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 {
82 /* A single OpenFlow request can execute any number of operations. The
83 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
84 * ofconn to which an error reply should be sent if necessary.
86 * ofproto initiates some operations internally. These operations are still
87 * assigned to groups but will not have an associated ofconn. */
89 struct ofproto *ofproto; /* Owning ofproto. */
90 struct list ofproto_node; /* In ofproto's "pending" list. */
91 struct list ops; /* List of "struct ofoperation"s. */
92 int n_running; /* Number of ops still pending. */
94 /* Data needed to send OpenFlow reply on failure or to send a buffered
97 * If list_is_empty(ofconn_node) then this ofopgroup never had an
98 * associated ofconn or its ofconn's connection dropped after it initiated
99 * the operation. In the latter case 'ofconn' is a wild pointer that
100 * refers to freed memory, so the 'ofconn' member must be used only if
101 * !list_is_empty(ofconn_node).
103 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
104 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
105 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
106 uint32_t buffer_id; /* Buffer id from original request. */
109 static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
110 static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
111 const struct ofp_header *,
113 static void ofopgroup_submit(struct ofopgroup *);
114 static void ofopgroup_complete(struct ofopgroup *);
116 /* A single flow table operation. */
118 struct ofopgroup *group; /* Owning group. */
119 struct list group_node; /* In ofopgroup's "ops" list. */
120 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
121 struct rule *rule; /* Rule being operated upon. */
122 enum ofoperation_type type; /* Type of operation. */
124 /* OFOPERATION_ADD. */
125 struct rule *victim; /* Rule being replaced, if any.. */
127 /* OFOPERATION_MODIFY: The old actions, if the actions are changing. */
128 struct ofpact *ofpacts;
132 /* OFOPERATION_DELETE. */
133 enum ofp_flow_removed_reason reason; /* Reason flow was removed. */
135 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
136 enum ofperr error; /* 0 if no error. */
139 static struct ofoperation *ofoperation_create(struct ofopgroup *,
141 enum ofoperation_type,
142 enum ofp_flow_removed_reason);
143 static void ofoperation_destroy(struct ofoperation *);
146 static void oftable_init(struct oftable *);
147 static void oftable_destroy(struct oftable *);
149 static void oftable_set_name(struct oftable *, const char *name);
151 static void oftable_disable_eviction(struct oftable *);
152 static void oftable_enable_eviction(struct oftable *,
153 const struct mf_subfield *fields,
156 static void oftable_remove_rule(struct rule *rule) OVS_RELEASES(rule->evict);
157 static void oftable_remove_rule__(struct ofproto *ofproto,
158 struct classifier *cls, struct rule *rule)
159 OVS_REQ_WRLOCK(cls->rwlock) OVS_RELEASES(rule->evict);
160 static struct rule *oftable_replace_rule(struct rule *);
161 static void oftable_substitute_rule(struct rule *old, struct rule *new);
163 /* A set of rules within a single OpenFlow table (oftable) that have the same
164 * values for the oftable's eviction_fields. A rule to be evicted, when one is
165 * needed, is taken from the eviction group that contains the greatest number
168 * An oftable owns any number of eviction groups, each of which contains any
171 * Membership in an eviction group is imprecise, based on the hash of the
172 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
173 * That is, if two rules have different eviction_fields, but those
174 * eviction_fields hash to the same value, then they will belong to the same
175 * eviction_group anyway.
177 * (When eviction is not enabled on an oftable, we don't track any eviction
178 * groups, to save time and space.) */
179 struct eviction_group {
180 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
181 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
182 struct heap rules; /* Contains "struct rule"s. */
185 static bool choose_rule_to_evict(struct oftable *table, struct rule **rulep)
186 OVS_TRY_WRLOCK(true, (*rulep)->evict);
187 static void ofproto_evict(struct ofproto *);
188 static uint32_t rule_eviction_priority(struct rule *);
189 static void eviction_group_add_rule(struct rule *rule);
192 static void ofport_destroy__(struct ofport *);
193 static void ofport_destroy(struct ofport *);
195 static void update_port(struct ofproto *, const char *devname);
196 static int init_ports(struct ofproto *);
197 static void reinit_ports(struct ofproto *);
200 static void ofproto_rule_destroy__(struct rule *);
201 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
202 static bool rule_is_modifiable(const struct rule *);
205 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
206 struct ofputil_flow_mod *,
207 const struct ofp_header *);
208 static void delete_flow__(struct rule *rule, struct ofopgroup *,
209 enum ofp_flow_removed_reason)
210 OVS_RELEASES(rule->evict);
211 static bool handle_openflow(struct ofconn *, const struct ofpbuf *);
212 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
213 struct ofputil_flow_mod *,
214 const struct ofp_header *);
215 static void calc_duration(long long int start, long long int now,
216 uint32_t *sec, uint32_t *nsec);
219 static uint64_t pick_datapath_id(const struct ofproto *);
220 static uint64_t pick_fallback_dpid(void);
221 static void ofproto_destroy__(struct ofproto *);
222 static void update_mtu(struct ofproto *, struct ofport *);
223 static void meter_delete(struct ofproto *, uint32_t first, uint32_t last);
226 static void ofproto_unixctl_init(void);
228 /* All registered ofproto classes, in probe order. */
229 static const struct ofproto_class **ofproto_classes;
230 static size_t n_ofproto_classes;
231 static size_t allocated_ofproto_classes;
233 unsigned flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT;
234 unsigned n_handler_threads;
235 enum ofproto_flow_miss_model flow_miss_model = OFPROTO_HANDLE_MISS_AUTO;
237 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
238 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
240 /* Initial mappings of port to OpenFlow number mappings. */
241 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
243 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
245 /* The default value of true waits for flow restore. */
246 static bool flow_restore_wait = true;
248 /* Must be called to initialize the ofproto library.
250 * The caller may pass in 'iface_hints', which contains an shash of
251 * "iface_hint" elements indexed by the interface's name. The provider
252 * may use these hints to describe the startup configuration in order to
253 * reinitialize its state. The caller owns the provided data, so a
254 * provider will make copies of anything required. An ofproto provider
255 * will remove any existing state that is not described by the hint, and
256 * may choose to remove it all. */
258 ofproto_init(const struct shash *iface_hints)
260 struct shash_node *node;
263 ofproto_class_register(&ofproto_dpif_class);
265 /* Make a local copy, since we don't own 'iface_hints' elements. */
266 SHASH_FOR_EACH(node, iface_hints) {
267 const struct iface_hint *orig_hint = node->data;
268 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
269 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
271 new_hint->br_name = xstrdup(orig_hint->br_name);
272 new_hint->br_type = xstrdup(br_type);
273 new_hint->ofp_port = orig_hint->ofp_port;
275 shash_add(&init_ofp_ports, node->name, new_hint);
278 for (i = 0; i < n_ofproto_classes; i++) {
279 ofproto_classes[i]->init(&init_ofp_ports);
283 /* 'type' should be a normalized datapath type, as returned by
284 * ofproto_normalize_type(). Returns the corresponding ofproto_class
285 * structure, or a null pointer if there is none registered for 'type'. */
286 static const struct ofproto_class *
287 ofproto_class_find__(const char *type)
291 for (i = 0; i < n_ofproto_classes; i++) {
292 const struct ofproto_class *class = ofproto_classes[i];
297 class->enumerate_types(&types);
298 found = sset_contains(&types, type);
299 sset_destroy(&types);
305 VLOG_WARN("unknown datapath type %s", type);
309 /* Registers a new ofproto class. After successful registration, new ofprotos
310 * of that type can be created using ofproto_create(). */
312 ofproto_class_register(const struct ofproto_class *new_class)
316 for (i = 0; i < n_ofproto_classes; i++) {
317 if (ofproto_classes[i] == new_class) {
322 if (n_ofproto_classes >= allocated_ofproto_classes) {
323 ofproto_classes = x2nrealloc(ofproto_classes,
324 &allocated_ofproto_classes,
325 sizeof *ofproto_classes);
327 ofproto_classes[n_ofproto_classes++] = new_class;
331 /* Unregisters a datapath provider. 'type' must have been previously
332 * registered and not currently be in use by any ofprotos. After
333 * unregistration new datapaths of that type cannot be opened using
334 * ofproto_create(). */
336 ofproto_class_unregister(const struct ofproto_class *class)
340 for (i = 0; i < n_ofproto_classes; i++) {
341 if (ofproto_classes[i] == class) {
342 for (i++; i < n_ofproto_classes; i++) {
343 ofproto_classes[i - 1] = ofproto_classes[i];
349 VLOG_WARN("attempted to unregister an ofproto class that is not "
354 /* Clears 'types' and enumerates all registered ofproto types into it. The
355 * caller must first initialize the sset. */
357 ofproto_enumerate_types(struct sset *types)
361 for (i = 0; i < n_ofproto_classes; i++) {
362 ofproto_classes[i]->enumerate_types(types);
366 /* Returns the fully spelled out name for the given ofproto 'type'.
368 * Normalized type string can be compared with strcmp(). Unnormalized type
369 * string might be the same even if they have different spellings. */
371 ofproto_normalize_type(const char *type)
373 return type && type[0] ? type : "system";
376 /* Clears 'names' and enumerates the names of all known created ofprotos with
377 * the given 'type'. The caller must first initialize the sset. Returns 0 if
378 * successful, otherwise a positive errno value.
380 * Some kinds of datapaths might not be practically enumerable. This is not
381 * considered an error. */
383 ofproto_enumerate_names(const char *type, struct sset *names)
385 const struct ofproto_class *class = ofproto_class_find__(type);
386 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
390 ofproto_create(const char *datapath_name, const char *datapath_type,
391 struct ofproto **ofprotop)
393 const struct ofproto_class *class;
394 struct ofproto *ofproto;
400 ofproto_unixctl_init();
402 datapath_type = ofproto_normalize_type(datapath_type);
403 class = ofproto_class_find__(datapath_type);
405 VLOG_WARN("could not create datapath %s of unknown type %s",
406 datapath_name, datapath_type);
410 ofproto = class->alloc();
412 VLOG_ERR("failed to allocate datapath %s of type %s",
413 datapath_name, datapath_type);
418 memset(ofproto, 0, sizeof *ofproto);
419 ofproto->ofproto_class = class;
420 ofproto->name = xstrdup(datapath_name);
421 ofproto->type = xstrdup(datapath_type);
422 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
423 hash_string(ofproto->name, 0));
424 ofproto->datapath_id = 0;
425 ofproto->forward_bpdu = false;
426 ofproto->fallback_dpid = pick_fallback_dpid();
427 ofproto->mfr_desc = NULL;
428 ofproto->hw_desc = NULL;
429 ofproto->sw_desc = NULL;
430 ofproto->serial_desc = NULL;
431 ofproto->dp_desc = NULL;
432 ofproto->frag_handling = OFPC_FRAG_NORMAL;
433 hmap_init(&ofproto->ports);
434 shash_init(&ofproto->port_by_name);
435 simap_init(&ofproto->ofp_requests);
436 ofproto->max_ports = ofp_to_u16(OFPP_MAX);
437 ofproto->tables = NULL;
438 ofproto->n_tables = 0;
439 hindex_init(&ofproto->cookies);
440 list_init(&ofproto->expirable);
441 ovs_mutex_init_recursive(&ofproto->expirable_mutex);
442 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
443 ofproto->state = S_OPENFLOW;
444 list_init(&ofproto->pending);
445 ofproto->n_pending = 0;
446 hmap_init(&ofproto->deletions);
447 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
448 ofproto->first_op = ofproto->last_op = LLONG_MIN;
449 ofproto->next_op_report = LLONG_MAX;
450 ofproto->op_backoff = LLONG_MIN;
451 ofproto->vlan_bitmap = NULL;
452 ofproto->vlans_changed = false;
453 ofproto->min_mtu = INT_MAX;
455 error = ofproto->ofproto_class->construct(ofproto);
457 VLOG_ERR("failed to open datapath %s: %s",
458 datapath_name, ovs_strerror(error));
459 ofproto_destroy__(ofproto);
463 /* The "max_ports" member should have been set by ->construct(ofproto).
464 * Port 0 is not a valid OpenFlow port, so mark that as unavailable. */
465 ofproto->ofp_port_ids = bitmap_allocate(ofproto->max_ports);
466 bitmap_set1(ofproto->ofp_port_ids, 0);
468 /* Check that hidden tables, if any, are at the end. */
469 ovs_assert(ofproto->n_tables);
470 for (i = 0; i + 1 < ofproto->n_tables; i++) {
471 enum oftable_flags flags = ofproto->tables[i].flags;
472 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
474 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
477 ofproto->datapath_id = pick_datapath_id(ofproto);
480 /* Initialize meters table. */
481 if (ofproto->ofproto_class->meter_get_features) {
482 ofproto->ofproto_class->meter_get_features(ofproto,
483 &ofproto->meter_features);
485 memset(&ofproto->meter_features, 0, sizeof ofproto->meter_features);
487 ofproto->meters = xzalloc((ofproto->meter_features.max_meters + 1)
488 * sizeof(struct meter *));
494 /* Must be called (only) by an ofproto implementation in its constructor
495 * function. See the large comment on 'construct' in struct ofproto_class for
498 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
500 struct oftable *table;
502 ovs_assert(!ofproto->n_tables);
503 ovs_assert(n_tables >= 1 && n_tables <= 255);
505 ofproto->n_tables = n_tables;
506 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
507 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
512 /* To be optionally called (only) by an ofproto implementation in its
513 * constructor function. See the large comment on 'construct' in struct
514 * ofproto_class for details.
516 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
517 * will then ensure that actions passed into the ofproto implementation will
518 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
519 * function is not called, there will be no such restriction.
521 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
522 * the 'max_ports' restriction. */
524 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
526 ovs_assert(max_ports <= ofp_to_u16(OFPP_MAX));
527 ofproto->max_ports = max_ports;
531 ofproto_get_datapath_id(const struct ofproto *ofproto)
533 return ofproto->datapath_id;
537 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
539 uint64_t old_dpid = p->datapath_id;
540 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
541 if (p->datapath_id != old_dpid) {
542 /* Force all active connections to reconnect, since there is no way to
543 * notify a controller that the datapath ID has changed. */
544 ofproto_reconnect_controllers(p);
549 ofproto_set_controllers(struct ofproto *p,
550 const struct ofproto_controller *controllers,
551 size_t n_controllers, uint32_t allowed_versions)
553 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
558 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
560 connmgr_set_fail_mode(p->connmgr, fail_mode);
563 /* Drops the connections between 'ofproto' and all of its controllers, forcing
564 * them to reconnect. */
566 ofproto_reconnect_controllers(struct ofproto *ofproto)
568 connmgr_reconnect(ofproto->connmgr);
571 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
572 * in-band control should guarantee access, in the same way that in-band
573 * control guarantees access to OpenFlow controllers. */
575 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
576 const struct sockaddr_in *extras, size_t n)
578 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
581 /* Sets the OpenFlow queue used by flows set up by in-band control on
582 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
583 * flows will use the default queue. */
585 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
587 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
590 /* Sets the number of flows at which eviction from the kernel flow table
593 ofproto_set_flow_eviction_threshold(unsigned threshold)
595 flow_eviction_threshold = MAX(OFPROTO_FLOW_EVICTION_THRESHOLD_MIN,
599 /* Sets the path for handling flow misses. */
601 ofproto_set_flow_miss_model(unsigned model)
603 flow_miss_model = model;
606 /* If forward_bpdu is true, the NORMAL action will forward frames with
607 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
608 * the NORMAL action will drop these frames. */
610 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
612 bool old_val = ofproto->forward_bpdu;
613 ofproto->forward_bpdu = forward_bpdu;
614 if (old_val != ofproto->forward_bpdu) {
615 if (ofproto->ofproto_class->forward_bpdu_changed) {
616 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
621 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
622 * 'idle_time', in seconds, and the maximum number of MAC table entries to
625 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
628 if (ofproto->ofproto_class->set_mac_table_config) {
629 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
634 /* Sets number of upcall handler threads. The default is
635 * (number of online cores - 1). */
637 ofproto_set_n_handler_threads(unsigned limit)
640 n_handler_threads = limit;
642 n_handler_threads = MAX(1, sysconf(_SC_NPROCESSORS_ONLN) - 1);
647 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
650 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
654 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
656 return connmgr_set_snoops(ofproto->connmgr, snoops);
660 ofproto_set_netflow(struct ofproto *ofproto,
661 const struct netflow_options *nf_options)
663 if (nf_options && sset_is_empty(&nf_options->collectors)) {
667 if (ofproto->ofproto_class->set_netflow) {
668 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
670 return nf_options ? EOPNOTSUPP : 0;
675 ofproto_set_sflow(struct ofproto *ofproto,
676 const struct ofproto_sflow_options *oso)
678 if (oso && sset_is_empty(&oso->targets)) {
682 if (ofproto->ofproto_class->set_sflow) {
683 return ofproto->ofproto_class->set_sflow(ofproto, oso);
685 return oso ? EOPNOTSUPP : 0;
690 ofproto_set_ipfix(struct ofproto *ofproto,
691 const struct ofproto_ipfix_bridge_exporter_options *bo,
692 const struct ofproto_ipfix_flow_exporter_options *fo,
695 if (ofproto->ofproto_class->set_ipfix) {
696 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
698 return (bo || fo) ? EOPNOTSUPP : 0;
703 ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
705 flow_restore_wait = flow_restore_wait_db;
709 ofproto_get_flow_restore_wait(void)
711 return flow_restore_wait;
715 /* Spanning Tree Protocol (STP) configuration. */
717 /* Configures STP on 'ofproto' using the settings defined in 's'. If
718 * 's' is NULL, disables STP.
720 * Returns 0 if successful, otherwise a positive errno value. */
722 ofproto_set_stp(struct ofproto *ofproto,
723 const struct ofproto_stp_settings *s)
725 return (ofproto->ofproto_class->set_stp
726 ? ofproto->ofproto_class->set_stp(ofproto, s)
730 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
731 * 'enabled' member of 's' is false, then the other members are not
734 * Returns 0 if successful, otherwise a positive errno value. */
736 ofproto_get_stp_status(struct ofproto *ofproto,
737 struct ofproto_stp_status *s)
739 return (ofproto->ofproto_class->get_stp_status
740 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
744 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
745 * in 's'. The caller is responsible for assigning STP port numbers
746 * (using the 'port_num' member in the range of 1 through 255, inclusive)
747 * and ensuring there are no duplicates. If the 's' is NULL, then STP
748 * is disabled on the port.
750 * Returns 0 if successful, otherwise a positive errno value.*/
752 ofproto_port_set_stp(struct ofproto *ofproto, ofp_port_t ofp_port,
753 const struct ofproto_port_stp_settings *s)
755 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
757 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
758 ofproto->name, ofp_port);
762 return (ofproto->ofproto_class->set_stp_port
763 ? ofproto->ofproto_class->set_stp_port(ofport, s)
767 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
768 * 's'. If the 'enabled' member in 's' is false, then the other members
769 * are not meaningful.
771 * Returns 0 if successful, otherwise a positive errno value.*/
773 ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
774 struct ofproto_port_stp_status *s)
776 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
778 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
779 "port %"PRIu16, ofproto->name, ofp_port);
783 return (ofproto->ofproto_class->get_stp_port_status
784 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
788 /* Queue DSCP configuration. */
790 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
791 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
792 * to implement QoS. Instead, it is used to implement features which require
793 * knowledge of what queues exist on a port, and some basic information about
796 * Returns 0 if successful, otherwise a positive errno value. */
798 ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port,
799 const struct ofproto_port_queue *queues,
802 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
805 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
806 ofproto->name, ofp_port);
810 return (ofproto->ofproto_class->set_queues
811 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
815 /* Connectivity Fault Management configuration. */
817 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
819 ofproto_port_clear_cfm(struct ofproto *ofproto, ofp_port_t ofp_port)
821 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
822 if (ofport && ofproto->ofproto_class->set_cfm) {
823 ofproto->ofproto_class->set_cfm(ofport, NULL);
827 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
828 * basic configuration from the configuration members in 'cfm', and the remote
829 * maintenance point ID from remote_mpid. Ignores the statistics members of
832 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
834 ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port,
835 const struct cfm_settings *s)
837 struct ofport *ofport;
840 ofport = ofproto_get_port(ofproto, ofp_port);
842 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
843 ofproto->name, ofp_port);
847 /* XXX: For configuration simplicity, we only support one remote_mpid
848 * outside of the CFM module. It's not clear if this is the correct long
849 * term solution or not. */
850 error = (ofproto->ofproto_class->set_cfm
851 ? ofproto->ofproto_class->set_cfm(ofport, s)
854 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
855 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
856 ovs_strerror(error));
860 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
861 * 'ofproto' does not have a port 'ofp_port'. */
863 ofproto_port_set_bfd(struct ofproto *ofproto, ofp_port_t ofp_port,
864 const struct smap *cfg)
866 struct ofport *ofport;
869 ofport = ofproto_get_port(ofproto, ofp_port);
871 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
872 ofproto->name, ofp_port);
876 error = (ofproto->ofproto_class->set_bfd
877 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
880 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
881 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
882 ovs_strerror(error));
886 /* Populates 'status' with key value pairs indicating the status of the BFD
887 * session on 'ofp_port'. This information is intended to be populated in the
888 * OVS database. Has no effect if 'ofp_port' is not na OpenFlow port in
891 ofproto_port_get_bfd_status(struct ofproto *ofproto, ofp_port_t ofp_port,
894 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
895 return (ofport && ofproto->ofproto_class->get_bfd_status
896 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
900 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
901 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
902 * 0 if LACP partner information is not current (generally indicating a
903 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
905 ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port)
907 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
908 return (ofport && ofproto->ofproto_class->port_is_lacp_current
909 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
915 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
916 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
917 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
918 * configuration plus, if there is more than one slave, a bonding
921 * If 'aux' is already registered then this function updates its configuration
922 * to 's'. Otherwise, this function registers a new bundle.
924 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
927 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
928 const struct ofproto_bundle_settings *s)
930 return (ofproto->ofproto_class->bundle_set
931 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
935 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
936 * If no such bundle has been registered, this has no effect. */
938 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
940 return ofproto_bundle_register(ofproto, aux, NULL);
944 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
945 * If 'aux' is already registered then this function updates its configuration
946 * to 's'. Otherwise, this function registers a new mirror. */
948 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
949 const struct ofproto_mirror_settings *s)
951 return (ofproto->ofproto_class->mirror_set
952 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
956 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
957 * If no mirror has been registered, this has no effect. */
959 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
961 return ofproto_mirror_register(ofproto, aux, NULL);
964 /* Retrieves statistics from mirror associated with client data pointer
965 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
966 * 'bytes', respectively. If a particular counters is not supported,
967 * the appropriate argument is set to UINT64_MAX. */
969 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
970 uint64_t *packets, uint64_t *bytes)
972 if (!ofproto->ofproto_class->mirror_get_stats) {
973 *packets = *bytes = UINT64_MAX;
977 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
981 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
982 * which all packets are flooded, instead of using MAC learning. If
983 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
985 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
988 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
990 return (ofproto->ofproto_class->set_flood_vlans
991 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
995 /* Returns true if 'aux' is a registered bundle that is currently in use as the
996 * output for a mirror. */
998 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
1000 return (ofproto->ofproto_class->is_mirror_output_bundle
1001 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
1005 /* Configuration of OpenFlow tables. */
1007 /* Returns the number of OpenFlow tables in 'ofproto'. */
1009 ofproto_get_n_tables(const struct ofproto *ofproto)
1011 return ofproto->n_tables;
1014 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
1015 * settings from 's'. 'table_id' must be in the range 0 through the number of
1016 * OpenFlow tables in 'ofproto' minus 1, inclusive.
1018 * For read-only tables, only the name may be configured. */
1020 ofproto_configure_table(struct ofproto *ofproto, int table_id,
1021 const struct ofproto_table_settings *s)
1023 struct oftable *table;
1025 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
1026 table = &ofproto->tables[table_id];
1028 oftable_set_name(table, s->name);
1030 if (table->flags & OFTABLE_READONLY) {
1035 oftable_enable_eviction(table, s->groups, s->n_groups);
1037 oftable_disable_eviction(table);
1040 table->max_flows = s->max_flows;
1041 ovs_rwlock_rdlock(&table->cls.rwlock);
1042 if (classifier_count(&table->cls) > table->max_flows
1043 && table->eviction_fields) {
1044 /* 'table' contains more flows than allowed. We might not be able to
1045 * evict them right away because of the asynchronous nature of flow
1046 * table changes. Schedule eviction for later. */
1047 switch (ofproto->state) {
1049 ofproto->state = S_EVICT;
1053 /* We're already deleting flows, nothing more to do. */
1057 ovs_rwlock_unlock(&table->cls.rwlock);
1061 ofproto_has_snoops(const struct ofproto *ofproto)
1063 return connmgr_has_snoops(ofproto->connmgr);
1067 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1069 connmgr_get_snoops(ofproto->connmgr, snoops);
1073 ofproto_flush__(struct ofproto *ofproto)
1075 struct ofopgroup *group;
1076 struct oftable *table;
1078 if (ofproto->ofproto_class->flush) {
1079 ofproto->ofproto_class->flush(ofproto);
1082 group = ofopgroup_create_unattached(ofproto);
1083 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1084 struct rule *rule, *next_rule;
1085 struct cls_cursor cursor;
1087 if (table->flags & OFTABLE_HIDDEN) {
1091 ovs_rwlock_wrlock(&table->cls.rwlock);
1092 cls_cursor_init(&cursor, &table->cls, NULL);
1093 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1094 if (!rule->pending) {
1095 ofoperation_create(group, rule, OFOPERATION_DELETE,
1097 ovs_rwlock_wrlock(&rule->evict);
1098 oftable_remove_rule__(ofproto, &table->cls, rule);
1099 ofproto->ofproto_class->rule_destruct(rule);
1102 ovs_rwlock_unlock(&table->cls.rwlock);
1104 ofopgroup_submit(group);
1108 ofproto_destroy__(struct ofproto *ofproto)
1110 struct oftable *table;
1112 ovs_assert(list_is_empty(&ofproto->pending));
1113 ovs_assert(!ofproto->n_pending);
1115 if (ofproto->meters) {
1116 meter_delete(ofproto, 1, ofproto->meter_features.max_meters);
1117 free(ofproto->meters);
1120 connmgr_destroy(ofproto->connmgr);
1122 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1123 free(ofproto->name);
1124 free(ofproto->type);
1125 free(ofproto->mfr_desc);
1126 free(ofproto->hw_desc);
1127 free(ofproto->sw_desc);
1128 free(ofproto->serial_desc);
1129 free(ofproto->dp_desc);
1130 hmap_destroy(&ofproto->ports);
1131 shash_destroy(&ofproto->port_by_name);
1132 bitmap_free(ofproto->ofp_port_ids);
1133 simap_destroy(&ofproto->ofp_requests);
1135 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1136 oftable_destroy(table);
1138 free(ofproto->tables);
1140 hmap_destroy(&ofproto->deletions);
1142 free(ofproto->vlan_bitmap);
1144 ovs_mutex_destroy(&ofproto->expirable_mutex);
1145 ofproto->ofproto_class->dealloc(ofproto);
1149 ofproto_destroy(struct ofproto *p)
1151 struct ofport *ofport, *next_ofport;
1158 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1159 ofport_destroy(ofport);
1162 p->ofproto_class->destruct(p);
1163 ofproto_destroy__(p);
1166 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1167 * kernel datapath, for example, this destroys the datapath in the kernel, and
1168 * with the netdev-based datapath, it tears down the data structures that
1169 * represent the datapath.
1171 * The datapath should not be currently open as an ofproto. */
1173 ofproto_delete(const char *name, const char *type)
1175 const struct ofproto_class *class = ofproto_class_find__(type);
1176 return (!class ? EAFNOSUPPORT
1177 : !class->del ? EACCES
1178 : class->del(type, name));
1182 process_port_change(struct ofproto *ofproto, int error, char *devname)
1184 if (error == ENOBUFS) {
1185 reinit_ports(ofproto);
1186 } else if (!error) {
1187 update_port(ofproto, devname);
1193 ofproto_type_run(const char *datapath_type)
1195 const struct ofproto_class *class;
1198 datapath_type = ofproto_normalize_type(datapath_type);
1199 class = ofproto_class_find__(datapath_type);
1201 error = class->type_run ? class->type_run(datapath_type) : 0;
1202 if (error && error != EAGAIN) {
1203 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1204 datapath_type, ovs_strerror(error));
1210 ofproto_type_run_fast(const char *datapath_type)
1212 const struct ofproto_class *class;
1215 datapath_type = ofproto_normalize_type(datapath_type);
1216 class = ofproto_class_find__(datapath_type);
1218 error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
1219 if (error && error != EAGAIN) {
1220 VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
1221 datapath_type, ovs_strerror(error));
1227 ofproto_type_wait(const char *datapath_type)
1229 const struct ofproto_class *class;
1231 datapath_type = ofproto_normalize_type(datapath_type);
1232 class = ofproto_class_find__(datapath_type);
1234 if (class->type_wait) {
1235 class->type_wait(datapath_type);
1240 ofproto_run(struct ofproto *p)
1242 struct sset changed_netdevs;
1243 const char *changed_netdev;
1244 struct ofport *ofport;
1247 error = p->ofproto_class->run(p);
1248 if (error && error != EAGAIN) {
1249 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, ovs_strerror(error));
1252 if (p->ofproto_class->port_poll) {
1255 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1256 process_port_change(p, error, devname);
1260 /* Update OpenFlow port status for any port whose netdev has changed.
1262 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1263 * destroyed, so it's not safe to update ports directly from the
1264 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1265 * need this two-phase approach. */
1266 sset_init(&changed_netdevs);
1267 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1268 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1269 if (ofport->change_seq != change_seq) {
1270 ofport->change_seq = change_seq;
1271 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1274 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1275 update_port(p, changed_netdev);
1277 sset_destroy(&changed_netdevs);
1281 connmgr_run(p->connmgr, handle_openflow);
1285 connmgr_run(p->connmgr, NULL);
1287 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1288 p->state = S_OPENFLOW;
1293 connmgr_run(p->connmgr, NULL);
1295 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1296 connmgr_flushed(p->connmgr);
1297 p->state = S_OPENFLOW;
1305 if (time_msec() >= p->next_op_report) {
1306 long long int ago = (time_msec() - p->first_op) / 1000;
1307 long long int interval = (p->last_op - p->first_op) / 1000;
1311 ds_put_format(&s, "%d flow_mods ",
1312 p->n_add + p->n_delete + p->n_modify);
1313 if (interval == ago) {
1314 ds_put_format(&s, "in the last %lld s", ago);
1315 } else if (interval) {
1316 ds_put_format(&s, "in the %lld s starting %lld s ago",
1319 ds_put_format(&s, "%lld s ago", ago);
1322 ds_put_cstr(&s, " (");
1324 ds_put_format(&s, "%d adds, ", p->n_add);
1327 ds_put_format(&s, "%d deletes, ", p->n_delete);
1330 ds_put_format(&s, "%d modifications, ", p->n_modify);
1333 ds_put_char(&s, ')');
1335 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1338 p->n_add = p->n_delete = p->n_modify = 0;
1339 p->next_op_report = LLONG_MAX;
1345 /* Performs periodic activity required by 'ofproto' that needs to be done
1346 * with the least possible latency.
1348 * It makes sense to call this function a couple of times per poll loop, to
1349 * provide a significant performance boost on some benchmarks with the
1350 * ofproto-dpif implementation. */
1352 ofproto_run_fast(struct ofproto *p)
1356 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1357 if (error && error != EAGAIN) {
1358 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1359 p->name, ovs_strerror(error));
1365 ofproto_wait(struct ofproto *p)
1367 struct ofport *ofport;
1369 p->ofproto_class->wait(p);
1370 if (p->ofproto_class->port_poll_wait) {
1371 p->ofproto_class->port_poll_wait(p);
1374 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1375 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1376 poll_immediate_wake();
1382 connmgr_wait(p->connmgr, true);
1387 connmgr_wait(p->connmgr, false);
1388 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1389 poll_immediate_wake();
1396 ofproto_is_alive(const struct ofproto *p)
1398 return connmgr_has_controllers(p->connmgr);
1401 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1402 * memory_report(). */
1404 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1406 const struct oftable *table;
1407 unsigned int n_rules;
1409 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1410 simap_increase(usage, "ops",
1411 ofproto->n_pending + hmap_count(&ofproto->deletions));
1414 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1415 ovs_rwlock_rdlock(&table->cls.rwlock);
1416 n_rules += classifier_count(&table->cls);
1417 ovs_rwlock_unlock(&table->cls.rwlock);
1419 simap_increase(usage, "rules", n_rules);
1421 if (ofproto->ofproto_class->get_memory_usage) {
1422 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1425 connmgr_get_memory_usage(ofproto->connmgr, usage);
1429 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1432 connmgr_get_controller_info(ofproto->connmgr, info);
1436 ofproto_free_ofproto_controller_info(struct shash *info)
1438 connmgr_free_controller_info(info);
1441 /* Makes a deep copy of 'old' into 'port'. */
1443 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1445 port->name = xstrdup(old->name);
1446 port->type = xstrdup(old->type);
1447 port->ofp_port = old->ofp_port;
1450 /* Frees memory allocated to members of 'ofproto_port'.
1452 * Do not call this function on an ofproto_port obtained from
1453 * ofproto_port_dump_next(): that function retains ownership of the data in the
1456 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1458 free(ofproto_port->name);
1459 free(ofproto_port->type);
1462 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1464 * This function provides no status indication. An error status for the entire
1465 * dump operation is provided when it is completed by calling
1466 * ofproto_port_dump_done().
1469 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1470 const struct ofproto *ofproto)
1472 dump->ofproto = ofproto;
1473 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1477 /* Attempts to retrieve another port from 'dump', which must have been created
1478 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1479 * 'port' and returns true. On failure, returns false.
1481 * Failure might indicate an actual error or merely that the last port has been
1482 * dumped. An error status for the entire dump operation is provided when it
1483 * is completed by calling ofproto_port_dump_done().
1485 * The ofproto owns the data stored in 'port'. It will remain valid until at
1486 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1487 * ofproto_port_dump_done(). */
1489 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1490 struct ofproto_port *port)
1492 const struct ofproto *ofproto = dump->ofproto;
1498 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1501 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1507 /* Completes port table dump operation 'dump', which must have been created
1508 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1509 * error-free, otherwise a positive errno value describing the problem. */
1511 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1513 const struct ofproto *ofproto = dump->ofproto;
1515 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1518 return dump->error == EOF ? 0 : dump->error;
1521 /* Returns the type to pass to netdev_open() when a datapath of type
1522 * 'datapath_type' has a port of type 'port_type', for a few special
1523 * cases when a netdev type differs from a port type. For example, when
1524 * using the userspace datapath, a port of type "internal" needs to be
1527 * Returns either 'type' itself or a string literal, which must not be
1530 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1532 const struct ofproto_class *class;
1534 datapath_type = ofproto_normalize_type(datapath_type);
1535 class = ofproto_class_find__(datapath_type);
1540 return (class->port_open_type
1541 ? class->port_open_type(datapath_type, port_type)
1545 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1546 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1547 * the port's OpenFlow port number.
1549 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1550 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1551 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1552 * 'ofp_portp' is non-null). */
1554 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1555 ofp_port_t *ofp_portp)
1557 ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1560 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1562 const char *netdev_name = netdev_get_name(netdev);
1564 simap_put(&ofproto->ofp_requests, netdev_name,
1565 ofp_to_u16(ofp_port));
1566 update_port(ofproto, netdev_name);
1569 struct ofproto_port ofproto_port;
1571 ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
1573 *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
1574 ofproto_port_destroy(&ofproto_port);
1579 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1580 * initializes '*port' appropriately; on failure, returns a positive errno
1583 * The caller owns the data in 'ofproto_port' and must free it with
1584 * ofproto_port_destroy() when it is no longer needed. */
1586 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1587 struct ofproto_port *port)
1591 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1593 memset(port, 0, sizeof *port);
1598 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1599 * Returns 0 if successful, otherwise a positive errno. */
1601 ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port)
1603 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1604 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1605 struct simap_node *ofp_request_node;
1608 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1609 if (ofp_request_node) {
1610 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1613 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1614 if (!error && ofport) {
1615 /* 'name' is the netdev's name and update_port() is going to close the
1616 * netdev. Just in case update_port() refers to 'name' after it
1617 * destroys 'ofport', make a copy of it around the update_port()
1619 char *devname = xstrdup(name);
1620 update_port(ofproto, devname);
1626 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1627 * performs the 'n_actions' actions in 'actions'. The new flow will not
1630 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1631 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1632 * controllers; otherwise, it will be hidden.
1634 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1636 * This is a helper function for in-band control and fail-open. */
1638 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1639 unsigned int priority,
1640 const struct ofpact *ofpacts, size_t ofpacts_len)
1642 const struct rule *rule;
1644 ovs_rwlock_rdlock(&ofproto->tables[0].cls.rwlock);
1645 rule = rule_from_cls_rule(classifier_find_match_exactly(
1646 &ofproto->tables[0].cls, match, priority));
1647 ovs_rwlock_unlock(&ofproto->tables[0].cls.rwlock);
1648 if (!rule || !ofpacts_equal(rule->ofpacts, rule->ofpacts_len,
1649 ofpacts, ofpacts_len)) {
1650 struct ofputil_flow_mod fm;
1652 memset(&fm, 0, sizeof fm);
1654 fm.priority = priority;
1655 fm.buffer_id = UINT32_MAX;
1656 fm.ofpacts = xmemdup(ofpacts, ofpacts_len);
1657 fm.ofpacts_len = ofpacts_len;
1658 add_flow(ofproto, NULL, &fm, NULL);
1663 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1664 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1665 * operation cannot be initiated now but may be retried later.
1667 * This is a helper function for in-band control and fail-open. */
1669 ofproto_flow_mod(struct ofproto *ofproto, struct ofputil_flow_mod *fm)
1671 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1674 /* Searches for a rule with matching criteria exactly equal to 'target' in
1675 * ofproto's table 0 and, if it finds one, deletes it.
1677 * This is a helper function for in-band control and fail-open. */
1679 ofproto_delete_flow(struct ofproto *ofproto,
1680 const struct match *target, unsigned int priority)
1684 ovs_rwlock_rdlock(&ofproto->tables[0].cls.rwlock);
1685 rule = rule_from_cls_rule(classifier_find_match_exactly(
1686 &ofproto->tables[0].cls, target, priority));
1687 ovs_rwlock_unlock(&ofproto->tables[0].cls.rwlock);
1689 /* No such rule -> success. */
1691 } else if (rule->pending) {
1692 /* An operation on the rule is already pending -> failure.
1693 * Caller must retry later if it's important. */
1696 /* Initiate deletion -> success. */
1697 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
1698 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1699 ovs_rwlock_wrlock(&rule->evict);
1700 oftable_remove_rule(rule);
1701 ofproto->ofproto_class->rule_destruct(rule);
1702 ofopgroup_submit(group);
1708 /* Starts the process of deleting all of the flows from all of ofproto's flow
1709 * tables and then reintroducing the flows required by in-band control and
1710 * fail-open. The process will complete in a later call to ofproto_run(). */
1712 ofproto_flush_flows(struct ofproto *ofproto)
1714 COVERAGE_INC(ofproto_flush);
1715 ofproto->state = S_FLUSH;
1719 reinit_ports(struct ofproto *p)
1721 struct ofproto_port_dump dump;
1722 struct sset devnames;
1723 struct ofport *ofport;
1724 struct ofproto_port ofproto_port;
1725 const char *devname;
1727 COVERAGE_INC(ofproto_reinit_ports);
1729 sset_init(&devnames);
1730 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1731 sset_add(&devnames, netdev_get_name(ofport->netdev));
1733 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1734 sset_add(&devnames, ofproto_port.name);
1737 SSET_FOR_EACH (devname, &devnames) {
1738 update_port(p, devname);
1740 sset_destroy(&devnames);
1744 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
1748 port_idx = simap_get(&ofproto->ofp_requests, netdev_name);
1749 port_idx = port_idx ? port_idx : UINT16_MAX;
1751 if (port_idx >= ofproto->max_ports
1752 || bitmap_is_set(ofproto->ofp_port_ids, port_idx)) {
1753 uint16_t end_port_no = ofproto->alloc_port_no;
1755 /* Search for a free OpenFlow port number. We try not to
1756 * immediately reuse them to prevent problems due to old
1759 if (++ofproto->alloc_port_no >= ofproto->max_ports) {
1760 ofproto->alloc_port_no = 0;
1762 if (!bitmap_is_set(ofproto->ofp_port_ids,
1763 ofproto->alloc_port_no)) {
1764 port_idx = ofproto->alloc_port_no;
1767 if (ofproto->alloc_port_no == end_port_no) {
1772 bitmap_set1(ofproto->ofp_port_ids, port_idx);
1773 return u16_to_ofp(port_idx);
1777 dealloc_ofp_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
1779 if (ofp_to_u16(ofp_port) < ofproto->max_ports) {
1780 bitmap_set0(ofproto->ofp_port_ids, ofp_to_u16(ofp_port));
1784 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1785 * pointer if the netdev cannot be opened. On success, also fills in
1787 static struct netdev *
1788 ofport_open(struct ofproto *ofproto,
1789 struct ofproto_port *ofproto_port,
1790 struct ofputil_phy_port *pp)
1792 enum netdev_flags flags;
1793 struct netdev *netdev;
1796 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1798 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
1799 "cannot be opened (%s)",
1801 ofproto_port->name, ofproto_port->ofp_port,
1802 ofproto_port->name, ovs_strerror(error));
1806 if (ofproto_port->ofp_port == OFPP_NONE) {
1807 if (!strcmp(ofproto->name, ofproto_port->name)) {
1808 ofproto_port->ofp_port = OFPP_LOCAL;
1810 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
1811 ofproto_port->name);
1814 pp->port_no = ofproto_port->ofp_port;
1815 netdev_get_etheraddr(netdev, pp->hw_addr);
1816 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
1817 netdev_get_flags(netdev, &flags);
1818 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
1819 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
1820 netdev_get_features(netdev, &pp->curr, &pp->advertised,
1821 &pp->supported, &pp->peer);
1822 pp->curr_speed = netdev_features_to_bps(pp->curr, 0);
1823 pp->max_speed = netdev_features_to_bps(pp->supported, 0);
1828 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1829 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1832 ofport_equal(const struct ofputil_phy_port *a,
1833 const struct ofputil_phy_port *b)
1835 return (eth_addr_equals(a->hw_addr, b->hw_addr)
1836 && a->state == b->state
1837 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
1838 && a->curr == b->curr
1839 && a->advertised == b->advertised
1840 && a->supported == b->supported
1841 && a->peer == b->peer
1842 && a->curr_speed == b->curr_speed
1843 && a->max_speed == b->max_speed);
1846 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1847 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1848 * one with the same name or port number). */
1850 ofport_install(struct ofproto *p,
1851 struct netdev *netdev, const struct ofputil_phy_port *pp)
1853 const char *netdev_name = netdev_get_name(netdev);
1854 struct ofport *ofport;
1857 /* Create ofport. */
1858 ofport = p->ofproto_class->port_alloc();
1863 ofport->ofproto = p;
1864 ofport->netdev = netdev;
1865 ofport->change_seq = netdev_change_seq(netdev);
1867 ofport->ofp_port = pp->port_no;
1868 ofport->created = time_msec();
1870 /* Add port to 'p'. */
1871 hmap_insert(&p->ports, &ofport->hmap_node,
1872 hash_ofp_port(ofport->ofp_port));
1873 shash_add(&p->port_by_name, netdev_name, ofport);
1875 update_mtu(p, ofport);
1877 /* Let the ofproto_class initialize its private data. */
1878 error = p->ofproto_class->port_construct(ofport);
1882 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
1886 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1887 p->name, netdev_name, ovs_strerror(error));
1889 ofport_destroy__(ofport);
1891 netdev_close(netdev);
1895 /* Removes 'ofport' from 'p' and destroys it. */
1897 ofport_remove(struct ofport *ofport)
1899 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
1901 ofport_destroy(ofport);
1904 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1907 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1909 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1911 ofport_remove(port);
1915 /* Updates 'port' with new 'pp' description.
1917 * Does not handle a name or port number change. The caller must implement
1918 * such a change as a delete followed by an add. */
1920 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
1922 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
1923 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
1924 | (pp->config & OFPUTIL_PC_PORT_DOWN));
1925 port->pp.state = pp->state;
1926 port->pp.curr = pp->curr;
1927 port->pp.advertised = pp->advertised;
1928 port->pp.supported = pp->supported;
1929 port->pp.peer = pp->peer;
1930 port->pp.curr_speed = pp->curr_speed;
1931 port->pp.max_speed = pp->max_speed;
1933 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
1936 /* Update OpenFlow 'state' in 'port' and notify controller. */
1938 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
1940 if (port->pp.state != state) {
1941 port->pp.state = state;
1942 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
1948 ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port)
1950 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1952 if (port->ofproto->ofproto_class->set_realdev) {
1953 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
1955 if (port->ofproto->ofproto_class->set_stp_port) {
1956 port->ofproto->ofproto_class->set_stp_port(port, NULL);
1958 if (port->ofproto->ofproto_class->set_cfm) {
1959 port->ofproto->ofproto_class->set_cfm(port, NULL);
1961 if (port->ofproto->ofproto_class->bundle_remove) {
1962 port->ofproto->ofproto_class->bundle_remove(port);
1968 ofport_destroy__(struct ofport *port)
1970 struct ofproto *ofproto = port->ofproto;
1971 const char *name = netdev_get_name(port->netdev);
1973 hmap_remove(&ofproto->ports, &port->hmap_node);
1974 shash_delete(&ofproto->port_by_name,
1975 shash_find(&ofproto->port_by_name, name));
1977 netdev_close(port->netdev);
1978 ofproto->ofproto_class->port_dealloc(port);
1982 ofport_destroy(struct ofport *port)
1985 dealloc_ofp_port(port->ofproto, port->ofp_port);
1986 port->ofproto->ofproto_class->port_destruct(port);
1987 ofport_destroy__(port);
1992 ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
1994 struct ofport *port;
1996 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_ofp_port(ofp_port),
1998 if (port->ofp_port == ofp_port) {
2006 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
2008 struct ofproto *ofproto = port->ofproto;
2011 if (ofproto->ofproto_class->port_get_stats) {
2012 error = ofproto->ofproto_class->port_get_stats(port, stats);
2021 update_port(struct ofproto *ofproto, const char *name)
2023 struct ofproto_port ofproto_port;
2024 struct ofputil_phy_port pp;
2025 struct netdev *netdev;
2026 struct ofport *port;
2028 COVERAGE_INC(ofproto_update_port);
2030 /* Fetch 'name''s location and properties from the datapath. */
2031 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
2032 ? ofport_open(ofproto, &ofproto_port, &pp)
2036 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
2037 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
2038 struct netdev *old_netdev = port->netdev;
2040 /* 'name' hasn't changed location. Any properties changed? */
2041 if (!ofport_equal(&port->pp, &pp)) {
2042 ofport_modified(port, &pp);
2045 update_mtu(ofproto, port);
2047 /* Install the newly opened netdev in case it has changed.
2048 * Don't close the old netdev yet in case port_modified has to
2049 * remove a retained reference to it.*/
2050 port->netdev = netdev;
2051 port->change_seq = netdev_change_seq(netdev);
2053 if (port->ofproto->ofproto_class->port_modified) {
2054 port->ofproto->ofproto_class->port_modified(port);
2057 netdev_close(old_netdev);
2059 /* If 'port' is nonnull then its name differs from 'name' and thus
2060 * we should delete it. If we think there's a port named 'name'
2061 * then its port number must be wrong now so delete it too. */
2063 ofport_remove(port);
2065 ofport_remove_with_name(ofproto, name);
2066 ofport_install(ofproto, netdev, &pp);
2069 /* Any port named 'name' is gone now. */
2070 ofport_remove_with_name(ofproto, name);
2072 ofproto_port_destroy(&ofproto_port);
2076 init_ports(struct ofproto *p)
2078 struct ofproto_port_dump dump;
2079 struct ofproto_port ofproto_port;
2080 struct shash_node *node, *next;
2082 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2083 const char *name = ofproto_port.name;
2085 if (shash_find(&p->port_by_name, name)) {
2086 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2089 struct ofputil_phy_port pp;
2090 struct netdev *netdev;
2092 /* Check if an OpenFlow port number had been requested. */
2093 node = shash_find(&init_ofp_ports, name);
2095 const struct iface_hint *iface_hint = node->data;
2096 simap_put(&p->ofp_requests, name,
2097 ofp_to_u16(iface_hint->ofp_port));
2100 netdev = ofport_open(p, &ofproto_port, &pp);
2102 ofport_install(p, netdev, &pp);
2103 if (ofp_to_u16(ofproto_port.ofp_port) < p->max_ports) {
2104 p->alloc_port_no = MAX(p->alloc_port_no,
2105 ofp_to_u16(ofproto_port.ofp_port));
2111 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2112 struct iface_hint *iface_hint = node->data;
2114 if (!strcmp(iface_hint->br_name, p->name)) {
2115 free(iface_hint->br_name);
2116 free(iface_hint->br_type);
2118 shash_delete(&init_ofp_ports, node);
2125 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2126 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2128 find_min_mtu(struct ofproto *p)
2130 struct ofport *ofport;
2133 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2134 struct netdev *netdev = ofport->netdev;
2137 /* Skip any internal ports, since that's what we're trying to
2139 if (!strcmp(netdev_get_type(netdev), "internal")) {
2143 if (netdev_get_mtu(netdev, &dev_mtu)) {
2146 if (!mtu || dev_mtu < mtu) {
2151 return mtu ? mtu: ETH_PAYLOAD_MAX;
2154 /* Update MTU of all datapath devices on 'p' to the minimum of the
2155 * non-datapath ports in event of 'port' added or changed. */
2157 update_mtu(struct ofproto *p, struct ofport *port)
2159 struct ofport *ofport;
2160 struct netdev *netdev = port->netdev;
2161 int dev_mtu, old_min;
2163 if (netdev_get_mtu(netdev, &dev_mtu)) {
2167 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2168 if (dev_mtu > p->min_mtu) {
2169 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2170 dev_mtu = p->min_mtu;
2173 port->mtu = dev_mtu;
2177 /* For non-internal port find new min mtu. */
2178 old_min = p->min_mtu;
2179 port->mtu = dev_mtu;
2180 p->min_mtu = find_min_mtu(p);
2181 if (p->min_mtu == old_min) {
2185 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2186 struct netdev *netdev = ofport->netdev;
2188 if (!strcmp(netdev_get_type(netdev), "internal")) {
2189 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2190 ofport->mtu = p->min_mtu;
2197 ofproto_rule_destroy__(struct rule *rule)
2200 cls_rule_destroy(&rule->cr);
2201 free(rule->ofpacts);
2202 ovs_mutex_destroy(&rule->timeout_mutex);
2203 ovs_rwlock_destroy(&rule->evict);
2204 rule->ofproto->ofproto_class->rule_dealloc(rule);
2208 /* This function allows an ofproto implementation to destroy any rules that
2209 * remain when its ->destruct() function is called. The caller must have
2210 * already uninitialized any derived members of 'rule' (step 5 described in the
2211 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
2212 * This function implements steps 6 and 7.
2214 * This function should only be called from an ofproto implementation's
2215 * ->destruct() function. It is not suitable elsewhere. */
2217 ofproto_rule_destroy(struct ofproto *ofproto, struct classifier *cls,
2218 struct rule *rule) OVS_REQ_WRLOCK(cls->rwlock)
2220 ovs_assert(!rule->pending);
2221 if (!ovs_rwlock_trywrlock(&rule->evict)) {
2222 oftable_remove_rule__(ofproto, cls, rule);
2226 ofproto_rule_destroy__(rule);
2229 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2230 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2232 ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port)
2234 return (port == OFPP_ANY
2235 || ofpacts_output_to_port(rule->ofpacts, rule->ofpacts_len, port));
2238 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2239 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2241 ofoperation_has_out_port(const struct ofoperation *op, ofp_port_t out_port)
2243 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2248 case OFOPERATION_ADD:
2249 return op->victim && ofproto_rule_has_out_port(op->victim, out_port);
2251 case OFOPERATION_DELETE:
2254 case OFOPERATION_MODIFY:
2255 return ofpacts_output_to_port(op->ofpacts, op->ofpacts_len, out_port);
2261 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2262 * statistics appropriately. 'packet' must have at least sizeof(struct
2263 * ofp10_packet_in) bytes of headroom.
2265 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2266 * with statistics for 'packet' either way.
2268 * Takes ownership of 'packet'. */
2270 rule_execute(struct rule *rule, ofp_port_t in_port, struct ofpbuf *packet)
2273 union flow_in_port in_port_;
2275 ovs_assert(ofpbuf_headroom(packet) >= sizeof(struct ofp10_packet_in));
2277 in_port_.ofp_port = in_port;
2278 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
2279 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
2282 /* Returns true if 'rule' should be hidden from the controller.
2284 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2285 * (e.g. by in-band control) and are intentionally hidden from the
2288 ofproto_rule_is_hidden(const struct rule *rule)
2290 return rule->cr.priority > UINT16_MAX;
2293 static enum oftable_flags
2294 rule_get_flags(const struct rule *rule)
2296 return rule->ofproto->tables[rule->table_id].flags;
2300 rule_is_modifiable(const struct rule *rule)
2302 return !(rule_get_flags(rule) & OFTABLE_READONLY);
2306 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2308 ofconn_send_reply(ofconn, make_echo_reply(oh));
2313 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2315 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2316 struct ofputil_switch_features features;
2317 struct ofport *port;
2323 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2325 ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2327 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2328 * if present, are always at the end.) */
2329 n_tables = ofproto->n_tables;
2330 for (i = 0; i < ofproto->n_tables; i++) {
2331 if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
2337 features.datapath_id = ofproto->datapath_id;
2338 features.n_buffers = pktbuf_capacity();
2339 features.n_tables = n_tables;
2340 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2341 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2343 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2345 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2346 features.auxiliary_id = 0;
2347 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2349 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2350 ofputil_put_switch_features_port(&port->pp, b);
2353 ofconn_send_reply(ofconn, b);
2358 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2360 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2361 struct ofp_switch_config *osc;
2362 enum ofp_config_flags flags;
2366 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2367 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2368 flags = ofproto->frag_handling;
2369 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2370 if (oh->version < OFP13_VERSION
2371 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2372 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2374 osc->flags = htons(flags);
2375 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2376 ofconn_send_reply(ofconn, buf);
2382 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2384 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2385 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2386 uint16_t flags = ntohs(osc->flags);
2388 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2389 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
2390 enum ofp_config_flags cur = ofproto->frag_handling;
2391 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2393 ovs_assert((cur & OFPC_FRAG_MASK) == cur);
2395 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2396 ofproto->frag_handling = next;
2398 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2400 ofputil_frag_handling_to_string(next));
2404 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2405 ofconn_set_invalid_ttl_to_controller(ofconn,
2406 (oh->version < OFP13_VERSION
2407 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2409 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2414 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2415 * error message code for the caller to propagate upward. Otherwise, returns
2418 * The log message mentions 'msg_type'. */
2420 reject_slave_controller(struct ofconn *ofconn)
2422 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2423 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
2424 return OFPERR_OFPBRC_EPERM;
2430 /* Finds the OFPACT_METER action, if any, in the 'ofpacts_len' bytes of
2431 * 'ofpacts'. If found, returns its meter ID; if not, returns 0.
2433 * This function relies on the order of 'ofpacts' being correct (as checked by
2434 * ofpacts_verify()). */
2436 find_meter(const struct ofpact ofpacts[], size_t ofpacts_len)
2438 const struct ofpact *a;
2440 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2441 enum ovs_instruction_type inst;
2443 inst = ovs_instruction_type_from_ofpact_type(a->type);
2444 if (a->type == OFPACT_METER) {
2445 return ofpact_get_METER(a)->meter_id;
2446 } else if (inst > OVSINST_OFPIT13_METER) {
2454 /* Checks that the 'ofpacts_len' bytes of actions in 'ofpacts' are appropriate
2455 * for a packet with the prerequisites satisfied by 'flow' in table 'table_id'.
2456 * 'flow' may be temporarily modified, but is restored at return.
2459 ofproto_check_ofpacts(struct ofproto *ofproto,
2460 const struct ofpact ofpacts[], size_t ofpacts_len,
2461 struct flow *flow, uint8_t table_id)
2466 error = ofpacts_check(ofpacts, ofpacts_len, flow,
2467 u16_to_ofp(ofproto->max_ports), table_id);
2472 mid = find_meter(ofpacts, ofpacts_len);
2473 if (mid && ofproto_get_provider_meter_id(ofproto, mid) == UINT32_MAX) {
2474 return OFPERR_OFPMMFC_INVALID_METER;
2480 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2482 struct ofproto *p = ofconn_get_ofproto(ofconn);
2483 struct ofputil_packet_out po;
2484 struct ofpbuf *payload;
2485 uint64_t ofpacts_stub[1024 / 8];
2486 struct ofpbuf ofpacts;
2488 union flow_in_port in_port_;
2491 COVERAGE_INC(ofproto_packet_out);
2493 error = reject_slave_controller(ofconn);
2498 /* Decode message. */
2499 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2500 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2502 goto exit_free_ofpacts;
2504 if (ofp_to_u16(po.in_port) >= p->max_ports
2505 && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) {
2506 error = OFPERR_OFPBRC_BAD_PORT;
2507 goto exit_free_ofpacts;
2512 if (po.buffer_id != UINT32_MAX) {
2513 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2514 if (error || !payload) {
2515 goto exit_free_ofpacts;
2518 payload = xmalloc(sizeof *payload);
2519 ofpbuf_use_const(payload, po.packet, po.packet_len);
2522 /* Verify actions against packet, then send packet if successful. */
2523 in_port_.ofp_port = po.in_port;
2524 flow_extract(payload, 0, 0, NULL, &in_port_, &flow);
2525 error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len, &flow, 0);
2527 error = p->ofproto_class->packet_out(p, payload, &flow,
2528 po.ofpacts, po.ofpacts_len);
2530 ofpbuf_delete(payload);
2533 ofpbuf_uninit(&ofpacts);
2539 update_port_config(struct ofport *port,
2540 enum ofputil_port_config config,
2541 enum ofputil_port_config mask)
2543 enum ofputil_port_config old_config = port->pp.config;
2544 enum ofputil_port_config toggle;
2546 toggle = (config ^ port->pp.config) & mask;
2547 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2548 if (config & OFPUTIL_PC_PORT_DOWN) {
2549 netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL);
2551 netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL);
2553 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2556 port->pp.config ^= toggle;
2557 if (port->pp.config != old_config) {
2558 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2563 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2565 struct ofproto *p = ofconn_get_ofproto(ofconn);
2566 struct ofputil_port_mod pm;
2567 struct ofport *port;
2570 error = reject_slave_controller(ofconn);
2575 error = ofputil_decode_port_mod(oh, &pm);
2580 port = ofproto_get_port(p, pm.port_no);
2582 return OFPERR_OFPPMFC_BAD_PORT;
2583 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2584 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2586 update_port_config(port, pm.config, pm.mask);
2588 netdev_set_advertisements(port->netdev, pm.advertise);
2595 handle_desc_stats_request(struct ofconn *ofconn,
2596 const struct ofp_header *request)
2598 static const char *default_mfr_desc = "Nicira, Inc.";
2599 static const char *default_hw_desc = "Open vSwitch";
2600 static const char *default_sw_desc = VERSION;
2601 static const char *default_serial_desc = "None";
2602 static const char *default_dp_desc = "None";
2604 struct ofproto *p = ofconn_get_ofproto(ofconn);
2605 struct ofp_desc_stats *ods;
2608 msg = ofpraw_alloc_stats_reply(request, 0);
2609 ods = ofpbuf_put_zeros(msg, sizeof *ods);
2610 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
2611 sizeof ods->mfr_desc);
2612 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
2613 sizeof ods->hw_desc);
2614 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
2615 sizeof ods->sw_desc);
2616 ovs_strlcpy(ods->serial_num,
2617 p->serial_desc ? p->serial_desc : default_serial_desc,
2618 sizeof ods->serial_num);
2619 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
2620 sizeof ods->dp_desc);
2621 ofconn_send_reply(ofconn, msg);
2627 handle_table_stats_request(struct ofconn *ofconn,
2628 const struct ofp_header *request)
2630 struct ofproto *p = ofconn_get_ofproto(ofconn);
2631 struct ofp12_table_stats *ots;
2636 /* Set up default values.
2638 * ofp12_table_stats is used as a generic structure as
2639 * it is able to hold all the fields for ofp10_table_stats
2640 * and ofp11_table_stats (and of course itself).
2642 ots = xcalloc(p->n_tables, sizeof *ots);
2643 for (i = 0; i < p->n_tables; i++) {
2644 ots[i].table_id = i;
2645 sprintf(ots[i].name, "table%zu", i);
2646 ots[i].match = htonll(OFPXMT12_MASK);
2647 ots[i].wildcards = htonll(OFPXMT12_MASK);
2648 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
2649 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
2650 ots[i].write_setfields = htonll(OFPXMT12_MASK);
2651 ots[i].apply_setfields = htonll(OFPXMT12_MASK);
2652 ots[i].metadata_match = htonll(UINT64_MAX);
2653 ots[i].metadata_write = htonll(UINT64_MAX);
2654 ots[i].instructions = htonl(OFPIT11_ALL);
2655 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
2656 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2657 ovs_rwlock_rdlock(&p->tables[i].cls.rwlock);
2658 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2659 ovs_rwlock_unlock(&p->tables[i].cls.rwlock);
2662 p->ofproto_class->get_tables(p, ots);
2664 /* Post-process the tables, dropping hidden tables. */
2665 n_tables = p->n_tables;
2666 for (i = 0; i < p->n_tables; i++) {
2667 const struct oftable *table = &p->tables[i];
2669 if (table->flags & OFTABLE_HIDDEN) {
2675 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2678 if (table->max_flows < ntohl(ots[i].max_entries)) {
2679 ots[i].max_entries = htonl(table->max_flows);
2683 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
2684 ofconn_send_reply(ofconn, msg);
2692 append_port_stat(struct ofport *port, struct list *replies)
2694 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
2696 calc_duration(port->created, time_msec(),
2697 &ops.duration_sec, &ops.duration_nsec);
2699 /* Intentionally ignore return value, since errors will set
2700 * 'stats' to all-1s, which is correct for OpenFlow, and
2701 * netdev_get_stats() will log errors. */
2702 ofproto_port_get_stats(port, &ops.stats);
2704 ofputil_append_port_stat(replies, &ops);
2708 handle_port_stats_request(struct ofconn *ofconn,
2709 const struct ofp_header *request)
2711 struct ofproto *p = ofconn_get_ofproto(ofconn);
2712 struct ofport *port;
2713 struct list replies;
2717 error = ofputil_decode_port_stats_request(request, &port_no);
2722 ofpmp_init(&replies, request);
2723 if (port_no != OFPP_ANY) {
2724 port = ofproto_get_port(p, port_no);
2726 append_port_stat(port, &replies);
2729 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2730 append_port_stat(port, &replies);
2734 ofconn_send_replies(ofconn, &replies);
2739 handle_port_desc_stats_request(struct ofconn *ofconn,
2740 const struct ofp_header *request)
2742 struct ofproto *p = ofconn_get_ofproto(ofconn);
2743 enum ofp_version version;
2744 struct ofport *port;
2745 struct list replies;
2747 ofpmp_init(&replies, request);
2749 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
2750 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2751 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
2754 ofconn_send_replies(ofconn, &replies);
2759 hash_cookie(ovs_be64 cookie)
2761 return hash_2words((OVS_FORCE uint64_t)cookie >> 32,
2762 (OVS_FORCE uint64_t)cookie);
2766 cookies_insert(struct ofproto *ofproto, struct rule *rule)
2768 hindex_insert(&ofproto->cookies, &rule->cookie_node,
2769 hash_cookie(rule->flow_cookie));
2773 cookies_remove(struct ofproto *ofproto, struct rule *rule)
2775 hindex_remove(&ofproto->cookies, &rule->cookie_node);
2779 ofproto_rule_change_cookie(struct ofproto *ofproto, struct rule *rule,
2780 ovs_be64 new_cookie)
2782 if (new_cookie != rule->flow_cookie) {
2783 cookies_remove(ofproto, rule);
2785 rule->flow_cookie = new_cookie;
2787 cookies_insert(ofproto, rule);
2792 calc_duration(long long int start, long long int now,
2793 uint32_t *sec, uint32_t *nsec)
2795 long long int msecs = now - start;
2796 *sec = msecs / 1000;
2797 *nsec = (msecs % 1000) * (1000 * 1000);
2800 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2801 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2803 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
2805 return (table_id == 0xff || table_id < ofproto->n_tables
2807 : OFPERR_OFPBRC_BAD_TABLE_ID);
2811 static struct oftable *
2812 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
2814 struct oftable *table;
2816 for (table = &ofproto->tables[table_id];
2817 table < &ofproto->tables[ofproto->n_tables];
2819 if (!(table->flags & OFTABLE_HIDDEN)) {
2827 static struct oftable *
2828 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
2830 if (table_id == 0xff) {
2831 return next_visible_table(ofproto, 0);
2832 } else if (table_id < ofproto->n_tables) {
2833 return &ofproto->tables[table_id];
2839 static struct oftable *
2840 next_matching_table(const struct ofproto *ofproto,
2841 const struct oftable *table, uint8_t table_id)
2843 return (table_id == 0xff
2844 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
2848 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2850 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2851 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2853 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2854 * only once, for that table. (This can be used to access tables marked
2857 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2858 * entered at all. (Perhaps you should have validated TABLE_ID with
2859 * check_table_id().)
2861 * All parameters are evaluated multiple times.
2863 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2864 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2866 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2868 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2869 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2870 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2873 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2874 * to 'out_port' are included.
2876 * Hidden rules are always omitted.
2878 * Returns 0 on success, otherwise an OpenFlow error code. */
2880 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
2881 const struct match *match,
2882 ovs_be64 cookie, ovs_be64 cookie_mask,
2883 ofp_port_t out_port, struct list *rules)
2885 struct oftable *table;
2889 error = check_table_id(ofproto, table_id);
2895 cls_rule_init(&cr, match, 0);
2897 if (cookie_mask == htonll(UINT64_MAX)) {
2900 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node, hash_cookie(cookie),
2901 &ofproto->cookies) {
2902 if (table_id != rule->table_id && table_id != 0xff) {
2905 if (ofproto_rule_is_hidden(rule)) {
2908 if (cls_rule_is_loose_match(&rule->cr, &cr.match)) {
2909 if (rule->pending) {
2910 error = OFPROTO_POSTPONE;
2913 if (rule->flow_cookie == cookie /* Hash collisions possible. */
2914 && ofproto_rule_has_out_port(rule, out_port)) {
2915 list_push_back(rules, &rule->ofproto_node);
2922 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2923 struct cls_cursor cursor;
2926 ovs_rwlock_rdlock(&table->cls.rwlock);
2927 cls_cursor_init(&cursor, &table->cls, &cr);
2928 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2929 if (rule->pending) {
2930 ovs_rwlock_unlock(&table->cls.rwlock);
2931 error = OFPROTO_POSTPONE;
2934 if (!ofproto_rule_is_hidden(rule)
2935 && ofproto_rule_has_out_port(rule, out_port)
2936 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2937 list_push_back(rules, &rule->ofproto_node);
2940 ovs_rwlock_unlock(&table->cls.rwlock);
2944 cls_rule_destroy(&cr);
2948 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2949 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2950 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2953 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2954 * to 'out_port' are included.
2956 * Hidden rules are always omitted.
2958 * Returns 0 on success, otherwise an OpenFlow error code. */
2960 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
2961 const struct match *match, unsigned int priority,
2962 ovs_be64 cookie, ovs_be64 cookie_mask,
2963 ofp_port_t out_port, struct list *rules)
2965 struct oftable *table;
2969 error = check_table_id(ofproto, table_id);
2975 cls_rule_init(&cr, match, priority);
2977 if (cookie_mask == htonll(UINT64_MAX)) {
2980 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node, hash_cookie(cookie),
2981 &ofproto->cookies) {
2982 if (table_id != rule->table_id && table_id != 0xff) {
2985 if (ofproto_rule_is_hidden(rule)) {
2988 if (cls_rule_equal(&rule->cr, &cr)) {
2989 if (rule->pending) {
2990 error = OFPROTO_POSTPONE;
2993 if (rule->flow_cookie == cookie /* Hash collisions possible. */
2994 && ofproto_rule_has_out_port(rule, out_port)) {
2995 list_push_back(rules, &rule->ofproto_node);
3002 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
3005 ovs_rwlock_rdlock(&table->cls.rwlock);
3006 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls,
3008 ovs_rwlock_unlock(&table->cls.rwlock);
3010 if (rule->pending) {
3011 error = OFPROTO_POSTPONE;
3014 if (!ofproto_rule_is_hidden(rule)
3015 && ofproto_rule_has_out_port(rule, out_port)
3016 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
3017 list_push_back(rules, &rule->ofproto_node);
3023 cls_rule_destroy(&cr);
3027 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
3028 * forced into the range of a uint16_t. */
3030 age_secs(long long int age_ms)
3032 return (age_ms < 0 ? 0
3033 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
3034 : (unsigned int) age_ms / 1000);
3038 handle_flow_stats_request(struct ofconn *ofconn,
3039 const struct ofp_header *request)
3041 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3042 struct ofputil_flow_stats_request fsr;
3043 struct list replies;
3048 error = ofputil_decode_flow_stats_request(&fsr, request);
3053 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
3054 fsr.cookie, fsr.cookie_mask,
3055 fsr.out_port, &rules);
3060 ofpmp_init(&replies, request);
3061 LIST_FOR_EACH (rule, ofproto_node, &rules) {
3062 long long int now = time_msec();
3063 struct ofputil_flow_stats fs;
3065 minimatch_expand(&rule->cr.match, &fs.match);
3066 fs.priority = rule->cr.priority;
3067 fs.cookie = rule->flow_cookie;
3068 fs.table_id = rule->table_id;
3069 calc_duration(rule->created, now, &fs.duration_sec, &fs.duration_nsec);
3070 fs.idle_age = age_secs(now - rule->used);
3071 fs.hard_age = age_secs(now - rule->modified);
3072 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
3074 fs.ofpacts = rule->ofpacts;
3075 fs.ofpacts_len = rule->ofpacts_len;
3078 ovs_mutex_lock(&rule->timeout_mutex);
3079 fs.idle_timeout = rule->idle_timeout;
3080 fs.hard_timeout = rule->hard_timeout;
3081 ovs_mutex_unlock(&rule->timeout_mutex);
3083 if (rule->send_flow_removed) {
3084 fs.flags |= OFPFF_SEND_FLOW_REM;
3085 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
3086 and OFPFF13_NO_BYT_COUNTS */
3088 ofputil_append_flow_stats_reply(&fs, &replies);
3090 ofconn_send_replies(ofconn, &replies);
3096 flow_stats_ds(struct rule *rule, struct ds *results)
3098 uint64_t packet_count, byte_count;
3100 rule->ofproto->ofproto_class->rule_get_stats(rule,
3101 &packet_count, &byte_count);
3103 if (rule->table_id != 0) {
3104 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
3106 ds_put_format(results, "duration=%llds, ",
3107 (time_msec() - rule->created) / 1000);
3108 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3109 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3110 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3111 cls_rule_format(&rule->cr, results);
3112 ds_put_char(results, ',');
3113 ofpacts_format(rule->ofpacts, rule->ofpacts_len, results);
3114 ds_put_cstr(results, "\n");
3117 /* Adds a pretty-printed description of all flows to 'results', including
3118 * hidden flows (e.g., set up by in-band control). */
3120 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3122 struct oftable *table;
3124 OFPROTO_FOR_EACH_TABLE (table, p) {
3125 struct cls_cursor cursor;
3128 ovs_rwlock_rdlock(&table->cls.rwlock);
3129 cls_cursor_init(&cursor, &table->cls, NULL);
3130 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3131 flow_stats_ds(rule, results);
3133 ovs_rwlock_unlock(&table->cls.rwlock);
3137 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
3138 * '*engine_type' and '*engine_id', respectively. */
3140 ofproto_get_netflow_ids(const struct ofproto *ofproto,
3141 uint8_t *engine_type, uint8_t *engine_id)
3143 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
3146 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
3147 * true if the port's CFM status was successfully stored into '*status'.
3148 * Returns false if the port did not have CFM configured, in which case
3149 * '*status' is indeterminate.
3151 * The caller must provide and owns '*status', and must free 'status->rmps'. */
3153 ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port,
3154 struct ofproto_cfm_status *status)
3156 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
3158 && ofproto->ofproto_class->get_cfm_status
3159 && ofproto->ofproto_class->get_cfm_status(ofport, status));
3163 handle_aggregate_stats_request(struct ofconn *ofconn,
3164 const struct ofp_header *oh)
3166 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3167 struct ofputil_flow_stats_request request;
3168 struct ofputil_aggregate_stats stats;
3169 bool unknown_packets, unknown_bytes;
3170 struct ofpbuf *reply;
3175 error = ofputil_decode_flow_stats_request(&request, oh);
3180 error = collect_rules_loose(ofproto, request.table_id, &request.match,
3181 request.cookie, request.cookie_mask,
3182 request.out_port, &rules);
3187 memset(&stats, 0, sizeof stats);
3188 unknown_packets = unknown_bytes = false;
3189 LIST_FOR_EACH (rule, ofproto_node, &rules) {
3190 uint64_t packet_count;
3191 uint64_t byte_count;
3193 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
3196 if (packet_count == UINT64_MAX) {
3197 unknown_packets = true;
3199 stats.packet_count += packet_count;
3202 if (byte_count == UINT64_MAX) {
3203 unknown_bytes = true;
3205 stats.byte_count += byte_count;
3210 if (unknown_packets) {
3211 stats.packet_count = UINT64_MAX;
3213 if (unknown_bytes) {
3214 stats.byte_count = UINT64_MAX;
3217 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
3218 ofconn_send_reply(ofconn, reply);
3223 struct queue_stats_cbdata {
3224 struct ofport *ofport;
3225 struct list replies;
3230 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3231 const struct netdev_queue_stats *stats)
3233 struct ofputil_queue_stats oqs;
3235 oqs.port_no = cbdata->ofport->pp.port_no;
3236 oqs.queue_id = queue_id;
3237 oqs.tx_bytes = stats->tx_bytes;
3238 oqs.tx_packets = stats->tx_packets;
3239 oqs.tx_errors = stats->tx_errors;
3240 if (stats->created != LLONG_MIN) {
3241 calc_duration(stats->created, cbdata->now,
3242 &oqs.duration_sec, &oqs.duration_nsec);
3244 oqs.duration_sec = oqs.duration_nsec = UINT32_MAX;
3246 ofputil_append_queue_stat(&cbdata->replies, &oqs);
3250 handle_queue_stats_dump_cb(uint32_t queue_id,
3251 struct netdev_queue_stats *stats,
3254 struct queue_stats_cbdata *cbdata = cbdata_;
3256 put_queue_stats(cbdata, queue_id, stats);
3260 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3261 struct queue_stats_cbdata *cbdata)
3263 cbdata->ofport = port;
3264 if (queue_id == OFPQ_ALL) {
3265 netdev_dump_queue_stats(port->netdev,
3266 handle_queue_stats_dump_cb, cbdata);
3268 struct netdev_queue_stats stats;
3270 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3271 put_queue_stats(cbdata, queue_id, &stats);
3273 return OFPERR_OFPQOFC_BAD_QUEUE;
3280 handle_queue_stats_request(struct ofconn *ofconn,
3281 const struct ofp_header *rq)
3283 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3284 struct queue_stats_cbdata cbdata;
3285 struct ofport *port;
3287 struct ofputil_queue_stats_request oqsr;
3289 COVERAGE_INC(ofproto_queue_req);
3291 ofpmp_init(&cbdata.replies, rq);
3292 cbdata.now = time_msec();
3294 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3299 if (oqsr.port_no == OFPP_ANY) {
3300 error = OFPERR_OFPQOFC_BAD_QUEUE;
3301 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3302 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3307 port = ofproto_get_port(ofproto, oqsr.port_no);
3309 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3310 : OFPERR_OFPQOFC_BAD_PORT);
3313 ofconn_send_replies(ofconn, &cbdata.replies);
3315 ofpbuf_list_delete(&cbdata.replies);
3322 is_flow_deletion_pending(const struct ofproto *ofproto,
3323 const struct cls_rule *cls_rule,
3326 if (!hmap_is_empty(&ofproto->deletions)) {
3327 struct ofoperation *op;
3329 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3330 cls_rule_hash(cls_rule, table_id),
3331 &ofproto->deletions) {
3332 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
3341 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3342 * in which no matching flow already exists in the flow table.
3344 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3345 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3346 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3347 * initiated now but may be retried later.
3349 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
3350 * ownership remains with the caller.
3352 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3355 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
3356 struct ofputil_flow_mod *fm, const struct ofp_header *request)
3358 struct oftable *table;
3359 struct ofopgroup *group;
3360 struct rule *victim;
3365 error = check_table_id(ofproto, fm->table_id);
3371 if (fm->table_id == 0xff) {
3372 if (ofproto->ofproto_class->rule_choose_table) {
3373 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3379 ovs_assert(table_id < ofproto->n_tables);
3383 } else if (fm->table_id < ofproto->n_tables) {
3384 table_id = fm->table_id;
3386 return OFPERR_OFPBRC_BAD_TABLE_ID;
3389 table = &ofproto->tables[table_id];
3391 if (table->flags & OFTABLE_READONLY) {
3392 return OFPERR_OFPBRC_EPERM;
3395 /* Verify actions. */
3396 error = ofproto_check_ofpacts(ofproto, fm->ofpacts, fm->ofpacts_len,
3397 &fm->match.flow, table_id);
3402 /* Allocate new rule and initialize classifier rule. */
3403 rule = ofproto->ofproto_class->rule_alloc();
3405 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
3406 ofproto->name, ovs_strerror(error));
3409 cls_rule_init(&rule->cr, &fm->match, fm->priority);
3411 /* Serialize against pending deletion. */
3412 if (is_flow_deletion_pending(ofproto, &rule->cr, table_id)) {
3413 cls_rule_destroy(&rule->cr);
3414 ofproto->ofproto_class->rule_dealloc(rule);
3415 return OFPROTO_POSTPONE;
3418 /* Check for overlap, if requested. */
3419 if (fm->flags & OFPFF_CHECK_OVERLAP) {
3422 ovs_rwlock_rdlock(&table->cls.rwlock);
3423 overlaps = classifier_rule_overlaps(&table->cls, &rule->cr);
3424 ovs_rwlock_unlock(&table->cls.rwlock);
3427 cls_rule_destroy(&rule->cr);
3428 ofproto->ofproto_class->rule_dealloc(rule);
3429 return OFPERR_OFPFMFC_OVERLAP;
3433 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3435 rule->ofproto = ofproto;
3436 rule->pending = NULL;
3437 rule->flow_cookie = fm->new_cookie;
3438 rule->created = rule->modified = rule->used = time_msec();
3440 ovs_mutex_init(&rule->timeout_mutex);
3441 ovs_mutex_lock(&rule->timeout_mutex);
3442 rule->idle_timeout = fm->idle_timeout;
3443 rule->hard_timeout = fm->hard_timeout;
3444 ovs_mutex_unlock(&rule->timeout_mutex);
3446 rule->table_id = table - ofproto->tables;
3447 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
3448 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
3449 and OFPFF13_NO_BYT_COUNTS */
3450 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3451 rule->ofpacts_len = fm->ofpacts_len;
3452 rule->meter_id = find_meter(rule->ofpacts, rule->ofpacts_len);
3453 list_init(&rule->meter_list_node);
3454 rule->eviction_group = NULL;
3455 list_init(&rule->expirable);
3456 rule->monitor_flags = 0;
3457 rule->add_seqno = 0;
3458 rule->modify_seqno = 0;
3459 ovs_rwlock_init(&rule->evict);
3461 /* Insert new rule. */
3462 victim = oftable_replace_rule(rule);
3463 if (victim && !rule_is_modifiable(victim)) {
3464 error = OFPERR_OFPBRC_EPERM;
3465 } else if (victim && victim->pending) {
3466 error = OFPROTO_POSTPONE;
3468 struct ofoperation *op;
3472 ovs_rwlock_rdlock(&table->cls.rwlock);
3473 n_rules = classifier_count(&table->cls);
3474 ovs_rwlock_unlock(&table->cls.rwlock);
3475 if (n_rules > table->max_flows) {
3476 ovs_rwlock_rdlock(&rule->evict);
3477 if (choose_rule_to_evict(table, &evict)) {
3478 ovs_rwlock_unlock(&rule->evict);
3479 ovs_rwlock_unlock(&evict->evict);
3480 if (evict->pending) {
3481 error = OFPROTO_POSTPONE;
3485 ovs_rwlock_unlock(&rule->evict);
3486 error = OFPERR_OFPFMFC_TABLE_FULL;
3493 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3494 op = ofoperation_create(group, rule, OFOPERATION_ADD, 0);
3495 op->victim = victim;
3497 error = ofproto->ofproto_class->rule_construct(rule);
3499 op->group->n_running--;
3500 ofoperation_destroy(rule->pending);
3502 /* It would be better if we maintained the lock we took in
3503 * choose_rule_to_evict() earlier, but that confuses the thread
3504 * safety analysis, and this code is fragile enough that we really
3505 * need it. In the worst case, we'll have to block a little while
3506 * before we perform the eviction, which doesn't seem like a big
3508 ovs_rwlock_wrlock(&evict->evict);
3509 delete_flow__(evict, group, OFPRR_EVICTION);
3511 ofopgroup_submit(group);
3515 /* Back out if an error occurred. */
3517 oftable_substitute_rule(rule, victim);
3518 ofproto_rule_destroy__(rule);
3523 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3525 /* Modifies the rules listed in 'rules', changing their actions to match those
3528 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3531 * Returns 0 on success, otherwise an OpenFlow error code. */
3533 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3534 struct ofputil_flow_mod *fm, const struct ofp_header *request,
3537 struct ofopgroup *group;
3541 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3542 error = OFPERR_OFPBRC_EPERM;
3543 LIST_FOR_EACH (rule, ofproto_node, rules) {
3544 struct ofoperation *op;
3545 bool actions_changed;
3547 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3549 if (rule_is_modifiable(rule)) {
3550 /* At least one rule is modifiable, don't report EPERM error. */
3556 /* Verify actions. */
3557 error = ofpacts_check(fm->ofpacts, fm->ofpacts_len, &fm->match.flow,
3558 u16_to_ofp(ofproto->max_ports), rule->table_id);
3563 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
3564 rule->ofpacts, rule->ofpacts_len);
3566 op = ofoperation_create(group, rule, OFOPERATION_MODIFY, 0);
3568 if (fm->modify_cookie && fm->new_cookie != htonll(UINT64_MAX)) {
3569 ofproto_rule_change_cookie(ofproto, rule, fm->new_cookie);
3571 if (actions_changed) {
3572 op->ofpacts = rule->ofpacts;
3573 op->ofpacts_len = rule->ofpacts_len;
3574 op->meter_id = rule->meter_id;
3575 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3576 rule->ofpacts_len = fm->ofpacts_len;
3577 rule->meter_id = find_meter(rule->ofpacts, rule->ofpacts_len);
3578 rule->ofproto->ofproto_class->rule_modify_actions(rule);
3580 ofoperation_complete(op, 0);
3583 ofopgroup_submit(group);
3589 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
3590 struct ofputil_flow_mod *fm, const struct ofp_header *request)
3592 if (fm->cookie_mask != htonll(0) || fm->new_cookie == htonll(UINT64_MAX)) {
3595 return add_flow(ofproto, ofconn, fm, request);
3598 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
3601 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3604 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3605 struct ofputil_flow_mod *fm,
3606 const struct ofp_header *request)
3611 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3612 fm->cookie, fm->cookie_mask,
3616 } else if (list_is_empty(&rules)) {
3617 return modify_flows_add(ofproto, ofconn, fm, request);
3619 return modify_flows__(ofproto, ofconn, fm, request, &rules);
3623 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3626 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3629 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3630 struct ofputil_flow_mod *fm,
3631 const struct ofp_header *request)
3636 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3637 fm->priority, fm->cookie, fm->cookie_mask,
3642 } else if (list_is_empty(&rules)) {
3643 return modify_flows_add(ofproto, ofconn, fm, request);
3645 return list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
3646 fm, request, &rules)
3651 /* OFPFC_DELETE implementation. */
3654 delete_flow__(struct rule *rule, struct ofopgroup *group,
3655 enum ofp_flow_removed_reason reason)
3657 struct ofproto *ofproto = rule->ofproto;
3659 ofproto_rule_send_removed(rule, reason);
3661 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
3662 oftable_remove_rule(rule);
3663 ofproto->ofproto_class->rule_destruct(rule);
3666 /* Deletes the rules listed in 'rules'.
3668 * Returns 0 on success, otherwise an OpenFlow error code. */
3670 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3671 const struct ofp_header *request, struct list *rules,
3672 enum ofp_flow_removed_reason reason)
3674 struct rule *rule, *next;
3675 struct ofopgroup *group;
3677 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
3678 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
3679 ovs_rwlock_wrlock(&rule->evict);
3680 delete_flow__(rule, group, reason);
3682 ofopgroup_submit(group);
3687 /* Implements OFPFC_DELETE. */
3689 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3690 const struct ofputil_flow_mod *fm,
3691 const struct ofp_header *request)
3696 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3697 fm->cookie, fm->cookie_mask,
3698 fm->out_port, &rules);
3699 return (error ? error
3700 : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
3701 &rules, OFPRR_DELETE)
3705 /* Implements OFPFC_DELETE_STRICT. */
3707 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3708 const struct ofputil_flow_mod *fm,
3709 const struct ofp_header *request)
3714 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3715 fm->priority, fm->cookie, fm->cookie_mask,
3716 fm->out_port, &rules);
3717 return (error ? error
3718 : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
3725 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
3727 struct ofputil_flow_removed fr;
3729 if (ofproto_rule_is_hidden(rule) || !rule->send_flow_removed) {
3733 minimatch_expand(&rule->cr.match, &fr.match);
3734 fr.priority = rule->cr.priority;
3735 fr.cookie = rule->flow_cookie;
3737 fr.table_id = rule->table_id;
3738 calc_duration(rule->created, time_msec(),
3739 &fr.duration_sec, &fr.duration_nsec);
3740 ovs_mutex_lock(&rule->timeout_mutex);
3741 fr.idle_timeout = rule->idle_timeout;
3742 fr.hard_timeout = rule->hard_timeout;
3743 ovs_mutex_unlock(&rule->timeout_mutex);
3744 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
3747 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
3751 ofproto_rule_update_used(struct rule *rule, long long int used)
3753 if (used > rule->used) {
3754 struct eviction_group *evg = rule->eviction_group;
3758 heap_change(&evg->rules, &rule->evg_node,
3759 rule_eviction_priority(rule));
3764 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3765 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3768 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
3771 * ofproto implementation ->run() functions should use this function to expire
3772 * OpenFlow flows. */
3774 ofproto_rule_expire(struct rule *rule, uint8_t reason)
3776 struct ofproto *ofproto = rule->ofproto;
3777 struct ofopgroup *group;
3779 ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
3781 ofproto_rule_send_removed(rule, reason);
3783 group = ofopgroup_create_unattached(ofproto);
3784 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
3785 oftable_remove_rule(rule);
3786 ofproto->ofproto_class->rule_destruct(rule);
3787 ofopgroup_submit(group);
3790 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
3791 * means "infinite". */
3793 reduce_timeout(uint16_t max, uint16_t *timeout)
3795 if (max && (!*timeout || *timeout > max)) {
3800 /* If 'idle_timeout' is nonzero, and 'rule' has no idle timeout or an idle
3801 * timeout greater than 'idle_timeout', lowers 'rule''s idle timeout to
3802 * 'idle_timeout' seconds. Similarly for 'hard_timeout'.
3804 * Suitable for implementing OFPACT_FIN_TIMEOUT. */
3806 ofproto_rule_reduce_timeouts(struct rule *rule,
3807 uint16_t idle_timeout, uint16_t hard_timeout)
3808 OVS_EXCLUDED(rule->ofproto->expirable_mutex, rule->timeout_mutex)
3810 if (!idle_timeout && !hard_timeout) {
3814 ovs_mutex_lock(&rule->ofproto->expirable_mutex);
3815 if (list_is_empty(&rule->expirable)) {
3816 list_insert(&rule->ofproto->expirable, &rule->expirable);
3818 ovs_mutex_unlock(&rule->ofproto->expirable_mutex);
3820 ovs_mutex_lock(&rule->timeout_mutex);
3821 reduce_timeout(idle_timeout, &rule->idle_timeout);
3822 reduce_timeout(hard_timeout, &rule->hard_timeout);
3823 ovs_mutex_unlock(&rule->timeout_mutex);
3825 if (!rule->eviction_group) {
3826 eviction_group_add_rule(rule);
3831 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3833 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3834 struct ofputil_flow_mod fm;
3835 uint64_t ofpacts_stub[1024 / 8];
3836 struct ofpbuf ofpacts;
3840 error = reject_slave_controller(ofconn);
3845 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3846 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
3849 error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
3852 goto exit_free_ofpacts;
3855 /* Record the operation for logging a summary report. */
3856 switch (fm.command) {
3862 case OFPFC_MODIFY_STRICT:
3863 ofproto->n_modify++;
3867 case OFPFC_DELETE_STRICT:
3868 ofproto->n_delete++;
3873 if (ofproto->next_op_report == LLONG_MAX) {
3874 ofproto->first_op = now;
3875 ofproto->next_op_report = MAX(now + 10 * 1000,
3876 ofproto->op_backoff);
3877 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
3879 ofproto->last_op = now;
3882 ofpbuf_uninit(&ofpacts);
3888 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
3889 struct ofputil_flow_mod *fm, const struct ofp_header *oh)
3891 if (ofproto->n_pending >= 50) {
3892 ovs_assert(!list_is_empty(&ofproto->pending));
3893 return OFPROTO_POSTPONE;
3896 switch (fm->command) {
3898 return add_flow(ofproto, ofconn, fm, oh);
3901 return modify_flows_loose(ofproto, ofconn, fm, oh);
3903 case OFPFC_MODIFY_STRICT:
3904 return modify_flow_strict(ofproto, ofconn, fm, oh);
3907 return delete_flows_loose(ofproto, ofconn, fm, oh);
3909 case OFPFC_DELETE_STRICT:
3910 return delete_flow_strict(ofproto, ofconn, fm, oh);
3913 if (fm->command > 0xff) {
3914 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
3915 "flow_mod_table_id extension is not enabled",
3918 return OFPERR_OFPFMFC_BAD_COMMAND;
3923 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3925 struct ofputil_role_request request;
3926 struct ofputil_role_request reply;
3930 error = ofputil_decode_role_message(oh, &request);
3935 if (request.role != OFPCR12_ROLE_NOCHANGE) {
3936 if (ofconn_get_role(ofconn) != request.role
3937 && ofconn_has_pending_opgroups(ofconn)) {
3938 return OFPROTO_POSTPONE;
3941 if (request.have_generation_id
3942 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
3943 return OFPERR_OFPRRFC_STALE;
3946 ofconn_set_role(ofconn, request.role);
3949 reply.role = ofconn_get_role(ofconn);
3950 reply.have_generation_id = ofconn_get_master_election_id(
3951 ofconn, &reply.generation_id);
3952 buf = ofputil_encode_role_reply(oh, &reply);
3953 ofconn_send_reply(ofconn, buf);
3959 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
3960 const struct ofp_header *oh)
3962 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
3963 enum ofputil_protocol cur, next;
3965 cur = ofconn_get_protocol(ofconn);
3966 next = ofputil_protocol_set_tid(cur, msg->set != 0);
3967 ofconn_set_protocol(ofconn, next);
3973 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3975 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
3976 enum ofputil_protocol cur, next;
3977 enum ofputil_protocol next_base;
3979 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
3981 return OFPERR_OFPBRC_EPERM;
3984 cur = ofconn_get_protocol(ofconn);
3985 next = ofputil_protocol_set_base(cur, next_base);
3986 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
3987 /* Avoid sending async messages in surprising protocol. */
3988 return OFPROTO_POSTPONE;
3991 ofconn_set_protocol(ofconn, next);
3996 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
3997 const struct ofp_header *oh)
3999 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
4002 format = ntohl(msg->format);
4003 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
4004 return OFPERR_OFPBRC_EPERM;
4007 if (format != ofconn_get_packet_in_format(ofconn)
4008 && ofconn_has_pending_opgroups(ofconn)) {
4009 /* Avoid sending async message in surprsing packet in format. */
4010 return OFPROTO_POSTPONE;
4013 ofconn_set_packet_in_format(ofconn, format);
4018 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
4020 const struct nx_async_config *msg = ofpmsg_body(oh);
4021 uint32_t master[OAM_N_TYPES];
4022 uint32_t slave[OAM_N_TYPES];
4024 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
4025 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
4026 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
4028 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
4029 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
4030 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
4032 ofconn_set_async_config(ofconn, master, slave);
4033 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
4034 !ofconn_get_miss_send_len(ofconn)) {
4035 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
4042 handle_nxt_set_controller_id(struct ofconn *ofconn,
4043 const struct ofp_header *oh)
4045 const struct nx_controller_id *nci = ofpmsg_body(oh);
4047 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
4048 return OFPERR_NXBRC_MUST_BE_ZERO;
4051 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
4056 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4060 if (ofconn_has_pending_opgroups(ofconn)) {
4061 return OFPROTO_POSTPONE;
4064 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
4065 ? OFPRAW_OFPT10_BARRIER_REPLY
4066 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
4067 ofconn_send_reply(ofconn, buf);
4072 ofproto_compose_flow_refresh_update(const struct rule *rule,
4073 enum nx_flow_monitor_flags flags,
4076 struct ofoperation *op = rule->pending;
4077 struct ofputil_flow_update fu;
4080 if (op && op->type == OFOPERATION_ADD && !op->victim) {
4081 /* We'll report the final flow when the operation completes. Reporting
4082 * it now would cause a duplicate report later. */
4086 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
4087 ? NXFME_ADDED : NXFME_MODIFIED);
4089 ovs_mutex_lock(&rule->timeout_mutex);
4090 fu.idle_timeout = rule->idle_timeout;
4091 fu.hard_timeout = rule->hard_timeout;
4092 ovs_mutex_unlock(&rule->timeout_mutex);
4093 fu.table_id = rule->table_id;
4094 fu.cookie = rule->flow_cookie;
4095 minimatch_expand(&rule->cr.match, &match);
4097 fu.priority = rule->cr.priority;
4098 if (!(flags & NXFMF_ACTIONS)) {
4102 fu.ofpacts = rule->ofpacts;
4103 fu.ofpacts_len = rule->ofpacts_len;
4105 /* An operation is in progress. Use the previous version of the flow's
4106 * actions, so that when the operation commits we report the change. */
4108 case OFOPERATION_ADD:
4109 /* We already verified that there was a victim. */
4110 fu.ofpacts = op->victim->ofpacts;
4111 fu.ofpacts_len = op->victim->ofpacts_len;
4114 case OFOPERATION_MODIFY:
4116 fu.ofpacts = op->ofpacts;
4117 fu.ofpacts_len = op->ofpacts_len;
4119 fu.ofpacts = rule->ofpacts;
4120 fu.ofpacts_len = rule->ofpacts_len;
4124 case OFOPERATION_DELETE:
4125 fu.ofpacts = rule->ofpacts;
4126 fu.ofpacts_len = rule->ofpacts_len;
4134 if (list_is_empty(msgs)) {
4135 ofputil_start_flow_update(msgs);
4137 ofputil_append_flow_update(&fu, msgs);
4141 ofmonitor_compose_refresh_updates(struct list *rules, struct list *msgs)
4145 LIST_FOR_EACH (rule, ofproto_node, rules) {
4146 enum nx_flow_monitor_flags flags = rule->monitor_flags;
4147 rule->monitor_flags = 0;
4149 ofproto_compose_flow_refresh_update(rule, flags, msgs);
4154 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
4155 struct rule *rule, uint64_t seqno,
4158 enum nx_flow_monitor_flags update;
4160 if (ofproto_rule_is_hidden(rule)) {
4165 ? ofoperation_has_out_port(rule->pending, m->out_port)
4166 : ofproto_rule_has_out_port(rule, m->out_port))) {
4171 if (rule->add_seqno > seqno) {
4172 update = NXFMF_ADD | NXFMF_MODIFY;
4173 } else if (rule->modify_seqno > seqno) {
4174 update = NXFMF_MODIFY;
4179 if (!(m->flags & update)) {
4183 update = NXFMF_INITIAL;
4186 if (!rule->monitor_flags) {
4187 list_push_back(rules, &rule->ofproto_node);
4189 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
4193 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
4197 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
4198 const struct ofoperation *op;
4199 const struct oftable *table;
4200 struct cls_rule target;
4202 cls_rule_init_from_minimatch(&target, &m->match, 0);
4203 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
4204 struct cls_cursor cursor;
4207 ovs_rwlock_rdlock(&table->cls.rwlock);
4208 cls_cursor_init(&cursor, &table->cls, &target);
4209 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4210 ovs_assert(!rule->pending); /* XXX */
4211 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4213 ovs_rwlock_unlock(&table->cls.rwlock);
4216 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
4217 struct rule *rule = op->rule;
4219 if (((m->table_id == 0xff
4220 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
4221 : m->table_id == rule->table_id))
4222 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
4223 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4226 cls_rule_destroy(&target);
4230 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
4233 if (m->flags & NXFMF_INITIAL) {
4234 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
4239 ofmonitor_collect_resume_rules(struct ofmonitor *m,
4240 uint64_t seqno, struct list *rules)
4242 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
4246 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
4248 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4249 struct ofmonitor **monitors;
4250 size_t n_monitors, allocated_monitors;
4251 struct list replies;
4258 ofpbuf_use_const(&b, oh, ntohs(oh->length));
4260 n_monitors = allocated_monitors = 0;
4262 struct ofputil_flow_monitor_request request;
4263 struct ofmonitor *m;
4266 retval = ofputil_decode_flow_monitor_request(&request, &b);
4267 if (retval == EOF) {
4269 } else if (retval) {
4274 if (request.table_id != 0xff
4275 && request.table_id >= ofproto->n_tables) {
4276 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4280 error = ofmonitor_create(&request, ofconn, &m);
4285 if (n_monitors >= allocated_monitors) {
4286 monitors = x2nrealloc(monitors, &allocated_monitors,
4289 monitors[n_monitors++] = m;
4293 for (i = 0; i < n_monitors; i++) {
4294 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
4297 ofpmp_init(&replies, oh);
4298 ofmonitor_compose_refresh_updates(&rules, &replies);
4299 ofconn_send_replies(ofconn, &replies);
4306 for (i = 0; i < n_monitors; i++) {
4307 ofmonitor_destroy(monitors[i]);
4314 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
4316 struct ofmonitor *m;
4319 id = ofputil_decode_flow_monitor_cancel(oh);
4320 m = ofmonitor_lookup(ofconn, id);
4322 return OFPERR_NXBRC_FM_BAD_ID;
4325 ofmonitor_destroy(m);
4329 /* Meters implementation.
4331 * Meter table entry, indexed by the OpenFlow meter_id.
4332 * These are always dynamically allocated to allocate enough space for
4334 * 'created' is used to compute the duration for meter stats.
4335 * 'list rules' is needed so that we can delete the dependent rules when the
4336 * meter table entry is deleted.
4337 * 'provider_meter_id' is for the provider's private use.
4340 long long int created; /* Time created. */
4341 struct list rules; /* List of "struct rule_dpif"s. */
4342 ofproto_meter_id provider_meter_id;
4343 uint16_t flags; /* Meter flags. */
4344 uint16_t n_bands; /* Number of meter bands. */
4345 struct ofputil_meter_band *bands;
4349 * This is used in instruction validation at flow set-up time,
4350 * as flows may not use non-existing meters.
4351 * This is also used by ofproto-providers to translate OpenFlow meter_ids
4352 * in METER instructions to the corresponding provider meter IDs.
4353 * Return value of UINT32_MAX signifies an invalid meter.
4356 ofproto_get_provider_meter_id(const struct ofproto * ofproto,
4357 uint32_t of_meter_id)
4359 if (of_meter_id && of_meter_id <= ofproto->meter_features.max_meters) {
4360 const struct meter *meter = ofproto->meters[of_meter_id];
4362 return meter->provider_meter_id.uint32;
4369 meter_update(struct meter *meter, const struct ofputil_meter_config *config)
4373 meter->flags = config->flags;
4374 meter->n_bands = config->n_bands;
4375 meter->bands = xmemdup(config->bands,
4376 config->n_bands * sizeof *meter->bands);
4379 static struct meter *
4380 meter_create(const struct ofputil_meter_config *config,
4381 ofproto_meter_id provider_meter_id)
4383 struct meter *meter;
4385 meter = xzalloc(sizeof *meter);
4386 meter->provider_meter_id = provider_meter_id;
4387 meter->created = time_msec();
4388 list_init(&meter->rules);
4390 meter_update(meter, config);
4396 meter_delete(struct ofproto *ofproto, uint32_t first, uint32_t last)
4399 for (mid = first; mid <= last; ++mid) {
4400 struct meter *meter = ofproto->meters[mid];
4402 ofproto->meters[mid] = NULL;
4403 ofproto->ofproto_class->meter_del(ofproto,
4404 meter->provider_meter_id);
4412 handle_add_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
4414 ofproto_meter_id provider_meter_id = { UINT32_MAX };
4415 struct meter **meterp = &ofproto->meters[mm->meter.meter_id];
4419 return OFPERR_OFPMMFC_METER_EXISTS;
4422 error = ofproto->ofproto_class->meter_set(ofproto, &provider_meter_id,
4425 ovs_assert(provider_meter_id.uint32 != UINT32_MAX);
4426 *meterp = meter_create(&mm->meter, provider_meter_id);
4432 handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
4434 struct meter *meter = ofproto->meters[mm->meter.meter_id];
4438 return OFPERR_OFPMMFC_UNKNOWN_METER;
4441 error = ofproto->ofproto_class->meter_set(ofproto,
4442 &meter->provider_meter_id,
4444 ovs_assert(meter->provider_meter_id.uint32 != UINT32_MAX);
4446 meter_update(meter, &mm->meter);
4452 handle_delete_meter(struct ofconn *ofconn, const struct ofp_header *oh,
4453 struct ofputil_meter_mod *mm)
4455 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4456 uint32_t meter_id = mm->meter.meter_id;
4457 uint32_t first, last;
4460 if (meter_id == OFPM13_ALL) {
4462 last = ofproto->meter_features.max_meters;
4464 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
4467 first = last = meter_id;
4470 /* First delete the rules that use this meter. If any of those rules are
4471 * currently being modified, postpone the whole operation until later. */
4473 for (meter_id = first; meter_id <= last; ++meter_id) {
4474 struct meter *meter = ofproto->meters[meter_id];
4475 if (meter && !list_is_empty(&meter->rules)) {
4478 LIST_FOR_EACH (rule, meter_list_node, &meter->rules) {
4479 if (rule->pending) {
4480 return OFPROTO_POSTPONE;
4482 list_push_back(&rules, &rule->ofproto_node);
4486 if (!list_is_empty(&rules)) {
4487 delete_flows__(ofproto, ofconn, oh, &rules, OFPRR_METER_DELETE);
4490 /* Delete the meters. */
4491 meter_delete(ofproto, first, last);
4497 handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4499 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4500 struct ofputil_meter_mod mm;
4501 uint64_t bands_stub[256 / 8];
4502 struct ofpbuf bands;
4506 error = reject_slave_controller(ofconn);
4511 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
4513 error = ofputil_decode_meter_mod(oh, &mm, &bands);
4515 goto exit_free_bands;
4518 meter_id = mm.meter.meter_id;
4520 if (mm.command != OFPMC13_DELETE) {
4521 /* Fails also when meters are not implemented by the provider. */
4522 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
4523 error = OFPERR_OFPMMFC_INVALID_METER;
4524 goto exit_free_bands;
4526 if (mm.meter.n_bands > ofproto->meter_features.max_bands) {
4527 error = OFPERR_OFPMMFC_OUT_OF_BANDS;
4528 goto exit_free_bands;
4532 switch (mm.command) {
4534 error = handle_add_meter(ofproto, &mm);
4537 case OFPMC13_MODIFY:
4538 error = handle_modify_meter(ofproto, &mm);
4541 case OFPMC13_DELETE:
4542 error = handle_delete_meter(ofconn, oh, &mm);
4546 error = OFPERR_OFPMMFC_BAD_COMMAND;
4551 ofpbuf_uninit(&bands);
4556 handle_meter_features_request(struct ofconn *ofconn,
4557 const struct ofp_header *request)
4559 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4560 struct ofputil_meter_features features;
4563 if (ofproto->ofproto_class->meter_get_features) {
4564 ofproto->ofproto_class->meter_get_features(ofproto, &features);
4566 memset(&features, 0, sizeof features);
4568 b = ofputil_encode_meter_features_reply(&features, request);
4570 ofconn_send_reply(ofconn, b);
4575 handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request,
4578 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4579 struct list replies;
4580 uint64_t bands_stub[256 / 8];
4581 struct ofpbuf bands;
4582 uint32_t meter_id, first, last;
4584 ofputil_decode_meter_request(request, &meter_id);
4586 if (meter_id == OFPM13_ALL) {
4588 last = ofproto->meter_features.max_meters;
4590 if (!meter_id || meter_id > ofproto->meter_features.max_meters ||
4591 !ofproto->meters[meter_id]) {
4592 return OFPERR_OFPMMFC_UNKNOWN_METER;
4594 first = last = meter_id;
4597 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
4598 ofpmp_init(&replies, request);
4600 for (meter_id = first; meter_id <= last; ++meter_id) {
4601 struct meter *meter = ofproto->meters[meter_id];
4603 continue; /* Skip non-existing meters. */
4605 if (type == OFPTYPE_METER_STATS_REQUEST) {
4606 struct ofputil_meter_stats stats;
4608 stats.meter_id = meter_id;
4610 /* Provider sets the packet and byte counts, we do the rest. */
4611 stats.flow_count = list_size(&meter->rules);
4612 calc_duration(meter->created, time_msec(),
4613 &stats.duration_sec, &stats.duration_nsec);
4614 stats.n_bands = meter->n_bands;
4615 ofpbuf_clear(&bands);
4617 = ofpbuf_put_uninit(&bands,
4618 meter->n_bands * sizeof *stats.bands);
4620 if (!ofproto->ofproto_class->meter_get(ofproto,
4621 meter->provider_meter_id,
4623 ofputil_append_meter_stats(&replies, &stats);
4625 } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */
4626 struct ofputil_meter_config config;
4628 config.meter_id = meter_id;
4629 config.flags = meter->flags;
4630 config.n_bands = meter->n_bands;
4631 config.bands = meter->bands;
4632 ofputil_append_meter_config(&replies, &config);
4636 ofconn_send_replies(ofconn, &replies);
4637 ofpbuf_uninit(&bands);
4642 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
4644 const struct ofp_header *oh = msg->data;
4648 error = ofptype_decode(&type, oh);
4654 /* OpenFlow requests. */
4655 case OFPTYPE_ECHO_REQUEST:
4656 return handle_echo_request(ofconn, oh);
4658 case OFPTYPE_FEATURES_REQUEST:
4659 return handle_features_request(ofconn, oh);
4661 case OFPTYPE_GET_CONFIG_REQUEST:
4662 return handle_get_config_request(ofconn, oh);
4664 case OFPTYPE_SET_CONFIG:
4665 return handle_set_config(ofconn, oh);
4667 case OFPTYPE_PACKET_OUT:
4668 return handle_packet_out(ofconn, oh);
4670 case OFPTYPE_PORT_MOD:
4671 return handle_port_mod(ofconn, oh);
4673 case OFPTYPE_FLOW_MOD:
4674 return handle_flow_mod(ofconn, oh);
4676 case OFPTYPE_METER_MOD:
4677 return handle_meter_mod(ofconn, oh);
4679 case OFPTYPE_BARRIER_REQUEST:
4680 return handle_barrier_request(ofconn, oh);
4682 case OFPTYPE_ROLE_REQUEST:
4683 return handle_role_request(ofconn, oh);
4685 /* OpenFlow replies. */
4686 case OFPTYPE_ECHO_REPLY:
4689 /* Nicira extension requests. */
4690 case OFPTYPE_FLOW_MOD_TABLE_ID:
4691 return handle_nxt_flow_mod_table_id(ofconn, oh);
4693 case OFPTYPE_SET_FLOW_FORMAT:
4694 return handle_nxt_set_flow_format(ofconn, oh);
4696 case OFPTYPE_SET_PACKET_IN_FORMAT:
4697 return handle_nxt_set_packet_in_format(ofconn, oh);
4699 case OFPTYPE_SET_CONTROLLER_ID:
4700 return handle_nxt_set_controller_id(ofconn, oh);
4702 case OFPTYPE_FLOW_AGE:
4703 /* Nothing to do. */
4706 case OFPTYPE_FLOW_MONITOR_CANCEL:
4707 return handle_flow_monitor_cancel(ofconn, oh);
4709 case OFPTYPE_SET_ASYNC_CONFIG:
4710 return handle_nxt_set_async_config(ofconn, oh);
4712 /* Statistics requests. */
4713 case OFPTYPE_DESC_STATS_REQUEST:
4714 return handle_desc_stats_request(ofconn, oh);
4716 case OFPTYPE_FLOW_STATS_REQUEST:
4717 return handle_flow_stats_request(ofconn, oh);
4719 case OFPTYPE_AGGREGATE_STATS_REQUEST:
4720 return handle_aggregate_stats_request(ofconn, oh);
4722 case OFPTYPE_TABLE_STATS_REQUEST:
4723 return handle_table_stats_request(ofconn, oh);
4725 case OFPTYPE_PORT_STATS_REQUEST:
4726 return handle_port_stats_request(ofconn, oh);
4728 case OFPTYPE_QUEUE_STATS_REQUEST:
4729 return handle_queue_stats_request(ofconn, oh);
4731 case OFPTYPE_PORT_DESC_STATS_REQUEST:
4732 return handle_port_desc_stats_request(ofconn, oh);
4734 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
4735 return handle_flow_monitor_request(ofconn, oh);
4737 case OFPTYPE_METER_STATS_REQUEST:
4738 case OFPTYPE_METER_CONFIG_STATS_REQUEST:
4739 return handle_meter_request(ofconn, oh, type);
4741 case OFPTYPE_METER_FEATURES_STATS_REQUEST:
4742 return handle_meter_features_request(ofconn, oh);
4744 /* FIXME: Change the following once they are implemented: */
4745 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
4746 case OFPTYPE_GET_ASYNC_REQUEST:
4747 case OFPTYPE_GROUP_STATS_REQUEST:
4748 case OFPTYPE_GROUP_DESC_STATS_REQUEST:
4749 case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
4750 case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
4751 return OFPERR_OFPBRC_BAD_TYPE;
4755 case OFPTYPE_FEATURES_REPLY:
4756 case OFPTYPE_GET_CONFIG_REPLY:
4757 case OFPTYPE_PACKET_IN:
4758 case OFPTYPE_FLOW_REMOVED:
4759 case OFPTYPE_PORT_STATUS:
4760 case OFPTYPE_BARRIER_REPLY:
4761 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
4762 case OFPTYPE_DESC_STATS_REPLY:
4763 case OFPTYPE_FLOW_STATS_REPLY:
4764 case OFPTYPE_QUEUE_STATS_REPLY:
4765 case OFPTYPE_PORT_STATS_REPLY:
4766 case OFPTYPE_TABLE_STATS_REPLY:
4767 case OFPTYPE_AGGREGATE_STATS_REPLY:
4768 case OFPTYPE_PORT_DESC_STATS_REPLY:
4769 case OFPTYPE_ROLE_REPLY:
4770 case OFPTYPE_FLOW_MONITOR_PAUSED:
4771 case OFPTYPE_FLOW_MONITOR_RESUMED:
4772 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
4773 case OFPTYPE_GET_ASYNC_REPLY:
4774 case OFPTYPE_GROUP_STATS_REPLY:
4775 case OFPTYPE_GROUP_DESC_STATS_REPLY:
4776 case OFPTYPE_GROUP_FEATURES_STATS_REPLY:
4777 case OFPTYPE_METER_STATS_REPLY:
4778 case OFPTYPE_METER_CONFIG_STATS_REPLY:
4779 case OFPTYPE_METER_FEATURES_STATS_REPLY:
4780 case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
4782 return OFPERR_OFPBRC_BAD_TYPE;
4787 handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
4789 int error = handle_openflow__(ofconn, ofp_msg);
4790 if (error && error != OFPROTO_POSTPONE) {
4791 ofconn_send_error(ofconn, ofp_msg->data, error);
4793 COVERAGE_INC(ofproto_recv_openflow);
4794 return error != OFPROTO_POSTPONE;
4797 /* Asynchronous operations. */
4799 /* Creates and returns a new ofopgroup that is not associated with any
4800 * OpenFlow connection.
4802 * The caller should add operations to the returned group with
4803 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4804 static struct ofopgroup *
4805 ofopgroup_create_unattached(struct ofproto *ofproto)
4807 struct ofopgroup *group = xzalloc(sizeof *group);
4808 group->ofproto = ofproto;
4809 list_init(&group->ofproto_node);
4810 list_init(&group->ops);
4811 list_init(&group->ofconn_node);
4815 /* Creates and returns a new ofopgroup for 'ofproto'.
4817 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
4818 * connection. The 'request' and 'buffer_id' arguments are ignored.
4820 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
4821 * If the ofopgroup eventually fails, then the error reply will include
4822 * 'request'. If the ofopgroup eventually succeeds, then the packet with
4823 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
4825 * The caller should add operations to the returned group with
4826 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4827 static struct ofopgroup *
4828 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
4829 const struct ofp_header *request, uint32_t buffer_id)
4831 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
4833 size_t request_len = ntohs(request->length);
4835 ovs_assert(ofconn_get_ofproto(ofconn) == ofproto);
4837 ofconn_add_opgroup(ofconn, &group->ofconn_node);
4838 group->ofconn = ofconn;
4839 group->request = xmemdup(request, MIN(request_len, 64));
4840 group->buffer_id = buffer_id;
4845 /* Submits 'group' for processing.
4847 * If 'group' contains no operations (e.g. none were ever added, or all of the
4848 * ones that were added completed synchronously), then it is destroyed
4849 * immediately. Otherwise it is added to the ofproto's list of pending
4852 ofopgroup_submit(struct ofopgroup *group)
4854 if (!group->n_running) {
4855 ofopgroup_complete(group);
4857 list_push_back(&group->ofproto->pending, &group->ofproto_node);
4858 group->ofproto->n_pending++;
4863 ofopgroup_complete(struct ofopgroup *group)
4865 struct ofproto *ofproto = group->ofproto;
4867 struct ofconn *abbrev_ofconn;
4868 ovs_be32 abbrev_xid;
4870 struct ofoperation *op, *next_op;
4873 ovs_assert(!group->n_running);
4876 LIST_FOR_EACH (op, group_node, &group->ops) {
4883 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
4884 LIST_FOR_EACH (op, group_node, &group->ops) {
4885 if (op->type != OFOPERATION_DELETE) {
4886 struct ofpbuf *packet;
4889 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
4893 error = rule_execute(op->rule, in_port, packet);
4900 if (!error && !list_is_empty(&group->ofconn_node)) {
4901 abbrev_ofconn = group->ofconn;
4902 abbrev_xid = group->request->xid;
4904 abbrev_ofconn = NULL;
4905 abbrev_xid = htonl(0);
4907 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
4908 struct rule *rule = op->rule;
4910 /* We generally want to report the change to active OpenFlow flow
4911 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
4913 - The operation failed.
4915 - The affected rule is not visible to controllers.
4917 - The operation's only effect was to update rule->modified. */
4919 || ofproto_rule_is_hidden(rule)
4920 || (op->type == OFOPERATION_MODIFY
4922 && rule->flow_cookie == op->flow_cookie))) {
4923 /* Check that we can just cast from ofoperation_type to
4924 * nx_flow_update_event. */
4925 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_ADD
4927 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_DELETE
4929 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_MODIFY
4932 ofmonitor_report(ofproto->connmgr, rule,
4933 (enum nx_flow_update_event) op->type,
4934 op->reason, abbrev_ofconn, abbrev_xid);
4937 rule->pending = NULL;
4940 case OFOPERATION_ADD:
4944 ofproto_rule_destroy__(op->victim);
4945 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
4946 if (vid_mask == VLAN_VID_MASK) {
4947 if (ofproto->vlan_bitmap) {
4948 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
4949 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
4950 bitmap_set1(ofproto->vlan_bitmap, vid);
4951 ofproto->vlans_changed = true;
4954 ofproto->vlans_changed = true;
4958 oftable_substitute_rule(rule, op->victim);
4959 ofproto_rule_destroy__(rule);
4963 case OFOPERATION_DELETE:
4964 ovs_assert(!op->error);
4965 ofproto_rule_destroy__(rule);
4969 case OFOPERATION_MODIFY:
4971 rule->modified = time_msec();
4973 ofproto_rule_change_cookie(ofproto, rule, op->flow_cookie);
4975 free(rule->ofpacts);
4976 rule->ofpacts = op->ofpacts;
4977 rule->ofpacts_len = op->ofpacts_len;
4979 op->ofpacts_len = 0;
4988 ofoperation_destroy(op);
4991 ofmonitor_flush(ofproto->connmgr);
4993 if (!list_is_empty(&group->ofproto_node)) {
4994 ovs_assert(ofproto->n_pending > 0);
4995 ofproto->n_pending--;
4996 list_remove(&group->ofproto_node);
4998 if (!list_is_empty(&group->ofconn_node)) {
4999 list_remove(&group->ofconn_node);
5001 ofconn_send_error(group->ofconn, group->request, error);
5003 connmgr_retry(ofproto->connmgr);
5005 free(group->request);
5009 /* Initiates a new operation on 'rule', of the specified 'type', within
5010 * 'group'. Prior to calling, 'rule' must not have any pending operation.
5012 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
5013 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
5015 * Returns the newly created ofoperation (which is also available as
5016 * rule->pending). */
5017 static struct ofoperation *
5018 ofoperation_create(struct ofopgroup *group, struct rule *rule,
5019 enum ofoperation_type type,
5020 enum ofp_flow_removed_reason reason)
5022 struct ofproto *ofproto = group->ofproto;
5023 struct ofoperation *op;
5025 ovs_assert(!rule->pending);
5027 op = rule->pending = xzalloc(sizeof *op);
5029 list_push_back(&group->ops, &op->group_node);
5032 op->reason = reason;
5033 op->flow_cookie = rule->flow_cookie;
5037 if (type == OFOPERATION_DELETE) {
5038 hmap_insert(&ofproto->deletions, &op->hmap_node,
5039 cls_rule_hash(&rule->cr, rule->table_id));
5046 ofoperation_destroy(struct ofoperation *op)
5048 struct ofopgroup *group = op->group;
5051 op->rule->pending = NULL;
5053 if (op->type == OFOPERATION_DELETE) {
5054 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
5056 list_remove(&op->group_node);
5061 /* Indicates that 'op' completed with status 'error', which is either 0 to
5062 * indicate success or an OpenFlow error code on failure.
5064 * If 'error' is 0, indicating success, the operation will be committed
5065 * permanently to the flow table. There is one interesting subcase:
5067 * - If 'op' is an "add flow" operation that is replacing an existing rule in
5068 * the flow table (the "victim" rule) by a new one, then the caller must
5069 * have uninitialized any derived state in the victim rule, as in step 5 in
5070 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
5071 * performs steps 6 and 7 for the victim rule, most notably by calling its
5072 * ->rule_dealloc() function.
5074 * If 'error' is nonzero, then generally the operation will be rolled back:
5076 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
5077 * restores the original rule. The caller must have uninitialized any
5078 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
5079 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
5080 * and 7 for the new rule, calling its ->rule_dealloc() function.
5082 * - If 'op' is a "modify flow" operation, ofproto restores the original
5085 * - 'op' must not be a "delete flow" operation. Removing a rule is not
5086 * allowed to fail. It must always succeed.
5088 * Please see the large comment in ofproto/ofproto-provider.h titled
5089 * "Asynchronous Operation Support" for more information. */
5091 ofoperation_complete(struct ofoperation *op, enum ofperr error)
5093 struct ofopgroup *group = op->group;
5095 ovs_assert(op->rule->pending == op);
5096 ovs_assert(group->n_running > 0);
5097 ovs_assert(!error || op->type != OFOPERATION_DELETE);
5100 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
5101 ofopgroup_complete(group);
5106 ofoperation_get_victim(struct ofoperation *op)
5108 ovs_assert(op->type == OFOPERATION_ADD);
5113 pick_datapath_id(const struct ofproto *ofproto)
5115 const struct ofport *port;
5117 port = ofproto_get_port(ofproto, OFPP_LOCAL);
5119 uint8_t ea[ETH_ADDR_LEN];
5122 error = netdev_get_etheraddr(port->netdev, ea);
5124 return eth_addr_to_uint64(ea);
5126 VLOG_WARN("%s: could not get MAC address for %s (%s)",
5127 ofproto->name, netdev_get_name(port->netdev),
5128 ovs_strerror(error));
5130 return ofproto->fallback_dpid;
5134 pick_fallback_dpid(void)
5136 uint8_t ea[ETH_ADDR_LEN];
5137 eth_addr_nicira_random(ea);
5138 return eth_addr_to_uint64(ea);
5141 /* Table overflow policy. */
5143 /* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep'
5144 * to NULL if the table is not configured to evict rules or if the table
5145 * contains no evictable rules. (Rules with a readlock on their evict rwlock,
5146 * or with no timeouts are not evictable.) */
5148 choose_rule_to_evict(struct oftable *table, struct rule **rulep)
5150 struct eviction_group *evg;
5153 if (!table->eviction_fields) {
5157 /* In the common case, the outer and inner loops here will each be entered
5160 * - The inner loop normally "return"s in its first iteration. If the
5161 * eviction group has any evictable rules, then it always returns in
5164 * - The outer loop only iterates more than once if the largest eviction
5165 * group has no evictable rules.
5167 * - The outer loop can exit only if table's 'max_flows' is all filled up
5168 * by unevictable rules. */
5169 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
5172 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
5173 if (!ovs_rwlock_trywrlock(&rule->evict)) {
5183 /* Searches 'ofproto' for tables that have more flows than their configured
5184 * maximum and that have flow eviction enabled, and evicts as many flows as
5185 * necessary and currently feasible from them.
5187 * This triggers only when an OpenFlow table has N flows in it and then the
5188 * client configures a maximum number of flows less than N. */
5190 ofproto_evict(struct ofproto *ofproto)
5192 struct ofopgroup *group;
5193 struct oftable *table;
5195 group = ofopgroup_create_unattached(ofproto);
5196 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
5197 while (table->eviction_fields) {
5201 ovs_rwlock_rdlock(&table->cls.rwlock);
5202 n_rules = classifier_count(&table->cls);
5203 ovs_rwlock_unlock(&table->cls.rwlock);
5205 if (n_rules <= table->max_flows) {
5209 if (!choose_rule_to_evict(table, &rule)) {
5213 if (rule->pending) {
5214 ovs_rwlock_unlock(&rule->evict);
5218 ofoperation_create(group, rule,
5219 OFOPERATION_DELETE, OFPRR_EVICTION);
5220 oftable_remove_rule(rule);
5221 ofproto->ofproto_class->rule_destruct(rule);
5224 ofopgroup_submit(group);
5227 /* Eviction groups. */
5229 /* Returns the priority to use for an eviction_group that contains 'n_rules'
5230 * rules. The priority contains low-order random bits to ensure that eviction
5231 * groups with the same number of rules are prioritized randomly. */
5233 eviction_group_priority(size_t n_rules)
5235 uint16_t size = MIN(UINT16_MAX, n_rules);
5236 return (size << 16) | random_uint16();
5239 /* Updates 'evg', an eviction_group within 'table', following a change that
5240 * adds or removes rules in 'evg'. */
5242 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
5244 heap_change(&table->eviction_groups_by_size, &evg->size_node,
5245 eviction_group_priority(heap_count(&evg->rules)));
5248 /* Destroys 'evg', an eviction_group within 'table':
5250 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
5251 * rules themselves, just removes them from the eviction group.)
5253 * - Removes 'evg' from 'table'.
5257 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
5259 while (!heap_is_empty(&evg->rules)) {
5262 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
5263 rule->eviction_group = NULL;
5265 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
5266 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
5267 heap_destroy(&evg->rules);
5271 /* Removes 'rule' from its eviction group, if any. */
5273 eviction_group_remove_rule(struct rule *rule)
5275 if (rule->eviction_group) {
5276 struct oftable *table = &rule->ofproto->tables[rule->table_id];
5277 struct eviction_group *evg = rule->eviction_group;
5279 rule->eviction_group = NULL;
5280 heap_remove(&evg->rules, &rule->evg_node);
5281 if (heap_is_empty(&evg->rules)) {
5282 eviction_group_destroy(table, evg);
5284 eviction_group_resized(table, evg);
5289 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
5290 * returns the hash value. */
5292 eviction_group_hash_rule(struct rule *rule)
5294 struct oftable *table = &rule->ofproto->tables[rule->table_id];
5295 const struct mf_subfield *sf;
5299 hash = table->eviction_group_id_basis;
5300 miniflow_expand(&rule->cr.match.flow, &flow);
5301 for (sf = table->eviction_fields;
5302 sf < &table->eviction_fields[table->n_eviction_fields];
5305 if (mf_are_prereqs_ok(sf->field, &flow)) {
5306 union mf_value value;
5308 mf_get_value(sf->field, &flow, &value);
5310 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
5312 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
5313 unsigned int start = sf->ofs + sf->n_bits;
5314 bitwise_zero(&value, sf->field->n_bytes, start,
5315 sf->field->n_bytes * 8 - start);
5317 hash = hash_bytes(&value, sf->field->n_bytes, hash);
5319 hash = hash_int(hash, 0);
5326 /* Returns an eviction group within 'table' with the given 'id', creating one
5328 static struct eviction_group *
5329 eviction_group_find(struct oftable *table, uint32_t id)
5331 struct eviction_group *evg;
5333 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
5337 evg = xmalloc(sizeof *evg);
5338 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
5339 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
5340 eviction_group_priority(0));
5341 heap_init(&evg->rules);
5346 /* Returns an eviction priority for 'rule'. The return value should be
5347 * interpreted so that higher priorities make a rule more attractive candidates
5350 rule_eviction_priority(struct rule *rule)
5352 long long int hard_expiration;
5353 long long int idle_expiration;
5354 long long int expiration;
5355 uint32_t expiration_offset;
5357 /* Calculate time of expiration. */
5358 ovs_mutex_lock(&rule->timeout_mutex);
5359 hard_expiration = (rule->hard_timeout
5360 ? rule->modified + rule->hard_timeout * 1000
5362 idle_expiration = (rule->idle_timeout
5363 ? rule->used + rule->idle_timeout * 1000
5365 expiration = MIN(hard_expiration, idle_expiration);
5366 ovs_mutex_unlock(&rule->timeout_mutex);
5367 if (expiration == LLONG_MAX) {
5371 /* Calculate the time of expiration as a number of (approximate) seconds
5372 * after program startup.
5374 * This should work OK for program runs that last UINT32_MAX seconds or
5375 * less. Therefore, please restart OVS at least once every 136 years. */
5376 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
5378 /* Invert the expiration offset because we're using a max-heap. */
5379 return UINT32_MAX - expiration_offset;
5382 /* Adds 'rule' to an appropriate eviction group for its oftable's
5383 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
5384 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
5387 * The caller must ensure that 'rule' is not already in an eviction group. */
5389 eviction_group_add_rule(struct rule *rule)
5391 struct ofproto *ofproto = rule->ofproto;
5392 struct oftable *table = &ofproto->tables[rule->table_id];
5395 ovs_mutex_lock(&rule->timeout_mutex);
5396 has_timeout = rule->hard_timeout || rule->idle_timeout;
5397 ovs_mutex_unlock(&rule->timeout_mutex);
5399 if (table->eviction_fields && has_timeout) {
5400 struct eviction_group *evg;
5402 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
5404 rule->eviction_group = evg;
5405 heap_insert(&evg->rules, &rule->evg_node,
5406 rule_eviction_priority(rule));
5407 eviction_group_resized(table, evg);
5413 /* Initializes 'table'. */
5415 oftable_init(struct oftable *table)
5417 memset(table, 0, sizeof *table);
5418 classifier_init(&table->cls);
5419 table->max_flows = UINT_MAX;
5422 /* Destroys 'table', including its classifier and eviction groups.
5424 * The caller is responsible for freeing 'table' itself. */
5426 oftable_destroy(struct oftable *table)
5428 ovs_rwlock_rdlock(&table->cls.rwlock);
5429 ovs_assert(classifier_is_empty(&table->cls));
5430 ovs_rwlock_unlock(&table->cls.rwlock);
5431 oftable_disable_eviction(table);
5432 classifier_destroy(&table->cls);
5436 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
5437 * string, then 'table' will use its default name.
5439 * This only affects the name exposed for a table exposed through the OpenFlow
5440 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
5442 oftable_set_name(struct oftable *table, const char *name)
5444 if (name && name[0]) {
5445 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
5446 if (!table->name || strncmp(name, table->name, len)) {
5448 table->name = xmemdup0(name, len);
5456 /* oftables support a choice of two policies when adding a rule would cause the
5457 * number of flows in the table to exceed the configured maximum number: either
5458 * they can refuse to add the new flow or they can evict some existing flow.
5459 * This function configures the former policy on 'table'. */
5461 oftable_disable_eviction(struct oftable *table)
5463 if (table->eviction_fields) {
5464 struct eviction_group *evg, *next;
5466 HMAP_FOR_EACH_SAFE (evg, next, id_node,
5467 &table->eviction_groups_by_id) {
5468 eviction_group_destroy(table, evg);
5470 hmap_destroy(&table->eviction_groups_by_id);
5471 heap_destroy(&table->eviction_groups_by_size);
5473 free(table->eviction_fields);
5474 table->eviction_fields = NULL;
5475 table->n_eviction_fields = 0;
5479 /* oftables support a choice of two policies when adding a rule would cause the
5480 * number of flows in the table to exceed the configured maximum number: either
5481 * they can refuse to add the new flow or they can evict some existing flow.
5482 * This function configures the latter policy on 'table', with fairness based
5483 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
5484 * 'n_fields' as 0 disables fairness.) */
5486 oftable_enable_eviction(struct oftable *table,
5487 const struct mf_subfield *fields, size_t n_fields)
5489 struct cls_cursor cursor;
5492 if (table->eviction_fields
5493 && n_fields == table->n_eviction_fields
5495 || !memcmp(fields, table->eviction_fields,
5496 n_fields * sizeof *fields))) {
5501 oftable_disable_eviction(table);
5503 table->n_eviction_fields = n_fields;
5504 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
5506 table->eviction_group_id_basis = random_uint32();
5507 hmap_init(&table->eviction_groups_by_id);
5508 heap_init(&table->eviction_groups_by_size);
5510 ovs_rwlock_rdlock(&table->cls.rwlock);
5511 cls_cursor_init(&cursor, &table->cls, NULL);
5512 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
5513 eviction_group_add_rule(rule);
5515 ovs_rwlock_unlock(&table->cls.rwlock);
5518 /* Removes 'rule' from the oftable that contains it. */
5520 oftable_remove_rule__(struct ofproto *ofproto, struct classifier *cls,
5522 OVS_REQ_WRLOCK(cls->rwlock) OVS_RELEASES(rule->evict)
5524 classifier_remove(cls, &rule->cr);
5525 if (rule->meter_id) {
5526 list_remove(&rule->meter_list_node);
5528 cookies_remove(ofproto, rule);
5529 eviction_group_remove_rule(rule);
5530 ovs_mutex_lock(&ofproto->expirable_mutex);
5531 if (!list_is_empty(&rule->expirable)) {
5532 list_remove(&rule->expirable);
5534 ovs_mutex_unlock(&ofproto->expirable_mutex);
5535 if (!list_is_empty(&rule->meter_list_node)) {
5536 list_remove(&rule->meter_list_node);
5538 ovs_rwlock_unlock(&rule->evict);
5542 oftable_remove_rule(struct rule *rule)
5544 struct ofproto *ofproto = rule->ofproto;
5545 struct oftable *table = &ofproto->tables[rule->table_id];
5547 ovs_rwlock_wrlock(&table->cls.rwlock);
5548 oftable_remove_rule__(ofproto, &table->cls, rule);
5549 ovs_rwlock_unlock(&table->cls.rwlock);
5552 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
5553 * oftable that has an identical cls_rule. Returns the rule that was removed,
5554 * if any, and otherwise NULL. */
5555 static struct rule *
5556 oftable_replace_rule(struct rule *rule)
5558 struct ofproto *ofproto = rule->ofproto;
5559 struct oftable *table = &ofproto->tables[rule->table_id];
5560 struct rule *victim;
5563 ovs_mutex_lock(&rule->timeout_mutex);
5564 may_expire = rule->hard_timeout || rule->idle_timeout;
5565 ovs_mutex_unlock(&rule->timeout_mutex);
5568 ovs_mutex_lock(&ofproto->expirable_mutex);
5569 list_insert(&ofproto->expirable, &rule->expirable);
5570 ovs_mutex_unlock(&ofproto->expirable_mutex);
5572 cookies_insert(ofproto, rule);
5573 if (rule->meter_id) {
5574 struct meter *meter = ofproto->meters[rule->meter_id];
5575 list_insert(&meter->rules, &rule->meter_list_node);
5577 ovs_rwlock_wrlock(&table->cls.rwlock);
5578 victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr));
5579 ovs_rwlock_unlock(&table->cls.rwlock);
5581 if (victim->meter_id) {
5582 list_remove(&victim->meter_list_node);
5584 cookies_remove(ofproto, victim);
5586 ovs_mutex_lock(&ofproto->expirable_mutex);
5587 if (!list_is_empty(&victim->expirable)) {
5588 list_remove(&victim->expirable);
5590 ovs_mutex_unlock(&ofproto->expirable_mutex);
5591 eviction_group_remove_rule(victim);
5593 eviction_group_add_rule(rule);
5597 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
5599 oftable_substitute_rule(struct rule *old, struct rule *new)
5602 oftable_replace_rule(new);
5604 ovs_rwlock_wrlock(&old->evict);
5605 oftable_remove_rule(old);
5609 /* unixctl commands. */
5612 ofproto_lookup(const char *name)
5614 struct ofproto *ofproto;
5616 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
5618 if (!strcmp(ofproto->name, name)) {
5626 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
5627 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5629 struct ofproto *ofproto;
5633 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
5634 ds_put_format(&results, "%s\n", ofproto->name);
5636 unixctl_command_reply(conn, ds_cstr(&results));
5637 ds_destroy(&results);
5641 ofproto_unixctl_init(void)
5643 static bool registered;
5649 unixctl_command_register("ofproto/list", "", 0, 0,
5650 ofproto_unixctl_list, NULL);
5653 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5655 * This is deprecated. It is only for compatibility with broken device drivers
5656 * in old versions of Linux that do not properly support VLANs when VLAN
5657 * devices are not used. When broken device drivers are no longer in
5658 * widespread use, we will delete these interfaces. */
5660 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
5661 * (exactly) by an OpenFlow rule in 'ofproto'. */
5663 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
5665 const struct oftable *oftable;
5667 free(ofproto->vlan_bitmap);
5668 ofproto->vlan_bitmap = bitmap_allocate(4096);
5669 ofproto->vlans_changed = false;
5671 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
5672 const struct cls_table *table;
5674 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
5675 if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
5676 const struct cls_rule *rule;
5678 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
5679 uint16_t vid = miniflow_get_vid(&rule->match.flow);
5680 bitmap_set1(vlan_bitmap, vid);
5681 bitmap_set1(ofproto->vlan_bitmap, vid);
5688 /* Returns true if new VLANs have come into use by the flow table since the
5689 * last call to ofproto_get_vlan_usage().
5691 * We don't track when old VLANs stop being used. */
5693 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
5695 return ofproto->vlans_changed;
5698 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
5699 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
5700 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
5701 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
5702 * then the VLAN device is un-enslaved. */
5704 ofproto_port_set_realdev(struct ofproto *ofproto, ofp_port_t vlandev_ofp_port,
5705 ofp_port_t realdev_ofp_port, int vid)
5707 struct ofport *ofport;
5710 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
5712 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
5714 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
5715 ofproto->name, vlandev_ofp_port);
5719 if (!ofproto->ofproto_class->set_realdev) {
5720 if (!vlandev_ofp_port) {
5723 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
5727 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
5729 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
5730 ofproto->name, vlandev_ofp_port,
5731 netdev_get_name(ofport->netdev), ovs_strerror(error));