_class->get_tables(p, ots);
3058 /* Post-process the tables, dropping hidden tables. */
3059 n_tables = p->n_tables;
3060 for (i = 0; i < p->n_tables; i++) {
3061 const struct oftable *table = &p->tables[i];
3063 if (table->flags & OFTABLE_HIDDEN) {
3069 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
3072 if (table->max_flows < ntohl(ots[i].max_entries)) {
3073 ots[i].max_entries = htonl(table->max_flows);
3077 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
3078 ofconn_send_reply(ofconn, msg);
3086 append_port_stat(struct ofport *port, struct list *replies)
3088 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
3090 calc_duration(port->created, time_msec(),
3091 &ops.duration_sec, &ops.duration_nsec);
3093 /* Intentionally ignore return value, since errors will set
3094 * 'stats' to all-1s, which is correct for OpenFlow, and
3095 * netdev_get_stats() will log errors. */
3096 ofproto_port_get_stats(port, &ops.stats);
3098 ofputil_append_port_stat(replies, &ops);
3102 handle_port_stats_request(struct ofconn *ofconn,
3103 const struct ofp_header *request)
3105 struct ofproto *p = ofconn_get_ofproto(ofconn);
3106 struct ofport *port;
3107 struct list replies;
3111 error = ofputil_decode_port_stats_request(request, &port_no);
3116 ofpmp_init(&replies, request);
3117 if (port_no != OFPP_ANY) {
3118 port = ofproto_get_port(p, port_no);
3120 append_port_stat(port, &replies);
3123 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3124 append_port_stat(port, &replies);
3128 ofconn_send_replies(ofconn, &replies);
3133 handle_port_desc_stats_request(struct ofconn *ofconn,
3134 const struct ofp_header *request)
3136 struct ofproto *p = ofconn_get_ofproto(ofconn);
3137 enum ofp_version version;
3138 struct ofport *port;
3139 struct list replies;
3141 ofpmp_init(&replies, request);
3143 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
3144 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3145 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
3148 ofconn_send_replies(ofconn, &replies);
3153 hash_cookie(ovs_be64 cookie)
3155 return hash_2words((OVS_FORCE uint64_t)cookie >> 32,
3156 (OVS_FORCE uint64_t)cookie);
3160 cookies_insert(struct ofproto *ofproto, struct rule *rule)
3161 OVS_REQUIRES(ofproto_mutex)
3163 hindex_insert(&ofproto->cookies, &rule->cookie_node,
3164 hash_cookie(rule->flow_cookie));
3168 cookies_remove(struct ofproto *ofproto, struct rule *rule)
3169 OVS_REQUIRES(ofproto_mutex)
3171 hindex_remove(&ofproto->cookies, &rule->cookie_node);
3175 ofproto_rule_change_cookie(struct ofproto *ofproto, struct rule *rule,
3176 ovs_be64 new_cookie)
3177 OVS_REQUIRES(ofproto_mutex)
3179 if (new_cookie != rule->flow_cookie) {
3180 cookies_remove(ofproto, rule);
3182 ovs_mutex_lock(&rule->mutex);
3183 rule->flow_cookie = new_cookie;
3184 ovs_mutex_unlock(&rule->mutex);
3186 cookies_insert(ofproto, rule);
3191 calc_duration(long long int start, long long int now,
3192 uint32_t *sec, uint32_t *nsec)
3194 long long int msecs = now - start;
3195 *sec = msecs / 1000;
3196 *nsec = (msecs % 1000) * (1000 * 1000);
3199 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
3200 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
3202 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
3204 return (table_id == 0xff || table_id < ofproto->n_tables
3206 : OFPERR_OFPBRC_BAD_TABLE_ID);
3210 static struct oftable *
3211 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
3213 struct oftable *table;
3215 for (table = &ofproto->tables[table_id];
3216 table < &ofproto->tables[ofproto->n_tables];
3218 if (!(table->flags & OFTABLE_HIDDEN)) {
3226 static struct oftable *
3227 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
3229 if (table_id == 0xff) {
3230 return next_visible_table(ofproto, 0);
3231 } else if (table_id < ofproto->n_tables) {
3232 return &ofproto->tables[table_id];
3238 static struct oftable *
3239 next_matching_table(const struct ofproto *ofproto,
3240 const struct oftable *table, uint8_t table_id)
3242 return (table_id == 0xff
3243 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
3247 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
3249 * - If TABLE_ID is 0xff, this iterates over every classifier table in
3250 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
3252 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
3253 * only once, for that table. (This can be used to access tables marked
3256 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
3257 * entered at all. (Perhaps you should have validated TABLE_ID with
3258 * check_table_id().)
3260 * All parameters are evaluated multiple times.
3262 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
3263 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
3265 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
3267 /* Initializes 'criteria' in a straightforward way based on the other
3270 * For "loose" matching, the 'priority' parameter is unimportant and may be
3273 rule_criteria_init(struct rule_criteria *criteria, uint8_t table_id,
3274 const struct match *match, unsigned int priority,
3275 ovs_be64 cookie, ovs_be64 cookie_mask,
3276 ofp_port_t out_port, uint32_t out_group)
3278 criteria->table_id = table_id;
3279 cls_rule_init(&criteria->cr, match, priority);
3280 criteria->cookie = cookie;
3281 criteria->cookie_mask = cookie_mask;
3282 criteria->out_port = out_port;
3283 criteria->out_group = out_group;
3287 rule_criteria_destroy(struct rule_criteria *criteria)
3289 cls_rule_destroy(&criteria->cr);
3293 rule_collection_init(struct rule_collection *rules)
3295 rules->rules = rules->stub;
3297 rules->capacity = ARRAY_SIZE(rules->stub);
3301 rule_collection_add(struct rule_collection *rules, struct rule *rule)
3303 if (rules->n >= rules->capacity) {
3304 size_t old_size, new_size;
3306 old_size = rules->capacity * sizeof *rules->rules;
3307 rules->capacity *= 2;
3308 new_size = rules->capacity * sizeof *rules->rules;
3310 if (rules->rules == rules->stub) {
3311 rules->rules = xmalloc(new_size);
3312 memcpy(rules->rules, rules->stub, old_size);
3314 rules->rules = xrealloc(rules->rules, new_size);
3318 rules->rules[rules->n++] = rule;
3322 rule_collection_ref(struct rule_collection *rules)
3323 OVS_REQUIRES(ofproto_mutex)
3327 for (i = 0; i < rules->n; i++) {
3328 ofproto_rule_ref(rules->rules[i]);
3333 rule_collection_unref(struct rule_collection *rules)
3337 for (i = 0; i < rules->n; i++) {
3338 ofproto_rule_unref(rules->rules[i]);
3343 rule_collection_destroy(struct rule_collection *rules)
3345 if (rules->rules != rules->stub) {
3351 collect_rule(struct rule *rule, const struct rule_criteria *c,
3352 struct rule_collection *rules)
3353 OVS_REQUIRES(ofproto_mutex)
3355 /* We ordinarily want to skip hidden rules, but there has to be a way for
3356 * code internal to OVS to modify and delete them, so if the criteria
3357 * specify a priority that can only be for a hidden flow, then allow hidden
3358 * rules to be selected. (This doesn't allow OpenFlow clients to meddle
3359 * with hidden flows because OpenFlow uses only a 16-bit field to specify
3361 if (ofproto_rule_is_hidden(rule) && c->cr.priority <= UINT16_MAX) {
3363 } else if (rule->pending) {
3364 return OFPROTO_POSTPONE;
3366 if ((c->table_id == rule->table_id || c->table_id == 0xff)
3367 && ofproto_rule_has_out_port(rule, c->out_port)
3368 && ofproto_rule_has_out_group(rule, c->out_group)
3369 && !((rule->flow_cookie ^ c->cookie) & c->cookie_mask)) {
3370 rule_collection_add(rules, rule);
3376 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3377 * on classifiers rules are done in the "loose" way required for OpenFlow
3378 * OFPFC_MODIFY and OFPFC_DELETE requests. Puts the selected rules on list
3381 * Hidden rules are always omitted.
3383 * Returns 0 on success, otherwise an OpenFlow error code. */
3385 collect_rules_loose(struct ofproto *ofproto,
3386 const struct rule_criteria *criteria,
3387 struct rule_collection *rules)
3388 OVS_REQUIRES(ofproto_mutex)
3390 struct oftable *table;
3393 rule_collection_init(rules);
3395 error = check_table_id(ofproto, criteria->table_id);
3400 if (criteria->cookie_mask == OVS_BE64_MAX) {
3403 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
3404 hash_cookie(criteria->cookie),
3405 &ofproto->cookies) {
3406 if (cls_rule_is_loose_match(&rule->cr, &criteria->cr.match)) {
3407 error = collect_rule(rule, criteria, rules);
3414 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
3415 struct cls_cursor cursor;
3418 ovs_rwlock_rdlock(&table->cls.rwlock);
3419 cls_cursor_init(&cursor, &table->cls, &criteria->cr);
3420 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3421 error = collect_rule(rule, criteria, rules);
3426 ovs_rwlock_unlock(&table->cls.rwlock);
3432 rule_collection_destroy(rules);
3437 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3438 * on classifiers rules are done in the "strict" way required for OpenFlow
3439 * OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests. Puts the selected
3440 * rules on list 'rules'.
3442 * Hidden rules are always omitted.
3444 * Returns 0 on success, otherwise an OpenFlow error code. */
3446 collect_rules_strict(struct ofproto *ofproto,
3447 const struct rule_criteria *criteria,
3448 struct rule_collection *rules)
3449 OVS_REQUIRES(ofproto_mutex)
3451 struct oftable *table;
3454 rule_collection_init(rules);
3456 error = check_table_id(ofproto, criteria->table_id);
3461 if (criteria->cookie_mask == OVS_BE64_MAX) {
3464 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
3465 hash_cookie(criteria->cookie),
3466 &ofproto->cookies) {
3467 if (cls_rule_equal(&rule->cr, &criteria->cr)) {
3468 error = collect_rule(rule, criteria, rules);
3475 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
3478 ovs_rwlock_rdlock(&table->cls.rwlock);
3479 rule = rule_from_cls_rule(classifier_find_rule_exactly(
3480 &table->cls, &criteria->cr));
3481 ovs_rwlock_unlock(&table->cls.rwlock);
3483 error = collect_rule(rule, criteria, rules);
3493 rule_collection_destroy(rules);
3498 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
3499 * forced into the range of a uint16_t. */
3501 age_secs(long long int age_ms)
3503 return (age_ms < 0 ? 0
3504 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
3505 : (unsigned int) age_ms / 1000);
3509 handle_flow_stats_request(struct ofconn *ofconn,
3510 const struct ofp_header *request)
3511 OVS_EXCLUDED(ofproto_mutex)
3513 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3514 struct ofputil_flow_stats_request fsr;
3515 struct rule_criteria criteria;
3516 struct rule_collection rules;
3517 struct list replies;
3521 error = ofputil_decode_flow_stats_request(&fsr, request);
3526 rule_criteria_init(&criteria, fsr.table_id, &fsr.match, 0, fsr.cookie,
3527 fsr.cookie_mask, fsr.out_port, fsr.out_group);
3529 ovs_mutex_lock(&ofproto_mutex);
3530 error = collect_rules_loose(ofproto, &criteria, &rules);
3531 rule_criteria_destroy(&criteria);
3533 rule_collection_ref(&rules);
3535 ovs_mutex_unlock(&ofproto_mutex);
3541 ofpmp_init(&replies, request);
3542 for (i = 0; i < rules.n; i++) {
3543 struct rule *rule = rules.rules[i];
3544 long long int now = time_msec();
3545 struct ofputil_flow_stats fs;
3546 long long int created, used, modified;
3547 struct rule_actions *actions;
3548 enum ofputil_flow_mod_flags flags;
3550 ovs_mutex_lock(&rule->mutex);
3551 fs.cookie = rule->flow_cookie;
3552 fs.idle_timeout = rule->idle_timeout;
3553 fs.hard_timeout = rule->hard_timeout;
3554 created = rule->created;
3556 modified = rule->modified;
3557 actions = rule_get_actions__(rule);
3558 flags = rule->flags;
3559 ovs_mutex_unlock(&rule->mutex);
3561 minimatch_expand(&rule->cr.match, &fs.match);
3562 fs.table_id = rule->table_id;
3563 calc_duration(created, now, &fs.duration_sec, &fs.duration_nsec);
3564 fs.priority = rule->cr.priority;
3565 fs.idle_age = age_secs(now - used);
3566 fs.hard_age = age_secs(now - modified);
3567 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
3569 fs.ofpacts = actions->ofpacts;
3570 fs.ofpacts_len = actions->ofpacts_len;
3573 ofputil_append_flow_stats_reply(&fs, &replies);
3575 rule_actions_unref(actions);
3578 rule_collection_unref(&rules);
3579 rule_collection_destroy(&rules);
3581 ofconn_send_replies(ofconn, &replies);
3587 flow_stats_ds(struct rule *rule, struct ds *results)
3589 uint64_t packet_count, byte_count;
3590 struct rule_actions *actions;
3591 long long int created;
3593 rule->ofproto->ofproto_class->rule_get_stats(rule,
3594 &packet_count, &byte_count);
3596 ovs_mutex_lock(&rule->mutex);
3597 actions = rule_get_actions__(rule);
3598 created = rule->created;
3599 ovs_mutex_unlock(&rule->mutex);
3601 if (rule->table_id != 0) {
3602 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
3604 ds_put_format(results, "duration=%llds, ", (time_msec() - created) / 1000);
3605 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3606 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3607 cls_rule_format(&rule->cr, results);
3608 ds_put_char(results, ',');
3610 ds_put_cstr(results, "actions=");
3611 ofpacts_format(actions->ofpacts, actions->ofpacts_len, results);
3613 ds_put_cstr(results, "\n");
3615 rule_actions_unref(actions);
3618 /* Adds a pretty-printed description of all flows to 'results', including
3619 * hidden flows (e.g., set up by in-band control). */
3621 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3623 struct oftable *table;
3625 OFPROTO_FOR_EACH_TABLE (table, p) {
3626 struct cls_cursor cursor;
3629 ovs_rwlock_rdlock(&table->cls.rwlock);
3630 cls_cursor_init(&cursor, &table->cls, NULL);
3631 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3632 flow_stats_ds(rule, results);
3634 ovs_rwlock_unlock(&table->cls.rwlock);
3638 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
3639 * '*engine_type' and '*engine_id', respectively. */
3641 ofproto_get_netflow_ids(const struct ofproto *ofproto,
3642 uint8_t *engine_type, uint8_t *engine_id)
3644 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
3647 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
3648 * true if the port's CFM status was successfully stored into '*status'.
3649 * Returns false if the port did not have CFM configured, in which case
3650 * '*status' is indeterminate.
3652 * The caller must provide and owns '*status', and must free 'status->rmps'. */
3654 ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port,
3655 struct ofproto_cfm_status *status)
3657 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
3659 && ofproto->ofproto_class->get_cfm_status
3660 && ofproto->ofproto_class->get_cfm_status(ofport, status));
3664 handle_aggregate_stats_request(struct ofconn *ofconn,
3665 const struct ofp_header *oh)
3666 OVS_EXCLUDED(ofproto_mutex)
3668 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3669 struct ofputil_flow_stats_request request;
3670 struct ofputil_aggregate_stats stats;
3671 bool unknown_packets, unknown_bytes;
3672 struct rule_criteria criteria;
3673 struct rule_collection rules;
3674 struct ofpbuf *reply;
3678 error = ofputil_decode_flow_stats_request(&request, oh);
3683 rule_criteria_init(&criteria, request.table_id, &request.match, 0,
3684 request.cookie, request.cookie_mask,
3685 request.out_port, request.out_group);
3687 ovs_mutex_lock(&ofproto_mutex);
3688 error = collect_rules_loose(ofproto, &criteria, &rules);
3689 rule_criteria_destroy(&criteria);
3691 rule_collection_ref(&rules);
3693 ovs_mutex_unlock(&ofproto_mutex);
3699 memset(&stats, 0, sizeof stats);
3700 unknown_packets = unknown_bytes = false;
3701 for (i = 0; i < rules.n; i++) {
3702 struct rule *rule = rules.rules[i];
3703 uint64_t packet_count;
3704 uint64_t byte_count;
3706 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
3709 if (packet_count == UINT64_MAX) {
3710 unknown_packets = true;
3712 stats.packet_count += packet_count;
3715 if (byte_count == UINT64_MAX) {
3716 unknown_bytes = true;
3718 stats.byte_count += byte_count;
3723 if (unknown_packets) {
3724 stats.packet_count = UINT64_MAX;
3726 if (unknown_bytes) {
3727 stats.byte_count = UINT64_MAX;
3730 rule_collection_unref(&rules);
3731 rule_collection_destroy(&rules);
3733 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
3734 ofconn_send_reply(ofconn, reply);
3739 struct queue_stats_cbdata {
3740 struct ofport *ofport;
3741 struct list replies;
3746 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3747 const struct netdev_queue_stats *stats)
3749 struct ofputil_queue_stats oqs;
3751 oqs.port_no = cbdata->ofport->pp.port_no;
3752 oqs.queue_id = queue_id;
3753 oqs.tx_bytes = stats->tx_bytes;
3754 oqs.tx_packets = stats->tx_packets;
3755 oqs.tx_errors = stats->tx_errors;
3756 if (stats->created != LLONG_MIN) {
3757 calc_duration(stats->created, cbdata->now,
3758 &oqs.duration_sec, &oqs.duration_nsec);
3760 oqs.duration_sec = oqs.duration_nsec = UINT32_MAX;
3762 ofputil_append_queue_stat(&cbdata->replies, &oqs);
3766 handle_queue_stats_dump_cb(uint32_t queue_id,
3767 struct netdev_queue_stats *stats,
3770 struct queue_stats_cbdata *cbdata = cbdata_;
3772 put_queue_stats(cbdata, queue_id, stats);
3776 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3777 struct queue_stats_cbdata *cbdata)
3779 cbdata->ofport = port;
3780 if (queue_id == OFPQ_ALL) {
3781 netdev_dump_queue_stats(port->netdev,
3782 handle_queue_stats_dump_cb, cbdata);
3784 struct netdev_queue_stats stats;
3786 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3787 put_queue_stats(cbdata, queue_id, &stats);
3789 return OFPERR_OFPQOFC_BAD_QUEUE;
3796 handle_queue_stats_request(struct ofconn *ofconn,
3797 const struct ofp_header *rq)
3799 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3800 struct queue_stats_cbdata cbdata;
3801 struct ofport *port;
3803 struct ofputil_queue_stats_request oqsr;
3805 COVERAGE_INC(ofproto_queue_req);
3807 ofpmp_init(&cbdata.replies, rq);
3808 cbdata.now = time_msec();
3810 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3815 if (oqsr.port_no == OFPP_ANY) {
3816 error = OFPERR_OFPQOFC_BAD_QUEUE;
3817 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3818 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3823 port = ofproto_get_port(ofproto, oqsr.port_no);
3825 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3826 : OFPERR_OFPQOFC_BAD_PORT);
3829 ofconn_send_replies(ofconn, &cbdata.replies);
3831 ofpbuf_list_delete(&cbdata.replies);
3838 is_flow_deletion_pending(const struct ofproto *ofproto,
3839 const struct cls_rule *cls_rule,
3841 OVS_REQUIRES(ofproto_mutex)
3843 if (!hmap_is_empty(&ofproto->deletions)) {
3844 struct ofoperation *op;
3846 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3847 cls_rule_hash(cls_rule, table_id),
3848 &ofproto->deletions) {
3849 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
3859 should_evict_a_rule(struct oftable *table, unsigned int extra_space)
3860 OVS_REQUIRES(ofproto_mutex)
3861 OVS_NO_THREAD_SAFETY_ANALYSIS
3863 return classifier_count(&table->cls) + extra_space > table->max_flows;
3867 evict_rules_from_table(struct ofproto *ofproto, struct oftable *table,
3868 unsigned int extra_space)
3869 OVS_REQUIRES(ofproto_mutex)
3871 while (should_evict_a_rule(table, extra_space)) {
3874 if (!choose_rule_to_evict(table, &rule)) {
3875 return OFPERR_OFPFMFC_TABLE_FULL;
3876 } else if (rule->pending) {
3877 return OFPROTO_POSTPONE;
3879 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
3880 delete_flow__(rule, group, OFPRR_EVICTION);
3881 ofopgroup_submit(group);
3888 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3889 * in which no matching flow already exists in the flow table.
3891 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3892 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3893 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3894 * initiated now but may be retried later.
3896 * The caller retains ownership of 'fm->ofpacts'.
3898 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3901 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
3902 struct ofputil_flow_mod *fm, const struct ofp_header *request)
3903 OVS_REQUIRES(ofproto_mutex)
3905 struct oftable *table;
3906 struct ofopgroup *group;
3912 error = check_table_id(ofproto, fm->table_id);
3918 if (fm->table_id == 0xff) {
3919 if (ofproto->ofproto_class->rule_choose_table) {
3920 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3926 ovs_assert(table_id < ofproto->n_tables);
3930 } else if (fm->table_id < ofproto->n_tables) {
3931 table_id = fm->table_id;
3933 return OFPERR_OFPBRC_BAD_TABLE_ID;
3936 table = &ofproto->tables[table_id];
3938 if (table->flags & OFTABLE_READONLY) {
3939 return OFPERR_OFPBRC_EPERM;
3942 cls_rule_init(&cr, &fm->match, fm->priority);
3944 /* Transform "add" into "modify" if there's an existing identical flow. */
3945 ovs_rwlock_rdlock(&table->cls.rwlock);
3946 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr));
3947 ovs_rwlock_unlock(&table->cls.rwlock);
3949 cls_rule_destroy(&cr);
3950 if (!rule_is_modifiable(rule)) {
3951 return OFPERR_OFPBRC_EPERM;
3952 } else if (rule->pending) {
3953 return OFPROTO_POSTPONE;
3955 struct rule_collection rules;
3957 rule_collection_init(&rules);
3958 rule_collection_add(&rules, rule);
3959 fm->modify_cookie = true;
3960 error = modify_flows__(ofproto, ofconn, fm, request, &rules);
3961 rule_collection_destroy(&rules);
3967 /* Serialize against pending deletion. */
3968 if (is_flow_deletion_pending(ofproto, &cr, table_id)) {
3969 cls_rule_destroy(&cr);
3970 return OFPROTO_POSTPONE;
3973 /* Check for overlap, if requested. */
3974 if (fm->flags & OFPUTIL_FF_CHECK_OVERLAP) {
3977 ovs_rwlock_rdlock(&table->cls.rwlock);
3978 overlaps = classifier_rule_overlaps(&table->cls, &cr);
3979 ovs_rwlock_unlock(&table->cls.rwlock);
3982 cls_rule_destroy(&cr);
3983 return OFPERR_OFPFMFC_OVERLAP;
3987 /* If necessary, evict an existing rule to clear out space. */
3988 error = evict_rules_from_table(ofproto, table, 1);
3990 cls_rule_destroy(&cr);
3994 /* Allocate new rule. */
3995 rule = ofproto->ofproto_class->rule_alloc();
3997 cls_rule_destroy(&cr);
3998 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
3999 ofproto->name, ovs_strerror(error));
4003 /* Initialize base state. */
4004 *CONST_CAST(struct ofproto **, &rule->ofproto) = ofproto;
4005 cls_rule_move(CONST_CAST(struct cls_rule *, &rule->cr), &cr);
4006 atomic_init(&rule->ref_count, 1);
4007 rule->pending = NULL;
4008 rule->flow_cookie = fm->new_cookie;
4009 rule->created = rule->modified = rule->used = time_msec();
4011 ovs_mutex_init(&rule->mutex);
4012 ovs_mutex_lock(&rule->mutex);
4013 rule->idle_timeout = fm->idle_timeout;
4014 rule->hard_timeout = fm->hard_timeout;
4015 ovs_mutex_unlock(&rule->mutex);
4017 *CONST_CAST(uint8_t *, &rule->table_id) = table - ofproto->tables;
4018 rule->flags = fm->flags & OFPUTIL_FF_STATE;
4019 rule->actions = rule_actions_create(ofproto, fm->ofpacts, fm->ofpacts_len);
4020 list_init(&rule->meter_list_node);
4021 rule->eviction_group = NULL;
4022 list_init(&rule->expirable);
4023 rule->monitor_flags = 0;
4024 rule->add_seqno = 0;
4025 rule->modify_seqno = 0;
4027 /* Construct rule, initializing derived state. */
4028 error = ofproto->ofproto_class->rule_construct(rule);
4030 ofproto_rule_destroy__(rule);
4035 oftable_insert_rule(rule);
4037 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
4038 ofoperation_create(group, rule, OFOPERATION_ADD, 0);
4039 ofproto->ofproto_class->rule_insert(rule);
4040 ofopgroup_submit(group);
4045 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
4047 /* Modifies the rules listed in 'rules', changing their actions to match those
4050 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4053 * Returns 0 on success, otherwise an OpenFlow error code. */
4055 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
4056 struct ofputil_flow_mod *fm, const struct ofp_header *request,
4057 const struct rule_collection *rules)
4058 OVS_REQUIRES(ofproto_mutex)
4060 enum ofoperation_type type;
4061 struct ofopgroup *group;
4065 type = fm->command == OFPFC_ADD ? OFOPERATION_REPLACE : OFOPERATION_MODIFY;
4066 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
4067 error = OFPERR_OFPBRC_EPERM;
4068 for (i = 0; i < rules->n; i++) {
4069 struct rule *rule = rules->rules[i];
4070 struct ofoperation *op;
4071 bool actions_changed;
4072 bool reset_counters;
4074 /* FIXME: Implement OFPFUTIL_FF_RESET_COUNTS */
4076 if (rule_is_modifiable(rule)) {
4077 /* At least one rule is modifiable, don't report EPERM error. */
4083 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
4084 rule->actions->ofpacts,
4085 rule->actions->ofpacts_len);
4087 op = ofoperation_create(group, rule, type, 0);
4089 if (fm->modify_cookie && fm->new_cookie != OVS_BE64_MAX) {
4090 ofproto_rule_change_cookie(ofproto, rule, fm->new_cookie);
4092 if (type == OFOPERATION_REPLACE) {
4093 ovs_mutex_lock(&rule->mutex);
4094 rule->idle_timeout = fm->idle_timeout;
4095 rule->hard_timeout = fm->hard_timeout;
4096 ovs_mutex_unlock(&rule->mutex);
4098 rule->flags = fm->flags & OFPUTIL_FF_STATE;
4099 if (fm->idle_timeout || fm->hard_timeout) {
4100 if (!rule->eviction_group) {
4101 eviction_group_add_rule(rule);
4104 eviction_group_remove_rule(rule);
4108 reset_counters = (fm->flags & OFPUTIL_FF_RESET_COUNTS) != 0;
4109 if (actions_changed || reset_counters) {
4110 struct rule_actions *new_actions;
4112 op->actions = rule->actions;
4113 new_actions = rule_actions_create(ofproto,
4114 fm->ofpacts, fm->ofpacts_len);
4116 ovs_mutex_lock(&rule->mutex);
4117 rule->actions = new_actions;
4118 ovs_mutex_unlock(&rule->mutex);
4120 rule->ofproto->ofproto_class->rule_modify_actions(rule,
4123 ofoperation_complete(op, 0);
4126 ofopgroup_submit(group);
4132 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
4133 struct ofputil_flow_mod *fm, const struct ofp_header *request)
4134 OVS_REQUIRES(ofproto_mutex)
4136 if (fm->cookie_mask != htonll(0) || fm->new_cookie == OVS_BE64_MAX) {
4139 return add_flow(ofproto, ofconn, fm, request);
4142 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
4145 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4148 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
4149 struct ofputil_flow_mod *fm,
4150 const struct ofp_header *request)
4151 OVS_REQUIRES(ofproto_mutex)
4153 struct rule_criteria criteria;
4154 struct rule_collection rules;
4157 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0,
4158 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY);
4159 error = collect_rules_loose(ofproto, &criteria, &rules);
4160 rule_criteria_destroy(&criteria);
4163 error = (rules.n > 0
4164 ? modify_flows__(ofproto, ofconn, fm, request, &rules)
4165 : modify_flows_add(ofproto, ofconn, fm, request));
4168 rule_collection_destroy(&rules);
4173 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4176 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4179 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
4180 struct ofputil_flow_mod *fm,
4181 const struct ofp_header *request)
4182 OVS_REQUIRES(ofproto_mutex)
4184 struct rule_criteria criteria;
4185 struct rule_collection rules;
4188 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
4189 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY);
4190 error = collect_rules_strict(ofproto, &criteria, &rules);
4191 rule_criteria_destroy(&criteria);
4195 error = modify_flows_add(ofproto, ofconn, fm, request);
4196 } else if (rules.n == 1) {
4197 error = modify_flows__(ofproto, ofconn, fm, request, &rules);
4201 rule_collection_destroy(&rules);
4206 /* OFPFC_DELETE implementation. */
4209 delete_flow__(struct rule *rule, struct ofopgroup *group,
4210 enum ofp_flow_removed_reason reason)
4211 OVS_REQUIRES(ofproto_mutex)
4213 struct ofproto *ofproto = rule->ofproto;
4215 ofproto_rule_send_removed(rule, reason);
4217 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
4218 oftable_remove_rule(rule);
4219 ofproto->ofproto_class->rule_delete(rule);
4222 /* Deletes the rules listed in 'rules'.
4224 * Returns 0 on success, otherwise an OpenFlow error code. */
4226 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
4227 const struct ofp_header *request,
4228 const struct rule_collection *rules,
4229 enum ofp_flow_removed_reason reason)
4230 OVS_REQUIRES(ofproto_mutex)
4232 struct ofopgroup *group;
4235 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
4236 for (i = 0; i < rules->n; i++) {
4237 delete_flow__(rules->rules[i], group, reason);
4239 ofopgroup_submit(group);
4244 /* Implements OFPFC_DELETE. */
4246 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
4247 const struct ofputil_flow_mod *fm,
4248 const struct ofp_header *request)
4249 OVS_REQUIRES(ofproto_mutex)
4251 struct rule_criteria criteria;
4252 struct rule_collection rules;
4255 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0,
4256 fm->cookie, fm->cookie_mask,
4257 fm->out_port, fm->out_group);
4258 error = collect_rules_loose(ofproto, &criteria, &rules);
4259 rule_criteria_destroy(&criteria);
4261 if (!error && rules.n > 0) {
4262 error = delete_flows__(ofproto, ofconn, request, &rules, OFPRR_DELETE);
4264 rule_collection_destroy(&rules);
4269 /* Implements OFPFC_DELETE_STRICT. */
4271 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
4272 const struct ofputil_flow_mod *fm,
4273 const struct ofp_header *request)
4274 OVS_REQUIRES(ofproto_mutex)
4276 struct rule_criteria criteria;
4277 struct rule_collection rules;
4280 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
4281 fm->cookie, fm->cookie_mask,
4282 fm->out_port, fm->out_group);
4283 error = collect_rules_strict(ofproto, &criteria, &rules);
4284 rule_criteria_destroy(&criteria);
4286 if (!error && rules.n > 0) {
4287 error = delete_flows__(ofproto, ofconn, request, &rules, OFPRR_DELETE);
4289 rule_collection_destroy(&rules);
4295 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
4296 OVS_REQUIRES(ofproto_mutex)
4298 struct ofputil_flow_removed fr;
4300 if (ofproto_rule_is_hidden(rule) ||
4301 !(rule->flags & OFPUTIL_FF_SEND_FLOW_REM)) {
4305 minimatch_expand(&rule->cr.match, &fr.match);
4306 fr.priority = rule->cr.priority;
4307 fr.cookie = rule->flow_cookie;
4309 fr.table_id = rule->table_id;
4310 calc_duration(rule->created, time_msec(),
4311 &fr.duration_sec, &fr.duration_nsec);
4312 ovs_mutex_lock(&rule->mutex);
4313 fr.idle_timeout = rule->idle_timeout;
4314 fr.hard_timeout = rule->hard_timeout;
4315 ovs_mutex_unlock(&rule->mutex);
4316 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
4319 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
4322 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
4323 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
4326 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
4329 * ofproto implementation ->run() functions should use this function to expire
4330 * OpenFlow flows. */
4332 ofproto_rule_expire(struct rule *rule, uint8_t reason)
4333 OVS_REQUIRES(ofproto_mutex)
4335 struct ofproto *ofproto = rule->ofproto;
4337 ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT
4338 || reason == OFPRR_DELETE || reason == OFPRR_GROUP_DELETE);
4340 ofproto_rule_delete__(ofproto, rule, reason);
4343 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
4344 * means "infinite". */
4346 reduce_timeout(uint16_t max, uint16_t *timeout)
4348 if (max && (!*timeout || *timeout > max)) {
4353 /* If 'idle_timeout' is nonzero, and 'rule' has no idle timeout or an idle
4354 * timeout greater than 'idle_timeout', lowers 'rule''s idle timeout to
4355 * 'idle_timeout' seconds. Similarly for 'hard_timeout'.
4357 * Suitable for implementing OFPACT_FIN_TIMEOUT. */
4359 ofproto_rule_reduce_timeouts(struct rule *rule,
4360 uint16_t idle_timeout, uint16_t hard_timeout)
4361 OVS_EXCLUDED(ofproto_mutex, rule->mutex)
4363 if (!idle_timeout && !hard_timeout) {
4367 ovs_mutex_lock(&ofproto_mutex);
4368 if (list_is_empty(&rule->expirable)) {
4369 list_insert(&rule->ofproto->expirable, &rule->expirable);
4371 ovs_mutex_unlock(&ofproto_mutex);
4373 ovs_mutex_lock(&rule->mutex);
4374 reduce_timeout(idle_timeout, &rule->idle_timeout);
4375 reduce_timeout(hard_timeout, &rule->hard_timeout);
4376 ovs_mutex_unlock(&rule->mutex);
4380 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4381 OVS_EXCLUDED(ofproto_mutex)
4383 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4384 struct ofputil_flow_mod fm;
4385 uint64_t ofpacts_stub[1024 / 8];
4386 struct ofpbuf ofpacts;
4390 error = reject_slave_controller(ofconn);
4395 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
4396 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
4398 u16_to_ofp(ofproto->max_ports),
4401 error = ofproto_check_ofpacts(ofproto, fm.ofpacts, fm.ofpacts_len);
4404 error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
4407 goto exit_free_ofpacts;
4410 /* Record the operation for logging a summary report. */
4411 switch (fm.command) {
4417 case OFPFC_MODIFY_STRICT:
4418 ofproto->n_modify++;
4422 case OFPFC_DELETE_STRICT:
4423 ofproto->n_delete++;
4428 if (ofproto->next_op_report == LLONG_MAX) {
4429 ofproto->first_op = now;
4430 ofproto->next_op_report = MAX(now + 10 * 1000,
4431 ofproto->op_backoff);
4432 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
4434 ofproto->last_op = now;
4437 ofpbuf_uninit(&ofpacts);
4443 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
4444 struct ofputil_flow_mod *fm, const struct ofp_header *oh)
4445 OVS_EXCLUDED(ofproto_mutex)
4449 ovs_mutex_lock(&ofproto_mutex);
4450 if (ofproto->n_pending < 50) {
4451 switch (fm->command) {
4453 error = add_flow(ofproto, ofconn, fm, oh);
4457 error = modify_flows_loose(ofproto, ofconn, fm, oh);
4460 case OFPFC_MODIFY_STRICT:
4461 error = modify_flow_strict(ofproto, ofconn, fm, oh);
4465 error = delete_flows_loose(ofproto, ofconn, fm, oh);
4468 case OFPFC_DELETE_STRICT:
4469 error = delete_flow_strict(ofproto, ofconn, fm, oh);
4473 if (fm->command > 0xff) {
4474 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
4475 "flow_mod_table_id extension is not enabled",
4478 error = OFPERR_OFPFMFC_BAD_COMMAND;
4482 ovs_assert(!list_is_empty(&ofproto->pending));
4483 error = OFPROTO_POSTPONE;
4485 ovs_mutex_unlock(&ofproto_mutex);
4487 run_rule_executes(ofproto);
4492 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4494 struct ofputil_role_request request;
4495 struct ofputil_role_request reply;
4499 error = ofputil_decode_role_message(oh, &request);
4504 if (request.role != OFPCR12_ROLE_NOCHANGE) {
4505 if (ofconn_get_role(ofconn) != request.role
4506 && ofconn_has_pending_opgroups(ofconn)) {
4507 return OFPROTO_POSTPONE;
4510 if (request.have_generation_id
4511 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
4512 return OFPERR_OFPRRFC_STALE;
4515 ofconn_set_role(ofconn, request.role);
4518 reply.role = ofconn_get_role(ofconn);
4519 reply.have_generation_id = ofconn_get_master_election_id(
4520 ofconn, &reply.generation_id);
4521 buf = ofputil_encode_role_reply(oh, &reply);
4522 ofconn_send_reply(ofconn, buf);
4528 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
4529 const struct ofp_header *oh)
4531 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
4532 enum ofputil_protocol cur, next;
4534 cur = ofconn_get_protocol(ofconn);
4535 next = ofputil_protocol_set_tid(cur, msg->set != 0);
4536 ofconn_set_protocol(ofconn, next);
4542 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4544 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
4545 enum ofputil_protocol cur, next;
4546 enum ofputil_protocol next_base;
4548 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
4550 return OFPERR_OFPBRC_EPERM;
4553 cur = ofconn_get_protocol(ofconn);
4554 next = ofputil_protocol_set_base(cur, next_base);
4555 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
4556 /* Avoid sending async messages in surprising protocol. */
4557 return OFPROTO_POSTPONE;
4560 ofconn_set_protocol(ofconn, next);
4565 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
4566 const struct ofp_header *oh)
4568 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
4571 format = ntohl(msg->format);
4572 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
4573 return OFPERR_OFPBRC_EPERM;
4576 if (format != ofconn_get_packet_in_format(ofconn)
4577 && ofconn_has_pending_opgroups(ofconn)) {
4578 /* Avoid sending async message in surprsing packet in format. */
4579 return OFPROTO_POSTPONE;
4582 ofconn_set_packet_in_format(ofconn, format);
4587 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
4589 const struct nx_async_config *msg = ofpmsg_body(oh);
4590 uint32_t master[OAM_N_TYPES];
4591 uint32_t slave[OAM_N_TYPES];
4593 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
4594 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
4595 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
4597 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
4598 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
4599 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
4601 ofconn_set_async_config(ofconn, master, slave);
4602 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
4603 !ofconn_get_miss_send_len(ofconn)) {
4604 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
4611 handle_nxt_get_async_request(struct ofconn *ofconn, const struct ofp_header *oh)
4614 uint32_t master[OAM_N_TYPES];
4615 uint32_t slave[OAM_N_TYPES];
4616 struct nx_async_config *msg;
4618 ofconn_get_async_config(ofconn, master, slave);
4619 buf = ofpraw_alloc_reply(OFPRAW_OFPT13_GET_ASYNC_REPLY, oh, 0);
4620 msg = ofpbuf_put_zeros(buf, sizeof *msg);
4622 msg->packet_in_mask[0] = htonl(master[OAM_PACKET_IN]);
4623 msg->port_status_mask[0] = htonl(master[OAM_PORT_STATUS]);
4624 msg->flow_removed_mask[0] = htonl(master[OAM_FLOW_REMOVED]);
4626 msg->packet_in_mask[1] = htonl(slave[OAM_PACKET_IN]);
4627 msg->port_status_mask[1] = htonl(slave[OAM_PORT_STATUS]);
4628 msg->flow_removed_mask[1] = htonl(slave[OAM_FLOW_REMOVED]);
4630 ofconn_send_reply(ofconn, buf);
4636 handle_nxt_set_controller_id(struct ofconn *ofconn,
4637 const struct ofp_header *oh)
4639 const struct nx_controller_id *nci = ofpmsg_body(oh);
4641 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
4642 return OFPERR_NXBRC_MUST_BE_ZERO;
4645 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
4650 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4654 if (ofconn_has_pending_opgroups(ofconn)) {
4655 return OFPROTO_POSTPONE;
4658 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
4659 ? OFPRAW_OFPT10_BARRIER_REPLY
4660 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
4661 ofconn_send_reply(ofconn, buf);
4666 ofproto_compose_flow_refresh_update(const struct rule *rule,
4667 enum nx_flow_monitor_flags flags,
4669 OVS_REQUIRES(ofproto_mutex)
4671 struct ofoperation *op = rule->pending;
4672 const struct rule_actions *actions;
4673 struct ofputil_flow_update fu;
4676 if (op && op->type == OFOPERATION_ADD) {
4677 /* We'll report the final flow when the operation completes. Reporting
4678 * it now would cause a duplicate report later. */
4682 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
4683 ? NXFME_ADDED : NXFME_MODIFIED);
4685 ovs_mutex_lock(&rule->mutex);
4686 fu.idle_timeout = rule->idle_timeout;
4687 fu.hard_timeout = rule->hard_timeout;
4688 ovs_mutex_unlock(&rule->mutex);
4689 fu.table_id = rule->table_id;
4690 fu.cookie = rule->flow_cookie;
4691 minimatch_expand(&rule->cr.match, &match);
4693 fu.priority = rule->cr.priority;
4695 if (!(flags & NXFMF_ACTIONS)) {
4698 actions = rule->actions;
4700 /* An operation is in progress. Use the previous version of the flow's
4701 * actions, so that when the operation commits we report the change. */
4703 case OFOPERATION_ADD:
4706 case OFOPERATION_MODIFY:
4707 case OFOPERATION_REPLACE:
4708 actions = op->actions ? op->actions : rule->actions;
4711 case OFOPERATION_DELETE:
4712 actions = rule->actions;
4719 fu.ofpacts = actions ? actions->ofpacts : NULL;
4720 fu.ofpacts_len = actions ? actions->ofpacts_len : 0;
4722 if (list_is_empty(msgs)) {
4723 ofputil_start_flow_update(msgs);
4725 ofputil_append_flow_update(&fu, msgs);
4729 ofmonitor_compose_refresh_updates(struct rule_collection *rules,
4731 OVS_REQUIRES(ofproto_mutex)
4735 for (i = 0; i < rules->n; i++) {
4736 struct rule *rule = rules->rules[i];
4737 enum nx_flow_monitor_flags flags = rule->monitor_flags;
4738 rule->monitor_flags = 0;
4740 ofproto_compose_flow_refresh_update(rule, flags, msgs);
4745 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
4746 struct rule *rule, uint64_t seqno,
4747 struct rule_collection *rules)
4748 OVS_REQUIRES(ofproto_mutex)
4750 enum nx_flow_monitor_flags update;
4752 if (ofproto_rule_is_hidden(rule)) {
4757 ? ofoperation_has_out_port(rule->pending, m->out_port)
4758 : ofproto_rule_has_out_port(rule, m->out_port))) {
4763 if (rule->add_seqno > seqno) {
4764 update = NXFMF_ADD | NXFMF_MODIFY;
4765 } else if (rule->modify_seqno > seqno) {
4766 update = NXFMF_MODIFY;
4771 if (!(m->flags & update)) {
4775 update = NXFMF_INITIAL;
4778 if (!rule->monitor_flags) {
4779 rule_collection_add(rules, rule);
4781 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
4785 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
4787 struct rule_collection *rules)
4788 OVS_REQUIRES(ofproto_mutex)
4790 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
4791 const struct ofoperation *op;
4792 const struct oftable *table;
4793 struct cls_rule target;
4795 cls_rule_init_from_minimatch(&target, &m->match, 0);
4796 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
4797 struct cls_cursor cursor;
4800 ovs_rwlock_rdlock(&table->cls.rwlock);
4801 cls_cursor_init(&cursor, &table->cls, &target);
4802 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4803 ovs_assert(!rule->pending); /* XXX */
4804 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4806 ovs_rwlock_unlock(&table->cls.rwlock);
4809 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
4810 struct rule *rule = op->rule;
4812 if (((m->table_id == 0xff
4813 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
4814 : m->table_id == rule->table_id))
4815 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
4816 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4819 cls_rule_destroy(&target);
4823 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
4824 struct rule_collection *rules)
4825 OVS_REQUIRES(ofproto_mutex)
4827 if (m->flags & NXFMF_INITIAL) {
4828 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
4833 ofmonitor_collect_resume_rules(struct ofmonitor *m,
4834 uint64_t seqno, struct rule_collection *rules)
4835 OVS_REQUIRES(ofproto_mutex)
4837 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
4841 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
4842 OVS_EXCLUDED(ofproto_mutex)
4844 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4845 struct ofmonitor **monitors;
4846 size_t n_monitors, allocated_monitors;
4847 struct rule_collection rules;
4848 struct list replies;
4854 ofpbuf_use_const(&b, oh, ntohs(oh->length));
4856 n_monitors = allocated_monitors = 0;
4858 ovs_mutex_lock(&ofproto_mutex);
4860 struct ofputil_flow_monitor_request request;
4861 struct ofmonitor *m;
4864 retval = ofputil_decode_flow_monitor_request(&request, &b);
4865 if (retval == EOF) {
4867 } else if (retval) {
4872 if (request.table_id != 0xff
4873 && request.table_id >= ofproto->n_tables) {
4874 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4878 error = ofmonitor_create(&request, ofconn, &m);
4883 if (n_monitors >= allocated_monitors) {
4884 monitors = x2nrealloc(monitors, &allocated_monitors,
4887 monitors[n_monitors++] = m;
4890 rule_collection_init(&rules);
4891 for (i = 0; i < n_monitors; i++) {
4892 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
4895 ofpmp_init(&replies, oh);
4896 ofmonitor_compose_refresh_updates(&rules, &replies);
4897 ovs_mutex_unlock(&ofproto_mutex);
4899 rule_collection_destroy(&rules);
4901 ofconn_send_replies(ofconn, &replies);
4907 for (i = 0; i < n_monitors; i++) {
4908 ofmonitor_destroy(monitors[i]);
4911 ovs_mutex_unlock(&ofproto_mutex);
4917 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
4918 OVS_EXCLUDED(ofproto_mutex)
4920 struct ofmonitor *m;
4924 id = ofputil_decode_flow_monitor_cancel(oh);
4926 ovs_mutex_lock(&ofproto_mutex);
4927 m = ofmonitor_lookup(ofconn, id);
4929 ofmonitor_destroy(m);
4932 error = OFPERR_NXBRC_FM_BAD_ID;
4934 ovs_mutex_unlock(&ofproto_mutex);
4939 /* Meters implementation.
4941 * Meter table entry, indexed by the OpenFlow meter_id.
4942 * These are always dynamically allocated to allocate enough space for
4944 * 'created' is used to compute the duration for meter stats.
4945 * 'list rules' is needed so that we can delete the dependent rules when the
4946 * meter table entry is deleted.
4947 * 'provider_meter_id' is for the provider's private use.
4950 long long int created; /* Time created. */
4951 struct list rules; /* List of "struct rule_dpif"s. */
4952 ofproto_meter_id provider_meter_id;
4953 uint16_t flags; /* Meter flags. */
4954 uint16_t n_bands; /* Number of meter bands. */
4955 struct ofputil_meter_band *bands;
4959 * This is used in instruction validation at flow set-up time,
4960 * as flows may not use non-existing meters.
4961 * Return value of UINT32_MAX signifies an invalid meter.
4964 get_provider_meter_id(const struct ofproto *ofproto, uint32_t of_meter_id)
4966 if (of_meter_id && of_meter_id <= ofproto->meter_features.max_meters) {
4967 const struct meter *meter = ofproto->meters[of_meter_id];
4969 return meter->provider_meter_id.uint32;
4976 meter_update(struct meter *meter, const struct ofputil_meter_config *config)
4980 meter->flags = config->flags;
4981 meter->n_bands = config->n_bands;
4982 meter->bands = xmemdup(config->bands,
4983 config->n_bands * sizeof *meter->bands);
4986 static struct meter *
4987 meter_create(const struct ofputil_meter_config *config,
4988 ofproto_meter_id provider_meter_id)
4990 struct meter *meter;
4992 meter = xzalloc(sizeof *meter);
4993 meter->provider_meter_id = provider_meter_id;
4994 meter->created = time_msec();
4995 list_init(&meter->rules);
4997 meter_update(meter, config);
5003 meter_delete(struct ofproto *ofproto, uint32_t first, uint32_t last)
5004 OVS_REQUIRES(ofproto_mutex)
5007 for (mid = first; mid <= last; ++mid) {
5008 struct meter *meter = ofproto->meters[mid];
5010 ofproto->meters[mid] = NULL;
5011 ofproto->ofproto_class->meter_del(ofproto,
5012 meter->provider_meter_id);
5020 handle_add_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
5022 ofproto_meter_id provider_meter_id = { UINT32_MAX };
5023 struct meter **meterp = &ofproto->meters[mm->meter.meter_id];
5027 return OFPERR_OFPMMFC_METER_EXISTS;
5030 error = ofproto->ofproto_class->meter_set(ofproto, &provider_meter_id,
5033 ovs_assert(provider_meter_id.uint32 != UINT32_MAX);
5034 *meterp = meter_create(&mm->meter, provider_meter_id);
5040 handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
5042 struct meter *meter = ofproto->meters[mm->meter.meter_id];
5044 uint32_t provider_meter_id;
5047 return OFPERR_OFPMMFC_UNKNOWN_METER;
5050 provider_meter_id = meter->provider_meter_id.uint32;
5051 error = ofproto->ofproto_class->meter_set(ofproto,
5052 &meter->provider_meter_id,
5054 ovs_assert(meter->provider_meter_id.uint32 == provider_meter_id);
5056 meter_update(meter, &mm->meter);
5062 handle_delete_meter(struct ofconn *ofconn, const struct ofp_header *oh,
5063 struct ofputil_meter_mod *mm)
5064 OVS_EXCLUDED(ofproto_mutex)
5066 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5067 uint32_t meter_id = mm->meter.meter_id;
5068 struct rule_collection rules;
5069 enum ofperr error = 0;
5070 uint32_t first, last;
5072 if (meter_id == OFPM13_ALL) {
5074 last = ofproto->meter_features.max_meters;
5076 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
5079 first = last = meter_id;
5082 /* First delete the rules that use this meter. If any of those rules are
5083 * currently being modified, postpone the whole operation until later. */
5084 rule_collection_init(&rules);
5085 ovs_mutex_lock(&ofproto_mutex);
5086 for (meter_id = first; meter_id <= last; ++meter_id) {
5087 struct meter *meter = ofproto->meters[meter_id];
5088 if (meter && !list_is_empty(&meter->rules)) {
5091 LIST_FOR_EACH (rule, meter_list_node, &meter->rules) {
5092 if (rule->pending) {
5093 error = OFPROTO_POSTPONE;
5096 rule_collection_add(&rules, rule);
5101 delete_flows__(ofproto, ofconn, oh, &rules, OFPRR_METER_DELETE);
5104 /* Delete the meters. */
5105 meter_delete(ofproto, first, last);
5108 ovs_mutex_unlock(&ofproto_mutex);
5109 rule_collection_destroy(&rules);
5115 handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5117 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5118 struct ofputil_meter_mod mm;
5119 uint64_t bands_stub[256 / 8];
5120 struct ofpbuf bands;
5124 error = reject_slave_controller(ofconn);
5129 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5131 error = ofputil_decode_meter_mod(oh, &mm, &bands);
5133 goto exit_free_bands;
5136 meter_id = mm.meter.meter_id;
5138 if (mm.command != OFPMC13_DELETE) {
5139 /* Fails also when meters are not implemented by the provider. */
5140 if (meter_id == 0 || meter_id > OFPM13_MAX) {
5141 error = OFPERR_OFPMMFC_INVALID_METER;
5142 goto exit_free_bands;
5143 } else if (meter_id > ofproto->meter_features.max_meters) {
5144 error = OFPERR_OFPMMFC_OUT_OF_METERS;
5145 goto exit_free_bands;
5147 if (mm.meter.n_bands > ofproto->meter_features.max_bands) {
5148 error = OFPERR_OFPMMFC_OUT_OF_BANDS;
5149 goto exit_free_bands;
5153 switch (mm.command) {
5155 error = handle_add_meter(ofproto, &mm);
5158 case OFPMC13_MODIFY:
5159 error = handle_modify_meter(ofproto, &mm);
5162 case OFPMC13_DELETE:
5163 error = handle_delete_meter(ofconn, oh, &mm);
5167 error = OFPERR_OFPMMFC_BAD_COMMAND;
5172 ofpbuf_uninit(&bands);
5177 handle_meter_features_request(struct ofconn *ofconn,
5178 const struct ofp_header *request)
5180 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5181 struct ofputil_meter_features features;
5184 if (ofproto->ofproto_class->meter_get_features) {
5185 ofproto->ofproto_class->meter_get_features(ofproto, &features);
5187 memset(&features, 0, sizeof features);
5189 b = ofputil_encode_meter_features_reply(&features, request);
5191 ofconn_send_reply(ofconn, b);
5196 handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request,
5199 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5200 struct list replies;
5201 uint64_t bands_stub[256 / 8];
5202 struct ofpbuf bands;
5203 uint32_t meter_id, first, last;
5205 ofputil_decode_meter_request(request, &meter_id);
5207 if (meter_id == OFPM13_ALL) {
5209 last = ofproto->meter_features.max_meters;
5211 if (!meter_id || meter_id > ofproto->meter_features.max_meters ||
5212 !ofproto->meters[meter_id]) {
5213 return OFPERR_OFPMMFC_UNKNOWN_METER;
5215 first = last = meter_id;
5218 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5219 ofpmp_init(&replies, request);
5221 for (meter_id = first; meter_id <= last; ++meter_id) {
5222 struct meter *meter = ofproto->meters[meter_id];
5224 continue; /* Skip non-existing meters. */
5226 if (type == OFPTYPE_METER_STATS_REQUEST) {
5227 struct ofputil_meter_stats stats;
5229 stats.meter_id = meter_id;
5231 /* Provider sets the packet and byte counts, we do the rest. */
5232 stats.flow_count = list_size(&meter->rules);
5233 calc_duration(meter->created, time_msec(),
5234 &stats.duration_sec, &stats.duration_nsec);
5235 stats.n_bands = meter->n_bands;
5236 ofpbuf_clear(&bands);
5238 = ofpbuf_put_uninit(&bands,
5239 meter->n_bands * sizeof *stats.bands);
5241 if (!ofproto->ofproto_class->meter_get(ofproto,
5242 meter->provider_meter_id,
5244 ofputil_append_meter_stats(&replies, &stats);
5246 } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */
5247 struct ofputil_meter_config config;
5249 config.meter_id = meter_id;
5250 config.flags = meter->flags;
5251 config.n_bands = meter->n_bands;
5252 config.bands = meter->bands;
5253 ofputil_append_meter_config(&replies, &config);
5257 ofconn_send_replies(ofconn, &replies);
5258 ofpbuf_uninit(&bands);
5263 ofproto_group_lookup(const struct ofproto *ofproto, uint32_t group_id,
5264 struct ofgroup **group)
5265 OVS_TRY_RDLOCK(true, (*group)->rwlock)
5267 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5268 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
5269 hash_int(group_id, 0), &ofproto->groups) {
5270 if ((*group)->group_id == group_id) {
5271 ovs_rwlock_rdlock(&(*group)->rwlock);
5272 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5276 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5281 ofproto_group_release(struct ofgroup *group)
5282 OVS_RELEASES(group->rwlock)
5284 ovs_rwlock_unlock(&group->rwlock);
5288 ofproto_group_write_lookup(const struct ofproto *ofproto, uint32_t group_id,
5289 struct ofgroup **group)
5290 OVS_TRY_WRLOCK(true, ofproto->groups_rwlock)
5291 OVS_TRY_WRLOCK(true, (*group)->rwlock)
5293 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5294 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
5295 hash_int(group_id, 0), &ofproto->groups) {
5296 if ((*group)->group_id == group_id) {
5297 ovs_rwlock_wrlock(&(*group)->rwlock);
5301 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5306 ofproto_group_exists__(const struct ofproto *ofproto, uint32_t group_id)
5307 OVS_REQ_RDLOCK(ofproto->groups_rwlock)
5309 struct ofgroup *grp;
5311 HMAP_FOR_EACH_IN_BUCKET (grp, hmap_node,
5312 hash_int(group_id, 0), &ofproto->groups) {
5313 if (grp->group_id == group_id) {
5321 ofproto_group_exists(const struct ofproto *ofproto, uint32_t group_id)
5322 OVS_EXCLUDED(ofproto->groups_rwlock)
5326 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5327 exists = ofproto_group_exists__(ofproto, group_id);
5328 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5334 group_get_ref_count(struct ofgroup *group)
5335 OVS_EXCLUDED(ofproto_mutex)
5337 struct ofproto *ofproto = group->ofproto;
5338 struct rule_criteria criteria;
5339 struct rule_collection rules;
5344 match_init_catchall(&match);
5345 rule_criteria_init(&criteria, 0xff, &match, 0, htonll(0), htonll(0),
5346 OFPP_ANY, group->group_id);
5347 ovs_mutex_lock(&ofproto_mutex);
5348 error = collect_rules_loose(ofproto, &criteria, &rules);
5349 ovs_mutex_unlock(&ofproto_mutex);
5350 rule_criteria_destroy(&criteria);
5352 count = !error && rules.n < UINT32_MAX ? rules.n : UINT32_MAX;
5354 rule_collection_destroy(&rules);
5359 append_group_stats(struct ofgroup *group, struct list *replies)
5360 OVS_REQ_RDLOCK(group->rwlock)
5362 struct ofputil_group_stats ogs;
5363 struct ofproto *ofproto = group->ofproto;
5364 long long int now = time_msec();
5367 ogs.bucket_stats = xmalloc(group->n_buckets * sizeof *ogs.bucket_stats);
5369 /* Provider sets the packet and byte counts, we do the rest. */
5370 ogs.ref_count = group_get_ref_count(group);
5371 ogs.n_buckets = group->n_buckets;
5373 error = (ofproto->ofproto_class->group_get_stats
5374 ? ofproto->ofproto_class->group_get_stats(group, &ogs)
5377 ogs.packet_count = UINT64_MAX;
5378 ogs.byte_count = UINT64_MAX;
5379 memset(ogs.bucket_stats, 0xff,
5380 ogs.n_buckets * sizeof *ogs.bucket_stats);
5383 ogs.group_id = group->group_id;
5384 calc_duration(group->created, now, &ogs.duration_sec, &ogs.duration_nsec);
5386 ofputil_append_group_stats(replies, &ogs);
5388 free(ogs.bucket_stats);
5392 handle_group_stats_request(struct ofconn *ofconn,
5393 const struct ofp_header *request)
5395 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5396 struct list replies;
5398 struct ofgroup *group;
5401 error = ofputil_decode_group_stats_request(request, &group_id);
5406 ofpmp_init(&replies, request);
5408 if (group_id == OFPG_ALL) {
5409 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5410 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
5411 ovs_rwlock_rdlock(&group->rwlock);
5412 append_group_stats(group, &replies);
5413 ovs_rwlock_unlock(&group->rwlock);
5415 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5417 if (ofproto_group_lookup(ofproto, group_id, &group)) {
5418 append_group_stats(group, &replies);
5419 ofproto_group_release(group);
5423 ofconn_send_replies(ofconn, &replies);
5429 handle_group_desc_stats_request(struct ofconn *ofconn,
5430 const struct ofp_header *request)
5432 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5433 struct list replies;
5434 struct ofputil_group_desc gds;
5435 struct ofgroup *group;
5437 ofpmp_init(&replies, request);
5439 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5440 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
5441 gds.group_id = group->group_id;
5442 gds.type = group->type;
5443 ofputil_append_group_desc_reply(&gds, &group->buckets, &replies);
5445 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5447 ofconn_send_replies(ofconn, &replies);
5453 handle_group_features_stats_request(struct ofconn *ofconn,
5454 const struct ofp_header *request)
5456 struct ofproto *p = ofconn_get_ofproto(ofconn);
5459 msg = ofputil_encode_group_features_reply(&p->ogf, request);
5461 ofconn_send_reply(ofconn, msg);
5468 handle_queue_get_config_request(struct ofconn *ofconn,
5469 const struct ofp_header *oh)
5471 struct ofproto *p = ofconn_get_ofproto(ofconn);
5472 struct netdev_queue_dump queue_dump;
5473 struct ofport *ofport;
5474 unsigned int queue_id;
5475 struct ofpbuf *reply;
5476 struct smap details;
5480 error = ofputil_decode_queue_get_config_request(oh, &request);
5485 ofport = ofproto_get_port(p, request);
5487 return OFPERR_OFPQOFC_BAD_PORT;
5490 reply = ofputil_encode_queue_get_config_reply(oh);
5492 smap_init(&details);
5493 NETDEV_QUEUE_FOR_EACH (&queue_id, &details, &queue_dump, ofport->netdev) {
5494 struct ofputil_queue_config queue;
5496 /* None of the existing queues have compatible properties, so we
5497 * hard-code omitting min_rate and max_rate. */
5498 queue.queue_id = queue_id;
5499 queue.min_rate = UINT16_MAX;
5500 queue.max_rate = UINT16_MAX;
5501 ofputil_append_queue_get_config_reply(reply, &queue);
5503 smap_destroy(&details);
5505 ofconn_send_reply(ofconn, reply);
5510 /* Implements OFPGC11_ADD
5511 * in which no matching flow already exists in the flow table.
5513 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
5514 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
5515 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
5516 * initiated now but may be retried later.
5518 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
5519 * ownership remains with the caller.
5521 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
5524 add_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
5526 struct ofgroup *ofgroup;
5529 if (gm->group_id > OFPG_MAX) {
5530 return OFPERR_OFPGMFC_INVALID_GROUP;
5532 if (gm->type > OFPGT11_FF) {
5533 return OFPERR_OFPGMFC_BAD_TYPE;
5536 /* Allocate new group and initialize it. */
5537 ofgroup = ofproto->ofproto_class->group_alloc();
5539 VLOG_WARN_RL(&rl, "%s: failed to create group", ofproto->name);
5540 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
5543 ovs_rwlock_init(&ofgroup->rwlock);
5544 ofgroup->ofproto = ofproto;
5545 ofgroup->group_id = gm->group_id;
5546 ofgroup->type = gm->type;
5547 ofgroup->created = ofgroup->modified = time_msec();
5549 list_move(&ofgroup->buckets, &gm->buckets);
5550 ofgroup->n_buckets = list_size(&ofgroup->buckets);
5552 /* Construct called BEFORE any locks are held. */
5553 error = ofproto->ofproto_class->group_construct(ofgroup);
5558 /* We wrlock as late as possible to minimize the time we jam any other
5559 * threads: No visible state changes before acquiring the lock. */
5560 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5562 if (ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
5563 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
5567 if (ofproto_group_exists__(ofproto, gm->group_id)) {
5568 error = OFPERR_OFPGMFC_GROUP_EXISTS;
5573 /* Insert new group. */
5574 hmap_insert(&ofproto->groups, &ofgroup->hmap_node,
5575 hash_int(ofgroup->group_id, 0));
5576 ofproto->n_groups[ofgroup->type]++;
5578 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5583 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5584 ofproto->ofproto_class->group_destruct(ofgroup);
5586 ofputil_bucket_list_destroy(&ofgroup->buckets);
5587 ofproto->ofproto_class->group_dealloc(ofgroup);
5592 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
5595 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
5598 modify_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
5600 struct ofgroup *ofgroup;
5601 struct ofgroup *victim;
5604 if (gm->group_id > OFPG_MAX) {
5605 return OFPERR_OFPGMFC_INVALID_GROUP;
5608 if (gm->type > OFPGT11_FF) {
5609 return OFPERR_OFPGMFC_BAD_TYPE;
5612 victim = ofproto->ofproto_class->group_alloc();
5614 VLOG_WARN_RL(&rl, "%s: failed to allocate group", ofproto->name);
5615 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
5618 if (!ofproto_group_write_lookup(ofproto, gm->group_id, &ofgroup)) {
5619 error = OFPERR_OFPGMFC_UNKNOWN_GROUP;
5622 /* Both group's and its container's write locks held now.
5623 * Also, n_groups[] is protected by ofproto->groups_rwlock. */
5624 if (ofgroup->type != gm->type
5625 && ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
5626 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
5631 list_move(&victim->buckets, &ofgroup->buckets);
5633 ofgroup->type = gm->type;
5634 list_move(&ofgroup->buckets, &gm->buckets);
5635 ofgroup->n_buckets = list_size(&ofgroup->buckets);
5637 error = ofproto->ofproto_class->group_modify(ofgroup, victim);
5639 ofputil_bucket_list_destroy(&victim->buckets);
5640 ofproto->n_groups[victim->type]--;
5641 ofproto->n_groups[ofgroup->type]++;
5642 ofgroup->modified = time_msec();
5644 ofputil_bucket_list_destroy(&ofgroup->buckets);
5647 list_move(&ofgroup->buckets, &victim->buckets);
5651 ovs_rwlock_unlock(&ofgroup->rwlock);
5652 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5654 ofproto->ofproto_class->group_dealloc(victim);
5659 delete_group__(struct ofproto *ofproto, struct ofgroup *ofgroup)
5660 OVS_RELEASES(ofproto->groups_rwlock)
5663 struct ofputil_flow_mod fm;
5665 /* Delete all flow entries containing this group in a group action */
5666 match_init_catchall(&match);
5667 flow_mod_init(&fm, &match, 0, NULL, 0, OFPFC_DELETE);
5668 fm.out_group = ofgroup->group_id;
5669 handle_flow_mod__(ofproto, NULL, &fm, NULL);
5671 /* Must wait until existing readers are done,
5672 * while holding the container's write lock at the same time. */
5673 ovs_rwlock_wrlock(&ofgroup->rwlock);
5674 hmap_remove(&ofproto->groups, &ofgroup->hmap_node);
5675 /* No-one can find this group any more. */
5676 ofproto->n_groups[ofgroup->type]--;
5677 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5679 ofproto->ofproto_class->group_destruct(ofgroup);
5680 ofputil_bucket_list_destroy(&ofgroup->buckets);
5681 ovs_rwlock_unlock(&ofgroup->rwlock);
5682 ovs_rwlock_destroy(&ofgroup->rwlock);
5683 ofproto->ofproto_class->group_dealloc(ofgroup);
5686 /* Implements OFPGC_DELETE. */
5688 delete_group(struct ofproto *ofproto, uint32_t group_id)
5690 struct ofgroup *ofgroup;
5692 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5693 if (group_id == OFPG_ALL) {
5695 struct hmap_node *node = hmap_first(&ofproto->groups);
5699 ofgroup = CONTAINER_OF(node, struct ofgroup, hmap_node);
5700 delete_group__(ofproto, ofgroup);
5701 /* Lock for each node separately, so that we will not jam the
5702 * other threads for too long time. */
5703 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5706 HMAP_FOR_EACH_IN_BUCKET (ofgroup, hmap_node,
5707 hash_int(group_id, 0), &ofproto->groups) {
5708 if (ofgroup->group_id == group_id) {
5709 delete_group__(ofproto, ofgroup);
5714 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5718 handle_group_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5720 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5721 struct ofputil_group_mod gm;
5724 error = reject_slave_controller(ofconn);
5729 error = ofputil_decode_group_mod(oh, &gm);
5734 switch (gm.command) {
5736 return add_group(ofproto, &gm);
5738 case OFPGC11_MODIFY:
5739 return modify_group(ofproto, &gm);
5741 case OFPGC11_DELETE:
5742 delete_group(ofproto, gm.group_id);
5746 if (gm.command > OFPGC11_DELETE) {
5747 VLOG_WARN_RL(&rl, "%s: Invalid group_mod command type %d",
5748 ofproto->name, gm.command);
5750 return OFPERR_OFPGMFC_BAD_COMMAND;
5755 handle_table_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5757 struct ofputil_table_mod tm;
5760 error = reject_slave_controller(ofconn);
5765 error = ofputil_decode_table_mod(oh, &tm);
5770 /* XXX Actual table mod support is not implemented yet. */
5775 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
5776 OVS_EXCLUDED(ofproto_mutex)
5778 const struct ofp_header *oh = msg->data;
5782 error = ofptype_decode(&type, oh);
5786 if (oh->version >= OFP13_VERSION && ofpmsg_is_stat_request(oh)
5787 && ofpmp_more(oh)) {
5788 /* We have no buffer implementation for multipart requests.
5789 * Report overflow for requests which consists of multiple
5791 return OFPERR_OFPBRC_MULTIPART_BUFFER_OVERFLOW;
5795 /* OpenFlow requests. */
5796 case OFPTYPE_ECHO_REQUEST:
5797 return handle_echo_request(ofconn, oh);
5799 case OFPTYPE_FEATURES_REQUEST:
5800 return handle_features_request(ofconn, oh);
5802 case OFPTYPE_GET_CONFIG_REQUEST:
5803 return handle_get_config_request(ofconn, oh);
5805 case OFPTYPE_SET_CONFIG:
5806 return handle_set_config(ofconn, oh);
5808 case OFPTYPE_PACKET_OUT:
5809 return handle_packet_out(ofconn, oh);
5811 case OFPTYPE_PORT_MOD:
5812 return handle_port_mod(ofconn, oh);
5814 case OFPTYPE_FLOW_MOD:
5815 return handle_flow_mod(ofconn, oh);
5817 case OFPTYPE_GROUP_MOD:
5818 return handle_group_mod(ofconn, oh);
5820 case OFPTYPE_TABLE_MOD:
5821 return handle_table_mod(ofconn, oh);
5823 case OFPTYPE_METER_MOD:
5824 return handle_meter_mod(ofconn, oh);
5826 case OFPTYPE_BARRIER_REQUEST:
5827 return handle_barrier_request(ofconn, oh);
5829 case OFPTYPE_ROLE_REQUEST:
5830 return handle_role_request(ofconn, oh);
5832 /* OpenFlow replies. */
5833 case OFPTYPE_ECHO_REPLY:
5836 /* Nicira extension requests. */
5837 case OFPTYPE_FLOW_MOD_TABLE_ID:
5838 return handle_nxt_flow_mod_table_id(ofconn, oh);
5840 case OFPTYPE_SET_FLOW_FORMAT:
5841 return handle_nxt_set_flow_format(ofconn, oh);
5843 case OFPTYPE_SET_PACKET_IN_FORMAT:
5844 return handle_nxt_set_packet_in_format(ofconn, oh);
5846 case OFPTYPE_SET_CONTROLLER_ID:
5847 return handle_nxt_set_controller_id(ofconn, oh);
5849 case OFPTYPE_FLOW_AGE:
5850 /* Nothing to do. */
5853 case OFPTYPE_FLOW_MONITOR_CANCEL:
5854 return handle_flow_monitor_cancel(ofconn, oh);
5856 case OFPTYPE_SET_ASYNC_CONFIG:
5857 return handle_nxt_set_async_config(ofconn, oh);
5859 case OFPTYPE_GET_ASYNC_REQUEST:
5860 return handle_nxt_get_async_request(ofconn, oh);
5862 /* Statistics requests. */
5863 case OFPTYPE_DESC_STATS_REQUEST:
5864 return handle_desc_stats_request(ofconn, oh);
5866 case OFPTYPE_FLOW_STATS_REQUEST:
5867 return handle_flow_stats_request(ofconn, oh);
5869 case OFPTYPE_AGGREGATE_STATS_REQUEST:
5870 return handle_aggregate_stats_request(ofconn, oh);
5872 case OFPTYPE_TABLE_STATS_REQUEST:
5873 return handle_table_stats_request(ofconn, oh);
5875 case OFPTYPE_PORT_STATS_REQUEST:
5876 return handle_port_stats_request(ofconn, oh);
5878 case OFPTYPE_QUEUE_STATS_REQUEST:
5879 return handle_queue_stats_request(ofconn, oh);
5881 case OFPTYPE_PORT_DESC_STATS_REQUEST:
5882 return handle_port_desc_stats_request(ofconn, oh);
5884 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
5885 return handle_flow_monitor_request(ofconn, oh);
5887 case OFPTYPE_METER_STATS_REQUEST:
5888 case OFPTYPE_METER_CONFIG_STATS_REQUEST:
5889 return handle_meter_request(ofconn, oh, type);
5891 case OFPTYPE_METER_FEATURES_STATS_REQUEST:
5892 return handle_meter_features_request(ofconn, oh);
5894 case OFPTYPE_GROUP_STATS_REQUEST:
5895 return handle_group_stats_request(ofconn, oh);
5897 case OFPTYPE_GROUP_DESC_STATS_REQUEST:
5898 return handle_group_desc_stats_request(ofconn, oh);
5900 case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
5901 return handle_group_features_stats_request(ofconn, oh);
5903 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
5904 return handle_queue_get_config_request(ofconn, oh);
5908 case OFPTYPE_FEATURES_REPLY:
5909 case OFPTYPE_GET_CONFIG_REPLY:
5910 case OFPTYPE_PACKET_IN:
5911 case OFPTYPE_FLOW_REMOVED:
5912 case OFPTYPE_PORT_STATUS:
5913 case OFPTYPE_BARRIER_REPLY:
5914 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
5915 case OFPTYPE_DESC_STATS_REPLY:
5916 case OFPTYPE_FLOW_STATS_REPLY:
5917 case OFPTYPE_QUEUE_STATS_REPLY:
5918 case OFPTYPE_PORT_STATS_REPLY:
5919 case OFPTYPE_TABLE_STATS_REPLY:
5920 case OFPTYPE_AGGREGATE_STATS_REPLY:
5921 case OFPTYPE_PORT_DESC_STATS_REPLY:
5922 case OFPTYPE_ROLE_REPLY:
5923 case OFPTYPE_FLOW_MONITOR_PAUSED:
5924 case OFPTYPE_FLOW_MONITOR_RESUMED:
5925 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
5926 case OFPTYPE_GET_ASYNC_REPLY:
5927 case OFPTYPE_GROUP_STATS_REPLY:
5928 case OFPTYPE_GROUP_DESC_STATS_REPLY:
5929 case OFPTYPE_GROUP_FEATURES_STATS_REPLY:
5930 case OFPTYPE_METER_STATS_REPLY:
5931 case OFPTYPE_METER_CONFIG_STATS_REPLY:
5932 case OFPTYPE_METER_FEATURES_STATS_REPLY:
5933 case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
5934 case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
5935 case OFPTYPE_ROLE_STATUS:
5937 if (ofpmsg_is_stat_request(oh)) {
5938 return OFPERR_OFPBRC_BAD_STAT;
5940 return OFPERR_OFPBRC_BAD_TYPE;
5946 handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
5947 OVS_EXCLUDED(ofproto_mutex)
5949 int error = handle_openflow__(ofconn, ofp_msg);
5950 if (error && error != OFPROTO_POSTPONE) {
5951 ofconn_send_error(ofconn, ofp_msg->data, error);
5953 COVERAGE_INC(ofproto_recv_openflow);
5954 return error != OFPROTO_POSTPONE;
5957 /* Asynchronous operations. */
5959 /* Creates and returns a new ofopgroup that is not associated with any
5960 * OpenFlow connection.
5962 * The caller should add operations to the returned group with
5963 * ofoperation_create() and then submit it with ofopgroup_submit(). */
5964 static struct ofopgroup *
5965 ofopgroup_create_unattached(struct ofproto *ofproto)
5966 OVS_REQUIRES(ofproto_mutex)
5968 struct ofopgroup *group = xzalloc(sizeof *group);
5969 group->ofproto = ofproto;
5970 list_init(&group->ofproto_node);
5971 list_init(&group->ops);
5972 list_init(&group->ofconn_node);
5976 /* Creates and returns a new ofopgroup for 'ofproto'.
5978 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
5979 * connection. The 'request' and 'buffer_id' arguments are ignored.
5981 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
5982 * If the ofopgroup eventually fails, then the error reply will include
5983 * 'request'. If the ofopgroup eventually succeeds, then the packet with
5984 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
5986 * The caller should add operations to the returned group with
5987 * ofoperation_create() and then submit it with ofopgroup_submit(). */
5988 static struct ofopgroup *
5989 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
5990 const struct ofp_header *request, uint32_t buffer_id)
5991 OVS_REQUIRES(ofproto_mutex)
5993 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
5995 size_t request_len = ntohs(request->length);
5997 ovs_assert(ofconn_get_ofproto(ofconn) == ofproto);
5999 ofconn_add_opgroup(ofconn, &group->ofconn_node);
6000 group->ofconn = ofconn;
6001 group->request = xmemdup(request, MIN(request_len, 64));
6002 group->buffer_id = buffer_id;
6007 /* Submits 'group' for processing.
6009 * If 'group' contains no operations (e.g. none were ever added, or all of the
6010 * ones that were added completed synchronously), then it is destroyed
6011 * immediately. Otherwise it is added to the ofproto's list of pending
6014 ofopgroup_submit(struct ofopgroup *group)
6015 OVS_REQUIRES(ofproto_mutex)
6017 if (!group->n_running) {
6018 ofopgroup_complete(group);
6020 list_push_back(&group->ofproto->pending, &group->ofproto_node);
6021 group->ofproto->n_pending++;
6026 ofopgroup_complete(struct ofopgroup *group)
6027 OVS_REQUIRES(ofproto_mutex)
6029 struct ofproto *ofproto = group->ofproto;
6031 struct ofconn *abbrev_ofconn;
6032 ovs_be32 abbrev_xid;
6034 struct ofoperation *op, *next_op;
6037 ovs_assert(!group->n_running);
6040 LIST_FOR_EACH (op, group_node, &group->ops) {
6047 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
6048 LIST_FOR_EACH (op, group_node, &group->ops) {
6049 if (op->type != OFOPERATION_DELETE) {
6050 struct ofpbuf *packet;
6053 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
6056 struct rule_execute *re;
6060 ofproto_rule_ref(op->rule);
6062 re = xmalloc(sizeof *re);
6063 re->rule = op->rule;
6064 re->in_port = in_port;
6065 re->packet = packet;
6067 if (!guarded_list_push_back(&ofproto->rule_executes,
6068 &re->list_node, 1024)) {
6069 ofproto_rule_unref(op->rule);
6070 ofpbuf_delete(re->packet);
6079 if (!error && !list_is_empty(&group->ofconn_node)) {
6080 abbrev_ofconn = group->ofconn;
6081 abbrev_xid = group->request->xid;
6083 abbrev_ofconn = NULL;
6084 abbrev_xid = htonl(0);
6086 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
6087 struct rule *rule = op->rule;
6089 /* We generally want to report the change to active OpenFlow flow
6090 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
6092 - The operation failed.
6094 - The affected rule is not visible to controllers.
6096 - The operation's only effect was to update rule->modified. */
6098 || ofproto_rule_is_hidden(rule)
6099 || (op->type == OFOPERATION_MODIFY
6101 && rule->flow_cookie == op->flow_cookie))) {
6102 /* Check that we can just cast from ofoperation_type to
6103 * nx_flow_update_event. */
6104 enum nx_flow_update_event event_type;
6107 case OFOPERATION_ADD:
6108 case OFOPERATION_REPLACE:
6109 event_type = NXFME_ADDED;
6112 case OFOPERATION_DELETE:
6113 event_type = NXFME_DELETED;
6116 case OFOPERATION_MODIFY:
6117 event_type = NXFME_MODIFIED;
6124 ofmonitor_report(ofproto->connmgr, rule, event_type,
6125 op->reason, abbrev_ofconn, abbrev_xid);
6128 rule->pending = NULL;
6131 case OFOPERATION_ADD:
6135 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
6136 if (vid_mask == VLAN_VID_MASK) {
6137 if (ofproto->vlan_bitmap) {
6138 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
6139 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
6140 bitmap_set1(ofproto->vlan_bitmap, vid);
6141 ofproto->vlans_changed = true;
6144 ofproto->vlans_changed = true;
6148 oftable_remove_rule(rule);
6149 ofproto_rule_unref(rule);
6153 case OFOPERATION_DELETE:
6154 ovs_assert(!op->error);
6155 ofproto_rule_unref(rule);
6159 case OFOPERATION_MODIFY:
6160 case OFOPERATION_REPLACE:
6162 long long int now = time_msec();
6164 rule->modified = now;
6165 if (op->type == OFOPERATION_REPLACE) {
6166 rule->created = rule->used = now;
6169 ofproto_rule_change_cookie(ofproto, rule, op->flow_cookie);
6170 ovs_mutex_lock(&rule->mutex);
6171 rule->idle_timeout = op->idle_timeout;
6172 rule->hard_timeout = op->hard_timeout;
6173 ovs_mutex_unlock(&rule->mutex);
6175 struct rule_actions *old_actions;
6177 ovs_mutex_lock(&rule->mutex);
6178 old_actions = rule->actions;
6179 rule->actions = op->actions;
6180 ovs_mutex_unlock(&rule->mutex);
6183 rule_actions_unref(old_actions);
6185 rule->flags = op->flags;
6193 ofoperation_destroy(op);
6196 ofmonitor_flush(ofproto->connmgr);
6198 if (!list_is_empty(&group->ofproto_node)) {
6199 ovs_assert(ofproto->n_pending > 0);
6200 ofproto->n_pending--;
6201 list_remove(&group->ofproto_node);
6203 if (!list_is_empty(&group->ofconn_node)) {
6204 list_remove(&group->ofconn_node);
6206 ofconn_send_error(group->ofconn, group->request, error);
6208 connmgr_retry(ofproto->connmgr);
6210 free(group->request);
6214 /* Initiates a new operation on 'rule', of the specified 'type', within
6215 * 'group'. Prior to calling, 'rule' must not have any pending operation.
6217 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
6218 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
6220 * Returns the newly created ofoperation (which is also available as
6221 * rule->pending). */
6222 static struct ofoperation *
6223 ofoperation_create(struct ofopgroup *group, struct rule *rule,
6224 enum ofoperation_type type,
6225 enum ofp_flow_removed_reason reason)
6226 OVS_REQUIRES(ofproto_mutex)
6228 struct ofproto *ofproto = group->ofproto;
6229 struct ofoperation *op;
6231 ovs_assert(!rule->pending);
6233 op = rule->pending = xzalloc(sizeof *op);
6235 list_push_back(&group->ops, &op->group_node);
6238 op->reason = reason;
6239 op->flow_cookie = rule->flow_cookie;
6240 ovs_mutex_lock(&rule->mutex);
6241 op->idle_timeout = rule->idle_timeout;
6242 op->hard_timeout = rule->hard_timeout;
6243 ovs_mutex_unlock(&rule->mutex);
6244 op->flags = rule->flags;
6248 if (type == OFOPERATION_DELETE) {
6249 hmap_insert(&ofproto->deletions, &op->hmap_node,
6250 cls_rule_hash(&rule->cr, rule->table_id));
6257 ofoperation_destroy(struct ofoperation *op)
6258 OVS_REQUIRES(ofproto_mutex)
6260 struct ofopgroup *group = op->group;
6263 op->rule->pending = NULL;
6265 if (op->type == OFOPERATION_DELETE) {
6266 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
6268 list_remove(&op->group_node);
6269 rule_actions_unref(op->actions);
6273 /* Indicates that 'op' completed with status 'error', which is either 0 to
6274 * indicate success or an OpenFlow error code on failure.
6276 * If 'error' is 0, indicating success, the operation will be committed
6277 * permanently to the flow table.
6279 * If 'error' is nonzero, then generally the operation will be rolled back:
6281 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
6282 * restores the original rule. The caller must have uninitialized any
6283 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
6284 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
6285 * and 7 for the new rule, calling its ->rule_dealloc() function.
6287 * - If 'op' is a "modify flow" operation, ofproto restores the original
6290 * - 'op' must not be a "delete flow" operation. Removing a rule is not
6291 * allowed to fail. It must always succeed.
6293 * Please see the large comment in ofproto/ofproto-provider.h titled
6294 * "Asynchronous Operation Support" for more information. */
6296 ofoperation_complete(struct ofoperation *op, enum ofperr error)
6298 struct ofopgroup *group = op->group;
6300 ovs_assert(group->n_running > 0);
6301 ovs_assert(!error || op->type != OFOPERATION_DELETE);
6304 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
6305 /* This function can be called from ->rule_construct(), in which case
6306 * ofproto_mutex is held, or it can be called from ->run(), in which
6307 * case ofproto_mutex is not held. But only in the latter case can we
6308 * arrive here, so we can safely take ofproto_mutex now. */
6309 ovs_mutex_lock(&ofproto_mutex);
6310 ovs_assert(op->rule->pending == op);
6311 ofopgroup_complete(group);
6312 ovs_mutex_unlock(&ofproto_mutex);
6317 pick_datapath_id(const struct ofproto *ofproto)
6319 const struct ofport *port;
6321 port = ofproto_get_port(ofproto, OFPP_LOCAL);
6323 uint8_t ea[ETH_ADDR_LEN];
6326 error = netdev_get_etheraddr(port->netdev, ea);
6328 return eth_addr_to_uint64(ea);
6330 VLOG_WARN("%s: could not get MAC address for %s (%s)",
6331 ofproto->name, netdev_get_name(port->netdev),
6332 ovs_strerror(error));
6334 return ofproto->fallback_dpid;
6338 pick_fallback_dpid(void)
6340 uint8_t ea[ETH_ADDR_LEN];
6341 eth_addr_nicira_random(ea);
6342 return eth_addr_to_uint64(ea);
6345 /* Table overflow policy. */
6347 /* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep'
6348 * to NULL if the table is not configured to evict rules or if the table
6349 * contains no evictable rules. (Rules with a readlock on their evict rwlock,
6350 * or with no timeouts are not evictable.) */
6352 choose_rule_to_evict(struct oftable *table, struct rule **rulep)
6353 OVS_REQUIRES(ofproto_mutex)
6355 struct eviction_group *evg;
6358 if (!table->eviction_fields) {
6362 /* In the common case, the outer and inner loops here will each be entered
6365 * - The inner loop normally "return"s in its first iteration. If the
6366 * eviction group has any evictable rules, then it always returns in
6369 * - The outer loop only iterates more than once if the largest eviction
6370 * group has no evictable rules.
6372 * - The outer loop can exit only if table's 'max_flows' is all filled up
6373 * by unevictable rules. */
6374 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
6377 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
6386 /* Searches 'ofproto' for tables that have more flows than their configured
6387 * maximum and that have flow eviction enabled, and evicts as many flows as
6388 * necessary and currently feasible from them.
6390 * This triggers only when an OpenFlow table has N flows in it and then the
6391 * client configures a maximum number of flows less than N. */
6393 ofproto_evict(struct ofproto *ofproto)
6395 struct oftable *table;
6397 ovs_mutex_lock(&ofproto_mutex);
6398 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
6399 evict_rules_from_table(ofproto, table, 0);
6401 ovs_mutex_unlock(&ofproto_mutex);
6404 /* Eviction groups. */
6406 /* Returns the priority to use for an eviction_group that contains 'n_rules'
6407 * rules. The priority contains low-order random bits to ensure that eviction
6408 * groups with the same number of rules are prioritized randomly. */
6410 eviction_group_priority(size_t n_rules)
6412 uint16_t size = MIN(UINT16_MAX, n_rules);
6413 return (size << 16) | random_uint16();
6416 /* Updates 'evg', an eviction_group within 'table', following a change that
6417 * adds or removes rules in 'evg'. */
6419 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
6420 OVS_REQUIRES(ofproto_mutex)
6422 heap_change(&table->eviction_groups_by_size, &evg->size_node,
6423 eviction_group_priority(heap_count(&evg->rules)));
6426 /* Destroys 'evg', an eviction_group within 'table':
6428 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
6429 * rules themselves, just removes them from the eviction group.)
6431 * - Removes 'evg' from 'table'.
6435 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
6436 OVS_REQUIRES(ofproto_mutex)
6438 while (!heap_is_empty(&evg->rules)) {
6441 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
6442 rule->eviction_group = NULL;
6444 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
6445 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
6446 heap_destroy(&evg->rules);
6450 /* Removes 'rule' from its eviction group, if any. */
6452 eviction_group_remove_rule(struct rule *rule)
6453 OVS_REQUIRES(ofproto_mutex)
6455 if (rule->eviction_group) {
6456 struct oftable *table = &rule->ofproto->tables[rule->table_id];
6457 struct eviction_group *evg = rule->eviction_group;
6459 rule->eviction_group = NULL;
6460 heap_remove(&evg->rules, &rule->evg_node);
6461 if (heap_is_empty(&evg->rules)) {
6462 eviction_group_destroy(table, evg);
6464 eviction_group_resized(table, evg);
6469 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
6470 * returns the hash value. */
6472 eviction_group_hash_rule(struct rule *rule)
6473 OVS_REQUIRES(ofproto_mutex)
6475 struct oftable *table = &rule->ofproto->tables[rule->table_id];
6476 const struct mf_subfield *sf;
6480 hash = table->eviction_group_id_basis;
6481 miniflow_expand(&rule->cr.match.flow, &flow);
6482 for (sf = table->eviction_fields;
6483 sf < &table->eviction_fields[table->n_eviction_fields];
6486 if (mf_are_prereqs_ok(sf->field, &flow)) {
6487 union mf_value value;
6489 mf_get_value(sf->field, &flow, &value);
6491 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
6493 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
6494 unsigned int start = sf->ofs + sf->n_bits;
6495 bitwise_zero(&value, sf->field->n_bytes, start,
6496 sf->field->n_bytes * 8 - start);
6498 hash = hash_bytes(&value, sf->field->n_bytes, hash);
6500 hash = hash_int(hash, 0);
6507 /* Returns an eviction group within 'table' with the given 'id', creating one
6509 static struct eviction_group *
6510 eviction_group_find(struct oftable *table, uint32_t id)
6511 OVS_REQUIRES(ofproto_mutex)
6513 struct eviction_group *evg;
6515 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
6519 evg = xmalloc(sizeof *evg);
6520 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
6521 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
6522 eviction_group_priority(0));
6523 heap_init(&evg->rules);
6528 /* Returns an eviction priority for 'rule'. The return value should be
6529 * interpreted so that higher priorities make a rule more attractive candidates
6532 rule_eviction_priority(struct rule *rule)
6533 OVS_REQUIRES(ofproto_mutex)
6535 long long int hard_expiration;
6536 long long int idle_expiration;
6537 long long int expiration;
6538 uint32_t expiration_offset;
6540 /* Calculate time of expiration. */
6541 ovs_mutex_lock(&rule->mutex);
6542 hard_expiration = (rule->hard_timeout
6543 ? rule->modified + rule->hard_timeout * 1000
6545 idle_expiration = (rule->idle_timeout
6546 ? rule->used + rule->idle_timeout * 1000
6548 expiration = MIN(hard_expiration, idle_expiration);
6549 ovs_mutex_unlock(&rule->mutex);
6550 if (expiration == LLONG_MAX) {
6554 /* Calculate the time of expiration as a number of (approximate) seconds
6555 * after program startup.
6557 * This should work OK for program runs that last UINT32_MAX seconds or
6558 * less. Therefore, please restart OVS at least once every 136 years. */
6559 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
6561 /* Invert the expiration offset because we're using a max-heap. */
6562 return UINT32_MAX - expiration_offset;
6565 /* Adds 'rule' to an appropriate eviction group for its oftable's
6566 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
6567 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
6570 * The caller must ensure that 'rule' is not already in an eviction group. */
6572 eviction_group_add_rule(struct rule *rule)
6573 OVS_REQUIRES(ofproto_mutex)
6575 struct ofproto *ofproto = rule->ofproto;
6576 struct oftable *table = &ofproto->tables[rule->table_id];
6579 ovs_mutex_lock(&rule->mutex);
6580 has_timeout = rule->hard_timeout || rule->idle_timeout;
6581 ovs_mutex_unlock(&rule->mutex);
6583 if (table->eviction_fields && has_timeout) {
6584 struct eviction_group *evg;
6586 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
6588 rule->eviction_group = evg;
6589 heap_insert(&evg->rules, &rule->evg_node,
6590 rule_eviction_priority(rule));
6591 eviction_group_resized(table, evg);
6597 /* Initializes 'table'. */
6599 oftable_init(struct oftable *table)
6601 memset(table, 0, sizeof *table);
6602 classifier_init(&table->cls);
6603 table->max_flows = UINT_MAX;
6606 /* Destroys 'table', including its classifier and eviction groups.
6608 * The caller is responsible for freeing 'table' itself. */
6610 oftable_destroy(struct oftable *table)
6612 ovs_rwlock_rdlock(&table->cls.rwlock);
6613 ovs_assert(classifier_is_empty(&table->cls));
6614 ovs_rwlock_unlock(&table->cls.rwlock);
6615 oftable_disable_eviction(table);
6616 classifier_destroy(&table->cls);
6620 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
6621 * string, then 'table' will use its default name.
6623 * This only affects the name exposed for a table exposed through the OpenFlow
6624 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
6626 oftable_set_name(struct oftable *table, const char *name)
6628 if (name && name[0]) {
6629 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
6630 if (!table->name || strncmp(name, table->name, len)) {
6632 table->name = xmemdup0(name, len);
6640 /* oftables support a choice of two policies when adding a rule would cause the
6641 * number of flows in the table to exceed the configured maximum number: either
6642 * they can refuse to add the new flow or they can evict some existing flow.
6643 * This function configures the former policy on 'table'. */
6645 oftable_disable_eviction(struct oftable *table)
6646 OVS_REQUIRES(ofproto_mutex)
6648 if (table->eviction_fields) {
6649 struct eviction_group *evg, *next;
6651 HMAP_FOR_EACH_SAFE (evg, next, id_node,
6652 &table->eviction_groups_by_id) {
6653 eviction_group_destroy(table, evg);
6655 hmap_destroy(&table->eviction_groups_by_id);
6656 heap_destroy(&table->eviction_groups_by_size);
6658 free(table->eviction_fields);
6659 table->eviction_fields = NULL;
6660 table->n_eviction_fields = 0;
6664 /* oftables support a choice of two policies when adding a rule would cause the
6665 * number of flows in the table to exceed the configured maximum number: either
6666 * they can refuse to add the new flow or they can evict some existing flow.
6667 * This function configures the latter policy on 'table', with fairness based
6668 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
6669 * 'n_fields' as 0 disables fairness.) */
6671 oftable_enable_eviction(struct oftable *table,
6672 const struct mf_subfield *fields, size_t n_fields)
6673 OVS_REQUIRES(ofproto_mutex)
6675 struct cls_cursor cursor;
6678 if (table->eviction_fields
6679 && n_fields == table->n_eviction_fields
6681 || !memcmp(fields, table->eviction_fields,
6682 n_fields * sizeof *fields))) {
6687 oftable_disable_eviction(table);
6689 table->n_eviction_fields = n_fields;
6690 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
6692 table->eviction_group_id_basis = random_uint32();
6693 hmap_init(&table->eviction_groups_by_id);
6694 heap_init(&table->eviction_groups_by_size);
6696 ovs_rwlock_rdlock(&table->cls.rwlock);
6697 cls_cursor_init(&cursor, &table->cls, NULL);
6698 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
6699 eviction_group_add_rule(rule);
6701 ovs_rwlock_unlock(&table->cls.rwlock);
6704 /* Removes 'rule' from the oftable that contains it. */
6706 oftable_remove_rule__(struct ofproto *ofproto, struct rule *rule)
6707 OVS_REQUIRES(ofproto_mutex)
6709 struct classifier *cls = &ofproto->tables[rule->table_id].cls;
6711 ovs_rwlock_wrlock(&cls->rwlock);
6712 classifier_remove(cls, CONST_CAST(struct cls_rule *, &rule->cr));
6713 ovs_rwlock_unlock(&cls->rwlock);
6715 cookies_remove(ofproto, rule);
6717 eviction_group_remove_rule(rule);
6718 if (!list_is_empty(&rule->expirable)) {
6719 list_remove(&rule->expirable);
6721 if (!list_is_empty(&rule->meter_list_node)) {
6722 list_remove(&rule->meter_list_node);
6723 list_init(&rule->meter_list_node);
6728 oftable_remove_rule(struct rule *rule)
6729 OVS_REQUIRES(ofproto_mutex)
6731 oftable_remove_rule__(rule->ofproto, rule);
6734 /* Inserts 'rule' into its oftable, which must not already contain any rule for
6735 * the same cls_rule. */
6737 oftable_insert_rule(struct rule *rule)
6738 OVS_REQUIRES(ofproto_mutex)
6740 struct ofproto *ofproto = rule->ofproto;
6741 struct oftable *table = &ofproto->tables[rule->table_id];
6744 ovs_mutex_lock(&rule->mutex);
6745 may_expire = rule->hard_timeout || rule->idle_timeout;
6746 ovs_mutex_unlock(&rule->mutex);
6749 list_insert(&ofproto->expirable, &rule->expirable);
6752 cookies_insert(ofproto, rule);
6754 if (rule->actions->provider_meter_id != UINT32_MAX) {
6755 uint32_t meter_id = ofpacts_get_meter(rule->actions->ofpacts,
6756 rule->actions->ofpacts_len);
6757 struct meter *meter = ofproto->meters[meter_id];
6758 list_insert(&meter->rules, &rule->meter_list_node);
6760 ovs_rwlock_wrlock(&table->cls.rwlock);
6761 classifier_insert(&table->cls, CONST_CAST(struct cls_rule *, &rule->cr));
6762 ovs_rwlock_unlock(&table->cls.rwlock);
6763 eviction_group_add_rule(rule);
6766 /* unixctl commands. */
6769 ofproto_lookup(const char *name)
6771 struct ofproto *ofproto;
6773 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
6775 if (!strcmp(ofproto->name, name)) {
6783 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
6784 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6786 struct ofproto *ofproto;
6790 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
6791 ds_put_format(&results, "%s\n", ofproto->name);
6793 unixctl_command_reply(conn, ds_cstr(&results));
6794 ds_destroy(&results);
6798 ofproto_unixctl_init(void)
6800 static bool registered;
6806 unixctl_command_register("ofproto/list", "", 0, 0,
6807 ofproto_unixctl_list, NULL);
6810 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6812 * This is deprecated. It is only for compatibility with broken device drivers
6813 * in old versions of Linux that do not properly support VLANs when VLAN
6814 * devices are not used. When broken device drivers are no longer in
6815 * widespread use, we will delete these interfaces. */
6817 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
6818 * (exactly) by an OpenFlow rule in 'ofproto'. */
6820 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
6822 const struct oftable *oftable;
6824 free(ofproto->vlan_bitmap);
6825 ofproto->vlan_bitmap = bitmap_allocate(4096);
6826 ofproto->vlans_changed = false;
6828 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
6829 const struct cls_subtable *table;
6831 ovs_rwlock_rdlock(&oftable->cls.rwlock);
6832 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.subtables) {
6833 if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
6834 const struct cls_rule *rule;
6836 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
6837 uint16_t vid = miniflow_get_vid(&rule->match.flow);
6838 bitmap_set1(vlan_bitmap, vid);
6839 bitmap_set1(ofproto->vlan_bitmap, vid);
6843 ovs_rwlock_unlock(&oftable->cls.rwlock);
6847 /* Returns true if new VLANs have come into use by the flow table since the
6848 * last call to ofproto_get_vlan_usage().
6850 * We don't track when old VLANs stop being used. */
6852 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
6854 return ofproto->vlans_changed;
6857 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
6858 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
6859 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
6860 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
6861 * then the VLAN device is un-enslaved. */
6863 ofproto_port_set_realdev(struct ofproto *ofproto, ofp_port_t vlandev_ofp_port,
6864 ofp_port_t realdev_ofp_port, int vid)
6866 struct ofport *ofport;
6869 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
6871 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
6873 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
6874 ofproto->name, vlandev_ofp_port);
6878 if (!ofproto->ofproto_class->set_realdev) {
6879 if (!vlandev_ofp_port) {
6882 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
6886 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
6888 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
6889 ofproto->name, vlandev_ofp_port,
6890 netdev_get_name(ofport->netdev), ovs_strerror(error));