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.actions[0] =
538 #define OFPAT11_ACTION(ENUM, STRUCT, EXTENSIBLE, NAME) (1 << ENUM) |
539 #include "ofp-util.def"
542 error = ofproto->ofproto_class->construct(ofproto);
544 VLOG_ERR("failed to open datapath %s: %s",
545 datapath_name, ovs_strerror(error));
546 ofproto_destroy__(ofproto);
550 /* Check that hidden tables, if any, are at the end. */
551 ovs_assert(ofproto->n_tables);
552 for (i = 0; i + 1 < ofproto->n_tables; i++) {
553 enum oftable_flags flags = ofproto->tables[i].flags;
554 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
556 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
559 ofproto->datapath_id = pick_datapath_id(ofproto);
562 /* Initialize meters table. */
563 if (ofproto->ofproto_class->meter_get_features) {
564 ofproto->ofproto_class->meter_get_features(ofproto,
565 &ofproto->meter_features);
567 memset(&ofproto->meter_features, 0, sizeof ofproto->meter_features);
569 ofproto->meters = xzalloc((ofproto->meter_features.max_meters + 1)
570 * sizeof(struct meter *));
576 /* Must be called (only) by an ofproto implementation in its constructor
577 * function. See the large comment on 'construct' in struct ofproto_class for
580 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
582 struct oftable *table;
584 ovs_assert(!ofproto->n_tables);
585 ovs_assert(n_tables >= 1 && n_tables <= 255);
587 ofproto->n_tables = n_tables;
588 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
589 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
594 /* To be optionally called (only) by an ofproto implementation in its
595 * constructor function. See the large comment on 'construct' in struct
596 * ofproto_class for details.
598 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
599 * will then ensure that actions passed into the ofproto implementation will
600 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
601 * function is not called, there will be no such restriction.
603 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
604 * the 'max_ports' restriction. */
606 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
608 ovs_assert(max_ports <= ofp_to_u16(OFPP_MAX));
609 ofproto->max_ports = max_ports;
613 ofproto_get_datapath_id(const struct ofproto *ofproto)
615 return ofproto->datapath_id;
619 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
621 uint64_t old_dpid = p->datapath_id;
622 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
623 if (p->datapath_id != old_dpid) {
624 /* Force all active connections to reconnect, since there is no way to
625 * notify a controller that the datapath ID has changed. */
626 ofproto_reconnect_controllers(p);
631 ofproto_set_controllers(struct ofproto *p,
632 const struct ofproto_controller *controllers,
633 size_t n_controllers, uint32_t allowed_versions)
635 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
640 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
642 connmgr_set_fail_mode(p->connmgr, fail_mode);
645 /* Drops the connections between 'ofproto' and all of its controllers, forcing
646 * them to reconnect. */
648 ofproto_reconnect_controllers(struct ofproto *ofproto)
650 connmgr_reconnect(ofproto->connmgr);
653 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
654 * in-band control should guarantee access, in the same way that in-band
655 * control guarantees access to OpenFlow controllers. */
657 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
658 const struct sockaddr_in *extras, size_t n)
660 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
663 /* Sets the OpenFlow queue used by flows set up by in-band control on
664 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
665 * flows will use the default queue. */
667 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
669 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
672 /* Sets the number of flows at which eviction from the kernel flow table
675 ofproto_set_flow_eviction_threshold(unsigned threshold)
677 flow_eviction_threshold = MAX(OFPROTO_FLOW_EVICTION_THRESHOLD_MIN,
681 /* Sets the path for handling flow misses. */
683 ofproto_set_flow_miss_model(unsigned model)
685 flow_miss_model = model;
688 /* If forward_bpdu is true, the NORMAL action will forward frames with
689 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
690 * the NORMAL action will drop these frames. */
692 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
694 bool old_val = ofproto->forward_bpdu;
695 ofproto->forward_bpdu = forward_bpdu;
696 if (old_val != ofproto->forward_bpdu) {
697 if (ofproto->ofproto_class->forward_bpdu_changed) {
698 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
703 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
704 * 'idle_time', in seconds, and the maximum number of MAC table entries to
707 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
710 if (ofproto->ofproto_class->set_mac_table_config) {
711 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
716 /* Sets number of upcall handler threads. The default is
717 * (number of online cores - 2). */
719 ofproto_set_n_handler_threads(unsigned limit)
722 n_handler_threads = limit;
724 int n_proc = sysconf(_SC_NPROCESSORS_ONLN);
725 n_handler_threads = n_proc > 2 ? n_proc - 2 : 1;
730 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
733 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
737 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
739 return connmgr_set_snoops(ofproto->connmgr, snoops);
743 ofproto_set_netflow(struct ofproto *ofproto,
744 const struct netflow_options *nf_options)
746 if (nf_options && sset_is_empty(&nf_options->collectors)) {
750 if (ofproto->ofproto_class->set_netflow) {
751 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
753 return nf_options ? EOPNOTSUPP : 0;
758 ofproto_set_sflow(struct ofproto *ofproto,
759 const struct ofproto_sflow_options *oso)
761 if (oso && sset_is_empty(&oso->targets)) {
765 if (ofproto->ofproto_class->set_sflow) {
766 return ofproto->ofproto_class->set_sflow(ofproto, oso);
768 return oso ? EOPNOTSUPP : 0;
773 ofproto_set_ipfix(struct ofproto *ofproto,
774 const struct ofproto_ipfix_bridge_exporter_options *bo,
775 const struct ofproto_ipfix_flow_exporter_options *fo,
778 if (ofproto->ofproto_class->set_ipfix) {
779 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
781 return (bo || fo) ? EOPNOTSUPP : 0;
786 ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
788 flow_restore_wait = flow_restore_wait_db;
792 ofproto_get_flow_restore_wait(void)
794 return flow_restore_wait;
798 /* Spanning Tree Protocol (STP) configuration. */
800 /* Configures STP on 'ofproto' using the settings defined in 's'. If
801 * 's' is NULL, disables STP.
803 * Returns 0 if successful, otherwise a positive errno value. */
805 ofproto_set_stp(struct ofproto *ofproto,
806 const struct ofproto_stp_settings *s)
808 return (ofproto->ofproto_class->set_stp
809 ? ofproto->ofproto_class->set_stp(ofproto, s)
813 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
814 * 'enabled' member of 's' is false, then the other members are not
817 * Returns 0 if successful, otherwise a positive errno value. */
819 ofproto_get_stp_status(struct ofproto *ofproto,
820 struct ofproto_stp_status *s)
822 return (ofproto->ofproto_class->get_stp_status
823 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
827 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
828 * in 's'. The caller is responsible for assigning STP port numbers
829 * (using the 'port_num' member in the range of 1 through 255, inclusive)
830 * and ensuring there are no duplicates. If the 's' is NULL, then STP
831 * is disabled on the port.
833 * Returns 0 if successful, otherwise a positive errno value.*/
835 ofproto_port_set_stp(struct ofproto *ofproto, ofp_port_t ofp_port,
836 const struct ofproto_port_stp_settings *s)
838 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
840 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
841 ofproto->name, ofp_port);
845 return (ofproto->ofproto_class->set_stp_port
846 ? ofproto->ofproto_class->set_stp_port(ofport, s)
850 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
851 * 's'. If the 'enabled' member in 's' is false, then the other members
852 * are not meaningful.
854 * Returns 0 if successful, otherwise a positive errno value.*/
856 ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
857 struct ofproto_port_stp_status *s)
859 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
861 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
862 "port %"PRIu16, ofproto->name, ofp_port);
866 return (ofproto->ofproto_class->get_stp_port_status
867 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
871 /* Queue DSCP configuration. */
873 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
874 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
875 * to implement QoS. Instead, it is used to implement features which require
876 * knowledge of what queues exist on a port, and some basic information about
879 * Returns 0 if successful, otherwise a positive errno value. */
881 ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port,
882 const struct ofproto_port_queue *queues,
885 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
888 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
889 ofproto->name, ofp_port);
893 return (ofproto->ofproto_class->set_queues
894 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
898 /* Connectivity Fault Management configuration. */
900 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
902 ofproto_port_clear_cfm(struct ofproto *ofproto, ofp_port_t ofp_port)
904 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
905 if (ofport && ofproto->ofproto_class->set_cfm) {
906 ofproto->ofproto_class->set_cfm(ofport, NULL);
910 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
911 * basic configuration from the configuration members in 'cfm', and the remote
912 * maintenance point ID from remote_mpid. Ignores the statistics members of
915 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
917 ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port,
918 const struct cfm_settings *s)
920 struct ofport *ofport;
923 ofport = ofproto_get_port(ofproto, ofp_port);
925 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
926 ofproto->name, ofp_port);
930 /* XXX: For configuration simplicity, we only support one remote_mpid
931 * outside of the CFM module. It's not clear if this is the correct long
932 * term solution or not. */
933 error = (ofproto->ofproto_class->set_cfm
934 ? ofproto->ofproto_class->set_cfm(ofport, s)
937 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
938 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
939 ovs_strerror(error));
943 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
944 * 'ofproto' does not have a port 'ofp_port'. */
946 ofproto_port_set_bfd(struct ofproto *ofproto, ofp_port_t ofp_port,
947 const struct smap *cfg)
949 struct ofport *ofport;
952 ofport = ofproto_get_port(ofproto, ofp_port);
954 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
955 ofproto->name, ofp_port);
959 error = (ofproto->ofproto_class->set_bfd
960 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
963 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
964 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
965 ovs_strerror(error));
969 /* Populates 'status' with key value pairs indicating the status of the BFD
970 * session on 'ofp_port'. This information is intended to be populated in the
971 * OVS database. Has no effect if 'ofp_port' is not na OpenFlow port in
974 ofproto_port_get_bfd_status(struct ofproto *ofproto, ofp_port_t ofp_port,
977 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
978 return (ofport && ofproto->ofproto_class->get_bfd_status
979 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
983 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
984 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
985 * 0 if LACP partner information is not current (generally indicating a
986 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
988 ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port)
990 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
991 return (ofport && ofproto->ofproto_class->port_is_lacp_current
992 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
998 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
999 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
1000 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
1001 * configuration plus, if there is more than one slave, a bonding
1004 * If 'aux' is already registered then this function updates its configuration
1005 * to 's'. Otherwise, this function registers a new bundle.
1007 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
1010 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
1011 const struct ofproto_bundle_settings *s)
1013 return (ofproto->ofproto_class->bundle_set
1014 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
1018 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
1019 * If no such bundle has been registered, this has no effect. */
1021 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
1023 return ofproto_bundle_register(ofproto, aux, NULL);
1027 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
1028 * If 'aux' is already registered then this function updates its configuration
1029 * to 's'. Otherwise, this function registers a new mirror. */
1031 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
1032 const struct ofproto_mirror_settings *s)
1034 return (ofproto->ofproto_class->mirror_set
1035 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
1039 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
1040 * If no mirror has been registered, this has no effect. */
1042 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
1044 return ofproto_mirror_register(ofproto, aux, NULL);
1047 /* Retrieves statistics from mirror associated with client data pointer
1048 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
1049 * 'bytes', respectively. If a particular counters is not supported,
1050 * the appropriate argument is set to UINT64_MAX. */
1052 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
1053 uint64_t *packets, uint64_t *bytes)
1055 if (!ofproto->ofproto_class->mirror_get_stats) {
1056 *packets = *bytes = UINT64_MAX;
1060 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
1064 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
1065 * which all packets are flooded, instead of using MAC learning. If
1066 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1068 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
1071 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
1073 return (ofproto->ofproto_class->set_flood_vlans
1074 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
1078 /* Returns true if 'aux' is a registered bundle that is currently in use as the
1079 * output for a mirror. */
1081 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
1083 return (ofproto->ofproto_class->is_mirror_output_bundle
1084 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
1088 /* Configuration of OpenFlow tables. */
1090 /* Returns the number of OpenFlow tables in 'ofproto'. */
1092 ofproto_get_n_tables(const struct ofproto *ofproto)
1094 return ofproto->n_tables;
1097 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
1098 * settings from 's'. 'table_id' must be in the range 0 through the number of
1099 * OpenFlow tables in 'ofproto' minus 1, inclusive.
1101 * For read-only tables, only the name may be configured. */
1103 ofproto_configure_table(struct ofproto *ofproto, int table_id,
1104 const struct ofproto_table_settings *s)
1106 struct oftable *table;
1108 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
1109 table = &ofproto->tables[table_id];
1111 oftable_set_name(table, s->name);
1113 if (table->flags & OFTABLE_READONLY) {
1118 oftable_enable_eviction(table, s->groups, s->n_groups);
1120 oftable_disable_eviction(table);
1123 table->max_flows = s->max_flows;
1124 ovs_rwlock_rdlock(&table->cls.rwlock);
1125 if (classifier_count(&table->cls) > table->max_flows
1126 && table->eviction_fields) {
1127 /* 'table' contains more flows than allowed. We might not be able to
1128 * evict them right away because of the asynchronous nature of flow
1129 * table changes. Schedule eviction for later. */
1130 switch (ofproto->state) {
1132 ofproto->state = S_EVICT;
1136 /* We're already deleting flows, nothing more to do. */
1140 ovs_rwlock_unlock(&table->cls.rwlock);
1144 ofproto_has_snoops(const struct ofproto *ofproto)
1146 return connmgr_has_snoops(ofproto->connmgr);
1150 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1152 connmgr_get_snoops(ofproto->connmgr, snoops);
1156 ofproto_rule_delete__(struct ofproto *ofproto, struct rule *rule,
1158 OVS_REQUIRES(ofproto_mutex)
1160 struct ofopgroup *group;
1162 ovs_assert(!rule->pending);
1164 group = ofopgroup_create_unattached(ofproto);
1165 delete_flow__(rule, group, reason);
1166 ofopgroup_submit(group);
1169 /* Deletes 'rule' from 'cls' within 'ofproto'.
1171 * Within an ofproto implementation, this function allows an ofproto
1172 * implementation to destroy any rules that remain when its ->destruct()
1173 * function is called. This function is not suitable for use elsewhere in an
1174 * ofproto implementation.
1176 * This function implements steps 4.4 and 4.5 in the section titled "Rule Life
1177 * Cycle" in ofproto-provider.h. */
1179 ofproto_rule_delete(struct ofproto *ofproto, struct rule *rule)
1180 OVS_EXCLUDED(ofproto_mutex)
1182 struct ofopgroup *group;
1184 ovs_mutex_lock(&ofproto_mutex);
1185 ovs_assert(!rule->pending);
1187 group = ofopgroup_create_unattached(ofproto);
1188 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1189 oftable_remove_rule__(ofproto, rule);
1190 ofproto->ofproto_class->rule_delete(rule);
1191 ofopgroup_submit(group);
1193 ovs_mutex_unlock(&ofproto_mutex);
1197 ofproto_flush__(struct ofproto *ofproto)
1198 OVS_EXCLUDED(ofproto_mutex)
1200 struct oftable *table;
1202 if (ofproto->ofproto_class->flush) {
1203 ofproto->ofproto_class->flush(ofproto);
1206 ovs_mutex_lock(&ofproto_mutex);
1207 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1208 struct rule *rule, *next_rule;
1209 struct cls_cursor cursor;
1211 if (table->flags & OFTABLE_HIDDEN) {
1215 ovs_rwlock_rdlock(&table->cls.rwlock);
1216 cls_cursor_init(&cursor, &table->cls, NULL);
1217 ovs_rwlock_unlock(&table->cls.rwlock);
1218 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1219 if (!rule->pending) {
1220 ofproto_rule_delete__(ofproto, rule, OFPRR_DELETE);
1224 ovs_mutex_unlock(&ofproto_mutex);
1227 static void delete_group(struct ofproto *ofproto, uint32_t group_id);
1230 ofproto_destroy__(struct ofproto *ofproto)
1231 OVS_EXCLUDED(ofproto_mutex)
1233 struct oftable *table;
1235 ovs_assert(list_is_empty(&ofproto->pending));
1237 destroy_rule_executes(ofproto);
1238 guarded_list_destroy(&ofproto->rule_executes);
1240 delete_group(ofproto, OFPG_ALL);
1241 ovs_rwlock_destroy(&ofproto->groups_rwlock);
1242 hmap_destroy(&ofproto->groups);
1244 connmgr_destroy(ofproto->connmgr);
1246 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1247 free(ofproto->name);
1248 free(ofproto->type);
1249 free(ofproto->mfr_desc);
1250 free(ofproto->hw_desc);
1251 free(ofproto->sw_desc);
1252 free(ofproto->serial_desc);
1253 free(ofproto->dp_desc);
1254 hmap_destroy(&ofproto->ports);
1255 hmap_destroy(&ofproto->ofport_usage);
1256 shash_destroy(&ofproto->port_by_name);
1257 simap_destroy(&ofproto->ofp_requests);
1259 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1260 oftable_destroy(table);
1262 free(ofproto->tables);
1264 hmap_destroy(&ofproto->deletions);
1266 free(ofproto->vlan_bitmap);
1268 ofproto->ofproto_class->dealloc(ofproto);
1272 ofproto_destroy(struct ofproto *p)
1273 OVS_EXCLUDED(ofproto_mutex)
1275 struct ofport *ofport, *next_ofport;
1276 struct ofport_usage *usage, *next_usage;
1283 meter_delete(p, 1, p->meter_features.max_meters);
1284 p->meter_features.max_meters = 0;
1290 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1291 ofport_destroy(ofport);
1294 HMAP_FOR_EACH_SAFE (usage, next_usage, hmap_node, &p->ofport_usage) {
1295 hmap_remove(&p->ofport_usage, &usage->hmap_node);
1299 p->ofproto_class->destruct(p);
1300 ofproto_destroy__(p);
1303 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1304 * kernel datapath, for example, this destroys the datapath in the kernel, and
1305 * with the netdev-based datapath, it tears down the data structures that
1306 * represent the datapath.
1308 * The datapath should not be currently open as an ofproto. */
1310 ofproto_delete(const char *name, const char *type)
1312 const struct ofproto_class *class = ofproto_class_find__(type);
1313 return (!class ? EAFNOSUPPORT
1314 : !class->del ? EACCES
1315 : class->del(type, name));
1319 process_port_change(struct ofproto *ofproto, int error, char *devname)
1321 if (error == ENOBUFS) {
1322 reinit_ports(ofproto);
1323 } else if (!error) {
1324 update_port(ofproto, devname);
1330 ofproto_type_run(const char *datapath_type)
1332 const struct ofproto_class *class;
1335 datapath_type = ofproto_normalize_type(datapath_type);
1336 class = ofproto_class_find__(datapath_type);
1338 error = class->type_run ? class->type_run(datapath_type) : 0;
1339 if (error && error != EAGAIN) {
1340 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1341 datapath_type, ovs_strerror(error));
1347 ofproto_type_run_fast(const char *datapath_type)
1349 const struct ofproto_class *class;
1352 datapath_type = ofproto_normalize_type(datapath_type);
1353 class = ofproto_class_find__(datapath_type);
1355 error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
1356 if (error && error != EAGAIN) {
1357 VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
1358 datapath_type, ovs_strerror(error));
1364 ofproto_type_wait(const char *datapath_type)
1366 const struct ofproto_class *class;
1368 datapath_type = ofproto_normalize_type(datapath_type);
1369 class = ofproto_class_find__(datapath_type);
1371 if (class->type_wait) {
1372 class->type_wait(datapath_type);
1377 any_pending_ops(const struct ofproto *p)
1378 OVS_EXCLUDED(ofproto_mutex)
1382 ovs_mutex_lock(&ofproto_mutex);
1383 b = !list_is_empty(&p->pending);
1384 ovs_mutex_unlock(&ofproto_mutex);
1390 ofproto_run(struct ofproto *p)
1392 struct sset changed_netdevs;
1393 const char *changed_netdev;
1394 struct ofport *ofport;
1397 error = p->ofproto_class->run(p);
1398 if (error && error != EAGAIN) {
1399 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, ovs_strerror(error));
1402 run_rule_executes(p);
1404 /* Restore the eviction group heap invariant occasionally. */
1405 if (p->eviction_group_timer < time_msec()) {
1408 p->eviction_group_timer = time_msec() + 1000;
1410 for (i = 0; i < p->n_tables; i++) {
1411 struct oftable *table = &p->tables[i];
1412 struct eviction_group *evg;
1413 struct cls_cursor cursor;
1416 if (!table->eviction_fields) {
1420 ovs_mutex_lock(&ofproto_mutex);
1421 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
1422 heap_rebuild(&evg->rules);
1425 ovs_rwlock_rdlock(&table->cls.rwlock);
1426 cls_cursor_init(&cursor, &table->cls, NULL);
1427 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1428 if (!rule->eviction_group
1429 && (rule->idle_timeout || rule->hard_timeout)) {
1430 eviction_group_add_rule(rule);
1433 ovs_rwlock_unlock(&table->cls.rwlock);
1434 ovs_mutex_unlock(&ofproto_mutex);
1438 if (p->ofproto_class->port_poll) {
1441 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1442 process_port_change(p, error, devname);
1446 /* Update OpenFlow port status for any port whose netdev has changed.
1448 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1449 * destroyed, so it's not safe to update ports directly from the
1450 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1451 * need this two-phase approach. */
1452 sset_init(&changed_netdevs);
1453 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1454 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1455 if (ofport->change_seq != change_seq) {
1456 ofport->change_seq = change_seq;
1457 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1460 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1461 update_port(p, changed_netdev);
1463 sset_destroy(&changed_netdevs);
1467 connmgr_run(p->connmgr, handle_openflow);
1471 connmgr_run(p->connmgr, NULL);
1473 if (!any_pending_ops(p)) {
1474 p->state = S_OPENFLOW;
1479 connmgr_run(p->connmgr, NULL);
1481 if (!any_pending_ops(p)) {
1482 connmgr_flushed(p->connmgr);
1483 p->state = S_OPENFLOW;
1491 if (time_msec() >= p->next_op_report) {
1492 long long int ago = (time_msec() - p->first_op) / 1000;
1493 long long int interval = (p->last_op - p->first_op) / 1000;
1497 ds_put_format(&s, "%d flow_mods ",
1498 p->n_add + p->n_delete + p->n_modify);
1499 if (interval == ago) {
1500 ds_put_format(&s, "in the last %lld s", ago);
1501 } else if (interval) {
1502 ds_put_format(&s, "in the %lld s starting %lld s ago",
1505 ds_put_format(&s, "%lld s ago", ago);
1508 ds_put_cstr(&s, " (");
1510 ds_put_format(&s, "%d adds, ", p->n_add);
1513 ds_put_format(&s, "%d deletes, ", p->n_delete);
1516 ds_put_format(&s, "%d modifications, ", p->n_modify);
1519 ds_put_char(&s, ')');
1521 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1524 p->n_add = p->n_delete = p->n_modify = 0;
1525 p->next_op_report = LLONG_MAX;
1531 /* Performs periodic activity required by 'ofproto' that needs to be done
1532 * with the least possible latency.
1534 * It makes sense to call this function a couple of times per poll loop, to
1535 * provide a significant performance boost on some benchmarks with the
1536 * ofproto-dpif implementation. */
1538 ofproto_run_fast(struct ofproto *p)
1542 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1543 if (error && error != EAGAIN) {
1544 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1545 p->name, ovs_strerror(error));
1551 ofproto_wait(struct ofproto *p)
1553 struct ofport *ofport;
1555 p->ofproto_class->wait(p);
1556 if (p->ofproto_class->port_poll_wait) {
1557 p->ofproto_class->port_poll_wait(p);
1560 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1561 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1562 poll_immediate_wake();
1568 connmgr_wait(p->connmgr, true);
1573 connmgr_wait(p->connmgr, false);
1574 if (!any_pending_ops(p)) {
1575 poll_immediate_wake();
1582 ofproto_is_alive(const struct ofproto *p)
1584 return connmgr_has_controllers(p->connmgr);
1587 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1588 * memory_report(). */
1590 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1592 const struct oftable *table;
1593 unsigned int n_rules;
1595 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1597 ovs_mutex_lock(&ofproto_mutex);
1598 simap_increase(usage, "ops",
1599 ofproto->n_pending + hmap_count(&ofproto->deletions));
1600 ovs_mutex_unlock(&ofproto_mutex);
1603 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1604 ovs_rwlock_rdlock(&table->cls.rwlock);
1605 n_rules += classifier_count(&table->cls);
1606 ovs_rwlock_unlock(&table->cls.rwlock);
1608 simap_increase(usage, "rules", n_rules);
1610 if (ofproto->ofproto_class->get_memory_usage) {
1611 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1614 connmgr_get_memory_usage(ofproto->connmgr, usage);
1618 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1621 connmgr_get_controller_info(ofproto->connmgr, info);
1625 ofproto_free_ofproto_controller_info(struct shash *info)
1627 connmgr_free_controller_info(info);
1630 /* Makes a deep copy of 'old' into 'port'. */
1632 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1634 port->name = xstrdup(old->name);
1635 port->type = xstrdup(old->type);
1636 port->ofp_port = old->ofp_port;
1639 /* Frees memory allocated to members of 'ofproto_port'.
1641 * Do not call this function on an ofproto_port obtained from
1642 * ofproto_port_dump_next(): that function retains ownership of the data in the
1645 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1647 free(ofproto_port->name);
1648 free(ofproto_port->type);
1651 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1653 * This function provides no status indication. An error status for the entire
1654 * dump operation is provided when it is completed by calling
1655 * ofproto_port_dump_done().
1658 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1659 const struct ofproto *ofproto)
1661 dump->ofproto = ofproto;
1662 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1666 /* Attempts to retrieve another port from 'dump', which must have been created
1667 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1668 * 'port' and returns true. On failure, returns false.
1670 * Failure might indicate an actual error or merely that the last port has been
1671 * dumped. An error status for the entire dump operation is provided when it
1672 * is completed by calling ofproto_port_dump_done().
1674 * The ofproto owns the data stored in 'port'. It will remain valid until at
1675 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1676 * ofproto_port_dump_done(). */
1678 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1679 struct ofproto_port *port)
1681 const struct ofproto *ofproto = dump->ofproto;
1687 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1690 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1696 /* Completes port table dump operation 'dump', which must have been created
1697 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1698 * error-free, otherwise a positive errno value describing the problem. */
1700 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1702 const struct ofproto *ofproto = dump->ofproto;
1704 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1707 return dump->error == EOF ? 0 : dump->error;
1710 /* Returns the type to pass to netdev_open() when a datapath of type
1711 * 'datapath_type' has a port of type 'port_type', for a few special
1712 * cases when a netdev type differs from a port type. For example, when
1713 * using the userspace datapath, a port of type "internal" needs to be
1716 * Returns either 'type' itself or a string literal, which must not be
1719 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1721 const struct ofproto_class *class;
1723 datapath_type = ofproto_normalize_type(datapath_type);
1724 class = ofproto_class_find__(datapath_type);
1729 return (class->port_open_type
1730 ? class->port_open_type(datapath_type, port_type)
1734 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1735 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1736 * the port's OpenFlow port number.
1738 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1739 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1740 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1741 * 'ofp_portp' is non-null). */
1743 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1744 ofp_port_t *ofp_portp)
1746 ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1749 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1751 const char *netdev_name = netdev_get_name(netdev);
1753 simap_put(&ofproto->ofp_requests, netdev_name,
1754 ofp_to_u16(ofp_port));
1755 update_port(ofproto, netdev_name);
1758 struct ofproto_port ofproto_port;
1760 ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
1762 *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
1763 ofproto_port_destroy(&ofproto_port);
1768 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1769 * initializes '*port' appropriately; on failure, returns a positive errno
1772 * The caller owns the data in 'ofproto_port' and must free it with
1773 * ofproto_port_destroy() when it is no longer needed. */
1775 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1776 struct ofproto_port *port)
1780 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1782 memset(port, 0, sizeof *port);
1787 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1788 * Returns 0 if successful, otherwise a positive errno. */
1790 ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port)
1792 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1793 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1794 struct simap_node *ofp_request_node;
1797 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1798 if (ofp_request_node) {
1799 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1802 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1803 if (!error && ofport) {
1804 /* 'name' is the netdev's name and update_port() is going to close the
1805 * netdev. Just in case update_port() refers to 'name' after it
1806 * destroys 'ofport', make a copy of it around the update_port()
1808 char *devname = xstrdup(name);
1809 update_port(ofproto, devname);
1816 flow_mod_init(struct ofputil_flow_mod *fm,
1817 const struct match *match, unsigned int priority,
1818 const struct ofpact *ofpacts, size_t ofpacts_len,
1819 enum ofp_flow_mod_command command)
1821 memset(fm, 0, sizeof *fm);
1823 fm->priority = priority;
1826 fm->modify_cookie = false;
1828 fm->command = command;
1829 fm->idle_timeout = 0;
1830 fm->hard_timeout = 0;
1831 fm->buffer_id = UINT32_MAX;
1832 fm->out_port = OFPP_ANY;
1833 fm->out_group = OFPG_ANY;
1835 fm->ofpacts = CONST_CAST(struct ofpact *, ofpacts);
1836 fm->ofpacts_len = ofpacts_len;
1840 simple_flow_mod(struct ofproto *ofproto,
1841 const struct match *match, unsigned int priority,
1842 const struct ofpact *ofpacts, size_t ofpacts_len,
1843 enum ofp_flow_mod_command command)
1845 struct ofputil_flow_mod fm;
1847 flow_mod_init(&fm, match, priority, ofpacts, ofpacts_len, command);
1849 return handle_flow_mod__(ofproto, NULL, &fm, NULL);
1852 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1853 * performs the 'n_actions' actions in 'actions'. The new flow will not
1856 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1857 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1858 * controllers; otherwise, it will be hidden.
1860 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1862 * This is a helper function for in-band control and fail-open. */
1864 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1865 unsigned int priority,
1866 const struct ofpact *ofpacts, size_t ofpacts_len)
1867 OVS_EXCLUDED(ofproto_mutex)
1869 const struct rule *rule;
1872 /* First do a cheap check whether the rule we're looking for already exists
1873 * with the actions that we want. If it does, then we're done. */
1874 ovs_rwlock_rdlock(&ofproto->tables[0].cls.rwlock);
1875 rule = rule_from_cls_rule(classifier_find_match_exactly(
1876 &ofproto->tables[0].cls, match, priority));
1878 ovs_mutex_lock(&rule->mutex);
1879 must_add = !ofpacts_equal(rule->actions->ofpacts,
1880 rule->actions->ofpacts_len,
1881 ofpacts, ofpacts_len);
1882 ovs_mutex_unlock(&rule->mutex);
1886 ovs_rwlock_unlock(&ofproto->tables[0].cls.rwlock);
1888 /* If there's no such rule or the rule doesn't have the actions we want,
1889 * fall back to a executing a full flow mod. We can't optimize this at
1890 * all because we didn't take enough locks above to ensure that the flow
1891 * table didn't already change beneath us. */
1893 simple_flow_mod(ofproto, match, priority, ofpacts, ofpacts_len,
1894 OFPFC_MODIFY_STRICT);
1898 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1899 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1900 * operation cannot be initiated now but may be retried later.
1902 * This is a helper function for in-band control and fail-open and the "learn"
1905 ofproto_flow_mod(struct ofproto *ofproto, struct ofputil_flow_mod *fm)
1906 OVS_EXCLUDED(ofproto_mutex)
1908 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1911 /* Searches for a rule with matching criteria exactly equal to 'target' in
1912 * ofproto's table 0 and, if it finds one, deletes it.
1914 * This is a helper function for in-band control and fail-open. */
1916 ofproto_delete_flow(struct ofproto *ofproto,
1917 const struct match *target, unsigned int priority)
1918 OVS_EXCLUDED(ofproto_mutex)
1920 struct classifier *cls = &ofproto->tables[0].cls;
1923 /* First do a cheap check whether the rule we're looking for has already
1924 * been deleted. If so, then we're done. */
1925 ovs_rwlock_rdlock(&cls->rwlock);
1926 rule = rule_from_cls_rule(classifier_find_match_exactly(cls, target,
1928 ovs_rwlock_unlock(&cls->rwlock);
1933 /* Fall back to a executing a full flow mod. We can't optimize this at all
1934 * because we didn't take enough locks above to ensure that the flow table
1935 * didn't already change beneath us. */
1936 return simple_flow_mod(ofproto, target, priority, NULL, 0,
1937 OFPFC_DELETE_STRICT) != OFPROTO_POSTPONE;
1940 /* Starts the process of deleting all of the flows from all of ofproto's flow
1941 * tables and then reintroducing the flows required by in-band control and
1942 * fail-open. The process will complete in a later call to ofproto_run(). */
1944 ofproto_flush_flows(struct ofproto *ofproto)
1946 COVERAGE_INC(ofproto_flush);
1947 ofproto->state = S_FLUSH;
1951 reinit_ports(struct ofproto *p)
1953 struct ofproto_port_dump dump;
1954 struct sset devnames;
1955 struct ofport *ofport;
1956 struct ofproto_port ofproto_port;
1957 const char *devname;
1959 COVERAGE_INC(ofproto_reinit_ports);
1961 sset_init(&devnames);
1962 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1963 sset_add(&devnames, netdev_get_name(ofport->netdev));
1965 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1966 sset_add(&devnames, ofproto_port.name);
1969 SSET_FOR_EACH (devname, &devnames) {
1970 update_port(p, devname);
1972 sset_destroy(&devnames);
1976 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
1980 port_idx = simap_get(&ofproto->ofp_requests, netdev_name);
1981 port_idx = port_idx ? port_idx : UINT16_MAX;
1983 if (port_idx >= ofproto->max_ports
1984 || ofport_get_usage(ofproto, u16_to_ofp(port_idx)) == LLONG_MAX) {
1985 uint16_t lru_ofport = 0, end_port_no = ofproto->alloc_port_no;
1986 long long int last_used_at, lru = LLONG_MAX;
1988 /* Search for a free OpenFlow port number. We try not to
1989 * immediately reuse them to prevent problems due to old
1992 if (++ofproto->alloc_port_no >= ofproto->max_ports) {
1993 ofproto->alloc_port_no = 1;
1995 last_used_at = ofport_get_usage(ofproto,
1996 u16_to_ofp(ofproto->alloc_port_no));
1997 if (!last_used_at) {
1998 port_idx = ofproto->alloc_port_no;
2000 } else if ( last_used_at < time_msec() - 60*60*1000) {
2001 /* If the port with ofport 'ofproto->alloc_port_no' was deleted
2002 * more than an hour ago, consider it usable. */
2003 ofport_remove_usage(ofproto,
2004 u16_to_ofp(ofproto->alloc_port_no));
2005 port_idx = ofproto->alloc_port_no;
2007 } else if (last_used_at < lru) {
2009 lru_ofport = ofproto->alloc_port_no;
2012 if (ofproto->alloc_port_no == end_port_no) {
2014 port_idx = lru_ofport;
2021 ofport_set_usage(ofproto, u16_to_ofp(port_idx), LLONG_MAX);
2022 return u16_to_ofp(port_idx);
2026 dealloc_ofp_port(struct ofproto *ofproto, ofp_port_t ofp_port)
2028 if (ofp_to_u16(ofp_port) < ofproto->max_ports) {
2029 ofport_set_usage(ofproto, ofp_port, time_msec());
2033 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
2034 * pointer if the netdev cannot be opened. On success, also fills in
2036 static struct netdev *
2037 ofport_open(struct ofproto *ofproto,
2038 struct ofproto_port *ofproto_port,
2039 struct ofputil_phy_port *pp)
2041 enum netdev_flags flags;
2042 struct netdev *netdev;
2045 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
2047 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
2048 "cannot be opened (%s)",
2050 ofproto_port->name, ofproto_port->ofp_port,
2051 ofproto_port->name, ovs_strerror(error));
2055 if (ofproto_port->ofp_port == OFPP_NONE) {
2056 if (!strcmp(ofproto->name, ofproto_port->name)) {
2057 ofproto_port->ofp_port = OFPP_LOCAL;
2059 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
2060 ofproto_port->name);
2063 pp->port_no = ofproto_port->ofp_port;
2064 netdev_get_etheraddr(netdev, pp->hw_addr);
2065 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
2066 netdev_get_flags(netdev, &flags);
2067 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
2068 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
2069 netdev_get_features(netdev, &pp->curr, &pp->advertised,
2070 &pp->supported, &pp->peer);
2071 pp->curr_speed = netdev_features_to_bps(pp->curr, 0) / 1000;
2072 pp->max_speed = netdev_features_to_bps(pp->supported, 0) / 1000;
2077 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
2078 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
2081 ofport_equal(const struct ofputil_phy_port *a,
2082 const struct ofputil_phy_port *b)
2084 return (eth_addr_equals(a->hw_addr, b->hw_addr)
2085 && a->state == b->state
2086 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
2087 && a->curr == b->curr
2088 && a->advertised == b->advertised
2089 && a->supported == b->supported
2090 && a->peer == b->peer
2091 && a->curr_speed == b->curr_speed
2092 && a->max_speed == b->max_speed);
2095 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
2096 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
2097 * one with the same name or port number). */
2099 ofport_install(struct ofproto *p,
2100 struct netdev *netdev, const struct ofputil_phy_port *pp)
2102 const char *netdev_name = netdev_get_name(netdev);
2103 struct ofport *ofport;
2106 /* Create ofport. */
2107 ofport = p->ofproto_class->port_alloc();
2112 ofport->ofproto = p;
2113 ofport->netdev = netdev;
2114 ofport->change_seq = netdev_change_seq(netdev);
2116 ofport->ofp_port = pp->port_no;
2117 ofport->created = time_msec();
2119 /* Add port to 'p'. */
2120 hmap_insert(&p->ports, &ofport->hmap_node,
2121 hash_ofp_port(ofport->ofp_port));
2122 shash_add(&p->port_by_name, netdev_name, ofport);
2124 update_mtu(p, ofport);
2126 /* Let the ofproto_class initialize its private data. */
2127 error = p->ofproto_class->port_construct(ofport);
2131 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
2135 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
2136 p->name, netdev_name, ovs_strerror(error));
2138 ofport_destroy__(ofport);
2140 netdev_close(netdev);
2144 /* Removes 'ofport' from 'p' and destroys it. */
2146 ofport_remove(struct ofport *ofport)
2148 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
2150 ofport_destroy(ofport);
2153 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
2156 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
2158 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
2160 ofport_remove(port);
2164 /* Updates 'port' with new 'pp' description.
2166 * Does not handle a name or port number change. The caller must implement
2167 * such a change as a delete followed by an add. */
2169 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
2171 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
2172 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
2173 | (pp->config & OFPUTIL_PC_PORT_DOWN));
2174 port->pp.state = pp->state;
2175 port->pp.curr = pp->curr;
2176 port->pp.advertised = pp->advertised;
2177 port->pp.supported = pp->supported;
2178 port->pp.peer = pp->peer;
2179 port->pp.curr_speed = pp->curr_speed;
2180 port->pp.max_speed = pp->max_speed;
2182 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
2185 /* Update OpenFlow 'state' in 'port' and notify controller. */
2187 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
2189 if (port->pp.state != state) {
2190 port->pp.state = state;
2191 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
2197 ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port)
2199 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
2201 if (port->ofproto->ofproto_class->set_realdev) {
2202 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
2204 if (port->ofproto->ofproto_class->set_stp_port) {
2205 port->ofproto->ofproto_class->set_stp_port(port, NULL);
2207 if (port->ofproto->ofproto_class->set_cfm) {
2208 port->ofproto->ofproto_class->set_cfm(port, NULL);
2210 if (port->ofproto->ofproto_class->bundle_remove) {
2211 port->ofproto->ofproto_class->bundle_remove(port);
2217 ofport_destroy__(struct ofport *port)
2219 struct ofproto *ofproto = port->ofproto;
2220 const char *name = netdev_get_name(port->netdev);
2222 hmap_remove(&ofproto->ports, &port->hmap_node);
2223 shash_delete(&ofproto->port_by_name,
2224 shash_find(&ofproto->port_by_name, name));
2226 netdev_close(port->netdev);
2227 ofproto->ofproto_class->port_dealloc(port);
2231 ofport_destroy(struct ofport *port)
2234 dealloc_ofp_port(port->ofproto, port->ofp_port);
2235 port->ofproto->ofproto_class->port_destruct(port);
2236 ofport_destroy__(port);
2241 ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
2243 struct ofport *port;
2245 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_ofp_port(ofp_port),
2247 if (port->ofp_port == ofp_port) {
2254 static long long int
2255 ofport_get_usage(const struct ofproto *ofproto, ofp_port_t ofp_port)
2257 struct ofport_usage *usage;
2259 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2260 &ofproto->ofport_usage) {
2261 if (usage->ofp_port == ofp_port) {
2262 return usage->last_used;
2269 ofport_set_usage(struct ofproto *ofproto, ofp_port_t ofp_port,
2270 long long int last_used)
2272 struct ofport_usage *usage;
2273 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2274 &ofproto->ofport_usage) {
2275 if (usage->ofp_port == ofp_port) {
2276 usage->last_used = last_used;
2280 ovs_assert(last_used == LLONG_MAX);
2282 usage = xmalloc(sizeof *usage);
2283 usage->ofp_port = ofp_port;
2284 usage->last_used = last_used;
2285 hmap_insert(&ofproto->ofport_usage, &usage->hmap_node,
2286 hash_ofp_port(ofp_port));
2290 ofport_remove_usage(struct ofproto *ofproto, ofp_port_t ofp_port)
2292 struct ofport_usage *usage;
2293 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2294 &ofproto->ofport_usage) {
2295 if (usage->ofp_port == ofp_port) {
2296 hmap_remove(&ofproto->ofport_usage, &usage->hmap_node);
2304 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
2306 struct ofproto *ofproto = port->ofproto;
2309 if (ofproto->ofproto_class->port_get_stats) {
2310 error = ofproto->ofproto_class->port_get_stats(port, stats);
2319 update_port(struct ofproto *ofproto, const char *name)
2321 struct ofproto_port ofproto_port;
2322 struct ofputil_phy_port pp;
2323 struct netdev *netdev;
2324 struct ofport *port;
2326 COVERAGE_INC(ofproto_update_port);
2328 /* Fetch 'name''s location and properties from the datapath. */
2329 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
2330 ? ofport_open(ofproto, &ofproto_port, &pp)
2334 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
2335 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
2336 struct netdev *old_netdev = port->netdev;
2338 /* 'name' hasn't changed location. Any properties changed? */
2339 if (!ofport_equal(&port->pp, &pp)) {
2340 ofport_modified(port, &pp);
2343 update_mtu(ofproto, port);
2345 /* Install the newly opened netdev in case it has changed.
2346 * Don't close the old netdev yet in case port_modified has to
2347 * remove a retained reference to it.*/
2348 port->netdev = netdev;
2349 port->change_seq = netdev_change_seq(netdev);
2351 if (port->ofproto->ofproto_class->port_modified) {
2352 port->ofproto->ofproto_class->port_modified(port);
2355 netdev_close(old_netdev);
2357 /* If 'port' is nonnull then its name differs from 'name' and thus
2358 * we should delete it. If we think there's a port named 'name'
2359 * then its port number must be wrong now so delete it too. */
2361 ofport_remove(port);
2363 ofport_remove_with_name(ofproto, name);
2364 ofport_install(ofproto, netdev, &pp);
2367 /* Any port named 'name' is gone now. */
2368 ofport_remove_with_name(ofproto, name);
2370 ofproto_port_destroy(&ofproto_port);
2374 init_ports(struct ofproto *p)
2376 struct ofproto_port_dump dump;
2377 struct ofproto_port ofproto_port;
2378 struct shash_node *node, *next;
2380 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2381 const char *name = ofproto_port.name;
2383 if (shash_find(&p->port_by_name, name)) {
2384 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2387 struct ofputil_phy_port pp;
2388 struct netdev *netdev;
2390 /* Check if an OpenFlow port number had been requested. */
2391 node = shash_find(&init_ofp_ports, name);
2393 const struct iface_hint *iface_hint = node->data;
2394 simap_put(&p->ofp_requests, name,
2395 ofp_to_u16(iface_hint->ofp_port));
2398 netdev = ofport_open(p, &ofproto_port, &pp);
2400 ofport_install(p, netdev, &pp);
2401 if (ofp_to_u16(ofproto_port.ofp_port) < p->max_ports) {
2402 p->alloc_port_no = MAX(p->alloc_port_no,
2403 ofp_to_u16(ofproto_port.ofp_port));
2409 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2410 struct iface_hint *iface_hint = node->data;
2412 if (!strcmp(iface_hint->br_name, p->name)) {
2413 free(iface_hint->br_name);
2414 free(iface_hint->br_type);
2416 shash_delete(&init_ofp_ports, node);
2423 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2424 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2426 find_min_mtu(struct ofproto *p)
2428 struct ofport *ofport;
2431 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2432 struct netdev *netdev = ofport->netdev;
2435 /* Skip any internal ports, since that's what we're trying to
2437 if (!strcmp(netdev_get_type(netdev), "internal")) {
2441 if (netdev_get_mtu(netdev, &dev_mtu)) {
2444 if (!mtu || dev_mtu < mtu) {
2449 return mtu ? mtu: ETH_PAYLOAD_MAX;
2452 /* Update MTU of all datapath devices on 'p' to the minimum of the
2453 * non-datapath ports in event of 'port' added or changed. */
2455 update_mtu(struct ofproto *p, struct ofport *port)
2457 struct ofport *ofport;
2458 struct netdev *netdev = port->netdev;
2459 int dev_mtu, old_min;
2461 if (netdev_get_mtu(netdev, &dev_mtu)) {
2465 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2466 if (dev_mtu > p->min_mtu) {
2467 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2468 dev_mtu = p->min_mtu;
2471 port->mtu = dev_mtu;
2475 /* For non-internal port find new min mtu. */
2476 old_min = p->min_mtu;
2477 port->mtu = dev_mtu;
2478 p->min_mtu = find_min_mtu(p);
2479 if (p->min_mtu == old_min) {
2483 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2484 struct netdev *netdev = ofport->netdev;
2486 if (!strcmp(netdev_get_type(netdev), "internal")) {
2487 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2488 ofport->mtu = p->min_mtu;
2495 ofproto_rule_ref(struct rule *rule)
2500 atomic_add(&rule->ref_count, 1, &orig);
2501 ovs_assert(orig != 0);
2506 ofproto_rule_unref(struct rule *rule)
2511 atomic_sub(&rule->ref_count, 1, &orig);
2513 rule->ofproto->ofproto_class->rule_destruct(rule);
2514 ofproto_rule_destroy__(rule);
2516 ovs_assert(orig != 0);
2521 struct rule_actions *
2522 rule_get_actions(const struct rule *rule)
2523 OVS_EXCLUDED(rule->mutex)
2525 struct rule_actions *actions;
2527 ovs_mutex_lock(&rule->mutex);
2528 actions = rule_get_actions__(rule);
2529 ovs_mutex_unlock(&rule->mutex);
2534 struct rule_actions *
2535 rule_get_actions__(const struct rule *rule)
2536 OVS_REQUIRES(rule->mutex)
2538 rule_actions_ref(rule->actions);
2539 return rule->actions;
2543 ofproto_rule_destroy__(struct rule *rule)
2544 OVS_NO_THREAD_SAFETY_ANALYSIS
2546 cls_rule_destroy(CONST_CAST(struct cls_rule *, &rule->cr));
2547 rule_actions_unref(rule->actions);
2548 ovs_mutex_destroy(&rule->mutex);
2549 rule->ofproto->ofproto_class->rule_dealloc(rule);
2552 static uint32_t get_provider_meter_id(const struct ofproto *,
2553 uint32_t of_meter_id);
2555 /* Creates and returns a new 'struct rule_actions', with a ref_count of 1,
2556 * whose actions are a copy of from the 'ofpacts_len' bytes of 'ofpacts'. */
2557 struct rule_actions *
2558 rule_actions_create(const struct ofproto *ofproto,
2559 const struct ofpact *ofpacts, size_t ofpacts_len)
2561 struct rule_actions *actions;
2563 actions = xmalloc(sizeof *actions);
2564 atomic_init(&actions->ref_count, 1);
2565 actions->ofpacts = xmemdup(ofpacts, ofpacts_len);
2566 actions->ofpacts_len = ofpacts_len;
2567 actions->provider_meter_id
2568 = get_provider_meter_id(ofproto,
2569 ofpacts_get_meter(ofpacts, ofpacts_len));
2574 /* Increments 'actions''s ref_count. */
2576 rule_actions_ref(struct rule_actions *actions)
2581 atomic_add(&actions->ref_count, 1, &orig);
2582 ovs_assert(orig != 0);
2586 /* Decrements 'actions''s ref_count and frees 'actions' if the ref_count
2589 rule_actions_unref(struct rule_actions *actions)
2594 atomic_sub(&actions->ref_count, 1, &orig);
2596 free(actions->ofpacts);
2599 ovs_assert(orig != 0);
2604 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2605 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2607 ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port)
2608 OVS_REQUIRES(ofproto_mutex)
2610 return (port == OFPP_ANY
2611 || ofpacts_output_to_port(rule->actions->ofpacts,
2612 rule->actions->ofpacts_len, port));
2615 /* Returns true if 'rule' has group and equals group_id. */
2617 ofproto_rule_has_out_group(const struct rule *rule, uint32_t group_id)
2618 OVS_REQUIRES(ofproto_mutex)
2620 return (group_id == OFPG11_ANY
2621 || ofpacts_output_to_group(rule->actions->ofpacts,
2622 rule->actions->ofpacts_len, group_id));
2625 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2626 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2628 ofoperation_has_out_port(const struct ofoperation *op, ofp_port_t out_port)
2629 OVS_REQUIRES(ofproto_mutex)
2631 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2636 case OFOPERATION_ADD:
2637 case OFOPERATION_DELETE:
2640 case OFOPERATION_MODIFY:
2641 case OFOPERATION_REPLACE:
2642 return ofpacts_output_to_port(op->actions->ofpacts,
2643 op->actions->ofpacts_len, out_port);
2650 rule_execute_destroy(struct rule_execute *e)
2652 ofproto_rule_unref(e->rule);
2653 list_remove(&e->list_node);
2657 /* Executes all "rule_execute" operations queued up in ofproto->rule_executes,
2658 * by passing them to the ofproto provider. */
2660 run_rule_executes(struct ofproto *ofproto)
2661 OVS_EXCLUDED(ofproto_mutex)
2663 struct rule_execute *e, *next;
2664 struct list executes;
2666 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2667 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2668 union flow_in_port in_port_;
2671 in_port_.ofp_port = e->in_port;
2672 flow_extract(e->packet, 0, 0, NULL, &in_port_, &flow);
2673 ofproto->ofproto_class->rule_execute(e->rule, &flow, e->packet);
2675 rule_execute_destroy(e);
2679 /* Destroys and discards all "rule_execute" operations queued up in
2680 * ofproto->rule_executes. */
2682 destroy_rule_executes(struct ofproto *ofproto)
2684 struct rule_execute *e, *next;
2685 struct list executes;
2687 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2688 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2689 ofpbuf_delete(e->packet);
2690 rule_execute_destroy(e);
2694 /* Returns true if 'rule' should be hidden from the controller.
2696 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2697 * (e.g. by in-band control) and are intentionally hidden from the
2700 ofproto_rule_is_hidden(const struct rule *rule)
2702 return rule->cr.priority > UINT16_MAX;
2705 static enum oftable_flags
2706 rule_get_flags(const struct rule *rule)
2708 return rule->ofproto->tables[rule->table_id].flags;
2712 rule_is_modifiable(const struct rule *rule)
2714 return !(rule_get_flags(rule) & OFTABLE_READONLY);
2718 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2720 ofconn_send_reply(ofconn, make_echo_reply(oh));
2725 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2727 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2728 struct ofputil_switch_features features;
2729 struct ofport *port;
2735 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2737 ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2739 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2740 * if present, are always at the end.) */
2741 n_tables = ofproto->n_tables;
2742 for (i = 0; i < ofproto->n_tables; i++) {
2743 if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
2749 features.datapath_id = ofproto->datapath_id;
2750 features.n_buffers = pktbuf_capacity();
2751 features.n_tables = n_tables;
2752 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2753 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2755 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2757 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2758 features.auxiliary_id = 0;
2759 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2761 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2762 ofputil_put_switch_features_port(&port->pp, b);
2765 ofconn_send_reply(ofconn, b);
2770 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2772 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2773 struct ofp_switch_config *osc;
2774 enum ofp_config_flags flags;
2778 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2779 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2780 flags = ofproto->frag_handling;
2781 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2782 if (oh->version < OFP13_VERSION
2783 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2784 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2786 osc->flags = htons(flags);
2787 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2788 ofconn_send_reply(ofconn, buf);
2794 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2796 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2797 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2798 uint16_t flags = ntohs(osc->flags);
2800 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2801 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
2802 enum ofp_config_flags cur = ofproto->frag_handling;
2803 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2805 ovs_assert((cur & OFPC_FRAG_MASK) == cur);
2807 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2808 ofproto->frag_handling = next;
2810 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2812 ofputil_frag_handling_to_string(next));
2816 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2817 ofconn_set_invalid_ttl_to_controller(ofconn,
2818 (oh->version < OFP13_VERSION
2819 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2821 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2826 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2827 * error message code for the caller to propagate upward. Otherwise, returns
2830 * The log message mentions 'msg_type'. */
2832 reject_slave_controller(struct ofconn *ofconn)
2834 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2835 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
2836 return OFPERR_OFPBRC_EPERM;
2842 /* Checks that the 'ofpacts_len' bytes of action in 'ofpacts' are appropriate
2845 * - If they use a meter, then 'ofproto' has that meter configured.
2847 * - If they use any groups, then 'ofproto' has that group configured.
2849 * Returns 0 if successful, otherwise an OpenFlow error. */
2851 ofproto_check_ofpacts(struct ofproto *ofproto,
2852 const struct ofpact ofpacts[], size_t ofpacts_len)
2854 const struct ofpact *a;
2857 mid = ofpacts_get_meter(ofpacts, ofpacts_len);
2858 if (mid && get_provider_meter_id(ofproto, mid) == UINT32_MAX) {
2859 return OFPERR_OFPMMFC_INVALID_METER;
2862 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2863 if (a->type == OFPACT_GROUP
2864 && !ofproto_group_exists(ofproto, ofpact_get_GROUP(a)->group_id)) {
2865 return OFPERR_OFPBAC_BAD_OUT_GROUP;
2873 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2875 struct ofproto *p = ofconn_get_ofproto(ofconn);
2876 struct ofputil_packet_out po;
2877 struct ofpbuf *payload;
2878 uint64_t ofpacts_stub[1024 / 8];
2879 struct ofpbuf ofpacts;
2881 union flow_in_port in_port_;
2884 COVERAGE_INC(ofproto_packet_out);
2886 error = reject_slave_controller(ofconn);
2891 /* Decode message. */
2892 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2893 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2895 goto exit_free_ofpacts;
2897 if (ofp_to_u16(po.in_port) >= p->max_ports
2898 && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) {
2899 error = OFPERR_OFPBRC_BAD_PORT;
2900 goto exit_free_ofpacts;
2905 if (po.buffer_id != UINT32_MAX) {
2906 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2907 if (error || !payload) {
2908 goto exit_free_ofpacts;
2911 /* Ensure that the L3 header is 32-bit aligned. */
2912 payload = ofpbuf_clone_data_with_headroom(po.packet, po.packet_len, 2);
2915 /* Verify actions against packet, then send packet if successful. */
2916 in_port_.ofp_port = po.in_port;
2917 flow_extract(payload, 0, 0, NULL, &in_port_, &flow);
2918 error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len);
2920 error = p->ofproto_class->packet_out(p, payload, &flow,
2921 po.ofpacts, po.ofpacts_len);
2923 ofpbuf_delete(payload);
2926 ofpbuf_uninit(&ofpacts);
2932 update_port_config(struct ofport *port,
2933 enum ofputil_port_config config,
2934 enum ofputil_port_config mask)
2936 enum ofputil_port_config old_config = port->pp.config;
2937 enum ofputil_port_config toggle;
2939 toggle = (config ^ port->pp.config) & mask;
2940 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2941 if (config & OFPUTIL_PC_PORT_DOWN) {
2942 netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL);
2944 netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL);
2946 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2949 port->pp.config ^= toggle;
2950 if (port->pp.config != old_config) {
2951 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2956 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2958 struct ofproto *p = ofconn_get_ofproto(ofconn);
2959 struct ofputil_port_mod pm;
2960 struct ofport *port;
2963 error = reject_slave_controller(ofconn);
2968 error = ofputil_decode_port_mod(oh, &pm);
2973 port = ofproto_get_port(p, pm.port_no);
2975 return OFPERR_OFPPMFC_BAD_PORT;
2976 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2977 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2979 update_port_config(port, pm.config, pm.mask);
2981 netdev_set_advertisements(port->netdev, pm.advertise);
2988 handle_desc_stats_request(struct ofconn *ofconn,
2989 const struct ofp_header *request)
2991 static const char *default_mfr_desc = "Nicira, Inc.";
2992 static const char *default_hw_desc = "Open vSwitch";
2993 static const char *default_sw_desc = VERSION;
2994 static const char *default_serial_desc = "None";
2995 static const char *default_dp_desc = "None";
2997 struct ofproto *p = ofconn_get_ofproto(ofconn);
2998 struct ofp_desc_stats *ods;
3001 msg = ofpraw_alloc_stats_reply(request, 0);
3002 ods = ofpbuf_put_zeros(msg, sizeof *ods);
3003 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
3004 sizeof ods->mfr_desc);
3005 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
3006 sizeof ods->hw_desc);
3007 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
3008 sizeof ods->sw_desc);
3009 ovs_strlcpy(ods->serial_num,
3010 p->serial_desc ? p->serial_desc : default_serial_desc,
3011 sizeof ods->serial_num);
3012 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
3013 sizeof ods->dp_desc);
3014 ofconn_send_reply(ofconn, msg);
3020 handle_table_stats_request(struct ofconn *ofconn,
3021 const struct ofp_header *request)
3023 struct ofproto *p = ofconn_get_ofproto(ofconn);
3024 struct ofp12_table_stats *ots;
3029 /* Set up default values.
3031 * ofp12_table_stats is used as a generic structure as
3032 * it is able to hold all the fields for ofp10_table_stats
3033 * and ofp11_table_stats (and of course itself).
3035 ots = xcalloc(p->n_tables, sizeof *ots);
3036 for (i = 0; i < p->n_tables; i++) {
3037 ots[i].table_id = i;
3038 sprintf(ots[i].name, "table%zu", i);
3039 ots[i].match = htonll(OFPXMT13_MASK);
3040 ots[i].wildcards = htonll(OFPXMT13_MASK);
3041 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
3042 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
3043 ots[i].write_setfields = htonll(OFPXMT13_MASK);
3044 ots[i].apply_setfields = htonll(OFPXMT13_MASK);
3045 ots[i].metadata_match = OVS_BE64_MAX;
3046 ots[i].metadata_write = OVS_BE64_MAX;
3047 ots[i].instructions = htonl(OFPIT11_ALL);
3048 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
3049 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
3050 ovs_rwlock_rdlock(&p->tables[i].cls.rwlock);
3051 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
3052 ovs_rwlock_unlock(&p->tables[i].cls.rwlock);
3055 p->ofproto_class->get_tables(p, ots);
3057 /* Post-process the tables, dropping hidden tables. */
3058 n_tables = p->n_tables;
3059 for (i = 0; i < p->n_tables; i++) {
3060 const struct oftable *table = &p->tables[i];
3062 if (table->flags & OFTABLE_HIDDEN) {
3068 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
3071 if (table->max_flows < ntohl(ots[i].max_entries)) {
3072 ots[i].max_entries = htonl(table->max_flows);
3076 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
3077 ofconn_send_reply(ofconn, msg);
3085 append_port_stat(struct ofport *port, struct list *replies)
3087 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
3089 calc_duration(port->created, time_msec(),
3090 &ops.duration_sec, &ops.duration_nsec);
3092 /* Intentionally ignore return value, since errors will set
3093 * 'stats' to all-1s, which is correct for OpenFlow, and
3094 * netdev_get_stats() will log errors. */
3095 ofproto_port_get_stats(port, &ops.stats);
3097 ofputil_append_port_stat(replies, &ops);
3101 handle_port_stats_request(struct ofconn *ofconn,
3102 const struct ofp_header *request)
3104 struct ofproto *p = ofconn_get_ofproto(ofconn);
3105 struct ofport *port;
3106 struct list replies;
3110 error = ofputil_decode_port_stats_request(request, &port_no);
3115 ofpmp_init(&replies, request);
3116 if (port_no != OFPP_ANY) {
3117 port = ofproto_get_port(p, port_no);
3119 append_port_stat(port, &replies);
3122 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3123 append_port_stat(port, &replies);
3127 ofconn_send_replies(ofconn, &replies);
3132 handle_port_desc_stats_request(struct ofconn *ofconn,
3133 const struct ofp_header *request)
3135 struct ofproto *p = ofconn_get_ofproto(ofconn);
3136 enum ofp_version version;
3137 struct ofport *port;
3138 struct list replies;
3140 ofpmp_init(&replies, request);
3142 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
3143 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3144 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
3147 ofconn_send_replies(ofconn, &replies);
3152 hash_cookie(ovs_be64 cookie)
3154 return hash_2words((OVS_FORCE uint64_t)cookie >> 32,
3155 (OVS_FORCE uint64_t)cookie);
3159 cookies_insert(struct ofproto *ofproto, struct rule *rule)
3160 OVS_REQUIRES(ofproto_mutex)
3162 hindex_insert(&ofproto->cookies, &rule->cookie_node,
3163 hash_cookie(rule->flow_cookie));
3167 cookies_remove(struct ofproto *ofproto, struct rule *rule)
3168 OVS_REQUIRES(ofproto_mutex)
3170 hindex_remove(&ofproto->cookies, &rule->cookie_node);
3174 ofproto_rule_change_cookie(struct ofproto *ofproto, struct rule *rule,
3175 ovs_be64 new_cookie)
3176 OVS_REQUIRES(ofproto_mutex)
3178 if (new_cookie != rule->flow_cookie) {
3179 cookies_remove(ofproto, rule);
3181 ovs_mutex_lock(&rule->mutex);
3182 rule->flow_cookie = new_cookie;
3183 ovs_mutex_unlock(&rule->mutex);
3185 cookies_insert(ofproto, rule);
3190 calc_duration(long long int start, long long int now,
3191 uint32_t *sec, uint32_t *nsec)
3193 long long int msecs = now - start;
3194 *sec = msecs / 1000;
3195 *nsec = (msecs % 1000) * (1000 * 1000);
3198 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
3199 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
3201 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
3203 return (table_id == 0xff || table_id < ofproto->n_tables
3205 : OFPERR_OFPBRC_BAD_TABLE_ID);
3209 static struct oftable *
3210 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
3212 struct oftable *table;
3214 for (table = &ofproto->tables[table_id];
3215 table < &ofproto->tables[ofproto->n_tables];
3217 if (!(table->flags & OFTABLE_HIDDEN)) {
3225 static struct oftable *
3226 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
3228 if (table_id == 0xff) {
3229 return next_visible_table(ofproto, 0);
3230 } else if (table_id < ofproto->n_tables) {
3231 return &ofproto->tables[table_id];
3237 static struct oftable *
3238 next_matching_table(const struct ofproto *ofproto,
3239 const struct oftable *table, uint8_t table_id)
3241 return (table_id == 0xff
3242 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
3246 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
3248 * - If TABLE_ID is 0xff, this iterates over every classifier table in
3249 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
3251 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
3252 * only once, for that table. (This can be used to access tables marked
3255 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
3256 * entered at all. (Perhaps you should have validated TABLE_ID with
3257 * check_table_id().)
3259 * All parameters are evaluated multiple times.
3261 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
3262 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
3264 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
3266 /* Initializes 'criteria' in a straightforward way based on the other
3269 * For "loose" matching, the 'priority' parameter is unimportant and may be
3272 rule_criteria_init(struct rule_criteria *criteria, uint8_t table_id,
3273 const struct match *match, unsigned int priority,
3274 ovs_be64 cookie, ovs_be64 cookie_mask,
3275 ofp_port_t out_port, uint32_t out_group)
3277 criteria->table_id = table_id;
3278 cls_rule_init(&criteria->cr, match, priority);
3279 criteria->cookie = cookie;
3280 criteria->cookie_mask = cookie_mask;
3281 criteria->out_port = out_port;
3282 criteria->out_group = out_group;
3286 rule_criteria_destroy(struct rule_criteria *criteria)
3288 cls_rule_destroy(&criteria->cr);
3292 rule_collection_init(struct rule_collection *rules)
3294 rules->rules = rules->stub;
3296 rules->capacity = ARRAY_SIZE(rules->stub);
3300 rule_collection_add(struct rule_collection *rules, struct rule *rule)
3302 if (rules->n >= rules->capacity) {
3303 size_t old_size, new_size;
3305 old_size = rules->capacity * sizeof *rules->rules;
3306 rules->capacity *= 2;
3307 new_size = rules->capacity * sizeof *rules->rules;
3309 if (rules->rules == rules->stub) {
3310 rules->rules = xmalloc(new_size);
3311 memcpy(rules->rules, rules->stub, old_size);
3313 rules->rules = xrealloc(rules->rules, new_size);
3317 rules->rules[rules->n++] = rule;
3321 rule_collection_ref(struct rule_collection *rules)
3322 OVS_REQUIRES(ofproto_mutex)
3326 for (i = 0; i < rules->n; i++) {
3327 ofproto_rule_ref(rules->rules[i]);
3332 rule_collection_unref(struct rule_collection *rules)
3336 for (i = 0; i < rules->n; i++) {
3337 ofproto_rule_unref(rules->rules[i]);
3342 rule_collection_destroy(struct rule_collection *rules)
3344 if (rules->rules != rules->stub) {
3350 collect_rule(struct rule *rule, const struct rule_criteria *c,
3351 struct rule_collection *rules)
3352 OVS_REQUIRES(ofproto_mutex)
3354 /* We ordinarily want to skip hidden rules, but there has to be a way for
3355 * code internal to OVS to modify and delete them, so if the criteria
3356 * specify a priority that can only be for a hidden flow, then allow hidden
3357 * rules to be selected. (This doesn't allow OpenFlow clients to meddle
3358 * with hidden flows because OpenFlow uses only a 16-bit field to specify
3360 if (ofproto_rule_is_hidden(rule) && c->cr.priority <= UINT16_MAX) {
3362 } else if (rule->pending) {
3363 return OFPROTO_POSTPONE;
3365 if ((c->table_id == rule->table_id || c->table_id == 0xff)
3366 && ofproto_rule_has_out_port(rule, c->out_port)
3367 && ofproto_rule_has_out_group(rule, c->out_group)
3368 && !((rule->flow_cookie ^ c->cookie) & c->cookie_mask)) {
3369 rule_collection_add(rules, rule);
3375 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3376 * on classifiers rules are done in the "loose" way required for OpenFlow
3377 * OFPFC_MODIFY and OFPFC_DELETE requests. Puts the selected rules on list
3380 * Hidden rules are always omitted.
3382 * Returns 0 on success, otherwise an OpenFlow error code. */
3384 collect_rules_loose(struct ofproto *ofproto,
3385 const struct rule_criteria *criteria,
3386 struct rule_collection *rules)
3387 OVS_REQUIRES(ofproto_mutex)
3389 struct oftable *table;
3392 rule_collection_init(rules);
3394 error = check_table_id(ofproto, criteria->table_id);
3399 if (criteria->cookie_mask == OVS_BE64_MAX) {
3402 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
3403 hash_cookie(criteria->cookie),
3404 &ofproto->cookies) {
3405 if (cls_rule_is_loose_match(&rule->cr, &criteria->cr.match)) {
3406 error = collect_rule(rule, criteria, rules);
3413 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
3414 struct cls_cursor cursor;
3417 ovs_rwlock_rdlock(&table->cls.rwlock);
3418 cls_cursor_init(&cursor, &table->cls, &criteria->cr);
3419 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3420 error = collect_rule(rule, criteria, rules);
3425 ovs_rwlock_unlock(&table->cls.rwlock);
3431 rule_collection_destroy(rules);
3436 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3437 * on classifiers rules are done in the "strict" way required for OpenFlow
3438 * OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests. Puts the selected
3439 * rules on list 'rules'.
3441 * Hidden rules are always omitted.
3443 * Returns 0 on success, otherwise an OpenFlow error code. */
3445 collect_rules_strict(struct ofproto *ofproto,
3446 const struct rule_criteria *criteria,
3447 struct rule_collection *rules)
3448 OVS_REQUIRES(ofproto_mutex)
3450 struct oftable *table;
3453 rule_collection_init(rules);
3455 error = check_table_id(ofproto, criteria->table_id);
3460 if (criteria->cookie_mask == OVS_BE64_MAX) {
3463 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
3464 hash_cookie(criteria->cookie),
3465 &ofproto->cookies) {
3466 if (cls_rule_equal(&rule->cr, &criteria->cr)) {
3467 error = collect_rule(rule, criteria, rules);
3474 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
3477 ovs_rwlock_rdlock(&table->cls.rwlock);
3478 rule = rule_from_cls_rule(classifier_find_rule_exactly(
3479 &table->cls, &criteria->cr));
3480 ovs_rwlock_unlock(&table->cls.rwlock);
3482 error = collect_rule(rule, criteria, rules);
3492 rule_collection_destroy(rules);
3497 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
3498 * forced into the range of a uint16_t. */
3500 age_secs(long long int age_ms)
3502 return (age_ms < 0 ? 0
3503 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
3504 : (unsigned int) age_ms / 1000);
3508 handle_flow_stats_request(struct ofconn *ofconn,
3509 const struct ofp_header *request)
3510 OVS_EXCLUDED(ofproto_mutex)
3512 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3513 struct ofputil_flow_stats_request fsr;
3514 struct rule_criteria criteria;
3515 struct rule_collection rules;
3516 struct list replies;
3520 error = ofputil_decode_flow_stats_request(&fsr, request);
3525 rule_criteria_init(&criteria, fsr.table_id, &fsr.match, 0, fsr.cookie,
3526 fsr.cookie_mask, fsr.out_port, fsr.out_group);
3528 ovs_mutex_lock(&ofproto_mutex);
3529 error = collect_rules_loose(ofproto, &criteria, &rules);
3530 rule_criteria_destroy(&criteria);
3532 rule_collection_ref(&rules);
3534 ovs_mutex_unlock(&ofproto_mutex);
3540 ofpmp_init(&replies, request);
3541 for (i = 0; i < rules.n; i++) {
3542 struct rule *rule = rules.rules[i];
3543 long long int now = time_msec();
3544 struct ofputil_flow_stats fs;
3545 long long int created, used, modified;
3546 struct rule_actions *actions;
3547 enum ofputil_flow_mod_flags flags;
3549 ovs_mutex_lock(&rule->mutex);
3550 fs.cookie = rule->flow_cookie;
3551 fs.idle_timeout = rule->idle_timeout;
3552 fs.hard_timeout = rule->hard_timeout;
3553 created = rule->created;
3555 modified = rule->modified;
3556 actions = rule_get_actions__(rule);
3557 flags = rule->flags;
3558 ovs_mutex_unlock(&rule->mutex);
3560 minimatch_expand(&rule->cr.match, &fs.match);
3561 fs.table_id = rule->table_id;
3562 calc_duration(created, now, &fs.duration_sec, &fs.duration_nsec);
3563 fs.priority = rule->cr.priority;
3564 fs.idle_age = age_secs(now - used);
3565 fs.hard_age = age_secs(now - modified);
3566 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
3568 fs.ofpacts = actions->ofpacts;
3569 fs.ofpacts_len = actions->ofpacts_len;
3572 ofputil_append_flow_stats_reply(&fs, &replies);
3574 rule_actions_unref(actions);
3577 rule_collection_unref(&rules);
3578 rule_collection_destroy(&rules);
3580 ofconn_send_replies(ofconn, &replies);
3586 flow_stats_ds(struct rule *rule, struct ds *results)
3588 uint64_t packet_count, byte_count;
3589 struct rule_actions *actions;
3590 long long int created;
3592 rule->ofproto->ofproto_class->rule_get_stats(rule,
3593 &packet_count, &byte_count);
3595 ovs_mutex_lock(&rule->mutex);
3596 actions = rule_get_actions__(rule);
3597 created = rule->created;
3598 ovs_mutex_unlock(&rule->mutex);
3600 if (rule->table_id != 0) {
3601 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
3603 ds_put_format(results, "duration=%llds, ", (time_msec() - created) / 1000);
3604 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3605 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3606 cls_rule_format(&rule->cr, results);
3607 ds_put_char(results, ',');
3609 ds_put_cstr(results, "actions=");
3610 ofpacts_format(actions->ofpacts, actions->ofpacts_len, results);
3612 ds_put_cstr(results, "\n");
3614 rule_actions_unref(actions);
3617 /* Adds a pretty-printed description of all flows to 'results', including
3618 * hidden flows (e.g., set up by in-band control). */
3620 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3622 struct oftable *table;
3624 OFPROTO_FOR_EACH_TABLE (table, p) {
3625 struct cls_cursor cursor;
3628 ovs_rwlock_rdlock(&table->cls.rwlock);
3629 cls_cursor_init(&cursor, &table->cls, NULL);
3630 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3631 flow_stats_ds(rule, results);
3633 ovs_rwlock_unlock(&table->cls.rwlock);
3637 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
3638 * '*engine_type' and '*engine_id', respectively. */
3640 ofproto_get_netflow_ids(const struct ofproto *ofproto,
3641 uint8_t *engine_type, uint8_t *engine_id)
3643 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
3646 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
3647 * true if the port's CFM status was successfully stored into '*status'.
3648 * Returns false if the port did not have CFM configured, in which case
3649 * '*status' is indeterminate.
3651 * The caller must provide and owns '*status', and must free 'status->rmps'. */
3653 ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port,
3654 struct ofproto_cfm_status *status)
3656 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
3658 && ofproto->ofproto_class->get_cfm_status
3659 && ofproto->ofproto_class->get_cfm_status(ofport, status));
3663 handle_aggregate_stats_request(struct ofconn *ofconn,
3664 const struct ofp_header *oh)
3665 OVS_EXCLUDED(ofproto_mutex)
3667 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3668 struct ofputil_flow_stats_request request;
3669 struct ofputil_aggregate_stats stats;
3670 bool unknown_packets, unknown_bytes;
3671 struct rule_criteria criteria;
3672 struct rule_collection rules;
3673 struct ofpbuf *reply;
3677 error = ofputil_decode_flow_stats_request(&request, oh);
3682 rule_criteria_init(&criteria, request.table_id, &request.match, 0,
3683 request.cookie, request.cookie_mask,
3684 request.out_port, request.out_group);
3686 ovs_mutex_lock(&ofproto_mutex);
3687 error = collect_rules_loose(ofproto, &criteria, &rules);
3688 rule_criteria_destroy(&criteria);
3690 rule_collection_ref(&rules);
3692 ovs_mutex_unlock(&ofproto_mutex);
3698 memset(&stats, 0, sizeof stats);
3699 unknown_packets = unknown_bytes = false;
3700 for (i = 0; i < rules.n; i++) {
3701 struct rule *rule = rules.rules[i];
3702 uint64_t packet_count;
3703 uint64_t byte_count;
3705 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
3708 if (packet_count == UINT64_MAX) {
3709 unknown_packets = true;
3711 stats.packet_count += packet_count;
3714 if (byte_count == UINT64_MAX) {
3715 unknown_bytes = true;
3717 stats.byte_count += byte_count;
3722 if (unknown_packets) {
3723 stats.packet_count = UINT64_MAX;
3725 if (unknown_bytes) {
3726 stats.byte_count = UINT64_MAX;
3729 rule_collection_unref(&rules);
3730 rule_collection_destroy(&rules);
3732 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
3733 ofconn_send_reply(ofconn, reply);
3738 struct queue_stats_cbdata {
3739 struct ofport *ofport;
3740 struct list replies;
3745 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3746 const struct netdev_queue_stats *stats)
3748 struct ofputil_queue_stats oqs;
3750 oqs.port_no = cbdata->ofport->pp.port_no;
3751 oqs.queue_id = queue_id;
3752 oqs.tx_bytes = stats->tx_bytes;
3753 oqs.tx_packets = stats->tx_packets;
3754 oqs.tx_errors = stats->tx_errors;
3755 if (stats->created != LLONG_MIN) {
3756 calc_duration(stats->created, cbdata->now,
3757 &oqs.duration_sec, &oqs.duration_nsec);
3759 oqs.duration_sec = oqs.duration_nsec = UINT32_MAX;
3761 ofputil_append_queue_stat(&cbdata->replies, &oqs);
3765 handle_queue_stats_dump_cb(uint32_t queue_id,
3766 struct netdev_queue_stats *stats,
3769 struct queue_stats_cbdata *cbdata = cbdata_;
3771 put_queue_stats(cbdata, queue_id, stats);
3775 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3776 struct queue_stats_cbdata *cbdata)
3778 cbdata->ofport = port;
3779 if (queue_id == OFPQ_ALL) {
3780 netdev_dump_queue_stats(port->netdev,
3781 handle_queue_stats_dump_cb, cbdata);
3783 struct netdev_queue_stats stats;
3785 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3786 put_queue_stats(cbdata, queue_id, &stats);
3788 return OFPERR_OFPQOFC_BAD_QUEUE;
3795 handle_queue_stats_request(struct ofconn *ofconn,
3796 const struct ofp_header *rq)
3798 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3799 struct queue_stats_cbdata cbdata;
3800 struct ofport *port;
3802 struct ofputil_queue_stats_request oqsr;
3804 COVERAGE_INC(ofproto_queue_req);
3806 ofpmp_init(&cbdata.replies, rq);
3807 cbdata.now = time_msec();
3809 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3814 if (oqsr.port_no == OFPP_ANY) {
3815 error = OFPERR_OFPQOFC_BAD_QUEUE;
3816 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3817 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3822 port = ofproto_get_port(ofproto, oqsr.port_no);
3824 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3825 : OFPERR_OFPQOFC_BAD_PORT);
3828 ofconn_send_replies(ofconn, &cbdata.replies);
3830 ofpbuf_list_delete(&cbdata.replies);
3837 is_flow_deletion_pending(const struct ofproto *ofproto,
3838 const struct cls_rule *cls_rule,
3840 OVS_REQUIRES(ofproto_mutex)
3842 if (!hmap_is_empty(&ofproto->deletions)) {
3843 struct ofoperation *op;
3845 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3846 cls_rule_hash(cls_rule, table_id),
3847 &ofproto->deletions) {
3848 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
3858 should_evict_a_rule(struct oftable *table, unsigned int extra_space)
3859 OVS_REQUIRES(ofproto_mutex)
3860 OVS_NO_THREAD_SAFETY_ANALYSIS
3862 return classifier_count(&table->cls) + extra_space > table->max_flows;
3866 evict_rules_from_table(struct ofproto *ofproto, struct oftable *table,
3867 unsigned int extra_space)
3868 OVS_REQUIRES(ofproto_mutex)
3870 while (should_evict_a_rule(table, extra_space)) {
3873 if (!choose_rule_to_evict(table, &rule)) {
3874 return OFPERR_OFPFMFC_TABLE_FULL;
3875 } else if (rule->pending) {
3876 return OFPROTO_POSTPONE;
3878 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
3879 delete_flow__(rule, group, OFPRR_EVICTION);
3880 ofopgroup_submit(group);
3887 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3888 * in which no matching flow already exists in the flow table.
3890 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3891 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3892 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3893 * initiated now but may be retried later.
3895 * The caller retains ownership of 'fm->ofpacts'.
3897 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3900 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
3901 struct ofputil_flow_mod *fm, const struct ofp_header *request)
3902 OVS_REQUIRES(ofproto_mutex)
3904 struct oftable *table;
3905 struct ofopgroup *group;
3911 error = check_table_id(ofproto, fm->table_id);
3917 if (fm->table_id == 0xff) {
3918 if (ofproto->ofproto_class->rule_choose_table) {
3919 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3925 ovs_assert(table_id < ofproto->n_tables);
3929 } else if (fm->table_id < ofproto->n_tables) {
3930 table_id = fm->table_id;
3932 return OFPERR_OFPBRC_BAD_TABLE_ID;
3935 table = &ofproto->tables[table_id];
3937 if (table->flags & OFTABLE_READONLY) {
3938 return OFPERR_OFPBRC_EPERM;
3941 cls_rule_init(&cr, &fm->match, fm->priority);
3943 /* Transform "add" into "modify" if there's an existing identical flow. */
3944 ovs_rwlock_rdlock(&table->cls.rwlock);
3945 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr));
3946 ovs_rwlock_unlock(&table->cls.rwlock);
3948 cls_rule_destroy(&cr);
3949 if (!rule_is_modifiable(rule)) {
3950 return OFPERR_OFPBRC_EPERM;
3951 } else if (rule->pending) {
3952 return OFPROTO_POSTPONE;
3954 struct rule_collection rules;
3956 rule_collection_init(&rules);
3957 rule_collection_add(&rules, rule);
3958 fm->modify_cookie = true;
3959 error = modify_flows__(ofproto, ofconn, fm, request, &rules);
3960 rule_collection_destroy(&rules);
3966 /* Serialize against pending deletion. */
3967 if (is_flow_deletion_pending(ofproto, &cr, table_id)) {
3968 cls_rule_destroy(&cr);
3969 return OFPROTO_POSTPONE;
3972 /* Check for overlap, if requested. */
3973 if (fm->flags & OFPUTIL_FF_CHECK_OVERLAP) {
3976 ovs_rwlock_rdlock(&table->cls.rwlock);
3977 overlaps = classifier_rule_overlaps(&table->cls, &cr);
3978 ovs_rwlock_unlock(&table->cls.rwlock);
3981 cls_rule_destroy(&cr);
3982 return OFPERR_OFPFMFC_OVERLAP;
3986 /* If necessary, evict an existing rule to clear out space. */
3987 error = evict_rules_from_table(ofproto, table, 1);
3989 cls_rule_destroy(&cr);
3993 /* Allocate new rule. */
3994 rule = ofproto->ofproto_class->rule_alloc();
3996 cls_rule_destroy(&cr);
3997 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
3998 ofproto->name, ovs_strerror(error));
4002 /* Initialize base state. */
4003 *CONST_CAST(struct ofproto **, &rule->ofproto) = ofproto;
4004 cls_rule_move(CONST_CAST(struct cls_rule *, &rule->cr), &cr);
4005 atomic_init(&rule->ref_count, 1);
4006 rule->pending = NULL;
4007 rule->flow_cookie = fm->new_cookie;
4008 rule->created = rule->modified = rule->used = time_msec();
4010 ovs_mutex_init(&rule->mutex);
4011 ovs_mutex_lock(&rule->mutex);
4012 rule->idle_timeout = fm->idle_timeout;
4013 rule->hard_timeout = fm->hard_timeout;
4014 ovs_mutex_unlock(&rule->mutex);
4016 *CONST_CAST(uint8_t *, &rule->table_id) = table - ofproto->tables;
4017 rule->flags = fm->flags & OFPUTIL_FF_STATE;
4018 rule->actions = rule_actions_create(ofproto, fm->ofpacts, fm->ofpacts_len);
4019 list_init(&rule->meter_list_node);
4020 rule->eviction_group = NULL;
4021 list_init(&rule->expirable);
4022 rule->monitor_flags = 0;
4023 rule->add_seqno = 0;
4024 rule->modify_seqno = 0;
4026 /* Construct rule, initializing derived state. */
4027 error = ofproto->ofproto_class->rule_construct(rule);
4029 ofproto_rule_destroy__(rule);
4034 oftable_insert_rule(rule);
4036 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
4037 ofoperation_create(group, rule, OFOPERATION_ADD, 0);
4038 ofproto->ofproto_class->rule_insert(rule);
4039 ofopgroup_submit(group);
4044 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
4046 /* Modifies the rules listed in 'rules', changing their actions to match those
4049 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4052 * Returns 0 on success, otherwise an OpenFlow error code. */
4054 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
4055 struct ofputil_flow_mod *fm, const struct ofp_header *request,
4056 const struct rule_collection *rules)
4057 OVS_REQUIRES(ofproto_mutex)
4059 enum ofoperation_type type;
4060 struct ofopgroup *group;
4064 type = fm->command == OFPFC_ADD ? OFOPERATION_REPLACE : OFOPERATION_MODIFY;
4065 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
4066 error = OFPERR_OFPBRC_EPERM;
4067 for (i = 0; i < rules->n; i++) {
4068 struct rule *rule = rules->rules[i];
4069 struct ofoperation *op;
4070 bool actions_changed;
4071 bool reset_counters;
4073 /* FIXME: Implement OFPFUTIL_FF_RESET_COUNTS */
4075 if (rule_is_modifiable(rule)) {
4076 /* At least one rule is modifiable, don't report EPERM error. */
4082 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
4083 rule->actions->ofpacts,
4084 rule->actions->ofpacts_len);
4086 op = ofoperation_create(group, rule, type, 0);
4088 if (fm->modify_cookie && fm->new_cookie != OVS_BE64_MAX) {
4089 ofproto_rule_change_cookie(ofproto, rule, fm->new_cookie);
4091 if (type == OFOPERATION_REPLACE) {
4092 ovs_mutex_lock(&rule->mutex);
4093 rule->idle_timeout = fm->idle_timeout;
4094 rule->hard_timeout = fm->hard_timeout;
4095 ovs_mutex_unlock(&rule->mutex);
4097 rule->flags = fm->flags & OFPUTIL_FF_STATE;
4098 if (fm->idle_timeout || fm->hard_timeout) {
4099 if (!rule->eviction_group) {
4100 eviction_group_add_rule(rule);
4103 eviction_group_remove_rule(rule);
4107 reset_counters = (fm->flags & OFPUTIL_FF_RESET_COUNTS) != 0;
4108 if (actions_changed || reset_counters) {
4109 struct rule_actions *new_actions;
4111 op->actions = rule->actions;
4112 new_actions = rule_actions_create(ofproto,
4113 fm->ofpacts, fm->ofpacts_len);
4115 ovs_mutex_lock(&rule->mutex);
4116 rule->actions = new_actions;
4117 ovs_mutex_unlock(&rule->mutex);
4119 rule->ofproto->ofproto_class->rule_modify_actions(rule,
4122 ofoperation_complete(op, 0);
4125 ofopgroup_submit(group);
4131 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
4132 struct ofputil_flow_mod *fm, const struct ofp_header *request)
4133 OVS_REQUIRES(ofproto_mutex)
4135 if (fm->cookie_mask != htonll(0) || fm->new_cookie == OVS_BE64_MAX) {
4138 return add_flow(ofproto, ofconn, fm, request);
4141 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
4144 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4147 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
4148 struct ofputil_flow_mod *fm,
4149 const struct ofp_header *request)
4150 OVS_REQUIRES(ofproto_mutex)
4152 struct rule_criteria criteria;
4153 struct rule_collection rules;
4156 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0,
4157 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY);
4158 error = collect_rules_loose(ofproto, &criteria, &rules);
4159 rule_criteria_destroy(&criteria);
4162 error = (rules.n > 0
4163 ? modify_flows__(ofproto, ofconn, fm, request, &rules)
4164 : modify_flows_add(ofproto, ofconn, fm, request));
4167 rule_collection_destroy(&rules);
4172 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4175 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4178 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
4179 struct ofputil_flow_mod *fm,
4180 const struct ofp_header *request)
4181 OVS_REQUIRES(ofproto_mutex)
4183 struct rule_criteria criteria;
4184 struct rule_collection rules;
4187 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
4188 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY);
4189 error = collect_rules_strict(ofproto, &criteria, &rules);
4190 rule_criteria_destroy(&criteria);
4194 error = modify_flows_add(ofproto, ofconn, fm, request);
4195 } else if (rules.n == 1) {
4196 error = modify_flows__(ofproto, ofconn, fm, request, &rules);
4200 rule_collection_destroy(&rules);
4205 /* OFPFC_DELETE implementation. */
4208 delete_flow__(struct rule *rule, struct ofopgroup *group,
4209 enum ofp_flow_removed_reason reason)
4210 OVS_REQUIRES(ofproto_mutex)
4212 struct ofproto *ofproto = rule->ofproto;
4214 ofproto_rule_send_removed(rule, reason);
4216 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
4217 oftable_remove_rule(rule);
4218 ofproto->ofproto_class->rule_delete(rule);
4221 /* Deletes the rules listed in 'rules'.
4223 * Returns 0 on success, otherwise an OpenFlow error code. */
4225 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
4226 const struct ofp_header *request,
4227 const struct rule_collection *rules,
4228 enum ofp_flow_removed_reason reason)
4229 OVS_REQUIRES(ofproto_mutex)
4231 struct ofopgroup *group;
4234 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
4235 for (i = 0; i < rules->n; i++) {
4236 delete_flow__(rules->rules[i], group, reason);
4238 ofopgroup_submit(group);
4243 /* Implements OFPFC_DELETE. */
4245 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
4246 const struct ofputil_flow_mod *fm,
4247 const struct ofp_header *request)
4248 OVS_REQUIRES(ofproto_mutex)
4250 struct rule_criteria criteria;
4251 struct rule_collection rules;
4254 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0,
4255 fm->cookie, fm->cookie_mask,
4256 fm->out_port, fm->out_group);
4257 error = collect_rules_loose(ofproto, &criteria, &rules);
4258 rule_criteria_destroy(&criteria);
4260 if (!error && rules.n > 0) {
4261 error = delete_flows__(ofproto, ofconn, request, &rules, OFPRR_DELETE);
4263 rule_collection_destroy(&rules);
4268 /* Implements OFPFC_DELETE_STRICT. */
4270 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
4271 const struct ofputil_flow_mod *fm,
4272 const struct ofp_header *request)
4273 OVS_REQUIRES(ofproto_mutex)
4275 struct rule_criteria criteria;
4276 struct rule_collection rules;
4279 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
4280 fm->cookie, fm->cookie_mask,
4281 fm->out_port, fm->out_group);
4282 error = collect_rules_strict(ofproto, &criteria, &rules);
4283 rule_criteria_destroy(&criteria);
4285 if (!error && rules.n > 0) {
4286 error = delete_flows__(ofproto, ofconn, request, &rules, OFPRR_DELETE);
4288 rule_collection_destroy(&rules);
4294 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
4295 OVS_REQUIRES(ofproto_mutex)
4297 struct ofputil_flow_removed fr;
4299 if (ofproto_rule_is_hidden(rule) ||
4300 !(rule->flags & OFPUTIL_FF_SEND_FLOW_REM)) {
4304 minimatch_expand(&rule->cr.match, &fr.match);
4305 fr.priority = rule->cr.priority;
4306 fr.cookie = rule->flow_cookie;
4308 fr.table_id = rule->table_id;
4309 calc_duration(rule->created, time_msec(),
4310 &fr.duration_sec, &fr.duration_nsec);
4311 ovs_mutex_lock(&rule->mutex);
4312 fr.idle_timeout = rule->idle_timeout;
4313 fr.hard_timeout = rule->hard_timeout;
4314 ovs_mutex_unlock(&rule->mutex);
4315 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
4318 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
4321 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
4322 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
4325 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
4328 * ofproto implementation ->run() functions should use this function to expire
4329 * OpenFlow flows. */
4331 ofproto_rule_expire(struct rule *rule, uint8_t reason)
4332 OVS_REQUIRES(ofproto_mutex)
4334 struct ofproto *ofproto = rule->ofproto;
4336 ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT
4337 || reason == OFPRR_DELETE || reason == OFPRR_GROUP_DELETE);
4339 ofproto_rule_delete__(ofproto, rule, reason);
4342 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
4343 * means "infinite". */
4345 reduce_timeout(uint16_t max, uint16_t *timeout)
4347 if (max && (!*timeout || *timeout > max)) {
4352 /* If 'idle_timeout' is nonzero, and 'rule' has no idle timeout or an idle
4353 * timeout greater than 'idle_timeout', lowers 'rule''s idle timeout to
4354 * 'idle_timeout' seconds. Similarly for 'hard_timeout'.
4356 * Suitable for implementing OFPACT_FIN_TIMEOUT. */
4358 ofproto_rule_reduce_timeouts(struct rule *rule,
4359 uint16_t idle_timeout, uint16_t hard_timeout)
4360 OVS_EXCLUDED(ofproto_mutex, rule->mutex)
4362 if (!idle_timeout && !hard_timeout) {
4366 ovs_mutex_lock(&ofproto_mutex);
4367 if (list_is_empty(&rule->expirable)) {
4368 list_insert(&rule->ofproto->expirable, &rule->expirable);
4370 ovs_mutex_unlock(&ofproto_mutex);
4372 ovs_mutex_lock(&rule->mutex);
4373 reduce_timeout(idle_timeout, &rule->idle_timeout);
4374 reduce_timeout(hard_timeout, &rule->hard_timeout);
4375 ovs_mutex_unlock(&rule->mutex);
4379 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4380 OVS_EXCLUDED(ofproto_mutex)
4382 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4383 struct ofputil_flow_mod fm;
4384 uint64_t ofpacts_stub[1024 / 8];
4385 struct ofpbuf ofpacts;
4389 error = reject_slave_controller(ofconn);
4394 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
4395 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
4397 u16_to_ofp(ofproto->max_ports),
4400 error = ofproto_check_ofpacts(ofproto, fm.ofpacts, fm.ofpacts_len);
4403 error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
4406 goto exit_free_ofpacts;
4409 /* Record the operation for logging a summary report. */
4410 switch (fm.command) {
4416 case OFPFC_MODIFY_STRICT:
4417 ofproto->n_modify++;
4421 case OFPFC_DELETE_STRICT:
4422 ofproto->n_delete++;
4427 if (ofproto->next_op_report == LLONG_MAX) {
4428 ofproto->first_op = now;
4429 ofproto->next_op_report = MAX(now + 10 * 1000,
4430 ofproto->op_backoff);
4431 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
4433 ofproto->last_op = now;
4436 ofpbuf_uninit(&ofpacts);
4442 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
4443 struct ofputil_flow_mod *fm, const struct ofp_header *oh)
4444 OVS_EXCLUDED(ofproto_mutex)
4448 ovs_mutex_lock(&ofproto_mutex);
4449 if (ofproto->n_pending < 50) {
4450 switch (fm->command) {
4452 error = add_flow(ofproto, ofconn, fm, oh);
4456 error = modify_flows_loose(ofproto, ofconn, fm, oh);
4459 case OFPFC_MODIFY_STRICT:
4460 error = modify_flow_strict(ofproto, ofconn, fm, oh);
4464 error = delete_flows_loose(ofproto, ofconn, fm, oh);
4467 case OFPFC_DELETE_STRICT:
4468 error = delete_flow_strict(ofproto, ofconn, fm, oh);
4472 if (fm->command > 0xff) {
4473 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
4474 "flow_mod_table_id extension is not enabled",
4477 error = OFPERR_OFPFMFC_BAD_COMMAND;
4481 ovs_assert(!list_is_empty(&ofproto->pending));
4482 error = OFPROTO_POSTPONE;
4484 ovs_mutex_unlock(&ofproto_mutex);
4486 run_rule_executes(ofproto);
4491 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4493 struct ofputil_role_request request;
4494 struct ofputil_role_request reply;
4498 error = ofputil_decode_role_message(oh, &request);
4503 if (request.role != OFPCR12_ROLE_NOCHANGE) {
4504 if (ofconn_get_role(ofconn) != request.role
4505 && ofconn_has_pending_opgroups(ofconn)) {
4506 return OFPROTO_POSTPONE;
4509 if (request.have_generation_id
4510 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
4511 return OFPERR_OFPRRFC_STALE;
4514 ofconn_set_role(ofconn, request.role);
4517 reply.role = ofconn_get_role(ofconn);
4518 reply.have_generation_id = ofconn_get_master_election_id(
4519 ofconn, &reply.generation_id);
4520 buf = ofputil_encode_role_reply(oh, &reply);
4521 ofconn_send_reply(ofconn, buf);
4527 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
4528 const struct ofp_header *oh)
4530 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
4531 enum ofputil_protocol cur, next;
4533 cur = ofconn_get_protocol(ofconn);
4534 next = ofputil_protocol_set_tid(cur, msg->set != 0);
4535 ofconn_set_protocol(ofconn, next);
4541 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4543 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
4544 enum ofputil_protocol cur, next;
4545 enum ofputil_protocol next_base;
4547 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
4549 return OFPERR_OFPBRC_EPERM;
4552 cur = ofconn_get_protocol(ofconn);
4553 next = ofputil_protocol_set_base(cur, next_base);
4554 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
4555 /* Avoid sending async messages in surprising protocol. */
4556 return OFPROTO_POSTPONE;
4559 ofconn_set_protocol(ofconn, next);
4564 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
4565 const struct ofp_header *oh)
4567 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
4570 format = ntohl(msg->format);
4571 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
4572 return OFPERR_OFPBRC_EPERM;
4575 if (format != ofconn_get_packet_in_format(ofconn)
4576 && ofconn_has_pending_opgroups(ofconn)) {
4577 /* Avoid sending async message in surprsing packet in format. */
4578 return OFPROTO_POSTPONE;
4581 ofconn_set_packet_in_format(ofconn, format);
4586 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
4588 const struct nx_async_config *msg = ofpmsg_body(oh);
4589 uint32_t master[OAM_N_TYPES];
4590 uint32_t slave[OAM_N_TYPES];
4592 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
4593 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
4594 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
4596 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
4597 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
4598 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
4600 ofconn_set_async_config(ofconn, master, slave);
4601 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
4602 !ofconn_get_miss_send_len(ofconn)) {
4603 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
4610 handle_nxt_get_async_request(struct ofconn *ofconn, const struct ofp_header *oh)
4613 uint32_t master[OAM_N_TYPES];
4614 uint32_t slave[OAM_N_TYPES];
4615 struct nx_async_config *msg;
4617 ofconn_get_async_config(ofconn, master, slave);
4618 buf = ofpraw_alloc_reply(OFPRAW_OFPT13_GET_ASYNC_REPLY, oh, 0);
4619 msg = ofpbuf_put_zeros(buf, sizeof *msg);
4621 msg->packet_in_mask[0] = htonl(master[OAM_PACKET_IN]);
4622 msg->port_status_mask[0] = htonl(master[OAM_PORT_STATUS]);
4623 msg->flow_removed_mask[0] = htonl(master[OAM_FLOW_REMOVED]);
4625 msg->packet_in_mask[1] = htonl(slave[OAM_PACKET_IN]);
4626 msg->port_status_mask[1] = htonl(slave[OAM_PORT_STATUS]);
4627 msg->flow_removed_mask[1] = htonl(slave[OAM_FLOW_REMOVED]);
4629 ofconn_send_reply(ofconn, buf);
4635 handle_nxt_set_controller_id(struct ofconn *ofconn,
4636 const struct ofp_header *oh)
4638 const struct nx_controller_id *nci = ofpmsg_body(oh);
4640 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
4641 return OFPERR_NXBRC_MUST_BE_ZERO;
4644 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
4649 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4653 if (ofconn_has_pending_opgroups(ofconn)) {
4654 return OFPROTO_POSTPONE;
4657 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
4658 ? OFPRAW_OFPT10_BARRIER_REPLY
4659 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
4660 ofconn_send_reply(ofconn, buf);
4665 ofproto_compose_flow_refresh_update(const struct rule *rule,
4666 enum nx_flow_monitor_flags flags,
4668 OVS_REQUIRES(ofproto_mutex)
4670 struct ofoperation *op = rule->pending;
4671 const struct rule_actions *actions;
4672 struct ofputil_flow_update fu;
4675 if (op && op->type == OFOPERATION_ADD) {
4676 /* We'll report the final flow when the operation completes. Reporting
4677 * it now would cause a duplicate report later. */
4681 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
4682 ? NXFME_ADDED : NXFME_MODIFIED);
4684 ovs_mutex_lock(&rule->mutex);
4685 fu.idle_timeout = rule->idle_timeout;
4686 fu.hard_timeout = rule->hard_timeout;
4687 ovs_mutex_unlock(&rule->mutex);
4688 fu.table_id = rule->table_id;
4689 fu.cookie = rule->flow_cookie;
4690 minimatch_expand(&rule->cr.match, &match);
4692 fu.priority = rule->cr.priority;
4694 if (!(flags & NXFMF_ACTIONS)) {
4697 actions = rule->actions;
4699 /* An operation is in progress. Use the previous version of the flow's
4700 * actions, so that when the operation commits we report the change. */
4702 case OFOPERATION_ADD:
4705 case OFOPERATION_MODIFY:
4706 case OFOPERATION_REPLACE:
4707 actions = op->actions ? op->actions : rule->actions;
4710 case OFOPERATION_DELETE:
4711 actions = rule->actions;
4718 fu.ofpacts = actions ? actions->ofpacts : NULL;
4719 fu.ofpacts_len = actions ? actions->ofpacts_len : 0;
4721 if (list_is_empty(msgs)) {
4722 ofputil_start_flow_update(msgs);
4724 ofputil_append_flow_update(&fu, msgs);
4728 ofmonitor_compose_refresh_updates(struct rule_collection *rules,
4730 OVS_REQUIRES(ofproto_mutex)
4734 for (i = 0; i < rules->n; i++) {
4735 struct rule *rule = rules->rules[i];
4736 enum nx_flow_monitor_flags flags = rule->monitor_flags;
4737 rule->monitor_flags = 0;
4739 ofproto_compose_flow_refresh_update(rule, flags, msgs);
4744 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
4745 struct rule *rule, uint64_t seqno,
4746 struct rule_collection *rules)
4747 OVS_REQUIRES(ofproto_mutex)
4749 enum nx_flow_monitor_flags update;
4751 if (ofproto_rule_is_hidden(rule)) {
4756 ? ofoperation_has_out_port(rule->pending, m->out_port)
4757 : ofproto_rule_has_out_port(rule, m->out_port))) {
4762 if (rule->add_seqno > seqno) {
4763 update = NXFMF_ADD | NXFMF_MODIFY;
4764 } else if (rule->modify_seqno > seqno) {
4765 update = NXFMF_MODIFY;
4770 if (!(m->flags & update)) {
4774 update = NXFMF_INITIAL;
4777 if (!rule->monitor_flags) {
4778 rule_collection_add(rules, rule);
4780 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
4784 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
4786 struct rule_collection *rules)
4787 OVS_REQUIRES(ofproto_mutex)
4789 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
4790 const struct ofoperation *op;
4791 const struct oftable *table;
4792 struct cls_rule target;
4794 cls_rule_init_from_minimatch(&target, &m->match, 0);
4795 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
4796 struct cls_cursor cursor;
4799 ovs_rwlock_rdlock(&table->cls.rwlock);
4800 cls_cursor_init(&cursor, &table->cls, &target);
4801 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4802 ovs_assert(!rule->pending); /* XXX */
4803 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4805 ovs_rwlock_unlock(&table->cls.rwlock);
4808 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
4809 struct rule *rule = op->rule;
4811 if (((m->table_id == 0xff
4812 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
4813 : m->table_id == rule->table_id))
4814 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
4815 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4818 cls_rule_destroy(&target);
4822 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
4823 struct rule_collection *rules)
4824 OVS_REQUIRES(ofproto_mutex)
4826 if (m->flags & NXFMF_INITIAL) {
4827 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
4832 ofmonitor_collect_resume_rules(struct ofmonitor *m,
4833 uint64_t seqno, struct rule_collection *rules)
4834 OVS_REQUIRES(ofproto_mutex)
4836 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
4840 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
4841 OVS_EXCLUDED(ofproto_mutex)
4843 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4844 struct ofmonitor **monitors;
4845 size_t n_monitors, allocated_monitors;
4846 struct rule_collection rules;
4847 struct list replies;
4853 ofpbuf_use_const(&b, oh, ntohs(oh->length));
4855 n_monitors = allocated_monitors = 0;
4857 ovs_mutex_lock(&ofproto_mutex);
4859 struct ofputil_flow_monitor_request request;
4860 struct ofmonitor *m;
4863 retval = ofputil_decode_flow_monitor_request(&request, &b);
4864 if (retval == EOF) {
4866 } else if (retval) {
4871 if (request.table_id != 0xff
4872 && request.table_id >= ofproto->n_tables) {
4873 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4877 error = ofmonitor_create(&request, ofconn, &m);
4882 if (n_monitors >= allocated_monitors) {
4883 monitors = x2nrealloc(monitors, &allocated_monitors,
4886 monitors[n_monitors++] = m;
4889 rule_collection_init(&rules);
4890 for (i = 0; i < n_monitors; i++) {
4891 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
4894 ofpmp_init(&replies, oh);
4895 ofmonitor_compose_refresh_updates(&rules, &replies);
4896 ovs_mutex_unlock(&ofproto_mutex);
4898 rule_collection_destroy(&rules);
4900 ofconn_send_replies(ofconn, &replies);
4906 for (i = 0; i < n_monitors; i++) {
4907 ofmonitor_destroy(monitors[i]);
4910 ovs_mutex_unlock(&ofproto_mutex);
4916 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
4917 OVS_EXCLUDED(ofproto_mutex)
4919 struct ofmonitor *m;
4923 id = ofputil_decode_flow_monitor_cancel(oh);
4925 ovs_mutex_lock(&ofproto_mutex);
4926 m = ofmonitor_lookup(ofconn, id);
4928 ofmonitor_destroy(m);
4931 error = OFPERR_NXBRC_FM_BAD_ID;
4933 ovs_mutex_unlock(&ofproto_mutex);
4938 /* Meters implementation.
4940 * Meter table entry, indexed by the OpenFlow meter_id.
4941 * These are always dynamically allocated to allocate enough space for
4943 * 'created' is used to compute the duration for meter stats.
4944 * 'list rules' is needed so that we can delete the dependent rules when the
4945 * meter table entry is deleted.
4946 * 'provider_meter_id' is for the provider's private use.
4949 long long int created; /* Time created. */
4950 struct list rules; /* List of "struct rule_dpif"s. */
4951 ofproto_meter_id provider_meter_id;
4952 uint16_t flags; /* Meter flags. */
4953 uint16_t n_bands; /* Number of meter bands. */
4954 struct ofputil_meter_band *bands;
4958 * This is used in instruction validation at flow set-up time,
4959 * as flows may not use non-existing meters.
4960 * Return value of UINT32_MAX signifies an invalid meter.
4963 get_provider_meter_id(const struct ofproto *ofproto, uint32_t of_meter_id)
4965 if (of_meter_id && of_meter_id <= ofproto->meter_features.max_meters) {
4966 const struct meter *meter = ofproto->meters[of_meter_id];
4968 return meter->provider_meter_id.uint32;
4975 meter_update(struct meter *meter, const struct ofputil_meter_config *config)
4979 meter->flags = config->flags;
4980 meter->n_bands = config->n_bands;
4981 meter->bands = xmemdup(config->bands,
4982 config->n_bands * sizeof *meter->bands);
4985 static struct meter *
4986 meter_create(const struct ofputil_meter_config *config,
4987 ofproto_meter_id provider_meter_id)
4989 struct meter *meter;
4991 meter = xzalloc(sizeof *meter);
4992 meter->provider_meter_id = provider_meter_id;
4993 meter->created = time_msec();
4994 list_init(&meter->rules);
4996 meter_update(meter, config);
5002 meter_delete(struct ofproto *ofproto, uint32_t first, uint32_t last)
5003 OVS_REQUIRES(ofproto_mutex)
5006 for (mid = first; mid <= last; ++mid) {
5007 struct meter *meter = ofproto->meters[mid];
5009 ofproto->meters[mid] = NULL;
5010 ofproto->ofproto_class->meter_del(ofproto,
5011 meter->provider_meter_id);
5019 handle_add_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
5021 ofproto_meter_id provider_meter_id = { UINT32_MAX };
5022 struct meter **meterp = &ofproto->meters[mm->meter.meter_id];
5026 return OFPERR_OFPMMFC_METER_EXISTS;
5029 error = ofproto->ofproto_class->meter_set(ofproto, &provider_meter_id,
5032 ovs_assert(provider_meter_id.uint32 != UINT32_MAX);
5033 *meterp = meter_create(&mm->meter, provider_meter_id);
5039 handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
5041 struct meter *meter = ofproto->meters[mm->meter.meter_id];
5043 uint32_t provider_meter_id;
5046 return OFPERR_OFPMMFC_UNKNOWN_METER;
5049 provider_meter_id = meter->provider_meter_id.uint32;
5050 error = ofproto->ofproto_class->meter_set(ofproto,
5051 &meter->provider_meter_id,
5053 ovs_assert(meter->provider_meter_id.uint32 == provider_meter_id);
5055 meter_update(meter, &mm->meter);
5061 handle_delete_meter(struct ofconn *ofconn, const struct ofp_header *oh,
5062 struct ofputil_meter_mod *mm)
5063 OVS_EXCLUDED(ofproto_mutex)
5065 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5066 uint32_t meter_id = mm->meter.meter_id;
5067 struct rule_collection rules;
5068 enum ofperr error = 0;
5069 uint32_t first, last;
5071 if (meter_id == OFPM13_ALL) {
5073 last = ofproto->meter_features.max_meters;
5075 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
5078 first = last = meter_id;
5081 /* First delete the rules that use this meter. If any of those rules are
5082 * currently being modified, postpone the whole operation until later. */
5083 rule_collection_init(&rules);
5084 ovs_mutex_lock(&ofproto_mutex);
5085 for (meter_id = first; meter_id <= last; ++meter_id) {
5086 struct meter *meter = ofproto->meters[meter_id];
5087 if (meter && !list_is_empty(&meter->rules)) {
5090 LIST_FOR_EACH (rule, meter_list_node, &meter->rules) {
5091 if (rule->pending) {
5092 error = OFPROTO_POSTPONE;
5095 rule_collection_add(&rules, rule);
5100 delete_flows__(ofproto, ofconn, oh, &rules, OFPRR_METER_DELETE);
5103 /* Delete the meters. */
5104 meter_delete(ofproto, first, last);
5107 ovs_mutex_unlock(&ofproto_mutex);
5108 rule_collection_destroy(&rules);
5114 handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5116 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5117 struct ofputil_meter_mod mm;
5118 uint64_t bands_stub[256 / 8];
5119 struct ofpbuf bands;
5123 error = reject_slave_controller(ofconn);
5128 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5130 error = ofputil_decode_meter_mod(oh, &mm, &bands);
5132 goto exit_free_bands;
5135 meter_id = mm.meter.meter_id;
5137 if (mm.command != OFPMC13_DELETE) {
5138 /* Fails also when meters are not implemented by the provider. */
5139 if (meter_id == 0 || meter_id > OFPM13_MAX) {
5140 error = OFPERR_OFPMMFC_INVALID_METER;
5141 goto exit_free_bands;
5142 } else if (meter_id > ofproto->meter_features.max_meters) {
5143 error = OFPERR_OFPMMFC_OUT_OF_METERS;
5144 goto exit_free_bands;
5146 if (mm.meter.n_bands > ofproto->meter_features.max_bands) {
5147 error = OFPERR_OFPMMFC_OUT_OF_BANDS;
5148 goto exit_free_bands;
5152 switch (mm.command) {
5154 error = handle_add_meter(ofproto, &mm);
5157 case OFPMC13_MODIFY:
5158 error = handle_modify_meter(ofproto, &mm);
5161 case OFPMC13_DELETE:
5162 error = handle_delete_meter(ofconn, oh, &mm);
5166 error = OFPERR_OFPMMFC_BAD_COMMAND;
5171 ofpbuf_uninit(&bands);
5176 handle_meter_features_request(struct ofconn *ofconn,
5177 const struct ofp_header *request)
5179 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5180 struct ofputil_meter_features features;
5183 if (ofproto->ofproto_class->meter_get_features) {
5184 ofproto->ofproto_class->meter_get_features(ofproto, &features);
5186 memset(&features, 0, sizeof features);
5188 b = ofputil_encode_meter_features_reply(&features, request);
5190 ofconn_send_reply(ofconn, b);
5195 handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request,
5198 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5199 struct list replies;
5200 uint64_t bands_stub[256 / 8];
5201 struct ofpbuf bands;
5202 uint32_t meter_id, first, last;
5204 ofputil_decode_meter_request(request, &meter_id);
5206 if (meter_id == OFPM13_ALL) {
5208 last = ofproto->meter_features.max_meters;
5210 if (!meter_id || meter_id > ofproto->meter_features.max_meters ||
5211 !ofproto->meters[meter_id]) {
5212 return OFPERR_OFPMMFC_UNKNOWN_METER;
5214 first = last = meter_id;
5217 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5218 ofpmp_init(&replies, request);
5220 for (meter_id = first; meter_id <= last; ++meter_id) {
5221 struct meter *meter = ofproto->meters[meter_id];
5223 continue; /* Skip non-existing meters. */
5225 if (type == OFPTYPE_METER_STATS_REQUEST) {
5226 struct ofputil_meter_stats stats;
5228 stats.meter_id = meter_id;
5230 /* Provider sets the packet and byte counts, we do the rest. */
5231 stats.flow_count = list_size(&meter->rules);
5232 calc_duration(meter->created, time_msec(),
5233 &stats.duration_sec, &stats.duration_nsec);
5234 stats.n_bands = meter->n_bands;
5235 ofpbuf_clear(&bands);
5237 = ofpbuf_put_uninit(&bands,
5238 meter->n_bands * sizeof *stats.bands);
5240 if (!ofproto->ofproto_class->meter_get(ofproto,
5241 meter->provider_meter_id,
5243 ofputil_append_meter_stats(&replies, &stats);
5245 } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */
5246 struct ofputil_meter_config config;
5248 config.meter_id = meter_id;
5249 config.flags = meter->flags;
5250 config.n_bands = meter->n_bands;
5251 config.bands = meter->bands;
5252 ofputil_append_meter_config(&replies, &config);
5256 ofconn_send_replies(ofconn, &replies);
5257 ofpbuf_uninit(&bands);
5262 ofproto_group_lookup(const struct ofproto *ofproto, uint32_t group_id,
5263 struct ofgroup **group)
5264 OVS_TRY_RDLOCK(true, (*group)->rwlock)
5266 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5267 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
5268 hash_int(group_id, 0), &ofproto->groups) {
5269 if ((*group)->group_id == group_id) {
5270 ovs_rwlock_rdlock(&(*group)->rwlock);
5271 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5275 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5280 ofproto_group_release(struct ofgroup *group)
5281 OVS_RELEASES(group->rwlock)
5283 ovs_rwlock_unlock(&group->rwlock);
5287 ofproto_group_write_lookup(const struct ofproto *ofproto, uint32_t group_id,
5288 struct ofgroup **group)
5289 OVS_TRY_WRLOCK(true, ofproto->groups_rwlock)
5290 OVS_TRY_WRLOCK(true, (*group)->rwlock)
5292 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5293 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
5294 hash_int(group_id, 0), &ofproto->groups) {
5295 if ((*group)->group_id == group_id) {
5296 ovs_rwlock_wrlock(&(*group)->rwlock);
5300 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5305 ofproto_group_exists__(const struct ofproto *ofproto, uint32_t group_id)
5306 OVS_REQ_RDLOCK(ofproto->groups_rwlock)
5308 struct ofgroup *grp;
5310 HMAP_FOR_EACH_IN_BUCKET (grp, hmap_node,
5311 hash_int(group_id, 0), &ofproto->groups) {
5312 if (grp->group_id == group_id) {
5320 ofproto_group_exists(const struct ofproto *ofproto, uint32_t group_id)
5321 OVS_EXCLUDED(ofproto->groups_rwlock)
5325 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5326 exists = ofproto_group_exists__(ofproto, group_id);
5327 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5333 group_get_ref_count(struct ofgroup *group)
5334 OVS_EXCLUDED(ofproto_mutex)
5336 struct ofproto *ofproto = group->ofproto;
5337 struct rule_criteria criteria;
5338 struct rule_collection rules;
5343 match_init_catchall(&match);
5344 rule_criteria_init(&criteria, 0xff, &match, 0, htonll(0), htonll(0),
5345 OFPP_ANY, group->group_id);
5346 ovs_mutex_lock(&ofproto_mutex);
5347 error = collect_rules_loose(ofproto, &criteria, &rules);
5348 ovs_mutex_unlock(&ofproto_mutex);
5349 rule_criteria_destroy(&criteria);
5351 count = !error && rules.n < UINT32_MAX ? rules.n : UINT32_MAX;
5353 rule_collection_destroy(&rules);
5358 append_group_stats(struct ofgroup *group, struct list *replies)
5359 OVS_REQ_RDLOCK(group->rwlock)
5361 struct ofputil_group_stats ogs;
5362 struct ofproto *ofproto = group->ofproto;
5363 long long int now = time_msec();
5366 ogs.bucket_stats = xmalloc(group->n_buckets * sizeof *ogs.bucket_stats);
5368 /* Provider sets the packet and byte counts, we do the rest. */
5369 ogs.ref_count = group_get_ref_count(group);
5370 ogs.n_buckets = group->n_buckets;
5372 error = (ofproto->ofproto_class->group_get_stats
5373 ? ofproto->ofproto_class->group_get_stats(group, &ogs)
5376 ogs.packet_count = UINT64_MAX;
5377 ogs.byte_count = UINT64_MAX;
5378 memset(ogs.bucket_stats, 0xff,
5379 ogs.n_buckets * sizeof *ogs.bucket_stats);
5382 ogs.group_id = group->group_id;
5383 calc_duration(group->created, now, &ogs.duration_sec, &ogs.duration_nsec);
5385 ofputil_append_group_stats(replies, &ogs);
5387 free(ogs.bucket_stats);
5391 handle_group_stats_request(struct ofconn *ofconn,
5392 const struct ofp_header *request)
5394 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5395 struct list replies;
5397 struct ofgroup *group;
5400 error = ofputil_decode_group_stats_request(request, &group_id);
5405 ofpmp_init(&replies, request);
5407 if (group_id == OFPG_ALL) {
5408 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5409 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
5410 ovs_rwlock_rdlock(&group->rwlock);
5411 append_group_stats(group, &replies);
5412 ovs_rwlock_unlock(&group->rwlock);
5414 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5416 if (ofproto_group_lookup(ofproto, group_id, &group)) {
5417 append_group_stats(group, &replies);
5418 ofproto_group_release(group);
5422 ofconn_send_replies(ofconn, &replies);
5428 handle_group_desc_stats_request(struct ofconn *ofconn,
5429 const struct ofp_header *request)
5431 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5432 struct list replies;
5433 struct ofputil_group_desc gds;
5434 struct ofgroup *group;
5436 ofpmp_init(&replies, request);
5438 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5439 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
5440 gds.group_id = group->group_id;
5441 gds.type = group->type;
5442 ofputil_append_group_desc_reply(&gds, &group->buckets, &replies);
5444 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5446 ofconn_send_replies(ofconn, &replies);
5452 handle_group_features_stats_request(struct ofconn *ofconn,
5453 const struct ofp_header *request)
5455 struct ofproto *p = ofconn_get_ofproto(ofconn);
5458 msg = ofputil_encode_group_features_reply(&p->ogf, request);
5460 ofconn_send_reply(ofconn, msg);
5467 handle_queue_get_config_request(struct ofconn *ofconn,
5468 const struct ofp_header *oh)
5470 struct ofproto *p = ofconn_get_ofproto(ofconn);
5471 struct netdev_queue_dump queue_dump;
5472 struct ofport *ofport;
5473 unsigned int queue_id;
5474 struct ofpbuf *reply;
5475 struct smap details;
5479 error = ofputil_decode_queue_get_config_request(oh, &request);
5484 ofport = ofproto_get_port(p, request);
5486 return OFPERR_OFPQOFC_BAD_PORT;
5489 reply = ofputil_encode_queue_get_config_reply(oh);
5491 smap_init(&details);
5492 NETDEV_QUEUE_FOR_EACH (&queue_id, &details, &queue_dump, ofport->netdev) {
5493 struct ofputil_queue_config queue;
5495 /* None of the existing queues have compatible properties, so we
5496 * hard-code omitting min_rate and max_rate. */
5497 queue.queue_id = queue_id;
5498 queue.min_rate = UINT16_MAX;
5499 queue.max_rate = UINT16_MAX;
5500 ofputil_append_queue_get_config_reply(reply, &queue);
5502 smap_destroy(&details);
5504 ofconn_send_reply(ofconn, reply);
5509 /* Implements OFPGC11_ADD
5510 * in which no matching flow already exists in the flow table.
5512 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
5513 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
5514 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
5515 * initiated now but may be retried later.
5517 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
5518 * ownership remains with the caller.
5520 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
5523 add_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
5525 struct ofgroup *ofgroup;
5528 if (gm->group_id > OFPG_MAX) {
5529 return OFPERR_OFPGMFC_INVALID_GROUP;
5531 if (gm->type > OFPGT11_FF) {
5532 return OFPERR_OFPGMFC_BAD_TYPE;
5535 /* Allocate new group and initialize it. */
5536 ofgroup = ofproto->ofproto_class->group_alloc();
5538 VLOG_WARN_RL(&rl, "%s: failed to create group", ofproto->name);
5539 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
5542 ovs_rwlock_init(&ofgroup->rwlock);
5543 ofgroup->ofproto = ofproto;
5544 ofgroup->group_id = gm->group_id;
5545 ofgroup->type = gm->type;
5546 ofgroup->created = ofgroup->modified = time_msec();
5548 list_move(&ofgroup->buckets, &gm->buckets);
5549 ofgroup->n_buckets = list_size(&ofgroup->buckets);
5551 /* Construct called BEFORE any locks are held. */
5552 error = ofproto->ofproto_class->group_construct(ofgroup);
5557 /* We wrlock as late as possible to minimize the time we jam any other
5558 * threads: No visible state changes before acquiring the lock. */
5559 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5561 if (ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
5562 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
5566 if (ofproto_group_exists__(ofproto, gm->group_id)) {
5567 error = OFPERR_OFPGMFC_GROUP_EXISTS;
5572 /* Insert new group. */
5573 hmap_insert(&ofproto->groups, &ofgroup->hmap_node,
5574 hash_int(ofgroup->group_id, 0));
5575 ofproto->n_groups[ofgroup->type]++;
5577 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5582 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5583 ofproto->ofproto_class->group_destruct(ofgroup);
5585 ofputil_bucket_list_destroy(&ofgroup->buckets);
5586 ofproto->ofproto_class->group_dealloc(ofgroup);
5591 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
5594 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
5597 modify_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
5599 struct ofgroup *ofgroup;
5600 struct ofgroup *victim;
5603 if (gm->group_id > OFPG_MAX) {
5604 return OFPERR_OFPGMFC_INVALID_GROUP;
5607 if (gm->type > OFPGT11_FF) {
5608 return OFPERR_OFPGMFC_BAD_TYPE;
5611 victim = ofproto->ofproto_class->group_alloc();
5613 VLOG_WARN_RL(&rl, "%s: failed to allocate group", ofproto->name);
5614 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
5617 if (!ofproto_group_write_lookup(ofproto, gm->group_id, &ofgroup)) {
5618 error = OFPERR_OFPGMFC_UNKNOWN_GROUP;
5621 /* Both group's and its container's write locks held now.
5622 * Also, n_groups[] is protected by ofproto->groups_rwlock. */
5623 if (ofgroup->type != gm->type
5624 && ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
5625 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
5630 list_move(&victim->buckets, &ofgroup->buckets);
5632 ofgroup->type = gm->type;
5633 list_move(&ofgroup->buckets, &gm->buckets);
5634 ofgroup->n_buckets = list_size(&ofgroup->buckets);
5636 error = ofproto->ofproto_class->group_modify(ofgroup, victim);
5638 ofputil_bucket_list_destroy(&victim->buckets);
5639 ofproto->n_groups[victim->type]--;
5640 ofproto->n_groups[ofgroup->type]++;
5641 ofgroup->modified = time_msec();
5643 ofputil_bucket_list_destroy(&ofgroup->buckets);
5646 list_move(&ofgroup->buckets, &victim->buckets);
5650 ovs_rwlock_unlock(&ofgroup->rwlock);
5651 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5653 ofproto->ofproto_class->group_dealloc(victim);
5658 delete_group__(struct ofproto *ofproto, struct ofgroup *ofgroup)
5659 OVS_RELEASES(ofproto->groups_rwlock)
5662 struct ofputil_flow_mod fm;
5664 /* Delete all flow entries containing this group in a group action */
5665 match_init_catchall(&match);
5666 flow_mod_init(&fm, &match, 0, NULL, 0, OFPFC_DELETE);
5667 fm.out_group = ofgroup->group_id;
5668 handle_flow_mod__(ofproto, NULL, &fm, NULL);
5670 /* Must wait until existing readers are done,
5671 * while holding the container's write lock at the same time. */
5672 ovs_rwlock_wrlock(&ofgroup->rwlock);
5673 hmap_remove(&ofproto->groups, &ofgroup->hmap_node);
5674 /* No-one can find this group any more. */
5675 ofproto->n_groups[ofgroup->type]--;
5676 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5678 ofproto->ofproto_class->group_destruct(ofgroup);
5679 ofputil_bucket_list_destroy(&ofgroup->buckets);
5680 ovs_rwlock_unlock(&ofgroup->rwlock);
5681 ovs_rwlock_destroy(&ofgroup->rwlock);
5682 ofproto->ofproto_class->group_dealloc(ofgroup);
5685 /* Implements OFPGC_DELETE. */
5687 delete_group(struct ofproto *ofproto, uint32_t group_id)
5689 struct ofgroup *ofgroup;
5691 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5692 if (group_id == OFPG_ALL) {
5694 struct hmap_node *node = hmap_first(&ofproto->groups);
5698 ofgroup = CONTAINER_OF(node, struct ofgroup, hmap_node);
5699 delete_group__(ofproto, ofgroup);
5700 /* Lock for each node separately, so that we will not jam the
5701 * other threads for too long time. */
5702 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5705 HMAP_FOR_EACH_IN_BUCKET (ofgroup, hmap_node,
5706 hash_int(group_id, 0), &ofproto->groups) {
5707 if (ofgroup->group_id == group_id) {
5708 delete_group__(ofproto, ofgroup);
5713 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5717 handle_group_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5719 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5720 struct ofputil_group_mod gm;
5723 error = reject_slave_controller(ofconn);
5728 error = ofputil_decode_group_mod(oh, &gm);
5733 switch (gm.command) {
5735 return add_group(ofproto, &gm);
5737 case OFPGC11_MODIFY:
5738 return modify_group(ofproto, &gm);
5740 case OFPGC11_DELETE:
5741 delete_group(ofproto, gm.group_id);
5745 if (gm.command > OFPGC11_DELETE) {
5746 VLOG_WARN_RL(&rl, "%s: Invalid group_mod command type %d",
5747 ofproto->name, gm.command);
5749 return OFPERR_OFPGMFC_BAD_COMMAND;
5754 handle_table_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5756 struct ofputil_table_mod tm;
5759 error = reject_slave_controller(ofconn);
5764 error = ofputil_decode_table_mod(oh, &tm);
5769 /* XXX Actual table mod support is not implemented yet. */
5774 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
5775 OVS_EXCLUDED(ofproto_mutex)
5777 const struct ofp_header *oh = msg->data;
5781 error = ofptype_decode(&type, oh);
5785 if (oh->version >= OFP13_VERSION && ofpmsg_is_stat_request(oh)
5786 && ofpmp_more(oh)) {
5787 /* We have no buffer implementation for multipart requests.
5788 * Report overflow for requests which consists of multiple
5790 return OFPERR_OFPBRC_MULTIPART_BUFFER_OVERFLOW;
5794 /* OpenFlow requests. */
5795 case OFPTYPE_ECHO_REQUEST:
5796 return handle_echo_request(ofconn, oh);
5798 case OFPTYPE_FEATURES_REQUEST:
5799 return handle_features_request(ofconn, oh);
5801 case OFPTYPE_GET_CONFIG_REQUEST:
5802 return handle_get_config_request(ofconn, oh);
5804 case OFPTYPE_SET_CONFIG:
5805 return handle_set_config(ofconn, oh);
5807 case OFPTYPE_PACKET_OUT:
5808 return handle_packet_out(ofconn, oh);
5810 case OFPTYPE_PORT_MOD:
5811 return handle_port_mod(ofconn, oh);
5813 case OFPTYPE_FLOW_MOD:
5814 return handle_flow_mod(ofconn, oh);
5816 case OFPTYPE_GROUP_MOD:
5817 return handle_group_mod(ofconn, oh);
5819 case OFPTYPE_TABLE_MOD:
5820 return handle_table_mod(ofconn, oh);
5822 case OFPTYPE_METER_MOD:
5823 return handle_meter_mod(ofconn, oh);
5825 case OFPTYPE_BARRIER_REQUEST:
5826 return handle_barrier_request(ofconn, oh);
5828 case OFPTYPE_ROLE_REQUEST:
5829 return handle_role_request(ofconn, oh);
5831 /* OpenFlow replies. */
5832 case OFPTYPE_ECHO_REPLY:
5835 /* Nicira extension requests. */
5836 case OFPTYPE_FLOW_MOD_TABLE_ID:
5837 return handle_nxt_flow_mod_table_id(ofconn, oh);
5839 case OFPTYPE_SET_FLOW_FORMAT:
5840 return handle_nxt_set_flow_format(ofconn, oh);
5842 case OFPTYPE_SET_PACKET_IN_FORMAT:
5843 return handle_nxt_set_packet_in_format(ofconn, oh);
5845 case OFPTYPE_SET_CONTROLLER_ID:
5846 return handle_nxt_set_controller_id(ofconn, oh);
5848 case OFPTYPE_FLOW_AGE:
5849 /* Nothing to do. */
5852 case OFPTYPE_FLOW_MONITOR_CANCEL:
5853 return handle_flow_monitor_cancel(ofconn, oh);
5855 case OFPTYPE_SET_ASYNC_CONFIG:
5856 return handle_nxt_set_async_config(ofconn, oh);
5858 case OFPTYPE_GET_ASYNC_REQUEST:
5859 return handle_nxt_get_async_request(ofconn, oh);
5861 /* Statistics requests. */
5862 case OFPTYPE_DESC_STATS_REQUEST:
5863 return handle_desc_stats_request(ofconn, oh);
5865 case OFPTYPE_FLOW_STATS_REQUEST:
5866 return handle_flow_stats_request(ofconn, oh);
5868 case OFPTYPE_AGGREGATE_STATS_REQUEST:
5869 return handle_aggregate_stats_request(ofconn, oh);
5871 case OFPTYPE_TABLE_STATS_REQUEST:
5872 return handle_table_stats_request(ofconn, oh);
5874 case OFPTYPE_PORT_STATS_REQUEST:
5875 return handle_port_stats_request(ofconn, oh);
5877 case OFPTYPE_QUEUE_STATS_REQUEST:
5878 return handle_queue_stats_request(ofconn, oh);
5880 case OFPTYPE_PORT_DESC_STATS_REQUEST:
5881 return handle_port_desc_stats_request(ofconn, oh);
5883 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
5884 return handle_flow_monitor_request(ofconn, oh);
5886 case OFPTYPE_METER_STATS_REQUEST:
5887 case OFPTYPE_METER_CONFIG_STATS_REQUEST:
5888 return handle_meter_request(ofconn, oh, type);
5890 case OFPTYPE_METER_FEATURES_STATS_REQUEST:
5891 return handle_meter_features_request(ofconn, oh);
5893 case OFPTYPE_GROUP_STATS_REQUEST:
5894 return handle_group_stats_request(ofconn, oh);
5896 case OFPTYPE_GROUP_DESC_STATS_REQUEST:
5897 return handle_group_desc_stats_request(ofconn, oh);
5899 case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
5900 return handle_group_features_stats_request(ofconn, oh);
5902 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
5903 return handle_queue_get_config_request(ofconn, oh);
5907 case OFPTYPE_FEATURES_REPLY:
5908 case OFPTYPE_GET_CONFIG_REPLY:
5909 case OFPTYPE_PACKET_IN:
5910 case OFPTYPE_FLOW_REMOVED:
5911 case OFPTYPE_PORT_STATUS:
5912 case OFPTYPE_BARRIER_REPLY:
5913 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
5914 case OFPTYPE_DESC_STATS_REPLY:
5915 case OFPTYPE_FLOW_STATS_REPLY:
5916 case OFPTYPE_QUEUE_STATS_REPLY:
5917 case OFPTYPE_PORT_STATS_REPLY:
5918 case OFPTYPE_TABLE_STATS_REPLY:
5919 case OFPTYPE_AGGREGATE_STATS_REPLY:
5920 case OFPTYPE_PORT_DESC_STATS_REPLY:
5921 case OFPTYPE_ROLE_REPLY:
5922 case OFPTYPE_FLOW_MONITOR_PAUSED:
5923 case OFPTYPE_FLOW_MONITOR_RESUMED:
5924 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
5925 case OFPTYPE_GET_ASYNC_REPLY:
5926 case OFPTYPE_GROUP_STATS_REPLY:
5927 case OFPTYPE_GROUP_DESC_STATS_REPLY:
5928 case OFPTYPE_GROUP_FEATURES_STATS_REPLY:
5929 case OFPTYPE_METER_STATS_REPLY:
5930 case OFPTYPE_METER_CONFIG_STATS_REPLY:
5931 case OFPTYPE_METER_FEATURES_STATS_REPLY:
5932 case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
5933 case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
5934 case OFPTYPE_ROLE_STATUS:
5936 if (ofpmsg_is_stat_request(oh)) {
5937 return OFPERR_OFPBRC_BAD_STAT;
5939 return OFPERR_OFPBRC_BAD_TYPE;
5945 handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
5946 OVS_EXCLUDED(ofproto_mutex)
5948 int error = handle_openflow__(ofconn, ofp_msg);
5949 if (error && error != OFPROTO_POSTPONE) {
5950 ofconn_send_error(ofconn, ofp_msg->data, error);
5952 COVERAGE_INC(ofproto_recv_openflow);
5953 return error != OFPROTO_POSTPONE;
5956 /* Asynchronous operations. */
5958 /* Creates and returns a new ofopgroup that is not associated with any
5959 * OpenFlow connection.
5961 * The caller should add operations to the returned group with
5962 * ofoperation_create() and then submit it with ofopgroup_submit(). */
5963 static struct ofopgroup *
5964 ofopgroup_create_unattached(struct ofproto *ofproto)
5965 OVS_REQUIRES(ofproto_mutex)
5967 struct ofopgroup *group = xzalloc(sizeof *group);
5968 group->ofproto = ofproto;
5969 list_init(&group->ofproto_node);
5970 list_init(&group->ops);
5971 list_init(&group->ofconn_node);
5975 /* Creates and returns a new ofopgroup for 'ofproto'.
5977 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
5978 * connection. The 'request' and 'buffer_id' arguments are ignored.
5980 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
5981 * If the ofopgroup eventually fails, then the error reply will include
5982 * 'request'. If the ofopgroup eventually succeeds, then the packet with
5983 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
5985 * The caller should add operations to the returned group with
5986 * ofoperation_create() and then submit it with ofopgroup_submit(). */
5987 static struct ofopgroup *
5988 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
5989 const struct ofp_header *request, uint32_t buffer_id)
5990 OVS_REQUIRES(ofproto_mutex)
5992 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
5994 size_t request_len = ntohs(request->length);
5996 ovs_assert(ofconn_get_ofproto(ofconn) == ofproto);
5998 ofconn_add_opgroup(ofconn, &group->ofconn_node);
5999 group->ofconn = ofconn;
6000 group->request = xmemdup(request, MIN(request_len, 64));
6001 group->buffer_id = buffer_id;
6006 /* Submits 'group' for processing.
6008 * If 'group' contains no operations (e.g. none were ever added, or all of the
6009 * ones that were added completed synchronously), then it is destroyed
6010 * immediately. Otherwise it is added to the ofproto's list of pending
6013 ofopgroup_submit(struct ofopgroup *group)
6014 OVS_REQUIRES(ofproto_mutex)
6016 if (!group->n_running) {
6017 ofopgroup_complete(group);
6019 list_push_back(&group->ofproto->pending, &group->ofproto_node);
6020 group->ofproto->n_pending++;
6025 ofopgroup_complete(struct ofopgroup *group)
6026 OVS_REQUIRES(ofproto_mutex)
6028 struct ofproto *ofproto = group->ofproto;
6030 struct ofconn *abbrev_ofconn;
6031 ovs_be32 abbrev_xid;
6033 struct ofoperation *op, *next_op;
6036 ovs_assert(!group->n_running);
6039 LIST_FOR_EACH (op, group_node, &group->ops) {
6046 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
6047 LIST_FOR_EACH (op, group_node, &group->ops) {
6048 if (op->type != OFOPERATION_DELETE) {
6049 struct ofpbuf *packet;
6052 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
6055 struct rule_execute *re;
6059 ofproto_rule_ref(op->rule);
6061 re = xmalloc(sizeof *re);
6062 re->rule = op->rule;
6063 re->in_port = in_port;
6064 re->packet = packet;
6066 if (!guarded_list_push_back(&ofproto->rule_executes,
6067 &re->list_node, 1024)) {
6068 ofproto_rule_unref(op->rule);
6069 ofpbuf_delete(re->packet);
6078 if (!error && !list_is_empty(&group->ofconn_node)) {
6079 abbrev_ofconn = group->ofconn;
6080 abbrev_xid = group->request->xid;
6082 abbrev_ofconn = NULL;
6083 abbrev_xid = htonl(0);
6085 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
6086 struct rule *rule = op->rule;
6088 /* We generally want to report the change to active OpenFlow flow
6089 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
6091 - The operation failed.
6093 - The affected rule is not visible to controllers.
6095 - The operation's only effect was to update rule->modified. */
6097 || ofproto_rule_is_hidden(rule)
6098 || (op->type == OFOPERATION_MODIFY
6100 && rule->flow_cookie == op->flow_cookie))) {
6101 /* Check that we can just cast from ofoperation_type to
6102 * nx_flow_update_event. */
6103 enum nx_flow_update_event event_type;
6106 case OFOPERATION_ADD:
6107 case OFOPERATION_REPLACE:
6108 event_type = NXFME_ADDED;
6111 case OFOPERATION_DELETE:
6112 event_type = NXFME_DELETED;
6115 case OFOPERATION_MODIFY:
6116 event_type = NXFME_MODIFIED;
6123 ofmonitor_report(ofproto->connmgr, rule, event_type,
6124 op->reason, abbrev_ofconn, abbrev_xid);
6127 rule->pending = NULL;
6130 case OFOPERATION_ADD:
6134 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
6135 if (vid_mask == VLAN_VID_MASK) {
6136 if (ofproto->vlan_bitmap) {
6137 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
6138 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
6139 bitmap_set1(ofproto->vlan_bitmap, vid);
6140 ofproto->vlans_changed = true;
6143 ofproto->vlans_changed = true;
6147 oftable_remove_rule(rule);
6148 ofproto_rule_unref(rule);
6152 case OFOPERATION_DELETE:
6153 ovs_assert(!op->error);
6154 ofproto_rule_unref(rule);
6158 case OFOPERATION_MODIFY:
6159 case OFOPERATION_REPLACE:
6161 long long int now = time_msec();
6163 rule->modified = now;
6164 if (op->type == OFOPERATION_REPLACE) {
6165 rule->created = rule->used = now;
6168 ofproto_rule_change_cookie(ofproto, rule, op->flow_cookie);
6169 ovs_mutex_lock(&rule->mutex);
6170 rule->idle_timeout = op->idle_timeout;
6171 rule->hard_timeout = op->hard_timeout;
6172 ovs_mutex_unlock(&rule->mutex);
6174 struct rule_actions *old_actions;
6176 ovs_mutex_lock(&rule->mutex);
6177 old_actions = rule->actions;
6178 rule->actions = op->actions;
6179 ovs_mutex_unlock(&rule->mutex);
6182 rule_actions_unref(old_actions);
6184 rule->flags = op->flags;
6192 ofoperation_destroy(op);
6195 ofmonitor_flush(ofproto->connmgr);
6197 if (!list_is_empty(&group->ofproto_node)) {
6198 ovs_assert(ofproto->n_pending > 0);
6199 ofproto->n_pending--;
6200 list_remove(&group->ofproto_node);
6202 if (!list_is_empty(&group->ofconn_node)) {
6203 list_remove(&group->ofconn_node);
6205 ofconn_send_error(group->ofconn, group->request, error);
6207 connmgr_retry(ofproto->connmgr);
6209 free(group->request);
6213 /* Initiates a new operation on 'rule', of the specified 'type', within
6214 * 'group'. Prior to calling, 'rule' must not have any pending operation.
6216 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
6217 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
6219 * Returns the newly created ofoperation (which is also available as
6220 * rule->pending). */
6221 static struct ofoperation *
6222 ofoperation_create(struct ofopgroup *group, struct rule *rule,
6223 enum ofoperation_type type,
6224 enum ofp_flow_removed_reason reason)
6225 OVS_REQUIRES(ofproto_mutex)
6227 struct ofproto *ofproto = group->ofproto;
6228 struct ofoperation *op;
6230 ovs_assert(!rule->pending);
6232 op = rule->pending = xzalloc(sizeof *op);
6234 list_push_back(&group->ops, &op->group_node);
6237 op->reason = reason;
6238 op->flow_cookie = rule->flow_cookie;
6239 ovs_mutex_lock(&rule->mutex);
6240 op->idle_timeout = rule->idle_timeout;
6241 op->hard_timeout = rule->hard_timeout;
6242 ovs_mutex_unlock(&rule->mutex);
6243 op->flags = rule->flags;
6247 if (type == OFOPERATION_DELETE) {
6248 hmap_insert(&ofproto->deletions, &op->hmap_node,
6249 cls_rule_hash(&rule->cr, rule->table_id));
6256 ofoperation_destroy(struct ofoperation *op)
6257 OVS_REQUIRES(ofproto_mutex)
6259 struct ofopgroup *group = op->group;
6262 op->rule->pending = NULL;
6264 if (op->type == OFOPERATION_DELETE) {
6265 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
6267 list_remove(&op->group_node);
6268 rule_actions_unref(op->actions);
6272 /* Indicates that 'op' completed with status 'error', which is either 0 to
6273 * indicate success or an OpenFlow error code on failure.
6275 * If 'error' is 0, indicating success, the operation will be committed
6276 * permanently to the flow table.
6278 * If 'error' is nonzero, then generally the operation will be rolled back:
6280 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
6281 * restores the original rule. The caller must have uninitialized any
6282 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
6283 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
6284 * and 7 for the new rule, calling its ->rule_dealloc() function.
6286 * - If 'op' is a "modify flow" operation, ofproto restores the original
6289 * - 'op' must not be a "delete flow" operation. Removing a rule is not
6290 * allowed to fail. It must always succeed.
6292 * Please see the large comment in ofproto/ofproto-provider.h titled
6293 * "Asynchronous Operation Support" for more information. */
6295 ofoperation_complete(struct ofoperation *op, enum ofperr error)
6297 struct ofopgroup *group = op->group;
6299 ovs_assert(group->n_running > 0);
6300 ovs_assert(!error || op->type != OFOPERATION_DELETE);
6303 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
6304 /* This function can be called from ->rule_construct(), in which case
6305 * ofproto_mutex is held, or it can be called from ->run(), in which
6306 * case ofproto_mutex is not held. But only in the latter case can we
6307 * arrive here, so we can safely take ofproto_mutex now. */
6308 ovs_mutex_lock(&ofproto_mutex);
6309 ovs_assert(op->rule->pending == op);
6310 ofopgroup_complete(group);
6311 ovs_mutex_unlock(&ofproto_mutex);
6316 pick_datapath_id(const struct ofproto *ofproto)
6318 const struct ofport *port;
6320 port = ofproto_get_port(ofproto, OFPP_LOCAL);
6322 uint8_t ea[ETH_ADDR_LEN];
6325 error = netdev_get_etheraddr(port->netdev, ea);
6327 return eth_addr_to_uint64(ea);
6329 VLOG_WARN("%s: could not get MAC address for %s (%s)",
6330 ofproto->name, netdev_get_name(port->netdev),
6331 ovs_strerror(error));
6333 return ofproto->fallback_dpid;
6337 pick_fallback_dpid(void)
6339 uint8_t ea[ETH_ADDR_LEN];
6340 eth_addr_nicira_random(ea);
6341 return eth_addr_to_uint64(ea);
6344 /* Table overflow policy. */
6346 /* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep'
6347 * to NULL if the table is not configured to evict rules or if the table
6348 * contains no evictable rules. (Rules with a readlock on their evict rwlock,
6349 * or with no timeouts are not evictable.) */
6351 choose_rule_to_evict(struct oftable *table, struct rule **rulep)
6352 OVS_REQUIRES(ofproto_mutex)
6354 struct eviction_group *evg;
6357 if (!table->eviction_fields) {
6361 /* In the common case, the outer and inner loops here will each be entered
6364 * - The inner loop normally "return"s in its first iteration. If the
6365 * eviction group has any evictable rules, then it always returns in
6368 * - The outer loop only iterates more than once if the largest eviction
6369 * group has no evictable rules.
6371 * - The outer loop can exit only if table's 'max_flows' is all filled up
6372 * by unevictable rules. */
6373 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
6376 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
6385 /* Searches 'ofproto' for tables that have more flows than their configured
6386 * maximum and that have flow eviction enabled, and evicts as many flows as
6387 * necessary and currently feasible from them.
6389 * This triggers only when an OpenFlow table has N flows in it and then the
6390 * client configures a maximum number of flows less than N. */
6392 ofproto_evict(struct ofproto *ofproto)
6394 struct oftable *table;
6396 ovs_mutex_lock(&ofproto_mutex);
6397 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
6398 evict_rules_from_table(ofproto, table, 0);
6400 ovs_mutex_unlock(&ofproto_mutex);
6403 /* Eviction groups. */
6405 /* Returns the priority to use for an eviction_group that contains 'n_rules'
6406 * rules. The priority contains low-order random bits to ensure that eviction
6407 * groups with the same number of rules are prioritized randomly. */
6409 eviction_group_priority(size_t n_rules)
6411 uint16_t size = MIN(UINT16_MAX, n_rules);
6412 return (size << 16) | random_uint16();
6415 /* Updates 'evg', an eviction_group within 'table', following a change that
6416 * adds or removes rules in 'evg'. */
6418 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
6419 OVS_REQUIRES(ofproto_mutex)
6421 heap_change(&table->eviction_groups_by_size, &evg->size_node,
6422 eviction_group_priority(heap_count(&evg->rules)));
6425 /* Destroys 'evg', an eviction_group within 'table':
6427 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
6428 * rules themselves, just removes them from the eviction group.)
6430 * - Removes 'evg' from 'table'.
6434 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
6435 OVS_REQUIRES(ofproto_mutex)
6437 while (!heap_is_empty(&evg->rules)) {
6440 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
6441 rule->eviction_group = NULL;
6443 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
6444 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
6445 heap_destroy(&evg->rules);
6449 /* Removes 'rule' from its eviction group, if any. */
6451 eviction_group_remove_rule(struct rule *rule)
6452 OVS_REQUIRES(ofproto_mutex)
6454 if (rule->eviction_group) {
6455 struct oftable *table = &rule->ofproto->tables[rule->table_id];
6456 struct eviction_group *evg = rule->eviction_group;
6458 rule->eviction_group = NULL;
6459 heap_remove(&evg->rules, &rule->evg_node);
6460 if (heap_is_empty(&evg->rules)) {
6461 eviction_group_destroy(table, evg);
6463 eviction_group_resized(table, evg);
6468 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
6469 * returns the hash value. */
6471 eviction_group_hash_rule(struct rule *rule)
6472 OVS_REQUIRES(ofproto_mutex)
6474 struct oftable *table = &rule->ofproto->tables[rule->table_id];
6475 const struct mf_subfield *sf;
6479 hash = table->eviction_group_id_basis;
6480 miniflow_expand(&rule->cr.match.flow, &flow);
6481 for (sf = table->eviction_fields;
6482 sf < &table->eviction_fields[table->n_eviction_fields];
6485 if (mf_are_prereqs_ok(sf->field, &flow)) {
6486 union mf_value value;
6488 mf_get_value(sf->field, &flow, &value);
6490 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
6492 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
6493 unsigned int start = sf->ofs + sf->n_bits;
6494 bitwise_zero(&value, sf->field->n_bytes, start,
6495 sf->field->n_bytes * 8 - start);
6497 hash = hash_bytes(&value, sf->field->n_bytes, hash);
6499 hash = hash_int(hash, 0);
6506 /* Returns an eviction group within 'table' with the given 'id', creating one
6508 static struct eviction_group *
6509 eviction_group_find(struct oftable *table, uint32_t id)
6510 OVS_REQUIRES(ofproto_mutex)
6512 struct eviction_group *evg;
6514 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
6518 evg = xmalloc(sizeof *evg);
6519 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
6520 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
6521 eviction_group_priority(0));
6522 heap_init(&evg->rules);
6527 /* Returns an eviction priority for 'rule'. The return value should be
6528 * interpreted so that higher priorities make a rule more attractive candidates
6531 rule_eviction_priority(struct rule *rule)
6532 OVS_REQUIRES(ofproto_mutex)
6534 long long int hard_expiration;
6535 long long int idle_expiration;
6536 long long int expiration;
6537 uint32_t expiration_offset;
6539 /* Calculate time of expiration. */
6540 ovs_mutex_lock(&rule->mutex);
6541 hard_expiration = (rule->hard_timeout
6542 ? rule->modified + rule->hard_timeout * 1000
6544 idle_expiration = (rule->idle_timeout
6545 ? rule->used + rule->idle_timeout * 1000
6547 expiration = MIN(hard_expiration, idle_expiration);
6548 ovs_mutex_unlock(&rule->mutex);
6549 if (expiration == LLONG_MAX) {
6553 /* Calculate the time of expiration as a number of (approximate) seconds
6554 * after program startup.
6556 * This should work OK for program runs that last UINT32_MAX seconds or
6557 * less. Therefore, please restart OVS at least once every 136 years. */
6558 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
6560 /* Invert the expiration offset because we're using a max-heap. */
6561 return UINT32_MAX - expiration_offset;
6564 /* Adds 'rule' to an appropriate eviction group for its oftable's
6565 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
6566 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
6569 * The caller must ensure that 'rule' is not already in an eviction group. */
6571 eviction_group_add_rule(struct rule *rule)
6572 OVS_REQUIRES(ofproto_mutex)
6574 struct ofproto *ofproto = rule->ofproto;
6575 struct oftable *table = &ofproto->tables[rule->table_id];
6578 ovs_mutex_lock(&rule->mutex);
6579 has_timeout = rule->hard_timeout || rule->idle_timeout;
6580 ovs_mutex_unlock(&rule->mutex);
6582 if (table->eviction_fields && has_timeout) {
6583 struct eviction_group *evg;
6585 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
6587 rule->eviction_group = evg;
6588 heap_insert(&evg->rules, &rule->evg_node,
6589 rule_eviction_priority(rule));
6590 eviction_group_resized(table, evg);
6596 /* Initializes 'table'. */
6598 oftable_init(struct oftable *table)
6600 memset(table, 0, sizeof *table);
6601 classifier_init(&table->cls);
6602 table->max_flows = UINT_MAX;
6605 /* Destroys 'table', including its classifier and eviction groups.
6607 * The caller is responsible for freeing 'table' itself. */
6609 oftable_destroy(struct oftable *table)
6611 ovs_rwlock_rdlock(&table->cls.rwlock);
6612 ovs_assert(classifier_is_empty(&table->cls));
6613 ovs_rwlock_unlock(&table->cls.rwlock);
6614 oftable_disable_eviction(table);
6615 classifier_destroy(&table->cls);
6619 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
6620 * string, then 'table' will use its default name.
6622 * This only affects the name exposed for a table exposed through the OpenFlow
6623 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
6625 oftable_set_name(struct oftable *table, const char *name)
6627 if (name && name[0]) {
6628 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
6629 if (!table->name || strncmp(name, table->name, len)) {
6631 table->name = xmemdup0(name, len);
6639 /* oftables support a choice of two policies when adding a rule would cause the
6640 * number of flows in the table to exceed the configured maximum number: either
6641 * they can refuse to add the new flow or they can evict some existing flow.
6642 * This function configures the former policy on 'table'. */
6644 oftable_disable_eviction(struct oftable *table)
6645 OVS_REQUIRES(ofproto_mutex)
6647 if (table->eviction_fields) {
6648 struct eviction_group *evg, *next;
6650 HMAP_FOR_EACH_SAFE (evg, next, id_node,
6651 &table->eviction_groups_by_id) {
6652 eviction_group_destroy(table, evg);
6654 hmap_destroy(&table->eviction_groups_by_id);
6655 heap_destroy(&table->eviction_groups_by_size);
6657 free(table->eviction_fields);
6658 table->eviction_fields = NULL;
6659 table->n_eviction_fields = 0;
6663 /* oftables support a choice of two policies when adding a rule would cause the
6664 * number of flows in the table to exceed the configured maximum number: either
6665 * they can refuse to add the new flow or they can evict some existing flow.
6666 * This function configures the latter policy on 'table', with fairness based
6667 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
6668 * 'n_fields' as 0 disables fairness.) */
6670 oftable_enable_eviction(struct oftable *table,
6671 const struct mf_subfield *fields, size_t n_fields)
6672 OVS_REQUIRES(ofproto_mutex)
6674 struct cls_cursor cursor;
6677 if (table->eviction_fields
6678 && n_fields == table->n_eviction_fields
6680 || !memcmp(fields, table->eviction_fields,
6681 n_fields * sizeof *fields))) {
6686 oftable_disable_eviction(table);
6688 table->n_eviction_fields = n_fields;
6689 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
6691 table->eviction_group_id_basis = random_uint32();
6692 hmap_init(&table->eviction_groups_by_id);
6693 heap_init(&table->eviction_groups_by_size);
6695 ovs_rwlock_rdlock(&table->cls.rwlock);
6696 cls_cursor_init(&cursor, &table->cls, NULL);
6697 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
6698 eviction_group_add_rule(rule);
6700 ovs_rwlock_unlock(&table->cls.rwlock);
6703 /* Removes 'rule' from the oftable that contains it. */
6705 oftable_remove_rule__(struct ofproto *ofproto, struct rule *rule)
6706 OVS_REQUIRES(ofproto_mutex)
6708 struct classifier *cls = &ofproto->tables[rule->table_id].cls;
6710 ovs_rwlock_wrlock(&cls->rwlock);
6711 classifier_remove(cls, CONST_CAST(struct cls_rule *, &rule->cr));
6712 ovs_rwlock_unlock(&cls->rwlock);
6714 cookies_remove(ofproto, rule);
6716 eviction_group_remove_rule(rule);
6717 if (!list_is_empty(&rule->expirable)) {
6718 list_remove(&rule->expirable);
6720 if (!list_is_empty(&rule->meter_list_node)) {
6721 list_remove(&rule->meter_list_node);
6722 list_init(&rule->meter_list_node);
6727 oftable_remove_rule(struct rule *rule)
6728 OVS_REQUIRES(ofproto_mutex)
6730 oftable_remove_rule__(rule->ofproto, rule);
6733 /* Inserts 'rule' into its oftable, which must not already contain any rule for
6734 * the same cls_rule. */
6736 oftable_insert_rule(struct rule *rule)
6737 OVS_REQUIRES(ofproto_mutex)
6739 struct ofproto *ofproto = rule->ofproto;
6740 struct oftable *table = &ofproto->tables[rule->table_id];
6743 ovs_mutex_lock(&rule->mutex);
6744 may_expire = rule->hard_timeout || rule->idle_timeout;
6745 ovs_mutex_unlock(&rule->mutex);
6748 list_insert(&ofproto->expirable, &rule->expirable);
6751 cookies_insert(ofproto, rule);
6753 if (rule->actions->provider_meter_id != UINT32_MAX) {
6754 uint32_t meter_id = ofpacts_get_meter(rule->actions->ofpacts,
6755 rule->actions->ofpacts_len);
6756 struct meter *meter = ofproto->meters[meter_id];
6757 list_insert(&meter->rules, &rule->meter_list_node);
6759 ovs_rwlock_wrlock(&table->cls.rwlock);
6760 classifier_insert(&table->cls, CONST_CAST(struct cls_rule *, &rule->cr));
6761 ovs_rwlock_unlock(&table->cls.rwlock);
6762 eviction_group_add_rule(rule);
6765 /* unixctl commands. */
6768 ofproto_lookup(const char *name)
6770 struct ofproto *ofproto;
6772 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
6774 if (!strcmp(ofproto->name, name)) {
6782 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
6783 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6785 struct ofproto *ofproto;
6789 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
6790 ds_put_format(&results, "%s\n", ofproto->name);
6792 unixctl_command_reply(conn, ds_cstr(&results));
6793 ds_destroy(&results);
6797 ofproto_unixctl_init(void)
6799 static bool registered;
6805 unixctl_command_register("ofproto/list", "", 0, 0,
6806 ofproto_unixctl_list, NULL);
6809 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6811 * This is deprecated. It is only for compatibility with broken device drivers
6812 * in old versions of Linux that do not properly support VLANs when VLAN
6813 * devices are not used. When broken device drivers are no longer in
6814 * widespread use, we will delete these interfaces. */
6816 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
6817 * (exactly) by an OpenFlow rule in 'ofproto'. */
6819 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
6821 const struct oftable *oftable;
6823 free(ofproto->vlan_bitmap);
6824 ofproto->vlan_bitmap = bitmap_allocate(4096);
6825 ofproto->vlans_changed = false;
6827 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
6828 const struct cls_subtable *table;
6830 ovs_rwlock_rdlock(&oftable->cls.rwlock);
6831 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.subtables) {
6832 if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
6833 const struct cls_rule *rule;
6835 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
6836 uint16_t vid = miniflow_get_vid(&rule->match.flow);
6837 bitmap_set1(vlan_bitmap, vid);
6838 bitmap_set1(ofproto->vlan_bitmap, vid);
6842 ovs_rwlock_unlock(&oftable->cls.rwlock);
6846 /* Returns true if new VLANs have come into use by the flow table since the
6847 * last call to ofproto_get_vlan_usage().
6849 * We don't track when old VLANs stop being used. */
6851 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
6853 return ofproto->vlans_changed;
6856 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
6857 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
6858 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
6859 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
6860 * then the VLAN device is un-enslaved. */
6862 ofproto_port_set_realdev(struct ofproto *ofproto, ofp_port_t vlandev_ofp_port,
6863 ofp_port_t realdev_ofp_port, int vid)
6865 struct ofport *ofport;
6868 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
6870 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
6872 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
6873 ofproto->name, vlandev_ofp_port);
6877 if (!ofproto->ofproto_class->set_realdev) {
6878 if (!vlandev_ofp_port) {
6881 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
6885 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
6887 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
6888 ofproto->name, vlandev_ofp_port,
6889 netdev_get_name(ofport->netdev), ovs_strerror(error));