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"
54 #include "unaligned.h"
58 VLOG_DEFINE_THIS_MODULE(ofproto);
60 COVERAGE_DEFINE(ofproto_error);
61 COVERAGE_DEFINE(ofproto_flush);
62 COVERAGE_DEFINE(ofproto_no_packet_in);
63 COVERAGE_DEFINE(ofproto_packet_out);
64 COVERAGE_DEFINE(ofproto_queue_req);
65 COVERAGE_DEFINE(ofproto_recv_openflow);
66 COVERAGE_DEFINE(ofproto_reinit_ports);
67 COVERAGE_DEFINE(ofproto_uninstallable);
68 COVERAGE_DEFINE(ofproto_update_port);
71 S_OPENFLOW, /* Processing OpenFlow commands. */
72 S_EVICT, /* Evicting flows from over-limit tables. */
73 S_FLUSH, /* Deleting all flow table rules. */
76 enum ofoperation_type {
83 /* A single OpenFlow request can execute any number of operations. The
84 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
85 * ofconn to which an error reply should be sent if necessary.
87 * ofproto initiates some operations internally. These operations are still
88 * assigned to groups but will not have an associated ofconn. */
90 struct ofproto *ofproto; /* Owning ofproto. */
91 struct list ofproto_node; /* In ofproto's "pending" list. */
92 struct list ops; /* List of "struct ofoperation"s. */
93 int n_running; /* Number of ops still pending. */
95 /* Data needed to send OpenFlow reply on failure or to send a buffered
98 * If list_is_empty(ofconn_node) then this ofopgroup never had an
99 * associated ofconn or its ofconn's connection dropped after it initiated
100 * the operation. In the latter case 'ofconn' is a wild pointer that
101 * refers to freed memory, so the 'ofconn' member must be used only if
102 * !list_is_empty(ofconn_node).
104 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
105 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
106 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
107 uint32_t buffer_id; /* Buffer id from original request. */
110 static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
111 static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
112 const struct ofp_header *,
114 static void ofopgroup_submit(struct ofopgroup *);
115 static void ofopgroup_complete(struct ofopgroup *);
117 /* A single flow table operation. */
119 struct ofopgroup *group; /* Owning group. */
120 struct list group_node; /* In ofopgroup's "ops" list. */
121 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
122 struct rule *rule; /* Rule being operated upon. */
123 enum ofoperation_type type; /* Type of operation. */
125 /* OFOPERATION_MODIFY, OFOPERATION_REPLACE: The old actions, if the actions
127 struct rule_actions *actions;
129 /* OFOPERATION_DELETE. */
130 enum ofp_flow_removed_reason reason; /* Reason flow was removed. */
132 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
133 uint16_t idle_timeout; /* Rule's old idle timeout. */
134 uint16_t hard_timeout; /* Rule's old hard timeout. */
135 bool send_flow_removed; /* Rule's old 'send_flow_removed'. */
136 enum ofperr error; /* 0 if no error. */
139 static struct ofoperation *ofoperation_create(struct ofopgroup *,
141 enum ofoperation_type,
142 enum ofp_flow_removed_reason);
143 static void ofoperation_destroy(struct ofoperation *);
146 static void oftable_init(struct oftable *);
147 static void oftable_destroy(struct oftable *);
149 static void oftable_set_name(struct oftable *, const char *name);
151 static void oftable_disable_eviction(struct oftable *);
152 static void oftable_enable_eviction(struct oftable *,
153 const struct mf_subfield *fields,
156 static void oftable_remove_rule(struct rule *rule) OVS_REQUIRES(ofproto_mutex);
157 static void oftable_remove_rule__(struct ofproto *, struct rule *)
158 OVS_REQUIRES(ofproto_mutex);
159 static void oftable_insert_rule(struct rule *);
161 /* A set of rules within a single OpenFlow table (oftable) that have the same
162 * values for the oftable's eviction_fields. A rule to be evicted, when one is
163 * needed, is taken from the eviction group that contains the greatest number
166 * An oftable owns any number of eviction groups, each of which contains any
169 * Membership in an eviction group is imprecise, based on the hash of the
170 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
171 * That is, if two rules have different eviction_fields, but those
172 * eviction_fields hash to the same value, then they will belong to the same
173 * eviction_group anyway.
175 * (When eviction is not enabled on an oftable, we don't track any eviction
176 * groups, to save time and space.) */
177 struct eviction_group {
178 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
179 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
180 struct heap rules; /* Contains "struct rule"s. */
183 static bool choose_rule_to_evict(struct oftable *table, struct rule **rulep);
184 static void ofproto_evict(struct ofproto *) OVS_EXCLUDED(ofproto_mutex);
185 static uint32_t rule_eviction_priority(struct rule *);
186 static void eviction_group_add_rule(struct rule *);
187 static void eviction_group_remove_rule(struct rule *);
189 /* Criteria that flow_mod and other operations use for selecting rules on
190 * which to operate. */
191 struct rule_criteria {
192 /* An OpenFlow table or 255 for all tables. */
195 /* OpenFlow matching criteria. Interpreted different in "loose" way by
196 * collect_rules_loose() and "strict" way by collect_rules_strict(), as
197 * defined in the OpenFlow spec. */
200 /* Matching criteria for the OpenFlow cookie. Consider a bit B in a rule's
201 * cookie and the corresponding bits C in 'cookie' and M in 'cookie_mask'.
202 * The rule will not be selected if M is 1 and B != C. */
204 ovs_be64 cookie_mask;
206 /* Selection based on actions within a rule:
208 * If out_port != OFPP_ANY, selects only rules that output to out_port. */
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);
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 *);
242 static void ofproto_rule_destroy__(struct rule *);
243 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
244 static bool rule_is_modifiable(const struct rule *);
247 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
248 struct ofputil_flow_mod *,
249 const struct ofp_header *);
250 static enum ofperr modify_flows__(struct ofproto *, struct ofconn *,
251 struct ofputil_flow_mod *,
252 const struct ofp_header *,
253 const struct rule_collection *);
254 static void delete_flow__(struct rule *rule, struct ofopgroup *,
255 enum ofp_flow_removed_reason)
256 OVS_REQUIRES(ofproto_mutex);
257 static bool handle_openflow(struct ofconn *, const struct ofpbuf *);
258 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
259 struct ofputil_flow_mod *,
260 const struct ofp_header *)
261 OVS_EXCLUDED(ofproto_mutex);
262 static void calc_duration(long long int start, long long int now,
263 uint32_t *sec, uint32_t *nsec);
266 static uint64_t pick_datapath_id(const struct ofproto *);
267 static uint64_t pick_fallback_dpid(void);
268 static void ofproto_destroy__(struct ofproto *);
269 static void update_mtu(struct ofproto *, struct ofport *);
270 static void meter_delete(struct ofproto *, uint32_t first, uint32_t last);
273 static void ofproto_unixctl_init(void);
275 /* All registered ofproto classes, in probe order. */
276 static const struct ofproto_class **ofproto_classes;
277 static size_t n_ofproto_classes;
278 static size_t allocated_ofproto_classes;
280 /* Global lock that protects all flow table operations. */
281 struct ovs_mutex ofproto_mutex = OVS_MUTEX_INITIALIZER;
283 unsigned flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT;
284 unsigned n_handler_threads;
285 enum ofproto_flow_miss_model flow_miss_model = OFPROTO_HANDLE_MISS_AUTO;
287 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
288 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
290 /* Initial mappings of port to OpenFlow number mappings. */
291 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
293 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
295 /* The default value of true waits for flow restore. */
296 static bool flow_restore_wait = true;
298 /* Must be called to initialize the ofproto library.
300 * The caller may pass in 'iface_hints', which contains an shash of
301 * "iface_hint" elements indexed by the interface's name. The provider
302 * may use these hints to describe the startup configuration in order to
303 * reinitialize its state. The caller owns the provided data, so a
304 * provider will make copies of anything required. An ofproto provider
305 * will remove any existing state that is not described by the hint, and
306 * may choose to remove it all. */
308 ofproto_init(const struct shash *iface_hints)
310 struct shash_node *node;
313 ofproto_class_register(&ofproto_dpif_class);
315 /* Make a local copy, since we don't own 'iface_hints' elements. */
316 SHASH_FOR_EACH(node, iface_hints) {
317 const struct iface_hint *orig_hint = node->data;
318 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
319 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
321 new_hint->br_name = xstrdup(orig_hint->br_name);
322 new_hint->br_type = xstrdup(br_type);
323 new_hint->ofp_port = orig_hint->ofp_port;
325 shash_add(&init_ofp_ports, node->name, new_hint);
328 for (i = 0; i < n_ofproto_classes; i++) {
329 ofproto_classes[i]->init(&init_ofp_ports);
333 /* 'type' should be a normalized datapath type, as returned by
334 * ofproto_normalize_type(). Returns the corresponding ofproto_class
335 * structure, or a null pointer if there is none registered for 'type'. */
336 static const struct ofproto_class *
337 ofproto_class_find__(const char *type)
341 for (i = 0; i < n_ofproto_classes; i++) {
342 const struct ofproto_class *class = ofproto_classes[i];
347 class->enumerate_types(&types);
348 found = sset_contains(&types, type);
349 sset_destroy(&types);
355 VLOG_WARN("unknown datapath type %s", type);
359 /* Registers a new ofproto class. After successful registration, new ofprotos
360 * of that type can be created using ofproto_create(). */
362 ofproto_class_register(const struct ofproto_class *new_class)
366 for (i = 0; i < n_ofproto_classes; i++) {
367 if (ofproto_classes[i] == new_class) {
372 if (n_ofproto_classes >= allocated_ofproto_classes) {
373 ofproto_classes = x2nrealloc(ofproto_classes,
374 &allocated_ofproto_classes,
375 sizeof *ofproto_classes);
377 ofproto_classes[n_ofproto_classes++] = new_class;
381 /* Unregisters a datapath provider. 'type' must have been previously
382 * registered and not currently be in use by any ofprotos. After
383 * unregistration new datapaths of that type cannot be opened using
384 * ofproto_create(). */
386 ofproto_class_unregister(const struct ofproto_class *class)
390 for (i = 0; i < n_ofproto_classes; i++) {
391 if (ofproto_classes[i] == class) {
392 for (i++; i < n_ofproto_classes; i++) {
393 ofproto_classes[i - 1] = ofproto_classes[i];
399 VLOG_WARN("attempted to unregister an ofproto class that is not "
404 /* Clears 'types' and enumerates all registered ofproto types into it. The
405 * caller must first initialize the sset. */
407 ofproto_enumerate_types(struct sset *types)
411 for (i = 0; i < n_ofproto_classes; i++) {
412 ofproto_classes[i]->enumerate_types(types);
416 /* Returns the fully spelled out name for the given ofproto 'type'.
418 * Normalized type string can be compared with strcmp(). Unnormalized type
419 * string might be the same even if they have different spellings. */
421 ofproto_normalize_type(const char *type)
423 return type && type[0] ? type : "system";
426 /* Clears 'names' and enumerates the names of all known created ofprotos with
427 * the given 'type'. The caller must first initialize the sset. Returns 0 if
428 * successful, otherwise a positive errno value.
430 * Some kinds of datapaths might not be practically enumerable. This is not
431 * considered an error. */
433 ofproto_enumerate_names(const char *type, struct sset *names)
435 const struct ofproto_class *class = ofproto_class_find__(type);
436 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
440 ofproto_create(const char *datapath_name, const char *datapath_type,
441 struct ofproto **ofprotop)
443 const struct ofproto_class *class;
444 struct ofproto *ofproto;
450 ofproto_unixctl_init();
452 datapath_type = ofproto_normalize_type(datapath_type);
453 class = ofproto_class_find__(datapath_type);
455 VLOG_WARN("could not create datapath %s of unknown type %s",
456 datapath_name, datapath_type);
460 ofproto = class->alloc();
462 VLOG_ERR("failed to allocate datapath %s of type %s",
463 datapath_name, datapath_type);
468 ovs_mutex_lock(&ofproto_mutex);
469 memset(ofproto, 0, sizeof *ofproto);
470 ofproto->ofproto_class = class;
471 ofproto->name = xstrdup(datapath_name);
472 ofproto->type = xstrdup(datapath_type);
473 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
474 hash_string(ofproto->name, 0));
475 ofproto->datapath_id = 0;
476 ofproto->forward_bpdu = false;
477 ofproto->fallback_dpid = pick_fallback_dpid();
478 ofproto->mfr_desc = NULL;
479 ofproto->hw_desc = NULL;
480 ofproto->sw_desc = NULL;
481 ofproto->serial_desc = NULL;
482 ofproto->dp_desc = NULL;
483 ofproto->frag_handling = OFPC_FRAG_NORMAL;
484 hmap_init(&ofproto->ports);
485 shash_init(&ofproto->port_by_name);
486 simap_init(&ofproto->ofp_requests);
487 ofproto->max_ports = ofp_to_u16(OFPP_MAX);
488 ofproto->eviction_group_timer = LLONG_MIN;
489 ofproto->tables = NULL;
490 ofproto->n_tables = 0;
491 hindex_init(&ofproto->cookies);
492 list_init(&ofproto->expirable);
493 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
494 ofproto->state = S_OPENFLOW;
495 list_init(&ofproto->pending);
496 ofproto->n_pending = 0;
497 hmap_init(&ofproto->deletions);
498 guarded_list_init(&ofproto->rule_executes);
499 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
500 ofproto->first_op = ofproto->last_op = LLONG_MIN;
501 ofproto->next_op_report = LLONG_MAX;
502 ofproto->op_backoff = LLONG_MIN;
503 ofproto->vlan_bitmap = NULL;
504 ofproto->vlans_changed = false;
505 ofproto->min_mtu = INT_MAX;
506 ovs_mutex_unlock(&ofproto_mutex);
508 error = ofproto->ofproto_class->construct(ofproto);
510 VLOG_ERR("failed to open datapath %s: %s",
511 datapath_name, ovs_strerror(error));
512 ofproto_destroy__(ofproto);
516 /* The "max_ports" member should have been set by ->construct(ofproto).
517 * Port 0 is not a valid OpenFlow port, so mark that as unavailable. */
518 ofproto->ofp_port_ids = bitmap_allocate(ofproto->max_ports);
519 bitmap_set1(ofproto->ofp_port_ids, 0);
521 /* Check that hidden tables, if any, are at the end. */
522 ovs_assert(ofproto->n_tables);
523 for (i = 0; i + 1 < ofproto->n_tables; i++) {
524 enum oftable_flags flags = ofproto->tables[i].flags;
525 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
527 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
530 ofproto->datapath_id = pick_datapath_id(ofproto);
533 /* Initialize meters table. */
534 if (ofproto->ofproto_class->meter_get_features) {
535 ofproto->ofproto_class->meter_get_features(ofproto,
536 &ofproto->meter_features);
538 memset(&ofproto->meter_features, 0, sizeof ofproto->meter_features);
540 ofproto->meters = xzalloc((ofproto->meter_features.max_meters + 1)
541 * sizeof(struct meter *));
547 /* Must be called (only) by an ofproto implementation in its constructor
548 * function. See the large comment on 'construct' in struct ofproto_class for
551 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
553 struct oftable *table;
555 ovs_assert(!ofproto->n_tables);
556 ovs_assert(n_tables >= 1 && n_tables <= 255);
558 ofproto->n_tables = n_tables;
559 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
560 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
565 /* To be optionally called (only) by an ofproto implementation in its
566 * constructor function. See the large comment on 'construct' in struct
567 * ofproto_class for details.
569 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
570 * will then ensure that actions passed into the ofproto implementation will
571 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
572 * function is not called, there will be no such restriction.
574 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
575 * the 'max_ports' restriction. */
577 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
579 ovs_assert(max_ports <= ofp_to_u16(OFPP_MAX));
580 ofproto->max_ports = max_ports;
584 ofproto_get_datapath_id(const struct ofproto *ofproto)
586 return ofproto->datapath_id;
590 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
592 uint64_t old_dpid = p->datapath_id;
593 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
594 if (p->datapath_id != old_dpid) {
595 /* Force all active connections to reconnect, since there is no way to
596 * notify a controller that the datapath ID has changed. */
597 ofproto_reconnect_controllers(p);
602 ofproto_set_controllers(struct ofproto *p,
603 const struct ofproto_controller *controllers,
604 size_t n_controllers, uint32_t allowed_versions)
606 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
611 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
613 connmgr_set_fail_mode(p->connmgr, fail_mode);
616 /* Drops the connections between 'ofproto' and all of its controllers, forcing
617 * them to reconnect. */
619 ofproto_reconnect_controllers(struct ofproto *ofproto)
621 connmgr_reconnect(ofproto->connmgr);
624 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
625 * in-band control should guarantee access, in the same way that in-band
626 * control guarantees access to OpenFlow controllers. */
628 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
629 const struct sockaddr_in *extras, size_t n)
631 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
634 /* Sets the OpenFlow queue used by flows set up by in-band control on
635 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
636 * flows will use the default queue. */
638 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
640 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
643 /* Sets the number of flows at which eviction from the kernel flow table
646 ofproto_set_flow_eviction_threshold(unsigned threshold)
648 flow_eviction_threshold = MAX(OFPROTO_FLOW_EVICTION_THRESHOLD_MIN,
652 /* Sets the path for handling flow misses. */
654 ofproto_set_flow_miss_model(unsigned model)
656 flow_miss_model = model;
659 /* If forward_bpdu is true, the NORMAL action will forward frames with
660 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
661 * the NORMAL action will drop these frames. */
663 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
665 bool old_val = ofproto->forward_bpdu;
666 ofproto->forward_bpdu = forward_bpdu;
667 if (old_val != ofproto->forward_bpdu) {
668 if (ofproto->ofproto_class->forward_bpdu_changed) {
669 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
674 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
675 * 'idle_time', in seconds, and the maximum number of MAC table entries to
678 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
681 if (ofproto->ofproto_class->set_mac_table_config) {
682 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
687 /* Sets number of upcall handler threads. The default is
688 * (number of online cores - 2). */
690 ofproto_set_n_handler_threads(unsigned limit)
693 n_handler_threads = limit;
695 int n_proc = sysconf(_SC_NPROCESSORS_ONLN);
696 n_handler_threads = n_proc > 2 ? n_proc - 2 : 1;
701 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
704 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
708 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
710 return connmgr_set_snoops(ofproto->connmgr, snoops);
714 ofproto_set_netflow(struct ofproto *ofproto,
715 const struct netflow_options *nf_options)
717 if (nf_options && sset_is_empty(&nf_options->collectors)) {
721 if (ofproto->ofproto_class->set_netflow) {
722 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
724 return nf_options ? EOPNOTSUPP : 0;
729 ofproto_set_sflow(struct ofproto *ofproto,
730 const struct ofproto_sflow_options *oso)
732 if (oso && sset_is_empty(&oso->targets)) {
736 if (ofproto->ofproto_class->set_sflow) {
737 return ofproto->ofproto_class->set_sflow(ofproto, oso);
739 return oso ? EOPNOTSUPP : 0;
744 ofproto_set_ipfix(struct ofproto *ofproto,
745 const struct ofproto_ipfix_bridge_exporter_options *bo,
746 const struct ofproto_ipfix_flow_exporter_options *fo,
749 if (ofproto->ofproto_class->set_ipfix) {
750 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
752 return (bo || fo) ? EOPNOTSUPP : 0;
757 ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
759 flow_restore_wait = flow_restore_wait_db;
763 ofproto_get_flow_restore_wait(void)
765 return flow_restore_wait;
769 /* Spanning Tree Protocol (STP) configuration. */
771 /* Configures STP on 'ofproto' using the settings defined in 's'. If
772 * 's' is NULL, disables STP.
774 * Returns 0 if successful, otherwise a positive errno value. */
776 ofproto_set_stp(struct ofproto *ofproto,
777 const struct ofproto_stp_settings *s)
779 return (ofproto->ofproto_class->set_stp
780 ? ofproto->ofproto_class->set_stp(ofproto, s)
784 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
785 * 'enabled' member of 's' is false, then the other members are not
788 * Returns 0 if successful, otherwise a positive errno value. */
790 ofproto_get_stp_status(struct ofproto *ofproto,
791 struct ofproto_stp_status *s)
793 return (ofproto->ofproto_class->get_stp_status
794 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
798 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
799 * in 's'. The caller is responsible for assigning STP port numbers
800 * (using the 'port_num' member in the range of 1 through 255, inclusive)
801 * and ensuring there are no duplicates. If the 's' is NULL, then STP
802 * is disabled on the port.
804 * Returns 0 if successful, otherwise a positive errno value.*/
806 ofproto_port_set_stp(struct ofproto *ofproto, ofp_port_t ofp_port,
807 const struct ofproto_port_stp_settings *s)
809 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
811 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
812 ofproto->name, ofp_port);
816 return (ofproto->ofproto_class->set_stp_port
817 ? ofproto->ofproto_class->set_stp_port(ofport, s)
821 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
822 * 's'. If the 'enabled' member in 's' is false, then the other members
823 * are not meaningful.
825 * Returns 0 if successful, otherwise a positive errno value.*/
827 ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
828 struct ofproto_port_stp_status *s)
830 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
832 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
833 "port %"PRIu16, ofproto->name, ofp_port);
837 return (ofproto->ofproto_class->get_stp_port_status
838 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
842 /* Queue DSCP configuration. */
844 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
845 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
846 * to implement QoS. Instead, it is used to implement features which require
847 * knowledge of what queues exist on a port, and some basic information about
850 * Returns 0 if successful, otherwise a positive errno value. */
852 ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port,
853 const struct ofproto_port_queue *queues,
856 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
859 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
860 ofproto->name, ofp_port);
864 return (ofproto->ofproto_class->set_queues
865 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
869 /* Connectivity Fault Management configuration. */
871 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
873 ofproto_port_clear_cfm(struct ofproto *ofproto, ofp_port_t ofp_port)
875 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
876 if (ofport && ofproto->ofproto_class->set_cfm) {
877 ofproto->ofproto_class->set_cfm(ofport, NULL);
881 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
882 * basic configuration from the configuration members in 'cfm', and the remote
883 * maintenance point ID from remote_mpid. Ignores the statistics members of
886 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
888 ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port,
889 const struct cfm_settings *s)
891 struct ofport *ofport;
894 ofport = ofproto_get_port(ofproto, ofp_port);
896 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
897 ofproto->name, ofp_port);
901 /* XXX: For configuration simplicity, we only support one remote_mpid
902 * outside of the CFM module. It's not clear if this is the correct long
903 * term solution or not. */
904 error = (ofproto->ofproto_class->set_cfm
905 ? ofproto->ofproto_class->set_cfm(ofport, s)
908 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
909 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
910 ovs_strerror(error));
914 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
915 * 'ofproto' does not have a port 'ofp_port'. */
917 ofproto_port_set_bfd(struct ofproto *ofproto, ofp_port_t ofp_port,
918 const struct smap *cfg)
920 struct ofport *ofport;
923 ofport = ofproto_get_port(ofproto, ofp_port);
925 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
926 ofproto->name, ofp_port);
930 error = (ofproto->ofproto_class->set_bfd
931 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
934 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
935 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
936 ovs_strerror(error));
940 /* Populates 'status' with key value pairs indicating the status of the BFD
941 * session on 'ofp_port'. This information is intended to be populated in the
942 * OVS database. Has no effect if 'ofp_port' is not na OpenFlow port in
945 ofproto_port_get_bfd_status(struct ofproto *ofproto, ofp_port_t ofp_port,
948 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
949 return (ofport && ofproto->ofproto_class->get_bfd_status
950 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
954 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
955 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
956 * 0 if LACP partner information is not current (generally indicating a
957 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
959 ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port)
961 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
962 return (ofport && ofproto->ofproto_class->port_is_lacp_current
963 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
969 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
970 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
971 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
972 * configuration plus, if there is more than one slave, a bonding
975 * If 'aux' is already registered then this function updates its configuration
976 * to 's'. Otherwise, this function registers a new bundle.
978 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
981 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
982 const struct ofproto_bundle_settings *s)
984 return (ofproto->ofproto_class->bundle_set
985 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
989 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
990 * If no such bundle has been registered, this has no effect. */
992 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
994 return ofproto_bundle_register(ofproto, aux, NULL);
998 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
999 * If 'aux' is already registered then this function updates its configuration
1000 * to 's'. Otherwise, this function registers a new mirror. */
1002 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
1003 const struct ofproto_mirror_settings *s)
1005 return (ofproto->ofproto_class->mirror_set
1006 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
1010 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
1011 * If no mirror has been registered, this has no effect. */
1013 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
1015 return ofproto_mirror_register(ofproto, aux, NULL);
1018 /* Retrieves statistics from mirror associated with client data pointer
1019 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
1020 * 'bytes', respectively. If a particular counters is not supported,
1021 * the appropriate argument is set to UINT64_MAX. */
1023 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
1024 uint64_t *packets, uint64_t *bytes)
1026 if (!ofproto->ofproto_class->mirror_get_stats) {
1027 *packets = *bytes = UINT64_MAX;
1031 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
1035 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
1036 * which all packets are flooded, instead of using MAC learning. If
1037 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1039 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
1042 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
1044 return (ofproto->ofproto_class->set_flood_vlans
1045 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
1049 /* Returns true if 'aux' is a registered bundle that is currently in use as the
1050 * output for a mirror. */
1052 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
1054 return (ofproto->ofproto_class->is_mirror_output_bundle
1055 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
1059 /* Configuration of OpenFlow tables. */
1061 /* Returns the number of OpenFlow tables in 'ofproto'. */
1063 ofproto_get_n_tables(const struct ofproto *ofproto)
1065 return ofproto->n_tables;
1068 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
1069 * settings from 's'. 'table_id' must be in the range 0 through the number of
1070 * OpenFlow tables in 'ofproto' minus 1, inclusive.
1072 * For read-only tables, only the name may be configured. */
1074 ofproto_configure_table(struct ofproto *ofproto, int table_id,
1075 const struct ofproto_table_settings *s)
1077 struct oftable *table;
1079 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
1080 table = &ofproto->tables[table_id];
1082 oftable_set_name(table, s->name);
1084 if (table->flags & OFTABLE_READONLY) {
1089 oftable_enable_eviction(table, s->groups, s->n_groups);
1091 oftable_disable_eviction(table);
1094 table->max_flows = s->max_flows;
1095 ovs_rwlock_rdlock(&table->cls.rwlock);
1096 if (classifier_count(&table->cls) > table->max_flows
1097 && table->eviction_fields) {
1098 /* 'table' contains more flows than allowed. We might not be able to
1099 * evict them right away because of the asynchronous nature of flow
1100 * table changes. Schedule eviction for later. */
1101 switch (ofproto->state) {
1103 ofproto->state = S_EVICT;
1107 /* We're already deleting flows, nothing more to do. */
1111 ovs_rwlock_unlock(&table->cls.rwlock);
1115 ofproto_has_snoops(const struct ofproto *ofproto)
1117 return connmgr_has_snoops(ofproto->connmgr);
1121 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1123 connmgr_get_snoops(ofproto->connmgr, snoops);
1127 ofproto_rule_delete__(struct ofproto *ofproto, struct rule *rule)
1128 OVS_REQUIRES(ofproto_mutex)
1130 struct ofopgroup *group;
1132 ovs_assert(!rule->pending);
1134 group = ofopgroup_create_unattached(ofproto);
1135 delete_flow__(rule, group, OFPRR_DELETE);
1136 ofopgroup_submit(group);
1139 /* Deletes 'rule' from 'cls' within 'ofproto'.
1141 * Within an ofproto implementation, this function allows an ofproto
1142 * implementation to destroy any rules that remain when its ->destruct()
1143 * function is called. This function is not suitable for use elsewhere in an
1144 * ofproto implementation.
1146 * This function implements steps 4.4 and 4.5 in the section titled "Rule Life
1147 * Cycle" in ofproto-provider.h. */
1149 ofproto_rule_delete(struct ofproto *ofproto, struct rule *rule)
1150 OVS_EXCLUDED(ofproto_mutex)
1152 struct ofopgroup *group;
1154 ovs_mutex_lock(&ofproto_mutex);
1155 ovs_assert(!rule->pending);
1157 group = ofopgroup_create_unattached(ofproto);
1158 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1159 oftable_remove_rule__(ofproto, rule);
1160 ofproto->ofproto_class->rule_delete(rule);
1161 ofopgroup_submit(group);
1163 ovs_mutex_unlock(&ofproto_mutex);
1167 ofproto_flush__(struct ofproto *ofproto)
1168 OVS_EXCLUDED(ofproto_mutex)
1170 struct oftable *table;
1172 if (ofproto->ofproto_class->flush) {
1173 ofproto->ofproto_class->flush(ofproto);
1176 ovs_mutex_lock(&ofproto_mutex);
1177 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1178 struct rule *rule, *next_rule;
1179 struct cls_cursor cursor;
1181 if (table->flags & OFTABLE_HIDDEN) {
1185 ovs_rwlock_rdlock(&table->cls.rwlock);
1186 cls_cursor_init(&cursor, &table->cls, NULL);
1187 ovs_rwlock_unlock(&table->cls.rwlock);
1188 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1189 if (!rule->pending) {
1190 ofproto_rule_delete__(ofproto, rule);
1194 ovs_mutex_unlock(&ofproto_mutex);
1198 ofproto_destroy__(struct ofproto *ofproto)
1199 OVS_EXCLUDED(ofproto_mutex)
1201 struct oftable *table;
1203 ovs_assert(list_is_empty(&ofproto->pending));
1205 destroy_rule_executes(ofproto);
1206 guarded_list_destroy(&ofproto->rule_executes);
1208 connmgr_destroy(ofproto->connmgr);
1210 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1211 free(ofproto->name);
1212 free(ofproto->type);
1213 free(ofproto->mfr_desc);
1214 free(ofproto->hw_desc);
1215 free(ofproto->sw_desc);
1216 free(ofproto->serial_desc);
1217 free(ofproto->dp_desc);
1218 hmap_destroy(&ofproto->ports);
1219 shash_destroy(&ofproto->port_by_name);
1220 bitmap_free(ofproto->ofp_port_ids);
1221 simap_destroy(&ofproto->ofp_requests);
1223 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1224 oftable_destroy(table);
1226 free(ofproto->tables);
1228 hmap_destroy(&ofproto->deletions);
1230 free(ofproto->vlan_bitmap);
1232 ofproto->ofproto_class->dealloc(ofproto);
1236 ofproto_destroy(struct ofproto *p)
1237 OVS_EXCLUDED(ofproto_mutex)
1239 struct ofport *ofport, *next_ofport;
1246 meter_delete(p, 1, p->meter_features.max_meters);
1247 p->meter_features.max_meters = 0;
1253 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1254 ofport_destroy(ofport);
1257 p->ofproto_class->destruct(p);
1258 ofproto_destroy__(p);
1261 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1262 * kernel datapath, for example, this destroys the datapath in the kernel, and
1263 * with the netdev-based datapath, it tears down the data structures that
1264 * represent the datapath.
1266 * The datapath should not be currently open as an ofproto. */
1268 ofproto_delete(const char *name, const char *type)
1270 const struct ofproto_class *class = ofproto_class_find__(type);
1271 return (!class ? EAFNOSUPPORT
1272 : !class->del ? EACCES
1273 : class->del(type, name));
1277 process_port_change(struct ofproto *ofproto, int error, char *devname)
1279 if (error == ENOBUFS) {
1280 reinit_ports(ofproto);
1281 } else if (!error) {
1282 update_port(ofproto, devname);
1288 ofproto_type_run(const char *datapath_type)
1290 const struct ofproto_class *class;
1293 datapath_type = ofproto_normalize_type(datapath_type);
1294 class = ofproto_class_find__(datapath_type);
1296 error = class->type_run ? class->type_run(datapath_type) : 0;
1297 if (error && error != EAGAIN) {
1298 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1299 datapath_type, ovs_strerror(error));
1305 ofproto_type_run_fast(const char *datapath_type)
1307 const struct ofproto_class *class;
1310 datapath_type = ofproto_normalize_type(datapath_type);
1311 class = ofproto_class_find__(datapath_type);
1313 error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
1314 if (error && error != EAGAIN) {
1315 VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
1316 datapath_type, ovs_strerror(error));
1322 ofproto_type_wait(const char *datapath_type)
1324 const struct ofproto_class *class;
1326 datapath_type = ofproto_normalize_type(datapath_type);
1327 class = ofproto_class_find__(datapath_type);
1329 if (class->type_wait) {
1330 class->type_wait(datapath_type);
1335 any_pending_ops(const struct ofproto *p)
1336 OVS_EXCLUDED(ofproto_mutex)
1340 ovs_mutex_lock(&ofproto_mutex);
1341 b = !list_is_empty(&p->pending);
1342 ovs_mutex_unlock(&ofproto_mutex);
1348 ofproto_run(struct ofproto *p)
1350 struct sset changed_netdevs;
1351 const char *changed_netdev;
1352 struct ofport *ofport;
1355 error = p->ofproto_class->run(p);
1356 if (error && error != EAGAIN) {
1357 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, ovs_strerror(error));
1360 run_rule_executes(p);
1362 /* Restore the eviction group heap invariant occasionally. */
1363 if (p->eviction_group_timer < time_msec()) {
1366 p->eviction_group_timer = time_msec() + 1000;
1368 for (i = 0; i < p->n_tables; i++) {
1369 struct oftable *table = &p->tables[i];
1370 struct eviction_group *evg;
1371 struct cls_cursor cursor;
1374 if (!table->eviction_fields) {
1378 ovs_mutex_lock(&ofproto_mutex);
1379 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
1380 heap_rebuild(&evg->rules);
1383 ovs_rwlock_rdlock(&table->cls.rwlock);
1384 cls_cursor_init(&cursor, &table->cls, NULL);
1385 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1386 if (!rule->eviction_group
1387 && (rule->idle_timeout || rule->hard_timeout)) {
1388 eviction_group_add_rule(rule);
1391 ovs_rwlock_unlock(&table->cls.rwlock);
1392 ovs_mutex_unlock(&ofproto_mutex);
1396 if (p->ofproto_class->port_poll) {
1399 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1400 process_port_change(p, error, devname);
1404 /* Update OpenFlow port status for any port whose netdev has changed.
1406 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1407 * destroyed, so it's not safe to update ports directly from the
1408 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1409 * need this two-phase approach. */
1410 sset_init(&changed_netdevs);
1411 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1412 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1413 if (ofport->change_seq != change_seq) {
1414 ofport->change_seq = change_seq;
1415 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1418 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1419 update_port(p, changed_netdev);
1421 sset_destroy(&changed_netdevs);
1425 connmgr_run(p->connmgr, handle_openflow);
1429 connmgr_run(p->connmgr, NULL);
1431 if (!any_pending_ops(p)) {
1432 p->state = S_OPENFLOW;
1437 connmgr_run(p->connmgr, NULL);
1439 if (!any_pending_ops(p)) {
1440 connmgr_flushed(p->connmgr);
1441 p->state = S_OPENFLOW;
1449 if (time_msec() >= p->next_op_report) {
1450 long long int ago = (time_msec() - p->first_op) / 1000;
1451 long long int interval = (p->last_op - p->first_op) / 1000;
1455 ds_put_format(&s, "%d flow_mods ",
1456 p->n_add + p->n_delete + p->n_modify);
1457 if (interval == ago) {
1458 ds_put_format(&s, "in the last %lld s", ago);
1459 } else if (interval) {
1460 ds_put_format(&s, "in the %lld s starting %lld s ago",
1463 ds_put_format(&s, "%lld s ago", ago);
1466 ds_put_cstr(&s, " (");
1468 ds_put_format(&s, "%d adds, ", p->n_add);
1471 ds_put_format(&s, "%d deletes, ", p->n_delete);
1474 ds_put_format(&s, "%d modifications, ", p->n_modify);
1477 ds_put_char(&s, ')');
1479 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1482 p->n_add = p->n_delete = p->n_modify = 0;
1483 p->next_op_report = LLONG_MAX;
1489 /* Performs periodic activity required by 'ofproto' that needs to be done
1490 * with the least possible latency.
1492 * It makes sense to call this function a couple of times per poll loop, to
1493 * provide a significant performance boost on some benchmarks with the
1494 * ofproto-dpif implementation. */
1496 ofproto_run_fast(struct ofproto *p)
1500 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1501 if (error && error != EAGAIN) {
1502 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1503 p->name, ovs_strerror(error));
1509 ofproto_wait(struct ofproto *p)
1511 struct ofport *ofport;
1513 p->ofproto_class->wait(p);
1514 if (p->ofproto_class->port_poll_wait) {
1515 p->ofproto_class->port_poll_wait(p);
1518 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1519 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1520 poll_immediate_wake();
1526 connmgr_wait(p->connmgr, true);
1531 connmgr_wait(p->connmgr, false);
1532 if (!any_pending_ops(p)) {
1533 poll_immediate_wake();
1540 ofproto_is_alive(const struct ofproto *p)
1542 return connmgr_has_controllers(p->connmgr);
1545 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1546 * memory_report(). */
1548 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1550 const struct oftable *table;
1551 unsigned int n_rules;
1553 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1555 ovs_mutex_lock(&ofproto_mutex);
1556 simap_increase(usage, "ops",
1557 ofproto->n_pending + hmap_count(&ofproto->deletions));
1558 ovs_mutex_unlock(&ofproto_mutex);
1561 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1562 ovs_rwlock_rdlock(&table->cls.rwlock);
1563 n_rules += classifier_count(&table->cls);
1564 ovs_rwlock_unlock(&table->cls.rwlock);
1566 simap_increase(usage, "rules", n_rules);
1568 if (ofproto->ofproto_class->get_memory_usage) {
1569 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1572 connmgr_get_memory_usage(ofproto->connmgr, usage);
1576 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1579 connmgr_get_controller_info(ofproto->connmgr, info);
1583 ofproto_free_ofproto_controller_info(struct shash *info)
1585 connmgr_free_controller_info(info);
1588 /* Makes a deep copy of 'old' into 'port'. */
1590 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1592 port->name = xstrdup(old->name);
1593 port->type = xstrdup(old->type);
1594 port->ofp_port = old->ofp_port;
1597 /* Frees memory allocated to members of 'ofproto_port'.
1599 * Do not call this function on an ofproto_port obtained from
1600 * ofproto_port_dump_next(): that function retains ownership of the data in the
1603 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1605 free(ofproto_port->name);
1606 free(ofproto_port->type);
1609 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1611 * This function provides no status indication. An error status for the entire
1612 * dump operation is provided when it is completed by calling
1613 * ofproto_port_dump_done().
1616 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1617 const struct ofproto *ofproto)
1619 dump->ofproto = ofproto;
1620 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1624 /* Attempts to retrieve another port from 'dump', which must have been created
1625 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1626 * 'port' and returns true. On failure, returns false.
1628 * Failure might indicate an actual error or merely that the last port has been
1629 * dumped. An error status for the entire dump operation is provided when it
1630 * is completed by calling ofproto_port_dump_done().
1632 * The ofproto owns the data stored in 'port'. It will remain valid until at
1633 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1634 * ofproto_port_dump_done(). */
1636 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1637 struct ofproto_port *port)
1639 const struct ofproto *ofproto = dump->ofproto;
1645 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1648 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1654 /* Completes port table dump operation 'dump', which must have been created
1655 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1656 * error-free, otherwise a positive errno value describing the problem. */
1658 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1660 const struct ofproto *ofproto = dump->ofproto;
1662 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1665 return dump->error == EOF ? 0 : dump->error;
1668 /* Returns the type to pass to netdev_open() when a datapath of type
1669 * 'datapath_type' has a port of type 'port_type', for a few special
1670 * cases when a netdev type differs from a port type. For example, when
1671 * using the userspace datapath, a port of type "internal" needs to be
1674 * Returns either 'type' itself or a string literal, which must not be
1677 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1679 const struct ofproto_class *class;
1681 datapath_type = ofproto_normalize_type(datapath_type);
1682 class = ofproto_class_find__(datapath_type);
1687 return (class->port_open_type
1688 ? class->port_open_type(datapath_type, port_type)
1692 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1693 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1694 * the port's OpenFlow port number.
1696 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1697 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1698 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1699 * 'ofp_portp' is non-null). */
1701 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1702 ofp_port_t *ofp_portp)
1704 ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1707 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1709 const char *netdev_name = netdev_get_name(netdev);
1711 simap_put(&ofproto->ofp_requests, netdev_name,
1712 ofp_to_u16(ofp_port));
1713 update_port(ofproto, netdev_name);
1716 struct ofproto_port ofproto_port;
1718 ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
1720 *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
1721 ofproto_port_destroy(&ofproto_port);
1726 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1727 * initializes '*port' appropriately; on failure, returns a positive errno
1730 * The caller owns the data in 'ofproto_port' and must free it with
1731 * ofproto_port_destroy() when it is no longer needed. */
1733 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1734 struct ofproto_port *port)
1738 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1740 memset(port, 0, sizeof *port);
1745 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1746 * Returns 0 if successful, otherwise a positive errno. */
1748 ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port)
1750 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1751 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1752 struct simap_node *ofp_request_node;
1755 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1756 if (ofp_request_node) {
1757 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1760 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1761 if (!error && ofport) {
1762 /* 'name' is the netdev's name and update_port() is going to close the
1763 * netdev. Just in case update_port() refers to 'name' after it
1764 * destroys 'ofport', make a copy of it around the update_port()
1766 char *devname = xstrdup(name);
1767 update_port(ofproto, devname);
1774 simple_flow_mod(struct ofproto *ofproto,
1775 const struct match *match, unsigned int priority,
1776 const struct ofpact *ofpacts, size_t ofpacts_len,
1777 enum ofp_flow_mod_command command)
1779 struct ofputil_flow_mod fm;
1781 memset(&fm, 0, sizeof fm);
1783 fm.priority = priority;
1786 fm.modify_cookie = false;
1788 fm.command = command;
1789 fm.idle_timeout = 0;
1790 fm.hard_timeout = 0;
1791 fm.buffer_id = UINT32_MAX;
1792 fm.out_port = OFPP_ANY;
1794 fm.ofpacts = CONST_CAST(struct ofpact *, ofpacts);
1795 fm.ofpacts_len = ofpacts_len;
1797 return handle_flow_mod__(ofproto, NULL, &fm, NULL);
1800 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1801 * performs the 'n_actions' actions in 'actions'. The new flow will not
1804 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1805 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1806 * controllers; otherwise, it will be hidden.
1808 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1810 * This is a helper function for in-band control and fail-open. */
1812 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1813 unsigned int priority,
1814 const struct ofpact *ofpacts, size_t ofpacts_len)
1815 OVS_EXCLUDED(ofproto_mutex)
1817 const struct rule *rule;
1820 /* First do a cheap check whether the rule we're looking for already exists
1821 * with the actions that we want. If it does, then we're done. */
1822 ovs_rwlock_rdlock(&ofproto->tables[0].cls.rwlock);
1823 rule = rule_from_cls_rule(classifier_find_match_exactly(
1824 &ofproto->tables[0].cls, match, priority));
1826 ovs_mutex_lock(&rule->mutex);
1827 must_add = !ofpacts_equal(rule->actions->ofpacts,
1828 rule->actions->ofpacts_len,
1829 ofpacts, ofpacts_len);
1830 ovs_mutex_unlock(&rule->mutex);
1834 ovs_rwlock_unlock(&ofproto->tables[0].cls.rwlock);
1836 /* If there's no such rule or the rule doesn't have the actions we want,
1837 * fall back to a executing a full flow mod. We can't optimize this at
1838 * all because we didn't take enough locks above to ensure that the flow
1839 * table didn't already change beneath us. */
1841 simple_flow_mod(ofproto, match, priority, ofpacts, ofpacts_len,
1842 OFPFC_MODIFY_STRICT);
1846 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1847 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1848 * operation cannot be initiated now but may be retried later.
1850 * This is a helper function for in-band control and fail-open and the "learn"
1853 ofproto_flow_mod(struct ofproto *ofproto, struct ofputil_flow_mod *fm)
1854 OVS_EXCLUDED(ofproto_mutex)
1856 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1859 /* Searches for a rule with matching criteria exactly equal to 'target' in
1860 * ofproto's table 0 and, if it finds one, deletes it.
1862 * This is a helper function for in-band control and fail-open. */
1864 ofproto_delete_flow(struct ofproto *ofproto,
1865 const struct match *target, unsigned int priority)
1866 OVS_EXCLUDED(ofproto_mutex)
1868 struct classifier *cls = &ofproto->tables[0].cls;
1871 /* First do a cheap check whether the rule we're looking for has already
1872 * been deleted. If so, then we're done. */
1873 ovs_rwlock_rdlock(&cls->rwlock);
1874 rule = rule_from_cls_rule(classifier_find_match_exactly(cls, target,
1876 ovs_rwlock_unlock(&cls->rwlock);
1881 /* Fall back to a executing a full flow mod. We can't optimize this at all
1882 * because we didn't take enough locks above to ensure that the flow table
1883 * didn't already change beneath us. */
1884 return simple_flow_mod(ofproto, target, priority, NULL, 0,
1885 OFPFC_DELETE_STRICT) != OFPROTO_POSTPONE;
1888 /* Starts the process of deleting all of the flows from all of ofproto's flow
1889 * tables and then reintroducing the flows required by in-band control and
1890 * fail-open. The process will complete in a later call to ofproto_run(). */
1892 ofproto_flush_flows(struct ofproto *ofproto)
1894 COVERAGE_INC(ofproto_flush);
1895 ofproto->state = S_FLUSH;
1899 reinit_ports(struct ofproto *p)
1901 struct ofproto_port_dump dump;
1902 struct sset devnames;
1903 struct ofport *ofport;
1904 struct ofproto_port ofproto_port;
1905 const char *devname;
1907 COVERAGE_INC(ofproto_reinit_ports);
1909 sset_init(&devnames);
1910 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1911 sset_add(&devnames, netdev_get_name(ofport->netdev));
1913 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1914 sset_add(&devnames, ofproto_port.name);
1917 SSET_FOR_EACH (devname, &devnames) {
1918 update_port(p, devname);
1920 sset_destroy(&devnames);
1924 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
1928 port_idx = simap_get(&ofproto->ofp_requests, netdev_name);
1929 port_idx = port_idx ? port_idx : UINT16_MAX;
1931 if (port_idx >= ofproto->max_ports
1932 || bitmap_is_set(ofproto->ofp_port_ids, port_idx)) {
1933 uint16_t end_port_no = ofproto->alloc_port_no;
1935 /* Search for a free OpenFlow port number. We try not to
1936 * immediately reuse them to prevent problems due to old
1939 if (++ofproto->alloc_port_no >= ofproto->max_ports) {
1940 ofproto->alloc_port_no = 0;
1942 if (!bitmap_is_set(ofproto->ofp_port_ids,
1943 ofproto->alloc_port_no)) {
1944 port_idx = ofproto->alloc_port_no;
1947 if (ofproto->alloc_port_no == end_port_no) {
1952 bitmap_set1(ofproto->ofp_port_ids, port_idx);
1953 return u16_to_ofp(port_idx);
1957 dealloc_ofp_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
1959 if (ofp_to_u16(ofp_port) < ofproto->max_ports) {
1960 bitmap_set0(ofproto->ofp_port_ids, ofp_to_u16(ofp_port));
1964 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1965 * pointer if the netdev cannot be opened. On success, also fills in
1967 static struct netdev *
1968 ofport_open(struct ofproto *ofproto,
1969 struct ofproto_port *ofproto_port,
1970 struct ofputil_phy_port *pp)
1972 enum netdev_flags flags;
1973 struct netdev *netdev;
1976 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1978 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
1979 "cannot be opened (%s)",
1981 ofproto_port->name, ofproto_port->ofp_port,
1982 ofproto_port->name, ovs_strerror(error));
1986 if (ofproto_port->ofp_port == OFPP_NONE) {
1987 if (!strcmp(ofproto->name, ofproto_port->name)) {
1988 ofproto_port->ofp_port = OFPP_LOCAL;
1990 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
1991 ofproto_port->name);
1994 pp->port_no = ofproto_port->ofp_port;
1995 netdev_get_etheraddr(netdev, pp->hw_addr);
1996 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
1997 netdev_get_flags(netdev, &flags);
1998 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
1999 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
2000 netdev_get_features(netdev, &pp->curr, &pp->advertised,
2001 &pp->supported, &pp->peer);
2002 pp->curr_speed = netdev_features_to_bps(pp->curr, 0) / 1000;
2003 pp->max_speed = netdev_features_to_bps(pp->supported, 0) / 1000;
2008 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
2009 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
2012 ofport_equal(const struct ofputil_phy_port *a,
2013 const struct ofputil_phy_port *b)
2015 return (eth_addr_equals(a->hw_addr, b->hw_addr)
2016 && a->state == b->state
2017 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
2018 && a->curr == b->curr
2019 && a->advertised == b->advertised
2020 && a->supported == b->supported
2021 && a->peer == b->peer
2022 && a->curr_speed == b->curr_speed
2023 && a->max_speed == b->max_speed);
2026 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
2027 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
2028 * one with the same name or port number). */
2030 ofport_install(struct ofproto *p,
2031 struct netdev *netdev, const struct ofputil_phy_port *pp)
2033 const char *netdev_name = netdev_get_name(netdev);
2034 struct ofport *ofport;
2037 /* Create ofport. */
2038 ofport = p->ofproto_class->port_alloc();
2043 ofport->ofproto = p;
2044 ofport->netdev = netdev;
2045 ofport->change_seq = netdev_change_seq(netdev);
2047 ofport->ofp_port = pp->port_no;
2048 ofport->created = time_msec();
2050 /* Add port to 'p'. */
2051 hmap_insert(&p->ports, &ofport->hmap_node,
2052 hash_ofp_port(ofport->ofp_port));
2053 shash_add(&p->port_by_name, netdev_name, ofport);
2055 update_mtu(p, ofport);
2057 /* Let the ofproto_class initialize its private data. */
2058 error = p->ofproto_class->port_construct(ofport);
2062 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
2066 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
2067 p->name, netdev_name, ovs_strerror(error));
2069 ofport_destroy__(ofport);
2071 netdev_close(netdev);
2075 /* Removes 'ofport' from 'p' and destroys it. */
2077 ofport_remove(struct ofport *ofport)
2079 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
2081 ofport_destroy(ofport);
2084 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
2087 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
2089 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
2091 ofport_remove(port);
2095 /* Updates 'port' with new 'pp' description.
2097 * Does not handle a name or port number change. The caller must implement
2098 * such a change as a delete followed by an add. */
2100 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
2102 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
2103 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
2104 | (pp->config & OFPUTIL_PC_PORT_DOWN));
2105 port->pp.state = pp->state;
2106 port->pp.curr = pp->curr;
2107 port->pp.advertised = pp->advertised;
2108 port->pp.supported = pp->supported;
2109 port->pp.peer = pp->peer;
2110 port->pp.curr_speed = pp->curr_speed;
2111 port->pp.max_speed = pp->max_speed;
2113 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
2116 /* Update OpenFlow 'state' in 'port' and notify controller. */
2118 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
2120 if (port->pp.state != state) {
2121 port->pp.state = state;
2122 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
2128 ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port)
2130 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
2132 if (port->ofproto->ofproto_class->set_realdev) {
2133 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
2135 if (port->ofproto->ofproto_class->set_stp_port) {
2136 port->ofproto->ofproto_class->set_stp_port(port, NULL);
2138 if (port->ofproto->ofproto_class->set_cfm) {
2139 port->ofproto->ofproto_class->set_cfm(port, NULL);
2141 if (port->ofproto->ofproto_class->bundle_remove) {
2142 port->ofproto->ofproto_class->bundle_remove(port);
2148 ofport_destroy__(struct ofport *port)
2150 struct ofproto *ofproto = port->ofproto;
2151 const char *name = netdev_get_name(port->netdev);
2153 hmap_remove(&ofproto->ports, &port->hmap_node);
2154 shash_delete(&ofproto->port_by_name,
2155 shash_find(&ofproto->port_by_name, name));
2157 netdev_close(port->netdev);
2158 ofproto->ofproto_class->port_dealloc(port);
2162 ofport_destroy(struct ofport *port)
2165 dealloc_ofp_port(port->ofproto, port->ofp_port);
2166 port->ofproto->ofproto_class->port_destruct(port);
2167 ofport_destroy__(port);
2172 ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
2174 struct ofport *port;
2176 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_ofp_port(ofp_port),
2178 if (port->ofp_port == ofp_port) {
2186 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
2188 struct ofproto *ofproto = port->ofproto;
2191 if (ofproto->ofproto_class->port_get_stats) {
2192 error = ofproto->ofproto_class->port_get_stats(port, stats);
2201 update_port(struct ofproto *ofproto, const char *name)
2203 struct ofproto_port ofproto_port;
2204 struct ofputil_phy_port pp;
2205 struct netdev *netdev;
2206 struct ofport *port;
2208 COVERAGE_INC(ofproto_update_port);
2210 /* Fetch 'name''s location and properties from the datapath. */
2211 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
2212 ? ofport_open(ofproto, &ofproto_port, &pp)
2216 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
2217 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
2218 struct netdev *old_netdev = port->netdev;
2220 /* 'name' hasn't changed location. Any properties changed? */
2221 if (!ofport_equal(&port->pp, &pp)) {
2222 ofport_modified(port, &pp);
2225 update_mtu(ofproto, port);
2227 /* Install the newly opened netdev in case it has changed.
2228 * Don't close the old netdev yet in case port_modified has to
2229 * remove a retained reference to it.*/
2230 port->netdev = netdev;
2231 port->change_seq = netdev_change_seq(netdev);
2233 if (port->ofproto->ofproto_class->port_modified) {
2234 port->ofproto->ofproto_class->port_modified(port);
2237 netdev_close(old_netdev);
2239 /* If 'port' is nonnull then its name differs from 'name' and thus
2240 * we should delete it. If we think there's a port named 'name'
2241 * then its port number must be wrong now so delete it too. */
2243 ofport_remove(port);
2245 ofport_remove_with_name(ofproto, name);
2246 ofport_install(ofproto, netdev, &pp);
2249 /* Any port named 'name' is gone now. */
2250 ofport_remove_with_name(ofproto, name);
2252 ofproto_port_destroy(&ofproto_port);
2256 init_ports(struct ofproto *p)
2258 struct ofproto_port_dump dump;
2259 struct ofproto_port ofproto_port;
2260 struct shash_node *node, *next;
2262 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2263 const char *name = ofproto_port.name;
2265 if (shash_find(&p->port_by_name, name)) {
2266 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2269 struct ofputil_phy_port pp;
2270 struct netdev *netdev;
2272 /* Check if an OpenFlow port number had been requested. */
2273 node = shash_find(&init_ofp_ports, name);
2275 const struct iface_hint *iface_hint = node->data;
2276 simap_put(&p->ofp_requests, name,
2277 ofp_to_u16(iface_hint->ofp_port));
2280 netdev = ofport_open(p, &ofproto_port, &pp);
2282 ofport_install(p, netdev, &pp);
2283 if (ofp_to_u16(ofproto_port.ofp_port) < p->max_ports) {
2284 p->alloc_port_no = MAX(p->alloc_port_no,
2285 ofp_to_u16(ofproto_port.ofp_port));
2291 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2292 struct iface_hint *iface_hint = node->data;
2294 if (!strcmp(iface_hint->br_name, p->name)) {
2295 free(iface_hint->br_name);
2296 free(iface_hint->br_type);
2298 shash_delete(&init_ofp_ports, node);
2305 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2306 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2308 find_min_mtu(struct ofproto *p)
2310 struct ofport *ofport;
2313 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2314 struct netdev *netdev = ofport->netdev;
2317 /* Skip any internal ports, since that's what we're trying to
2319 if (!strcmp(netdev_get_type(netdev), "internal")) {
2323 if (netdev_get_mtu(netdev, &dev_mtu)) {
2326 if (!mtu || dev_mtu < mtu) {
2331 return mtu ? mtu: ETH_PAYLOAD_MAX;
2334 /* Update MTU of all datapath devices on 'p' to the minimum of the
2335 * non-datapath ports in event of 'port' added or changed. */
2337 update_mtu(struct ofproto *p, struct ofport *port)
2339 struct ofport *ofport;
2340 struct netdev *netdev = port->netdev;
2341 int dev_mtu, old_min;
2343 if (netdev_get_mtu(netdev, &dev_mtu)) {
2347 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2348 if (dev_mtu > p->min_mtu) {
2349 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2350 dev_mtu = p->min_mtu;
2353 port->mtu = dev_mtu;
2357 /* For non-internal port find new min mtu. */
2358 old_min = p->min_mtu;
2359 port->mtu = dev_mtu;
2360 p->min_mtu = find_min_mtu(p);
2361 if (p->min_mtu == old_min) {
2365 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2366 struct netdev *netdev = ofport->netdev;
2368 if (!strcmp(netdev_get_type(netdev), "internal")) {
2369 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2370 ofport->mtu = p->min_mtu;
2377 ofproto_rule_ref(struct rule *rule)
2382 atomic_add(&rule->ref_count, 1, &orig);
2383 ovs_assert(orig != 0);
2388 ofproto_rule_unref(struct rule *rule)
2393 atomic_sub(&rule->ref_count, 1, &orig);
2395 rule->ofproto->ofproto_class->rule_destruct(rule);
2396 ofproto_rule_destroy__(rule);
2398 ovs_assert(orig != 0);
2403 struct rule_actions *
2404 rule_get_actions(const struct rule *rule)
2405 OVS_EXCLUDED(rule->mutex)
2407 struct rule_actions *actions;
2409 ovs_mutex_lock(&rule->mutex);
2410 actions = rule_get_actions__(rule);
2411 ovs_mutex_unlock(&rule->mutex);
2416 struct rule_actions *
2417 rule_get_actions__(const struct rule *rule)
2418 OVS_REQUIRES(rule->mutex)
2420 rule_actions_ref(rule->actions);
2421 return rule->actions;
2425 ofproto_rule_destroy__(struct rule *rule)
2426 OVS_NO_THREAD_SAFETY_ANALYSIS
2428 cls_rule_destroy(CONST_CAST(struct cls_rule *, &rule->cr));
2429 rule_actions_unref(rule->actions);
2430 ovs_mutex_destroy(&rule->mutex);
2431 rule->ofproto->ofproto_class->rule_dealloc(rule);
2434 /* Creates and returns a new 'struct rule_actions', with a ref_count of 1,
2435 * whose actions are a copy of from the 'ofpacts_len' bytes of 'ofpacts'. */
2436 struct rule_actions *
2437 rule_actions_create(const struct ofpact *ofpacts, size_t ofpacts_len)
2439 struct rule_actions *actions;
2441 actions = xmalloc(sizeof *actions);
2442 atomic_init(&actions->ref_count, 1);
2443 actions->ofpacts = xmemdup(ofpacts, ofpacts_len);
2444 actions->ofpacts_len = ofpacts_len;
2445 actions->meter_id = ofpacts_get_meter(ofpacts, ofpacts_len);
2449 /* Increments 'actions''s ref_count. */
2451 rule_actions_ref(struct rule_actions *actions)
2456 atomic_add(&actions->ref_count, 1, &orig);
2457 ovs_assert(orig != 0);
2461 /* Decrements 'actions''s ref_count and frees 'actions' if the ref_count
2464 rule_actions_unref(struct rule_actions *actions)
2469 atomic_sub(&actions->ref_count, 1, &orig);
2473 ovs_assert(orig != 0);
2478 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2479 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2481 ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port)
2482 OVS_REQUIRES(ofproto_mutex)
2484 return (port == OFPP_ANY
2485 || ofpacts_output_to_port(rule->actions->ofpacts,
2486 rule->actions->ofpacts_len, port));
2489 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2490 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2492 ofoperation_has_out_port(const struct ofoperation *op, ofp_port_t out_port)
2493 OVS_REQUIRES(ofproto_mutex)
2495 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2500 case OFOPERATION_ADD:
2501 case OFOPERATION_DELETE:
2504 case OFOPERATION_MODIFY:
2505 case OFOPERATION_REPLACE:
2506 return ofpacts_output_to_port(op->actions->ofpacts,
2507 op->actions->ofpacts_len, out_port);
2514 rule_execute_destroy(struct rule_execute *e)
2516 ofproto_rule_unref(e->rule);
2517 list_remove(&e->list_node);
2521 /* Executes all "rule_execute" operations queued up in ofproto->rule_executes,
2522 * by passing them to the ofproto provider. */
2524 run_rule_executes(struct ofproto *ofproto)
2525 OVS_EXCLUDED(ofproto_mutex)
2527 struct rule_execute *e, *next;
2528 struct list executes;
2530 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2531 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2532 union flow_in_port in_port_;
2535 in_port_.ofp_port = e->in_port;
2536 flow_extract(e->packet, 0, 0, NULL, &in_port_, &flow);
2537 ofproto->ofproto_class->rule_execute(e->rule, &flow, e->packet);
2539 rule_execute_destroy(e);
2543 /* Destroys and discards all "rule_execute" operations queued up in
2544 * ofproto->rule_executes. */
2546 destroy_rule_executes(struct ofproto *ofproto)
2548 struct rule_execute *e, *next;
2549 struct list executes;
2551 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2552 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2553 ofpbuf_delete(e->packet);
2554 rule_execute_destroy(e);
2558 /* Returns true if 'rule' should be hidden from the controller.
2560 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2561 * (e.g. by in-band control) and are intentionally hidden from the
2564 ofproto_rule_is_hidden(const struct rule *rule)
2566 return rule->cr.priority > UINT16_MAX;
2569 static enum oftable_flags
2570 rule_get_flags(const struct rule *rule)
2572 return rule->ofproto->tables[rule->table_id].flags;
2576 rule_is_modifiable(const struct rule *rule)
2578 return !(rule_get_flags(rule) & OFTABLE_READONLY);
2582 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2584 ofconn_send_reply(ofconn, make_echo_reply(oh));
2589 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2591 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2592 struct ofputil_switch_features features;
2593 struct ofport *port;
2599 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2601 ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2603 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2604 * if present, are always at the end.) */
2605 n_tables = ofproto->n_tables;
2606 for (i = 0; i < ofproto->n_tables; i++) {
2607 if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
2613 features.datapath_id = ofproto->datapath_id;
2614 features.n_buffers = pktbuf_capacity();
2615 features.n_tables = n_tables;
2616 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2617 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2619 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2621 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2622 features.auxiliary_id = 0;
2623 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2625 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2626 ofputil_put_switch_features_port(&port->pp, b);
2629 ofconn_send_reply(ofconn, b);
2634 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2636 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2637 struct ofp_switch_config *osc;
2638 enum ofp_config_flags flags;
2642 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2643 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2644 flags = ofproto->frag_handling;
2645 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2646 if (oh->version < OFP13_VERSION
2647 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2648 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2650 osc->flags = htons(flags);
2651 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2652 ofconn_send_reply(ofconn, buf);
2658 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2660 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2661 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2662 uint16_t flags = ntohs(osc->flags);
2664 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2665 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
2666 enum ofp_config_flags cur = ofproto->frag_handling;
2667 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2669 ovs_assert((cur & OFPC_FRAG_MASK) == cur);
2671 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2672 ofproto->frag_handling = next;
2674 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2676 ofputil_frag_handling_to_string(next));
2680 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2681 ofconn_set_invalid_ttl_to_controller(ofconn,
2682 (oh->version < OFP13_VERSION
2683 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2685 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2690 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2691 * error message code for the caller to propagate upward. Otherwise, returns
2694 * The log message mentions 'msg_type'. */
2696 reject_slave_controller(struct ofconn *ofconn)
2698 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2699 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
2700 return OFPERR_OFPBRC_EPERM;
2706 /* Checks that the 'ofpacts_len' bytes of actions in 'ofpacts' are appropriate
2707 * for a packet with the prerequisites satisfied by 'flow' in table 'table_id'.
2708 * 'flow' may be temporarily modified, but is restored at return.
2711 ofproto_check_ofpacts(struct ofproto *ofproto,
2712 const struct ofpact ofpacts[], size_t ofpacts_len,
2713 struct flow *flow, uint8_t table_id)
2718 error = ofpacts_check(ofpacts, ofpacts_len, flow,
2719 u16_to_ofp(ofproto->max_ports), table_id);
2724 mid = ofpacts_get_meter(ofpacts, ofpacts_len);
2725 if (mid && ofproto_get_provider_meter_id(ofproto, mid) == UINT32_MAX) {
2726 return OFPERR_OFPMMFC_INVALID_METER;
2732 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2734 struct ofproto *p = ofconn_get_ofproto(ofconn);
2735 struct ofputil_packet_out po;
2736 struct ofpbuf *payload;
2737 uint64_t ofpacts_stub[1024 / 8];
2738 struct ofpbuf ofpacts;
2740 union flow_in_port in_port_;
2743 COVERAGE_INC(ofproto_packet_out);
2745 error = reject_slave_controller(ofconn);
2750 /* Decode message. */
2751 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2752 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2754 goto exit_free_ofpacts;
2756 if (ofp_to_u16(po.in_port) >= p->max_ports
2757 && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) {
2758 error = OFPERR_OFPBRC_BAD_PORT;
2759 goto exit_free_ofpacts;
2764 if (po.buffer_id != UINT32_MAX) {
2765 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2766 if (error || !payload) {
2767 goto exit_free_ofpacts;
2770 /* Ensure that the L3 header is 32-bit aligned. */
2771 payload = ofpbuf_clone_data_with_headroom(po.packet, po.packet_len, 2);
2774 /* Verify actions against packet, then send packet if successful. */
2775 in_port_.ofp_port = po.in_port;
2776 flow_extract(payload, 0, 0, NULL, &in_port_, &flow);
2777 error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len, &flow, 0);
2779 error = p->ofproto_class->packet_out(p, payload, &flow,
2780 po.ofpacts, po.ofpacts_len);
2782 ofpbuf_delete(payload);
2785 ofpbuf_uninit(&ofpacts);
2791 update_port_config(struct ofport *port,
2792 enum ofputil_port_config config,
2793 enum ofputil_port_config mask)
2795 enum ofputil_port_config old_config = port->pp.config;
2796 enum ofputil_port_config toggle;
2798 toggle = (config ^ port->pp.config) & mask;
2799 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2800 if (config & OFPUTIL_PC_PORT_DOWN) {
2801 netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL);
2803 netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL);
2805 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2808 port->pp.config ^= toggle;
2809 if (port->pp.config != old_config) {
2810 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2815 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2817 struct ofproto *p = ofconn_get_ofproto(ofconn);
2818 struct ofputil_port_mod pm;
2819 struct ofport *port;
2822 error = reject_slave_controller(ofconn);
2827 error = ofputil_decode_port_mod(oh, &pm);
2832 port = ofproto_get_port(p, pm.port_no);
2834 return OFPERR_OFPPMFC_BAD_PORT;
2835 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2836 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2838 update_port_config(port, pm.config, pm.mask);
2840 netdev_set_advertisements(port->netdev, pm.advertise);
2847 handle_desc_stats_request(struct ofconn *ofconn,
2848 const struct ofp_header *request)
2850 static const char *default_mfr_desc = "Nicira, Inc.";
2851 static const char *default_hw_desc = "Open vSwitch";
2852 static const char *default_sw_desc = VERSION;
2853 static const char *default_serial_desc = "None";
2854 static const char *default_dp_desc = "None";
2856 struct ofproto *p = ofconn_get_ofproto(ofconn);
2857 struct ofp_desc_stats *ods;
2860 msg = ofpraw_alloc_stats_reply(request, 0);
2861 ods = ofpbuf_put_zeros(msg, sizeof *ods);
2862 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
2863 sizeof ods->mfr_desc);
2864 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
2865 sizeof ods->hw_desc);
2866 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
2867 sizeof ods->sw_desc);
2868 ovs_strlcpy(ods->serial_num,
2869 p->serial_desc ? p->serial_desc : default_serial_desc,
2870 sizeof ods->serial_num);
2871 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
2872 sizeof ods->dp_desc);
2873 ofconn_send_reply(ofconn, msg);
2879 handle_table_stats_request(struct ofconn *ofconn,
2880 const struct ofp_header *request)
2882 struct ofproto *p = ofconn_get_ofproto(ofconn);
2883 struct ofp12_table_stats *ots;
2888 /* Set up default values.
2890 * ofp12_table_stats is used as a generic structure as
2891 * it is able to hold all the fields for ofp10_table_stats
2892 * and ofp11_table_stats (and of course itself).
2894 ots = xcalloc(p->n_tables, sizeof *ots);
2895 for (i = 0; i < p->n_tables; i++) {
2896 ots[i].table_id = i;
2897 sprintf(ots[i].name, "table%zu", i);
2898 ots[i].match = htonll(OFPXMT12_MASK);
2899 ots[i].wildcards = htonll(OFPXMT12_MASK);
2900 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
2901 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
2902 ots[i].write_setfields = htonll(OFPXMT12_MASK);
2903 ots[i].apply_setfields = htonll(OFPXMT12_MASK);
2904 ots[i].metadata_match = htonll(UINT64_MAX);
2905 ots[i].metadata_write = htonll(UINT64_MAX);
2906 ots[i].instructions = htonl(OFPIT11_ALL);
2907 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
2908 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2909 ovs_rwlock_rdlock(&p->tables[i].cls.rwlock);
2910 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2911 ovs_rwlock_unlock(&p->tables[i].cls.rwlock);
2914 p->ofproto_class->get_tables(p, ots);
2916 /* Post-process the tables, dropping hidden tables. */
2917 n_tables = p->n_tables;
2918 for (i = 0; i < p->n_tables; i++) {
2919 const struct oftable *table = &p->tables[i];
2921 if (table->flags & OFTABLE_HIDDEN) {
2927 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2930 if (table->max_flows < ntohl(ots[i].max_entries)) {
2931 ots[i].max_entries = htonl(table->max_flows);
2935 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
2936 ofconn_send_reply(ofconn, msg);
2944 append_port_stat(struct ofport *port, struct list *replies)
2946 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
2948 calc_duration(port->created, time_msec(),
2949 &ops.duration_sec, &ops.duration_nsec);
2951 /* Intentionally ignore return value, since errors will set
2952 * 'stats' to all-1s, which is correct for OpenFlow, and
2953 * netdev_get_stats() will log errors. */
2954 ofproto_port_get_stats(port, &ops.stats);
2956 ofputil_append_port_stat(replies, &ops);
2960 handle_port_stats_request(struct ofconn *ofconn,
2961 const struct ofp_header *request)
2963 struct ofproto *p = ofconn_get_ofproto(ofconn);
2964 struct ofport *port;
2965 struct list replies;
2969 error = ofputil_decode_port_stats_request(request, &port_no);
2974 ofpmp_init(&replies, request);
2975 if (port_no != OFPP_ANY) {
2976 port = ofproto_get_port(p, port_no);
2978 append_port_stat(port, &replies);
2981 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2982 append_port_stat(port, &replies);
2986 ofconn_send_replies(ofconn, &replies);
2991 handle_port_desc_stats_request(struct ofconn *ofconn,
2992 const struct ofp_header *request)
2994 struct ofproto *p = ofconn_get_ofproto(ofconn);
2995 enum ofp_version version;
2996 struct ofport *port;
2997 struct list replies;
2999 ofpmp_init(&replies, request);
3001 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
3002 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3003 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
3006 ofconn_send_replies(ofconn, &replies);
3011 hash_cookie(ovs_be64 cookie)
3013 return hash_2words((OVS_FORCE uint64_t)cookie >> 32,
3014 (OVS_FORCE uint64_t)cookie);
3018 cookies_insert(struct ofproto *ofproto, struct rule *rule)
3019 OVS_REQUIRES(ofproto_mutex)
3021 hindex_insert(&ofproto->cookies, &rule->cookie_node,
3022 hash_cookie(rule->flow_cookie));
3026 cookies_remove(struct ofproto *ofproto, struct rule *rule)
3027 OVS_REQUIRES(ofproto_mutex)
3029 hindex_remove(&ofproto->cookies, &rule->cookie_node);
3033 ofproto_rule_change_cookie(struct ofproto *ofproto, struct rule *rule,
3034 ovs_be64 new_cookie)
3035 OVS_REQUIRES(ofproto_mutex)
3037 if (new_cookie != rule->flow_cookie) {
3038 cookies_remove(ofproto, rule);
3040 ovs_mutex_lock(&rule->mutex);
3041 rule->flow_cookie = new_cookie;
3042 ovs_mutex_unlock(&rule->mutex);
3044 cookies_insert(ofproto, rule);
3049 calc_duration(long long int start, long long int now,
3050 uint32_t *sec, uint32_t *nsec)
3052 long long int msecs = now - start;
3053 *sec = msecs / 1000;
3054 *nsec = (msecs % 1000) * (1000 * 1000);
3057 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
3058 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
3060 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
3062 return (table_id == 0xff || table_id < ofproto->n_tables
3064 : OFPERR_OFPBRC_BAD_TABLE_ID);
3068 static struct oftable *
3069 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
3071 struct oftable *table;
3073 for (table = &ofproto->tables[table_id];
3074 table < &ofproto->tables[ofproto->n_tables];
3076 if (!(table->flags & OFTABLE_HIDDEN)) {
3084 static struct oftable *
3085 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
3087 if (table_id == 0xff) {
3088 return next_visible_table(ofproto, 0);
3089 } else if (table_id < ofproto->n_tables) {
3090 return &ofproto->tables[table_id];
3096 static struct oftable *
3097 next_matching_table(const struct ofproto *ofproto,
3098 const struct oftable *table, uint8_t table_id)
3100 return (table_id == 0xff
3101 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
3105 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
3107 * - If TABLE_ID is 0xff, this iterates over every classifier table in
3108 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
3110 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
3111 * only once, for that table. (This can be used to access tables marked
3114 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
3115 * entered at all. (Perhaps you should have validated TABLE_ID with
3116 * check_table_id().)
3118 * All parameters are evaluated multiple times.
3120 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
3121 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
3123 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
3125 /* Initializes 'criteria' in a straightforward way based on the other
3128 * For "loose" matching, the 'priority' parameter is unimportant and may be
3131 rule_criteria_init(struct rule_criteria *criteria, uint8_t table_id,
3132 const struct match *match, unsigned int priority,
3133 ovs_be64 cookie, ovs_be64 cookie_mask,
3134 ofp_port_t out_port)
3136 criteria->table_id = table_id;
3137 cls_rule_init(&criteria->cr, match, priority);
3138 criteria->cookie = cookie;
3139 criteria->cookie_mask = cookie_mask;
3140 criteria->out_port = out_port;
3144 rule_criteria_destroy(struct rule_criteria *criteria)
3146 cls_rule_destroy(&criteria->cr);
3150 rule_collection_init(struct rule_collection *rules)
3152 rules->rules = rules->stub;
3154 rules->capacity = ARRAY_SIZE(rules->stub);
3158 rule_collection_add(struct rule_collection *rules, struct rule *rule)
3160 if (rules->n >= rules->capacity) {
3161 size_t old_size, new_size;
3163 old_size = rules->capacity * sizeof *rules->rules;
3164 rules->capacity *= 2;
3165 new_size = rules->capacity * sizeof *rules->rules;
3167 if (rules->rules == rules->stub) {
3168 rules->rules = xmalloc(new_size);
3169 memcpy(rules->rules, rules->stub, old_size);
3171 rules->rules = xrealloc(rules->rules, new_size);
3175 rules->rules[rules->n++] = rule;
3179 rule_collection_ref(struct rule_collection *rules)
3180 OVS_REQUIRES(ofproto_mutex)
3184 for (i = 0; i < rules->n; i++) {
3185 ofproto_rule_ref(rules->rules[i]);
3190 rule_collection_unref(struct rule_collection *rules)
3194 for (i = 0; i < rules->n; i++) {
3195 ofproto_rule_unref(rules->rules[i]);
3200 rule_collection_destroy(struct rule_collection *rules)
3202 if (rules->rules != rules->stub) {
3208 collect_rule(struct rule *rule, const struct rule_criteria *c,
3209 struct rule_collection *rules)
3210 OVS_REQUIRES(ofproto_mutex)
3212 if (ofproto_rule_is_hidden(rule)) {
3214 } else if (rule->pending) {
3215 return OFPROTO_POSTPONE;
3217 if ((c->table_id == rule->table_id || c->table_id == 0xff)
3218 && ofproto_rule_has_out_port(rule, c->out_port)
3219 && !((rule->flow_cookie ^ c->cookie) & c->cookie_mask)) {
3220 rule_collection_add(rules, rule);
3226 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3227 * on classifiers rules are done in the "loose" way required for OpenFlow
3228 * OFPFC_MODIFY and OFPFC_DELETE requests. Puts the selected rules on list
3231 * Hidden rules are always omitted.
3233 * Returns 0 on success, otherwise an OpenFlow error code. */
3235 collect_rules_loose(struct ofproto *ofproto,
3236 const struct rule_criteria *criteria,
3237 struct rule_collection *rules)
3238 OVS_REQUIRES(ofproto_mutex)
3240 struct oftable *table;
3243 rule_collection_init(rules);
3245 error = check_table_id(ofproto, criteria->table_id);
3250 if (criteria->cookie_mask == htonll(UINT64_MAX)) {
3253 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
3254 hash_cookie(criteria->cookie),
3255 &ofproto->cookies) {
3256 if (cls_rule_is_loose_match(&rule->cr, &criteria->cr.match)) {
3257 error = collect_rule(rule, criteria, rules);
3264 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
3265 struct cls_cursor cursor;
3268 ovs_rwlock_rdlock(&table->cls.rwlock);
3269 cls_cursor_init(&cursor, &table->cls, &criteria->cr);
3270 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3271 error = collect_rule(rule, criteria, rules);
3276 ovs_rwlock_unlock(&table->cls.rwlock);
3282 rule_collection_destroy(rules);
3287 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3288 * on classifiers rules are done in the "strict" way required for OpenFlow
3289 * OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests. Puts the selected
3290 * rules on list 'rules'.
3292 * Hidden rules are always omitted.
3294 * Returns 0 on success, otherwise an OpenFlow error code. */
3296 collect_rules_strict(struct ofproto *ofproto,
3297 const struct rule_criteria *criteria,
3298 struct rule_collection *rules)
3299 OVS_REQUIRES(ofproto_mutex)
3301 struct oftable *table;
3304 rule_collection_init(rules);
3306 error = check_table_id(ofproto, criteria->table_id);
3311 if (criteria->cookie_mask == htonll(UINT64_MAX)) {
3314 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
3315 hash_cookie(criteria->cookie),
3316 &ofproto->cookies) {
3317 if (cls_rule_equal(&rule->cr, &criteria->cr)) {
3318 error = collect_rule(rule, criteria, rules);
3325 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
3328 ovs_rwlock_rdlock(&table->cls.rwlock);
3329 rule = rule_from_cls_rule(classifier_find_rule_exactly(
3330 &table->cls, &criteria->cr));
3331 ovs_rwlock_unlock(&table->cls.rwlock);
3333 error = collect_rule(rule, criteria, rules);
3343 rule_collection_destroy(rules);
3348 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
3349 * forced into the range of a uint16_t. */
3351 age_secs(long long int age_ms)
3353 return (age_ms < 0 ? 0
3354 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
3355 : (unsigned int) age_ms / 1000);
3359 handle_flow_stats_request(struct ofconn *ofconn,
3360 const struct ofp_header *request)
3361 OVS_EXCLUDED(ofproto_mutex)
3363 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3364 struct ofputil_flow_stats_request fsr;
3365 struct rule_criteria criteria;
3366 struct rule_collection rules;
3367 struct list replies;
3371 error = ofputil_decode_flow_stats_request(&fsr, request);
3376 rule_criteria_init(&criteria, fsr.table_id, &fsr.match, 0, fsr.cookie,
3377 fsr.cookie_mask, fsr.out_port);
3379 ovs_mutex_lock(&ofproto_mutex);
3380 error = collect_rules_loose(ofproto, &criteria, &rules);
3381 rule_criteria_destroy(&criteria);
3383 rule_collection_ref(&rules);
3385 ovs_mutex_unlock(&ofproto_mutex);
3391 ofpmp_init(&replies, request);
3392 for (i = 0; i < rules.n; i++) {
3393 struct rule *rule = rules.rules[i];
3394 long long int now = time_msec();
3395 struct ofputil_flow_stats fs;
3396 long long int created, used, modified;
3397 struct rule_actions *actions;
3398 bool send_flow_removed;
3400 ovs_mutex_lock(&rule->mutex);
3401 fs.cookie = rule->flow_cookie;
3402 fs.idle_timeout = rule->idle_timeout;
3403 fs.hard_timeout = rule->hard_timeout;
3404 created = rule->created;
3406 modified = rule->modified;
3407 actions = rule_get_actions__(rule);
3408 send_flow_removed = rule->send_flow_removed;
3409 ovs_mutex_unlock(&rule->mutex);
3411 minimatch_expand(&rule->cr.match, &fs.match);
3412 fs.table_id = rule->table_id;
3413 calc_duration(created, now, &fs.duration_sec, &fs.duration_nsec);
3414 fs.priority = rule->cr.priority;
3415 fs.idle_age = age_secs(now - used);
3416 fs.hard_age = age_secs(now - modified);
3417 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
3419 fs.ofpacts = actions->ofpacts;
3420 fs.ofpacts_len = actions->ofpacts_len;
3423 if (send_flow_removed) {
3424 fs.flags |= OFPUTIL_FF_SEND_FLOW_REM;
3425 /* FIXME: Implement OFPUTIL_FF_NO_PKT_COUNTS and
3426 OFPUTIL_FF_NO_BYT_COUNTS. */
3428 ofputil_append_flow_stats_reply(&fs, &replies);
3430 rule_actions_unref(actions);
3433 rule_collection_unref(&rules);
3434 rule_collection_destroy(&rules);
3436 ofconn_send_replies(ofconn, &replies);
3442 flow_stats_ds(struct rule *rule, struct ds *results)
3444 uint64_t packet_count, byte_count;
3445 struct rule_actions *actions;
3446 long long int created;
3448 rule->ofproto->ofproto_class->rule_get_stats(rule,
3449 &packet_count, &byte_count);
3451 ovs_mutex_lock(&rule->mutex);
3452 actions = rule_get_actions__(rule);
3453 created = rule->created;
3454 ovs_mutex_unlock(&rule->mutex);
3456 if (rule->table_id != 0) {
3457 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
3459 ds_put_format(results, "duration=%llds, ", (time_msec() - created) / 1000);
3460 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3461 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3462 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3463 cls_rule_format(&rule->cr, results);
3464 ds_put_char(results, ',');
3466 ofpacts_format(actions->ofpacts, actions->ofpacts_len, results);
3468 ds_put_cstr(results, "\n");
3470 rule_actions_unref(actions);
3473 /* Adds a pretty-printed description of all flows to 'results', including
3474 * hidden flows (e.g., set up by in-band control). */
3476 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3478 struct oftable *table;
3480 OFPROTO_FOR_EACH_TABLE (table, p) {
3481 struct cls_cursor cursor;
3484 ovs_rwlock_rdlock(&table->cls.rwlock);
3485 cls_cursor_init(&cursor, &table->cls, NULL);
3486 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3487 flow_stats_ds(rule, results);
3489 ovs_rwlock_unlock(&table->cls.rwlock);
3493 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
3494 * '*engine_type' and '*engine_id', respectively. */
3496 ofproto_get_netflow_ids(const struct ofproto *ofproto,
3497 uint8_t *engine_type, uint8_t *engine_id)
3499 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
3502 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
3503 * true if the port's CFM status was successfully stored into '*status'.
3504 * Returns false if the port did not have CFM configured, in which case
3505 * '*status' is indeterminate.
3507 * The caller must provide and owns '*status', and must free 'status->rmps'. */
3509 ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port,
3510 struct ofproto_cfm_status *status)
3512 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
3514 && ofproto->ofproto_class->get_cfm_status
3515 && ofproto->ofproto_class->get_cfm_status(ofport, status));
3519 handle_aggregate_stats_request(struct ofconn *ofconn,
3520 const struct ofp_header *oh)
3521 OVS_EXCLUDED(ofproto_mutex)
3523 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3524 struct ofputil_flow_stats_request request;
3525 struct ofputil_aggregate_stats stats;
3526 bool unknown_packets, unknown_bytes;
3527 struct rule_criteria criteria;
3528 struct rule_collection rules;
3529 struct ofpbuf *reply;
3533 error = ofputil_decode_flow_stats_request(&request, oh);
3538 rule_criteria_init(&criteria, request.table_id, &request.match, 0,
3539 request.cookie, request.cookie_mask,
3542 ovs_mutex_lock(&ofproto_mutex);
3543 error = collect_rules_loose(ofproto, &criteria, &rules);
3544 rule_criteria_destroy(&criteria);
3546 rule_collection_ref(&rules);
3548 ovs_mutex_unlock(&ofproto_mutex);
3554 memset(&stats, 0, sizeof stats);
3555 unknown_packets = unknown_bytes = false;
3556 for (i = 0; i < rules.n; i++) {
3557 struct rule *rule = rules.rules[i];
3558 uint64_t packet_count;
3559 uint64_t byte_count;
3561 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
3564 if (packet_count == UINT64_MAX) {
3565 unknown_packets = true;
3567 stats.packet_count += packet_count;
3570 if (byte_count == UINT64_MAX) {
3571 unknown_bytes = true;
3573 stats.byte_count += byte_count;
3578 if (unknown_packets) {
3579 stats.packet_count = UINT64_MAX;
3581 if (unknown_bytes) {
3582 stats.byte_count = UINT64_MAX;
3585 rule_collection_unref(&rules);
3586 rule_collection_destroy(&rules);
3588 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
3589 ofconn_send_reply(ofconn, reply);
3594 struct queue_stats_cbdata {
3595 struct ofport *ofport;
3596 struct list replies;
3601 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3602 const struct netdev_queue_stats *stats)
3604 struct ofputil_queue_stats oqs;
3606 oqs.port_no = cbdata->ofport->pp.port_no;
3607 oqs.queue_id = queue_id;
3608 oqs.tx_bytes = stats->tx_bytes;
3609 oqs.tx_packets = stats->tx_packets;
3610 oqs.tx_errors = stats->tx_errors;
3611 if (stats->created != LLONG_MIN) {
3612 calc_duration(stats->created, cbdata->now,
3613 &oqs.duration_sec, &oqs.duration_nsec);
3615 oqs.duration_sec = oqs.duration_nsec = UINT32_MAX;
3617 ofputil_append_queue_stat(&cbdata->replies, &oqs);
3621 handle_queue_stats_dump_cb(uint32_t queue_id,
3622 struct netdev_queue_stats *stats,
3625 struct queue_stats_cbdata *cbdata = cbdata_;
3627 put_queue_stats(cbdata, queue_id, stats);
3631 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3632 struct queue_stats_cbdata *cbdata)
3634 cbdata->ofport = port;
3635 if (queue_id == OFPQ_ALL) {
3636 netdev_dump_queue_stats(port->netdev,
3637 handle_queue_stats_dump_cb, cbdata);
3639 struct netdev_queue_stats stats;
3641 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3642 put_queue_stats(cbdata, queue_id, &stats);
3644 return OFPERR_OFPQOFC_BAD_QUEUE;
3651 handle_queue_stats_request(struct ofconn *ofconn,
3652 const struct ofp_header *rq)
3654 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3655 struct queue_stats_cbdata cbdata;
3656 struct ofport *port;
3658 struct ofputil_queue_stats_request oqsr;
3660 COVERAGE_INC(ofproto_queue_req);
3662 ofpmp_init(&cbdata.replies, rq);
3663 cbdata.now = time_msec();
3665 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3670 if (oqsr.port_no == OFPP_ANY) {
3671 error = OFPERR_OFPQOFC_BAD_QUEUE;
3672 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3673 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3678 port = ofproto_get_port(ofproto, oqsr.port_no);
3680 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3681 : OFPERR_OFPQOFC_BAD_PORT);
3684 ofconn_send_replies(ofconn, &cbdata.replies);
3686 ofpbuf_list_delete(&cbdata.replies);
3693 is_flow_deletion_pending(const struct ofproto *ofproto,
3694 const struct cls_rule *cls_rule,
3696 OVS_REQUIRES(ofproto_mutex)
3698 if (!hmap_is_empty(&ofproto->deletions)) {
3699 struct ofoperation *op;
3701 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3702 cls_rule_hash(cls_rule, table_id),
3703 &ofproto->deletions) {
3704 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
3714 should_evict_a_rule(struct oftable *table, unsigned int extra_space)
3715 OVS_REQUIRES(ofproto_mutex)
3716 OVS_NO_THREAD_SAFETY_ANALYSIS
3718 return classifier_count(&table->cls) + extra_space > table->max_flows;
3722 evict_rules_from_table(struct ofproto *ofproto, struct oftable *table,
3723 unsigned int extra_space)
3724 OVS_REQUIRES(ofproto_mutex)
3726 while (should_evict_a_rule(table, extra_space)) {
3729 if (!choose_rule_to_evict(table, &rule)) {
3730 return OFPERR_OFPFMFC_TABLE_FULL;
3731 } else if (rule->pending) {
3732 return OFPROTO_POSTPONE;
3734 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
3735 delete_flow__(rule, group, OFPRR_EVICTION);
3736 ofopgroup_submit(group);
3743 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3744 * in which no matching flow already exists in the flow table.
3746 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3747 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3748 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3749 * initiated now but may be retried later.
3751 * The caller retains ownership of 'fm->ofpacts'.
3753 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3756 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
3757 struct ofputil_flow_mod *fm, const struct ofp_header *request)
3758 OVS_REQUIRES(ofproto_mutex)
3760 struct oftable *table;
3761 struct ofopgroup *group;
3767 error = check_table_id(ofproto, fm->table_id);
3773 if (fm->table_id == 0xff) {
3774 if (ofproto->ofproto_class->rule_choose_table) {
3775 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3781 ovs_assert(table_id < ofproto->n_tables);
3785 } else if (fm->table_id < ofproto->n_tables) {
3786 table_id = fm->table_id;
3788 return OFPERR_OFPBRC_BAD_TABLE_ID;
3791 table = &ofproto->tables[table_id];
3793 if (table->flags & OFTABLE_READONLY) {
3794 return OFPERR_OFPBRC_EPERM;
3797 cls_rule_init(&cr, &fm->match, fm->priority);
3799 /* Transform "add" into "modify" if there's an existing identical flow. */
3800 ovs_rwlock_rdlock(&table->cls.rwlock);
3801 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr));
3802 ovs_rwlock_unlock(&table->cls.rwlock);
3804 cls_rule_destroy(&cr);
3805 if (!rule_is_modifiable(rule)) {
3806 return OFPERR_OFPBRC_EPERM;
3807 } else if (rule->pending) {
3808 return OFPROTO_POSTPONE;
3810 struct rule_collection rules;
3812 rule_collection_init(&rules);
3813 rule_collection_add(&rules, rule);
3814 fm->modify_cookie = true;
3815 error = modify_flows__(ofproto, ofconn, fm, request, &rules);
3816 rule_collection_destroy(&rules);
3822 /* Verify actions. */
3823 error = ofproto_check_ofpacts(ofproto, fm->ofpacts, fm->ofpacts_len,
3824 &fm->match.flow, table_id);
3826 cls_rule_destroy(&cr);
3830 /* Serialize against pending deletion. */
3831 if (is_flow_deletion_pending(ofproto, &cr, table_id)) {
3832 cls_rule_destroy(&cr);
3833 return OFPROTO_POSTPONE;
3836 /* Check for overlap, if requested. */
3837 if (fm->flags & OFPUTIL_FF_CHECK_OVERLAP) {
3840 ovs_rwlock_rdlock(&table->cls.rwlock);
3841 overlaps = classifier_rule_overlaps(&table->cls, &cr);
3842 ovs_rwlock_unlock(&table->cls.rwlock);
3845 cls_rule_destroy(&cr);
3846 return OFPERR_OFPFMFC_OVERLAP;
3850 /* If necessary, evict an existing rule to clear out space. */
3851 error = evict_rules_from_table(ofproto, table, 1);
3853 cls_rule_destroy(&cr);
3857 /* Allocate new rule. */
3858 rule = ofproto->ofproto_class->rule_alloc();
3860 cls_rule_destroy(&cr);
3861 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
3862 ofproto->name, ovs_strerror(error));
3866 /* Initialize base state. */
3867 *CONST_CAST(struct ofproto **, &rule->ofproto) = ofproto;
3868 cls_rule_move(CONST_CAST(struct cls_rule *, &rule->cr), &cr);
3869 atomic_init(&rule->ref_count, 1);
3870 rule->pending = NULL;
3871 rule->flow_cookie = fm->new_cookie;
3872 rule->created = rule->modified = rule->used = time_msec();
3874 ovs_mutex_init(&rule->mutex);
3875 ovs_mutex_lock(&rule->mutex);
3876 rule->idle_timeout = fm->idle_timeout;
3877 rule->hard_timeout = fm->hard_timeout;
3878 ovs_mutex_unlock(&rule->mutex);
3880 *CONST_CAST(uint8_t *, &rule->table_id) = table - ofproto->tables;
3881 rule->send_flow_removed = (fm->flags & OFPUTIL_FF_SEND_FLOW_REM) != 0;
3882 rule->actions = rule_actions_create(fm->ofpacts, fm->ofpacts_len);
3883 list_init(&rule->meter_list_node);
3884 rule->eviction_group = NULL;
3885 list_init(&rule->expirable);
3886 rule->monitor_flags = 0;
3887 rule->add_seqno = 0;
3888 rule->modify_seqno = 0;
3890 /* Construct rule, initializing derived state. */
3891 error = ofproto->ofproto_class->rule_construct(rule);
3893 ofproto_rule_destroy__(rule);
3898 oftable_insert_rule(rule);
3900 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3901 ofoperation_create(group, rule, OFOPERATION_ADD, 0);
3902 ofproto->ofproto_class->rule_insert(rule);
3903 ofopgroup_submit(group);
3908 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3910 /* Modifies the rules listed in 'rules', changing their actions to match those
3913 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3916 * Returns 0 on success, otherwise an OpenFlow error code. */
3918 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3919 struct ofputil_flow_mod *fm, const struct ofp_header *request,
3920 const struct rule_collection *rules)
3921 OVS_REQUIRES(ofproto_mutex)
3923 enum ofoperation_type type;
3924 struct ofopgroup *group;
3928 type = fm->command == OFPFC_ADD ? OFOPERATION_REPLACE : OFOPERATION_MODIFY;
3929 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3930 error = OFPERR_OFPBRC_EPERM;
3931 for (i = 0; i < rules->n; i++) {
3932 struct rule *rule = rules->rules[i];
3933 struct ofoperation *op;
3934 bool actions_changed;
3935 bool reset_counters;
3937 /* FIXME: Implement OFPFUTIL_FF_RESET_COUNTS */
3939 if (rule_is_modifiable(rule)) {
3940 /* At least one rule is modifiable, don't report EPERM error. */
3946 /* Verify actions. */
3947 error = ofpacts_check(fm->ofpacts, fm->ofpacts_len, &fm->match.flow,
3948 u16_to_ofp(ofproto->max_ports), rule->table_id);
3953 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
3954 rule->actions->ofpacts,
3955 rule->actions->ofpacts_len);
3957 op = ofoperation_create(group, rule, type, 0);
3959 if (fm->modify_cookie && fm->new_cookie != htonll(UINT64_MAX)) {
3960 ofproto_rule_change_cookie(ofproto, rule, fm->new_cookie);
3962 if (type == OFOPERATION_REPLACE) {
3963 ovs_mutex_lock(&rule->mutex);
3964 rule->idle_timeout = fm->idle_timeout;
3965 rule->hard_timeout = fm->hard_timeout;
3966 ovs_mutex_unlock(&rule->mutex);
3968 rule->send_flow_removed = (fm->flags
3969 & OFPUTIL_FF_SEND_FLOW_REM) != 0;
3971 if (fm->idle_timeout || fm->hard_timeout) {
3972 if (!rule->eviction_group) {
3973 eviction_group_add_rule(rule);
3976 eviction_group_remove_rule(rule);
3980 reset_counters = (fm->flags & OFPUTIL_FF_RESET_COUNTS) != 0;
3981 if (actions_changed || reset_counters) {
3982 struct rule_actions *new_actions;
3984 op->actions = rule->actions;
3985 new_actions = rule_actions_create(fm->ofpacts, fm->ofpacts_len);
3987 ovs_mutex_lock(&rule->mutex);
3988 rule->actions = new_actions;
3989 ovs_mutex_unlock(&rule->mutex);
3991 rule->ofproto->ofproto_class->rule_modify_actions(rule,
3994 ofoperation_complete(op, 0);
3997 ofopgroup_submit(group);
4003 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
4004 struct ofputil_flow_mod *fm, const struct ofp_header *request)
4005 OVS_REQUIRES(ofproto_mutex)
4007 if (fm->cookie_mask != htonll(0) || fm->new_cookie == htonll(UINT64_MAX)) {
4010 return add_flow(ofproto, ofconn, fm, request);
4013 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
4016 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4019 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
4020 struct ofputil_flow_mod *fm,
4021 const struct ofp_header *request)
4022 OVS_REQUIRES(ofproto_mutex)
4024 struct rule_criteria criteria;
4025 struct rule_collection rules;
4028 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0,
4029 fm->cookie, fm->cookie_mask, OFPP_ANY);
4030 error = collect_rules_loose(ofproto, &criteria, &rules);
4031 rule_criteria_destroy(&criteria);
4034 error = (rules.n > 0
4035 ? modify_flows__(ofproto, ofconn, fm, request, &rules)
4036 : modify_flows_add(ofproto, ofconn, fm, request));
4039 rule_collection_destroy(&rules);
4044 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4047 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4050 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
4051 struct ofputil_flow_mod *fm,
4052 const struct ofp_header *request)
4053 OVS_REQUIRES(ofproto_mutex)
4055 struct rule_criteria criteria;
4056 struct rule_collection rules;
4059 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
4060 fm->cookie, fm->cookie_mask, OFPP_ANY);
4061 error = collect_rules_strict(ofproto, &criteria, &rules);
4062 rule_criteria_destroy(&criteria);
4066 error = modify_flows_add(ofproto, ofconn, fm, request);
4067 } else if (rules.n == 1) {
4068 error = modify_flows__(ofproto, ofconn, fm, request, &rules);
4072 rule_collection_destroy(&rules);
4077 /* OFPFC_DELETE implementation. */
4080 delete_flow__(struct rule *rule, struct ofopgroup *group,
4081 enum ofp_flow_removed_reason reason)
4082 OVS_REQUIRES(ofproto_mutex)
4084 struct ofproto *ofproto = rule->ofproto;
4086 ofproto_rule_send_removed(rule, reason);
4088 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
4089 oftable_remove_rule(rule);
4090 ofproto->ofproto_class->rule_delete(rule);
4093 /* Deletes the rules listed in 'rules'.
4095 * Returns 0 on success, otherwise an OpenFlow error code. */
4097 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
4098 const struct ofp_header *request,
4099 const struct rule_collection *rules,
4100 enum ofp_flow_removed_reason reason)
4101 OVS_REQUIRES(ofproto_mutex)
4103 struct ofopgroup *group;
4106 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
4107 for (i = 0; i < rules->n; i++) {
4108 delete_flow__(rules->rules[i], group, reason);
4110 ofopgroup_submit(group);
4115 /* Implements OFPFC_DELETE. */
4117 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
4118 const struct ofputil_flow_mod *fm,
4119 const struct ofp_header *request)
4120 OVS_REQUIRES(ofproto_mutex)
4122 struct rule_criteria criteria;
4123 struct rule_collection rules;
4126 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0,
4127 fm->cookie, fm->cookie_mask,
4129 error = collect_rules_loose(ofproto, &criteria, &rules);
4130 rule_criteria_destroy(&criteria);
4132 if (!error && rules.n > 0) {
4133 error = delete_flows__(ofproto, ofconn, request, &rules, OFPRR_DELETE);
4135 rule_collection_destroy(&rules);
4140 /* Implements OFPFC_DELETE_STRICT. */
4142 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
4143 const struct ofputil_flow_mod *fm,
4144 const struct ofp_header *request)
4145 OVS_REQUIRES(ofproto_mutex)
4147 struct rule_criteria criteria;
4148 struct rule_collection rules;
4151 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
4152 fm->cookie, fm->cookie_mask, fm->out_port);
4153 error = collect_rules_strict(ofproto, &criteria, &rules);
4154 rule_criteria_destroy(&criteria);
4156 if (!error && rules.n > 0) {
4157 error = delete_flows__(ofproto, ofconn, request, &rules, OFPRR_DELETE);
4159 rule_collection_destroy(&rules);
4165 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
4166 OVS_REQUIRES(ofproto_mutex)
4168 struct ofputil_flow_removed fr;
4170 if (ofproto_rule_is_hidden(rule) || !rule->send_flow_removed) {
4174 minimatch_expand(&rule->cr.match, &fr.match);
4175 fr.priority = rule->cr.priority;
4176 fr.cookie = rule->flow_cookie;
4178 fr.table_id = rule->table_id;
4179 calc_duration(rule->created, time_msec(),
4180 &fr.duration_sec, &fr.duration_nsec);
4181 ovs_mutex_lock(&rule->mutex);
4182 fr.idle_timeout = rule->idle_timeout;
4183 fr.hard_timeout = rule->hard_timeout;
4184 ovs_mutex_unlock(&rule->mutex);
4185 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
4188 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
4191 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
4192 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
4195 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
4198 * ofproto implementation ->run() functions should use this function to expire
4199 * OpenFlow flows. */
4201 ofproto_rule_expire(struct rule *rule, uint8_t reason)
4202 OVS_REQUIRES(ofproto_mutex)
4204 struct ofproto *ofproto = rule->ofproto;
4206 ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT
4207 || reason == OFPRR_DELETE);
4209 ofproto_rule_send_removed(rule, reason);
4210 ofproto_rule_delete__(ofproto, rule);
4213 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
4214 * means "infinite". */
4216 reduce_timeout(uint16_t max, uint16_t *timeout)
4218 if (max && (!*timeout || *timeout > max)) {
4223 /* If 'idle_timeout' is nonzero, and 'rule' has no idle timeout or an idle
4224 * timeout greater than 'idle_timeout', lowers 'rule''s idle timeout to
4225 * 'idle_timeout' seconds. Similarly for 'hard_timeout'.
4227 * Suitable for implementing OFPACT_FIN_TIMEOUT. */
4229 ofproto_rule_reduce_timeouts(struct rule *rule,
4230 uint16_t idle_timeout, uint16_t hard_timeout)
4231 OVS_EXCLUDED(ofproto_mutex, rule->mutex)
4233 if (!idle_timeout && !hard_timeout) {
4237 ovs_mutex_lock(&ofproto_mutex);
4238 if (list_is_empty(&rule->expirable)) {
4239 list_insert(&rule->ofproto->expirable, &rule->expirable);
4241 ovs_mutex_unlock(&ofproto_mutex);
4243 ovs_mutex_lock(&rule->mutex);
4244 reduce_timeout(idle_timeout, &rule->idle_timeout);
4245 reduce_timeout(hard_timeout, &rule->hard_timeout);
4246 ovs_mutex_unlock(&rule->mutex);
4250 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4251 OVS_EXCLUDED(ofproto_mutex)
4253 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4254 struct ofputil_flow_mod fm;
4255 uint64_t ofpacts_stub[1024 / 8];
4256 struct ofpbuf ofpacts;
4260 error = reject_slave_controller(ofconn);
4265 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
4266 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
4269 error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
4272 goto exit_free_ofpacts;
4275 /* Record the operation for logging a summary report. */
4276 switch (fm.command) {
4282 case OFPFC_MODIFY_STRICT:
4283 ofproto->n_modify++;
4287 case OFPFC_DELETE_STRICT:
4288 ofproto->n_delete++;
4293 if (ofproto->next_op_report == LLONG_MAX) {
4294 ofproto->first_op = now;
4295 ofproto->next_op_report = MAX(now + 10 * 1000,
4296 ofproto->op_backoff);
4297 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
4299 ofproto->last_op = now;
4302 ofpbuf_uninit(&ofpacts);
4308 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
4309 struct ofputil_flow_mod *fm, const struct ofp_header *oh)
4310 OVS_EXCLUDED(ofproto_mutex)
4314 ovs_mutex_lock(&ofproto_mutex);
4315 if (ofproto->n_pending < 50) {
4316 switch (fm->command) {
4318 error = add_flow(ofproto, ofconn, fm, oh);
4322 error = modify_flows_loose(ofproto, ofconn, fm, oh);
4325 case OFPFC_MODIFY_STRICT:
4326 error = modify_flow_strict(ofproto, ofconn, fm, oh);
4330 error = delete_flows_loose(ofproto, ofconn, fm, oh);
4333 case OFPFC_DELETE_STRICT:
4334 error = delete_flow_strict(ofproto, ofconn, fm, oh);
4338 if (fm->command > 0xff) {
4339 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
4340 "flow_mod_table_id extension is not enabled",
4343 error = OFPERR_OFPFMFC_BAD_COMMAND;
4347 ovs_assert(!list_is_empty(&ofproto->pending));
4348 error = OFPROTO_POSTPONE;
4350 ovs_mutex_unlock(&ofproto_mutex);
4352 run_rule_executes(ofproto);
4357 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4359 struct ofputil_role_request request;
4360 struct ofputil_role_request reply;
4364 error = ofputil_decode_role_message(oh, &request);
4369 if (request.role != OFPCR12_ROLE_NOCHANGE) {
4370 if (ofconn_get_role(ofconn) != request.role
4371 && ofconn_has_pending_opgroups(ofconn)) {
4372 return OFPROTO_POSTPONE;
4375 if (request.have_generation_id
4376 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
4377 return OFPERR_OFPRRFC_STALE;
4380 ofconn_set_role(ofconn, request.role);
4383 reply.role = ofconn_get_role(ofconn);
4384 reply.have_generation_id = ofconn_get_master_election_id(
4385 ofconn, &reply.generation_id);
4386 buf = ofputil_encode_role_reply(oh, &reply);
4387 ofconn_send_reply(ofconn, buf);
4393 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
4394 const struct ofp_header *oh)
4396 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
4397 enum ofputil_protocol cur, next;
4399 cur = ofconn_get_protocol(ofconn);
4400 next = ofputil_protocol_set_tid(cur, msg->set != 0);
4401 ofconn_set_protocol(ofconn, next);
4407 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4409 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
4410 enum ofputil_protocol cur, next;
4411 enum ofputil_protocol next_base;
4413 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
4415 return OFPERR_OFPBRC_EPERM;
4418 cur = ofconn_get_protocol(ofconn);
4419 next = ofputil_protocol_set_base(cur, next_base);
4420 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
4421 /* Avoid sending async messages in surprising protocol. */
4422 return OFPROTO_POSTPONE;
4425 ofconn_set_protocol(ofconn, next);
4430 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
4431 const struct ofp_header *oh)
4433 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
4436 format = ntohl(msg->format);
4437 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
4438 return OFPERR_OFPBRC_EPERM;
4441 if (format != ofconn_get_packet_in_format(ofconn)
4442 && ofconn_has_pending_opgroups(ofconn)) {
4443 /* Avoid sending async message in surprsing packet in format. */
4444 return OFPROTO_POSTPONE;
4447 ofconn_set_packet_in_format(ofconn, format);
4452 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
4454 const struct nx_async_config *msg = ofpmsg_body(oh);
4455 uint32_t master[OAM_N_TYPES];
4456 uint32_t slave[OAM_N_TYPES];
4458 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
4459 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
4460 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
4462 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
4463 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
4464 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
4466 ofconn_set_async_config(ofconn, master, slave);
4467 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
4468 !ofconn_get_miss_send_len(ofconn)) {
4469 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
4476 handle_nxt_set_controller_id(struct ofconn *ofconn,
4477 const struct ofp_header *oh)
4479 const struct nx_controller_id *nci = ofpmsg_body(oh);
4481 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
4482 return OFPERR_NXBRC_MUST_BE_ZERO;
4485 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
4490 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4494 if (ofconn_has_pending_opgroups(ofconn)) {
4495 return OFPROTO_POSTPONE;
4498 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
4499 ? OFPRAW_OFPT10_BARRIER_REPLY
4500 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
4501 ofconn_send_reply(ofconn, buf);
4506 ofproto_compose_flow_refresh_update(const struct rule *rule,
4507 enum nx_flow_monitor_flags flags,
4509 OVS_REQUIRES(ofproto_mutex)
4511 struct ofoperation *op = rule->pending;
4512 const struct rule_actions *actions;
4513 struct ofputil_flow_update fu;
4516 if (op && op->type == OFOPERATION_ADD) {
4517 /* We'll report the final flow when the operation completes. Reporting
4518 * it now would cause a duplicate report later. */
4522 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
4523 ? NXFME_ADDED : NXFME_MODIFIED);
4525 ovs_mutex_lock(&rule->mutex);
4526 fu.idle_timeout = rule->idle_timeout;
4527 fu.hard_timeout = rule->hard_timeout;
4528 ovs_mutex_unlock(&rule->mutex);
4529 fu.table_id = rule->table_id;
4530 fu.cookie = rule->flow_cookie;
4531 minimatch_expand(&rule->cr.match, &match);
4533 fu.priority = rule->cr.priority;
4535 if (!(flags & NXFMF_ACTIONS)) {
4538 actions = rule->actions;
4540 /* An operation is in progress. Use the previous version of the flow's
4541 * actions, so that when the operation commits we report the change. */
4543 case OFOPERATION_ADD:
4546 case OFOPERATION_MODIFY:
4547 case OFOPERATION_REPLACE:
4548 actions = op->actions ? op->actions : rule->actions;
4551 case OFOPERATION_DELETE:
4552 actions = rule->actions;
4559 fu.ofpacts = actions ? actions->ofpacts : NULL;
4560 fu.ofpacts_len = actions ? actions->ofpacts_len : 0;
4562 if (list_is_empty(msgs)) {
4563 ofputil_start_flow_update(msgs);
4565 ofputil_append_flow_update(&fu, msgs);
4569 ofmonitor_compose_refresh_updates(struct rule_collection *rules,
4571 OVS_REQUIRES(ofproto_mutex)
4575 for (i = 0; i < rules->n; i++) {
4576 struct rule *rule = rules->rules[i];
4577 enum nx_flow_monitor_flags flags = rule->monitor_flags;
4578 rule->monitor_flags = 0;
4580 ofproto_compose_flow_refresh_update(rule, flags, msgs);
4585 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
4586 struct rule *rule, uint64_t seqno,
4587 struct rule_collection *rules)
4588 OVS_REQUIRES(ofproto_mutex)
4590 enum nx_flow_monitor_flags update;
4592 if (ofproto_rule_is_hidden(rule)) {
4597 ? ofoperation_has_out_port(rule->pending, m->out_port)
4598 : ofproto_rule_has_out_port(rule, m->out_port))) {
4603 if (rule->add_seqno > seqno) {
4604 update = NXFMF_ADD | NXFMF_MODIFY;
4605 } else if (rule->modify_seqno > seqno) {
4606 update = NXFMF_MODIFY;
4611 if (!(m->flags & update)) {
4615 update = NXFMF_INITIAL;
4618 if (!rule->monitor_flags) {
4619 rule_collection_add(rules, rule);
4621 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
4625 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
4627 struct rule_collection *rules)
4628 OVS_REQUIRES(ofproto_mutex)
4630 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
4631 const struct ofoperation *op;
4632 const struct oftable *table;
4633 struct cls_rule target;
4635 cls_rule_init_from_minimatch(&target, &m->match, 0);
4636 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
4637 struct cls_cursor cursor;
4640 ovs_rwlock_rdlock(&table->cls.rwlock);
4641 cls_cursor_init(&cursor, &table->cls, &target);
4642 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4643 ovs_assert(!rule->pending); /* XXX */
4644 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4646 ovs_rwlock_unlock(&table->cls.rwlock);
4649 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
4650 struct rule *rule = op->rule;
4652 if (((m->table_id == 0xff
4653 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
4654 : m->table_id == rule->table_id))
4655 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
4656 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4659 cls_rule_destroy(&target);
4663 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
4664 struct rule_collection *rules)
4665 OVS_REQUIRES(ofproto_mutex)
4667 if (m->flags & NXFMF_INITIAL) {
4668 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
4673 ofmonitor_collect_resume_rules(struct ofmonitor *m,
4674 uint64_t seqno, struct rule_collection *rules)
4675 OVS_REQUIRES(ofproto_mutex)
4677 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
4681 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
4682 OVS_EXCLUDED(ofproto_mutex)
4684 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4685 struct ofmonitor **monitors;
4686 size_t n_monitors, allocated_monitors;
4687 struct rule_collection rules;
4688 struct list replies;
4694 ofpbuf_use_const(&b, oh, ntohs(oh->length));
4696 n_monitors = allocated_monitors = 0;
4698 ovs_mutex_lock(&ofproto_mutex);
4700 struct ofputil_flow_monitor_request request;
4701 struct ofmonitor *m;
4704 retval = ofputil_decode_flow_monitor_request(&request, &b);
4705 if (retval == EOF) {
4707 } else if (retval) {
4712 if (request.table_id != 0xff
4713 && request.table_id >= ofproto->n_tables) {
4714 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4718 error = ofmonitor_create(&request, ofconn, &m);
4723 if (n_monitors >= allocated_monitors) {
4724 monitors = x2nrealloc(monitors, &allocated_monitors,
4727 monitors[n_monitors++] = m;
4730 rule_collection_init(&rules);
4731 for (i = 0; i < n_monitors; i++) {
4732 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
4735 ofpmp_init(&replies, oh);
4736 ofmonitor_compose_refresh_updates(&rules, &replies);
4737 ovs_mutex_unlock(&ofproto_mutex);
4739 rule_collection_destroy(&rules);
4741 ofconn_send_replies(ofconn, &replies);
4747 ovs_mutex_unlock(&ofproto_mutex);
4749 for (i = 0; i < n_monitors; i++) {
4750 ofmonitor_destroy(monitors[i]);
4757 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
4758 OVS_EXCLUDED(ofproto_mutex)
4760 struct ofmonitor *m;
4764 id = ofputil_decode_flow_monitor_cancel(oh);
4766 ovs_mutex_lock(&ofproto_mutex);
4767 m = ofmonitor_lookup(ofconn, id);
4769 ofmonitor_destroy(m);
4772 error = OFPERR_NXBRC_FM_BAD_ID;
4774 ovs_mutex_unlock(&ofproto_mutex);
4779 /* Meters implementation.
4781 * Meter table entry, indexed by the OpenFlow meter_id.
4782 * These are always dynamically allocated to allocate enough space for
4784 * 'created' is used to compute the duration for meter stats.
4785 * 'list rules' is needed so that we can delete the dependent rules when the
4786 * meter table entry is deleted.
4787 * 'provider_meter_id' is for the provider's private use.
4790 long long int created; /* Time created. */
4791 struct list rules; /* List of "struct rule_dpif"s. */
4792 ofproto_meter_id provider_meter_id;
4793 uint16_t flags; /* Meter flags. */
4794 uint16_t n_bands; /* Number of meter bands. */
4795 struct ofputil_meter_band *bands;
4799 * This is used in instruction validation at flow set-up time,
4800 * as flows may not use non-existing meters.
4801 * This is also used by ofproto-providers to translate OpenFlow meter_ids
4802 * in METER instructions to the corresponding provider meter IDs.
4803 * Return value of UINT32_MAX signifies an invalid meter.
4806 ofproto_get_provider_meter_id(const struct ofproto * ofproto,
4807 uint32_t of_meter_id)
4809 if (of_meter_id && of_meter_id <= ofproto->meter_features.max_meters) {
4810 const struct meter *meter = ofproto->meters[of_meter_id];
4812 return meter->provider_meter_id.uint32;
4819 meter_update(struct meter *meter, const struct ofputil_meter_config *config)
4823 meter->flags = config->flags;
4824 meter->n_bands = config->n_bands;
4825 meter->bands = xmemdup(config->bands,
4826 config->n_bands * sizeof *meter->bands);
4829 static struct meter *
4830 meter_create(const struct ofputil_meter_config *config,
4831 ofproto_meter_id provider_meter_id)
4833 struct meter *meter;
4835 meter = xzalloc(sizeof *meter);
4836 meter->provider_meter_id = provider_meter_id;
4837 meter->created = time_msec();
4838 list_init(&meter->rules);
4840 meter_update(meter, config);
4846 meter_delete(struct ofproto *ofproto, uint32_t first, uint32_t last)
4847 OVS_REQUIRES(ofproto_mutex)
4850 for (mid = first; mid <= last; ++mid) {
4851 struct meter *meter = ofproto->meters[mid];
4853 ofproto->meters[mid] = NULL;
4854 ofproto->ofproto_class->meter_del(ofproto,
4855 meter->provider_meter_id);
4863 handle_add_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
4865 ofproto_meter_id provider_meter_id = { UINT32_MAX };
4866 struct meter **meterp = &ofproto->meters[mm->meter.meter_id];
4870 return OFPERR_OFPMMFC_METER_EXISTS;
4873 error = ofproto->ofproto_class->meter_set(ofproto, &provider_meter_id,
4876 ovs_assert(provider_meter_id.uint32 != UINT32_MAX);
4877 *meterp = meter_create(&mm->meter, provider_meter_id);
4883 handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
4885 struct meter *meter = ofproto->meters[mm->meter.meter_id];
4889 return OFPERR_OFPMMFC_UNKNOWN_METER;
4892 error = ofproto->ofproto_class->meter_set(ofproto,
4893 &meter->provider_meter_id,
4895 ovs_assert(meter->provider_meter_id.uint32 != UINT32_MAX);
4897 meter_update(meter, &mm->meter);
4903 handle_delete_meter(struct ofconn *ofconn, const struct ofp_header *oh,
4904 struct ofputil_meter_mod *mm)
4905 OVS_EXCLUDED(ofproto_mutex)
4907 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4908 uint32_t meter_id = mm->meter.meter_id;
4909 struct rule_collection rules;
4910 enum ofperr error = 0;
4911 uint32_t first, last;
4913 if (meter_id == OFPM13_ALL) {
4915 last = ofproto->meter_features.max_meters;
4917 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
4920 first = last = meter_id;
4923 /* First delete the rules that use this meter. If any of those rules are
4924 * currently being modified, postpone the whole operation until later. */
4925 rule_collection_init(&rules);
4926 ovs_mutex_lock(&ofproto_mutex);
4927 for (meter_id = first; meter_id <= last; ++meter_id) {
4928 struct meter *meter = ofproto->meters[meter_id];
4929 if (meter && !list_is_empty(&meter->rules)) {
4932 LIST_FOR_EACH (rule, meter_list_node, &meter->rules) {
4933 if (rule->pending) {
4934 error = OFPROTO_POSTPONE;
4937 rule_collection_add(&rules, rule);
4942 delete_flows__(ofproto, ofconn, oh, &rules, OFPRR_METER_DELETE);
4945 /* Delete the meters. */
4946 meter_delete(ofproto, first, last);
4949 ovs_mutex_unlock(&ofproto_mutex);
4950 rule_collection_destroy(&rules);
4956 handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4958 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4959 struct ofputil_meter_mod mm;
4960 uint64_t bands_stub[256 / 8];
4961 struct ofpbuf bands;
4965 error = reject_slave_controller(ofconn);
4970 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
4972 error = ofputil_decode_meter_mod(oh, &mm, &bands);
4974 goto exit_free_bands;
4977 meter_id = mm.meter.meter_id;
4979 if (mm.command != OFPMC13_DELETE) {
4980 /* Fails also when meters are not implemented by the provider. */
4981 if (meter_id == 0 || meter_id > OFPM13_MAX) {
4982 error = OFPERR_OFPMMFC_INVALID_METER;
4983 goto exit_free_bands;
4984 } else if (meter_id > ofproto->meter_features.max_meters) {
4985 error = OFPERR_OFPMMFC_OUT_OF_METERS;
4986 goto exit_free_bands;
4988 if (mm.meter.n_bands > ofproto->meter_features.max_bands) {
4989 error = OFPERR_OFPMMFC_OUT_OF_BANDS;
4990 goto exit_free_bands;
4994 switch (mm.command) {
4996 error = handle_add_meter(ofproto, &mm);
4999 case OFPMC13_MODIFY:
5000 error = handle_modify_meter(ofproto, &mm);
5003 case OFPMC13_DELETE:
5004 error = handle_delete_meter(ofconn, oh, &mm);
5008 error = OFPERR_OFPMMFC_BAD_COMMAND;
5013 ofpbuf_uninit(&bands);
5018 handle_meter_features_request(struct ofconn *ofconn,
5019 const struct ofp_header *request)
5021 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5022 struct ofputil_meter_features features;
5025 if (ofproto->ofproto_class->meter_get_features) {
5026 ofproto->ofproto_class->meter_get_features(ofproto, &features);
5028 memset(&features, 0, sizeof features);
5030 b = ofputil_encode_meter_features_reply(&features, request);
5032 ofconn_send_reply(ofconn, b);
5037 handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request,
5040 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5041 struct list replies;
5042 uint64_t bands_stub[256 / 8];
5043 struct ofpbuf bands;
5044 uint32_t meter_id, first, last;
5046 ofputil_decode_meter_request(request, &meter_id);
5048 if (meter_id == OFPM13_ALL) {
5050 last = ofproto->meter_features.max_meters;
5052 if (!meter_id || meter_id > ofproto->meter_features.max_meters ||
5053 !ofproto->meters[meter_id]) {
5054 return OFPERR_OFPMMFC_UNKNOWN_METER;
5056 first = last = meter_id;
5059 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5060 ofpmp_init(&replies, request);
5062 for (meter_id = first; meter_id <= last; ++meter_id) {
5063 struct meter *meter = ofproto->meters[meter_id];
5065 continue; /* Skip non-existing meters. */
5067 if (type == OFPTYPE_METER_STATS_REQUEST) {
5068 struct ofputil_meter_stats stats;
5070 stats.meter_id = meter_id;
5072 /* Provider sets the packet and byte counts, we do the rest. */
5073 stats.flow_count = list_size(&meter->rules);
5074 calc_duration(meter->created, time_msec(),
5075 &stats.duration_sec, &stats.duration_nsec);
5076 stats.n_bands = meter->n_bands;
5077 ofpbuf_clear(&bands);
5079 = ofpbuf_put_uninit(&bands,
5080 meter->n_bands * sizeof *stats.bands);
5082 if (!ofproto->ofproto_class->meter_get(ofproto,
5083 meter->provider_meter_id,
5085 ofputil_append_meter_stats(&replies, &stats);
5087 } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */
5088 struct ofputil_meter_config config;
5090 config.meter_id = meter_id;
5091 config.flags = meter->flags;
5092 config.n_bands = meter->n_bands;
5093 config.bands = meter->bands;
5094 ofputil_append_meter_config(&replies, &config);
5098 ofconn_send_replies(ofconn, &replies);
5099 ofpbuf_uninit(&bands);
5104 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
5105 OVS_EXCLUDED(ofproto_mutex)
5107 const struct ofp_header *oh = msg->data;
5111 error = ofptype_decode(&type, oh);
5117 /* OpenFlow requests. */
5118 case OFPTYPE_ECHO_REQUEST:
5119 return handle_echo_request(ofconn, oh);
5121 case OFPTYPE_FEATURES_REQUEST:
5122 return handle_features_request(ofconn, oh);
5124 case OFPTYPE_GET_CONFIG_REQUEST:
5125 return handle_get_config_request(ofconn, oh);
5127 case OFPTYPE_SET_CONFIG:
5128 return handle_set_config(ofconn, oh);
5130 case OFPTYPE_PACKET_OUT:
5131 return handle_packet_out(ofconn, oh);
5133 case OFPTYPE_PORT_MOD:
5134 return handle_port_mod(ofconn, oh);
5136 case OFPTYPE_FLOW_MOD:
5137 return handle_flow_mod(ofconn, oh);
5139 case OFPTYPE_METER_MOD:
5140 return handle_meter_mod(ofconn, oh);
5142 case OFPTYPE_BARRIER_REQUEST:
5143 return handle_barrier_request(ofconn, oh);
5145 case OFPTYPE_ROLE_REQUEST:
5146 return handle_role_request(ofconn, oh);
5148 /* OpenFlow replies. */
5149 case OFPTYPE_ECHO_REPLY:
5152 /* Nicira extension requests. */
5153 case OFPTYPE_FLOW_MOD_TABLE_ID:
5154 return handle_nxt_flow_mod_table_id(ofconn, oh);
5156 case OFPTYPE_SET_FLOW_FORMAT:
5157 return handle_nxt_set_flow_format(ofconn, oh);
5159 case OFPTYPE_SET_PACKET_IN_FORMAT:
5160 return handle_nxt_set_packet_in_format(ofconn, oh);
5162 case OFPTYPE_SET_CONTROLLER_ID:
5163 return handle_nxt_set_controller_id(ofconn, oh);
5165 case OFPTYPE_FLOW_AGE:
5166 /* Nothing to do. */
5169 case OFPTYPE_FLOW_MONITOR_CANCEL:
5170 return handle_flow_monitor_cancel(ofconn, oh);
5172 case OFPTYPE_SET_ASYNC_CONFIG:
5173 return handle_nxt_set_async_config(ofconn, oh);
5175 /* Statistics requests. */
5176 case OFPTYPE_DESC_STATS_REQUEST:
5177 return handle_desc_stats_request(ofconn, oh);
5179 case OFPTYPE_FLOW_STATS_REQUEST:
5180 return handle_flow_stats_request(ofconn, oh);
5182 case OFPTYPE_AGGREGATE_STATS_REQUEST:
5183 return handle_aggregate_stats_request(ofconn, oh);
5185 case OFPTYPE_TABLE_STATS_REQUEST:
5186 return handle_table_stats_request(ofconn, oh);
5188 case OFPTYPE_PORT_STATS_REQUEST:
5189 return handle_port_stats_request(ofconn, oh);
5191 case OFPTYPE_QUEUE_STATS_REQUEST:
5192 return handle_queue_stats_request(ofconn, oh);
5194 case OFPTYPE_PORT_DESC_STATS_REQUEST:
5195 return handle_port_desc_stats_request(ofconn, oh);
5197 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
5198 return handle_flow_monitor_request(ofconn, oh);
5200 case OFPTYPE_METER_STATS_REQUEST:
5201 case OFPTYPE_METER_CONFIG_STATS_REQUEST:
5202 return handle_meter_request(ofconn, oh, type);
5204 case OFPTYPE_METER_FEATURES_STATS_REQUEST:
5205 return handle_meter_features_request(ofconn, oh);
5207 /* FIXME: Change the following once they are implemented: */
5208 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
5209 case OFPTYPE_GET_ASYNC_REQUEST:
5210 case OFPTYPE_GROUP_STATS_REQUEST:
5211 case OFPTYPE_GROUP_DESC_STATS_REQUEST:
5212 case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
5213 case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
5214 return OFPERR_OFPBRC_BAD_TYPE;
5218 case OFPTYPE_FEATURES_REPLY:
5219 case OFPTYPE_GET_CONFIG_REPLY:
5220 case OFPTYPE_PACKET_IN:
5221 case OFPTYPE_FLOW_REMOVED:
5222 case OFPTYPE_PORT_STATUS:
5223 case OFPTYPE_BARRIER_REPLY:
5224 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
5225 case OFPTYPE_DESC_STATS_REPLY:
5226 case OFPTYPE_FLOW_STATS_REPLY:
5227 case OFPTYPE_QUEUE_STATS_REPLY:
5228 case OFPTYPE_PORT_STATS_REPLY:
5229 case OFPTYPE_TABLE_STATS_REPLY:
5230 case OFPTYPE_AGGREGATE_STATS_REPLY:
5231 case OFPTYPE_PORT_DESC_STATS_REPLY:
5232 case OFPTYPE_ROLE_REPLY:
5233 case OFPTYPE_FLOW_MONITOR_PAUSED:
5234 case OFPTYPE_FLOW_MONITOR_RESUMED:
5235 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
5236 case OFPTYPE_GET_ASYNC_REPLY:
5237 case OFPTYPE_GROUP_STATS_REPLY:
5238 case OFPTYPE_GROUP_DESC_STATS_REPLY:
5239 case OFPTYPE_GROUP_FEATURES_STATS_REPLY:
5240 case OFPTYPE_METER_STATS_REPLY:
5241 case OFPTYPE_METER_CONFIG_STATS_REPLY:
5242 case OFPTYPE_METER_FEATURES_STATS_REPLY:
5243 case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
5245 return OFPERR_OFPBRC_BAD_TYPE;
5250 handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
5251 OVS_EXCLUDED(ofproto_mutex)
5253 int error = handle_openflow__(ofconn, ofp_msg);
5254 if (error && error != OFPROTO_POSTPONE) {
5255 ofconn_send_error(ofconn, ofp_msg->data, error);
5257 COVERAGE_INC(ofproto_recv_openflow);
5258 return error != OFPROTO_POSTPONE;
5261 /* Asynchronous operations. */
5263 /* Creates and returns a new ofopgroup that is not associated with any
5264 * OpenFlow connection.
5266 * The caller should add operations to the returned group with
5267 * ofoperation_create() and then submit it with ofopgroup_submit(). */
5268 static struct ofopgroup *
5269 ofopgroup_create_unattached(struct ofproto *ofproto)
5270 OVS_REQUIRES(ofproto_mutex)
5272 struct ofopgroup *group = xzalloc(sizeof *group);
5273 group->ofproto = ofproto;
5274 list_init(&group->ofproto_node);
5275 list_init(&group->ops);
5276 list_init(&group->ofconn_node);
5280 /* Creates and returns a new ofopgroup for 'ofproto'.
5282 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
5283 * connection. The 'request' and 'buffer_id' arguments are ignored.
5285 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
5286 * If the ofopgroup eventually fails, then the error reply will include
5287 * 'request'. If the ofopgroup eventually succeeds, then the packet with
5288 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
5290 * The caller should add operations to the returned group with
5291 * ofoperation_create() and then submit it with ofopgroup_submit(). */
5292 static struct ofopgroup *
5293 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
5294 const struct ofp_header *request, uint32_t buffer_id)
5295 OVS_REQUIRES(ofproto_mutex)
5297 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
5299 size_t request_len = ntohs(request->length);
5301 ovs_assert(ofconn_get_ofproto(ofconn) == ofproto);
5303 ofconn_add_opgroup(ofconn, &group->ofconn_node);
5304 group->ofconn = ofconn;
5305 group->request = xmemdup(request, MIN(request_len, 64));
5306 group->buffer_id = buffer_id;
5311 /* Submits 'group' for processing.
5313 * If 'group' contains no operations (e.g. none were ever added, or all of the
5314 * ones that were added completed synchronously), then it is destroyed
5315 * immediately. Otherwise it is added to the ofproto's list of pending
5318 ofopgroup_submit(struct ofopgroup *group)
5319 OVS_REQUIRES(ofproto_mutex)
5321 if (!group->n_running) {
5322 ofopgroup_complete(group);
5324 list_push_back(&group->ofproto->pending, &group->ofproto_node);
5325 group->ofproto->n_pending++;
5330 ofopgroup_complete(struct ofopgroup *group)
5331 OVS_REQUIRES(ofproto_mutex)
5333 struct ofproto *ofproto = group->ofproto;
5335 struct ofconn *abbrev_ofconn;
5336 ovs_be32 abbrev_xid;
5338 struct ofoperation *op, *next_op;
5341 ovs_assert(!group->n_running);
5344 LIST_FOR_EACH (op, group_node, &group->ops) {
5351 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
5352 LIST_FOR_EACH (op, group_node, &group->ops) {
5353 if (op->type != OFOPERATION_DELETE) {
5354 struct ofpbuf *packet;
5357 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
5360 struct rule_execute *re;
5364 ofproto_rule_ref(op->rule);
5366 re = xmalloc(sizeof *re);
5367 re->rule = op->rule;
5368 re->in_port = in_port;
5369 re->packet = packet;
5371 if (!guarded_list_push_back(&ofproto->rule_executes,
5372 &re->list_node, 1024)) {
5373 ofproto_rule_unref(op->rule);
5374 ofpbuf_delete(re->packet);
5383 if (!error && !list_is_empty(&group->ofconn_node)) {
5384 abbrev_ofconn = group->ofconn;
5385 abbrev_xid = group->request->xid;
5387 abbrev_ofconn = NULL;
5388 abbrev_xid = htonl(0);
5390 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
5391 struct rule *rule = op->rule;
5393 /* We generally want to report the change to active OpenFlow flow
5394 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
5396 - The operation failed.
5398 - The affected rule is not visible to controllers.
5400 - The operation's only effect was to update rule->modified. */
5402 || ofproto_rule_is_hidden(rule)
5403 || (op->type == OFOPERATION_MODIFY
5405 && rule->flow_cookie == op->flow_cookie))) {
5406 /* Check that we can just cast from ofoperation_type to
5407 * nx_flow_update_event. */
5408 enum nx_flow_update_event event_type;
5411 case OFOPERATION_ADD:
5412 case OFOPERATION_REPLACE:
5413 event_type = NXFME_ADDED;
5416 case OFOPERATION_DELETE:
5417 event_type = NXFME_DELETED;
5420 case OFOPERATION_MODIFY:
5421 event_type = NXFME_MODIFIED;
5428 ofmonitor_report(ofproto->connmgr, rule, event_type,
5429 op->reason, abbrev_ofconn, abbrev_xid);
5432 rule->pending = NULL;
5435 case OFOPERATION_ADD:
5439 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
5440 if (vid_mask == VLAN_VID_MASK) {
5441 if (ofproto->vlan_bitmap) {
5442 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
5443 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
5444 bitmap_set1(ofproto->vlan_bitmap, vid);
5445 ofproto->vlans_changed = true;
5448 ofproto->vlans_changed = true;
5452 oftable_remove_rule(rule);
5453 ofproto_rule_unref(rule);
5457 case OFOPERATION_DELETE:
5458 ovs_assert(!op->error);
5459 ofproto_rule_unref(rule);
5463 case OFOPERATION_MODIFY:
5464 case OFOPERATION_REPLACE:
5466 long long int now = time_msec();
5468 rule->modified = now;
5469 if (op->type == OFOPERATION_REPLACE) {
5470 rule->created = rule->used = now;
5473 ofproto_rule_change_cookie(ofproto, rule, op->flow_cookie);
5474 ovs_mutex_lock(&rule->mutex);
5475 rule->idle_timeout = op->idle_timeout;
5476 rule->hard_timeout = op->hard_timeout;
5477 ovs_mutex_unlock(&rule->mutex);
5479 struct rule_actions *old_actions;
5481 ovs_mutex_lock(&rule->mutex);
5482 old_actions = rule->actions;
5483 rule->actions = op->actions;
5484 ovs_mutex_unlock(&rule->mutex);
5487 rule_actions_unref(old_actions);
5489 rule->send_flow_removed = op->send_flow_removed;
5497 ofoperation_destroy(op);
5500 ofmonitor_flush(ofproto->connmgr);
5502 if (!list_is_empty(&group->ofproto_node)) {
5503 ovs_assert(ofproto->n_pending > 0);
5504 ofproto->n_pending--;
5505 list_remove(&group->ofproto_node);
5507 if (!list_is_empty(&group->ofconn_node)) {
5508 list_remove(&group->ofconn_node);
5510 ofconn_send_error(group->ofconn, group->request, error);
5512 connmgr_retry(ofproto->connmgr);
5514 free(group->request);
5518 /* Initiates a new operation on 'rule', of the specified 'type', within
5519 * 'group'. Prior to calling, 'rule' must not have any pending operation.
5521 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
5522 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
5524 * Returns the newly created ofoperation (which is also available as
5525 * rule->pending). */
5526 static struct ofoperation *
5527 ofoperation_create(struct ofopgroup *group, struct rule *rule,
5528 enum ofoperation_type type,
5529 enum ofp_flow_removed_reason reason)
5530 OVS_REQUIRES(ofproto_mutex)
5532 struct ofproto *ofproto = group->ofproto;
5533 struct ofoperation *op;
5535 ovs_assert(!rule->pending);
5537 op = rule->pending = xzalloc(sizeof *op);
5539 list_push_back(&group->ops, &op->group_node);
5542 op->reason = reason;
5543 op->flow_cookie = rule->flow_cookie;
5544 ovs_mutex_lock(&rule->mutex);
5545 op->idle_timeout = rule->idle_timeout;
5546 op->hard_timeout = rule->hard_timeout;
5547 ovs_mutex_unlock(&rule->mutex);
5548 op->send_flow_removed = rule->send_flow_removed;
5552 if (type == OFOPERATION_DELETE) {
5553 hmap_insert(&ofproto->deletions, &op->hmap_node,
5554 cls_rule_hash(&rule->cr, rule->table_id));
5561 ofoperation_destroy(struct ofoperation *op)
5562 OVS_REQUIRES(ofproto_mutex)
5564 struct ofopgroup *group = op->group;
5567 op->rule->pending = NULL;
5569 if (op->type == OFOPERATION_DELETE) {
5570 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
5572 list_remove(&op->group_node);
5573 rule_actions_unref(op->actions);
5577 /* Indicates that 'op' completed with status 'error', which is either 0 to
5578 * indicate success or an OpenFlow error code on failure.
5580 * If 'error' is 0, indicating success, the operation will be committed
5581 * permanently to the flow table.
5583 * If 'error' is nonzero, then generally the operation will be rolled back:
5585 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
5586 * restores the original rule. The caller must have uninitialized any
5587 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
5588 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
5589 * and 7 for the new rule, calling its ->rule_dealloc() function.
5591 * - If 'op' is a "modify flow" operation, ofproto restores the original
5594 * - 'op' must not be a "delete flow" operation. Removing a rule is not
5595 * allowed to fail. It must always succeed.
5597 * Please see the large comment in ofproto/ofproto-provider.h titled
5598 * "Asynchronous Operation Support" for more information. */
5600 ofoperation_complete(struct ofoperation *op, enum ofperr error)
5602 struct ofopgroup *group = op->group;
5604 ovs_assert(group->n_running > 0);
5605 ovs_assert(!error || op->type != OFOPERATION_DELETE);
5608 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
5609 /* This function can be called from ->rule_construct(), in which case
5610 * ofproto_mutex is held, or it can be called from ->run(), in which
5611 * case ofproto_mutex is not held. But only in the latter case can we
5612 * arrive here, so we can safely take ofproto_mutex now. */
5613 ovs_mutex_lock(&ofproto_mutex);
5614 ovs_assert(op->rule->pending == op);
5615 ofopgroup_complete(group);
5616 ovs_mutex_unlock(&ofproto_mutex);
5621 pick_datapath_id(const struct ofproto *ofproto)
5623 const struct ofport *port;
5625 port = ofproto_get_port(ofproto, OFPP_LOCAL);
5627 uint8_t ea[ETH_ADDR_LEN];
5630 error = netdev_get_etheraddr(port->netdev, ea);
5632 return eth_addr_to_uint64(ea);
5634 VLOG_WARN("%s: could not get MAC address for %s (%s)",
5635 ofproto->name, netdev_get_name(port->netdev),
5636 ovs_strerror(error));
5638 return ofproto->fallback_dpid;
5642 pick_fallback_dpid(void)
5644 uint8_t ea[ETH_ADDR_LEN];
5645 eth_addr_nicira_random(ea);
5646 return eth_addr_to_uint64(ea);
5649 /* Table overflow policy. */
5651 /* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep'
5652 * to NULL if the table is not configured to evict rules or if the table
5653 * contains no evictable rules. (Rules with a readlock on their evict rwlock,
5654 * or with no timeouts are not evictable.) */
5656 choose_rule_to_evict(struct oftable *table, struct rule **rulep)
5657 OVS_REQUIRES(ofproto_mutex)
5659 struct eviction_group *evg;
5662 if (!table->eviction_fields) {
5666 /* In the common case, the outer and inner loops here will each be entered
5669 * - The inner loop normally "return"s in its first iteration. If the
5670 * eviction group has any evictable rules, then it always returns in
5673 * - The outer loop only iterates more than once if the largest eviction
5674 * group has no evictable rules.
5676 * - The outer loop can exit only if table's 'max_flows' is all filled up
5677 * by unevictable rules. */
5678 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
5681 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
5690 /* Searches 'ofproto' for tables that have more flows than their configured
5691 * maximum and that have flow eviction enabled, and evicts as many flows as
5692 * necessary and currently feasible from them.
5694 * This triggers only when an OpenFlow table has N flows in it and then the
5695 * client configures a maximum number of flows less than N. */
5697 ofproto_evict(struct ofproto *ofproto)
5699 struct oftable *table;
5701 ovs_mutex_lock(&ofproto_mutex);
5702 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
5703 evict_rules_from_table(ofproto, table, 0);
5705 ovs_mutex_unlock(&ofproto_mutex);
5708 /* Eviction groups. */
5710 /* Returns the priority to use for an eviction_group that contains 'n_rules'
5711 * rules. The priority contains low-order random bits to ensure that eviction
5712 * groups with the same number of rules are prioritized randomly. */
5714 eviction_group_priority(size_t n_rules)
5716 uint16_t size = MIN(UINT16_MAX, n_rules);
5717 return (size << 16) | random_uint16();
5720 /* Updates 'evg', an eviction_group within 'table', following a change that
5721 * adds or removes rules in 'evg'. */
5723 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
5724 OVS_REQUIRES(ofproto_mutex)
5726 heap_change(&table->eviction_groups_by_size, &evg->size_node,
5727 eviction_group_priority(heap_count(&evg->rules)));
5730 /* Destroys 'evg', an eviction_group within 'table':
5732 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
5733 * rules themselves, just removes them from the eviction group.)
5735 * - Removes 'evg' from 'table'.
5739 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
5740 OVS_REQUIRES(ofproto_mutex)
5742 while (!heap_is_empty(&evg->rules)) {
5745 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
5746 rule->eviction_group = NULL;
5748 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
5749 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
5750 heap_destroy(&evg->rules);
5754 /* Removes 'rule' from its eviction group, if any. */
5756 eviction_group_remove_rule(struct rule *rule)
5757 OVS_REQUIRES(ofproto_mutex)
5759 if (rule->eviction_group) {
5760 struct oftable *table = &rule->ofproto->tables[rule->table_id];
5761 struct eviction_group *evg = rule->eviction_group;
5763 rule->eviction_group = NULL;
5764 heap_remove(&evg->rules, &rule->evg_node);
5765 if (heap_is_empty(&evg->rules)) {
5766 eviction_group_destroy(table, evg);
5768 eviction_group_resized(table, evg);
5773 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
5774 * returns the hash value. */
5776 eviction_group_hash_rule(struct rule *rule)
5777 OVS_REQUIRES(ofproto_mutex)
5779 struct oftable *table = &rule->ofproto->tables[rule->table_id];
5780 const struct mf_subfield *sf;
5784 hash = table->eviction_group_id_basis;
5785 miniflow_expand(&rule->cr.match.flow, &flow);
5786 for (sf = table->eviction_fields;
5787 sf < &table->eviction_fields[table->n_eviction_fields];
5790 if (mf_are_prereqs_ok(sf->field, &flow)) {
5791 union mf_value value;
5793 mf_get_value(sf->field, &flow, &value);
5795 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
5797 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
5798 unsigned int start = sf->ofs + sf->n_bits;
5799 bitwise_zero(&value, sf->field->n_bytes, start,
5800 sf->field->n_bytes * 8 - start);
5802 hash = hash_bytes(&value, sf->field->n_bytes, hash);
5804 hash = hash_int(hash, 0);
5811 /* Returns an eviction group within 'table' with the given 'id', creating one
5813 static struct eviction_group *
5814 eviction_group_find(struct oftable *table, uint32_t id)
5815 OVS_REQUIRES(ofproto_mutex)
5817 struct eviction_group *evg;
5819 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
5823 evg = xmalloc(sizeof *evg);
5824 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
5825 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
5826 eviction_group_priority(0));
5827 heap_init(&evg->rules);
5832 /* Returns an eviction priority for 'rule'. The return value should be
5833 * interpreted so that higher priorities make a rule more attractive candidates
5836 rule_eviction_priority(struct rule *rule)
5837 OVS_REQUIRES(ofproto_mutex)
5839 long long int hard_expiration;
5840 long long int idle_expiration;
5841 long long int expiration;
5842 uint32_t expiration_offset;
5844 /* Calculate time of expiration. */
5845 ovs_mutex_lock(&rule->mutex);
5846 hard_expiration = (rule->hard_timeout
5847 ? rule->modified + rule->hard_timeout * 1000
5849 idle_expiration = (rule->idle_timeout
5850 ? rule->used + rule->idle_timeout * 1000
5852 expiration = MIN(hard_expiration, idle_expiration);
5853 ovs_mutex_unlock(&rule->mutex);
5854 if (expiration == LLONG_MAX) {
5858 /* Calculate the time of expiration as a number of (approximate) seconds
5859 * after program startup.
5861 * This should work OK for program runs that last UINT32_MAX seconds or
5862 * less. Therefore, please restart OVS at least once every 136 years. */
5863 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
5865 /* Invert the expiration offset because we're using a max-heap. */
5866 return UINT32_MAX - expiration_offset;
5869 /* Adds 'rule' to an appropriate eviction group for its oftable's
5870 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
5871 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
5874 * The caller must ensure that 'rule' is not already in an eviction group. */
5876 eviction_group_add_rule(struct rule *rule)
5877 OVS_REQUIRES(ofproto_mutex)
5879 struct ofproto *ofproto = rule->ofproto;
5880 struct oftable *table = &ofproto->tables[rule->table_id];
5883 ovs_mutex_lock(&rule->mutex);
5884 has_timeout = rule->hard_timeout || rule->idle_timeout;
5885 ovs_mutex_unlock(&rule->mutex);
5887 if (table->eviction_fields && has_timeout) {
5888 struct eviction_group *evg;
5890 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
5892 rule->eviction_group = evg;
5893 heap_insert(&evg->rules, &rule->evg_node,
5894 rule_eviction_priority(rule));
5895 eviction_group_resized(table, evg);
5901 /* Initializes 'table'. */
5903 oftable_init(struct oftable *table)
5905 memset(table, 0, sizeof *table);
5906 classifier_init(&table->cls);
5907 table->max_flows = UINT_MAX;
5910 /* Destroys 'table', including its classifier and eviction groups.
5912 * The caller is responsible for freeing 'table' itself. */
5914 oftable_destroy(struct oftable *table)
5916 ovs_rwlock_rdlock(&table->cls.rwlock);
5917 ovs_assert(classifier_is_empty(&table->cls));
5918 ovs_rwlock_unlock(&table->cls.rwlock);
5919 oftable_disable_eviction(table);
5920 classifier_destroy(&table->cls);
5924 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
5925 * string, then 'table' will use its default name.
5927 * This only affects the name exposed for a table exposed through the OpenFlow
5928 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
5930 oftable_set_name(struct oftable *table, const char *name)
5932 if (name && name[0]) {
5933 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
5934 if (!table->name || strncmp(name, table->name, len)) {
5936 table->name = xmemdup0(name, len);
5944 /* oftables support a choice of two policies when adding a rule would cause the
5945 * number of flows in the table to exceed the configured maximum number: either
5946 * they can refuse to add the new flow or they can evict some existing flow.
5947 * This function configures the former policy on 'table'. */
5949 oftable_disable_eviction(struct oftable *table)
5950 OVS_REQUIRES(ofproto_mutex)
5952 if (table->eviction_fields) {
5953 struct eviction_group *evg, *next;
5955 HMAP_FOR_EACH_SAFE (evg, next, id_node,
5956 &table->eviction_groups_by_id) {
5957 eviction_group_destroy(table, evg);
5959 hmap_destroy(&table->eviction_groups_by_id);
5960 heap_destroy(&table->eviction_groups_by_size);
5962 free(table->eviction_fields);
5963 table->eviction_fields = NULL;
5964 table->n_eviction_fields = 0;
5968 /* oftables support a choice of two policies when adding a rule would cause the
5969 * number of flows in the table to exceed the configured maximum number: either
5970 * they can refuse to add the new flow or they can evict some existing flow.
5971 * This function configures the latter policy on 'table', with fairness based
5972 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
5973 * 'n_fields' as 0 disables fairness.) */
5975 oftable_enable_eviction(struct oftable *table,
5976 const struct mf_subfield *fields, size_t n_fields)
5977 OVS_REQUIRES(ofproto_mutex)
5979 struct cls_cursor cursor;
5982 if (table->eviction_fields
5983 && n_fields == table->n_eviction_fields
5985 || !memcmp(fields, table->eviction_fields,
5986 n_fields * sizeof *fields))) {
5991 oftable_disable_eviction(table);
5993 table->n_eviction_fields = n_fields;
5994 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
5996 table->eviction_group_id_basis = random_uint32();
5997 hmap_init(&table->eviction_groups_by_id);
5998 heap_init(&table->eviction_groups_by_size);
6000 ovs_rwlock_rdlock(&table->cls.rwlock);
6001 cls_cursor_init(&cursor, &table->cls, NULL);
6002 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
6003 eviction_group_add_rule(rule);
6005 ovs_rwlock_unlock(&table->cls.rwlock);
6008 /* Removes 'rule' from the oftable that contains it. */
6010 oftable_remove_rule__(struct ofproto *ofproto, struct rule *rule)
6011 OVS_REQUIRES(ofproto_mutex)
6013 struct classifier *cls = &ofproto->tables[rule->table_id].cls;
6015 ovs_rwlock_wrlock(&cls->rwlock);
6016 classifier_remove(cls, CONST_CAST(struct cls_rule *, &rule->cr));
6017 ovs_rwlock_unlock(&cls->rwlock);
6019 cookies_remove(ofproto, rule);
6021 eviction_group_remove_rule(rule);
6022 if (!list_is_empty(&rule->expirable)) {
6023 list_remove(&rule->expirable);
6025 if (!list_is_empty(&rule->meter_list_node)) {
6026 list_remove(&rule->meter_list_node);
6027 list_init(&rule->meter_list_node);
6032 oftable_remove_rule(struct rule *rule)
6033 OVS_REQUIRES(ofproto_mutex)
6035 oftable_remove_rule__(rule->ofproto, rule);
6038 /* Inserts 'rule' into its oftable, which must not already contain any rule for
6039 * the same cls_rule. */
6041 oftable_insert_rule(struct rule *rule)
6042 OVS_REQUIRES(ofproto_mutex)
6044 struct ofproto *ofproto = rule->ofproto;
6045 struct oftable *table = &ofproto->tables[rule->table_id];
6048 ovs_mutex_lock(&rule->mutex);
6049 may_expire = rule->hard_timeout || rule->idle_timeout;
6050 ovs_mutex_unlock(&rule->mutex);
6053 list_insert(&ofproto->expirable, &rule->expirable);
6056 cookies_insert(ofproto, rule);
6058 if (rule->actions->meter_id) {
6059 struct meter *meter = ofproto->meters[rule->actions->meter_id];
6060 list_insert(&meter->rules, &rule->meter_list_node);
6062 ovs_rwlock_wrlock(&table->cls.rwlock);
6063 classifier_insert(&table->cls, CONST_CAST(struct cls_rule *, &rule->cr));
6064 ovs_rwlock_unlock(&table->cls.rwlock);
6065 eviction_group_add_rule(rule);
6068 /* unixctl commands. */
6071 ofproto_lookup(const char *name)
6073 struct ofproto *ofproto;
6075 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
6077 if (!strcmp(ofproto->name, name)) {
6085 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
6086 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6088 struct ofproto *ofproto;
6092 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
6093 ds_put_format(&results, "%s\n", ofproto->name);
6095 unixctl_command_reply(conn, ds_cstr(&results));
6096 ds_destroy(&results);
6100 ofproto_unixctl_init(void)
6102 static bool registered;
6108 unixctl_command_register("ofproto/list", "", 0, 0,
6109 ofproto_unixctl_list, NULL);
6112 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6114 * This is deprecated. It is only for compatibility with broken device drivers
6115 * in old versions of Linux that do not properly support VLANs when VLAN
6116 * devices are not used. When broken device drivers are no longer in
6117 * widespread use, we will delete these interfaces. */
6119 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
6120 * (exactly) by an OpenFlow rule in 'ofproto'. */
6122 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
6124 const struct oftable *oftable;
6126 free(ofproto->vlan_bitmap);
6127 ofproto->vlan_bitmap = bitmap_allocate(4096);
6128 ofproto->vlans_changed = false;
6130 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
6131 const struct cls_table *table;
6133 ovs_rwlock_rdlock(&oftable->cls.rwlock);
6134 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
6135 if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
6136 const struct cls_rule *rule;
6138 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
6139 uint16_t vid = miniflow_get_vid(&rule->match.flow);
6140 bitmap_set1(vlan_bitmap, vid);
6141 bitmap_set1(ofproto->vlan_bitmap, vid);
6145 ovs_rwlock_unlock(&oftable->cls.rwlock);
6149 /* Returns true if new VLANs have come into use by the flow table since the
6150 * last call to ofproto_get_vlan_usage().
6152 * We don't track when old VLANs stop being used. */
6154 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
6156 return ofproto->vlans_changed;
6159 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
6160 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
6161 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
6162 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
6163 * then the VLAN device is un-enslaved. */
6165 ofproto_port_set_realdev(struct ofproto *ofproto, ofp_port_t vlandev_ofp_port,
6166 ofp_port_t realdev_ofp_port, int vid)
6168 struct ofport *ofport;
6171 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
6173 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
6175 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
6176 ofproto->name, vlandev_ofp_port);
6180 if (!ofproto->ofproto_class->set_realdev) {
6181 if (!vlandev_ofp_port) {
6184 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
6188 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
6190 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
6191 ofproto->name, vlandev_ofp_port,
6192 netdev_get_name(ofport->netdev), ovs_strerror(error));