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 enum ofputil_flow_mod_flags flags; /* Rule's old flags. */
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_RELEASES(rule->rwlock);
157 static void oftable_remove_rule__(struct ofproto *ofproto,
158 struct classifier *cls, struct rule *rule)
159 OVS_REQ_WRLOCK(cls->rwlock) OVS_RELEASES(rule->rwlock);
160 static void oftable_insert_rule(struct rule *);
162 /* A set of rules within a single OpenFlow table (oftable) that have the same
163 * values for the oftable's eviction_fields. A rule to be evicted, when one is
164 * needed, is taken from the eviction group that contains the greatest number
167 * An oftable owns any number of eviction groups, each of which contains any
170 * Membership in an eviction group is imprecise, based on the hash of the
171 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
172 * That is, if two rules have different eviction_fields, but those
173 * eviction_fields hash to the same value, then they will belong to the same
174 * eviction_group anyway.
176 * (When eviction is not enabled on an oftable, we don't track any eviction
177 * groups, to save time and space.) */
178 struct eviction_group {
179 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
180 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
181 struct heap rules; /* Contains "struct rule"s. */
184 static bool choose_rule_to_evict(struct oftable *table, struct rule **rulep)
185 OVS_TRY_WRLOCK(true, (*rulep)->rwlock);
186 static void ofproto_evict(struct ofproto *);
187 static uint32_t rule_eviction_priority(struct rule *);
188 static void eviction_group_add_rule(struct rule *);
189 static void eviction_group_remove_rule(struct rule *);
191 /* Criteria that flow_mod and other operations use for selecting rules on
192 * which to operate. */
193 struct rule_criteria {
194 /* An OpenFlow table or 255 for all tables. */
197 /* OpenFlow matching criteria. Interpreted different in "loose" way by
198 * collect_rules_loose() and "strict" way by collect_rules_strict(), as
199 * defined in the OpenFlow spec. */
202 /* Matching criteria for the OpenFlow cookie. Consider a bit B in a rule's
203 * cookie and the corresponding bits C in 'cookie' and M in 'cookie_mask'.
204 * The rule will not be selected if M is 1 and B != C. */
206 ovs_be64 cookie_mask;
208 /* Selection based on actions within a rule:
210 * If out_port != OFPP_ANY, selects only rules that output to out_port.
211 * If out_group != OFPG_ALL, select only rules that output to out_group. */
216 static void rule_criteria_init(struct rule_criteria *, uint8_t table_id,
217 const struct match *match,
218 unsigned int priority,
219 ovs_be64 cookie, ovs_be64 cookie_mask,
220 ofp_port_t out_port, uint32_t out_group);
221 static void rule_criteria_destroy(struct rule_criteria *);
223 /* A packet that needs to be passed to rule_execute(). */
224 struct rule_execute {
225 struct list list_node; /* In struct ofproto's "rule_executes" list. */
226 struct rule *rule; /* Owns a reference to the rule. */
228 struct ofpbuf *packet; /* Owns the packet. */
231 static void run_rule_executes(struct ofproto *);
232 static void destroy_rule_executes(struct ofproto *);
235 static void ofport_destroy__(struct ofport *);
236 static void ofport_destroy(struct ofport *);
238 static void update_port(struct ofproto *, const char *devname);
239 static int init_ports(struct ofproto *);
240 static void reinit_ports(struct ofproto *);
243 static void ofproto_rule_destroy__(struct rule *);
244 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
245 static bool rule_is_modifiable(const struct rule *);
248 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
249 struct ofputil_flow_mod *,
250 const struct ofp_header *);
251 static enum ofperr modify_flows__(struct ofproto *, struct ofconn *,
252 struct ofputil_flow_mod *,
253 const struct ofp_header *,
254 const struct rule_collection *);
255 static void delete_flow__(struct rule *rule, struct ofopgroup *,
256 enum ofp_flow_removed_reason)
257 OVS_RELEASES(rule->rwlock);
258 static enum ofperr add_group(struct ofproto *, struct ofputil_group_mod *);
259 static bool handle_openflow(struct ofconn *, const struct ofpbuf *);
260 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
261 struct ofputil_flow_mod *,
262 const struct ofp_header *);
263 static void calc_duration(long long int start, long long int now,
264 uint32_t *sec, uint32_t *nsec);
267 static uint64_t pick_datapath_id(const struct ofproto *);
268 static uint64_t pick_fallback_dpid(void);
269 static void ofproto_destroy__(struct ofproto *);
270 static void update_mtu(struct ofproto *, struct ofport *);
271 static void meter_delete(struct ofproto *, uint32_t first, uint32_t last);
274 static void ofproto_unixctl_init(void);
276 /* All registered ofproto classes, in probe order. */
277 static const struct ofproto_class **ofproto_classes;
278 static size_t n_ofproto_classes;
279 static size_t allocated_ofproto_classes;
281 struct ovs_mutex ofproto_mutex;
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 ovs_mutex_init_recursive(&ofproto_mutex);
315 ofproto_class_register(&ofproto_dpif_class);
317 /* Make a local copy, since we don't own 'iface_hints' elements. */
318 SHASH_FOR_EACH(node, iface_hints) {
319 const struct iface_hint *orig_hint = node->data;
320 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
321 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
323 new_hint->br_name = xstrdup(orig_hint->br_name);
324 new_hint->br_type = xstrdup(br_type);
325 new_hint->ofp_port = orig_hint->ofp_port;
327 shash_add(&init_ofp_ports, node->name, new_hint);
330 for (i = 0; i < n_ofproto_classes; i++) {
331 ofproto_classes[i]->init(&init_ofp_ports);
335 /* 'type' should be a normalized datapath type, as returned by
336 * ofproto_normalize_type(). Returns the corresponding ofproto_class
337 * structure, or a null pointer if there is none registered for 'type'. */
338 static const struct ofproto_class *
339 ofproto_class_find__(const char *type)
343 for (i = 0; i < n_ofproto_classes; i++) {
344 const struct ofproto_class *class = ofproto_classes[i];
349 class->enumerate_types(&types);
350 found = sset_contains(&types, type);
351 sset_destroy(&types);
357 VLOG_WARN("unknown datapath type %s", type);
361 /* Registers a new ofproto class. After successful registration, new ofprotos
362 * of that type can be created using ofproto_create(). */
364 ofproto_class_register(const struct ofproto_class *new_class)
368 for (i = 0; i < n_ofproto_classes; i++) {
369 if (ofproto_classes[i] == new_class) {
374 if (n_ofproto_classes >= allocated_ofproto_classes) {
375 ofproto_classes = x2nrealloc(ofproto_classes,
376 &allocated_ofproto_classes,
377 sizeof *ofproto_classes);
379 ofproto_classes[n_ofproto_classes++] = new_class;
383 /* Unregisters a datapath provider. 'type' must have been previously
384 * registered and not currently be in use by any ofprotos. After
385 * unregistration new datapaths of that type cannot be opened using
386 * ofproto_create(). */
388 ofproto_class_unregister(const struct ofproto_class *class)
392 for (i = 0; i < n_ofproto_classes; i++) {
393 if (ofproto_classes[i] == class) {
394 for (i++; i < n_ofproto_classes; i++) {
395 ofproto_classes[i - 1] = ofproto_classes[i];
401 VLOG_WARN("attempted to unregister an ofproto class that is not "
406 /* Clears 'types' and enumerates all registered ofproto types into it. The
407 * caller must first initialize the sset. */
409 ofproto_enumerate_types(struct sset *types)
413 for (i = 0; i < n_ofproto_classes; i++) {
414 ofproto_classes[i]->enumerate_types(types);
418 /* Returns the fully spelled out name for the given ofproto 'type'.
420 * Normalized type string can be compared with strcmp(). Unnormalized type
421 * string might be the same even if they have different spellings. */
423 ofproto_normalize_type(const char *type)
425 return type && type[0] ? type : "system";
428 /* Clears 'names' and enumerates the names of all known created ofprotos with
429 * the given 'type'. The caller must first initialize the sset. Returns 0 if
430 * successful, otherwise a positive errno value.
432 * Some kinds of datapaths might not be practically enumerable. This is not
433 * considered an error. */
435 ofproto_enumerate_names(const char *type, struct sset *names)
437 const struct ofproto_class *class = ofproto_class_find__(type);
438 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
442 ofproto_create(const char *datapath_name, const char *datapath_type,
443 struct ofproto **ofprotop)
445 const struct ofproto_class *class;
446 struct ofproto *ofproto;
452 ofproto_unixctl_init();
454 datapath_type = ofproto_normalize_type(datapath_type);
455 class = ofproto_class_find__(datapath_type);
457 VLOG_WARN("could not create datapath %s of unknown type %s",
458 datapath_name, datapath_type);
462 ofproto = class->alloc();
464 VLOG_ERR("failed to allocate datapath %s of type %s",
465 datapath_name, datapath_type);
470 ovs_mutex_lock(&ofproto_mutex);
471 memset(ofproto, 0, sizeof *ofproto);
472 ofproto->ofproto_class = class;
473 ofproto->name = xstrdup(datapath_name);
474 ofproto->type = xstrdup(datapath_type);
475 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
476 hash_string(ofproto->name, 0));
477 ofproto->datapath_id = 0;
478 ofproto->forward_bpdu = false;
479 ofproto->fallback_dpid = pick_fallback_dpid();
480 ofproto->mfr_desc = NULL;
481 ofproto->hw_desc = NULL;
482 ofproto->sw_desc = NULL;
483 ofproto->serial_desc = NULL;
484 ofproto->dp_desc = NULL;
485 ofproto->frag_handling = OFPC_FRAG_NORMAL;
486 hmap_init(&ofproto->ports);
487 shash_init(&ofproto->port_by_name);
488 simap_init(&ofproto->ofp_requests);
489 ofproto->max_ports = ofp_to_u16(OFPP_MAX);
490 ofproto->eviction_group_timer = LLONG_MIN;
491 ofproto->tables = NULL;
492 ofproto->n_tables = 0;
493 hindex_init(&ofproto->cookies);
494 list_init(&ofproto->expirable);
495 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
496 ofproto->state = S_OPENFLOW;
497 list_init(&ofproto->pending);
498 ofproto->n_pending = 0;
499 hmap_init(&ofproto->deletions);
500 guarded_list_init(&ofproto->rule_executes);
501 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
502 ofproto->first_op = ofproto->last_op = LLONG_MIN;
503 ofproto->next_op_report = LLONG_MAX;
504 ofproto->op_backoff = LLONG_MIN;
505 ofproto->vlan_bitmap = NULL;
506 ofproto->vlans_changed = false;
507 ofproto->min_mtu = INT_MAX;
508 ovs_rwlock_init(&ofproto->groups_rwlock);
509 hmap_init(&ofproto->groups);
510 ovs_mutex_unlock(&ofproto_mutex);
512 error = ofproto->ofproto_class->construct(ofproto);
514 VLOG_ERR("failed to open datapath %s: %s",
515 datapath_name, ovs_strerror(error));
516 ofproto_destroy__(ofproto);
520 /* The "max_ports" member should have been set by ->construct(ofproto).
521 * Port 0 is not a valid OpenFlow port, so mark that as unavailable. */
522 ofproto->ofp_port_ids = bitmap_allocate(ofproto->max_ports);
523 bitmap_set1(ofproto->ofp_port_ids, 0);
525 /* Check that hidden tables, if any, are at the end. */
526 ovs_assert(ofproto->n_tables);
527 for (i = 0; i + 1 < ofproto->n_tables; i++) {
528 enum oftable_flags flags = ofproto->tables[i].flags;
529 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
531 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
534 ofproto->datapath_id = pick_datapath_id(ofproto);
537 /* Initialize meters table. */
538 if (ofproto->ofproto_class->meter_get_features) {
539 ofproto->ofproto_class->meter_get_features(ofproto,
540 &ofproto->meter_features);
542 memset(&ofproto->meter_features, 0, sizeof ofproto->meter_features);
544 ofproto->meters = xzalloc((ofproto->meter_features.max_meters + 1)
545 * sizeof(struct meter *));
551 /* Must be called (only) by an ofproto implementation in its constructor
552 * function. See the large comment on 'construct' in struct ofproto_class for
555 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
557 struct oftable *table;
559 ovs_assert(!ofproto->n_tables);
560 ovs_assert(n_tables >= 1 && n_tables <= 255);
562 ofproto->n_tables = n_tables;
563 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
564 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
569 /* To be optionally called (only) by an ofproto implementation in its
570 * constructor function. See the large comment on 'construct' in struct
571 * ofproto_class for details.
573 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
574 * will then ensure that actions passed into the ofproto implementation will
575 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
576 * function is not called, there will be no such restriction.
578 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
579 * the 'max_ports' restriction. */
581 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
583 ovs_assert(max_ports <= ofp_to_u16(OFPP_MAX));
584 ofproto->max_ports = max_ports;
588 ofproto_get_datapath_id(const struct ofproto *ofproto)
590 return ofproto->datapath_id;
594 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
596 uint64_t old_dpid = p->datapath_id;
597 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
598 if (p->datapath_id != old_dpid) {
599 /* Force all active connections to reconnect, since there is no way to
600 * notify a controller that the datapath ID has changed. */
601 ofproto_reconnect_controllers(p);
606 ofproto_set_controllers(struct ofproto *p,
607 const struct ofproto_controller *controllers,
608 size_t n_controllers, uint32_t allowed_versions)
610 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
615 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
617 connmgr_set_fail_mode(p->connmgr, fail_mode);
620 /* Drops the connections between 'ofproto' and all of its controllers, forcing
621 * them to reconnect. */
623 ofproto_reconnect_controllers(struct ofproto *ofproto)
625 connmgr_reconnect(ofproto->connmgr);
628 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
629 * in-band control should guarantee access, in the same way that in-band
630 * control guarantees access to OpenFlow controllers. */
632 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
633 const struct sockaddr_in *extras, size_t n)
635 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
638 /* Sets the OpenFlow queue used by flows set up by in-band control on
639 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
640 * flows will use the default queue. */
642 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
644 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
647 /* Sets the number of flows at which eviction from the kernel flow table
650 ofproto_set_flow_eviction_threshold(unsigned threshold)
652 flow_eviction_threshold = MAX(OFPROTO_FLOW_EVICTION_THRESHOLD_MIN,
656 /* Sets the path for handling flow misses. */
658 ofproto_set_flow_miss_model(unsigned model)
660 flow_miss_model = model;
663 /* If forward_bpdu is true, the NORMAL action will forward frames with
664 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
665 * the NORMAL action will drop these frames. */
667 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
669 bool old_val = ofproto->forward_bpdu;
670 ofproto->forward_bpdu = forward_bpdu;
671 if (old_val != ofproto->forward_bpdu) {
672 if (ofproto->ofproto_class->forward_bpdu_changed) {
673 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
678 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
679 * 'idle_time', in seconds, and the maximum number of MAC table entries to
682 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
685 if (ofproto->ofproto_class->set_mac_table_config) {
686 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
691 /* Sets number of upcall handler threads. The default is
692 * (number of online cores - 2). */
694 ofproto_set_n_handler_threads(unsigned limit)
697 n_handler_threads = limit;
699 int n_proc = sysconf(_SC_NPROCESSORS_ONLN);
700 n_handler_threads = n_proc > 2 ? n_proc - 2 : 1;
705 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
708 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
712 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
714 return connmgr_set_snoops(ofproto->connmgr, snoops);
718 ofproto_set_netflow(struct ofproto *ofproto,
719 const struct netflow_options *nf_options)
721 if (nf_options && sset_is_empty(&nf_options->collectors)) {
725 if (ofproto->ofproto_class->set_netflow) {
726 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
728 return nf_options ? EOPNOTSUPP : 0;
733 ofproto_set_sflow(struct ofproto *ofproto,
734 const struct ofproto_sflow_options *oso)
736 if (oso && sset_is_empty(&oso->targets)) {
740 if (ofproto->ofproto_class->set_sflow) {
741 return ofproto->ofproto_class->set_sflow(ofproto, oso);
743 return oso ? EOPNOTSUPP : 0;
748 ofproto_set_ipfix(struct ofproto *ofproto,
749 const struct ofproto_ipfix_bridge_exporter_options *bo,
750 const struct ofproto_ipfix_flow_exporter_options *fo,
753 if (ofproto->ofproto_class->set_ipfix) {
754 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
756 return (bo || fo) ? EOPNOTSUPP : 0;
761 ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
763 flow_restore_wait = flow_restore_wait_db;
767 ofproto_get_flow_restore_wait(void)
769 return flow_restore_wait;
773 /* Spanning Tree Protocol (STP) configuration. */
775 /* Configures STP on 'ofproto' using the settings defined in 's'. If
776 * 's' is NULL, disables STP.
778 * Returns 0 if successful, otherwise a positive errno value. */
780 ofproto_set_stp(struct ofproto *ofproto,
781 const struct ofproto_stp_settings *s)
783 return (ofproto->ofproto_class->set_stp
784 ? ofproto->ofproto_class->set_stp(ofproto, s)
788 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
789 * 'enabled' member of 's' is false, then the other members are not
792 * Returns 0 if successful, otherwise a positive errno value. */
794 ofproto_get_stp_status(struct ofproto *ofproto,
795 struct ofproto_stp_status *s)
797 return (ofproto->ofproto_class->get_stp_status
798 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
802 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
803 * in 's'. The caller is responsible for assigning STP port numbers
804 * (using the 'port_num' member in the range of 1 through 255, inclusive)
805 * and ensuring there are no duplicates. If the 's' is NULL, then STP
806 * is disabled on the port.
808 * Returns 0 if successful, otherwise a positive errno value.*/
810 ofproto_port_set_stp(struct ofproto *ofproto, ofp_port_t ofp_port,
811 const struct ofproto_port_stp_settings *s)
813 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
815 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
816 ofproto->name, ofp_port);
820 return (ofproto->ofproto_class->set_stp_port
821 ? ofproto->ofproto_class->set_stp_port(ofport, s)
825 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
826 * 's'. If the 'enabled' member in 's' is false, then the other members
827 * are not meaningful.
829 * Returns 0 if successful, otherwise a positive errno value.*/
831 ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
832 struct ofproto_port_stp_status *s)
834 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
836 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
837 "port %"PRIu16, ofproto->name, ofp_port);
841 return (ofproto->ofproto_class->get_stp_port_status
842 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
846 /* Queue DSCP configuration. */
848 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
849 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
850 * to implement QoS. Instead, it is used to implement features which require
851 * knowledge of what queues exist on a port, and some basic information about
854 * Returns 0 if successful, otherwise a positive errno value. */
856 ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port,
857 const struct ofproto_port_queue *queues,
860 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
863 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
864 ofproto->name, ofp_port);
868 return (ofproto->ofproto_class->set_queues
869 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
873 /* Connectivity Fault Management configuration. */
875 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
877 ofproto_port_clear_cfm(struct ofproto *ofproto, ofp_port_t ofp_port)
879 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
880 if (ofport && ofproto->ofproto_class->set_cfm) {
881 ofproto->ofproto_class->set_cfm(ofport, NULL);
885 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
886 * basic configuration from the configuration members in 'cfm', and the remote
887 * maintenance point ID from remote_mpid. Ignores the statistics members of
890 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
892 ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port,
893 const struct cfm_settings *s)
895 struct ofport *ofport;
898 ofport = ofproto_get_port(ofproto, ofp_port);
900 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
901 ofproto->name, ofp_port);
905 /* XXX: For configuration simplicity, we only support one remote_mpid
906 * outside of the CFM module. It's not clear if this is the correct long
907 * term solution or not. */
908 error = (ofproto->ofproto_class->set_cfm
909 ? ofproto->ofproto_class->set_cfm(ofport, s)
912 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
913 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
914 ovs_strerror(error));
918 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
919 * 'ofproto' does not have a port 'ofp_port'. */
921 ofproto_port_set_bfd(struct ofproto *ofproto, ofp_port_t ofp_port,
922 const struct smap *cfg)
924 struct ofport *ofport;
927 ofport = ofproto_get_port(ofproto, ofp_port);
929 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
930 ofproto->name, ofp_port);
934 error = (ofproto->ofproto_class->set_bfd
935 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
938 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
939 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
940 ovs_strerror(error));
944 /* Populates 'status' with key value pairs indicating the status of the BFD
945 * session on 'ofp_port'. This information is intended to be populated in the
946 * OVS database. Has no effect if 'ofp_port' is not na OpenFlow port in
949 ofproto_port_get_bfd_status(struct ofproto *ofproto, ofp_port_t ofp_port,
952 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
953 return (ofport && ofproto->ofproto_class->get_bfd_status
954 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
958 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
959 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
960 * 0 if LACP partner information is not current (generally indicating a
961 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
963 ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port)
965 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
966 return (ofport && ofproto->ofproto_class->port_is_lacp_current
967 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
973 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
974 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
975 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
976 * configuration plus, if there is more than one slave, a bonding
979 * If 'aux' is already registered then this function updates its configuration
980 * to 's'. Otherwise, this function registers a new bundle.
982 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
985 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
986 const struct ofproto_bundle_settings *s)
988 return (ofproto->ofproto_class->bundle_set
989 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
993 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
994 * If no such bundle has been registered, this has no effect. */
996 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
998 return ofproto_bundle_register(ofproto, aux, NULL);
1002 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
1003 * If 'aux' is already registered then this function updates its configuration
1004 * to 's'. Otherwise, this function registers a new mirror. */
1006 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
1007 const struct ofproto_mirror_settings *s)
1009 return (ofproto->ofproto_class->mirror_set
1010 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
1014 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
1015 * If no mirror has been registered, this has no effect. */
1017 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
1019 return ofproto_mirror_register(ofproto, aux, NULL);
1022 /* Retrieves statistics from mirror associated with client data pointer
1023 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
1024 * 'bytes', respectively. If a particular counters is not supported,
1025 * the appropriate argument is set to UINT64_MAX. */
1027 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
1028 uint64_t *packets, uint64_t *bytes)
1030 if (!ofproto->ofproto_class->mirror_get_stats) {
1031 *packets = *bytes = UINT64_MAX;
1035 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
1039 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
1040 * which all packets are flooded, instead of using MAC learning. If
1041 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1043 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
1046 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
1048 return (ofproto->ofproto_class->set_flood_vlans
1049 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
1053 /* Returns true if 'aux' is a registered bundle that is currently in use as the
1054 * output for a mirror. */
1056 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
1058 return (ofproto->ofproto_class->is_mirror_output_bundle
1059 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
1063 /* Configuration of OpenFlow tables. */
1065 /* Returns the number of OpenFlow tables in 'ofproto'. */
1067 ofproto_get_n_tables(const struct ofproto *ofproto)
1069 return ofproto->n_tables;
1072 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
1073 * settings from 's'. 'table_id' must be in the range 0 through the number of
1074 * OpenFlow tables in 'ofproto' minus 1, inclusive.
1076 * For read-only tables, only the name may be configured. */
1078 ofproto_configure_table(struct ofproto *ofproto, int table_id,
1079 const struct ofproto_table_settings *s)
1081 struct oftable *table;
1083 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
1084 table = &ofproto->tables[table_id];
1086 oftable_set_name(table, s->name);
1088 if (table->flags & OFTABLE_READONLY) {
1093 oftable_enable_eviction(table, s->groups, s->n_groups);
1095 oftable_disable_eviction(table);
1098 table->max_flows = s->max_flows;
1099 ovs_rwlock_rdlock(&table->cls.rwlock);
1100 if (classifier_count(&table->cls) > table->max_flows
1101 && table->eviction_fields) {
1102 /* 'table' contains more flows than allowed. We might not be able to
1103 * evict them right away because of the asynchronous nature of flow
1104 * table changes. Schedule eviction for later. */
1105 switch (ofproto->state) {
1107 ofproto->state = S_EVICT;
1111 /* We're already deleting flows, nothing more to do. */
1115 ovs_rwlock_unlock(&table->cls.rwlock);
1119 ofproto_has_snoops(const struct ofproto *ofproto)
1121 return connmgr_has_snoops(ofproto->connmgr);
1125 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1127 connmgr_get_snoops(ofproto->connmgr, snoops);
1130 /* Deletes 'rule' from 'cls' within 'ofproto'.
1132 * Within an ofproto implementation, this function allows an ofproto
1133 * implementation to destroy any rules that remain when its ->destruct()
1134 * function is called. This function is not suitable for use elsewhere in an
1135 * ofproto implementation.
1137 * This function is also used internally in ofproto.c.
1139 * This function implements steps 4.4 and 4.5 in the section titled "Rule Life
1140 * Cycle" in ofproto-provider.h.
1142 * The 'cls' argument is redundant (it is &ofproto->tables[rule->table_id].cls)
1143 * but it allows Clang to do better checking. */
1145 ofproto_rule_delete(struct ofproto *ofproto, struct classifier *cls,
1147 OVS_REQ_WRLOCK(cls->rwlock)
1149 struct ofopgroup *group;
1151 ovs_assert(!rule->pending);
1152 ovs_assert(cls == &ofproto->tables[rule->table_id].cls);
1154 group = ofopgroup_create_unattached(ofproto);
1155 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1156 ovs_rwlock_wrlock(&rule->rwlock);
1157 oftable_remove_rule__(ofproto, cls, rule);
1158 ofproto->ofproto_class->rule_delete(rule);
1159 ofopgroup_submit(group);
1163 ofproto_flush__(struct ofproto *ofproto)
1165 struct oftable *table;
1167 if (ofproto->ofproto_class->flush) {
1168 ofproto->ofproto_class->flush(ofproto);
1171 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1172 struct rule *rule, *next_rule;
1173 struct cls_cursor cursor;
1175 if (table->flags & OFTABLE_HIDDEN) {
1179 ovs_rwlock_wrlock(&table->cls.rwlock);
1180 cls_cursor_init(&cursor, &table->cls, NULL);
1181 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1182 if (!rule->pending) {
1183 ofproto_rule_delete(ofproto, &table->cls, rule);
1186 ovs_rwlock_unlock(&table->cls.rwlock);
1190 static void delete_group(struct ofproto *ofproto, uint32_t group_id);
1193 ofproto_destroy__(struct ofproto *ofproto)
1195 struct oftable *table;
1197 ovs_assert(list_is_empty(&ofproto->pending));
1198 ovs_assert(!ofproto->n_pending);
1200 destroy_rule_executes(ofproto);
1201 guarded_list_destroy(&ofproto->rule_executes);
1203 delete_group(ofproto, OFPG_ALL);
1204 ovs_rwlock_destroy(&ofproto->groups_rwlock);
1205 hmap_destroy(&ofproto->groups);
1207 connmgr_destroy(ofproto->connmgr);
1209 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1210 free(ofproto->name);
1211 free(ofproto->type);
1212 free(ofproto->mfr_desc);
1213 free(ofproto->hw_desc);
1214 free(ofproto->sw_desc);
1215 free(ofproto->serial_desc);
1216 free(ofproto->dp_desc);
1217 hmap_destroy(&ofproto->ports);
1218 shash_destroy(&ofproto->port_by_name);
1219 bitmap_free(ofproto->ofp_port_ids);
1220 simap_destroy(&ofproto->ofp_requests);
1222 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1223 oftable_destroy(table);
1225 free(ofproto->tables);
1227 hmap_destroy(&ofproto->deletions);
1229 free(ofproto->vlan_bitmap);
1231 ofproto->ofproto_class->dealloc(ofproto);
1235 ofproto_destroy(struct ofproto *p)
1237 struct ofport *ofport, *next_ofport;
1244 meter_delete(p, 1, p->meter_features.max_meters);
1245 p->meter_features.max_meters = 0;
1251 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1252 ofport_destroy(ofport);
1255 p->ofproto_class->destruct(p);
1256 ofproto_destroy__(p);
1259 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1260 * kernel datapath, for example, this destroys the datapath in the kernel, and
1261 * with the netdev-based datapath, it tears down the data structures that
1262 * represent the datapath.
1264 * The datapath should not be currently open as an ofproto. */
1266 ofproto_delete(const char *name, const char *type)
1268 const struct ofproto_class *class = ofproto_class_find__(type);
1269 return (!class ? EAFNOSUPPORT
1270 : !class->del ? EACCES
1271 : class->del(type, name));
1275 process_port_change(struct ofproto *ofproto, int error, char *devname)
1277 if (error == ENOBUFS) {
1278 reinit_ports(ofproto);
1279 } else if (!error) {
1280 update_port(ofproto, devname);
1286 ofproto_type_run(const char *datapath_type)
1288 const struct ofproto_class *class;
1291 datapath_type = ofproto_normalize_type(datapath_type);
1292 class = ofproto_class_find__(datapath_type);
1294 error = class->type_run ? class->type_run(datapath_type) : 0;
1295 if (error && error != EAGAIN) {
1296 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1297 datapath_type, ovs_strerror(error));
1303 ofproto_type_run_fast(const char *datapath_type)
1305 const struct ofproto_class *class;
1308 datapath_type = ofproto_normalize_type(datapath_type);
1309 class = ofproto_class_find__(datapath_type);
1311 error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
1312 if (error && error != EAGAIN) {
1313 VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
1314 datapath_type, ovs_strerror(error));
1320 ofproto_type_wait(const char *datapath_type)
1322 const struct ofproto_class *class;
1324 datapath_type = ofproto_normalize_type(datapath_type);
1325 class = ofproto_class_find__(datapath_type);
1327 if (class->type_wait) {
1328 class->type_wait(datapath_type);
1333 ofproto_run(struct ofproto *p)
1335 struct sset changed_netdevs;
1336 const char *changed_netdev;
1337 struct ofport *ofport;
1340 error = p->ofproto_class->run(p);
1341 if (error && error != EAGAIN) {
1342 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, ovs_strerror(error));
1345 run_rule_executes(p);
1347 /* Restore the eviction group heap invariant occasionally. */
1348 if (p->eviction_group_timer < time_msec()) {
1351 p->eviction_group_timer = time_msec() + 1000;
1353 for (i = 0; i < p->n_tables; i++) {
1354 struct oftable *table = &p->tables[i];
1355 struct eviction_group *evg;
1356 struct cls_cursor cursor;
1359 if (!table->eviction_fields) {
1363 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
1364 heap_rebuild(&evg->rules);
1367 ovs_rwlock_rdlock(&table->cls.rwlock);
1368 cls_cursor_init(&cursor, &table->cls, NULL);
1369 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1370 if (!rule->eviction_group
1371 && (rule->idle_timeout || rule->hard_timeout)) {
1372 eviction_group_add_rule(rule);
1375 ovs_rwlock_unlock(&table->cls.rwlock);
1379 if (p->ofproto_class->port_poll) {
1382 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1383 process_port_change(p, error, devname);
1387 /* Update OpenFlow port status for any port whose netdev has changed.
1389 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1390 * destroyed, so it's not safe to update ports directly from the
1391 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1392 * need this two-phase approach. */
1393 sset_init(&changed_netdevs);
1394 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1395 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1396 if (ofport->change_seq != change_seq) {
1397 ofport->change_seq = change_seq;
1398 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1401 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1402 update_port(p, changed_netdev);
1404 sset_destroy(&changed_netdevs);
1408 connmgr_run(p->connmgr, handle_openflow);
1412 connmgr_run(p->connmgr, NULL);
1414 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1415 p->state = S_OPENFLOW;
1420 connmgr_run(p->connmgr, NULL);
1422 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1423 connmgr_flushed(p->connmgr);
1424 p->state = S_OPENFLOW;
1432 if (time_msec() >= p->next_op_report) {
1433 long long int ago = (time_msec() - p->first_op) / 1000;
1434 long long int interval = (p->last_op - p->first_op) / 1000;
1438 ds_put_format(&s, "%d flow_mods ",
1439 p->n_add + p->n_delete + p->n_modify);
1440 if (interval == ago) {
1441 ds_put_format(&s, "in the last %lld s", ago);
1442 } else if (interval) {
1443 ds_put_format(&s, "in the %lld s starting %lld s ago",
1446 ds_put_format(&s, "%lld s ago", ago);
1449 ds_put_cstr(&s, " (");
1451 ds_put_format(&s, "%d adds, ", p->n_add);
1454 ds_put_format(&s, "%d deletes, ", p->n_delete);
1457 ds_put_format(&s, "%d modifications, ", p->n_modify);
1460 ds_put_char(&s, ')');
1462 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1465 p->n_add = p->n_delete = p->n_modify = 0;
1466 p->next_op_report = LLONG_MAX;
1472 /* Performs periodic activity required by 'ofproto' that needs to be done
1473 * with the least possible latency.
1475 * It makes sense to call this function a couple of times per poll loop, to
1476 * provide a significant performance boost on some benchmarks with the
1477 * ofproto-dpif implementation. */
1479 ofproto_run_fast(struct ofproto *p)
1483 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1484 if (error && error != EAGAIN) {
1485 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1486 p->name, ovs_strerror(error));
1492 ofproto_wait(struct ofproto *p)
1494 struct ofport *ofport;
1496 p->ofproto_class->wait(p);
1497 if (p->ofproto_class->port_poll_wait) {
1498 p->ofproto_class->port_poll_wait(p);
1501 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1502 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1503 poll_immediate_wake();
1509 connmgr_wait(p->connmgr, true);
1514 connmgr_wait(p->connmgr, false);
1515 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1516 poll_immediate_wake();
1523 ofproto_is_alive(const struct ofproto *p)
1525 return connmgr_has_controllers(p->connmgr);
1528 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1529 * memory_report(). */
1531 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1533 const struct oftable *table;
1534 unsigned int n_rules;
1536 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1537 simap_increase(usage, "ops",
1538 ofproto->n_pending + hmap_count(&ofproto->deletions));
1541 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1542 ovs_rwlock_rdlock(&table->cls.rwlock);
1543 n_rules += classifier_count(&table->cls);
1544 ovs_rwlock_unlock(&table->cls.rwlock);
1546 simap_increase(usage, "rules", n_rules);
1548 if (ofproto->ofproto_class->get_memory_usage) {
1549 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1552 connmgr_get_memory_usage(ofproto->connmgr, usage);
1556 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1559 connmgr_get_controller_info(ofproto->connmgr, info);
1563 ofproto_free_ofproto_controller_info(struct shash *info)
1565 connmgr_free_controller_info(info);
1568 /* Makes a deep copy of 'old' into 'port'. */
1570 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1572 port->name = xstrdup(old->name);
1573 port->type = xstrdup(old->type);
1574 port->ofp_port = old->ofp_port;
1577 /* Frees memory allocated to members of 'ofproto_port'.
1579 * Do not call this function on an ofproto_port obtained from
1580 * ofproto_port_dump_next(): that function retains ownership of the data in the
1583 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1585 free(ofproto_port->name);
1586 free(ofproto_port->type);
1589 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1591 * This function provides no status indication. An error status for the entire
1592 * dump operation is provided when it is completed by calling
1593 * ofproto_port_dump_done().
1596 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1597 const struct ofproto *ofproto)
1599 dump->ofproto = ofproto;
1600 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1604 /* Attempts to retrieve another port from 'dump', which must have been created
1605 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1606 * 'port' and returns true. On failure, returns false.
1608 * Failure might indicate an actual error or merely that the last port has been
1609 * dumped. An error status for the entire dump operation is provided when it
1610 * is completed by calling ofproto_port_dump_done().
1612 * The ofproto owns the data stored in 'port'. It will remain valid until at
1613 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1614 * ofproto_port_dump_done(). */
1616 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1617 struct ofproto_port *port)
1619 const struct ofproto *ofproto = dump->ofproto;
1625 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1628 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1634 /* Completes port table dump operation 'dump', which must have been created
1635 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1636 * error-free, otherwise a positive errno value describing the problem. */
1638 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1640 const struct ofproto *ofproto = dump->ofproto;
1642 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1645 return dump->error == EOF ? 0 : dump->error;
1648 /* Returns the type to pass to netdev_open() when a datapath of type
1649 * 'datapath_type' has a port of type 'port_type', for a few special
1650 * cases when a netdev type differs from a port type. For example, when
1651 * using the userspace datapath, a port of type "internal" needs to be
1654 * Returns either 'type' itself or a string literal, which must not be
1657 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1659 const struct ofproto_class *class;
1661 datapath_type = ofproto_normalize_type(datapath_type);
1662 class = ofproto_class_find__(datapath_type);
1667 return (class->port_open_type
1668 ? class->port_open_type(datapath_type, port_type)
1672 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1673 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1674 * the port's OpenFlow port number.
1676 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1677 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1678 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1679 * 'ofp_portp' is non-null). */
1681 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1682 ofp_port_t *ofp_portp)
1684 ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1687 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1689 const char *netdev_name = netdev_get_name(netdev);
1691 simap_put(&ofproto->ofp_requests, netdev_name,
1692 ofp_to_u16(ofp_port));
1693 update_port(ofproto, netdev_name);
1696 struct ofproto_port ofproto_port;
1698 ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
1700 *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
1701 ofproto_port_destroy(&ofproto_port);
1706 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1707 * initializes '*port' appropriately; on failure, returns a positive errno
1710 * The caller owns the data in 'ofproto_port' and must free it with
1711 * ofproto_port_destroy() when it is no longer needed. */
1713 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1714 struct ofproto_port *port)
1718 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1720 memset(port, 0, sizeof *port);
1725 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1726 * Returns 0 if successful, otherwise a positive errno. */
1728 ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port)
1730 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1731 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1732 struct simap_node *ofp_request_node;
1735 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1736 if (ofp_request_node) {
1737 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1740 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1741 if (!error && ofport) {
1742 /* 'name' is the netdev's name and update_port() is going to close the
1743 * netdev. Just in case update_port() refers to 'name' after it
1744 * destroys 'ofport', make a copy of it around the update_port()
1746 char *devname = xstrdup(name);
1747 update_port(ofproto, devname);
1754 simple_flow_mod(struct ofproto *ofproto,
1755 const struct match *match, unsigned int priority,
1756 const struct ofpact *ofpacts, size_t ofpacts_len,
1757 enum ofp_flow_mod_command command)
1759 struct ofputil_flow_mod fm;
1761 memset(&fm, 0, sizeof fm);
1763 fm.priority = priority;
1766 fm.modify_cookie = false;
1768 fm.command = command;
1769 fm.idle_timeout = 0;
1770 fm.hard_timeout = 0;
1771 fm.buffer_id = UINT32_MAX;
1772 fm.out_port = OFPP_ANY;
1773 fm.out_group = OFPG_ANY;
1775 fm.ofpacts = CONST_CAST(struct ofpact *, ofpacts);
1776 fm.ofpacts_len = ofpacts_len;
1777 return handle_flow_mod__(ofproto, NULL, &fm, NULL);
1780 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1781 * performs the 'n_actions' actions in 'actions'. The new flow will not
1784 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1785 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1786 * controllers; otherwise, it will be hidden.
1788 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1790 * This is a helper function for in-band control and fail-open. */
1792 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1793 unsigned int priority,
1794 const struct ofpact *ofpacts, size_t ofpacts_len)
1796 const struct rule *rule;
1799 /* First do a cheap check whether the rule we're looking for already exists
1800 * with the actions that we want. If it does, then we're done. */
1801 ovs_rwlock_rdlock(&ofproto->tables[0].cls.rwlock);
1802 rule = rule_from_cls_rule(classifier_find_match_exactly(
1803 &ofproto->tables[0].cls, match, priority));
1805 ovs_rwlock_rdlock(&rule->rwlock);
1806 must_add = !ofpacts_equal(rule->actions->ofpacts,
1807 rule->actions->ofpacts_len,
1808 ofpacts, ofpacts_len);
1809 ovs_rwlock_unlock(&rule->rwlock);
1813 ovs_rwlock_unlock(&ofproto->tables[0].cls.rwlock);
1815 /* If there's no such rule or the rule doesn't have the actions we want,
1816 * fall back to a executing a full flow mod. We can't optimize this at
1817 * all because we didn't take enough locks above to ensure that the flow
1818 * table didn't already change beneath us. */
1820 simple_flow_mod(ofproto, match, priority, ofpacts, ofpacts_len,
1821 OFPFC_MODIFY_STRICT);
1825 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1826 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1827 * operation cannot be initiated now but may be retried later.
1829 * This is a helper function for in-band control and fail-open. */
1831 ofproto_flow_mod(struct ofproto *ofproto, struct ofputil_flow_mod *fm)
1833 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1836 /* Searches for a rule with matching criteria exactly equal to 'target' in
1837 * ofproto's table 0 and, if it finds one, deletes it.
1839 * This is a helper function for in-band control and fail-open. */
1841 ofproto_delete_flow(struct ofproto *ofproto,
1842 const struct match *target, unsigned int priority)
1844 struct classifier *cls = &ofproto->tables[0].cls;
1847 /* First do a cheap check whether the rule we're looking for has already
1848 * been deleted. If so, then we're done. */
1849 ovs_rwlock_rdlock(&cls->rwlock);
1850 rule = rule_from_cls_rule(classifier_find_match_exactly(cls, target,
1852 ovs_rwlock_unlock(&cls->rwlock);
1857 /* Fall back to a executing a full flow mod. We can't optimize this at all
1858 * because we didn't take enough locks above to ensure that the flow table
1859 * didn't already change beneath us. */
1860 return simple_flow_mod(ofproto, target, priority, NULL, 0,
1861 OFPFC_DELETE_STRICT) != OFPROTO_POSTPONE;
1864 /* Starts the process of deleting all of the flows from all of ofproto's flow
1865 * tables and then reintroducing the flows required by in-band control and
1866 * fail-open. The process will complete in a later call to ofproto_run(). */
1868 ofproto_flush_flows(struct ofproto *ofproto)
1870 COVERAGE_INC(ofproto_flush);
1871 ofproto->state = S_FLUSH;
1875 reinit_ports(struct ofproto *p)
1877 struct ofproto_port_dump dump;
1878 struct sset devnames;
1879 struct ofport *ofport;
1880 struct ofproto_port ofproto_port;
1881 const char *devname;
1883 COVERAGE_INC(ofproto_reinit_ports);
1885 sset_init(&devnames);
1886 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1887 sset_add(&devnames, netdev_get_name(ofport->netdev));
1889 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1890 sset_add(&devnames, ofproto_port.name);
1893 SSET_FOR_EACH (devname, &devnames) {
1894 update_port(p, devname);
1896 sset_destroy(&devnames);
1900 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
1904 port_idx = simap_get(&ofproto->ofp_requests, netdev_name);
1905 port_idx = port_idx ? port_idx : UINT16_MAX;
1907 if (port_idx >= ofproto->max_ports
1908 || bitmap_is_set(ofproto->ofp_port_ids, port_idx)) {
1909 uint16_t end_port_no = ofproto->alloc_port_no;
1911 /* Search for a free OpenFlow port number. We try not to
1912 * immediately reuse them to prevent problems due to old
1915 if (++ofproto->alloc_port_no >= ofproto->max_ports) {
1916 ofproto->alloc_port_no = 0;
1918 if (!bitmap_is_set(ofproto->ofp_port_ids,
1919 ofproto->alloc_port_no)) {
1920 port_idx = ofproto->alloc_port_no;
1923 if (ofproto->alloc_port_no == end_port_no) {
1928 bitmap_set1(ofproto->ofp_port_ids, port_idx);
1929 return u16_to_ofp(port_idx);
1933 dealloc_ofp_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
1935 if (ofp_to_u16(ofp_port) < ofproto->max_ports) {
1936 bitmap_set0(ofproto->ofp_port_ids, ofp_to_u16(ofp_port));
1940 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1941 * pointer if the netdev cannot be opened. On success, also fills in
1943 static struct netdev *
1944 ofport_open(struct ofproto *ofproto,
1945 struct ofproto_port *ofproto_port,
1946 struct ofputil_phy_port *pp)
1948 enum netdev_flags flags;
1949 struct netdev *netdev;
1952 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1954 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
1955 "cannot be opened (%s)",
1957 ofproto_port->name, ofproto_port->ofp_port,
1958 ofproto_port->name, ovs_strerror(error));
1962 if (ofproto_port->ofp_port == OFPP_NONE) {
1963 if (!strcmp(ofproto->name, ofproto_port->name)) {
1964 ofproto_port->ofp_port = OFPP_LOCAL;
1966 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
1967 ofproto_port->name);
1970 pp->port_no = ofproto_port->ofp_port;
1971 netdev_get_etheraddr(netdev, pp->hw_addr);
1972 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
1973 netdev_get_flags(netdev, &flags);
1974 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
1975 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
1976 netdev_get_features(netdev, &pp->curr, &pp->advertised,
1977 &pp->supported, &pp->peer);
1978 pp->curr_speed = netdev_features_to_bps(pp->curr, 0) / 1000;
1979 pp->max_speed = netdev_features_to_bps(pp->supported, 0) / 1000;
1984 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1985 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1988 ofport_equal(const struct ofputil_phy_port *a,
1989 const struct ofputil_phy_port *b)
1991 return (eth_addr_equals(a->hw_addr, b->hw_addr)
1992 && a->state == b->state
1993 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
1994 && a->curr == b->curr
1995 && a->advertised == b->advertised
1996 && a->supported == b->supported
1997 && a->peer == b->peer
1998 && a->curr_speed == b->curr_speed
1999 && a->max_speed == b->max_speed);
2002 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
2003 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
2004 * one with the same name or port number). */
2006 ofport_install(struct ofproto *p,
2007 struct netdev *netdev, const struct ofputil_phy_port *pp)
2009 const char *netdev_name = netdev_get_name(netdev);
2010 struct ofport *ofport;
2013 /* Create ofport. */
2014 ofport = p->ofproto_class->port_alloc();
2019 ofport->ofproto = p;
2020 ofport->netdev = netdev;
2021 ofport->change_seq = netdev_change_seq(netdev);
2023 ofport->ofp_port = pp->port_no;
2024 ofport->created = time_msec();
2026 /* Add port to 'p'. */
2027 hmap_insert(&p->ports, &ofport->hmap_node,
2028 hash_ofp_port(ofport->ofp_port));
2029 shash_add(&p->port_by_name, netdev_name, ofport);
2031 update_mtu(p, ofport);
2033 /* Let the ofproto_class initialize its private data. */
2034 error = p->ofproto_class->port_construct(ofport);
2038 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
2042 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
2043 p->name, netdev_name, ovs_strerror(error));
2045 ofport_destroy__(ofport);
2047 netdev_close(netdev);
2051 /* Removes 'ofport' from 'p' and destroys it. */
2053 ofport_remove(struct ofport *ofport)
2055 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
2057 ofport_destroy(ofport);
2060 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
2063 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
2065 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
2067 ofport_remove(port);
2071 /* Updates 'port' with new 'pp' description.
2073 * Does not handle a name or port number change. The caller must implement
2074 * such a change as a delete followed by an add. */
2076 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
2078 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
2079 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
2080 | (pp->config & OFPUTIL_PC_PORT_DOWN));
2081 port->pp.state = pp->state;
2082 port->pp.curr = pp->curr;
2083 port->pp.advertised = pp->advertised;
2084 port->pp.supported = pp->supported;
2085 port->pp.peer = pp->peer;
2086 port->pp.curr_speed = pp->curr_speed;
2087 port->pp.max_speed = pp->max_speed;
2089 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
2092 /* Update OpenFlow 'state' in 'port' and notify controller. */
2094 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
2096 if (port->pp.state != state) {
2097 port->pp.state = state;
2098 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
2104 ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port)
2106 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
2108 if (port->ofproto->ofproto_class->set_realdev) {
2109 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
2111 if (port->ofproto->ofproto_class->set_stp_port) {
2112 port->ofproto->ofproto_class->set_stp_port(port, NULL);
2114 if (port->ofproto->ofproto_class->set_cfm) {
2115 port->ofproto->ofproto_class->set_cfm(port, NULL);
2117 if (port->ofproto->ofproto_class->bundle_remove) {
2118 port->ofproto->ofproto_class->bundle_remove(port);
2124 ofport_destroy__(struct ofport *port)
2126 struct ofproto *ofproto = port->ofproto;
2127 const char *name = netdev_get_name(port->netdev);
2129 hmap_remove(&ofproto->ports, &port->hmap_node);
2130 shash_delete(&ofproto->port_by_name,
2131 shash_find(&ofproto->port_by_name, name));
2133 netdev_close(port->netdev);
2134 ofproto->ofproto_class->port_dealloc(port);
2138 ofport_destroy(struct ofport *port)
2141 dealloc_ofp_port(port->ofproto, port->ofp_port);
2142 port->ofproto->ofproto_class->port_destruct(port);
2143 ofport_destroy__(port);
2148 ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
2150 struct ofport *port;
2152 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_ofp_port(ofp_port),
2154 if (port->ofp_port == ofp_port) {
2162 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
2164 struct ofproto *ofproto = port->ofproto;
2167 if (ofproto->ofproto_class->port_get_stats) {
2168 error = ofproto->ofproto_class->port_get_stats(port, stats);
2177 update_port(struct ofproto *ofproto, const char *name)
2179 struct ofproto_port ofproto_port;
2180 struct ofputil_phy_port pp;
2181 struct netdev *netdev;
2182 struct ofport *port;
2184 COVERAGE_INC(ofproto_update_port);
2186 /* Fetch 'name''s location and properties from the datapath. */
2187 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
2188 ? ofport_open(ofproto, &ofproto_port, &pp)
2192 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
2193 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
2194 struct netdev *old_netdev = port->netdev;
2196 /* 'name' hasn't changed location. Any properties changed? */
2197 if (!ofport_equal(&port->pp, &pp)) {
2198 ofport_modified(port, &pp);
2201 update_mtu(ofproto, port);
2203 /* Install the newly opened netdev in case it has changed.
2204 * Don't close the old netdev yet in case port_modified has to
2205 * remove a retained reference to it.*/
2206 port->netdev = netdev;
2207 port->change_seq = netdev_change_seq(netdev);
2209 if (port->ofproto->ofproto_class->port_modified) {
2210 port->ofproto->ofproto_class->port_modified(port);
2213 netdev_close(old_netdev);
2215 /* If 'port' is nonnull then its name differs from 'name' and thus
2216 * we should delete it. If we think there's a port named 'name'
2217 * then its port number must be wrong now so delete it too. */
2219 ofport_remove(port);
2221 ofport_remove_with_name(ofproto, name);
2222 ofport_install(ofproto, netdev, &pp);
2225 /* Any port named 'name' is gone now. */
2226 ofport_remove_with_name(ofproto, name);
2228 ofproto_port_destroy(&ofproto_port);
2232 init_ports(struct ofproto *p)
2234 struct ofproto_port_dump dump;
2235 struct ofproto_port ofproto_port;
2236 struct shash_node *node, *next;
2238 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2239 const char *name = ofproto_port.name;
2241 if (shash_find(&p->port_by_name, name)) {
2242 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2245 struct ofputil_phy_port pp;
2246 struct netdev *netdev;
2248 /* Check if an OpenFlow port number had been requested. */
2249 node = shash_find(&init_ofp_ports, name);
2251 const struct iface_hint *iface_hint = node->data;
2252 simap_put(&p->ofp_requests, name,
2253 ofp_to_u16(iface_hint->ofp_port));
2256 netdev = ofport_open(p, &ofproto_port, &pp);
2258 ofport_install(p, netdev, &pp);
2259 if (ofp_to_u16(ofproto_port.ofp_port) < p->max_ports) {
2260 p->alloc_port_no = MAX(p->alloc_port_no,
2261 ofp_to_u16(ofproto_port.ofp_port));
2267 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2268 struct iface_hint *iface_hint = node->data;
2270 if (!strcmp(iface_hint->br_name, p->name)) {
2271 free(iface_hint->br_name);
2272 free(iface_hint->br_type);
2274 shash_delete(&init_ofp_ports, node);
2281 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2282 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2284 find_min_mtu(struct ofproto *p)
2286 struct ofport *ofport;
2289 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2290 struct netdev *netdev = ofport->netdev;
2293 /* Skip any internal ports, since that's what we're trying to
2295 if (!strcmp(netdev_get_type(netdev), "internal")) {
2299 if (netdev_get_mtu(netdev, &dev_mtu)) {
2302 if (!mtu || dev_mtu < mtu) {
2307 return mtu ? mtu: ETH_PAYLOAD_MAX;
2310 /* Update MTU of all datapath devices on 'p' to the minimum of the
2311 * non-datapath ports in event of 'port' added or changed. */
2313 update_mtu(struct ofproto *p, struct ofport *port)
2315 struct ofport *ofport;
2316 struct netdev *netdev = port->netdev;
2317 int dev_mtu, old_min;
2319 if (netdev_get_mtu(netdev, &dev_mtu)) {
2323 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2324 if (dev_mtu > p->min_mtu) {
2325 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2326 dev_mtu = p->min_mtu;
2329 port->mtu = dev_mtu;
2333 /* For non-internal port find new min mtu. */
2334 old_min = p->min_mtu;
2335 port->mtu = dev_mtu;
2336 p->min_mtu = find_min_mtu(p);
2337 if (p->min_mtu == old_min) {
2341 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2342 struct netdev *netdev = ofport->netdev;
2344 if (!strcmp(netdev_get_type(netdev), "internal")) {
2345 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2346 ofport->mtu = p->min_mtu;
2353 ofproto_rule_ref(struct rule *rule)
2358 atomic_add(&rule->ref_count, 1, &orig);
2359 ovs_assert(orig != 0);
2364 ofproto_rule_unref(struct rule *rule)
2369 atomic_sub(&rule->ref_count, 1, &orig);
2371 rule->ofproto->ofproto_class->rule_destruct(rule);
2372 ofproto_rule_destroy__(rule);
2374 ovs_assert(orig != 0);
2380 ofproto_rule_destroy__(struct rule *rule)
2382 cls_rule_destroy(&rule->cr);
2383 rule_actions_unref(rule->actions);
2384 ovs_mutex_destroy(&rule->timeout_mutex);
2385 ovs_rwlock_destroy(&rule->rwlock);
2386 rule->ofproto->ofproto_class->rule_dealloc(rule);
2389 /* Creates and returns a new 'struct rule_actions', with a ref_count of 1,
2390 * whose actions are a copy of from the 'ofpacts_len' bytes of 'ofpacts'. */
2391 struct rule_actions *
2392 rule_actions_create(const struct ofpact *ofpacts, size_t ofpacts_len)
2394 struct rule_actions *actions;
2396 actions = xmalloc(sizeof *actions);
2397 atomic_init(&actions->ref_count, 1);
2398 actions->ofpacts = xmemdup(ofpacts, ofpacts_len);
2399 actions->ofpacts_len = ofpacts_len;
2400 actions->meter_id = ofpacts_get_meter(ofpacts, ofpacts_len);
2404 /* Increments 'actions''s ref_count. */
2406 rule_actions_ref(struct rule_actions *actions)
2411 atomic_add(&actions->ref_count, 1, &orig);
2412 ovs_assert(orig != 0);
2416 /* Decrements 'actions''s ref_count and frees 'actions' if the ref_count
2419 rule_actions_unref(struct rule_actions *actions)
2424 atomic_sub(&actions->ref_count, 1, &orig);
2428 ovs_assert(orig != 0);
2433 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2434 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2436 ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port)
2438 return (port == OFPP_ANY
2439 || ofpacts_output_to_port(rule->actions->ofpacts,
2440 rule->actions->ofpacts_len, port));
2443 /* Returns true if 'rule' has group and equals group_id. */
2445 ofproto_rule_has_out_group(const struct rule *rule, uint32_t group_id)
2447 return (group_id == OFPG11_ANY
2448 || ofpacts_output_to_group(rule->actions->ofpacts,
2449 rule->actions->ofpacts_len, group_id));
2452 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2453 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2455 ofoperation_has_out_port(const struct ofoperation *op, ofp_port_t out_port)
2457 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2462 case OFOPERATION_ADD:
2463 case OFOPERATION_DELETE:
2466 case OFOPERATION_MODIFY:
2467 case OFOPERATION_REPLACE:
2468 return ofpacts_output_to_port(op->actions->ofpacts,
2469 op->actions->ofpacts_len, out_port);
2476 rule_execute_destroy(struct rule_execute *e)
2478 ofproto_rule_unref(e->rule);
2479 list_remove(&e->list_node);
2483 /* Executes all "rule_execute" operations queued up in ofproto->rule_executes,
2484 * by passing them to the ofproto provider. */
2486 run_rule_executes(struct ofproto *ofproto)
2488 struct rule_execute *e, *next;
2489 struct list executes;
2491 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2492 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2493 union flow_in_port in_port_;
2496 in_port_.ofp_port = e->in_port;
2497 flow_extract(e->packet, 0, 0, NULL, &in_port_, &flow);
2498 ofproto->ofproto_class->rule_execute(e->rule, &flow, e->packet);
2500 rule_execute_destroy(e);
2504 /* Destroys and discards all "rule_execute" operations queued up in
2505 * ofproto->rule_executes. */
2507 destroy_rule_executes(struct ofproto *ofproto)
2509 struct rule_execute *e, *next;
2510 struct list executes;
2512 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2513 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2514 ofpbuf_delete(e->packet);
2515 rule_execute_destroy(e);
2519 /* Returns true if 'rule' should be hidden from the controller.
2521 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2522 * (e.g. by in-band control) and are intentionally hidden from the
2525 ofproto_rule_is_hidden(const struct rule *rule)
2527 return rule->cr.priority > UINT16_MAX;
2530 static enum oftable_flags
2531 rule_get_flags(const struct rule *rule)
2533 return rule->ofproto->tables[rule->table_id].flags;
2537 rule_is_modifiable(const struct rule *rule)
2539 return !(rule_get_flags(rule) & OFTABLE_READONLY);
2543 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2545 ofconn_send_reply(ofconn, make_echo_reply(oh));
2550 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2552 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2553 struct ofputil_switch_features features;
2554 struct ofport *port;
2560 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2562 ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2564 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2565 * if present, are always at the end.) */
2566 n_tables = ofproto->n_tables;
2567 for (i = 0; i < ofproto->n_tables; i++) {
2568 if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
2574 features.datapath_id = ofproto->datapath_id;
2575 features.n_buffers = pktbuf_capacity();
2576 features.n_tables = n_tables;
2577 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2578 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2580 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2582 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2583 features.auxiliary_id = 0;
2584 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2586 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2587 ofputil_put_switch_features_port(&port->pp, b);
2590 ofconn_send_reply(ofconn, b);
2595 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2597 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2598 struct ofp_switch_config *osc;
2599 enum ofp_config_flags flags;
2603 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2604 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2605 flags = ofproto->frag_handling;
2606 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2607 if (oh->version < OFP13_VERSION
2608 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2609 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2611 osc->flags = htons(flags);
2612 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2613 ofconn_send_reply(ofconn, buf);
2619 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2621 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2622 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2623 uint16_t flags = ntohs(osc->flags);
2625 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2626 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
2627 enum ofp_config_flags cur = ofproto->frag_handling;
2628 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2630 ovs_assert((cur & OFPC_FRAG_MASK) == cur);
2632 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2633 ofproto->frag_handling = next;
2635 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2637 ofputil_frag_handling_to_string(next));
2641 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2642 ofconn_set_invalid_ttl_to_controller(ofconn,
2643 (oh->version < OFP13_VERSION
2644 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2646 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2651 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2652 * error message code for the caller to propagate upward. Otherwise, returns
2655 * The log message mentions 'msg_type'. */
2657 reject_slave_controller(struct ofconn *ofconn)
2659 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2660 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
2661 return OFPERR_OFPBRC_EPERM;
2667 /* Checks that the 'ofpacts_len' bytes of actions in 'ofpacts' are appropriate
2668 * for a packet with the prerequisites satisfied by 'flow' in table 'table_id'.
2669 * 'flow' may be temporarily modified, but is restored at return.
2672 ofproto_check_ofpacts(struct ofproto *ofproto,
2673 const struct ofpact ofpacts[], size_t ofpacts_len,
2674 struct flow *flow, uint8_t table_id)
2679 error = ofpacts_check(ofpacts, ofpacts_len, flow,
2680 u16_to_ofp(ofproto->max_ports), table_id);
2685 mid = ofpacts_get_meter(ofpacts, ofpacts_len);
2686 if (mid && ofproto_get_provider_meter_id(ofproto, mid) == UINT32_MAX) {
2687 return OFPERR_OFPMMFC_INVALID_METER;
2693 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2695 struct ofproto *p = ofconn_get_ofproto(ofconn);
2696 struct ofputil_packet_out po;
2697 struct ofpbuf *payload;
2698 uint64_t ofpacts_stub[1024 / 8];
2699 struct ofpbuf ofpacts;
2701 union flow_in_port in_port_;
2704 COVERAGE_INC(ofproto_packet_out);
2706 error = reject_slave_controller(ofconn);
2711 /* Decode message. */
2712 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2713 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2715 goto exit_free_ofpacts;
2717 if (ofp_to_u16(po.in_port) >= p->max_ports
2718 && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) {
2719 error = OFPERR_OFPBRC_BAD_PORT;
2720 goto exit_free_ofpacts;
2725 if (po.buffer_id != UINT32_MAX) {
2726 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2727 if (error || !payload) {
2728 goto exit_free_ofpacts;
2731 /* Ensure that the L3 header is 32-bit aligned. */
2732 payload = ofpbuf_clone_data_with_headroom(po.packet, po.packet_len, 2);
2735 /* Verify actions against packet, then send packet if successful. */
2736 in_port_.ofp_port = po.in_port;
2737 flow_extract(payload, 0, 0, NULL, &in_port_, &flow);
2738 error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len, &flow, 0);
2740 error = p->ofproto_class->packet_out(p, payload, &flow,
2741 po.ofpacts, po.ofpacts_len);
2743 ofpbuf_delete(payload);
2746 ofpbuf_uninit(&ofpacts);
2752 update_port_config(struct ofport *port,
2753 enum ofputil_port_config config,
2754 enum ofputil_port_config mask)
2756 enum ofputil_port_config old_config = port->pp.config;
2757 enum ofputil_port_config toggle;
2759 toggle = (config ^ port->pp.config) & mask;
2760 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2761 if (config & OFPUTIL_PC_PORT_DOWN) {
2762 netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL);
2764 netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL);
2766 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2769 port->pp.config ^= toggle;
2770 if (port->pp.config != old_config) {
2771 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2776 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2778 struct ofproto *p = ofconn_get_ofproto(ofconn);
2779 struct ofputil_port_mod pm;
2780 struct ofport *port;
2783 error = reject_slave_controller(ofconn);
2788 error = ofputil_decode_port_mod(oh, &pm);
2793 port = ofproto_get_port(p, pm.port_no);
2795 return OFPERR_OFPPMFC_BAD_PORT;
2796 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2797 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2799 update_port_config(port, pm.config, pm.mask);
2801 netdev_set_advertisements(port->netdev, pm.advertise);
2808 handle_desc_stats_request(struct ofconn *ofconn,
2809 const struct ofp_header *request)
2811 static const char *default_mfr_desc = "Nicira, Inc.";
2812 static const char *default_hw_desc = "Open vSwitch";
2813 static const char *default_sw_desc = VERSION;
2814 static const char *default_serial_desc = "None";
2815 static const char *default_dp_desc = "None";
2817 struct ofproto *p = ofconn_get_ofproto(ofconn);
2818 struct ofp_desc_stats *ods;
2821 msg = ofpraw_alloc_stats_reply(request, 0);
2822 ods = ofpbuf_put_zeros(msg, sizeof *ods);
2823 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
2824 sizeof ods->mfr_desc);
2825 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
2826 sizeof ods->hw_desc);
2827 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
2828 sizeof ods->sw_desc);
2829 ovs_strlcpy(ods->serial_num,
2830 p->serial_desc ? p->serial_desc : default_serial_desc,
2831 sizeof ods->serial_num);
2832 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
2833 sizeof ods->dp_desc);
2834 ofconn_send_reply(ofconn, msg);
2840 handle_table_stats_request(struct ofconn *ofconn,
2841 const struct ofp_header *request)
2843 struct ofproto *p = ofconn_get_ofproto(ofconn);
2844 struct ofp12_table_stats *ots;
2849 /* Set up default values.
2851 * ofp12_table_stats is used as a generic structure as
2852 * it is able to hold all the fields for ofp10_table_stats
2853 * and ofp11_table_stats (and of course itself).
2855 ots = xcalloc(p->n_tables, sizeof *ots);
2856 for (i = 0; i < p->n_tables; i++) {
2857 ots[i].table_id = i;
2858 sprintf(ots[i].name, "table%zu", i);
2859 ots[i].match = htonll(OFPXMT12_MASK);
2860 ots[i].wildcards = htonll(OFPXMT12_MASK);
2861 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
2862 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
2863 ots[i].write_setfields = htonll(OFPXMT12_MASK);
2864 ots[i].apply_setfields = htonll(OFPXMT12_MASK);
2865 ots[i].metadata_match = htonll(UINT64_MAX);
2866 ots[i].metadata_write = htonll(UINT64_MAX);
2867 ots[i].instructions = htonl(OFPIT11_ALL);
2868 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
2869 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2870 ovs_rwlock_rdlock(&p->tables[i].cls.rwlock);
2871 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2872 ovs_rwlock_unlock(&p->tables[i].cls.rwlock);
2875 p->ofproto_class->get_tables(p, ots);
2877 /* Post-process the tables, dropping hidden tables. */
2878 n_tables = p->n_tables;
2879 for (i = 0; i < p->n_tables; i++) {
2880 const struct oftable *table = &p->tables[i];
2882 if (table->flags & OFTABLE_HIDDEN) {
2888 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2891 if (table->max_flows < ntohl(ots[i].max_entries)) {
2892 ots[i].max_entries = htonl(table->max_flows);
2896 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
2897 ofconn_send_reply(ofconn, msg);
2905 append_port_stat(struct ofport *port, struct list *replies)
2907 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
2909 calc_duration(port->created, time_msec(),
2910 &ops.duration_sec, &ops.duration_nsec);
2912 /* Intentionally ignore return value, since errors will set
2913 * 'stats' to all-1s, which is correct for OpenFlow, and
2914 * netdev_get_stats() will log errors. */
2915 ofproto_port_get_stats(port, &ops.stats);
2917 ofputil_append_port_stat(replies, &ops);
2921 handle_port_stats_request(struct ofconn *ofconn,
2922 const struct ofp_header *request)
2924 struct ofproto *p = ofconn_get_ofproto(ofconn);
2925 struct ofport *port;
2926 struct list replies;
2930 error = ofputil_decode_port_stats_request(request, &port_no);
2935 ofpmp_init(&replies, request);
2936 if (port_no != OFPP_ANY) {
2937 port = ofproto_get_port(p, port_no);
2939 append_port_stat(port, &replies);
2942 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2943 append_port_stat(port, &replies);
2947 ofconn_send_replies(ofconn, &replies);
2952 handle_port_desc_stats_request(struct ofconn *ofconn,
2953 const struct ofp_header *request)
2955 struct ofproto *p = ofconn_get_ofproto(ofconn);
2956 enum ofp_version version;
2957 struct ofport *port;
2958 struct list replies;
2960 ofpmp_init(&replies, request);
2962 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
2963 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2964 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
2967 ofconn_send_replies(ofconn, &replies);
2972 hash_cookie(ovs_be64 cookie)
2974 return hash_2words((OVS_FORCE uint64_t)cookie >> 32,
2975 (OVS_FORCE uint64_t)cookie);
2979 cookies_insert(struct ofproto *ofproto, struct rule *rule)
2981 hindex_insert(&ofproto->cookies, &rule->cookie_node,
2982 hash_cookie(rule->flow_cookie));
2986 cookies_remove(struct ofproto *ofproto, struct rule *rule)
2988 hindex_remove(&ofproto->cookies, &rule->cookie_node);
2992 ofproto_rule_change_cookie(struct ofproto *ofproto, struct rule *rule,
2993 ovs_be64 new_cookie)
2995 if (new_cookie != rule->flow_cookie) {
2996 cookies_remove(ofproto, rule);
2998 ovs_rwlock_wrlock(&rule->rwlock);
2999 rule->flow_cookie = new_cookie;
3000 ovs_rwlock_unlock(&rule->rwlock);
3002 cookies_insert(ofproto, rule);
3007 calc_duration(long long int start, long long int now,
3008 uint32_t *sec, uint32_t *nsec)
3010 long long int msecs = now - start;
3011 *sec = msecs / 1000;
3012 *nsec = (msecs % 1000) * (1000 * 1000);
3015 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
3016 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
3018 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
3020 return (table_id == 0xff || table_id < ofproto->n_tables
3022 : OFPERR_OFPBRC_BAD_TABLE_ID);
3026 static struct oftable *
3027 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
3029 struct oftable *table;
3031 for (table = &ofproto->tables[table_id];
3032 table < &ofproto->tables[ofproto->n_tables];
3034 if (!(table->flags & OFTABLE_HIDDEN)) {
3042 static struct oftable *
3043 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
3045 if (table_id == 0xff) {
3046 return next_visible_table(ofproto, 0);
3047 } else if (table_id < ofproto->n_tables) {
3048 return &ofproto->tables[table_id];
3054 static struct oftable *
3055 next_matching_table(const struct ofproto *ofproto,
3056 const struct oftable *table, uint8_t table_id)
3058 return (table_id == 0xff
3059 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
3063 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
3065 * - If TABLE_ID is 0xff, this iterates over every classifier table in
3066 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
3068 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
3069 * only once, for that table. (This can be used to access tables marked
3072 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
3073 * entered at all. (Perhaps you should have validated TABLE_ID with
3074 * check_table_id().)
3076 * All parameters are evaluated multiple times.
3078 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
3079 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
3081 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
3083 /* Initializes 'criteria' in a straightforward way based on the other
3086 * For "loose" matching, the 'priority' parameter is unimportant and may be
3089 rule_criteria_init(struct rule_criteria *criteria, uint8_t table_id,
3090 const struct match *match, unsigned int priority,
3091 ovs_be64 cookie, ovs_be64 cookie_mask,
3092 ofp_port_t out_port, uint32_t out_group)
3094 criteria->table_id = table_id;
3095 cls_rule_init(&criteria->cr, match, priority);
3096 criteria->cookie = cookie;
3097 criteria->cookie_mask = cookie_mask;
3098 criteria->out_port = out_port;
3099 criteria->out_group = out_group;
3103 rule_criteria_destroy(struct rule_criteria *criteria)
3105 cls_rule_destroy(&criteria->cr);
3109 rule_collection_init(struct rule_collection *rules)
3111 rules->rules = rules->stub;
3113 rules->capacity = ARRAY_SIZE(rules->stub);
3117 rule_collection_add(struct rule_collection *rules, struct rule *rule)
3119 if (rules->n >= rules->capacity) {
3120 size_t old_size, new_size;
3122 old_size = rules->capacity * sizeof *rules->rules;
3123 rules->capacity *= 2;
3124 new_size = rules->capacity * sizeof *rules->rules;
3126 if (rules->rules == rules->stub) {
3127 rules->rules = xmalloc(new_size);
3128 memcpy(rules->rules, rules->stub, old_size);
3130 rules->rules = xrealloc(rules->rules, new_size);
3134 rules->rules[rules->n++] = rule;
3138 rule_collection_destroy(struct rule_collection *rules)
3140 if (rules->rules != rules->stub) {
3146 collect_rule(struct rule *rule, const struct rule_criteria *c,
3147 struct rule_collection *rules)
3149 if (ofproto_rule_is_hidden(rule)) {
3151 } else if (rule->pending) {
3152 return OFPROTO_POSTPONE;
3154 if ((c->table_id == rule->table_id || c->table_id == 0xff)
3155 && ofproto_rule_has_out_port(rule, c->out_port)
3156 && ofproto_rule_has_out_group(rule, c->out_group)
3157 && !((rule->flow_cookie ^ c->cookie) & c->cookie_mask)) {
3158 rule_collection_add(rules, rule);
3164 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3165 * on classifiers rules are done in the "loose" way required for OpenFlow
3166 * OFPFC_MODIFY and OFPFC_DELETE requests. Puts the selected rules on list
3169 * Hidden rules are always omitted.
3171 * Returns 0 on success, otherwise an OpenFlow error code. */
3173 collect_rules_loose(struct ofproto *ofproto,
3174 const struct rule_criteria *criteria,
3175 struct rule_collection *rules)
3177 struct oftable *table;
3180 rule_collection_init(rules);
3182 error = check_table_id(ofproto, criteria->table_id);
3187 if (criteria->cookie_mask == htonll(UINT64_MAX)) {
3190 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
3191 hash_cookie(criteria->cookie),
3192 &ofproto->cookies) {
3193 if (cls_rule_is_loose_match(&rule->cr, &criteria->cr.match)) {
3194 error = collect_rule(rule, criteria, rules);
3201 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
3202 struct cls_cursor cursor;
3205 ovs_rwlock_rdlock(&table->cls.rwlock);
3206 cls_cursor_init(&cursor, &table->cls, &criteria->cr);
3207 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3208 error = collect_rule(rule, criteria, rules);
3213 ovs_rwlock_unlock(&table->cls.rwlock);
3219 rule_collection_destroy(rules);
3224 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3225 * on classifiers rules are done in the "strict" way required for OpenFlow
3226 * OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests. Puts the selected
3227 * rules on list 'rules'.
3229 * Hidden rules are always omitted.
3231 * Returns 0 on success, otherwise an OpenFlow error code. */
3233 collect_rules_strict(struct ofproto *ofproto,
3234 const struct rule_criteria *criteria,
3235 struct rule_collection *rules)
3237 struct oftable *table;
3240 rule_collection_init(rules);
3242 error = check_table_id(ofproto, criteria->table_id);
3247 if (criteria->cookie_mask == htonll(UINT64_MAX)) {
3250 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
3251 hash_cookie(criteria->cookie),
3252 &ofproto->cookies) {
3253 if (cls_rule_equal(&rule->cr, &criteria->cr)) {
3254 error = collect_rule(rule, criteria, rules);
3261 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
3264 ovs_rwlock_rdlock(&table->cls.rwlock);
3265 rule = rule_from_cls_rule(classifier_find_rule_exactly(
3266 &table->cls, &criteria->cr));
3267 ovs_rwlock_unlock(&table->cls.rwlock);
3269 error = collect_rule(rule, criteria, rules);
3279 rule_collection_destroy(rules);
3284 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
3285 * forced into the range of a uint16_t. */
3287 age_secs(long long int age_ms)
3289 return (age_ms < 0 ? 0
3290 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
3291 : (unsigned int) age_ms / 1000);
3295 handle_flow_stats_request(struct ofconn *ofconn,
3296 const struct ofp_header *request)
3298 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3299 struct ofputil_flow_stats_request fsr;
3300 struct rule_criteria criteria;
3301 struct rule_collection rules;
3302 struct list replies;
3306 error = ofputil_decode_flow_stats_request(&fsr, request);
3311 rule_criteria_init(&criteria, fsr.table_id, &fsr.match, 0, fsr.cookie,
3312 fsr.cookie_mask, fsr.out_port, fsr.out_group);
3313 error = collect_rules_loose(ofproto, &criteria, &rules);
3314 rule_criteria_destroy(&criteria);
3319 ofpmp_init(&replies, request);
3320 for (i = 0; i < rules.n; i++) {
3321 struct rule *rule = rules.rules[i];
3322 long long int now = time_msec();
3323 struct ofputil_flow_stats fs;
3325 minimatch_expand(&rule->cr.match, &fs.match);
3326 fs.priority = rule->cr.priority;
3327 fs.cookie = rule->flow_cookie;
3328 fs.table_id = rule->table_id;
3329 calc_duration(rule->created, now, &fs.duration_sec, &fs.duration_nsec);
3330 fs.idle_age = age_secs(now - rule->used);
3331 fs.hard_age = age_secs(now - rule->modified);
3332 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
3334 fs.ofpacts = rule->actions->ofpacts;
3335 fs.ofpacts_len = rule->actions->ofpacts_len;
3337 ovs_mutex_lock(&rule->timeout_mutex);
3338 fs.idle_timeout = rule->idle_timeout;
3339 fs.hard_timeout = rule->hard_timeout;
3340 ovs_mutex_unlock(&rule->timeout_mutex);
3342 fs.flags = rule->flags;
3344 ofputil_append_flow_stats_reply(&fs, &replies);
3346 rule_collection_destroy(&rules);
3348 ofconn_send_replies(ofconn, &replies);
3354 flow_stats_ds(struct rule *rule, struct ds *results)
3356 uint64_t packet_count, byte_count;
3358 rule->ofproto->ofproto_class->rule_get_stats(rule,
3359 &packet_count, &byte_count);
3361 if (rule->table_id != 0) {
3362 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
3364 ds_put_format(results, "duration=%llds, ",
3365 (time_msec() - rule->created) / 1000);
3366 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3367 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3368 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3369 cls_rule_format(&rule->cr, results);
3370 ds_put_char(results, ',');
3371 ofpacts_format(rule->actions->ofpacts, rule->actions->ofpacts_len,
3373 ds_put_cstr(results, "\n");
3376 /* Adds a pretty-printed description of all flows to 'results', including
3377 * hidden flows (e.g., set up by in-band control). */
3379 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3381 struct oftable *table;
3383 OFPROTO_FOR_EACH_TABLE (table, p) {
3384 struct cls_cursor cursor;
3387 ovs_rwlock_rdlock(&table->cls.rwlock);
3388 cls_cursor_init(&cursor, &table->cls, NULL);
3389 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3390 flow_stats_ds(rule, results);
3392 ovs_rwlock_unlock(&table->cls.rwlock);
3396 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
3397 * '*engine_type' and '*engine_id', respectively. */
3399 ofproto_get_netflow_ids(const struct ofproto *ofproto,
3400 uint8_t *engine_type, uint8_t *engine_id)
3402 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
3405 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
3406 * true if the port's CFM status was successfully stored into '*status'.
3407 * Returns false if the port did not have CFM configured, in which case
3408 * '*status' is indeterminate.
3410 * The caller must provide and owns '*status', and must free 'status->rmps'. */
3412 ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port,
3413 struct ofproto_cfm_status *status)
3415 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
3417 && ofproto->ofproto_class->get_cfm_status
3418 && ofproto->ofproto_class->get_cfm_status(ofport, status));
3422 handle_aggregate_stats_request(struct ofconn *ofconn,
3423 const struct ofp_header *oh)
3425 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3426 struct ofputil_flow_stats_request request;
3427 struct ofputil_aggregate_stats stats;
3428 bool unknown_packets, unknown_bytes;
3429 struct rule_criteria criteria;
3430 struct rule_collection rules;
3431 struct ofpbuf *reply;
3435 error = ofputil_decode_flow_stats_request(&request, oh);
3440 rule_criteria_init(&criteria, request.table_id, &request.match, 0,
3441 request.cookie, request.cookie_mask,
3442 request.out_port, request.out_group);
3443 error = collect_rules_loose(ofproto, &criteria, &rules);
3444 rule_criteria_destroy(&criteria);
3449 memset(&stats, 0, sizeof stats);
3450 unknown_packets = unknown_bytes = false;
3451 for (i = 0; i < rules.n; i++) {
3452 struct rule *rule = rules.rules[i];
3453 uint64_t packet_count;
3454 uint64_t byte_count;
3456 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
3459 if (packet_count == UINT64_MAX) {
3460 unknown_packets = true;
3462 stats.packet_count += packet_count;
3465 if (byte_count == UINT64_MAX) {
3466 unknown_bytes = true;
3468 stats.byte_count += byte_count;
3473 if (unknown_packets) {
3474 stats.packet_count = UINT64_MAX;
3476 if (unknown_bytes) {
3477 stats.byte_count = UINT64_MAX;
3480 rule_collection_destroy(&rules);
3482 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
3483 ofconn_send_reply(ofconn, reply);
3488 struct queue_stats_cbdata {
3489 struct ofport *ofport;
3490 struct list replies;
3495 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3496 const struct netdev_queue_stats *stats)
3498 struct ofputil_queue_stats oqs;
3500 oqs.port_no = cbdata->ofport->pp.port_no;
3501 oqs.queue_id = queue_id;
3502 oqs.tx_bytes = stats->tx_bytes;
3503 oqs.tx_packets = stats->tx_packets;
3504 oqs.tx_errors = stats->tx_errors;
3505 if (stats->created != LLONG_MIN) {
3506 calc_duration(stats->created, cbdata->now,
3507 &oqs.duration_sec, &oqs.duration_nsec);
3509 oqs.duration_sec = oqs.duration_nsec = UINT32_MAX;
3511 ofputil_append_queue_stat(&cbdata->replies, &oqs);
3515 handle_queue_stats_dump_cb(uint32_t queue_id,
3516 struct netdev_queue_stats *stats,
3519 struct queue_stats_cbdata *cbdata = cbdata_;
3521 put_queue_stats(cbdata, queue_id, stats);
3525 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3526 struct queue_stats_cbdata *cbdata)
3528 cbdata->ofport = port;
3529 if (queue_id == OFPQ_ALL) {
3530 netdev_dump_queue_stats(port->netdev,
3531 handle_queue_stats_dump_cb, cbdata);
3533 struct netdev_queue_stats stats;
3535 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3536 put_queue_stats(cbdata, queue_id, &stats);
3538 return OFPERR_OFPQOFC_BAD_QUEUE;
3545 handle_queue_stats_request(struct ofconn *ofconn,
3546 const struct ofp_header *rq)
3548 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3549 struct queue_stats_cbdata cbdata;
3550 struct ofport *port;
3552 struct ofputil_queue_stats_request oqsr;
3554 COVERAGE_INC(ofproto_queue_req);
3556 ofpmp_init(&cbdata.replies, rq);
3557 cbdata.now = time_msec();
3559 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3564 if (oqsr.port_no == OFPP_ANY) {
3565 error = OFPERR_OFPQOFC_BAD_QUEUE;
3566 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3567 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3572 port = ofproto_get_port(ofproto, oqsr.port_no);
3574 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3575 : OFPERR_OFPQOFC_BAD_PORT);
3578 ofconn_send_replies(ofconn, &cbdata.replies);
3580 ofpbuf_list_delete(&cbdata.replies);
3587 is_flow_deletion_pending(const struct ofproto *ofproto,
3588 const struct cls_rule *cls_rule,
3591 if (!hmap_is_empty(&ofproto->deletions)) {
3592 struct ofoperation *op;
3594 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3595 cls_rule_hash(cls_rule, table_id),
3596 &ofproto->deletions) {
3597 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
3607 evict_rule_from_table(struct ofproto *ofproto, struct oftable *table)
3612 ovs_rwlock_rdlock(&table->cls.rwlock);
3613 n_rules = classifier_count(&table->cls);
3614 ovs_rwlock_unlock(&table->cls.rwlock);
3616 if (n_rules < table->max_flows) {
3618 } else if (!choose_rule_to_evict(table, &rule)) {
3619 return OFPERR_OFPFMFC_TABLE_FULL;
3620 } else if (rule->pending) {
3621 ovs_rwlock_unlock(&rule->rwlock);
3622 return OFPROTO_POSTPONE;
3624 struct ofopgroup *group;
3626 group = ofopgroup_create_unattached(ofproto);
3627 delete_flow__(rule, group, OFPRR_EVICTION);
3628 ofopgroup_submit(group);
3634 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3635 * in which no matching flow already exists in the flow table.
3637 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3638 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3639 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3640 * initiated now but may be retried later.
3642 * The caller retains ownership of 'fm->ofpacts'.
3644 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3647 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
3648 struct ofputil_flow_mod *fm, const struct ofp_header *request)
3650 struct oftable *table;
3651 struct ofopgroup *group;
3657 error = check_table_id(ofproto, fm->table_id);
3663 if (fm->table_id == 0xff) {
3664 if (ofproto->ofproto_class->rule_choose_table) {
3665 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3671 ovs_assert(table_id < ofproto->n_tables);
3675 } else if (fm->table_id < ofproto->n_tables) {
3676 table_id = fm->table_id;
3678 return OFPERR_OFPBRC_BAD_TABLE_ID;
3681 table = &ofproto->tables[table_id];
3683 if (table->flags & OFTABLE_READONLY) {
3684 return OFPERR_OFPBRC_EPERM;
3687 cls_rule_init(&cr, &fm->match, fm->priority);
3689 /* Transform "add" into "modify" if there's an existing identical flow. */
3690 ovs_rwlock_rdlock(&table->cls.rwlock);
3691 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr));
3692 ovs_rwlock_unlock(&table->cls.rwlock);
3694 cls_rule_destroy(&cr);
3695 if (!rule_is_modifiable(rule)) {
3696 return OFPERR_OFPBRC_EPERM;
3697 } else if (rule->pending) {
3698 return OFPROTO_POSTPONE;
3700 struct rule_collection rules;
3702 rule_collection_init(&rules);
3703 rule_collection_add(&rules, rule);
3704 fm->modify_cookie = true;
3705 error = modify_flows__(ofproto, ofconn, fm, request, &rules);
3706 rule_collection_destroy(&rules);
3712 /* Verify actions. */
3713 error = ofproto_check_ofpacts(ofproto, fm->ofpacts, fm->ofpacts_len,
3714 &fm->match.flow, table_id);
3716 cls_rule_destroy(&cr);
3720 /* Serialize against pending deletion. */
3721 if (is_flow_deletion_pending(ofproto, &cr, table_id)) {
3722 cls_rule_destroy(&cr);
3723 return OFPROTO_POSTPONE;
3726 /* Check for overlap, if requested. */
3727 if (fm->flags & OFPUTIL_FF_CHECK_OVERLAP) {
3730 ovs_rwlock_rdlock(&table->cls.rwlock);
3731 overlaps = classifier_rule_overlaps(&table->cls, &cr);
3732 ovs_rwlock_unlock(&table->cls.rwlock);
3735 cls_rule_destroy(&cr);
3736 return OFPERR_OFPFMFC_OVERLAP;
3740 /* If necessary, evict an existing rule to clear out space. */
3741 error = evict_rule_from_table(ofproto, table);
3743 cls_rule_destroy(&cr);
3747 /* Allocate new rule. */
3748 rule = ofproto->ofproto_class->rule_alloc();
3750 cls_rule_destroy(&cr);
3751 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
3752 ofproto->name, ovs_strerror(error));
3756 /* Initialize base state. */
3757 rule->ofproto = ofproto;
3758 cls_rule_move(&rule->cr, &cr);
3759 atomic_init(&rule->ref_count, 1);
3760 rule->pending = NULL;
3761 rule->flow_cookie = fm->new_cookie;
3762 rule->created = rule->modified = rule->used = time_msec();
3764 ovs_mutex_init(&rule->timeout_mutex);
3765 ovs_mutex_lock(&rule->timeout_mutex);
3766 rule->idle_timeout = fm->idle_timeout;
3767 rule->hard_timeout = fm->hard_timeout;
3768 ovs_mutex_unlock(&rule->timeout_mutex);
3770 rule->table_id = table - ofproto->tables;
3771 rule->flags = fm->flags & OFPUTIL_FF_STATE;
3772 rule->actions = rule_actions_create(fm->ofpacts, fm->ofpacts_len);
3773 list_init(&rule->meter_list_node);
3774 rule->eviction_group = NULL;
3775 list_init(&rule->expirable);
3776 rule->monitor_flags = 0;
3777 rule->add_seqno = 0;
3778 rule->modify_seqno = 0;
3779 ovs_rwlock_init(&rule->rwlock);
3781 /* Construct rule, initializing derived state. */
3782 error = ofproto->ofproto_class->rule_construct(rule);
3784 ofproto_rule_destroy__(rule);
3789 oftable_insert_rule(rule);
3791 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3792 ofoperation_create(group, rule, OFOPERATION_ADD, 0);
3793 ofproto->ofproto_class->rule_insert(rule);
3794 ofopgroup_submit(group);
3799 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3801 /* Modifies the rules listed in 'rules', changing their actions to match those
3804 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3807 * Returns 0 on success, otherwise an OpenFlow error code. */
3809 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3810 struct ofputil_flow_mod *fm, const struct ofp_header *request,
3811 const struct rule_collection *rules)
3813 enum ofoperation_type type;
3814 struct ofopgroup *group;
3818 type = fm->command == OFPFC_ADD ? OFOPERATION_REPLACE : OFOPERATION_MODIFY;
3819 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3820 error = OFPERR_OFPBRC_EPERM;
3821 for (i = 0; i < rules->n; i++) {
3822 struct rule *rule = rules->rules[i];
3823 struct ofoperation *op;
3824 bool actions_changed;
3825 bool reset_counters;
3827 /* FIXME: Implement OFPFUTIL_FF_RESET_COUNTS */
3829 if (rule_is_modifiable(rule)) {
3830 /* At least one rule is modifiable, don't report EPERM error. */
3836 /* Verify actions. */
3837 error = ofpacts_check(fm->ofpacts, fm->ofpacts_len, &fm->match.flow,
3838 u16_to_ofp(ofproto->max_ports), rule->table_id);
3843 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
3844 rule->actions->ofpacts,
3845 rule->actions->ofpacts_len);
3847 op = ofoperation_create(group, rule, type, 0);
3849 if (fm->modify_cookie && fm->new_cookie != htonll(UINT64_MAX)) {
3850 ofproto_rule_change_cookie(ofproto, rule, fm->new_cookie);
3852 if (type == OFOPERATION_REPLACE) {
3853 ovs_mutex_lock(&rule->timeout_mutex);
3854 rule->idle_timeout = fm->idle_timeout;
3855 rule->hard_timeout = fm->hard_timeout;
3856 ovs_mutex_unlock(&rule->timeout_mutex);
3858 rule->flags = fm->flags & OFPUTIL_FF_STATE;
3859 if (fm->idle_timeout || fm->hard_timeout) {
3860 if (!rule->eviction_group) {
3861 eviction_group_add_rule(rule);
3864 eviction_group_remove_rule(rule);
3868 reset_counters = (fm->flags & OFPUTIL_FF_RESET_COUNTS) != 0;
3869 if (actions_changed || reset_counters) {
3870 struct rule_actions *new_actions;
3872 op->actions = rule->actions;
3873 new_actions = rule_actions_create(fm->ofpacts, fm->ofpacts_len);
3875 ovs_rwlock_wrlock(&rule->rwlock);
3876 rule->actions = new_actions;
3877 ovs_rwlock_unlock(&rule->rwlock);
3879 rule->ofproto->ofproto_class->rule_modify_actions(rule,
3882 ofoperation_complete(op, 0);
3885 ofopgroup_submit(group);
3891 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
3892 struct ofputil_flow_mod *fm, const struct ofp_header *request)
3894 if (fm->cookie_mask != htonll(0) || fm->new_cookie == htonll(UINT64_MAX)) {
3897 return add_flow(ofproto, ofconn, fm, request);
3900 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
3903 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3906 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3907 struct ofputil_flow_mod *fm,
3908 const struct ofp_header *request)
3910 struct rule_criteria criteria;
3911 struct rule_collection rules;
3914 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0,
3915 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY);
3916 error = collect_rules_loose(ofproto, &criteria, &rules);
3917 rule_criteria_destroy(&criteria);
3920 error = (rules.n > 0
3921 ? modify_flows__(ofproto, ofconn, fm, request, &rules)
3922 : modify_flows_add(ofproto, ofconn, fm, request));
3925 rule_collection_destroy(&rules);
3930 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3933 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3936 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3937 struct ofputil_flow_mod *fm,
3938 const struct ofp_header *request)
3940 struct rule_criteria criteria;
3941 struct rule_collection rules;
3944 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
3945 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY);
3946 error = collect_rules_strict(ofproto, &criteria, &rules);
3947 rule_criteria_destroy(&criteria);
3951 error = modify_flows_add(ofproto, ofconn, fm, request);
3952 } else if (rules.n == 1) {
3953 error = modify_flows__(ofproto, ofconn, fm, request, &rules);
3957 rule_collection_destroy(&rules);
3962 /* OFPFC_DELETE implementation. */
3965 delete_flow__(struct rule *rule, struct ofopgroup *group,
3966 enum ofp_flow_removed_reason reason)
3968 struct ofproto *ofproto = rule->ofproto;
3970 ofproto_rule_send_removed(rule, reason);
3972 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
3973 oftable_remove_rule(rule);
3974 ofproto->ofproto_class->rule_delete(rule);
3977 /* Deletes the rules listed in 'rules'.
3979 * Returns 0 on success, otherwise an OpenFlow error code. */
3981 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3982 const struct ofp_header *request,
3983 const struct rule_collection *rules,
3984 enum ofp_flow_removed_reason reason)
3986 struct ofopgroup *group;
3989 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
3990 for (i = 0; i < rules->n; i++) {
3991 struct rule *rule = rules->rules[i];
3992 ovs_rwlock_wrlock(&rule->rwlock);
3993 delete_flow__(rule, group, reason);
3995 ofopgroup_submit(group);
4000 /* Implements OFPFC_DELETE. */
4002 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
4003 const struct ofputil_flow_mod *fm,
4004 const struct ofp_header *request)
4006 struct rule_criteria criteria;
4007 struct rule_collection rules;
4010 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0,
4011 fm->cookie, fm->cookie_mask,
4012 fm->out_port, fm->out_group);
4013 error = collect_rules_loose(ofproto, &criteria, &rules);
4014 rule_criteria_destroy(&criteria);
4016 if (!error && rules.n > 0) {
4017 error = delete_flows__(ofproto, ofconn, request, &rules, OFPRR_DELETE);
4019 rule_collection_destroy(&rules);
4024 /* Implements OFPFC_DELETE_STRICT. */
4026 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
4027 const struct ofputil_flow_mod *fm,
4028 const struct ofp_header *request)
4030 struct rule_criteria criteria;
4031 struct rule_collection rules;
4034 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
4035 fm->cookie, fm->cookie_mask,
4036 fm->out_port, fm->out_group);
4037 error = collect_rules_strict(ofproto, &criteria, &rules);
4038 rule_criteria_destroy(&criteria);
4040 if (!error && rules.n > 0) {
4041 error = delete_flows__(ofproto, ofconn, request, &rules, OFPRR_DELETE);
4043 rule_collection_destroy(&rules);
4049 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
4051 struct ofputil_flow_removed fr;
4053 if (ofproto_rule_is_hidden(rule) ||
4054 !(rule->flags & OFPUTIL_FF_SEND_FLOW_REM)) {
4058 minimatch_expand(&rule->cr.match, &fr.match);
4059 fr.priority = rule->cr.priority;
4060 fr.cookie = rule->flow_cookie;
4062 fr.table_id = rule->table_id;
4063 calc_duration(rule->created, time_msec(),
4064 &fr.duration_sec, &fr.duration_nsec);
4065 ovs_mutex_lock(&rule->timeout_mutex);
4066 fr.idle_timeout = rule->idle_timeout;
4067 fr.hard_timeout = rule->hard_timeout;
4068 ovs_mutex_unlock(&rule->timeout_mutex);
4069 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
4072 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
4075 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
4076 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
4079 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
4082 * ofproto implementation ->run() functions should use this function to expire
4083 * OpenFlow flows. */
4085 ofproto_rule_expire(struct rule *rule, uint8_t reason)
4087 struct ofproto *ofproto = rule->ofproto;
4088 struct classifier *cls = &ofproto->tables[rule->table_id].cls;
4090 ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT
4091 || reason == OFPRR_DELETE || reason == OFPRR_GROUP_DELETE);
4092 ofproto_rule_send_removed(rule, reason);
4094 ovs_rwlock_wrlock(&cls->rwlock);
4095 ofproto_rule_delete(ofproto, cls, rule);
4096 ovs_rwlock_unlock(&cls->rwlock);
4099 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
4100 * means "infinite". */
4102 reduce_timeout(uint16_t max, uint16_t *timeout)
4104 if (max && (!*timeout || *timeout > max)) {
4109 /* If 'idle_timeout' is nonzero, and 'rule' has no idle timeout or an idle
4110 * timeout greater than 'idle_timeout', lowers 'rule''s idle timeout to
4111 * 'idle_timeout' seconds. Similarly for 'hard_timeout'.
4113 * Suitable for implementing OFPACT_FIN_TIMEOUT. */
4115 ofproto_rule_reduce_timeouts(struct rule *rule,
4116 uint16_t idle_timeout, uint16_t hard_timeout)
4117 OVS_EXCLUDED(ofproto_mutex, rule->timeout_mutex)
4119 if (!idle_timeout && !hard_timeout) {
4123 ovs_mutex_lock(&ofproto_mutex);
4124 if (list_is_empty(&rule->expirable)) {
4125 list_insert(&rule->ofproto->expirable, &rule->expirable);
4127 ovs_mutex_unlock(&ofproto_mutex);
4129 ovs_mutex_lock(&rule->timeout_mutex);
4130 reduce_timeout(idle_timeout, &rule->idle_timeout);
4131 reduce_timeout(hard_timeout, &rule->hard_timeout);
4132 ovs_mutex_unlock(&rule->timeout_mutex);
4136 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4138 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4139 struct ofputil_flow_mod fm;
4140 uint64_t ofpacts_stub[1024 / 8];
4141 struct ofpbuf ofpacts;
4145 error = reject_slave_controller(ofconn);
4150 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
4151 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
4154 error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
4157 goto exit_free_ofpacts;
4160 /* Record the operation for logging a summary report. */
4161 switch (fm.command) {
4167 case OFPFC_MODIFY_STRICT:
4168 ofproto->n_modify++;
4172 case OFPFC_DELETE_STRICT:
4173 ofproto->n_delete++;
4178 if (ofproto->next_op_report == LLONG_MAX) {
4179 ofproto->first_op = now;
4180 ofproto->next_op_report = MAX(now + 10 * 1000,
4181 ofproto->op_backoff);
4182 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
4184 ofproto->last_op = now;
4187 ofpbuf_uninit(&ofpacts);
4193 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
4194 struct ofputil_flow_mod *fm, const struct ofp_header *oh)
4198 if (ofproto->n_pending < 50) {
4199 switch (fm->command) {
4201 error = add_flow(ofproto, ofconn, fm, oh);
4205 error = modify_flows_loose(ofproto, ofconn, fm, oh);
4208 case OFPFC_MODIFY_STRICT:
4209 error = modify_flow_strict(ofproto, ofconn, fm, oh);
4213 error = delete_flows_loose(ofproto, ofconn, fm, oh);
4216 case OFPFC_DELETE_STRICT:
4217 error = delete_flow_strict(ofproto, ofconn, fm, oh);
4221 if (fm->command > 0xff) {
4222 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
4223 "flow_mod_table_id extension is not enabled",
4226 error = OFPERR_OFPFMFC_BAD_COMMAND;
4230 ovs_assert(!list_is_empty(&ofproto->pending));
4231 error = OFPROTO_POSTPONE;
4234 run_rule_executes(ofproto);
4239 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4241 struct ofputil_role_request request;
4242 struct ofputil_role_request reply;
4246 error = ofputil_decode_role_message(oh, &request);
4251 if (request.role != OFPCR12_ROLE_NOCHANGE) {
4252 if (ofconn_get_role(ofconn) != request.role
4253 && ofconn_has_pending_opgroups(ofconn)) {
4254 return OFPROTO_POSTPONE;
4257 if (request.have_generation_id
4258 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
4259 return OFPERR_OFPRRFC_STALE;
4262 ofconn_set_role(ofconn, request.role);
4265 reply.role = ofconn_get_role(ofconn);
4266 reply.have_generation_id = ofconn_get_master_election_id(
4267 ofconn, &reply.generation_id);
4268 buf = ofputil_encode_role_reply(oh, &reply);
4269 ofconn_send_reply(ofconn, buf);
4275 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
4276 const struct ofp_header *oh)
4278 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
4279 enum ofputil_protocol cur, next;
4281 cur = ofconn_get_protocol(ofconn);
4282 next = ofputil_protocol_set_tid(cur, msg->set != 0);
4283 ofconn_set_protocol(ofconn, next);
4289 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4291 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
4292 enum ofputil_protocol cur, next;
4293 enum ofputil_protocol next_base;
4295 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
4297 return OFPERR_OFPBRC_EPERM;
4300 cur = ofconn_get_protocol(ofconn);
4301 next = ofputil_protocol_set_base(cur, next_base);
4302 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
4303 /* Avoid sending async messages in surprising protocol. */
4304 return OFPROTO_POSTPONE;
4307 ofconn_set_protocol(ofconn, next);
4312 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
4313 const struct ofp_header *oh)
4315 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
4318 format = ntohl(msg->format);
4319 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
4320 return OFPERR_OFPBRC_EPERM;
4323 if (format != ofconn_get_packet_in_format(ofconn)
4324 && ofconn_has_pending_opgroups(ofconn)) {
4325 /* Avoid sending async message in surprsing packet in format. */
4326 return OFPROTO_POSTPONE;
4329 ofconn_set_packet_in_format(ofconn, format);
4334 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
4336 const struct nx_async_config *msg = ofpmsg_body(oh);
4337 uint32_t master[OAM_N_TYPES];
4338 uint32_t slave[OAM_N_TYPES];
4340 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
4341 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
4342 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
4344 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
4345 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
4346 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
4348 ofconn_set_async_config(ofconn, master, slave);
4349 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
4350 !ofconn_get_miss_send_len(ofconn)) {
4351 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
4358 handle_nxt_get_async_request(struct ofconn *ofconn, const struct ofp_header *oh)
4361 uint32_t master[OAM_N_TYPES];
4362 uint32_t slave[OAM_N_TYPES];
4363 struct nx_async_config *msg;
4365 ofconn_get_async_config(ofconn, master, slave);
4366 buf = ofpraw_alloc_reply(OFPRAW_OFPT13_GET_ASYNC_REPLY, oh, 0);
4367 msg = ofpbuf_put_zeros(buf, sizeof *msg);
4369 msg->packet_in_mask[0] = htonl(master[OAM_PACKET_IN]);
4370 msg->port_status_mask[0] = htonl(master[OAM_PORT_STATUS]);
4371 msg->flow_removed_mask[0] = htonl(master[OAM_FLOW_REMOVED]);
4373 msg->packet_in_mask[1] = htonl(slave[OAM_PACKET_IN]);
4374 msg->port_status_mask[1] = htonl(slave[OAM_PORT_STATUS]);
4375 msg->flow_removed_mask[1] = htonl(slave[OAM_FLOW_REMOVED]);
4377 ofconn_send_reply(ofconn, buf);
4383 handle_nxt_set_controller_id(struct ofconn *ofconn,
4384 const struct ofp_header *oh)
4386 const struct nx_controller_id *nci = ofpmsg_body(oh);
4388 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
4389 return OFPERR_NXBRC_MUST_BE_ZERO;
4392 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
4397 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4401 if (ofconn_has_pending_opgroups(ofconn)) {
4402 return OFPROTO_POSTPONE;
4405 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
4406 ? OFPRAW_OFPT10_BARRIER_REPLY
4407 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
4408 ofconn_send_reply(ofconn, buf);
4413 ofproto_compose_flow_refresh_update(const struct rule *rule,
4414 enum nx_flow_monitor_flags flags,
4417 struct ofoperation *op = rule->pending;
4418 const struct rule_actions *actions;
4419 struct ofputil_flow_update fu;
4422 if (op && op->type == OFOPERATION_ADD) {
4423 /* We'll report the final flow when the operation completes. Reporting
4424 * it now would cause a duplicate report later. */
4428 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
4429 ? NXFME_ADDED : NXFME_MODIFIED);
4431 ovs_mutex_lock(&rule->timeout_mutex);
4432 fu.idle_timeout = rule->idle_timeout;
4433 fu.hard_timeout = rule->hard_timeout;
4434 ovs_mutex_unlock(&rule->timeout_mutex);
4435 fu.table_id = rule->table_id;
4436 fu.cookie = rule->flow_cookie;
4437 minimatch_expand(&rule->cr.match, &match);
4439 fu.priority = rule->cr.priority;
4441 if (!(flags & NXFMF_ACTIONS)) {
4444 actions = rule->actions;
4446 /* An operation is in progress. Use the previous version of the flow's
4447 * actions, so that when the operation commits we report the change. */
4449 case OFOPERATION_ADD:
4452 case OFOPERATION_MODIFY:
4453 case OFOPERATION_REPLACE:
4454 actions = op->actions ? op->actions : rule->actions;
4457 case OFOPERATION_DELETE:
4458 actions = rule->actions;
4465 fu.ofpacts = actions ? actions->ofpacts : NULL;
4466 fu.ofpacts_len = actions ? actions->ofpacts_len : 0;
4468 if (list_is_empty(msgs)) {
4469 ofputil_start_flow_update(msgs);
4471 ofputil_append_flow_update(&fu, msgs);
4475 ofmonitor_compose_refresh_updates(struct rule_collection *rules,
4480 for (i = 0; i < rules->n; i++) {
4481 struct rule *rule = rules->rules[i];
4482 enum nx_flow_monitor_flags flags = rule->monitor_flags;
4483 rule->monitor_flags = 0;
4485 ofproto_compose_flow_refresh_update(rule, flags, msgs);
4490 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
4491 struct rule *rule, uint64_t seqno,
4492 struct rule_collection *rules)
4494 enum nx_flow_monitor_flags update;
4496 if (ofproto_rule_is_hidden(rule)) {
4501 ? ofoperation_has_out_port(rule->pending, m->out_port)
4502 : ofproto_rule_has_out_port(rule, m->out_port))) {
4507 if (rule->add_seqno > seqno) {
4508 update = NXFMF_ADD | NXFMF_MODIFY;
4509 } else if (rule->modify_seqno > seqno) {
4510 update = NXFMF_MODIFY;
4515 if (!(m->flags & update)) {
4519 update = NXFMF_INITIAL;
4522 if (!rule->monitor_flags) {
4523 rule_collection_add(rules, rule);
4525 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
4529 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
4531 struct rule_collection *rules)
4533 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
4534 const struct ofoperation *op;
4535 const struct oftable *table;
4536 struct cls_rule target;
4538 cls_rule_init_from_minimatch(&target, &m->match, 0);
4539 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
4540 struct cls_cursor cursor;
4543 ovs_rwlock_rdlock(&table->cls.rwlock);
4544 cls_cursor_init(&cursor, &table->cls, &target);
4545 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4546 ovs_assert(!rule->pending); /* XXX */
4547 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4549 ovs_rwlock_unlock(&table->cls.rwlock);
4552 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
4553 struct rule *rule = op->rule;
4555 if (((m->table_id == 0xff
4556 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
4557 : m->table_id == rule->table_id))
4558 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
4559 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4562 cls_rule_destroy(&target);
4566 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
4567 struct rule_collection *rules)
4569 if (m->flags & NXFMF_INITIAL) {
4570 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
4575 ofmonitor_collect_resume_rules(struct ofmonitor *m,
4576 uint64_t seqno, struct rule_collection *rules)
4578 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
4582 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
4584 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4585 struct ofmonitor **monitors;
4586 size_t n_monitors, allocated_monitors;
4587 struct rule_collection rules;
4588 struct list replies;
4594 ofpbuf_use_const(&b, oh, ntohs(oh->length));
4596 n_monitors = allocated_monitors = 0;
4598 struct ofputil_flow_monitor_request request;
4599 struct ofmonitor *m;
4602 retval = ofputil_decode_flow_monitor_request(&request, &b);
4603 if (retval == EOF) {
4605 } else if (retval) {
4610 if (request.table_id != 0xff
4611 && request.table_id >= ofproto->n_tables) {
4612 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4616 error = ofmonitor_create(&request, ofconn, &m);
4621 if (n_monitors >= allocated_monitors) {
4622 monitors = x2nrealloc(monitors, &allocated_monitors,
4625 monitors[n_monitors++] = m;
4628 rule_collection_init(&rules);
4629 for (i = 0; i < n_monitors; i++) {
4630 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
4633 ofpmp_init(&replies, oh);
4634 ofmonitor_compose_refresh_updates(&rules, &replies);
4635 rule_collection_destroy(&rules);
4637 ofconn_send_replies(ofconn, &replies);
4644 for (i = 0; i < n_monitors; i++) {
4645 ofmonitor_destroy(monitors[i]);
4652 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
4654 struct ofmonitor *m;
4657 id = ofputil_decode_flow_monitor_cancel(oh);
4658 m = ofmonitor_lookup(ofconn, id);
4660 return OFPERR_NXBRC_FM_BAD_ID;
4663 ofmonitor_destroy(m);
4667 /* Meters implementation.
4669 * Meter table entry, indexed by the OpenFlow meter_id.
4670 * These are always dynamically allocated to allocate enough space for
4672 * 'created' is used to compute the duration for meter stats.
4673 * 'list rules' is needed so that we can delete the dependent rules when the
4674 * meter table entry is deleted.
4675 * 'provider_meter_id' is for the provider's private use.
4678 long long int created; /* Time created. */
4679 struct list rules; /* List of "struct rule_dpif"s. */
4680 ofproto_meter_id provider_meter_id;
4681 uint16_t flags; /* Meter flags. */
4682 uint16_t n_bands; /* Number of meter bands. */
4683 struct ofputil_meter_band *bands;
4687 * This is used in instruction validation at flow set-up time,
4688 * as flows may not use non-existing meters.
4689 * This is also used by ofproto-providers to translate OpenFlow meter_ids
4690 * in METER instructions to the corresponding provider meter IDs.
4691 * Return value of UINT32_MAX signifies an invalid meter.
4694 ofproto_get_provider_meter_id(const struct ofproto * ofproto,
4695 uint32_t of_meter_id)
4697 if (of_meter_id && of_meter_id <= ofproto->meter_features.max_meters) {
4698 const struct meter *meter = ofproto->meters[of_meter_id];
4700 return meter->provider_meter_id.uint32;
4707 meter_update(struct meter *meter, const struct ofputil_meter_config *config)
4711 meter->flags = config->flags;
4712 meter->n_bands = config->n_bands;
4713 meter->bands = xmemdup(config->bands,
4714 config->n_bands * sizeof *meter->bands);
4717 static struct meter *
4718 meter_create(const struct ofputil_meter_config *config,
4719 ofproto_meter_id provider_meter_id)
4721 struct meter *meter;
4723 meter = xzalloc(sizeof *meter);
4724 meter->provider_meter_id = provider_meter_id;
4725 meter->created = time_msec();
4726 list_init(&meter->rules);
4728 meter_update(meter, config);
4734 meter_delete(struct ofproto *ofproto, uint32_t first, uint32_t last)
4737 for (mid = first; mid <= last; ++mid) {
4738 struct meter *meter = ofproto->meters[mid];
4740 ofproto->meters[mid] = NULL;
4741 ofproto->ofproto_class->meter_del(ofproto,
4742 meter->provider_meter_id);
4750 handle_add_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
4752 ofproto_meter_id provider_meter_id = { UINT32_MAX };
4753 struct meter **meterp = &ofproto->meters[mm->meter.meter_id];
4757 return OFPERR_OFPMMFC_METER_EXISTS;
4760 error = ofproto->ofproto_class->meter_set(ofproto, &provider_meter_id,
4763 ovs_assert(provider_meter_id.uint32 != UINT32_MAX);
4764 *meterp = meter_create(&mm->meter, provider_meter_id);
4770 handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
4772 struct meter *meter = ofproto->meters[mm->meter.meter_id];
4776 return OFPERR_OFPMMFC_UNKNOWN_METER;
4779 error = ofproto->ofproto_class->meter_set(ofproto,
4780 &meter->provider_meter_id,
4782 ovs_assert(meter->provider_meter_id.uint32 != UINT32_MAX);
4784 meter_update(meter, &mm->meter);
4790 handle_delete_meter(struct ofconn *ofconn, const struct ofp_header *oh,
4791 struct ofputil_meter_mod *mm)
4793 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4794 uint32_t meter_id = mm->meter.meter_id;
4795 struct rule_collection rules;
4796 enum ofperr error = 0;
4797 uint32_t first, last;
4799 if (meter_id == OFPM13_ALL) {
4801 last = ofproto->meter_features.max_meters;
4803 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
4806 first = last = meter_id;
4809 /* First delete the rules that use this meter. If any of those rules are
4810 * currently being modified, postpone the whole operation until later. */
4811 rule_collection_init(&rules);
4812 for (meter_id = first; meter_id <= last; ++meter_id) {
4813 struct meter *meter = ofproto->meters[meter_id];
4814 if (meter && !list_is_empty(&meter->rules)) {
4817 LIST_FOR_EACH (rule, meter_list_node, &meter->rules) {
4818 if (rule->pending) {
4819 error = OFPROTO_POSTPONE;
4822 rule_collection_add(&rules, rule);
4827 delete_flows__(ofproto, ofconn, oh, &rules, OFPRR_METER_DELETE);
4830 /* Delete the meters. */
4831 meter_delete(ofproto, first, last);
4834 rule_collection_destroy(&rules);
4840 handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4842 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4843 struct ofputil_meter_mod mm;
4844 uint64_t bands_stub[256 / 8];
4845 struct ofpbuf bands;
4849 error = reject_slave_controller(ofconn);
4854 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
4856 error = ofputil_decode_meter_mod(oh, &mm, &bands);
4858 goto exit_free_bands;
4861 meter_id = mm.meter.meter_id;
4863 if (mm.command != OFPMC13_DELETE) {
4864 /* Fails also when meters are not implemented by the provider. */
4865 if (meter_id == 0 || meter_id > OFPM13_MAX) {
4866 error = OFPERR_OFPMMFC_INVALID_METER;
4867 goto exit_free_bands;
4868 } else if (meter_id > ofproto->meter_features.max_meters) {
4869 error = OFPERR_OFPMMFC_OUT_OF_METERS;
4870 goto exit_free_bands;
4872 if (mm.meter.n_bands > ofproto->meter_features.max_bands) {
4873 error = OFPERR_OFPMMFC_OUT_OF_BANDS;
4874 goto exit_free_bands;
4878 switch (mm.command) {
4880 error = handle_add_meter(ofproto, &mm);
4883 case OFPMC13_MODIFY:
4884 error = handle_modify_meter(ofproto, &mm);
4887 case OFPMC13_DELETE:
4888 error = handle_delete_meter(ofconn, oh, &mm);
4892 error = OFPERR_OFPMMFC_BAD_COMMAND;
4897 ofpbuf_uninit(&bands);
4902 handle_meter_features_request(struct ofconn *ofconn,
4903 const struct ofp_header *request)
4905 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4906 struct ofputil_meter_features features;
4909 if (ofproto->ofproto_class->meter_get_features) {
4910 ofproto->ofproto_class->meter_get_features(ofproto, &features);
4912 memset(&features, 0, sizeof features);
4914 b = ofputil_encode_meter_features_reply(&features, request);
4916 ofconn_send_reply(ofconn, b);
4921 handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request,
4924 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4925 struct list replies;
4926 uint64_t bands_stub[256 / 8];
4927 struct ofpbuf bands;
4928 uint32_t meter_id, first, last;
4930 ofputil_decode_meter_request(request, &meter_id);
4932 if (meter_id == OFPM13_ALL) {
4934 last = ofproto->meter_features.max_meters;
4936 if (!meter_id || meter_id > ofproto->meter_features.max_meters ||
4937 !ofproto->meters[meter_id]) {
4938 return OFPERR_OFPMMFC_UNKNOWN_METER;
4940 first = last = meter_id;
4943 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
4944 ofpmp_init(&replies, request);
4946 for (meter_id = first; meter_id <= last; ++meter_id) {
4947 struct meter *meter = ofproto->meters[meter_id];
4949 continue; /* Skip non-existing meters. */
4951 if (type == OFPTYPE_METER_STATS_REQUEST) {
4952 struct ofputil_meter_stats stats;
4954 stats.meter_id = meter_id;
4956 /* Provider sets the packet and byte counts, we do the rest. */
4957 stats.flow_count = list_size(&meter->rules);
4958 calc_duration(meter->created, time_msec(),
4959 &stats.duration_sec, &stats.duration_nsec);
4960 stats.n_bands = meter->n_bands;
4961 ofpbuf_clear(&bands);
4963 = ofpbuf_put_uninit(&bands,
4964 meter->n_bands * sizeof *stats.bands);
4966 if (!ofproto->ofproto_class->meter_get(ofproto,
4967 meter->provider_meter_id,
4969 ofputil_append_meter_stats(&replies, &stats);
4971 } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */
4972 struct ofputil_meter_config config;
4974 config.meter_id = meter_id;
4975 config.flags = meter->flags;
4976 config.n_bands = meter->n_bands;
4977 config.bands = meter->bands;
4978 ofputil_append_meter_config(&replies, &config);
4982 ofconn_send_replies(ofconn, &replies);
4983 ofpbuf_uninit(&bands);
4988 ofproto_group_lookup(const struct ofproto *ofproto, uint32_t group_id,
4989 struct ofgroup **group)
4990 OVS_TRY_RDLOCK(true, (*group)->rwlock)
4992 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
4993 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
4994 hash_int(group_id, 0), &ofproto->groups) {
4995 if ((*group)->group_id == group_id) {
4996 ovs_rwlock_rdlock(&(*group)->rwlock);
4997 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5001 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5006 ofproto_group_release(struct ofgroup *group)
5007 OVS_RELEASES(group->rwlock)
5009 ovs_rwlock_unlock(&group->rwlock);
5013 ofproto_group_write_lookup(const struct ofproto *ofproto, uint32_t group_id,
5014 struct ofgroup **group)
5015 OVS_TRY_WRLOCK(true, ofproto->groups_rwlock)
5016 OVS_TRY_WRLOCK(true, (*group)->rwlock)
5018 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5019 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
5020 hash_int(group_id, 0), &ofproto->groups) {
5021 if ((*group)->group_id == group_id) {
5022 ovs_rwlock_wrlock(&(*group)->rwlock);
5026 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5031 ofproto_group_exists(const struct ofproto *ofproto, uint32_t group_id)
5032 OVS_REQ_RDLOCK(ofproto->groups_rwlock)
5034 struct ofgroup *grp;
5036 HMAP_FOR_EACH_IN_BUCKET (grp, hmap_node,
5037 hash_int(group_id, 0), &ofproto->groups) {
5038 if (grp->group_id == group_id) {
5046 append_group_stats(struct ofgroup *group, struct list *replies)
5047 OVS_REQ_RDLOCK(group->rwlock)
5049 struct ofputil_group_stats ogs;
5050 struct ofproto *ofproto = group->ofproto;
5051 long long int now = time_msec();
5054 ogs.bucket_stats = xmalloc(group->n_buckets * sizeof *ogs.bucket_stats);
5056 error = (ofproto->ofproto_class->group_get_stats
5057 ? ofproto->ofproto_class->group_get_stats(group, &ogs)
5060 ogs.ref_count = UINT32_MAX;
5061 ogs.packet_count = UINT64_MAX;
5062 ogs.byte_count = UINT64_MAX;
5063 ogs.n_buckets = group->n_buckets;
5064 memset(ogs.bucket_stats, 0xff,
5065 ogs.n_buckets * sizeof *ogs.bucket_stats);
5068 ogs.group_id = group->group_id;
5069 calc_duration(group->created, now, &ogs.duration_sec, &ogs.duration_nsec);
5071 ofputil_append_group_stats(replies, &ogs);
5073 free(ogs.bucket_stats);
5077 handle_group_stats_request(struct ofconn *ofconn,
5078 const struct ofp_header *request)
5080 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5081 struct list replies;
5083 struct ofgroup *group;
5086 error = ofputil_decode_group_stats_request(request, &group_id);
5091 ofpmp_init(&replies, request);
5093 if (group_id == OFPG_ALL) {
5094 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5095 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
5096 ovs_rwlock_rdlock(&group->rwlock);
5097 append_group_stats(group, &replies);
5098 ovs_rwlock_unlock(&group->rwlock);
5100 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5102 if (ofproto_group_lookup(ofproto, group_id, &group)) {
5103 append_group_stats(group, &replies);
5104 ofproto_group_release(group);
5108 ofconn_send_replies(ofconn, &replies);
5114 handle_group_desc_stats_request(struct ofconn *ofconn,
5115 const struct ofp_header *request)
5117 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5118 struct list replies;
5119 struct ofputil_group_desc gds;
5120 struct ofgroup *group;
5122 ofpmp_init(&replies, request);
5124 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5125 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
5126 gds.group_id = group->group_id;
5127 gds.type = group->type;
5128 ofputil_append_group_desc_reply(&gds, &group->buckets, &replies);
5130 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5132 ofconn_send_replies(ofconn, &replies);
5138 handle_group_features_stats_request(struct ofconn *ofconn,
5139 const struct ofp_header *request)
5141 struct ofproto *p = ofconn_get_ofproto(ofconn);
5144 msg = ofputil_encode_group_features_reply(&p->ogf, request);
5146 ofconn_send_reply(ofconn, msg);
5152 /* Implements OFPGC11_ADD
5153 * in which no matching flow already exists in the flow table.
5155 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
5156 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
5157 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
5158 * initiated now but may be retried later.
5160 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
5161 * ownership remains with the caller.
5163 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
5166 add_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
5168 struct ofgroup *ofgroup;
5171 if (gm->group_id > OFPG_MAX) {
5172 return OFPERR_OFPGMFC_INVALID_GROUP;
5174 if (gm->type > OFPGT11_FF) {
5175 return OFPERR_OFPGMFC_BAD_TYPE;
5178 /* Allocate new group and initialize it. */
5179 ofgroup = ofproto->ofproto_class->group_alloc();
5181 VLOG_WARN_RL(&rl, "%s: failed to create group", ofproto->name);
5182 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
5185 ovs_rwlock_init(&ofgroup->rwlock);
5186 ofgroup->ofproto = ofproto;
5187 ofgroup->group_id = gm->group_id;
5188 ofgroup->type = gm->type;
5189 ofgroup->created = ofgroup->modified = time_msec();
5191 list_move(&ofgroup->buckets, &gm->buckets);
5192 ofgroup->n_buckets = list_size(&ofgroup->buckets);
5194 /* Construct called BEFORE any locks are held. */
5195 error = ofproto->ofproto_class->group_construct(ofgroup);
5200 /* We wrlock as late as possible to minimize the time we jam any other
5201 * threads: No visible state changes before acquiring the lock. */
5202 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5204 if (ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
5205 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
5209 if (ofproto_group_exists(ofproto, gm->group_id)) {
5210 error = OFPERR_OFPGMFC_GROUP_EXISTS;
5215 /* Insert new group. */
5216 hmap_insert(&ofproto->groups, &ofgroup->hmap_node,
5217 hash_int(ofgroup->group_id, 0));
5218 ofproto->n_groups[ofgroup->type]++;
5220 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5225 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5226 ofproto->ofproto_class->group_destruct(ofgroup);
5228 ofputil_bucket_list_destroy(&ofgroup->buckets);
5229 ofproto->ofproto_class->group_dealloc(ofgroup);
5234 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
5237 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
5240 modify_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
5242 struct ofgroup *ofgroup;
5243 struct ofgroup *victim;
5246 if (gm->group_id > OFPG_MAX) {
5247 return OFPERR_OFPGMFC_INVALID_GROUP;
5250 if (gm->type > OFPGT11_FF) {
5251 return OFPERR_OFPGMFC_BAD_TYPE;
5254 victim = ofproto->ofproto_class->group_alloc();
5256 VLOG_WARN_RL(&rl, "%s: failed to allocate group", ofproto->name);
5257 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
5260 if (!ofproto_group_write_lookup(ofproto, gm->group_id, &ofgroup)) {
5261 error = OFPERR_OFPGMFC_UNKNOWN_GROUP;
5264 /* Both group's and its container's write locks held now.
5265 * Also, n_groups[] is protected by ofproto->groups_rwlock. */
5266 if (ofgroup->type != gm->type
5267 && ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
5268 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
5273 list_move(&victim->buckets, &ofgroup->buckets);
5275 ofgroup->type = gm->type;
5276 list_move(&ofgroup->buckets, &gm->buckets);
5277 ofgroup->n_buckets = list_size(&ofgroup->buckets);
5279 error = ofproto->ofproto_class->group_modify(ofgroup, victim);
5281 ofputil_bucket_list_destroy(&victim->buckets);
5282 ofproto->n_groups[victim->type]--;
5283 ofproto->n_groups[ofgroup->type]++;
5284 ofgroup->modified = time_msec();
5286 ofputil_bucket_list_destroy(&ofgroup->buckets);
5289 list_move(&ofgroup->buckets, &victim->buckets);
5293 ovs_rwlock_unlock(&ofgroup->rwlock);
5294 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5296 ofproto->ofproto_class->group_dealloc(victim);
5301 delete_group__(struct ofproto *ofproto, struct ofgroup *ofgroup)
5302 OVS_RELEASES(ofproto->groups_rwlock)
5304 /* Must wait until existing readers are done,
5305 * while holding the container's write lock at the same time. */
5306 ovs_rwlock_wrlock(&ofgroup->rwlock);
5307 hmap_remove(&ofproto->groups, &ofgroup->hmap_node);
5308 /* No-one can find this group any more. */
5309 ofproto->n_groups[ofgroup->type]--;
5310 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5312 ofproto->ofproto_class->group_destruct(ofgroup);
5313 ofputil_bucket_list_destroy(&ofgroup->buckets);
5314 ovs_rwlock_unlock(&ofgroup->rwlock);
5315 ovs_rwlock_destroy(&ofgroup->rwlock);
5316 ofproto->ofproto_class->group_dealloc(ofgroup);
5319 /* Implements OFPGC_DELETE. */
5321 delete_group(struct ofproto *ofproto, uint32_t group_id)
5323 struct ofgroup *ofgroup;
5325 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5326 if (group_id == OFPG_ALL) {
5328 struct hmap_node *node = hmap_first(&ofproto->groups);
5332 ofgroup = CONTAINER_OF(node, struct ofgroup, hmap_node);
5333 delete_group__(ofproto, ofgroup);
5334 /* Lock for each node separately, so that we will not jam the
5335 * other threads for too long time. */
5336 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5339 HMAP_FOR_EACH_IN_BUCKET (ofgroup, hmap_node,
5340 hash_int(group_id, 0), &ofproto->groups) {
5341 if (ofgroup->group_id == group_id) {
5342 delete_group__(ofproto, ofgroup);
5347 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5351 handle_group_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5353 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5354 struct ofputil_group_mod gm;
5357 error = reject_slave_controller(ofconn);
5362 error = ofputil_decode_group_mod(oh, &gm);
5367 switch (gm.command) {
5369 return add_group(ofproto, &gm);
5371 case OFPGC11_MODIFY:
5372 return modify_group(ofproto, &gm);
5374 case OFPGC11_DELETE:
5375 delete_group(ofproto, gm.group_id);
5379 if (gm.command > OFPGC11_DELETE) {
5380 VLOG_WARN_RL(&rl, "%s: Invalid group_mod command type %d",
5381 ofproto->name, gm.command);
5383 return OFPERR_OFPGMFC_BAD_COMMAND;
5388 handle_table_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5390 struct ofputil_table_mod tm;
5393 error = reject_slave_controller(ofconn);
5398 error = ofputil_decode_table_mod(oh, &tm);
5403 /* XXX Actual table mod support is not implemented yet. */
5408 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
5410 const struct ofp_header *oh = msg->data;
5414 error = ofptype_decode(&type, oh);
5420 /* OpenFlow requests. */
5421 case OFPTYPE_ECHO_REQUEST:
5422 return handle_echo_request(ofconn, oh);
5424 case OFPTYPE_FEATURES_REQUEST:
5425 return handle_features_request(ofconn, oh);
5427 case OFPTYPE_GET_CONFIG_REQUEST:
5428 return handle_get_config_request(ofconn, oh);
5430 case OFPTYPE_SET_CONFIG:
5431 return handle_set_config(ofconn, oh);
5433 case OFPTYPE_PACKET_OUT:
5434 return handle_packet_out(ofconn, oh);
5436 case OFPTYPE_PORT_MOD:
5437 return handle_port_mod(ofconn, oh);
5439 case OFPTYPE_FLOW_MOD:
5440 return handle_flow_mod(ofconn, oh);
5442 case OFPTYPE_GROUP_MOD:
5443 return handle_group_mod(ofconn, oh);
5445 case OFPTYPE_TABLE_MOD:
5446 return handle_table_mod(ofconn, oh);
5448 case OFPTYPE_METER_MOD:
5449 return handle_meter_mod(ofconn, oh);
5451 case OFPTYPE_BARRIER_REQUEST:
5452 return handle_barrier_request(ofconn, oh);
5454 case OFPTYPE_ROLE_REQUEST:
5455 return handle_role_request(ofconn, oh);
5457 /* OpenFlow replies. */
5458 case OFPTYPE_ECHO_REPLY:
5461 /* Nicira extension requests. */
5462 case OFPTYPE_FLOW_MOD_TABLE_ID:
5463 return handle_nxt_flow_mod_table_id(ofconn, oh);
5465 case OFPTYPE_SET_FLOW_FORMAT:
5466 return handle_nxt_set_flow_format(ofconn, oh);
5468 case OFPTYPE_SET_PACKET_IN_FORMAT:
5469 return handle_nxt_set_packet_in_format(ofconn, oh);
5471 case OFPTYPE_SET_CONTROLLER_ID:
5472 return handle_nxt_set_controller_id(ofconn, oh);
5474 case OFPTYPE_FLOW_AGE:
5475 /* Nothing to do. */
5478 case OFPTYPE_FLOW_MONITOR_CANCEL:
5479 return handle_flow_monitor_cancel(ofconn, oh);
5481 case OFPTYPE_SET_ASYNC_CONFIG:
5482 return handle_nxt_set_async_config(ofconn, oh);
5484 case OFPTYPE_GET_ASYNC_REQUEST:
5485 return handle_nxt_get_async_request(ofconn, oh);
5487 /* Statistics requests. */
5488 case OFPTYPE_DESC_STATS_REQUEST:
5489 return handle_desc_stats_request(ofconn, oh);
5491 case OFPTYPE_FLOW_STATS_REQUEST:
5492 return handle_flow_stats_request(ofconn, oh);
5494 case OFPTYPE_AGGREGATE_STATS_REQUEST:
5495 return handle_aggregate_stats_request(ofconn, oh);
5497 case OFPTYPE_TABLE_STATS_REQUEST:
5498 return handle_table_stats_request(ofconn, oh);
5500 case OFPTYPE_PORT_STATS_REQUEST:
5501 return handle_port_stats_request(ofconn, oh);
5503 case OFPTYPE_QUEUE_STATS_REQUEST:
5504 return handle_queue_stats_request(ofconn, oh);
5506 case OFPTYPE_PORT_DESC_STATS_REQUEST:
5507 return handle_port_desc_stats_request(ofconn, oh);
5509 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
5510 return handle_flow_monitor_request(ofconn, oh);
5512 case OFPTYPE_METER_STATS_REQUEST:
5513 case OFPTYPE_METER_CONFIG_STATS_REQUEST:
5514 return handle_meter_request(ofconn, oh, type);
5516 case OFPTYPE_METER_FEATURES_STATS_REQUEST:
5517 return handle_meter_features_request(ofconn, oh);
5519 case OFPTYPE_GROUP_STATS_REQUEST:
5520 return handle_group_stats_request(ofconn, oh);
5522 case OFPTYPE_GROUP_DESC_STATS_REQUEST:
5523 return handle_group_desc_stats_request(ofconn, oh);
5525 case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
5526 return handle_group_features_stats_request(ofconn, oh);
5528 /* FIXME: Change the following once they are implemented: */
5529 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
5530 case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
5531 return OFPERR_OFPBRC_BAD_TYPE;
5535 case OFPTYPE_FEATURES_REPLY:
5536 case OFPTYPE_GET_CONFIG_REPLY:
5537 case OFPTYPE_PACKET_IN:
5538 case OFPTYPE_FLOW_REMOVED:
5539 case OFPTYPE_PORT_STATUS:
5540 case OFPTYPE_BARRIER_REPLY:
5541 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
5542 case OFPTYPE_DESC_STATS_REPLY:
5543 case OFPTYPE_FLOW_STATS_REPLY:
5544 case OFPTYPE_QUEUE_STATS_REPLY:
5545 case OFPTYPE_PORT_STATS_REPLY:
5546 case OFPTYPE_TABLE_STATS_REPLY:
5547 case OFPTYPE_AGGREGATE_STATS_REPLY:
5548 case OFPTYPE_PORT_DESC_STATS_REPLY:
5549 case OFPTYPE_ROLE_REPLY:
5550 case OFPTYPE_FLOW_MONITOR_PAUSED:
5551 case OFPTYPE_FLOW_MONITOR_RESUMED:
5552 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
5553 case OFPTYPE_GET_ASYNC_REPLY:
5554 case OFPTYPE_GROUP_STATS_REPLY:
5555 case OFPTYPE_GROUP_DESC_STATS_REPLY:
5556 case OFPTYPE_GROUP_FEATURES_STATS_REPLY:
5557 case OFPTYPE_METER_STATS_REPLY:
5558 case OFPTYPE_METER_CONFIG_STATS_REPLY:
5559 case OFPTYPE_METER_FEATURES_STATS_REPLY:
5560 case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
5562 return OFPERR_OFPBRC_BAD_TYPE;
5567 handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
5569 int error = handle_openflow__(ofconn, ofp_msg);
5570 if (error && error != OFPROTO_POSTPONE) {
5571 ofconn_send_error(ofconn, ofp_msg->data, error);
5573 COVERAGE_INC(ofproto_recv_openflow);
5574 return error != OFPROTO_POSTPONE;
5577 /* Asynchronous operations. */
5579 /* Creates and returns a new ofopgroup that is not associated with any
5580 * OpenFlow connection.
5582 * The caller should add operations to the returned group with
5583 * ofoperation_create() and then submit it with ofopgroup_submit(). */
5584 static struct ofopgroup *
5585 ofopgroup_create_unattached(struct ofproto *ofproto)
5587 struct ofopgroup *group = xzalloc(sizeof *group);
5588 group->ofproto = ofproto;
5589 list_init(&group->ofproto_node);
5590 list_init(&group->ops);
5591 list_init(&group->ofconn_node);
5595 /* Creates and returns a new ofopgroup for 'ofproto'.
5597 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
5598 * connection. The 'request' and 'buffer_id' arguments are ignored.
5600 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
5601 * If the ofopgroup eventually fails, then the error reply will include
5602 * 'request'. If the ofopgroup eventually succeeds, then the packet with
5603 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
5605 * The caller should add operations to the returned group with
5606 * ofoperation_create() and then submit it with ofopgroup_submit(). */
5607 static struct ofopgroup *
5608 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
5609 const struct ofp_header *request, uint32_t buffer_id)
5611 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
5613 size_t request_len = ntohs(request->length);
5615 ovs_assert(ofconn_get_ofproto(ofconn) == ofproto);
5617 ofconn_add_opgroup(ofconn, &group->ofconn_node);
5618 group->ofconn = ofconn;
5619 group->request = xmemdup(request, MIN(request_len, 64));
5620 group->buffer_id = buffer_id;
5625 /* Submits 'group' for processing.
5627 * If 'group' contains no operations (e.g. none were ever added, or all of the
5628 * ones that were added completed synchronously), then it is destroyed
5629 * immediately. Otherwise it is added to the ofproto's list of pending
5632 ofopgroup_submit(struct ofopgroup *group)
5634 if (!group->n_running) {
5635 ofopgroup_complete(group);
5637 list_push_back(&group->ofproto->pending, &group->ofproto_node);
5638 group->ofproto->n_pending++;
5643 ofopgroup_complete(struct ofopgroup *group)
5645 struct ofproto *ofproto = group->ofproto;
5647 struct ofconn *abbrev_ofconn;
5648 ovs_be32 abbrev_xid;
5650 struct ofoperation *op, *next_op;
5653 ovs_assert(!group->n_running);
5656 LIST_FOR_EACH (op, group_node, &group->ops) {
5663 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
5664 LIST_FOR_EACH (op, group_node, &group->ops) {
5665 if (op->type != OFOPERATION_DELETE) {
5666 struct ofpbuf *packet;
5669 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
5672 struct rule_execute *re;
5676 ofproto_rule_ref(op->rule);
5678 re = xmalloc(sizeof *re);
5679 re->rule = op->rule;
5680 re->in_port = in_port;
5681 re->packet = packet;
5683 if (!guarded_list_push_back(&ofproto->rule_executes,
5684 &re->list_node, 1024)) {
5685 ofproto_rule_unref(op->rule);
5686 ofpbuf_delete(re->packet);
5695 if (!error && !list_is_empty(&group->ofconn_node)) {
5696 abbrev_ofconn = group->ofconn;
5697 abbrev_xid = group->request->xid;
5699 abbrev_ofconn = NULL;
5700 abbrev_xid = htonl(0);
5702 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
5703 struct rule *rule = op->rule;
5705 /* We generally want to report the change to active OpenFlow flow
5706 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
5708 - The operation failed.
5710 - The affected rule is not visible to controllers.
5712 - The operation's only effect was to update rule->modified. */
5714 || ofproto_rule_is_hidden(rule)
5715 || (op->type == OFOPERATION_MODIFY
5717 && rule->flow_cookie == op->flow_cookie))) {
5718 /* Check that we can just cast from ofoperation_type to
5719 * nx_flow_update_event. */
5720 enum nx_flow_update_event event_type;
5723 case OFOPERATION_ADD:
5724 case OFOPERATION_REPLACE:
5725 event_type = NXFME_ADDED;
5728 case OFOPERATION_DELETE:
5729 event_type = NXFME_DELETED;
5732 case OFOPERATION_MODIFY:
5733 event_type = NXFME_MODIFIED;
5740 ofmonitor_report(ofproto->connmgr, rule, event_type,
5741 op->reason, abbrev_ofconn, abbrev_xid);
5744 rule->pending = NULL;
5747 case OFOPERATION_ADD:
5751 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
5752 if (vid_mask == VLAN_VID_MASK) {
5753 if (ofproto->vlan_bitmap) {
5754 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
5755 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
5756 bitmap_set1(ofproto->vlan_bitmap, vid);
5757 ofproto->vlans_changed = true;
5760 ofproto->vlans_changed = true;
5764 ovs_rwlock_wrlock(&rule->rwlock);
5765 oftable_remove_rule(rule);
5766 ofproto_rule_unref(rule);
5770 case OFOPERATION_DELETE:
5771 ovs_assert(!op->error);
5772 ofproto_rule_unref(rule);
5776 case OFOPERATION_MODIFY:
5777 case OFOPERATION_REPLACE:
5779 long long int now = time_msec();
5781 rule->modified = now;
5782 if (op->type == OFOPERATION_REPLACE) {
5783 rule->created = rule->used = now;
5786 ofproto_rule_change_cookie(ofproto, rule, op->flow_cookie);
5787 ovs_mutex_lock(&rule->timeout_mutex);
5788 rule->idle_timeout = op->idle_timeout;
5789 rule->hard_timeout = op->hard_timeout;
5790 ovs_mutex_unlock(&rule->timeout_mutex);
5792 struct rule_actions *old_actions;
5794 ovs_rwlock_wrlock(&rule->rwlock);
5795 old_actions = rule->actions;
5796 rule->actions = op->actions;
5797 ovs_rwlock_unlock(&rule->rwlock);
5800 rule_actions_unref(old_actions);
5802 rule->flags = op->flags;
5810 ofoperation_destroy(op);
5813 ofmonitor_flush(ofproto->connmgr);
5815 if (!list_is_empty(&group->ofproto_node)) {
5816 ovs_assert(ofproto->n_pending > 0);
5817 ofproto->n_pending--;
5818 list_remove(&group->ofproto_node);
5820 if (!list_is_empty(&group->ofconn_node)) {
5821 list_remove(&group->ofconn_node);
5823 ofconn_send_error(group->ofconn, group->request, error);
5825 connmgr_retry(ofproto->connmgr);
5827 free(group->request);
5831 /* Initiates a new operation on 'rule', of the specified 'type', within
5832 * 'group'. Prior to calling, 'rule' must not have any pending operation.
5834 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
5835 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
5837 * Returns the newly created ofoperation (which is also available as
5838 * rule->pending). */
5839 static struct ofoperation *
5840 ofoperation_create(struct ofopgroup *group, struct rule *rule,
5841 enum ofoperation_type type,
5842 enum ofp_flow_removed_reason reason)
5844 struct ofproto *ofproto = group->ofproto;
5845 struct ofoperation *op;
5847 ovs_assert(!rule->pending);
5849 op = rule->pending = xzalloc(sizeof *op);
5851 list_push_back(&group->ops, &op->group_node);
5854 op->reason = reason;
5855 op->flow_cookie = rule->flow_cookie;
5856 ovs_mutex_lock(&rule->timeout_mutex);
5857 op->idle_timeout = rule->idle_timeout;
5858 op->hard_timeout = rule->hard_timeout;
5859 ovs_mutex_unlock(&rule->timeout_mutex);
5860 op->flags = rule->flags;
5864 if (type == OFOPERATION_DELETE) {
5865 hmap_insert(&ofproto->deletions, &op->hmap_node,
5866 cls_rule_hash(&rule->cr, rule->table_id));
5873 ofoperation_destroy(struct ofoperation *op)
5875 struct ofopgroup *group = op->group;
5878 op->rule->pending = NULL;
5880 if (op->type == OFOPERATION_DELETE) {
5881 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
5883 list_remove(&op->group_node);
5884 rule_actions_unref(op->actions);
5888 /* Indicates that 'op' completed with status 'error', which is either 0 to
5889 * indicate success or an OpenFlow error code on failure.
5891 * If 'error' is 0, indicating success, the operation will be committed
5892 * permanently to the flow table.
5894 * If 'error' is nonzero, then generally the operation will be rolled back:
5896 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
5897 * restores the original rule. The caller must have uninitialized any
5898 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
5899 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
5900 * and 7 for the new rule, calling its ->rule_dealloc() function.
5902 * - If 'op' is a "modify flow" operation, ofproto restores the original
5905 * - 'op' must not be a "delete flow" operation. Removing a rule is not
5906 * allowed to fail. It must always succeed.
5908 * Please see the large comment in ofproto/ofproto-provider.h titled
5909 * "Asynchronous Operation Support" for more information. */
5911 ofoperation_complete(struct ofoperation *op, enum ofperr error)
5913 struct ofopgroup *group = op->group;
5915 ovs_assert(op->rule->pending == op);
5916 ovs_assert(group->n_running > 0);
5917 ovs_assert(!error || op->type != OFOPERATION_DELETE);
5920 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
5921 ofopgroup_complete(group);
5926 pick_datapath_id(const struct ofproto *ofproto)
5928 const struct ofport *port;
5930 port = ofproto_get_port(ofproto, OFPP_LOCAL);
5932 uint8_t ea[ETH_ADDR_LEN];
5935 error = netdev_get_etheraddr(port->netdev, ea);
5937 return eth_addr_to_uint64(ea);
5939 VLOG_WARN("%s: could not get MAC address for %s (%s)",
5940 ofproto->name, netdev_get_name(port->netdev),
5941 ovs_strerror(error));
5943 return ofproto->fallback_dpid;
5947 pick_fallback_dpid(void)
5949 uint8_t ea[ETH_ADDR_LEN];
5950 eth_addr_nicira_random(ea);
5951 return eth_addr_to_uint64(ea);
5954 /* Table overflow policy. */
5956 /* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep'
5957 * to NULL if the table is not configured to evict rules or if the table
5958 * contains no evictable rules. (Rules with a readlock on their evict rwlock,
5959 * or with no timeouts are not evictable.) */
5961 choose_rule_to_evict(struct oftable *table, struct rule **rulep)
5963 struct eviction_group *evg;
5966 if (!table->eviction_fields) {
5970 /* In the common case, the outer and inner loops here will each be entered
5973 * - The inner loop normally "return"s in its first iteration. If the
5974 * eviction group has any evictable rules, then it always returns in
5977 * - The outer loop only iterates more than once if the largest eviction
5978 * group has no evictable rules.
5980 * - The outer loop can exit only if table's 'max_flows' is all filled up
5981 * by unevictable rules. */
5982 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
5985 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
5986 if (!ovs_rwlock_trywrlock(&rule->rwlock)) {
5996 /* Searches 'ofproto' for tables that have more flows than their configured
5997 * maximum and that have flow eviction enabled, and evicts as many flows as
5998 * necessary and currently feasible from them.
6000 * This triggers only when an OpenFlow table has N flows in it and then the
6001 * client configures a maximum number of flows less than N. */
6003 ofproto_evict(struct ofproto *ofproto)
6005 struct ofopgroup *group;
6006 struct oftable *table;
6008 group = ofopgroup_create_unattached(ofproto);
6009 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
6010 while (table->eviction_fields) {
6014 ovs_rwlock_rdlock(&table->cls.rwlock);
6015 n_rules = classifier_count(&table->cls);
6016 ovs_rwlock_unlock(&table->cls.rwlock);
6018 if (n_rules <= table->max_flows) {
6022 if (!choose_rule_to_evict(table, &rule)) {
6026 if (rule->pending) {
6027 ovs_rwlock_unlock(&rule->rwlock);
6031 ofoperation_create(group, rule,
6032 OFOPERATION_DELETE, OFPRR_EVICTION);
6033 oftable_remove_rule(rule);
6034 ofproto->ofproto_class->rule_delete(rule);
6037 ofopgroup_submit(group);
6040 /* Eviction groups. */
6042 /* Returns the priority to use for an eviction_group that contains 'n_rules'
6043 * rules. The priority contains low-order random bits to ensure that eviction
6044 * groups with the same number of rules are prioritized randomly. */
6046 eviction_group_priority(size_t n_rules)
6048 uint16_t size = MIN(UINT16_MAX, n_rules);
6049 return (size << 16) | random_uint16();
6052 /* Updates 'evg', an eviction_group within 'table', following a change that
6053 * adds or removes rules in 'evg'. */
6055 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
6057 heap_change(&table->eviction_groups_by_size, &evg->size_node,
6058 eviction_group_priority(heap_count(&evg->rules)));
6061 /* Destroys 'evg', an eviction_group within 'table':
6063 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
6064 * rules themselves, just removes them from the eviction group.)
6066 * - Removes 'evg' from 'table'.
6070 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
6072 while (!heap_is_empty(&evg->rules)) {
6075 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
6076 rule->eviction_group = NULL;
6078 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
6079 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
6080 heap_destroy(&evg->rules);
6084 /* Removes 'rule' from its eviction group, if any. */
6086 eviction_group_remove_rule(struct rule *rule)
6088 if (rule->eviction_group) {
6089 struct oftable *table = &rule->ofproto->tables[rule->table_id];
6090 struct eviction_group *evg = rule->eviction_group;
6092 rule->eviction_group = NULL;
6093 heap_remove(&evg->rules, &rule->evg_node);
6094 if (heap_is_empty(&evg->rules)) {
6095 eviction_group_destroy(table, evg);
6097 eviction_group_resized(table, evg);
6102 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
6103 * returns the hash value. */
6105 eviction_group_hash_rule(struct rule *rule)
6107 struct oftable *table = &rule->ofproto->tables[rule->table_id];
6108 const struct mf_subfield *sf;
6112 hash = table->eviction_group_id_basis;
6113 miniflow_expand(&rule->cr.match.flow, &flow);
6114 for (sf = table->eviction_fields;
6115 sf < &table->eviction_fields[table->n_eviction_fields];
6118 if (mf_are_prereqs_ok(sf->field, &flow)) {
6119 union mf_value value;
6121 mf_get_value(sf->field, &flow, &value);
6123 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
6125 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
6126 unsigned int start = sf->ofs + sf->n_bits;
6127 bitwise_zero(&value, sf->field->n_bytes, start,
6128 sf->field->n_bytes * 8 - start);
6130 hash = hash_bytes(&value, sf->field->n_bytes, hash);
6132 hash = hash_int(hash, 0);
6139 /* Returns an eviction group within 'table' with the given 'id', creating one
6141 static struct eviction_group *
6142 eviction_group_find(struct oftable *table, uint32_t id)
6144 struct eviction_group *evg;
6146 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
6150 evg = xmalloc(sizeof *evg);
6151 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
6152 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
6153 eviction_group_priority(0));
6154 heap_init(&evg->rules);
6159 /* Returns an eviction priority for 'rule'. The return value should be
6160 * interpreted so that higher priorities make a rule more attractive candidates
6163 rule_eviction_priority(struct rule *rule)
6165 long long int hard_expiration;
6166 long long int idle_expiration;
6167 long long int expiration;
6168 uint32_t expiration_offset;
6170 /* Calculate time of expiration. */
6171 ovs_mutex_lock(&rule->timeout_mutex);
6172 hard_expiration = (rule->hard_timeout
6173 ? rule->modified + rule->hard_timeout * 1000
6175 idle_expiration = (rule->idle_timeout
6176 ? rule->used + rule->idle_timeout * 1000
6178 expiration = MIN(hard_expiration, idle_expiration);
6179 ovs_mutex_unlock(&rule->timeout_mutex);
6180 if (expiration == LLONG_MAX) {
6184 /* Calculate the time of expiration as a number of (approximate) seconds
6185 * after program startup.
6187 * This should work OK for program runs that last UINT32_MAX seconds or
6188 * less. Therefore, please restart OVS at least once every 136 years. */
6189 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
6191 /* Invert the expiration offset because we're using a max-heap. */
6192 return UINT32_MAX - expiration_offset;
6195 /* Adds 'rule' to an appropriate eviction group for its oftable's
6196 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
6197 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
6200 * The caller must ensure that 'rule' is not already in an eviction group. */
6202 eviction_group_add_rule(struct rule *rule)
6204 struct ofproto *ofproto = rule->ofproto;
6205 struct oftable *table = &ofproto->tables[rule->table_id];
6208 ovs_mutex_lock(&rule->timeout_mutex);
6209 has_timeout = rule->hard_timeout || rule->idle_timeout;
6210 ovs_mutex_unlock(&rule->timeout_mutex);
6212 if (table->eviction_fields && has_timeout) {
6213 struct eviction_group *evg;
6215 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
6217 rule->eviction_group = evg;
6218 heap_insert(&evg->rules, &rule->evg_node,
6219 rule_eviction_priority(rule));
6220 eviction_group_resized(table, evg);
6226 /* Initializes 'table'. */
6228 oftable_init(struct oftable *table)
6230 memset(table, 0, sizeof *table);
6231 classifier_init(&table->cls);
6232 table->max_flows = UINT_MAX;
6235 /* Destroys 'table', including its classifier and eviction groups.
6237 * The caller is responsible for freeing 'table' itself. */
6239 oftable_destroy(struct oftable *table)
6241 ovs_rwlock_rdlock(&table->cls.rwlock);
6242 ovs_assert(classifier_is_empty(&table->cls));
6243 ovs_rwlock_unlock(&table->cls.rwlock);
6244 oftable_disable_eviction(table);
6245 classifier_destroy(&table->cls);
6249 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
6250 * string, then 'table' will use its default name.
6252 * This only affects the name exposed for a table exposed through the OpenFlow
6253 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
6255 oftable_set_name(struct oftable *table, const char *name)
6257 if (name && name[0]) {
6258 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
6259 if (!table->name || strncmp(name, table->name, len)) {
6261 table->name = xmemdup0(name, len);
6269 /* oftables support a choice of two policies when adding a rule would cause the
6270 * number of flows in the table to exceed the configured maximum number: either
6271 * they can refuse to add the new flow or they can evict some existing flow.
6272 * This function configures the former policy on 'table'. */
6274 oftable_disable_eviction(struct oftable *table)
6276 if (table->eviction_fields) {
6277 struct eviction_group *evg, *next;
6279 HMAP_FOR_EACH_SAFE (evg, next, id_node,
6280 &table->eviction_groups_by_id) {
6281 eviction_group_destroy(table, evg);
6283 hmap_destroy(&table->eviction_groups_by_id);
6284 heap_destroy(&table->eviction_groups_by_size);
6286 free(table->eviction_fields);
6287 table->eviction_fields = NULL;
6288 table->n_eviction_fields = 0;
6292 /* oftables support a choice of two policies when adding a rule would cause the
6293 * number of flows in the table to exceed the configured maximum number: either
6294 * they can refuse to add the new flow or they can evict some existing flow.
6295 * This function configures the latter policy on 'table', with fairness based
6296 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
6297 * 'n_fields' as 0 disables fairness.) */
6299 oftable_enable_eviction(struct oftable *table,
6300 const struct mf_subfield *fields, size_t n_fields)
6302 struct cls_cursor cursor;
6305 if (table->eviction_fields
6306 && n_fields == table->n_eviction_fields
6308 || !memcmp(fields, table->eviction_fields,
6309 n_fields * sizeof *fields))) {
6314 oftable_disable_eviction(table);
6316 table->n_eviction_fields = n_fields;
6317 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
6319 table->eviction_group_id_basis = random_uint32();
6320 hmap_init(&table->eviction_groups_by_id);
6321 heap_init(&table->eviction_groups_by_size);
6323 ovs_rwlock_rdlock(&table->cls.rwlock);
6324 cls_cursor_init(&cursor, &table->cls, NULL);
6325 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
6326 eviction_group_add_rule(rule);
6328 ovs_rwlock_unlock(&table->cls.rwlock);
6331 /* Removes 'rule' from the oftable that contains it. */
6333 oftable_remove_rule__(struct ofproto *ofproto, struct classifier *cls,
6335 OVS_REQ_WRLOCK(cls->rwlock) OVS_RELEASES(rule->rwlock)
6337 classifier_remove(cls, &rule->cr);
6338 cookies_remove(ofproto, rule);
6339 eviction_group_remove_rule(rule);
6340 ovs_mutex_lock(&ofproto_mutex);
6341 if (!list_is_empty(&rule->expirable)) {
6342 list_remove(&rule->expirable);
6344 ovs_mutex_unlock(&ofproto_mutex);
6345 if (!list_is_empty(&rule->meter_list_node)) {
6346 list_remove(&rule->meter_list_node);
6347 list_init(&rule->meter_list_node);
6349 ovs_rwlock_unlock(&rule->rwlock);
6353 oftable_remove_rule(struct rule *rule)
6355 struct ofproto *ofproto = rule->ofproto;
6356 struct oftable *table = &ofproto->tables[rule->table_id];
6358 ovs_rwlock_wrlock(&table->cls.rwlock);
6359 oftable_remove_rule__(ofproto, &table->cls, rule);
6360 ovs_rwlock_unlock(&table->cls.rwlock);
6363 /* Inserts 'rule' into its oftable, which must not already contain any rule for
6364 * the same cls_rule. */
6366 oftable_insert_rule(struct rule *rule)
6368 struct ofproto *ofproto = rule->ofproto;
6369 struct oftable *table = &ofproto->tables[rule->table_id];
6372 ovs_mutex_lock(&rule->timeout_mutex);
6373 may_expire = rule->hard_timeout || rule->idle_timeout;
6374 ovs_mutex_unlock(&rule->timeout_mutex);
6377 ovs_mutex_lock(&ofproto_mutex);
6378 list_insert(&ofproto->expirable, &rule->expirable);
6379 ovs_mutex_unlock(&ofproto_mutex);
6381 cookies_insert(ofproto, rule);
6383 if (rule->actions->meter_id) {
6384 struct meter *meter = ofproto->meters[rule->actions->meter_id];
6385 list_insert(&meter->rules, &rule->meter_list_node);
6387 ovs_rwlock_wrlock(&table->cls.rwlock);
6388 classifier_insert(&table->cls, &rule->cr);
6389 ovs_rwlock_unlock(&table->cls.rwlock);
6390 eviction_group_add_rule(rule);
6393 /* unixctl commands. */
6396 ofproto_lookup(const char *name)
6398 struct ofproto *ofproto;
6400 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
6402 if (!strcmp(ofproto->name, name)) {
6410 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
6411 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6413 struct ofproto *ofproto;
6417 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
6418 ds_put_format(&results, "%s\n", ofproto->name);
6420 unixctl_command_reply(conn, ds_cstr(&results));
6421 ds_destroy(&results);
6425 ofproto_unixctl_init(void)
6427 static bool registered;
6433 unixctl_command_register("ofproto/list", "", 0, 0,
6434 ofproto_unixctl_list, NULL);
6437 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6439 * This is deprecated. It is only for compatibility with broken device drivers
6440 * in old versions of Linux that do not properly support VLANs when VLAN
6441 * devices are not used. When broken device drivers are no longer in
6442 * widespread use, we will delete these interfaces. */
6444 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
6445 * (exactly) by an OpenFlow rule in 'ofproto'. */
6447 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
6449 const struct oftable *oftable;
6451 free(ofproto->vlan_bitmap);
6452 ofproto->vlan_bitmap = bitmap_allocate(4096);
6453 ofproto->vlans_changed = false;
6455 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
6456 const struct cls_table *table;
6458 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
6459 if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
6460 const struct cls_rule *rule;
6462 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
6463 uint16_t vid = miniflow_get_vid(&rule->match.flow);
6464 bitmap_set1(vlan_bitmap, vid);
6465 bitmap_set1(ofproto->vlan_bitmap, vid);
6472 /* Returns true if new VLANs have come into use by the flow table since the
6473 * last call to ofproto_get_vlan_usage().
6475 * We don't track when old VLANs stop being used. */
6477 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
6479 return ofproto->vlans_changed;
6482 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
6483 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
6484 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
6485 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
6486 * then the VLAN device is un-enslaved. */
6488 ofproto_port_set_realdev(struct ofproto *ofproto, ofp_port_t vlandev_ofp_port,
6489 ofp_port_t realdev_ofp_port, int vid)
6491 struct ofport *ofport;
6494 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
6496 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
6498 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
6499 ofproto->name, vlandev_ofp_port);
6503 if (!ofproto->ofproto_class->set_realdev) {
6504 if (!vlandev_ofp_port) {
6507 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
6511 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
6513 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
6514 ofproto->name, vlandev_ofp_port,
6515 netdev_get_name(ofport->netdev), ovs_strerror(error));