2 * Copyright (c) 2009-2016 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.
27 #include "byte-order.h"
28 #include "classifier.h"
29 #include "connectivity.h"
32 #include "dp-packet.h"
38 #include "ofproto-provider.h"
39 #include "openflow/nicira-ext.h"
40 #include "openflow/openflow.h"
41 #include "openvswitch/dynamic-string.h"
42 #include "openvswitch/meta-flow.h"
43 #include "openvswitch/ofp-actions.h"
44 #include "openvswitch/ofp-errors.h"
45 #include "openvswitch/ofp-msgs.h"
46 #include "openvswitch/ofp-print.h"
47 #include "openvswitch/ofp-util.h"
48 #include "openvswitch/ofpbuf.h"
49 #include "openvswitch/vlog.h"
54 #include "poll-loop.h"
62 #include "tun-metadata.h"
63 #include "unaligned.h"
66 VLOG_DEFINE_THIS_MODULE(ofproto);
68 COVERAGE_DEFINE(ofproto_flush);
69 COVERAGE_DEFINE(ofproto_packet_out);
70 COVERAGE_DEFINE(ofproto_queue_req);
71 COVERAGE_DEFINE(ofproto_recv_openflow);
72 COVERAGE_DEFINE(ofproto_reinit_ports);
73 COVERAGE_DEFINE(ofproto_update_port);
75 /* Default fields to use for prefix tries in each flow table, unless something
76 * else is configured. */
77 const enum mf_field_id default_prefix_fields[2] =
78 { MFF_IPV4_DST, MFF_IPV4_SRC };
81 static void oftable_init(struct oftable *);
82 static void oftable_destroy(struct oftable *);
84 static void oftable_set_name(struct oftable *, const char *name);
86 static enum ofperr evict_rules_from_table(struct oftable *)
87 OVS_REQUIRES(ofproto_mutex);
88 static void oftable_configure_eviction(struct oftable *,
89 unsigned int eviction,
90 const struct mf_subfield *fields,
92 OVS_REQUIRES(ofproto_mutex);
94 /* This is the only combination of OpenFlow eviction flags that OVS supports: a
95 * combination of OF1.4+ importance, the remaining lifetime of the flow, and
96 * fairness based on user-specified fields. */
97 #define OFPROTO_EVICTION_FLAGS \
98 (OFPTMPEF14_OTHER | OFPTMPEF14_IMPORTANCE | OFPTMPEF14_LIFETIME)
100 /* A set of rules within a single OpenFlow table (oftable) that have the same
101 * values for the oftable's eviction_fields. A rule to be evicted, when one is
102 * needed, is taken from the eviction group that contains the greatest number
105 * An oftable owns any number of eviction groups, each of which contains any
108 * Membership in an eviction group is imprecise, based on the hash of the
109 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
110 * That is, if two rules have different eviction_fields, but those
111 * eviction_fields hash to the same value, then they will belong to the same
112 * eviction_group anyway.
114 * (When eviction is not enabled on an oftable, we don't track any eviction
115 * groups, to save time and space.) */
116 struct eviction_group {
117 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
118 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
119 struct heap rules; /* Contains "struct rule"s. */
122 static bool choose_rule_to_evict(struct oftable *table, struct rule **rulep)
123 OVS_REQUIRES(ofproto_mutex);
124 static uint64_t rule_eviction_priority(struct ofproto *ofproto, struct rule *)
125 OVS_REQUIRES(ofproto_mutex);
126 static void eviction_group_add_rule(struct rule *)
127 OVS_REQUIRES(ofproto_mutex);
128 static void eviction_group_remove_rule(struct rule *)
129 OVS_REQUIRES(ofproto_mutex);
131 /* Criteria that flow_mod and other operations use for selecting rules on
132 * which to operate. */
133 struct rule_criteria {
134 /* An OpenFlow table or 255 for all tables. */
137 /* OpenFlow matching criteria. Interpreted different in "loose" way by
138 * collect_rules_loose() and "strict" way by collect_rules_strict(), as
139 * defined in the OpenFlow spec. */
141 cls_version_t version;
143 /* Matching criteria for the OpenFlow cookie. Consider a bit B in a rule's
144 * cookie and the corresponding bits C in 'cookie' and M in 'cookie_mask'.
145 * The rule will not be selected if M is 1 and B != C. */
147 ovs_be64 cookie_mask;
149 /* Selection based on actions within a rule:
151 * If out_port != OFPP_ANY, selects only rules that output to out_port.
152 * If out_group != OFPG_ALL, select only rules that output to out_group. */
156 /* If true, collects only rules that are modifiable. */
158 bool include_readonly;
161 static void rule_criteria_init(struct rule_criteria *, uint8_t table_id,
162 const struct match *match, int priority,
163 cls_version_t version,
164 ovs_be64 cookie, ovs_be64 cookie_mask,
165 ofp_port_t out_port, uint32_t out_group);
166 static void rule_criteria_require_rw(struct rule_criteria *,
167 bool can_write_readonly);
168 static void rule_criteria_destroy(struct rule_criteria *);
170 static enum ofperr collect_rules_loose(struct ofproto *,
171 const struct rule_criteria *,
172 struct rule_collection *);
174 /* A packet that needs to be passed to rule_execute().
176 * (We can't do this immediately from ofopgroup_complete() because that holds
177 * ofproto_mutex, which rule_execute() needs released.) */
178 struct rule_execute {
179 struct ovs_list list_node; /* In struct ofproto's "rule_executes" list. */
180 struct rule *rule; /* Owns a reference to the rule. */
182 struct dp_packet *packet; /* Owns the packet. */
185 static void run_rule_executes(struct ofproto *) OVS_EXCLUDED(ofproto_mutex);
186 static void destroy_rule_executes(struct ofproto *);
188 struct learned_cookie {
190 /* In struct ofproto's 'learned_cookies' hmap. */
191 struct hmap_node hmap_node OVS_GUARDED_BY(ofproto_mutex);
193 /* In 'dead_cookies' list when removed from hmap. */
194 struct ovs_list list_node;
198 ovs_be64 cookie OVS_GUARDED_BY(ofproto_mutex);
199 uint8_t table_id OVS_GUARDED_BY(ofproto_mutex);
201 /* Number of references from "learn" actions.
203 * When this drops to 0, all of the flows in 'table_id' with the specified
204 * 'cookie' are deleted. */
205 int n OVS_GUARDED_BY(ofproto_mutex);
208 static const struct ofpact_learn *next_learn_with_delete(
209 const struct rule_actions *, const struct ofpact_learn *start);
211 static void learned_cookies_inc(struct ofproto *, const struct rule_actions *)
212 OVS_REQUIRES(ofproto_mutex);
213 static void learned_cookies_dec(struct ofproto *, const struct rule_actions *,
214 struct ovs_list *dead_cookies)
215 OVS_REQUIRES(ofproto_mutex);
216 static void learned_cookies_flush(struct ofproto *, struct ovs_list *dead_cookies)
217 OVS_REQUIRES(ofproto_mutex);
220 static void ofport_destroy__(struct ofport *) OVS_EXCLUDED(ofproto_mutex);
221 static void ofport_destroy(struct ofport *, bool del);
223 static int update_port(struct ofproto *, const char *devname);
224 static int init_ports(struct ofproto *);
225 static void reinit_ports(struct ofproto *);
227 static long long int ofport_get_usage(const struct ofproto *,
228 ofp_port_t ofp_port);
229 static void ofport_set_usage(struct ofproto *, ofp_port_t ofp_port,
230 long long int last_used);
231 static void ofport_remove_usage(struct ofproto *, ofp_port_t ofp_port);
235 * Keeps track of the currently used and recently used ofport values and is
236 * used to prevent immediate recycling of ofport values. */
237 struct ofport_usage {
238 struct hmap_node hmap_node; /* In struct ofproto's "ofport_usage" hmap. */
239 ofp_port_t ofp_port; /* OpenFlow port number. */
240 long long int last_used; /* Last time the 'ofp_port' was used. LLONG_MAX
241 represents in-use ofports. */
245 static void ofproto_rule_send_removed(struct rule *)
246 OVS_EXCLUDED(ofproto_mutex);
247 static bool rule_is_readonly(const struct rule *);
248 static void ofproto_rule_insert__(struct ofproto *, struct rule *)
249 OVS_REQUIRES(ofproto_mutex);
250 static void ofproto_rule_remove__(struct ofproto *, struct rule *)
251 OVS_REQUIRES(ofproto_mutex);
253 /* The source of a flow_mod request, in the code that processes flow_mods.
255 * A flow table modification request can be generated externally, via OpenFlow,
256 * or internally through a function call. This structure indicates the source
257 * of an OpenFlow-generated flow_mod. For an internal flow_mod, it isn't
258 * meaningful and thus supplied as NULL. */
259 struct flow_mod_requester {
260 struct ofconn *ofconn; /* Connection on which flow_mod arrived. */
261 const struct ofp_header *request;
265 static enum ofperr replace_rule_create(struct ofproto *,
266 struct ofputil_flow_mod *,
267 struct cls_rule *cr, uint8_t table_id,
268 struct rule *old_rule,
269 struct rule **new_rule)
270 OVS_REQUIRES(ofproto_mutex);
272 static void replace_rule_start(struct ofproto *, cls_version_t version,
273 struct rule *old_rule, struct rule *new_rule,
274 struct cls_conjunction *, size_t n_conjs)
275 OVS_REQUIRES(ofproto_mutex);
277 static void replace_rule_revert(struct ofproto *, struct rule *old_rule,
278 struct rule *new_rule)
279 OVS_REQUIRES(ofproto_mutex);
281 static void replace_rule_finish(struct ofproto *, struct ofputil_flow_mod *,
282 const struct flow_mod_requester *,
283 struct rule *old_rule, struct rule *new_rule,
284 struct ovs_list *dead_cookies)
285 OVS_REQUIRES(ofproto_mutex);
286 static void delete_flows__(struct rule_collection *,
287 enum ofp_flow_removed_reason,
288 const struct flow_mod_requester *)
289 OVS_REQUIRES(ofproto_mutex);
291 static void send_buffered_packet(const struct flow_mod_requester *,
292 uint32_t buffer_id, struct rule *)
293 OVS_REQUIRES(ofproto_mutex);
295 static bool ofproto_group_exists__(const struct ofproto *ofproto,
297 OVS_REQ_RDLOCK(ofproto->groups_rwlock);
298 static bool ofproto_group_exists(const struct ofproto *ofproto,
300 OVS_EXCLUDED(ofproto->groups_rwlock);
301 static enum ofperr add_group(struct ofproto *,
302 const struct ofputil_group_mod *);
303 static void handle_openflow(struct ofconn *, const struct ofpbuf *);
304 static enum ofperr ofproto_flow_mod_start(struct ofproto *,
305 struct ofproto_flow_mod *)
306 OVS_REQUIRES(ofproto_mutex);
307 static void ofproto_flow_mod_finish(struct ofproto *,
308 struct ofproto_flow_mod *,
309 const struct flow_mod_requester *)
310 OVS_REQUIRES(ofproto_mutex);
311 static enum ofperr handle_flow_mod__(struct ofproto *,
312 struct ofproto_flow_mod *,
313 const struct flow_mod_requester *)
314 OVS_EXCLUDED(ofproto_mutex);
315 static void calc_duration(long long int start, long long int now,
316 uint32_t *sec, uint32_t *nsec);
319 static uint64_t pick_datapath_id(const struct ofproto *);
320 static uint64_t pick_fallback_dpid(void);
321 static void ofproto_destroy__(struct ofproto *);
322 static void update_mtu(struct ofproto *, struct ofport *);
323 static void meter_delete(struct ofproto *, uint32_t first, uint32_t last);
324 static void meter_insert_rule(struct rule *);
327 static void ofproto_unixctl_init(void);
329 /* All registered ofproto classes, in probe order. */
330 static const struct ofproto_class **ofproto_classes;
331 static size_t n_ofproto_classes;
332 static size_t allocated_ofproto_classes;
334 /* Global lock that protects all flow table operations. */
335 struct ovs_mutex ofproto_mutex = OVS_MUTEX_INITIALIZER;
337 unsigned ofproto_flow_limit = OFPROTO_FLOW_LIMIT_DEFAULT;
338 unsigned ofproto_max_idle = OFPROTO_MAX_IDLE_DEFAULT;
340 size_t n_handlers, n_revalidators;
343 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
344 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
346 /* Initial mappings of port to OpenFlow number mappings. */
347 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
349 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
351 /* The default value of true waits for flow restore. */
352 static bool flow_restore_wait = true;
354 /* Must be called to initialize the ofproto library.
356 * The caller may pass in 'iface_hints', which contains an shash of
357 * "iface_hint" elements indexed by the interface's name. The provider
358 * may use these hints to describe the startup configuration in order to
359 * reinitialize its state. The caller owns the provided data, so a
360 * provider will make copies of anything required. An ofproto provider
361 * will remove any existing state that is not described by the hint, and
362 * may choose to remove it all. */
364 ofproto_init(const struct shash *iface_hints)
366 struct shash_node *node;
369 ofproto_class_register(&ofproto_dpif_class);
371 /* Make a local copy, since we don't own 'iface_hints' elements. */
372 SHASH_FOR_EACH(node, iface_hints) {
373 const struct iface_hint *orig_hint = node->data;
374 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
375 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
377 new_hint->br_name = xstrdup(orig_hint->br_name);
378 new_hint->br_type = xstrdup(br_type);
379 new_hint->ofp_port = orig_hint->ofp_port;
381 shash_add(&init_ofp_ports, node->name, new_hint);
384 for (i = 0; i < n_ofproto_classes; i++) {
385 ofproto_classes[i]->init(&init_ofp_ports);
388 ofproto_unixctl_init();
391 /* 'type' should be a normalized datapath type, as returned by
392 * ofproto_normalize_type(). Returns the corresponding ofproto_class
393 * structure, or a null pointer if there is none registered for 'type'. */
394 static const struct ofproto_class *
395 ofproto_class_find__(const char *type)
399 for (i = 0; i < n_ofproto_classes; i++) {
400 const struct ofproto_class *class = ofproto_classes[i];
405 class->enumerate_types(&types);
406 found = sset_contains(&types, type);
407 sset_destroy(&types);
413 VLOG_WARN("unknown datapath type %s", type);
417 /* Registers a new ofproto class. After successful registration, new ofprotos
418 * of that type can be created using ofproto_create(). */
420 ofproto_class_register(const struct ofproto_class *new_class)
424 for (i = 0; i < n_ofproto_classes; i++) {
425 if (ofproto_classes[i] == new_class) {
430 if (n_ofproto_classes >= allocated_ofproto_classes) {
431 ofproto_classes = x2nrealloc(ofproto_classes,
432 &allocated_ofproto_classes,
433 sizeof *ofproto_classes);
435 ofproto_classes[n_ofproto_classes++] = new_class;
439 /* Unregisters a datapath provider. 'type' must have been previously
440 * registered and not currently be in use by any ofprotos. After
441 * unregistration new datapaths of that type cannot be opened using
442 * ofproto_create(). */
444 ofproto_class_unregister(const struct ofproto_class *class)
448 for (i = 0; i < n_ofproto_classes; i++) {
449 if (ofproto_classes[i] == class) {
450 for (i++; i < n_ofproto_classes; i++) {
451 ofproto_classes[i - 1] = ofproto_classes[i];
457 VLOG_WARN("attempted to unregister an ofproto class that is not "
462 /* Clears 'types' and enumerates all registered ofproto types into it. The
463 * caller must first initialize the sset. */
465 ofproto_enumerate_types(struct sset *types)
470 for (i = 0; i < n_ofproto_classes; i++) {
471 ofproto_classes[i]->enumerate_types(types);
475 /* Returns the fully spelled out name for the given ofproto 'type'.
477 * Normalized type string can be compared with strcmp(). Unnormalized type
478 * string might be the same even if they have different spellings. */
480 ofproto_normalize_type(const char *type)
482 return type && type[0] ? type : "system";
485 /* Clears 'names' and enumerates the names of all known created ofprotos with
486 * the given 'type'. The caller must first initialize the sset. Returns 0 if
487 * successful, otherwise a positive errno value.
489 * Some kinds of datapaths might not be practically enumerable. This is not
490 * considered an error. */
492 ofproto_enumerate_names(const char *type, struct sset *names)
494 const struct ofproto_class *class = ofproto_class_find__(type);
495 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
499 ofproto_bump_tables_version(struct ofproto *ofproto)
501 ++ofproto->tables_version;
502 ofproto->ofproto_class->set_tables_version(ofproto,
503 ofproto->tables_version);
507 ofproto_create(const char *datapath_name, const char *datapath_type,
508 struct ofproto **ofprotop)
510 const struct ofproto_class *class;
511 struct ofproto *ofproto;
517 datapath_type = ofproto_normalize_type(datapath_type);
518 class = ofproto_class_find__(datapath_type);
520 VLOG_WARN("could not create datapath %s of unknown type %s",
521 datapath_name, datapath_type);
525 ofproto = class->alloc();
527 VLOG_ERR("failed to allocate datapath %s of type %s",
528 datapath_name, datapath_type);
533 ovs_mutex_lock(&ofproto_mutex);
534 memset(ofproto, 0, sizeof *ofproto);
535 ofproto->ofproto_class = class;
536 ofproto->name = xstrdup(datapath_name);
537 ofproto->type = xstrdup(datapath_type);
538 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
539 hash_string(ofproto->name, 0));
540 ofproto->datapath_id = 0;
541 ofproto->forward_bpdu = false;
542 ofproto->fallback_dpid = pick_fallback_dpid();
543 ofproto->mfr_desc = NULL;
544 ofproto->hw_desc = NULL;
545 ofproto->sw_desc = NULL;
546 ofproto->serial_desc = NULL;
547 ofproto->dp_desc = NULL;
548 ofproto->frag_handling = OFPUTIL_FRAG_NORMAL;
549 hmap_init(&ofproto->ports);
550 hmap_init(&ofproto->ofport_usage);
551 shash_init(&ofproto->port_by_name);
552 simap_init(&ofproto->ofp_requests);
553 ofproto->max_ports = ofp_to_u16(OFPP_MAX);
554 ofproto->eviction_group_timer = LLONG_MIN;
555 ofproto->tables = NULL;
556 ofproto->n_tables = 0;
557 ofproto->tables_version = CLS_MIN_VERSION;
558 hindex_init(&ofproto->cookies);
559 hmap_init(&ofproto->learned_cookies);
560 ovs_list_init(&ofproto->expirable);
561 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
562 guarded_list_init(&ofproto->rule_executes);
563 ofproto->vlan_bitmap = NULL;
564 ofproto->vlans_changed = false;
565 ofproto->min_mtu = INT_MAX;
566 ovs_rwlock_init(&ofproto->groups_rwlock);
567 hmap_init(&ofproto->groups);
568 ovs_mutex_unlock(&ofproto_mutex);
569 ofproto->ogf.types = 0xf;
570 ofproto->ogf.capabilities = OFPGFC_CHAINING | OFPGFC_SELECT_LIVENESS |
571 OFPGFC_SELECT_WEIGHT;
572 for (i = 0; i < 4; i++) {
573 ofproto->ogf.max_groups[i] = OFPG_MAX;
574 ofproto->ogf.ofpacts[i] = (UINT64_C(1) << N_OFPACTS) - 1;
578 error = ofproto->ofproto_class->construct(ofproto);
580 VLOG_ERR("failed to open datapath %s: %s",
581 datapath_name, ovs_strerror(error));
582 connmgr_destroy(ofproto->connmgr);
583 ofproto_destroy__(ofproto);
587 /* Check that hidden tables, if any, are at the end. */
588 ovs_assert(ofproto->n_tables);
589 for (i = 0; i + 1 < ofproto->n_tables; i++) {
590 enum oftable_flags flags = ofproto->tables[i].flags;
591 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
593 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
596 ofproto->datapath_id = pick_datapath_id(ofproto);
599 /* Initialize meters table. */
600 if (ofproto->ofproto_class->meter_get_features) {
601 ofproto->ofproto_class->meter_get_features(ofproto,
602 &ofproto->meter_features);
604 memset(&ofproto->meter_features, 0, sizeof ofproto->meter_features);
606 ofproto->meters = xzalloc((ofproto->meter_features.max_meters + 1)
607 * sizeof(struct meter *));
609 /* Set the initial tables version. */
610 ofproto_bump_tables_version(ofproto);
616 /* Must be called (only) by an ofproto implementation in its constructor
617 * function. See the large comment on 'construct' in struct ofproto_class for
620 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
622 struct oftable *table;
624 ovs_assert(!ofproto->n_tables);
625 ovs_assert(n_tables >= 1 && n_tables <= 255);
627 ofproto->n_tables = n_tables;
628 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
629 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
634 /* To be optionally called (only) by an ofproto implementation in its
635 * constructor function. See the large comment on 'construct' in struct
636 * ofproto_class for details.
638 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
639 * will then ensure that actions passed into the ofproto implementation will
640 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
641 * function is not called, there will be no such restriction.
643 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
644 * the 'max_ports' restriction. */
646 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
648 ovs_assert(max_ports <= ofp_to_u16(OFPP_MAX));
649 ofproto->max_ports = max_ports;
653 ofproto_get_datapath_id(const struct ofproto *ofproto)
655 return ofproto->datapath_id;
659 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
661 uint64_t old_dpid = p->datapath_id;
662 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
663 if (p->datapath_id != old_dpid) {
664 /* Force all active connections to reconnect, since there is no way to
665 * notify a controller that the datapath ID has changed. */
666 ofproto_reconnect_controllers(p);
671 ofproto_set_controllers(struct ofproto *p,
672 const struct ofproto_controller *controllers,
673 size_t n_controllers, uint32_t allowed_versions)
675 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
680 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
682 connmgr_set_fail_mode(p->connmgr, fail_mode);
685 /* Drops the connections between 'ofproto' and all of its controllers, forcing
686 * them to reconnect. */
688 ofproto_reconnect_controllers(struct ofproto *ofproto)
690 connmgr_reconnect(ofproto->connmgr);
693 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
694 * in-band control should guarantee access, in the same way that in-band
695 * control guarantees access to OpenFlow controllers. */
697 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
698 const struct sockaddr_in *extras, size_t n)
700 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
703 /* Sets the OpenFlow queue used by flows set up by in-band control on
704 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
705 * flows will use the default queue. */
707 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
709 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
712 /* Sets the number of flows at which eviction from the kernel flow table
715 ofproto_set_flow_limit(unsigned limit)
717 ofproto_flow_limit = limit;
720 /* Sets the maximum idle time for flows in the datapath before they are
723 ofproto_set_max_idle(unsigned max_idle)
725 ofproto_max_idle = max_idle;
728 /* If forward_bpdu is true, the NORMAL action will forward frames with
729 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
730 * the NORMAL action will drop these frames. */
732 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
734 bool old_val = ofproto->forward_bpdu;
735 ofproto->forward_bpdu = forward_bpdu;
736 if (old_val != ofproto->forward_bpdu) {
737 if (ofproto->ofproto_class->forward_bpdu_changed) {
738 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
743 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
744 * 'idle_time', in seconds, and the maximum number of MAC table entries to
747 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
750 if (ofproto->ofproto_class->set_mac_table_config) {
751 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
756 /* Multicast snooping configuration. */
758 /* Configures multicast snooping on 'ofproto' using the settings
759 * defined in 's'. If 's' is NULL, disables multicast snooping.
761 * Returns 0 if successful, otherwise a positive errno value. */
763 ofproto_set_mcast_snooping(struct ofproto *ofproto,
764 const struct ofproto_mcast_snooping_settings *s)
766 return (ofproto->ofproto_class->set_mcast_snooping
767 ? ofproto->ofproto_class->set_mcast_snooping(ofproto, s)
771 /* Configures multicast snooping flood settings on 'ofp_port' of 'ofproto'.
773 * Returns 0 if successful, otherwise a positive errno value.*/
775 ofproto_port_set_mcast_snooping(struct ofproto *ofproto, void *aux,
776 const struct ofproto_mcast_snooping_port_settings *s)
778 return (ofproto->ofproto_class->set_mcast_snooping_port
779 ? ofproto->ofproto_class->set_mcast_snooping_port(ofproto, aux, s)
784 ofproto_set_cpu_mask(const char *cmask)
788 pmd_cpu_mask = cmask ? xstrdup(cmask) : NULL;
792 ofproto_set_threads(int n_handlers_, int n_revalidators_)
794 int threads = MAX(count_cpu_cores(), 2);
796 n_revalidators = MAX(n_revalidators_, 0);
797 n_handlers = MAX(n_handlers_, 0);
799 if (!n_revalidators) {
800 n_revalidators = n_handlers
801 ? MAX(threads - (int) n_handlers, 1)
806 n_handlers = MAX(threads - (int) n_revalidators, 1);
811 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
814 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
818 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
820 return connmgr_set_snoops(ofproto->connmgr, snoops);
824 ofproto_set_netflow(struct ofproto *ofproto,
825 const struct netflow_options *nf_options)
827 if (nf_options && sset_is_empty(&nf_options->collectors)) {
831 if (ofproto->ofproto_class->set_netflow) {
832 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
834 return nf_options ? EOPNOTSUPP : 0;
839 ofproto_set_sflow(struct ofproto *ofproto,
840 const struct ofproto_sflow_options *oso)
842 if (oso && sset_is_empty(&oso->targets)) {
846 if (ofproto->ofproto_class->set_sflow) {
847 return ofproto->ofproto_class->set_sflow(ofproto, oso);
849 return oso ? EOPNOTSUPP : 0;
854 ofproto_set_ipfix(struct ofproto *ofproto,
855 const struct ofproto_ipfix_bridge_exporter_options *bo,
856 const struct ofproto_ipfix_flow_exporter_options *fo,
859 if (ofproto->ofproto_class->set_ipfix) {
860 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
862 return (bo || fo) ? EOPNOTSUPP : 0;
867 ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
869 flow_restore_wait = flow_restore_wait_db;
873 ofproto_get_flow_restore_wait(void)
875 return flow_restore_wait;
879 /* Spanning Tree Protocol (STP) configuration. */
881 /* Configures STP on 'ofproto' using the settings defined in 's'. If
882 * 's' is NULL, disables STP.
884 * Returns 0 if successful, otherwise a positive errno value. */
886 ofproto_set_stp(struct ofproto *ofproto,
887 const struct ofproto_stp_settings *s)
889 return (ofproto->ofproto_class->set_stp
890 ? ofproto->ofproto_class->set_stp(ofproto, s)
894 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
895 * 'enabled' member of 's' is false, then the other members are not
898 * Returns 0 if successful, otherwise a positive errno value. */
900 ofproto_get_stp_status(struct ofproto *ofproto,
901 struct ofproto_stp_status *s)
903 return (ofproto->ofproto_class->get_stp_status
904 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
908 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
909 * in 's'. The caller is responsible for assigning STP port numbers
910 * (using the 'port_num' member in the range of 1 through 255, inclusive)
911 * and ensuring there are no duplicates. If the 's' is NULL, then STP
912 * is disabled on the port.
914 * Returns 0 if successful, otherwise a positive errno value.*/
916 ofproto_port_set_stp(struct ofproto *ofproto, ofp_port_t ofp_port,
917 const struct ofproto_port_stp_settings *s)
919 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
921 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
922 ofproto->name, ofp_port);
926 return (ofproto->ofproto_class->set_stp_port
927 ? ofproto->ofproto_class->set_stp_port(ofport, s)
931 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
932 * 's'. If the 'enabled' member in 's' is false, then the other members
933 * are not meaningful.
935 * Returns 0 if successful, otherwise a positive errno value.*/
937 ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
938 struct ofproto_port_stp_status *s)
940 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
942 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
943 "port %"PRIu16, ofproto->name, ofp_port);
947 return (ofproto->ofproto_class->get_stp_port_status
948 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
952 /* Retrieves STP port statistics of 'ofp_port' on 'ofproto' and stores it in
953 * 's'. If the 'enabled' member in 's' is false, then the other members
954 * are not meaningful.
956 * Returns 0 if successful, otherwise a positive errno value.*/
958 ofproto_port_get_stp_stats(struct ofproto *ofproto, ofp_port_t ofp_port,
959 struct ofproto_port_stp_stats *s)
961 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
963 VLOG_WARN_RL(&rl, "%s: cannot get STP stats on nonexistent "
964 "port %"PRIu16, ofproto->name, ofp_port);
968 return (ofproto->ofproto_class->get_stp_port_stats
969 ? ofproto->ofproto_class->get_stp_port_stats(ofport, s)
973 /* Rapid Spanning Tree Protocol (RSTP) configuration. */
975 /* Configures RSTP on 'ofproto' using the settings defined in 's'. If
976 * 's' is NULL, disables RSTP.
978 * Returns 0 if successful, otherwise a positive errno value. */
980 ofproto_set_rstp(struct ofproto *ofproto,
981 const struct ofproto_rstp_settings *s)
983 if (!ofproto->ofproto_class->set_rstp) {
986 ofproto->ofproto_class->set_rstp(ofproto, s);
990 /* Retrieves RSTP status of 'ofproto' and stores it in 's'. If the
991 * 'enabled' member of 's' is false, then the other members are not
994 * Returns 0 if successful, otherwise a positive errno value. */
996 ofproto_get_rstp_status(struct ofproto *ofproto,
997 struct ofproto_rstp_status *s)
999 if (!ofproto->ofproto_class->get_rstp_status) {
1002 ofproto->ofproto_class->get_rstp_status(ofproto, s);
1006 /* Configures RSTP on 'ofp_port' of 'ofproto' using the settings defined
1007 * in 's'. The caller is responsible for assigning RSTP port numbers
1008 * (using the 'port_num' member in the range of 1 through 255, inclusive)
1009 * and ensuring there are no duplicates. If the 's' is NULL, then RSTP
1010 * is disabled on the port.
1012 * Returns 0 if successful, otherwise a positive errno value.*/
1014 ofproto_port_set_rstp(struct ofproto *ofproto, ofp_port_t ofp_port,
1015 const struct ofproto_port_rstp_settings *s)
1017 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1019 VLOG_WARN("%s: cannot configure RSTP on nonexistent port %"PRIu16,
1020 ofproto->name, ofp_port);
1024 if (!ofproto->ofproto_class->set_rstp_port) {
1027 ofproto->ofproto_class->set_rstp_port(ofport, s);
1031 /* Retrieves RSTP port status of 'ofp_port' on 'ofproto' and stores it in
1032 * 's'. If the 'enabled' member in 's' is false, then the other members
1033 * are not meaningful.
1035 * Returns 0 if successful, otherwise a positive errno value.*/
1037 ofproto_port_get_rstp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
1038 struct ofproto_port_rstp_status *s)
1040 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1042 VLOG_WARN_RL(&rl, "%s: cannot get RSTP status on nonexistent "
1043 "port %"PRIu16, ofproto->name, ofp_port);
1047 if (!ofproto->ofproto_class->get_rstp_port_status) {
1050 ofproto->ofproto_class->get_rstp_port_status(ofport, s);
1054 /* Queue DSCP configuration. */
1056 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
1057 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
1058 * to implement QoS. Instead, it is used to implement features which require
1059 * knowledge of what queues exist on a port, and some basic information about
1062 * Returns 0 if successful, otherwise a positive errno value. */
1064 ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port,
1065 const struct ofproto_port_queue *queues,
1068 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1071 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
1072 ofproto->name, ofp_port);
1076 return (ofproto->ofproto_class->set_queues
1077 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
1081 /* LLDP configuration. */
1083 ofproto_port_set_lldp(struct ofproto *ofproto,
1084 ofp_port_t ofp_port,
1085 const struct smap *cfg)
1087 struct ofport *ofport;
1090 ofport = ofproto_get_port(ofproto, ofp_port);
1092 VLOG_WARN("%s: cannot configure LLDP on nonexistent port %"PRIu16,
1093 ofproto->name, ofp_port);
1096 error = (ofproto->ofproto_class->set_lldp
1097 ? ofproto->ofproto_class->set_lldp(ofport, cfg)
1100 VLOG_WARN("%s: lldp configuration on port %"PRIu16" (%s) failed (%s)",
1101 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
1102 ovs_strerror(error));
1107 ofproto_set_aa(struct ofproto *ofproto, void *aux OVS_UNUSED,
1108 const struct aa_settings *s)
1110 if (!ofproto->ofproto_class->set_aa) {
1113 ofproto->ofproto_class->set_aa(ofproto, s);
1118 ofproto_aa_mapping_register(struct ofproto *ofproto, void *aux,
1119 const struct aa_mapping_settings *s)
1121 if (!ofproto->ofproto_class->aa_mapping_set) {
1124 ofproto->ofproto_class->aa_mapping_set(ofproto, aux, s);
1129 ofproto_aa_mapping_unregister(struct ofproto *ofproto, void *aux)
1131 if (!ofproto->ofproto_class->aa_mapping_unset) {
1134 ofproto->ofproto_class->aa_mapping_unset(ofproto, aux);
1139 ofproto_aa_vlan_get_queued(struct ofproto *ofproto,
1140 struct ovs_list *list)
1142 if (!ofproto->ofproto_class->aa_vlan_get_queued) {
1145 ofproto->ofproto_class->aa_vlan_get_queued(ofproto, list);
1150 ofproto_aa_vlan_get_queue_size(struct ofproto *ofproto)
1152 if (!ofproto->ofproto_class->aa_vlan_get_queue_size) {
1155 return ofproto->ofproto_class->aa_vlan_get_queue_size(ofproto);
1158 /* Connectivity Fault Management configuration. */
1160 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
1162 ofproto_port_clear_cfm(struct ofproto *ofproto, ofp_port_t ofp_port)
1164 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1165 if (ofport && ofproto->ofproto_class->set_cfm) {
1166 ofproto->ofproto_class->set_cfm(ofport, NULL);
1170 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
1171 * basic configuration from the configuration members in 'cfm', and the remote
1172 * maintenance point ID from remote_mpid. Ignores the statistics members of
1175 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
1177 ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port,
1178 const struct cfm_settings *s)
1180 struct ofport *ofport;
1183 ofport = ofproto_get_port(ofproto, ofp_port);
1185 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
1186 ofproto->name, ofp_port);
1190 /* XXX: For configuration simplicity, we only support one remote_mpid
1191 * outside of the CFM module. It's not clear if this is the correct long
1192 * term solution or not. */
1193 error = (ofproto->ofproto_class->set_cfm
1194 ? ofproto->ofproto_class->set_cfm(ofport, s)
1197 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
1198 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
1199 ovs_strerror(error));
1203 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
1204 * 'ofproto' does not have a port 'ofp_port'. */
1206 ofproto_port_set_bfd(struct ofproto *ofproto, ofp_port_t ofp_port,
1207 const struct smap *cfg)
1209 struct ofport *ofport;
1212 ofport = ofproto_get_port(ofproto, ofp_port);
1214 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
1215 ofproto->name, ofp_port);
1219 error = (ofproto->ofproto_class->set_bfd
1220 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
1223 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
1224 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
1225 ovs_strerror(error));
1229 /* Checks the status change of BFD on 'ofport'.
1231 * Returns true if 'ofproto_class' does not support 'bfd_status_changed'. */
1233 ofproto_port_bfd_status_changed(struct ofproto *ofproto, ofp_port_t ofp_port)
1235 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1236 return (ofport && ofproto->ofproto_class->bfd_status_changed
1237 ? ofproto->ofproto_class->bfd_status_changed(ofport)
1241 /* Populates 'status' with the status of BFD on 'ofport'. Returns 0 on
1242 * success. Returns a positive errno otherwise. Has no effect if 'ofp_port'
1243 * is not an OpenFlow port in 'ofproto'.
1245 * The caller must provide and own '*status'. */
1247 ofproto_port_get_bfd_status(struct ofproto *ofproto, ofp_port_t ofp_port,
1248 struct smap *status)
1250 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1251 return (ofport && ofproto->ofproto_class->get_bfd_status
1252 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
1256 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
1257 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
1258 * 0 if LACP partner information is not current (generally indicating a
1259 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
1261 ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port)
1263 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1264 return (ofport && ofproto->ofproto_class->port_is_lacp_current
1265 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
1270 ofproto_port_get_lacp_stats(const struct ofport *port, struct lacp_slave_stats *stats)
1272 struct ofproto *ofproto = port->ofproto;
1275 if (ofproto->ofproto_class->port_get_lacp_stats) {
1276 error = ofproto->ofproto_class->port_get_lacp_stats(port, stats);
1286 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
1287 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
1288 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
1289 * configuration plus, if there is more than one slave, a bonding
1292 * If 'aux' is already registered then this function updates its configuration
1293 * to 's'. Otherwise, this function registers a new bundle.
1295 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
1298 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
1299 const struct ofproto_bundle_settings *s)
1301 return (ofproto->ofproto_class->bundle_set
1302 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
1306 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
1307 * If no such bundle has been registered, this has no effect. */
1309 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
1311 return ofproto_bundle_register(ofproto, aux, NULL);
1315 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
1316 * If 'aux' is already registered then this function updates its configuration
1317 * to 's'. Otherwise, this function registers a new mirror. */
1319 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
1320 const struct ofproto_mirror_settings *s)
1322 return (ofproto->ofproto_class->mirror_set
1323 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
1327 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
1328 * If no mirror has been registered, this has no effect. */
1330 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
1332 return ofproto_mirror_register(ofproto, aux, NULL);
1335 /* Retrieves statistics from mirror associated with client data pointer
1336 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
1337 * 'bytes', respectively. If a particular counters is not supported,
1338 * the appropriate argument is set to UINT64_MAX.
1341 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
1342 uint64_t *packets, uint64_t *bytes)
1344 if (!ofproto->ofproto_class->mirror_get_stats) {
1345 *packets = *bytes = UINT64_MAX;
1349 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
1353 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
1354 * which all packets are flooded, instead of using MAC learning. If
1355 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1357 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
1360 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
1362 return (ofproto->ofproto_class->set_flood_vlans
1363 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
1367 /* Returns true if 'aux' is a registered bundle that is currently in use as the
1368 * output for a mirror. */
1370 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
1372 return (ofproto->ofproto_class->is_mirror_output_bundle
1373 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
1377 /* Configuration of OpenFlow tables. */
1379 /* Returns the number of OpenFlow tables in 'ofproto'. */
1381 ofproto_get_n_tables(const struct ofproto *ofproto)
1383 return ofproto->n_tables;
1386 /* Returns the number of Controller visible OpenFlow tables
1387 * in 'ofproto'. This number will exclude Hidden tables.
1388 * This funtion's return value should be less or equal to that of
1389 * ofproto_get_n_tables() . */
1391 ofproto_get_n_visible_tables(const struct ofproto *ofproto)
1393 uint8_t n = ofproto->n_tables;
1395 /* Count only non-hidden tables in the number of tables. (Hidden tables,
1396 * if present, are always at the end.) */
1397 while(n && (ofproto->tables[n - 1].flags & OFTABLE_HIDDEN)) {
1404 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
1405 * settings from 's'. 'table_id' must be in the range 0 through the number of
1406 * OpenFlow tables in 'ofproto' minus 1, inclusive.
1408 * For read-only tables, only the name may be configured. */
1410 ofproto_configure_table(struct ofproto *ofproto, int table_id,
1411 const struct ofproto_table_settings *s)
1413 struct oftable *table;
1415 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
1416 table = &ofproto->tables[table_id];
1418 oftable_set_name(table, s->name);
1420 if (table->flags & OFTABLE_READONLY) {
1424 if (classifier_set_prefix_fields(&table->cls,
1425 s->prefix_fields, s->n_prefix_fields)) {
1426 /* XXX: Trigger revalidation. */
1429 ovs_mutex_lock(&ofproto_mutex);
1430 unsigned int new_eviction = (s->enable_eviction
1431 ? table->eviction | EVICTION_CLIENT
1432 : table->eviction & ~EVICTION_CLIENT);
1433 oftable_configure_eviction(table, new_eviction, s->groups, s->n_groups);
1434 table->max_flows = s->max_flows;
1435 evict_rules_from_table(table);
1436 ovs_mutex_unlock(&ofproto_mutex);
1440 ofproto_has_snoops(const struct ofproto *ofproto)
1442 return connmgr_has_snoops(ofproto->connmgr);
1446 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1448 connmgr_get_snoops(ofproto->connmgr, snoops);
1451 /* Deletes 'rule' from 'ofproto'.
1453 * Within an ofproto implementation, this function allows an ofproto
1454 * implementation to destroy any rules that remain when its ->destruct()
1455 * function is called. This function is not suitable for use elsewhere in an
1456 * ofproto implementation.
1458 * This function implements steps 4.4 and 4.5 in the section titled "Rule Life
1459 * Cycle" in ofproto-provider.h. */
1461 ofproto_rule_delete(struct ofproto *ofproto, struct rule *rule)
1462 OVS_EXCLUDED(ofproto_mutex)
1464 /* This skips the ofmonitor and flow-removed notifications because the
1465 * switch is being deleted and any OpenFlow channels have been or soon will
1467 ovs_mutex_lock(&ofproto_mutex);
1469 if (!rule->removed) {
1470 /* Make sure there is no postponed removal of the rule. */
1471 ovs_assert(cls_rule_visible_in_version(&rule->cr, CLS_MAX_VERSION));
1473 if (!classifier_remove(&rule->ofproto->tables[rule->table_id].cls,
1477 ofproto_rule_remove__(rule->ofproto, rule);
1478 ofproto->ofproto_class->rule_delete(rule);
1479 ofproto_rule_unref(rule);
1481 ovs_mutex_unlock(&ofproto_mutex);
1485 ofproto_flush__(struct ofproto *ofproto)
1486 OVS_EXCLUDED(ofproto_mutex)
1488 struct oftable *table;
1490 /* This will flush all datapath flows. */
1491 if (ofproto->ofproto_class->flush) {
1492 ofproto->ofproto_class->flush(ofproto);
1495 /* XXX: There is a small race window here, where new datapath flows can be
1496 * created by upcall handlers based on the existing flow table. We can not
1497 * call ofproto class flush while holding 'ofproto_mutex' to prevent this,
1498 * as then we could deadlock on syncing with the handler threads waiting on
1499 * the same mutex. */
1501 ovs_mutex_lock(&ofproto_mutex);
1502 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1503 struct rule_collection rules;
1506 if (table->flags & OFTABLE_HIDDEN) {
1510 rule_collection_init(&rules);
1512 CLS_FOR_EACH (rule, cr, &table->cls) {
1513 rule_collection_add(&rules, rule);
1515 delete_flows__(&rules, OFPRR_DELETE, NULL);
1517 /* XXX: Concurrent handler threads may insert new learned flows based on
1518 * learn actions of the now deleted flows right after we release
1519 * 'ofproto_mutex'. */
1520 ovs_mutex_unlock(&ofproto_mutex);
1523 static void delete_group(struct ofproto *ofproto, uint32_t group_id);
1526 ofproto_destroy__(struct ofproto *ofproto)
1527 OVS_EXCLUDED(ofproto_mutex)
1529 struct oftable *table;
1531 destroy_rule_executes(ofproto);
1533 guarded_list_destroy(&ofproto->rule_executes);
1534 ovs_rwlock_destroy(&ofproto->groups_rwlock);
1535 hmap_destroy(&ofproto->groups);
1537 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1538 free(ofproto->name);
1539 free(ofproto->type);
1540 free(ofproto->mfr_desc);
1541 free(ofproto->hw_desc);
1542 free(ofproto->sw_desc);
1543 free(ofproto->serial_desc);
1544 free(ofproto->dp_desc);
1545 hmap_destroy(&ofproto->ports);
1546 hmap_destroy(&ofproto->ofport_usage);
1547 shash_destroy(&ofproto->port_by_name);
1548 simap_destroy(&ofproto->ofp_requests);
1550 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1551 oftable_destroy(table);
1553 free(ofproto->tables);
1555 ovs_assert(hindex_is_empty(&ofproto->cookies));
1556 hindex_destroy(&ofproto->cookies);
1558 ovs_assert(hmap_is_empty(&ofproto->learned_cookies));
1559 hmap_destroy(&ofproto->learned_cookies);
1561 free(ofproto->vlan_bitmap);
1563 ofproto->ofproto_class->dealloc(ofproto);
1566 /* Destroying rules is doubly deferred, must have 'ofproto' around for them.
1567 * - 1st we defer the removal of the rules from the classifier
1568 * - 2nd we defer the actual destruction of the rules. */
1570 ofproto_destroy_defer__(struct ofproto *ofproto)
1571 OVS_EXCLUDED(ofproto_mutex)
1573 ovsrcu_postpone(ofproto_destroy__, ofproto);
1577 ofproto_destroy(struct ofproto *p, bool del)
1578 OVS_EXCLUDED(ofproto_mutex)
1580 struct ofport *ofport, *next_ofport;
1581 struct ofport_usage *usage, *next_usage;
1588 meter_delete(p, 1, p->meter_features.max_meters);
1589 p->meter_features.max_meters = 0;
1595 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1596 ofport_destroy(ofport, del);
1599 HMAP_FOR_EACH_SAFE (usage, next_usage, hmap_node, &p->ofport_usage) {
1600 hmap_remove(&p->ofport_usage, &usage->hmap_node);
1604 p->ofproto_class->destruct(p);
1606 /* We should not postpone this because it involves deleting a listening
1607 * socket which we may want to reopen soon. 'connmgr' should not be used
1608 * by other threads */
1609 connmgr_destroy(p->connmgr);
1611 /* Destroying rules is deferred, must have 'ofproto' around for them. */
1612 ovsrcu_postpone(ofproto_destroy_defer__, p);
1615 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1616 * kernel datapath, for example, this destroys the datapath in the kernel, and
1617 * with the netdev-based datapath, it tears down the data structures that
1618 * represent the datapath.
1620 * The datapath should not be currently open as an ofproto. */
1622 ofproto_delete(const char *name, const char *type)
1624 const struct ofproto_class *class = ofproto_class_find__(type);
1625 return (!class ? EAFNOSUPPORT
1626 : !class->del ? EACCES
1627 : class->del(type, name));
1631 process_port_change(struct ofproto *ofproto, int error, char *devname)
1633 if (error == ENOBUFS) {
1634 reinit_ports(ofproto);
1635 } else if (!error) {
1636 update_port(ofproto, devname);
1642 ofproto_type_run(const char *datapath_type)
1644 const struct ofproto_class *class;
1647 datapath_type = ofproto_normalize_type(datapath_type);
1648 class = ofproto_class_find__(datapath_type);
1650 error = class->type_run ? class->type_run(datapath_type) : 0;
1651 if (error && error != EAGAIN) {
1652 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1653 datapath_type, ovs_strerror(error));
1659 ofproto_type_wait(const char *datapath_type)
1661 const struct ofproto_class *class;
1663 datapath_type = ofproto_normalize_type(datapath_type);
1664 class = ofproto_class_find__(datapath_type);
1666 if (class->type_wait) {
1667 class->type_wait(datapath_type);
1672 ofproto_run(struct ofproto *p)
1677 error = p->ofproto_class->run(p);
1678 if (error && error != EAGAIN) {
1679 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, ovs_strerror(error));
1682 run_rule_executes(p);
1684 /* Restore the eviction group heap invariant occasionally. */
1685 if (p->eviction_group_timer < time_msec()) {
1688 p->eviction_group_timer = time_msec() + 1000;
1690 for (i = 0; i < p->n_tables; i++) {
1691 struct oftable *table = &p->tables[i];
1692 struct eviction_group *evg;
1695 if (!table->eviction) {
1699 if (table->n_flows > 100000) {
1700 static struct vlog_rate_limit count_rl =
1701 VLOG_RATE_LIMIT_INIT(1, 1);
1702 VLOG_WARN_RL(&count_rl, "Table %"PRIuSIZE" has an excessive"
1703 " number of rules: %d", i, table->n_flows);
1706 ovs_mutex_lock(&ofproto_mutex);
1707 CLS_FOR_EACH (rule, cr, &table->cls) {
1708 if (rule->idle_timeout || rule->hard_timeout) {
1709 if (!rule->eviction_group) {
1710 eviction_group_add_rule(rule);
1712 heap_raw_change(&rule->evg_node,
1713 rule_eviction_priority(p, rule));
1718 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
1719 heap_rebuild(&evg->rules);
1721 ovs_mutex_unlock(&ofproto_mutex);
1725 if (p->ofproto_class->port_poll) {
1728 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1729 process_port_change(p, error, devname);
1733 new_seq = seq_read(connectivity_seq_get());
1734 if (new_seq != p->change_seq) {
1735 struct sset devnames;
1736 const char *devname;
1737 struct ofport *ofport;
1739 /* Update OpenFlow port status for any port whose netdev has changed.
1741 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1742 * destroyed, so it's not safe to update ports directly from the
1743 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1744 * need this two-phase approach. */
1745 sset_init(&devnames);
1746 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1747 uint64_t port_change_seq;
1749 port_change_seq = netdev_get_change_seq(ofport->netdev);
1750 if (ofport->change_seq != port_change_seq) {
1751 ofport->change_seq = port_change_seq;
1752 sset_add(&devnames, netdev_get_name(ofport->netdev));
1755 SSET_FOR_EACH (devname, &devnames) {
1756 update_port(p, devname);
1758 sset_destroy(&devnames);
1760 p->change_seq = new_seq;
1763 connmgr_run(p->connmgr, handle_openflow);
1769 ofproto_wait(struct ofproto *p)
1771 p->ofproto_class->wait(p);
1772 if (p->ofproto_class->port_poll_wait) {
1773 p->ofproto_class->port_poll_wait(p);
1775 seq_wait(connectivity_seq_get(), p->change_seq);
1776 connmgr_wait(p->connmgr);
1780 ofproto_is_alive(const struct ofproto *p)
1782 return connmgr_has_controllers(p->connmgr);
1785 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1786 * memory_report(). */
1788 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1790 const struct oftable *table;
1791 unsigned int n_rules;
1793 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1796 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1797 n_rules += table->n_flows;
1799 simap_increase(usage, "rules", n_rules);
1801 if (ofproto->ofproto_class->get_memory_usage) {
1802 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1805 connmgr_get_memory_usage(ofproto->connmgr, usage);
1809 ofproto_type_get_memory_usage(const char *datapath_type, struct simap *usage)
1811 const struct ofproto_class *class;
1813 datapath_type = ofproto_normalize_type(datapath_type);
1814 class = ofproto_class_find__(datapath_type);
1816 if (class && class->type_get_memory_usage) {
1817 class->type_get_memory_usage(datapath_type, usage);
1822 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1825 connmgr_get_controller_info(ofproto->connmgr, info);
1829 ofproto_free_ofproto_controller_info(struct shash *info)
1831 connmgr_free_controller_info(info);
1834 /* Makes a deep copy of 'old' into 'port'. */
1836 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1838 port->name = xstrdup(old->name);
1839 port->type = xstrdup(old->type);
1840 port->ofp_port = old->ofp_port;
1843 /* Frees memory allocated to members of 'ofproto_port'.
1845 * Do not call this function on an ofproto_port obtained from
1846 * ofproto_port_dump_next(): that function retains ownership of the data in the
1849 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1851 free(ofproto_port->name);
1852 free(ofproto_port->type);
1855 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1857 * This function provides no status indication. An error status for the entire
1858 * dump operation is provided when it is completed by calling
1859 * ofproto_port_dump_done().
1862 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1863 const struct ofproto *ofproto)
1865 dump->ofproto = ofproto;
1866 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1870 /* Attempts to retrieve another port from 'dump', which must have been created
1871 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1872 * 'port' and returns true. On failure, returns false.
1874 * Failure might indicate an actual error or merely that the last port has been
1875 * dumped. An error status for the entire dump operation is provided when it
1876 * is completed by calling ofproto_port_dump_done().
1878 * The ofproto owns the data stored in 'port'. It will remain valid until at
1879 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1880 * ofproto_port_dump_done(). */
1882 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1883 struct ofproto_port *port)
1885 const struct ofproto *ofproto = dump->ofproto;
1891 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1894 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1900 /* Completes port table dump operation 'dump', which must have been created
1901 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1902 * error-free, otherwise a positive errno value describing the problem. */
1904 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1906 const struct ofproto *ofproto = dump->ofproto;
1908 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1911 return dump->error == EOF ? 0 : dump->error;
1914 /* Returns the type to pass to netdev_open() when a datapath of type
1915 * 'datapath_type' has a port of type 'port_type', for a few special
1916 * cases when a netdev type differs from a port type. For example, when
1917 * using the userspace datapath, a port of type "internal" needs to be
1920 * Returns either 'type' itself or a string literal, which must not be
1923 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1925 const struct ofproto_class *class;
1927 datapath_type = ofproto_normalize_type(datapath_type);
1928 class = ofproto_class_find__(datapath_type);
1933 return (class->port_open_type
1934 ? class->port_open_type(datapath_type, port_type)
1938 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1939 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1940 * the port's OpenFlow port number.
1942 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1943 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1944 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1945 * 'ofp_portp' is non-null). */
1947 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1948 ofp_port_t *ofp_portp)
1950 ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1953 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1955 const char *netdev_name = netdev_get_name(netdev);
1957 simap_put(&ofproto->ofp_requests, netdev_name,
1958 ofp_to_u16(ofp_port));
1959 error = update_port(ofproto, netdev_name);
1962 *ofp_portp = OFPP_NONE;
1964 struct ofproto_port ofproto_port;
1966 error = ofproto_port_query_by_name(ofproto,
1967 netdev_get_name(netdev),
1970 *ofp_portp = ofproto_port.ofp_port;
1971 ofproto_port_destroy(&ofproto_port);
1978 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1979 * initializes '*port' appropriately; on failure, returns a positive errno
1982 * The caller owns the data in 'ofproto_port' and must free it with
1983 * ofproto_port_destroy() when it is no longer needed. */
1985 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1986 struct ofproto_port *port)
1990 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1992 memset(port, 0, sizeof *port);
1997 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1998 * Returns 0 if successful, otherwise a positive errno. */
2000 ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port)
2002 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2003 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
2004 struct simap_node *ofp_request_node;
2007 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
2008 if (ofp_request_node) {
2009 simap_delete(&ofproto->ofp_requests, ofp_request_node);
2012 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
2013 if (!error && ofport) {
2014 /* 'name' is the netdev's name and update_port() is going to close the
2015 * netdev. Just in case update_port() refers to 'name' after it
2016 * destroys 'ofport', make a copy of it around the update_port()
2018 char *devname = xstrdup(name);
2019 update_port(ofproto, devname);
2026 flow_mod_init(struct ofputil_flow_mod *fm,
2027 const struct match *match, int priority,
2028 const struct ofpact *ofpacts, size_t ofpacts_len,
2029 enum ofp_flow_mod_command command)
2031 *fm = (struct ofputil_flow_mod) {
2033 .priority = priority,
2036 .buffer_id = UINT32_MAX,
2037 .out_port = OFPP_ANY,
2038 .out_group = OFPG_ANY,
2039 .ofpacts = CONST_CAST(struct ofpact *, ofpacts),
2040 .ofpacts_len = ofpacts_len,
2041 .delete_reason = OFPRR_DELETE,
2046 simple_flow_mod(struct ofproto *ofproto,
2047 const struct match *match, int priority,
2048 const struct ofpact *ofpacts, size_t ofpacts_len,
2049 enum ofp_flow_mod_command command)
2051 struct ofproto_flow_mod ofm;
2053 flow_mod_init(&ofm.fm, match, priority, ofpacts, ofpacts_len, command);
2055 return handle_flow_mod__(ofproto, &ofm, NULL);
2058 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
2059 * performs the 'n_actions' actions in 'actions'. The new flow will not
2062 * If cls_rule->priority is in the range of priorities supported by OpenFlow
2063 * (0...65535, inclusive) then the flow will be visible to OpenFlow
2064 * controllers; otherwise, it will be hidden.
2066 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
2068 * This is a helper function for in-band control and fail-open. */
2070 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
2072 const struct ofpact *ofpacts, size_t ofpacts_len)
2073 OVS_EXCLUDED(ofproto_mutex)
2075 const struct rule *rule;
2078 /* First do a cheap check whether the rule we're looking for already exists
2079 * with the actions that we want. If it does, then we're done. */
2080 rule = rule_from_cls_rule(classifier_find_match_exactly(
2081 &ofproto->tables[0].cls, match, priority,
2084 const struct rule_actions *actions = rule_get_actions(rule);
2085 must_add = !ofpacts_equal(actions->ofpacts, actions->ofpacts_len,
2086 ofpacts, ofpacts_len);
2091 /* If there's no such rule or the rule doesn't have the actions we want,
2092 * fall back to a executing a full flow mod. We can't optimize this at
2093 * all because we didn't take enough locks above to ensure that the flow
2094 * table didn't already change beneath us. */
2096 simple_flow_mod(ofproto, match, priority, ofpacts, ofpacts_len,
2097 OFPFC_MODIFY_STRICT);
2101 /* Executes the flow modification specified in 'fm'. Returns 0 on success, or
2102 * an OFPERR_* OpenFlow error code on failure.
2104 * This is a helper function for in-band control and fail-open and the "learn"
2107 ofproto_flow_mod(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
2108 OVS_EXCLUDED(ofproto_mutex)
2110 struct ofputil_flow_mod *fm = &ofm->fm;
2112 /* Optimize for the most common case of a repeated learn action.
2113 * If an identical flow already exists we only need to update its
2114 * 'modified' time. */
2115 if (fm->command == OFPFC_MODIFY_STRICT && fm->table_id != OFPTT_ALL
2116 && !(fm->flags & OFPUTIL_FF_RESET_COUNTS)) {
2117 struct oftable *table = &ofproto->tables[fm->table_id];
2121 rule = rule_from_cls_rule(classifier_find_match_exactly(
2122 &table->cls, &fm->match, fm->priority,
2125 /* Reading many of the rule fields and writing on 'modified'
2126 * requires the rule->mutex. Also, rule->actions may change
2127 * if rule->mutex is not held. */
2128 const struct rule_actions *actions;
2130 ovs_mutex_lock(&rule->mutex);
2131 actions = rule_get_actions(rule);
2132 if (rule->idle_timeout == fm->idle_timeout
2133 && rule->hard_timeout == fm->hard_timeout
2134 && rule->importance == fm->importance
2135 && rule->flags == (fm->flags & OFPUTIL_FF_STATE)
2136 && (!fm->modify_cookie || (fm->new_cookie == rule->flow_cookie))
2137 && ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
2138 actions->ofpacts, actions->ofpacts_len)) {
2139 /* Rule already exists and need not change, only update the
2140 modified timestamp. */
2141 rule->modified = time_msec();
2144 ovs_mutex_unlock(&rule->mutex);
2152 return handle_flow_mod__(ofproto, ofm, NULL);
2155 /* Searches for a rule with matching criteria exactly equal to 'target' in
2156 * ofproto's table 0 and, if it finds one, deletes it.
2158 * This is a helper function for in-band control and fail-open. */
2160 ofproto_delete_flow(struct ofproto *ofproto,
2161 const struct match *target, int priority)
2162 OVS_EXCLUDED(ofproto_mutex)
2164 struct classifier *cls = &ofproto->tables[0].cls;
2167 /* First do a cheap check whether the rule we're looking for has already
2168 * been deleted. If so, then we're done. */
2169 rule = rule_from_cls_rule(classifier_find_match_exactly(
2170 cls, target, priority, CLS_MAX_VERSION));
2175 /* Execute a flow mod. We can't optimize this at all because we didn't
2176 * take enough locks above to ensure that the flow table didn't already
2177 * change beneath us. */
2178 simple_flow_mod(ofproto, target, priority, NULL, 0, OFPFC_DELETE_STRICT);
2181 /* Delete all of the flows from all of ofproto's flow tables, then reintroduce
2182 * the flows required by in-band control and fail-open. */
2184 ofproto_flush_flows(struct ofproto *ofproto)
2186 COVERAGE_INC(ofproto_flush);
2187 ofproto_flush__(ofproto);
2188 connmgr_flushed(ofproto->connmgr);
2192 reinit_ports(struct ofproto *p)
2194 struct ofproto_port_dump dump;
2195 struct sset devnames;
2196 struct ofport *ofport;
2197 struct ofproto_port ofproto_port;
2198 const char *devname;
2200 COVERAGE_INC(ofproto_reinit_ports);
2202 sset_init(&devnames);
2203 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2204 sset_add(&devnames, netdev_get_name(ofport->netdev));
2206 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2207 sset_add(&devnames, ofproto_port.name);
2210 SSET_FOR_EACH (devname, &devnames) {
2211 update_port(p, devname);
2213 sset_destroy(&devnames);
2217 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
2221 port_idx = simap_get(&ofproto->ofp_requests, netdev_name);
2222 port_idx = port_idx ? port_idx : UINT16_MAX;
2224 if (port_idx >= ofproto->max_ports
2225 || ofport_get_usage(ofproto, u16_to_ofp(port_idx)) == LLONG_MAX) {
2226 uint16_t lru_ofport = 0, end_port_no = ofproto->alloc_port_no;
2227 long long int last_used_at, lru = LLONG_MAX;
2229 /* Search for a free OpenFlow port number. We try not to
2230 * immediately reuse them to prevent problems due to old
2233 * We limit the automatically assigned port numbers to the lower half
2234 * of the port range, to reserve the upper half for assignment by
2237 if (++ofproto->alloc_port_no >= MIN(ofproto->max_ports, 32768)) {
2238 ofproto->alloc_port_no = 1;
2240 last_used_at = ofport_get_usage(ofproto,
2241 u16_to_ofp(ofproto->alloc_port_no));
2242 if (!last_used_at) {
2243 port_idx = ofproto->alloc_port_no;
2245 } else if ( last_used_at < time_msec() - 60*60*1000) {
2246 /* If the port with ofport 'ofproto->alloc_port_no' was deleted
2247 * more than an hour ago, consider it usable. */
2248 ofport_remove_usage(ofproto,
2249 u16_to_ofp(ofproto->alloc_port_no));
2250 port_idx = ofproto->alloc_port_no;
2252 } else if (last_used_at < lru) {
2254 lru_ofport = ofproto->alloc_port_no;
2257 if (ofproto->alloc_port_no == end_port_no) {
2259 port_idx = lru_ofport;
2266 ofport_set_usage(ofproto, u16_to_ofp(port_idx), LLONG_MAX);
2267 return u16_to_ofp(port_idx);
2271 dealloc_ofp_port(struct ofproto *ofproto, ofp_port_t ofp_port)
2273 if (ofp_to_u16(ofp_port) < ofproto->max_ports) {
2274 ofport_set_usage(ofproto, ofp_port, time_msec());
2278 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
2279 * pointer if the netdev cannot be opened. On success, also fills in
2281 static struct netdev *
2282 ofport_open(struct ofproto *ofproto,
2283 struct ofproto_port *ofproto_port,
2284 struct ofputil_phy_port *pp)
2286 enum netdev_flags flags;
2287 struct netdev *netdev;
2290 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
2292 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
2293 "cannot be opened (%s)",
2295 ofproto_port->name, ofproto_port->ofp_port,
2296 ofproto_port->name, ovs_strerror(error));
2300 if (ofproto_port->ofp_port == OFPP_NONE) {
2301 if (!strcmp(ofproto->name, ofproto_port->name)) {
2302 ofproto_port->ofp_port = OFPP_LOCAL;
2304 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
2305 ofproto_port->name);
2308 pp->port_no = ofproto_port->ofp_port;
2309 netdev_get_etheraddr(netdev, &pp->hw_addr);
2310 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
2311 netdev_get_flags(netdev, &flags);
2312 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
2313 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
2314 netdev_get_features(netdev, &pp->curr, &pp->advertised,
2315 &pp->supported, &pp->peer);
2316 pp->curr_speed = netdev_features_to_bps(pp->curr, 0) / 1000;
2317 pp->max_speed = netdev_features_to_bps(pp->supported, 0) / 1000;
2322 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
2323 * port number, and 'config' bits other than OFPUTIL_PC_PORT_DOWN are
2326 ofport_equal(const struct ofputil_phy_port *a,
2327 const struct ofputil_phy_port *b)
2329 return (eth_addr_equals(a->hw_addr, b->hw_addr)
2330 && a->state == b->state
2331 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
2332 && a->curr == b->curr
2333 && a->advertised == b->advertised
2334 && a->supported == b->supported
2335 && a->peer == b->peer
2336 && a->curr_speed == b->curr_speed
2337 && a->max_speed == b->max_speed);
2340 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
2341 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
2342 * one with the same name or port number). */
2344 ofport_install(struct ofproto *p,
2345 struct netdev *netdev, const struct ofputil_phy_port *pp)
2347 const char *netdev_name = netdev_get_name(netdev);
2348 struct ofport *ofport;
2351 /* Create ofport. */
2352 ofport = p->ofproto_class->port_alloc();
2357 ofport->ofproto = p;
2358 ofport->netdev = netdev;
2359 ofport->change_seq = netdev_get_change_seq(netdev);
2361 ofport->ofp_port = pp->port_no;
2362 ofport->created = time_msec();
2364 /* Add port to 'p'. */
2365 hmap_insert(&p->ports, &ofport->hmap_node,
2366 hash_ofp_port(ofport->ofp_port));
2367 shash_add(&p->port_by_name, netdev_name, ofport);
2369 update_mtu(p, ofport);
2371 /* Let the ofproto_class initialize its private data. */
2372 error = p->ofproto_class->port_construct(ofport);
2376 connmgr_send_port_status(p->connmgr, NULL, pp, OFPPR_ADD);
2380 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
2381 p->name, netdev_name, ovs_strerror(error));
2383 ofport_destroy__(ofport);
2385 netdev_close(netdev);
2390 /* Removes 'ofport' from 'p' and destroys it. */
2392 ofport_remove(struct ofport *ofport)
2394 connmgr_send_port_status(ofport->ofproto->connmgr, NULL, &ofport->pp,
2396 ofport_destroy(ofport, true);
2399 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
2402 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
2404 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
2406 ofport_remove(port);
2410 /* Updates 'port' with new 'pp' description.
2412 * Does not handle a name or port number change. The caller must implement
2413 * such a change as a delete followed by an add. */
2415 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
2417 port->pp.hw_addr = pp->hw_addr;
2418 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
2419 | (pp->config & OFPUTIL_PC_PORT_DOWN));
2420 port->pp.state = ((port->pp.state & ~OFPUTIL_PS_LINK_DOWN)
2421 | (pp->state & OFPUTIL_PS_LINK_DOWN));
2422 port->pp.curr = pp->curr;
2423 port->pp.advertised = pp->advertised;
2424 port->pp.supported = pp->supported;
2425 port->pp.peer = pp->peer;
2426 port->pp.curr_speed = pp->curr_speed;
2427 port->pp.max_speed = pp->max_speed;
2429 connmgr_send_port_status(port->ofproto->connmgr, NULL,
2430 &port->pp, OFPPR_MODIFY);
2433 /* Update OpenFlow 'state' in 'port' and notify controller. */
2435 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
2437 if (port->pp.state != state) {
2438 port->pp.state = state;
2439 connmgr_send_port_status(port->ofproto->connmgr, NULL,
2440 &port->pp, OFPPR_MODIFY);
2445 ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port)
2447 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
2449 if (port->ofproto->ofproto_class->set_realdev) {
2450 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
2452 if (port->ofproto->ofproto_class->set_stp_port) {
2453 port->ofproto->ofproto_class->set_stp_port(port, NULL);
2455 if (port->ofproto->ofproto_class->set_rstp_port) {
2456 port->ofproto->ofproto_class->set_rstp_port(port, NULL);
2458 if (port->ofproto->ofproto_class->set_cfm) {
2459 port->ofproto->ofproto_class->set_cfm(port, NULL);
2461 if (port->ofproto->ofproto_class->bundle_remove) {
2462 port->ofproto->ofproto_class->bundle_remove(port);
2468 ofport_destroy__(struct ofport *port)
2470 struct ofproto *ofproto = port->ofproto;
2471 const char *name = netdev_get_name(port->netdev);
2473 hmap_remove(&ofproto->ports, &port->hmap_node);
2474 shash_delete(&ofproto->port_by_name,
2475 shash_find(&ofproto->port_by_name, name));
2477 netdev_close(port->netdev);
2478 ofproto->ofproto_class->port_dealloc(port);
2482 ofport_destroy(struct ofport *port, bool del)
2485 dealloc_ofp_port(port->ofproto, port->ofp_port);
2486 port->ofproto->ofproto_class->port_destruct(port, del);
2487 ofport_destroy__(port);
2492 ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
2494 struct ofport *port;
2496 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_ofp_port(ofp_port),
2498 if (port->ofp_port == ofp_port) {
2505 static long long int
2506 ofport_get_usage(const struct ofproto *ofproto, ofp_port_t ofp_port)
2508 struct ofport_usage *usage;
2510 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2511 &ofproto->ofport_usage) {
2512 if (usage->ofp_port == ofp_port) {
2513 return usage->last_used;
2520 ofport_set_usage(struct ofproto *ofproto, ofp_port_t ofp_port,
2521 long long int last_used)
2523 struct ofport_usage *usage;
2524 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2525 &ofproto->ofport_usage) {
2526 if (usage->ofp_port == ofp_port) {
2527 usage->last_used = last_used;
2531 ovs_assert(last_used == LLONG_MAX);
2533 usage = xmalloc(sizeof *usage);
2534 usage->ofp_port = ofp_port;
2535 usage->last_used = last_used;
2536 hmap_insert(&ofproto->ofport_usage, &usage->hmap_node,
2537 hash_ofp_port(ofp_port));
2541 ofport_remove_usage(struct ofproto *ofproto, ofp_port_t ofp_port)
2543 struct ofport_usage *usage;
2544 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2545 &ofproto->ofport_usage) {
2546 if (usage->ofp_port == ofp_port) {
2547 hmap_remove(&ofproto->ofport_usage, &usage->hmap_node);
2555 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
2557 struct ofproto *ofproto = port->ofproto;
2560 if (ofproto->ofproto_class->port_get_stats) {
2561 error = ofproto->ofproto_class->port_get_stats(port, stats);
2570 update_port(struct ofproto *ofproto, const char *name)
2572 struct ofproto_port ofproto_port;
2573 struct ofputil_phy_port pp;
2574 struct netdev *netdev;
2575 struct ofport *port;
2578 COVERAGE_INC(ofproto_update_port);
2580 /* Fetch 'name''s location and properties from the datapath. */
2581 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
2582 ? ofport_open(ofproto, &ofproto_port, &pp)
2586 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
2587 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
2588 struct netdev *old_netdev = port->netdev;
2590 /* 'name' hasn't changed location. Any properties changed? */
2591 if (!ofport_equal(&port->pp, &pp)) {
2592 ofport_modified(port, &pp);
2595 update_mtu(ofproto, port);
2597 /* Install the newly opened netdev in case it has changed.
2598 * Don't close the old netdev yet in case port_modified has to
2599 * remove a retained reference to it.*/
2600 port->netdev = netdev;
2601 port->change_seq = netdev_get_change_seq(netdev);
2603 if (port->ofproto->ofproto_class->port_modified) {
2604 port->ofproto->ofproto_class->port_modified(port);
2607 netdev_close(old_netdev);
2609 /* If 'port' is nonnull then its name differs from 'name' and thus
2610 * we should delete it. If we think there's a port named 'name'
2611 * then its port number must be wrong now so delete it too. */
2613 ofport_remove(port);
2615 ofport_remove_with_name(ofproto, name);
2616 error = ofport_install(ofproto, netdev, &pp);
2619 /* Any port named 'name' is gone now. */
2620 ofport_remove_with_name(ofproto, name);
2622 ofproto_port_destroy(&ofproto_port);
2628 init_ports(struct ofproto *p)
2630 struct ofproto_port_dump dump;
2631 struct ofproto_port ofproto_port;
2632 struct shash_node *node, *next;
2634 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2635 const char *name = ofproto_port.name;
2637 if (shash_find(&p->port_by_name, name)) {
2638 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2641 struct ofputil_phy_port pp;
2642 struct netdev *netdev;
2644 /* Check if an OpenFlow port number had been requested. */
2645 node = shash_find(&init_ofp_ports, name);
2647 const struct iface_hint *iface_hint = node->data;
2648 simap_put(&p->ofp_requests, name,
2649 ofp_to_u16(iface_hint->ofp_port));
2652 netdev = ofport_open(p, &ofproto_port, &pp);
2654 ofport_install(p, netdev, &pp);
2655 if (ofp_to_u16(ofproto_port.ofp_port) < p->max_ports) {
2656 p->alloc_port_no = MAX(p->alloc_port_no,
2657 ofp_to_u16(ofproto_port.ofp_port));
2663 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2664 struct iface_hint *iface_hint = node->data;
2666 if (!strcmp(iface_hint->br_name, p->name)) {
2667 free(iface_hint->br_name);
2668 free(iface_hint->br_type);
2670 shash_delete(&init_ofp_ports, node);
2677 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2678 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2680 find_min_mtu(struct ofproto *p)
2682 struct ofport *ofport;
2685 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2686 struct netdev *netdev = ofport->netdev;
2689 /* Skip any internal ports, since that's what we're trying to
2691 if (!strcmp(netdev_get_type(netdev), "internal")) {
2695 if (netdev_get_mtu(netdev, &dev_mtu)) {
2698 if (!mtu || dev_mtu < mtu) {
2703 return mtu ? mtu: ETH_PAYLOAD_MAX;
2706 /* Update MTU of all datapath devices on 'p' to the minimum of the
2707 * non-datapath ports in event of 'port' added or changed. */
2709 update_mtu(struct ofproto *p, struct ofport *port)
2711 struct ofport *ofport;
2712 struct netdev *netdev = port->netdev;
2713 int dev_mtu, old_min;
2715 if (netdev_get_mtu(netdev, &dev_mtu)) {
2719 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2720 if (dev_mtu > p->min_mtu) {
2721 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2722 dev_mtu = p->min_mtu;
2725 port->mtu = dev_mtu;
2729 /* For non-internal port find new min mtu. */
2730 old_min = p->min_mtu;
2731 port->mtu = dev_mtu;
2732 p->min_mtu = find_min_mtu(p);
2733 if (p->min_mtu == old_min) {
2737 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2738 struct netdev *netdev = ofport->netdev;
2740 if (!strcmp(netdev_get_type(netdev), "internal")) {
2741 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2742 ofport->mtu = p->min_mtu;
2749 ofproto_rule_destroy__(struct rule *rule)
2750 OVS_NO_THREAD_SAFETY_ANALYSIS
2752 cls_rule_destroy(CONST_CAST(struct cls_rule *, &rule->cr));
2753 rule_actions_destroy(rule_get_actions(rule));
2754 ovs_mutex_destroy(&rule->mutex);
2755 rule->ofproto->ofproto_class->rule_dealloc(rule);
2759 rule_destroy_cb(struct rule *rule)
2760 OVS_NO_THREAD_SAFETY_ANALYSIS
2762 /* Send rule removed if needed. */
2763 if (rule->flags & OFPUTIL_FF_SEND_FLOW_REM
2764 && rule->removed_reason != OVS_OFPRR_NONE
2765 && !rule_is_hidden(rule)) {
2766 ofproto_rule_send_removed(rule);
2768 rule->ofproto->ofproto_class->rule_destruct(rule);
2769 ofproto_rule_destroy__(rule);
2773 ofproto_rule_ref(struct rule *rule)
2776 ovs_refcount_ref(&rule->ref_count);
2781 ofproto_rule_try_ref(struct rule *rule)
2784 return ovs_refcount_try_ref_rcu(&rule->ref_count);
2789 /* Decrements 'rule''s ref_count and schedules 'rule' to be destroyed if the
2790 * ref_count reaches 0.
2792 * Use of RCU allows short term use (between RCU quiescent periods) without
2793 * keeping a reference. A reference must be taken if the rule needs to
2794 * stay around accross the RCU quiescent periods. */
2796 ofproto_rule_unref(struct rule *rule)
2798 if (rule && ovs_refcount_unref_relaxed(&rule->ref_count) == 1) {
2799 ovsrcu_postpone(rule_destroy_cb, rule);
2804 remove_rule_rcu__(struct rule *rule)
2805 OVS_REQUIRES(ofproto_mutex)
2807 struct ofproto *ofproto = rule->ofproto;
2808 struct oftable *table = &ofproto->tables[rule->table_id];
2810 ovs_assert(!cls_rule_visible_in_version(&rule->cr, CLS_MAX_VERSION));
2811 if (!classifier_remove(&table->cls, &rule->cr)) {
2814 ofproto->ofproto_class->rule_delete(rule);
2815 ofproto_rule_unref(rule);
2819 remove_rule_rcu(struct rule *rule)
2820 OVS_EXCLUDED(ofproto_mutex)
2822 ovs_mutex_lock(&ofproto_mutex);
2823 remove_rule_rcu__(rule);
2824 ovs_mutex_unlock(&ofproto_mutex);
2827 /* Removes and deletes rules from a NULL-terminated array of rule pointers. */
2829 remove_rules_rcu(struct rule **rules)
2830 OVS_EXCLUDED(ofproto_mutex)
2832 struct rule **orig_rules = rules;
2835 struct ofproto *ofproto = rules[0]->ofproto;
2836 unsigned long tables[BITMAP_N_LONGS(256)];
2840 memset(tables, 0, sizeof tables);
2842 ovs_mutex_lock(&ofproto_mutex);
2843 while ((rule = *rules++)) {
2844 /* Defer once for each new table. This defers the subtable cleanup
2845 * until later, so that when removing large number of flows the
2846 * operation is faster. */
2847 if (!bitmap_is_set(tables, rule->table_id)) {
2848 struct classifier *cls = &ofproto->tables[rule->table_id].cls;
2850 bitmap_set1(tables, rule->table_id);
2851 classifier_defer(cls);
2853 remove_rule_rcu__(rule);
2856 BITMAP_FOR_EACH_1(table_id, 256, tables) {
2857 struct classifier *cls = &ofproto->tables[table_id].cls;
2859 classifier_publish(cls);
2861 ovs_mutex_unlock(&ofproto_mutex);
2868 ofproto_group_ref(struct ofgroup *group)
2871 ovs_refcount_ref(&group->ref_count);
2876 ofproto_group_unref(struct ofgroup *group)
2878 if (group && ovs_refcount_unref(&group->ref_count) == 1) {
2879 group->ofproto->ofproto_class->group_destruct(group);
2880 ofputil_bucket_list_destroy(&group->buckets);
2881 group->ofproto->ofproto_class->group_dealloc(group);
2885 static uint32_t get_provider_meter_id(const struct ofproto *,
2886 uint32_t of_meter_id);
2888 /* Creates and returns a new 'struct rule_actions', whose actions are a copy
2889 * of from the 'ofpacts_len' bytes of 'ofpacts'. */
2890 const struct rule_actions *
2891 rule_actions_create(const struct ofpact *ofpacts, size_t ofpacts_len)
2893 struct rule_actions *actions;
2895 actions = xmalloc(sizeof *actions + ofpacts_len);
2896 actions->ofpacts_len = ofpacts_len;
2897 actions->has_meter = ofpacts_get_meter(ofpacts, ofpacts_len) != 0;
2898 memcpy(actions->ofpacts, ofpacts, ofpacts_len);
2900 actions->has_learn_with_delete = (next_learn_with_delete(actions, NULL)
2906 /* Free the actions after the RCU quiescent period is reached. */
2908 rule_actions_destroy(const struct rule_actions *actions)
2911 ovsrcu_postpone(free, CONST_CAST(struct rule_actions *, actions));
2915 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2916 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2918 ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port)
2919 OVS_REQUIRES(ofproto_mutex)
2921 if (port == OFPP_ANY) {
2924 const struct rule_actions *actions = rule_get_actions(rule);
2925 return ofpacts_output_to_port(actions->ofpacts,
2926 actions->ofpacts_len, port);
2930 /* Returns true if 'rule' has group and equals group_id. */
2932 ofproto_rule_has_out_group(const struct rule *rule, uint32_t group_id)
2933 OVS_REQUIRES(ofproto_mutex)
2935 if (group_id == OFPG_ANY) {
2938 const struct rule_actions *actions = rule_get_actions(rule);
2939 return ofpacts_output_to_group(actions->ofpacts,
2940 actions->ofpacts_len, group_id);
2945 rule_execute_destroy(struct rule_execute *e)
2947 ofproto_rule_unref(e->rule);
2948 ovs_list_remove(&e->list_node);
2952 /* Executes all "rule_execute" operations queued up in ofproto->rule_executes,
2953 * by passing them to the ofproto provider. */
2955 run_rule_executes(struct ofproto *ofproto)
2956 OVS_EXCLUDED(ofproto_mutex)
2958 struct rule_execute *e, *next;
2959 struct ovs_list executes;
2961 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2962 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2965 flow_extract(e->packet, &flow);
2966 flow.in_port.ofp_port = e->in_port;
2967 ofproto->ofproto_class->rule_execute(e->rule, &flow, e->packet);
2969 rule_execute_destroy(e);
2973 /* Destroys and discards all "rule_execute" operations queued up in
2974 * ofproto->rule_executes. */
2976 destroy_rule_executes(struct ofproto *ofproto)
2978 struct rule_execute *e, *next;
2979 struct ovs_list executes;
2981 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2982 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2983 dp_packet_delete(e->packet);
2984 rule_execute_destroy(e);
2989 rule_is_readonly(const struct rule *rule)
2991 const struct oftable *table = &rule->ofproto->tables[rule->table_id];
2992 return (table->flags & OFTABLE_READONLY) != 0;
2996 hash_learned_cookie(ovs_be64 cookie_, uint8_t table_id)
2998 uint64_t cookie = (OVS_FORCE uint64_t) cookie_;
2999 return hash_3words(cookie, cookie >> 32, table_id);
3003 learned_cookies_update_one__(struct ofproto *ofproto,
3004 const struct ofpact_learn *learn,
3005 int delta, struct ovs_list *dead_cookies)
3006 OVS_REQUIRES(ofproto_mutex)
3008 uint32_t hash = hash_learned_cookie(learn->cookie, learn->table_id);
3009 struct learned_cookie *c;
3011 HMAP_FOR_EACH_WITH_HASH (c, u.hmap_node, hash, &ofproto->learned_cookies) {
3012 if (c->cookie == learn->cookie && c->table_id == learn->table_id) {
3014 ovs_assert(c->n >= 0);
3017 hmap_remove(&ofproto->learned_cookies, &c->u.hmap_node);
3018 ovs_list_push_back(dead_cookies, &c->u.list_node);
3025 ovs_assert(delta > 0);
3026 c = xmalloc(sizeof *c);
3027 hmap_insert(&ofproto->learned_cookies, &c->u.hmap_node, hash);
3028 c->cookie = learn->cookie;
3029 c->table_id = learn->table_id;
3033 static const struct ofpact_learn *
3034 next_learn_with_delete(const struct rule_actions *actions,
3035 const struct ofpact_learn *start)
3037 const struct ofpact *pos;
3039 for (pos = start ? ofpact_next(&start->ofpact) : actions->ofpacts;
3040 pos < ofpact_end(actions->ofpacts, actions->ofpacts_len);
3041 pos = ofpact_next(pos)) {
3042 if (pos->type == OFPACT_LEARN) {
3043 const struct ofpact_learn *learn = ofpact_get_LEARN(pos);
3044 if (learn->flags & NX_LEARN_F_DELETE_LEARNED) {
3054 learned_cookies_update__(struct ofproto *ofproto,
3055 const struct rule_actions *actions,
3056 int delta, struct ovs_list *dead_cookies)
3057 OVS_REQUIRES(ofproto_mutex)
3059 if (actions->has_learn_with_delete) {
3060 const struct ofpact_learn *learn;
3062 for (learn = next_learn_with_delete(actions, NULL); learn;
3063 learn = next_learn_with_delete(actions, learn)) {
3064 learned_cookies_update_one__(ofproto, learn, delta, dead_cookies);
3070 learned_cookies_inc(struct ofproto *ofproto,
3071 const struct rule_actions *actions)
3072 OVS_REQUIRES(ofproto_mutex)
3074 learned_cookies_update__(ofproto, actions, +1, NULL);
3078 learned_cookies_dec(struct ofproto *ofproto,
3079 const struct rule_actions *actions,
3080 struct ovs_list *dead_cookies)
3081 OVS_REQUIRES(ofproto_mutex)
3083 learned_cookies_update__(ofproto, actions, -1, dead_cookies);
3087 learned_cookies_flush(struct ofproto *ofproto, struct ovs_list *dead_cookies)
3088 OVS_REQUIRES(ofproto_mutex)
3090 struct learned_cookie *c;
3092 LIST_FOR_EACH_POP (c, u.list_node, dead_cookies) {
3093 struct rule_criteria criteria;
3094 struct rule_collection rules;
3097 match_init_catchall(&match);
3098 rule_criteria_init(&criteria, c->table_id, &match, 0, CLS_MAX_VERSION,
3099 c->cookie, OVS_BE64_MAX, OFPP_ANY, OFPG_ANY);
3100 rule_criteria_require_rw(&criteria, false);
3101 collect_rules_loose(ofproto, &criteria, &rules);
3102 rule_criteria_destroy(&criteria);
3103 delete_flows__(&rules, OFPRR_DELETE, NULL);
3110 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
3112 ofconn_send_reply(ofconn, make_echo_reply(oh));
3117 query_tables(struct ofproto *ofproto,
3118 struct ofputil_table_features **featuresp,
3119 struct ofputil_table_stats **statsp)
3121 struct mf_bitmap rw_fields = oxm_writable_fields();
3122 struct mf_bitmap match = oxm_matchable_fields();
3123 struct mf_bitmap mask = oxm_maskable_fields();
3125 struct ofputil_table_features *features;
3126 struct ofputil_table_stats *stats;
3129 features = *featuresp = xcalloc(ofproto->n_tables, sizeof *features);
3130 for (i = 0; i < ofproto->n_tables; i++) {
3131 struct ofputil_table_features *f = &features[i];
3134 sprintf(f->name, "table%d", i);
3135 f->metadata_match = OVS_BE64_MAX;
3136 f->metadata_write = OVS_BE64_MAX;
3137 atomic_read_relaxed(&ofproto->tables[i].miss_config, &f->miss_config);
3138 f->max_entries = 1000000;
3140 bool more_tables = false;
3141 for (int j = i + 1; j < ofproto->n_tables; j++) {
3142 if (!(ofproto->tables[j].flags & OFTABLE_HIDDEN)) {
3143 bitmap_set1(f->nonmiss.next, j);
3147 f->nonmiss.instructions = (1u << N_OVS_INSTRUCTIONS) - 1;
3149 f->nonmiss.instructions &= ~(1u << OVSINST_OFPIT11_GOTO_TABLE);
3151 f->nonmiss.write.ofpacts = (UINT64_C(1) << N_OFPACTS) - 1;
3152 f->nonmiss.write.set_fields = rw_fields;
3153 f->nonmiss.apply = f->nonmiss.write;
3154 f->miss = f->nonmiss;
3158 f->wildcard = match;
3162 stats = *statsp = xcalloc(ofproto->n_tables, sizeof *stats);
3163 for (i = 0; i < ofproto->n_tables; i++) {
3164 struct ofputil_table_stats *s = &stats[i];
3167 s->active_count = ofproto->tables[i].n_flows;
3169 s->active_count -= connmgr_count_hidden_rules(
3177 ofproto->ofproto_class->query_tables(ofproto, features, stats);
3179 for (i = 0; i < ofproto->n_tables; i++) {
3180 const struct oftable *table = &ofproto->tables[i];
3181 struct ofputil_table_features *f = &features[i];
3184 ovs_strzcpy(f->name, table->name, sizeof f->name);
3187 if (table->max_flows < f->max_entries) {
3188 f->max_entries = table->max_flows;
3194 query_switch_features(struct ofproto *ofproto,
3195 bool *arp_match_ip, uint64_t *ofpacts)
3197 struct ofputil_table_features *features, *f;
3199 *arp_match_ip = false;
3202 query_tables(ofproto, &features, NULL);
3203 for (f = features; f < &features[ofproto->n_tables]; f++) {
3204 *ofpacts |= f->nonmiss.apply.ofpacts | f->miss.apply.ofpacts;
3205 if (bitmap_is_set(f->match.bm, MFF_ARP_SPA) ||
3206 bitmap_is_set(f->match.bm, MFF_ARP_TPA)) {
3207 *arp_match_ip = true;
3213 ovs_assert(*ofpacts & (UINT64_C(1) << OFPACT_OUTPUT));
3217 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
3219 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3220 struct ofputil_switch_features features;
3221 struct ofport *port;
3225 query_switch_features(ofproto, &arp_match_ip, &features.ofpacts);
3227 features.datapath_id = ofproto->datapath_id;
3228 features.n_buffers = pktbuf_capacity();
3229 features.n_tables = ofproto_get_n_visible_tables(ofproto);
3230 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
3231 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS |
3232 OFPUTIL_C_GROUP_STATS);
3234 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
3236 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
3237 features.auxiliary_id = 0;
3238 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
3240 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3241 ofputil_put_switch_features_port(&port->pp, b);
3244 ofconn_send_reply(ofconn, b);
3249 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
3251 struct ofputil_switch_config config;
3252 config.frag = ofconn_get_ofproto(ofconn)->frag_handling;
3253 config.invalid_ttl_to_controller
3254 = ofconn_get_invalid_ttl_to_controller(ofconn);
3255 config.miss_send_len = ofconn_get_miss_send_len(ofconn);
3257 ofconn_send_reply(ofconn, ofputil_encode_get_config_reply(oh, &config));
3263 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
3265 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3266 struct ofputil_switch_config config;
3269 error = ofputil_decode_set_config(oh, &config);
3274 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
3275 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
3276 enum ofputil_frag_handling cur = ofproto->frag_handling;
3277 enum ofputil_frag_handling next = config.frag;
3280 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
3281 ofproto->frag_handling = next;
3283 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
3285 ofputil_frag_handling_to_string(next));
3290 if (config.invalid_ttl_to_controller >= 0) {
3291 ofconn_set_invalid_ttl_to_controller(ofconn,
3292 config.invalid_ttl_to_controller);
3295 ofconn_set_miss_send_len(ofconn, config.miss_send_len);
3300 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3301 * error message code for the caller to propagate upward. Otherwise, returns
3304 * The log message mentions 'msg_type'. */
3306 reject_slave_controller(struct ofconn *ofconn)
3308 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
3309 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
3310 return OFPERR_OFPBRC_IS_SLAVE;
3316 /* Checks that the 'ofpacts_len' bytes of action in 'ofpacts' are appropriate
3319 * - If they use a meter, then 'ofproto' has that meter configured.
3321 * - If they use any groups, then 'ofproto' has that group configured.
3323 * Returns 0 if successful, otherwise an OpenFlow error. */
3325 ofproto_check_ofpacts(struct ofproto *ofproto,
3326 const struct ofpact ofpacts[], size_t ofpacts_len)
3328 const struct ofpact *a;
3331 mid = ofpacts_get_meter(ofpacts, ofpacts_len);
3332 if (mid && get_provider_meter_id(ofproto, mid) == UINT32_MAX) {
3333 return OFPERR_OFPMMFC_INVALID_METER;
3336 OFPACT_FOR_EACH_FLATTENED (a, ofpacts, ofpacts_len) {
3337 if (a->type == OFPACT_GROUP
3338 && !ofproto_group_exists(ofproto, ofpact_get_GROUP(a)->group_id)) {
3339 return OFPERR_OFPBAC_BAD_OUT_GROUP;
3347 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
3349 struct ofproto *p = ofconn_get_ofproto(ofconn);
3350 struct ofputil_packet_out po;
3351 struct dp_packet *payload;
3352 uint64_t ofpacts_stub[1024 / 8];
3353 struct ofpbuf ofpacts;
3357 COVERAGE_INC(ofproto_packet_out);
3359 error = reject_slave_controller(ofconn);
3364 /* Decode message. */
3365 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3366 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
3368 goto exit_free_ofpacts;
3370 if (ofp_to_u16(po.in_port) >= p->max_ports
3371 && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) {
3372 error = OFPERR_OFPBRC_BAD_PORT;
3373 goto exit_free_ofpacts;
3377 if (po.buffer_id != UINT32_MAX) {
3378 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
3380 goto exit_free_ofpacts;
3383 /* Ensure that the L3 header is 32-bit aligned. */
3384 payload = dp_packet_clone_data_with_headroom(po.packet, po.packet_len, 2);
3387 /* Verify actions against packet, then send packet if successful. */
3388 flow_extract(payload, &flow);
3389 flow.in_port.ofp_port = po.in_port;
3391 /* Check actions like for flow mods. We pass a 'table_id' of 0 to
3392 * ofproto_check_consistency(), which isn't strictly correct because these
3393 * actions aren't in any table. This is OK as 'table_id' is only used to
3394 * check instructions (e.g., goto-table), which can't appear on the action
3395 * list of a packet-out. */
3396 error = ofpacts_check_consistency(po.ofpacts, po.ofpacts_len,
3397 &flow, u16_to_ofp(p->max_ports),
3399 ofconn_get_protocol(ofconn));
3401 error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len);
3403 error = p->ofproto_class->packet_out(p, payload, &flow,
3404 po.ofpacts, po.ofpacts_len);
3407 dp_packet_delete(payload);
3410 ofpbuf_uninit(&ofpacts);
3416 handle_nxt_resume(struct ofconn *ofconn, const struct ofp_header *oh)
3418 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3419 struct ofputil_packet_in_private pin;
3422 error = ofputil_decode_packet_in_private(oh, false, &pin, NULL, NULL);
3427 error = (ofproto->ofproto_class->nxt_resume
3428 ? ofproto->ofproto_class->nxt_resume(ofproto, &pin)
3429 : OFPERR_NXR_NOT_SUPPORTED);
3431 ofputil_packet_in_private_destroy(&pin);
3437 update_port_config(struct ofconn *ofconn, struct ofport *port,
3438 enum ofputil_port_config config,
3439 enum ofputil_port_config mask)
3441 enum ofputil_port_config toggle = (config ^ port->pp.config) & mask;
3443 if (toggle & OFPUTIL_PC_PORT_DOWN
3444 && (config & OFPUTIL_PC_PORT_DOWN
3445 ? netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL)
3446 : netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL))) {
3447 /* We tried to bring the port up or down, but it failed, so don't
3448 * update the "down" bit. */
3449 toggle &= ~OFPUTIL_PC_PORT_DOWN;
3453 enum ofputil_port_config old_config = port->pp.config;
3454 port->pp.config ^= toggle;
3455 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
3456 connmgr_send_port_status(port->ofproto->connmgr, ofconn, &port->pp,
3462 port_mod_start(struct ofconn *ofconn, struct ofputil_port_mod *pm,
3463 struct ofport **port)
3465 struct ofproto *p = ofconn_get_ofproto(ofconn);
3467 *port = ofproto_get_port(p, pm->port_no);
3469 return OFPERR_OFPPMFC_BAD_PORT;
3471 if (!eth_addr_equals((*port)->pp.hw_addr, pm->hw_addr)) {
3472 return OFPERR_OFPPMFC_BAD_HW_ADDR;
3478 port_mod_finish(struct ofconn *ofconn, struct ofputil_port_mod *pm,
3479 struct ofport *port)
3481 update_port_config(ofconn, port, pm->config, pm->mask);
3482 if (pm->advertise) {
3483 netdev_set_advertisements(port->netdev, pm->advertise);
3488 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3490 struct ofputil_port_mod pm;
3491 struct ofport *port;
3494 error = reject_slave_controller(ofconn);
3499 error = ofputil_decode_port_mod(oh, &pm, false);
3504 error = port_mod_start(ofconn, &pm, &port);
3506 port_mod_finish(ofconn, &pm, port);
3512 handle_desc_stats_request(struct ofconn *ofconn,
3513 const struct ofp_header *request)
3515 static const char *default_mfr_desc = "Nicira, Inc.";
3516 static const char *default_hw_desc = "Open vSwitch";
3517 static const char *default_sw_desc = VERSION;
3518 static const char *default_serial_desc = "None";
3519 static const char *default_dp_desc = "None";
3521 struct ofproto *p = ofconn_get_ofproto(ofconn);
3522 struct ofp_desc_stats *ods;
3525 msg = ofpraw_alloc_stats_reply(request, 0);
3526 ods = ofpbuf_put_zeros(msg, sizeof *ods);
3527 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
3528 sizeof ods->mfr_desc);
3529 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
3530 sizeof ods->hw_desc);
3531 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
3532 sizeof ods->sw_desc);
3533 ovs_strlcpy(ods->serial_num,
3534 p->serial_desc ? p->serial_desc : default_serial_desc,
3535 sizeof ods->serial_num);
3536 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
3537 sizeof ods->dp_desc);
3538 ofconn_send_reply(ofconn, msg);
3544 handle_table_stats_request(struct ofconn *ofconn,
3545 const struct ofp_header *request)
3547 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3548 struct ofputil_table_features *features;
3549 struct ofputil_table_stats *stats;
3550 struct ofpbuf *reply;
3553 query_tables(ofproto, &features, &stats);
3555 reply = ofputil_encode_table_stats_reply(request);
3556 for (i = 0; i < ofproto->n_tables; i++) {
3557 if (!(ofproto->tables[i].flags & OFTABLE_HIDDEN)) {
3558 ofputil_append_table_stats_reply(reply, &stats[i], &features[i]);
3561 ofconn_send_reply(ofconn, reply);
3570 handle_table_features_request(struct ofconn *ofconn,
3571 const struct ofp_header *request)
3573 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3574 struct ofpbuf msg = ofpbuf_const_initializer(request,
3575 ntohs(request->length));
3576 ofpraw_pull_assert(&msg);
3577 if (msg.size || ofpmp_more(request)) {
3578 return OFPERR_OFPTFFC_EPERM;
3581 struct ofputil_table_features *features;
3582 query_tables(ofproto, &features, NULL);
3584 struct ovs_list replies;
3585 ofpmp_init(&replies, request);
3586 for (size_t i = 0; i < ofproto->n_tables; i++) {
3587 if (!(ofproto->tables[i].flags & OFTABLE_HIDDEN)) {
3588 ofputil_append_table_features_reply(&features[i], &replies);
3591 ofconn_send_replies(ofconn, &replies);
3598 /* Returns the vacancy of 'oftable', a number that ranges from 0 (if the table
3599 * is full) to 100 (if the table is empty).
3601 * A table without a limit on flows is considered to be empty. */
3603 oftable_vacancy(const struct oftable *t)
3605 return (!t->max_flows ? 100
3606 : t->n_flows >= t->max_flows ? 0
3607 : (t->max_flows - t->n_flows) * 100.0 / t->max_flows);
3611 query_table_desc__(struct ofputil_table_desc *td,
3612 struct ofproto *ofproto, uint8_t table_id)
3614 const struct oftable *t = &ofproto->tables[table_id];
3616 td->table_id = table_id;
3617 td->eviction = (t->eviction & EVICTION_OPENFLOW
3618 ? OFPUTIL_TABLE_EVICTION_ON
3619 : OFPUTIL_TABLE_EVICTION_OFF);
3620 td->eviction_flags = OFPROTO_EVICTION_FLAGS;
3621 td->vacancy = (t->vacancy_event
3622 ? OFPUTIL_TABLE_VACANCY_ON
3623 : OFPUTIL_TABLE_VACANCY_OFF);
3624 td->table_vacancy.vacancy_down = t->vacancy_down;
3625 td->table_vacancy.vacancy_up = t->vacancy_up;
3626 td->table_vacancy.vacancy = oftable_vacancy(t);
3629 /* This function queries the database for dumping table-desc. */
3631 query_tables_desc(struct ofproto *ofproto, struct ofputil_table_desc **descp)
3633 struct ofputil_table_desc *table_desc;
3636 table_desc = *descp = xcalloc(ofproto->n_tables, sizeof *table_desc);
3637 for (i = 0; i < ofproto->n_tables; i++) {
3638 struct ofputil_table_desc *td = &table_desc[i];
3639 query_table_desc__(td, ofproto, i);
3643 /* Function to handle dump-table-desc request. */
3645 handle_table_desc_request(struct ofconn *ofconn,
3646 const struct ofp_header *request)
3648 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3649 struct ofputil_table_desc *table_desc;
3650 struct ovs_list replies;
3653 query_tables_desc(ofproto, &table_desc);
3654 ofpmp_init(&replies, request);
3655 for (i = 0; i < ofproto->n_tables; i++) {
3656 if (!(ofproto->tables[i].flags & OFTABLE_HIDDEN)) {
3657 ofputil_append_table_desc_reply(&table_desc[i], &replies,
3661 ofconn_send_replies(ofconn, &replies);
3666 /* This function determines and sends the vacancy event, based on the value
3667 * of current vacancy and threshold vacancy. If the current vacancy is less
3668 * than or equal to vacancy_down, vacancy up events must be enabled, and when
3669 * the current vacancy is greater or equal to vacancy_up, vacancy down events
3670 * must be enabled. */
3672 send_table_status(struct ofproto *ofproto, uint8_t table_id)
3674 struct oftable *t = &ofproto->tables[table_id];
3675 if (!t->vacancy_event) {
3679 uint8_t vacancy = oftable_vacancy(t);
3680 enum ofp14_table_reason event;
3681 if (vacancy < t->vacancy_down) {
3682 event = OFPTR_VACANCY_DOWN;
3683 } else if (vacancy > t->vacancy_up) {
3684 event = OFPTR_VACANCY_UP;
3689 if (event == t->vacancy_event) {
3690 struct ofputil_table_desc td;
3691 query_table_desc__(&td, ofproto, table_id);
3692 connmgr_send_table_status(ofproto->connmgr, &td, event);
3694 t->vacancy_event = (event == OFPTR_VACANCY_DOWN
3696 : OFPTR_VACANCY_DOWN);
3701 append_port_stat(struct ofport *port, struct ovs_list *replies)
3703 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
3705 calc_duration(port->created, time_msec(),
3706 &ops.duration_sec, &ops.duration_nsec);
3708 /* Intentionally ignore return value, since errors will set
3709 * 'stats' to all-1s, which is correct for OpenFlow, and
3710 * netdev_get_stats() will log errors. */
3711 ofproto_port_get_stats(port, &ops.stats);
3713 ofputil_append_port_stat(replies, &ops);
3717 handle_port_request(struct ofconn *ofconn,
3718 const struct ofp_header *request, ofp_port_t port_no,
3719 void (*cb)(struct ofport *, struct ovs_list *replies))
3721 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3722 struct ofport *port;
3723 struct ovs_list replies;
3725 ofpmp_init(&replies, request);
3726 if (port_no != OFPP_ANY) {
3727 port = ofproto_get_port(ofproto, port_no);
3732 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3737 ofconn_send_replies(ofconn, &replies);
3741 handle_port_stats_request(struct ofconn *ofconn,
3742 const struct ofp_header *request)
3747 error = ofputil_decode_port_stats_request(request, &port_no);
3749 handle_port_request(ofconn, request, port_no, append_port_stat);
3755 append_port_desc(struct ofport *port, struct ovs_list *replies)
3757 ofputil_append_port_desc_stats_reply(&port->pp, replies);
3761 handle_port_desc_stats_request(struct ofconn *ofconn,
3762 const struct ofp_header *request)
3767 error = ofputil_decode_port_desc_stats_request(request, &port_no);
3769 handle_port_request(ofconn, request, port_no, append_port_desc);
3775 hash_cookie(ovs_be64 cookie)
3777 return hash_uint64((OVS_FORCE uint64_t)cookie);
3781 cookies_insert(struct ofproto *ofproto, struct rule *rule)
3782 OVS_REQUIRES(ofproto_mutex)
3784 hindex_insert(&ofproto->cookies, &rule->cookie_node,
3785 hash_cookie(rule->flow_cookie));
3789 cookies_remove(struct ofproto *ofproto, struct rule *rule)
3790 OVS_REQUIRES(ofproto_mutex)
3792 hindex_remove(&ofproto->cookies, &rule->cookie_node);
3796 calc_duration(long long int start, long long int now,
3797 uint32_t *sec, uint32_t *nsec)
3799 long long int msecs = now - start;
3800 *sec = msecs / 1000;
3801 *nsec = (msecs % 1000) * (1000 * 1000);
3804 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
3805 * true if 'table_id' is OK, false otherwise. */
3807 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
3809 return table_id == OFPTT_ALL || table_id < ofproto->n_tables;
3812 static struct oftable *
3813 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
3815 struct oftable *table;
3817 for (table = &ofproto->tables[table_id];
3818 table < &ofproto->tables[ofproto->n_tables];
3820 if (!(table->flags & OFTABLE_HIDDEN)) {
3828 static struct oftable *
3829 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
3831 if (table_id == 0xff) {
3832 return next_visible_table(ofproto, 0);
3833 } else if (table_id < ofproto->n_tables) {
3834 return &ofproto->tables[table_id];
3840 static struct oftable *
3841 next_matching_table(const struct ofproto *ofproto,
3842 const struct oftable *table, uint8_t table_id)
3844 return (table_id == 0xff
3845 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
3849 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
3851 * - If TABLE_ID is 0xff, this iterates over every classifier table in
3852 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
3854 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
3855 * only once, for that table. (This can be used to access tables marked
3858 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
3859 * entered at all. (Perhaps you should have validated TABLE_ID with
3860 * check_table_id().)
3862 * All parameters are evaluated multiple times.
3864 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
3865 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
3867 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
3869 /* Initializes 'criteria' in a straightforward way based on the other
3872 * By default, the criteria include flows that are read-only, on the assumption
3873 * that the collected flows won't be modified. Call rule_criteria_require_rw()
3874 * if flows will be modified.
3876 * For "loose" matching, the 'priority' parameter is unimportant and may be
3879 rule_criteria_init(struct rule_criteria *criteria, uint8_t table_id,
3880 const struct match *match, int priority,
3881 cls_version_t version, ovs_be64 cookie,
3882 ovs_be64 cookie_mask, ofp_port_t out_port,
3885 criteria->table_id = table_id;
3886 cls_rule_init(&criteria->cr, match, priority);
3887 criteria->version = version;
3888 criteria->cookie = cookie;
3889 criteria->cookie_mask = cookie_mask;
3890 criteria->out_port = out_port;
3891 criteria->out_group = out_group;
3893 /* We ordinarily want to skip hidden rules, but there has to be a way for
3894 * code internal to OVS to modify and delete them, so if the criteria
3895 * specify a priority that can only be for a hidden flow, then allow hidden
3896 * rules to be selected. (This doesn't allow OpenFlow clients to meddle
3897 * with hidden flows because OpenFlow uses only a 16-bit field to specify
3899 criteria->include_hidden = priority > UINT16_MAX;
3901 /* We assume that the criteria are being used to collect flows for reading
3902 * but not modification. Thus, we should collect read-only flows. */
3903 criteria->include_readonly = true;
3906 /* By default, criteria initialized by rule_criteria_init() will match flows
3907 * that are read-only, on the assumption that the collected flows won't be
3908 * modified. Call this function to match only flows that are be modifiable.
3910 * Specify 'can_write_readonly' as false in ordinary circumstances, true if the
3911 * caller has special privileges that allow it to modify even "read-only"
3914 rule_criteria_require_rw(struct rule_criteria *criteria,
3915 bool can_write_readonly)
3917 criteria->include_readonly = can_write_readonly;
3921 rule_criteria_destroy(struct rule_criteria *criteria)
3923 cls_rule_destroy(&criteria->cr);
3927 rule_collection_init(struct rule_collection *rules)
3929 rules->rules = rules->stub;
3931 rules->capacity = ARRAY_SIZE(rules->stub);
3935 rule_collection_add(struct rule_collection *rules, struct rule *rule)
3937 if (rules->n >= rules->capacity) {
3938 size_t old_size, new_size;
3940 old_size = rules->capacity * sizeof *rules->rules;
3941 rules->capacity *= 2;
3942 new_size = rules->capacity * sizeof *rules->rules;
3944 if (rules->rules == rules->stub) {
3945 rules->rules = xmalloc(new_size);
3946 memcpy(rules->rules, rules->stub, old_size);
3948 rules->rules = xrealloc(rules->rules, new_size);
3952 rules->rules[rules->n++] = rule;
3956 rule_collection_ref(struct rule_collection *rules)
3957 OVS_REQUIRES(ofproto_mutex)
3961 for (i = 0; i < rules->n; i++) {
3962 ofproto_rule_ref(rules->rules[i]);
3967 rule_collection_unref(struct rule_collection *rules)
3971 for (i = 0; i < rules->n; i++) {
3972 ofproto_rule_unref(rules->rules[i]);
3976 /* Returns a NULL-terminated array of rule pointers,
3977 * destroys 'rules'. */
3978 static struct rule **
3979 rule_collection_detach(struct rule_collection *rules)
3981 struct rule **rule_array;
3983 rule_collection_add(rules, NULL);
3985 if (rules->rules == rules->stub) {
3986 rules->rules = xmemdup(rules->rules, rules->n * sizeof *rules->rules);
3989 rule_array = rules->rules;
3990 rule_collection_init(rules);
3996 rule_collection_destroy(struct rule_collection *rules)
3998 if (rules->rules != rules->stub) {
4002 /* Make repeated destruction harmless. */
4003 rule_collection_init(rules);
4006 /* Schedules postponed removal of rules, destroys 'rules'. */
4008 rule_collection_remove_postponed(struct rule_collection *rules)
4009 OVS_REQUIRES(ofproto_mutex)
4012 if (rules->n == 1) {
4013 ovsrcu_postpone(remove_rule_rcu, rules->rules[0]);
4015 ovsrcu_postpone(remove_rules_rcu, rule_collection_detach(rules));
4020 /* Checks whether 'rule' matches 'c' and, if so, adds it to 'rules'. This
4021 * function verifies most of the criteria in 'c' itself, but the caller must
4022 * check 'c->cr' itself.
4024 * Rules that have already been marked for removal are not collected.
4026 * Increments '*n_readonly' if 'rule' wasn't added because it's read-only (and
4027 * 'c' only includes modifiable rules). */
4029 collect_rule(struct rule *rule, const struct rule_criteria *c,
4030 struct rule_collection *rules, size_t *n_readonly)
4031 OVS_REQUIRES(ofproto_mutex)
4033 if ((c->table_id == rule->table_id || c->table_id == 0xff)
4034 && ofproto_rule_has_out_port(rule, c->out_port)
4035 && ofproto_rule_has_out_group(rule, c->out_group)
4036 && !((rule->flow_cookie ^ c->cookie) & c->cookie_mask)
4037 && (!rule_is_hidden(rule) || c->include_hidden)
4038 && cls_rule_visible_in_version(&rule->cr, c->version)) {
4039 /* Rule matches all the criteria... */
4040 if (!rule_is_readonly(rule) || c->include_readonly) {
4042 rule_collection_add(rules, rule);
4044 /* ...except it's read-only. */
4050 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
4051 * on classifiers rules are done in the "loose" way required for OpenFlow
4052 * OFPFC_MODIFY and OFPFC_DELETE requests. Puts the selected rules on list
4055 * Returns 0 on success, otherwise an OpenFlow error code. */
4057 collect_rules_loose(struct ofproto *ofproto,
4058 const struct rule_criteria *criteria,
4059 struct rule_collection *rules)
4060 OVS_REQUIRES(ofproto_mutex)
4062 struct oftable *table;
4063 enum ofperr error = 0;
4064 size_t n_readonly = 0;
4066 rule_collection_init(rules);
4068 if (!check_table_id(ofproto, criteria->table_id)) {
4069 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4073 if (criteria->cookie_mask == OVS_BE64_MAX) {
4076 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
4077 hash_cookie(criteria->cookie),
4078 &ofproto->cookies) {
4079 if (cls_rule_is_loose_match(&rule->cr, &criteria->cr.match)) {
4080 collect_rule(rule, criteria, rules, &n_readonly);
4084 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
4087 CLS_FOR_EACH_TARGET (rule, cr, &table->cls, &criteria->cr,
4088 criteria->version) {
4089 collect_rule(rule, criteria, rules, &n_readonly);
4095 if (!error && !rules->n && n_readonly) {
4096 /* We didn't find any rules to modify. We did find some read-only
4097 * rules that we're not allowed to modify, so report that. */
4098 error = OFPERR_OFPBRC_EPERM;
4101 rule_collection_destroy(rules);
4106 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
4107 * on classifiers rules are done in the "strict" way required for OpenFlow
4108 * OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests. Puts the selected
4109 * rules on list 'rules'.
4111 * Returns 0 on success, otherwise an OpenFlow error code. */
4113 collect_rules_strict(struct ofproto *ofproto,
4114 const struct rule_criteria *criteria,
4115 struct rule_collection *rules)
4116 OVS_REQUIRES(ofproto_mutex)
4118 struct oftable *table;
4119 size_t n_readonly = 0;
4120 enum ofperr error = 0;
4122 rule_collection_init(rules);
4124 if (!check_table_id(ofproto, criteria->table_id)) {
4125 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4129 if (criteria->cookie_mask == OVS_BE64_MAX) {
4132 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
4133 hash_cookie(criteria->cookie),
4134 &ofproto->cookies) {
4135 if (cls_rule_equal(&rule->cr, &criteria->cr)) {
4136 collect_rule(rule, criteria, rules, &n_readonly);
4140 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
4143 rule = rule_from_cls_rule(classifier_find_rule_exactly(
4144 &table->cls, &criteria->cr,
4145 criteria->version));
4147 collect_rule(rule, criteria, rules, &n_readonly);
4153 if (!error && !rules->n && n_readonly) {
4154 /* We didn't find any rules to modify. We did find some read-only
4155 * rules that we're not allowed to modify, so report that. */
4156 error = OFPERR_OFPBRC_EPERM;
4159 rule_collection_destroy(rules);
4164 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
4165 * forced into the range of a uint16_t. */
4167 age_secs(long long int age_ms)
4169 return (age_ms < 0 ? 0
4170 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
4171 : (unsigned int) age_ms / 1000);
4175 handle_flow_stats_request(struct ofconn *ofconn,
4176 const struct ofp_header *request)
4177 OVS_EXCLUDED(ofproto_mutex)
4179 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4180 struct ofputil_flow_stats_request fsr;
4181 struct rule_criteria criteria;
4182 struct rule_collection rules;
4183 struct ovs_list replies;
4187 error = ofputil_decode_flow_stats_request(&fsr, request);
4192 rule_criteria_init(&criteria, fsr.table_id, &fsr.match, 0, CLS_MAX_VERSION,
4193 fsr.cookie, fsr.cookie_mask, fsr.out_port,
4196 ovs_mutex_lock(&ofproto_mutex);
4197 error = collect_rules_loose(ofproto, &criteria, &rules);
4198 rule_criteria_destroy(&criteria);
4200 rule_collection_ref(&rules);
4202 ovs_mutex_unlock(&ofproto_mutex);
4208 ofpmp_init(&replies, request);
4209 for (i = 0; i < rules.n; i++) {
4210 struct rule *rule = rules.rules[i];
4211 long long int now = time_msec();
4212 struct ofputil_flow_stats fs;
4213 long long int created, used, modified;
4214 const struct rule_actions *actions;
4215 enum ofputil_flow_mod_flags flags;
4217 ovs_mutex_lock(&rule->mutex);
4218 fs.cookie = rule->flow_cookie;
4219 fs.idle_timeout = rule->idle_timeout;
4220 fs.hard_timeout = rule->hard_timeout;
4221 fs.importance = rule->importance;
4222 created = rule->created;
4223 modified = rule->modified;
4224 actions = rule_get_actions(rule);
4225 flags = rule->flags;
4226 ovs_mutex_unlock(&rule->mutex);
4228 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
4229 &fs.byte_count, &used);
4231 minimatch_expand(&rule->cr.match, &fs.match);
4232 fs.table_id = rule->table_id;
4233 calc_duration(created, now, &fs.duration_sec, &fs.duration_nsec);
4234 fs.priority = rule->cr.priority;
4235 fs.idle_age = age_secs(now - used);
4236 fs.hard_age = age_secs(now - modified);
4237 fs.ofpacts = actions->ofpacts;
4238 fs.ofpacts_len = actions->ofpacts_len;
4241 ofputil_append_flow_stats_reply(&fs, &replies);
4244 rule_collection_unref(&rules);
4245 rule_collection_destroy(&rules);
4247 ofconn_send_replies(ofconn, &replies);
4253 flow_stats_ds(struct rule *rule, struct ds *results)
4255 uint64_t packet_count, byte_count;
4256 const struct rule_actions *actions;
4257 long long int created, used;
4259 rule->ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
4260 &byte_count, &used);
4262 ovs_mutex_lock(&rule->mutex);
4263 actions = rule_get_actions(rule);
4264 created = rule->created;
4265 ovs_mutex_unlock(&rule->mutex);
4267 if (rule->table_id != 0) {
4268 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
4270 ds_put_format(results, "duration=%llds, ", (time_msec() - created) / 1000);
4271 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
4272 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
4273 cls_rule_format(&rule->cr, results);
4274 ds_put_char(results, ',');
4276 ds_put_cstr(results, "actions=");
4277 ofpacts_format(actions->ofpacts, actions->ofpacts_len, results);
4279 ds_put_cstr(results, "\n");
4282 /* Adds a pretty-printed description of all flows to 'results', including
4283 * hidden flows (e.g., set up by in-band control). */
4285 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
4287 struct oftable *table;
4289 OFPROTO_FOR_EACH_TABLE (table, p) {
4292 CLS_FOR_EACH (rule, cr, &table->cls) {
4293 flow_stats_ds(rule, results);
4298 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
4299 * '*engine_type' and '*engine_id', respectively. */
4301 ofproto_get_netflow_ids(const struct ofproto *ofproto,
4302 uint8_t *engine_type, uint8_t *engine_id)
4304 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
4307 /* Checks the status change of CFM on 'ofport'.
4309 * Returns true if 'ofproto_class' does not support 'cfm_status_changed'. */
4311 ofproto_port_cfm_status_changed(struct ofproto *ofproto, ofp_port_t ofp_port)
4313 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
4314 return (ofport && ofproto->ofproto_class->cfm_status_changed
4315 ? ofproto->ofproto_class->cfm_status_changed(ofport)
4319 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'.
4320 * Returns 0 if the port's CFM status was successfully stored into
4321 * '*status'. Returns positive errno if the port did not have CFM
4324 * The caller must provide and own '*status', and must free 'status->rmps'.
4325 * '*status' is indeterminate if the return value is non-zero. */
4327 ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port,
4328 struct cfm_status *status)
4330 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
4331 return (ofport && ofproto->ofproto_class->get_cfm_status
4332 ? ofproto->ofproto_class->get_cfm_status(ofport, status)
4337 handle_aggregate_stats_request(struct ofconn *ofconn,
4338 const struct ofp_header *oh)
4339 OVS_EXCLUDED(ofproto_mutex)
4341 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4342 struct ofputil_flow_stats_request request;
4343 struct ofputil_aggregate_stats stats;
4344 bool unknown_packets, unknown_bytes;
4345 struct rule_criteria criteria;
4346 struct rule_collection rules;
4347 struct ofpbuf *reply;
4351 error = ofputil_decode_flow_stats_request(&request, oh);
4356 rule_criteria_init(&criteria, request.table_id, &request.match, 0,
4357 CLS_MAX_VERSION, request.cookie, request.cookie_mask,
4358 request.out_port, request.out_group);
4360 ovs_mutex_lock(&ofproto_mutex);
4361 error = collect_rules_loose(ofproto, &criteria, &rules);
4362 rule_criteria_destroy(&criteria);
4364 rule_collection_ref(&rules);
4366 ovs_mutex_unlock(&ofproto_mutex);
4372 memset(&stats, 0, sizeof stats);
4373 unknown_packets = unknown_bytes = false;
4374 for (i = 0; i < rules.n; i++) {
4375 struct rule *rule = rules.rules[i];
4376 uint64_t packet_count;
4377 uint64_t byte_count;
4380 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
4381 &byte_count, &used);
4383 if (packet_count == UINT64_MAX) {
4384 unknown_packets = true;
4386 stats.packet_count += packet_count;
4389 if (byte_count == UINT64_MAX) {
4390 unknown_bytes = true;
4392 stats.byte_count += byte_count;
4397 if (unknown_packets) {
4398 stats.packet_count = UINT64_MAX;
4400 if (unknown_bytes) {
4401 stats.byte_count = UINT64_MAX;
4404 rule_collection_unref(&rules);
4405 rule_collection_destroy(&rules);
4407 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
4408 ofconn_send_reply(ofconn, reply);
4413 struct queue_stats_cbdata {
4414 struct ofport *ofport;
4415 struct ovs_list replies;
4420 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
4421 const struct netdev_queue_stats *stats)
4423 struct ofputil_queue_stats oqs;
4425 oqs.port_no = cbdata->ofport->pp.port_no;
4426 oqs.queue_id = queue_id;
4427 oqs.tx_bytes = stats->tx_bytes;
4428 oqs.tx_packets = stats->tx_packets;
4429 oqs.tx_errors = stats->tx_errors;
4430 if (stats->created != LLONG_MIN) {
4431 calc_duration(stats->created, cbdata->now,
4432 &oqs.duration_sec, &oqs.duration_nsec);
4434 oqs.duration_sec = oqs.duration_nsec = UINT32_MAX;
4436 ofputil_append_queue_stat(&cbdata->replies, &oqs);
4440 handle_queue_stats_dump_cb(uint32_t queue_id,
4441 struct netdev_queue_stats *stats,
4444 struct queue_stats_cbdata *cbdata = cbdata_;
4446 put_queue_stats(cbdata, queue_id, stats);
4450 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
4451 struct queue_stats_cbdata *cbdata)
4453 cbdata->ofport = port;
4454 if (queue_id == OFPQ_ALL) {
4455 netdev_dump_queue_stats(port->netdev,
4456 handle_queue_stats_dump_cb, cbdata);
4458 struct netdev_queue_stats stats;
4460 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
4461 put_queue_stats(cbdata, queue_id, &stats);
4463 return OFPERR_OFPQOFC_BAD_QUEUE;
4470 handle_queue_stats_request(struct ofconn *ofconn,
4471 const struct ofp_header *rq)
4473 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4474 struct queue_stats_cbdata cbdata;
4475 struct ofport *port;
4477 struct ofputil_queue_stats_request oqsr;
4479 COVERAGE_INC(ofproto_queue_req);
4481 ofpmp_init(&cbdata.replies, rq);
4482 cbdata.now = time_msec();
4484 error = ofputil_decode_queue_stats_request(rq, &oqsr);
4489 if (oqsr.port_no == OFPP_ANY) {
4490 error = OFPERR_OFPQOFC_BAD_QUEUE;
4491 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
4492 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
4497 port = ofproto_get_port(ofproto, oqsr.port_no);
4499 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
4500 : OFPERR_OFPQOFC_BAD_PORT);
4503 ofconn_send_replies(ofconn, &cbdata.replies);
4505 ofpbuf_list_delete(&cbdata.replies);
4512 evict_rules_from_table(struct oftable *table)
4513 OVS_REQUIRES(ofproto_mutex)
4515 enum ofperr error = 0;
4516 struct rule_collection rules;
4517 unsigned int count = table->n_flows;
4518 unsigned int max_flows = table->max_flows;
4520 rule_collection_init(&rules);
4522 while (count-- > max_flows) {
4525 if (!choose_rule_to_evict(table, &rule)) {
4526 error = OFPERR_OFPFMFC_TABLE_FULL;
4529 eviction_group_remove_rule(rule);
4530 rule_collection_add(&rules, rule);
4533 delete_flows__(&rules, OFPRR_EVICTION, NULL);
4539 get_conjunctions(const struct ofputil_flow_mod *fm,
4540 struct cls_conjunction **conjsp, size_t *n_conjsp)
4541 OVS_REQUIRES(ofproto_mutex)
4543 struct cls_conjunction *conjs = NULL;
4546 const struct ofpact *ofpact;
4547 OFPACT_FOR_EACH (ofpact, fm->ofpacts, fm->ofpacts_len) {
4548 if (ofpact->type == OFPACT_CONJUNCTION) {
4550 } else if (ofpact->type != OFPACT_NOTE) {
4551 /* "conjunction" may appear with "note" actions but not with any
4552 * other type of actions. */
4553 ovs_assert(!n_conjs);
4560 conjs = xzalloc(n_conjs * sizeof *conjs);
4561 OFPACT_FOR_EACH (ofpact, fm->ofpacts, fm->ofpacts_len) {
4562 if (ofpact->type == OFPACT_CONJUNCTION) {
4563 struct ofpact_conjunction *oc = ofpact_get_CONJUNCTION(ofpact);
4564 conjs[i].clause = oc->clause;
4565 conjs[i].n_clauses = oc->n_clauses;
4566 conjs[i].id = oc->id;
4573 *n_conjsp = n_conjs;
4576 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
4577 * in which no matching flow already exists in the flow table.
4579 * Adds the flow specified by 'fm', to the ofproto's flow table. Returns 0 on
4580 * success, or an OpenFlow error code on failure.
4582 * On successful return the caller must complete the operation either by
4583 * calling add_flow_finish(), or add_flow_revert() if the operation needs to
4586 * The caller retains ownership of 'fm->ofpacts'. */
4588 add_flow_start(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
4589 OVS_REQUIRES(ofproto_mutex)
4591 struct ofputil_flow_mod *fm = &ofm->fm;
4592 struct rule **old_rule = &ofm->old_rules.stub[0];
4593 struct rule **new_rule = &ofm->new_rules.stub[0];
4594 struct oftable *table;
4598 struct cls_conjunction *conjs;
4602 if (!check_table_id(ofproto, fm->table_id)) {
4603 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4608 if (fm->table_id == 0xff) {
4609 if (ofproto->ofproto_class->rule_choose_table) {
4610 error = ofproto->ofproto_class->rule_choose_table(ofproto,
4616 ovs_assert(table_id < ofproto->n_tables);
4620 } else if (fm->table_id < ofproto->n_tables) {
4621 table_id = fm->table_id;
4623 return OFPERR_OFPBRC_BAD_TABLE_ID;
4626 table = &ofproto->tables[table_id];
4627 if (table->flags & OFTABLE_READONLY
4628 && !(fm->flags & OFPUTIL_FF_NO_READONLY)) {
4629 return OFPERR_OFPBRC_EPERM;
4632 if (!(fm->flags & OFPUTIL_FF_HIDDEN_FIELDS)
4633 && !match_has_default_hidden_fields(&fm->match)) {
4634 VLOG_WARN_RL(&rl, "%s: (add_flow) only internal flows can set "
4635 "non-default values to hidden fields", ofproto->name);
4636 return OFPERR_OFPBRC_EPERM;
4639 cls_rule_init(&cr, &fm->match, fm->priority);
4641 /* Check for the existence of an identical rule.
4642 * This will not return rules earlier marked for removal. */
4643 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr,
4647 /* Check for overlap, if requested. */
4648 if (fm->flags & OFPUTIL_FF_CHECK_OVERLAP
4649 && classifier_rule_overlaps(&table->cls, &cr, ofm->version)) {
4650 cls_rule_destroy(&cr);
4651 return OFPERR_OFPFMFC_OVERLAP;
4654 /* If necessary, evict an existing rule to clear out space. */
4655 if (table->n_flows >= table->max_flows) {
4656 if (!choose_rule_to_evict(table, &rule)) {
4657 error = OFPERR_OFPFMFC_TABLE_FULL;
4658 cls_rule_destroy(&cr);
4661 eviction_group_remove_rule(rule);
4662 /* Marks '*old_rule' as an evicted rule rather than replaced rule.
4664 fm->delete_reason = OFPRR_EVICTION;
4668 fm->modify_cookie = true;
4671 /* Allocate new rule. */
4672 error = replace_rule_create(ofproto, fm, &cr, table - ofproto->tables,
4678 get_conjunctions(fm, &conjs, &n_conjs);
4679 replace_rule_start(ofproto, ofm->version, rule, *new_rule, conjs, n_conjs);
4685 /* Revert the effects of add_flow_start(). */
4687 add_flow_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
4688 OVS_REQUIRES(ofproto_mutex)
4690 struct ofputil_flow_mod *fm = &ofm->fm;
4691 struct rule *old_rule = ofm->old_rules.stub[0];
4692 struct rule *new_rule = ofm->new_rules.stub[0];
4694 if (old_rule && fm->delete_reason == OFPRR_EVICTION) {
4695 /* Revert the eviction. */
4696 eviction_group_add_rule(old_rule);
4699 replace_rule_revert(ofproto, old_rule, new_rule);
4702 /* To be called after version bump. */
4704 add_flow_finish(struct ofproto *ofproto, struct ofproto_flow_mod *ofm,
4705 const struct flow_mod_requester *req)
4706 OVS_REQUIRES(ofproto_mutex)
4708 struct ofputil_flow_mod *fm = &ofm->fm;
4709 struct rule *old_rule = ofm->old_rules.stub[0];
4710 struct rule *new_rule = ofm->new_rules.stub[0];
4711 struct ovs_list dead_cookies = OVS_LIST_INITIALIZER(&dead_cookies);
4713 replace_rule_finish(ofproto, fm, req, old_rule, new_rule, &dead_cookies);
4714 learned_cookies_flush(ofproto, &dead_cookies);
4717 ovsrcu_postpone(remove_rule_rcu, old_rule);
4719 if (minimask_get_vid_mask(new_rule->cr.match.mask) == VLAN_VID_MASK) {
4720 if (ofproto->vlan_bitmap) {
4721 uint16_t vid = miniflow_get_vid(new_rule->cr.match.flow);
4723 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
4724 bitmap_set1(ofproto->vlan_bitmap, vid);
4725 ofproto->vlans_changed = true;
4728 ofproto->vlans_changed = true;
4732 ofmonitor_report(ofproto->connmgr, new_rule, NXFME_ADDED, 0,
4733 req ? req->ofconn : NULL,
4734 req ? req->request->xid : 0, NULL);
4736 /* Send Vacancy Events for OF1.4+. */
4737 send_table_status(ofproto, new_rule->table_id);
4740 send_buffered_packet(req, fm->buffer_id, new_rule);
4743 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
4745 /* Create a new rule based on attributes in 'fm', match in 'cr', 'table_id',
4746 * and 'old_rule'. Note that the rule is NOT inserted into a any data
4747 * structures yet. Takes ownership of 'cr'. */
4749 replace_rule_create(struct ofproto *ofproto, struct ofputil_flow_mod *fm,
4750 struct cls_rule *cr, uint8_t table_id,
4751 struct rule *old_rule, struct rule **new_rule)
4756 /* Allocate new rule. */
4757 rule = ofproto->ofproto_class->rule_alloc();
4759 cls_rule_destroy(cr);
4760 VLOG_WARN_RL(&rl, "%s: failed to allocate a rule.", ofproto->name);
4761 return OFPERR_OFPFMFC_UNKNOWN;
4764 /* Initialize base state. */
4765 *CONST_CAST(struct ofproto **, &rule->ofproto) = ofproto;
4766 cls_rule_move(CONST_CAST(struct cls_rule *, &rule->cr), cr);
4767 ovs_refcount_init(&rule->ref_count);
4768 rule->flow_cookie = fm->new_cookie;
4769 rule->created = rule->modified = time_msec();
4771 ovs_mutex_init(&rule->mutex);
4772 ovs_mutex_lock(&rule->mutex);
4773 rule->idle_timeout = fm->idle_timeout;
4774 rule->hard_timeout = fm->hard_timeout;
4775 *CONST_CAST(uint16_t *, &rule->importance) = fm->importance;
4776 rule->removed_reason = OVS_OFPRR_NONE;
4778 *CONST_CAST(uint8_t *, &rule->table_id) = table_id;
4779 rule->flags = fm->flags & OFPUTIL_FF_STATE;
4780 *CONST_CAST(const struct rule_actions **, &rule->actions)
4781 = rule_actions_create(fm->ofpacts, fm->ofpacts_len);
4782 ovs_list_init(&rule->meter_list_node);
4783 rule->eviction_group = NULL;
4784 ovs_list_init(&rule->expirable);
4785 rule->monitor_flags = 0;
4786 rule->add_seqno = 0;
4787 rule->modify_seqno = 0;
4789 /* Copy values from old rule for modify semantics. */
4790 if (old_rule && fm->delete_reason != OFPRR_EVICTION) {
4791 bool change_cookie = (fm->modify_cookie
4792 && fm->new_cookie != OVS_BE64_MAX
4793 && fm->new_cookie != old_rule->flow_cookie);
4795 ovs_mutex_lock(&old_rule->mutex);
4796 if (fm->command != OFPFC_ADD) {
4797 rule->idle_timeout = old_rule->idle_timeout;
4798 rule->hard_timeout = old_rule->hard_timeout;
4799 *CONST_CAST(uint16_t *, &rule->importance) = old_rule->importance;
4800 rule->flags = old_rule->flags;
4801 rule->created = old_rule->created;
4803 if (!change_cookie) {
4804 rule->flow_cookie = old_rule->flow_cookie;
4806 ovs_mutex_unlock(&old_rule->mutex);
4808 ovs_mutex_unlock(&rule->mutex);
4810 /* Construct rule, initializing derived state. */
4811 error = ofproto->ofproto_class->rule_construct(rule);
4813 ofproto_rule_destroy__(rule);
4817 rule->removed = true; /* Not yet in ofproto data structures. */
4824 replace_rule_start(struct ofproto *ofproto, cls_version_t version,
4825 struct rule *old_rule, struct rule *new_rule,
4826 struct cls_conjunction *conjs, size_t n_conjs)
4828 struct oftable *table = &ofproto->tables[new_rule->table_id];
4830 /* 'old_rule' may be either an evicted rule or replaced rule. */
4832 /* Mark the old rule for removal in the next version. */
4833 cls_rule_make_invisible_in_version(&old_rule->cr, version);
4837 /* Insert flow to the classifier, so that later flow_mods may relate
4838 * to it. This is reversible, in case later errors require this to
4840 ofproto_rule_insert__(ofproto, new_rule);
4841 /* Make the new rule visible for classifier lookups only from the next
4843 classifier_insert(&table->cls, &new_rule->cr, version, conjs, n_conjs);
4846 static void replace_rule_revert(struct ofproto *ofproto,
4847 struct rule *old_rule, struct rule *new_rule)
4849 struct oftable *table = &ofproto->tables[new_rule->table_id];
4852 /* Restore the original visibility of the old rule. */
4853 cls_rule_restore_visibility(&old_rule->cr);
4855 /* Restore table's rule count. */
4859 /* Remove the new rule immediately. It was never visible to lookups. */
4860 if (!classifier_remove(&table->cls, &new_rule->cr)) {
4863 ofproto_rule_remove__(ofproto, new_rule);
4864 /* The rule was not inserted to the ofproto provider, so we can
4865 * release it without deleting it from the ofproto provider. */
4866 ofproto_rule_unref(new_rule);
4869 /* Adds the 'new_rule', replacing the 'old_rule'. */
4871 replace_rule_finish(struct ofproto *ofproto, struct ofputil_flow_mod *fm,
4872 const struct flow_mod_requester *req,
4873 struct rule *old_rule, struct rule *new_rule,
4874 struct ovs_list *dead_cookies)
4875 OVS_REQUIRES(ofproto_mutex)
4877 bool forward_stats = !(fm->flags & OFPUTIL_FF_RESET_COUNTS);
4878 struct rule *replaced_rule;
4880 replaced_rule = fm->delete_reason != OFPRR_EVICTION ? old_rule : NULL;
4882 /* Insert the new flow to the ofproto provider. A non-NULL 'replaced_rule'
4883 * is a duplicate rule the 'new_rule' is replacing. The provider should
4884 * link the stats from the old rule to the new one if 'forward_stats' is
4885 * 'true'. The 'replaced_rule' will be deleted right after this call. */
4886 ofproto->ofproto_class->rule_insert(new_rule, replaced_rule,
4888 learned_cookies_inc(ofproto, rule_get_actions(new_rule));
4891 const struct rule_actions *old_actions = rule_get_actions(old_rule);
4893 /* Remove the old rule from data structures. Removal from the
4894 * classifier and the deletion of the rule is RCU postponed by the
4896 ofproto_rule_remove__(ofproto, old_rule);
4897 learned_cookies_dec(ofproto, old_actions, dead_cookies);
4899 if (replaced_rule) {
4900 enum nx_flow_update_event event = fm->command == OFPFC_ADD
4901 ? NXFME_ADDED : NXFME_MODIFIED;
4903 bool change_cookie = (fm->modify_cookie
4904 && fm->new_cookie != OVS_BE64_MAX
4905 && fm->new_cookie != old_rule->flow_cookie);
4907 bool change_actions = !ofpacts_equal(fm->ofpacts,
4909 old_actions->ofpacts,
4910 old_actions->ofpacts_len);
4912 if (event != NXFME_MODIFIED || change_actions || change_cookie) {
4913 ofmonitor_report(ofproto->connmgr, new_rule, event, 0,
4914 req ? req->ofconn : NULL,
4915 req ? req->request->xid : 0,
4916 change_actions ? old_actions : NULL);
4919 /* XXX: This is slight duplication with delete_flows_finish__() */
4921 old_rule->removed_reason = OFPRR_EVICTION;
4923 ofmonitor_report(ofproto->connmgr, old_rule, NXFME_DELETED,
4925 req ? req->ofconn : NULL,
4926 req ? req->request->xid : 0, NULL);
4932 modify_flows_start__(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
4933 OVS_REQUIRES(ofproto_mutex)
4935 struct ofputil_flow_mod *fm = &ofm->fm;
4936 struct rule_collection *old_rules = &ofm->old_rules;
4937 struct rule_collection *new_rules = &ofm->new_rules;
4940 rule_collection_init(new_rules);
4942 if (old_rules->n > 0) {
4943 struct cls_conjunction *conjs;
4947 /* Create a new 'modified' rule for each old rule. */
4948 for (i = 0; i < old_rules->n; i++) {
4949 struct rule *old_rule = old_rules->rules[i];
4950 struct rule *new_rule;
4953 cls_rule_clone(&cr, &old_rule->cr);
4954 error = replace_rule_create(ofproto, fm, &cr, old_rule->table_id,
4955 old_rule, &new_rule);
4957 rule_collection_add(new_rules, new_rule);
4959 rule_collection_unref(new_rules);
4960 rule_collection_destroy(new_rules);
4964 ovs_assert(new_rules->n == old_rules->n);
4966 get_conjunctions(fm, &conjs, &n_conjs);
4967 for (i = 0; i < old_rules->n; i++) {
4968 replace_rule_start(ofproto, ofm->version, old_rules->rules[i],
4969 new_rules->rules[i], conjs, n_conjs);
4972 } else if (!(fm->cookie_mask != htonll(0)
4973 || fm->new_cookie == OVS_BE64_MAX)) {
4974 /* No match, add a new flow. */
4975 error = add_flow_start(ofproto, ofm);
4977 ovs_assert(fm->delete_reason == OFPRR_EVICTION
4978 || !old_rules->rules[0]);
4988 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
4991 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4994 modify_flows_start_loose(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
4995 OVS_REQUIRES(ofproto_mutex)
4997 struct ofputil_flow_mod *fm = &ofm->fm;
4998 struct rule_collection *old_rules = &ofm->old_rules;
4999 struct rule_criteria criteria;
5002 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0, CLS_MAX_VERSION,
5003 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG_ANY);
5004 rule_criteria_require_rw(&criteria,
5005 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
5006 error = collect_rules_loose(ofproto, &criteria, old_rules);
5007 rule_criteria_destroy(&criteria);
5010 error = modify_flows_start__(ofproto, ofm);
5014 rule_collection_destroy(old_rules);
5020 modify_flows_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
5021 OVS_REQUIRES(ofproto_mutex)
5023 struct rule_collection *old_rules = &ofm->old_rules;
5024 struct rule_collection *new_rules = &ofm->new_rules;
5026 /* Old rules were not changed yet, only need to revert new rules. */
5027 if (old_rules->n == 0 && new_rules->n == 1) {
5028 add_flow_revert(ofproto, ofm);
5029 } else if (old_rules->n > 0) {
5030 for (size_t i = 0; i < old_rules->n; i++) {
5031 replace_rule_revert(ofproto, old_rules->rules[i],
5032 new_rules->rules[i]);
5034 rule_collection_destroy(new_rules);
5035 rule_collection_destroy(old_rules);
5040 modify_flows_finish(struct ofproto *ofproto, struct ofproto_flow_mod *ofm,
5041 const struct flow_mod_requester *req)
5042 OVS_REQUIRES(ofproto_mutex)
5044 struct ofputil_flow_mod *fm = &ofm->fm;
5045 struct rule_collection *old_rules = &ofm->old_rules;
5046 struct rule_collection *new_rules = &ofm->new_rules;
5048 if (old_rules->n == 0 && new_rules->n == 1) {
5049 add_flow_finish(ofproto, ofm, req);
5050 } else if (old_rules->n > 0) {
5051 struct ovs_list dead_cookies = OVS_LIST_INITIALIZER(&dead_cookies);
5053 ovs_assert(new_rules->n == old_rules->n);
5055 for (size_t i = 0; i < old_rules->n; i++) {
5056 replace_rule_finish(ofproto, fm, req, old_rules->rules[i],
5057 new_rules->rules[i], &dead_cookies);
5059 learned_cookies_flush(ofproto, &dead_cookies);
5060 rule_collection_remove_postponed(old_rules);
5062 send_buffered_packet(req, fm->buffer_id, new_rules->rules[0]);
5063 rule_collection_destroy(new_rules);
5067 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
5068 * code on failure. */
5070 modify_flow_start_strict(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
5071 OVS_REQUIRES(ofproto_mutex)
5073 struct ofputil_flow_mod *fm = &ofm->fm;
5074 struct rule_collection *old_rules = &ofm->old_rules;
5075 struct rule_criteria criteria;
5078 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
5079 CLS_MAX_VERSION, fm->cookie, fm->cookie_mask, OFPP_ANY,
5081 rule_criteria_require_rw(&criteria,
5082 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
5083 error = collect_rules_strict(ofproto, &criteria, old_rules);
5084 rule_criteria_destroy(&criteria);
5087 /* collect_rules_strict() can return max 1 rule. */
5088 error = modify_flows_start__(ofproto, ofm);
5092 rule_collection_destroy(old_rules);
5097 /* OFPFC_DELETE implementation. */
5100 delete_flows_start__(struct ofproto *ofproto, cls_version_t version,
5101 const struct rule_collection *rules)
5102 OVS_REQUIRES(ofproto_mutex)
5104 for (size_t i = 0; i < rules->n; i++) {
5105 struct rule *rule = rules->rules[i];
5106 struct oftable *table = &ofproto->tables[rule->table_id];
5109 cls_rule_make_invisible_in_version(&rule->cr, version);
5114 delete_flows_finish__(struct ofproto *ofproto,
5115 struct rule_collection *rules,
5116 enum ofp_flow_removed_reason reason,
5117 const struct flow_mod_requester *req)
5118 OVS_REQUIRES(ofproto_mutex)
5121 struct ovs_list dead_cookies = OVS_LIST_INITIALIZER(&dead_cookies);
5123 for (size_t i = 0; i < rules->n; i++) {
5124 struct rule *rule = rules->rules[i];
5126 /* This value will be used to send the flow removed message right
5127 * before the rule is actually destroyed. */
5128 rule->removed_reason = reason;
5130 ofmonitor_report(ofproto->connmgr, rule, NXFME_DELETED, reason,
5131 req ? req->ofconn : NULL,
5132 req ? req->request->xid : 0, NULL);
5134 /* Send Vacancy Event for OF1.4+. */
5135 send_table_status(ofproto, rule->table_id);
5137 ofproto_rule_remove__(ofproto, rule);
5138 learned_cookies_dec(ofproto, rule_get_actions(rule),
5141 rule_collection_remove_postponed(rules);
5143 learned_cookies_flush(ofproto, &dead_cookies);
5147 /* Deletes the rules listed in 'rules'.
5148 * The deleted rules will become invisible to the lookups in the next version.
5149 * Destroys 'rules'. */
5151 delete_flows__(struct rule_collection *rules,
5152 enum ofp_flow_removed_reason reason,
5153 const struct flow_mod_requester *req)
5154 OVS_REQUIRES(ofproto_mutex)
5157 struct ofproto *ofproto = rules->rules[0]->ofproto;
5159 delete_flows_start__(ofproto, ofproto->tables_version + 1, rules);
5160 ofproto_bump_tables_version(ofproto);
5161 delete_flows_finish__(ofproto, rules, reason, req);
5162 ofmonitor_flush(ofproto->connmgr);
5166 /* Implements OFPFC_DELETE. */
5168 delete_flows_start_loose(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
5169 OVS_REQUIRES(ofproto_mutex)
5171 const struct ofputil_flow_mod *fm = &ofm->fm;
5172 struct rule_collection *rules = &ofm->old_rules;
5173 struct rule_criteria criteria;
5176 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0, CLS_MAX_VERSION,
5177 fm->cookie, fm->cookie_mask, fm->out_port,
5179 rule_criteria_require_rw(&criteria,
5180 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
5181 error = collect_rules_loose(ofproto, &criteria, rules);
5182 rule_criteria_destroy(&criteria);
5185 delete_flows_start__(ofproto, ofm->version, rules);
5192 delete_flows_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
5193 OVS_REQUIRES(ofproto_mutex)
5195 struct rule_collection *rules = &ofm->old_rules;
5197 for (size_t i = 0; i < rules->n; i++) {
5198 struct rule *rule = rules->rules[i];
5199 struct oftable *table = &ofproto->tables[rule->table_id];
5201 /* Restore table's rule count. */
5204 /* Restore the original visibility of the rule. */
5205 cls_rule_restore_visibility(&rule->cr);
5207 rule_collection_destroy(rules);
5211 delete_flows_finish(struct ofproto *ofproto,
5212 struct ofproto_flow_mod *ofm,
5213 const struct flow_mod_requester *req)
5214 OVS_REQUIRES(ofproto_mutex)
5216 delete_flows_finish__(ofproto, &ofm->old_rules, ofm->fm.delete_reason,
5220 /* Implements OFPFC_DELETE_STRICT. */
5222 delete_flow_start_strict(struct ofproto *ofproto,
5223 struct ofproto_flow_mod *ofm)
5224 OVS_REQUIRES(ofproto_mutex)
5226 const struct ofputil_flow_mod *fm = &ofm->fm;
5227 struct rule_collection *rules = &ofm->old_rules;
5228 struct rule_criteria criteria;
5231 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
5232 CLS_MAX_VERSION, fm->cookie, fm->cookie_mask,
5233 fm->out_port, fm->out_group);
5234 rule_criteria_require_rw(&criteria,
5235 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
5236 error = collect_rules_strict(ofproto, &criteria, rules);
5237 rule_criteria_destroy(&criteria);
5240 delete_flows_start__(ofproto, ofm->version, rules);
5246 /* This may only be called by rule_destroy_cb()! */
5248 ofproto_rule_send_removed(struct rule *rule)
5249 OVS_EXCLUDED(ofproto_mutex)
5251 struct ofputil_flow_removed fr;
5254 minimatch_expand(&rule->cr.match, &fr.match);
5255 fr.priority = rule->cr.priority;
5257 ovs_mutex_lock(&ofproto_mutex);
5258 fr.cookie = rule->flow_cookie;
5259 fr.reason = rule->removed_reason;
5260 fr.table_id = rule->table_id;
5261 calc_duration(rule->created, time_msec(),
5262 &fr.duration_sec, &fr.duration_nsec);
5263 ovs_mutex_lock(&rule->mutex);
5264 fr.idle_timeout = rule->idle_timeout;
5265 fr.hard_timeout = rule->hard_timeout;
5266 ovs_mutex_unlock(&rule->mutex);
5267 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
5268 &fr.byte_count, &used);
5269 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
5270 ovs_mutex_unlock(&ofproto_mutex);
5273 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
5274 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
5277 * ofproto implementation ->run() functions should use this function to expire
5278 * OpenFlow flows. */
5280 ofproto_rule_expire(struct rule *rule, uint8_t reason)
5281 OVS_REQUIRES(ofproto_mutex)
5283 struct rule_collection rules;
5285 rules.rules = rules.stub;
5287 rules.stub[0] = rule;
5288 delete_flows__(&rules, reason, NULL);
5291 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5292 * means "infinite". */
5294 reduce_timeout(uint16_t max, uint16_t *timeout)
5296 if (max && (!*timeout || *timeout > max)) {
5301 /* If 'idle_timeout' is nonzero, and 'rule' has no idle timeout or an idle
5302 * timeout greater than 'idle_timeout', lowers 'rule''s idle timeout to
5303 * 'idle_timeout' seconds. Similarly for 'hard_timeout'.
5305 * Suitable for implementing OFPACT_FIN_TIMEOUT. */
5307 ofproto_rule_reduce_timeouts(struct rule *rule,
5308 uint16_t idle_timeout, uint16_t hard_timeout)
5309 OVS_EXCLUDED(ofproto_mutex, rule->mutex)
5311 if (!idle_timeout && !hard_timeout) {
5315 ovs_mutex_lock(&ofproto_mutex);
5316 if (ovs_list_is_empty(&rule->expirable)) {
5317 ovs_list_insert(&rule->ofproto->expirable, &rule->expirable);
5319 ovs_mutex_unlock(&ofproto_mutex);
5321 ovs_mutex_lock(&rule->mutex);
5322 reduce_timeout(idle_timeout, &rule->idle_timeout);
5323 reduce_timeout(hard_timeout, &rule->hard_timeout);
5324 ovs_mutex_unlock(&rule->mutex);
5328 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5329 OVS_EXCLUDED(ofproto_mutex)
5331 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5332 struct ofproto_flow_mod ofm;
5333 uint64_t ofpacts_stub[1024 / 8];
5334 struct ofpbuf ofpacts;
5337 error = reject_slave_controller(ofconn);
5342 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
5343 error = ofputil_decode_flow_mod(&ofm.fm, oh, ofconn_get_protocol(ofconn),
5345 u16_to_ofp(ofproto->max_ports),
5348 error = ofproto_check_ofpacts(ofproto, ofm.fm.ofpacts,
5349 ofm.fm.ofpacts_len);
5352 struct flow_mod_requester req;
5354 req.ofconn = ofconn;
5356 error = handle_flow_mod__(ofproto, &ofm, &req);
5359 goto exit_free_ofpacts;
5362 ofconn_report_flow_mod(ofconn, ofm.fm.command);
5365 ofpbuf_uninit(&ofpacts);
5371 handle_flow_mod__(struct ofproto *ofproto, struct ofproto_flow_mod *ofm,
5372 const struct flow_mod_requester *req)
5373 OVS_EXCLUDED(ofproto_mutex)
5377 ovs_mutex_lock(&ofproto_mutex);
5378 ofm->version = ofproto->tables_version + 1;
5379 error = ofproto_flow_mod_start(ofproto, ofm);
5381 ofproto_bump_tables_version(ofproto);
5382 ofproto_flow_mod_finish(ofproto, ofm, req);
5384 ofmonitor_flush(ofproto->connmgr);
5385 ovs_mutex_unlock(&ofproto_mutex);
5387 run_rule_executes(ofproto);
5392 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
5394 struct ofputil_role_request request;
5395 struct ofputil_role_request reply;
5399 error = ofputil_decode_role_message(oh, &request);
5404 if (request.role != OFPCR12_ROLE_NOCHANGE) {
5405 if (request.role != OFPCR12_ROLE_EQUAL
5406 && request.have_generation_id
5407 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
5408 return OFPERR_OFPRRFC_STALE;
5411 ofconn_set_role(ofconn, request.role);
5414 reply.role = ofconn_get_role(ofconn);
5415 reply.have_generation_id = ofconn_get_master_election_id(
5416 ofconn, &reply.generation_id);
5417 buf = ofputil_encode_role_reply(oh, &reply);
5418 ofconn_send_reply(ofconn, buf);
5424 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
5425 const struct ofp_header *oh)
5427 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
5428 enum ofputil_protocol cur, next;
5430 cur = ofconn_get_protocol(ofconn);
5431 next = ofputil_protocol_set_tid(cur, msg->set != 0);
5432 ofconn_set_protocol(ofconn, next);
5438 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
5440 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
5441 enum ofputil_protocol cur, next;
5442 enum ofputil_protocol next_base;
5444 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
5446 return OFPERR_OFPBRC_EPERM;
5449 cur = ofconn_get_protocol(ofconn);
5450 next = ofputil_protocol_set_base(cur, next_base);
5451 ofconn_set_protocol(ofconn, next);
5457 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
5458 const struct ofp_header *oh)
5460 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
5463 format = ntohl(msg->format);
5464 if (!ofputil_packet_in_format_is_valid(format)) {
5465 return OFPERR_OFPBRC_EPERM;
5468 ofconn_set_packet_in_format(ofconn, format);
5473 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
5475 struct ofputil_async_cfg basis = ofconn_get_async_config(ofconn);
5476 struct ofputil_async_cfg ac;
5479 error = ofputil_decode_set_async_config(oh, false, &basis, &ac);
5484 ofconn_set_async_config(ofconn, &ac);
5485 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
5486 !ofconn_get_miss_send_len(ofconn)) {
5487 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
5494 handle_nxt_get_async_request(struct ofconn *ofconn, const struct ofp_header *oh)
5496 struct ofputil_async_cfg ac = ofconn_get_async_config(ofconn);
5497 ofconn_send_reply(ofconn, ofputil_encode_get_async_reply(oh, &ac));
5503 handle_nxt_set_controller_id(struct ofconn *ofconn,
5504 const struct ofp_header *oh)
5506 const struct nx_controller_id *nci = ofpmsg_body(oh);
5508 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
5509 return OFPERR_NXBRC_MUST_BE_ZERO;
5512 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
5517 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
5521 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
5522 ? OFPRAW_OFPT10_BARRIER_REPLY
5523 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
5524 ofconn_send_reply(ofconn, buf);
5529 ofproto_compose_flow_refresh_update(const struct rule *rule,
5530 enum nx_flow_monitor_flags flags,
5531 struct ovs_list *msgs)
5532 OVS_REQUIRES(ofproto_mutex)
5534 const struct rule_actions *actions;
5535 struct ofputil_flow_update fu;
5538 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
5539 ? NXFME_ADDED : NXFME_MODIFIED);
5541 ovs_mutex_lock(&rule->mutex);
5542 fu.idle_timeout = rule->idle_timeout;
5543 fu.hard_timeout = rule->hard_timeout;
5544 ovs_mutex_unlock(&rule->mutex);
5545 fu.table_id = rule->table_id;
5546 fu.cookie = rule->flow_cookie;
5547 minimatch_expand(&rule->cr.match, &match);
5549 fu.priority = rule->cr.priority;
5551 actions = flags & NXFMF_ACTIONS ? rule_get_actions(rule) : NULL;
5552 fu.ofpacts = actions ? actions->ofpacts : NULL;
5553 fu.ofpacts_len = actions ? actions->ofpacts_len : 0;
5555 if (ovs_list_is_empty(msgs)) {
5556 ofputil_start_flow_update(msgs);
5558 ofputil_append_flow_update(&fu, msgs);
5562 ofmonitor_compose_refresh_updates(struct rule_collection *rules,
5563 struct ovs_list *msgs)
5564 OVS_REQUIRES(ofproto_mutex)
5568 for (i = 0; i < rules->n; i++) {
5569 struct rule *rule = rules->rules[i];
5570 enum nx_flow_monitor_flags flags = rule->monitor_flags;
5571 rule->monitor_flags = 0;
5573 ofproto_compose_flow_refresh_update(rule, flags, msgs);
5578 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
5579 struct rule *rule, uint64_t seqno,
5580 struct rule_collection *rules)
5581 OVS_REQUIRES(ofproto_mutex)
5583 enum nx_flow_monitor_flags update;
5585 if (rule_is_hidden(rule)) {
5589 if (!ofproto_rule_has_out_port(rule, m->out_port)) {
5594 if (rule->add_seqno > seqno) {
5595 update = NXFMF_ADD | NXFMF_MODIFY;
5596 } else if (rule->modify_seqno > seqno) {
5597 update = NXFMF_MODIFY;
5602 if (!(m->flags & update)) {
5606 update = NXFMF_INITIAL;
5609 if (!rule->monitor_flags) {
5610 rule_collection_add(rules, rule);
5612 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
5616 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
5618 struct rule_collection *rules)
5619 OVS_REQUIRES(ofproto_mutex)
5621 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
5622 const struct oftable *table;
5623 struct cls_rule target;
5625 cls_rule_init_from_minimatch(&target, &m->match, 0);
5626 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
5629 CLS_FOR_EACH_TARGET (rule, cr, &table->cls, &target, CLS_MAX_VERSION) {
5630 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
5633 cls_rule_destroy(&target);
5637 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
5638 struct rule_collection *rules)
5639 OVS_REQUIRES(ofproto_mutex)
5641 if (m->flags & NXFMF_INITIAL) {
5642 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
5647 ofmonitor_collect_resume_rules(struct ofmonitor *m,
5648 uint64_t seqno, struct rule_collection *rules)
5649 OVS_REQUIRES(ofproto_mutex)
5651 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
5655 flow_monitor_delete(struct ofconn *ofconn, uint32_t id)
5656 OVS_REQUIRES(ofproto_mutex)
5658 struct ofmonitor *m;
5661 m = ofmonitor_lookup(ofconn, id);
5663 ofmonitor_destroy(m);
5666 error = OFPERR_OFPMOFC_UNKNOWN_MONITOR;
5673 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
5674 OVS_EXCLUDED(ofproto_mutex)
5676 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5678 struct ofpbuf b = ofpbuf_const_initializer(oh, ntohs(oh->length));
5680 struct ofmonitor **monitors = NULL;
5681 size_t allocated_monitors = 0;
5682 size_t n_monitors = 0;
5686 ovs_mutex_lock(&ofproto_mutex);
5688 struct ofputil_flow_monitor_request request;
5689 struct ofmonitor *m;
5692 retval = ofputil_decode_flow_monitor_request(&request, &b);
5693 if (retval == EOF) {
5695 } else if (retval) {
5700 if (request.table_id != 0xff
5701 && request.table_id >= ofproto->n_tables) {
5702 error = OFPERR_OFPBRC_BAD_TABLE_ID;
5706 error = ofmonitor_create(&request, ofconn, &m);
5711 if (n_monitors >= allocated_monitors) {
5712 monitors = x2nrealloc(monitors, &allocated_monitors,
5715 monitors[n_monitors++] = m;
5718 struct rule_collection rules;
5719 rule_collection_init(&rules);
5720 for (size_t i = 0; i < n_monitors; i++) {
5721 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
5724 struct ovs_list replies;
5725 ofpmp_init(&replies, oh);
5726 ofmonitor_compose_refresh_updates(&rules, &replies);
5727 ovs_mutex_unlock(&ofproto_mutex);
5729 rule_collection_destroy(&rules);
5731 ofconn_send_replies(ofconn, &replies);
5737 for (size_t i = 0; i < n_monitors; i++) {
5738 ofmonitor_destroy(monitors[i]);
5741 ovs_mutex_unlock(&ofproto_mutex);
5747 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
5748 OVS_EXCLUDED(ofproto_mutex)
5753 id = ofputil_decode_flow_monitor_cancel(oh);
5755 ovs_mutex_lock(&ofproto_mutex);
5756 error = flow_monitor_delete(ofconn, id);
5757 ovs_mutex_unlock(&ofproto_mutex);
5762 /* Meters implementation.
5764 * Meter table entry, indexed by the OpenFlow meter_id.
5765 * 'created' is used to compute the duration for meter stats.
5766 * 'list rules' is needed so that we can delete the dependent rules when the
5767 * meter table entry is deleted.
5768 * 'provider_meter_id' is for the provider's private use.
5771 long long int created; /* Time created. */
5772 struct ovs_list rules; /* List of "struct rule_dpif"s. */
5773 ofproto_meter_id provider_meter_id;
5774 uint16_t flags; /* Meter flags. */
5775 uint16_t n_bands; /* Number of meter bands. */
5776 struct ofputil_meter_band *bands;
5780 * This is used in instruction validation at flow set-up time,
5781 * as flows may not use non-existing meters.
5782 * Return value of UINT32_MAX signifies an invalid meter.
5785 get_provider_meter_id(const struct ofproto *ofproto, uint32_t of_meter_id)
5787 if (of_meter_id && of_meter_id <= ofproto->meter_features.max_meters) {
5788 const struct meter *meter = ofproto->meters[of_meter_id];
5790 return meter->provider_meter_id.uint32;
5796 /* Finds the meter invoked by 'rule''s actions and adds 'rule' to the meter's
5799 meter_insert_rule(struct rule *rule)
5801 const struct rule_actions *a = rule_get_actions(rule);
5802 uint32_t meter_id = ofpacts_get_meter(a->ofpacts, a->ofpacts_len);
5803 struct meter *meter = rule->ofproto->meters[meter_id];
5805 ovs_list_insert(&meter->rules, &rule->meter_list_node);
5809 meter_update(struct meter *meter, const struct ofputil_meter_config *config)
5813 meter->flags = config->flags;
5814 meter->n_bands = config->n_bands;
5815 meter->bands = xmemdup(config->bands,
5816 config->n_bands * sizeof *meter->bands);
5819 static struct meter *
5820 meter_create(const struct ofputil_meter_config *config,
5821 ofproto_meter_id provider_meter_id)
5823 struct meter *meter;
5825 meter = xzalloc(sizeof *meter);
5826 meter->provider_meter_id = provider_meter_id;
5827 meter->created = time_msec();
5828 ovs_list_init(&meter->rules);
5830 meter_update(meter, config);
5836 meter_delete(struct ofproto *ofproto, uint32_t first, uint32_t last)
5837 OVS_REQUIRES(ofproto_mutex)
5840 for (mid = first; mid <= last; ++mid) {
5841 struct meter *meter = ofproto->meters[mid];
5843 ofproto->meters[mid] = NULL;
5844 ofproto->ofproto_class->meter_del(ofproto,
5845 meter->provider_meter_id);
5853 handle_add_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
5855 ofproto_meter_id provider_meter_id = { UINT32_MAX };
5856 struct meter **meterp = &ofproto->meters[mm->meter.meter_id];
5860 return OFPERR_OFPMMFC_METER_EXISTS;
5863 error = ofproto->ofproto_class->meter_set(ofproto, &provider_meter_id,
5866 ovs_assert(provider_meter_id.uint32 != UINT32_MAX);
5867 *meterp = meter_create(&mm->meter, provider_meter_id);
5873 handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
5875 struct meter *meter = ofproto->meters[mm->meter.meter_id];
5877 uint32_t provider_meter_id;
5880 return OFPERR_OFPMMFC_UNKNOWN_METER;
5883 provider_meter_id = meter->provider_meter_id.uint32;
5884 error = ofproto->ofproto_class->meter_set(ofproto,
5885 &meter->provider_meter_id,
5887 ovs_assert(meter->provider_meter_id.uint32 == provider_meter_id);
5889 meter_update(meter, &mm->meter);
5895 handle_delete_meter(struct ofconn *ofconn, struct ofputil_meter_mod *mm)
5896 OVS_EXCLUDED(ofproto_mutex)
5898 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5899 uint32_t meter_id = mm->meter.meter_id;
5900 struct rule_collection rules;
5901 enum ofperr error = 0;
5902 uint32_t first, last;
5904 if (meter_id == OFPM13_ALL) {
5906 last = ofproto->meter_features.max_meters;
5908 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
5911 first = last = meter_id;
5914 /* First delete the rules that use this meter. If any of those rules are
5915 * currently being modified, postpone the whole operation until later. */
5916 rule_collection_init(&rules);
5917 ovs_mutex_lock(&ofproto_mutex);
5918 for (meter_id = first; meter_id <= last; ++meter_id) {
5919 struct meter *meter = ofproto->meters[meter_id];
5920 if (meter && !ovs_list_is_empty(&meter->rules)) {
5923 LIST_FOR_EACH (rule, meter_list_node, &meter->rules) {
5924 rule_collection_add(&rules, rule);
5928 delete_flows__(&rules, OFPRR_METER_DELETE, NULL);
5930 /* Delete the meters. */
5931 meter_delete(ofproto, first, last);
5933 ovs_mutex_unlock(&ofproto_mutex);
5939 handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5941 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5942 struct ofputil_meter_mod mm;
5943 uint64_t bands_stub[256 / 8];
5944 struct ofpbuf bands;
5948 error = reject_slave_controller(ofconn);
5953 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5955 error = ofputil_decode_meter_mod(oh, &mm, &bands);
5957 goto exit_free_bands;
5960 meter_id = mm.meter.meter_id;
5962 if (mm.command != OFPMC13_DELETE) {
5963 /* Fails also when meters are not implemented by the provider. */
5964 if (meter_id == 0 || meter_id > OFPM13_MAX) {
5965 error = OFPERR_OFPMMFC_INVALID_METER;
5966 goto exit_free_bands;
5967 } else if (meter_id > ofproto->meter_features.max_meters) {
5968 error = OFPERR_OFPMMFC_OUT_OF_METERS;
5969 goto exit_free_bands;
5971 if (mm.meter.n_bands > ofproto->meter_features.max_bands) {
5972 error = OFPERR_OFPMMFC_OUT_OF_BANDS;
5973 goto exit_free_bands;
5977 switch (mm.command) {
5979 error = handle_add_meter(ofproto, &mm);
5982 case OFPMC13_MODIFY:
5983 error = handle_modify_meter(ofproto, &mm);
5986 case OFPMC13_DELETE:
5987 error = handle_delete_meter(ofconn, &mm);
5991 error = OFPERR_OFPMMFC_BAD_COMMAND;
5996 struct ofputil_requestforward rf;
5998 rf.reason = OFPRFR_METER_MOD;
6000 connmgr_send_requestforward(ofproto->connmgr, ofconn, &rf);
6004 ofpbuf_uninit(&bands);
6009 handle_meter_features_request(struct ofconn *ofconn,
6010 const struct ofp_header *request)
6012 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6013 struct ofputil_meter_features features;
6016 if (ofproto->ofproto_class->meter_get_features) {
6017 ofproto->ofproto_class->meter_get_features(ofproto, &features);
6019 memset(&features, 0, sizeof features);
6021 b = ofputil_encode_meter_features_reply(&features, request);
6023 ofconn_send_reply(ofconn, b);
6028 handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request,
6031 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6032 struct ovs_list replies;
6033 uint64_t bands_stub[256 / 8];
6034 struct ofpbuf bands;
6035 uint32_t meter_id, first, last;
6037 ofputil_decode_meter_request(request, &meter_id);
6039 if (meter_id == OFPM13_ALL) {
6041 last = ofproto->meter_features.max_meters;
6043 if (!meter_id || meter_id > ofproto->meter_features.max_meters ||
6044 !ofproto->meters[meter_id]) {
6045 return OFPERR_OFPMMFC_UNKNOWN_METER;
6047 first = last = meter_id;
6050 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
6051 ofpmp_init(&replies, request);
6053 for (meter_id = first; meter_id <= last; ++meter_id) {
6054 struct meter *meter = ofproto->meters[meter_id];
6056 continue; /* Skip non-existing meters. */
6058 if (type == OFPTYPE_METER_STATS_REQUEST) {
6059 struct ofputil_meter_stats stats;
6061 stats.meter_id = meter_id;
6063 /* Provider sets the packet and byte counts, we do the rest. */
6064 stats.flow_count = ovs_list_size(&meter->rules);
6065 calc_duration(meter->created, time_msec(),
6066 &stats.duration_sec, &stats.duration_nsec);
6067 stats.n_bands = meter->n_bands;
6068 ofpbuf_clear(&bands);
6070 = ofpbuf_put_uninit(&bands,
6071 meter->n_bands * sizeof *stats.bands);
6073 if (!ofproto->ofproto_class->meter_get(ofproto,
6074 meter->provider_meter_id,
6076 ofputil_append_meter_stats(&replies, &stats);
6078 } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */
6079 struct ofputil_meter_config config;
6081 config.meter_id = meter_id;
6082 config.flags = meter->flags;
6083 config.n_bands = meter->n_bands;
6084 config.bands = meter->bands;
6085 ofputil_append_meter_config(&replies, &config);
6089 ofconn_send_replies(ofconn, &replies);
6090 ofpbuf_uninit(&bands);
6095 ofproto_group_lookup__(const struct ofproto *ofproto, uint32_t group_id,
6096 struct ofgroup **group)
6097 OVS_REQ_RDLOCK(ofproto->groups_rwlock)
6099 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
6100 hash_int(group_id, 0), &ofproto->groups) {
6101 if ((*group)->group_id == group_id) {
6109 /* If the group exists, this function increments the groups's reference count.
6111 * Make sure to call ofproto_group_unref() after no longer needing to maintain
6112 * a reference to the group. */
6114 ofproto_group_lookup(const struct ofproto *ofproto, uint32_t group_id,
6115 struct ofgroup **group)
6119 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
6120 found = ofproto_group_lookup__(ofproto, group_id, group);
6122 ofproto_group_ref(*group);
6124 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6129 ofproto_group_exists__(const struct ofproto *ofproto, uint32_t group_id)
6130 OVS_REQ_RDLOCK(ofproto->groups_rwlock)
6132 struct ofgroup *grp;
6134 HMAP_FOR_EACH_IN_BUCKET (grp, hmap_node,
6135 hash_int(group_id, 0), &ofproto->groups) {
6136 if (grp->group_id == group_id) {
6144 ofproto_group_exists(const struct ofproto *ofproto, uint32_t group_id)
6145 OVS_EXCLUDED(ofproto->groups_rwlock)
6149 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
6150 exists = ofproto_group_exists__(ofproto, group_id);
6151 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6157 group_get_ref_count(struct ofgroup *group)
6158 OVS_EXCLUDED(ofproto_mutex)
6160 struct ofproto *ofproto = CONST_CAST(struct ofproto *, group->ofproto);
6161 struct rule_criteria criteria;
6162 struct rule_collection rules;
6167 match_init_catchall(&match);
6168 rule_criteria_init(&criteria, 0xff, &match, 0, CLS_MAX_VERSION, htonll(0),
6169 htonll(0), OFPP_ANY, group->group_id);
6170 ovs_mutex_lock(&ofproto_mutex);
6171 error = collect_rules_loose(ofproto, &criteria, &rules);
6172 ovs_mutex_unlock(&ofproto_mutex);
6173 rule_criteria_destroy(&criteria);
6175 count = !error && rules.n < UINT32_MAX ? rules.n : UINT32_MAX;
6177 rule_collection_destroy(&rules);
6182 append_group_stats(struct ofgroup *group, struct ovs_list *replies)
6184 struct ofputil_group_stats ogs;
6185 const struct ofproto *ofproto = group->ofproto;
6186 long long int now = time_msec();
6189 ogs.bucket_stats = xmalloc(group->n_buckets * sizeof *ogs.bucket_stats);
6191 /* Provider sets the packet and byte counts, we do the rest. */
6192 ogs.ref_count = group_get_ref_count(group);
6193 ogs.n_buckets = group->n_buckets;
6195 error = (ofproto->ofproto_class->group_get_stats
6196 ? ofproto->ofproto_class->group_get_stats(group, &ogs)
6199 ogs.packet_count = UINT64_MAX;
6200 ogs.byte_count = UINT64_MAX;
6201 memset(ogs.bucket_stats, 0xff,
6202 ogs.n_buckets * sizeof *ogs.bucket_stats);
6205 ogs.group_id = group->group_id;
6206 calc_duration(group->created, now, &ogs.duration_sec, &ogs.duration_nsec);
6208 ofputil_append_group_stats(replies, &ogs);
6210 free(ogs.bucket_stats);
6214 handle_group_request(struct ofconn *ofconn,
6215 const struct ofp_header *request, uint32_t group_id,
6216 void (*cb)(struct ofgroup *, struct ovs_list *replies))
6218 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6219 struct ofgroup *group;
6220 struct ovs_list replies;
6222 ofpmp_init(&replies, request);
6223 if (group_id == OFPG_ALL) {
6224 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
6225 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
6226 cb(group, &replies);
6228 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6230 if (ofproto_group_lookup(ofproto, group_id, &group)) {
6231 cb(group, &replies);
6232 ofproto_group_unref(group);
6235 ofconn_send_replies(ofconn, &replies);
6239 handle_group_stats_request(struct ofconn *ofconn,
6240 const struct ofp_header *request)
6245 error = ofputil_decode_group_stats_request(request, &group_id);
6250 handle_group_request(ofconn, request, group_id, append_group_stats);
6255 append_group_desc(struct ofgroup *group, struct ovs_list *replies)
6257 struct ofputil_group_desc gds;
6259 gds.group_id = group->group_id;
6260 gds.type = group->type;
6261 gds.props = group->props;
6263 ofputil_append_group_desc_reply(&gds, &group->buckets, replies);
6267 handle_group_desc_stats_request(struct ofconn *ofconn,
6268 const struct ofp_header *request)
6270 handle_group_request(ofconn, request,
6271 ofputil_decode_group_desc_request(request),
6277 handle_group_features_stats_request(struct ofconn *ofconn,
6278 const struct ofp_header *request)
6280 struct ofproto *p = ofconn_get_ofproto(ofconn);
6283 msg = ofputil_encode_group_features_reply(&p->ogf, request);
6285 ofconn_send_reply(ofconn, msg);
6292 put_queue_get_config_reply(struct ofport *port, uint32_t queue,
6293 struct ovs_list *replies)
6295 struct ofputil_queue_config qc;
6297 /* None of the existing queues have compatible properties, so we hard-code
6298 * omitting min_rate and max_rate. */
6299 qc.port = port->ofp_port;
6301 qc.min_rate = UINT16_MAX;
6302 qc.max_rate = UINT16_MAX;
6303 ofputil_append_queue_get_config_reply(&qc, replies);
6307 handle_queue_get_config_request_for_port(struct ofport *port, uint32_t queue,
6308 struct ovs_list *replies)
6310 struct smap details = SMAP_INITIALIZER(&details);
6311 if (queue != OFPQ_ALL) {
6312 int error = netdev_get_queue(port->netdev, queue, &details);
6315 put_queue_get_config_reply(port, queue, replies);
6319 return OFPERR_OFPQOFC_BAD_QUEUE;
6321 return OFPERR_NXQOFC_QUEUE_ERROR;
6324 struct netdev_queue_dump queue_dump;
6327 NETDEV_QUEUE_FOR_EACH (&queue_id, &details, &queue_dump,
6329 put_queue_get_config_reply(port, queue_id, replies);
6332 smap_destroy(&details);
6337 handle_queue_get_config_request(struct ofconn *ofconn,
6338 const struct ofp_header *oh)
6340 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6341 struct ovs_list replies;
6342 struct ofport *port;
6343 ofp_port_t req_port;
6347 error = ofputil_decode_queue_get_config_request(oh, &req_port, &req_queue);
6352 ofputil_start_queue_get_config_reply(oh, &replies);
6353 if (req_port == OFPP_ANY) {
6354 error = OFPERR_OFPQOFC_BAD_QUEUE;
6355 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
6356 if (!handle_queue_get_config_request_for_port(port, req_queue,
6362 port = ofproto_get_port(ofproto, req_port);
6364 ? handle_queue_get_config_request_for_port(port, req_queue,
6366 : OFPERR_OFPQOFC_BAD_PORT);
6369 ofconn_send_replies(ofconn, &replies);
6371 ofpbuf_list_delete(&replies);
6378 init_group(struct ofproto *ofproto, const struct ofputil_group_mod *gm,
6379 struct ofgroup **ofgroup)
6382 const long long int now = time_msec();
6384 if (gm->group_id > OFPG_MAX) {
6385 return OFPERR_OFPGMFC_INVALID_GROUP;
6387 if (gm->type > OFPGT11_FF) {
6388 return OFPERR_OFPGMFC_BAD_TYPE;
6391 *ofgroup = ofproto->ofproto_class->group_alloc();
6393 VLOG_WARN_RL(&rl, "%s: failed to allocate group", ofproto->name);
6394 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
6397 (*ofgroup)->ofproto = ofproto;
6398 *CONST_CAST(uint32_t *, &((*ofgroup)->group_id)) = gm->group_id;
6399 *CONST_CAST(enum ofp11_group_type *, &(*ofgroup)->type) = gm->type;
6400 *CONST_CAST(long long int *, &((*ofgroup)->created)) = now;
6401 *CONST_CAST(long long int *, &((*ofgroup)->modified)) = now;
6402 ovs_refcount_init(&(*ofgroup)->ref_count);
6404 ovs_list_init(&(*ofgroup)->buckets);
6405 ofputil_bucket_clone_list(&(*ofgroup)->buckets, &gm->buckets, NULL);
6407 *CONST_CAST(uint32_t *, &(*ofgroup)->n_buckets) =
6408 ovs_list_size(&(*ofgroup)->buckets);
6410 memcpy(CONST_CAST(struct ofputil_group_props *, &(*ofgroup)->props),
6411 &gm->props, sizeof (struct ofputil_group_props));
6413 /* Construct called BEFORE any locks are held. */
6414 error = ofproto->ofproto_class->group_construct(*ofgroup);
6416 ofputil_bucket_list_destroy(&(*ofgroup)->buckets);
6417 ofproto->ofproto_class->group_dealloc(*ofgroup);
6422 /* Implements the OFPGC11_ADD operation specified by 'gm', adding a group to
6423 * 'ofproto''s group table. Returns 0 on success or an OpenFlow error code on
6426 add_group(struct ofproto *ofproto, const struct ofputil_group_mod *gm)
6428 struct ofgroup *ofgroup;
6431 /* Allocate new group and initialize it. */
6432 error = init_group(ofproto, gm, &ofgroup);
6437 /* We wrlock as late as possible to minimize the time we jam any other
6438 * threads: No visible state changes before acquiring the lock. */
6439 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
6441 if (ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
6442 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
6446 if (ofproto_group_exists__(ofproto, gm->group_id)) {
6447 error = OFPERR_OFPGMFC_GROUP_EXISTS;
6452 /* Insert new group. */
6453 hmap_insert(&ofproto->groups, &ofgroup->hmap_node,
6454 hash_int(ofgroup->group_id, 0));
6455 ofproto->n_groups[ofgroup->type]++;
6457 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6462 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6463 ofproto->ofproto_class->group_destruct(ofgroup);
6464 ofputil_bucket_list_destroy(&ofgroup->buckets);
6465 ofproto->ofproto_class->group_dealloc(ofgroup);
6470 /* Adds all of the buckets from 'ofgroup' to 'new_ofgroup'. The buckets
6471 * already in 'new_ofgroup' will be placed just after the (copy of the) bucket
6472 * in 'ofgroup' with bucket ID 'command_bucket_id'. Special
6473 * 'command_bucket_id' values OFPG15_BUCKET_FIRST and OFPG15_BUCKET_LAST are
6476 copy_buckets_for_insert_bucket(const struct ofgroup *ofgroup,
6477 struct ofgroup *new_ofgroup,
6478 uint32_t command_bucket_id)
6480 struct ofputil_bucket *last = NULL;
6482 if (command_bucket_id <= OFPG15_BUCKET_MAX) {
6483 /* Check here to ensure that a bucket corresponding to
6484 * command_bucket_id exists in the old bucket list.
6486 * The subsequent search of below of new_ofgroup covers
6487 * both buckets in the old bucket list and buckets added
6488 * by the insert buckets group mod message this function processes. */
6489 if (!ofputil_bucket_find(&ofgroup->buckets, command_bucket_id)) {
6490 return OFPERR_OFPGMFC_UNKNOWN_BUCKET;
6493 if (!ovs_list_is_empty(&new_ofgroup->buckets)) {
6494 last = ofputil_bucket_list_back(&new_ofgroup->buckets);
6498 ofputil_bucket_clone_list(&new_ofgroup->buckets, &ofgroup->buckets, NULL);
6500 if (ofputil_bucket_check_duplicate_id(&new_ofgroup->buckets)) {
6501 VLOG_INFO_RL(&rl, "Duplicate bucket id");
6502 return OFPERR_OFPGMFC_BUCKET_EXISTS;
6505 /* Rearrange list according to command_bucket_id */
6506 if (command_bucket_id == OFPG15_BUCKET_LAST) {
6507 if (!ovs_list_is_empty(&ofgroup->buckets)) {
6508 struct ofputil_bucket *new_first;
6509 const struct ofputil_bucket *first;
6511 first = ofputil_bucket_list_front(&ofgroup->buckets);
6512 new_first = ofputil_bucket_find(&new_ofgroup->buckets,
6515 ovs_list_splice(new_ofgroup->buckets.next, &new_first->list_node,
6516 &new_ofgroup->buckets);
6518 } else if (command_bucket_id <= OFPG15_BUCKET_MAX && last) {
6519 struct ofputil_bucket *after;
6521 /* Presence of bucket is checked above so after should never be NULL */
6522 after = ofputil_bucket_find(&new_ofgroup->buckets, command_bucket_id);
6524 ovs_list_splice(after->list_node.next, new_ofgroup->buckets.next,
6525 last->list_node.next);
6531 /* Appends all of the a copy of all the buckets from 'ofgroup' to 'new_ofgroup'
6532 * with the exception of the bucket whose bucket id is 'command_bucket_id'.
6533 * Special 'command_bucket_id' values OFPG15_BUCKET_FIRST, OFPG15_BUCKET_LAST
6534 * and OFPG15_BUCKET_ALL are also honored. */
6536 copy_buckets_for_remove_bucket(const struct ofgroup *ofgroup,
6537 struct ofgroup *new_ofgroup,
6538 uint32_t command_bucket_id)
6540 const struct ofputil_bucket *skip = NULL;
6542 if (command_bucket_id == OFPG15_BUCKET_ALL) {
6546 if (command_bucket_id == OFPG15_BUCKET_FIRST) {
6547 if (!ovs_list_is_empty(&ofgroup->buckets)) {
6548 skip = ofputil_bucket_list_front(&ofgroup->buckets);
6550 } else if (command_bucket_id == OFPG15_BUCKET_LAST) {
6551 if (!ovs_list_is_empty(&ofgroup->buckets)) {
6552 skip = ofputil_bucket_list_back(&ofgroup->buckets);
6555 skip = ofputil_bucket_find(&ofgroup->buckets, command_bucket_id);
6557 return OFPERR_OFPGMFC_UNKNOWN_BUCKET;
6561 ofputil_bucket_clone_list(&new_ofgroup->buckets, &ofgroup->buckets, skip);
6566 /* Implements OFPGC11_MODIFY, OFPGC15_INSERT_BUCKET and
6567 * OFPGC15_REMOVE_BUCKET. Returns 0 on success or an OpenFlow error code
6570 * Note that the group is re-created and then replaces the old group in
6571 * ofproto's ofgroup hash map. Thus, the group is never altered while users of
6572 * the xlate module hold a pointer to the group. */
6574 modify_group(struct ofproto *ofproto, const struct ofputil_group_mod *gm)
6576 struct ofgroup *ofgroup, *new_ofgroup, *retiring;
6579 error = init_group(ofproto, gm, &new_ofgroup);
6584 retiring = new_ofgroup;
6586 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
6587 if (!ofproto_group_lookup__(ofproto, gm->group_id, &ofgroup)) {
6588 error = OFPERR_OFPGMFC_UNKNOWN_GROUP;
6592 /* Ofproto's group write lock is held now. */
6593 if (ofgroup->type != gm->type
6594 && ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
6595 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
6599 /* Manipulate bucket list for bucket commands */
6600 if (gm->command == OFPGC15_INSERT_BUCKET) {
6601 error = copy_buckets_for_insert_bucket(ofgroup, new_ofgroup,
6602 gm->command_bucket_id);
6603 } else if (gm->command == OFPGC15_REMOVE_BUCKET) {
6604 error = copy_buckets_for_remove_bucket(ofgroup, new_ofgroup,
6605 gm->command_bucket_id);
6611 /* The group creation time does not change during modification. */
6612 *CONST_CAST(long long int *, &(new_ofgroup->created)) = ofgroup->created;
6613 *CONST_CAST(long long int *, &(new_ofgroup->modified)) = time_msec();
6615 error = ofproto->ofproto_class->group_modify(new_ofgroup);
6621 /* Replace ofgroup in ofproto's groups hash map with new_ofgroup. */
6622 hmap_remove(&ofproto->groups, &ofgroup->hmap_node);
6623 hmap_insert(&ofproto->groups, &new_ofgroup->hmap_node,
6624 hash_int(new_ofgroup->group_id, 0));
6625 if (ofgroup->type != new_ofgroup->type) {
6626 ofproto->n_groups[ofgroup->type]--;
6627 ofproto->n_groups[new_ofgroup->type]++;
6631 ofproto_group_unref(retiring);
6632 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6637 delete_group__(struct ofproto *ofproto, struct ofgroup *ofgroup)
6638 OVS_RELEASES(ofproto->groups_rwlock)
6641 struct ofproto_flow_mod ofm;
6643 /* Delete all flow entries containing this group in a group action */
6644 match_init_catchall(&match);
6645 flow_mod_init(&ofm.fm, &match, 0, NULL, 0, OFPFC_DELETE);
6646 ofm.fm.delete_reason = OFPRR_GROUP_DELETE;
6647 ofm.fm.out_group = ofgroup->group_id;
6648 ofm.fm.table_id = OFPTT_ALL;
6649 handle_flow_mod__(ofproto, &ofm, NULL);
6651 hmap_remove(&ofproto->groups, &ofgroup->hmap_node);
6652 /* No-one can find this group any more. */
6653 ofproto->n_groups[ofgroup->type]--;
6654 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6655 ofproto_group_unref(ofgroup);
6658 /* Implements OFPGC11_DELETE. */
6660 delete_group(struct ofproto *ofproto, uint32_t group_id)
6662 struct ofgroup *ofgroup;
6664 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
6665 if (group_id == OFPG_ALL) {
6667 struct hmap_node *node = hmap_first(&ofproto->groups);
6671 ofgroup = CONTAINER_OF(node, struct ofgroup, hmap_node);
6672 delete_group__(ofproto, ofgroup);
6673 /* Lock for each node separately, so that we will not jam the
6674 * other threads for too long time. */
6675 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
6678 HMAP_FOR_EACH_IN_BUCKET (ofgroup, hmap_node,
6679 hash_int(group_id, 0), &ofproto->groups) {
6680 if (ofgroup->group_id == group_id) {
6681 delete_group__(ofproto, ofgroup);
6686 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6689 /* Delete all groups from 'ofproto'.
6691 * This is intended for use within an ofproto provider's 'destruct'
6694 ofproto_group_delete_all(struct ofproto *ofproto)
6696 delete_group(ofproto, OFPG_ALL);
6700 handle_group_mod(struct ofconn *ofconn, const struct ofp_header *oh)
6702 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6703 struct ofputil_group_mod gm;
6706 error = reject_slave_controller(ofconn);
6711 error = ofputil_decode_group_mod(oh, &gm);
6716 switch (gm.command) {
6718 error = add_group(ofproto, &gm);
6721 case OFPGC11_MODIFY:
6722 error = modify_group(ofproto, &gm);
6725 case OFPGC11_DELETE:
6726 delete_group(ofproto, gm.group_id);
6730 case OFPGC15_INSERT_BUCKET:
6731 error = modify_group(ofproto, &gm);
6734 case OFPGC15_REMOVE_BUCKET:
6735 error = modify_group(ofproto, &gm);
6739 if (gm.command > OFPGC11_DELETE) {
6740 VLOG_INFO_RL(&rl, "%s: Invalid group_mod command type %d",
6741 ofproto->name, gm.command);
6743 error = OFPERR_OFPGMFC_BAD_COMMAND;
6747 struct ofputil_requestforward rf;
6749 rf.reason = OFPRFR_GROUP_MOD;
6751 connmgr_send_requestforward(ofproto->connmgr, ofconn, &rf);
6753 ofputil_bucket_list_destroy(&gm.buckets);
6758 enum ofputil_table_miss
6759 ofproto_table_get_miss_config(const struct ofproto *ofproto, uint8_t table_id)
6761 enum ofputil_table_miss miss;
6763 atomic_read_relaxed(&ofproto->tables[table_id].miss_config, &miss);
6768 table_mod__(struct oftable *oftable,
6769 const struct ofputil_table_mod *tm)
6771 if (tm->miss == OFPUTIL_TABLE_MISS_DEFAULT) {
6772 /* This is how an OFPT_TABLE_MOD decodes if it doesn't specify any
6773 * table-miss configuration (because the protocol used doesn't have
6774 * such a concept), so there's nothing to do. */
6776 atomic_store_relaxed(&oftable->miss_config, tm->miss);
6779 unsigned int new_eviction = oftable->eviction;
6780 if (tm->eviction == OFPUTIL_TABLE_EVICTION_ON) {
6781 new_eviction |= EVICTION_OPENFLOW;
6782 } else if (tm->eviction == OFPUTIL_TABLE_EVICTION_OFF) {
6783 new_eviction &= ~EVICTION_OPENFLOW;
6786 if (new_eviction != oftable->eviction) {
6787 ovs_mutex_lock(&ofproto_mutex);
6788 oftable_configure_eviction(oftable, new_eviction,
6789 oftable->eviction_fields,
6790 oftable->n_eviction_fields);
6791 ovs_mutex_unlock(&ofproto_mutex);
6794 if (tm->vacancy != OFPUTIL_TABLE_VACANCY_DEFAULT) {
6795 ovs_mutex_lock(&ofproto_mutex);
6796 oftable->vacancy_down = tm->table_vacancy.vacancy_down;
6797 oftable->vacancy_up = tm->table_vacancy.vacancy_up;
6798 if (tm->vacancy == OFPUTIL_TABLE_VACANCY_OFF) {
6799 oftable->vacancy_event = 0;
6800 } else if (!oftable->vacancy_event) {
6801 uint8_t vacancy = oftable_vacancy(oftable);
6802 oftable->vacancy_event = (vacancy < oftable->vacancy_up
6804 : OFPTR_VACANCY_DOWN);
6806 ovs_mutex_unlock(&ofproto_mutex);
6811 table_mod(struct ofproto *ofproto, const struct ofputil_table_mod *tm)
6813 if (!check_table_id(ofproto, tm->table_id)) {
6814 return OFPERR_OFPTMFC_BAD_TABLE;
6817 /* Don't allow the eviction flags to be changed (except to the only fixed
6818 * value that OVS supports). OF1.4 says this is normal: "The
6819 * OFPTMPT_EVICTION property usually cannot be modified using a
6820 * OFP_TABLE_MOD request, because the eviction mechanism is switch
6822 if (tm->eviction_flags != UINT32_MAX
6823 && tm->eviction_flags != OFPROTO_EVICTION_FLAGS) {
6824 return OFPERR_OFPTMFC_BAD_CONFIG;
6827 if (tm->table_id == OFPTT_ALL) {
6828 struct oftable *oftable;
6829 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
6830 if (!(oftable->flags & (OFTABLE_HIDDEN | OFTABLE_READONLY))) {
6831 table_mod__(oftable, tm);
6835 struct oftable *oftable = &ofproto->tables[tm->table_id];
6836 if (oftable->flags & OFTABLE_READONLY) {
6837 return OFPERR_OFPTMFC_EPERM;
6839 table_mod__(oftable, tm);
6846 handle_table_mod(struct ofconn *ofconn, const struct ofp_header *oh)
6848 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6849 struct ofputil_table_mod tm;
6852 error = reject_slave_controller(ofconn);
6857 error = ofputil_decode_table_mod(oh, &tm);
6862 return table_mod(ofproto, &tm);
6866 ofproto_flow_mod_start(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
6867 OVS_REQUIRES(ofproto_mutex)
6869 switch (ofm->fm.command) {
6871 return add_flow_start(ofproto, ofm);
6872 /* , &be->old_rules.stub[0],
6873 &be->new_rules.stub[0]); */
6875 return modify_flows_start_loose(ofproto, ofm);
6876 case OFPFC_MODIFY_STRICT:
6877 return modify_flow_start_strict(ofproto, ofm);
6879 return delete_flows_start_loose(ofproto, ofm);
6881 case OFPFC_DELETE_STRICT:
6882 return delete_flow_start_strict(ofproto, ofm);
6885 return OFPERR_OFPFMFC_BAD_COMMAND;
6889 ofproto_flow_mod_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
6890 OVS_REQUIRES(ofproto_mutex)
6892 switch (ofm->fm.command) {
6894 add_flow_revert(ofproto, ofm);
6898 case OFPFC_MODIFY_STRICT:
6899 modify_flows_revert(ofproto, ofm);
6903 case OFPFC_DELETE_STRICT:
6904 delete_flows_revert(ofproto, ofm);
6913 ofproto_flow_mod_finish(struct ofproto *ofproto,
6914 struct ofproto_flow_mod *ofm,
6915 const struct flow_mod_requester *req)
6916 OVS_REQUIRES(ofproto_mutex)
6918 switch (ofm->fm.command) {
6920 add_flow_finish(ofproto, ofm, req);
6924 case OFPFC_MODIFY_STRICT:
6925 modify_flows_finish(ofproto, ofm, req);
6929 case OFPFC_DELETE_STRICT:
6930 delete_flows_finish(ofproto, ofm, req);
6938 /* Commit phases (all while locking ofproto_mutex):
6940 * 1. Begin: Gather resources and make changes visible in the next version.
6941 * - Mark affected rules for removal in the next version.
6942 * - Create new replacement rules, make visible in the next
6944 * - Do not send any events or notifications.
6946 * 2. Revert: Fail if any errors are found. After this point no errors are
6947 * possible. No visible changes were made, so rollback is minimal (remove
6948 * added invisible rules, restore visibility of rules marked for removal).
6950 * 3. Finish: Make the changes visible for lookups. Insert replacement rules to
6951 * the ofproto provider. Remove replaced and deleted rules from ofproto data
6952 * structures, and Schedule postponed removal of deleted rules from the
6953 * classifier. Send notifications, buffered packets, etc.
6956 do_bundle_commit(struct ofconn *ofconn, uint32_t id, uint16_t flags)
6958 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6959 cls_version_t version = ofproto->tables_version + 1;
6960 struct ofp_bundle *bundle;
6961 struct ofp_bundle_entry *be;
6964 bundle = ofconn_get_bundle(ofconn, id);
6967 return OFPERR_OFPBFC_BAD_ID;
6969 if (bundle->flags != flags) {
6970 error = OFPERR_OFPBFC_BAD_FLAGS;
6972 bool prev_is_port_mod = false;
6975 ovs_mutex_lock(&ofproto_mutex);
6978 LIST_FOR_EACH (be, node, &bundle->msg_list) {
6979 if (be->type == OFPTYPE_PORT_MOD) {
6980 /* Our port mods are not atomic. */
6981 if (flags & OFPBF_ATOMIC) {
6982 error = OFPERR_OFPBFC_MSG_FAILED;
6984 prev_is_port_mod = true;
6985 error = port_mod_start(ofconn, &be->opm.pm, &be->opm.port);
6987 } else if (be->type == OFPTYPE_FLOW_MOD) {
6988 /* Flow mods between port mods are applied as a single
6989 * version, but the versions are published only after
6990 * we know the commit is successful. */
6991 if (prev_is_port_mod) {
6994 prev_is_port_mod = false;
6995 /* Store the version in which the changes should take
6997 be->ofm.version = version;
6998 error = ofproto_flow_mod_start(ofproto, &be->ofm);
7008 /* Send error referring to the original message. */
7010 ofconn_send_error(ofconn, be->ofp_msg, error);
7011 error = OFPERR_OFPBFC_MSG_FAILED;
7014 /* 2. Revert. Undo all the changes made above. */
7015 LIST_FOR_EACH_REVERSE_CONTINUE(be, node, &bundle->msg_list) {
7016 if (be->type == OFPTYPE_FLOW_MOD) {
7017 ofproto_flow_mod_revert(ofproto, &be->ofm);
7019 /* Nothing needs to be reverted for a port mod. */
7023 LIST_FOR_EACH (be, node, &bundle->msg_list) {
7024 if (be->type == OFPTYPE_FLOW_MOD) {
7025 struct flow_mod_requester req = { ofconn, be->ofp_msg };
7027 /* Bump the lookup version to the one of the current
7028 * message. This makes all the changes in the bundle at
7029 * this version visible to lookups at once. */
7030 if (ofproto->tables_version < be->ofm.version) {
7031 ofproto->tables_version = be->ofm.version;
7032 ofproto->ofproto_class->set_tables_version(
7033 ofproto, ofproto->tables_version);
7036 ofproto_flow_mod_finish(ofproto, &be->ofm, &req);
7037 } else if (be->type == OFPTYPE_PORT_MOD) {
7038 /* Perform the actual port mod. This is not atomic, i.e.,
7039 * the effects will be immediately seen by upcall
7040 * processing regardless of the lookup version. It should
7041 * be noted that port configuration changes can originate
7042 * also from OVSDB changes asynchronously to all upcall
7044 port_mod_finish(ofconn, &be->opm.pm, be->opm.port);
7049 ofmonitor_flush(ofproto->connmgr);
7050 ovs_mutex_unlock(&ofproto_mutex);
7052 run_rule_executes(ofproto);
7055 /* The bundle is discarded regardless the outcome. */
7056 ofp_bundle_remove__(ofconn, bundle, !error);
7061 handle_bundle_control(struct ofconn *ofconn, const struct ofp_header *oh)
7063 struct ofputil_bundle_ctrl_msg bctrl;
7064 struct ofputil_bundle_ctrl_msg reply;
7068 error = reject_slave_controller(ofconn);
7073 error = ofputil_decode_bundle_ctrl(oh, &bctrl);
7078 reply.bundle_id = bctrl.bundle_id;
7080 switch (bctrl.type) {
7081 case OFPBCT_OPEN_REQUEST:
7082 error = ofp_bundle_open(ofconn, bctrl.bundle_id, bctrl.flags);
7083 reply.type = OFPBCT_OPEN_REPLY;
7085 case OFPBCT_CLOSE_REQUEST:
7086 error = ofp_bundle_close(ofconn, bctrl.bundle_id, bctrl.flags);
7087 reply.type = OFPBCT_CLOSE_REPLY;
7089 case OFPBCT_COMMIT_REQUEST:
7090 error = do_bundle_commit(ofconn, bctrl.bundle_id, bctrl.flags);
7091 reply.type = OFPBCT_COMMIT_REPLY;
7093 case OFPBCT_DISCARD_REQUEST:
7094 error = ofp_bundle_discard(ofconn, bctrl.bundle_id);
7095 reply.type = OFPBCT_DISCARD_REPLY;
7098 case OFPBCT_OPEN_REPLY:
7099 case OFPBCT_CLOSE_REPLY:
7100 case OFPBCT_COMMIT_REPLY:
7101 case OFPBCT_DISCARD_REPLY:
7102 return OFPERR_OFPBFC_BAD_TYPE;
7107 buf = ofputil_encode_bundle_ctrl_reply(oh, &reply);
7108 ofconn_send_reply(ofconn, buf);
7114 handle_bundle_add(struct ofconn *ofconn, const struct ofp_header *oh)
7116 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
7118 struct ofputil_bundle_add_msg badd;
7119 struct ofp_bundle_entry *bmsg;
7122 error = reject_slave_controller(ofconn);
7127 error = ofputil_decode_bundle_add(oh, &badd, &type);
7132 bmsg = ofp_bundle_entry_alloc(type, badd.msg);
7134 if (type == OFPTYPE_PORT_MOD) {
7135 error = ofputil_decode_port_mod(badd.msg, &bmsg->opm.pm, false);
7136 } else if (type == OFPTYPE_FLOW_MOD) {
7137 struct ofpbuf ofpacts;
7138 uint64_t ofpacts_stub[1024 / 8];
7140 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
7141 error = ofputil_decode_flow_mod(&bmsg->ofm.fm, badd.msg,
7142 ofconn_get_protocol(ofconn),
7144 u16_to_ofp(ofproto->max_ports),
7146 /* Move actions to heap. */
7147 bmsg->ofm.fm.ofpacts = ofpbuf_steal_data(&ofpacts);
7149 if (!error && bmsg->ofm.fm.ofpacts_len) {
7150 error = ofproto_check_ofpacts(ofproto, bmsg->ofm.fm.ofpacts,
7151 bmsg->ofm.fm.ofpacts_len);
7158 error = ofp_bundle_add_message(ofconn, badd.bundle_id, badd.flags,
7163 ofp_bundle_entry_free(bmsg);
7170 handle_tlv_table_mod(struct ofconn *ofconn, const struct ofp_header *oh)
7172 struct ofputil_tlv_table_mod ttm;
7175 error = reject_slave_controller(ofconn);
7180 error = ofputil_decode_tlv_table_mod(oh, &ttm);
7185 error = tun_metadata_table_mod(&ttm);
7187 ofputil_uninit_tlv_table(&ttm.mappings);
7192 handle_tlv_table_request(struct ofconn *ofconn, const struct ofp_header *oh)
7194 struct ofputil_tlv_table_reply ttr;
7197 tun_metadata_table_request(&ttr);
7198 b = ofputil_encode_tlv_table_reply(oh, &ttr);
7199 ofputil_uninit_tlv_table(&ttr.mappings);
7201 ofconn_send_reply(ofconn, b);
7206 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
7207 OVS_EXCLUDED(ofproto_mutex)
7209 const struct ofp_header *oh = msg->data;
7213 error = ofptype_decode(&type, oh);
7217 if (oh->version >= OFP13_VERSION && ofpmsg_is_stat_request(oh)
7218 && ofpmp_more(oh)) {
7219 /* We have no buffer implementation for multipart requests.
7220 * Report overflow for requests which consists of multiple
7222 return OFPERR_OFPBRC_MULTIPART_BUFFER_OVERFLOW;
7226 /* OpenFlow requests. */
7227 case OFPTYPE_ECHO_REQUEST:
7228 return handle_echo_request(ofconn, oh);
7230 case OFPTYPE_FEATURES_REQUEST:
7231 return handle_features_request(ofconn, oh);
7233 case OFPTYPE_GET_CONFIG_REQUEST:
7234 return handle_get_config_request(ofconn, oh);
7236 case OFPTYPE_SET_CONFIG:
7237 return handle_set_config(ofconn, oh);
7239 case OFPTYPE_PACKET_OUT:
7240 return handle_packet_out(ofconn, oh);
7242 case OFPTYPE_PORT_MOD:
7243 return handle_port_mod(ofconn, oh);
7245 case OFPTYPE_FLOW_MOD:
7246 return handle_flow_mod(ofconn, oh);
7248 case OFPTYPE_GROUP_MOD:
7249 return handle_group_mod(ofconn, oh);
7251 case OFPTYPE_TABLE_MOD:
7252 return handle_table_mod(ofconn, oh);
7254 case OFPTYPE_METER_MOD:
7255 return handle_meter_mod(ofconn, oh);
7257 case OFPTYPE_BARRIER_REQUEST:
7258 return handle_barrier_request(ofconn, oh);
7260 case OFPTYPE_ROLE_REQUEST:
7261 return handle_role_request(ofconn, oh);
7263 /* OpenFlow replies. */
7264 case OFPTYPE_ECHO_REPLY:
7267 /* Nicira extension requests. */
7268 case OFPTYPE_FLOW_MOD_TABLE_ID:
7269 return handle_nxt_flow_mod_table_id(ofconn, oh);
7271 case OFPTYPE_SET_FLOW_FORMAT:
7272 return handle_nxt_set_flow_format(ofconn, oh);
7274 case OFPTYPE_SET_PACKET_IN_FORMAT:
7275 return handle_nxt_set_packet_in_format(ofconn, oh);
7277 case OFPTYPE_SET_CONTROLLER_ID:
7278 return handle_nxt_set_controller_id(ofconn, oh);
7280 case OFPTYPE_FLOW_AGE:
7281 /* Nothing to do. */
7284 case OFPTYPE_FLOW_MONITOR_CANCEL:
7285 return handle_flow_monitor_cancel(ofconn, oh);
7287 case OFPTYPE_SET_ASYNC_CONFIG:
7288 return handle_nxt_set_async_config(ofconn, oh);
7290 case OFPTYPE_GET_ASYNC_REQUEST:
7291 return handle_nxt_get_async_request(ofconn, oh);
7293 case OFPTYPE_NXT_RESUME:
7294 return handle_nxt_resume(ofconn, oh);
7296 /* Statistics requests. */
7297 case OFPTYPE_DESC_STATS_REQUEST:
7298 return handle_desc_stats_request(ofconn, oh);
7300 case OFPTYPE_FLOW_STATS_REQUEST:
7301 return handle_flow_stats_request(ofconn, oh);
7303 case OFPTYPE_AGGREGATE_STATS_REQUEST:
7304 return handle_aggregate_stats_request(ofconn, oh);
7306 case OFPTYPE_TABLE_STATS_REQUEST:
7307 return handle_table_stats_request(ofconn, oh);
7309 case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
7310 return handle_table_features_request(ofconn, oh);
7312 case OFPTYPE_TABLE_DESC_REQUEST:
7313 return handle_table_desc_request(ofconn, oh);
7315 case OFPTYPE_PORT_STATS_REQUEST:
7316 return handle_port_stats_request(ofconn, oh);
7318 case OFPTYPE_QUEUE_STATS_REQUEST:
7319 return handle_queue_stats_request(ofconn, oh);
7321 case OFPTYPE_PORT_DESC_STATS_REQUEST:
7322 return handle_port_desc_stats_request(ofconn, oh);
7324 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
7325 return handle_flow_monitor_request(ofconn, oh);
7327 case OFPTYPE_METER_STATS_REQUEST:
7328 case OFPTYPE_METER_CONFIG_STATS_REQUEST:
7329 return handle_meter_request(ofconn, oh, type);
7331 case OFPTYPE_METER_FEATURES_STATS_REQUEST:
7332 return handle_meter_features_request(ofconn, oh);
7334 case OFPTYPE_GROUP_STATS_REQUEST:
7335 return handle_group_stats_request(ofconn, oh);
7337 case OFPTYPE_GROUP_DESC_STATS_REQUEST:
7338 return handle_group_desc_stats_request(ofconn, oh);
7340 case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
7341 return handle_group_features_stats_request(ofconn, oh);
7343 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
7344 return handle_queue_get_config_request(ofconn, oh);
7346 case OFPTYPE_BUNDLE_CONTROL:
7347 return handle_bundle_control(ofconn, oh);
7349 case OFPTYPE_BUNDLE_ADD_MESSAGE:
7350 return handle_bundle_add(ofconn, oh);
7352 case OFPTYPE_NXT_TLV_TABLE_MOD:
7353 return handle_tlv_table_mod(ofconn, oh);
7355 case OFPTYPE_NXT_TLV_TABLE_REQUEST:
7356 return handle_tlv_table_request(ofconn, oh);
7360 case OFPTYPE_FEATURES_REPLY:
7361 case OFPTYPE_GET_CONFIG_REPLY:
7362 case OFPTYPE_PACKET_IN:
7363 case OFPTYPE_FLOW_REMOVED:
7364 case OFPTYPE_PORT_STATUS:
7365 case OFPTYPE_BARRIER_REPLY:
7366 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
7367 case OFPTYPE_DESC_STATS_REPLY:
7368 case OFPTYPE_FLOW_STATS_REPLY:
7369 case OFPTYPE_QUEUE_STATS_REPLY:
7370 case OFPTYPE_PORT_STATS_REPLY:
7371 case OFPTYPE_TABLE_STATS_REPLY:
7372 case OFPTYPE_AGGREGATE_STATS_REPLY:
7373 case OFPTYPE_PORT_DESC_STATS_REPLY:
7374 case OFPTYPE_ROLE_REPLY:
7375 case OFPTYPE_FLOW_MONITOR_PAUSED:
7376 case OFPTYPE_FLOW_MONITOR_RESUMED:
7377 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
7378 case OFPTYPE_GET_ASYNC_REPLY:
7379 case OFPTYPE_GROUP_STATS_REPLY:
7380 case OFPTYPE_GROUP_DESC_STATS_REPLY:
7381 case OFPTYPE_GROUP_FEATURES_STATS_REPLY:
7382 case OFPTYPE_METER_STATS_REPLY:
7383 case OFPTYPE_METER_CONFIG_STATS_REPLY:
7384 case OFPTYPE_METER_FEATURES_STATS_REPLY:
7385 case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
7386 case OFPTYPE_TABLE_DESC_REPLY:
7387 case OFPTYPE_ROLE_STATUS:
7388 case OFPTYPE_REQUESTFORWARD:
7389 case OFPTYPE_TABLE_STATUS:
7390 case OFPTYPE_NXT_TLV_TABLE_REPLY:
7392 if (ofpmsg_is_stat_request(oh)) {
7393 return OFPERR_OFPBRC_BAD_STAT;
7395 return OFPERR_OFPBRC_BAD_TYPE;
7401 handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
7402 OVS_EXCLUDED(ofproto_mutex)
7404 enum ofperr error = handle_openflow__(ofconn, ofp_msg);
7407 ofconn_send_error(ofconn, ofp_msg->data, error);
7409 COVERAGE_INC(ofproto_recv_openflow);
7412 /* Asynchronous operations. */
7415 send_buffered_packet(const struct flow_mod_requester *req, uint32_t buffer_id,
7417 OVS_REQUIRES(ofproto_mutex)
7419 if (req && req->ofconn && buffer_id != UINT32_MAX) {
7420 struct ofproto *ofproto = ofconn_get_ofproto(req->ofconn);
7421 struct dp_packet *packet;
7425 error = ofconn_pktbuf_retrieve(req->ofconn, buffer_id, &packet,
7428 struct rule_execute *re;
7430 ofproto_rule_ref(rule);
7432 re = xmalloc(sizeof *re);
7434 re->in_port = in_port;
7435 re->packet = packet;
7437 if (!guarded_list_push_back(&ofproto->rule_executes,
7438 &re->list_node, 1024)) {
7439 ofproto_rule_unref(rule);
7440 dp_packet_delete(re->packet);
7444 ofconn_send_error(req->ofconn, req->request, error);
7450 pick_datapath_id(const struct ofproto *ofproto)
7452 const struct ofport *port;
7454 port = ofproto_get_port(ofproto, OFPP_LOCAL);
7459 error = netdev_get_etheraddr(port->netdev, &ea);
7461 return eth_addr_to_uint64(ea);
7463 VLOG_WARN("%s: could not get MAC address for %s (%s)",
7464 ofproto->name, netdev_get_name(port->netdev),
7465 ovs_strerror(error));
7467 return ofproto->fallback_dpid;
7471 pick_fallback_dpid(void)
7474 eth_addr_nicira_random(&ea);
7475 return eth_addr_to_uint64(ea);
7478 /* Table overflow policy. */
7480 /* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep'
7481 * to NULL if the table is not configured to evict rules or if the table
7482 * contains no evictable rules. (Rules with a readlock on their evict rwlock,
7483 * or with no timeouts are not evictable.) */
7485 choose_rule_to_evict(struct oftable *table, struct rule **rulep)
7486 OVS_REQUIRES(ofproto_mutex)
7488 struct eviction_group *evg;
7491 if (!table->eviction) {
7495 /* In the common case, the outer and inner loops here will each be entered
7498 * - The inner loop normally "return"s in its first iteration. If the
7499 * eviction group has any evictable rules, then it always returns in
7502 * - The outer loop only iterates more than once if the largest eviction
7503 * group has no evictable rules.
7505 * - The outer loop can exit only if table's 'max_flows' is all filled up
7506 * by unevictable rules. */
7507 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
7510 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
7519 /* Eviction groups. */
7521 /* Returns the priority to use for an eviction_group that contains 'n_rules'
7522 * rules. The priority contains low-order random bits to ensure that eviction
7523 * groups with the same number of rules are prioritized randomly. */
7525 eviction_group_priority(size_t n_rules)
7527 uint16_t size = MIN(UINT16_MAX, n_rules);
7528 return (size << 16) | random_uint16();
7531 /* Updates 'evg', an eviction_group within 'table', following a change that
7532 * adds or removes rules in 'evg'. */
7534 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
7535 OVS_REQUIRES(ofproto_mutex)
7537 heap_change(&table->eviction_groups_by_size, &evg->size_node,
7538 eviction_group_priority(heap_count(&evg->rules)));
7541 /* Destroys 'evg', an eviction_group within 'table':
7543 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
7544 * rules themselves, just removes them from the eviction group.)
7546 * - Removes 'evg' from 'table'.
7550 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
7551 OVS_REQUIRES(ofproto_mutex)
7553 while (!heap_is_empty(&evg->rules)) {
7556 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
7557 rule->eviction_group = NULL;
7559 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
7560 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
7561 heap_destroy(&evg->rules);
7565 /* Removes 'rule' from its eviction group, if any. */
7567 eviction_group_remove_rule(struct rule *rule)
7568 OVS_REQUIRES(ofproto_mutex)
7570 if (rule->eviction_group) {
7571 struct oftable *table = &rule->ofproto->tables[rule->table_id];
7572 struct eviction_group *evg = rule->eviction_group;
7574 rule->eviction_group = NULL;
7575 heap_remove(&evg->rules, &rule->evg_node);
7576 if (heap_is_empty(&evg->rules)) {
7577 eviction_group_destroy(table, evg);
7579 eviction_group_resized(table, evg);
7584 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
7585 * returns the hash value. */
7587 eviction_group_hash_rule(struct rule *rule)
7588 OVS_REQUIRES(ofproto_mutex)
7590 struct oftable *table = &rule->ofproto->tables[rule->table_id];
7591 const struct mf_subfield *sf;
7595 hash = table->eviction_group_id_basis;
7596 miniflow_expand(rule->cr.match.flow, &flow);
7597 for (sf = table->eviction_fields;
7598 sf < &table->eviction_fields[table->n_eviction_fields];
7601 if (mf_are_prereqs_ok(sf->field, &flow)) {
7602 union mf_value value;
7604 mf_get_value(sf->field, &flow, &value);
7606 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
7608 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
7609 unsigned int start = sf->ofs + sf->n_bits;
7610 bitwise_zero(&value, sf->field->n_bytes, start,
7611 sf->field->n_bytes * 8 - start);
7613 hash = hash_bytes(&value, sf->field->n_bytes, hash);
7615 hash = hash_int(hash, 0);
7622 /* Returns an eviction group within 'table' with the given 'id', creating one
7624 static struct eviction_group *
7625 eviction_group_find(struct oftable *table, uint32_t id)
7626 OVS_REQUIRES(ofproto_mutex)
7628 struct eviction_group *evg;
7630 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
7634 evg = xmalloc(sizeof *evg);
7635 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
7636 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
7637 eviction_group_priority(0));
7638 heap_init(&evg->rules);
7643 /* Returns an eviction priority for 'rule'. The return value should be
7644 * interpreted so that higher priorities make a rule a more attractive
7645 * candidate for eviction. */
7647 rule_eviction_priority(struct ofproto *ofproto, struct rule *rule)
7648 OVS_REQUIRES(ofproto_mutex)
7650 /* Calculate absolute time when this flow will expire. If it will never
7651 * expire, then return 0 to make it unevictable. */
7652 long long int expiration = LLONG_MAX;
7653 if (rule->hard_timeout) {
7654 /* 'modified' needs protection even when we hold 'ofproto_mutex'. */
7655 ovs_mutex_lock(&rule->mutex);
7656 long long int modified = rule->modified;
7657 ovs_mutex_unlock(&rule->mutex);
7659 expiration = modified + rule->hard_timeout * 1000;
7661 if (rule->idle_timeout) {
7662 uint64_t packets, bytes;
7664 long long int idle_expiration;
7666 ofproto->ofproto_class->rule_get_stats(rule, &packets, &bytes, &used);
7667 idle_expiration = used + rule->idle_timeout * 1000;
7668 expiration = MIN(expiration, idle_expiration);
7670 if (expiration == LLONG_MAX) {
7674 /* Calculate the time of expiration as a number of (approximate) seconds
7675 * after program startup.
7677 * This should work OK for program runs that last UINT32_MAX seconds or
7678 * less. Therefore, please restart OVS at least once every 136 years. */
7679 uint32_t expiration_ofs = (expiration >> 10) - (time_boot_msec() >> 10);
7681 /* Combine expiration time with OpenFlow "importance" to form a single
7682 * priority value. We want flows with relatively low "importance" to be
7683 * evicted before even considering expiration time, so put "importance" in
7684 * the most significant bits and expiration time in the least significant
7687 * Small 'priority' should be evicted before those with large 'priority'.
7688 * The caller expects the opposite convention (a large return value being
7689 * more attractive for eviction) so we invert it before returning. */
7690 uint64_t priority = ((uint64_t) rule->importance << 32) + expiration_ofs;
7691 return UINT64_MAX - priority;
7694 /* Adds 'rule' to an appropriate eviction group for its oftable's
7695 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
7696 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
7699 * The caller must ensure that 'rule' is not already in an eviction group. */
7701 eviction_group_add_rule(struct rule *rule)
7702 OVS_REQUIRES(ofproto_mutex)
7704 struct ofproto *ofproto = rule->ofproto;
7705 struct oftable *table = &ofproto->tables[rule->table_id];
7708 /* Timeouts may be modified only when holding 'ofproto_mutex'. We have it
7709 * so no additional protection is needed. */
7710 has_timeout = rule->hard_timeout || rule->idle_timeout;
7712 if (table->eviction && has_timeout) {
7713 struct eviction_group *evg;
7715 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
7717 rule->eviction_group = evg;
7718 heap_insert(&evg->rules, &rule->evg_node,
7719 rule_eviction_priority(ofproto, rule));
7720 eviction_group_resized(table, evg);
7726 /* Initializes 'table'. */
7728 oftable_init(struct oftable *table)
7730 memset(table, 0, sizeof *table);
7731 classifier_init(&table->cls, flow_segment_u64s);
7732 table->max_flows = UINT_MAX;
7734 hmap_init(&table->eviction_groups_by_id);
7735 heap_init(&table->eviction_groups_by_size);
7736 atomic_init(&table->miss_config, OFPUTIL_TABLE_MISS_DEFAULT);
7738 classifier_set_prefix_fields(&table->cls, default_prefix_fields,
7739 ARRAY_SIZE(default_prefix_fields));
7741 atomic_init(&table->n_matched, 0);
7742 atomic_init(&table->n_missed, 0);
7745 /* Destroys 'table', including its classifier and eviction groups.
7747 * The caller is responsible for freeing 'table' itself. */
7749 oftable_destroy(struct oftable *table)
7751 ovs_assert(classifier_is_empty(&table->cls));
7753 ovs_mutex_lock(&ofproto_mutex);
7754 oftable_configure_eviction(table, 0, NULL, 0);
7755 ovs_mutex_unlock(&ofproto_mutex);
7757 hmap_destroy(&table->eviction_groups_by_id);
7758 heap_destroy(&table->eviction_groups_by_size);
7759 classifier_destroy(&table->cls);
7763 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
7764 * string, then 'table' will use its default name.
7766 * This only affects the name exposed for a table exposed through the OpenFlow
7767 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
7769 oftable_set_name(struct oftable *table, const char *name)
7771 if (name && name[0]) {
7772 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
7773 if (!table->name || strncmp(name, table->name, len)) {
7775 table->name = xmemdup0(name, len);
7783 /* oftables support a choice of two policies when adding a rule would cause the
7784 * number of flows in the table to exceed the configured maximum number: either
7785 * they can refuse to add the new flow or they can evict some existing flow.
7786 * This function configures the latter policy on 'table', with fairness based
7787 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
7788 * 'n_fields' as 0 disables fairness.) */
7790 oftable_configure_eviction(struct oftable *table, unsigned int eviction,
7791 const struct mf_subfield *fields, size_t n_fields)
7792 OVS_REQUIRES(ofproto_mutex)
7796 if ((table->eviction != 0) == (eviction != 0)
7797 && n_fields == table->n_eviction_fields
7799 || !memcmp(fields, table->eviction_fields,
7800 n_fields * sizeof *fields))) {
7801 /* The set of eviction fields did not change. If 'eviction' changed,
7802 * it remains nonzero, so that we can just update table->eviction
7803 * without fussing with the eviction groups. */
7804 table->eviction = eviction;
7808 /* Destroy existing eviction groups, then destroy and recreate data
7809 * structures to recover memory. */
7810 struct eviction_group *evg, *next;
7811 HMAP_FOR_EACH_SAFE (evg, next, id_node, &table->eviction_groups_by_id) {
7812 eviction_group_destroy(table, evg);
7814 hmap_destroy(&table->eviction_groups_by_id);
7815 hmap_init(&table->eviction_groups_by_id);
7816 heap_destroy(&table->eviction_groups_by_size);
7817 heap_init(&table->eviction_groups_by_size);
7819 /* Replace eviction groups by the new ones, if there is a change. Free the
7820 * old fields only after allocating the new ones, because 'fields ==
7821 * table->eviction_fields' is possible. */
7822 struct mf_subfield *old_fields = table->eviction_fields;
7823 table->n_eviction_fields = n_fields;
7824 table->eviction_fields = (fields
7825 ? xmemdup(fields, n_fields * sizeof *fields)
7829 /* Add the new eviction groups, if enabled. */
7830 table->eviction = eviction;
7831 if (table->eviction) {
7832 table->eviction_group_id_basis = random_uint32();
7833 CLS_FOR_EACH (rule, cr, &table->cls) {
7834 eviction_group_add_rule(rule);
7839 /* Inserts 'rule' from the ofproto data structures BEFORE caller has inserted
7840 * it to the classifier. */
7842 ofproto_rule_insert__(struct ofproto *ofproto, struct rule *rule)
7843 OVS_REQUIRES(ofproto_mutex)
7845 const struct rule_actions *actions = rule_get_actions(rule);
7847 ovs_assert(rule->removed);
7849 if (rule->hard_timeout || rule->idle_timeout) {
7850 ovs_list_insert(&ofproto->expirable, &rule->expirable);
7852 cookies_insert(ofproto, rule);
7853 eviction_group_add_rule(rule);
7854 if (actions->has_meter) {
7855 meter_insert_rule(rule);
7857 rule->removed = false;
7860 /* Removes 'rule' from the ofproto data structures. Caller may have deferred
7861 * the removal from the classifier. */
7863 ofproto_rule_remove__(struct ofproto *ofproto, struct rule *rule)
7864 OVS_REQUIRES(ofproto_mutex)
7866 ovs_assert(!rule->removed);
7868 cookies_remove(ofproto, rule);
7870 eviction_group_remove_rule(rule);
7871 if (!ovs_list_is_empty(&rule->expirable)) {
7872 ovs_list_remove(&rule->expirable);
7874 if (!ovs_list_is_empty(&rule->meter_list_node)) {
7875 ovs_list_remove(&rule->meter_list_node);
7876 ovs_list_init(&rule->meter_list_node);
7879 rule->removed = true;
7882 /* unixctl commands. */
7885 ofproto_lookup(const char *name)
7887 struct ofproto *ofproto;
7889 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
7891 if (!strcmp(ofproto->name, name)) {
7899 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
7900 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7902 struct ofproto *ofproto;
7906 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
7907 ds_put_format(&results, "%s\n", ofproto->name);
7909 unixctl_command_reply(conn, ds_cstr(&results));
7910 ds_destroy(&results);
7914 ofproto_unixctl_init(void)
7916 static bool registered;
7922 unixctl_command_register("ofproto/list", "", 0, 0,
7923 ofproto_unixctl_list, NULL);
7926 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
7928 * This is deprecated. It is only for compatibility with broken device drivers
7929 * in old versions of Linux that do not properly support VLANs when VLAN
7930 * devices are not used. When broken device drivers are no longer in
7931 * widespread use, we will delete these interfaces. */
7933 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
7934 * (exactly) by an OpenFlow rule in 'ofproto'. */
7936 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
7939 struct cls_rule target;
7940 const struct oftable *oftable;
7942 match_init_catchall(&match);
7943 match_set_vlan_vid_masked(&match, htons(VLAN_CFI), htons(VLAN_CFI));
7944 cls_rule_init(&target, &match, 0);
7946 free(ofproto->vlan_bitmap);
7947 ofproto->vlan_bitmap = bitmap_allocate(4096);
7948 ofproto->vlans_changed = false;
7950 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
7953 CLS_FOR_EACH_TARGET (rule, cr, &oftable->cls, &target,
7955 if (minimask_get_vid_mask(rule->cr.match.mask) == VLAN_VID_MASK) {
7956 uint16_t vid = miniflow_get_vid(rule->cr.match.flow);
7958 bitmap_set1(vlan_bitmap, vid);
7959 bitmap_set1(ofproto->vlan_bitmap, vid);
7964 cls_rule_destroy(&target);
7967 /* Returns true if new VLANs have come into use by the flow table since the
7968 * last call to ofproto_get_vlan_usage().
7970 * We don't track when old VLANs stop being used. */
7972 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
7974 return ofproto->vlans_changed;
7977 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
7978 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
7979 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
7980 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
7981 * then the VLAN device is un-enslaved. */
7983 ofproto_port_set_realdev(struct ofproto *ofproto, ofp_port_t vlandev_ofp_port,
7984 ofp_port_t realdev_ofp_port, int vid)
7986 struct ofport *ofport;
7989 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
7991 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
7993 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
7994 ofproto->name, vlandev_ofp_port);
7998 if (!ofproto->ofproto_class->set_realdev) {
7999 if (!vlandev_ofp_port) {
8002 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
8006 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
8008 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
8009 ofproto->name, vlandev_ofp_port,
8010 netdev_get_name(ofport->netdev), ovs_strerror(error));