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);
222 static inline bool ofport_is_internal(const struct ofport *);
224 static int update_port(struct ofproto *, const char *devname);
225 static int init_ports(struct ofproto *);
226 static void reinit_ports(struct ofproto *);
228 static long long int ofport_get_usage(const struct ofproto *,
229 ofp_port_t ofp_port);
230 static void ofport_set_usage(struct ofproto *, ofp_port_t ofp_port,
231 long long int last_used);
232 static void ofport_remove_usage(struct ofproto *, ofp_port_t ofp_port);
236 * Keeps track of the currently used and recently used ofport values and is
237 * used to prevent immediate recycling of ofport values. */
238 struct ofport_usage {
239 struct hmap_node hmap_node; /* In struct ofproto's "ofport_usage" hmap. */
240 ofp_port_t ofp_port; /* OpenFlow port number. */
241 long long int last_used; /* Last time the 'ofp_port' was used. LLONG_MAX
242 represents in-use ofports. */
246 static void ofproto_rule_send_removed(struct rule *)
247 OVS_EXCLUDED(ofproto_mutex);
248 static bool rule_is_readonly(const struct rule *);
249 static void ofproto_rule_insert__(struct ofproto *, struct rule *)
250 OVS_REQUIRES(ofproto_mutex);
251 static void ofproto_rule_remove__(struct ofproto *, struct rule *)
252 OVS_REQUIRES(ofproto_mutex);
254 /* The source of a flow_mod request, in the code that processes flow_mods.
256 * A flow table modification request can be generated externally, via OpenFlow,
257 * or internally through a function call. This structure indicates the source
258 * of an OpenFlow-generated flow_mod. For an internal flow_mod, it isn't
259 * meaningful and thus supplied as NULL. */
260 struct flow_mod_requester {
261 struct ofconn *ofconn; /* Connection on which flow_mod arrived. */
262 const struct ofp_header *request;
266 static enum ofperr replace_rule_create(struct ofproto *,
267 struct ofputil_flow_mod *,
268 struct cls_rule *cr, uint8_t table_id,
269 struct rule *old_rule,
270 struct rule **new_rule)
271 OVS_REQUIRES(ofproto_mutex);
273 static void replace_rule_start(struct ofproto *, cls_version_t version,
274 struct rule *old_rule, struct rule *new_rule,
275 struct cls_conjunction *, size_t n_conjs)
276 OVS_REQUIRES(ofproto_mutex);
278 static void replace_rule_revert(struct ofproto *, struct rule *old_rule,
279 struct rule *new_rule)
280 OVS_REQUIRES(ofproto_mutex);
282 static void replace_rule_finish(struct ofproto *, struct ofputil_flow_mod *,
283 const struct flow_mod_requester *,
284 struct rule *old_rule, struct rule *new_rule,
285 struct ovs_list *dead_cookies)
286 OVS_REQUIRES(ofproto_mutex);
287 static void delete_flows__(struct rule_collection *,
288 enum ofp_flow_removed_reason,
289 const struct flow_mod_requester *)
290 OVS_REQUIRES(ofproto_mutex);
292 static void send_buffered_packet(const struct flow_mod_requester *,
293 uint32_t buffer_id, struct rule *)
294 OVS_REQUIRES(ofproto_mutex);
296 static bool ofproto_group_exists__(const struct ofproto *ofproto,
298 OVS_REQ_RDLOCK(ofproto->groups_rwlock);
299 static bool ofproto_group_exists(const struct ofproto *ofproto,
301 OVS_EXCLUDED(ofproto->groups_rwlock);
302 static enum ofperr add_group(struct ofproto *,
303 const struct ofputil_group_mod *);
304 static void handle_openflow(struct ofconn *, const struct ofpbuf *);
305 static enum ofperr ofproto_flow_mod_start(struct ofproto *,
306 struct ofproto_flow_mod *)
307 OVS_REQUIRES(ofproto_mutex);
308 static void ofproto_flow_mod_finish(struct ofproto *,
309 struct ofproto_flow_mod *,
310 const struct flow_mod_requester *)
311 OVS_REQUIRES(ofproto_mutex);
312 static enum ofperr handle_flow_mod__(struct ofproto *,
313 struct ofproto_flow_mod *,
314 const struct flow_mod_requester *)
315 OVS_EXCLUDED(ofproto_mutex);
316 static void calc_duration(long long int start, long long int now,
317 uint32_t *sec, uint32_t *nsec);
320 static uint64_t pick_datapath_id(const struct ofproto *);
321 static uint64_t pick_fallback_dpid(void);
322 static void ofproto_destroy__(struct ofproto *);
323 static void update_mtu(struct ofproto *, struct ofport *);
324 static void update_mtu_ofproto(struct ofproto *);
325 static void meter_delete(struct ofproto *, uint32_t first, uint32_t last);
326 static void meter_insert_rule(struct rule *);
329 static void ofproto_unixctl_init(void);
331 /* All registered ofproto classes, in probe order. */
332 static const struct ofproto_class **ofproto_classes;
333 static size_t n_ofproto_classes;
334 static size_t allocated_ofproto_classes;
336 /* Global lock that protects all flow table operations. */
337 struct ovs_mutex ofproto_mutex = OVS_MUTEX_INITIALIZER;
339 unsigned ofproto_flow_limit = OFPROTO_FLOW_LIMIT_DEFAULT;
340 unsigned ofproto_max_idle = OFPROTO_MAX_IDLE_DEFAULT;
342 size_t n_handlers, n_revalidators;
345 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
346 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
348 /* Initial mappings of port to OpenFlow number mappings. */
349 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
351 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
353 /* The default value of true waits for flow restore. */
354 static bool flow_restore_wait = true;
356 /* Must be called to initialize the ofproto library.
358 * The caller may pass in 'iface_hints', which contains an shash of
359 * "iface_hint" elements indexed by the interface's name. The provider
360 * may use these hints to describe the startup configuration in order to
361 * reinitialize its state. The caller owns the provided data, so a
362 * provider will make copies of anything required. An ofproto provider
363 * will remove any existing state that is not described by the hint, and
364 * may choose to remove it all. */
366 ofproto_init(const struct shash *iface_hints)
368 struct shash_node *node;
371 ofproto_class_register(&ofproto_dpif_class);
373 /* Make a local copy, since we don't own 'iface_hints' elements. */
374 SHASH_FOR_EACH(node, iface_hints) {
375 const struct iface_hint *orig_hint = node->data;
376 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
377 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
379 new_hint->br_name = xstrdup(orig_hint->br_name);
380 new_hint->br_type = xstrdup(br_type);
381 new_hint->ofp_port = orig_hint->ofp_port;
383 shash_add(&init_ofp_ports, node->name, new_hint);
386 for (i = 0; i < n_ofproto_classes; i++) {
387 ofproto_classes[i]->init(&init_ofp_ports);
390 ofproto_unixctl_init();
393 /* 'type' should be a normalized datapath type, as returned by
394 * ofproto_normalize_type(). Returns the corresponding ofproto_class
395 * structure, or a null pointer if there is none registered for 'type'. */
396 static const struct ofproto_class *
397 ofproto_class_find__(const char *type)
401 for (i = 0; i < n_ofproto_classes; i++) {
402 const struct ofproto_class *class = ofproto_classes[i];
407 class->enumerate_types(&types);
408 found = sset_contains(&types, type);
409 sset_destroy(&types);
415 VLOG_WARN("unknown datapath type %s", type);
419 /* Registers a new ofproto class. After successful registration, new ofprotos
420 * of that type can be created using ofproto_create(). */
422 ofproto_class_register(const struct ofproto_class *new_class)
426 for (i = 0; i < n_ofproto_classes; i++) {
427 if (ofproto_classes[i] == new_class) {
432 if (n_ofproto_classes >= allocated_ofproto_classes) {
433 ofproto_classes = x2nrealloc(ofproto_classes,
434 &allocated_ofproto_classes,
435 sizeof *ofproto_classes);
437 ofproto_classes[n_ofproto_classes++] = new_class;
441 /* Unregisters a datapath provider. 'type' must have been previously
442 * registered and not currently be in use by any ofprotos. After
443 * unregistration new datapaths of that type cannot be opened using
444 * ofproto_create(). */
446 ofproto_class_unregister(const struct ofproto_class *class)
450 for (i = 0; i < n_ofproto_classes; i++) {
451 if (ofproto_classes[i] == class) {
452 for (i++; i < n_ofproto_classes; i++) {
453 ofproto_classes[i - 1] = ofproto_classes[i];
459 VLOG_WARN("attempted to unregister an ofproto class that is not "
464 /* Clears 'types' and enumerates all registered ofproto types into it. The
465 * caller must first initialize the sset. */
467 ofproto_enumerate_types(struct sset *types)
472 for (i = 0; i < n_ofproto_classes; i++) {
473 ofproto_classes[i]->enumerate_types(types);
477 /* Returns the fully spelled out name for the given ofproto 'type'.
479 * Normalized type string can be compared with strcmp(). Unnormalized type
480 * string might be the same even if they have different spellings. */
482 ofproto_normalize_type(const char *type)
484 return type && type[0] ? type : "system";
487 /* Clears 'names' and enumerates the names of all known created ofprotos with
488 * the given 'type'. The caller must first initialize the sset. Returns 0 if
489 * successful, otherwise a positive errno value.
491 * Some kinds of datapaths might not be practically enumerable. This is not
492 * considered an error. */
494 ofproto_enumerate_names(const char *type, struct sset *names)
496 const struct ofproto_class *class = ofproto_class_find__(type);
497 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
501 ofproto_bump_tables_version(struct ofproto *ofproto)
503 ++ofproto->tables_version;
504 ofproto->ofproto_class->set_tables_version(ofproto,
505 ofproto->tables_version);
509 ofproto_create(const char *datapath_name, const char *datapath_type,
510 struct ofproto **ofprotop)
512 const struct ofproto_class *class;
513 struct ofproto *ofproto;
519 datapath_type = ofproto_normalize_type(datapath_type);
520 class = ofproto_class_find__(datapath_type);
522 VLOG_WARN("could not create datapath %s of unknown type %s",
523 datapath_name, datapath_type);
527 ofproto = class->alloc();
529 VLOG_ERR("failed to allocate datapath %s of type %s",
530 datapath_name, datapath_type);
535 ovs_mutex_lock(&ofproto_mutex);
536 memset(ofproto, 0, sizeof *ofproto);
537 ofproto->ofproto_class = class;
538 ofproto->name = xstrdup(datapath_name);
539 ofproto->type = xstrdup(datapath_type);
540 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
541 hash_string(ofproto->name, 0));
542 ofproto->datapath_id = 0;
543 ofproto->forward_bpdu = false;
544 ofproto->fallback_dpid = pick_fallback_dpid();
545 ofproto->mfr_desc = NULL;
546 ofproto->hw_desc = NULL;
547 ofproto->sw_desc = NULL;
548 ofproto->serial_desc = NULL;
549 ofproto->dp_desc = NULL;
550 ofproto->frag_handling = OFPUTIL_FRAG_NORMAL;
551 hmap_init(&ofproto->ports);
552 hmap_init(&ofproto->ofport_usage);
553 shash_init(&ofproto->port_by_name);
554 simap_init(&ofproto->ofp_requests);
555 ofproto->max_ports = ofp_to_u16(OFPP_MAX);
556 ofproto->eviction_group_timer = LLONG_MIN;
557 ofproto->tables = NULL;
558 ofproto->n_tables = 0;
559 ofproto->tables_version = CLS_MIN_VERSION;
560 hindex_init(&ofproto->cookies);
561 hmap_init(&ofproto->learned_cookies);
562 ovs_list_init(&ofproto->expirable);
563 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
564 guarded_list_init(&ofproto->rule_executes);
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 ofproto->ofproto_class->dealloc(ofproto);
1564 /* Destroying rules is doubly deferred, must have 'ofproto' around for them.
1565 * - 1st we defer the removal of the rules from the classifier
1566 * - 2nd we defer the actual destruction of the rules. */
1568 ofproto_destroy_defer__(struct ofproto *ofproto)
1569 OVS_EXCLUDED(ofproto_mutex)
1571 ovsrcu_postpone(ofproto_destroy__, ofproto);
1575 ofproto_destroy(struct ofproto *p, bool del)
1576 OVS_EXCLUDED(ofproto_mutex)
1578 struct ofport *ofport, *next_ofport;
1579 struct ofport_usage *usage;
1586 meter_delete(p, 1, p->meter_features.max_meters);
1587 p->meter_features.max_meters = 0;
1593 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1594 ofport_destroy(ofport, del);
1597 HMAP_FOR_EACH_POP (usage, hmap_node, &p->ofport_usage) {
1601 p->ofproto_class->destruct(p);
1603 /* We should not postpone this because it involves deleting a listening
1604 * socket which we may want to reopen soon. 'connmgr' should not be used
1605 * by other threads */
1606 connmgr_destroy(p->connmgr);
1608 /* Destroying rules is deferred, must have 'ofproto' around for them. */
1609 ovsrcu_postpone(ofproto_destroy_defer__, p);
1612 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1613 * kernel datapath, for example, this destroys the datapath in the kernel, and
1614 * with the netdev-based datapath, it tears down the data structures that
1615 * represent the datapath.
1617 * The datapath should not be currently open as an ofproto. */
1619 ofproto_delete(const char *name, const char *type)
1621 const struct ofproto_class *class = ofproto_class_find__(type);
1622 return (!class ? EAFNOSUPPORT
1623 : !class->del ? EACCES
1624 : class->del(type, name));
1628 process_port_change(struct ofproto *ofproto, int error, char *devname)
1630 if (error == ENOBUFS) {
1631 reinit_ports(ofproto);
1632 } else if (!error) {
1633 update_port(ofproto, devname);
1639 ofproto_type_run(const char *datapath_type)
1641 const struct ofproto_class *class;
1644 datapath_type = ofproto_normalize_type(datapath_type);
1645 class = ofproto_class_find__(datapath_type);
1647 error = class->type_run ? class->type_run(datapath_type) : 0;
1648 if (error && error != EAGAIN) {
1649 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1650 datapath_type, ovs_strerror(error));
1656 ofproto_type_wait(const char *datapath_type)
1658 const struct ofproto_class *class;
1660 datapath_type = ofproto_normalize_type(datapath_type);
1661 class = ofproto_class_find__(datapath_type);
1663 if (class->type_wait) {
1664 class->type_wait(datapath_type);
1669 ofproto_run(struct ofproto *p)
1674 error = p->ofproto_class->run(p);
1675 if (error && error != EAGAIN) {
1676 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, ovs_strerror(error));
1679 run_rule_executes(p);
1681 /* Restore the eviction group heap invariant occasionally. */
1682 if (p->eviction_group_timer < time_msec()) {
1685 p->eviction_group_timer = time_msec() + 1000;
1687 for (i = 0; i < p->n_tables; i++) {
1688 struct oftable *table = &p->tables[i];
1689 struct eviction_group *evg;
1692 if (!table->eviction) {
1696 if (table->n_flows > 100000) {
1697 static struct vlog_rate_limit count_rl =
1698 VLOG_RATE_LIMIT_INIT(1, 1);
1699 VLOG_WARN_RL(&count_rl, "Table %"PRIuSIZE" has an excessive"
1700 " number of rules: %d", i, table->n_flows);
1703 ovs_mutex_lock(&ofproto_mutex);
1704 CLS_FOR_EACH (rule, cr, &table->cls) {
1705 if (rule->idle_timeout || rule->hard_timeout) {
1706 if (!rule->eviction_group) {
1707 eviction_group_add_rule(rule);
1709 heap_raw_change(&rule->evg_node,
1710 rule_eviction_priority(p, rule));
1715 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
1716 heap_rebuild(&evg->rules);
1718 ovs_mutex_unlock(&ofproto_mutex);
1722 if (p->ofproto_class->port_poll) {
1725 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1726 process_port_change(p, error, devname);
1730 new_seq = seq_read(connectivity_seq_get());
1731 if (new_seq != p->change_seq) {
1732 struct sset devnames;
1733 const char *devname;
1734 struct ofport *ofport;
1736 /* Update OpenFlow port status for any port whose netdev has changed.
1738 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1739 * destroyed, so it's not safe to update ports directly from the
1740 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1741 * need this two-phase approach. */
1742 sset_init(&devnames);
1743 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1744 uint64_t port_change_seq;
1746 port_change_seq = netdev_get_change_seq(ofport->netdev);
1747 if (ofport->change_seq != port_change_seq) {
1748 ofport->change_seq = port_change_seq;
1749 sset_add(&devnames, netdev_get_name(ofport->netdev));
1752 SSET_FOR_EACH (devname, &devnames) {
1753 update_port(p, devname);
1755 sset_destroy(&devnames);
1757 p->change_seq = new_seq;
1760 connmgr_run(p->connmgr, handle_openflow);
1766 ofproto_wait(struct ofproto *p)
1768 p->ofproto_class->wait(p);
1769 if (p->ofproto_class->port_poll_wait) {
1770 p->ofproto_class->port_poll_wait(p);
1772 seq_wait(connectivity_seq_get(), p->change_seq);
1773 connmgr_wait(p->connmgr);
1777 ofproto_is_alive(const struct ofproto *p)
1779 return connmgr_has_controllers(p->connmgr);
1782 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1783 * memory_report(). */
1785 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1787 const struct oftable *table;
1788 unsigned int n_rules;
1790 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1793 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1794 n_rules += table->n_flows;
1796 simap_increase(usage, "rules", n_rules);
1798 if (ofproto->ofproto_class->get_memory_usage) {
1799 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1802 connmgr_get_memory_usage(ofproto->connmgr, usage);
1806 ofproto_type_get_memory_usage(const char *datapath_type, struct simap *usage)
1808 const struct ofproto_class *class;
1810 datapath_type = ofproto_normalize_type(datapath_type);
1811 class = ofproto_class_find__(datapath_type);
1813 if (class && class->type_get_memory_usage) {
1814 class->type_get_memory_usage(datapath_type, usage);
1819 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1822 connmgr_get_controller_info(ofproto->connmgr, info);
1826 ofproto_free_ofproto_controller_info(struct shash *info)
1828 connmgr_free_controller_info(info);
1831 /* Makes a deep copy of 'old' into 'port'. */
1833 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1835 port->name = xstrdup(old->name);
1836 port->type = xstrdup(old->type);
1837 port->ofp_port = old->ofp_port;
1840 /* Frees memory allocated to members of 'ofproto_port'.
1842 * Do not call this function on an ofproto_port obtained from
1843 * ofproto_port_dump_next(): that function retains ownership of the data in the
1846 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1848 free(ofproto_port->name);
1849 free(ofproto_port->type);
1852 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1854 * This function provides no status indication. An error status for the entire
1855 * dump operation is provided when it is completed by calling
1856 * ofproto_port_dump_done().
1859 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1860 const struct ofproto *ofproto)
1862 dump->ofproto = ofproto;
1863 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1867 /* Attempts to retrieve another port from 'dump', which must have been created
1868 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1869 * 'port' and returns true. On failure, returns false.
1871 * Failure might indicate an actual error or merely that the last port has been
1872 * dumped. An error status for the entire dump operation is provided when it
1873 * is completed by calling ofproto_port_dump_done().
1875 * The ofproto owns the data stored in 'port'. It will remain valid until at
1876 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1877 * ofproto_port_dump_done(). */
1879 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1880 struct ofproto_port *port)
1882 const struct ofproto *ofproto = dump->ofproto;
1888 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1891 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1897 /* Completes port table dump operation 'dump', which must have been created
1898 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1899 * error-free, otherwise a positive errno value describing the problem. */
1901 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1903 const struct ofproto *ofproto = dump->ofproto;
1905 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1908 return dump->error == EOF ? 0 : dump->error;
1911 /* Returns the type to pass to netdev_open() when a datapath of type
1912 * 'datapath_type' has a port of type 'port_type', for a few special
1913 * cases when a netdev type differs from a port type. For example, when
1914 * using the userspace datapath, a port of type "internal" needs to be
1917 * Returns either 'type' itself or a string literal, which must not be
1920 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1922 const struct ofproto_class *class;
1924 datapath_type = ofproto_normalize_type(datapath_type);
1925 class = ofproto_class_find__(datapath_type);
1930 return (class->port_open_type
1931 ? class->port_open_type(datapath_type, port_type)
1935 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1936 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1937 * the port's OpenFlow port number.
1939 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1940 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1941 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1942 * 'ofp_portp' is non-null). */
1944 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1945 ofp_port_t *ofp_portp)
1947 ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1950 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1952 const char *netdev_name = netdev_get_name(netdev);
1954 simap_put(&ofproto->ofp_requests, netdev_name,
1955 ofp_to_u16(ofp_port));
1956 error = update_port(ofproto, netdev_name);
1959 *ofp_portp = OFPP_NONE;
1961 struct ofproto_port ofproto_port;
1963 error = ofproto_port_query_by_name(ofproto,
1964 netdev_get_name(netdev),
1967 *ofp_portp = ofproto_port.ofp_port;
1968 ofproto_port_destroy(&ofproto_port);
1975 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1976 * initializes '*port' appropriately; on failure, returns a positive errno
1979 * The caller owns the data in 'ofproto_port' and must free it with
1980 * ofproto_port_destroy() when it is no longer needed. */
1982 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1983 struct ofproto_port *port)
1987 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1989 memset(port, 0, sizeof *port);
1994 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1995 * Returns 0 if successful, otherwise a positive errno. */
1997 ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port)
1999 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2000 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
2001 struct simap_node *ofp_request_node;
2004 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
2005 if (ofp_request_node) {
2006 simap_delete(&ofproto->ofp_requests, ofp_request_node);
2009 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
2010 if (!error && ofport) {
2011 /* 'name' is the netdev's name and update_port() is going to close the
2012 * netdev. Just in case update_port() refers to 'name' after it
2013 * destroys 'ofport', make a copy of it around the update_port()
2015 char *devname = xstrdup(name);
2016 update_port(ofproto, devname);
2023 flow_mod_init(struct ofputil_flow_mod *fm,
2024 const struct match *match, int priority,
2025 const struct ofpact *ofpacts, size_t ofpacts_len,
2026 enum ofp_flow_mod_command command)
2028 *fm = (struct ofputil_flow_mod) {
2030 .priority = priority,
2033 .buffer_id = UINT32_MAX,
2034 .out_port = OFPP_ANY,
2035 .out_group = OFPG_ANY,
2036 .ofpacts = CONST_CAST(struct ofpact *, ofpacts),
2037 .ofpacts_len = ofpacts_len,
2038 .delete_reason = OFPRR_DELETE,
2043 simple_flow_mod(struct ofproto *ofproto,
2044 const struct match *match, int priority,
2045 const struct ofpact *ofpacts, size_t ofpacts_len,
2046 enum ofp_flow_mod_command command)
2048 struct ofproto_flow_mod ofm;
2050 flow_mod_init(&ofm.fm, match, priority, ofpacts, ofpacts_len, command);
2052 return handle_flow_mod__(ofproto, &ofm, NULL);
2055 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
2056 * performs the 'n_actions' actions in 'actions'. The new flow will not
2059 * If cls_rule->priority is in the range of priorities supported by OpenFlow
2060 * (0...65535, inclusive) then the flow will be visible to OpenFlow
2061 * controllers; otherwise, it will be hidden.
2063 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
2065 * This is a helper function for in-band control and fail-open. */
2067 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
2069 const struct ofpact *ofpacts, size_t ofpacts_len)
2070 OVS_EXCLUDED(ofproto_mutex)
2072 const struct rule *rule;
2075 /* First do a cheap check whether the rule we're looking for already exists
2076 * with the actions that we want. If it does, then we're done. */
2077 rule = rule_from_cls_rule(classifier_find_match_exactly(
2078 &ofproto->tables[0].cls, match, priority,
2081 const struct rule_actions *actions = rule_get_actions(rule);
2082 must_add = !ofpacts_equal(actions->ofpacts, actions->ofpacts_len,
2083 ofpacts, ofpacts_len);
2088 /* If there's no such rule or the rule doesn't have the actions we want,
2089 * fall back to a executing a full flow mod. We can't optimize this at
2090 * all because we didn't take enough locks above to ensure that the flow
2091 * table didn't already change beneath us. */
2093 simple_flow_mod(ofproto, match, priority, ofpacts, ofpacts_len,
2094 OFPFC_MODIFY_STRICT);
2098 /* Executes the flow modification specified in 'fm'. Returns 0 on success, or
2099 * an OFPERR_* OpenFlow error code on failure.
2101 * This is a helper function for in-band control and fail-open and the "learn"
2104 ofproto_flow_mod(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
2105 OVS_EXCLUDED(ofproto_mutex)
2107 struct ofputil_flow_mod *fm = &ofm->fm;
2109 /* Optimize for the most common case of a repeated learn action.
2110 * If an identical flow already exists we only need to update its
2111 * 'modified' time. */
2112 if (fm->command == OFPFC_MODIFY_STRICT && fm->table_id != OFPTT_ALL
2113 && !(fm->flags & OFPUTIL_FF_RESET_COUNTS)) {
2114 struct oftable *table = &ofproto->tables[fm->table_id];
2118 rule = rule_from_cls_rule(classifier_find_match_exactly(
2119 &table->cls, &fm->match, fm->priority,
2122 /* Reading many of the rule fields and writing on 'modified'
2123 * requires the rule->mutex. Also, rule->actions may change
2124 * if rule->mutex is not held. */
2125 const struct rule_actions *actions;
2127 ovs_mutex_lock(&rule->mutex);
2128 actions = rule_get_actions(rule);
2129 if (rule->idle_timeout == fm->idle_timeout
2130 && rule->hard_timeout == fm->hard_timeout
2131 && rule->importance == fm->importance
2132 && rule->flags == (fm->flags & OFPUTIL_FF_STATE)
2133 && (!fm->modify_cookie || (fm->new_cookie == rule->flow_cookie))
2134 && ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
2135 actions->ofpacts, actions->ofpacts_len)) {
2136 /* Rule already exists and need not change, only update the
2137 modified timestamp. */
2138 rule->modified = time_msec();
2141 ovs_mutex_unlock(&rule->mutex);
2149 return handle_flow_mod__(ofproto, ofm, NULL);
2152 /* Searches for a rule with matching criteria exactly equal to 'target' in
2153 * ofproto's table 0 and, if it finds one, deletes it.
2155 * This is a helper function for in-band control and fail-open. */
2157 ofproto_delete_flow(struct ofproto *ofproto,
2158 const struct match *target, int priority)
2159 OVS_EXCLUDED(ofproto_mutex)
2161 struct classifier *cls = &ofproto->tables[0].cls;
2164 /* First do a cheap check whether the rule we're looking for has already
2165 * been deleted. If so, then we're done. */
2166 rule = rule_from_cls_rule(classifier_find_match_exactly(
2167 cls, target, priority, CLS_MAX_VERSION));
2172 /* Execute a flow mod. We can't optimize this at all because we didn't
2173 * take enough locks above to ensure that the flow table didn't already
2174 * change beneath us. */
2175 simple_flow_mod(ofproto, target, priority, NULL, 0, OFPFC_DELETE_STRICT);
2178 /* Delete all of the flows from all of ofproto's flow tables, then reintroduce
2179 * the flows required by in-band control and fail-open. */
2181 ofproto_flush_flows(struct ofproto *ofproto)
2183 COVERAGE_INC(ofproto_flush);
2184 ofproto_flush__(ofproto);
2185 connmgr_flushed(ofproto->connmgr);
2189 reinit_ports(struct ofproto *p)
2191 struct ofproto_port_dump dump;
2192 struct sset devnames;
2193 struct ofport *ofport;
2194 struct ofproto_port ofproto_port;
2195 const char *devname;
2197 COVERAGE_INC(ofproto_reinit_ports);
2199 sset_init(&devnames);
2200 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2201 sset_add(&devnames, netdev_get_name(ofport->netdev));
2203 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2204 sset_add(&devnames, ofproto_port.name);
2207 SSET_FOR_EACH (devname, &devnames) {
2208 update_port(p, devname);
2210 sset_destroy(&devnames);
2214 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
2218 port_idx = simap_get(&ofproto->ofp_requests, netdev_name);
2219 port_idx = port_idx ? port_idx : UINT16_MAX;
2221 if (port_idx >= ofproto->max_ports
2222 || ofport_get_usage(ofproto, u16_to_ofp(port_idx)) == LLONG_MAX) {
2223 uint16_t lru_ofport = 0, end_port_no = ofproto->alloc_port_no;
2224 long long int last_used_at, lru = LLONG_MAX;
2226 /* Search for a free OpenFlow port number. We try not to
2227 * immediately reuse them to prevent problems due to old
2230 * We limit the automatically assigned port numbers to the lower half
2231 * of the port range, to reserve the upper half for assignment by
2234 if (++ofproto->alloc_port_no >= MIN(ofproto->max_ports, 32768)) {
2235 ofproto->alloc_port_no = 1;
2237 last_used_at = ofport_get_usage(ofproto,
2238 u16_to_ofp(ofproto->alloc_port_no));
2239 if (!last_used_at) {
2240 port_idx = ofproto->alloc_port_no;
2242 } else if ( last_used_at < time_msec() - 60*60*1000) {
2243 /* If the port with ofport 'ofproto->alloc_port_no' was deleted
2244 * more than an hour ago, consider it usable. */
2245 ofport_remove_usage(ofproto,
2246 u16_to_ofp(ofproto->alloc_port_no));
2247 port_idx = ofproto->alloc_port_no;
2249 } else if (last_used_at < lru) {
2251 lru_ofport = ofproto->alloc_port_no;
2254 if (ofproto->alloc_port_no == end_port_no) {
2256 port_idx = lru_ofport;
2263 ofport_set_usage(ofproto, u16_to_ofp(port_idx), LLONG_MAX);
2264 return u16_to_ofp(port_idx);
2268 dealloc_ofp_port(struct ofproto *ofproto, ofp_port_t ofp_port)
2270 if (ofp_to_u16(ofp_port) < ofproto->max_ports) {
2271 ofport_set_usage(ofproto, ofp_port, time_msec());
2275 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
2276 * pointer if the netdev cannot be opened. On success, also fills in
2278 static struct netdev *
2279 ofport_open(struct ofproto *ofproto,
2280 struct ofproto_port *ofproto_port,
2281 struct ofputil_phy_port *pp)
2283 enum netdev_flags flags;
2284 struct netdev *netdev;
2287 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
2289 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
2290 "cannot be opened (%s)",
2292 ofproto_port->name, ofproto_port->ofp_port,
2293 ofproto_port->name, ovs_strerror(error));
2297 if (ofproto_port->ofp_port == OFPP_NONE) {
2298 if (!strcmp(ofproto->name, ofproto_port->name)) {
2299 ofproto_port->ofp_port = OFPP_LOCAL;
2301 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
2302 ofproto_port->name);
2305 pp->port_no = ofproto_port->ofp_port;
2306 netdev_get_etheraddr(netdev, &pp->hw_addr);
2307 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
2308 netdev_get_flags(netdev, &flags);
2309 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
2310 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
2311 netdev_get_features(netdev, &pp->curr, &pp->advertised,
2312 &pp->supported, &pp->peer);
2313 pp->curr_speed = netdev_features_to_bps(pp->curr, 0) / 1000;
2314 pp->max_speed = netdev_features_to_bps(pp->supported, 0) / 1000;
2319 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
2320 * port number, and 'config' bits other than OFPUTIL_PC_PORT_DOWN are
2323 ofport_equal(const struct ofputil_phy_port *a,
2324 const struct ofputil_phy_port *b)
2326 return (eth_addr_equals(a->hw_addr, b->hw_addr)
2327 && a->state == b->state
2328 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
2329 && a->curr == b->curr
2330 && a->advertised == b->advertised
2331 && a->supported == b->supported
2332 && a->peer == b->peer
2333 && a->curr_speed == b->curr_speed
2334 && a->max_speed == b->max_speed);
2337 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
2338 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
2339 * one with the same name or port number). */
2341 ofport_install(struct ofproto *p,
2342 struct netdev *netdev, const struct ofputil_phy_port *pp)
2344 const char *netdev_name = netdev_get_name(netdev);
2345 struct ofport *ofport;
2348 /* Create ofport. */
2349 ofport = p->ofproto_class->port_alloc();
2354 ofport->ofproto = p;
2355 ofport->netdev = netdev;
2356 ofport->change_seq = netdev_get_change_seq(netdev);
2358 ofport->ofp_port = pp->port_no;
2359 ofport->created = time_msec();
2361 /* Add port to 'p'. */
2362 hmap_insert(&p->ports, &ofport->hmap_node,
2363 hash_ofp_port(ofport->ofp_port));
2364 shash_add(&p->port_by_name, netdev_name, ofport);
2366 update_mtu(p, ofport);
2368 /* Let the ofproto_class initialize its private data. */
2369 error = p->ofproto_class->port_construct(ofport);
2373 connmgr_send_port_status(p->connmgr, NULL, pp, OFPPR_ADD);
2377 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
2378 p->name, netdev_name, ovs_strerror(error));
2380 ofport_destroy__(ofport);
2382 netdev_close(netdev);
2387 /* Removes 'ofport' from 'p' and destroys it. */
2389 ofport_remove(struct ofport *ofport)
2391 struct ofproto *p = ofport->ofproto;
2392 bool is_internal = ofport_is_internal(ofport);
2394 connmgr_send_port_status(ofport->ofproto->connmgr, NULL, &ofport->pp,
2396 ofport_destroy(ofport, true);
2398 update_mtu_ofproto(p);
2402 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
2405 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
2407 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
2409 ofport_remove(port);
2413 /* Updates 'port' with new 'pp' description.
2415 * Does not handle a name or port number change. The caller must implement
2416 * such a change as a delete followed by an add. */
2418 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
2420 port->pp.hw_addr = pp->hw_addr;
2421 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
2422 | (pp->config & OFPUTIL_PC_PORT_DOWN));
2423 port->pp.state = ((port->pp.state & ~OFPUTIL_PS_LINK_DOWN)
2424 | (pp->state & OFPUTIL_PS_LINK_DOWN));
2425 port->pp.curr = pp->curr;
2426 port->pp.advertised = pp->advertised;
2427 port->pp.supported = pp->supported;
2428 port->pp.peer = pp->peer;
2429 port->pp.curr_speed = pp->curr_speed;
2430 port->pp.max_speed = pp->max_speed;
2432 connmgr_send_port_status(port->ofproto->connmgr, NULL,
2433 &port->pp, OFPPR_MODIFY);
2436 /* Update OpenFlow 'state' in 'port' and notify controller. */
2438 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
2440 if (port->pp.state != state) {
2441 port->pp.state = state;
2442 connmgr_send_port_status(port->ofproto->connmgr, NULL,
2443 &port->pp, OFPPR_MODIFY);
2448 ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port)
2450 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
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);
2678 ofport_is_internal(const struct ofport *port)
2680 return !strcmp(netdev_get_type(port->netdev), "internal");
2683 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2684 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2686 find_min_mtu(struct ofproto *p)
2688 struct ofport *ofport;
2691 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2692 struct netdev *netdev = ofport->netdev;
2695 /* Skip any internal ports, since that's what we're trying to
2697 if (ofport_is_internal(ofport)) {
2701 if (netdev_get_mtu(netdev, &dev_mtu)) {
2704 if (!mtu || dev_mtu < mtu) {
2709 return mtu ? mtu: ETH_PAYLOAD_MAX;
2712 /* Update MTU of all datapath devices on 'p' to the minimum of the
2713 * non-datapath ports in event of 'port' added or changed. */
2715 update_mtu(struct ofproto *p, struct ofport *port)
2717 struct netdev *netdev = port->netdev;
2720 if (netdev_get_mtu(netdev, &dev_mtu)) {
2724 if (ofport_is_internal(port)) {
2725 if (dev_mtu > p->min_mtu) {
2726 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2727 dev_mtu = p->min_mtu;
2730 port->mtu = dev_mtu;
2734 port->mtu = dev_mtu;
2735 /* For non-internal port find new min mtu. */
2736 update_mtu_ofproto(p);
2740 update_mtu_ofproto(struct ofproto *p)
2742 /* For non-internal port find new min mtu. */
2743 struct ofport *ofport;
2744 int old_min = p->min_mtu;
2746 p->min_mtu = find_min_mtu(p);
2747 if (p->min_mtu == old_min) {
2751 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2752 struct netdev *netdev = ofport->netdev;
2754 if (ofport_is_internal(ofport)) {
2755 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2756 ofport->mtu = p->min_mtu;
2763 ofproto_rule_destroy__(struct rule *rule)
2764 OVS_NO_THREAD_SAFETY_ANALYSIS
2766 cls_rule_destroy(CONST_CAST(struct cls_rule *, &rule->cr));
2767 rule_actions_destroy(rule_get_actions(rule));
2768 ovs_mutex_destroy(&rule->mutex);
2769 rule->ofproto->ofproto_class->rule_dealloc(rule);
2773 rule_destroy_cb(struct rule *rule)
2774 OVS_NO_THREAD_SAFETY_ANALYSIS
2776 /* Send rule removed if needed. */
2777 if (rule->flags & OFPUTIL_FF_SEND_FLOW_REM
2778 && rule->removed_reason != OVS_OFPRR_NONE
2779 && !rule_is_hidden(rule)) {
2780 ofproto_rule_send_removed(rule);
2782 rule->ofproto->ofproto_class->rule_destruct(rule);
2783 ofproto_rule_destroy__(rule);
2787 ofproto_rule_ref(struct rule *rule)
2790 ovs_refcount_ref(&rule->ref_count);
2795 ofproto_rule_try_ref(struct rule *rule)
2798 return ovs_refcount_try_ref_rcu(&rule->ref_count);
2803 /* Decrements 'rule''s ref_count and schedules 'rule' to be destroyed if the
2804 * ref_count reaches 0.
2806 * Use of RCU allows short term use (between RCU quiescent periods) without
2807 * keeping a reference. A reference must be taken if the rule needs to
2808 * stay around accross the RCU quiescent periods. */
2810 ofproto_rule_unref(struct rule *rule)
2812 if (rule && ovs_refcount_unref_relaxed(&rule->ref_count) == 1) {
2813 ovsrcu_postpone(rule_destroy_cb, rule);
2818 remove_rule_rcu__(struct rule *rule)
2819 OVS_REQUIRES(ofproto_mutex)
2821 struct ofproto *ofproto = rule->ofproto;
2822 struct oftable *table = &ofproto->tables[rule->table_id];
2824 ovs_assert(!cls_rule_visible_in_version(&rule->cr, CLS_MAX_VERSION));
2825 if (!classifier_remove(&table->cls, &rule->cr)) {
2828 ofproto->ofproto_class->rule_delete(rule);
2829 ofproto_rule_unref(rule);
2833 remove_rule_rcu(struct rule *rule)
2834 OVS_EXCLUDED(ofproto_mutex)
2836 ovs_mutex_lock(&ofproto_mutex);
2837 remove_rule_rcu__(rule);
2838 ovs_mutex_unlock(&ofproto_mutex);
2841 /* Removes and deletes rules from a NULL-terminated array of rule pointers. */
2843 remove_rules_rcu(struct rule **rules)
2844 OVS_EXCLUDED(ofproto_mutex)
2846 struct rule **orig_rules = rules;
2849 struct ofproto *ofproto = rules[0]->ofproto;
2850 unsigned long tables[BITMAP_N_LONGS(256)];
2854 memset(tables, 0, sizeof tables);
2856 ovs_mutex_lock(&ofproto_mutex);
2857 while ((rule = *rules++)) {
2858 /* Defer once for each new table. This defers the subtable cleanup
2859 * until later, so that when removing large number of flows the
2860 * operation is faster. */
2861 if (!bitmap_is_set(tables, rule->table_id)) {
2862 struct classifier *cls = &ofproto->tables[rule->table_id].cls;
2864 bitmap_set1(tables, rule->table_id);
2865 classifier_defer(cls);
2867 remove_rule_rcu__(rule);
2870 BITMAP_FOR_EACH_1(table_id, 256, tables) {
2871 struct classifier *cls = &ofproto->tables[table_id].cls;
2873 classifier_publish(cls);
2875 ovs_mutex_unlock(&ofproto_mutex);
2882 ofproto_group_ref(struct ofgroup *group)
2885 ovs_refcount_ref(&group->ref_count);
2890 ofproto_group_unref(struct ofgroup *group)
2892 if (group && ovs_refcount_unref(&group->ref_count) == 1) {
2893 group->ofproto->ofproto_class->group_destruct(group);
2894 ofputil_bucket_list_destroy(&group->buckets);
2895 group->ofproto->ofproto_class->group_dealloc(group);
2899 static uint32_t get_provider_meter_id(const struct ofproto *,
2900 uint32_t of_meter_id);
2902 /* Creates and returns a new 'struct rule_actions', whose actions are a copy
2903 * of from the 'ofpacts_len' bytes of 'ofpacts'. */
2904 const struct rule_actions *
2905 rule_actions_create(const struct ofpact *ofpacts, size_t ofpacts_len)
2907 struct rule_actions *actions;
2909 actions = xmalloc(sizeof *actions + ofpacts_len);
2910 actions->ofpacts_len = ofpacts_len;
2911 actions->has_meter = ofpacts_get_meter(ofpacts, ofpacts_len) != 0;
2912 memcpy(actions->ofpacts, ofpacts, ofpacts_len);
2914 actions->has_learn_with_delete = (next_learn_with_delete(actions, NULL)
2920 /* Free the actions after the RCU quiescent period is reached. */
2922 rule_actions_destroy(const struct rule_actions *actions)
2925 ovsrcu_postpone(free, CONST_CAST(struct rule_actions *, actions));
2929 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2930 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2932 ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port)
2933 OVS_REQUIRES(ofproto_mutex)
2935 if (port == OFPP_ANY) {
2938 const struct rule_actions *actions = rule_get_actions(rule);
2939 return ofpacts_output_to_port(actions->ofpacts,
2940 actions->ofpacts_len, port);
2944 /* Returns true if 'rule' has group and equals group_id. */
2946 ofproto_rule_has_out_group(const struct rule *rule, uint32_t group_id)
2947 OVS_REQUIRES(ofproto_mutex)
2949 if (group_id == OFPG_ANY) {
2952 const struct rule_actions *actions = rule_get_actions(rule);
2953 return ofpacts_output_to_group(actions->ofpacts,
2954 actions->ofpacts_len, group_id);
2959 rule_execute_destroy(struct rule_execute *e)
2961 ofproto_rule_unref(e->rule);
2962 ovs_list_remove(&e->list_node);
2966 /* Executes all "rule_execute" operations queued up in ofproto->rule_executes,
2967 * by passing them to the ofproto provider. */
2969 run_rule_executes(struct ofproto *ofproto)
2970 OVS_EXCLUDED(ofproto_mutex)
2972 struct rule_execute *e, *next;
2973 struct ovs_list executes;
2975 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2976 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2979 flow_extract(e->packet, &flow);
2980 flow.in_port.ofp_port = e->in_port;
2981 ofproto->ofproto_class->rule_execute(e->rule, &flow, e->packet);
2983 rule_execute_destroy(e);
2987 /* Destroys and discards all "rule_execute" operations queued up in
2988 * ofproto->rule_executes. */
2990 destroy_rule_executes(struct ofproto *ofproto)
2992 struct rule_execute *e, *next;
2993 struct ovs_list executes;
2995 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2996 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2997 dp_packet_delete(e->packet);
2998 rule_execute_destroy(e);
3003 rule_is_readonly(const struct rule *rule)
3005 const struct oftable *table = &rule->ofproto->tables[rule->table_id];
3006 return (table->flags & OFTABLE_READONLY) != 0;
3010 hash_learned_cookie(ovs_be64 cookie_, uint8_t table_id)
3012 uint64_t cookie = (OVS_FORCE uint64_t) cookie_;
3013 return hash_3words(cookie, cookie >> 32, table_id);
3017 learned_cookies_update_one__(struct ofproto *ofproto,
3018 const struct ofpact_learn *learn,
3019 int delta, struct ovs_list *dead_cookies)
3020 OVS_REQUIRES(ofproto_mutex)
3022 uint32_t hash = hash_learned_cookie(learn->cookie, learn->table_id);
3023 struct learned_cookie *c;
3025 HMAP_FOR_EACH_WITH_HASH (c, u.hmap_node, hash, &ofproto->learned_cookies) {
3026 if (c->cookie == learn->cookie && c->table_id == learn->table_id) {
3028 ovs_assert(c->n >= 0);
3031 hmap_remove(&ofproto->learned_cookies, &c->u.hmap_node);
3032 ovs_list_push_back(dead_cookies, &c->u.list_node);
3039 ovs_assert(delta > 0);
3040 c = xmalloc(sizeof *c);
3041 hmap_insert(&ofproto->learned_cookies, &c->u.hmap_node, hash);
3042 c->cookie = learn->cookie;
3043 c->table_id = learn->table_id;
3047 static const struct ofpact_learn *
3048 next_learn_with_delete(const struct rule_actions *actions,
3049 const struct ofpact_learn *start)
3051 const struct ofpact *pos;
3053 for (pos = start ? ofpact_next(&start->ofpact) : actions->ofpacts;
3054 pos < ofpact_end(actions->ofpacts, actions->ofpacts_len);
3055 pos = ofpact_next(pos)) {
3056 if (pos->type == OFPACT_LEARN) {
3057 const struct ofpact_learn *learn = ofpact_get_LEARN(pos);
3058 if (learn->flags & NX_LEARN_F_DELETE_LEARNED) {
3068 learned_cookies_update__(struct ofproto *ofproto,
3069 const struct rule_actions *actions,
3070 int delta, struct ovs_list *dead_cookies)
3071 OVS_REQUIRES(ofproto_mutex)
3073 if (actions->has_learn_with_delete) {
3074 const struct ofpact_learn *learn;
3076 for (learn = next_learn_with_delete(actions, NULL); learn;
3077 learn = next_learn_with_delete(actions, learn)) {
3078 learned_cookies_update_one__(ofproto, learn, delta, dead_cookies);
3084 learned_cookies_inc(struct ofproto *ofproto,
3085 const struct rule_actions *actions)
3086 OVS_REQUIRES(ofproto_mutex)
3088 learned_cookies_update__(ofproto, actions, +1, NULL);
3092 learned_cookies_dec(struct ofproto *ofproto,
3093 const struct rule_actions *actions,
3094 struct ovs_list *dead_cookies)
3095 OVS_REQUIRES(ofproto_mutex)
3097 learned_cookies_update__(ofproto, actions, -1, dead_cookies);
3101 learned_cookies_flush(struct ofproto *ofproto, struct ovs_list *dead_cookies)
3102 OVS_REQUIRES(ofproto_mutex)
3104 struct learned_cookie *c;
3106 LIST_FOR_EACH_POP (c, u.list_node, dead_cookies) {
3107 struct rule_criteria criteria;
3108 struct rule_collection rules;
3111 match_init_catchall(&match);
3112 rule_criteria_init(&criteria, c->table_id, &match, 0, CLS_MAX_VERSION,
3113 c->cookie, OVS_BE64_MAX, OFPP_ANY, OFPG_ANY);
3114 rule_criteria_require_rw(&criteria, false);
3115 collect_rules_loose(ofproto, &criteria, &rules);
3116 rule_criteria_destroy(&criteria);
3117 delete_flows__(&rules, OFPRR_DELETE, NULL);
3124 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
3126 ofconn_send_reply(ofconn, make_echo_reply(oh));
3131 query_tables(struct ofproto *ofproto,
3132 struct ofputil_table_features **featuresp,
3133 struct ofputil_table_stats **statsp)
3135 struct mf_bitmap rw_fields = oxm_writable_fields();
3136 struct mf_bitmap match = oxm_matchable_fields();
3137 struct mf_bitmap mask = oxm_maskable_fields();
3139 struct ofputil_table_features *features;
3140 struct ofputil_table_stats *stats;
3143 features = *featuresp = xcalloc(ofproto->n_tables, sizeof *features);
3144 for (i = 0; i < ofproto->n_tables; i++) {
3145 struct ofputil_table_features *f = &features[i];
3148 sprintf(f->name, "table%d", i);
3149 f->metadata_match = OVS_BE64_MAX;
3150 f->metadata_write = OVS_BE64_MAX;
3151 atomic_read_relaxed(&ofproto->tables[i].miss_config, &f->miss_config);
3152 f->max_entries = 1000000;
3154 bool more_tables = false;
3155 for (int j = i + 1; j < ofproto->n_tables; j++) {
3156 if (!(ofproto->tables[j].flags & OFTABLE_HIDDEN)) {
3157 bitmap_set1(f->nonmiss.next, j);
3161 f->nonmiss.instructions = (1u << N_OVS_INSTRUCTIONS) - 1;
3163 f->nonmiss.instructions &= ~(1u << OVSINST_OFPIT11_GOTO_TABLE);
3165 f->nonmiss.write.ofpacts = (UINT64_C(1) << N_OFPACTS) - 1;
3166 f->nonmiss.write.set_fields = rw_fields;
3167 f->nonmiss.apply = f->nonmiss.write;
3168 f->miss = f->nonmiss;
3172 f->wildcard = match;
3176 stats = *statsp = xcalloc(ofproto->n_tables, sizeof *stats);
3177 for (i = 0; i < ofproto->n_tables; i++) {
3178 struct ofputil_table_stats *s = &stats[i];
3181 s->active_count = ofproto->tables[i].n_flows;
3183 s->active_count -= connmgr_count_hidden_rules(
3191 ofproto->ofproto_class->query_tables(ofproto, features, stats);
3193 for (i = 0; i < ofproto->n_tables; i++) {
3194 const struct oftable *table = &ofproto->tables[i];
3195 struct ofputil_table_features *f = &features[i];
3198 ovs_strzcpy(f->name, table->name, sizeof f->name);
3201 if (table->max_flows < f->max_entries) {
3202 f->max_entries = table->max_flows;
3208 query_switch_features(struct ofproto *ofproto,
3209 bool *arp_match_ip, uint64_t *ofpacts)
3211 struct ofputil_table_features *features, *f;
3213 *arp_match_ip = false;
3216 query_tables(ofproto, &features, NULL);
3217 for (f = features; f < &features[ofproto->n_tables]; f++) {
3218 *ofpacts |= f->nonmiss.apply.ofpacts | f->miss.apply.ofpacts;
3219 if (bitmap_is_set(f->match.bm, MFF_ARP_SPA) ||
3220 bitmap_is_set(f->match.bm, MFF_ARP_TPA)) {
3221 *arp_match_ip = true;
3227 ovs_assert(*ofpacts & (UINT64_C(1) << OFPACT_OUTPUT));
3231 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
3233 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3234 struct ofputil_switch_features features;
3235 struct ofport *port;
3239 query_switch_features(ofproto, &arp_match_ip, &features.ofpacts);
3241 features.datapath_id = ofproto->datapath_id;
3242 features.n_buffers = pktbuf_capacity();
3243 features.n_tables = ofproto_get_n_visible_tables(ofproto);
3244 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
3245 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS |
3246 OFPUTIL_C_GROUP_STATS);
3248 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
3250 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
3251 features.auxiliary_id = 0;
3252 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
3254 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3255 ofputil_put_switch_features_port(&port->pp, b);
3258 ofconn_send_reply(ofconn, b);
3263 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
3265 struct ofputil_switch_config config;
3266 config.frag = ofconn_get_ofproto(ofconn)->frag_handling;
3267 config.invalid_ttl_to_controller
3268 = ofconn_get_invalid_ttl_to_controller(ofconn);
3269 config.miss_send_len = ofconn_get_miss_send_len(ofconn);
3271 ofconn_send_reply(ofconn, ofputil_encode_get_config_reply(oh, &config));
3277 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
3279 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3280 struct ofputil_switch_config config;
3283 error = ofputil_decode_set_config(oh, &config);
3288 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
3289 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
3290 enum ofputil_frag_handling cur = ofproto->frag_handling;
3291 enum ofputil_frag_handling next = config.frag;
3294 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
3295 ofproto->frag_handling = next;
3297 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
3299 ofputil_frag_handling_to_string(next));
3304 if (config.invalid_ttl_to_controller >= 0) {
3305 ofconn_set_invalid_ttl_to_controller(ofconn,
3306 config.invalid_ttl_to_controller);
3309 ofconn_set_miss_send_len(ofconn, config.miss_send_len);
3314 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3315 * error message code for the caller to propagate upward. Otherwise, returns
3318 * The log message mentions 'msg_type'. */
3320 reject_slave_controller(struct ofconn *ofconn)
3322 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
3323 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
3324 return OFPERR_OFPBRC_IS_SLAVE;
3330 /* Checks that the 'ofpacts_len' bytes of action in 'ofpacts' are appropriate
3333 * - If they use a meter, then 'ofproto' has that meter configured.
3335 * - If they use any groups, then 'ofproto' has that group configured.
3337 * Returns 0 if successful, otherwise an OpenFlow error. */
3339 ofproto_check_ofpacts(struct ofproto *ofproto,
3340 const struct ofpact ofpacts[], size_t ofpacts_len)
3342 const struct ofpact *a;
3345 mid = ofpacts_get_meter(ofpacts, ofpacts_len);
3346 if (mid && get_provider_meter_id(ofproto, mid) == UINT32_MAX) {
3347 return OFPERR_OFPMMFC_INVALID_METER;
3350 OFPACT_FOR_EACH_FLATTENED (a, ofpacts, ofpacts_len) {
3351 if (a->type == OFPACT_GROUP
3352 && !ofproto_group_exists(ofproto, ofpact_get_GROUP(a)->group_id)) {
3353 return OFPERR_OFPBAC_BAD_OUT_GROUP;
3361 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
3363 struct ofproto *p = ofconn_get_ofproto(ofconn);
3364 struct ofputil_packet_out po;
3365 struct dp_packet *payload;
3366 uint64_t ofpacts_stub[1024 / 8];
3367 struct ofpbuf ofpacts;
3371 COVERAGE_INC(ofproto_packet_out);
3373 error = reject_slave_controller(ofconn);
3378 /* Decode message. */
3379 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3380 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
3382 goto exit_free_ofpacts;
3384 if (ofp_to_u16(po.in_port) >= p->max_ports
3385 && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) {
3386 error = OFPERR_OFPBRC_BAD_PORT;
3387 goto exit_free_ofpacts;
3391 if (po.buffer_id != UINT32_MAX) {
3392 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
3394 goto exit_free_ofpacts;
3397 /* Ensure that the L3 header is 32-bit aligned. */
3398 payload = dp_packet_clone_data_with_headroom(po.packet, po.packet_len, 2);
3401 /* Verify actions against packet, then send packet if successful. */
3402 flow_extract(payload, &flow);
3403 flow.in_port.ofp_port = po.in_port;
3405 /* Check actions like for flow mods. We pass a 'table_id' of 0 to
3406 * ofproto_check_consistency(), which isn't strictly correct because these
3407 * actions aren't in any table. This is OK as 'table_id' is only used to
3408 * check instructions (e.g., goto-table), which can't appear on the action
3409 * list of a packet-out. */
3410 error = ofpacts_check_consistency(po.ofpacts, po.ofpacts_len,
3411 &flow, u16_to_ofp(p->max_ports),
3413 ofconn_get_protocol(ofconn));
3415 error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len);
3417 error = p->ofproto_class->packet_out(p, payload, &flow,
3418 po.ofpacts, po.ofpacts_len);
3421 dp_packet_delete(payload);
3424 ofpbuf_uninit(&ofpacts);
3430 handle_nxt_resume(struct ofconn *ofconn, const struct ofp_header *oh)
3432 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3433 struct ofputil_packet_in_private pin;
3436 error = ofputil_decode_packet_in_private(oh, false, &pin, NULL, NULL);
3441 error = (ofproto->ofproto_class->nxt_resume
3442 ? ofproto->ofproto_class->nxt_resume(ofproto, &pin)
3443 : OFPERR_NXR_NOT_SUPPORTED);
3445 ofputil_packet_in_private_destroy(&pin);
3451 update_port_config(struct ofconn *ofconn, struct ofport *port,
3452 enum ofputil_port_config config,
3453 enum ofputil_port_config mask)
3455 enum ofputil_port_config toggle = (config ^ port->pp.config) & mask;
3457 if (toggle & OFPUTIL_PC_PORT_DOWN
3458 && (config & OFPUTIL_PC_PORT_DOWN
3459 ? netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL)
3460 : netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL))) {
3461 /* We tried to bring the port up or down, but it failed, so don't
3462 * update the "down" bit. */
3463 toggle &= ~OFPUTIL_PC_PORT_DOWN;
3467 enum ofputil_port_config old_config = port->pp.config;
3468 port->pp.config ^= toggle;
3469 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
3470 connmgr_send_port_status(port->ofproto->connmgr, ofconn, &port->pp,
3476 port_mod_start(struct ofconn *ofconn, struct ofputil_port_mod *pm,
3477 struct ofport **port)
3479 struct ofproto *p = ofconn_get_ofproto(ofconn);
3481 *port = ofproto_get_port(p, pm->port_no);
3483 return OFPERR_OFPPMFC_BAD_PORT;
3485 if (!eth_addr_equals((*port)->pp.hw_addr, pm->hw_addr)) {
3486 return OFPERR_OFPPMFC_BAD_HW_ADDR;
3492 port_mod_finish(struct ofconn *ofconn, struct ofputil_port_mod *pm,
3493 struct ofport *port)
3495 update_port_config(ofconn, port, pm->config, pm->mask);
3496 if (pm->advertise) {
3497 netdev_set_advertisements(port->netdev, pm->advertise);
3502 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3504 struct ofputil_port_mod pm;
3505 struct ofport *port;
3508 error = reject_slave_controller(ofconn);
3513 error = ofputil_decode_port_mod(oh, &pm, false);
3518 error = port_mod_start(ofconn, &pm, &port);
3520 port_mod_finish(ofconn, &pm, port);
3526 handle_desc_stats_request(struct ofconn *ofconn,
3527 const struct ofp_header *request)
3529 static const char *default_mfr_desc = "Nicira, Inc.";
3530 static const char *default_hw_desc = "Open vSwitch";
3531 static const char *default_sw_desc = VERSION;
3532 static const char *default_serial_desc = "None";
3533 static const char *default_dp_desc = "None";
3535 struct ofproto *p = ofconn_get_ofproto(ofconn);
3536 struct ofp_desc_stats *ods;
3539 msg = ofpraw_alloc_stats_reply(request, 0);
3540 ods = ofpbuf_put_zeros(msg, sizeof *ods);
3541 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
3542 sizeof ods->mfr_desc);
3543 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
3544 sizeof ods->hw_desc);
3545 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
3546 sizeof ods->sw_desc);
3547 ovs_strlcpy(ods->serial_num,
3548 p->serial_desc ? p->serial_desc : default_serial_desc,
3549 sizeof ods->serial_num);
3550 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
3551 sizeof ods->dp_desc);
3552 ofconn_send_reply(ofconn, msg);
3558 handle_table_stats_request(struct ofconn *ofconn,
3559 const struct ofp_header *request)
3561 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3562 struct ofputil_table_features *features;
3563 struct ofputil_table_stats *stats;
3564 struct ofpbuf *reply;
3567 query_tables(ofproto, &features, &stats);
3569 reply = ofputil_encode_table_stats_reply(request);
3570 for (i = 0; i < ofproto->n_tables; i++) {
3571 if (!(ofproto->tables[i].flags & OFTABLE_HIDDEN)) {
3572 ofputil_append_table_stats_reply(reply, &stats[i], &features[i]);
3575 ofconn_send_reply(ofconn, reply);
3584 handle_table_features_request(struct ofconn *ofconn,
3585 const struct ofp_header *request)
3587 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3588 struct ofpbuf msg = ofpbuf_const_initializer(request,
3589 ntohs(request->length));
3590 ofpraw_pull_assert(&msg);
3591 if (msg.size || ofpmp_more(request)) {
3592 return OFPERR_OFPTFFC_EPERM;
3595 struct ofputil_table_features *features;
3596 query_tables(ofproto, &features, NULL);
3598 struct ovs_list replies;
3599 ofpmp_init(&replies, request);
3600 for (size_t i = 0; i < ofproto->n_tables; i++) {
3601 if (!(ofproto->tables[i].flags & OFTABLE_HIDDEN)) {
3602 ofputil_append_table_features_reply(&features[i], &replies);
3605 ofconn_send_replies(ofconn, &replies);
3612 /* Returns the vacancy of 'oftable', a number that ranges from 0 (if the table
3613 * is full) to 100 (if the table is empty).
3615 * A table without a limit on flows is considered to be empty. */
3617 oftable_vacancy(const struct oftable *t)
3619 return (!t->max_flows ? 100
3620 : t->n_flows >= t->max_flows ? 0
3621 : (t->max_flows - t->n_flows) * 100.0 / t->max_flows);
3625 query_table_desc__(struct ofputil_table_desc *td,
3626 struct ofproto *ofproto, uint8_t table_id)
3628 const struct oftable *t = &ofproto->tables[table_id];
3630 td->table_id = table_id;
3631 td->eviction = (t->eviction & EVICTION_OPENFLOW
3632 ? OFPUTIL_TABLE_EVICTION_ON
3633 : OFPUTIL_TABLE_EVICTION_OFF);
3634 td->eviction_flags = OFPROTO_EVICTION_FLAGS;
3635 td->vacancy = (t->vacancy_event
3636 ? OFPUTIL_TABLE_VACANCY_ON
3637 : OFPUTIL_TABLE_VACANCY_OFF);
3638 td->table_vacancy.vacancy_down = t->vacancy_down;
3639 td->table_vacancy.vacancy_up = t->vacancy_up;
3640 td->table_vacancy.vacancy = oftable_vacancy(t);
3643 /* This function queries the database for dumping table-desc. */
3645 query_tables_desc(struct ofproto *ofproto, struct ofputil_table_desc **descp)
3647 struct ofputil_table_desc *table_desc;
3650 table_desc = *descp = xcalloc(ofproto->n_tables, sizeof *table_desc);
3651 for (i = 0; i < ofproto->n_tables; i++) {
3652 struct ofputil_table_desc *td = &table_desc[i];
3653 query_table_desc__(td, ofproto, i);
3657 /* Function to handle dump-table-desc request. */
3659 handle_table_desc_request(struct ofconn *ofconn,
3660 const struct ofp_header *request)
3662 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3663 struct ofputil_table_desc *table_desc;
3664 struct ovs_list replies;
3667 query_tables_desc(ofproto, &table_desc);
3668 ofpmp_init(&replies, request);
3669 for (i = 0; i < ofproto->n_tables; i++) {
3670 if (!(ofproto->tables[i].flags & OFTABLE_HIDDEN)) {
3671 ofputil_append_table_desc_reply(&table_desc[i], &replies,
3675 ofconn_send_replies(ofconn, &replies);
3680 /* This function determines and sends the vacancy event, based on the value
3681 * of current vacancy and threshold vacancy. If the current vacancy is less
3682 * than or equal to vacancy_down, vacancy up events must be enabled, and when
3683 * the current vacancy is greater or equal to vacancy_up, vacancy down events
3684 * must be enabled. */
3686 send_table_status(struct ofproto *ofproto, uint8_t table_id)
3688 struct oftable *t = &ofproto->tables[table_id];
3689 if (!t->vacancy_event) {
3693 uint8_t vacancy = oftable_vacancy(t);
3694 enum ofp14_table_reason event;
3695 if (vacancy < t->vacancy_down) {
3696 event = OFPTR_VACANCY_DOWN;
3697 } else if (vacancy > t->vacancy_up) {
3698 event = OFPTR_VACANCY_UP;
3703 if (event == t->vacancy_event) {
3704 struct ofputil_table_desc td;
3705 query_table_desc__(&td, ofproto, table_id);
3706 connmgr_send_table_status(ofproto->connmgr, &td, event);
3708 t->vacancy_event = (event == OFPTR_VACANCY_DOWN
3710 : OFPTR_VACANCY_DOWN);
3715 append_port_stat(struct ofport *port, struct ovs_list *replies)
3717 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
3719 calc_duration(port->created, time_msec(),
3720 &ops.duration_sec, &ops.duration_nsec);
3722 /* Intentionally ignore return value, since errors will set
3723 * 'stats' to all-1s, which is correct for OpenFlow, and
3724 * netdev_get_stats() will log errors. */
3725 ofproto_port_get_stats(port, &ops.stats);
3727 ofputil_append_port_stat(replies, &ops);
3731 handle_port_request(struct ofconn *ofconn,
3732 const struct ofp_header *request, ofp_port_t port_no,
3733 void (*cb)(struct ofport *, struct ovs_list *replies))
3735 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3736 struct ofport *port;
3737 struct ovs_list replies;
3739 ofpmp_init(&replies, request);
3740 if (port_no != OFPP_ANY) {
3741 port = ofproto_get_port(ofproto, port_no);
3746 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3751 ofconn_send_replies(ofconn, &replies);
3755 handle_port_stats_request(struct ofconn *ofconn,
3756 const struct ofp_header *request)
3761 error = ofputil_decode_port_stats_request(request, &port_no);
3763 handle_port_request(ofconn, request, port_no, append_port_stat);
3769 append_port_desc(struct ofport *port, struct ovs_list *replies)
3771 ofputil_append_port_desc_stats_reply(&port->pp, replies);
3775 handle_port_desc_stats_request(struct ofconn *ofconn,
3776 const struct ofp_header *request)
3781 error = ofputil_decode_port_desc_stats_request(request, &port_no);
3783 handle_port_request(ofconn, request, port_no, append_port_desc);
3789 hash_cookie(ovs_be64 cookie)
3791 return hash_uint64((OVS_FORCE uint64_t)cookie);
3795 cookies_insert(struct ofproto *ofproto, struct rule *rule)
3796 OVS_REQUIRES(ofproto_mutex)
3798 hindex_insert(&ofproto->cookies, &rule->cookie_node,
3799 hash_cookie(rule->flow_cookie));
3803 cookies_remove(struct ofproto *ofproto, struct rule *rule)
3804 OVS_REQUIRES(ofproto_mutex)
3806 hindex_remove(&ofproto->cookies, &rule->cookie_node);
3810 calc_duration(long long int start, long long int now,
3811 uint32_t *sec, uint32_t *nsec)
3813 long long int msecs = now - start;
3814 *sec = msecs / 1000;
3815 *nsec = (msecs % 1000) * (1000 * 1000);
3818 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
3819 * true if 'table_id' is OK, false otherwise. */
3821 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
3823 return table_id == OFPTT_ALL || table_id < ofproto->n_tables;
3826 static struct oftable *
3827 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
3829 struct oftable *table;
3831 for (table = &ofproto->tables[table_id];
3832 table < &ofproto->tables[ofproto->n_tables];
3834 if (!(table->flags & OFTABLE_HIDDEN)) {
3842 static struct oftable *
3843 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
3845 if (table_id == 0xff) {
3846 return next_visible_table(ofproto, 0);
3847 } else if (table_id < ofproto->n_tables) {
3848 return &ofproto->tables[table_id];
3854 static struct oftable *
3855 next_matching_table(const struct ofproto *ofproto,
3856 const struct oftable *table, uint8_t table_id)
3858 return (table_id == 0xff
3859 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
3863 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
3865 * - If TABLE_ID is 0xff, this iterates over every classifier table in
3866 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
3868 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
3869 * only once, for that table. (This can be used to access tables marked
3872 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
3873 * entered at all. (Perhaps you should have validated TABLE_ID with
3874 * check_table_id().)
3876 * All parameters are evaluated multiple times.
3878 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
3879 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
3881 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
3883 /* Initializes 'criteria' in a straightforward way based on the other
3886 * By default, the criteria include flows that are read-only, on the assumption
3887 * that the collected flows won't be modified. Call rule_criteria_require_rw()
3888 * if flows will be modified.
3890 * For "loose" matching, the 'priority' parameter is unimportant and may be
3893 rule_criteria_init(struct rule_criteria *criteria, uint8_t table_id,
3894 const struct match *match, int priority,
3895 cls_version_t version, ovs_be64 cookie,
3896 ovs_be64 cookie_mask, ofp_port_t out_port,
3899 criteria->table_id = table_id;
3900 cls_rule_init(&criteria->cr, match, priority);
3901 criteria->version = version;
3902 criteria->cookie = cookie;
3903 criteria->cookie_mask = cookie_mask;
3904 criteria->out_port = out_port;
3905 criteria->out_group = out_group;
3907 /* We ordinarily want to skip hidden rules, but there has to be a way for
3908 * code internal to OVS to modify and delete them, so if the criteria
3909 * specify a priority that can only be for a hidden flow, then allow hidden
3910 * rules to be selected. (This doesn't allow OpenFlow clients to meddle
3911 * with hidden flows because OpenFlow uses only a 16-bit field to specify
3913 criteria->include_hidden = priority > UINT16_MAX;
3915 /* We assume that the criteria are being used to collect flows for reading
3916 * but not modification. Thus, we should collect read-only flows. */
3917 criteria->include_readonly = true;
3920 /* By default, criteria initialized by rule_criteria_init() will match flows
3921 * that are read-only, on the assumption that the collected flows won't be
3922 * modified. Call this function to match only flows that are be modifiable.
3924 * Specify 'can_write_readonly' as false in ordinary circumstances, true if the
3925 * caller has special privileges that allow it to modify even "read-only"
3928 rule_criteria_require_rw(struct rule_criteria *criteria,
3929 bool can_write_readonly)
3931 criteria->include_readonly = can_write_readonly;
3935 rule_criteria_destroy(struct rule_criteria *criteria)
3937 cls_rule_destroy(&criteria->cr);
3941 rule_collection_init(struct rule_collection *rules)
3943 rules->rules = rules->stub;
3945 rules->capacity = ARRAY_SIZE(rules->stub);
3949 rule_collection_add(struct rule_collection *rules, struct rule *rule)
3951 if (rules->n >= rules->capacity) {
3952 size_t old_size, new_size;
3954 old_size = rules->capacity * sizeof *rules->rules;
3955 rules->capacity *= 2;
3956 new_size = rules->capacity * sizeof *rules->rules;
3958 if (rules->rules == rules->stub) {
3959 rules->rules = xmalloc(new_size);
3960 memcpy(rules->rules, rules->stub, old_size);
3962 rules->rules = xrealloc(rules->rules, new_size);
3966 rules->rules[rules->n++] = rule;
3970 rule_collection_ref(struct rule_collection *rules)
3971 OVS_REQUIRES(ofproto_mutex)
3975 for (i = 0; i < rules->n; i++) {
3976 ofproto_rule_ref(rules->rules[i]);
3981 rule_collection_unref(struct rule_collection *rules)
3985 for (i = 0; i < rules->n; i++) {
3986 ofproto_rule_unref(rules->rules[i]);
3990 /* Returns a NULL-terminated array of rule pointers,
3991 * destroys 'rules'. */
3992 static struct rule **
3993 rule_collection_detach(struct rule_collection *rules)
3995 struct rule **rule_array;
3997 rule_collection_add(rules, NULL);
3999 if (rules->rules == rules->stub) {
4000 rules->rules = xmemdup(rules->rules, rules->n * sizeof *rules->rules);
4003 rule_array = rules->rules;
4004 rule_collection_init(rules);
4010 rule_collection_destroy(struct rule_collection *rules)
4012 if (rules->rules != rules->stub) {
4016 /* Make repeated destruction harmless. */
4017 rule_collection_init(rules);
4020 /* Schedules postponed removal of rules, destroys 'rules'. */
4022 rule_collection_remove_postponed(struct rule_collection *rules)
4023 OVS_REQUIRES(ofproto_mutex)
4026 if (rules->n == 1) {
4027 ovsrcu_postpone(remove_rule_rcu, rules->rules[0]);
4029 ovsrcu_postpone(remove_rules_rcu, rule_collection_detach(rules));
4034 /* Checks whether 'rule' matches 'c' and, if so, adds it to 'rules'. This
4035 * function verifies most of the criteria in 'c' itself, but the caller must
4036 * check 'c->cr' itself.
4038 * Rules that have already been marked for removal are not collected.
4040 * Increments '*n_readonly' if 'rule' wasn't added because it's read-only (and
4041 * 'c' only includes modifiable rules). */
4043 collect_rule(struct rule *rule, const struct rule_criteria *c,
4044 struct rule_collection *rules, size_t *n_readonly)
4045 OVS_REQUIRES(ofproto_mutex)
4047 if ((c->table_id == rule->table_id || c->table_id == 0xff)
4048 && ofproto_rule_has_out_port(rule, c->out_port)
4049 && ofproto_rule_has_out_group(rule, c->out_group)
4050 && !((rule->flow_cookie ^ c->cookie) & c->cookie_mask)
4051 && (!rule_is_hidden(rule) || c->include_hidden)
4052 && cls_rule_visible_in_version(&rule->cr, c->version)) {
4053 /* Rule matches all the criteria... */
4054 if (!rule_is_readonly(rule) || c->include_readonly) {
4056 rule_collection_add(rules, rule);
4058 /* ...except it's read-only. */
4064 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
4065 * on classifiers rules are done in the "loose" way required for OpenFlow
4066 * OFPFC_MODIFY and OFPFC_DELETE requests. Puts the selected rules on list
4069 * Returns 0 on success, otherwise an OpenFlow error code. */
4071 collect_rules_loose(struct ofproto *ofproto,
4072 const struct rule_criteria *criteria,
4073 struct rule_collection *rules)
4074 OVS_REQUIRES(ofproto_mutex)
4076 struct oftable *table;
4077 enum ofperr error = 0;
4078 size_t n_readonly = 0;
4080 rule_collection_init(rules);
4082 if (!check_table_id(ofproto, criteria->table_id)) {
4083 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4087 if (criteria->cookie_mask == OVS_BE64_MAX) {
4090 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
4091 hash_cookie(criteria->cookie),
4092 &ofproto->cookies) {
4093 if (cls_rule_is_loose_match(&rule->cr, &criteria->cr.match)) {
4094 collect_rule(rule, criteria, rules, &n_readonly);
4098 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
4101 CLS_FOR_EACH_TARGET (rule, cr, &table->cls, &criteria->cr,
4102 criteria->version) {
4103 collect_rule(rule, criteria, rules, &n_readonly);
4109 if (!error && !rules->n && n_readonly) {
4110 /* We didn't find any rules to modify. We did find some read-only
4111 * rules that we're not allowed to modify, so report that. */
4112 error = OFPERR_OFPBRC_EPERM;
4115 rule_collection_destroy(rules);
4120 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
4121 * on classifiers rules are done in the "strict" way required for OpenFlow
4122 * OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests. Puts the selected
4123 * rules on list 'rules'.
4125 * Returns 0 on success, otherwise an OpenFlow error code. */
4127 collect_rules_strict(struct ofproto *ofproto,
4128 const struct rule_criteria *criteria,
4129 struct rule_collection *rules)
4130 OVS_REQUIRES(ofproto_mutex)
4132 struct oftable *table;
4133 size_t n_readonly = 0;
4134 enum ofperr error = 0;
4136 rule_collection_init(rules);
4138 if (!check_table_id(ofproto, criteria->table_id)) {
4139 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4143 if (criteria->cookie_mask == OVS_BE64_MAX) {
4146 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
4147 hash_cookie(criteria->cookie),
4148 &ofproto->cookies) {
4149 if (cls_rule_equal(&rule->cr, &criteria->cr)) {
4150 collect_rule(rule, criteria, rules, &n_readonly);
4154 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
4157 rule = rule_from_cls_rule(classifier_find_rule_exactly(
4158 &table->cls, &criteria->cr,
4159 criteria->version));
4161 collect_rule(rule, criteria, rules, &n_readonly);
4167 if (!error && !rules->n && n_readonly) {
4168 /* We didn't find any rules to modify. We did find some read-only
4169 * rules that we're not allowed to modify, so report that. */
4170 error = OFPERR_OFPBRC_EPERM;
4173 rule_collection_destroy(rules);
4178 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
4179 * forced into the range of a uint16_t. */
4181 age_secs(long long int age_ms)
4183 return (age_ms < 0 ? 0
4184 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
4185 : (unsigned int) age_ms / 1000);
4189 handle_flow_stats_request(struct ofconn *ofconn,
4190 const struct ofp_header *request)
4191 OVS_EXCLUDED(ofproto_mutex)
4193 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4194 struct ofputil_flow_stats_request fsr;
4195 struct rule_criteria criteria;
4196 struct rule_collection rules;
4197 struct ovs_list replies;
4201 error = ofputil_decode_flow_stats_request(&fsr, request);
4206 rule_criteria_init(&criteria, fsr.table_id, &fsr.match, 0, CLS_MAX_VERSION,
4207 fsr.cookie, fsr.cookie_mask, fsr.out_port,
4210 ovs_mutex_lock(&ofproto_mutex);
4211 error = collect_rules_loose(ofproto, &criteria, &rules);
4212 rule_criteria_destroy(&criteria);
4214 rule_collection_ref(&rules);
4216 ovs_mutex_unlock(&ofproto_mutex);
4222 ofpmp_init(&replies, request);
4223 for (i = 0; i < rules.n; i++) {
4224 struct rule *rule = rules.rules[i];
4225 long long int now = time_msec();
4226 struct ofputil_flow_stats fs;
4227 long long int created, used, modified;
4228 const struct rule_actions *actions;
4229 enum ofputil_flow_mod_flags flags;
4231 ovs_mutex_lock(&rule->mutex);
4232 fs.cookie = rule->flow_cookie;
4233 fs.idle_timeout = rule->idle_timeout;
4234 fs.hard_timeout = rule->hard_timeout;
4235 fs.importance = rule->importance;
4236 created = rule->created;
4237 modified = rule->modified;
4238 actions = rule_get_actions(rule);
4239 flags = rule->flags;
4240 ovs_mutex_unlock(&rule->mutex);
4242 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
4243 &fs.byte_count, &used);
4245 minimatch_expand(&rule->cr.match, &fs.match);
4246 fs.table_id = rule->table_id;
4247 calc_duration(created, now, &fs.duration_sec, &fs.duration_nsec);
4248 fs.priority = rule->cr.priority;
4249 fs.idle_age = age_secs(now - used);
4250 fs.hard_age = age_secs(now - modified);
4251 fs.ofpacts = actions->ofpacts;
4252 fs.ofpacts_len = actions->ofpacts_len;
4255 ofputil_append_flow_stats_reply(&fs, &replies);
4258 rule_collection_unref(&rules);
4259 rule_collection_destroy(&rules);
4261 ofconn_send_replies(ofconn, &replies);
4267 flow_stats_ds(struct rule *rule, struct ds *results)
4269 uint64_t packet_count, byte_count;
4270 const struct rule_actions *actions;
4271 long long int created, used;
4273 rule->ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
4274 &byte_count, &used);
4276 ovs_mutex_lock(&rule->mutex);
4277 actions = rule_get_actions(rule);
4278 created = rule->created;
4279 ovs_mutex_unlock(&rule->mutex);
4281 if (rule->table_id != 0) {
4282 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
4284 ds_put_format(results, "duration=%llds, ", (time_msec() - created) / 1000);
4285 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
4286 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
4287 cls_rule_format(&rule->cr, results);
4288 ds_put_char(results, ',');
4290 ds_put_cstr(results, "actions=");
4291 ofpacts_format(actions->ofpacts, actions->ofpacts_len, results);
4293 ds_put_cstr(results, "\n");
4296 /* Adds a pretty-printed description of all flows to 'results', including
4297 * hidden flows (e.g., set up by in-band control). */
4299 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
4301 struct oftable *table;
4303 OFPROTO_FOR_EACH_TABLE (table, p) {
4306 CLS_FOR_EACH (rule, cr, &table->cls) {
4307 flow_stats_ds(rule, results);
4312 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
4313 * '*engine_type' and '*engine_id', respectively. */
4315 ofproto_get_netflow_ids(const struct ofproto *ofproto,
4316 uint8_t *engine_type, uint8_t *engine_id)
4318 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
4321 /* Checks the status change of CFM on 'ofport'.
4323 * Returns true if 'ofproto_class' does not support 'cfm_status_changed'. */
4325 ofproto_port_cfm_status_changed(struct ofproto *ofproto, ofp_port_t ofp_port)
4327 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
4328 return (ofport && ofproto->ofproto_class->cfm_status_changed
4329 ? ofproto->ofproto_class->cfm_status_changed(ofport)
4333 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'.
4334 * Returns 0 if the port's CFM status was successfully stored into
4335 * '*status'. Returns positive errno if the port did not have CFM
4338 * The caller must provide and own '*status', and must free 'status->rmps'.
4339 * '*status' is indeterminate if the return value is non-zero. */
4341 ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port,
4342 struct cfm_status *status)
4344 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
4345 return (ofport && ofproto->ofproto_class->get_cfm_status
4346 ? ofproto->ofproto_class->get_cfm_status(ofport, status)
4351 handle_aggregate_stats_request(struct ofconn *ofconn,
4352 const struct ofp_header *oh)
4353 OVS_EXCLUDED(ofproto_mutex)
4355 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4356 struct ofputil_flow_stats_request request;
4357 struct ofputil_aggregate_stats stats;
4358 bool unknown_packets, unknown_bytes;
4359 struct rule_criteria criteria;
4360 struct rule_collection rules;
4361 struct ofpbuf *reply;
4365 error = ofputil_decode_flow_stats_request(&request, oh);
4370 rule_criteria_init(&criteria, request.table_id, &request.match, 0,
4371 CLS_MAX_VERSION, request.cookie, request.cookie_mask,
4372 request.out_port, request.out_group);
4374 ovs_mutex_lock(&ofproto_mutex);
4375 error = collect_rules_loose(ofproto, &criteria, &rules);
4376 rule_criteria_destroy(&criteria);
4378 rule_collection_ref(&rules);
4380 ovs_mutex_unlock(&ofproto_mutex);
4386 memset(&stats, 0, sizeof stats);
4387 unknown_packets = unknown_bytes = false;
4388 for (i = 0; i < rules.n; i++) {
4389 struct rule *rule = rules.rules[i];
4390 uint64_t packet_count;
4391 uint64_t byte_count;
4394 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
4395 &byte_count, &used);
4397 if (packet_count == UINT64_MAX) {
4398 unknown_packets = true;
4400 stats.packet_count += packet_count;
4403 if (byte_count == UINT64_MAX) {
4404 unknown_bytes = true;
4406 stats.byte_count += byte_count;
4411 if (unknown_packets) {
4412 stats.packet_count = UINT64_MAX;
4414 if (unknown_bytes) {
4415 stats.byte_count = UINT64_MAX;
4418 rule_collection_unref(&rules);
4419 rule_collection_destroy(&rules);
4421 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
4422 ofconn_send_reply(ofconn, reply);
4427 struct queue_stats_cbdata {
4428 struct ofport *ofport;
4429 struct ovs_list replies;
4434 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
4435 const struct netdev_queue_stats *stats)
4437 struct ofputil_queue_stats oqs;
4439 oqs.port_no = cbdata->ofport->pp.port_no;
4440 oqs.queue_id = queue_id;
4441 oqs.tx_bytes = stats->tx_bytes;
4442 oqs.tx_packets = stats->tx_packets;
4443 oqs.tx_errors = stats->tx_errors;
4444 if (stats->created != LLONG_MIN) {
4445 calc_duration(stats->created, cbdata->now,
4446 &oqs.duration_sec, &oqs.duration_nsec);
4448 oqs.duration_sec = oqs.duration_nsec = UINT32_MAX;
4450 ofputil_append_queue_stat(&cbdata->replies, &oqs);
4454 handle_queue_stats_dump_cb(uint32_t queue_id,
4455 struct netdev_queue_stats *stats,
4458 struct queue_stats_cbdata *cbdata = cbdata_;
4460 put_queue_stats(cbdata, queue_id, stats);
4464 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
4465 struct queue_stats_cbdata *cbdata)
4467 cbdata->ofport = port;
4468 if (queue_id == OFPQ_ALL) {
4469 netdev_dump_queue_stats(port->netdev,
4470 handle_queue_stats_dump_cb, cbdata);
4472 struct netdev_queue_stats stats;
4474 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
4475 put_queue_stats(cbdata, queue_id, &stats);
4477 return OFPERR_OFPQOFC_BAD_QUEUE;
4484 handle_queue_stats_request(struct ofconn *ofconn,
4485 const struct ofp_header *rq)
4487 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4488 struct queue_stats_cbdata cbdata;
4489 struct ofport *port;
4491 struct ofputil_queue_stats_request oqsr;
4493 COVERAGE_INC(ofproto_queue_req);
4495 ofpmp_init(&cbdata.replies, rq);
4496 cbdata.now = time_msec();
4498 error = ofputil_decode_queue_stats_request(rq, &oqsr);
4503 if (oqsr.port_no == OFPP_ANY) {
4504 error = OFPERR_OFPQOFC_BAD_QUEUE;
4505 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
4506 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
4511 port = ofproto_get_port(ofproto, oqsr.port_no);
4513 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
4514 : OFPERR_OFPQOFC_BAD_PORT);
4517 ofconn_send_replies(ofconn, &cbdata.replies);
4519 ofpbuf_list_delete(&cbdata.replies);
4526 evict_rules_from_table(struct oftable *table)
4527 OVS_REQUIRES(ofproto_mutex)
4529 enum ofperr error = 0;
4530 struct rule_collection rules;
4531 unsigned int count = table->n_flows;
4532 unsigned int max_flows = table->max_flows;
4534 rule_collection_init(&rules);
4536 while (count-- > max_flows) {
4539 if (!choose_rule_to_evict(table, &rule)) {
4540 error = OFPERR_OFPFMFC_TABLE_FULL;
4543 eviction_group_remove_rule(rule);
4544 rule_collection_add(&rules, rule);
4547 delete_flows__(&rules, OFPRR_EVICTION, NULL);
4553 get_conjunctions(const struct ofputil_flow_mod *fm,
4554 struct cls_conjunction **conjsp, size_t *n_conjsp)
4555 OVS_REQUIRES(ofproto_mutex)
4557 struct cls_conjunction *conjs = NULL;
4560 const struct ofpact *ofpact;
4561 OFPACT_FOR_EACH (ofpact, fm->ofpacts, fm->ofpacts_len) {
4562 if (ofpact->type == OFPACT_CONJUNCTION) {
4564 } else if (ofpact->type != OFPACT_NOTE) {
4565 /* "conjunction" may appear with "note" actions but not with any
4566 * other type of actions. */
4567 ovs_assert(!n_conjs);
4574 conjs = xzalloc(n_conjs * sizeof *conjs);
4575 OFPACT_FOR_EACH (ofpact, fm->ofpacts, fm->ofpacts_len) {
4576 if (ofpact->type == OFPACT_CONJUNCTION) {
4577 struct ofpact_conjunction *oc = ofpact_get_CONJUNCTION(ofpact);
4578 conjs[i].clause = oc->clause;
4579 conjs[i].n_clauses = oc->n_clauses;
4580 conjs[i].id = oc->id;
4587 *n_conjsp = n_conjs;
4590 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
4591 * in which no matching flow already exists in the flow table.
4593 * Adds the flow specified by 'fm', to the ofproto's flow table. Returns 0 on
4594 * success, or an OpenFlow error code on failure.
4596 * On successful return the caller must complete the operation either by
4597 * calling add_flow_finish(), or add_flow_revert() if the operation needs to
4600 * The caller retains ownership of 'fm->ofpacts'. */
4602 add_flow_start(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
4603 OVS_REQUIRES(ofproto_mutex)
4605 struct ofputil_flow_mod *fm = &ofm->fm;
4606 struct rule **old_rule = &ofm->old_rules.stub[0];
4607 struct rule **new_rule = &ofm->new_rules.stub[0];
4608 struct oftable *table;
4612 struct cls_conjunction *conjs;
4616 if (!check_table_id(ofproto, fm->table_id)) {
4617 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4622 if (fm->table_id == 0xff) {
4623 if (ofproto->ofproto_class->rule_choose_table) {
4624 error = ofproto->ofproto_class->rule_choose_table(ofproto,
4630 ovs_assert(table_id < ofproto->n_tables);
4634 } else if (fm->table_id < ofproto->n_tables) {
4635 table_id = fm->table_id;
4637 return OFPERR_OFPBRC_BAD_TABLE_ID;
4640 table = &ofproto->tables[table_id];
4641 if (table->flags & OFTABLE_READONLY
4642 && !(fm->flags & OFPUTIL_FF_NO_READONLY)) {
4643 return OFPERR_OFPBRC_EPERM;
4646 if (!(fm->flags & OFPUTIL_FF_HIDDEN_FIELDS)
4647 && !match_has_default_hidden_fields(&fm->match)) {
4648 VLOG_WARN_RL(&rl, "%s: (add_flow) only internal flows can set "
4649 "non-default values to hidden fields", ofproto->name);
4650 return OFPERR_OFPBRC_EPERM;
4653 cls_rule_init(&cr, &fm->match, fm->priority);
4655 /* Check for the existence of an identical rule.
4656 * This will not return rules earlier marked for removal. */
4657 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr,
4661 /* Check for overlap, if requested. */
4662 if (fm->flags & OFPUTIL_FF_CHECK_OVERLAP
4663 && classifier_rule_overlaps(&table->cls, &cr, ofm->version)) {
4664 cls_rule_destroy(&cr);
4665 return OFPERR_OFPFMFC_OVERLAP;
4668 /* If necessary, evict an existing rule to clear out space. */
4669 if (table->n_flows >= table->max_flows) {
4670 if (!choose_rule_to_evict(table, &rule)) {
4671 error = OFPERR_OFPFMFC_TABLE_FULL;
4672 cls_rule_destroy(&cr);
4675 eviction_group_remove_rule(rule);
4676 /* Marks '*old_rule' as an evicted rule rather than replaced rule.
4678 fm->delete_reason = OFPRR_EVICTION;
4682 fm->modify_cookie = true;
4685 /* Allocate new rule. */
4686 error = replace_rule_create(ofproto, fm, &cr, table - ofproto->tables,
4692 get_conjunctions(fm, &conjs, &n_conjs);
4693 replace_rule_start(ofproto, ofm->version, rule, *new_rule, conjs, n_conjs);
4699 /* Revert the effects of add_flow_start(). */
4701 add_flow_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
4702 OVS_REQUIRES(ofproto_mutex)
4704 struct ofputil_flow_mod *fm = &ofm->fm;
4705 struct rule *old_rule = ofm->old_rules.stub[0];
4706 struct rule *new_rule = ofm->new_rules.stub[0];
4708 if (old_rule && fm->delete_reason == OFPRR_EVICTION) {
4709 /* Revert the eviction. */
4710 eviction_group_add_rule(old_rule);
4713 replace_rule_revert(ofproto, old_rule, new_rule);
4716 /* To be called after version bump. */
4718 add_flow_finish(struct ofproto *ofproto, struct ofproto_flow_mod *ofm,
4719 const struct flow_mod_requester *req)
4720 OVS_REQUIRES(ofproto_mutex)
4722 struct ofputil_flow_mod *fm = &ofm->fm;
4723 struct rule *old_rule = ofm->old_rules.stub[0];
4724 struct rule *new_rule = ofm->new_rules.stub[0];
4725 struct ovs_list dead_cookies = OVS_LIST_INITIALIZER(&dead_cookies);
4727 replace_rule_finish(ofproto, fm, req, old_rule, new_rule, &dead_cookies);
4728 learned_cookies_flush(ofproto, &dead_cookies);
4731 ovsrcu_postpone(remove_rule_rcu, old_rule);
4733 ofmonitor_report(ofproto->connmgr, new_rule, NXFME_ADDED, 0,
4734 req ? req->ofconn : NULL,
4735 req ? req->request->xid : 0, NULL);
4737 /* Send Vacancy Events for OF1.4+. */
4738 send_table_status(ofproto, new_rule->table_id);
4741 send_buffered_packet(req, fm->buffer_id, new_rule);
4744 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
4746 /* Create a new rule based on attributes in 'fm', match in 'cr', 'table_id',
4747 * and 'old_rule'. Note that the rule is NOT inserted into a any data
4748 * structures yet. Takes ownership of 'cr'. */
4750 replace_rule_create(struct ofproto *ofproto, struct ofputil_flow_mod *fm,
4751 struct cls_rule *cr, uint8_t table_id,
4752 struct rule *old_rule, struct rule **new_rule)
4757 /* Allocate new rule. */
4758 rule = ofproto->ofproto_class->rule_alloc();
4760 cls_rule_destroy(cr);
4761 VLOG_WARN_RL(&rl, "%s: failed to allocate a rule.", ofproto->name);
4762 return OFPERR_OFPFMFC_UNKNOWN;
4765 /* Initialize base state. */
4766 *CONST_CAST(struct ofproto **, &rule->ofproto) = ofproto;
4767 cls_rule_move(CONST_CAST(struct cls_rule *, &rule->cr), cr);
4768 ovs_refcount_init(&rule->ref_count);
4769 rule->flow_cookie = fm->new_cookie;
4770 rule->created = rule->modified = time_msec();
4772 ovs_mutex_init(&rule->mutex);
4773 ovs_mutex_lock(&rule->mutex);
4774 rule->idle_timeout = fm->idle_timeout;
4775 rule->hard_timeout = fm->hard_timeout;
4776 *CONST_CAST(uint16_t *, &rule->importance) = fm->importance;
4777 rule->removed_reason = OVS_OFPRR_NONE;
4779 *CONST_CAST(uint8_t *, &rule->table_id) = table_id;
4780 rule->flags = fm->flags & OFPUTIL_FF_STATE;
4781 *CONST_CAST(const struct rule_actions **, &rule->actions)
4782 = rule_actions_create(fm->ofpacts, fm->ofpacts_len);
4783 ovs_list_init(&rule->meter_list_node);
4784 rule->eviction_group = NULL;
4785 ovs_list_init(&rule->expirable);
4786 rule->monitor_flags = 0;
4787 rule->add_seqno = 0;
4788 rule->modify_seqno = 0;
4790 /* Copy values from old rule for modify semantics. */
4791 if (old_rule && fm->delete_reason != OFPRR_EVICTION) {
4792 bool change_cookie = (fm->modify_cookie
4793 && fm->new_cookie != OVS_BE64_MAX
4794 && fm->new_cookie != old_rule->flow_cookie);
4796 ovs_mutex_lock(&old_rule->mutex);
4797 if (fm->command != OFPFC_ADD) {
4798 rule->idle_timeout = old_rule->idle_timeout;
4799 rule->hard_timeout = old_rule->hard_timeout;
4800 *CONST_CAST(uint16_t *, &rule->importance) = old_rule->importance;
4801 rule->flags = old_rule->flags;
4802 rule->created = old_rule->created;
4804 if (!change_cookie) {
4805 rule->flow_cookie = old_rule->flow_cookie;
4807 ovs_mutex_unlock(&old_rule->mutex);
4809 ovs_mutex_unlock(&rule->mutex);
4811 /* Construct rule, initializing derived state. */
4812 error = ofproto->ofproto_class->rule_construct(rule);
4814 ofproto_rule_destroy__(rule);
4818 rule->removed = true; /* Not yet in ofproto data structures. */
4825 replace_rule_start(struct ofproto *ofproto, cls_version_t version,
4826 struct rule *old_rule, struct rule *new_rule,
4827 struct cls_conjunction *conjs, size_t n_conjs)
4829 struct oftable *table = &ofproto->tables[new_rule->table_id];
4831 /* 'old_rule' may be either an evicted rule or replaced rule. */
4833 /* Mark the old rule for removal in the next version. */
4834 cls_rule_make_invisible_in_version(&old_rule->cr, version);
4838 /* Insert flow to the classifier, so that later flow_mods may relate
4839 * to it. This is reversible, in case later errors require this to
4841 ofproto_rule_insert__(ofproto, new_rule);
4842 /* Make the new rule visible for classifier lookups only from the next
4844 classifier_insert(&table->cls, &new_rule->cr, version, conjs, n_conjs);
4847 static void replace_rule_revert(struct ofproto *ofproto,
4848 struct rule *old_rule, struct rule *new_rule)
4850 struct oftable *table = &ofproto->tables[new_rule->table_id];
4853 /* Restore the original visibility of the old rule. */
4854 cls_rule_restore_visibility(&old_rule->cr);
4856 /* Restore table's rule count. */
4860 /* Remove the new rule immediately. It was never visible to lookups. */
4861 if (!classifier_remove(&table->cls, &new_rule->cr)) {
4864 ofproto_rule_remove__(ofproto, new_rule);
4865 /* The rule was not inserted to the ofproto provider, so we can
4866 * release it without deleting it from the ofproto provider. */
4867 ofproto_rule_unref(new_rule);
4870 /* Adds the 'new_rule', replacing the 'old_rule'. */
4872 replace_rule_finish(struct ofproto *ofproto, struct ofputil_flow_mod *fm,
4873 const struct flow_mod_requester *req,
4874 struct rule *old_rule, struct rule *new_rule,
4875 struct ovs_list *dead_cookies)
4876 OVS_REQUIRES(ofproto_mutex)
4878 bool forward_stats = !(fm->flags & OFPUTIL_FF_RESET_COUNTS);
4879 struct rule *replaced_rule;
4881 replaced_rule = fm->delete_reason != OFPRR_EVICTION ? old_rule : NULL;
4883 /* Insert the new flow to the ofproto provider. A non-NULL 'replaced_rule'
4884 * is a duplicate rule the 'new_rule' is replacing. The provider should
4885 * link the stats from the old rule to the new one if 'forward_stats' is
4886 * 'true'. The 'replaced_rule' will be deleted right after this call. */
4887 ofproto->ofproto_class->rule_insert(new_rule, replaced_rule,
4889 learned_cookies_inc(ofproto, rule_get_actions(new_rule));
4892 const struct rule_actions *old_actions = rule_get_actions(old_rule);
4894 /* Remove the old rule from data structures. Removal from the
4895 * classifier and the deletion of the rule is RCU postponed by the
4897 ofproto_rule_remove__(ofproto, old_rule);
4898 learned_cookies_dec(ofproto, old_actions, dead_cookies);
4900 if (replaced_rule) {
4901 enum nx_flow_update_event event = fm->command == OFPFC_ADD
4902 ? NXFME_ADDED : NXFME_MODIFIED;
4904 bool change_cookie = (fm->modify_cookie
4905 && fm->new_cookie != OVS_BE64_MAX
4906 && fm->new_cookie != old_rule->flow_cookie);
4908 bool change_actions = !ofpacts_equal(fm->ofpacts,
4910 old_actions->ofpacts,
4911 old_actions->ofpacts_len);
4913 if (event != NXFME_MODIFIED || change_actions || change_cookie) {
4914 ofmonitor_report(ofproto->connmgr, new_rule, event, 0,
4915 req ? req->ofconn : NULL,
4916 req ? req->request->xid : 0,
4917 change_actions ? old_actions : NULL);
4920 /* XXX: This is slight duplication with delete_flows_finish__() */
4922 old_rule->removed_reason = OFPRR_EVICTION;
4924 ofmonitor_report(ofproto->connmgr, old_rule, NXFME_DELETED,
4926 req ? req->ofconn : NULL,
4927 req ? req->request->xid : 0, NULL);
4933 modify_flows_start__(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
4934 OVS_REQUIRES(ofproto_mutex)
4936 struct ofputil_flow_mod *fm = &ofm->fm;
4937 struct rule_collection *old_rules = &ofm->old_rules;
4938 struct rule_collection *new_rules = &ofm->new_rules;
4941 rule_collection_init(new_rules);
4943 if (old_rules->n > 0) {
4944 struct cls_conjunction *conjs;
4948 /* Create a new 'modified' rule for each old rule. */
4949 for (i = 0; i < old_rules->n; i++) {
4950 struct rule *old_rule = old_rules->rules[i];
4951 struct rule *new_rule;
4954 cls_rule_clone(&cr, &old_rule->cr);
4955 error = replace_rule_create(ofproto, fm, &cr, old_rule->table_id,
4956 old_rule, &new_rule);
4958 rule_collection_add(new_rules, new_rule);
4960 rule_collection_unref(new_rules);
4961 rule_collection_destroy(new_rules);
4965 ovs_assert(new_rules->n == old_rules->n);
4967 get_conjunctions(fm, &conjs, &n_conjs);
4968 for (i = 0; i < old_rules->n; i++) {
4969 replace_rule_start(ofproto, ofm->version, old_rules->rules[i],
4970 new_rules->rules[i], conjs, n_conjs);
4973 } else if (!(fm->cookie_mask != htonll(0)
4974 || fm->new_cookie == OVS_BE64_MAX)) {
4975 /* No match, add a new flow. */
4976 error = add_flow_start(ofproto, ofm);
4978 ovs_assert(fm->delete_reason == OFPRR_EVICTION
4979 || !old_rules->rules[0]);
4989 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
4992 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4995 modify_flows_start_loose(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
4996 OVS_REQUIRES(ofproto_mutex)
4998 struct ofputil_flow_mod *fm = &ofm->fm;
4999 struct rule_collection *old_rules = &ofm->old_rules;
5000 struct rule_criteria criteria;
5003 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0, CLS_MAX_VERSION,
5004 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG_ANY);
5005 rule_criteria_require_rw(&criteria,
5006 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
5007 error = collect_rules_loose(ofproto, &criteria, old_rules);
5008 rule_criteria_destroy(&criteria);
5011 error = modify_flows_start__(ofproto, ofm);
5015 rule_collection_destroy(old_rules);
5021 modify_flows_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
5022 OVS_REQUIRES(ofproto_mutex)
5024 struct rule_collection *old_rules = &ofm->old_rules;
5025 struct rule_collection *new_rules = &ofm->new_rules;
5027 /* Old rules were not changed yet, only need to revert new rules. */
5028 if (old_rules->n == 0 && new_rules->n == 1) {
5029 add_flow_revert(ofproto, ofm);
5030 } else if (old_rules->n > 0) {
5031 for (size_t i = 0; i < old_rules->n; i++) {
5032 replace_rule_revert(ofproto, old_rules->rules[i],
5033 new_rules->rules[i]);
5035 rule_collection_destroy(new_rules);
5036 rule_collection_destroy(old_rules);
5041 modify_flows_finish(struct ofproto *ofproto, struct ofproto_flow_mod *ofm,
5042 const struct flow_mod_requester *req)
5043 OVS_REQUIRES(ofproto_mutex)
5045 struct ofputil_flow_mod *fm = &ofm->fm;
5046 struct rule_collection *old_rules = &ofm->old_rules;
5047 struct rule_collection *new_rules = &ofm->new_rules;
5049 if (old_rules->n == 0 && new_rules->n == 1) {
5050 add_flow_finish(ofproto, ofm, req);
5051 } else if (old_rules->n > 0) {
5052 struct ovs_list dead_cookies = OVS_LIST_INITIALIZER(&dead_cookies);
5054 ovs_assert(new_rules->n == old_rules->n);
5056 for (size_t i = 0; i < old_rules->n; i++) {
5057 replace_rule_finish(ofproto, fm, req, old_rules->rules[i],
5058 new_rules->rules[i], &dead_cookies);
5060 learned_cookies_flush(ofproto, &dead_cookies);
5061 rule_collection_remove_postponed(old_rules);
5063 send_buffered_packet(req, fm->buffer_id, new_rules->rules[0]);
5064 rule_collection_destroy(new_rules);
5068 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
5069 * code on failure. */
5071 modify_flow_start_strict(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
5072 OVS_REQUIRES(ofproto_mutex)
5074 struct ofputil_flow_mod *fm = &ofm->fm;
5075 struct rule_collection *old_rules = &ofm->old_rules;
5076 struct rule_criteria criteria;
5079 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
5080 CLS_MAX_VERSION, fm->cookie, fm->cookie_mask, OFPP_ANY,
5082 rule_criteria_require_rw(&criteria,
5083 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
5084 error = collect_rules_strict(ofproto, &criteria, old_rules);
5085 rule_criteria_destroy(&criteria);
5088 /* collect_rules_strict() can return max 1 rule. */
5089 error = modify_flows_start__(ofproto, ofm);
5093 rule_collection_destroy(old_rules);
5098 /* OFPFC_DELETE implementation. */
5101 delete_flows_start__(struct ofproto *ofproto, cls_version_t version,
5102 const struct rule_collection *rules)
5103 OVS_REQUIRES(ofproto_mutex)
5105 for (size_t i = 0; i < rules->n; i++) {
5106 struct rule *rule = rules->rules[i];
5107 struct oftable *table = &ofproto->tables[rule->table_id];
5110 cls_rule_make_invisible_in_version(&rule->cr, version);
5115 delete_flows_finish__(struct ofproto *ofproto,
5116 struct rule_collection *rules,
5117 enum ofp_flow_removed_reason reason,
5118 const struct flow_mod_requester *req)
5119 OVS_REQUIRES(ofproto_mutex)
5122 struct ovs_list dead_cookies = OVS_LIST_INITIALIZER(&dead_cookies);
5124 for (size_t i = 0; i < rules->n; i++) {
5125 struct rule *rule = rules->rules[i];
5127 /* This value will be used to send the flow removed message right
5128 * before the rule is actually destroyed. */
5129 rule->removed_reason = reason;
5131 ofmonitor_report(ofproto->connmgr, rule, NXFME_DELETED, reason,
5132 req ? req->ofconn : NULL,
5133 req ? req->request->xid : 0, NULL);
5135 /* Send Vacancy Event for OF1.4+. */
5136 send_table_status(ofproto, rule->table_id);
5138 ofproto_rule_remove__(ofproto, rule);
5139 learned_cookies_dec(ofproto, rule_get_actions(rule),
5142 rule_collection_remove_postponed(rules);
5144 learned_cookies_flush(ofproto, &dead_cookies);
5148 /* Deletes the rules listed in 'rules'.
5149 * The deleted rules will become invisible to the lookups in the next version.
5150 * Destroys 'rules'. */
5152 delete_flows__(struct rule_collection *rules,
5153 enum ofp_flow_removed_reason reason,
5154 const struct flow_mod_requester *req)
5155 OVS_REQUIRES(ofproto_mutex)
5158 struct ofproto *ofproto = rules->rules[0]->ofproto;
5160 delete_flows_start__(ofproto, ofproto->tables_version + 1, rules);
5161 ofproto_bump_tables_version(ofproto);
5162 delete_flows_finish__(ofproto, rules, reason, req);
5163 ofmonitor_flush(ofproto->connmgr);
5167 /* Implements OFPFC_DELETE. */
5169 delete_flows_start_loose(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
5170 OVS_REQUIRES(ofproto_mutex)
5172 const struct ofputil_flow_mod *fm = &ofm->fm;
5173 struct rule_collection *rules = &ofm->old_rules;
5174 struct rule_criteria criteria;
5177 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0, CLS_MAX_VERSION,
5178 fm->cookie, fm->cookie_mask, fm->out_port,
5180 rule_criteria_require_rw(&criteria,
5181 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
5182 error = collect_rules_loose(ofproto, &criteria, rules);
5183 rule_criteria_destroy(&criteria);
5186 delete_flows_start__(ofproto, ofm->version, rules);
5193 delete_flows_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
5194 OVS_REQUIRES(ofproto_mutex)
5196 struct rule_collection *rules = &ofm->old_rules;
5198 for (size_t i = 0; i < rules->n; i++) {
5199 struct rule *rule = rules->rules[i];
5200 struct oftable *table = &ofproto->tables[rule->table_id];
5202 /* Restore table's rule count. */
5205 /* Restore the original visibility of the rule. */
5206 cls_rule_restore_visibility(&rule->cr);
5208 rule_collection_destroy(rules);
5212 delete_flows_finish(struct ofproto *ofproto,
5213 struct ofproto_flow_mod *ofm,
5214 const struct flow_mod_requester *req)
5215 OVS_REQUIRES(ofproto_mutex)
5217 delete_flows_finish__(ofproto, &ofm->old_rules, ofm->fm.delete_reason,
5221 /* Implements OFPFC_DELETE_STRICT. */
5223 delete_flow_start_strict(struct ofproto *ofproto,
5224 struct ofproto_flow_mod *ofm)
5225 OVS_REQUIRES(ofproto_mutex)
5227 const struct ofputil_flow_mod *fm = &ofm->fm;
5228 struct rule_collection *rules = &ofm->old_rules;
5229 struct rule_criteria criteria;
5232 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
5233 CLS_MAX_VERSION, fm->cookie, fm->cookie_mask,
5234 fm->out_port, fm->out_group);
5235 rule_criteria_require_rw(&criteria,
5236 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
5237 error = collect_rules_strict(ofproto, &criteria, rules);
5238 rule_criteria_destroy(&criteria);
5241 delete_flows_start__(ofproto, ofm->version, rules);
5247 /* This may only be called by rule_destroy_cb()! */
5249 ofproto_rule_send_removed(struct rule *rule)
5250 OVS_EXCLUDED(ofproto_mutex)
5252 struct ofputil_flow_removed fr;
5255 minimatch_expand(&rule->cr.match, &fr.match);
5256 fr.priority = rule->cr.priority;
5258 ovs_mutex_lock(&ofproto_mutex);
5259 fr.cookie = rule->flow_cookie;
5260 fr.reason = rule->removed_reason;
5261 fr.table_id = rule->table_id;
5262 calc_duration(rule->created, time_msec(),
5263 &fr.duration_sec, &fr.duration_nsec);
5264 ovs_mutex_lock(&rule->mutex);
5265 fr.idle_timeout = rule->idle_timeout;
5266 fr.hard_timeout = rule->hard_timeout;
5267 ovs_mutex_unlock(&rule->mutex);
5268 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
5269 &fr.byte_count, &used);
5270 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
5271 ovs_mutex_unlock(&ofproto_mutex);
5274 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
5275 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
5278 * ofproto implementation ->run() functions should use this function to expire
5279 * OpenFlow flows. */
5281 ofproto_rule_expire(struct rule *rule, uint8_t reason)
5282 OVS_REQUIRES(ofproto_mutex)
5284 struct rule_collection rules;
5286 rules.rules = rules.stub;
5288 rules.stub[0] = rule;
5289 delete_flows__(&rules, reason, NULL);
5292 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5293 * means "infinite". */
5295 reduce_timeout(uint16_t max, uint16_t *timeout)
5297 if (max && (!*timeout || *timeout > max)) {
5302 /* If 'idle_timeout' is nonzero, and 'rule' has no idle timeout or an idle
5303 * timeout greater than 'idle_timeout', lowers 'rule''s idle timeout to
5304 * 'idle_timeout' seconds. Similarly for 'hard_timeout'.
5306 * Suitable for implementing OFPACT_FIN_TIMEOUT. */
5308 ofproto_rule_reduce_timeouts(struct rule *rule,
5309 uint16_t idle_timeout, uint16_t hard_timeout)
5310 OVS_EXCLUDED(ofproto_mutex, rule->mutex)
5312 if (!idle_timeout && !hard_timeout) {
5316 ovs_mutex_lock(&ofproto_mutex);
5317 if (ovs_list_is_empty(&rule->expirable)) {
5318 ovs_list_insert(&rule->ofproto->expirable, &rule->expirable);
5320 ovs_mutex_unlock(&ofproto_mutex);
5322 ovs_mutex_lock(&rule->mutex);
5323 reduce_timeout(idle_timeout, &rule->idle_timeout);
5324 reduce_timeout(hard_timeout, &rule->hard_timeout);
5325 ovs_mutex_unlock(&rule->mutex);
5329 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5330 OVS_EXCLUDED(ofproto_mutex)
5332 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5333 struct ofproto_flow_mod ofm;
5334 uint64_t ofpacts_stub[1024 / 8];
5335 struct ofpbuf ofpacts;
5338 error = reject_slave_controller(ofconn);
5343 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
5344 error = ofputil_decode_flow_mod(&ofm.fm, oh, ofconn_get_protocol(ofconn),
5346 u16_to_ofp(ofproto->max_ports),
5349 error = ofproto_check_ofpacts(ofproto, ofm.fm.ofpacts,
5350 ofm.fm.ofpacts_len);
5353 struct flow_mod_requester req;
5355 req.ofconn = ofconn;
5357 error = handle_flow_mod__(ofproto, &ofm, &req);
5360 goto exit_free_ofpacts;
5363 ofconn_report_flow_mod(ofconn, ofm.fm.command);
5366 ofpbuf_uninit(&ofpacts);
5372 handle_flow_mod__(struct ofproto *ofproto, struct ofproto_flow_mod *ofm,
5373 const struct flow_mod_requester *req)
5374 OVS_EXCLUDED(ofproto_mutex)
5378 ovs_mutex_lock(&ofproto_mutex);
5379 ofm->version = ofproto->tables_version + 1;
5380 error = ofproto_flow_mod_start(ofproto, ofm);
5382 ofproto_bump_tables_version(ofproto);
5383 ofproto_flow_mod_finish(ofproto, ofm, req);
5385 ofmonitor_flush(ofproto->connmgr);
5386 ovs_mutex_unlock(&ofproto_mutex);
5388 run_rule_executes(ofproto);
5393 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
5395 struct ofputil_role_request request;
5396 struct ofputil_role_request reply;
5400 error = ofputil_decode_role_message(oh, &request);
5405 if (request.role != OFPCR12_ROLE_NOCHANGE) {
5406 if (request.role != OFPCR12_ROLE_EQUAL
5407 && request.have_generation_id
5408 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
5409 return OFPERR_OFPRRFC_STALE;
5412 ofconn_set_role(ofconn, request.role);
5415 reply.role = ofconn_get_role(ofconn);
5416 reply.have_generation_id = ofconn_get_master_election_id(
5417 ofconn, &reply.generation_id);
5418 buf = ofputil_encode_role_reply(oh, &reply);
5419 ofconn_send_reply(ofconn, buf);
5425 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
5426 const struct ofp_header *oh)
5428 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
5429 enum ofputil_protocol cur, next;
5431 cur = ofconn_get_protocol(ofconn);
5432 next = ofputil_protocol_set_tid(cur, msg->set != 0);
5433 ofconn_set_protocol(ofconn, next);
5439 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
5441 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
5442 enum ofputil_protocol cur, next;
5443 enum ofputil_protocol next_base;
5445 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
5447 return OFPERR_OFPBRC_EPERM;
5450 cur = ofconn_get_protocol(ofconn);
5451 next = ofputil_protocol_set_base(cur, next_base);
5452 ofconn_set_protocol(ofconn, next);
5458 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
5459 const struct ofp_header *oh)
5461 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
5464 format = ntohl(msg->format);
5465 if (!ofputil_packet_in_format_is_valid(format)) {
5466 return OFPERR_OFPBRC_EPERM;
5469 ofconn_set_packet_in_format(ofconn, format);
5474 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
5476 struct ofputil_async_cfg basis = ofconn_get_async_config(ofconn);
5477 struct ofputil_async_cfg ac;
5480 error = ofputil_decode_set_async_config(oh, false, &basis, &ac);
5485 ofconn_set_async_config(ofconn, &ac);
5486 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
5487 !ofconn_get_miss_send_len(ofconn)) {
5488 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
5495 handle_nxt_get_async_request(struct ofconn *ofconn, const struct ofp_header *oh)
5497 struct ofputil_async_cfg ac = ofconn_get_async_config(ofconn);
5498 ofconn_send_reply(ofconn, ofputil_encode_get_async_reply(oh, &ac));
5504 handle_nxt_set_controller_id(struct ofconn *ofconn,
5505 const struct ofp_header *oh)
5507 const struct nx_controller_id *nci = ofpmsg_body(oh);
5509 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
5510 return OFPERR_NXBRC_MUST_BE_ZERO;
5513 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
5518 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
5522 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
5523 ? OFPRAW_OFPT10_BARRIER_REPLY
5524 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
5525 ofconn_send_reply(ofconn, buf);
5530 ofproto_compose_flow_refresh_update(const struct rule *rule,
5531 enum nx_flow_monitor_flags flags,
5532 struct ovs_list *msgs)
5533 OVS_REQUIRES(ofproto_mutex)
5535 const struct rule_actions *actions;
5536 struct ofputil_flow_update fu;
5539 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
5540 ? NXFME_ADDED : NXFME_MODIFIED);
5542 ovs_mutex_lock(&rule->mutex);
5543 fu.idle_timeout = rule->idle_timeout;
5544 fu.hard_timeout = rule->hard_timeout;
5545 ovs_mutex_unlock(&rule->mutex);
5546 fu.table_id = rule->table_id;
5547 fu.cookie = rule->flow_cookie;
5548 minimatch_expand(&rule->cr.match, &match);
5550 fu.priority = rule->cr.priority;
5552 actions = flags & NXFMF_ACTIONS ? rule_get_actions(rule) : NULL;
5553 fu.ofpacts = actions ? actions->ofpacts : NULL;
5554 fu.ofpacts_len = actions ? actions->ofpacts_len : 0;
5556 if (ovs_list_is_empty(msgs)) {
5557 ofputil_start_flow_update(msgs);
5559 ofputil_append_flow_update(&fu, msgs);
5563 ofmonitor_compose_refresh_updates(struct rule_collection *rules,
5564 struct ovs_list *msgs)
5565 OVS_REQUIRES(ofproto_mutex)
5569 for (i = 0; i < rules->n; i++) {
5570 struct rule *rule = rules->rules[i];
5571 enum nx_flow_monitor_flags flags = rule->monitor_flags;
5572 rule->monitor_flags = 0;
5574 ofproto_compose_flow_refresh_update(rule, flags, msgs);
5579 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
5580 struct rule *rule, uint64_t seqno,
5581 struct rule_collection *rules)
5582 OVS_REQUIRES(ofproto_mutex)
5584 enum nx_flow_monitor_flags update;
5586 if (rule_is_hidden(rule)) {
5590 if (!ofproto_rule_has_out_port(rule, m->out_port)) {
5595 if (rule->add_seqno > seqno) {
5596 update = NXFMF_ADD | NXFMF_MODIFY;
5597 } else if (rule->modify_seqno > seqno) {
5598 update = NXFMF_MODIFY;
5603 if (!(m->flags & update)) {
5607 update = NXFMF_INITIAL;
5610 if (!rule->monitor_flags) {
5611 rule_collection_add(rules, rule);
5613 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
5617 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
5619 struct rule_collection *rules)
5620 OVS_REQUIRES(ofproto_mutex)
5622 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
5623 const struct oftable *table;
5624 struct cls_rule target;
5626 cls_rule_init_from_minimatch(&target, &m->match, 0);
5627 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
5630 CLS_FOR_EACH_TARGET (rule, cr, &table->cls, &target, CLS_MAX_VERSION) {
5631 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
5634 cls_rule_destroy(&target);
5638 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
5639 struct rule_collection *rules)
5640 OVS_REQUIRES(ofproto_mutex)
5642 if (m->flags & NXFMF_INITIAL) {
5643 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
5648 ofmonitor_collect_resume_rules(struct ofmonitor *m,
5649 uint64_t seqno, struct rule_collection *rules)
5650 OVS_REQUIRES(ofproto_mutex)
5652 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
5656 flow_monitor_delete(struct ofconn *ofconn, uint32_t id)
5657 OVS_REQUIRES(ofproto_mutex)
5659 struct ofmonitor *m;
5662 m = ofmonitor_lookup(ofconn, id);
5664 ofmonitor_destroy(m);
5667 error = OFPERR_OFPMOFC_UNKNOWN_MONITOR;
5674 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
5675 OVS_EXCLUDED(ofproto_mutex)
5677 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5679 struct ofpbuf b = ofpbuf_const_initializer(oh, ntohs(oh->length));
5681 struct ofmonitor **monitors = NULL;
5682 size_t allocated_monitors = 0;
5683 size_t n_monitors = 0;
5687 ovs_mutex_lock(&ofproto_mutex);
5689 struct ofputil_flow_monitor_request request;
5690 struct ofmonitor *m;
5693 retval = ofputil_decode_flow_monitor_request(&request, &b);
5694 if (retval == EOF) {
5696 } else if (retval) {
5701 if (request.table_id != 0xff
5702 && request.table_id >= ofproto->n_tables) {
5703 error = OFPERR_OFPBRC_BAD_TABLE_ID;
5707 error = ofmonitor_create(&request, ofconn, &m);
5712 if (n_monitors >= allocated_monitors) {
5713 monitors = x2nrealloc(monitors, &allocated_monitors,
5716 monitors[n_monitors++] = m;
5719 struct rule_collection rules;
5720 rule_collection_init(&rules);
5721 for (size_t i = 0; i < n_monitors; i++) {
5722 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
5725 struct ovs_list replies;
5726 ofpmp_init(&replies, oh);
5727 ofmonitor_compose_refresh_updates(&rules, &replies);
5728 ovs_mutex_unlock(&ofproto_mutex);
5730 rule_collection_destroy(&rules);
5732 ofconn_send_replies(ofconn, &replies);
5738 for (size_t i = 0; i < n_monitors; i++) {
5739 ofmonitor_destroy(monitors[i]);
5742 ovs_mutex_unlock(&ofproto_mutex);
5748 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
5749 OVS_EXCLUDED(ofproto_mutex)
5754 id = ofputil_decode_flow_monitor_cancel(oh);
5756 ovs_mutex_lock(&ofproto_mutex);
5757 error = flow_monitor_delete(ofconn, id);
5758 ovs_mutex_unlock(&ofproto_mutex);
5763 /* Meters implementation.
5765 * Meter table entry, indexed by the OpenFlow meter_id.
5766 * 'created' is used to compute the duration for meter stats.
5767 * 'list rules' is needed so that we can delete the dependent rules when the
5768 * meter table entry is deleted.
5769 * 'provider_meter_id' is for the provider's private use.
5772 long long int created; /* Time created. */
5773 struct ovs_list rules; /* List of "struct rule_dpif"s. */
5774 ofproto_meter_id provider_meter_id;
5775 uint16_t flags; /* Meter flags. */
5776 uint16_t n_bands; /* Number of meter bands. */
5777 struct ofputil_meter_band *bands;
5781 * This is used in instruction validation at flow set-up time,
5782 * as flows may not use non-existing meters.
5783 * Return value of UINT32_MAX signifies an invalid meter.
5786 get_provider_meter_id(const struct ofproto *ofproto, uint32_t of_meter_id)
5788 if (of_meter_id && of_meter_id <= ofproto->meter_features.max_meters) {
5789 const struct meter *meter = ofproto->meters[of_meter_id];
5791 return meter->provider_meter_id.uint32;
5797 /* Finds the meter invoked by 'rule''s actions and adds 'rule' to the meter's
5800 meter_insert_rule(struct rule *rule)
5802 const struct rule_actions *a = rule_get_actions(rule);
5803 uint32_t meter_id = ofpacts_get_meter(a->ofpacts, a->ofpacts_len);
5804 struct meter *meter = rule->ofproto->meters[meter_id];
5806 ovs_list_insert(&meter->rules, &rule->meter_list_node);
5810 meter_update(struct meter *meter, const struct ofputil_meter_config *config)
5814 meter->flags = config->flags;
5815 meter->n_bands = config->n_bands;
5816 meter->bands = xmemdup(config->bands,
5817 config->n_bands * sizeof *meter->bands);
5820 static struct meter *
5821 meter_create(const struct ofputil_meter_config *config,
5822 ofproto_meter_id provider_meter_id)
5824 struct meter *meter;
5826 meter = xzalloc(sizeof *meter);
5827 meter->provider_meter_id = provider_meter_id;
5828 meter->created = time_msec();
5829 ovs_list_init(&meter->rules);
5831 meter_update(meter, config);
5837 meter_delete(struct ofproto *ofproto, uint32_t first, uint32_t last)
5838 OVS_REQUIRES(ofproto_mutex)
5841 for (mid = first; mid <= last; ++mid) {
5842 struct meter *meter = ofproto->meters[mid];
5844 ofproto->meters[mid] = NULL;
5845 ofproto->ofproto_class->meter_del(ofproto,
5846 meter->provider_meter_id);
5854 handle_add_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
5856 ofproto_meter_id provider_meter_id = { UINT32_MAX };
5857 struct meter **meterp = &ofproto->meters[mm->meter.meter_id];
5861 return OFPERR_OFPMMFC_METER_EXISTS;
5864 error = ofproto->ofproto_class->meter_set(ofproto, &provider_meter_id,
5867 ovs_assert(provider_meter_id.uint32 != UINT32_MAX);
5868 *meterp = meter_create(&mm->meter, provider_meter_id);
5874 handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
5876 struct meter *meter = ofproto->meters[mm->meter.meter_id];
5878 uint32_t provider_meter_id;
5881 return OFPERR_OFPMMFC_UNKNOWN_METER;
5884 provider_meter_id = meter->provider_meter_id.uint32;
5885 error = ofproto->ofproto_class->meter_set(ofproto,
5886 &meter->provider_meter_id,
5888 ovs_assert(meter->provider_meter_id.uint32 == provider_meter_id);
5890 meter_update(meter, &mm->meter);
5896 handle_delete_meter(struct ofconn *ofconn, struct ofputil_meter_mod *mm)
5897 OVS_EXCLUDED(ofproto_mutex)
5899 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5900 uint32_t meter_id = mm->meter.meter_id;
5901 struct rule_collection rules;
5902 enum ofperr error = 0;
5903 uint32_t first, last;
5905 if (meter_id == OFPM13_ALL) {
5907 last = ofproto->meter_features.max_meters;
5909 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
5912 first = last = meter_id;
5915 /* First delete the rules that use this meter. If any of those rules are
5916 * currently being modified, postpone the whole operation until later. */
5917 rule_collection_init(&rules);
5918 ovs_mutex_lock(&ofproto_mutex);
5919 for (meter_id = first; meter_id <= last; ++meter_id) {
5920 struct meter *meter = ofproto->meters[meter_id];
5921 if (meter && !ovs_list_is_empty(&meter->rules)) {
5924 LIST_FOR_EACH (rule, meter_list_node, &meter->rules) {
5925 rule_collection_add(&rules, rule);
5929 delete_flows__(&rules, OFPRR_METER_DELETE, NULL);
5931 /* Delete the meters. */
5932 meter_delete(ofproto, first, last);
5934 ovs_mutex_unlock(&ofproto_mutex);
5940 handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5942 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5943 struct ofputil_meter_mod mm;
5944 uint64_t bands_stub[256 / 8];
5945 struct ofpbuf bands;
5949 error = reject_slave_controller(ofconn);
5954 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5956 error = ofputil_decode_meter_mod(oh, &mm, &bands);
5958 goto exit_free_bands;
5961 meter_id = mm.meter.meter_id;
5963 if (mm.command != OFPMC13_DELETE) {
5964 /* Fails also when meters are not implemented by the provider. */
5965 if (meter_id == 0 || meter_id > OFPM13_MAX) {
5966 error = OFPERR_OFPMMFC_INVALID_METER;
5967 goto exit_free_bands;
5968 } else if (meter_id > ofproto->meter_features.max_meters) {
5969 error = OFPERR_OFPMMFC_OUT_OF_METERS;
5970 goto exit_free_bands;
5972 if (mm.meter.n_bands > ofproto->meter_features.max_bands) {
5973 error = OFPERR_OFPMMFC_OUT_OF_BANDS;
5974 goto exit_free_bands;
5978 switch (mm.command) {
5980 error = handle_add_meter(ofproto, &mm);
5983 case OFPMC13_MODIFY:
5984 error = handle_modify_meter(ofproto, &mm);
5987 case OFPMC13_DELETE:
5988 error = handle_delete_meter(ofconn, &mm);
5992 error = OFPERR_OFPMMFC_BAD_COMMAND;
5997 struct ofputil_requestforward rf;
5999 rf.reason = OFPRFR_METER_MOD;
6001 connmgr_send_requestforward(ofproto->connmgr, ofconn, &rf);
6005 ofpbuf_uninit(&bands);
6010 handle_meter_features_request(struct ofconn *ofconn,
6011 const struct ofp_header *request)
6013 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6014 struct ofputil_meter_features features;
6017 if (ofproto->ofproto_class->meter_get_features) {
6018 ofproto->ofproto_class->meter_get_features(ofproto, &features);
6020 memset(&features, 0, sizeof features);
6022 b = ofputil_encode_meter_features_reply(&features, request);
6024 ofconn_send_reply(ofconn, b);
6029 handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request,
6032 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6033 struct ovs_list replies;
6034 uint64_t bands_stub[256 / 8];
6035 struct ofpbuf bands;
6036 uint32_t meter_id, first, last;
6038 ofputil_decode_meter_request(request, &meter_id);
6040 if (meter_id == OFPM13_ALL) {
6042 last = ofproto->meter_features.max_meters;
6044 if (!meter_id || meter_id > ofproto->meter_features.max_meters ||
6045 !ofproto->meters[meter_id]) {
6046 return OFPERR_OFPMMFC_UNKNOWN_METER;
6048 first = last = meter_id;
6051 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
6052 ofpmp_init(&replies, request);
6054 for (meter_id = first; meter_id <= last; ++meter_id) {
6055 struct meter *meter = ofproto->meters[meter_id];
6057 continue; /* Skip non-existing meters. */
6059 if (type == OFPTYPE_METER_STATS_REQUEST) {
6060 struct ofputil_meter_stats stats;
6062 stats.meter_id = meter_id;
6064 /* Provider sets the packet and byte counts, we do the rest. */
6065 stats.flow_count = ovs_list_size(&meter->rules);
6066 calc_duration(meter->created, time_msec(),
6067 &stats.duration_sec, &stats.duration_nsec);
6068 stats.n_bands = meter->n_bands;
6069 ofpbuf_clear(&bands);
6071 = ofpbuf_put_uninit(&bands,
6072 meter->n_bands * sizeof *stats.bands);
6074 if (!ofproto->ofproto_class->meter_get(ofproto,
6075 meter->provider_meter_id,
6077 ofputil_append_meter_stats(&replies, &stats);
6079 } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */
6080 struct ofputil_meter_config config;
6082 config.meter_id = meter_id;
6083 config.flags = meter->flags;
6084 config.n_bands = meter->n_bands;
6085 config.bands = meter->bands;
6086 ofputil_append_meter_config(&replies, &config);
6090 ofconn_send_replies(ofconn, &replies);
6091 ofpbuf_uninit(&bands);
6096 ofproto_group_lookup__(const struct ofproto *ofproto, uint32_t group_id,
6097 struct ofgroup **group)
6098 OVS_REQ_RDLOCK(ofproto->groups_rwlock)
6100 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
6101 hash_int(group_id, 0), &ofproto->groups) {
6102 if ((*group)->group_id == group_id) {
6110 /* If the group exists, this function increments the groups's reference count.
6112 * Make sure to call ofproto_group_unref() after no longer needing to maintain
6113 * a reference to the group. */
6115 ofproto_group_lookup(const struct ofproto *ofproto, uint32_t group_id,
6116 struct ofgroup **group)
6120 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
6121 found = ofproto_group_lookup__(ofproto, group_id, group);
6123 ofproto_group_ref(*group);
6125 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6130 ofproto_group_exists__(const struct ofproto *ofproto, uint32_t group_id)
6131 OVS_REQ_RDLOCK(ofproto->groups_rwlock)
6133 struct ofgroup *grp;
6135 HMAP_FOR_EACH_IN_BUCKET (grp, hmap_node,
6136 hash_int(group_id, 0), &ofproto->groups) {
6137 if (grp->group_id == group_id) {
6145 ofproto_group_exists(const struct ofproto *ofproto, uint32_t group_id)
6146 OVS_EXCLUDED(ofproto->groups_rwlock)
6150 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
6151 exists = ofproto_group_exists__(ofproto, group_id);
6152 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6158 group_get_ref_count(struct ofgroup *group)
6159 OVS_EXCLUDED(ofproto_mutex)
6161 struct ofproto *ofproto = CONST_CAST(struct ofproto *, group->ofproto);
6162 struct rule_criteria criteria;
6163 struct rule_collection rules;
6168 match_init_catchall(&match);
6169 rule_criteria_init(&criteria, 0xff, &match, 0, CLS_MAX_VERSION, htonll(0),
6170 htonll(0), OFPP_ANY, group->group_id);
6171 ovs_mutex_lock(&ofproto_mutex);
6172 error = collect_rules_loose(ofproto, &criteria, &rules);
6173 ovs_mutex_unlock(&ofproto_mutex);
6174 rule_criteria_destroy(&criteria);
6176 count = !error && rules.n < UINT32_MAX ? rules.n : UINT32_MAX;
6178 rule_collection_destroy(&rules);
6183 append_group_stats(struct ofgroup *group, struct ovs_list *replies)
6185 struct ofputil_group_stats ogs;
6186 const struct ofproto *ofproto = group->ofproto;
6187 long long int now = time_msec();
6190 ogs.bucket_stats = xmalloc(group->n_buckets * sizeof *ogs.bucket_stats);
6192 /* Provider sets the packet and byte counts, we do the rest. */
6193 ogs.ref_count = group_get_ref_count(group);
6194 ogs.n_buckets = group->n_buckets;
6196 error = (ofproto->ofproto_class->group_get_stats
6197 ? ofproto->ofproto_class->group_get_stats(group, &ogs)
6200 ogs.packet_count = UINT64_MAX;
6201 ogs.byte_count = UINT64_MAX;
6202 memset(ogs.bucket_stats, 0xff,
6203 ogs.n_buckets * sizeof *ogs.bucket_stats);
6206 ogs.group_id = group->group_id;
6207 calc_duration(group->created, now, &ogs.duration_sec, &ogs.duration_nsec);
6209 ofputil_append_group_stats(replies, &ogs);
6211 free(ogs.bucket_stats);
6215 handle_group_request(struct ofconn *ofconn,
6216 const struct ofp_header *request, uint32_t group_id,
6217 void (*cb)(struct ofgroup *, struct ovs_list *replies))
6219 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6220 struct ofgroup *group;
6221 struct ovs_list replies;
6223 ofpmp_init(&replies, request);
6224 if (group_id == OFPG_ALL) {
6225 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
6226 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
6227 cb(group, &replies);
6229 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6231 if (ofproto_group_lookup(ofproto, group_id, &group)) {
6232 cb(group, &replies);
6233 ofproto_group_unref(group);
6236 ofconn_send_replies(ofconn, &replies);
6240 handle_group_stats_request(struct ofconn *ofconn,
6241 const struct ofp_header *request)
6246 error = ofputil_decode_group_stats_request(request, &group_id);
6251 handle_group_request(ofconn, request, group_id, append_group_stats);
6256 append_group_desc(struct ofgroup *group, struct ovs_list *replies)
6258 struct ofputil_group_desc gds;
6260 gds.group_id = group->group_id;
6261 gds.type = group->type;
6262 gds.props = group->props;
6264 ofputil_append_group_desc_reply(&gds, &group->buckets, replies);
6268 handle_group_desc_stats_request(struct ofconn *ofconn,
6269 const struct ofp_header *request)
6271 handle_group_request(ofconn, request,
6272 ofputil_decode_group_desc_request(request),
6278 handle_group_features_stats_request(struct ofconn *ofconn,
6279 const struct ofp_header *request)
6281 struct ofproto *p = ofconn_get_ofproto(ofconn);
6284 msg = ofputil_encode_group_features_reply(&p->ogf, request);
6286 ofconn_send_reply(ofconn, msg);
6293 put_queue_get_config_reply(struct ofport *port, uint32_t queue,
6294 struct ovs_list *replies)
6296 struct ofputil_queue_config qc;
6298 /* None of the existing queues have compatible properties, so we hard-code
6299 * omitting min_rate and max_rate. */
6300 qc.port = port->ofp_port;
6302 qc.min_rate = UINT16_MAX;
6303 qc.max_rate = UINT16_MAX;
6304 ofputil_append_queue_get_config_reply(&qc, replies);
6308 handle_queue_get_config_request_for_port(struct ofport *port, uint32_t queue,
6309 struct ovs_list *replies)
6311 struct smap details = SMAP_INITIALIZER(&details);
6312 if (queue != OFPQ_ALL) {
6313 int error = netdev_get_queue(port->netdev, queue, &details);
6316 put_queue_get_config_reply(port, queue, replies);
6320 return OFPERR_OFPQOFC_BAD_QUEUE;
6322 return OFPERR_NXQOFC_QUEUE_ERROR;
6325 struct netdev_queue_dump queue_dump;
6328 NETDEV_QUEUE_FOR_EACH (&queue_id, &details, &queue_dump,
6330 put_queue_get_config_reply(port, queue_id, replies);
6333 smap_destroy(&details);
6338 handle_queue_get_config_request(struct ofconn *ofconn,
6339 const struct ofp_header *oh)
6341 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6342 struct ovs_list replies;
6343 struct ofport *port;
6344 ofp_port_t req_port;
6348 error = ofputil_decode_queue_get_config_request(oh, &req_port, &req_queue);
6353 ofputil_start_queue_get_config_reply(oh, &replies);
6354 if (req_port == OFPP_ANY) {
6355 error = OFPERR_OFPQOFC_BAD_QUEUE;
6356 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
6357 if (!handle_queue_get_config_request_for_port(port, req_queue,
6363 port = ofproto_get_port(ofproto, req_port);
6365 ? handle_queue_get_config_request_for_port(port, req_queue,
6367 : OFPERR_OFPQOFC_BAD_PORT);
6370 ofconn_send_replies(ofconn, &replies);
6372 ofpbuf_list_delete(&replies);
6379 init_group(struct ofproto *ofproto, const struct ofputil_group_mod *gm,
6380 struct ofgroup **ofgroup)
6383 const long long int now = time_msec();
6385 if (gm->group_id > OFPG_MAX) {
6386 return OFPERR_OFPGMFC_INVALID_GROUP;
6388 if (gm->type > OFPGT11_FF) {
6389 return OFPERR_OFPGMFC_BAD_TYPE;
6392 *ofgroup = ofproto->ofproto_class->group_alloc();
6394 VLOG_WARN_RL(&rl, "%s: failed to allocate group", ofproto->name);
6395 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
6398 (*ofgroup)->ofproto = ofproto;
6399 *CONST_CAST(uint32_t *, &((*ofgroup)->group_id)) = gm->group_id;
6400 *CONST_CAST(enum ofp11_group_type *, &(*ofgroup)->type) = gm->type;
6401 *CONST_CAST(long long int *, &((*ofgroup)->created)) = now;
6402 *CONST_CAST(long long int *, &((*ofgroup)->modified)) = now;
6403 ovs_refcount_init(&(*ofgroup)->ref_count);
6405 ovs_list_init(&(*ofgroup)->buckets);
6406 ofputil_bucket_clone_list(&(*ofgroup)->buckets, &gm->buckets, NULL);
6408 *CONST_CAST(uint32_t *, &(*ofgroup)->n_buckets) =
6409 ovs_list_size(&(*ofgroup)->buckets);
6411 memcpy(CONST_CAST(struct ofputil_group_props *, &(*ofgroup)->props),
6412 &gm->props, sizeof (struct ofputil_group_props));
6414 /* Construct called BEFORE any locks are held. */
6415 error = ofproto->ofproto_class->group_construct(*ofgroup);
6417 ofputil_bucket_list_destroy(&(*ofgroup)->buckets);
6418 ofproto->ofproto_class->group_dealloc(*ofgroup);
6423 /* Implements the OFPGC11_ADD operation specified by 'gm', adding a group to
6424 * 'ofproto''s group table. Returns 0 on success or an OpenFlow error code on
6427 add_group(struct ofproto *ofproto, const struct ofputil_group_mod *gm)
6429 struct ofgroup *ofgroup;
6432 /* Allocate new group and initialize it. */
6433 error = init_group(ofproto, gm, &ofgroup);
6438 /* We wrlock as late as possible to minimize the time we jam any other
6439 * threads: No visible state changes before acquiring the lock. */
6440 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
6442 if (ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
6443 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
6447 if (ofproto_group_exists__(ofproto, gm->group_id)) {
6448 error = OFPERR_OFPGMFC_GROUP_EXISTS;
6453 /* Insert new group. */
6454 hmap_insert(&ofproto->groups, &ofgroup->hmap_node,
6455 hash_int(ofgroup->group_id, 0));
6456 ofproto->n_groups[ofgroup->type]++;
6458 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6463 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6464 ofproto->ofproto_class->group_destruct(ofgroup);
6465 ofputil_bucket_list_destroy(&ofgroup->buckets);
6466 ofproto->ofproto_class->group_dealloc(ofgroup);
6471 /* Adds all of the buckets from 'ofgroup' to 'new_ofgroup'. The buckets
6472 * already in 'new_ofgroup' will be placed just after the (copy of the) bucket
6473 * in 'ofgroup' with bucket ID 'command_bucket_id'. Special
6474 * 'command_bucket_id' values OFPG15_BUCKET_FIRST and OFPG15_BUCKET_LAST are
6477 copy_buckets_for_insert_bucket(const struct ofgroup *ofgroup,
6478 struct ofgroup *new_ofgroup,
6479 uint32_t command_bucket_id)
6481 struct ofputil_bucket *last = NULL;
6483 if (command_bucket_id <= OFPG15_BUCKET_MAX) {
6484 /* Check here to ensure that a bucket corresponding to
6485 * command_bucket_id exists in the old bucket list.
6487 * The subsequent search of below of new_ofgroup covers
6488 * both buckets in the old bucket list and buckets added
6489 * by the insert buckets group mod message this function processes. */
6490 if (!ofputil_bucket_find(&ofgroup->buckets, command_bucket_id)) {
6491 return OFPERR_OFPGMFC_UNKNOWN_BUCKET;
6494 if (!ovs_list_is_empty(&new_ofgroup->buckets)) {
6495 last = ofputil_bucket_list_back(&new_ofgroup->buckets);
6499 ofputil_bucket_clone_list(&new_ofgroup->buckets, &ofgroup->buckets, NULL);
6501 if (ofputil_bucket_check_duplicate_id(&new_ofgroup->buckets)) {
6502 VLOG_INFO_RL(&rl, "Duplicate bucket id");
6503 return OFPERR_OFPGMFC_BUCKET_EXISTS;
6506 /* Rearrange list according to command_bucket_id */
6507 if (command_bucket_id == OFPG15_BUCKET_LAST) {
6508 if (!ovs_list_is_empty(&ofgroup->buckets)) {
6509 struct ofputil_bucket *new_first;
6510 const struct ofputil_bucket *first;
6512 first = ofputil_bucket_list_front(&ofgroup->buckets);
6513 new_first = ofputil_bucket_find(&new_ofgroup->buckets,
6516 ovs_list_splice(new_ofgroup->buckets.next, &new_first->list_node,
6517 &new_ofgroup->buckets);
6519 } else if (command_bucket_id <= OFPG15_BUCKET_MAX && last) {
6520 struct ofputil_bucket *after;
6522 /* Presence of bucket is checked above so after should never be NULL */
6523 after = ofputil_bucket_find(&new_ofgroup->buckets, command_bucket_id);
6525 ovs_list_splice(after->list_node.next, new_ofgroup->buckets.next,
6526 last->list_node.next);
6532 /* Appends all of the a copy of all the buckets from 'ofgroup' to 'new_ofgroup'
6533 * with the exception of the bucket whose bucket id is 'command_bucket_id'.
6534 * Special 'command_bucket_id' values OFPG15_BUCKET_FIRST, OFPG15_BUCKET_LAST
6535 * and OFPG15_BUCKET_ALL are also honored. */
6537 copy_buckets_for_remove_bucket(const struct ofgroup *ofgroup,
6538 struct ofgroup *new_ofgroup,
6539 uint32_t command_bucket_id)
6541 const struct ofputil_bucket *skip = NULL;
6543 if (command_bucket_id == OFPG15_BUCKET_ALL) {
6547 if (command_bucket_id == OFPG15_BUCKET_FIRST) {
6548 if (!ovs_list_is_empty(&ofgroup->buckets)) {
6549 skip = ofputil_bucket_list_front(&ofgroup->buckets);
6551 } else if (command_bucket_id == OFPG15_BUCKET_LAST) {
6552 if (!ovs_list_is_empty(&ofgroup->buckets)) {
6553 skip = ofputil_bucket_list_back(&ofgroup->buckets);
6556 skip = ofputil_bucket_find(&ofgroup->buckets, command_bucket_id);
6558 return OFPERR_OFPGMFC_UNKNOWN_BUCKET;
6562 ofputil_bucket_clone_list(&new_ofgroup->buckets, &ofgroup->buckets, skip);
6567 /* Implements OFPGC11_MODIFY, OFPGC15_INSERT_BUCKET and
6568 * OFPGC15_REMOVE_BUCKET. Returns 0 on success or an OpenFlow error code
6571 * Note that the group is re-created and then replaces the old group in
6572 * ofproto's ofgroup hash map. Thus, the group is never altered while users of
6573 * the xlate module hold a pointer to the group. */
6575 modify_group(struct ofproto *ofproto, const struct ofputil_group_mod *gm)
6577 struct ofgroup *ofgroup, *new_ofgroup, *retiring;
6580 error = init_group(ofproto, gm, &new_ofgroup);
6585 retiring = new_ofgroup;
6587 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
6588 if (!ofproto_group_lookup__(ofproto, gm->group_id, &ofgroup)) {
6589 error = OFPERR_OFPGMFC_UNKNOWN_GROUP;
6593 /* Ofproto's group write lock is held now. */
6594 if (ofgroup->type != gm->type
6595 && ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
6596 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
6600 /* Manipulate bucket list for bucket commands */
6601 if (gm->command == OFPGC15_INSERT_BUCKET) {
6602 error = copy_buckets_for_insert_bucket(ofgroup, new_ofgroup,
6603 gm->command_bucket_id);
6604 } else if (gm->command == OFPGC15_REMOVE_BUCKET) {
6605 error = copy_buckets_for_remove_bucket(ofgroup, new_ofgroup,
6606 gm->command_bucket_id);
6612 /* The group creation time does not change during modification. */
6613 *CONST_CAST(long long int *, &(new_ofgroup->created)) = ofgroup->created;
6614 *CONST_CAST(long long int *, &(new_ofgroup->modified)) = time_msec();
6616 error = ofproto->ofproto_class->group_modify(new_ofgroup);
6622 /* Replace ofgroup in ofproto's groups hash map with new_ofgroup. */
6623 hmap_remove(&ofproto->groups, &ofgroup->hmap_node);
6624 hmap_insert(&ofproto->groups, &new_ofgroup->hmap_node,
6625 hash_int(new_ofgroup->group_id, 0));
6626 if (ofgroup->type != new_ofgroup->type) {
6627 ofproto->n_groups[ofgroup->type]--;
6628 ofproto->n_groups[new_ofgroup->type]++;
6632 ofproto_group_unref(retiring);
6633 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6638 delete_group__(struct ofproto *ofproto, struct ofgroup *ofgroup)
6639 OVS_RELEASES(ofproto->groups_rwlock)
6642 struct ofproto_flow_mod ofm;
6644 /* Delete all flow entries containing this group in a group action */
6645 match_init_catchall(&match);
6646 flow_mod_init(&ofm.fm, &match, 0, NULL, 0, OFPFC_DELETE);
6647 ofm.fm.delete_reason = OFPRR_GROUP_DELETE;
6648 ofm.fm.out_group = ofgroup->group_id;
6649 ofm.fm.table_id = OFPTT_ALL;
6650 handle_flow_mod__(ofproto, &ofm, NULL);
6652 hmap_remove(&ofproto->groups, &ofgroup->hmap_node);
6653 /* No-one can find this group any more. */
6654 ofproto->n_groups[ofgroup->type]--;
6655 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6656 ofproto_group_unref(ofgroup);
6659 /* Implements OFPGC11_DELETE. */
6661 delete_group(struct ofproto *ofproto, uint32_t group_id)
6663 struct ofgroup *ofgroup;
6665 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
6666 if (group_id == OFPG_ALL) {
6668 struct hmap_node *node = hmap_first(&ofproto->groups);
6672 ofgroup = CONTAINER_OF(node, struct ofgroup, hmap_node);
6673 delete_group__(ofproto, ofgroup);
6674 /* Lock for each node separately, so that we will not jam the
6675 * other threads for too long time. */
6676 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
6679 HMAP_FOR_EACH_IN_BUCKET (ofgroup, hmap_node,
6680 hash_int(group_id, 0), &ofproto->groups) {
6681 if (ofgroup->group_id == group_id) {
6682 delete_group__(ofproto, ofgroup);
6687 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6690 /* Delete all groups from 'ofproto'.
6692 * This is intended for use within an ofproto provider's 'destruct'
6695 ofproto_group_delete_all(struct ofproto *ofproto)
6697 delete_group(ofproto, OFPG_ALL);
6701 handle_group_mod(struct ofconn *ofconn, const struct ofp_header *oh)
6703 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6704 struct ofputil_group_mod gm;
6707 error = reject_slave_controller(ofconn);
6712 error = ofputil_decode_group_mod(oh, &gm);
6717 switch (gm.command) {
6719 error = add_group(ofproto, &gm);
6722 case OFPGC11_MODIFY:
6723 error = modify_group(ofproto, &gm);
6726 case OFPGC11_DELETE:
6727 delete_group(ofproto, gm.group_id);
6731 case OFPGC15_INSERT_BUCKET:
6732 error = modify_group(ofproto, &gm);
6735 case OFPGC15_REMOVE_BUCKET:
6736 error = modify_group(ofproto, &gm);
6740 if (gm.command > OFPGC11_DELETE) {
6741 VLOG_INFO_RL(&rl, "%s: Invalid group_mod command type %d",
6742 ofproto->name, gm.command);
6744 error = OFPERR_OFPGMFC_BAD_COMMAND;
6748 struct ofputil_requestforward rf;
6750 rf.reason = OFPRFR_GROUP_MOD;
6752 connmgr_send_requestforward(ofproto->connmgr, ofconn, &rf);
6754 ofputil_bucket_list_destroy(&gm.buckets);
6759 enum ofputil_table_miss
6760 ofproto_table_get_miss_config(const struct ofproto *ofproto, uint8_t table_id)
6762 enum ofputil_table_miss miss;
6764 atomic_read_relaxed(&ofproto->tables[table_id].miss_config, &miss);
6769 table_mod__(struct oftable *oftable,
6770 const struct ofputil_table_mod *tm)
6772 if (tm->miss == OFPUTIL_TABLE_MISS_DEFAULT) {
6773 /* This is how an OFPT_TABLE_MOD decodes if it doesn't specify any
6774 * table-miss configuration (because the protocol used doesn't have
6775 * such a concept), so there's nothing to do. */
6777 atomic_store_relaxed(&oftable->miss_config, tm->miss);
6780 unsigned int new_eviction = oftable->eviction;
6781 if (tm->eviction == OFPUTIL_TABLE_EVICTION_ON) {
6782 new_eviction |= EVICTION_OPENFLOW;
6783 } else if (tm->eviction == OFPUTIL_TABLE_EVICTION_OFF) {
6784 new_eviction &= ~EVICTION_OPENFLOW;
6787 if (new_eviction != oftable->eviction) {
6788 ovs_mutex_lock(&ofproto_mutex);
6789 oftable_configure_eviction(oftable, new_eviction,
6790 oftable->eviction_fields,
6791 oftable->n_eviction_fields);
6792 ovs_mutex_unlock(&ofproto_mutex);
6795 if (tm->vacancy != OFPUTIL_TABLE_VACANCY_DEFAULT) {
6796 ovs_mutex_lock(&ofproto_mutex);
6797 oftable->vacancy_down = tm->table_vacancy.vacancy_down;
6798 oftable->vacancy_up = tm->table_vacancy.vacancy_up;
6799 if (tm->vacancy == OFPUTIL_TABLE_VACANCY_OFF) {
6800 oftable->vacancy_event = 0;
6801 } else if (!oftable->vacancy_event) {
6802 uint8_t vacancy = oftable_vacancy(oftable);
6803 oftable->vacancy_event = (vacancy < oftable->vacancy_up
6805 : OFPTR_VACANCY_DOWN);
6807 ovs_mutex_unlock(&ofproto_mutex);
6812 table_mod(struct ofproto *ofproto, const struct ofputil_table_mod *tm)
6814 if (!check_table_id(ofproto, tm->table_id)) {
6815 return OFPERR_OFPTMFC_BAD_TABLE;
6818 /* Don't allow the eviction flags to be changed (except to the only fixed
6819 * value that OVS supports). OF1.4 says this is normal: "The
6820 * OFPTMPT_EVICTION property usually cannot be modified using a
6821 * OFP_TABLE_MOD request, because the eviction mechanism is switch
6823 if (tm->eviction_flags != UINT32_MAX
6824 && tm->eviction_flags != OFPROTO_EVICTION_FLAGS) {
6825 return OFPERR_OFPTMFC_BAD_CONFIG;
6828 if (tm->table_id == OFPTT_ALL) {
6829 struct oftable *oftable;
6830 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
6831 if (!(oftable->flags & (OFTABLE_HIDDEN | OFTABLE_READONLY))) {
6832 table_mod__(oftable, tm);
6836 struct oftable *oftable = &ofproto->tables[tm->table_id];
6837 if (oftable->flags & OFTABLE_READONLY) {
6838 return OFPERR_OFPTMFC_EPERM;
6840 table_mod__(oftable, tm);
6847 handle_table_mod(struct ofconn *ofconn, const struct ofp_header *oh)
6849 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6850 struct ofputil_table_mod tm;
6853 error = reject_slave_controller(ofconn);
6858 error = ofputil_decode_table_mod(oh, &tm);
6863 return table_mod(ofproto, &tm);
6867 ofproto_flow_mod_start(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
6868 OVS_REQUIRES(ofproto_mutex)
6870 switch (ofm->fm.command) {
6872 return add_flow_start(ofproto, ofm);
6873 /* , &be->old_rules.stub[0],
6874 &be->new_rules.stub[0]); */
6876 return modify_flows_start_loose(ofproto, ofm);
6877 case OFPFC_MODIFY_STRICT:
6878 return modify_flow_start_strict(ofproto, ofm);
6880 return delete_flows_start_loose(ofproto, ofm);
6882 case OFPFC_DELETE_STRICT:
6883 return delete_flow_start_strict(ofproto, ofm);
6886 return OFPERR_OFPFMFC_BAD_COMMAND;
6890 ofproto_flow_mod_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
6891 OVS_REQUIRES(ofproto_mutex)
6893 switch (ofm->fm.command) {
6895 add_flow_revert(ofproto, ofm);
6899 case OFPFC_MODIFY_STRICT:
6900 modify_flows_revert(ofproto, ofm);
6904 case OFPFC_DELETE_STRICT:
6905 delete_flows_revert(ofproto, ofm);
6914 ofproto_flow_mod_finish(struct ofproto *ofproto,
6915 struct ofproto_flow_mod *ofm,
6916 const struct flow_mod_requester *req)
6917 OVS_REQUIRES(ofproto_mutex)
6919 switch (ofm->fm.command) {
6921 add_flow_finish(ofproto, ofm, req);
6925 case OFPFC_MODIFY_STRICT:
6926 modify_flows_finish(ofproto, ofm, req);
6930 case OFPFC_DELETE_STRICT:
6931 delete_flows_finish(ofproto, ofm, req);
6939 /* Commit phases (all while locking ofproto_mutex):
6941 * 1. Begin: Gather resources and make changes visible in the next version.
6942 * - Mark affected rules for removal in the next version.
6943 * - Create new replacement rules, make visible in the next
6945 * - Do not send any events or notifications.
6947 * 2. Revert: Fail if any errors are found. After this point no errors are
6948 * possible. No visible changes were made, so rollback is minimal (remove
6949 * added invisible rules, restore visibility of rules marked for removal).
6951 * 3. Finish: Make the changes visible for lookups. Insert replacement rules to
6952 * the ofproto provider. Remove replaced and deleted rules from ofproto data
6953 * structures, and Schedule postponed removal of deleted rules from the
6954 * classifier. Send notifications, buffered packets, etc.
6957 do_bundle_commit(struct ofconn *ofconn, uint32_t id, uint16_t flags)
6959 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6960 cls_version_t version = ofproto->tables_version + 1;
6961 struct ofp_bundle *bundle;
6962 struct ofp_bundle_entry *be;
6965 bundle = ofconn_get_bundle(ofconn, id);
6968 return OFPERR_OFPBFC_BAD_ID;
6970 if (bundle->flags != flags) {
6971 error = OFPERR_OFPBFC_BAD_FLAGS;
6973 bool prev_is_port_mod = false;
6976 ovs_mutex_lock(&ofproto_mutex);
6979 LIST_FOR_EACH (be, node, &bundle->msg_list) {
6980 if (be->type == OFPTYPE_PORT_MOD) {
6981 /* Our port mods are not atomic. */
6982 if (flags & OFPBF_ATOMIC) {
6983 error = OFPERR_OFPBFC_MSG_FAILED;
6985 prev_is_port_mod = true;
6986 error = port_mod_start(ofconn, &be->opm.pm, &be->opm.port);
6988 } else if (be->type == OFPTYPE_FLOW_MOD) {
6989 /* Flow mods between port mods are applied as a single
6990 * version, but the versions are published only after
6991 * we know the commit is successful. */
6992 if (prev_is_port_mod) {
6995 prev_is_port_mod = false;
6996 /* Store the version in which the changes should take
6998 be->ofm.version = version;
6999 error = ofproto_flow_mod_start(ofproto, &be->ofm);
7009 /* Send error referring to the original message. */
7011 ofconn_send_error(ofconn, be->ofp_msg, error);
7012 error = OFPERR_OFPBFC_MSG_FAILED;
7015 /* 2. Revert. Undo all the changes made above. */
7016 LIST_FOR_EACH_REVERSE_CONTINUE(be, node, &bundle->msg_list) {
7017 if (be->type == OFPTYPE_FLOW_MOD) {
7018 ofproto_flow_mod_revert(ofproto, &be->ofm);
7020 /* Nothing needs to be reverted for a port mod. */
7024 LIST_FOR_EACH (be, node, &bundle->msg_list) {
7025 if (be->type == OFPTYPE_FLOW_MOD) {
7026 struct flow_mod_requester req = { ofconn, be->ofp_msg };
7028 /* Bump the lookup version to the one of the current
7029 * message. This makes all the changes in the bundle at
7030 * this version visible to lookups at once. */
7031 if (ofproto->tables_version < be->ofm.version) {
7032 ofproto->tables_version = be->ofm.version;
7033 ofproto->ofproto_class->set_tables_version(
7034 ofproto, ofproto->tables_version);
7037 ofproto_flow_mod_finish(ofproto, &be->ofm, &req);
7038 } else if (be->type == OFPTYPE_PORT_MOD) {
7039 /* Perform the actual port mod. This is not atomic, i.e.,
7040 * the effects will be immediately seen by upcall
7041 * processing regardless of the lookup version. It should
7042 * be noted that port configuration changes can originate
7043 * also from OVSDB changes asynchronously to all upcall
7045 port_mod_finish(ofconn, &be->opm.pm, be->opm.port);
7050 ofmonitor_flush(ofproto->connmgr);
7051 ovs_mutex_unlock(&ofproto_mutex);
7053 run_rule_executes(ofproto);
7056 /* The bundle is discarded regardless the outcome. */
7057 ofp_bundle_remove__(ofconn, bundle, !error);
7062 handle_bundle_control(struct ofconn *ofconn, const struct ofp_header *oh)
7064 struct ofputil_bundle_ctrl_msg bctrl;
7065 struct ofputil_bundle_ctrl_msg reply;
7069 error = reject_slave_controller(ofconn);
7074 error = ofputil_decode_bundle_ctrl(oh, &bctrl);
7079 reply.bundle_id = bctrl.bundle_id;
7081 switch (bctrl.type) {
7082 case OFPBCT_OPEN_REQUEST:
7083 error = ofp_bundle_open(ofconn, bctrl.bundle_id, bctrl.flags);
7084 reply.type = OFPBCT_OPEN_REPLY;
7086 case OFPBCT_CLOSE_REQUEST:
7087 error = ofp_bundle_close(ofconn, bctrl.bundle_id, bctrl.flags);
7088 reply.type = OFPBCT_CLOSE_REPLY;
7090 case OFPBCT_COMMIT_REQUEST:
7091 error = do_bundle_commit(ofconn, bctrl.bundle_id, bctrl.flags);
7092 reply.type = OFPBCT_COMMIT_REPLY;
7094 case OFPBCT_DISCARD_REQUEST:
7095 error = ofp_bundle_discard(ofconn, bctrl.bundle_id);
7096 reply.type = OFPBCT_DISCARD_REPLY;
7099 case OFPBCT_OPEN_REPLY:
7100 case OFPBCT_CLOSE_REPLY:
7101 case OFPBCT_COMMIT_REPLY:
7102 case OFPBCT_DISCARD_REPLY:
7103 return OFPERR_OFPBFC_BAD_TYPE;
7108 buf = ofputil_encode_bundle_ctrl_reply(oh, &reply);
7109 ofconn_send_reply(ofconn, buf);
7115 handle_bundle_add(struct ofconn *ofconn, const struct ofp_header *oh)
7117 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
7119 struct ofputil_bundle_add_msg badd;
7120 struct ofp_bundle_entry *bmsg;
7123 error = reject_slave_controller(ofconn);
7128 error = ofputil_decode_bundle_add(oh, &badd, &type);
7133 bmsg = ofp_bundle_entry_alloc(type, badd.msg);
7135 if (type == OFPTYPE_PORT_MOD) {
7136 error = ofputil_decode_port_mod(badd.msg, &bmsg->opm.pm, false);
7137 } else if (type == OFPTYPE_FLOW_MOD) {
7138 struct ofpbuf ofpacts;
7139 uint64_t ofpacts_stub[1024 / 8];
7141 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
7142 error = ofputil_decode_flow_mod(&bmsg->ofm.fm, badd.msg,
7143 ofconn_get_protocol(ofconn),
7145 u16_to_ofp(ofproto->max_ports),
7147 /* Move actions to heap. */
7148 bmsg->ofm.fm.ofpacts = ofpbuf_steal_data(&ofpacts);
7150 if (!error && bmsg->ofm.fm.ofpacts_len) {
7151 error = ofproto_check_ofpacts(ofproto, bmsg->ofm.fm.ofpacts,
7152 bmsg->ofm.fm.ofpacts_len);
7159 error = ofp_bundle_add_message(ofconn, badd.bundle_id, badd.flags,
7164 ofp_bundle_entry_free(bmsg);
7171 handle_tlv_table_mod(struct ofconn *ofconn, const struct ofp_header *oh)
7173 struct ofputil_tlv_table_mod ttm;
7176 error = reject_slave_controller(ofconn);
7181 error = ofputil_decode_tlv_table_mod(oh, &ttm);
7186 error = tun_metadata_table_mod(&ttm);
7188 ofputil_uninit_tlv_table(&ttm.mappings);
7193 handle_tlv_table_request(struct ofconn *ofconn, const struct ofp_header *oh)
7195 struct ofputil_tlv_table_reply ttr;
7198 tun_metadata_table_request(&ttr);
7199 b = ofputil_encode_tlv_table_reply(oh, &ttr);
7200 ofputil_uninit_tlv_table(&ttr.mappings);
7202 ofconn_send_reply(ofconn, b);
7207 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
7208 OVS_EXCLUDED(ofproto_mutex)
7210 const struct ofp_header *oh = msg->data;
7214 error = ofptype_decode(&type, oh);
7218 if (oh->version >= OFP13_VERSION && ofpmsg_is_stat_request(oh)
7219 && ofpmp_more(oh)) {
7220 /* We have no buffer implementation for multipart requests.
7221 * Report overflow for requests which consists of multiple
7223 return OFPERR_OFPBRC_MULTIPART_BUFFER_OVERFLOW;
7227 /* OpenFlow requests. */
7228 case OFPTYPE_ECHO_REQUEST:
7229 return handle_echo_request(ofconn, oh);
7231 case OFPTYPE_FEATURES_REQUEST:
7232 return handle_features_request(ofconn, oh);
7234 case OFPTYPE_GET_CONFIG_REQUEST:
7235 return handle_get_config_request(ofconn, oh);
7237 case OFPTYPE_SET_CONFIG:
7238 return handle_set_config(ofconn, oh);
7240 case OFPTYPE_PACKET_OUT:
7241 return handle_packet_out(ofconn, oh);
7243 case OFPTYPE_PORT_MOD:
7244 return handle_port_mod(ofconn, oh);
7246 case OFPTYPE_FLOW_MOD:
7247 return handle_flow_mod(ofconn, oh);
7249 case OFPTYPE_GROUP_MOD:
7250 return handle_group_mod(ofconn, oh);
7252 case OFPTYPE_TABLE_MOD:
7253 return handle_table_mod(ofconn, oh);
7255 case OFPTYPE_METER_MOD:
7256 return handle_meter_mod(ofconn, oh);
7258 case OFPTYPE_BARRIER_REQUEST:
7259 return handle_barrier_request(ofconn, oh);
7261 case OFPTYPE_ROLE_REQUEST:
7262 return handle_role_request(ofconn, oh);
7264 /* OpenFlow replies. */
7265 case OFPTYPE_ECHO_REPLY:
7268 /* Nicira extension requests. */
7269 case OFPTYPE_FLOW_MOD_TABLE_ID:
7270 return handle_nxt_flow_mod_table_id(ofconn, oh);
7272 case OFPTYPE_SET_FLOW_FORMAT:
7273 return handle_nxt_set_flow_format(ofconn, oh);
7275 case OFPTYPE_SET_PACKET_IN_FORMAT:
7276 return handle_nxt_set_packet_in_format(ofconn, oh);
7278 case OFPTYPE_SET_CONTROLLER_ID:
7279 return handle_nxt_set_controller_id(ofconn, oh);
7281 case OFPTYPE_FLOW_AGE:
7282 /* Nothing to do. */
7285 case OFPTYPE_FLOW_MONITOR_CANCEL:
7286 return handle_flow_monitor_cancel(ofconn, oh);
7288 case OFPTYPE_SET_ASYNC_CONFIG:
7289 return handle_nxt_set_async_config(ofconn, oh);
7291 case OFPTYPE_GET_ASYNC_REQUEST:
7292 return handle_nxt_get_async_request(ofconn, oh);
7294 case OFPTYPE_NXT_RESUME:
7295 return handle_nxt_resume(ofconn, oh);
7297 /* Statistics requests. */
7298 case OFPTYPE_DESC_STATS_REQUEST:
7299 return handle_desc_stats_request(ofconn, oh);
7301 case OFPTYPE_FLOW_STATS_REQUEST:
7302 return handle_flow_stats_request(ofconn, oh);
7304 case OFPTYPE_AGGREGATE_STATS_REQUEST:
7305 return handle_aggregate_stats_request(ofconn, oh);
7307 case OFPTYPE_TABLE_STATS_REQUEST:
7308 return handle_table_stats_request(ofconn, oh);
7310 case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
7311 return handle_table_features_request(ofconn, oh);
7313 case OFPTYPE_TABLE_DESC_REQUEST:
7314 return handle_table_desc_request(ofconn, oh);
7316 case OFPTYPE_PORT_STATS_REQUEST:
7317 return handle_port_stats_request(ofconn, oh);
7319 case OFPTYPE_QUEUE_STATS_REQUEST:
7320 return handle_queue_stats_request(ofconn, oh);
7322 case OFPTYPE_PORT_DESC_STATS_REQUEST:
7323 return handle_port_desc_stats_request(ofconn, oh);
7325 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
7326 return handle_flow_monitor_request(ofconn, oh);
7328 case OFPTYPE_METER_STATS_REQUEST:
7329 case OFPTYPE_METER_CONFIG_STATS_REQUEST:
7330 return handle_meter_request(ofconn, oh, type);
7332 case OFPTYPE_METER_FEATURES_STATS_REQUEST:
7333 return handle_meter_features_request(ofconn, oh);
7335 case OFPTYPE_GROUP_STATS_REQUEST:
7336 return handle_group_stats_request(ofconn, oh);
7338 case OFPTYPE_GROUP_DESC_STATS_REQUEST:
7339 return handle_group_desc_stats_request(ofconn, oh);
7341 case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
7342 return handle_group_features_stats_request(ofconn, oh);
7344 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
7345 return handle_queue_get_config_request(ofconn, oh);
7347 case OFPTYPE_BUNDLE_CONTROL:
7348 return handle_bundle_control(ofconn, oh);
7350 case OFPTYPE_BUNDLE_ADD_MESSAGE:
7351 return handle_bundle_add(ofconn, oh);
7353 case OFPTYPE_NXT_TLV_TABLE_MOD:
7354 return handle_tlv_table_mod(ofconn, oh);
7356 case OFPTYPE_NXT_TLV_TABLE_REQUEST:
7357 return handle_tlv_table_request(ofconn, oh);
7361 case OFPTYPE_FEATURES_REPLY:
7362 case OFPTYPE_GET_CONFIG_REPLY:
7363 case OFPTYPE_PACKET_IN:
7364 case OFPTYPE_FLOW_REMOVED:
7365 case OFPTYPE_PORT_STATUS:
7366 case OFPTYPE_BARRIER_REPLY:
7367 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
7368 case OFPTYPE_DESC_STATS_REPLY:
7369 case OFPTYPE_FLOW_STATS_REPLY:
7370 case OFPTYPE_QUEUE_STATS_REPLY:
7371 case OFPTYPE_PORT_STATS_REPLY:
7372 case OFPTYPE_TABLE_STATS_REPLY:
7373 case OFPTYPE_AGGREGATE_STATS_REPLY:
7374 case OFPTYPE_PORT_DESC_STATS_REPLY:
7375 case OFPTYPE_ROLE_REPLY:
7376 case OFPTYPE_FLOW_MONITOR_PAUSED:
7377 case OFPTYPE_FLOW_MONITOR_RESUMED:
7378 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
7379 case OFPTYPE_GET_ASYNC_REPLY:
7380 case OFPTYPE_GROUP_STATS_REPLY:
7381 case OFPTYPE_GROUP_DESC_STATS_REPLY:
7382 case OFPTYPE_GROUP_FEATURES_STATS_REPLY:
7383 case OFPTYPE_METER_STATS_REPLY:
7384 case OFPTYPE_METER_CONFIG_STATS_REPLY:
7385 case OFPTYPE_METER_FEATURES_STATS_REPLY:
7386 case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
7387 case OFPTYPE_TABLE_DESC_REPLY:
7388 case OFPTYPE_ROLE_STATUS:
7389 case OFPTYPE_REQUESTFORWARD:
7390 case OFPTYPE_TABLE_STATUS:
7391 case OFPTYPE_NXT_TLV_TABLE_REPLY:
7393 if (ofpmsg_is_stat_request(oh)) {
7394 return OFPERR_OFPBRC_BAD_STAT;
7396 return OFPERR_OFPBRC_BAD_TYPE;
7402 handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
7403 OVS_EXCLUDED(ofproto_mutex)
7405 enum ofperr error = handle_openflow__(ofconn, ofp_msg);
7408 ofconn_send_error(ofconn, ofp_msg->data, error);
7410 COVERAGE_INC(ofproto_recv_openflow);
7413 /* Asynchronous operations. */
7416 send_buffered_packet(const struct flow_mod_requester *req, uint32_t buffer_id,
7418 OVS_REQUIRES(ofproto_mutex)
7420 if (req && req->ofconn && buffer_id != UINT32_MAX) {
7421 struct ofproto *ofproto = ofconn_get_ofproto(req->ofconn);
7422 struct dp_packet *packet;
7426 error = ofconn_pktbuf_retrieve(req->ofconn, buffer_id, &packet,
7429 struct rule_execute *re;
7431 ofproto_rule_ref(rule);
7433 re = xmalloc(sizeof *re);
7435 re->in_port = in_port;
7436 re->packet = packet;
7438 if (!guarded_list_push_back(&ofproto->rule_executes,
7439 &re->list_node, 1024)) {
7440 ofproto_rule_unref(rule);
7441 dp_packet_delete(re->packet);
7445 ofconn_send_error(req->ofconn, req->request, error);
7451 pick_datapath_id(const struct ofproto *ofproto)
7453 const struct ofport *port;
7455 port = ofproto_get_port(ofproto, OFPP_LOCAL);
7460 error = netdev_get_etheraddr(port->netdev, &ea);
7462 return eth_addr_to_uint64(ea);
7464 VLOG_WARN("%s: could not get MAC address for %s (%s)",
7465 ofproto->name, netdev_get_name(port->netdev),
7466 ovs_strerror(error));
7468 return ofproto->fallback_dpid;
7472 pick_fallback_dpid(void)
7475 eth_addr_nicira_random(&ea);
7476 return eth_addr_to_uint64(ea);
7479 /* Table overflow policy. */
7481 /* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep'
7482 * to NULL if the table is not configured to evict rules or if the table
7483 * contains no evictable rules. (Rules with a readlock on their evict rwlock,
7484 * or with no timeouts are not evictable.) */
7486 choose_rule_to_evict(struct oftable *table, struct rule **rulep)
7487 OVS_REQUIRES(ofproto_mutex)
7489 struct eviction_group *evg;
7492 if (!table->eviction) {
7496 /* In the common case, the outer and inner loops here will each be entered
7499 * - The inner loop normally "return"s in its first iteration. If the
7500 * eviction group has any evictable rules, then it always returns in
7503 * - The outer loop only iterates more than once if the largest eviction
7504 * group has no evictable rules.
7506 * - The outer loop can exit only if table's 'max_flows' is all filled up
7507 * by unevictable rules. */
7508 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
7511 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
7520 /* Eviction groups. */
7522 /* Returns the priority to use for an eviction_group that contains 'n_rules'
7523 * rules. The priority contains low-order random bits to ensure that eviction
7524 * groups with the same number of rules are prioritized randomly. */
7526 eviction_group_priority(size_t n_rules)
7528 uint16_t size = MIN(UINT16_MAX, n_rules);
7529 return (size << 16) | random_uint16();
7532 /* Updates 'evg', an eviction_group within 'table', following a change that
7533 * adds or removes rules in 'evg'. */
7535 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
7536 OVS_REQUIRES(ofproto_mutex)
7538 heap_change(&table->eviction_groups_by_size, &evg->size_node,
7539 eviction_group_priority(heap_count(&evg->rules)));
7542 /* Destroys 'evg', an eviction_group within 'table':
7544 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
7545 * rules themselves, just removes them from the eviction group.)
7547 * - Removes 'evg' from 'table'.
7551 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
7552 OVS_REQUIRES(ofproto_mutex)
7554 while (!heap_is_empty(&evg->rules)) {
7557 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
7558 rule->eviction_group = NULL;
7560 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
7561 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
7562 heap_destroy(&evg->rules);
7566 /* Removes 'rule' from its eviction group, if any. */
7568 eviction_group_remove_rule(struct rule *rule)
7569 OVS_REQUIRES(ofproto_mutex)
7571 if (rule->eviction_group) {
7572 struct oftable *table = &rule->ofproto->tables[rule->table_id];
7573 struct eviction_group *evg = rule->eviction_group;
7575 rule->eviction_group = NULL;
7576 heap_remove(&evg->rules, &rule->evg_node);
7577 if (heap_is_empty(&evg->rules)) {
7578 eviction_group_destroy(table, evg);
7580 eviction_group_resized(table, evg);
7585 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
7586 * returns the hash value. */
7588 eviction_group_hash_rule(struct rule *rule)
7589 OVS_REQUIRES(ofproto_mutex)
7591 struct oftable *table = &rule->ofproto->tables[rule->table_id];
7592 const struct mf_subfield *sf;
7596 hash = table->eviction_group_id_basis;
7597 miniflow_expand(rule->cr.match.flow, &flow);
7598 for (sf = table->eviction_fields;
7599 sf < &table->eviction_fields[table->n_eviction_fields];
7602 if (mf_are_prereqs_ok(sf->field, &flow)) {
7603 union mf_value value;
7605 mf_get_value(sf->field, &flow, &value);
7607 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
7609 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
7610 unsigned int start = sf->ofs + sf->n_bits;
7611 bitwise_zero(&value, sf->field->n_bytes, start,
7612 sf->field->n_bytes * 8 - start);
7614 hash = hash_bytes(&value, sf->field->n_bytes, hash);
7616 hash = hash_int(hash, 0);
7623 /* Returns an eviction group within 'table' with the given 'id', creating one
7625 static struct eviction_group *
7626 eviction_group_find(struct oftable *table, uint32_t id)
7627 OVS_REQUIRES(ofproto_mutex)
7629 struct eviction_group *evg;
7631 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
7635 evg = xmalloc(sizeof *evg);
7636 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
7637 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
7638 eviction_group_priority(0));
7639 heap_init(&evg->rules);
7644 /* Returns an eviction priority for 'rule'. The return value should be
7645 * interpreted so that higher priorities make a rule a more attractive
7646 * candidate for eviction. */
7648 rule_eviction_priority(struct ofproto *ofproto, struct rule *rule)
7649 OVS_REQUIRES(ofproto_mutex)
7651 /* Calculate absolute time when this flow will expire. If it will never
7652 * expire, then return 0 to make it unevictable. */
7653 long long int expiration = LLONG_MAX;
7654 if (rule->hard_timeout) {
7655 /* 'modified' needs protection even when we hold 'ofproto_mutex'. */
7656 ovs_mutex_lock(&rule->mutex);
7657 long long int modified = rule->modified;
7658 ovs_mutex_unlock(&rule->mutex);
7660 expiration = modified + rule->hard_timeout * 1000;
7662 if (rule->idle_timeout) {
7663 uint64_t packets, bytes;
7665 long long int idle_expiration;
7667 ofproto->ofproto_class->rule_get_stats(rule, &packets, &bytes, &used);
7668 idle_expiration = used + rule->idle_timeout * 1000;
7669 expiration = MIN(expiration, idle_expiration);
7671 if (expiration == LLONG_MAX) {
7675 /* Calculate the time of expiration as a number of (approximate) seconds
7676 * after program startup.
7678 * This should work OK for program runs that last UINT32_MAX seconds or
7679 * less. Therefore, please restart OVS at least once every 136 years. */
7680 uint32_t expiration_ofs = (expiration >> 10) - (time_boot_msec() >> 10);
7682 /* Combine expiration time with OpenFlow "importance" to form a single
7683 * priority value. We want flows with relatively low "importance" to be
7684 * evicted before even considering expiration time, so put "importance" in
7685 * the most significant bits and expiration time in the least significant
7688 * Small 'priority' should be evicted before those with large 'priority'.
7689 * The caller expects the opposite convention (a large return value being
7690 * more attractive for eviction) so we invert it before returning. */
7691 uint64_t priority = ((uint64_t) rule->importance << 32) + expiration_ofs;
7692 return UINT64_MAX - priority;
7695 /* Adds 'rule' to an appropriate eviction group for its oftable's
7696 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
7697 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
7700 * The caller must ensure that 'rule' is not already in an eviction group. */
7702 eviction_group_add_rule(struct rule *rule)
7703 OVS_REQUIRES(ofproto_mutex)
7705 struct ofproto *ofproto = rule->ofproto;
7706 struct oftable *table = &ofproto->tables[rule->table_id];
7709 /* Timeouts may be modified only when holding 'ofproto_mutex'. We have it
7710 * so no additional protection is needed. */
7711 has_timeout = rule->hard_timeout || rule->idle_timeout;
7713 if (table->eviction && has_timeout) {
7714 struct eviction_group *evg;
7716 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
7718 rule->eviction_group = evg;
7719 heap_insert(&evg->rules, &rule->evg_node,
7720 rule_eviction_priority(ofproto, rule));
7721 eviction_group_resized(table, evg);
7727 /* Initializes 'table'. */
7729 oftable_init(struct oftable *table)
7731 memset(table, 0, sizeof *table);
7732 classifier_init(&table->cls, flow_segment_u64s);
7733 table->max_flows = UINT_MAX;
7735 hmap_init(&table->eviction_groups_by_id);
7736 heap_init(&table->eviction_groups_by_size);
7737 atomic_init(&table->miss_config, OFPUTIL_TABLE_MISS_DEFAULT);
7739 classifier_set_prefix_fields(&table->cls, default_prefix_fields,
7740 ARRAY_SIZE(default_prefix_fields));
7742 atomic_init(&table->n_matched, 0);
7743 atomic_init(&table->n_missed, 0);
7746 /* Destroys 'table', including its classifier and eviction groups.
7748 * The caller is responsible for freeing 'table' itself. */
7750 oftable_destroy(struct oftable *table)
7752 ovs_assert(classifier_is_empty(&table->cls));
7754 ovs_mutex_lock(&ofproto_mutex);
7755 oftable_configure_eviction(table, 0, NULL, 0);
7756 ovs_mutex_unlock(&ofproto_mutex);
7758 hmap_destroy(&table->eviction_groups_by_id);
7759 heap_destroy(&table->eviction_groups_by_size);
7760 classifier_destroy(&table->cls);
7764 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
7765 * string, then 'table' will use its default name.
7767 * This only affects the name exposed for a table exposed through the OpenFlow
7768 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
7770 oftable_set_name(struct oftable *table, const char *name)
7772 if (name && name[0]) {
7773 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
7774 if (!table->name || strncmp(name, table->name, len)) {
7776 table->name = xmemdup0(name, len);
7784 /* oftables support a choice of two policies when adding a rule would cause the
7785 * number of flows in the table to exceed the configured maximum number: either
7786 * they can refuse to add the new flow or they can evict some existing flow.
7787 * This function configures the latter policy on 'table', with fairness based
7788 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
7789 * 'n_fields' as 0 disables fairness.) */
7791 oftable_configure_eviction(struct oftable *table, unsigned int eviction,
7792 const struct mf_subfield *fields, size_t n_fields)
7793 OVS_REQUIRES(ofproto_mutex)
7797 if ((table->eviction != 0) == (eviction != 0)
7798 && n_fields == table->n_eviction_fields
7800 || !memcmp(fields, table->eviction_fields,
7801 n_fields * sizeof *fields))) {
7802 /* The set of eviction fields did not change. If 'eviction' changed,
7803 * it remains nonzero, so that we can just update table->eviction
7804 * without fussing with the eviction groups. */
7805 table->eviction = eviction;
7809 /* Destroy existing eviction groups, then destroy and recreate data
7810 * structures to recover memory. */
7811 struct eviction_group *evg, *next;
7812 HMAP_FOR_EACH_SAFE (evg, next, id_node, &table->eviction_groups_by_id) {
7813 eviction_group_destroy(table, evg);
7815 hmap_destroy(&table->eviction_groups_by_id);
7816 hmap_init(&table->eviction_groups_by_id);
7817 heap_destroy(&table->eviction_groups_by_size);
7818 heap_init(&table->eviction_groups_by_size);
7820 /* Replace eviction groups by the new ones, if there is a change. Free the
7821 * old fields only after allocating the new ones, because 'fields ==
7822 * table->eviction_fields' is possible. */
7823 struct mf_subfield *old_fields = table->eviction_fields;
7824 table->n_eviction_fields = n_fields;
7825 table->eviction_fields = (fields
7826 ? xmemdup(fields, n_fields * sizeof *fields)
7830 /* Add the new eviction groups, if enabled. */
7831 table->eviction = eviction;
7832 if (table->eviction) {
7833 table->eviction_group_id_basis = random_uint32();
7834 CLS_FOR_EACH (rule, cr, &table->cls) {
7835 eviction_group_add_rule(rule);
7840 /* Inserts 'rule' from the ofproto data structures BEFORE caller has inserted
7841 * it to the classifier. */
7843 ofproto_rule_insert__(struct ofproto *ofproto, struct rule *rule)
7844 OVS_REQUIRES(ofproto_mutex)
7846 const struct rule_actions *actions = rule_get_actions(rule);
7848 ovs_assert(rule->removed);
7850 if (rule->hard_timeout || rule->idle_timeout) {
7851 ovs_list_insert(&ofproto->expirable, &rule->expirable);
7853 cookies_insert(ofproto, rule);
7854 eviction_group_add_rule(rule);
7855 if (actions->has_meter) {
7856 meter_insert_rule(rule);
7858 rule->removed = false;
7861 /* Removes 'rule' from the ofproto data structures. Caller may have deferred
7862 * the removal from the classifier. */
7864 ofproto_rule_remove__(struct ofproto *ofproto, struct rule *rule)
7865 OVS_REQUIRES(ofproto_mutex)
7867 ovs_assert(!rule->removed);
7869 cookies_remove(ofproto, rule);
7871 eviction_group_remove_rule(rule);
7872 if (!ovs_list_is_empty(&rule->expirable)) {
7873 ovs_list_remove(&rule->expirable);
7875 if (!ovs_list_is_empty(&rule->meter_list_node)) {
7876 ovs_list_remove(&rule->meter_list_node);
7877 ovs_list_init(&rule->meter_list_node);
7880 rule->removed = true;
7883 /* unixctl commands. */
7886 ofproto_lookup(const char *name)
7888 struct ofproto *ofproto;
7890 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
7892 if (!strcmp(ofproto->name, name)) {
7900 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
7901 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7903 struct ofproto *ofproto;
7907 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
7908 ds_put_format(&results, "%s\n", ofproto->name);
7910 unixctl_command_reply(conn, ds_cstr(&results));
7911 ds_destroy(&results);
7915 ofproto_unixctl_init(void)
7917 static bool registered;
7923 unixctl_command_register("ofproto/list", "", 0, 0,
7924 ofproto_unixctl_list, NULL);