2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
26 #include "byte-order.h"
27 #include "classifier.h"
30 #include "dynamic-string.h"
33 #include "meta-flow.h"
36 #include "ofp-actions.h"
37 #include "ofp-errors.h"
39 #include "ofp-print.h"
42 #include "ofproto-provider.h"
43 #include "openflow/nicira-ext.h"
44 #include "openflow/openflow.h"
48 #include "poll-loop.h"
54 #include "unaligned.h"
58 VLOG_DEFINE_THIS_MODULE(ofproto);
60 COVERAGE_DEFINE(ofproto_error);
61 COVERAGE_DEFINE(ofproto_flush);
62 COVERAGE_DEFINE(ofproto_no_packet_in);
63 COVERAGE_DEFINE(ofproto_packet_out);
64 COVERAGE_DEFINE(ofproto_queue_req);
65 COVERAGE_DEFINE(ofproto_recv_openflow);
66 COVERAGE_DEFINE(ofproto_reinit_ports);
67 COVERAGE_DEFINE(ofproto_uninstallable);
68 COVERAGE_DEFINE(ofproto_update_port);
71 S_OPENFLOW, /* Processing OpenFlow commands. */
72 S_EVICT, /* Evicting flows from over-limit tables. */
73 S_FLUSH, /* Deleting all flow table rules. */
76 enum ofoperation_type {
83 /* A single OpenFlow request can execute any number of operations. The
84 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
85 * ofconn to which an error reply should be sent if necessary.
87 * ofproto initiates some operations internally. These operations are still
88 * assigned to groups but will not have an associated ofconn. */
90 struct ofproto *ofproto; /* Owning ofproto. */
91 struct list ofproto_node; /* In ofproto's "pending" list. */
92 struct list ops; /* List of "struct ofoperation"s. */
93 int n_running; /* Number of ops still pending. */
95 /* Data needed to send OpenFlow reply on failure or to send a buffered
98 * If list_is_empty(ofconn_node) then this ofopgroup never had an
99 * associated ofconn or its ofconn's connection dropped after it initiated
100 * the operation. In the latter case 'ofconn' is a wild pointer that
101 * refers to freed memory, so the 'ofconn' member must be used only if
102 * !list_is_empty(ofconn_node).
104 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
105 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
106 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
107 uint32_t buffer_id; /* Buffer id from original request. */
110 static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
111 static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
112 const struct ofp_header *,
114 static void ofopgroup_submit(struct ofopgroup *);
115 static void ofopgroup_complete(struct ofopgroup *);
117 /* A single flow table operation. */
119 struct ofopgroup *group; /* Owning group. */
120 struct list group_node; /* In ofopgroup's "ops" list. */
121 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
122 struct rule *rule; /* Rule being operated upon. */
123 enum ofoperation_type type; /* Type of operation. */
125 /* OFOPERATION_MODIFY, OFOPERATION_REPLACE: The old actions, if the actions
127 struct ofpact *ofpacts;
131 /* OFOPERATION_DELETE. */
132 enum ofp_flow_removed_reason reason; /* Reason flow was removed. */
134 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
135 uint16_t idle_timeout; /* Rule's old idle timeout. */
136 uint16_t hard_timeout; /* Rule's old hard timeout. */
137 bool send_flow_removed; /* Rule's old 'send_flow_removed'. */
138 enum ofperr error; /* 0 if no error. */
141 static struct ofoperation *ofoperation_create(struct ofopgroup *,
143 enum ofoperation_type,
144 enum ofp_flow_removed_reason);
145 static void ofoperation_destroy(struct ofoperation *);
148 static void oftable_init(struct oftable *);
149 static void oftable_destroy(struct oftable *);
151 static void oftable_set_name(struct oftable *, const char *name);
153 static void oftable_disable_eviction(struct oftable *);
154 static void oftable_enable_eviction(struct oftable *,
155 const struct mf_subfield *fields,
158 static void oftable_remove_rule(struct rule *rule) OVS_RELEASES(rule->evict);
159 static void oftable_remove_rule__(struct ofproto *ofproto,
160 struct classifier *cls, struct rule *rule)
161 OVS_REQ_WRLOCK(cls->rwlock) OVS_RELEASES(rule->evict);
162 static void oftable_insert_rule(struct rule *);
164 /* A set of rules within a single OpenFlow table (oftable) that have the same
165 * values for the oftable's eviction_fields. A rule to be evicted, when one is
166 * needed, is taken from the eviction group that contains the greatest number
169 * An oftable owns any number of eviction groups, each of which contains any
172 * Membership in an eviction group is imprecise, based on the hash of the
173 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
174 * That is, if two rules have different eviction_fields, but those
175 * eviction_fields hash to the same value, then they will belong to the same
176 * eviction_group anyway.
178 * (When eviction is not enabled on an oftable, we don't track any eviction
179 * groups, to save time and space.) */
180 struct eviction_group {
181 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
182 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
183 struct heap rules; /* Contains "struct rule"s. */
186 static bool choose_rule_to_evict(struct oftable *table, struct rule **rulep)
187 OVS_TRY_WRLOCK(true, (*rulep)->evict);
188 static void ofproto_evict(struct ofproto *);
189 static uint32_t rule_eviction_priority(struct rule *);
190 static void eviction_group_add_rule(struct rule *);
191 static void eviction_group_remove_rule(struct rule *);
194 static void ofport_destroy__(struct ofport *);
195 static void ofport_destroy(struct ofport *);
197 static void update_port(struct ofproto *, const char *devname);
198 static int init_ports(struct ofproto *);
199 static void reinit_ports(struct ofproto *);
202 static void ofproto_rule_destroy__(struct rule *);
203 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
204 static bool rule_is_modifiable(const struct rule *);
207 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
208 struct ofputil_flow_mod *,
209 const struct ofp_header *);
210 static enum ofperr modify_flows__(struct ofproto *, struct ofconn *,
211 struct ofputil_flow_mod *,
212 const struct ofp_header *, struct list *);
213 static void delete_flow__(struct rule *rule, struct ofopgroup *,
214 enum ofp_flow_removed_reason)
215 OVS_RELEASES(rule->evict);
216 static bool handle_openflow(struct ofconn *, const struct ofpbuf *);
217 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
218 struct ofputil_flow_mod *,
219 const struct ofp_header *);
220 static void calc_duration(long long int start, long long int now,
221 uint32_t *sec, uint32_t *nsec);
224 static uint64_t pick_datapath_id(const struct ofproto *);
225 static uint64_t pick_fallback_dpid(void);
226 static void ofproto_destroy__(struct ofproto *);
227 static void update_mtu(struct ofproto *, struct ofport *);
228 static void meter_delete(struct ofproto *, uint32_t first, uint32_t last);
231 static void ofproto_unixctl_init(void);
233 /* All registered ofproto classes, in probe order. */
234 static const struct ofproto_class **ofproto_classes;
235 static size_t n_ofproto_classes;
236 static size_t allocated_ofproto_classes;
238 unsigned flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT;
239 unsigned n_handler_threads;
240 enum ofproto_flow_miss_model flow_miss_model = OFPROTO_HANDLE_MISS_AUTO;
242 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
243 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
245 /* Initial mappings of port to OpenFlow number mappings. */
246 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
248 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
250 /* The default value of true waits for flow restore. */
251 static bool flow_restore_wait = true;
253 /* Must be called to initialize the ofproto library.
255 * The caller may pass in 'iface_hints', which contains an shash of
256 * "iface_hint" elements indexed by the interface's name. The provider
257 * may use these hints to describe the startup configuration in order to
258 * reinitialize its state. The caller owns the provided data, so a
259 * provider will make copies of anything required. An ofproto provider
260 * will remove any existing state that is not described by the hint, and
261 * may choose to remove it all. */
263 ofproto_init(const struct shash *iface_hints)
265 struct shash_node *node;
268 ofproto_class_register(&ofproto_dpif_class);
270 /* Make a local copy, since we don't own 'iface_hints' elements. */
271 SHASH_FOR_EACH(node, iface_hints) {
272 const struct iface_hint *orig_hint = node->data;
273 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
274 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
276 new_hint->br_name = xstrdup(orig_hint->br_name);
277 new_hint->br_type = xstrdup(br_type);
278 new_hint->ofp_port = orig_hint->ofp_port;
280 shash_add(&init_ofp_ports, node->name, new_hint);
283 for (i = 0; i < n_ofproto_classes; i++) {
284 ofproto_classes[i]->init(&init_ofp_ports);
288 /* 'type' should be a normalized datapath type, as returned by
289 * ofproto_normalize_type(). Returns the corresponding ofproto_class
290 * structure, or a null pointer if there is none registered for 'type'. */
291 static const struct ofproto_class *
292 ofproto_class_find__(const char *type)
296 for (i = 0; i < n_ofproto_classes; i++) {
297 const struct ofproto_class *class = ofproto_classes[i];
302 class->enumerate_types(&types);
303 found = sset_contains(&types, type);
304 sset_destroy(&types);
310 VLOG_WARN("unknown datapath type %s", type);
314 /* Registers a new ofproto class. After successful registration, new ofprotos
315 * of that type can be created using ofproto_create(). */
317 ofproto_class_register(const struct ofproto_class *new_class)
321 for (i = 0; i < n_ofproto_classes; i++) {
322 if (ofproto_classes[i] == new_class) {
327 if (n_ofproto_classes >= allocated_ofproto_classes) {
328 ofproto_classes = x2nrealloc(ofproto_classes,
329 &allocated_ofproto_classes,
330 sizeof *ofproto_classes);
332 ofproto_classes[n_ofproto_classes++] = new_class;
336 /* Unregisters a datapath provider. 'type' must have been previously
337 * registered and not currently be in use by any ofprotos. After
338 * unregistration new datapaths of that type cannot be opened using
339 * ofproto_create(). */
341 ofproto_class_unregister(const struct ofproto_class *class)
345 for (i = 0; i < n_ofproto_classes; i++) {
346 if (ofproto_classes[i] == class) {
347 for (i++; i < n_ofproto_classes; i++) {
348 ofproto_classes[i - 1] = ofproto_classes[i];
354 VLOG_WARN("attempted to unregister an ofproto class that is not "
359 /* Clears 'types' and enumerates all registered ofproto types into it. The
360 * caller must first initialize the sset. */
362 ofproto_enumerate_types(struct sset *types)
366 for (i = 0; i < n_ofproto_classes; i++) {
367 ofproto_classes[i]->enumerate_types(types);
371 /* Returns the fully spelled out name for the given ofproto 'type'.
373 * Normalized type string can be compared with strcmp(). Unnormalized type
374 * string might be the same even if they have different spellings. */
376 ofproto_normalize_type(const char *type)
378 return type && type[0] ? type : "system";
381 /* Clears 'names' and enumerates the names of all known created ofprotos with
382 * the given 'type'. The caller must first initialize the sset. Returns 0 if
383 * successful, otherwise a positive errno value.
385 * Some kinds of datapaths might not be practically enumerable. This is not
386 * considered an error. */
388 ofproto_enumerate_names(const char *type, struct sset *names)
390 const struct ofproto_class *class = ofproto_class_find__(type);
391 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
395 ofproto_create(const char *datapath_name, const char *datapath_type,
396 struct ofproto **ofprotop)
398 const struct ofproto_class *class;
399 struct ofproto *ofproto;
405 ofproto_unixctl_init();
407 datapath_type = ofproto_normalize_type(datapath_type);
408 class = ofproto_class_find__(datapath_type);
410 VLOG_WARN("could not create datapath %s of unknown type %s",
411 datapath_name, datapath_type);
415 ofproto = class->alloc();
417 VLOG_ERR("failed to allocate datapath %s of type %s",
418 datapath_name, datapath_type);
423 memset(ofproto, 0, sizeof *ofproto);
424 ofproto->ofproto_class = class;
425 ofproto->name = xstrdup(datapath_name);
426 ofproto->type = xstrdup(datapath_type);
427 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
428 hash_string(ofproto->name, 0));
429 ofproto->datapath_id = 0;
430 ofproto->forward_bpdu = false;
431 ofproto->fallback_dpid = pick_fallback_dpid();
432 ofproto->mfr_desc = NULL;
433 ofproto->hw_desc = NULL;
434 ofproto->sw_desc = NULL;
435 ofproto->serial_desc = NULL;
436 ofproto->dp_desc = NULL;
437 ofproto->frag_handling = OFPC_FRAG_NORMAL;
438 hmap_init(&ofproto->ports);
439 shash_init(&ofproto->port_by_name);
440 simap_init(&ofproto->ofp_requests);
441 ofproto->max_ports = ofp_to_u16(OFPP_MAX);
442 ofproto->tables = NULL;
443 ofproto->n_tables = 0;
444 hindex_init(&ofproto->cookies);
445 list_init(&ofproto->expirable);
446 ovs_mutex_init_recursive(&ofproto->expirable_mutex);
447 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
448 ofproto->state = S_OPENFLOW;
449 list_init(&ofproto->pending);
450 ofproto->n_pending = 0;
451 hmap_init(&ofproto->deletions);
452 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
453 ofproto->first_op = ofproto->last_op = LLONG_MIN;
454 ofproto->next_op_report = LLONG_MAX;
455 ofproto->op_backoff = LLONG_MIN;
456 ofproto->vlan_bitmap = NULL;
457 ofproto->vlans_changed = false;
458 ofproto->min_mtu = INT_MAX;
460 error = ofproto->ofproto_class->construct(ofproto);
462 VLOG_ERR("failed to open datapath %s: %s",
463 datapath_name, ovs_strerror(error));
464 ofproto_destroy__(ofproto);
468 /* The "max_ports" member should have been set by ->construct(ofproto).
469 * Port 0 is not a valid OpenFlow port, so mark that as unavailable. */
470 ofproto->ofp_port_ids = bitmap_allocate(ofproto->max_ports);
471 bitmap_set1(ofproto->ofp_port_ids, 0);
473 /* Check that hidden tables, if any, are at the end. */
474 ovs_assert(ofproto->n_tables);
475 for (i = 0; i + 1 < ofproto->n_tables; i++) {
476 enum oftable_flags flags = ofproto->tables[i].flags;
477 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
479 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
482 ofproto->datapath_id = pick_datapath_id(ofproto);
485 /* Initialize meters table. */
486 if (ofproto->ofproto_class->meter_get_features) {
487 ofproto->ofproto_class->meter_get_features(ofproto,
488 &ofproto->meter_features);
490 memset(&ofproto->meter_features, 0, sizeof ofproto->meter_features);
492 ofproto->meters = xzalloc((ofproto->meter_features.max_meters + 1)
493 * sizeof(struct meter *));
499 /* Must be called (only) by an ofproto implementation in its constructor
500 * function. See the large comment on 'construct' in struct ofproto_class for
503 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
505 struct oftable *table;
507 ovs_assert(!ofproto->n_tables);
508 ovs_assert(n_tables >= 1 && n_tables <= 255);
510 ofproto->n_tables = n_tables;
511 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
512 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
517 /* To be optionally called (only) by an ofproto implementation in its
518 * constructor function. See the large comment on 'construct' in struct
519 * ofproto_class for details.
521 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
522 * will then ensure that actions passed into the ofproto implementation will
523 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
524 * function is not called, there will be no such restriction.
526 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
527 * the 'max_ports' restriction. */
529 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
531 ovs_assert(max_ports <= ofp_to_u16(OFPP_MAX));
532 ofproto->max_ports = max_ports;
536 ofproto_get_datapath_id(const struct ofproto *ofproto)
538 return ofproto->datapath_id;
542 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
544 uint64_t old_dpid = p->datapath_id;
545 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
546 if (p->datapath_id != old_dpid) {
547 /* Force all active connections to reconnect, since there is no way to
548 * notify a controller that the datapath ID has changed. */
549 ofproto_reconnect_controllers(p);
554 ofproto_set_controllers(struct ofproto *p,
555 const struct ofproto_controller *controllers,
556 size_t n_controllers, uint32_t allowed_versions)
558 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
563 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
565 connmgr_set_fail_mode(p->connmgr, fail_mode);
568 /* Drops the connections between 'ofproto' and all of its controllers, forcing
569 * them to reconnect. */
571 ofproto_reconnect_controllers(struct ofproto *ofproto)
573 connmgr_reconnect(ofproto->connmgr);
576 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
577 * in-band control should guarantee access, in the same way that in-band
578 * control guarantees access to OpenFlow controllers. */
580 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
581 const struct sockaddr_in *extras, size_t n)
583 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
586 /* Sets the OpenFlow queue used by flows set up by in-band control on
587 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
588 * flows will use the default queue. */
590 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
592 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
595 /* Sets the number of flows at which eviction from the kernel flow table
598 ofproto_set_flow_eviction_threshold(unsigned threshold)
600 flow_eviction_threshold = MAX(OFPROTO_FLOW_EVICTION_THRESHOLD_MIN,
604 /* Sets the path for handling flow misses. */
606 ofproto_set_flow_miss_model(unsigned model)
608 flow_miss_model = model;
611 /* If forward_bpdu is true, the NORMAL action will forward frames with
612 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
613 * the NORMAL action will drop these frames. */
615 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
617 bool old_val = ofproto->forward_bpdu;
618 ofproto->forward_bpdu = forward_bpdu;
619 if (old_val != ofproto->forward_bpdu) {
620 if (ofproto->ofproto_class->forward_bpdu_changed) {
621 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
626 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
627 * 'idle_time', in seconds, and the maximum number of MAC table entries to
630 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
633 if (ofproto->ofproto_class->set_mac_table_config) {
634 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
639 /* Sets number of upcall handler threads. The default is
640 * (number of online cores - 1). */
642 ofproto_set_n_handler_threads(unsigned limit)
645 n_handler_threads = limit;
647 n_handler_threads = MAX(1, sysconf(_SC_NPROCESSORS_ONLN) - 1);
652 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
655 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
659 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
661 return connmgr_set_snoops(ofproto->connmgr, snoops);
665 ofproto_set_netflow(struct ofproto *ofproto,
666 const struct netflow_options *nf_options)
668 if (nf_options && sset_is_empty(&nf_options->collectors)) {
672 if (ofproto->ofproto_class->set_netflow) {
673 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
675 return nf_options ? EOPNOTSUPP : 0;
680 ofproto_set_sflow(struct ofproto *ofproto,
681 const struct ofproto_sflow_options *oso)
683 if (oso && sset_is_empty(&oso->targets)) {
687 if (ofproto->ofproto_class->set_sflow) {
688 return ofproto->ofproto_class->set_sflow(ofproto, oso);
690 return oso ? EOPNOTSUPP : 0;
695 ofproto_set_ipfix(struct ofproto *ofproto,
696 const struct ofproto_ipfix_bridge_exporter_options *bo,
697 const struct ofproto_ipfix_flow_exporter_options *fo,
700 if (ofproto->ofproto_class->set_ipfix) {
701 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
703 return (bo || fo) ? EOPNOTSUPP : 0;
708 ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
710 flow_restore_wait = flow_restore_wait_db;
714 ofproto_get_flow_restore_wait(void)
716 return flow_restore_wait;
720 /* Spanning Tree Protocol (STP) configuration. */
722 /* Configures STP on 'ofproto' using the settings defined in 's'. If
723 * 's' is NULL, disables STP.
725 * Returns 0 if successful, otherwise a positive errno value. */
727 ofproto_set_stp(struct ofproto *ofproto,
728 const struct ofproto_stp_settings *s)
730 return (ofproto->ofproto_class->set_stp
731 ? ofproto->ofproto_class->set_stp(ofproto, s)
735 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
736 * 'enabled' member of 's' is false, then the other members are not
739 * Returns 0 if successful, otherwise a positive errno value. */
741 ofproto_get_stp_status(struct ofproto *ofproto,
742 struct ofproto_stp_status *s)
744 return (ofproto->ofproto_class->get_stp_status
745 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
749 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
750 * in 's'. The caller is responsible for assigning STP port numbers
751 * (using the 'port_num' member in the range of 1 through 255, inclusive)
752 * and ensuring there are no duplicates. If the 's' is NULL, then STP
753 * is disabled on the port.
755 * Returns 0 if successful, otherwise a positive errno value.*/
757 ofproto_port_set_stp(struct ofproto *ofproto, ofp_port_t ofp_port,
758 const struct ofproto_port_stp_settings *s)
760 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
762 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
763 ofproto->name, ofp_port);
767 return (ofproto->ofproto_class->set_stp_port
768 ? ofproto->ofproto_class->set_stp_port(ofport, s)
772 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
773 * 's'. If the 'enabled' member in 's' is false, then the other members
774 * are not meaningful.
776 * Returns 0 if successful, otherwise a positive errno value.*/
778 ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
779 struct ofproto_port_stp_status *s)
781 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
783 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
784 "port %"PRIu16, ofproto->name, ofp_port);
788 return (ofproto->ofproto_class->get_stp_port_status
789 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
793 /* Queue DSCP configuration. */
795 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
796 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
797 * to implement QoS. Instead, it is used to implement features which require
798 * knowledge of what queues exist on a port, and some basic information about
801 * Returns 0 if successful, otherwise a positive errno value. */
803 ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port,
804 const struct ofproto_port_queue *queues,
807 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
810 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
811 ofproto->name, ofp_port);
815 return (ofproto->ofproto_class->set_queues
816 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
820 /* Connectivity Fault Management configuration. */
822 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
824 ofproto_port_clear_cfm(struct ofproto *ofproto, ofp_port_t ofp_port)
826 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
827 if (ofport && ofproto->ofproto_class->set_cfm) {
828 ofproto->ofproto_class->set_cfm(ofport, NULL);
832 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
833 * basic configuration from the configuration members in 'cfm', and the remote
834 * maintenance point ID from remote_mpid. Ignores the statistics members of
837 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
839 ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port,
840 const struct cfm_settings *s)
842 struct ofport *ofport;
845 ofport = ofproto_get_port(ofproto, ofp_port);
847 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
848 ofproto->name, ofp_port);
852 /* XXX: For configuration simplicity, we only support one remote_mpid
853 * outside of the CFM module. It's not clear if this is the correct long
854 * term solution or not. */
855 error = (ofproto->ofproto_class->set_cfm
856 ? ofproto->ofproto_class->set_cfm(ofport, s)
859 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
860 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
861 ovs_strerror(error));
865 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
866 * 'ofproto' does not have a port 'ofp_port'. */
868 ofproto_port_set_bfd(struct ofproto *ofproto, ofp_port_t ofp_port,
869 const struct smap *cfg)
871 struct ofport *ofport;
874 ofport = ofproto_get_port(ofproto, ofp_port);
876 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
877 ofproto->name, ofp_port);
881 error = (ofproto->ofproto_class->set_bfd
882 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
885 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
886 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
887 ovs_strerror(error));
891 /* Populates 'status' with key value pairs indicating the status of the BFD
892 * session on 'ofp_port'. This information is intended to be populated in the
893 * OVS database. Has no effect if 'ofp_port' is not na OpenFlow port in
896 ofproto_port_get_bfd_status(struct ofproto *ofproto, ofp_port_t ofp_port,
899 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
900 return (ofport && ofproto->ofproto_class->get_bfd_status
901 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
905 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
906 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
907 * 0 if LACP partner information is not current (generally indicating a
908 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
910 ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port)
912 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
913 return (ofport && ofproto->ofproto_class->port_is_lacp_current
914 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
920 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
921 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
922 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
923 * configuration plus, if there is more than one slave, a bonding
926 * If 'aux' is already registered then this function updates its configuration
927 * to 's'. Otherwise, this function registers a new bundle.
929 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
932 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
933 const struct ofproto_bundle_settings *s)
935 return (ofproto->ofproto_class->bundle_set
936 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
940 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
941 * If no such bundle has been registered, this has no effect. */
943 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
945 return ofproto_bundle_register(ofproto, aux, NULL);
949 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
950 * If 'aux' is already registered then this function updates its configuration
951 * to 's'. Otherwise, this function registers a new mirror. */
953 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
954 const struct ofproto_mirror_settings *s)
956 return (ofproto->ofproto_class->mirror_set
957 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
961 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
962 * If no mirror has been registered, this has no effect. */
964 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
966 return ofproto_mirror_register(ofproto, aux, NULL);
969 /* Retrieves statistics from mirror associated with client data pointer
970 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
971 * 'bytes', respectively. If a particular counters is not supported,
972 * the appropriate argument is set to UINT64_MAX. */
974 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
975 uint64_t *packets, uint64_t *bytes)
977 if (!ofproto->ofproto_class->mirror_get_stats) {
978 *packets = *bytes = UINT64_MAX;
982 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
986 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
987 * which all packets are flooded, instead of using MAC learning. If
988 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
990 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
993 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
995 return (ofproto->ofproto_class->set_flood_vlans
996 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
1000 /* Returns true if 'aux' is a registered bundle that is currently in use as the
1001 * output for a mirror. */
1003 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
1005 return (ofproto->ofproto_class->is_mirror_output_bundle
1006 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
1010 /* Configuration of OpenFlow tables. */
1012 /* Returns the number of OpenFlow tables in 'ofproto'. */
1014 ofproto_get_n_tables(const struct ofproto *ofproto)
1016 return ofproto->n_tables;
1019 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
1020 * settings from 's'. 'table_id' must be in the range 0 through the number of
1021 * OpenFlow tables in 'ofproto' minus 1, inclusive.
1023 * For read-only tables, only the name may be configured. */
1025 ofproto_configure_table(struct ofproto *ofproto, int table_id,
1026 const struct ofproto_table_settings *s)
1028 struct oftable *table;
1030 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
1031 table = &ofproto->tables[table_id];
1033 oftable_set_name(table, s->name);
1035 if (table->flags & OFTABLE_READONLY) {
1040 oftable_enable_eviction(table, s->groups, s->n_groups);
1042 oftable_disable_eviction(table);
1045 table->max_flows = s->max_flows;
1046 ovs_rwlock_rdlock(&table->cls.rwlock);
1047 if (classifier_count(&table->cls) > table->max_flows
1048 && table->eviction_fields) {
1049 /* 'table' contains more flows than allowed. We might not be able to
1050 * evict them right away because of the asynchronous nature of flow
1051 * table changes. Schedule eviction for later. */
1052 switch (ofproto->state) {
1054 ofproto->state = S_EVICT;
1058 /* We're already deleting flows, nothing more to do. */
1062 ovs_rwlock_unlock(&table->cls.rwlock);
1066 ofproto_has_snoops(const struct ofproto *ofproto)
1068 return connmgr_has_snoops(ofproto->connmgr);
1072 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1074 connmgr_get_snoops(ofproto->connmgr, snoops);
1077 /* Deletes 'rule' from 'cls' within 'ofproto'.
1079 * The 'cls' argument is redundant (it is &ofproto->tables[rule->table_id].cls)
1080 * but it allows Clang to do better checking. */
1082 ofproto_delete_rule(struct ofproto *ofproto, struct classifier *cls,
1084 OVS_REQ_WRLOCK(cls->rwlock)
1086 struct ofopgroup *group;
1088 ovs_assert(!rule->pending);
1089 ovs_assert(cls == &ofproto->tables[rule->table_id].cls);
1091 group = ofopgroup_create_unattached(ofproto);
1092 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1093 ovs_rwlock_wrlock(&rule->evict);
1094 oftable_remove_rule__(ofproto, cls, rule);
1095 ofproto->ofproto_class->rule_delete(rule);
1096 ofopgroup_submit(group);
1100 ofproto_flush__(struct ofproto *ofproto)
1102 struct oftable *table;
1104 if (ofproto->ofproto_class->flush) {
1105 ofproto->ofproto_class->flush(ofproto);
1108 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1109 struct rule *rule, *next_rule;
1110 struct cls_cursor cursor;
1112 if (table->flags & OFTABLE_HIDDEN) {
1116 ovs_rwlock_wrlock(&table->cls.rwlock);
1117 cls_cursor_init(&cursor, &table->cls, NULL);
1118 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1119 if (!rule->pending) {
1120 ofproto_delete_rule(ofproto, &table->cls, rule);
1123 ovs_rwlock_unlock(&table->cls.rwlock);
1128 ofproto_destroy__(struct ofproto *ofproto)
1130 struct oftable *table;
1132 ovs_assert(list_is_empty(&ofproto->pending));
1133 ovs_assert(!ofproto->n_pending);
1135 if (ofproto->meters) {
1136 meter_delete(ofproto, 1, ofproto->meter_features.max_meters);
1137 free(ofproto->meters);
1140 connmgr_destroy(ofproto->connmgr);
1142 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1143 free(ofproto->name);
1144 free(ofproto->type);
1145 free(ofproto->mfr_desc);
1146 free(ofproto->hw_desc);
1147 free(ofproto->sw_desc);
1148 free(ofproto->serial_desc);
1149 free(ofproto->dp_desc);
1150 hmap_destroy(&ofproto->ports);
1151 shash_destroy(&ofproto->port_by_name);
1152 bitmap_free(ofproto->ofp_port_ids);
1153 simap_destroy(&ofproto->ofp_requests);
1155 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1156 oftable_destroy(table);
1158 free(ofproto->tables);
1160 hmap_destroy(&ofproto->deletions);
1162 free(ofproto->vlan_bitmap);
1164 ovs_mutex_destroy(&ofproto->expirable_mutex);
1165 ofproto->ofproto_class->dealloc(ofproto);
1169 ofproto_destroy(struct ofproto *p)
1171 struct ofport *ofport, *next_ofport;
1178 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1179 ofport_destroy(ofport);
1182 p->ofproto_class->destruct(p);
1183 ofproto_destroy__(p);
1186 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1187 * kernel datapath, for example, this destroys the datapath in the kernel, and
1188 * with the netdev-based datapath, it tears down the data structures that
1189 * represent the datapath.
1191 * The datapath should not be currently open as an ofproto. */
1193 ofproto_delete(const char *name, const char *type)
1195 const struct ofproto_class *class = ofproto_class_find__(type);
1196 return (!class ? EAFNOSUPPORT
1197 : !class->del ? EACCES
1198 : class->del(type, name));
1202 process_port_change(struct ofproto *ofproto, int error, char *devname)
1204 if (error == ENOBUFS) {
1205 reinit_ports(ofproto);
1206 } else if (!error) {
1207 update_port(ofproto, devname);
1213 ofproto_type_run(const char *datapath_type)
1215 const struct ofproto_class *class;
1218 datapath_type = ofproto_normalize_type(datapath_type);
1219 class = ofproto_class_find__(datapath_type);
1221 error = class->type_run ? class->type_run(datapath_type) : 0;
1222 if (error && error != EAGAIN) {
1223 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1224 datapath_type, ovs_strerror(error));
1230 ofproto_type_run_fast(const char *datapath_type)
1232 const struct ofproto_class *class;
1235 datapath_type = ofproto_normalize_type(datapath_type);
1236 class = ofproto_class_find__(datapath_type);
1238 error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
1239 if (error && error != EAGAIN) {
1240 VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
1241 datapath_type, ovs_strerror(error));
1247 ofproto_type_wait(const char *datapath_type)
1249 const struct ofproto_class *class;
1251 datapath_type = ofproto_normalize_type(datapath_type);
1252 class = ofproto_class_find__(datapath_type);
1254 if (class->type_wait) {
1255 class->type_wait(datapath_type);
1260 ofproto_run(struct ofproto *p)
1262 struct sset changed_netdevs;
1263 const char *changed_netdev;
1264 struct ofport *ofport;
1267 error = p->ofproto_class->run(p);
1268 if (error && error != EAGAIN) {
1269 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, ovs_strerror(error));
1272 if (p->ofproto_class->port_poll) {
1275 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1276 process_port_change(p, error, devname);
1280 /* Update OpenFlow port status for any port whose netdev has changed.
1282 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1283 * destroyed, so it's not safe to update ports directly from the
1284 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1285 * need this two-phase approach. */
1286 sset_init(&changed_netdevs);
1287 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1288 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1289 if (ofport->change_seq != change_seq) {
1290 ofport->change_seq = change_seq;
1291 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1294 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1295 update_port(p, changed_netdev);
1297 sset_destroy(&changed_netdevs);
1301 connmgr_run(p->connmgr, handle_openflow);
1305 connmgr_run(p->connmgr, NULL);
1307 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1308 p->state = S_OPENFLOW;
1313 connmgr_run(p->connmgr, NULL);
1315 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1316 connmgr_flushed(p->connmgr);
1317 p->state = S_OPENFLOW;
1325 if (time_msec() >= p->next_op_report) {
1326 long long int ago = (time_msec() - p->first_op) / 1000;
1327 long long int interval = (p->last_op - p->first_op) / 1000;
1331 ds_put_format(&s, "%d flow_mods ",
1332 p->n_add + p->n_delete + p->n_modify);
1333 if (interval == ago) {
1334 ds_put_format(&s, "in the last %lld s", ago);
1335 } else if (interval) {
1336 ds_put_format(&s, "in the %lld s starting %lld s ago",
1339 ds_put_format(&s, "%lld s ago", ago);
1342 ds_put_cstr(&s, " (");
1344 ds_put_format(&s, "%d adds, ", p->n_add);
1347 ds_put_format(&s, "%d deletes, ", p->n_delete);
1350 ds_put_format(&s, "%d modifications, ", p->n_modify);
1353 ds_put_char(&s, ')');
1355 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1358 p->n_add = p->n_delete = p->n_modify = 0;
1359 p->next_op_report = LLONG_MAX;
1365 /* Performs periodic activity required by 'ofproto' that needs to be done
1366 * with the least possible latency.
1368 * It makes sense to call this function a couple of times per poll loop, to
1369 * provide a significant performance boost on some benchmarks with the
1370 * ofproto-dpif implementation. */
1372 ofproto_run_fast(struct ofproto *p)
1376 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1377 if (error && error != EAGAIN) {
1378 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1379 p->name, ovs_strerror(error));
1385 ofproto_wait(struct ofproto *p)
1387 struct ofport *ofport;
1389 p->ofproto_class->wait(p);
1390 if (p->ofproto_class->port_poll_wait) {
1391 p->ofproto_class->port_poll_wait(p);
1394 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1395 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1396 poll_immediate_wake();
1402 connmgr_wait(p->connmgr, true);
1407 connmgr_wait(p->connmgr, false);
1408 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1409 poll_immediate_wake();
1416 ofproto_is_alive(const struct ofproto *p)
1418 return connmgr_has_controllers(p->connmgr);
1421 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1422 * memory_report(). */
1424 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1426 const struct oftable *table;
1427 unsigned int n_rules;
1429 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1430 simap_increase(usage, "ops",
1431 ofproto->n_pending + hmap_count(&ofproto->deletions));
1434 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1435 ovs_rwlock_rdlock(&table->cls.rwlock);
1436 n_rules += classifier_count(&table->cls);
1437 ovs_rwlock_unlock(&table->cls.rwlock);
1439 simap_increase(usage, "rules", n_rules);
1441 if (ofproto->ofproto_class->get_memory_usage) {
1442 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1445 connmgr_get_memory_usage(ofproto->connmgr, usage);
1449 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1452 connmgr_get_controller_info(ofproto->connmgr, info);
1456 ofproto_free_ofproto_controller_info(struct shash *info)
1458 connmgr_free_controller_info(info);
1461 /* Makes a deep copy of 'old' into 'port'. */
1463 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1465 port->name = xstrdup(old->name);
1466 port->type = xstrdup(old->type);
1467 port->ofp_port = old->ofp_port;
1470 /* Frees memory allocated to members of 'ofproto_port'.
1472 * Do not call this function on an ofproto_port obtained from
1473 * ofproto_port_dump_next(): that function retains ownership of the data in the
1476 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1478 free(ofproto_port->name);
1479 free(ofproto_port->type);
1482 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1484 * This function provides no status indication. An error status for the entire
1485 * dump operation is provided when it is completed by calling
1486 * ofproto_port_dump_done().
1489 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1490 const struct ofproto *ofproto)
1492 dump->ofproto = ofproto;
1493 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1497 /* Attempts to retrieve another port from 'dump', which must have been created
1498 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1499 * 'port' and returns true. On failure, returns false.
1501 * Failure might indicate an actual error or merely that the last port has been
1502 * dumped. An error status for the entire dump operation is provided when it
1503 * is completed by calling ofproto_port_dump_done().
1505 * The ofproto owns the data stored in 'port'. It will remain valid until at
1506 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1507 * ofproto_port_dump_done(). */
1509 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1510 struct ofproto_port *port)
1512 const struct ofproto *ofproto = dump->ofproto;
1518 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1521 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1527 /* Completes port table dump operation 'dump', which must have been created
1528 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1529 * error-free, otherwise a positive errno value describing the problem. */
1531 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1533 const struct ofproto *ofproto = dump->ofproto;
1535 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1538 return dump->error == EOF ? 0 : dump->error;
1541 /* Returns the type to pass to netdev_open() when a datapath of type
1542 * 'datapath_type' has a port of type 'port_type', for a few special
1543 * cases when a netdev type differs from a port type. For example, when
1544 * using the userspace datapath, a port of type "internal" needs to be
1547 * Returns either 'type' itself or a string literal, which must not be
1550 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1552 const struct ofproto_class *class;
1554 datapath_type = ofproto_normalize_type(datapath_type);
1555 class = ofproto_class_find__(datapath_type);
1560 return (class->port_open_type
1561 ? class->port_open_type(datapath_type, port_type)
1565 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1566 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1567 * the port's OpenFlow port number.
1569 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1570 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1571 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1572 * 'ofp_portp' is non-null). */
1574 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1575 ofp_port_t *ofp_portp)
1577 ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1580 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1582 const char *netdev_name = netdev_get_name(netdev);
1584 simap_put(&ofproto->ofp_requests, netdev_name,
1585 ofp_to_u16(ofp_port));
1586 update_port(ofproto, netdev_name);
1589 struct ofproto_port ofproto_port;
1591 ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
1593 *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
1594 ofproto_port_destroy(&ofproto_port);
1599 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1600 * initializes '*port' appropriately; on failure, returns a positive errno
1603 * The caller owns the data in 'ofproto_port' and must free it with
1604 * ofproto_port_destroy() when it is no longer needed. */
1606 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1607 struct ofproto_port *port)
1611 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1613 memset(port, 0, sizeof *port);
1618 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1619 * Returns 0 if successful, otherwise a positive errno. */
1621 ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port)
1623 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1624 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1625 struct simap_node *ofp_request_node;
1628 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1629 if (ofp_request_node) {
1630 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1633 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1634 if (!error && ofport) {
1635 /* 'name' is the netdev's name and update_port() is going to close the
1636 * netdev. Just in case update_port() refers to 'name' after it
1637 * destroys 'ofport', make a copy of it around the update_port()
1639 char *devname = xstrdup(name);
1640 update_port(ofproto, devname);
1646 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1647 * performs the 'n_actions' actions in 'actions'. The new flow will not
1650 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1651 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1652 * controllers; otherwise, it will be hidden.
1654 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1656 * This is a helper function for in-band control and fail-open. */
1658 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1659 unsigned int priority,
1660 const struct ofpact *ofpacts, size_t ofpacts_len)
1662 const struct rule *rule;
1664 ovs_rwlock_rdlock(&ofproto->tables[0].cls.rwlock);
1665 rule = rule_from_cls_rule(classifier_find_match_exactly(
1666 &ofproto->tables[0].cls, match, priority));
1667 ovs_rwlock_unlock(&ofproto->tables[0].cls.rwlock);
1668 if (!rule || !ofpacts_equal(rule->ofpacts, rule->ofpacts_len,
1669 ofpacts, ofpacts_len)) {
1670 struct ofputil_flow_mod fm;
1672 memset(&fm, 0, sizeof fm);
1674 fm.priority = priority;
1675 fm.buffer_id = UINT32_MAX;
1676 fm.ofpacts = xmemdup(ofpacts, ofpacts_len);
1677 fm.ofpacts_len = ofpacts_len;
1678 add_flow(ofproto, NULL, &fm, NULL);
1683 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1684 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1685 * operation cannot be initiated now but may be retried later.
1687 * This is a helper function for in-band control and fail-open. */
1689 ofproto_flow_mod(struct ofproto *ofproto, struct ofputil_flow_mod *fm)
1691 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1694 /* Searches for a rule with matching criteria exactly equal to 'target' in
1695 * ofproto's table 0 and, if it finds one, deletes it.
1697 * This is a helper function for in-band control and fail-open. */
1699 ofproto_delete_flow(struct ofproto *ofproto,
1700 const struct match *target, unsigned int priority)
1702 struct classifier *cls = &ofproto->tables[0].cls;
1705 ovs_rwlock_rdlock(&cls->rwlock);
1706 rule = rule_from_cls_rule(classifier_find_match_exactly(cls, target,
1708 ovs_rwlock_unlock(&cls->rwlock);
1710 /* No such rule -> success. */
1712 } else if (rule->pending) {
1713 /* An operation on the rule is already pending -> failure.
1714 * Caller must retry later if it's important. */
1717 /* Initiate deletion -> success. */
1718 ovs_rwlock_wrlock(&cls->rwlock);
1719 ofproto_delete_rule(ofproto, cls, rule);
1720 ovs_rwlock_unlock(&cls->rwlock);
1727 /* Starts the process of deleting all of the flows from all of ofproto's flow
1728 * tables and then reintroducing the flows required by in-band control and
1729 * fail-open. The process will complete in a later call to ofproto_run(). */
1731 ofproto_flush_flows(struct ofproto *ofproto)
1733 COVERAGE_INC(ofproto_flush);
1734 ofproto->state = S_FLUSH;
1738 reinit_ports(struct ofproto *p)
1740 struct ofproto_port_dump dump;
1741 struct sset devnames;
1742 struct ofport *ofport;
1743 struct ofproto_port ofproto_port;
1744 const char *devname;
1746 COVERAGE_INC(ofproto_reinit_ports);
1748 sset_init(&devnames);
1749 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1750 sset_add(&devnames, netdev_get_name(ofport->netdev));
1752 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1753 sset_add(&devnames, ofproto_port.name);
1756 SSET_FOR_EACH (devname, &devnames) {
1757 update_port(p, devname);
1759 sset_destroy(&devnames);
1763 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
1767 port_idx = simap_get(&ofproto->ofp_requests, netdev_name);
1768 port_idx = port_idx ? port_idx : UINT16_MAX;
1770 if (port_idx >= ofproto->max_ports
1771 || bitmap_is_set(ofproto->ofp_port_ids, port_idx)) {
1772 uint16_t end_port_no = ofproto->alloc_port_no;
1774 /* Search for a free OpenFlow port number. We try not to
1775 * immediately reuse them to prevent problems due to old
1778 if (++ofproto->alloc_port_no >= ofproto->max_ports) {
1779 ofproto->alloc_port_no = 0;
1781 if (!bitmap_is_set(ofproto->ofp_port_ids,
1782 ofproto->alloc_port_no)) {
1783 port_idx = ofproto->alloc_port_no;
1786 if (ofproto->alloc_port_no == end_port_no) {
1791 bitmap_set1(ofproto->ofp_port_ids, port_idx);
1792 return u16_to_ofp(port_idx);
1796 dealloc_ofp_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
1798 if (ofp_to_u16(ofp_port) < ofproto->max_ports) {
1799 bitmap_set0(ofproto->ofp_port_ids, ofp_to_u16(ofp_port));
1803 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1804 * pointer if the netdev cannot be opened. On success, also fills in
1806 static struct netdev *
1807 ofport_open(struct ofproto *ofproto,
1808 struct ofproto_port *ofproto_port,
1809 struct ofputil_phy_port *pp)
1811 enum netdev_flags flags;
1812 struct netdev *netdev;
1815 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1817 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
1818 "cannot be opened (%s)",
1820 ofproto_port->name, ofproto_port->ofp_port,
1821 ofproto_port->name, ovs_strerror(error));
1825 if (ofproto_port->ofp_port == OFPP_NONE) {
1826 if (!strcmp(ofproto->name, ofproto_port->name)) {
1827 ofproto_port->ofp_port = OFPP_LOCAL;
1829 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
1830 ofproto_port->name);
1833 pp->port_no = ofproto_port->ofp_port;
1834 netdev_get_etheraddr(netdev, pp->hw_addr);
1835 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
1836 netdev_get_flags(netdev, &flags);
1837 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
1838 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
1839 netdev_get_features(netdev, &pp->curr, &pp->advertised,
1840 &pp->supported, &pp->peer);
1841 pp->curr_speed = netdev_features_to_bps(pp->curr, 0);
1842 pp->max_speed = netdev_features_to_bps(pp->supported, 0);
1847 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1848 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1851 ofport_equal(const struct ofputil_phy_port *a,
1852 const struct ofputil_phy_port *b)
1854 return (eth_addr_equals(a->hw_addr, b->hw_addr)
1855 && a->state == b->state
1856 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
1857 && a->curr == b->curr
1858 && a->advertised == b->advertised
1859 && a->supported == b->supported
1860 && a->peer == b->peer
1861 && a->curr_speed == b->curr_speed
1862 && a->max_speed == b->max_speed);
1865 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1866 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1867 * one with the same name or port number). */
1869 ofport_install(struct ofproto *p,
1870 struct netdev *netdev, const struct ofputil_phy_port *pp)
1872 const char *netdev_name = netdev_get_name(netdev);
1873 struct ofport *ofport;
1876 /* Create ofport. */
1877 ofport = p->ofproto_class->port_alloc();
1882 ofport->ofproto = p;
1883 ofport->netdev = netdev;
1884 ofport->change_seq = netdev_change_seq(netdev);
1886 ofport->ofp_port = pp->port_no;
1887 ofport->created = time_msec();
1889 /* Add port to 'p'. */
1890 hmap_insert(&p->ports, &ofport->hmap_node,
1891 hash_ofp_port(ofport->ofp_port));
1892 shash_add(&p->port_by_name, netdev_name, ofport);
1894 update_mtu(p, ofport);
1896 /* Let the ofproto_class initialize its private data. */
1897 error = p->ofproto_class->port_construct(ofport);
1901 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
1905 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1906 p->name, netdev_name, ovs_strerror(error));
1908 ofport_destroy__(ofport);
1910 netdev_close(netdev);
1914 /* Removes 'ofport' from 'p' and destroys it. */
1916 ofport_remove(struct ofport *ofport)
1918 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
1920 ofport_destroy(ofport);
1923 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1926 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1928 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1930 ofport_remove(port);
1934 /* Updates 'port' with new 'pp' description.
1936 * Does not handle a name or port number change. The caller must implement
1937 * such a change as a delete followed by an add. */
1939 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
1941 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
1942 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
1943 | (pp->config & OFPUTIL_PC_PORT_DOWN));
1944 port->pp.state = pp->state;
1945 port->pp.curr = pp->curr;
1946 port->pp.advertised = pp->advertised;
1947 port->pp.supported = pp->supported;
1948 port->pp.peer = pp->peer;
1949 port->pp.curr_speed = pp->curr_speed;
1950 port->pp.max_speed = pp->max_speed;
1952 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
1955 /* Update OpenFlow 'state' in 'port' and notify controller. */
1957 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
1959 if (port->pp.state != state) {
1960 port->pp.state = state;
1961 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
1967 ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port)
1969 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1971 if (port->ofproto->ofproto_class->set_realdev) {
1972 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
1974 if (port->ofproto->ofproto_class->set_stp_port) {
1975 port->ofproto->ofproto_class->set_stp_port(port, NULL);
1977 if (port->ofproto->ofproto_class->set_cfm) {
1978 port->ofproto->ofproto_class->set_cfm(port, NULL);
1980 if (port->ofproto->ofproto_class->bundle_remove) {
1981 port->ofproto->ofproto_class->bundle_remove(port);
1987 ofport_destroy__(struct ofport *port)
1989 struct ofproto *ofproto = port->ofproto;
1990 const char *name = netdev_get_name(port->netdev);
1992 hmap_remove(&ofproto->ports, &port->hmap_node);
1993 shash_delete(&ofproto->port_by_name,
1994 shash_find(&ofproto->port_by_name, name));
1996 netdev_close(port->netdev);
1997 ofproto->ofproto_class->port_dealloc(port);
2001 ofport_destroy(struct ofport *port)
2004 dealloc_ofp_port(port->ofproto, port->ofp_port);
2005 port->ofproto->ofproto_class->port_destruct(port);
2006 ofport_destroy__(port);
2011 ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
2013 struct ofport *port;
2015 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_ofp_port(ofp_port),
2017 if (port->ofp_port == ofp_port) {
2025 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
2027 struct ofproto *ofproto = port->ofproto;
2030 if (ofproto->ofproto_class->port_get_stats) {
2031 error = ofproto->ofproto_class->port_get_stats(port, stats);
2040 update_port(struct ofproto *ofproto, const char *name)
2042 struct ofproto_port ofproto_port;
2043 struct ofputil_phy_port pp;
2044 struct netdev *netdev;
2045 struct ofport *port;
2047 COVERAGE_INC(ofproto_update_port);
2049 /* Fetch 'name''s location and properties from the datapath. */
2050 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
2051 ? ofport_open(ofproto, &ofproto_port, &pp)
2055 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
2056 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
2057 struct netdev *old_netdev = port->netdev;
2059 /* 'name' hasn't changed location. Any properties changed? */
2060 if (!ofport_equal(&port->pp, &pp)) {
2061 ofport_modified(port, &pp);
2064 update_mtu(ofproto, port);
2066 /* Install the newly opened netdev in case it has changed.
2067 * Don't close the old netdev yet in case port_modified has to
2068 * remove a retained reference to it.*/
2069 port->netdev = netdev;
2070 port->change_seq = netdev_change_seq(netdev);
2072 if (port->ofproto->ofproto_class->port_modified) {
2073 port->ofproto->ofproto_class->port_modified(port);
2076 netdev_close(old_netdev);
2078 /* If 'port' is nonnull then its name differs from 'name' and thus
2079 * we should delete it. If we think there's a port named 'name'
2080 * then its port number must be wrong now so delete it too. */
2082 ofport_remove(port);
2084 ofport_remove_with_name(ofproto, name);
2085 ofport_install(ofproto, netdev, &pp);
2088 /* Any port named 'name' is gone now. */
2089 ofport_remove_with_name(ofproto, name);
2091 ofproto_port_destroy(&ofproto_port);
2095 init_ports(struct ofproto *p)
2097 struct ofproto_port_dump dump;
2098 struct ofproto_port ofproto_port;
2099 struct shash_node *node, *next;
2101 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2102 const char *name = ofproto_port.name;
2104 if (shash_find(&p->port_by_name, name)) {
2105 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2108 struct ofputil_phy_port pp;
2109 struct netdev *netdev;
2111 /* Check if an OpenFlow port number had been requested. */
2112 node = shash_find(&init_ofp_ports, name);
2114 const struct iface_hint *iface_hint = node->data;
2115 simap_put(&p->ofp_requests, name,
2116 ofp_to_u16(iface_hint->ofp_port));
2119 netdev = ofport_open(p, &ofproto_port, &pp);
2121 ofport_install(p, netdev, &pp);
2122 if (ofp_to_u16(ofproto_port.ofp_port) < p->max_ports) {
2123 p->alloc_port_no = MAX(p->alloc_port_no,
2124 ofp_to_u16(ofproto_port.ofp_port));
2130 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2131 struct iface_hint *iface_hint = node->data;
2133 if (!strcmp(iface_hint->br_name, p->name)) {
2134 free(iface_hint->br_name);
2135 free(iface_hint->br_type);
2137 shash_delete(&init_ofp_ports, node);
2144 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2145 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2147 find_min_mtu(struct ofproto *p)
2149 struct ofport *ofport;
2152 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2153 struct netdev *netdev = ofport->netdev;
2156 /* Skip any internal ports, since that's what we're trying to
2158 if (!strcmp(netdev_get_type(netdev), "internal")) {
2162 if (netdev_get_mtu(netdev, &dev_mtu)) {
2165 if (!mtu || dev_mtu < mtu) {
2170 return mtu ? mtu: ETH_PAYLOAD_MAX;
2173 /* Update MTU of all datapath devices on 'p' to the minimum of the
2174 * non-datapath ports in event of 'port' added or changed. */
2176 update_mtu(struct ofproto *p, struct ofport *port)
2178 struct ofport *ofport;
2179 struct netdev *netdev = port->netdev;
2180 int dev_mtu, old_min;
2182 if (netdev_get_mtu(netdev, &dev_mtu)) {
2186 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2187 if (dev_mtu > p->min_mtu) {
2188 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2189 dev_mtu = p->min_mtu;
2192 port->mtu = dev_mtu;
2196 /* For non-internal port find new min mtu. */
2197 old_min = p->min_mtu;
2198 port->mtu = dev_mtu;
2199 p->min_mtu = find_min_mtu(p);
2200 if (p->min_mtu == old_min) {
2204 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2205 struct netdev *netdev = ofport->netdev;
2207 if (!strcmp(netdev_get_type(netdev), "internal")) {
2208 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2209 ofport->mtu = p->min_mtu;
2216 ofproto_rule_destroy__(struct rule *rule)
2219 rule->ofproto->ofproto_class->rule_destruct(rule);
2220 cls_rule_destroy(&rule->cr);
2221 free(rule->ofpacts);
2222 ovs_mutex_destroy(&rule->timeout_mutex);
2223 ovs_rwlock_destroy(&rule->evict);
2224 rule->ofproto->ofproto_class->rule_dealloc(rule);
2228 /* This function allows an ofproto implementation to destroy any rules that
2229 * remain when its ->destruct() function is called.. This function implements
2230 * steps 4.4 and 4.5 in the section titled "Rule Life Cycle" in
2231 * ofproto-provider.h.
2233 * This function should only be called from an ofproto implementation's
2234 * ->destruct() function. It is not suitable elsewhere. */
2236 ofproto_rule_delete(struct ofproto *ofproto, struct classifier *cls,
2238 OVS_REQ_WRLOCK(cls->rwlock)
2240 ofproto_delete_rule(ofproto, cls, rule);
2243 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2244 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2246 ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port)
2248 return (port == OFPP_ANY
2249 || ofpacts_output_to_port(rule->ofpacts, rule->ofpacts_len, port));
2252 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2253 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2255 ofoperation_has_out_port(const struct ofoperation *op, ofp_port_t out_port)
2257 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2262 case OFOPERATION_ADD:
2263 case OFOPERATION_DELETE:
2266 case OFOPERATION_MODIFY:
2267 case OFOPERATION_REPLACE:
2268 return ofpacts_output_to_port(op->ofpacts, op->ofpacts_len, out_port);
2274 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2275 * statistics appropriately.
2277 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2278 * with statistics for 'packet' either way.
2280 * Takes ownership of 'packet'. */
2282 rule_execute(struct rule *rule, ofp_port_t in_port, struct ofpbuf *packet)
2285 union flow_in_port in_port_;
2287 in_port_.ofp_port = in_port;
2288 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
2289 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
2292 /* Returns true if 'rule' should be hidden from the controller.
2294 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2295 * (e.g. by in-band control) and are intentionally hidden from the
2298 ofproto_rule_is_hidden(const struct rule *rule)
2300 return rule->cr.priority > UINT16_MAX;
2303 static enum oftable_flags
2304 rule_get_flags(const struct rule *rule)
2306 return rule->ofproto->tables[rule->table_id].flags;
2310 rule_is_modifiable(const struct rule *rule)
2312 return !(rule_get_flags(rule) & OFTABLE_READONLY);
2316 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2318 ofconn_send_reply(ofconn, make_echo_reply(oh));
2323 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2325 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2326 struct ofputil_switch_features features;
2327 struct ofport *port;
2333 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2335 ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2337 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2338 * if present, are always at the end.) */
2339 n_tables = ofproto->n_tables;
2340 for (i = 0; i < ofproto->n_tables; i++) {
2341 if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
2347 features.datapath_id = ofproto->datapath_id;
2348 features.n_buffers = pktbuf_capacity();
2349 features.n_tables = n_tables;
2350 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2351 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2353 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2355 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2356 features.auxiliary_id = 0;
2357 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2359 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2360 ofputil_put_switch_features_port(&port->pp, b);
2363 ofconn_send_reply(ofconn, b);
2368 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2370 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2371 struct ofp_switch_config *osc;
2372 enum ofp_config_flags flags;
2376 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2377 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2378 flags = ofproto->frag_handling;
2379 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2380 if (oh->version < OFP13_VERSION
2381 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2382 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2384 osc->flags = htons(flags);
2385 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2386 ofconn_send_reply(ofconn, buf);
2392 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2394 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2395 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2396 uint16_t flags = ntohs(osc->flags);
2398 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2399 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
2400 enum ofp_config_flags cur = ofproto->frag_handling;
2401 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2403 ovs_assert((cur & OFPC_FRAG_MASK) == cur);
2405 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2406 ofproto->frag_handling = next;
2408 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2410 ofputil_frag_handling_to_string(next));
2414 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2415 ofconn_set_invalid_ttl_to_controller(ofconn,
2416 (oh->version < OFP13_VERSION
2417 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2419 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2424 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2425 * error message code for the caller to propagate upward. Otherwise, returns
2428 * The log message mentions 'msg_type'. */
2430 reject_slave_controller(struct ofconn *ofconn)
2432 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2433 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
2434 return OFPERR_OFPBRC_EPERM;
2440 /* Finds the OFPACT_METER action, if any, in the 'ofpacts_len' bytes of
2441 * 'ofpacts'. If found, returns its meter ID; if not, returns 0.
2443 * This function relies on the order of 'ofpacts' being correct (as checked by
2444 * ofpacts_verify()). */
2446 find_meter(const struct ofpact ofpacts[], size_t ofpacts_len)
2448 const struct ofpact *a;
2450 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2451 enum ovs_instruction_type inst;
2453 inst = ovs_instruction_type_from_ofpact_type(a->type);
2454 if (a->type == OFPACT_METER) {
2455 return ofpact_get_METER(a)->meter_id;
2456 } else if (inst > OVSINST_OFPIT13_METER) {
2464 /* Checks that the 'ofpacts_len' bytes of actions in 'ofpacts' are appropriate
2465 * for a packet with the prerequisites satisfied by 'flow' in table 'table_id'.
2466 * 'flow' may be temporarily modified, but is restored at return.
2469 ofproto_check_ofpacts(struct ofproto *ofproto,
2470 const struct ofpact ofpacts[], size_t ofpacts_len,
2471 struct flow *flow, uint8_t table_id)
2476 error = ofpacts_check(ofpacts, ofpacts_len, flow,
2477 u16_to_ofp(ofproto->max_ports), table_id);
2482 mid = find_meter(ofpacts, ofpacts_len);
2483 if (mid && ofproto_get_provider_meter_id(ofproto, mid) == UINT32_MAX) {
2484 return OFPERR_OFPMMFC_INVALID_METER;
2490 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2492 struct ofproto *p = ofconn_get_ofproto(ofconn);
2493 struct ofputil_packet_out po;
2494 struct ofpbuf *payload;
2495 uint64_t ofpacts_stub[1024 / 8];
2496 struct ofpbuf ofpacts;
2498 union flow_in_port in_port_;
2501 COVERAGE_INC(ofproto_packet_out);
2503 error = reject_slave_controller(ofconn);
2508 /* Decode message. */
2509 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2510 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2512 goto exit_free_ofpacts;
2514 if (ofp_to_u16(po.in_port) >= p->max_ports
2515 && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) {
2516 error = OFPERR_OFPBRC_BAD_PORT;
2517 goto exit_free_ofpacts;
2522 if (po.buffer_id != UINT32_MAX) {
2523 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2524 if (error || !payload) {
2525 goto exit_free_ofpacts;
2528 /* Ensure that the L3 header is 32-bit aligned. */
2529 payload = ofpbuf_clone_data_with_headroom(po.packet, po.packet_len, 2);
2532 /* Verify actions against packet, then send packet if successful. */
2533 in_port_.ofp_port = po.in_port;
2534 flow_extract(payload, 0, 0, NULL, &in_port_, &flow);
2535 error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len, &flow, 0);
2537 error = p->ofproto_class->packet_out(p, payload, &flow,
2538 po.ofpacts, po.ofpacts_len);
2540 ofpbuf_delete(payload);
2543 ofpbuf_uninit(&ofpacts);
2549 update_port_config(struct ofport *port,
2550 enum ofputil_port_config config,
2551 enum ofputil_port_config mask)
2553 enum ofputil_port_config old_config = port->pp.config;
2554 enum ofputil_port_config toggle;
2556 toggle = (config ^ port->pp.config) & mask;
2557 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2558 if (config & OFPUTIL_PC_PORT_DOWN) {
2559 netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL);
2561 netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL);
2563 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2566 port->pp.config ^= toggle;
2567 if (port->pp.config != old_config) {
2568 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2573 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2575 struct ofproto *p = ofconn_get_ofproto(ofconn);
2576 struct ofputil_port_mod pm;
2577 struct ofport *port;
2580 error = reject_slave_controller(ofconn);
2585 error = ofputil_decode_port_mod(oh, &pm);
2590 port = ofproto_get_port(p, pm.port_no);
2592 return OFPERR_OFPPMFC_BAD_PORT;
2593 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2594 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2596 update_port_config(port, pm.config, pm.mask);
2598 netdev_set_advertisements(port->netdev, pm.advertise);
2605 handle_desc_stats_request(struct ofconn *ofconn,
2606 const struct ofp_header *request)
2608 static const char *default_mfr_desc = "Nicira, Inc.";
2609 static const char *default_hw_desc = "Open vSwitch";
2610 static const char *default_sw_desc = VERSION;
2611 static const char *default_serial_desc = "None";
2612 static const char *default_dp_desc = "None";
2614 struct ofproto *p = ofconn_get_ofproto(ofconn);
2615 struct ofp_desc_stats *ods;
2618 msg = ofpraw_alloc_stats_reply(request, 0);
2619 ods = ofpbuf_put_zeros(msg, sizeof *ods);
2620 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
2621 sizeof ods->mfr_desc);
2622 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
2623 sizeof ods->hw_desc);
2624 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
2625 sizeof ods->sw_desc);
2626 ovs_strlcpy(ods->serial_num,
2627 p->serial_desc ? p->serial_desc : default_serial_desc,
2628 sizeof ods->serial_num);
2629 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
2630 sizeof ods->dp_desc);
2631 ofconn_send_reply(ofconn, msg);
2637 handle_table_stats_request(struct ofconn *ofconn,
2638 const struct ofp_header *request)
2640 struct ofproto *p = ofconn_get_ofproto(ofconn);
2641 struct ofp12_table_stats *ots;
2646 /* Set up default values.
2648 * ofp12_table_stats is used as a generic structure as
2649 * it is able to hold all the fields for ofp10_table_stats
2650 * and ofp11_table_stats (and of course itself).
2652 ots = xcalloc(p->n_tables, sizeof *ots);
2653 for (i = 0; i < p->n_tables; i++) {
2654 ots[i].table_id = i;
2655 sprintf(ots[i].name, "table%zu", i);
2656 ots[i].match = htonll(OFPXMT12_MASK);
2657 ots[i].wildcards = htonll(OFPXMT12_MASK);
2658 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
2659 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
2660 ots[i].write_setfields = htonll(OFPXMT12_MASK);
2661 ots[i].apply_setfields = htonll(OFPXMT12_MASK);
2662 ots[i].metadata_match = htonll(UINT64_MAX);
2663 ots[i].metadata_write = htonll(UINT64_MAX);
2664 ots[i].instructions = htonl(OFPIT11_ALL);
2665 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
2666 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2667 ovs_rwlock_rdlock(&p->tables[i].cls.rwlock);
2668 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2669 ovs_rwlock_unlock(&p->tables[i].cls.rwlock);
2672 p->ofproto_class->get_tables(p, ots);
2674 /* Post-process the tables, dropping hidden tables. */
2675 n_tables = p->n_tables;
2676 for (i = 0; i < p->n_tables; i++) {
2677 const struct oftable *table = &p->tables[i];
2679 if (table->flags & OFTABLE_HIDDEN) {
2685 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2688 if (table->max_flows < ntohl(ots[i].max_entries)) {
2689 ots[i].max_entries = htonl(table->max_flows);
2693 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
2694 ofconn_send_reply(ofconn, msg);
2702 append_port_stat(struct ofport *port, struct list *replies)
2704 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
2706 calc_duration(port->created, time_msec(),
2707 &ops.duration_sec, &ops.duration_nsec);
2709 /* Intentionally ignore return value, since errors will set
2710 * 'stats' to all-1s, which is correct for OpenFlow, and
2711 * netdev_get_stats() will log errors. */
2712 ofproto_port_get_stats(port, &ops.stats);
2714 ofputil_append_port_stat(replies, &ops);
2718 handle_port_stats_request(struct ofconn *ofconn,
2719 const struct ofp_header *request)
2721 struct ofproto *p = ofconn_get_ofproto(ofconn);
2722 struct ofport *port;
2723 struct list replies;
2727 error = ofputil_decode_port_stats_request(request, &port_no);
2732 ofpmp_init(&replies, request);
2733 if (port_no != OFPP_ANY) {
2734 port = ofproto_get_port(p, port_no);
2736 append_port_stat(port, &replies);
2739 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2740 append_port_stat(port, &replies);
2744 ofconn_send_replies(ofconn, &replies);
2749 handle_port_desc_stats_request(struct ofconn *ofconn,
2750 const struct ofp_header *request)
2752 struct ofproto *p = ofconn_get_ofproto(ofconn);
2753 enum ofp_version version;
2754 struct ofport *port;
2755 struct list replies;
2757 ofpmp_init(&replies, request);
2759 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
2760 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2761 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
2764 ofconn_send_replies(ofconn, &replies);
2769 hash_cookie(ovs_be64 cookie)
2771 return hash_2words((OVS_FORCE uint64_t)cookie >> 32,
2772 (OVS_FORCE uint64_t)cookie);
2776 cookies_insert(struct ofproto *ofproto, struct rule *rule)
2778 hindex_insert(&ofproto->cookies, &rule->cookie_node,
2779 hash_cookie(rule->flow_cookie));
2783 cookies_remove(struct ofproto *ofproto, struct rule *rule)
2785 hindex_remove(&ofproto->cookies, &rule->cookie_node);
2789 ofproto_rule_change_cookie(struct ofproto *ofproto, struct rule *rule,
2790 ovs_be64 new_cookie)
2792 if (new_cookie != rule->flow_cookie) {
2793 cookies_remove(ofproto, rule);
2795 rule->flow_cookie = new_cookie;
2797 cookies_insert(ofproto, rule);
2802 calc_duration(long long int start, long long int now,
2803 uint32_t *sec, uint32_t *nsec)
2805 long long int msecs = now - start;
2806 *sec = msecs / 1000;
2807 *nsec = (msecs % 1000) * (1000 * 1000);
2810 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2811 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2813 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
2815 return (table_id == 0xff || table_id < ofproto->n_tables
2817 : OFPERR_OFPBRC_BAD_TABLE_ID);
2821 static struct oftable *
2822 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
2824 struct oftable *table;
2826 for (table = &ofproto->tables[table_id];
2827 table < &ofproto->tables[ofproto->n_tables];
2829 if (!(table->flags & OFTABLE_HIDDEN)) {
2837 static struct oftable *
2838 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
2840 if (table_id == 0xff) {
2841 return next_visible_table(ofproto, 0);
2842 } else if (table_id < ofproto->n_tables) {
2843 return &ofproto->tables[table_id];
2849 static struct oftable *
2850 next_matching_table(const struct ofproto *ofproto,
2851 const struct oftable *table, uint8_t table_id)
2853 return (table_id == 0xff
2854 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
2858 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2860 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2861 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2863 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2864 * only once, for that table. (This can be used to access tables marked
2867 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2868 * entered at all. (Perhaps you should have validated TABLE_ID with
2869 * check_table_id().)
2871 * All parameters are evaluated multiple times.
2873 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2874 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2876 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2878 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2879 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2880 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2883 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2884 * to 'out_port' are included.
2886 * Hidden rules are always omitted.
2888 * Returns 0 on success, otherwise an OpenFlow error code. */
2890 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
2891 const struct match *match,
2892 ovs_be64 cookie, ovs_be64 cookie_mask,
2893 ofp_port_t out_port, struct list *rules)
2895 struct oftable *table;
2899 error = check_table_id(ofproto, table_id);
2905 cls_rule_init(&cr, match, 0);
2907 if (cookie_mask == htonll(UINT64_MAX)) {
2910 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node, hash_cookie(cookie),
2911 &ofproto->cookies) {
2912 if (table_id != rule->table_id && table_id != 0xff) {
2915 if (ofproto_rule_is_hidden(rule)) {
2918 if (cls_rule_is_loose_match(&rule->cr, &cr.match)) {
2919 if (rule->pending) {
2920 error = OFPROTO_POSTPONE;
2923 if (rule->flow_cookie == cookie /* Hash collisions possible. */
2924 && ofproto_rule_has_out_port(rule, out_port)) {
2925 list_push_back(rules, &rule->ofproto_node);
2932 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2933 struct cls_cursor cursor;
2936 ovs_rwlock_rdlock(&table->cls.rwlock);
2937 cls_cursor_init(&cursor, &table->cls, &cr);
2938 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2939 if (rule->pending) {
2940 ovs_rwlock_unlock(&table->cls.rwlock);
2941 error = OFPROTO_POSTPONE;
2944 if (!ofproto_rule_is_hidden(rule)
2945 && ofproto_rule_has_out_port(rule, out_port)
2946 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2947 list_push_back(rules, &rule->ofproto_node);
2950 ovs_rwlock_unlock(&table->cls.rwlock);
2954 cls_rule_destroy(&cr);
2958 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2959 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2960 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2963 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2964 * to 'out_port' are included.
2966 * Hidden rules are always omitted.
2968 * Returns 0 on success, otherwise an OpenFlow error code. */
2970 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
2971 const struct match *match, unsigned int priority,
2972 ovs_be64 cookie, ovs_be64 cookie_mask,
2973 ofp_port_t out_port, struct list *rules)
2975 struct oftable *table;
2979 error = check_table_id(ofproto, table_id);
2985 cls_rule_init(&cr, match, priority);
2987 if (cookie_mask == htonll(UINT64_MAX)) {
2990 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node, hash_cookie(cookie),
2991 &ofproto->cookies) {
2992 if (table_id != rule->table_id && table_id != 0xff) {
2995 if (ofproto_rule_is_hidden(rule)) {
2998 if (cls_rule_equal(&rule->cr, &cr)) {
2999 if (rule->pending) {
3000 error = OFPROTO_POSTPONE;
3003 if (rule->flow_cookie == cookie /* Hash collisions possible. */
3004 && ofproto_rule_has_out_port(rule, out_port)) {
3005 list_push_back(rules, &rule->ofproto_node);
3012 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
3015 ovs_rwlock_rdlock(&table->cls.rwlock);
3016 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls,
3018 ovs_rwlock_unlock(&table->cls.rwlock);
3020 if (rule->pending) {
3021 error = OFPROTO_POSTPONE;
3024 if (!ofproto_rule_is_hidden(rule)
3025 && ofproto_rule_has_out_port(rule, out_port)
3026 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
3027 list_push_back(rules, &rule->ofproto_node);
3033 cls_rule_destroy(&cr);
3037 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
3038 * forced into the range of a uint16_t. */
3040 age_secs(long long int age_ms)
3042 return (age_ms < 0 ? 0
3043 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
3044 : (unsigned int) age_ms / 1000);
3048 handle_flow_stats_request(struct ofconn *ofconn,
3049 const struct ofp_header *request)
3051 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3052 struct ofputil_flow_stats_request fsr;
3053 struct list replies;
3058 error = ofputil_decode_flow_stats_request(&fsr, request);
3063 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
3064 fsr.cookie, fsr.cookie_mask,
3065 fsr.out_port, &rules);
3070 ofpmp_init(&replies, request);
3071 LIST_FOR_EACH (rule, ofproto_node, &rules) {
3072 long long int now = time_msec();
3073 struct ofputil_flow_stats fs;
3075 minimatch_expand(&rule->cr.match, &fs.match);
3076 fs.priority = rule->cr.priority;
3077 fs.cookie = rule->flow_cookie;
3078 fs.table_id = rule->table_id;
3079 calc_duration(rule->created, now, &fs.duration_sec, &fs.duration_nsec);
3080 fs.idle_age = age_secs(now - rule->used);
3081 fs.hard_age = age_secs(now - rule->modified);
3082 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
3084 fs.ofpacts = rule->ofpacts;
3085 fs.ofpacts_len = rule->ofpacts_len;
3087 ovs_mutex_lock(&rule->timeout_mutex);
3088 fs.idle_timeout = rule->idle_timeout;
3089 fs.hard_timeout = rule->hard_timeout;
3090 ovs_mutex_unlock(&rule->timeout_mutex);
3093 if (rule->send_flow_removed) {
3094 fs.flags |= OFPUTIL_FF_SEND_FLOW_REM;
3095 /* FIXME: Implement OFPUTIL_FF_NO_PKT_COUNTS and
3096 OFPUTIL_FF_NO_BYT_COUNTS. */
3098 ofputil_append_flow_stats_reply(&fs, &replies);
3100 ofconn_send_replies(ofconn, &replies);
3106 flow_stats_ds(struct rule *rule, struct ds *results)
3108 uint64_t packet_count, byte_count;
3110 rule->ofproto->ofproto_class->rule_get_stats(rule,
3111 &packet_count, &byte_count);
3113 if (rule->table_id != 0) {
3114 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
3116 ds_put_format(results, "duration=%llds, ",
3117 (time_msec() - rule->created) / 1000);
3118 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3119 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3120 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3121 cls_rule_format(&rule->cr, results);
3122 ds_put_char(results, ',');
3123 ofpacts_format(rule->ofpacts, rule->ofpacts_len, results);
3124 ds_put_cstr(results, "\n");
3127 /* Adds a pretty-printed description of all flows to 'results', including
3128 * hidden flows (e.g., set up by in-band control). */
3130 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3132 struct oftable *table;
3134 OFPROTO_FOR_EACH_TABLE (table, p) {
3135 struct cls_cursor cursor;
3138 ovs_rwlock_rdlock(&table->cls.rwlock);
3139 cls_cursor_init(&cursor, &table->cls, NULL);
3140 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3141 flow_stats_ds(rule, results);
3143 ovs_rwlock_unlock(&table->cls.rwlock);
3147 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
3148 * '*engine_type' and '*engine_id', respectively. */
3150 ofproto_get_netflow_ids(const struct ofproto *ofproto,
3151 uint8_t *engine_type, uint8_t *engine_id)
3153 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
3156 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
3157 * true if the port's CFM status was successfully stored into '*status'.
3158 * Returns false if the port did not have CFM configured, in which case
3159 * '*status' is indeterminate.
3161 * The caller must provide and owns '*status', and must free 'status->rmps'. */
3163 ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port,
3164 struct ofproto_cfm_status *status)
3166 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
3168 && ofproto->ofproto_class->get_cfm_status
3169 && ofproto->ofproto_class->get_cfm_status(ofport, status));
3173 handle_aggregate_stats_request(struct ofconn *ofconn,
3174 const struct ofp_header *oh)
3176 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3177 struct ofputil_flow_stats_request request;
3178 struct ofputil_aggregate_stats stats;
3179 bool unknown_packets, unknown_bytes;
3180 struct ofpbuf *reply;
3185 error = ofputil_decode_flow_stats_request(&request, oh);
3190 error = collect_rules_loose(ofproto, request.table_id, &request.match,
3191 request.cookie, request.cookie_mask,
3192 request.out_port, &rules);
3197 memset(&stats, 0, sizeof stats);
3198 unknown_packets = unknown_bytes = false;
3199 LIST_FOR_EACH (rule, ofproto_node, &rules) {
3200 uint64_t packet_count;
3201 uint64_t byte_count;
3203 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
3206 if (packet_count == UINT64_MAX) {
3207 unknown_packets = true;
3209 stats.packet_count += packet_count;
3212 if (byte_count == UINT64_MAX) {
3213 unknown_bytes = true;
3215 stats.byte_count += byte_count;
3220 if (unknown_packets) {
3221 stats.packet_count = UINT64_MAX;
3223 if (unknown_bytes) {
3224 stats.byte_count = UINT64_MAX;
3227 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
3228 ofconn_send_reply(ofconn, reply);
3233 struct queue_stats_cbdata {
3234 struct ofport *ofport;
3235 struct list replies;
3240 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3241 const struct netdev_queue_stats *stats)
3243 struct ofputil_queue_stats oqs;
3245 oqs.port_no = cbdata->ofport->pp.port_no;
3246 oqs.queue_id = queue_id;
3247 oqs.tx_bytes = stats->tx_bytes;
3248 oqs.tx_packets = stats->tx_packets;
3249 oqs.tx_errors = stats->tx_errors;
3250 if (stats->created != LLONG_MIN) {
3251 calc_duration(stats->created, cbdata->now,
3252 &oqs.duration_sec, &oqs.duration_nsec);
3254 oqs.duration_sec = oqs.duration_nsec = UINT32_MAX;
3256 ofputil_append_queue_stat(&cbdata->replies, &oqs);
3260 handle_queue_stats_dump_cb(uint32_t queue_id,
3261 struct netdev_queue_stats *stats,
3264 struct queue_stats_cbdata *cbdata = cbdata_;
3266 put_queue_stats(cbdata, queue_id, stats);
3270 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3271 struct queue_stats_cbdata *cbdata)
3273 cbdata->ofport = port;
3274 if (queue_id == OFPQ_ALL) {
3275 netdev_dump_queue_stats(port->netdev,
3276 handle_queue_stats_dump_cb, cbdata);
3278 struct netdev_queue_stats stats;
3280 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3281 put_queue_stats(cbdata, queue_id, &stats);
3283 return OFPERR_OFPQOFC_BAD_QUEUE;
3290 handle_queue_stats_request(struct ofconn *ofconn,
3291 const struct ofp_header *rq)
3293 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3294 struct queue_stats_cbdata cbdata;
3295 struct ofport *port;
3297 struct ofputil_queue_stats_request oqsr;
3299 COVERAGE_INC(ofproto_queue_req);
3301 ofpmp_init(&cbdata.replies, rq);
3302 cbdata.now = time_msec();
3304 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3309 if (oqsr.port_no == OFPP_ANY) {
3310 error = OFPERR_OFPQOFC_BAD_QUEUE;
3311 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3312 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3317 port = ofproto_get_port(ofproto, oqsr.port_no);
3319 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3320 : OFPERR_OFPQOFC_BAD_PORT);
3323 ofconn_send_replies(ofconn, &cbdata.replies);
3325 ofpbuf_list_delete(&cbdata.replies);
3332 is_flow_deletion_pending(const struct ofproto *ofproto,
3333 const struct cls_rule *cls_rule,
3336 if (!hmap_is_empty(&ofproto->deletions)) {
3337 struct ofoperation *op;
3339 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3340 cls_rule_hash(cls_rule, table_id),
3341 &ofproto->deletions) {
3342 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
3352 evict_rule_from_table(struct ofproto *ofproto, struct oftable *table)
3357 ovs_rwlock_rdlock(&table->cls.rwlock);
3358 n_rules = classifier_count(&table->cls);
3359 ovs_rwlock_unlock(&table->cls.rwlock);
3361 if (n_rules < table->max_flows) {
3363 } else if (!choose_rule_to_evict(table, &rule)) {
3364 return OFPERR_OFPFMFC_TABLE_FULL;
3365 } else if (rule->pending) {
3366 ovs_rwlock_unlock(&rule->evict);
3367 return OFPROTO_POSTPONE;
3369 struct ofopgroup *group;
3371 group = ofopgroup_create_unattached(ofproto);
3372 delete_flow__(rule, group, OFPRR_EVICTION);
3373 ofopgroup_submit(group);
3379 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3380 * in which no matching flow already exists in the flow table.
3382 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3383 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3384 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3385 * initiated now but may be retried later.
3387 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
3388 * ownership remains with the caller.
3390 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3393 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
3394 struct ofputil_flow_mod *fm, const struct ofp_header *request)
3396 struct oftable *table;
3397 struct ofopgroup *group;
3403 error = check_table_id(ofproto, fm->table_id);
3409 if (fm->table_id == 0xff) {
3410 if (ofproto->ofproto_class->rule_choose_table) {
3411 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3417 ovs_assert(table_id < ofproto->n_tables);
3421 } else if (fm->table_id < ofproto->n_tables) {
3422 table_id = fm->table_id;
3424 return OFPERR_OFPBRC_BAD_TABLE_ID;
3427 table = &ofproto->tables[table_id];
3429 if (table->flags & OFTABLE_READONLY) {
3430 return OFPERR_OFPBRC_EPERM;
3433 cls_rule_init(&cr, &fm->match, fm->priority);
3435 /* Transform "add" into "modify" if there's an existing identical flow. */
3436 ovs_rwlock_rdlock(&table->cls.rwlock);
3437 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr));
3438 ovs_rwlock_unlock(&table->cls.rwlock);
3440 cls_rule_destroy(&cr);
3441 if (!rule_is_modifiable(rule)) {
3442 return OFPERR_OFPBRC_EPERM;
3443 } else if (rule->pending) {
3444 return OFPROTO_POSTPONE;
3449 list_push_back(&rules, &rule->ofproto_node);
3450 fm->modify_cookie = true;
3451 return modify_flows__(ofproto, ofconn, fm, request, &rules);
3455 /* Verify actions. */
3456 error = ofproto_check_ofpacts(ofproto, fm->ofpacts, fm->ofpacts_len,
3457 &fm->match.flow, table_id);
3462 /* Serialize against pending deletion. */
3463 if (is_flow_deletion_pending(ofproto, &cr, table_id)) {
3464 cls_rule_destroy(&cr);
3465 return OFPROTO_POSTPONE;
3468 /* Check for overlap, if requested. */
3469 if (fm->flags & OFPUTIL_FF_CHECK_OVERLAP) {
3472 ovs_rwlock_rdlock(&table->cls.rwlock);
3473 overlaps = classifier_rule_overlaps(&table->cls, &cr);
3474 ovs_rwlock_unlock(&table->cls.rwlock);
3477 cls_rule_destroy(&cr);
3478 return OFPERR_OFPFMFC_OVERLAP;
3482 /* If necessary, evict an existing rule to clear out space. */
3483 error = evict_rule_from_table(ofproto, table);
3485 cls_rule_destroy(&cr);
3489 /* Allocate new rule. */
3490 rule = ofproto->ofproto_class->rule_alloc();
3492 cls_rule_destroy(&cr);
3493 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
3494 ofproto->name, ovs_strerror(error));
3498 /* Initialize base state. */
3499 rule->ofproto = ofproto;
3500 cls_rule_move(&rule->cr, &cr);
3501 rule->pending = NULL;
3502 rule->flow_cookie = fm->new_cookie;
3503 rule->created = rule->modified = rule->used = time_msec();
3505 ovs_mutex_init(&rule->timeout_mutex);
3506 ovs_mutex_lock(&rule->timeout_mutex);
3507 rule->idle_timeout = fm->idle_timeout;
3508 rule->hard_timeout = fm->hard_timeout;
3509 ovs_mutex_unlock(&rule->timeout_mutex);
3511 rule->table_id = table - ofproto->tables;
3512 rule->send_flow_removed = (fm->flags & OFPUTIL_FF_SEND_FLOW_REM) != 0;
3513 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3514 rule->ofpacts_len = fm->ofpacts_len;
3515 rule->meter_id = find_meter(rule->ofpacts, rule->ofpacts_len);
3516 list_init(&rule->meter_list_node);
3517 rule->eviction_group = NULL;
3518 list_init(&rule->expirable);
3519 rule->monitor_flags = 0;
3520 rule->add_seqno = 0;
3521 rule->modify_seqno = 0;
3522 ovs_rwlock_init(&rule->evict);
3524 /* Construct rule, initializing derived state. */
3525 error = ofproto->ofproto_class->rule_construct(rule);
3527 ofproto_rule_destroy__(rule);
3532 oftable_insert_rule(rule);
3534 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3535 ofoperation_create(group, rule, OFOPERATION_ADD, 0);
3536 ofproto->ofproto_class->rule_insert(rule);
3537 ofopgroup_submit(group);
3542 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3544 /* Modifies the rules listed in 'rules', changing their actions to match those
3547 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3550 * Returns 0 on success, otherwise an OpenFlow error code. */
3552 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3553 struct ofputil_flow_mod *fm, const struct ofp_header *request,
3556 enum ofoperation_type type;
3557 struct ofopgroup *group;
3561 type = fm->command == OFPFC_ADD ? OFOPERATION_REPLACE : OFOPERATION_MODIFY;
3562 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3563 error = OFPERR_OFPBRC_EPERM;
3564 LIST_FOR_EACH (rule, ofproto_node, rules) {
3565 struct ofoperation *op;
3566 bool actions_changed;
3567 bool reset_counters;
3569 /* FIXME: Implement OFPFUTIL_FF_RESET_COUNTS */
3571 if (rule_is_modifiable(rule)) {
3572 /* At least one rule is modifiable, don't report EPERM error. */
3578 /* Verify actions. */
3579 error = ofpacts_check(fm->ofpacts, fm->ofpacts_len, &fm->match.flow,
3580 u16_to_ofp(ofproto->max_ports), rule->table_id);
3585 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
3586 rule->ofpacts, rule->ofpacts_len);
3588 op = ofoperation_create(group, rule, type, 0);
3590 if (fm->modify_cookie && fm->new_cookie != htonll(UINT64_MAX)) {
3591 ofproto_rule_change_cookie(ofproto, rule, fm->new_cookie);
3593 if (type == OFOPERATION_REPLACE) {
3594 ovs_mutex_lock(&rule->timeout_mutex);
3595 rule->idle_timeout = fm->idle_timeout;
3596 rule->hard_timeout = fm->hard_timeout;
3597 ovs_mutex_unlock(&rule->timeout_mutex);
3599 rule->send_flow_removed = (fm->flags
3600 & OFPUTIL_FF_SEND_FLOW_REM) != 0;
3602 if (fm->idle_timeout || fm->hard_timeout) {
3603 if (!rule->eviction_group) {
3604 eviction_group_add_rule(rule);
3607 eviction_group_remove_rule(rule);
3611 reset_counters = (fm->flags & OFPUTIL_FF_RESET_COUNTS) != 0;
3612 if (actions_changed || reset_counters) {
3613 op->ofpacts = rule->ofpacts;
3614 op->ofpacts_len = rule->ofpacts_len;
3615 op->meter_id = rule->meter_id;
3616 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3617 rule->ofpacts_len = fm->ofpacts_len;
3618 rule->meter_id = find_meter(rule->ofpacts, rule->ofpacts_len);
3619 rule->ofproto->ofproto_class->rule_modify_actions(rule,
3622 ofoperation_complete(op, 0);
3625 ofopgroup_submit(group);
3631 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
3632 struct ofputil_flow_mod *fm, const struct ofp_header *request)
3634 if (fm->cookie_mask != htonll(0) || fm->new_cookie == htonll(UINT64_MAX)) {
3637 return add_flow(ofproto, ofconn, fm, request);
3640 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
3643 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3646 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3647 struct ofputil_flow_mod *fm,
3648 const struct ofp_header *request)
3653 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3654 fm->cookie, fm->cookie_mask,
3658 } else if (list_is_empty(&rules)) {
3659 return modify_flows_add(ofproto, ofconn, fm, request);
3661 return modify_flows__(ofproto, ofconn, fm, request, &rules);
3665 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3668 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3671 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3672 struct ofputil_flow_mod *fm,
3673 const struct ofp_header *request)
3678 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3679 fm->priority, fm->cookie, fm->cookie_mask,
3684 } else if (list_is_empty(&rules)) {
3685 return modify_flows_add(ofproto, ofconn, fm, request);
3687 return list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
3688 fm, request, &rules)
3693 /* OFPFC_DELETE implementation. */
3696 delete_flow__(struct rule *rule, struct ofopgroup *group,
3697 enum ofp_flow_removed_reason reason)
3699 struct ofproto *ofproto = rule->ofproto;
3701 ofproto_rule_send_removed(rule, reason);
3703 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
3704 oftable_remove_rule(rule);
3705 ofproto->ofproto_class->rule_delete(rule);
3708 /* Deletes the rules listed in 'rules'.
3710 * Returns 0 on success, otherwise an OpenFlow error code. */
3712 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3713 const struct ofp_header *request, struct list *rules,
3714 enum ofp_flow_removed_reason reason)
3716 struct rule *rule, *next;
3717 struct ofopgroup *group;
3719 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
3720 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
3721 ovs_rwlock_wrlock(&rule->evict);
3722 delete_flow__(rule, group, reason);
3724 ofopgroup_submit(group);
3729 /* Implements OFPFC_DELETE. */
3731 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3732 const struct ofputil_flow_mod *fm,
3733 const struct ofp_header *request)
3738 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3739 fm->cookie, fm->cookie_mask,
3740 fm->out_port, &rules);
3741 return (error ? error
3742 : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
3743 &rules, OFPRR_DELETE)
3747 /* Implements OFPFC_DELETE_STRICT. */
3749 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3750 const struct ofputil_flow_mod *fm,
3751 const struct ofp_header *request)
3756 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3757 fm->priority, fm->cookie, fm->cookie_mask,
3758 fm->out_port, &rules);
3759 return (error ? error
3760 : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
3767 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
3769 struct ofputil_flow_removed fr;
3771 if (ofproto_rule_is_hidden(rule) || !rule->send_flow_removed) {
3775 minimatch_expand(&rule->cr.match, &fr.match);
3776 fr.priority = rule->cr.priority;
3777 fr.cookie = rule->flow_cookie;
3779 fr.table_id = rule->table_id;
3780 calc_duration(rule->created, time_msec(),
3781 &fr.duration_sec, &fr.duration_nsec);
3782 ovs_mutex_lock(&rule->timeout_mutex);
3783 fr.idle_timeout = rule->idle_timeout;
3784 fr.hard_timeout = rule->hard_timeout;
3785 ovs_mutex_unlock(&rule->timeout_mutex);
3786 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
3789 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
3793 ofproto_rule_update_used(struct rule *rule, long long int used)
3795 if (used > rule->used) {
3796 struct eviction_group *evg = rule->eviction_group;
3800 heap_change(&evg->rules, &rule->evg_node,
3801 rule_eviction_priority(rule));
3806 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3807 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3810 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
3813 * ofproto implementation ->run() functions should use this function to expire
3814 * OpenFlow flows. */
3816 ofproto_rule_expire(struct rule *rule, uint8_t reason)
3818 struct ofproto *ofproto = rule->ofproto;
3819 struct classifier *cls = &ofproto->tables[rule->table_id].cls;
3821 ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
3822 ofproto_rule_send_removed(rule, reason);
3824 ovs_rwlock_wrlock(&cls->rwlock);
3825 ofproto_delete_rule(ofproto, cls, rule);
3826 ovs_rwlock_unlock(&cls->rwlock);
3829 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
3830 * means "infinite". */
3832 reduce_timeout(uint16_t max, uint16_t *timeout)
3834 if (max && (!*timeout || *timeout > max)) {
3839 /* If 'idle_timeout' is nonzero, and 'rule' has no idle timeout or an idle
3840 * timeout greater than 'idle_timeout', lowers 'rule''s idle timeout to
3841 * 'idle_timeout' seconds. Similarly for 'hard_timeout'.
3843 * Suitable for implementing OFPACT_FIN_TIMEOUT. */
3845 ofproto_rule_reduce_timeouts(struct rule *rule,
3846 uint16_t idle_timeout, uint16_t hard_timeout)
3847 OVS_EXCLUDED(rule->ofproto->expirable_mutex, rule->timeout_mutex)
3849 if (!idle_timeout && !hard_timeout) {
3853 ovs_mutex_lock(&rule->ofproto->expirable_mutex);
3854 if (list_is_empty(&rule->expirable)) {
3855 list_insert(&rule->ofproto->expirable, &rule->expirable);
3857 ovs_mutex_unlock(&rule->ofproto->expirable_mutex);
3859 ovs_mutex_lock(&rule->timeout_mutex);
3860 reduce_timeout(idle_timeout, &rule->idle_timeout);
3861 reduce_timeout(hard_timeout, &rule->hard_timeout);
3862 ovs_mutex_unlock(&rule->timeout_mutex);
3864 if (!rule->eviction_group) {
3865 eviction_group_add_rule(rule);
3870 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3872 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3873 struct ofputil_flow_mod fm;
3874 uint64_t ofpacts_stub[1024 / 8];
3875 struct ofpbuf ofpacts;
3879 error = reject_slave_controller(ofconn);
3884 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3885 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
3888 error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
3891 goto exit_free_ofpacts;
3894 /* Record the operation for logging a summary report. */
3895 switch (fm.command) {
3901 case OFPFC_MODIFY_STRICT:
3902 ofproto->n_modify++;
3906 case OFPFC_DELETE_STRICT:
3907 ofproto->n_delete++;
3912 if (ofproto->next_op_report == LLONG_MAX) {
3913 ofproto->first_op = now;
3914 ofproto->next_op_report = MAX(now + 10 * 1000,
3915 ofproto->op_backoff);
3916 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
3918 ofproto->last_op = now;
3921 ofpbuf_uninit(&ofpacts);
3927 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
3928 struct ofputil_flow_mod *fm, const struct ofp_header *oh)
3930 if (ofproto->n_pending >= 50) {
3931 ovs_assert(!list_is_empty(&ofproto->pending));
3932 return OFPROTO_POSTPONE;
3935 switch (fm->command) {
3937 return add_flow(ofproto, ofconn, fm, oh);
3940 return modify_flows_loose(ofproto, ofconn, fm, oh);
3942 case OFPFC_MODIFY_STRICT:
3943 return modify_flow_strict(ofproto, ofconn, fm, oh);
3946 return delete_flows_loose(ofproto, ofconn, fm, oh);
3948 case OFPFC_DELETE_STRICT:
3949 return delete_flow_strict(ofproto, ofconn, fm, oh);
3952 if (fm->command > 0xff) {
3953 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
3954 "flow_mod_table_id extension is not enabled",
3957 return OFPERR_OFPFMFC_BAD_COMMAND;
3962 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3964 struct ofputil_role_request request;
3965 struct ofputil_role_request reply;
3969 error = ofputil_decode_role_message(oh, &request);
3974 if (request.role != OFPCR12_ROLE_NOCHANGE) {
3975 if (ofconn_get_role(ofconn) != request.role
3976 && ofconn_has_pending_opgroups(ofconn)) {
3977 return OFPROTO_POSTPONE;
3980 if (request.have_generation_id
3981 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
3982 return OFPERR_OFPRRFC_STALE;
3985 ofconn_set_role(ofconn, request.role);
3988 reply.role = ofconn_get_role(ofconn);
3989 reply.have_generation_id = ofconn_get_master_election_id(
3990 ofconn, &reply.generation_id);
3991 buf = ofputil_encode_role_reply(oh, &reply);
3992 ofconn_send_reply(ofconn, buf);
3998 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
3999 const struct ofp_header *oh)
4001 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
4002 enum ofputil_protocol cur, next;
4004 cur = ofconn_get_protocol(ofconn);
4005 next = ofputil_protocol_set_tid(cur, msg->set != 0);
4006 ofconn_set_protocol(ofconn, next);
4012 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4014 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
4015 enum ofputil_protocol cur, next;
4016 enum ofputil_protocol next_base;
4018 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
4020 return OFPERR_OFPBRC_EPERM;
4023 cur = ofconn_get_protocol(ofconn);
4024 next = ofputil_protocol_set_base(cur, next_base);
4025 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
4026 /* Avoid sending async messages in surprising protocol. */
4027 return OFPROTO_POSTPONE;
4030 ofconn_set_protocol(ofconn, next);
4035 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
4036 const struct ofp_header *oh)
4038 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
4041 format = ntohl(msg->format);
4042 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
4043 return OFPERR_OFPBRC_EPERM;
4046 if (format != ofconn_get_packet_in_format(ofconn)
4047 && ofconn_has_pending_opgroups(ofconn)) {
4048 /* Avoid sending async message in surprsing packet in format. */
4049 return OFPROTO_POSTPONE;
4052 ofconn_set_packet_in_format(ofconn, format);
4057 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
4059 const struct nx_async_config *msg = ofpmsg_body(oh);
4060 uint32_t master[OAM_N_TYPES];
4061 uint32_t slave[OAM_N_TYPES];
4063 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
4064 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
4065 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
4067 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
4068 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
4069 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
4071 ofconn_set_async_config(ofconn, master, slave);
4072 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
4073 !ofconn_get_miss_send_len(ofconn)) {
4074 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
4081 handle_nxt_set_controller_id(struct ofconn *ofconn,
4082 const struct ofp_header *oh)
4084 const struct nx_controller_id *nci = ofpmsg_body(oh);
4086 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
4087 return OFPERR_NXBRC_MUST_BE_ZERO;
4090 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
4095 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4099 if (ofconn_has_pending_opgroups(ofconn)) {
4100 return OFPROTO_POSTPONE;
4103 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
4104 ? OFPRAW_OFPT10_BARRIER_REPLY
4105 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
4106 ofconn_send_reply(ofconn, buf);
4111 ofproto_compose_flow_refresh_update(const struct rule *rule,
4112 enum nx_flow_monitor_flags flags,
4115 struct ofoperation *op = rule->pending;
4116 struct ofputil_flow_update fu;
4119 if (op && op->type == OFOPERATION_ADD) {
4120 /* We'll report the final flow when the operation completes. Reporting
4121 * it now would cause a duplicate report later. */
4125 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
4126 ? NXFME_ADDED : NXFME_MODIFIED);
4128 ovs_mutex_lock(&rule->timeout_mutex);
4129 fu.idle_timeout = rule->idle_timeout;
4130 fu.hard_timeout = rule->hard_timeout;
4131 ovs_mutex_unlock(&rule->timeout_mutex);
4132 fu.table_id = rule->table_id;
4133 fu.cookie = rule->flow_cookie;
4134 minimatch_expand(&rule->cr.match, &match);
4136 fu.priority = rule->cr.priority;
4137 if (!(flags & NXFMF_ACTIONS)) {
4141 fu.ofpacts = rule->ofpacts;
4142 fu.ofpacts_len = rule->ofpacts_len;
4144 /* An operation is in progress. Use the previous version of the flow's
4145 * actions, so that when the operation commits we report the change. */
4147 case OFOPERATION_ADD:
4150 case OFOPERATION_MODIFY:
4151 case OFOPERATION_REPLACE:
4153 fu.ofpacts = op->ofpacts;
4154 fu.ofpacts_len = op->ofpacts_len;
4156 fu.ofpacts = rule->ofpacts;
4157 fu.ofpacts_len = rule->ofpacts_len;
4161 case OFOPERATION_DELETE:
4162 fu.ofpacts = rule->ofpacts;
4163 fu.ofpacts_len = rule->ofpacts_len;
4171 if (list_is_empty(msgs)) {
4172 ofputil_start_flow_update(msgs);
4174 ofputil_append_flow_update(&fu, msgs);
4178 ofmonitor_compose_refresh_updates(struct list *rules, struct list *msgs)
4182 LIST_FOR_EACH (rule, ofproto_node, rules) {
4183 enum nx_flow_monitor_flags flags = rule->monitor_flags;
4184 rule->monitor_flags = 0;
4186 ofproto_compose_flow_refresh_update(rule, flags, msgs);
4191 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
4192 struct rule *rule, uint64_t seqno,
4195 enum nx_flow_monitor_flags update;
4197 if (ofproto_rule_is_hidden(rule)) {
4202 ? ofoperation_has_out_port(rule->pending, m->out_port)
4203 : ofproto_rule_has_out_port(rule, m->out_port))) {
4208 if (rule->add_seqno > seqno) {
4209 update = NXFMF_ADD | NXFMF_MODIFY;
4210 } else if (rule->modify_seqno > seqno) {
4211 update = NXFMF_MODIFY;
4216 if (!(m->flags & update)) {
4220 update = NXFMF_INITIAL;
4223 if (!rule->monitor_flags) {
4224 list_push_back(rules, &rule->ofproto_node);
4226 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
4230 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
4234 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
4235 const struct ofoperation *op;
4236 const struct oftable *table;
4237 struct cls_rule target;
4239 cls_rule_init_from_minimatch(&target, &m->match, 0);
4240 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
4241 struct cls_cursor cursor;
4244 ovs_rwlock_rdlock(&table->cls.rwlock);
4245 cls_cursor_init(&cursor, &table->cls, &target);
4246 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4247 ovs_assert(!rule->pending); /* XXX */
4248 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4250 ovs_rwlock_unlock(&table->cls.rwlock);
4253 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
4254 struct rule *rule = op->rule;
4256 if (((m->table_id == 0xff
4257 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
4258 : m->table_id == rule->table_id))
4259 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
4260 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4263 cls_rule_destroy(&target);
4267 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
4270 if (m->flags & NXFMF_INITIAL) {
4271 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
4276 ofmonitor_collect_resume_rules(struct ofmonitor *m,
4277 uint64_t seqno, struct list *rules)
4279 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
4283 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
4285 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4286 struct ofmonitor **monitors;
4287 size_t n_monitors, allocated_monitors;
4288 struct list replies;
4295 ofpbuf_use_const(&b, oh, ntohs(oh->length));
4297 n_monitors = allocated_monitors = 0;
4299 struct ofputil_flow_monitor_request request;
4300 struct ofmonitor *m;
4303 retval = ofputil_decode_flow_monitor_request(&request, &b);
4304 if (retval == EOF) {
4306 } else if (retval) {
4311 if (request.table_id != 0xff
4312 && request.table_id >= ofproto->n_tables) {
4313 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4317 error = ofmonitor_create(&request, ofconn, &m);
4322 if (n_monitors >= allocated_monitors) {
4323 monitors = x2nrealloc(monitors, &allocated_monitors,
4326 monitors[n_monitors++] = m;
4330 for (i = 0; i < n_monitors; i++) {
4331 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
4334 ofpmp_init(&replies, oh);
4335 ofmonitor_compose_refresh_updates(&rules, &replies);
4336 ofconn_send_replies(ofconn, &replies);
4343 for (i = 0; i < n_monitors; i++) {
4344 ofmonitor_destroy(monitors[i]);
4351 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
4353 struct ofmonitor *m;
4356 id = ofputil_decode_flow_monitor_cancel(oh);
4357 m = ofmonitor_lookup(ofconn, id);
4359 return OFPERR_NXBRC_FM_BAD_ID;
4362 ofmonitor_destroy(m);
4366 /* Meters implementation.
4368 * Meter table entry, indexed by the OpenFlow meter_id.
4369 * These are always dynamically allocated to allocate enough space for
4371 * 'created' is used to compute the duration for meter stats.
4372 * 'list rules' is needed so that we can delete the dependent rules when the
4373 * meter table entry is deleted.
4374 * 'provider_meter_id' is for the provider's private use.
4377 long long int created; /* Time created. */
4378 struct list rules; /* List of "struct rule_dpif"s. */
4379 ofproto_meter_id provider_meter_id;
4380 uint16_t flags; /* Meter flags. */
4381 uint16_t n_bands; /* Number of meter bands. */
4382 struct ofputil_meter_band *bands;
4386 * This is used in instruction validation at flow set-up time,
4387 * as flows may not use non-existing meters.
4388 * This is also used by ofproto-providers to translate OpenFlow meter_ids
4389 * in METER instructions to the corresponding provider meter IDs.
4390 * Return value of UINT32_MAX signifies an invalid meter.
4393 ofproto_get_provider_meter_id(const struct ofproto * ofproto,
4394 uint32_t of_meter_id)
4396 if (of_meter_id && of_meter_id <= ofproto->meter_features.max_meters) {
4397 const struct meter *meter = ofproto->meters[of_meter_id];
4399 return meter->provider_meter_id.uint32;
4406 meter_update(struct meter *meter, const struct ofputil_meter_config *config)
4410 meter->flags = config->flags;
4411 meter->n_bands = config->n_bands;
4412 meter->bands = xmemdup(config->bands,
4413 config->n_bands * sizeof *meter->bands);
4416 static struct meter *
4417 meter_create(const struct ofputil_meter_config *config,
4418 ofproto_meter_id provider_meter_id)
4420 struct meter *meter;
4422 meter = xzalloc(sizeof *meter);
4423 meter->provider_meter_id = provider_meter_id;
4424 meter->created = time_msec();
4425 list_init(&meter->rules);
4427 meter_update(meter, config);
4433 meter_delete(struct ofproto *ofproto, uint32_t first, uint32_t last)
4436 for (mid = first; mid <= last; ++mid) {
4437 struct meter *meter = ofproto->meters[mid];
4439 ofproto->meters[mid] = NULL;
4440 ofproto->ofproto_class->meter_del(ofproto,
4441 meter->provider_meter_id);
4449 handle_add_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
4451 ofproto_meter_id provider_meter_id = { UINT32_MAX };
4452 struct meter **meterp = &ofproto->meters[mm->meter.meter_id];
4456 return OFPERR_OFPMMFC_METER_EXISTS;
4459 error = ofproto->ofproto_class->meter_set(ofproto, &provider_meter_id,
4462 ovs_assert(provider_meter_id.uint32 != UINT32_MAX);
4463 *meterp = meter_create(&mm->meter, provider_meter_id);
4469 handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
4471 struct meter *meter = ofproto->meters[mm->meter.meter_id];
4475 return OFPERR_OFPMMFC_UNKNOWN_METER;
4478 error = ofproto->ofproto_class->meter_set(ofproto,
4479 &meter->provider_meter_id,
4481 ovs_assert(meter->provider_meter_id.uint32 != UINT32_MAX);
4483 meter_update(meter, &mm->meter);
4489 handle_delete_meter(struct ofconn *ofconn, const struct ofp_header *oh,
4490 struct ofputil_meter_mod *mm)
4492 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4493 uint32_t meter_id = mm->meter.meter_id;
4494 uint32_t first, last;
4497 if (meter_id == OFPM13_ALL) {
4499 last = ofproto->meter_features.max_meters;
4501 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
4504 first = last = meter_id;
4507 /* First delete the rules that use this meter. If any of those rules are
4508 * currently being modified, postpone the whole operation until later. */
4510 for (meter_id = first; meter_id <= last; ++meter_id) {
4511 struct meter *meter = ofproto->meters[meter_id];
4512 if (meter && !list_is_empty(&meter->rules)) {
4515 LIST_FOR_EACH (rule, meter_list_node, &meter->rules) {
4516 if (rule->pending) {
4517 return OFPROTO_POSTPONE;
4519 list_push_back(&rules, &rule->ofproto_node);
4523 if (!list_is_empty(&rules)) {
4524 delete_flows__(ofproto, ofconn, oh, &rules, OFPRR_METER_DELETE);
4527 /* Delete the meters. */
4528 meter_delete(ofproto, first, last);
4534 handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4536 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4537 struct ofputil_meter_mod mm;
4538 uint64_t bands_stub[256 / 8];
4539 struct ofpbuf bands;
4543 error = reject_slave_controller(ofconn);
4548 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
4550 error = ofputil_decode_meter_mod(oh, &mm, &bands);
4552 goto exit_free_bands;
4555 meter_id = mm.meter.meter_id;
4557 if (mm.command != OFPMC13_DELETE) {
4558 /* Fails also when meters are not implemented by the provider. */
4559 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
4560 error = OFPERR_OFPMMFC_INVALID_METER;
4561 goto exit_free_bands;
4563 if (mm.meter.n_bands > ofproto->meter_features.max_bands) {
4564 error = OFPERR_OFPMMFC_OUT_OF_BANDS;
4565 goto exit_free_bands;
4569 switch (mm.command) {
4571 error = handle_add_meter(ofproto, &mm);
4574 case OFPMC13_MODIFY:
4575 error = handle_modify_meter(ofproto, &mm);
4578 case OFPMC13_DELETE:
4579 error = handle_delete_meter(ofconn, oh, &mm);
4583 error = OFPERR_OFPMMFC_BAD_COMMAND;
4588 ofpbuf_uninit(&bands);
4593 handle_meter_features_request(struct ofconn *ofconn,
4594 const struct ofp_header *request)
4596 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4597 struct ofputil_meter_features features;
4600 if (ofproto->ofproto_class->meter_get_features) {
4601 ofproto->ofproto_class->meter_get_features(ofproto, &features);
4603 memset(&features, 0, sizeof features);
4605 b = ofputil_encode_meter_features_reply(&features, request);
4607 ofconn_send_reply(ofconn, b);
4612 handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request,
4615 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4616 struct list replies;
4617 uint64_t bands_stub[256 / 8];
4618 struct ofpbuf bands;
4619 uint32_t meter_id, first, last;
4621 ofputil_decode_meter_request(request, &meter_id);
4623 if (meter_id == OFPM13_ALL) {
4625 last = ofproto->meter_features.max_meters;
4627 if (!meter_id || meter_id > ofproto->meter_features.max_meters ||
4628 !ofproto->meters[meter_id]) {
4629 return OFPERR_OFPMMFC_UNKNOWN_METER;
4631 first = last = meter_id;
4634 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
4635 ofpmp_init(&replies, request);
4637 for (meter_id = first; meter_id <= last; ++meter_id) {
4638 struct meter *meter = ofproto->meters[meter_id];
4640 continue; /* Skip non-existing meters. */
4642 if (type == OFPTYPE_METER_STATS_REQUEST) {
4643 struct ofputil_meter_stats stats;
4645 stats.meter_id = meter_id;
4647 /* Provider sets the packet and byte counts, we do the rest. */
4648 stats.flow_count = list_size(&meter->rules);
4649 calc_duration(meter->created, time_msec(),
4650 &stats.duration_sec, &stats.duration_nsec);
4651 stats.n_bands = meter->n_bands;
4652 ofpbuf_clear(&bands);
4654 = ofpbuf_put_uninit(&bands,
4655 meter->n_bands * sizeof *stats.bands);
4657 if (!ofproto->ofproto_class->meter_get(ofproto,
4658 meter->provider_meter_id,
4660 ofputil_append_meter_stats(&replies, &stats);
4662 } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */
4663 struct ofputil_meter_config config;
4665 config.meter_id = meter_id;
4666 config.flags = meter->flags;
4667 config.n_bands = meter->n_bands;
4668 config.bands = meter->bands;
4669 ofputil_append_meter_config(&replies, &config);
4673 ofconn_send_replies(ofconn, &replies);
4674 ofpbuf_uninit(&bands);
4679 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
4681 const struct ofp_header *oh = msg->data;
4685 error = ofptype_decode(&type, oh);
4691 /* OpenFlow requests. */
4692 case OFPTYPE_ECHO_REQUEST:
4693 return handle_echo_request(ofconn, oh);
4695 case OFPTYPE_FEATURES_REQUEST:
4696 return handle_features_request(ofconn, oh);
4698 case OFPTYPE_GET_CONFIG_REQUEST:
4699 return handle_get_config_request(ofconn, oh);
4701 case OFPTYPE_SET_CONFIG:
4702 return handle_set_config(ofconn, oh);
4704 case OFPTYPE_PACKET_OUT:
4705 return handle_packet_out(ofconn, oh);
4707 case OFPTYPE_PORT_MOD:
4708 return handle_port_mod(ofconn, oh);
4710 case OFPTYPE_FLOW_MOD:
4711 return handle_flow_mod(ofconn, oh);
4713 case OFPTYPE_METER_MOD:
4714 return handle_meter_mod(ofconn, oh);
4716 case OFPTYPE_BARRIER_REQUEST:
4717 return handle_barrier_request(ofconn, oh);
4719 case OFPTYPE_ROLE_REQUEST:
4720 return handle_role_request(ofconn, oh);
4722 /* OpenFlow replies. */
4723 case OFPTYPE_ECHO_REPLY:
4726 /* Nicira extension requests. */
4727 case OFPTYPE_FLOW_MOD_TABLE_ID:
4728 return handle_nxt_flow_mod_table_id(ofconn, oh);
4730 case OFPTYPE_SET_FLOW_FORMAT:
4731 return handle_nxt_set_flow_format(ofconn, oh);
4733 case OFPTYPE_SET_PACKET_IN_FORMAT:
4734 return handle_nxt_set_packet_in_format(ofconn, oh);
4736 case OFPTYPE_SET_CONTROLLER_ID:
4737 return handle_nxt_set_controller_id(ofconn, oh);
4739 case OFPTYPE_FLOW_AGE:
4740 /* Nothing to do. */
4743 case OFPTYPE_FLOW_MONITOR_CANCEL:
4744 return handle_flow_monitor_cancel(ofconn, oh);
4746 case OFPTYPE_SET_ASYNC_CONFIG:
4747 return handle_nxt_set_async_config(ofconn, oh);
4749 /* Statistics requests. */
4750 case OFPTYPE_DESC_STATS_REQUEST:
4751 return handle_desc_stats_request(ofconn, oh);
4753 case OFPTYPE_FLOW_STATS_REQUEST:
4754 return handle_flow_stats_request(ofconn, oh);
4756 case OFPTYPE_AGGREGATE_STATS_REQUEST:
4757 return handle_aggregate_stats_request(ofconn, oh);
4759 case OFPTYPE_TABLE_STATS_REQUEST:
4760 return handle_table_stats_request(ofconn, oh);
4762 case OFPTYPE_PORT_STATS_REQUEST:
4763 return handle_port_stats_request(ofconn, oh);
4765 case OFPTYPE_QUEUE_STATS_REQUEST:
4766 return handle_queue_stats_request(ofconn, oh);
4768 case OFPTYPE_PORT_DESC_STATS_REQUEST:
4769 return handle_port_desc_stats_request(ofconn, oh);
4771 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
4772 return handle_flow_monitor_request(ofconn, oh);
4774 case OFPTYPE_METER_STATS_REQUEST:
4775 case OFPTYPE_METER_CONFIG_STATS_REQUEST:
4776 return handle_meter_request(ofconn, oh, type);
4778 case OFPTYPE_METER_FEATURES_STATS_REQUEST:
4779 return handle_meter_features_request(ofconn, oh);
4781 /* FIXME: Change the following once they are implemented: */
4782 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
4783 case OFPTYPE_GET_ASYNC_REQUEST:
4784 case OFPTYPE_GROUP_STATS_REQUEST:
4785 case OFPTYPE_GROUP_DESC_STATS_REQUEST:
4786 case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
4787 case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
4788 return OFPERR_OFPBRC_BAD_TYPE;
4792 case OFPTYPE_FEATURES_REPLY:
4793 case OFPTYPE_GET_CONFIG_REPLY:
4794 case OFPTYPE_PACKET_IN:
4795 case OFPTYPE_FLOW_REMOVED:
4796 case OFPTYPE_PORT_STATUS:
4797 case OFPTYPE_BARRIER_REPLY:
4798 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
4799 case OFPTYPE_DESC_STATS_REPLY:
4800 case OFPTYPE_FLOW_STATS_REPLY:
4801 case OFPTYPE_QUEUE_STATS_REPLY:
4802 case OFPTYPE_PORT_STATS_REPLY:
4803 case OFPTYPE_TABLE_STATS_REPLY:
4804 case OFPTYPE_AGGREGATE_STATS_REPLY:
4805 case OFPTYPE_PORT_DESC_STATS_REPLY:
4806 case OFPTYPE_ROLE_REPLY:
4807 case OFPTYPE_FLOW_MONITOR_PAUSED:
4808 case OFPTYPE_FLOW_MONITOR_RESUMED:
4809 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
4810 case OFPTYPE_GET_ASYNC_REPLY:
4811 case OFPTYPE_GROUP_STATS_REPLY:
4812 case OFPTYPE_GROUP_DESC_STATS_REPLY:
4813 case OFPTYPE_GROUP_FEATURES_STATS_REPLY:
4814 case OFPTYPE_METER_STATS_REPLY:
4815 case OFPTYPE_METER_CONFIG_STATS_REPLY:
4816 case OFPTYPE_METER_FEATURES_STATS_REPLY:
4817 case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
4819 return OFPERR_OFPBRC_BAD_TYPE;
4824 handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
4826 int error = handle_openflow__(ofconn, ofp_msg);
4827 if (error && error != OFPROTO_POSTPONE) {
4828 ofconn_send_error(ofconn, ofp_msg->data, error);
4830 COVERAGE_INC(ofproto_recv_openflow);
4831 return error != OFPROTO_POSTPONE;
4834 /* Asynchronous operations. */
4836 /* Creates and returns a new ofopgroup that is not associated with any
4837 * OpenFlow connection.
4839 * The caller should add operations to the returned group with
4840 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4841 static struct ofopgroup *
4842 ofopgroup_create_unattached(struct ofproto *ofproto)
4844 struct ofopgroup *group = xzalloc(sizeof *group);
4845 group->ofproto = ofproto;
4846 list_init(&group->ofproto_node);
4847 list_init(&group->ops);
4848 list_init(&group->ofconn_node);
4852 /* Creates and returns a new ofopgroup for 'ofproto'.
4854 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
4855 * connection. The 'request' and 'buffer_id' arguments are ignored.
4857 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
4858 * If the ofopgroup eventually fails, then the error reply will include
4859 * 'request'. If the ofopgroup eventually succeeds, then the packet with
4860 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
4862 * The caller should add operations to the returned group with
4863 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4864 static struct ofopgroup *
4865 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
4866 const struct ofp_header *request, uint32_t buffer_id)
4868 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
4870 size_t request_len = ntohs(request->length);
4872 ovs_assert(ofconn_get_ofproto(ofconn) == ofproto);
4874 ofconn_add_opgroup(ofconn, &group->ofconn_node);
4875 group->ofconn = ofconn;
4876 group->request = xmemdup(request, MIN(request_len, 64));
4877 group->buffer_id = buffer_id;
4882 /* Submits 'group' for processing.
4884 * If 'group' contains no operations (e.g. none were ever added, or all of the
4885 * ones that were added completed synchronously), then it is destroyed
4886 * immediately. Otherwise it is added to the ofproto's list of pending
4889 ofopgroup_submit(struct ofopgroup *group)
4891 if (!group->n_running) {
4892 ofopgroup_complete(group);
4894 list_push_back(&group->ofproto->pending, &group->ofproto_node);
4895 group->ofproto->n_pending++;
4900 ofopgroup_complete(struct ofopgroup *group)
4902 struct ofproto *ofproto = group->ofproto;
4904 struct ofconn *abbrev_ofconn;
4905 ovs_be32 abbrev_xid;
4907 struct ofoperation *op, *next_op;
4910 ovs_assert(!group->n_running);
4913 LIST_FOR_EACH (op, group_node, &group->ops) {
4920 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
4921 LIST_FOR_EACH (op, group_node, &group->ops) {
4922 if (op->type != OFOPERATION_DELETE) {
4923 struct ofpbuf *packet;
4926 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
4930 error = rule_execute(op->rule, in_port, packet);
4937 if (!error && !list_is_empty(&group->ofconn_node)) {
4938 abbrev_ofconn = group->ofconn;
4939 abbrev_xid = group->request->xid;
4941 abbrev_ofconn = NULL;
4942 abbrev_xid = htonl(0);
4944 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
4945 struct rule *rule = op->rule;
4947 /* We generally want to report the change to active OpenFlow flow
4948 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
4950 - The operation failed.
4952 - The affected rule is not visible to controllers.
4954 - The operation's only effect was to update rule->modified. */
4956 || ofproto_rule_is_hidden(rule)
4957 || (op->type == OFOPERATION_MODIFY
4959 && rule->flow_cookie == op->flow_cookie))) {
4960 /* Check that we can just cast from ofoperation_type to
4961 * nx_flow_update_event. */
4962 enum nx_flow_update_event event_type;
4965 case OFOPERATION_ADD:
4966 case OFOPERATION_REPLACE:
4967 event_type = NXFME_ADDED;
4970 case OFOPERATION_DELETE:
4971 event_type = NXFME_DELETED;
4974 case OFOPERATION_MODIFY:
4975 event_type = NXFME_MODIFIED;
4982 ofmonitor_report(ofproto->connmgr, rule, event_type,
4983 op->reason, abbrev_ofconn, abbrev_xid);
4986 rule->pending = NULL;
4989 case OFOPERATION_ADD:
4993 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
4994 if (vid_mask == VLAN_VID_MASK) {
4995 if (ofproto->vlan_bitmap) {
4996 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
4997 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
4998 bitmap_set1(ofproto->vlan_bitmap, vid);
4999 ofproto->vlans_changed = true;
5002 ofproto->vlans_changed = true;
5006 ovs_rwlock_wrlock(&rule->evict);
5007 oftable_remove_rule(rule);
5008 ofproto_rule_destroy__(rule);
5012 case OFOPERATION_DELETE:
5013 ovs_assert(!op->error);
5014 ofproto_rule_destroy__(rule);
5018 case OFOPERATION_MODIFY:
5019 case OFOPERATION_REPLACE:
5021 long long int now = time_msec();
5023 rule->modified = now;
5024 if (op->type == OFOPERATION_REPLACE) {
5025 rule->created = rule->used = now;
5028 ofproto_rule_change_cookie(ofproto, rule, op->flow_cookie);
5029 ovs_mutex_lock(&rule->timeout_mutex);
5030 rule->idle_timeout = op->idle_timeout;
5031 rule->hard_timeout = op->hard_timeout;
5032 ovs_mutex_unlock(&rule->timeout_mutex);
5034 free(rule->ofpacts);
5035 rule->ofpacts = op->ofpacts;
5036 rule->ofpacts_len = op->ofpacts_len;
5038 op->ofpacts_len = 0;
5040 rule->send_flow_removed = op->send_flow_removed;
5048 ofoperation_destroy(op);
5051 ofmonitor_flush(ofproto->connmgr);
5053 if (!list_is_empty(&group->ofproto_node)) {
5054 ovs_assert(ofproto->n_pending > 0);
5055 ofproto->n_pending--;
5056 list_remove(&group->ofproto_node);
5058 if (!list_is_empty(&group->ofconn_node)) {
5059 list_remove(&group->ofconn_node);
5061 ofconn_send_error(group->ofconn, group->request, error);
5063 connmgr_retry(ofproto->connmgr);
5065 free(group->request);
5069 /* Initiates a new operation on 'rule', of the specified 'type', within
5070 * 'group'. Prior to calling, 'rule' must not have any pending operation.
5072 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
5073 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
5075 * Returns the newly created ofoperation (which is also available as
5076 * rule->pending). */
5077 static struct ofoperation *
5078 ofoperation_create(struct ofopgroup *group, struct rule *rule,
5079 enum ofoperation_type type,
5080 enum ofp_flow_removed_reason reason)
5082 struct ofproto *ofproto = group->ofproto;
5083 struct ofoperation *op;
5085 ovs_assert(!rule->pending);
5087 op = rule->pending = xzalloc(sizeof *op);
5089 list_push_back(&group->ops, &op->group_node);
5092 op->reason = reason;
5093 op->flow_cookie = rule->flow_cookie;
5094 ovs_mutex_lock(&rule->timeout_mutex);
5095 op->idle_timeout = rule->idle_timeout;
5096 op->hard_timeout = rule->hard_timeout;
5097 ovs_mutex_unlock(&rule->timeout_mutex);
5098 op->send_flow_removed = rule->send_flow_removed;
5102 if (type == OFOPERATION_DELETE) {
5103 hmap_insert(&ofproto->deletions, &op->hmap_node,
5104 cls_rule_hash(&rule->cr, rule->table_id));
5111 ofoperation_destroy(struct ofoperation *op)
5113 struct ofopgroup *group = op->group;
5116 op->rule->pending = NULL;
5118 if (op->type == OFOPERATION_DELETE) {
5119 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
5121 list_remove(&op->group_node);
5126 /* Indicates that 'op' completed with status 'error', which is either 0 to
5127 * indicate success or an OpenFlow error code on failure.
5129 * If 'error' is 0, indicating success, the operation will be committed
5130 * permanently to the flow table.
5132 * If 'error' is nonzero, then generally the operation will be rolled back:
5134 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
5135 * restores the original rule. The caller must have uninitialized any
5136 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
5137 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
5138 * and 7 for the new rule, calling its ->rule_dealloc() function.
5140 * - If 'op' is a "modify flow" operation, ofproto restores the original
5143 * - 'op' must not be a "delete flow" operation. Removing a rule is not
5144 * allowed to fail. It must always succeed.
5146 * Please see the large comment in ofproto/ofproto-provider.h titled
5147 * "Asynchronous Operation Support" for more information. */
5149 ofoperation_complete(struct ofoperation *op, enum ofperr error)
5151 struct ofopgroup *group = op->group;
5153 ovs_assert(op->rule->pending == op);
5154 ovs_assert(group->n_running > 0);
5155 ovs_assert(!error || op->type != OFOPERATION_DELETE);
5158 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
5159 ofopgroup_complete(group);
5164 pick_datapath_id(const struct ofproto *ofproto)
5166 const struct ofport *port;
5168 port = ofproto_get_port(ofproto, OFPP_LOCAL);
5170 uint8_t ea[ETH_ADDR_LEN];
5173 error = netdev_get_etheraddr(port->netdev, ea);
5175 return eth_addr_to_uint64(ea);
5177 VLOG_WARN("%s: could not get MAC address for %s (%s)",
5178 ofproto->name, netdev_get_name(port->netdev),
5179 ovs_strerror(error));
5181 return ofproto->fallback_dpid;
5185 pick_fallback_dpid(void)
5187 uint8_t ea[ETH_ADDR_LEN];
5188 eth_addr_nicira_random(ea);
5189 return eth_addr_to_uint64(ea);
5192 /* Table overflow policy. */
5194 /* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep'
5195 * to NULL if the table is not configured to evict rules or if the table
5196 * contains no evictable rules. (Rules with a readlock on their evict rwlock,
5197 * or with no timeouts are not evictable.) */
5199 choose_rule_to_evict(struct oftable *table, struct rule **rulep)
5201 struct eviction_group *evg;
5204 if (!table->eviction_fields) {
5208 /* In the common case, the outer and inner loops here will each be entered
5211 * - The inner loop normally "return"s in its first iteration. If the
5212 * eviction group has any evictable rules, then it always returns in
5215 * - The outer loop only iterates more than once if the largest eviction
5216 * group has no evictable rules.
5218 * - The outer loop can exit only if table's 'max_flows' is all filled up
5219 * by unevictable rules. */
5220 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
5223 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
5224 if (!ovs_rwlock_trywrlock(&rule->evict)) {
5234 /* Searches 'ofproto' for tables that have more flows than their configured
5235 * maximum and that have flow eviction enabled, and evicts as many flows as
5236 * necessary and currently feasible from them.
5238 * This triggers only when an OpenFlow table has N flows in it and then the
5239 * client configures a maximum number of flows less than N. */
5241 ofproto_evict(struct ofproto *ofproto)
5243 struct ofopgroup *group;
5244 struct oftable *table;
5246 group = ofopgroup_create_unattached(ofproto);
5247 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
5248 while (table->eviction_fields) {
5252 ovs_rwlock_rdlock(&table->cls.rwlock);
5253 n_rules = classifier_count(&table->cls);
5254 ovs_rwlock_unlock(&table->cls.rwlock);
5256 if (n_rules <= table->max_flows) {
5260 if (!choose_rule_to_evict(table, &rule)) {
5264 if (rule->pending) {
5265 ovs_rwlock_unlock(&rule->evict);
5269 ofoperation_create(group, rule,
5270 OFOPERATION_DELETE, OFPRR_EVICTION);
5271 oftable_remove_rule(rule);
5272 ofproto->ofproto_class->rule_delete(rule);
5275 ofopgroup_submit(group);
5278 /* Eviction groups. */
5280 /* Returns the priority to use for an eviction_group that contains 'n_rules'
5281 * rules. The priority contains low-order random bits to ensure that eviction
5282 * groups with the same number of rules are prioritized randomly. */
5284 eviction_group_priority(size_t n_rules)
5286 uint16_t size = MIN(UINT16_MAX, n_rules);
5287 return (size << 16) | random_uint16();
5290 /* Updates 'evg', an eviction_group within 'table', following a change that
5291 * adds or removes rules in 'evg'. */
5293 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
5295 heap_change(&table->eviction_groups_by_size, &evg->size_node,
5296 eviction_group_priority(heap_count(&evg->rules)));
5299 /* Destroys 'evg', an eviction_group within 'table':
5301 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
5302 * rules themselves, just removes them from the eviction group.)
5304 * - Removes 'evg' from 'table'.
5308 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
5310 while (!heap_is_empty(&evg->rules)) {
5313 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
5314 rule->eviction_group = NULL;
5316 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
5317 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
5318 heap_destroy(&evg->rules);
5322 /* Removes 'rule' from its eviction group, if any. */
5324 eviction_group_remove_rule(struct rule *rule)
5326 if (rule->eviction_group) {
5327 struct oftable *table = &rule->ofproto->tables[rule->table_id];
5328 struct eviction_group *evg = rule->eviction_group;
5330 rule->eviction_group = NULL;
5331 heap_remove(&evg->rules, &rule->evg_node);
5332 if (heap_is_empty(&evg->rules)) {
5333 eviction_group_destroy(table, evg);
5335 eviction_group_resized(table, evg);
5340 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
5341 * returns the hash value. */
5343 eviction_group_hash_rule(struct rule *rule)
5345 struct oftable *table = &rule->ofproto->tables[rule->table_id];
5346 const struct mf_subfield *sf;
5350 hash = table->eviction_group_id_basis;
5351 miniflow_expand(&rule->cr.match.flow, &flow);
5352 for (sf = table->eviction_fields;
5353 sf < &table->eviction_fields[table->n_eviction_fields];
5356 if (mf_are_prereqs_ok(sf->field, &flow)) {
5357 union mf_value value;
5359 mf_get_value(sf->field, &flow, &value);
5361 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
5363 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
5364 unsigned int start = sf->ofs + sf->n_bits;
5365 bitwise_zero(&value, sf->field->n_bytes, start,
5366 sf->field->n_bytes * 8 - start);
5368 hash = hash_bytes(&value, sf->field->n_bytes, hash);
5370 hash = hash_int(hash, 0);
5377 /* Returns an eviction group within 'table' with the given 'id', creating one
5379 static struct eviction_group *
5380 eviction_group_find(struct oftable *table, uint32_t id)
5382 struct eviction_group *evg;
5384 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
5388 evg = xmalloc(sizeof *evg);
5389 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
5390 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
5391 eviction_group_priority(0));
5392 heap_init(&evg->rules);
5397 /* Returns an eviction priority for 'rule'. The return value should be
5398 * interpreted so that higher priorities make a rule more attractive candidates
5401 rule_eviction_priority(struct rule *rule)
5403 long long int hard_expiration;
5404 long long int idle_expiration;
5405 long long int expiration;
5406 uint32_t expiration_offset;
5408 /* Calculate time of expiration. */
5409 ovs_mutex_lock(&rule->timeout_mutex);
5410 hard_expiration = (rule->hard_timeout
5411 ? rule->modified + rule->hard_timeout * 1000
5413 idle_expiration = (rule->idle_timeout
5414 ? rule->used + rule->idle_timeout * 1000
5416 expiration = MIN(hard_expiration, idle_expiration);
5417 ovs_mutex_unlock(&rule->timeout_mutex);
5418 if (expiration == LLONG_MAX) {
5422 /* Calculate the time of expiration as a number of (approximate) seconds
5423 * after program startup.
5425 * This should work OK for program runs that last UINT32_MAX seconds or
5426 * less. Therefore, please restart OVS at least once every 136 years. */
5427 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
5429 /* Invert the expiration offset because we're using a max-heap. */
5430 return UINT32_MAX - expiration_offset;
5433 /* Adds 'rule' to an appropriate eviction group for its oftable's
5434 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
5435 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
5438 * The caller must ensure that 'rule' is not already in an eviction group. */
5440 eviction_group_add_rule(struct rule *rule)
5442 struct ofproto *ofproto = rule->ofproto;
5443 struct oftable *table = &ofproto->tables[rule->table_id];
5446 ovs_mutex_lock(&rule->timeout_mutex);
5447 has_timeout = rule->hard_timeout || rule->idle_timeout;
5448 ovs_mutex_unlock(&rule->timeout_mutex);
5450 if (table->eviction_fields && has_timeout) {
5451 struct eviction_group *evg;
5453 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
5455 rule->eviction_group = evg;
5456 heap_insert(&evg->rules, &rule->evg_node,
5457 rule_eviction_priority(rule));
5458 eviction_group_resized(table, evg);
5464 /* Initializes 'table'. */
5466 oftable_init(struct oftable *table)
5468 memset(table, 0, sizeof *table);
5469 classifier_init(&table->cls);
5470 table->max_flows = UINT_MAX;
5473 /* Destroys 'table', including its classifier and eviction groups.
5475 * The caller is responsible for freeing 'table' itself. */
5477 oftable_destroy(struct oftable *table)
5479 ovs_rwlock_rdlock(&table->cls.rwlock);
5480 ovs_assert(classifier_is_empty(&table->cls));
5481 ovs_rwlock_unlock(&table->cls.rwlock);
5482 oftable_disable_eviction(table);
5483 classifier_destroy(&table->cls);
5487 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
5488 * string, then 'table' will use its default name.
5490 * This only affects the name exposed for a table exposed through the OpenFlow
5491 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
5493 oftable_set_name(struct oftable *table, const char *name)
5495 if (name && name[0]) {
5496 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
5497 if (!table->name || strncmp(name, table->name, len)) {
5499 table->name = xmemdup0(name, len);
5507 /* oftables support a choice of two policies when adding a rule would cause the
5508 * number of flows in the table to exceed the configured maximum number: either
5509 * they can refuse to add the new flow or they can evict some existing flow.
5510 * This function configures the former policy on 'table'. */
5512 oftable_disable_eviction(struct oftable *table)
5514 if (table->eviction_fields) {
5515 struct eviction_group *evg, *next;
5517 HMAP_FOR_EACH_SAFE (evg, next, id_node,
5518 &table->eviction_groups_by_id) {
5519 eviction_group_destroy(table, evg);
5521 hmap_destroy(&table->eviction_groups_by_id);
5522 heap_destroy(&table->eviction_groups_by_size);
5524 free(table->eviction_fields);
5525 table->eviction_fields = NULL;
5526 table->n_eviction_fields = 0;
5530 /* oftables support a choice of two policies when adding a rule would cause the
5531 * number of flows in the table to exceed the configured maximum number: either
5532 * they can refuse to add the new flow or they can evict some existing flow.
5533 * This function configures the latter policy on 'table', with fairness based
5534 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
5535 * 'n_fields' as 0 disables fairness.) */
5537 oftable_enable_eviction(struct oftable *table,
5538 const struct mf_subfield *fields, size_t n_fields)
5540 struct cls_cursor cursor;
5543 if (table->eviction_fields
5544 && n_fields == table->n_eviction_fields
5546 || !memcmp(fields, table->eviction_fields,
5547 n_fields * sizeof *fields))) {
5552 oftable_disable_eviction(table);
5554 table->n_eviction_fields = n_fields;
5555 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
5557 table->eviction_group_id_basis = random_uint32();
5558 hmap_init(&table->eviction_groups_by_id);
5559 heap_init(&table->eviction_groups_by_size);
5561 ovs_rwlock_rdlock(&table->cls.rwlock);
5562 cls_cursor_init(&cursor, &table->cls, NULL);
5563 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
5564 eviction_group_add_rule(rule);
5566 ovs_rwlock_unlock(&table->cls.rwlock);
5569 /* Removes 'rule' from the oftable that contains it. */
5571 oftable_remove_rule__(struct ofproto *ofproto, struct classifier *cls,
5573 OVS_REQ_WRLOCK(cls->rwlock) OVS_RELEASES(rule->evict)
5575 classifier_remove(cls, &rule->cr);
5576 if (rule->meter_id) {
5577 list_remove(&rule->meter_list_node);
5579 cookies_remove(ofproto, rule);
5580 eviction_group_remove_rule(rule);
5581 ovs_mutex_lock(&ofproto->expirable_mutex);
5582 if (!list_is_empty(&rule->expirable)) {
5583 list_remove(&rule->expirable);
5585 ovs_mutex_unlock(&ofproto->expirable_mutex);
5586 if (!list_is_empty(&rule->meter_list_node)) {
5587 list_remove(&rule->meter_list_node);
5589 ovs_rwlock_unlock(&rule->evict);
5593 oftable_remove_rule(struct rule *rule)
5595 struct ofproto *ofproto = rule->ofproto;
5596 struct oftable *table = &ofproto->tables[rule->table_id];
5598 ovs_rwlock_wrlock(&table->cls.rwlock);
5599 oftable_remove_rule__(ofproto, &table->cls, rule);
5600 ovs_rwlock_unlock(&table->cls.rwlock);
5603 /* Inserts 'rule' into its oftable, which must not already contain any rule for
5604 * the same cls_rule. */
5606 oftable_insert_rule(struct rule *rule)
5608 struct ofproto *ofproto = rule->ofproto;
5609 struct oftable *table = &ofproto->tables[rule->table_id];
5612 ovs_mutex_lock(&rule->timeout_mutex);
5613 may_expire = rule->hard_timeout || rule->idle_timeout;
5614 ovs_mutex_unlock(&rule->timeout_mutex);
5617 ovs_mutex_lock(&ofproto->expirable_mutex);
5618 list_insert(&ofproto->expirable, &rule->expirable);
5619 ovs_mutex_unlock(&ofproto->expirable_mutex);
5621 cookies_insert(ofproto, rule);
5622 if (rule->meter_id) {
5623 struct meter *meter = ofproto->meters[rule->meter_id];
5624 list_insert(&meter->rules, &rule->meter_list_node);
5626 ovs_rwlock_wrlock(&table->cls.rwlock);
5627 classifier_insert(&table->cls, &rule->cr);
5628 ovs_rwlock_unlock(&table->cls.rwlock);
5629 eviction_group_add_rule(rule);
5632 /* unixctl commands. */
5635 ofproto_lookup(const char *name)
5637 struct ofproto *ofproto;
5639 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
5641 if (!strcmp(ofproto->name, name)) {
5649 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
5650 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5652 struct ofproto *ofproto;
5656 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
5657 ds_put_format(&results, "%s\n", ofproto->name);
5659 unixctl_command_reply(conn, ds_cstr(&results));
5660 ds_destroy(&results);
5664 ofproto_unixctl_init(void)
5666 static bool registered;
5672 unixctl_command_register("ofproto/list", "", 0, 0,
5673 ofproto_unixctl_list, NULL);
5676 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5678 * This is deprecated. It is only for compatibility with broken device drivers
5679 * in old versions of Linux that do not properly support VLANs when VLAN
5680 * devices are not used. When broken device drivers are no longer in
5681 * widespread use, we will delete these interfaces. */
5683 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
5684 * (exactly) by an OpenFlow rule in 'ofproto'. */
5686 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
5688 const struct oftable *oftable;
5690 free(ofproto->vlan_bitmap);
5691 ofproto->vlan_bitmap = bitmap_allocate(4096);
5692 ofproto->vlans_changed = false;
5694 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
5695 const struct cls_table *table;
5697 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
5698 if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
5699 const struct cls_rule *rule;
5701 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
5702 uint16_t vid = miniflow_get_vid(&rule->match.flow);
5703 bitmap_set1(vlan_bitmap, vid);
5704 bitmap_set1(ofproto->vlan_bitmap, vid);
5711 /* Returns true if new VLANs have come into use by the flow table since the
5712 * last call to ofproto_get_vlan_usage().
5714 * We don't track when old VLANs stop being used. */
5716 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
5718 return ofproto->vlans_changed;
5721 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
5722 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
5723 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
5724 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
5725 * then the VLAN device is un-enslaved. */
5727 ofproto_port_set_realdev(struct ofproto *ofproto, ofp_port_t vlandev_ofp_port,
5728 ofp_port_t realdev_ofp_port, int vid)
5730 struct ofport *ofport;
5733 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
5735 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
5737 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
5738 ofproto->name, vlandev_ofp_port);
5742 if (!ofproto->ofproto_class->set_realdev) {
5743 if (!vlandev_ofp_port) {
5746 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
5750 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
5752 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
5753 ofproto->name, vlandev_ofp_port,
5754 netdev_get_name(ofport->netdev), ovs_strerror(error));