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 *);
191 static void ofport_destroy__(struct ofport *);
192 static void ofport_destroy(struct ofport *);
194 static void update_port(struct ofproto *, const char *devname);
195 static int init_ports(struct ofproto *);
196 static void reinit_ports(struct ofproto *);
199 static void ofproto_rule_destroy__(struct rule *);
200 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
201 static bool rule_is_modifiable(const struct rule *);
204 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
205 struct ofputil_flow_mod *,
206 const struct ofp_header *);
207 static void delete_flow__(struct rule *rule, struct ofopgroup *,
208 enum ofp_flow_removed_reason)
209 OVS_RELEASES(rule->evict);
210 static bool handle_openflow(struct ofconn *, const struct ofpbuf *);
211 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
212 struct ofputil_flow_mod *,
213 const struct ofp_header *);
214 static void calc_duration(long long int start, long long int now,
215 uint32_t *sec, uint32_t *nsec);
218 static uint64_t pick_datapath_id(const struct ofproto *);
219 static uint64_t pick_fallback_dpid(void);
220 static void ofproto_destroy__(struct ofproto *);
221 static void update_mtu(struct ofproto *, struct ofport *);
222 static void meter_delete(struct ofproto *, uint32_t first, uint32_t last);
225 static void ofproto_unixctl_init(void);
227 /* All registered ofproto classes, in probe order. */
228 static const struct ofproto_class **ofproto_classes;
229 static size_t n_ofproto_classes;
230 static size_t allocated_ofproto_classes;
232 unsigned flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT;
233 unsigned n_handler_threads;
234 enum ofproto_flow_miss_model flow_miss_model = OFPROTO_HANDLE_MISS_AUTO;
236 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
237 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
239 /* Initial mappings of port to OpenFlow number mappings. */
240 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
242 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
244 /* The default value of true waits for flow restore. */
245 static bool flow_restore_wait = true;
247 /* Must be called to initialize the ofproto library.
249 * The caller may pass in 'iface_hints', which contains an shash of
250 * "iface_hint" elements indexed by the interface's name. The provider
251 * may use these hints to describe the startup configuration in order to
252 * reinitialize its state. The caller owns the provided data, so a
253 * provider will make copies of anything required. An ofproto provider
254 * will remove any existing state that is not described by the hint, and
255 * may choose to remove it all. */
257 ofproto_init(const struct shash *iface_hints)
259 struct shash_node *node;
262 ofproto_class_register(&ofproto_dpif_class);
264 /* Make a local copy, since we don't own 'iface_hints' elements. */
265 SHASH_FOR_EACH(node, iface_hints) {
266 const struct iface_hint *orig_hint = node->data;
267 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
268 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
270 new_hint->br_name = xstrdup(orig_hint->br_name);
271 new_hint->br_type = xstrdup(br_type);
272 new_hint->ofp_port = orig_hint->ofp_port;
274 shash_add(&init_ofp_ports, node->name, new_hint);
277 for (i = 0; i < n_ofproto_classes; i++) {
278 ofproto_classes[i]->init(&init_ofp_ports);
282 /* 'type' should be a normalized datapath type, as returned by
283 * ofproto_normalize_type(). Returns the corresponding ofproto_class
284 * structure, or a null pointer if there is none registered for 'type'. */
285 static const struct ofproto_class *
286 ofproto_class_find__(const char *type)
290 for (i = 0; i < n_ofproto_classes; i++) {
291 const struct ofproto_class *class = ofproto_classes[i];
296 class->enumerate_types(&types);
297 found = sset_contains(&types, type);
298 sset_destroy(&types);
304 VLOG_WARN("unknown datapath type %s", type);
308 /* Registers a new ofproto class. After successful registration, new ofprotos
309 * of that type can be created using ofproto_create(). */
311 ofproto_class_register(const struct ofproto_class *new_class)
315 for (i = 0; i < n_ofproto_classes; i++) {
316 if (ofproto_classes[i] == new_class) {
321 if (n_ofproto_classes >= allocated_ofproto_classes) {
322 ofproto_classes = x2nrealloc(ofproto_classes,
323 &allocated_ofproto_classes,
324 sizeof *ofproto_classes);
326 ofproto_classes[n_ofproto_classes++] = new_class;
330 /* Unregisters a datapath provider. 'type' must have been previously
331 * registered and not currently be in use by any ofprotos. After
332 * unregistration new datapaths of that type cannot be opened using
333 * ofproto_create(). */
335 ofproto_class_unregister(const struct ofproto_class *class)
339 for (i = 0; i < n_ofproto_classes; i++) {
340 if (ofproto_classes[i] == class) {
341 for (i++; i < n_ofproto_classes; i++) {
342 ofproto_classes[i - 1] = ofproto_classes[i];
348 VLOG_WARN("attempted to unregister an ofproto class that is not "
353 /* Clears 'types' and enumerates all registered ofproto types into it. The
354 * caller must first initialize the sset. */
356 ofproto_enumerate_types(struct sset *types)
360 for (i = 0; i < n_ofproto_classes; i++) {
361 ofproto_classes[i]->enumerate_types(types);
365 /* Returns the fully spelled out name for the given ofproto 'type'.
367 * Normalized type string can be compared with strcmp(). Unnormalized type
368 * string might be the same even if they have different spellings. */
370 ofproto_normalize_type(const char *type)
372 return type && type[0] ? type : "system";
375 /* Clears 'names' and enumerates the names of all known created ofprotos with
376 * the given 'type'. The caller must first initialize the sset. Returns 0 if
377 * successful, otherwise a positive errno value.
379 * Some kinds of datapaths might not be practically enumerable. This is not
380 * considered an error. */
382 ofproto_enumerate_names(const char *type, struct sset *names)
384 const struct ofproto_class *class = ofproto_class_find__(type);
385 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
389 ofproto_create(const char *datapath_name, const char *datapath_type,
390 struct ofproto **ofprotop)
392 const struct ofproto_class *class;
393 struct ofproto *ofproto;
399 ofproto_unixctl_init();
401 datapath_type = ofproto_normalize_type(datapath_type);
402 class = ofproto_class_find__(datapath_type);
404 VLOG_WARN("could not create datapath %s of unknown type %s",
405 datapath_name, datapath_type);
409 ofproto = class->alloc();
411 VLOG_ERR("failed to allocate datapath %s of type %s",
412 datapath_name, datapath_type);
417 memset(ofproto, 0, sizeof *ofproto);
418 ofproto->ofproto_class = class;
419 ofproto->name = xstrdup(datapath_name);
420 ofproto->type = xstrdup(datapath_type);
421 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
422 hash_string(ofproto->name, 0));
423 ofproto->datapath_id = 0;
424 ofproto->forward_bpdu = false;
425 ofproto->fallback_dpid = pick_fallback_dpid();
426 ofproto->mfr_desc = NULL;
427 ofproto->hw_desc = NULL;
428 ofproto->sw_desc = NULL;
429 ofproto->serial_desc = NULL;
430 ofproto->dp_desc = NULL;
431 ofproto->frag_handling = OFPC_FRAG_NORMAL;
432 hmap_init(&ofproto->ports);
433 shash_init(&ofproto->port_by_name);
434 simap_init(&ofproto->ofp_requests);
435 ofproto->max_ports = OFPP_MAX;
436 ofproto->tables = NULL;
437 ofproto->n_tables = 0;
438 hindex_init(&ofproto->cookies);
439 list_init(&ofproto->expirable);
440 ovs_mutex_init(&ofproto->expirable_mutex, PTHREAD_MUTEX_RECURSIVE);
441 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
442 ofproto->state = S_OPENFLOW;
443 list_init(&ofproto->pending);
444 ofproto->n_pending = 0;
445 hmap_init(&ofproto->deletions);
446 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
447 ofproto->first_op = ofproto->last_op = LLONG_MIN;
448 ofproto->next_op_report = LLONG_MAX;
449 ofproto->op_backoff = LLONG_MIN;
450 ofproto->vlan_bitmap = NULL;
451 ofproto->vlans_changed = false;
452 ofproto->min_mtu = INT_MAX;
454 error = ofproto->ofproto_class->construct(ofproto);
456 VLOG_ERR("failed to open datapath %s: %s",
457 datapath_name, ovs_strerror(error));
458 ofproto_destroy__(ofproto);
462 /* The "max_ports" member should have been set by ->construct(ofproto).
463 * Port 0 is not a valid OpenFlow port, so mark that as unavailable. */
464 ofproto->ofp_port_ids = bitmap_allocate(ofp_to_u16(ofproto->max_ports));
465 bitmap_set1(ofproto->ofp_port_ids, 0);
467 /* Check that hidden tables, if any, are at the end. */
468 ovs_assert(ofproto->n_tables);
469 for (i = 0; i + 1 < ofproto->n_tables; i++) {
470 enum oftable_flags flags = ofproto->tables[i].flags;
471 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
473 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
476 ofproto->datapath_id = pick_datapath_id(ofproto);
479 /* Initialize meters table. */
480 if (ofproto->ofproto_class->meter_get_features) {
481 ofproto->ofproto_class->meter_get_features(ofproto,
482 &ofproto->meter_features);
484 memset(&ofproto->meter_features, 0, sizeof ofproto->meter_features);
486 ofproto->meters = xzalloc((ofproto->meter_features.max_meters + 1)
487 * sizeof(struct meter *));
493 /* Must be called (only) by an ofproto implementation in its constructor
494 * function. See the large comment on 'construct' in struct ofproto_class for
497 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
499 struct oftable *table;
501 ovs_assert(!ofproto->n_tables);
502 ovs_assert(n_tables >= 1 && n_tables <= 255);
504 ofproto->n_tables = n_tables;
505 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
506 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
511 /* To be optionally called (only) by an ofproto implementation in its
512 * constructor function. See the large comment on 'construct' in struct
513 * ofproto_class for details.
515 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
516 * will then ensure that actions passed into the ofproto implementation will
517 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
518 * function is not called, there will be no such restriction.
520 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
521 * the 'max_ports' restriction. */
523 ofproto_init_max_ports(struct ofproto *ofproto, ofp_port_t max_ports)
525 ovs_assert(ofp_to_u16(max_ports) <= ofp_to_u16(OFPP_MAX));
526 ofproto->max_ports = max_ports;
530 ofproto_get_datapath_id(const struct ofproto *ofproto)
532 return ofproto->datapath_id;
536 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
538 uint64_t old_dpid = p->datapath_id;
539 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
540 if (p->datapath_id != old_dpid) {
541 /* Force all active connections to reconnect, since there is no way to
542 * notify a controller that the datapath ID has changed. */
543 ofproto_reconnect_controllers(p);
548 ofproto_set_controllers(struct ofproto *p,
549 const struct ofproto_controller *controllers,
550 size_t n_controllers, uint32_t allowed_versions)
552 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
557 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
559 connmgr_set_fail_mode(p->connmgr, fail_mode);
562 /* Drops the connections between 'ofproto' and all of its controllers, forcing
563 * them to reconnect. */
565 ofproto_reconnect_controllers(struct ofproto *ofproto)
567 connmgr_reconnect(ofproto->connmgr);
570 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
571 * in-band control should guarantee access, in the same way that in-band
572 * control guarantees access to OpenFlow controllers. */
574 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
575 const struct sockaddr_in *extras, size_t n)
577 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
580 /* Sets the OpenFlow queue used by flows set up by in-band control on
581 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
582 * flows will use the default queue. */
584 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
586 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
589 /* Sets the number of flows at which eviction from the kernel flow table
592 ofproto_set_flow_eviction_threshold(unsigned threshold)
594 flow_eviction_threshold = MAX(OFPROTO_FLOW_EVICTION_THRESHOLD_MIN,
598 /* Sets the path for handling flow misses. */
600 ofproto_set_flow_miss_model(unsigned model)
602 flow_miss_model = model;
605 /* If forward_bpdu is true, the NORMAL action will forward frames with
606 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
607 * the NORMAL action will drop these frames. */
609 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
611 bool old_val = ofproto->forward_bpdu;
612 ofproto->forward_bpdu = forward_bpdu;
613 if (old_val != ofproto->forward_bpdu) {
614 if (ofproto->ofproto_class->forward_bpdu_changed) {
615 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
620 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
621 * 'idle_time', in seconds, and the maximum number of MAC table entries to
624 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
627 if (ofproto->ofproto_class->set_mac_table_config) {
628 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
633 /* Sets number of upcall handler threads. The default is
634 * (number of online cores - 1). */
636 ofproto_set_n_handler_threads(unsigned limit)
639 n_handler_threads = limit;
641 n_handler_threads = MAX(1, sysconf(_SC_NPROCESSORS_ONLN) - 1);
646 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
649 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
653 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
655 return connmgr_set_snoops(ofproto->connmgr, snoops);
659 ofproto_set_netflow(struct ofproto *ofproto,
660 const struct netflow_options *nf_options)
662 if (nf_options && sset_is_empty(&nf_options->collectors)) {
666 if (ofproto->ofproto_class->set_netflow) {
667 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
669 return nf_options ? EOPNOTSUPP : 0;
674 ofproto_set_sflow(struct ofproto *ofproto,
675 const struct ofproto_sflow_options *oso)
677 if (oso && sset_is_empty(&oso->targets)) {
681 if (ofproto->ofproto_class->set_sflow) {
682 return ofproto->ofproto_class->set_sflow(ofproto, oso);
684 return oso ? EOPNOTSUPP : 0;
689 ofproto_set_ipfix(struct ofproto *ofproto,
690 const struct ofproto_ipfix_bridge_exporter_options *bo,
691 const struct ofproto_ipfix_flow_exporter_options *fo,
694 if (ofproto->ofproto_class->set_ipfix) {
695 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
697 return (bo || fo) ? EOPNOTSUPP : 0;
702 ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
704 flow_restore_wait = flow_restore_wait_db;
708 ofproto_get_flow_restore_wait(void)
710 return flow_restore_wait;
714 /* Spanning Tree Protocol (STP) configuration. */
716 /* Configures STP on 'ofproto' using the settings defined in 's'. If
717 * 's' is NULL, disables STP.
719 * Returns 0 if successful, otherwise a positive errno value. */
721 ofproto_set_stp(struct ofproto *ofproto,
722 const struct ofproto_stp_settings *s)
724 return (ofproto->ofproto_class->set_stp
725 ? ofproto->ofproto_class->set_stp(ofproto, s)
729 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
730 * 'enabled' member of 's' is false, then the other members are not
733 * Returns 0 if successful, otherwise a positive errno value. */
735 ofproto_get_stp_status(struct ofproto *ofproto,
736 struct ofproto_stp_status *s)
738 return (ofproto->ofproto_class->get_stp_status
739 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
743 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
744 * in 's'. The caller is responsible for assigning STP port numbers
745 * (using the 'port_num' member in the range of 1 through 255, inclusive)
746 * and ensuring there are no duplicates. If the 's' is NULL, then STP
747 * is disabled on the port.
749 * Returns 0 if successful, otherwise a positive errno value.*/
751 ofproto_port_set_stp(struct ofproto *ofproto, ofp_port_t ofp_port,
752 const struct ofproto_port_stp_settings *s)
754 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
756 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
757 ofproto->name, ofp_port);
761 return (ofproto->ofproto_class->set_stp_port
762 ? ofproto->ofproto_class->set_stp_port(ofport, s)
766 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
767 * 's'. If the 'enabled' member in 's' is false, then the other members
768 * are not meaningful.
770 * Returns 0 if successful, otherwise a positive errno value.*/
772 ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
773 struct ofproto_port_stp_status *s)
775 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
777 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
778 "port %"PRIu16, ofproto->name, ofp_port);
782 return (ofproto->ofproto_class->get_stp_port_status
783 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
787 /* Queue DSCP configuration. */
789 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
790 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
791 * to implement QoS. Instead, it is used to implement features which require
792 * knowledge of what queues exist on a port, and some basic information about
795 * Returns 0 if successful, otherwise a positive errno value. */
797 ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port,
798 const struct ofproto_port_queue *queues,
801 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
804 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
805 ofproto->name, ofp_port);
809 return (ofproto->ofproto_class->set_queues
810 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
814 /* Connectivity Fault Management configuration. */
816 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
818 ofproto_port_clear_cfm(struct ofproto *ofproto, ofp_port_t ofp_port)
820 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
821 if (ofport && ofproto->ofproto_class->set_cfm) {
822 ofproto->ofproto_class->set_cfm(ofport, NULL);
826 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
827 * basic configuration from the configuration members in 'cfm', and the remote
828 * maintenance point ID from remote_mpid. Ignores the statistics members of
831 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
833 ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port,
834 const struct cfm_settings *s)
836 struct ofport *ofport;
839 ofport = ofproto_get_port(ofproto, ofp_port);
841 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
842 ofproto->name, ofp_port);
846 /* XXX: For configuration simplicity, we only support one remote_mpid
847 * outside of the CFM module. It's not clear if this is the correct long
848 * term solution or not. */
849 error = (ofproto->ofproto_class->set_cfm
850 ? ofproto->ofproto_class->set_cfm(ofport, s)
853 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
854 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
855 ovs_strerror(error));
859 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
860 * 'ofproto' does not have a port 'ofp_port'. */
862 ofproto_port_set_bfd(struct ofproto *ofproto, ofp_port_t ofp_port,
863 const struct smap *cfg)
865 struct ofport *ofport;
868 ofport = ofproto_get_port(ofproto, ofp_port);
870 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
871 ofproto->name, ofp_port);
875 error = (ofproto->ofproto_class->set_bfd
876 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
879 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
880 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
881 ovs_strerror(error));
885 /* Populates 'status' with key value pairs indicating the status of the BFD
886 * session on 'ofp_port'. This information is intended to be populated in the
887 * OVS database. Has no effect if 'ofp_port' is not na OpenFlow port in
890 ofproto_port_get_bfd_status(struct ofproto *ofproto, ofp_port_t ofp_port,
893 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
894 return (ofport && ofproto->ofproto_class->get_bfd_status
895 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
899 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
900 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
901 * 0 if LACP partner information is not current (generally indicating a
902 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
904 ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port)
906 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
907 return (ofport && ofproto->ofproto_class->port_is_lacp_current
908 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
914 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
915 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
916 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
917 * configuration plus, if there is more than one slave, a bonding
920 * If 'aux' is already registered then this function updates its configuration
921 * to 's'. Otherwise, this function registers a new bundle.
923 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
926 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
927 const struct ofproto_bundle_settings *s)
929 return (ofproto->ofproto_class->bundle_set
930 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
934 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
935 * If no such bundle has been registered, this has no effect. */
937 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
939 return ofproto_bundle_register(ofproto, aux, NULL);
943 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
944 * If 'aux' is already registered then this function updates its configuration
945 * to 's'. Otherwise, this function registers a new mirror. */
947 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
948 const struct ofproto_mirror_settings *s)
950 return (ofproto->ofproto_class->mirror_set
951 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
955 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
956 * If no mirror has been registered, this has no effect. */
958 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
960 return ofproto_mirror_register(ofproto, aux, NULL);
963 /* Retrieves statistics from mirror associated with client data pointer
964 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
965 * 'bytes', respectively. If a particular counters is not supported,
966 * the appropriate argument is set to UINT64_MAX. */
968 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
969 uint64_t *packets, uint64_t *bytes)
971 if (!ofproto->ofproto_class->mirror_get_stats) {
972 *packets = *bytes = UINT64_MAX;
976 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
980 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
981 * which all packets are flooded, instead of using MAC learning. If
982 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
984 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
987 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
989 return (ofproto->ofproto_class->set_flood_vlans
990 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
994 /* Returns true if 'aux' is a registered bundle that is currently in use as the
995 * output for a mirror. */
997 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
999 return (ofproto->ofproto_class->is_mirror_output_bundle
1000 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
1004 /* Configuration of OpenFlow tables. */
1006 /* Returns the number of OpenFlow tables in 'ofproto'. */
1008 ofproto_get_n_tables(const struct ofproto *ofproto)
1010 return ofproto->n_tables;
1013 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
1014 * settings from 's'. 'table_id' must be in the range 0 through the number of
1015 * OpenFlow tables in 'ofproto' minus 1, inclusive.
1017 * For read-only tables, only the name may be configured. */
1019 ofproto_configure_table(struct ofproto *ofproto, int table_id,
1020 const struct ofproto_table_settings *s)
1022 struct oftable *table;
1024 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
1025 table = &ofproto->tables[table_id];
1027 oftable_set_name(table, s->name);
1029 if (table->flags & OFTABLE_READONLY) {
1034 oftable_enable_eviction(table, s->groups, s->n_groups);
1036 oftable_disable_eviction(table);
1039 table->max_flows = s->max_flows;
1040 ovs_rwlock_rdlock(&table->cls.rwlock);
1041 if (classifier_count(&table->cls) > table->max_flows
1042 && table->eviction_fields) {
1043 /* 'table' contains more flows than allowed. We might not be able to
1044 * evict them right away because of the asynchronous nature of flow
1045 * table changes. Schedule eviction for later. */
1046 switch (ofproto->state) {
1048 ofproto->state = S_EVICT;
1052 /* We're already deleting flows, nothing more to do. */
1056 ovs_rwlock_unlock(&table->cls.rwlock);
1060 ofproto_has_snoops(const struct ofproto *ofproto)
1062 return connmgr_has_snoops(ofproto->connmgr);
1066 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1068 connmgr_get_snoops(ofproto->connmgr, snoops);
1072 ofproto_flush__(struct ofproto *ofproto)
1074 struct ofopgroup *group;
1075 struct oftable *table;
1077 if (ofproto->ofproto_class->flush) {
1078 ofproto->ofproto_class->flush(ofproto);
1081 group = ofopgroup_create_unattached(ofproto);
1082 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1083 struct rule *rule, *next_rule;
1084 struct cls_cursor cursor;
1086 if (table->flags & OFTABLE_HIDDEN) {
1090 ovs_rwlock_wrlock(&table->cls.rwlock);
1091 cls_cursor_init(&cursor, &table->cls, NULL);
1092 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1093 if (!rule->pending) {
1094 ofoperation_create(group, rule, OFOPERATION_DELETE,
1096 ovs_rwlock_wrlock(&rule->evict);
1097 oftable_remove_rule__(ofproto, &table->cls, rule);
1098 ofproto->ofproto_class->rule_destruct(rule);
1101 ovs_rwlock_unlock(&table->cls.rwlock);
1103 ofopgroup_submit(group);
1107 ofproto_destroy__(struct ofproto *ofproto)
1109 struct oftable *table;
1111 ovs_assert(list_is_empty(&ofproto->pending));
1112 ovs_assert(!ofproto->n_pending);
1114 if (ofproto->meters) {
1115 meter_delete(ofproto, 1, ofproto->meter_features.max_meters);
1116 free(ofproto->meters);
1119 connmgr_destroy(ofproto->connmgr);
1121 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1122 free(ofproto->name);
1123 free(ofproto->type);
1124 free(ofproto->mfr_desc);
1125 free(ofproto->hw_desc);
1126 free(ofproto->sw_desc);
1127 free(ofproto->serial_desc);
1128 free(ofproto->dp_desc);
1129 hmap_destroy(&ofproto->ports);
1130 shash_destroy(&ofproto->port_by_name);
1131 bitmap_free(ofproto->ofp_port_ids);
1132 simap_destroy(&ofproto->ofp_requests);
1134 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1135 oftable_destroy(table);
1137 free(ofproto->tables);
1139 hmap_destroy(&ofproto->deletions);
1141 free(ofproto->vlan_bitmap);
1143 ovs_mutex_destroy(&ofproto->expirable_mutex);
1144 ofproto->ofproto_class->dealloc(ofproto);
1148 ofproto_destroy(struct ofproto *p)
1150 struct ofport *ofport, *next_ofport;
1157 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1158 ofport_destroy(ofport);
1161 p->ofproto_class->destruct(p);
1162 ofproto_destroy__(p);
1165 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1166 * kernel datapath, for example, this destroys the datapath in the kernel, and
1167 * with the netdev-based datapath, it tears down the data structures that
1168 * represent the datapath.
1170 * The datapath should not be currently open as an ofproto. */
1172 ofproto_delete(const char *name, const char *type)
1174 const struct ofproto_class *class = ofproto_class_find__(type);
1175 return (!class ? EAFNOSUPPORT
1176 : !class->del ? EACCES
1177 : class->del(type, name));
1181 process_port_change(struct ofproto *ofproto, int error, char *devname)
1183 if (error == ENOBUFS) {
1184 reinit_ports(ofproto);
1185 } else if (!error) {
1186 update_port(ofproto, devname);
1192 ofproto_type_run(const char *datapath_type)
1194 const struct ofproto_class *class;
1197 datapath_type = ofproto_normalize_type(datapath_type);
1198 class = ofproto_class_find__(datapath_type);
1200 error = class->type_run ? class->type_run(datapath_type) : 0;
1201 if (error && error != EAGAIN) {
1202 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1203 datapath_type, ovs_strerror(error));
1209 ofproto_type_run_fast(const char *datapath_type)
1211 const struct ofproto_class *class;
1214 datapath_type = ofproto_normalize_type(datapath_type);
1215 class = ofproto_class_find__(datapath_type);
1217 error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
1218 if (error && error != EAGAIN) {
1219 VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
1220 datapath_type, ovs_strerror(error));
1226 ofproto_type_wait(const char *datapath_type)
1228 const struct ofproto_class *class;
1230 datapath_type = ofproto_normalize_type(datapath_type);
1231 class = ofproto_class_find__(datapath_type);
1233 if (class->type_wait) {
1234 class->type_wait(datapath_type);
1239 ofproto_run(struct ofproto *p)
1241 struct sset changed_netdevs;
1242 const char *changed_netdev;
1243 struct ofport *ofport;
1246 error = p->ofproto_class->run(p);
1247 if (error && error != EAGAIN) {
1248 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, ovs_strerror(error));
1251 if (p->ofproto_class->port_poll) {
1254 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1255 process_port_change(p, error, devname);
1259 /* Update OpenFlow port status for any port whose netdev has changed.
1261 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1262 * destroyed, so it's not safe to update ports directly from the
1263 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1264 * need this two-phase approach. */
1265 sset_init(&changed_netdevs);
1266 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1267 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1268 if (ofport->change_seq != change_seq) {
1269 ofport->change_seq = change_seq;
1270 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1273 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1274 update_port(p, changed_netdev);
1276 sset_destroy(&changed_netdevs);
1280 connmgr_run(p->connmgr, handle_openflow);
1284 connmgr_run(p->connmgr, NULL);
1286 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1287 p->state = S_OPENFLOW;
1292 connmgr_run(p->connmgr, NULL);
1294 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1295 connmgr_flushed(p->connmgr);
1296 p->state = S_OPENFLOW;
1304 if (time_msec() >= p->next_op_report) {
1305 long long int ago = (time_msec() - p->first_op) / 1000;
1306 long long int interval = (p->last_op - p->first_op) / 1000;
1310 ds_put_format(&s, "%d flow_mods ",
1311 p->n_add + p->n_delete + p->n_modify);
1312 if (interval == ago) {
1313 ds_put_format(&s, "in the last %lld s", ago);
1314 } else if (interval) {
1315 ds_put_format(&s, "in the %lld s starting %lld s ago",
1318 ds_put_format(&s, "%lld s ago", ago);
1321 ds_put_cstr(&s, " (");
1323 ds_put_format(&s, "%d adds, ", p->n_add);
1326 ds_put_format(&s, "%d deletes, ", p->n_delete);
1329 ds_put_format(&s, "%d modifications, ", p->n_modify);
1332 ds_put_char(&s, ')');
1334 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1337 p->n_add = p->n_delete = p->n_modify = 0;
1338 p->next_op_report = LLONG_MAX;
1344 /* Performs periodic activity required by 'ofproto' that needs to be done
1345 * with the least possible latency.
1347 * It makes sense to call this function a couple of times per poll loop, to
1348 * provide a significant performance boost on some benchmarks with the
1349 * ofproto-dpif implementation. */
1351 ofproto_run_fast(struct ofproto *p)
1355 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1356 if (error && error != EAGAIN) {
1357 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1358 p->name, ovs_strerror(error));
1364 ofproto_wait(struct ofproto *p)
1366 struct ofport *ofport;
1368 p->ofproto_class->wait(p);
1369 if (p->ofproto_class->port_poll_wait) {
1370 p->ofproto_class->port_poll_wait(p);
1373 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1374 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1375 poll_immediate_wake();
1381 connmgr_wait(p->connmgr, true);
1386 connmgr_wait(p->connmgr, false);
1387 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1388 poll_immediate_wake();
1395 ofproto_is_alive(const struct ofproto *p)
1397 return connmgr_has_controllers(p->connmgr);
1400 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1401 * memory_report(). */
1403 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1405 const struct oftable *table;
1406 unsigned int n_rules;
1408 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1409 simap_increase(usage, "ops",
1410 ofproto->n_pending + hmap_count(&ofproto->deletions));
1413 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1414 ovs_rwlock_rdlock(&table->cls.rwlock);
1415 n_rules += classifier_count(&table->cls);
1416 ovs_rwlock_unlock(&table->cls.rwlock);
1418 simap_increase(usage, "rules", n_rules);
1420 if (ofproto->ofproto_class->get_memory_usage) {
1421 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1424 connmgr_get_memory_usage(ofproto->connmgr, usage);
1428 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1431 connmgr_get_controller_info(ofproto->connmgr, info);
1435 ofproto_free_ofproto_controller_info(struct shash *info)
1437 connmgr_free_controller_info(info);
1440 /* Makes a deep copy of 'old' into 'port'. */
1442 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1444 port->name = xstrdup(old->name);
1445 port->type = xstrdup(old->type);
1446 port->ofp_port = old->ofp_port;
1449 /* Frees memory allocated to members of 'ofproto_port'.
1451 * Do not call this function on an ofproto_port obtained from
1452 * ofproto_port_dump_next(): that function retains ownership of the data in the
1455 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1457 free(ofproto_port->name);
1458 free(ofproto_port->type);
1461 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1463 * This function provides no status indication. An error status for the entire
1464 * dump operation is provided when it is completed by calling
1465 * ofproto_port_dump_done().
1468 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1469 const struct ofproto *ofproto)
1471 dump->ofproto = ofproto;
1472 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1476 /* Attempts to retrieve another port from 'dump', which must have been created
1477 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1478 * 'port' and returns true. On failure, returns false.
1480 * Failure might indicate an actual error or merely that the last port has been
1481 * dumped. An error status for the entire dump operation is provided when it
1482 * is completed by calling ofproto_port_dump_done().
1484 * The ofproto owns the data stored in 'port'. It will remain valid until at
1485 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1486 * ofproto_port_dump_done(). */
1488 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1489 struct ofproto_port *port)
1491 const struct ofproto *ofproto = dump->ofproto;
1497 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1500 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1506 /* Completes port table dump operation 'dump', which must have been created
1507 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1508 * error-free, otherwise a positive errno value describing the problem. */
1510 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1512 const struct ofproto *ofproto = dump->ofproto;
1514 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1517 return dump->error == EOF ? 0 : dump->error;
1520 /* Returns the type to pass to netdev_open() when a datapath of type
1521 * 'datapath_type' has a port of type 'port_type', for a few special
1522 * cases when a netdev type differs from a port type. For example, when
1523 * using the userspace datapath, a port of type "internal" needs to be
1526 * Returns either 'type' itself or a string literal, which must not be
1529 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1531 const struct ofproto_class *class;
1533 datapath_type = ofproto_normalize_type(datapath_type);
1534 class = ofproto_class_find__(datapath_type);
1539 return (class->port_open_type
1540 ? class->port_open_type(datapath_type, port_type)
1544 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1545 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1546 * the port's OpenFlow port number.
1548 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1549 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1550 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1551 * 'ofp_portp' is non-null). */
1553 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1554 ofp_port_t *ofp_portp)
1556 ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1559 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1561 const char *netdev_name = netdev_get_name(netdev);
1563 simap_put(&ofproto->ofp_requests, netdev_name,
1564 ofp_to_u16(ofp_port));
1565 update_port(ofproto, netdev_name);
1568 struct ofproto_port ofproto_port;
1570 ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
1572 *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
1573 ofproto_port_destroy(&ofproto_port);
1578 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1579 * initializes '*port' appropriately; on failure, returns a positive errno
1582 * The caller owns the data in 'ofproto_port' and must free it with
1583 * ofproto_port_destroy() when it is no longer needed. */
1585 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1586 struct ofproto_port *port)
1590 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1592 memset(port, 0, sizeof *port);
1597 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1598 * Returns 0 if successful, otherwise a positive errno. */
1600 ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port)
1602 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1603 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1604 struct simap_node *ofp_request_node;
1607 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1608 if (ofp_request_node) {
1609 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1612 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1613 if (!error && ofport) {
1614 /* 'name' is the netdev's name and update_port() is going to close the
1615 * netdev. Just in case update_port() refers to 'name' after it
1616 * destroys 'ofport', make a copy of it around the update_port()
1618 char *devname = xstrdup(name);
1619 update_port(ofproto, devname);
1625 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1626 * performs the 'n_actions' actions in 'actions'. The new flow will not
1629 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1630 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1631 * controllers; otherwise, it will be hidden.
1633 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1635 * This is a helper function for in-band control and fail-open. */
1637 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1638 unsigned int priority,
1639 const struct ofpact *ofpacts, size_t ofpacts_len)
1641 const struct rule *rule;
1643 ovs_rwlock_rdlock(&ofproto->tables[0].cls.rwlock);
1644 rule = rule_from_cls_rule(classifier_find_match_exactly(
1645 &ofproto->tables[0].cls, match, priority));
1646 ovs_rwlock_unlock(&ofproto->tables[0].cls.rwlock);
1647 if (!rule || !ofpacts_equal(rule->ofpacts, rule->ofpacts_len,
1648 ofpacts, ofpacts_len)) {
1649 struct ofputil_flow_mod fm;
1651 memset(&fm, 0, sizeof fm);
1653 fm.priority = priority;
1654 fm.buffer_id = UINT32_MAX;
1655 fm.ofpacts = xmemdup(ofpacts, ofpacts_len);
1656 fm.ofpacts_len = ofpacts_len;
1657 add_flow(ofproto, NULL, &fm, NULL);
1662 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1663 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1664 * operation cannot be initiated now but may be retried later.
1666 * This is a helper function for in-band control and fail-open. */
1668 ofproto_flow_mod(struct ofproto *ofproto, struct ofputil_flow_mod *fm)
1670 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1673 /* Searches for a rule with matching criteria exactly equal to 'target' in
1674 * ofproto's table 0 and, if it finds one, deletes it.
1676 * This is a helper function for in-band control and fail-open. */
1678 ofproto_delete_flow(struct ofproto *ofproto,
1679 const struct match *target, unsigned int priority)
1683 ovs_rwlock_rdlock(&ofproto->tables[0].cls.rwlock);
1684 rule = rule_from_cls_rule(classifier_find_match_exactly(
1685 &ofproto->tables[0].cls, target, priority));
1686 ovs_rwlock_unlock(&ofproto->tables[0].cls.rwlock);
1688 /* No such rule -> success. */
1690 } else if (rule->pending) {
1691 /* An operation on the rule is already pending -> failure.
1692 * Caller must retry later if it's important. */
1695 /* Initiate deletion -> success. */
1696 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
1697 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1698 ovs_rwlock_wrlock(&rule->evict);
1699 oftable_remove_rule(rule);
1700 ofproto->ofproto_class->rule_destruct(rule);
1701 ofopgroup_submit(group);
1707 /* Starts the process of deleting all of the flows from all of ofproto's flow
1708 * tables and then reintroducing the flows required by in-band control and
1709 * fail-open. The process will complete in a later call to ofproto_run(). */
1711 ofproto_flush_flows(struct ofproto *ofproto)
1713 COVERAGE_INC(ofproto_flush);
1714 ofproto->state = S_FLUSH;
1718 reinit_ports(struct ofproto *p)
1720 struct ofproto_port_dump dump;
1721 struct sset devnames;
1722 struct ofport *ofport;
1723 struct ofproto_port ofproto_port;
1724 const char *devname;
1726 COVERAGE_INC(ofproto_reinit_ports);
1728 sset_init(&devnames);
1729 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1730 sset_add(&devnames, netdev_get_name(ofport->netdev));
1732 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1733 sset_add(&devnames, ofproto_port.name);
1736 SSET_FOR_EACH (devname, &devnames) {
1737 update_port(p, devname);
1739 sset_destroy(&devnames);
1743 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
1745 uint16_t max_ports = ofp_to_u16(ofproto->max_ports);
1748 port_idx = simap_get(&ofproto->ofp_requests, netdev_name);
1750 port_idx = UINT16_MAX;
1753 if (port_idx >= max_ports
1754 || bitmap_is_set(ofproto->ofp_port_ids, port_idx)) {
1755 uint16_t end_port_no = ofp_to_u16(ofproto->alloc_port_no);
1756 uint16_t alloc_port_no = end_port_no;
1758 /* Search for a free OpenFlow port number. We try not to
1759 * immediately reuse them to prevent problems due to old
1762 if (++alloc_port_no >= max_ports) {
1765 if (!bitmap_is_set(ofproto->ofp_port_ids, alloc_port_no)) {
1766 port_idx = alloc_port_no;
1767 ofproto->alloc_port_no = u16_to_ofp(alloc_port_no);
1770 if (alloc_port_no == end_port_no) {
1775 bitmap_set1(ofproto->ofp_port_ids, port_idx);
1776 return u16_to_ofp(port_idx);
1780 dealloc_ofp_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
1782 if (ofp_to_u16(ofp_port) < ofp_to_u16(ofproto->max_ports)) {
1783 bitmap_set0(ofproto->ofp_port_ids, ofp_to_u16(ofp_port));
1787 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1788 * pointer if the netdev cannot be opened. On success, also fills in
1790 static struct netdev *
1791 ofport_open(struct ofproto *ofproto,
1792 struct ofproto_port *ofproto_port,
1793 struct ofputil_phy_port *pp)
1795 enum netdev_flags flags;
1796 struct netdev *netdev;
1799 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1801 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
1802 "cannot be opened (%s)",
1804 ofproto_port->name, ofproto_port->ofp_port,
1805 ofproto_port->name, ovs_strerror(error));
1809 if (ofproto_port->ofp_port == OFPP_NONE) {
1810 if (!strcmp(ofproto->name, ofproto_port->name)) {
1811 ofproto_port->ofp_port = OFPP_LOCAL;
1813 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
1814 ofproto_port->name);
1817 pp->port_no = ofproto_port->ofp_port;
1818 netdev_get_etheraddr(netdev, pp->hw_addr);
1819 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
1820 netdev_get_flags(netdev, &flags);
1821 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
1822 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
1823 netdev_get_features(netdev, &pp->curr, &pp->advertised,
1824 &pp->supported, &pp->peer);
1825 pp->curr_speed = netdev_features_to_bps(pp->curr, 0);
1826 pp->max_speed = netdev_features_to_bps(pp->supported, 0);
1831 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1832 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1835 ofport_equal(const struct ofputil_phy_port *a,
1836 const struct ofputil_phy_port *b)
1838 return (eth_addr_equals(a->hw_addr, b->hw_addr)
1839 && a->state == b->state
1840 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
1841 && a->curr == b->curr
1842 && a->advertised == b->advertised
1843 && a->supported == b->supported
1844 && a->peer == b->peer
1845 && a->curr_speed == b->curr_speed
1846 && a->max_speed == b->max_speed);
1849 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1850 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1851 * one with the same name or port number). */
1853 ofport_install(struct ofproto *p,
1854 struct netdev *netdev, const struct ofputil_phy_port *pp)
1856 const char *netdev_name = netdev_get_name(netdev);
1857 struct ofport *ofport;
1860 /* Create ofport. */
1861 ofport = p->ofproto_class->port_alloc();
1866 ofport->ofproto = p;
1867 ofport->netdev = netdev;
1868 ofport->change_seq = netdev_change_seq(netdev);
1870 ofport->ofp_port = pp->port_no;
1871 ofport->created = time_msec();
1873 /* Add port to 'p'. */
1874 hmap_insert(&p->ports, &ofport->hmap_node,
1875 hash_ofp_port(ofport->ofp_port));
1876 shash_add(&p->port_by_name, netdev_name, ofport);
1878 update_mtu(p, ofport);
1880 /* Let the ofproto_class initialize its private data. */
1881 error = p->ofproto_class->port_construct(ofport);
1885 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
1889 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1890 p->name, netdev_name, ovs_strerror(error));
1892 ofport_destroy__(ofport);
1894 netdev_close(netdev);
1898 /* Removes 'ofport' from 'p' and destroys it. */
1900 ofport_remove(struct ofport *ofport)
1902 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
1904 ofport_destroy(ofport);
1907 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1910 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1912 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1914 ofport_remove(port);
1918 /* Updates 'port' with new 'pp' description.
1920 * Does not handle a name or port number change. The caller must implement
1921 * such a change as a delete followed by an add. */
1923 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
1925 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
1926 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
1927 | (pp->config & OFPUTIL_PC_PORT_DOWN));
1928 port->pp.state = pp->state;
1929 port->pp.curr = pp->curr;
1930 port->pp.advertised = pp->advertised;
1931 port->pp.supported = pp->supported;
1932 port->pp.peer = pp->peer;
1933 port->pp.curr_speed = pp->curr_speed;
1934 port->pp.max_speed = pp->max_speed;
1936 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
1939 /* Update OpenFlow 'state' in 'port' and notify controller. */
1941 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
1943 if (port->pp.state != state) {
1944 port->pp.state = state;
1945 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
1951 ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port)
1953 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1955 if (port->ofproto->ofproto_class->set_realdev) {
1956 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
1958 if (port->ofproto->ofproto_class->set_stp_port) {
1959 port->ofproto->ofproto_class->set_stp_port(port, NULL);
1961 if (port->ofproto->ofproto_class->set_cfm) {
1962 port->ofproto->ofproto_class->set_cfm(port, NULL);
1964 if (port->ofproto->ofproto_class->bundle_remove) {
1965 port->ofproto->ofproto_class->bundle_remove(port);
1971 ofport_destroy__(struct ofport *port)
1973 struct ofproto *ofproto = port->ofproto;
1974 const char *name = netdev_get_name(port->netdev);
1976 hmap_remove(&ofproto->ports, &port->hmap_node);
1977 shash_delete(&ofproto->port_by_name,
1978 shash_find(&ofproto->port_by_name, name));
1980 netdev_close(port->netdev);
1981 ofproto->ofproto_class->port_dealloc(port);
1985 ofport_destroy(struct ofport *port)
1988 dealloc_ofp_port(port->ofproto, port->ofp_port);
1989 port->ofproto->ofproto_class->port_destruct(port);
1990 ofport_destroy__(port);
1995 ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
1997 struct ofport *port;
1999 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_ofp_port(ofp_port),
2001 if (port->ofp_port == ofp_port) {
2009 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
2011 struct ofproto *ofproto = port->ofproto;
2014 if (ofproto->ofproto_class->port_get_stats) {
2015 error = ofproto->ofproto_class->port_get_stats(port, stats);
2024 update_port(struct ofproto *ofproto, const char *name)
2026 struct ofproto_port ofproto_port;
2027 struct ofputil_phy_port pp;
2028 struct netdev *netdev;
2029 struct ofport *port;
2031 COVERAGE_INC(ofproto_update_port);
2033 /* Fetch 'name''s location and properties from the datapath. */
2034 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
2035 ? ofport_open(ofproto, &ofproto_port, &pp)
2039 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
2040 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
2041 struct netdev *old_netdev = port->netdev;
2043 /* 'name' hasn't changed location. Any properties changed? */
2044 if (!ofport_equal(&port->pp, &pp)) {
2045 ofport_modified(port, &pp);
2048 update_mtu(ofproto, port);
2050 /* Install the newly opened netdev in case it has changed.
2051 * Don't close the old netdev yet in case port_modified has to
2052 * remove a retained reference to it.*/
2053 port->netdev = netdev;
2054 port->change_seq = netdev_change_seq(netdev);
2056 if (port->ofproto->ofproto_class->port_modified) {
2057 port->ofproto->ofproto_class->port_modified(port);
2060 netdev_close(old_netdev);
2062 /* If 'port' is nonnull then its name differs from 'name' and thus
2063 * we should delete it. If we think there's a port named 'name'
2064 * then its port number must be wrong now so delete it too. */
2066 ofport_remove(port);
2068 ofport_remove_with_name(ofproto, name);
2069 ofport_install(ofproto, netdev, &pp);
2072 /* Any port named 'name' is gone now. */
2073 ofport_remove_with_name(ofproto, name);
2075 ofproto_port_destroy(&ofproto_port);
2079 init_ports(struct ofproto *p)
2081 struct ofproto_port_dump dump;
2082 struct ofproto_port ofproto_port;
2083 struct shash_node *node, *next;
2085 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2086 const char *name = ofproto_port.name;
2088 if (shash_find(&p->port_by_name, name)) {
2089 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2092 struct ofputil_phy_port pp;
2093 struct netdev *netdev;
2095 /* Check if an OpenFlow port number had been requested. */
2096 node = shash_find(&init_ofp_ports, name);
2098 const struct iface_hint *iface_hint = node->data;
2099 simap_put(&p->ofp_requests, name,
2100 ofp_to_u16(iface_hint->ofp_port));
2103 netdev = ofport_open(p, &ofproto_port, &pp);
2105 ofport_install(p, netdev, &pp);
2110 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2111 struct iface_hint *iface_hint = node->data;
2113 if (!strcmp(iface_hint->br_name, p->name)) {
2114 free(iface_hint->br_name);
2115 free(iface_hint->br_type);
2117 shash_delete(&init_ofp_ports, node);
2124 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2125 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2127 find_min_mtu(struct ofproto *p)
2129 struct ofport *ofport;
2132 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2133 struct netdev *netdev = ofport->netdev;
2136 /* Skip any internal ports, since that's what we're trying to
2138 if (!strcmp(netdev_get_type(netdev), "internal")) {
2142 if (netdev_get_mtu(netdev, &dev_mtu)) {
2145 if (!mtu || dev_mtu < mtu) {
2150 return mtu ? mtu: ETH_PAYLOAD_MAX;
2153 /* Update MTU of all datapath devices on 'p' to the minimum of the
2154 * non-datapath ports in event of 'port' added or changed. */
2156 update_mtu(struct ofproto *p, struct ofport *port)
2158 struct ofport *ofport;
2159 struct netdev *netdev = port->netdev;
2160 int dev_mtu, old_min;
2162 if (netdev_get_mtu(netdev, &dev_mtu)) {
2166 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2167 if (dev_mtu > p->min_mtu) {
2168 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2169 dev_mtu = p->min_mtu;
2172 port->mtu = dev_mtu;
2176 /* For non-internal port find new min mtu. */
2177 old_min = p->min_mtu;
2178 port->mtu = dev_mtu;
2179 p->min_mtu = find_min_mtu(p);
2180 if (p->min_mtu == old_min) {
2184 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2185 struct netdev *netdev = ofport->netdev;
2187 if (!strcmp(netdev_get_type(netdev), "internal")) {
2188 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2189 ofport->mtu = p->min_mtu;
2196 ofproto_rule_destroy__(struct rule *rule)
2199 cls_rule_destroy(&rule->cr);
2200 free(rule->ofpacts);
2201 ovs_mutex_destroy(&rule->timeout_mutex);
2202 ovs_rwlock_destroy(&rule->evict);
2203 rule->ofproto->ofproto_class->rule_dealloc(rule);
2207 /* This function allows an ofproto implementation to destroy any rules that
2208 * remain when its ->destruct() function is called. The caller must have
2209 * already uninitialized any derived members of 'rule' (step 5 described in the
2210 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
2211 * This function implements steps 6 and 7.
2213 * This function should only be called from an ofproto implementation's
2214 * ->destruct() function. It is not suitable elsewhere. */
2216 ofproto_rule_destroy(struct ofproto *ofproto, struct classifier *cls,
2217 struct rule *rule) OVS_REQ_WRLOCK(cls->rwlock)
2219 ovs_assert(!rule->pending);
2220 if (!ovs_rwlock_trywrlock(&rule->evict)) {
2221 oftable_remove_rule__(ofproto, cls, rule);
2225 ofproto_rule_destroy__(rule);
2228 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2229 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2231 ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port)
2233 return (port == OFPP_ANY
2234 || ofpacts_output_to_port(rule->ofpacts, rule->ofpacts_len, port));
2237 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2238 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2240 ofoperation_has_out_port(const struct ofoperation *op, ofp_port_t out_port)
2242 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2247 case OFOPERATION_ADD:
2248 return op->victim && ofproto_rule_has_out_port(op->victim, out_port);
2250 case OFOPERATION_DELETE:
2253 case OFOPERATION_MODIFY:
2254 return ofpacts_output_to_port(op->ofpacts, op->ofpacts_len, out_port);
2260 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2261 * statistics appropriately. 'packet' must have at least sizeof(struct
2262 * ofp10_packet_in) bytes of headroom.
2264 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2265 * with statistics for 'packet' either way.
2267 * Takes ownership of 'packet'. */
2269 rule_execute(struct rule *rule, ofp_port_t in_port, struct ofpbuf *packet)
2272 union flow_in_port in_port_;
2274 ovs_assert(ofpbuf_headroom(packet) >= sizeof(struct ofp10_packet_in));
2276 in_port_.ofp_port = in_port;
2277 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
2278 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
2281 /* Returns true if 'rule' should be hidden from the controller.
2283 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2284 * (e.g. by in-band control) and are intentionally hidden from the
2287 ofproto_rule_is_hidden(const struct rule *rule)
2289 return rule->cr.priority > UINT16_MAX;
2292 static enum oftable_flags
2293 rule_get_flags(const struct rule *rule)
2295 return rule->ofproto->tables[rule->table_id].flags;
2299 rule_is_modifiable(const struct rule *rule)
2301 return !(rule_get_flags(rule) & OFTABLE_READONLY);
2305 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2307 ofconn_send_reply(ofconn, make_echo_reply(oh));
2312 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2314 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2315 struct ofputil_switch_features features;
2316 struct ofport *port;
2322 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2324 ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2326 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2327 * if present, are always at the end.) */
2328 n_tables = ofproto->n_tables;
2329 for (i = 0; i < ofproto->n_tables; i++) {
2330 if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
2336 features.datapath_id = ofproto->datapath_id;
2337 features.n_buffers = pktbuf_capacity();
2338 features.n_tables = n_tables;
2339 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2340 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2342 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2344 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2345 features.auxiliary_id = 0;
2346 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2348 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2349 ofputil_put_switch_features_port(&port->pp, b);
2352 ofconn_send_reply(ofconn, b);
2357 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2359 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2360 struct ofp_switch_config *osc;
2361 enum ofp_config_flags flags;
2365 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2366 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2367 flags = ofproto->frag_handling;
2368 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2369 if (oh->version < OFP13_VERSION
2370 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2371 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2373 osc->flags = htons(flags);
2374 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2375 ofconn_send_reply(ofconn, buf);
2381 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2383 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2384 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2385 uint16_t flags = ntohs(osc->flags);
2387 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2388 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
2389 enum ofp_config_flags cur = ofproto->frag_handling;
2390 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2392 ovs_assert((cur & OFPC_FRAG_MASK) == cur);
2394 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2395 ofproto->frag_handling = next;
2397 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2399 ofputil_frag_handling_to_string(next));
2403 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2404 ofconn_set_invalid_ttl_to_controller(ofconn,
2405 (oh->version < OFP13_VERSION
2406 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2408 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2413 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2414 * error message code for the caller to propagate upward. Otherwise, returns
2417 * The log message mentions 'msg_type'. */
2419 reject_slave_controller(struct ofconn *ofconn)
2421 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2422 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
2423 return OFPERR_OFPBRC_EPERM;
2429 /* Finds the OFPACT_METER action, if any, in the 'ofpacts_len' bytes of
2430 * 'ofpacts'. If found, returns its meter ID; if not, returns 0.
2432 * This function relies on the order of 'ofpacts' being correct (as checked by
2433 * ofpacts_verify()). */
2435 find_meter(const struct ofpact ofpacts[], size_t ofpacts_len)
2437 const struct ofpact *a;
2439 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2440 enum ovs_instruction_type inst;
2442 inst = ovs_instruction_type_from_ofpact_type(a->type);
2443 if (a->type == OFPACT_METER) {
2444 return ofpact_get_METER(a)->meter_id;
2445 } else if (inst > OVSINST_OFPIT13_METER) {
2453 /* Checks that the 'ofpacts_len' bytes of actions in 'ofpacts' are appropriate
2454 * for a packet with the prerequisites satisfied by 'flow' in table 'table_id'.
2455 * 'flow' may be temporarily modified, but is restored at return.
2458 ofproto_check_ofpacts(struct ofproto *ofproto,
2459 const struct ofpact ofpacts[], size_t ofpacts_len,
2460 struct flow *flow, uint8_t table_id)
2465 error = ofpacts_check(ofpacts, ofpacts_len, flow, ofproto->max_ports,
2471 mid = find_meter(ofpacts, ofpacts_len);
2472 if (mid && ofproto_get_provider_meter_id(ofproto, mid) == UINT32_MAX) {
2473 return OFPERR_OFPMMFC_INVALID_METER;
2479 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2481 struct ofproto *p = ofconn_get_ofproto(ofconn);
2482 struct ofputil_packet_out po;
2483 struct ofpbuf *payload;
2484 uint64_t ofpacts_stub[1024 / 8];
2485 struct ofpbuf ofpacts;
2487 union flow_in_port in_port_;
2490 COVERAGE_INC(ofproto_packet_out);
2492 error = reject_slave_controller(ofconn);
2497 /* Decode message. */
2498 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2499 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2501 goto exit_free_ofpacts;
2503 if (ofp_to_u16(po.in_port) >= ofp_to_u16(p->max_ports)
2504 && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) {
2505 error = OFPERR_OFPBRC_BAD_PORT;
2506 goto exit_free_ofpacts;
2511 if (po.buffer_id != UINT32_MAX) {
2512 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2513 if (error || !payload) {
2514 goto exit_free_ofpacts;
2517 payload = xmalloc(sizeof *payload);
2518 ofpbuf_use_const(payload, po.packet, po.packet_len);
2521 /* Verify actions against packet, then send packet if successful. */
2522 in_port_.ofp_port = po.in_port;
2523 flow_extract(payload, 0, 0, NULL, &in_port_, &flow);
2524 error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len, &flow, 0);
2526 error = p->ofproto_class->packet_out(p, payload, &flow,
2527 po.ofpacts, po.ofpacts_len);
2529 ofpbuf_delete(payload);
2532 ofpbuf_uninit(&ofpacts);
2538 update_port_config(struct ofport *port,
2539 enum ofputil_port_config config,
2540 enum ofputil_port_config mask)
2542 enum ofputil_port_config old_config = port->pp.config;
2543 enum ofputil_port_config toggle;
2545 toggle = (config ^ port->pp.config) & mask;
2546 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2547 if (config & OFPUTIL_PC_PORT_DOWN) {
2548 netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL);
2550 netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL);
2552 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2555 port->pp.config ^= toggle;
2556 if (port->pp.config != old_config) {
2557 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2562 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2564 struct ofproto *p = ofconn_get_ofproto(ofconn);
2565 struct ofputil_port_mod pm;
2566 struct ofport *port;
2569 error = reject_slave_controller(ofconn);
2574 error = ofputil_decode_port_mod(oh, &pm);
2579 port = ofproto_get_port(p, pm.port_no);
2581 return OFPERR_OFPPMFC_BAD_PORT;
2582 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2583 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2585 update_port_config(port, pm.config, pm.mask);
2587 netdev_set_advertisements(port->netdev, pm.advertise);
2594 handle_desc_stats_request(struct ofconn *ofconn,
2595 const struct ofp_header *request)
2597 static const char *default_mfr_desc = "Nicira, Inc.";
2598 static const char *default_hw_desc = "Open vSwitch";
2599 static const char *default_sw_desc = VERSION;
2600 static const char *default_serial_desc = "None";
2601 static const char *default_dp_desc = "None";
2603 struct ofproto *p = ofconn_get_ofproto(ofconn);
2604 struct ofp_desc_stats *ods;
2607 msg = ofpraw_alloc_stats_reply(request, 0);
2608 ods = ofpbuf_put_zeros(msg, sizeof *ods);
2609 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
2610 sizeof ods->mfr_desc);
2611 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
2612 sizeof ods->hw_desc);
2613 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
2614 sizeof ods->sw_desc);
2615 ovs_strlcpy(ods->serial_num,
2616 p->serial_desc ? p->serial_desc : default_serial_desc,
2617 sizeof ods->serial_num);
2618 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
2619 sizeof ods->dp_desc);
2620 ofconn_send_reply(ofconn, msg);
2626 handle_table_stats_request(struct ofconn *ofconn,
2627 const struct ofp_header *request)
2629 struct ofproto *p = ofconn_get_ofproto(ofconn);
2630 struct ofp12_table_stats *ots;
2635 /* Set up default values.
2637 * ofp12_table_stats is used as a generic structure as
2638 * it is able to hold all the fields for ofp10_table_stats
2639 * and ofp11_table_stats (and of course itself).
2641 ots = xcalloc(p->n_tables, sizeof *ots);
2642 for (i = 0; i < p->n_tables; i++) {
2643 ots[i].table_id = i;
2644 sprintf(ots[i].name, "table%zu", i);
2645 ots[i].match = htonll(OFPXMT12_MASK);
2646 ots[i].wildcards = htonll(OFPXMT12_MASK);
2647 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
2648 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
2649 ots[i].write_setfields = htonll(OFPXMT12_MASK);
2650 ots[i].apply_setfields = htonll(OFPXMT12_MASK);
2651 ots[i].metadata_match = htonll(UINT64_MAX);
2652 ots[i].metadata_write = htonll(UINT64_MAX);
2653 ots[i].instructions = htonl(OFPIT11_ALL);
2654 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
2655 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2656 ovs_rwlock_rdlock(&p->tables[i].cls.rwlock);
2657 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2658 ovs_rwlock_unlock(&p->tables[i].cls.rwlock);
2661 p->ofproto_class->get_tables(p, ots);
2663 /* Post-process the tables, dropping hidden tables. */
2664 n_tables = p->n_tables;
2665 for (i = 0; i < p->n_tables; i++) {
2666 const struct oftable *table = &p->tables[i];
2668 if (table->flags & OFTABLE_HIDDEN) {
2674 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2677 if (table->max_flows < ntohl(ots[i].max_entries)) {
2678 ots[i].max_entries = htonl(table->max_flows);
2682 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
2683 ofconn_send_reply(ofconn, msg);
2691 append_port_stat(struct ofport *port, struct list *replies)
2693 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
2695 calc_duration(port->created, time_msec(),
2696 &ops.duration_sec, &ops.duration_nsec);
2698 /* Intentionally ignore return value, since errors will set
2699 * 'stats' to all-1s, which is correct for OpenFlow, and
2700 * netdev_get_stats() will log errors. */
2701 ofproto_port_get_stats(port, &ops.stats);
2703 ofputil_append_port_stat(replies, &ops);
2707 handle_port_stats_request(struct ofconn *ofconn,
2708 const struct ofp_header *request)
2710 struct ofproto *p = ofconn_get_ofproto(ofconn);
2711 struct ofport *port;
2712 struct list replies;
2716 error = ofputil_decode_port_stats_request(request, &port_no);
2721 ofpmp_init(&replies, request);
2722 if (port_no != OFPP_ANY) {
2723 port = ofproto_get_port(p, port_no);
2725 append_port_stat(port, &replies);
2728 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2729 append_port_stat(port, &replies);
2733 ofconn_send_replies(ofconn, &replies);
2738 handle_port_desc_stats_request(struct ofconn *ofconn,
2739 const struct ofp_header *request)
2741 struct ofproto *p = ofconn_get_ofproto(ofconn);
2742 enum ofp_version version;
2743 struct ofport *port;
2744 struct list replies;
2746 ofpmp_init(&replies, request);
2748 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
2749 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2750 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
2753 ofconn_send_replies(ofconn, &replies);
2758 hash_cookie(ovs_be64 cookie)
2760 return hash_2words((OVS_FORCE uint64_t)cookie >> 32,
2761 (OVS_FORCE uint64_t)cookie);
2765 cookies_insert(struct ofproto *ofproto, struct rule *rule)
2767 hindex_insert(&ofproto->cookies, &rule->cookie_node,
2768 hash_cookie(rule->flow_cookie));
2772 cookies_remove(struct ofproto *ofproto, struct rule *rule)
2774 hindex_remove(&ofproto->cookies, &rule->cookie_node);
2778 ofproto_rule_change_cookie(struct ofproto *ofproto, struct rule *rule,
2779 ovs_be64 new_cookie)
2781 if (new_cookie != rule->flow_cookie) {
2782 cookies_remove(ofproto, rule);
2784 rule->flow_cookie = new_cookie;
2786 cookies_insert(ofproto, rule);
2791 calc_duration(long long int start, long long int now,
2792 uint32_t *sec, uint32_t *nsec)
2794 long long int msecs = now - start;
2795 *sec = msecs / 1000;
2796 *nsec = (msecs % 1000) * (1000 * 1000);
2799 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2800 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2802 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
2804 return (table_id == 0xff || table_id < ofproto->n_tables
2806 : OFPERR_OFPBRC_BAD_TABLE_ID);
2810 static struct oftable *
2811 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
2813 struct oftable *table;
2815 for (table = &ofproto->tables[table_id];
2816 table < &ofproto->tables[ofproto->n_tables];
2818 if (!(table->flags & OFTABLE_HIDDEN)) {
2826 static struct oftable *
2827 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
2829 if (table_id == 0xff) {
2830 return next_visible_table(ofproto, 0);
2831 } else if (table_id < ofproto->n_tables) {
2832 return &ofproto->tables[table_id];
2838 static struct oftable *
2839 next_matching_table(const struct ofproto *ofproto,
2840 const struct oftable *table, uint8_t table_id)
2842 return (table_id == 0xff
2843 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
2847 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2849 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2850 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2852 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2853 * only once, for that table. (This can be used to access tables marked
2856 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2857 * entered at all. (Perhaps you should have validated TABLE_ID with
2858 * check_table_id().)
2860 * All parameters are evaluated multiple times.
2862 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2863 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2865 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2867 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2868 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2869 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2872 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2873 * to 'out_port' are included.
2875 * Hidden rules are always omitted.
2877 * Returns 0 on success, otherwise an OpenFlow error code. */
2879 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
2880 const struct match *match,
2881 ovs_be64 cookie, ovs_be64 cookie_mask,
2882 ofp_port_t out_port, struct list *rules)
2884 struct oftable *table;
2888 error = check_table_id(ofproto, table_id);
2894 cls_rule_init(&cr, match, 0);
2896 if (cookie_mask == htonll(UINT64_MAX)) {
2899 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node, hash_cookie(cookie),
2900 &ofproto->cookies) {
2901 if (table_id != rule->table_id && table_id != 0xff) {
2904 if (ofproto_rule_is_hidden(rule)) {
2907 if (cls_rule_is_loose_match(&rule->cr, &cr.match)) {
2908 if (rule->pending) {
2909 error = OFPROTO_POSTPONE;
2912 if (rule->flow_cookie == cookie /* Hash collisions possible. */
2913 && ofproto_rule_has_out_port(rule, out_port)) {
2914 list_push_back(rules, &rule->ofproto_node);
2921 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2922 struct cls_cursor cursor;
2925 ovs_rwlock_rdlock(&table->cls.rwlock);
2926 cls_cursor_init(&cursor, &table->cls, &cr);
2927 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2928 if (rule->pending) {
2929 ovs_rwlock_unlock(&table->cls.rwlock);
2930 error = OFPROTO_POSTPONE;
2933 if (!ofproto_rule_is_hidden(rule)
2934 && ofproto_rule_has_out_port(rule, out_port)
2935 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2936 list_push_back(rules, &rule->ofproto_node);
2939 ovs_rwlock_unlock(&table->cls.rwlock);
2943 cls_rule_destroy(&cr);
2947 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2948 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2949 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2952 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2953 * to 'out_port' are included.
2955 * Hidden rules are always omitted.
2957 * Returns 0 on success, otherwise an OpenFlow error code. */
2959 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
2960 const struct match *match, unsigned int priority,
2961 ovs_be64 cookie, ovs_be64 cookie_mask,
2962 ofp_port_t out_port, struct list *rules)
2964 struct oftable *table;
2968 error = check_table_id(ofproto, table_id);
2974 cls_rule_init(&cr, match, priority);
2976 if (cookie_mask == htonll(UINT64_MAX)) {
2979 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node, hash_cookie(cookie),
2980 &ofproto->cookies) {
2981 if (table_id != rule->table_id && table_id != 0xff) {
2984 if (ofproto_rule_is_hidden(rule)) {
2987 if (cls_rule_equal(&rule->cr, &cr)) {
2988 if (rule->pending) {
2989 error = OFPROTO_POSTPONE;
2992 if (rule->flow_cookie == cookie /* Hash collisions possible. */
2993 && ofproto_rule_has_out_port(rule, out_port)) {
2994 list_push_back(rules, &rule->ofproto_node);
3001 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
3004 ovs_rwlock_rdlock(&table->cls.rwlock);
3005 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls,
3007 ovs_rwlock_unlock(&table->cls.rwlock);
3009 if (rule->pending) {
3010 error = OFPROTO_POSTPONE;
3013 if (!ofproto_rule_is_hidden(rule)
3014 && ofproto_rule_has_out_port(rule, out_port)
3015 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
3016 list_push_back(rules, &rule->ofproto_node);
3022 cls_rule_destroy(&cr);
3026 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
3027 * forced into the range of a uint16_t. */
3029 age_secs(long long int age_ms)
3031 return (age_ms < 0 ? 0
3032 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
3033 : (unsigned int) age_ms / 1000);
3037 handle_flow_stats_request(struct ofconn *ofconn,
3038 const struct ofp_header *request)
3040 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3041 struct ofputil_flow_stats_request fsr;
3042 struct list replies;
3047 error = ofputil_decode_flow_stats_request(&fsr, request);
3052 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
3053 fsr.cookie, fsr.cookie_mask,
3054 fsr.out_port, &rules);
3059 ofpmp_init(&replies, request);
3060 LIST_FOR_EACH (rule, ofproto_node, &rules) {
3061 long long int now = time_msec();
3062 struct ofputil_flow_stats fs;
3064 minimatch_expand(&rule->cr.match, &fs.match);
3065 fs.priority = rule->cr.priority;
3066 fs.cookie = rule->flow_cookie;
3067 fs.table_id = rule->table_id;
3068 calc_duration(rule->created, now, &fs.duration_sec, &fs.duration_nsec);
3069 fs.idle_age = age_secs(now - rule->used);
3070 fs.hard_age = age_secs(now - rule->modified);
3071 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
3073 fs.ofpacts = rule->ofpacts;
3074 fs.ofpacts_len = rule->ofpacts_len;
3077 ovs_mutex_lock(&rule->timeout_mutex);
3078 fs.idle_timeout = rule->idle_timeout;
3079 fs.hard_timeout = rule->hard_timeout;
3080 ovs_mutex_unlock(&rule->timeout_mutex);
3082 if (rule->send_flow_removed) {
3083 fs.flags |= OFPFF_SEND_FLOW_REM;
3084 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
3085 and OFPFF13_NO_BYT_COUNTS */
3087 ofputil_append_flow_stats_reply(&fs, &replies);
3089 ofconn_send_replies(ofconn, &replies);
3095 flow_stats_ds(struct rule *rule, struct ds *results)
3097 uint64_t packet_count, byte_count;
3099 rule->ofproto->ofproto_class->rule_get_stats(rule,
3100 &packet_count, &byte_count);
3102 if (rule->table_id != 0) {
3103 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
3105 ds_put_format(results, "duration=%llds, ",
3106 (time_msec() - rule->created) / 1000);
3107 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3108 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3109 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3110 cls_rule_format(&rule->cr, results);
3111 ds_put_char(results, ',');
3112 ofpacts_format(rule->ofpacts, rule->ofpacts_len, results);
3113 ds_put_cstr(results, "\n");
3116 /* Adds a pretty-printed description of all flows to 'results', including
3117 * hidden flows (e.g., set up by in-band control). */
3119 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3121 struct oftable *table;
3123 OFPROTO_FOR_EACH_TABLE (table, p) {
3124 struct cls_cursor cursor;
3127 ovs_rwlock_rdlock(&table->cls.rwlock);
3128 cls_cursor_init(&cursor, &table->cls, NULL);
3129 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3130 flow_stats_ds(rule, results);
3132 ovs_rwlock_unlock(&table->cls.rwlock);
3136 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
3137 * '*engine_type' and '*engine_id', respectively. */
3139 ofproto_get_netflow_ids(const struct ofproto *ofproto,
3140 uint8_t *engine_type, uint8_t *engine_id)
3142 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
3145 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
3146 * true if the port's CFM status was successfully stored into '*status'.
3147 * Returns false if the port did not have CFM configured, in which case
3148 * '*status' is indeterminate.
3150 * The caller must provide and owns '*status', and must free 'status->rmps'. */
3152 ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port,
3153 struct ofproto_cfm_status *status)
3155 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
3157 && ofproto->ofproto_class->get_cfm_status
3158 && ofproto->ofproto_class->get_cfm_status(ofport, status));
3162 handle_aggregate_stats_request(struct ofconn *ofconn,
3163 const struct ofp_header *oh)
3165 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3166 struct ofputil_flow_stats_request request;
3167 struct ofputil_aggregate_stats stats;
3168 bool unknown_packets, unknown_bytes;
3169 struct ofpbuf *reply;
3174 error = ofputil_decode_flow_stats_request(&request, oh);
3179 error = collect_rules_loose(ofproto, request.table_id, &request.match,
3180 request.cookie, request.cookie_mask,
3181 request.out_port, &rules);
3186 memset(&stats, 0, sizeof stats);
3187 unknown_packets = unknown_bytes = false;
3188 LIST_FOR_EACH (rule, ofproto_node, &rules) {
3189 uint64_t packet_count;
3190 uint64_t byte_count;
3192 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
3195 if (packet_count == UINT64_MAX) {
3196 unknown_packets = true;
3198 stats.packet_count += packet_count;
3201 if (byte_count == UINT64_MAX) {
3202 unknown_bytes = true;
3204 stats.byte_count += byte_count;
3209 if (unknown_packets) {
3210 stats.packet_count = UINT64_MAX;
3212 if (unknown_bytes) {
3213 stats.byte_count = UINT64_MAX;
3216 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
3217 ofconn_send_reply(ofconn, reply);
3222 struct queue_stats_cbdata {
3223 struct ofport *ofport;
3224 struct list replies;
3229 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3230 const struct netdev_queue_stats *stats)
3232 struct ofputil_queue_stats oqs;
3234 oqs.port_no = cbdata->ofport->pp.port_no;
3235 oqs.queue_id = queue_id;
3236 oqs.tx_bytes = stats->tx_bytes;
3237 oqs.tx_packets = stats->tx_packets;
3238 oqs.tx_errors = stats->tx_errors;
3239 if (stats->created != LLONG_MIN) {
3240 calc_duration(stats->created, cbdata->now,
3241 &oqs.duration_sec, &oqs.duration_nsec);
3243 oqs.duration_sec = oqs.duration_nsec = UINT32_MAX;
3245 ofputil_append_queue_stat(&cbdata->replies, &oqs);
3249 handle_queue_stats_dump_cb(uint32_t queue_id,
3250 struct netdev_queue_stats *stats,
3253 struct queue_stats_cbdata *cbdata = cbdata_;
3255 put_queue_stats(cbdata, queue_id, stats);
3259 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3260 struct queue_stats_cbdata *cbdata)
3262 cbdata->ofport = port;
3263 if (queue_id == OFPQ_ALL) {
3264 netdev_dump_queue_stats(port->netdev,
3265 handle_queue_stats_dump_cb, cbdata);
3267 struct netdev_queue_stats stats;
3269 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3270 put_queue_stats(cbdata, queue_id, &stats);
3272 return OFPERR_OFPQOFC_BAD_QUEUE;
3279 handle_queue_stats_request(struct ofconn *ofconn,
3280 const struct ofp_header *rq)
3282 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3283 struct queue_stats_cbdata cbdata;
3284 struct ofport *port;
3286 struct ofputil_queue_stats_request oqsr;
3288 COVERAGE_INC(ofproto_queue_req);
3290 ofpmp_init(&cbdata.replies, rq);
3291 cbdata.now = time_msec();
3293 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3298 if (oqsr.port_no == OFPP_ANY) {
3299 error = OFPERR_OFPQOFC_BAD_QUEUE;
3300 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3301 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3306 port = ofproto_get_port(ofproto, oqsr.port_no);
3308 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3309 : OFPERR_OFPQOFC_BAD_PORT);
3312 ofconn_send_replies(ofconn, &cbdata.replies);
3314 ofpbuf_list_delete(&cbdata.replies);
3321 is_flow_deletion_pending(const struct ofproto *ofproto,
3322 const struct cls_rule *cls_rule,
3325 if (!hmap_is_empty(&ofproto->deletions)) {
3326 struct ofoperation *op;
3328 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3329 cls_rule_hash(cls_rule, table_id),
3330 &ofproto->deletions) {
3331 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
3340 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3341 * in which no matching flow already exists in the flow table.
3343 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3344 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3345 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3346 * initiated now but may be retried later.
3348 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
3349 * ownership remains with the caller.
3351 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3354 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
3355 struct ofputil_flow_mod *fm, const struct ofp_header *request)
3357 struct oftable *table;
3358 struct ofopgroup *group;
3359 struct rule *victim;
3364 error = check_table_id(ofproto, fm->table_id);
3370 if (fm->table_id == 0xff) {
3371 if (ofproto->ofproto_class->rule_choose_table) {
3372 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3378 ovs_assert(table_id < ofproto->n_tables);
3382 } else if (fm->table_id < ofproto->n_tables) {
3383 table_id = fm->table_id;
3385 return OFPERR_OFPBRC_BAD_TABLE_ID;
3388 table = &ofproto->tables[table_id];
3390 if (table->flags & OFTABLE_READONLY) {
3391 return OFPERR_OFPBRC_EPERM;
3394 /* Verify actions. */
3395 error = ofproto_check_ofpacts(ofproto, fm->ofpacts, fm->ofpacts_len,
3396 &fm->match.flow, table_id);
3401 /* Allocate new rule and initialize classifier rule. */
3402 rule = ofproto->ofproto_class->rule_alloc();
3404 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
3405 ofproto->name, ovs_strerror(error));
3408 cls_rule_init(&rule->cr, &fm->match, fm->priority);
3410 /* Serialize against pending deletion. */
3411 if (is_flow_deletion_pending(ofproto, &rule->cr, table_id)) {
3412 cls_rule_destroy(&rule->cr);
3413 ofproto->ofproto_class->rule_dealloc(rule);
3414 return OFPROTO_POSTPONE;
3417 /* Check for overlap, if requested. */
3418 if (fm->flags & OFPFF_CHECK_OVERLAP) {
3421 ovs_rwlock_rdlock(&table->cls.rwlock);
3422 overlaps = classifier_rule_overlaps(&table->cls, &rule->cr);
3423 ovs_rwlock_unlock(&table->cls.rwlock);
3426 cls_rule_destroy(&rule->cr);
3427 ofproto->ofproto_class->rule_dealloc(rule);
3428 return OFPERR_OFPFMFC_OVERLAP;
3432 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3434 rule->ofproto = ofproto;
3435 rule->pending = NULL;
3436 rule->flow_cookie = fm->new_cookie;
3437 rule->created = rule->modified = rule->used = time_msec();
3439 ovs_mutex_init(&rule->timeout_mutex, OVS_MUTEX_ADAPTIVE);
3440 ovs_mutex_lock(&rule->timeout_mutex);
3441 rule->idle_timeout = fm->idle_timeout;
3442 rule->hard_timeout = fm->hard_timeout;
3443 ovs_mutex_unlock(&rule->timeout_mutex);
3445 rule->table_id = table - ofproto->tables;
3446 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
3447 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
3448 and OFPFF13_NO_BYT_COUNTS */
3449 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3450 rule->ofpacts_len = fm->ofpacts_len;
3451 rule->meter_id = find_meter(rule->ofpacts, rule->ofpacts_len);
3452 list_init(&rule->meter_list_node);
3453 rule->eviction_group = NULL;
3454 list_init(&rule->expirable);
3455 rule->monitor_flags = 0;
3456 rule->add_seqno = 0;
3457 rule->modify_seqno = 0;
3458 ovs_rwlock_init(&rule->evict);
3460 /* Insert new rule. */
3461 victim = oftable_replace_rule(rule);
3462 if (victim && !rule_is_modifiable(victim)) {
3463 error = OFPERR_OFPBRC_EPERM;
3464 } else if (victim && victim->pending) {
3465 error = OFPROTO_POSTPONE;
3467 struct ofoperation *op;
3471 ovs_rwlock_rdlock(&table->cls.rwlock);
3472 n_rules = classifier_count(&table->cls);
3473 ovs_rwlock_unlock(&table->cls.rwlock);
3474 if (n_rules > table->max_flows) {
3475 ovs_rwlock_rdlock(&rule->evict);
3476 if (choose_rule_to_evict(table, &evict)) {
3477 ovs_rwlock_unlock(&rule->evict);
3478 ovs_rwlock_unlock(&evict->evict);
3479 if (evict->pending) {
3480 error = OFPROTO_POSTPONE;
3484 ovs_rwlock_unlock(&rule->evict);
3485 error = OFPERR_OFPFMFC_TABLE_FULL;
3492 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3493 op = ofoperation_create(group, rule, OFOPERATION_ADD, 0);
3494 op->victim = victim;
3496 error = ofproto->ofproto_class->rule_construct(rule);
3498 op->group->n_running--;
3499 ofoperation_destroy(rule->pending);
3501 /* It would be better if we maintained the lock we took in
3502 * choose_rule_to_evict() earlier, but that confuses the thread
3503 * safety analysis, and this code is fragile enough that we really
3504 * need it. In the worst case, we'll have to block a little while
3505 * before we perform the eviction, which doesn't seem like a big
3507 ovs_rwlock_wrlock(&evict->evict);
3508 delete_flow__(evict, group, OFPRR_EVICTION);
3510 ofopgroup_submit(group);
3514 /* Back out if an error occurred. */
3516 oftable_substitute_rule(rule, victim);
3517 ofproto_rule_destroy__(rule);
3522 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3524 /* Modifies the rules listed in 'rules', changing their actions to match those
3527 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3530 * Returns 0 on success, otherwise an OpenFlow error code. */
3532 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3533 struct ofputil_flow_mod *fm, const struct ofp_header *request,
3536 struct ofopgroup *group;
3540 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3541 error = OFPERR_OFPBRC_EPERM;
3542 LIST_FOR_EACH (rule, ofproto_node, rules) {
3543 struct ofoperation *op;
3544 bool actions_changed;
3546 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3548 if (rule_is_modifiable(rule)) {
3549 /* At least one rule is modifiable, don't report EPERM error. */
3555 /* Verify actions. */
3556 error = ofpacts_check(fm->ofpacts, fm->ofpacts_len, &fm->match.flow,
3557 ofproto->max_ports, rule->table_id);
3562 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
3563 rule->ofpacts, rule->ofpacts_len);
3565 op = ofoperation_create(group, rule, OFOPERATION_MODIFY, 0);
3567 if (fm->modify_cookie && fm->new_cookie != htonll(UINT64_MAX)) {
3568 ofproto_rule_change_cookie(ofproto, rule, fm->new_cookie);
3570 if (actions_changed) {
3571 op->ofpacts = rule->ofpacts;
3572 op->ofpacts_len = rule->ofpacts_len;
3573 op->meter_id = rule->meter_id;
3574 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3575 rule->ofpacts_len = fm->ofpacts_len;
3576 rule->meter_id = find_meter(rule->ofpacts, rule->ofpacts_len);
3577 rule->ofproto->ofproto_class->rule_modify_actions(rule);
3579 ofoperation_complete(op, 0);
3582 ofopgroup_submit(group);
3588 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
3589 struct ofputil_flow_mod *fm, const struct ofp_header *request)
3591 if (fm->cookie_mask != htonll(0) || fm->new_cookie == htonll(UINT64_MAX)) {
3594 return add_flow(ofproto, ofconn, fm, request);
3597 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
3600 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3603 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3604 struct ofputil_flow_mod *fm,
3605 const struct ofp_header *request)
3610 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3611 fm->cookie, fm->cookie_mask,
3615 } else if (list_is_empty(&rules)) {
3616 return modify_flows_add(ofproto, ofconn, fm, request);
3618 return modify_flows__(ofproto, ofconn, fm, request, &rules);
3622 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3625 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3628 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3629 struct ofputil_flow_mod *fm,
3630 const struct ofp_header *request)
3635 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3636 fm->priority, fm->cookie, fm->cookie_mask,
3641 } else if (list_is_empty(&rules)) {
3642 return modify_flows_add(ofproto, ofconn, fm, request);
3644 return list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
3645 fm, request, &rules)
3650 /* OFPFC_DELETE implementation. */
3653 delete_flow__(struct rule *rule, struct ofopgroup *group,
3654 enum ofp_flow_removed_reason reason)
3656 struct ofproto *ofproto = rule->ofproto;
3658 ofproto_rule_send_removed(rule, reason);
3660 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
3661 oftable_remove_rule(rule);
3662 ofproto->ofproto_class->rule_destruct(rule);
3665 /* Deletes the rules listed in 'rules'.
3667 * Returns 0 on success, otherwise an OpenFlow error code. */
3669 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3670 const struct ofp_header *request, struct list *rules,
3671 enum ofp_flow_removed_reason reason)
3673 struct rule *rule, *next;
3674 struct ofopgroup *group;
3676 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
3677 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
3678 ovs_rwlock_wrlock(&rule->evict);
3679 delete_flow__(rule, group, reason);
3681 ofopgroup_submit(group);
3686 /* Implements OFPFC_DELETE. */
3688 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3689 const struct ofputil_flow_mod *fm,
3690 const struct ofp_header *request)
3695 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3696 fm->cookie, fm->cookie_mask,
3697 fm->out_port, &rules);
3698 return (error ? error
3699 : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
3700 &rules, OFPRR_DELETE)
3704 /* Implements OFPFC_DELETE_STRICT. */
3706 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3707 const struct ofputil_flow_mod *fm,
3708 const struct ofp_header *request)
3713 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3714 fm->priority, fm->cookie, fm->cookie_mask,
3715 fm->out_port, &rules);
3716 return (error ? error
3717 : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
3724 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
3726 struct ofputil_flow_removed fr;
3728 if (ofproto_rule_is_hidden(rule) || !rule->send_flow_removed) {
3732 minimatch_expand(&rule->cr.match, &fr.match);
3733 fr.priority = rule->cr.priority;
3734 fr.cookie = rule->flow_cookie;
3736 fr.table_id = rule->table_id;
3737 calc_duration(rule->created, time_msec(),
3738 &fr.duration_sec, &fr.duration_nsec);
3739 ovs_mutex_lock(&rule->timeout_mutex);
3740 fr.idle_timeout = rule->idle_timeout;
3741 fr.hard_timeout = rule->hard_timeout;
3742 ovs_mutex_unlock(&rule->timeout_mutex);
3743 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
3746 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
3750 ofproto_rule_update_used(struct rule *rule, long long int used)
3752 if (used > rule->used) {
3753 struct eviction_group *evg = rule->eviction_group;
3757 heap_change(&evg->rules, &rule->evg_node,
3758 rule_eviction_priority(rule));
3763 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3764 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3767 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
3770 * ofproto implementation ->run() functions should use this function to expire
3771 * OpenFlow flows. */
3773 ofproto_rule_expire(struct rule *rule, uint8_t reason)
3775 struct ofproto *ofproto = rule->ofproto;
3776 struct ofopgroup *group;
3778 ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
3780 ofproto_rule_send_removed(rule, reason);
3782 group = ofopgroup_create_unattached(ofproto);
3783 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
3784 oftable_remove_rule(rule);
3785 ofproto->ofproto_class->rule_destruct(rule);
3786 ofopgroup_submit(group);
3790 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3792 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3793 struct ofputil_flow_mod fm;
3794 uint64_t ofpacts_stub[1024 / 8];
3795 struct ofpbuf ofpacts;
3799 error = reject_slave_controller(ofconn);
3804 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3805 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
3808 error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
3811 goto exit_free_ofpacts;
3814 /* Record the operation for logging a summary report. */
3815 switch (fm.command) {
3821 case OFPFC_MODIFY_STRICT:
3822 ofproto->n_modify++;
3826 case OFPFC_DELETE_STRICT:
3827 ofproto->n_delete++;
3832 if (ofproto->next_op_report == LLONG_MAX) {
3833 ofproto->first_op = now;
3834 ofproto->next_op_report = MAX(now + 10 * 1000,
3835 ofproto->op_backoff);
3836 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
3838 ofproto->last_op = now;
3841 ofpbuf_uninit(&ofpacts);
3847 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
3848 struct ofputil_flow_mod *fm, const struct ofp_header *oh)
3850 if (ofproto->n_pending >= 50) {
3851 ovs_assert(!list_is_empty(&ofproto->pending));
3852 return OFPROTO_POSTPONE;
3855 switch (fm->command) {
3857 return add_flow(ofproto, ofconn, fm, oh);
3860 return modify_flows_loose(ofproto, ofconn, fm, oh);
3862 case OFPFC_MODIFY_STRICT:
3863 return modify_flow_strict(ofproto, ofconn, fm, oh);
3866 return delete_flows_loose(ofproto, ofconn, fm, oh);
3868 case OFPFC_DELETE_STRICT:
3869 return delete_flow_strict(ofproto, ofconn, fm, oh);
3872 if (fm->command > 0xff) {
3873 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
3874 "flow_mod_table_id extension is not enabled",
3877 return OFPERR_OFPFMFC_BAD_COMMAND;
3882 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3884 struct ofputil_role_request request;
3885 struct ofputil_role_request reply;
3889 error = ofputil_decode_role_message(oh, &request);
3894 if (request.role != OFPCR12_ROLE_NOCHANGE) {
3895 if (ofconn_get_role(ofconn) != request.role
3896 && ofconn_has_pending_opgroups(ofconn)) {
3897 return OFPROTO_POSTPONE;
3900 if (request.have_generation_id
3901 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
3902 return OFPERR_OFPRRFC_STALE;
3905 ofconn_set_role(ofconn, request.role);
3908 reply.role = ofconn_get_role(ofconn);
3909 reply.have_generation_id = ofconn_get_master_election_id(
3910 ofconn, &reply.generation_id);
3911 buf = ofputil_encode_role_reply(oh, &reply);
3912 ofconn_send_reply(ofconn, buf);
3918 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
3919 const struct ofp_header *oh)
3921 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
3922 enum ofputil_protocol cur, next;
3924 cur = ofconn_get_protocol(ofconn);
3925 next = ofputil_protocol_set_tid(cur, msg->set != 0);
3926 ofconn_set_protocol(ofconn, next);
3932 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3934 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
3935 enum ofputil_protocol cur, next;
3936 enum ofputil_protocol next_base;
3938 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
3940 return OFPERR_OFPBRC_EPERM;
3943 cur = ofconn_get_protocol(ofconn);
3944 next = ofputil_protocol_set_base(cur, next_base);
3945 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
3946 /* Avoid sending async messages in surprising protocol. */
3947 return OFPROTO_POSTPONE;
3950 ofconn_set_protocol(ofconn, next);
3955 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
3956 const struct ofp_header *oh)
3958 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
3961 format = ntohl(msg->format);
3962 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
3963 return OFPERR_OFPBRC_EPERM;
3966 if (format != ofconn_get_packet_in_format(ofconn)
3967 && ofconn_has_pending_opgroups(ofconn)) {
3968 /* Avoid sending async message in surprsing packet in format. */
3969 return OFPROTO_POSTPONE;
3972 ofconn_set_packet_in_format(ofconn, format);
3977 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
3979 const struct nx_async_config *msg = ofpmsg_body(oh);
3980 uint32_t master[OAM_N_TYPES];
3981 uint32_t slave[OAM_N_TYPES];
3983 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
3984 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
3985 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
3987 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
3988 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
3989 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
3991 ofconn_set_async_config(ofconn, master, slave);
3992 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
3993 !ofconn_get_miss_send_len(ofconn)) {
3994 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
4001 handle_nxt_set_controller_id(struct ofconn *ofconn,
4002 const struct ofp_header *oh)
4004 const struct nx_controller_id *nci = ofpmsg_body(oh);
4006 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
4007 return OFPERR_NXBRC_MUST_BE_ZERO;
4010 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
4015 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4019 if (ofconn_has_pending_opgroups(ofconn)) {
4020 return OFPROTO_POSTPONE;
4023 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
4024 ? OFPRAW_OFPT10_BARRIER_REPLY
4025 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
4026 ofconn_send_reply(ofconn, buf);
4031 ofproto_compose_flow_refresh_update(const struct rule *rule,
4032 enum nx_flow_monitor_flags flags,
4035 struct ofoperation *op = rule->pending;
4036 struct ofputil_flow_update fu;
4039 if (op && op->type == OFOPERATION_ADD && !op->victim) {
4040 /* We'll report the final flow when the operation completes. Reporting
4041 * it now would cause a duplicate report later. */
4045 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
4046 ? NXFME_ADDED : NXFME_MODIFIED);
4048 ovs_mutex_lock(&rule->timeout_mutex);
4049 fu.idle_timeout = rule->idle_timeout;
4050 fu.hard_timeout = rule->hard_timeout;
4051 ovs_mutex_unlock(&rule->timeout_mutex);
4052 fu.table_id = rule->table_id;
4053 fu.cookie = rule->flow_cookie;
4054 minimatch_expand(&rule->cr.match, &match);
4056 fu.priority = rule->cr.priority;
4057 if (!(flags & NXFMF_ACTIONS)) {
4061 fu.ofpacts = rule->ofpacts;
4062 fu.ofpacts_len = rule->ofpacts_len;
4064 /* An operation is in progress. Use the previous version of the flow's
4065 * actions, so that when the operation commits we report the change. */
4067 case OFOPERATION_ADD:
4068 /* We already verified that there was a victim. */
4069 fu.ofpacts = op->victim->ofpacts;
4070 fu.ofpacts_len = op->victim->ofpacts_len;
4073 case OFOPERATION_MODIFY:
4075 fu.ofpacts = op->ofpacts;
4076 fu.ofpacts_len = op->ofpacts_len;
4078 fu.ofpacts = rule->ofpacts;
4079 fu.ofpacts_len = rule->ofpacts_len;
4083 case OFOPERATION_DELETE:
4084 fu.ofpacts = rule->ofpacts;
4085 fu.ofpacts_len = rule->ofpacts_len;
4093 if (list_is_empty(msgs)) {
4094 ofputil_start_flow_update(msgs);
4096 ofputil_append_flow_update(&fu, msgs);
4100 ofmonitor_compose_refresh_updates(struct list *rules, struct list *msgs)
4104 LIST_FOR_EACH (rule, ofproto_node, rules) {
4105 enum nx_flow_monitor_flags flags = rule->monitor_flags;
4106 rule->monitor_flags = 0;
4108 ofproto_compose_flow_refresh_update(rule, flags, msgs);
4113 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
4114 struct rule *rule, uint64_t seqno,
4117 enum nx_flow_monitor_flags update;
4119 if (ofproto_rule_is_hidden(rule)) {
4124 ? ofoperation_has_out_port(rule->pending, m->out_port)
4125 : ofproto_rule_has_out_port(rule, m->out_port))) {
4130 if (rule->add_seqno > seqno) {
4131 update = NXFMF_ADD | NXFMF_MODIFY;
4132 } else if (rule->modify_seqno > seqno) {
4133 update = NXFMF_MODIFY;
4138 if (!(m->flags & update)) {
4142 update = NXFMF_INITIAL;
4145 if (!rule->monitor_flags) {
4146 list_push_back(rules, &rule->ofproto_node);
4148 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
4152 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
4156 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
4157 const struct ofoperation *op;
4158 const struct oftable *table;
4159 struct cls_rule target;
4161 cls_rule_init_from_minimatch(&target, &m->match, 0);
4162 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
4163 struct cls_cursor cursor;
4166 ovs_rwlock_rdlock(&table->cls.rwlock);
4167 cls_cursor_init(&cursor, &table->cls, &target);
4168 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4169 ovs_assert(!rule->pending); /* XXX */
4170 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4172 ovs_rwlock_unlock(&table->cls.rwlock);
4175 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
4176 struct rule *rule = op->rule;
4178 if (((m->table_id == 0xff
4179 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
4180 : m->table_id == rule->table_id))
4181 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
4182 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4185 cls_rule_destroy(&target);
4189 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
4192 if (m->flags & NXFMF_INITIAL) {
4193 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
4198 ofmonitor_collect_resume_rules(struct ofmonitor *m,
4199 uint64_t seqno, struct list *rules)
4201 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
4205 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
4207 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4208 struct ofmonitor **monitors;
4209 size_t n_monitors, allocated_monitors;
4210 struct list replies;
4217 ofpbuf_use_const(&b, oh, ntohs(oh->length));
4219 n_monitors = allocated_monitors = 0;
4221 struct ofputil_flow_monitor_request request;
4222 struct ofmonitor *m;
4225 retval = ofputil_decode_flow_monitor_request(&request, &b);
4226 if (retval == EOF) {
4228 } else if (retval) {
4233 if (request.table_id != 0xff
4234 && request.table_id >= ofproto->n_tables) {
4235 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4239 error = ofmonitor_create(&request, ofconn, &m);
4244 if (n_monitors >= allocated_monitors) {
4245 monitors = x2nrealloc(monitors, &allocated_monitors,
4248 monitors[n_monitors++] = m;
4252 for (i = 0; i < n_monitors; i++) {
4253 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
4256 ofpmp_init(&replies, oh);
4257 ofmonitor_compose_refresh_updates(&rules, &replies);
4258 ofconn_send_replies(ofconn, &replies);
4265 for (i = 0; i < n_monitors; i++) {
4266 ofmonitor_destroy(monitors[i]);
4273 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
4275 struct ofmonitor *m;
4278 id = ofputil_decode_flow_monitor_cancel(oh);
4279 m = ofmonitor_lookup(ofconn, id);
4281 return OFPERR_NXBRC_FM_BAD_ID;
4284 ofmonitor_destroy(m);
4288 /* Meters implementation.
4290 * Meter table entry, indexed by the OpenFlow meter_id.
4291 * These are always dynamically allocated to allocate enough space for
4293 * 'created' is used to compute the duration for meter stats.
4294 * 'list rules' is needed so that we can delete the dependent rules when the
4295 * meter table entry is deleted.
4296 * 'provider_meter_id' is for the provider's private use.
4299 long long int created; /* Time created. */
4300 struct list rules; /* List of "struct rule_dpif"s. */
4301 ofproto_meter_id provider_meter_id;
4302 uint16_t flags; /* Meter flags. */
4303 uint16_t n_bands; /* Number of meter bands. */
4304 struct ofputil_meter_band *bands;
4308 * This is used in instruction validation at flow set-up time,
4309 * as flows may not use non-existing meters.
4310 * This is also used by ofproto-providers to translate OpenFlow meter_ids
4311 * in METER instructions to the corresponding provider meter IDs.
4312 * Return value of UINT32_MAX signifies an invalid meter.
4315 ofproto_get_provider_meter_id(const struct ofproto * ofproto,
4316 uint32_t of_meter_id)
4318 if (of_meter_id && of_meter_id <= ofproto->meter_features.max_meters) {
4319 const struct meter *meter = ofproto->meters[of_meter_id];
4321 return meter->provider_meter_id.uint32;
4328 meter_update(struct meter *meter, const struct ofputil_meter_config *config)
4332 meter->flags = config->flags;
4333 meter->n_bands = config->n_bands;
4334 meter->bands = xmemdup(config->bands,
4335 config->n_bands * sizeof *meter->bands);
4338 static struct meter *
4339 meter_create(const struct ofputil_meter_config *config,
4340 ofproto_meter_id provider_meter_id)
4342 struct meter *meter;
4344 meter = xzalloc(sizeof *meter);
4345 meter->provider_meter_id = provider_meter_id;
4346 meter->created = time_msec();
4347 list_init(&meter->rules);
4349 meter_update(meter, config);
4355 meter_delete(struct ofproto *ofproto, uint32_t first, uint32_t last)
4358 for (mid = first; mid <= last; ++mid) {
4359 struct meter *meter = ofproto->meters[mid];
4361 ofproto->meters[mid] = NULL;
4362 ofproto->ofproto_class->meter_del(ofproto,
4363 meter->provider_meter_id);
4371 handle_add_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
4373 ofproto_meter_id provider_meter_id = { UINT32_MAX };
4374 struct meter **meterp = &ofproto->meters[mm->meter.meter_id];
4378 return OFPERR_OFPMMFC_METER_EXISTS;
4381 error = ofproto->ofproto_class->meter_set(ofproto, &provider_meter_id,
4384 ovs_assert(provider_meter_id.uint32 != UINT32_MAX);
4385 *meterp = meter_create(&mm->meter, provider_meter_id);
4391 handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
4393 struct meter *meter = ofproto->meters[mm->meter.meter_id];
4397 return OFPERR_OFPMMFC_UNKNOWN_METER;
4400 error = ofproto->ofproto_class->meter_set(ofproto,
4401 &meter->provider_meter_id,
4403 ovs_assert(meter->provider_meter_id.uint32 != UINT32_MAX);
4405 meter_update(meter, &mm->meter);
4411 handle_delete_meter(struct ofconn *ofconn, const struct ofp_header *oh,
4412 struct ofputil_meter_mod *mm)
4414 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4415 uint32_t meter_id = mm->meter.meter_id;
4416 uint32_t first, last;
4419 if (meter_id == OFPM13_ALL) {
4421 last = ofproto->meter_features.max_meters;
4423 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
4426 first = last = meter_id;
4429 /* First delete the rules that use this meter. If any of those rules are
4430 * currently being modified, postpone the whole operation until later. */
4432 for (meter_id = first; meter_id <= last; ++meter_id) {
4433 struct meter *meter = ofproto->meters[meter_id];
4434 if (meter && !list_is_empty(&meter->rules)) {
4437 LIST_FOR_EACH (rule, meter_list_node, &meter->rules) {
4438 if (rule->pending) {
4439 return OFPROTO_POSTPONE;
4441 list_push_back(&rules, &rule->ofproto_node);
4445 if (!list_is_empty(&rules)) {
4446 delete_flows__(ofproto, ofconn, oh, &rules, OFPRR_METER_DELETE);
4449 /* Delete the meters. */
4450 meter_delete(ofproto, first, last);
4456 handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4458 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4459 struct ofputil_meter_mod mm;
4460 uint64_t bands_stub[256 / 8];
4461 struct ofpbuf bands;
4465 error = reject_slave_controller(ofconn);
4470 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
4472 error = ofputil_decode_meter_mod(oh, &mm, &bands);
4474 goto exit_free_bands;
4477 meter_id = mm.meter.meter_id;
4479 if (mm.command != OFPMC13_DELETE) {
4480 /* Fails also when meters are not implemented by the provider. */
4481 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
4482 error = OFPERR_OFPMMFC_INVALID_METER;
4483 goto exit_free_bands;
4485 if (mm.meter.n_bands > ofproto->meter_features.max_bands) {
4486 error = OFPERR_OFPMMFC_OUT_OF_BANDS;
4487 goto exit_free_bands;
4491 switch (mm.command) {
4493 error = handle_add_meter(ofproto, &mm);
4496 case OFPMC13_MODIFY:
4497 error = handle_modify_meter(ofproto, &mm);
4500 case OFPMC13_DELETE:
4501 error = handle_delete_meter(ofconn, oh, &mm);
4505 error = OFPERR_OFPMMFC_BAD_COMMAND;
4510 ofpbuf_uninit(&bands);
4515 handle_meter_features_request(struct ofconn *ofconn,
4516 const struct ofp_header *request)
4518 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4519 struct ofputil_meter_features features;
4522 if (ofproto->ofproto_class->meter_get_features) {
4523 ofproto->ofproto_class->meter_get_features(ofproto, &features);
4525 memset(&features, 0, sizeof features);
4527 b = ofputil_encode_meter_features_reply(&features, request);
4529 ofconn_send_reply(ofconn, b);
4534 handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request,
4537 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4538 struct list replies;
4539 uint64_t bands_stub[256 / 8];
4540 struct ofpbuf bands;
4541 uint32_t meter_id, first, last;
4543 ofputil_decode_meter_request(request, &meter_id);
4545 if (meter_id == OFPM13_ALL) {
4547 last = ofproto->meter_features.max_meters;
4549 if (!meter_id || meter_id > ofproto->meter_features.max_meters ||
4550 !ofproto->meters[meter_id]) {
4551 return OFPERR_OFPMMFC_UNKNOWN_METER;
4553 first = last = meter_id;
4556 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
4557 ofpmp_init(&replies, request);
4559 for (meter_id = first; meter_id <= last; ++meter_id) {
4560 struct meter *meter = ofproto->meters[meter_id];
4562 continue; /* Skip non-existing meters. */
4564 if (type == OFPTYPE_METER_STATS_REQUEST) {
4565 struct ofputil_meter_stats stats;
4567 stats.meter_id = meter_id;
4569 /* Provider sets the packet and byte counts, we do the rest. */
4570 stats.flow_count = list_size(&meter->rules);
4571 calc_duration(meter->created, time_msec(),
4572 &stats.duration_sec, &stats.duration_nsec);
4573 stats.n_bands = meter->n_bands;
4574 ofpbuf_clear(&bands);
4576 = ofpbuf_put_uninit(&bands,
4577 meter->n_bands * sizeof *stats.bands);
4579 if (!ofproto->ofproto_class->meter_get(ofproto,
4580 meter->provider_meter_id,
4582 ofputil_append_meter_stats(&replies, &stats);
4584 } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */
4585 struct ofputil_meter_config config;
4587 config.meter_id = meter_id;
4588 config.flags = meter->flags;
4589 config.n_bands = meter->n_bands;
4590 config.bands = meter->bands;
4591 ofputil_append_meter_config(&replies, &config);
4595 ofconn_send_replies(ofconn, &replies);
4596 ofpbuf_uninit(&bands);
4601 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
4603 const struct ofp_header *oh = msg->data;
4607 error = ofptype_decode(&type, oh);
4613 /* OpenFlow requests. */
4614 case OFPTYPE_ECHO_REQUEST:
4615 return handle_echo_request(ofconn, oh);
4617 case OFPTYPE_FEATURES_REQUEST:
4618 return handle_features_request(ofconn, oh);
4620 case OFPTYPE_GET_CONFIG_REQUEST:
4621 return handle_get_config_request(ofconn, oh);
4623 case OFPTYPE_SET_CONFIG:
4624 return handle_set_config(ofconn, oh);
4626 case OFPTYPE_PACKET_OUT:
4627 return handle_packet_out(ofconn, oh);
4629 case OFPTYPE_PORT_MOD:
4630 return handle_port_mod(ofconn, oh);
4632 case OFPTYPE_FLOW_MOD:
4633 return handle_flow_mod(ofconn, oh);
4635 case OFPTYPE_METER_MOD:
4636 return handle_meter_mod(ofconn, oh);
4638 case OFPTYPE_BARRIER_REQUEST:
4639 return handle_barrier_request(ofconn, oh);
4641 case OFPTYPE_ROLE_REQUEST:
4642 return handle_role_request(ofconn, oh);
4644 /* OpenFlow replies. */
4645 case OFPTYPE_ECHO_REPLY:
4648 /* Nicira extension requests. */
4649 case OFPTYPE_FLOW_MOD_TABLE_ID:
4650 return handle_nxt_flow_mod_table_id(ofconn, oh);
4652 case OFPTYPE_SET_FLOW_FORMAT:
4653 return handle_nxt_set_flow_format(ofconn, oh);
4655 case OFPTYPE_SET_PACKET_IN_FORMAT:
4656 return handle_nxt_set_packet_in_format(ofconn, oh);
4658 case OFPTYPE_SET_CONTROLLER_ID:
4659 return handle_nxt_set_controller_id(ofconn, oh);
4661 case OFPTYPE_FLOW_AGE:
4662 /* Nothing to do. */
4665 case OFPTYPE_FLOW_MONITOR_CANCEL:
4666 return handle_flow_monitor_cancel(ofconn, oh);
4668 case OFPTYPE_SET_ASYNC_CONFIG:
4669 return handle_nxt_set_async_config(ofconn, oh);
4671 /* Statistics requests. */
4672 case OFPTYPE_DESC_STATS_REQUEST:
4673 return handle_desc_stats_request(ofconn, oh);
4675 case OFPTYPE_FLOW_STATS_REQUEST:
4676 return handle_flow_stats_request(ofconn, oh);
4678 case OFPTYPE_AGGREGATE_STATS_REQUEST:
4679 return handle_aggregate_stats_request(ofconn, oh);
4681 case OFPTYPE_TABLE_STATS_REQUEST:
4682 return handle_table_stats_request(ofconn, oh);
4684 case OFPTYPE_PORT_STATS_REQUEST:
4685 return handle_port_stats_request(ofconn, oh);
4687 case OFPTYPE_QUEUE_STATS_REQUEST:
4688 return handle_queue_stats_request(ofconn, oh);
4690 case OFPTYPE_PORT_DESC_STATS_REQUEST:
4691 return handle_port_desc_stats_request(ofconn, oh);
4693 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
4694 return handle_flow_monitor_request(ofconn, oh);
4696 case OFPTYPE_METER_STATS_REQUEST:
4697 case OFPTYPE_METER_CONFIG_STATS_REQUEST:
4698 return handle_meter_request(ofconn, oh, type);
4700 case OFPTYPE_METER_FEATURES_STATS_REQUEST:
4701 return handle_meter_features_request(ofconn, oh);
4703 /* FIXME: Change the following once they are implemented: */
4704 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
4705 case OFPTYPE_GET_ASYNC_REQUEST:
4706 case OFPTYPE_GROUP_STATS_REQUEST:
4707 case OFPTYPE_GROUP_DESC_STATS_REQUEST:
4708 case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
4709 case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
4710 return OFPERR_OFPBRC_BAD_TYPE;
4714 case OFPTYPE_FEATURES_REPLY:
4715 case OFPTYPE_GET_CONFIG_REPLY:
4716 case OFPTYPE_PACKET_IN:
4717 case OFPTYPE_FLOW_REMOVED:
4718 case OFPTYPE_PORT_STATUS:
4719 case OFPTYPE_BARRIER_REPLY:
4720 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
4721 case OFPTYPE_DESC_STATS_REPLY:
4722 case OFPTYPE_FLOW_STATS_REPLY:
4723 case OFPTYPE_QUEUE_STATS_REPLY:
4724 case OFPTYPE_PORT_STATS_REPLY:
4725 case OFPTYPE_TABLE_STATS_REPLY:
4726 case OFPTYPE_AGGREGATE_STATS_REPLY:
4727 case OFPTYPE_PORT_DESC_STATS_REPLY:
4728 case OFPTYPE_ROLE_REPLY:
4729 case OFPTYPE_FLOW_MONITOR_PAUSED:
4730 case OFPTYPE_FLOW_MONITOR_RESUMED:
4731 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
4732 case OFPTYPE_GET_ASYNC_REPLY:
4733 case OFPTYPE_GROUP_STATS_REPLY:
4734 case OFPTYPE_GROUP_DESC_STATS_REPLY:
4735 case OFPTYPE_GROUP_FEATURES_STATS_REPLY:
4736 case OFPTYPE_METER_STATS_REPLY:
4737 case OFPTYPE_METER_CONFIG_STATS_REPLY:
4738 case OFPTYPE_METER_FEATURES_STATS_REPLY:
4739 case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
4741 return OFPERR_OFPBRC_BAD_TYPE;
4746 handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
4748 int error = handle_openflow__(ofconn, ofp_msg);
4749 if (error && error != OFPROTO_POSTPONE) {
4750 ofconn_send_error(ofconn, ofp_msg->data, error);
4752 COVERAGE_INC(ofproto_recv_openflow);
4753 return error != OFPROTO_POSTPONE;
4756 /* Asynchronous operations. */
4758 /* Creates and returns a new ofopgroup that is not associated with any
4759 * OpenFlow connection.
4761 * The caller should add operations to the returned group with
4762 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4763 static struct ofopgroup *
4764 ofopgroup_create_unattached(struct ofproto *ofproto)
4766 struct ofopgroup *group = xzalloc(sizeof *group);
4767 group->ofproto = ofproto;
4768 list_init(&group->ofproto_node);
4769 list_init(&group->ops);
4770 list_init(&group->ofconn_node);
4774 /* Creates and returns a new ofopgroup for 'ofproto'.
4776 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
4777 * connection. The 'request' and 'buffer_id' arguments are ignored.
4779 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
4780 * If the ofopgroup eventually fails, then the error reply will include
4781 * 'request'. If the ofopgroup eventually succeeds, then the packet with
4782 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
4784 * The caller should add operations to the returned group with
4785 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4786 static struct ofopgroup *
4787 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
4788 const struct ofp_header *request, uint32_t buffer_id)
4790 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
4792 size_t request_len = ntohs(request->length);
4794 ovs_assert(ofconn_get_ofproto(ofconn) == ofproto);
4796 ofconn_add_opgroup(ofconn, &group->ofconn_node);
4797 group->ofconn = ofconn;
4798 group->request = xmemdup(request, MIN(request_len, 64));
4799 group->buffer_id = buffer_id;
4804 /* Submits 'group' for processing.
4806 * If 'group' contains no operations (e.g. none were ever added, or all of the
4807 * ones that were added completed synchronously), then it is destroyed
4808 * immediately. Otherwise it is added to the ofproto's list of pending
4811 ofopgroup_submit(struct ofopgroup *group)
4813 if (!group->n_running) {
4814 ofopgroup_complete(group);
4816 list_push_back(&group->ofproto->pending, &group->ofproto_node);
4817 group->ofproto->n_pending++;
4822 ofopgroup_complete(struct ofopgroup *group)
4824 struct ofproto *ofproto = group->ofproto;
4826 struct ofconn *abbrev_ofconn;
4827 ovs_be32 abbrev_xid;
4829 struct ofoperation *op, *next_op;
4832 ovs_assert(!group->n_running);
4835 LIST_FOR_EACH (op, group_node, &group->ops) {
4842 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
4843 LIST_FOR_EACH (op, group_node, &group->ops) {
4844 if (op->type != OFOPERATION_DELETE) {
4845 struct ofpbuf *packet;
4848 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
4852 error = rule_execute(op->rule, in_port, packet);
4859 if (!error && !list_is_empty(&group->ofconn_node)) {
4860 abbrev_ofconn = group->ofconn;
4861 abbrev_xid = group->request->xid;
4863 abbrev_ofconn = NULL;
4864 abbrev_xid = htonl(0);
4866 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
4867 struct rule *rule = op->rule;
4869 /* We generally want to report the change to active OpenFlow flow
4870 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
4872 - The operation failed.
4874 - The affected rule is not visible to controllers.
4876 - The operation's only effect was to update rule->modified. */
4878 || ofproto_rule_is_hidden(rule)
4879 || (op->type == OFOPERATION_MODIFY
4881 && rule->flow_cookie == op->flow_cookie))) {
4882 /* Check that we can just cast from ofoperation_type to
4883 * nx_flow_update_event. */
4884 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_ADD
4886 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_DELETE
4888 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_MODIFY
4891 ofmonitor_report(ofproto->connmgr, rule,
4892 (enum nx_flow_update_event) op->type,
4893 op->reason, abbrev_ofconn, abbrev_xid);
4896 rule->pending = NULL;
4899 case OFOPERATION_ADD:
4903 ofproto_rule_destroy__(op->victim);
4904 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
4905 if (vid_mask == VLAN_VID_MASK) {
4906 if (ofproto->vlan_bitmap) {
4907 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
4908 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
4909 bitmap_set1(ofproto->vlan_bitmap, vid);
4910 ofproto->vlans_changed = true;
4913 ofproto->vlans_changed = true;
4917 oftable_substitute_rule(rule, op->victim);
4918 ofproto_rule_destroy__(rule);
4922 case OFOPERATION_DELETE:
4923 ovs_assert(!op->error);
4924 ofproto_rule_destroy__(rule);
4928 case OFOPERATION_MODIFY:
4930 rule->modified = time_msec();
4932 ofproto_rule_change_cookie(ofproto, rule, op->flow_cookie);
4934 free(rule->ofpacts);
4935 rule->ofpacts = op->ofpacts;
4936 rule->ofpacts_len = op->ofpacts_len;
4938 op->ofpacts_len = 0;
4947 ofoperation_destroy(op);
4950 ofmonitor_flush(ofproto->connmgr);
4952 if (!list_is_empty(&group->ofproto_node)) {
4953 ovs_assert(ofproto->n_pending > 0);
4954 ofproto->n_pending--;
4955 list_remove(&group->ofproto_node);
4957 if (!list_is_empty(&group->ofconn_node)) {
4958 list_remove(&group->ofconn_node);
4960 ofconn_send_error(group->ofconn, group->request, error);
4962 connmgr_retry(ofproto->connmgr);
4964 free(group->request);
4968 /* Initiates a new operation on 'rule', of the specified 'type', within
4969 * 'group'. Prior to calling, 'rule' must not have any pending operation.
4971 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
4972 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
4974 * Returns the newly created ofoperation (which is also available as
4975 * rule->pending). */
4976 static struct ofoperation *
4977 ofoperation_create(struct ofopgroup *group, struct rule *rule,
4978 enum ofoperation_type type,
4979 enum ofp_flow_removed_reason reason)
4981 struct ofproto *ofproto = group->ofproto;
4982 struct ofoperation *op;
4984 ovs_assert(!rule->pending);
4986 op = rule->pending = xzalloc(sizeof *op);
4988 list_push_back(&group->ops, &op->group_node);
4991 op->reason = reason;
4992 op->flow_cookie = rule->flow_cookie;
4996 if (type == OFOPERATION_DELETE) {
4997 hmap_insert(&ofproto->deletions, &op->hmap_node,
4998 cls_rule_hash(&rule->cr, rule->table_id));
5005 ofoperation_destroy(struct ofoperation *op)
5007 struct ofopgroup *group = op->group;
5010 op->rule->pending = NULL;
5012 if (op->type == OFOPERATION_DELETE) {
5013 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
5015 list_remove(&op->group_node);
5020 /* Indicates that 'op' completed with status 'error', which is either 0 to
5021 * indicate success or an OpenFlow error code on failure.
5023 * If 'error' is 0, indicating success, the operation will be committed
5024 * permanently to the flow table. There is one interesting subcase:
5026 * - If 'op' is an "add flow" operation that is replacing an existing rule in
5027 * the flow table (the "victim" rule) by a new one, then the caller must
5028 * have uninitialized any derived state in the victim rule, as in step 5 in
5029 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
5030 * performs steps 6 and 7 for the victim rule, most notably by calling its
5031 * ->rule_dealloc() function.
5033 * If 'error' is nonzero, then generally the operation will be rolled back:
5035 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
5036 * restores the original rule. The caller must have uninitialized any
5037 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
5038 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
5039 * and 7 for the new rule, calling its ->rule_dealloc() function.
5041 * - If 'op' is a "modify flow" operation, ofproto restores the original
5044 * - 'op' must not be a "delete flow" operation. Removing a rule is not
5045 * allowed to fail. It must always succeed.
5047 * Please see the large comment in ofproto/ofproto-provider.h titled
5048 * "Asynchronous Operation Support" for more information. */
5050 ofoperation_complete(struct ofoperation *op, enum ofperr error)
5052 struct ofopgroup *group = op->group;
5054 ovs_assert(op->rule->pending == op);
5055 ovs_assert(group->n_running > 0);
5056 ovs_assert(!error || op->type != OFOPERATION_DELETE);
5059 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
5060 ofopgroup_complete(group);
5065 ofoperation_get_victim(struct ofoperation *op)
5067 ovs_assert(op->type == OFOPERATION_ADD);
5072 pick_datapath_id(const struct ofproto *ofproto)
5074 const struct ofport *port;
5076 port = ofproto_get_port(ofproto, OFPP_LOCAL);
5078 uint8_t ea[ETH_ADDR_LEN];
5081 error = netdev_get_etheraddr(port->netdev, ea);
5083 return eth_addr_to_uint64(ea);
5085 VLOG_WARN("%s: could not get MAC address for %s (%s)",
5086 ofproto->name, netdev_get_name(port->netdev),
5087 ovs_strerror(error));
5089 return ofproto->fallback_dpid;
5093 pick_fallback_dpid(void)
5095 uint8_t ea[ETH_ADDR_LEN];
5096 eth_addr_nicira_random(ea);
5097 return eth_addr_to_uint64(ea);
5100 /* Table overflow policy. */
5102 /* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep'
5103 * to NULL if the table is not configured to evict rules or if the table
5104 * contains no evictable rules. (Rules with a readlock on their evict rwlock,
5105 * or with no timeouts are not evictable.) */
5107 choose_rule_to_evict(struct oftable *table, struct rule **rulep)
5109 struct eviction_group *evg;
5112 if (!table->eviction_fields) {
5116 /* In the common case, the outer and inner loops here will each be entered
5119 * - The inner loop normally "return"s in its first iteration. If the
5120 * eviction group has any evictable rules, then it always returns in
5123 * - The outer loop only iterates more than once if the largest eviction
5124 * group has no evictable rules.
5126 * - The outer loop can exit only if table's 'max_flows' is all filled up
5127 * by unevictable rules'. */
5128 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
5131 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
5132 if (!ovs_rwlock_trywrlock(&rule->evict)) {
5142 /* Searches 'ofproto' for tables that have more flows than their configured
5143 * maximum and that have flow eviction enabled, and evicts as many flows as
5144 * necessary and currently feasible from them.
5146 * This triggers only when an OpenFlow table has N flows in it and then the
5147 * client configures a maximum number of flows less than N. */
5149 ofproto_evict(struct ofproto *ofproto)
5151 struct ofopgroup *group;
5152 struct oftable *table;
5154 group = ofopgroup_create_unattached(ofproto);
5155 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
5156 while (table->eviction_fields) {
5160 ovs_rwlock_rdlock(&table->cls.rwlock);
5161 n_rules = classifier_count(&table->cls);
5162 ovs_rwlock_unlock(&table->cls.rwlock);
5164 if (n_rules <= table->max_flows) {
5168 if (!choose_rule_to_evict(table, &rule)) {
5172 if (rule->pending) {
5173 ovs_rwlock_unlock(&rule->evict);
5177 ofoperation_create(group, rule,
5178 OFOPERATION_DELETE, OFPRR_EVICTION);
5179 oftable_remove_rule(rule);
5180 ofproto->ofproto_class->rule_destruct(rule);
5183 ofopgroup_submit(group);
5186 /* Eviction groups. */
5188 /* Returns the priority to use for an eviction_group that contains 'n_rules'
5189 * rules. The priority contains low-order random bits to ensure that eviction
5190 * groups with the same number of rules are prioritized randomly. */
5192 eviction_group_priority(size_t n_rules)
5194 uint16_t size = MIN(UINT16_MAX, n_rules);
5195 return (size << 16) | random_uint16();
5198 /* Updates 'evg', an eviction_group within 'table', following a change that
5199 * adds or removes rules in 'evg'. */
5201 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
5203 heap_change(&table->eviction_groups_by_size, &evg->size_node,
5204 eviction_group_priority(heap_count(&evg->rules)));
5207 /* Destroys 'evg', an eviction_group within 'table':
5209 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
5210 * rules themselves, just removes them from the eviction group.)
5212 * - Removes 'evg' from 'table'.
5216 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
5218 while (!heap_is_empty(&evg->rules)) {
5221 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
5222 rule->eviction_group = NULL;
5224 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
5225 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
5226 heap_destroy(&evg->rules);
5230 /* Removes 'rule' from its eviction group, if any. */
5232 eviction_group_remove_rule(struct rule *rule)
5234 if (rule->eviction_group) {
5235 struct oftable *table = &rule->ofproto->tables[rule->table_id];
5236 struct eviction_group *evg = rule->eviction_group;
5238 rule->eviction_group = NULL;
5239 heap_remove(&evg->rules, &rule->evg_node);
5240 if (heap_is_empty(&evg->rules)) {
5241 eviction_group_destroy(table, evg);
5243 eviction_group_resized(table, evg);
5248 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
5249 * returns the hash value. */
5251 eviction_group_hash_rule(struct rule *rule)
5253 struct oftable *table = &rule->ofproto->tables[rule->table_id];
5254 const struct mf_subfield *sf;
5258 hash = table->eviction_group_id_basis;
5259 miniflow_expand(&rule->cr.match.flow, &flow);
5260 for (sf = table->eviction_fields;
5261 sf < &table->eviction_fields[table->n_eviction_fields];
5264 if (mf_are_prereqs_ok(sf->field, &flow)) {
5265 union mf_value value;
5267 mf_get_value(sf->field, &flow, &value);
5269 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
5271 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
5272 unsigned int start = sf->ofs + sf->n_bits;
5273 bitwise_zero(&value, sf->field->n_bytes, start,
5274 sf->field->n_bytes * 8 - start);
5276 hash = hash_bytes(&value, sf->field->n_bytes, hash);
5278 hash = hash_int(hash, 0);
5285 /* Returns an eviction group within 'table' with the given 'id', creating one
5287 static struct eviction_group *
5288 eviction_group_find(struct oftable *table, uint32_t id)
5290 struct eviction_group *evg;
5292 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
5296 evg = xmalloc(sizeof *evg);
5297 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
5298 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
5299 eviction_group_priority(0));
5300 heap_init(&evg->rules);
5305 /* Returns an eviction priority for 'rule'. The return value should be
5306 * interpreted so that higher priorities make a rule more attractive candidates
5309 rule_eviction_priority(struct rule *rule)
5311 long long int hard_expiration;
5312 long long int idle_expiration;
5313 long long int expiration;
5314 uint32_t expiration_offset;
5316 /* Calculate time of expiration. */
5317 ovs_mutex_lock(&rule->timeout_mutex);
5318 hard_expiration = (rule->hard_timeout
5319 ? rule->modified + rule->hard_timeout * 1000
5321 idle_expiration = (rule->idle_timeout
5322 ? rule->used + rule->idle_timeout * 1000
5324 expiration = MIN(hard_expiration, idle_expiration);
5325 ovs_mutex_unlock(&rule->timeout_mutex);
5326 if (expiration == LLONG_MAX) {
5330 /* Calculate the time of expiration as a number of (approximate) seconds
5331 * after program startup.
5333 * This should work OK for program runs that last UINT32_MAX seconds or
5334 * less. Therefore, please restart OVS at least once every 136 years. */
5335 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
5337 /* Invert the expiration offset because we're using a max-heap. */
5338 return UINT32_MAX - expiration_offset;
5341 /* Adds 'rule' to an appropriate eviction group for its oftable's
5342 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
5343 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
5346 * The caller must ensure that 'rule' is not already in an eviction group. */
5348 eviction_group_add_rule(struct rule *rule)
5350 struct ofproto *ofproto = rule->ofproto;
5351 struct oftable *table = &ofproto->tables[rule->table_id];
5354 ovs_mutex_lock(&rule->timeout_mutex);
5355 has_timeout = rule->hard_timeout || rule->idle_timeout;
5356 ovs_mutex_unlock(&rule->timeout_mutex);
5358 if (table->eviction_fields && has_timeout) {
5359 struct eviction_group *evg;
5361 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
5363 rule->eviction_group = evg;
5364 heap_insert(&evg->rules, &rule->evg_node,
5365 rule_eviction_priority(rule));
5366 eviction_group_resized(table, evg);
5372 /* Initializes 'table'. */
5374 oftable_init(struct oftable *table)
5376 memset(table, 0, sizeof *table);
5377 classifier_init(&table->cls);
5378 table->max_flows = UINT_MAX;
5381 /* Destroys 'table', including its classifier and eviction groups.
5383 * The caller is responsible for freeing 'table' itself. */
5385 oftable_destroy(struct oftable *table)
5387 ovs_rwlock_rdlock(&table->cls.rwlock);
5388 ovs_assert(classifier_is_empty(&table->cls));
5389 ovs_rwlock_unlock(&table->cls.rwlock);
5390 oftable_disable_eviction(table);
5391 classifier_destroy(&table->cls);
5395 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
5396 * string, then 'table' will use its default name.
5398 * This only affects the name exposed for a table exposed through the OpenFlow
5399 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
5401 oftable_set_name(struct oftable *table, const char *name)
5403 if (name && name[0]) {
5404 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
5405 if (!table->name || strncmp(name, table->name, len)) {
5407 table->name = xmemdup0(name, len);
5415 /* oftables support a choice of two policies when adding a rule would cause the
5416 * number of flows in the table to exceed the configured maximum number: either
5417 * they can refuse to add the new flow or they can evict some existing flow.
5418 * This function configures the former policy on 'table'. */
5420 oftable_disable_eviction(struct oftable *table)
5422 if (table->eviction_fields) {
5423 struct eviction_group *evg, *next;
5425 HMAP_FOR_EACH_SAFE (evg, next, id_node,
5426 &table->eviction_groups_by_id) {
5427 eviction_group_destroy(table, evg);
5429 hmap_destroy(&table->eviction_groups_by_id);
5430 heap_destroy(&table->eviction_groups_by_size);
5432 free(table->eviction_fields);
5433 table->eviction_fields = NULL;
5434 table->n_eviction_fields = 0;
5438 /* oftables support a choice of two policies when adding a rule would cause the
5439 * number of flows in the table to exceed the configured maximum number: either
5440 * they can refuse to add the new flow or they can evict some existing flow.
5441 * This function configures the latter policy on 'table', with fairness based
5442 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
5443 * 'n_fields' as 0 disables fairness.) */
5445 oftable_enable_eviction(struct oftable *table,
5446 const struct mf_subfield *fields, size_t n_fields)
5448 struct cls_cursor cursor;
5451 if (table->eviction_fields
5452 && n_fields == table->n_eviction_fields
5454 || !memcmp(fields, table->eviction_fields,
5455 n_fields * sizeof *fields))) {
5460 oftable_disable_eviction(table);
5462 table->n_eviction_fields = n_fields;
5463 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
5465 table->eviction_group_id_basis = random_uint32();
5466 hmap_init(&table->eviction_groups_by_id);
5467 heap_init(&table->eviction_groups_by_size);
5469 ovs_rwlock_rdlock(&table->cls.rwlock);
5470 cls_cursor_init(&cursor, &table->cls, NULL);
5471 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
5472 eviction_group_add_rule(rule);
5474 ovs_rwlock_unlock(&table->cls.rwlock);
5477 /* Removes 'rule' from the oftable that contains it. */
5479 oftable_remove_rule__(struct ofproto *ofproto, struct classifier *cls,
5481 OVS_REQ_WRLOCK(cls->rwlock) OVS_RELEASES(rule->evict)
5483 classifier_remove(cls, &rule->cr);
5484 if (rule->meter_id) {
5485 list_remove(&rule->meter_list_node);
5487 cookies_remove(ofproto, rule);
5488 eviction_group_remove_rule(rule);
5489 ovs_mutex_lock(&ofproto->expirable_mutex);
5490 if (!list_is_empty(&rule->expirable)) {
5491 list_remove(&rule->expirable);
5493 ovs_mutex_unlock(&ofproto->expirable_mutex);
5494 if (!list_is_empty(&rule->meter_list_node)) {
5495 list_remove(&rule->meter_list_node);
5497 ovs_rwlock_unlock(&rule->evict);
5501 oftable_remove_rule(struct rule *rule)
5503 struct ofproto *ofproto = rule->ofproto;
5504 struct oftable *table = &ofproto->tables[rule->table_id];
5506 ovs_rwlock_wrlock(&table->cls.rwlock);
5507 oftable_remove_rule__(ofproto, &table->cls, rule);
5508 ovs_rwlock_unlock(&table->cls.rwlock);
5511 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
5512 * oftable that has an identical cls_rule. Returns the rule that was removed,
5513 * if any, and otherwise NULL. */
5514 static struct rule *
5515 oftable_replace_rule(struct rule *rule)
5517 struct ofproto *ofproto = rule->ofproto;
5518 struct oftable *table = &ofproto->tables[rule->table_id];
5519 struct rule *victim;
5522 ovs_mutex_lock(&rule->timeout_mutex);
5523 may_expire = rule->hard_timeout || rule->idle_timeout;
5524 ovs_mutex_unlock(&rule->timeout_mutex);
5527 ovs_mutex_lock(&ofproto->expirable_mutex);
5528 list_insert(&ofproto->expirable, &rule->expirable);
5529 ovs_mutex_unlock(&ofproto->expirable_mutex);
5531 cookies_insert(ofproto, rule);
5532 if (rule->meter_id) {
5533 struct meter *meter = ofproto->meters[rule->meter_id];
5534 list_insert(&meter->rules, &rule->meter_list_node);
5536 ovs_rwlock_wrlock(&table->cls.rwlock);
5537 victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr));
5538 ovs_rwlock_unlock(&table->cls.rwlock);
5540 if (victim->meter_id) {
5541 list_remove(&victim->meter_list_node);
5543 cookies_remove(ofproto, victim);
5545 ovs_mutex_lock(&ofproto->expirable_mutex);
5546 if (!list_is_empty(&victim->expirable)) {
5547 list_remove(&victim->expirable);
5549 ovs_mutex_unlock(&ofproto->expirable_mutex);
5550 eviction_group_remove_rule(victim);
5552 eviction_group_add_rule(rule);
5556 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
5558 oftable_substitute_rule(struct rule *old, struct rule *new)
5561 oftable_replace_rule(new);
5563 ovs_rwlock_wrlock(&old->evict);
5564 oftable_remove_rule(old);
5568 /* unixctl commands. */
5571 ofproto_lookup(const char *name)
5573 struct ofproto *ofproto;
5575 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
5577 if (!strcmp(ofproto->name, name)) {
5585 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
5586 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5588 struct ofproto *ofproto;
5592 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
5593 ds_put_format(&results, "%s\n", ofproto->name);
5595 unixctl_command_reply(conn, ds_cstr(&results));
5596 ds_destroy(&results);
5600 ofproto_unixctl_init(void)
5602 static bool registered;
5608 unixctl_command_register("ofproto/list", "", 0, 0,
5609 ofproto_unixctl_list, NULL);
5612 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5614 * This is deprecated. It is only for compatibility with broken device drivers
5615 * in old versions of Linux that do not properly support VLANs when VLAN
5616 * devices are not used. When broken device drivers are no longer in
5617 * widespread use, we will delete these interfaces. */
5619 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
5620 * (exactly) by an OpenFlow rule in 'ofproto'. */
5622 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
5624 const struct oftable *oftable;
5626 free(ofproto->vlan_bitmap);
5627 ofproto->vlan_bitmap = bitmap_allocate(4096);
5628 ofproto->vlans_changed = false;
5630 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
5631 const struct cls_table *table;
5633 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
5634 if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
5635 const struct cls_rule *rule;
5637 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
5638 uint16_t vid = miniflow_get_vid(&rule->match.flow);
5639 bitmap_set1(vlan_bitmap, vid);
5640 bitmap_set1(ofproto->vlan_bitmap, vid);
5647 /* Returns true if new VLANs have come into use by the flow table since the
5648 * last call to ofproto_get_vlan_usage().
5650 * We don't track when old VLANs stop being used. */
5652 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
5654 return ofproto->vlans_changed;
5657 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
5658 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
5659 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
5660 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
5661 * then the VLAN device is un-enslaved. */
5663 ofproto_port_set_realdev(struct ofproto *ofproto, ofp_port_t vlandev_ofp_port,
5664 ofp_port_t realdev_ofp_port, int vid)
5666 struct ofport *ofport;
5669 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
5671 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
5673 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
5674 ofproto->name, vlandev_ofp_port);
5678 if (!ofproto->ofproto_class->set_realdev) {
5679 if (!vlandev_ofp_port) {
5682 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
5686 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
5688 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
5689 ofproto->name, vlandev_ofp_port,
5690 netdev_get_name(ofport->netdev), ovs_strerror(error));